{
  "metadata": {
    "timestamp": "2026-09-05T13:01:03.947452+00:00",
    "platform": "macOS-26.6.2-arm64-arm-64bit-Mach-O",
    "torch_version": "2.14.0",
    "torch_git_version": "08187d9e0fba026dc8217405802ab5381dc88d90",
    "torch_config": "PyTorch built with:\n  - GCC 4.2\n  - C++ Version: 202002\n  - clang 21.0.0\n  - OpenMP 202011\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: DEFAULT\n  - Build settings: BLAS_INFO=accelerate, BUILD_TYPE=Release, COMMIT_SHA=08187d9e0fba026dc8217405802ab5381dc88d90, CUDA_FLAGS= -DLIBCUDACXX_ENABLE_SIMPLIFIED_COMPLEX_OPERATIONS -Xfatbin -compress-all  -Wno-deprecated-gpu-targets --expt-extended-lambda -DCUB_WRAPPED_NAMESPACE=at_cuda_detail -DDISABLE_CUSPARSE_DEPRECATED -DCUDA_HAS_FP16=1 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -DC10_NODEPRECATED, CXX_COMPILER=/usr/bin/c++, CXX_FLAGS= -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DUSE_PYTORCH_QNNPACK -DAT_BUILD_ARM_VEC256_WITH_SLEEF -DUSE_PYTORCH_METAL_EXPORT -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DUSE_COREML_DELEGATE -O2 -fPIC -DC10_NODEPRECATED -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=braced-scalar-init -Werror=range-loop-construct -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-unknown-pragmas -Wno-unused-parameter -Wno-strict-overflow -Wno-strict-aliasing -Wvla-extension -Wsuggest-override -Wnewline-eof -Winconsistent-missing-override -Winconsistent-missing-destructor-override -Wno-pass-failed -Wno-error=old-style-cast -Wconstant-conversion -Qunused-arguments -faligned-new -fno-math-errno -fno-trapping-math -Werror=format -fclang-abi-compat=17 -DUSE_MPS -Wno-missing-braces, LAPACK_INFO=accelerate, TORCH_VERSION=2.14.0, USE_CUDA=OFF, USE_CUDNN=OFF, USE_CUSPARSELT=OFF, USE_EIGEN_FOR_BLAS=ON, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=ON, USE_HIPSPARSELT=OFF, USE_MKL=OFF, USE_MKLDNN=OFF, USE_MPI=OFF, USE_NCCL=OFF, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF, USE_XCCL=OFF, USE_XPU=OFF, \n",
    "threads": 8,
    "thread_environment": {
      "TVM_NUM_THREADS": "8",
      "OMP_NUM_THREADS": "8",
      "VECLIB_MAXIMUM_THREADS": "8"
    },
    "git_revision": "86dbcb072d0a4a166c91aeb237f943a39e9d0f17",
    "worktree_dirty": true,
    "removed_environment": {
      "PYTORCH_MPS_FAST_MATH": null,
      "PYTORCH_MPS_PREFER_METAL": null,
      "PYTORCH_ENABLE_MPS_FALLBACK": null,
      "LUISA_ENABLE_VALIDATION": null,
      "MTL_DEBUG_LAYER": null,
      "MTL_SHADER_VALIDATION": null,
      "DYLD_PRINT_LIBRARIES": null,
      "LUISA_TILE_BENCH_DUMP_SOURCE": null
    },
    "artifacts_sha256": {
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx": "9cbdc7873355118a9874c58eec499cdb0692dd9286c14b52afceae620c62ad87",
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libSPIRV.16.3.0.dylib": "c0ff450913316a86f6544729b2efc15b23ca1a73994d6b19c20d715ec8646901",
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libglslang-default-resource-limits.16.3.0.dylib": "bf2ea542cee8832f0638b0db27ddb23fac6488060a367f2f10c3e98e3fb88999",
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libglslang.16.3.0.dylib": "bbdbf823956fb94d1131177bb87c90b82209fdcbaced1466bbf337386048459d",
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libluisa-ast.dylib": "c90d64cebb541f658c5df764b3e05ec61cffbc884a01662eed43ba22620cfea7",
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libluisa-backend-metal.so": "e5a069b6b0381058cef33aec8c0ea6788fe5536072f886b36e0253e2258859dd",
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libluisa-benchmark-metal-timing.dylib": "eb568b9177926113e8af2ea1ec995b9c90c13623fc1e73d483864c0b8217eae0",
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libluisa-core.dylib": "89ba28beebc6d47fee75f43b1f51e678ba35a0dcf728bb6f4a53ae3791adb87a",
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libluisa-coro.dylib": "bbfb0d881eb8352ff99b95935619fcd8c36ab9c934c18a541a2118d320496487",
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libluisa-dsl.dylib": "20ac9be7d1393baea0cb951872134db82057c1a71cd5118b3363a40e8c144ff5",
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libluisa-osl.dylib": "0239b750e65ce07da91de82175f2577004a3a5cc62746562d73a2cc98cda9929",
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libluisa-runtime.dylib": "371b29d248be11edba1a5fda9e859b9ac06cc2f87361d74a4bdda375a06cf799",
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libluisa-tile-bridge-tirx.dylib": "1a06ad07d6d4eeb4d645b6324ea3719b5192940537da786df65860c77754b420",
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libluisa-tile-bridge-xir.dylib": "77b6343363ba3b2dc242d7c0a22d4b68c277b057a0795f52ba98090200954d95",
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libluisa-tile.dylib": "3c826ec66f1c3bc303b7a3eaec0641cabc401fc5f9489c99f09201746e57883f",
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libluisa-validation-layer.so": "d86328e83c026f502b4df69f48d6b3fb756491a4900c244b99900146e812f282",
      "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libluisa-xir.dylib": "9efd84887588f8943c9e0c25cc59eaf45ad4f860f738f9cb61cf817e88a831d9",
      "/private/tmp/luisa-tvmx-venv/lib/python3.14/site-packages/tvm/lib/libtvm_compiler.dylib": "44a277c13f8400925b6eb7170148b0c0e03ca727a70d29f33713d0bc0c8d5c89",
      "/private/tmp/luisa-tvmx-venv/lib/python3.14/site-packages/tvm/lib/libtvm_runtime.dylib": "bf5e2c96f4f946e27d590f99d859860d3c6a8a7803b20bbd17cd3534bb759f9d",
      "/private/tmp/luisa-tvmx-venv/lib/python3.14/site-packages/tvm/lib/libtvm_runtime_metal.dylib": "83dc06e34c4a531f0eb0ab1912579c8ee7c2a05f2736d7cdfd7d7d9ce700ac60",
      "/private/tmp/luisa-tvmx-venv/lib/python3.14/site-packages/tvm_ffi/lib/libtvm_ffi.dylib": "3dec87d55293c7dea456d66be3d26060b53a769c0518f6c7c18b488dad0fd6cd"
    },
    "capture_sources": true,
    "metal_device_timing": "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/libluisa-benchmark-metal-timing.dylib",
    "native_sha256": "9cbdc7873355118a9874c58eec499cdb0692dd9286c14b52afceae620c62ad87",
    "adjacent_tile_library_sha256": {
      "libluisa-tile-bridge-tirx.dylib": "1a06ad07d6d4eeb4d645b6324ea3719b5192940537da786df65860c77754b420",
      "libluisa-tile-bridge-xir.dylib": "77b6343363ba3b2dc242d7c0a22d4b68c277b057a0795f52ba98090200954d95",
      "libluisa-tile.dylib": "3c826ec66f1c3bc303b7a3eaec0641cabc401fc5f9489c99f09201746e57883f"
    },
    "native_variants": {
      "reference": {
        "binary": "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "sha256": "9cbdc7873355118a9874c58eec499cdb0692dd9286c14b52afceae620c62ad87",
        "adjacent_tile_library_sha256": {
          "libluisa-tile-bridge-tirx.dylib": "1a06ad07d6d4eeb4d645b6324ea3719b5192940537da786df65860c77754b420",
          "libluisa-tile-bridge-xir.dylib": "77b6343363ba3b2dc242d7c0a22d4b68c277b057a0795f52ba98090200954d95",
          "libluisa-tile.dylib": "3c826ec66f1c3bc303b7a3eaec0641cabc401fc5f9489c99f09201746e57883f"
        }
      },
      "candidate": {
        "binary": "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "sha256": "9cbdc7873355118a9874c58eec499cdb0692dd9286c14b52afceae620c62ad87",
        "adjacent_tile_library_sha256": {
          "libluisa-tile-bridge-tirx.dylib": "1a06ad07d6d4eeb4d645b6324ea3719b5192940537da786df65860c77754b420",
          "libluisa-tile-bridge-xir.dylib": "77b6343363ba3b2dc242d7c0a22d4b68c277b057a0795f52ba98090200954d95",
          "libluisa-tile.dylib": "3c826ec66f1c3bc303b7a3eaec0641cabc401fc5f9489c99f09201746e57883f"
        }
      }
    },
    "rounds": 4,
    "samples": 9,
    "sample_ms": 30,
    "warmup_ms": 200,
    "candidate_vector_mode": "reported",
    "candidate_subgroup_fences": "reported",
    "candidate_cpu_stack_bytes": null,
    "shared_cpu_stack_bytes": null,
    "candidate_cpu_vector_lanes": null,
    "shared_cpu_vector_lanes": null,
    "candidate_cpu_input_views": false,
    "candidate_cpu_model": null,
    "timing": "synchronized device-resident host wall time including dispatch; no profiler",
    "source_reports": {
      "reference": {
        "path": "scripts/benchmark/tile_torch/results/m1-max-20260905-cooperating-packing/reference/results.json",
        "sha256": "9db930c33983bd52311fc19417706a59cb4526b1ec4887994489672fbde58cc3"
      },
      "candidate": {
        "path": "scripts/benchmark/tile_torch/results/m1-max-20260905-cooperating-packing/candidate/results.json",
        "sha256": "898a666aa7a9b99e42c06ab658958f35e9b5f663277924b880ee2ecd5937c344"
      }
    },
    "frozen_plans": {
      "reference": [
        {
          "backend": "metal",
          "name": "softmax_37x1537",
          "case": {
            "operation": "softmax",
            "m": 37,
            "n": 1537,
            "k": 1
          },
          "gemm_block": [
            1,
            1537,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 256,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 1,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "softmax_256x3072",
          "case": {
            "operation": "softmax",
            "m": 256,
            "n": 3072,
            "k": 1
          },
          "gemm_block": [
            1,
            3072,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 256,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 1,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "softmax_768x6144",
          "case": {
            "operation": "softmax",
            "m": 768,
            "n": 6144,
            "k": 1
          },
          "gemm_block": [
            1,
            6144,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 256,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 1,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "softmax_1024x4096",
          "case": {
            "operation": "softmax",
            "m": 1024,
            "n": 4096,
            "k": 1
          },
          "gemm_block": [
            1,
            4096,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 256,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 1,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "rmsnorm_37x1537",
          "case": {
            "operation": "rmsnorm",
            "m": 37,
            "n": 1537,
            "k": 1
          },
          "gemm_block": [
            1,
            1537,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 256,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 1,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "rmsnorm_256x3072",
          "case": {
            "operation": "rmsnorm",
            "m": 256,
            "n": 3072,
            "k": 1
          },
          "gemm_block": [
            1,
            3072,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 256,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 1,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "rmsnorm_768x6144",
          "case": {
            "operation": "rmsnorm",
            "m": 768,
            "n": 6144,
            "k": 1
          },
          "gemm_block": [
            1,
            6144,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 256,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 1,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "rmsnorm_1024x4096",
          "case": {
            "operation": "rmsnorm",
            "m": 1024,
            "n": 4096,
            "k": 1
          },
          "gemm_block": [
            1,
            4096,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 256,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 1,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "layernorm_37x1537",
          "case": {
            "operation": "layernorm",
            "m": 37,
            "n": 1537,
            "k": 1
          },
          "gemm_block": [
            1,
            1537,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 256,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 1,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "layernorm_256x3072",
          "case": {
            "operation": "layernorm",
            "m": 256,
            "n": 3072,
            "k": 1
          },
          "gemm_block": [
            1,
            3072,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 256,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 1,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "layernorm_768x6144",
          "case": {
            "operation": "layernorm",
            "m": 768,
            "n": 6144,
            "k": 1
          },
          "gemm_block": [
            1,
            6144,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 256,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 1,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "layernorm_1024x4096",
          "case": {
            "operation": "layernorm",
            "m": 1024,
            "n": 4096,
            "k": 1
          },
          "gemm_block": [
            1,
            4096,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 256,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 1,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        }
      ],
      "candidate": [
        {
          "backend": "metal",
          "name": "softmax_37x1537",
          "case": {
            "operation": "softmax",
            "m": 37,
            "n": 1537,
            "k": 1
          },
          "gemm_block": [
            1,
            1537,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 512,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 2,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "softmax_256x3072",
          "case": {
            "operation": "softmax",
            "m": 256,
            "n": 3072,
            "k": 1
          },
          "gemm_block": [
            1,
            3072,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 512,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 2,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "softmax_768x6144",
          "case": {
            "operation": "softmax",
            "m": 768,
            "n": 6144,
            "k": 1
          },
          "gemm_block": [
            1,
            6144,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 512,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 2,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "softmax_1024x4096",
          "case": {
            "operation": "softmax",
            "m": 1024,
            "n": 4096,
            "k": 1
          },
          "gemm_block": [
            1,
            4096,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 512,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 2,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "rmsnorm_37x1537",
          "case": {
            "operation": "rmsnorm",
            "m": 37,
            "n": 1537,
            "k": 1
          },
          "gemm_block": [
            1,
            1537,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 512,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 2,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "rmsnorm_256x3072",
          "case": {
            "operation": "rmsnorm",
            "m": 256,
            "n": 3072,
            "k": 1
          },
          "gemm_block": [
            1,
            3072,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 512,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 2,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "rmsnorm_768x6144",
          "case": {
            "operation": "rmsnorm",
            "m": 768,
            "n": 6144,
            "k": 1
          },
          "gemm_block": [
            1,
            6144,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 512,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 2,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "rmsnorm_1024x4096",
          "case": {
            "operation": "rmsnorm",
            "m": 1024,
            "n": 4096,
            "k": 1
          },
          "gemm_block": [
            1,
            4096,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 512,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 2,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "layernorm_37x1537",
          "case": {
            "operation": "layernorm",
            "m": 37,
            "n": 1537,
            "k": 1
          },
          "gemm_block": [
            1,
            1537,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 512,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 2,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "layernorm_256x3072",
          "case": {
            "operation": "layernorm",
            "m": 256,
            "n": 3072,
            "k": 1
          },
          "gemm_block": [
            1,
            3072,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 512,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 2,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "layernorm_768x6144",
          "case": {
            "operation": "layernorm",
            "m": 768,
            "n": 6144,
            "k": 1
          },
          "gemm_block": [
            1,
            6144,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 512,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 2,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        },
        {
          "backend": "metal",
          "name": "layernorm_1024x4096",
          "case": {
            "operation": "layernorm",
            "m": 1024,
            "n": 4096,
            "k": 1
          },
          "gemm_block": [
            1,
            4096,
            1
          ],
          "execution_scope": "auto",
          "pipeline_window": 2,
          "cooperative_matrix": false,
          "no_vectorize": false,
          "auto_vectorize": false,
          "group_threads": 512,
          "copy_batch": 1,
          "cpu_stack_bytes": 0,
          "cpu_vector_lanes": 16,
          "cpu_input_views": false,
          "input_views": false,
          "cpu_model": null,
          "cpu_matrix_backend": "reference",
          "cpu_math_backend": "reference",
          "shared_tile_materialization": "preserve",
          "metal_subgroup_reductions": true,
          "reduction_programs_per_group": 2,
          "reduction_unroll": 1,
          "reduction_lane_elements": 4,
          "cache_reduction_inputs": true,
          "reduction_cost_profile": "analytic",
          "element_grid": "auto",
          "expected_cpu_model": null,
          "elide_independent_subgroup_barriers": false,
          "matrix_realization": "simdgroup"
        }
      ]
    },
    "artifacts_unchanged": true
  },
  "results": [
    {
      "case": {
        "operation": "softmax",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "softmax_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-bdkk4hy3/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "1ca0ba2a59ee9b200a0f7933e3c823dd873803c26a844f54bebdc043f01bc250",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.454666,
        "capture_ms": 0.073958,
        "compile_ms": 55.092542,
        "allocation_upload_ms": 1.251292,
        "cold_call_ms": 2.818083,
        "warmup_ms": 200.451625,
        "download_ms": 0.502,
        "repetitions": 4892,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          5.7342260834,
          5.79366823385,
          5.55178515944,
          5.8515944399,
          5.58693601799,
          5.56814697465,
          6.05683605887,
          5.59391189697,
          5.58516455437
        ],
        "latency_us": [
          256.291,
          452.333,
          1981.833,
          390.667,
          307.125,
          477.083,
          247.916,
          213.375,
          237.417
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 37,
            "reduction_scalar_rounds": 40,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 6148,
              "global_write_bytes": 6148,
              "private_read_bytes": 36888,
              "private_write_bytes": 12296
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 32,
              "global_write_bytes": 32,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 2,
            "reduction_elements": 3074,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 88,
            "normalized_kernel_cost": 88,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 4611,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 242833,
              "compute_span_ns": 242833,
              "calibration_cpu_ns": 689458,
              "calibration_gpu_ticks": 689458,
              "compute_passes": 1,
              "command_buffer_ns": 242916.692514,
              "command_buffers": 2
            },
            {
              "compute_ns": 241083,
              "compute_span_ns": 241083,
              "calibration_cpu_ns": 710333,
              "calibration_gpu_ticks": 710333,
              "compute_passes": 1,
              "command_buffer_ns": 241125.002503,
              "command_buffers": 2
            },
            {
              "compute_ns": 260667,
              "compute_span_ns": 260667,
              "calibration_cpu_ns": 729333,
              "calibration_gpu_ticks": 729333,
              "compute_passes": 1,
              "command_buffer_ns": 260750.006419,
              "command_buffers": 2
            },
            {
              "compute_ns": 634958,
              "compute_span_ns": 634958,
              "calibration_cpu_ns": 2097458,
              "calibration_gpu_ticks": 2097458,
              "compute_passes": 1,
              "command_buffer_ns": 634999.94576,
              "command_buffers": 2
            },
            {
              "compute_ns": 715625,
              "compute_span_ns": 715625,
              "calibration_cpu_ns": 2022334,
              "calibration_gpu_ticks": 2022334,
              "compute_passes": 1,
              "command_buffer_ns": 715666.683391,
              "command_buffers": 2
            },
            {
              "compute_ns": 242709,
              "compute_span_ns": 242709,
              "calibration_cpu_ns": 692208,
              "calibration_gpu_ticks": 692208,
              "compute_passes": 1,
              "command_buffer_ns": 242875.015829,
              "command_buffers": 2
            },
            {
              "compute_ns": 282167,
              "compute_span_ns": 282167,
              "calibration_cpu_ns": 769750,
              "calibration_gpu_ticks": 769750,
              "compute_passes": 1,
              "command_buffer_ns": 282333.348878,
              "command_buffers": 2
            },
            {
              "compute_ns": 262333,
              "compute_span_ns": 262333,
              "calibration_cpu_ns": 829167,
              "calibration_gpu_ticks": 829167,
              "compute_passes": 1,
              "command_buffer_ns": 262374.931481,
              "command_buffers": 2
            },
            {
              "compute_ns": 241666,
              "compute_span_ns": 241666,
              "calibration_cpu_ns": 896750,
              "calibration_gpu_ticks": 896750,
              "compute_passes": 1,
              "command_buffer_ns": 241625.006311,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 7250,
              "compute_span_ns": 7250,
              "calibration_cpu_ns": 789250,
              "calibration_gpu_ticks": 789250,
              "compute_passes": 1,
              "command_buffer_ns": 7375.02705306,
              "command_buffers": 2
            },
            {
              "compute_ns": 6083,
              "compute_span_ns": 6083,
              "calibration_cpu_ns": 426083,
              "calibration_gpu_ticks": 426083,
              "compute_passes": 1,
              "command_buffer_ns": 6166.69422016,
              "command_buffers": 2
            },
            {
              "compute_ns": 9125,
              "compute_span_ns": 9125,
              "calibration_cpu_ns": 1193500,
              "calibration_gpu_ticks": 1193500,
              "compute_passes": 1,
              "command_buffer_ns": 9166.65885597,
              "command_buffers": 2
            },
            {
              "compute_ns": 7375,
              "compute_span_ns": 7375,
              "calibration_cpu_ns": 773666,
              "calibration_gpu_ticks": 773666,
              "compute_passes": 1,
              "command_buffer_ns": 7500.0571087,
              "command_buffers": 2
            },
            {
              "compute_ns": 30167,
              "compute_span_ns": 30167,
              "calibration_cpu_ns": 1327791,
              "calibration_gpu_ticks": 1327791,
              "compute_passes": 1,
              "command_buffer_ns": 30249.9393001,
              "command_buffers": 2
            },
            {
              "compute_ns": 7375,
              "compute_span_ns": 7375,
              "calibration_cpu_ns": 941125,
              "calibration_gpu_ticks": 941125,
              "compute_passes": 1,
              "command_buffer_ns": 7416.70373827,
              "command_buffers": 2
            },
            {
              "compute_ns": 61750,
              "compute_span_ns": 61750,
              "calibration_cpu_ns": 1241583,
              "calibration_gpu_ticks": 1241583,
              "compute_passes": 1,
              "command_buffer_ns": 61833.3579041,
              "command_buffers": 2
            },
            {
              "compute_ns": 7333,
              "compute_span_ns": 7333,
              "calibration_cpu_ns": 773334,
              "calibration_gpu_ticks": 773334,
              "compute_passes": 1,
              "command_buffer_ns": 7374.9688454,
              "command_buffers": 2
            },
            {
              "compute_ns": 26416,
              "compute_span_ns": 26416,
              "calibration_cpu_ns": 1375416,
              "calibration_gpu_ticks": 1375416,
              "compute_passes": 1,
              "command_buffer_ns": 26458.3504759,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 508375.000209,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 646583.328489,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 262124.987785,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 242708.367296,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 240624.998696,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 240791.705437,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 679749.937262,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 621958.286501,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 241583.271418,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 9749.95782599,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 13958.2552947,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6708.37471262,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6083.34084973,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6499.99128655,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6250.04759058,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6416.69612378,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6208.31269771,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6666.63981974,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              7.943359378265625,
              10.102864507640625,
              4.0957029341406255,
              3.792318239,
              3.759765604625,
              3.762370397453125,
              10.62109276971875,
              9.718098226578126,
              3.7747386159062497
            ],
            "command_buffer_throughput_us_p50": 4.0957029341406255,
            "command_buffer_throughput_us_p90": 10.20651016005625,
            "command_buffer_latency_us": [
              9.74995782599,
              13.9582552947,
              6.7083747126199995,
              6.08334084973,
              6.49999128655,
              6.25004759058,
              6.4166961237799995,
              6.20831269771,
              6.66663981974
            ],
            "command_buffer_latency_us_p50": 6.49999128655,
            "command_buffer_latency_us_p90": 10.591617319732
          },
          "compute_throughput_us": [
            3.794265625,
            3.766921875,
            4.072921875,
            9.92121875,
            11.181640625,
            3.792328125,
            4.408859375,
            4.098953125,
            3.77603125
          ],
          "compute_throughput_us_p50": 4.072921875,
          "compute_throughput_us_p90": 10.173303125,
          "command_buffer_throughput_us": [
            3.79557332053125,
            3.767578164109375,
            4.074218850296875,
            9.921874152500001,
            11.182291927984375,
            3.794922122328125,
            4.41145857621875,
            4.099608304390625,
            3.775390723609375
          ],
          "command_buffer_throughput_us_p50": 4.074218850296875,
          "command_buffer_throughput_us_p90": 10.173957707596877,
          "compute_latency_us": [
            7.25,
            6.083,
            9.125,
            7.375,
            30.167,
            7.375,
            61.75,
            7.333,
            26.416
          ],
          "compute_latency_us_p50": 7.375,
          "compute_latency_us_p90": 36.48360000000001,
          "command_buffer_latency_us": [
            7.37502705306,
            6.16669422016,
            9.166658855969999,
            7.5000571087,
            30.2499393001,
            7.41670373827,
            61.8333579041,
            7.3749688454,
            26.458350475899998
          ],
          "command_buffer_latency_us_p50": 7.5000571087,
          "command_buffer_latency_us_p90": 36.5666230209,
          "counter_control_throughput_ratio": 0.9947544819072046,
          "counter_control_latency_ratio": 1.2328845767425443
        },
        "correctness": {
          "max_abs_error": 2.6856349761260334e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 5.59391189697,
        "throughput_us_p90": 5.892642763694,
        "latency_us_p50": 307.125,
        "latency_us_p90": 778.0330000000004
      },
      "torch": {
        "cold_call_ms": 52.690458,
        "warmup_ms": 201.158834,
        "repetitions": 884,
        "throughput_us": [
          32.890789592760186,
          31.201923076923077,
          30.148472850678733,
          33.456259049773756,
          30.68627488687783,
          29.66963574660634,
          28.952064479638008,
          32.25358257918552,
          30.46431900452489
        ],
        "latency_us": [
          344.708,
          402.167,
          340.667,
          311.0,
          321.0,
          345.04200000000003,
          330.667,
          2338.834,
          581.0419999999999
        ],
        "throughput_us_p50": 30.68627488687783,
        "throughput_us_p90": 33.0038834841629,
        "latency_us_p50": 344.708,
        "latency_us_p90": 932.6004000000003,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1053083.3969824016,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1075541.7170003057,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1039749.8263046145,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1139624.9756217003,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1040291.681420058,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1071541.6283346713,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1040374.9765828252,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1099000.044632703,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1038958.3185315132,
                "command_buffers": 7
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 85000.0069476664,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 53833.29698815942,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 71458.3438821137,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 54000.00372901559,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 131916.66221246123,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 57625.0022277236,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 57958.357501775026,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 54124.9755769968,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 100166.7114906013,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              16.454428077850025,
              16.805339328129776,
              16.246091036009602,
              17.806640244089067,
              16.254557522188406,
              16.74283794272924,
              16.255859009106643,
              17.171875697385985,
              16.233723727054894
            ],
            "command_buffer_throughput_us_p50": 16.454428077850025,
            "command_buffer_throughput_us_p90": 17.2988286067266,
            "command_buffer_latency_us": [
              85.0000069476664,
              53.83329698815942,
              71.4583438821137,
              54.00000372901559,
              131.91666221246123,
              57.6250022277236,
              57.958357501775026,
              54.1249755769968,
              100.1667114906013
            ],
            "command_buffer_latency_us_p50": 57.958357501775026,
            "command_buffer_latency_us_p90": 106.51670163497329
          },
          "throughput": [
            {
              "compute_ns": 3486751.0,
              "compute_span_ns": 4592417.0,
              "command_buffer_ns": 4423375.008627772,
              "calibration_cpu_ns": 6099209.0,
              "calibration_gpu_ticks": 6099209.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3495542.0,
              "compute_span_ns": 4603667.0,
              "command_buffer_ns": 4447125.014849007,
              "calibration_cpu_ns": 6084458.0,
              "calibration_gpu_ticks": 6084458.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3513044.0,
              "compute_span_ns": 4719458.0,
              "command_buffer_ns": 4604750.080034137,
              "calibration_cpu_ns": 6055792.0,
              "calibration_gpu_ticks": 6055792.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3501080.0,
              "compute_span_ns": 4596375.0,
              "command_buffer_ns": 4448875.0281743705,
              "calibration_cpu_ns": 6220084.0,
              "calibration_gpu_ticks": 6220084.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3483668.0,
              "compute_span_ns": 4593541.0,
              "command_buffer_ns": 4490458.290092647,
              "calibration_cpu_ns": 6128250.0,
              "calibration_gpu_ticks": 6128250.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3523911.0,
              "compute_span_ns": 4511791.0,
              "command_buffer_ns": 4417999.996803701,
              "calibration_cpu_ns": 6022583.0,
              "calibration_gpu_ticks": 6022583.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3496584.0,
              "compute_span_ns": 4649875.0,
              "command_buffer_ns": 4550625.046249479,
              "calibration_cpu_ns": 6209750.0,
              "calibration_gpu_ticks": 6209750.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3500957.0,
              "compute_span_ns": 4580042.0,
              "command_buffer_ns": 4474791.639950126,
              "calibration_cpu_ns": 6097667.0,
              "calibration_gpu_ticks": 6097667.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3498997.0,
              "compute_span_ns": 4568042.0,
              "command_buffer_ns": 4436916.555278003,
              "calibration_cpu_ns": 6207250.0,
              "calibration_gpu_ticks": 6207250.0,
              "compute_passes": 64,
              "command_buffers": 7
            }
          ],
          "latency": [
            {
              "compute_ns": 59083.0,
              "compute_span_ns": 59083.0,
              "command_buffer_ns": 59083.33696424961,
              "calibration_cpu_ns": 962958.0,
              "calibration_gpu_ticks": 962958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 56833.0,
              "compute_span_ns": 56833.0,
              "command_buffer_ns": 56833.31983163953,
              "calibration_cpu_ns": 564917.0,
              "calibration_gpu_ticks": 564917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 54750.0,
              "compute_span_ns": 54750.0,
              "command_buffer_ns": 54750.009439885616,
              "calibration_cpu_ns": 957291.0,
              "calibration_gpu_ticks": 957291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 54709.0,
              "compute_span_ns": 54709.0,
              "command_buffer_ns": 54750.009439885616,
              "calibration_cpu_ns": 531375.0,
              "calibration_gpu_ticks": 531375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 53917.0,
              "compute_span_ns": 53917.0,
              "command_buffer_ns": 53958.327043801546,
              "calibration_cpu_ns": 964875.0,
              "calibration_gpu_ticks": 964875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 57000.0,
              "compute_span_ns": 57000.0,
              "command_buffer_ns": 57000.0265724957,
              "calibration_cpu_ns": 504083.0,
              "calibration_gpu_ticks": 504083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 58125.0,
              "compute_span_ns": 58125.0,
              "command_buffer_ns": 58125.00603497028,
              "calibration_cpu_ns": 558792.0,
              "calibration_gpu_ticks": 558792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 55375.0,
              "compute_span_ns": 55375.0,
              "command_buffer_ns": 55374.985095113516,
              "calibration_cpu_ns": 603667.0,
              "calibration_gpu_ticks": 603667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 56916.0,
              "compute_span_ns": 56916.0,
              "command_buffer_ns": 56874.99651685357,
              "calibration_cpu_ns": 925125.0,
              "calibration_gpu_ticks": 925125.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            54.480484375,
            54.61784375,
            54.8913125,
            54.704375,
            54.4323125,
            55.061109375,
            54.634125,
            54.702453125,
            54.671828125
          ],
          "compute_throughput_us_p50": 54.671828125,
          "compute_throughput_us_p90": 54.925271875,
          "command_buffer_throughput_us": [
            69.11523450980894,
            69.48632835701574,
            71.94922000053339,
            69.51367231522454,
            70.16341078269761,
            69.03124995005783,
            71.10351634764811,
            69.91861937422073,
            69.3268211762188
          ],
          "command_buffer_throughput_us_p50": 69.51367231522454,
          "command_buffer_throughput_us_p90": 71.27265707822517,
          "compute_latency_us": [
            59.083,
            56.833,
            54.75,
            54.709,
            53.917,
            57.0,
            58.125,
            55.375,
            56.916
          ],
          "compute_latency_us_p50": 56.833,
          "compute_latency_us_p90": 58.3166,
          "command_buffer_latency_us": [
            59.08333696424961,
            56.83331983163953,
            54.750009439885616,
            54.750009439885616,
            53.958327043801546,
            57.0000265724957,
            58.12500603497028,
            55.374985095113516,
            56.87499651685357
          ],
          "command_buffer_latency_us_p50": 56.83331983163953,
          "command_buffer_latency_us_p90": 58.31667222082615,
          "counter_control_throughput_ratio": 4.200403331115942,
          "counter_control_latency_ratio": 0.9891544359033929
        },
        "download_ms": 0.315709,
        "allocation_upload_ms": 4.196667,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 4.2122453832241225e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.182293612293817,
      "device_compute_slowdown": 0.07449763460786048,
      "valid": true,
      "round": 0,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "softmax_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 1.96275,
        "warmup_ms": 200.669417,
        "repetitions": 913,
        "throughput_us": [
          33.62267250821468,
          31.059191675794086,
          33.51620153340635,
          30.31147316538883,
          32.62732749178532,
          30.261318729463305,
          29.285779846659366,
          31.65197152245345,
          37.89882256297919
        ],
        "latency_us": [
          350.95799999999997,
          360.625,
          356.45799999999997,
          327.458,
          337.75,
          313.625,
          418.833,
          292.66700000000003,
          300.875
        ],
        "throughput_us_p50": 31.65197152245345,
        "throughput_us_p90": 34.47790251916758,
        "latency_us_p50": 337.75,
        "latency_us_p90": 372.26660000000004,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1080750.022083521,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1023291.7265966535,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1029583.4508724511,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1025750.0107400119,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1041458.3957754076,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1049416.6053831577,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1020708.295982331,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1034500.1355744898,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1021500.036586076,
                "command_buffers": 7
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 60458.31833034754,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 54583.302699029446,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 54000.00372901559,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 58749.98169019818,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 59375.015553086996,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 54500.007536262274,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 57708.355598151684,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 55374.985095113516,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 54583.302699029446,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              16.886719095055014,
              15.98893322807271,
              16.087241419882048,
              16.027343917812686,
              16.272787433990743,
              16.39713445911184,
              15.948567124723922,
              16.164064618351404,
              15.960938071657438
            ],
            "command_buffer_throughput_us_p50": 16.087241419882048,
            "command_buffer_throughput_us_p90": 16.495051386300474,
            "command_buffer_latency_us": [
              60.45831833034754,
              54.583302699029446,
              54.00000372901559,
              58.74998169019818,
              59.375015553086996,
              54.500007536262274,
              57.708355598151684,
              55.374985095113516,
              54.583302699029446
            ],
            "command_buffer_latency_us_p50": 55.374985095113516,
            "command_buffer_latency_us_p90": 59.591676108539104
          },
          "throughput": [
            {
              "compute_ns": 3527793.0,
              "compute_span_ns": 4484125.0,
              "command_buffer_ns": 4385791.718959808,
              "calibration_cpu_ns": 6479250.0,
              "calibration_gpu_ticks": 6479250.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3524499.0,
              "compute_span_ns": 4857291.0,
              "command_buffer_ns": 4773875.058162957,
              "calibration_cpu_ns": 6858083.0,
              "calibration_gpu_ticks": 6858083.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3667836.0,
              "compute_span_ns": 4680833.0,
              "command_buffer_ns": 4594374.971929938,
              "calibration_cpu_ns": 6590083.0,
              "calibration_gpu_ticks": 6590083.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3494247.0,
              "compute_span_ns": 4543833.0,
              "command_buffer_ns": 4436124.989297241,
              "calibration_cpu_ns": 6599250.0,
              "calibration_gpu_ticks": 6599250.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3475039.0,
              "compute_span_ns": 4288375.0,
              "command_buffer_ns": 4195624.904241413,
              "calibration_cpu_ns": 6120458.0,
              "calibration_gpu_ticks": 6120458.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 5136918.0,
              "compute_span_ns": 6165917.0,
              "command_buffer_ns": 5839749.99608472,
              "calibration_cpu_ns": 9992541.0,
              "calibration_gpu_ticks": 9992541.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3513667.0,
              "compute_span_ns": 4461083.0,
              "command_buffer_ns": 4378916.695713997,
              "calibration_cpu_ns": 6887083.0,
              "calibration_gpu_ticks": 6887083.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3637044.0,
              "compute_span_ns": 4747084.0,
              "command_buffer_ns": 4659333.382733166,
              "calibration_cpu_ns": 7065375.0,
              "calibration_gpu_ticks": 7065375.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3727796.0,
              "compute_span_ns": 4743833.0,
              "command_buffer_ns": 4642250.016331673,
              "calibration_cpu_ns": 7429291.0,
              "calibration_gpu_ticks": 7429291.0,
              "compute_passes": 64,
              "command_buffers": 7
            }
          ],
          "latency": [
            {
              "compute_ns": 53417.0,
              "compute_span_ns": 53417.0,
              "command_buffer_ns": 53458.32323655486,
              "calibration_cpu_ns": 843375.0,
              "calibration_gpu_ticks": 843375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 54042.0,
              "compute_span_ns": 54042.0,
              "command_buffer_ns": 54083.29889178276,
              "calibration_cpu_ns": 620209.0,
              "calibration_gpu_ticks": 620209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 56333.0,
              "compute_span_ns": 56333.0,
              "command_buffer_ns": 56333.37423205376,
              "calibration_cpu_ns": 3197417.0,
              "calibration_gpu_ticks": 3197417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 61167.0,
              "compute_span_ns": 61167.0,
              "command_buffer_ns": 61166.705563664436,
              "calibration_cpu_ns": 1862166.0,
              "calibration_gpu_ticks": 1862166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 53625.0,
              "compute_span_ns": 53625.0,
              "command_buffer_ns": 53624.97176975012,
              "calibration_cpu_ns": 871958.0,
              "calibration_gpu_ticks": 871958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 56167.0,
              "compute_span_ns": 56167.0,
              "command_buffer_ns": 56166.667491197586,
              "calibration_cpu_ns": 1986584.0,
              "calibration_gpu_ticks": 1986584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 60584.0,
              "compute_span_ns": 60584.0,
              "command_buffer_ns": 60625.02507120371,
              "calibration_cpu_ns": 1863750.0,
              "calibration_gpu_ticks": 1863750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 53625.0,
              "compute_span_ns": 53625.0,
              "command_buffer_ns": 53625.02997741103,
              "calibration_cpu_ns": 847333.0,
              "calibration_gpu_ticks": 847333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 61666.0,
              "compute_span_ns": 61666.0,
              "command_buffer_ns": 61624.974478036165,
              "calibration_cpu_ns": 1969083.0,
              "calibration_gpu_ticks": 1969083.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            55.121765625,
            55.070296875,
            57.3099375,
            54.597609375,
            54.297484375,
            80.26434375,
            54.901046875,
            56.8288125,
            58.2468125
          ],
          "compute_throughput_us_p50": 55.121765625,
          "compute_throughput_us_p90": 62.65031875,
          "command_buffer_throughput_us": [
            68.527995608747,
            74.59179778379621,
            71.78710893640528,
            69.31445295776939,
            65.55663912877208,
            91.24609368882375,
            68.4205733705312,
            72.80208410520572,
            72.53515650518239
          ],
          "command_buffer_throughput_us_p50": 71.78710893640528,
          "command_buffer_throughput_us_p90": 77.92265696480172,
          "compute_latency_us": [
            53.417,
            54.042,
            56.333,
            61.167,
            53.625,
            56.167,
            60.584,
            53.625,
            61.666
          ],
          "compute_latency_us_p50": 56.167,
          "compute_latency_us_p90": 61.2668,
          "command_buffer_latency_us": [
            53.45832323655486,
            54.08329889178276,
            56.33337423205376,
            61.166705563664436,
            53.62497176975012,
            56.166667491197586,
            60.62502507120371,
            53.62502997741103,
            61.624974478036165
          ],
          "command_buffer_latency_us_p50": 56.166667491197586,
          "command_buffer_latency_us_p90": 61.25835934653878,
          "counter_control_throughput_ratio": 4.462362878926176,
          "counter_control_latency_ratio": 1.0305809123755878
        },
        "download_ms": 0.630542,
        "allocation_upload_ms": 0.920625,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 4.2122453832241225e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-a5we3h6i/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "d45e9e72129e03853b1c708b1bdcbc59be9547b0d0e02ca6bdc2eb0b41c08926",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 21.545125,
        "capture_ms": 0.079458,
        "compile_ms": 60.768416,
        "allocation_upload_ms": 1.039333,
        "cold_call_ms": 62.879958,
        "warmup_ms": 200.001917,
        "download_ms": 0.416375,
        "repetitions": 4941,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          6.49815320785,
          6.48848067193,
          5.96120056669,
          6.32830722526,
          6.48353066181,
          6.47326796195,
          5.95102206031,
          6.5340856102,
          6.42978830196
        ],
        "latency_us": [
          279.625,
          229.375,
          216.708,
          190.084,
          208.125,
          208.084,
          211,
          201.667,
          191.167
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 19,
            "reduction_scalar_rounds": 40,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 6148,
              "global_write_bytes": 6148,
              "private_read_bytes": 36888,
              "private_write_bytes": 12296
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 32,
              "global_write_bytes": 32,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 2,
            "reduction_elements": 3074,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 80,
            "normalized_kernel_cost": 80,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 4611,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 263666,
              "compute_span_ns": 263666,
              "calibration_cpu_ns": 744833,
              "calibration_gpu_ticks": 744833,
              "compute_passes": 1,
              "command_buffer_ns": 263624.999207,
              "command_buffers": 2
            },
            {
              "compute_ns": 641917,
              "compute_span_ns": 641917,
              "calibration_cpu_ns": 2036208,
              "calibration_gpu_ticks": 2036208,
              "compute_passes": 1,
              "command_buffer_ns": 641958.264168,
              "command_buffers": 2
            },
            {
              "compute_ns": 287709,
              "compute_span_ns": 287709,
              "calibration_cpu_ns": 857458,
              "calibration_gpu_ticks": 857458,
              "compute_passes": 1,
              "command_buffer_ns": 287791.597657,
              "command_buffers": 2
            },
            {
              "compute_ns": 264875,
              "compute_span_ns": 264875,
              "calibration_cpu_ns": 693500,
              "calibration_gpu_ticks": 693500,
              "compute_passes": 1,
              "command_buffer_ns": 264875.008725,
              "command_buffers": 2
            },
            {
              "compute_ns": 792250,
              "compute_span_ns": 792250,
              "calibration_cpu_ns": 2452916,
              "calibration_gpu_ticks": 2452916,
              "compute_passes": 1,
              "command_buffer_ns": 792333.332356,
              "command_buffers": 2
            },
            {
              "compute_ns": 285208,
              "compute_span_ns": 285208,
              "calibration_cpu_ns": 861000,
              "calibration_gpu_ticks": 861000,
              "compute_passes": 1,
              "command_buffer_ns": 285208.341666,
              "command_buffers": 2
            },
            {
              "compute_ns": 265750,
              "compute_span_ns": 265750,
              "calibration_cpu_ns": 716500,
              "calibration_gpu_ticks": 716500,
              "compute_passes": 1,
              "command_buffer_ns": 265750.044491,
              "command_buffers": 2
            },
            {
              "compute_ns": 658750,
              "compute_span_ns": 658750,
              "calibration_cpu_ns": 1866333,
              "calibration_gpu_ticks": 1866333,
              "compute_passes": 1,
              "command_buffer_ns": 658875.040244,
              "command_buffers": 2
            },
            {
              "compute_ns": 627250,
              "compute_span_ns": 627250,
              "calibration_cpu_ns": 1826875,
              "calibration_gpu_ticks": 1826875,
              "compute_passes": 1,
              "command_buffer_ns": 627291.62164,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 7000,
              "compute_span_ns": 7000,
              "calibration_cpu_ns": 751333,
              "calibration_gpu_ticks": 751333,
              "compute_passes": 1,
              "command_buffer_ns": 7041.72998667,
              "command_buffers": 2
            },
            {
              "compute_ns": 48292,
              "compute_span_ns": 48292,
              "calibration_cpu_ns": 1004333,
              "calibration_gpu_ticks": 1004333,
              "compute_passes": 1,
              "command_buffer_ns": 48458.3433717,
              "command_buffers": 2
            },
            {
              "compute_ns": 7250,
              "compute_span_ns": 7250,
              "calibration_cpu_ns": 448459,
              "calibration_gpu_ticks": 448459,
              "compute_passes": 1,
              "command_buffer_ns": 7291.67368263,
              "command_buffers": 2
            },
            {
              "compute_ns": 88083,
              "compute_span_ns": 88083,
              "calibration_cpu_ns": 1101500,
              "calibration_gpu_ticks": 1101500,
              "compute_passes": 1,
              "command_buffer_ns": 88208.2968019,
              "command_buffers": 2
            },
            {
              "compute_ns": 7208,
              "compute_span_ns": 7208,
              "calibration_cpu_ns": 461583,
              "calibration_gpu_ticks": 461583,
              "compute_passes": 1,
              "command_buffer_ns": 7291.61547497,
              "command_buffers": 2
            },
            {
              "compute_ns": 22666,
              "compute_span_ns": 22666,
              "calibration_cpu_ns": 1400875,
              "calibration_gpu_ticks": 1400875,
              "compute_passes": 1,
              "command_buffer_ns": 22749.9986067,
              "command_buffers": 2
            },
            {
              "compute_ns": 7125,
              "compute_span_ns": 7125,
              "calibration_cpu_ns": 410042,
              "calibration_gpu_ticks": 410042,
              "compute_passes": 1,
              "command_buffer_ns": 7125.02514943,
              "command_buffers": 2
            },
            {
              "compute_ns": 66625,
              "compute_span_ns": 66625,
              "calibration_cpu_ns": 1373541,
              "calibration_gpu_ticks": 1373541,
              "compute_passes": 1,
              "command_buffer_ns": 66708.3077133,
              "command_buffers": 2
            },
            {
              "compute_ns": 165208,
              "compute_span_ns": 165208,
              "calibration_cpu_ns": 1412000,
              "calibration_gpu_ticks": 1412000,
              "compute_passes": 1,
              "command_buffer_ns": 165291.596204,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 492250.022944,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 276041.624602,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 833833.299112,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 746708.305087,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 291958.334856,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 771125.021856,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 295166.682918,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 299291.685224,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 265791.662969,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 11291.7041406,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7666.64743423,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6583.34465697,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7583.35227147,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6750.05139783,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7541.67558625,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6541.66797176,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7625.02895668,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6625.02134219,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              7.6914066085,
              4.31315038440625,
              13.028645298625001,
              11.667317266984375,
              4.5618489821249995,
              12.0488284665,
              4.61197942059375,
              4.676432581625,
              4.152994733890625
            ],
            "command_buffer_throughput_us_p50": 4.676432581625,
            "command_buffer_throughput_us_p90": 12.244791832925,
            "command_buffer_latency_us": [
              11.2917041406,
              7.66664743423,
              6.58334465697,
              7.58335227147,
              6.750051397829999,
              7.54167558625,
              6.54166797176,
              7.6250289566800005,
              6.62502134219
            ],
            "command_buffer_latency_us_p50": 7.54167558625,
            "command_buffer_latency_us_p90": 8.391658775504
          },
          "compute_throughput_us": [
            4.11978125,
            10.029953125,
            4.495453125,
            4.138671875,
            12.37890625,
            4.456375,
            4.15234375,
            10.29296875,
            9.80078125
          ],
          "compute_throughput_us_p50": 4.495453125,
          "compute_throughput_us_p90": 10.71015625,
          "command_buffer_throughput_us": [
            4.1191406126093755,
            10.030597877624999,
            4.496743713390625,
            4.138672011328126,
            12.380208318062499,
            4.45638033853125,
            4.152344445171875,
            10.2949225038125,
            9.801431588125
          ],
          "command_buffer_throughput_us_p50": 4.496743713390625,
          "command_buffer_throughput_us_p90": 10.7119796666625,
          "compute_latency_us": [
            7.0,
            48.292,
            7.25,
            88.083,
            7.208,
            22.666,
            7.125,
            66.625,
            165.208
          ],
          "compute_latency_us_p50": 22.666,
          "compute_latency_us_p90": 103.50800000000001,
          "command_buffer_latency_us": [
            7.041729986669999,
            48.4583433717,
            7.29167368263,
            88.20829680189999,
            7.2916154749699995,
            22.7499986067,
            7.12502514943,
            66.7083077133,
            165.291596204
          ],
          "command_buffer_latency_us_p50": 22.7499986067,
          "command_buffer_latency_us_p90": 103.62495668232,
          "counter_control_throughput_ratio": 0.9003388928039274,
          "counter_control_latency_ratio": 3.0165708331684074
        },
        "correctness": {
          "max_abs_error": 2.6856349761260334e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 6.47326796195,
        "throughput_us_p90": 6.505339688319999,
        "latency_us_p50": 208.125,
        "latency_us_p90": 239.425
      },
      "slowdown": 0.20451389441437978,
      "device_compute_slowdown": 0.0815549551801934,
      "valid": true,
      "round": 0,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "softmax_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-i85kc8hw/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "7bac43b2781c01105a8e678fa6161555c0e89cbd2b0a98679fdc42bb1bd74b45",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 34.507542,
        "capture_ms": 0.075417,
        "compile_ms": 100.677542,
        "allocation_upload_ms": 1.788459,
        "cold_call_ms": 6.088709,
        "warmup_ms": 201.514125,
        "download_ms": 0.642209,
        "repetitions": 1370,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          20.0888080292,
          22.1710766423,
          20.191910219,
          21.0010335766,
          20.8618613139,
          20.3702554745,
          21.7501824818,
          20.0105226277,
          22.0835459854
        ],
        "latency_us": [
          275.875,
          234.666,
          258.167,
          240.958,
          224.5,
          213.416,
          240.917,
          234.417,
          221.583
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 128,
            "reduction_scalar_rounds": 60,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 12288,
              "global_write_bytes": 12288,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 48,
              "global_write_bytes": 48,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 2,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 100,
            "normalized_kernel_cost": 400,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 9216,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 1378708,
              "compute_span_ns": 1378708,
              "calibration_cpu_ns": 2433375,
              "calibration_gpu_ticks": 2433375,
              "compute_passes": 1,
              "command_buffer_ns": 1378791.63958,
              "command_buffers": 2
            },
            {
              "compute_ns": 919834,
              "compute_span_ns": 919834,
              "calibration_cpu_ns": 1414167,
              "calibration_gpu_ticks": 1414167,
              "compute_passes": 1,
              "command_buffer_ns": 919999.962207,
              "command_buffers": 2
            },
            {
              "compute_ns": 1468125,
              "compute_span_ns": 1468125,
              "calibration_cpu_ns": 2774000,
              "calibration_gpu_ticks": 2774000,
              "compute_passes": 1,
              "command_buffer_ns": 1468166.65074,
              "command_buffers": 2
            },
            {
              "compute_ns": 941625,
              "compute_span_ns": 941625,
              "calibration_cpu_ns": 1470417,
              "calibration_gpu_ticks": 1470417,
              "compute_passes": 1,
              "command_buffer_ns": 941708.392929,
              "command_buffers": 2
            },
            {
              "compute_ns": 1488750,
              "compute_span_ns": 1488750,
              "calibration_cpu_ns": 2809875,
              "calibration_gpu_ticks": 2809875,
              "compute_passes": 1,
              "command_buffer_ns": 1488833.28075,
              "command_buffers": 2
            },
            {
              "compute_ns": 901041,
              "compute_span_ns": 901041,
              "calibration_cpu_ns": 1410833,
              "calibration_gpu_ticks": 1410833,
              "compute_passes": 1,
              "command_buffer_ns": 901125.02221,
              "command_buffers": 2
            },
            {
              "compute_ns": 1457166,
              "compute_span_ns": 1457166,
              "calibration_cpu_ns": 2809417,
              "calibration_gpu_ticks": 2809417,
              "compute_passes": 1,
              "command_buffer_ns": 1457166.68339,
              "command_buffers": 2
            },
            {
              "compute_ns": 904917,
              "compute_span_ns": 904917,
              "calibration_cpu_ns": 1459583,
              "calibration_gpu_ticks": 1459583,
              "compute_passes": 1,
              "command_buffer_ns": 905041.64109,
              "command_buffers": 2
            },
            {
              "compute_ns": 1301500,
              "compute_span_ns": 1301500,
              "calibration_cpu_ns": 2362959,
              "calibration_gpu_ticks": 2362959,
              "compute_passes": 1,
              "command_buffer_ns": 1301541.63344,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 91500,
              "compute_span_ns": 91500,
              "calibration_cpu_ns": 1086000,
              "calibration_gpu_ticks": 1086000,
              "compute_passes": 1,
              "command_buffer_ns": 91541.6749194,
              "command_buffers": 2
            },
            {
              "compute_ns": 18083,
              "compute_span_ns": 18083,
              "calibration_cpu_ns": 985917,
              "calibration_gpu_ticks": 985917,
              "compute_passes": 1,
              "command_buffer_ns": 18083.3158083,
              "command_buffers": 2
            },
            {
              "compute_ns": 51042,
              "compute_span_ns": 51042,
              "calibration_cpu_ns": 1351041,
              "calibration_gpu_ticks": 1351041,
              "compute_passes": 1,
              "command_buffer_ns": 51208.3643116,
              "command_buffers": 2
            },
            {
              "compute_ns": 20292,
              "compute_span_ns": 20292,
              "calibration_cpu_ns": 961791,
              "calibration_gpu_ticks": 961791,
              "compute_passes": 1,
              "command_buffer_ns": 20374.9514185,
              "command_buffers": 2
            },
            {
              "compute_ns": 69583,
              "compute_span_ns": 69583,
              "calibration_cpu_ns": 1524916,
              "calibration_gpu_ticks": 1524916,
              "compute_passes": 1,
              "command_buffer_ns": 69583.3587088,
              "command_buffers": 2
            },
            {
              "compute_ns": 20625,
              "compute_span_ns": 20625,
              "calibration_cpu_ns": 763792,
              "calibration_gpu_ticks": 763792,
              "compute_passes": 1,
              "command_buffer_ns": 20624.9533221,
              "command_buffers": 2
            },
            {
              "compute_ns": 20917,
              "compute_span_ns": 20917,
              "calibration_cpu_ns": 1042583,
              "calibration_gpu_ticks": 1042583,
              "compute_passes": 1,
              "command_buffer_ns": 20958.3668038,
              "command_buffers": 2
            },
            {
              "compute_ns": 16167,
              "compute_span_ns": 16167,
              "calibration_cpu_ns": 623917,
              "calibration_gpu_ticks": 623917,
              "compute_passes": 1,
              "command_buffer_ns": 16249.9491125,
              "command_buffers": 2
            },
            {
              "compute_ns": 18958,
              "compute_span_ns": 18958,
              "calibration_cpu_ns": 546583,
              "calibration_gpu_ticks": 546583,
              "compute_passes": 1,
              "command_buffer_ns": 19083.3816305,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1224624.98257,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1154166.6463,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 897791.7023,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1441625.0051,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 926666.602027,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1287958.3519,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 922083.389014,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1476916.65916,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 922708.364669,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 16083.358787,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 16166.6539498,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 29416.6966341,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 16250.0073202,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18666.672986,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 16166.7121574,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 69000.0015311,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 109958.346002,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20833.3367482,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              19.13476535265625,
              18.0338538484375,
              14.0279953484375,
              22.5253907046875,
              14.479165656671876,
              20.124349248437497,
              14.40755295334375,
              23.076822799375,
              14.417318197953126
            ],
            "command_buffer_throughput_us_p50": 18.0338538484375,
            "command_buffer_throughput_us_p90": 22.635677123625,
            "command_buffer_latency_us": [
              16.083358786999998,
              16.1666539498,
              29.4166966341,
              16.2500073202,
              18.666672986000002,
              16.1667121574,
              69.0000015311,
              109.958346002,
              20.8333367482
            ],
            "command_buffer_latency_us_p50": 18.666672986000002,
            "command_buffer_latency_us_p90": 77.19167042528001
          },
          "compute_throughput_us": [
            21.5423125,
            14.37240625,
            22.939453125,
            14.712890625,
            23.26171875,
            14.078765625,
            22.76821875,
            14.139328125,
            20.3359375
          ],
          "compute_throughput_us_p50": 20.3359375,
          "compute_throughput_us_p90": 23.00390625,
          "command_buffer_throughput_us": [
            21.5436193684375,
            14.374999409484374,
            22.9401039178125,
            14.714193639515624,
            23.263020011718748,
            14.08007847203125,
            22.76822942796875,
            14.141275642031252,
            20.3365880225
          ],
          "command_buffer_throughput_us_p50": 20.3365880225,
          "command_buffer_throughput_us_p90": 23.00468713659375,
          "compute_latency_us": [
            91.5,
            18.083,
            51.042,
            20.292,
            69.583,
            20.625,
            20.917,
            16.167,
            18.958
          ],
          "compute_latency_us_p50": 20.625,
          "compute_latency_us_p90": 73.96640000000001,
          "command_buffer_latency_us": [
            91.5416749194,
            18.0833158083,
            51.2083643116,
            20.3749514185,
            69.5833587088,
            20.624953322099998,
            20.9583668038,
            16.2499491125,
            19.0833816305
          ],
          "command_buffer_latency_us_p50": 20.624953322099998,
          "command_buffer_latency_us_p90": 73.97502195092001,
          "counter_control_throughput_ratio": 1.1258888714539088,
          "counter_control_latency_ratio": 1.2538425993920863
        },
        "correctness": {
          "max_abs_error": 1.569250362249905e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 20.8618613139,
        "throughput_us_p90": 22.10105211678,
        "latency_us_p50": 234.666,
        "latency_us_p90": 261.7086
      },
      "torch": {
        "cold_call_ms": 6.969041,
        "warmup_ms": 200.304959,
        "repetitions": 542,
        "throughput_us": [
          44.799276752767526,
          45.68373431734318,
          46.19072878228782,
          46.43888376383764,
          45.9330405904059,
          46.379612546125465,
          45.52775092250923,
          45.66013099630997,
          45.70940959409594
        ],
        "latency_us": [
          517.4159999999999,
          327.41700000000003,
          2321.125,
          671.959,
          366.125,
          570.125,
          487.79200000000003,
          378.33299999999997,
          373.042
        ],
        "throughput_us_p50": 45.70940959409594,
        "throughput_us_p90": 46.3914667896679,
        "latency_us_p50": 487.79200000000003,
        "latency_us_p90": 1001.7922000000003,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 2041250.0016391277,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1931624.8944960535,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 2017625.025473535,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1955625.0772438943,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1872125.1981332898,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1920458.220411092,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1885291.6546165943,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1895041.6706502438,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1888708.269689232,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 71041.6934452951,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 63749.96155500412,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49875.001423060894,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 65250.03118440509,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 52916.70095175505,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49041.700549423695,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49541.646149009466,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 50125.00332668424,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 119875.01056864858,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              31.89453127561137,
              30.181638976500835,
              31.525391023023985,
              30.55664183193585,
              29.251956220832653,
              30.007159693923313,
              29.457682103384286,
              29.61002610391006,
              29.51106671389425
            ],
            "command_buffer_throughput_us_p50": 30.007159693923313,
            "command_buffer_throughput_us_p90": 31.599219073541462,
            "command_buffer_latency_us": [
              71.0416934452951,
              63.74996155500412,
              49.875001423060894,
              65.25003118440509,
              52.91670095175505,
              49.041700549423695,
              49.541646149009466,
              50.12500332668424,
              119.87501056864858
            ],
            "command_buffer_latency_us_p50": 52.91670095175505,
            "command_buffer_latency_us_p90": 80.8083568699658
          },
          "throughput": [
            {
              "compute_ns": 3048001.0,
              "compute_span_ns": 4395667.0,
              "command_buffer_ns": 4264708.142727613,
              "calibration_cpu_ns": 5566708.0,
              "calibration_gpu_ticks": 5566708.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3024037.0,
              "compute_span_ns": 4320833.0,
              "command_buffer_ns": 4186583.450064063,
              "calibration_cpu_ns": 5617042.0,
              "calibration_gpu_ticks": 5617042.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3084456.0,
              "compute_span_ns": 4312709.0,
              "command_buffer_ns": 4161208.344157785,
              "calibration_cpu_ns": 5913959.0,
              "calibration_gpu_ticks": 5913959.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3036916.0,
              "compute_span_ns": 4340625.0,
              "command_buffer_ns": 4185833.2697302103,
              "calibration_cpu_ns": 5776375.0,
              "calibration_gpu_ticks": 5776375.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3081956.0,
              "compute_span_ns": 4317000.0,
              "command_buffer_ns": 4155833.332333714,
              "calibration_cpu_ns": 5876625.0,
              "calibration_gpu_ticks": 5876625.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3062420.0,
              "compute_span_ns": 4380250.0,
              "command_buffer_ns": 4228541.627526283,
              "calibration_cpu_ns": 5793292.0,
              "calibration_gpu_ticks": 5793292.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2981495.0,
              "compute_span_ns": 3994416.0,
              "command_buffer_ns": 3832416.725344956,
              "calibration_cpu_ns": 5755708.0,
              "calibration_gpu_ticks": 5755708.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3034212.0,
              "compute_span_ns": 4141583.0,
              "command_buffer_ns": 3864000.0275336206,
              "calibration_cpu_ns": 5903791.0,
              "calibration_gpu_ticks": 5903791.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3029711.0,
              "compute_span_ns": 4345125.0,
              "command_buffer_ns": 4180666.641332209,
              "calibration_cpu_ns": 5978000.0,
              "calibration_gpu_ticks": 5978000.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 47084.0,
              "compute_span_ns": 47084.0,
              "command_buffer_ns": 47125.03869086504,
              "calibration_cpu_ns": 506708.0,
              "calibration_gpu_ticks": 506708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49792.0,
              "compute_span_ns": 49792.0,
              "command_buffer_ns": 49833.382945507765,
              "calibration_cpu_ns": 641083.0,
              "calibration_gpu_ticks": 641083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49708.0,
              "compute_span_ns": 49708.0,
              "command_buffer_ns": 49666.676204651594,
              "calibration_cpu_ns": 550084.0,
              "calibration_gpu_ticks": 550084.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47959.0,
              "compute_span_ns": 47959.0,
              "command_buffer_ns": 48000.01624971628,
              "calibration_cpu_ns": 544333.0,
              "calibration_gpu_ticks": 544333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49875.0,
              "compute_span_ns": 49875.0,
              "command_buffer_ns": 49875.001423060894,
              "calibration_cpu_ns": 652459.0,
              "calibration_gpu_ticks": 652459.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 72167.0,
              "compute_span_ns": 72167.0,
              "command_buffer_ns": 72208.34959298372,
              "calibration_cpu_ns": 1435125.0,
              "calibration_gpu_ticks": 1435125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 52708.0,
              "compute_span_ns": 52708.0,
              "command_buffer_ns": 52708.31752568483,
              "calibration_cpu_ns": 1448959.0,
              "calibration_gpu_ticks": 1448959.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49084.0,
              "compute_span_ns": 49084.0,
              "command_buffer_ns": 49124.99571219087,
              "calibration_cpu_ns": 1101709.0,
              "calibration_gpu_ticks": 1101709.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 50959.0,
              "compute_span_ns": 50959.0,
              "command_buffer_ns": 50999.98088553548,
              "calibration_cpu_ns": 1202875.0,
              "calibration_gpu_ticks": 1202875.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            47.625015625,
            47.250578125,
            48.194625,
            47.4518125,
            48.1555625,
            47.8503125,
            46.585859375,
            47.4095625,
            47.339234375
          ],
          "compute_throughput_us_p50": 47.4518125,
          "compute_throughput_us_p90": 48.163375,
          "command_buffer_throughput_us": [
            66.63606473011896,
            65.41536640725099,
            65.0188803774654,
            65.40364483953454,
            64.93489581771428,
            66.07096293009818,
            59.881511333514936,
            60.37500043021282,
            65.32291627081577
          ],
          "command_buffer_throughput_us_p50": 65.32291627081577,
          "command_buffer_throughput_us_p90": 66.18398329010233,
          "compute_latency_us": [
            47.084,
            49.792,
            49.708,
            47.959,
            49.875,
            72.167,
            52.708,
            49.084,
            50.959
          ],
          "compute_latency_us_p50": 49.792,
          "compute_latency_us_p90": 56.5998,
          "command_buffer_latency_us": [
            47.12503869086504,
            49.833382945507765,
            49.666676204651594,
            48.00001624971628,
            49.875001423060894,
            72.20834959298372,
            52.70831752568483,
            49.12499571219087,
            50.99998088553548
          ],
          "command_buffer_latency_us_p50": 49.833382945507765,
          "command_buffer_latency_us_p90": 56.60832393914461,
          "counter_control_throughput_ratio": 2.1404068287104385,
          "counter_control_latency_ratio": 0.9425191012669646
        },
        "download_ms": 0.443917,
        "allocation_upload_ms": 1.77675,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 2.673614807194649e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.45640189840900125,
      "device_compute_slowdown": 0.4285597626012705,
      "valid": true,
      "round": 0,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "softmax_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.366459,
        "warmup_ms": 200.652084,
        "repetitions": 654,
        "throughput_us": [
          44.59403669724771,
          44.17214678899083,
          42.923483180428136,
          46.17730733944954,
          44.53911773700305,
          44.3236498470948,
          45.418642201834864,
          44.000126911314986,
          44.196801223241586
        ],
        "latency_us": [
          2580.7090000000003,
          984.0419999999999,
          345.75,
          348.125,
          383.70799999999997,
          301.70799999999997,
          343.209,
          363.042,
          360.16700000000003
        ],
        "throughput_us_p50": 44.3236498470948,
        "throughput_us_p90": 45.5703752293578,
        "latency_us_p50": 360.16700000000003,
        "latency_us_p90": 1303.3754000000004,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1885333.1566788256,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1945833.32631737,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 2020625.0483170152,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1943749.8995102942,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1905208.3953283727,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1913083.3679810166,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 2030166.7391322553,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1929291.698615998,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 2008999.8724870384,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 68499.99772384763,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 126958.35903286934,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 50916.62751510739,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 48833.375331014395,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 50416.62370786071,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 48374.99000132084,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 48000.01624971628,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 48124.9880976975,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 121124.96187910438,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              29.45833057310665,
              30.403645723708905,
              31.572266379953362,
              30.371092179848347,
              29.768881177005824,
              29.891927624703385,
              31.72135529894149,
              30.14518279087497,
              31.390623007609975
            ],
            "command_buffer_throughput_us_p50": 30.371092179848347,
            "command_buffer_throughput_us_p90": 31.602084163750987,
            "command_buffer_latency_us": [
              68.49999772384763,
              126.95835903286934,
              50.91662751510739,
              48.833375331014395,
              50.41662370786071,
              48.37499000132084,
              48.00001624971628,
              48.124988097697496,
              121.12496187910438
            ],
            "command_buffer_latency_us_p50": 50.41662370786071,
            "command_buffer_latency_us_p90": 122.29164130985737
          },
          "throughput": [
            {
              "compute_ns": 3036163.0,
              "compute_span_ns": 4336875.0,
              "command_buffer_ns": 3905791.7310856283,
              "calibration_cpu_ns": 6098041.0,
              "calibration_gpu_ticks": 6098041.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3087663.0,
              "compute_span_ns": 4443708.0,
              "command_buffer_ns": 3932833.3223238587,
              "calibration_cpu_ns": 5899791.0,
              "calibration_gpu_ticks": 5899791.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2989752.0,
              "compute_span_ns": 3926500.0,
              "command_buffer_ns": 3769333.241507411,
              "calibration_cpu_ns": 5833417.0,
              "calibration_gpu_ticks": 5833417.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3054710.0,
              "compute_span_ns": 4370083.0,
              "command_buffer_ns": 4219333.291985095,
              "calibration_cpu_ns": 5698167.0,
              "calibration_gpu_ticks": 5698167.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3002752.0,
              "compute_span_ns": 4403167.0,
              "command_buffer_ns": 4217166.628222913,
              "calibration_cpu_ns": 5951584.0,
              "calibration_gpu_ticks": 5951584.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3045416.0,
              "compute_span_ns": 4536083.0,
              "command_buffer_ns": 4366750.072222203,
              "calibration_cpu_ns": 6014709.0,
              "calibration_gpu_ticks": 6014709.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3000540.0,
              "compute_span_ns": 4322042.0,
              "command_buffer_ns": 4182708.275038749,
              "calibration_cpu_ns": 5706125.0,
              "calibration_gpu_ticks": 5706125.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3014915.0,
              "compute_span_ns": 4231708.0,
              "command_buffer_ns": 4058624.9670013785,
              "calibration_cpu_ns": 5854833.0,
              "calibration_gpu_ticks": 5854833.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2999250.0,
              "compute_span_ns": 4206917.0,
              "command_buffer_ns": 3884624.980855733,
              "calibration_cpu_ns": 6052833.0,
              "calibration_gpu_ticks": 6052833.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 48625.0,
              "compute_span_ns": 48625.0,
              "command_buffer_ns": 48625.050112605095,
              "calibration_cpu_ns": 1009000.0,
              "calibration_gpu_ticks": 1009000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 50083.0,
              "compute_span_ns": 50083.0,
              "command_buffer_ns": 50041.708163917065,
              "calibration_cpu_ns": 769750.0,
              "calibration_gpu_ticks": 769750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 67708.0,
              "compute_span_ns": 67708.0,
              "command_buffer_ns": 67708.31532776356,
              "calibration_cpu_ns": 504083.0,
              "calibration_gpu_ticks": 504083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 119167.0,
              "compute_span_ns": 119167.0,
              "command_buffer_ns": 119208.35822820663,
              "calibration_cpu_ns": 1498209.0,
              "calibration_gpu_ticks": 1498209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 62542.0,
              "compute_span_ns": 62542.0,
              "command_buffer_ns": 62583.363614976406,
              "calibration_cpu_ns": 1434084.0,
              "calibration_gpu_ticks": 1434084.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 51583.0,
              "compute_span_ns": 51583.0,
              "command_buffer_ns": 51583.33806321025,
              "calibration_cpu_ns": 1246584.0,
              "calibration_gpu_ticks": 1246584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 62917.0,
              "compute_span_ns": 62917.0,
              "command_buffer_ns": 62916.66068136692,
              "calibration_cpu_ns": 2082916.0,
              "calibration_gpu_ticks": 2082916.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70167.0,
              "compute_span_ns": 70167.0,
              "command_buffer_ns": 70208.33436399698,
              "calibration_cpu_ns": 658584.0,
              "calibration_gpu_ticks": 658584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 51042.0,
              "compute_span_ns": 51042.0,
              "command_buffer_ns": 51083.334255963564,
              "calibration_cpu_ns": 1103167.0,
              "calibration_gpu_ticks": 1103167.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            47.440046875,
            48.244734375,
            46.714875,
            47.72984375,
            46.918,
            47.584625,
            46.8834375,
            47.108046875,
            46.86328125
          ],
          "compute_throughput_us_p50": 47.108046875,
          "compute_throughput_us_p90": 47.832821875,
          "command_buffer_throughput_us": [
            61.02799579821294,
            61.45052066131029,
            58.8958318985533,
            65.9270826872671,
            65.89322856598301,
            68.23046987847192,
            65.35481679748045,
            63.41601510939654,
            60.69726532587083
          ],
          "command_buffer_throughput_us_p50": 63.41601510939654,
          "command_buffer_throughput_us_p90": 66.38776012550807,
          "compute_latency_us": [
            48.625,
            50.083,
            67.708,
            119.167,
            62.542,
            51.583,
            62.917,
            70.167,
            51.042
          ],
          "compute_latency_us_p50": 62.542,
          "compute_latency_us_p90": 79.96700000000001,
          "command_buffer_latency_us": [
            48.625050112605095,
            50.041708163917065,
            67.70831532776356,
            119.20835822820663,
            62.583363614976406,
            51.58333806321025,
            62.91666068136692,
            70.20833436399698,
            51.083334255963564
          ],
          "command_buffer_latency_us_p50": 62.583363614976406,
          "command_buffer_latency_us_p90": 80.00833913683893,
          "counter_control_throughput_ratio": 2.071671904378965,
          "counter_control_latency_ratio": 1.24132397237895
        },
        "download_ms": 0.415417,
        "allocation_upload_ms": 0.521583,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 2.673614807194649e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-rncm118l/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "0d8a82e58cbe97c43d15ef559cf043ad575745e0d3fb0f4f40ed60b2caca21c0",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 22.367125,
        "capture_ms": 0.067333,
        "compile_ms": 53.919083,
        "allocation_upload_ms": 2.063958,
        "cold_call_ms": 5.710458,
        "warmup_ms": 200.187,
        "download_ms": 0.820584,
        "repetitions": 1237,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          20.0511649151,
          19.9629482619,
          20.1505658852,
          20.391,
          20.4376515764,
          20.145109135,
          20.2671107518,
          20.1908852061,
          20.2612845594
        ],
        "latency_us": [
          307.5,
          306.5,
          244.791,
          243.167,
          248.583,
          268.375,
          230.958,
          207.541,
          228.542
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 256,
            "reduction_scalar_rounds": 60,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 12288,
              "global_write_bytes": 12288,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 48,
              "global_write_bytes": 48,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 2,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 108,
            "normalized_kernel_cost": 432,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 9216,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 837292,
              "compute_span_ns": 837292,
              "calibration_cpu_ns": 1301042,
              "calibration_gpu_ticks": 1301042,
              "compute_passes": 1,
              "command_buffer_ns": 837333.325762,
              "command_buffers": 2
            },
            {
              "compute_ns": 883792,
              "compute_span_ns": 883792,
              "calibration_cpu_ns": 1380458,
              "calibration_gpu_ticks": 1380458,
              "compute_passes": 1,
              "command_buffer_ns": 883833.33059,
              "command_buffers": 2
            },
            {
              "compute_ns": 1379458,
              "compute_span_ns": 1379458,
              "calibration_cpu_ns": 2542708,
              "calibration_gpu_ticks": 2542708,
              "compute_passes": 1,
              "command_buffer_ns": 1379541.7035,
              "command_buffers": 2
            },
            {
              "compute_ns": 856083,
              "compute_span_ns": 856083,
              "calibration_cpu_ns": 1596166,
              "calibration_gpu_ticks": 1596166,
              "compute_passes": 1,
              "command_buffer_ns": 856083.352119,
              "command_buffers": 2
            },
            {
              "compute_ns": 1327166,
              "compute_span_ns": 1327166,
              "calibration_cpu_ns": 2199667,
              "calibration_gpu_ticks": 2199667,
              "compute_passes": 1,
              "command_buffer_ns": 1327125.00636,
              "command_buffers": 2
            },
            {
              "compute_ns": 873667,
              "compute_span_ns": 873667,
              "calibration_cpu_ns": 1490333,
              "calibration_gpu_ticks": 1490333,
              "compute_passes": 1,
              "command_buffer_ns": 873749.959283,
              "command_buffers": 2
            },
            {
              "compute_ns": 1358583,
              "compute_span_ns": 1358583,
              "calibration_cpu_ns": 2630625,
              "calibration_gpu_ticks": 2630625,
              "compute_passes": 1,
              "command_buffer_ns": 1358625.01338,
              "command_buffers": 2
            },
            {
              "compute_ns": 873125,
              "compute_span_ns": 873125,
              "calibration_cpu_ns": 1423917,
              "calibration_gpu_ticks": 1423917,
              "compute_passes": 1,
              "command_buffer_ns": 873208.395205,
              "command_buffers": 2
            },
            {
              "compute_ns": 1123625,
              "compute_span_ns": 1123625,
              "calibration_cpu_ns": 2349875,
              "calibration_gpu_ticks": 2349875,
              "compute_passes": 1,
              "command_buffer_ns": 1123708.32358,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 23416,
              "compute_span_ns": 23416,
              "calibration_cpu_ns": 1064375,
              "calibration_gpu_ticks": 1064375,
              "compute_passes": 1,
              "command_buffer_ns": 23375.0324696,
              "command_buffers": 2
            },
            {
              "compute_ns": 18584,
              "compute_span_ns": 18584,
              "calibration_cpu_ns": 925125,
              "calibration_gpu_ticks": 925125,
              "compute_passes": 1,
              "command_buffer_ns": 18708.2914636,
              "command_buffers": 2
            },
            {
              "compute_ns": 87333,
              "compute_span_ns": 87333,
              "calibration_cpu_ns": 1451666,
              "calibration_gpu_ticks": 1451666,
              "compute_passes": 1,
              "command_buffer_ns": 87416.6144058,
              "command_buffers": 2
            },
            {
              "compute_ns": 15875,
              "compute_span_ns": 15875,
              "calibration_cpu_ns": 735375,
              "calibration_gpu_ticks": 735375,
              "compute_passes": 1,
              "command_buffer_ns": 15958.386939,
              "command_buffers": 2
            },
            {
              "compute_ns": 17541,
              "compute_span_ns": 17541,
              "calibration_cpu_ns": 1001250,
              "calibration_gpu_ticks": 1001250,
              "compute_passes": 1,
              "command_buffer_ns": 17625.046894,
              "command_buffers": 2
            },
            {
              "compute_ns": 18125,
              "compute_span_ns": 18125,
              "calibration_cpu_ns": 498000,
              "calibration_gpu_ticks": 498000,
              "compute_passes": 1,
              "command_buffer_ns": 18208.345864,
              "command_buffers": 2
            },
            {
              "compute_ns": 15792,
              "compute_span_ns": 15792,
              "calibration_cpu_ns": 513291,
              "calibration_gpu_ticks": 513291,
              "compute_passes": 1,
              "command_buffer_ns": 15874.9171533,
              "command_buffers": 2
            },
            {
              "compute_ns": 101041,
              "compute_span_ns": 101041,
              "calibration_cpu_ns": 1393667,
              "calibration_gpu_ticks": 1393667,
              "compute_passes": 1,
              "command_buffer_ns": 101083.307527,
              "command_buffers": 2
            },
            {
              "compute_ns": 18125,
              "compute_span_ns": 18125,
              "calibration_cpu_ns": 485750,
              "calibration_gpu_ticks": 485750,
              "compute_passes": 1,
              "command_buffer_ns": 18166.6109711,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 918374.978937,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 863083.347213,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 867541.646585,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1094250.00846,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 880499.952473,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1354750.01298,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 921041.646507,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1149708.28872,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 850624.986924,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 15541.6200869,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 15666.6501425,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18124.9924935,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 15666.6501425,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 42208.3539888,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65583.3864585,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20291.6562557,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 17749.9605343,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 15875.0335686,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              14.349609045890626,
              13.485677300203124,
              13.555338227890624,
              17.0976563821875,
              13.757811757390625,
              21.1679689528125,
              14.391275726671875,
              17.964192011250002,
              13.291015420687499
            ],
            "command_buffer_throughput_us_p50": 14.349609045890626,
            "command_buffer_throughput_us_p90": 18.6049473995625,
            "command_buffer_latency_us": [
              15.5416200869,
              15.6666501425,
              18.124992493500002,
              15.6666501425,
              42.2083539888,
              65.5833864585,
              20.2916562557,
              17.7499605343,
              15.8750335686
            ],
            "command_buffer_latency_us_p50": 17.7499605343,
            "command_buffer_latency_us_p90": 46.88336048274
          },
          "compute_throughput_us": [
            13.0826875,
            13.80925,
            21.55403125,
            13.376296875,
            20.73696875,
            13.651046875,
            21.227859375,
            13.642578125,
            17.556640625
          ],
          "compute_throughput_us_p50": 13.80925,
          "compute_throughput_us_p90": 21.29309375,
          "command_buffer_throughput_us": [
            13.083333215031251,
            13.809895790468751,
            21.5553391171875,
            13.376302376859377,
            20.736328224375,
            13.652343113796874,
            21.2285158340625,
            13.643881175078125,
            17.5579425559375
          ],
          "command_buffer_throughput_us_p50": 13.809895790468751,
          "command_buffer_throughput_us_p90": 21.2938804906875,
          "compute_latency_us": [
            23.416,
            18.584,
            87.333,
            15.875,
            17.541,
            18.125,
            15.792,
            101.041,
            18.125
          ],
          "compute_latency_us_p50": 18.125,
          "compute_latency_us_p90": 90.0746,
          "command_buffer_latency_us": [
            23.375032469599997,
            18.7082914636,
            87.4166144058,
            15.958386939,
            17.625046894,
            18.208345863999998,
            15.8749171533,
            101.083307527,
            18.1666109711
          ],
          "command_buffer_latency_us_p50": 18.208345863999998,
          "command_buffer_latency_us_p90": 90.14995303004001,
          "counter_control_throughput_ratio": 1.024041691273507,
          "counter_control_latency_ratio": 1.1443510272024005
        },
        "correctness": {
          "max_abs_error": 1.569250362249905e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 20.1908852061,
        "throughput_us_p90": 20.400330315279998,
        "latency_us_p50": 244.791,
        "latency_us_p90": 306.7
      },
      "slowdown": 0.45553300045807066,
      "device_compute_slowdown": 0.29313993926860293,
      "valid": true,
      "round": 0,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "softmax_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-u3w4634t/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "7f45b1ebfa78e24e3d477c2ad514fde5c49a3d7b63ed786757d9dc5d1979c8b1",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.456084,
        "capture_ms": 0.074834,
        "compile_ms": 53.858458,
        "allocation_upload_ms": 6.973708,
        "cold_call_ms": 5.659125,
        "warmup_ms": 201.25525,
        "download_ms": 3.388709,
        "repetitions": 373,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          86.0979651475,
          86.4674932976,
          87.2794906166,
          83.2625093834,
          80.498772118,
          86.6539329759,
          86.6568364611,
          86.1267882038,
          84.3385817694
        ],
        "latency_us": [
          1113.5,
          383.291,
          289.833,
          290.25,
          275.916,
          274.459,
          285.375,
          267.833,
          270.208
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 768,
            "reduction_scalar_rounds": 120,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 24576,
              "global_write_bytes": 24576,
              "private_read_bytes": 147456,
              "private_write_bytes": 49152
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 96,
              "private_read_bytes": 576,
              "private_write_bytes": 192
            },
            "striped_storage_scalars_per_worker": 48,
            "reduction_operations": 2,
            "reduction_elements": 12288,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 168,
            "normalized_kernel_cost": 2016,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 18432,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3901292,
              "compute_span_ns": 3901292,
              "calibration_cpu_ns": 5033500,
              "calibration_gpu_ticks": 5033500,
              "compute_passes": 1,
              "command_buffer_ns": 3901416.66867,
              "command_buffers": 2
            },
            {
              "compute_ns": 4022625,
              "compute_span_ns": 4022625,
              "calibration_cpu_ns": 4991708,
              "calibration_gpu_ticks": 4991708,
              "compute_passes": 1,
              "command_buffer_ns": 4022624.98392,
              "command_buffers": 2
            },
            {
              "compute_ns": 4083333,
              "compute_span_ns": 4083333,
              "calibration_cpu_ns": 5244833,
              "calibration_gpu_ticks": 5244833,
              "compute_passes": 1,
              "command_buffer_ns": 4083458.33421,
              "command_buffers": 2
            },
            {
              "compute_ns": 4340833,
              "compute_span_ns": 4340833,
              "calibration_cpu_ns": 5531709,
              "calibration_gpu_ticks": 5531709,
              "compute_passes": 1,
              "command_buffer_ns": 4340916.6974,
              "command_buffers": 2
            },
            {
              "compute_ns": 4028459,
              "compute_span_ns": 4028459,
              "calibration_cpu_ns": 5155166,
              "calibration_gpu_ticks": 5155166,
              "compute_passes": 1,
              "command_buffer_ns": 4028541.67623,
              "command_buffers": 2
            },
            {
              "compute_ns": 4184292,
              "compute_span_ns": 4184292,
              "calibration_cpu_ns": 5380375,
              "calibration_gpu_ticks": 5380375,
              "compute_passes": 1,
              "command_buffer_ns": 4184374.93499,
              "command_buffers": 2
            },
            {
              "compute_ns": 3976875,
              "compute_span_ns": 3976875,
              "calibration_cpu_ns": 5115334,
              "calibration_gpu_ticks": 5115334,
              "compute_passes": 1,
              "command_buffer_ns": 3976916.66149,
              "command_buffers": 2
            },
            {
              "compute_ns": 4157417,
              "compute_span_ns": 4157417,
              "calibration_cpu_ns": 5374417,
              "calibration_gpu_ticks": 5374417,
              "compute_passes": 1,
              "command_buffer_ns": 4157458.3156,
              "command_buffers": 2
            },
            {
              "compute_ns": 3624500,
              "compute_span_ns": 3624500,
              "calibration_cpu_ns": 4587458,
              "calibration_gpu_ticks": 4587458,
              "compute_passes": 1,
              "command_buffer_ns": 3624500.0083,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 97750,
              "compute_span_ns": 97750,
              "calibration_cpu_ns": 1019917,
              "calibration_gpu_ticks": 1019917,
              "compute_passes": 1,
              "command_buffer_ns": 97833.3409876,
              "command_buffers": 2
            },
            {
              "compute_ns": 131500,
              "compute_span_ns": 131500,
              "calibration_cpu_ns": 1561875,
              "calibration_gpu_ticks": 1561875,
              "compute_passes": 1,
              "command_buffer_ns": 131583.365146,
              "command_buffers": 2
            },
            {
              "compute_ns": 112042,
              "compute_span_ns": 112042,
              "calibration_cpu_ns": 1449666,
              "calibration_gpu_ticks": 1449666,
              "compute_passes": 1,
              "command_buffer_ns": 112125.009764,
              "command_buffers": 2
            },
            {
              "compute_ns": 71625,
              "compute_span_ns": 71625,
              "calibration_cpu_ns": 649458,
              "calibration_gpu_ticks": 649458,
              "compute_passes": 1,
              "command_buffer_ns": 71624.9924153,
              "command_buffers": 2
            },
            {
              "compute_ns": 123584,
              "compute_span_ns": 123584,
              "calibration_cpu_ns": 1533458,
              "calibration_gpu_ticks": 1533458,
              "compute_passes": 1,
              "command_buffer_ns": 123708.392493,
              "command_buffers": 2
            },
            {
              "compute_ns": 99625,
              "compute_span_ns": 99625,
              "calibration_cpu_ns": 936667,
              "calibration_gpu_ticks": 936667,
              "compute_passes": 1,
              "command_buffer_ns": 99708.3261609,
              "command_buffers": 2
            },
            {
              "compute_ns": 116084,
              "compute_span_ns": 116084,
              "calibration_cpu_ns": 2428542,
              "calibration_gpu_ticks": 2428542,
              "compute_passes": 1,
              "command_buffer_ns": 116166.600492,
              "command_buffers": 2
            },
            {
              "compute_ns": 81250,
              "compute_span_ns": 81250,
              "calibration_cpu_ns": 606666,
              "calibration_gpu_ticks": 606666,
              "compute_passes": 1,
              "command_buffer_ns": 81291.6550785,
              "command_buffers": 2
            },
            {
              "compute_ns": 116375,
              "compute_span_ns": 116375,
              "calibration_cpu_ns": 635792,
              "calibration_gpu_ticks": 635792,
              "compute_passes": 1,
              "command_buffer_ns": 116416.602395,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 3800458.33299,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4115958.34885,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4229000.01286,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4047041.64248,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4054041.69578,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4116666.67787,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3928958.32192,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3956416.68001,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3996583.34209,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 157166.679855,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 97875.0176728,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 64583.2624286,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 111458.357424,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65999.9786876,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 83208.3169371,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 114874.972496,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 110499.968287,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 61458.3259448,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              59.38216145296875,
              64.31184920078125,
              66.0781252009375,
              63.23502566375,
              63.344401496562504,
              64.32291684171875,
              61.38997378,
              61.819010625156245,
              62.44661472015625
            ],
            "command_buffer_throughput_us_p50": 63.23502566375,
            "command_buffer_throughput_us_p90": 64.6739585135625,
            "command_buffer_latency_us": [
              157.16667985499998,
              97.8750176728,
              64.5832624286,
              111.458357424,
              65.9999786876,
              83.2083169371,
              114.874972496,
              110.499968287,
              61.4583259448
            ],
            "command_buffer_latency_us_p50": 97.8750176728,
            "command_buffer_latency_us_p90": 123.3333139678
          },
          "compute_throughput_us": [
            60.9576875,
            62.853515625,
            63.802078125,
            67.825515625,
            62.944671875,
            65.3795625,
            62.138671875,
            64.959640625,
            56.6328125
          ],
          "compute_throughput_us_p50": 62.944671875,
          "compute_throughput_us_p90": 65.868753125,
          "command_buffer_throughput_us": [
            60.95963544796875,
            62.85351537375,
            63.80403647203125,
            67.826823396875,
            62.945963691093745,
            65.38085835921875,
            62.13932283578125,
            64.96028618125,
            56.6328126296875
          ],
          "command_buffer_throughput_us_p50": 62.945963691093745,
          "command_buffer_throughput_us_p90": 65.87005136675,
          "compute_latency_us": [
            97.75,
            131.5,
            112.042,
            71.625,
            123.584,
            99.625,
            116.084,
            81.25,
            116.375
          ],
          "compute_latency_us_p50": 112.042,
          "compute_latency_us_p90": 125.16720000000001,
          "command_buffer_latency_us": [
            97.8333409876,
            131.583365146,
            112.125009764,
            71.6249924153,
            123.708392493,
            99.7083261609,
            116.166600492,
            81.2916550785,
            116.416602395
          ],
          "command_buffer_latency_us_p50": 112.125009764,
          "command_buffer_latency_us_p90": 125.2833870236,
          "counter_control_throughput_ratio": 1.0122063752375372,
          "counter_control_latency_ratio": 1.1982975960957476
        },
        "correctness": {
          "max_abs_error": 7.116789644580346e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 86.1267882038,
        "throughput_us_p90": 86.7813672922,
        "latency_us_p50": 285.375,
        "latency_us_p90": 529.3328000000001
      },
      "torch": {
        "cold_call_ms": 46.657041,
        "warmup_ms": 201.901917,
        "repetitions": 171,
        "throughput_us": [
          187.6430292397661,
          187.58625730994152,
          187.87743859649123,
          176.36062573099417,
          185.37597660818713,
          189.14790643274856,
          188.14644444444446,
          185.4712514619883,
          181.1676432748538
        ],
        "latency_us": [
          2069.166,
          739.375,
          549.458,
          470.417,
          430.417,
          410.0,
          394.667,
          439.583,
          418.5
        ],
        "throughput_us_p50": 187.58625730994152,
        "throughput_us_p90": 188.3467368421053,
        "latency_us_p50": 439.583,
        "latency_us_p90": 1005.3332000000003,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 10494166.635908186,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10536291.694734246,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10585250.158328563,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10537166.672293097,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10606958.414427936,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10382458.33478868,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10436249.955091625,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10462749.807629734,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10572624.974884093,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 167041.66773706675,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 140374.99204277992,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 189333.33922177553,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 236000.05079060793,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 152875.02901628613,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 148666.67333990335,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 137750.00115856528,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 160874.97351691127,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 143375.01488626003,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              163.9713536860654,
              164.6295577302226,
              165.3945337238838,
              164.64322925457964,
              165.7337252254365,
              162.22591148107313,
              163.06640554830665,
              163.4804657442146,
              165.19726523256395
            ],
            "command_buffer_throughput_us_p50": 164.6295577302226,
            "command_buffer_throughput_us_p90": 165.46237202419434,
            "command_buffer_latency_us": [
              167.04166773706675,
              140.37499204277992,
              189.33333922177553,
              236.00005079060793,
              152.87502901628613,
              148.66667333990335,
              137.75000115856528,
              160.87497351691127,
              143.37501488626003
            ],
            "command_buffer_latency_us_p50": 152.87502901628613,
            "command_buffer_latency_us_p90": 198.666681535542
          },
          "throughput": [
            {
              "compute_ns": 11160246.0,
              "compute_span_ns": 12288625.0,
              "command_buffer_ns": 12106041.715014726,
              "calibration_cpu_ns": 13961041.0,
              "calibration_gpu_ticks": 13961041.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11050546.0,
              "compute_span_ns": 12263542.0,
              "command_buffer_ns": 12078124.971594661,
              "calibration_cpu_ns": 13724167.0,
              "calibration_gpu_ticks": 13724167.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11118912.0,
              "compute_span_ns": 12327625.0,
              "command_buffer_ns": 12158958.415966481,
              "calibration_cpu_ns": 13809750.0,
              "calibration_gpu_ticks": 13809750.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10961082.0,
              "compute_span_ns": 12259834.0,
              "command_buffer_ns": 12047125.084791332,
              "calibration_cpu_ns": 13837000.0,
              "calibration_gpu_ticks": 13837000.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11041045.0,
              "compute_span_ns": 12271625.0,
              "command_buffer_ns": 12053708.254825324,
              "calibration_cpu_ns": 13887583.0,
              "calibration_gpu_ticks": 13887583.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11062377.0,
              "compute_span_ns": 12280167.0,
              "command_buffer_ns": 12066458.410117775,
              "calibration_cpu_ns": 13794375.0,
              "calibration_gpu_ticks": 13794375.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10999705.0,
              "compute_span_ns": 12261000.0,
              "command_buffer_ns": 12085750.058759004,
              "calibration_cpu_ns": 13831292.0,
              "calibration_gpu_ticks": 13831292.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10927373.0,
              "compute_span_ns": 12183042.0,
              "command_buffer_ns": 11998916.801530868,
              "calibration_cpu_ns": 14081042.0,
              "calibration_gpu_ticks": 14081042.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11092044.0,
              "compute_span_ns": 12288084.0,
              "command_buffer_ns": 12101041.618734598,
              "calibration_cpu_ns": 13626792.0,
              "calibration_gpu_ticks": 13626792.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 150583.0,
              "compute_span_ns": 150583.0,
              "command_buffer_ns": 150541.65851324797,
              "calibration_cpu_ns": 885625.0,
              "calibration_gpu_ticks": 885625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 139834.0,
              "compute_span_ns": 139834.0,
              "command_buffer_ns": 139874.98823553324,
              "calibration_cpu_ns": 642833.0,
              "calibration_gpu_ticks": 642833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 142375.0,
              "compute_span_ns": 142375.0,
              "command_buffer_ns": 142374.94906410575,
              "calibration_cpu_ns": 828166.0,
              "calibration_gpu_ticks": 828166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 139333.0,
              "compute_span_ns": 139333.0,
              "command_buffer_ns": 139333.3077430725,
              "calibration_cpu_ns": 633708.0,
              "calibration_gpu_ticks": 633708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 142625.0,
              "compute_span_ns": 142625.0,
              "command_buffer_ns": 142625.00917539,
              "calibration_cpu_ns": 833583.0,
              "calibration_gpu_ticks": 833583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 203291.0,
              "compute_span_ns": 203291.0,
              "command_buffer_ns": 203249.9760389328,
              "calibration_cpu_ns": 1494625.0,
              "calibration_gpu_ticks": 1494625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 140083.0,
              "compute_span_ns": 140083.0,
              "command_buffer_ns": 140041.6367687285,
              "calibration_cpu_ns": 645333.0,
              "calibration_gpu_ticks": 645333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 337584.0,
              "compute_span_ns": 337584.0,
              "command_buffer_ns": 337625.03881007433,
              "calibration_cpu_ns": 1428167.0,
              "calibration_gpu_ticks": 1428167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 139834.0,
              "compute_span_ns": 139834.0,
              "command_buffer_ns": 139874.98823553324,
              "calibration_cpu_ns": 627875.0,
              "calibration_gpu_ticks": 627875.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            174.37884375,
            172.66478125,
            173.733,
            171.26690625,
            172.516328125,
            172.849640625,
            171.870390625,
            170.740203125,
            173.3131875
          ],
          "compute_throughput_us_p50": 172.66478125,
          "compute_throughput_us_p90": 173.86216875,
          "command_buffer_throughput_us": [
            189.1569017971051,
            188.72070268116659,
            189.98372524947627,
            188.23632944986457,
            188.33919148164568,
            188.53841265809024,
            188.83984466810944,
            187.4830750239198,
            189.0787752927281
          ],
          "command_buffer_throughput_us_p50": 188.72070268116659,
          "command_buffer_throughput_us_p90": 189.32226648757933,
          "compute_latency_us": [
            150.583,
            139.834,
            142.375,
            139.333,
            142.625,
            203.291,
            140.083,
            337.584,
            139.834
          ],
          "compute_latency_us_p50": 142.375,
          "compute_latency_us_p90": 230.14960000000002,
          "command_buffer_latency_us": [
            150.54165851324797,
            139.87498823553324,
            142.37494906410575,
            139.3333077430725,
            142.62500917539,
            203.2499760389328,
            140.0416367687285,
            337.62503881007433,
            139.87498823553324
          ],
          "command_buffer_latency_us_p50": 142.37494906410575,
          "command_buffer_latency_us_p90": 230.12498859316113,
          "counter_control_throughput_ratio": 1.1468224914239005,
          "counter_control_latency_ratio": 0.9755883083708596
        },
        "download_ms": 0.687167,
        "allocation_upload_ms": 2.439667,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 7.553236784944503e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.4591316519604954,
      "device_compute_slowdown": 0.3645484123589911,
      "valid": true,
      "round": 0,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "softmax_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.719333,
        "warmup_ms": 200.986834,
        "repetitions": 167,
        "throughput_us": [
          188.9930119760479,
          174.28517964071855,
          191.57709580838323,
          189.75399401197606,
          181.47480239520957,
          180.55813173652695,
          188.49974850299404,
          179.99625748502993,
          181.36776646706585
        ],
        "latency_us": [
          569.4169999999999,
          807.958,
          1946.2079999999999,
          671.2080000000001,
          676.2080000000001,
          361.834,
          423.917,
          435.542,
          399.625
        ],
        "throughput_us_p50": 181.47480239520957,
        "throughput_us_p90": 190.1186143712575,
        "latency_us_p50": 569.4169999999999,
        "latency_us_p90": 1035.6080000000002,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 10605083.429254591,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10438541.64890945,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10410041.722934693,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10604333.3071284,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10668500.035535544,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10632999.998051673,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10604916.722513735,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10512208.33323896,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10645958.303939551,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 216624.99057129025,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 316874.9972246587,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 159375.02030283213,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 156124.99555572867,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 141208.35112407804,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 137249.9973513186,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 315749.95955452323,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 350416.696164757,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 161500.00737980008,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              165.704428582103,
              163.10221326421015,
              162.65690192085458,
              165.69270792388124,
              166.69531305524288,
              166.1406249695574,
              165.7018237892771,
              164.25325520685874,
              166.3430984990555
            ],
            "command_buffer_throughput_us_p50": 165.7018237892771,
            "command_buffer_throughput_us_p90": 166.41354141029296,
            "command_buffer_latency_us": [
              216.62499057129025,
              316.8749972246587,
              159.37502030283213,
              156.12499555572867,
              141.20835112407804,
              137.2499973513186,
              315.74995955452323,
              350.416696164757,
              161.50000737980008
            ],
            "command_buffer_latency_us_p50": 161.50000737980008,
            "command_buffer_latency_us_p90": 323.5833370126784
          },
          "throughput": [
            {
              "compute_ns": 10893285.0,
              "compute_span_ns": 12170916.0,
              "command_buffer_ns": 11952958.36078003,
              "calibration_cpu_ns": 14268042.0,
              "calibration_gpu_ticks": 14268042.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10938753.0,
              "compute_span_ns": 12161125.0,
              "command_buffer_ns": 11949458.392336965,
              "calibration_cpu_ns": 13804583.0,
              "calibration_gpu_ticks": 13804583.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11041004.0,
              "compute_span_ns": 12272709.0,
              "command_buffer_ns": 12078791.565727443,
              "calibration_cpu_ns": 14089041.0,
              "calibration_gpu_ticks": 14089041.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10931628.0,
              "compute_span_ns": 12227542.0,
              "command_buffer_ns": 12073541.700374335,
              "calibration_cpu_ns": 13753584.0,
              "calibration_gpu_ticks": 13753584.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11077415.0,
              "compute_span_ns": 12307667.0,
              "command_buffer_ns": 12121749.925427139,
              "calibration_cpu_ns": 14598541.0,
              "calibration_gpu_ticks": 14598541.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10881084.0,
              "compute_span_ns": 12068708.0,
              "command_buffer_ns": 11890624.882653356,
              "calibration_cpu_ns": 13805666.0,
              "calibration_gpu_ticks": 13805666.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10969834.0,
              "compute_span_ns": 12298875.0,
              "command_buffer_ns": 12083000.037819147,
              "calibration_cpu_ns": 13577167.0,
              "calibration_gpu_ticks": 13577167.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10923915.0,
              "compute_span_ns": 12152291.0,
              "command_buffer_ns": 11937999.981455505,
              "calibration_cpu_ns": 13641125.0,
              "calibration_gpu_ticks": 13641125.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11049497.0,
              "compute_span_ns": 12280709.0,
              "command_buffer_ns": 12092208.431567997,
              "calibration_cpu_ns": 14057417.0,
              "calibration_gpu_ticks": 14057417.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 148334.0,
              "compute_span_ns": 148334.0,
              "command_buffer_ns": 148374.99475106597,
              "calibration_cpu_ns": 718584.0,
              "calibration_gpu_ticks": 718584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 140542.0,
              "compute_span_ns": 140542.0,
              "command_buffer_ns": 140583.31726118922,
              "calibration_cpu_ns": 779709.0,
              "calibration_gpu_ticks": 779709.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 143750.0,
              "compute_span_ns": 143750.0,
              "command_buffer_ns": 143749.9886378646,
              "calibration_cpu_ns": 822584.0,
              "calibration_gpu_ticks": 822584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 297084.0,
              "compute_span_ns": 297084.0,
              "command_buffer_ns": 297124.9632537365,
              "calibration_cpu_ns": 1918500.0,
              "calibration_gpu_ticks": 1918500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 299500.0,
              "compute_span_ns": 299500.0,
              "command_buffer_ns": 299500.0104419887,
              "calibration_cpu_ns": 1849917.0,
              "calibration_gpu_ticks": 1849917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 161458.0,
              "compute_span_ns": 161458.0,
              "command_buffer_ns": 161458.38890224695,
              "calibration_cpu_ns": 863750.0,
              "calibration_gpu_ticks": 863750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 162750.0,
              "compute_span_ns": 162750.0,
              "command_buffer_ns": 162750.0168979168,
              "calibration_cpu_ns": 951500.0,
              "calibration_gpu_ticks": 951500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 145958.0,
              "compute_span_ns": 145958.0,
              "command_buffer_ns": 145958.32908526063,
              "calibration_cpu_ns": 740167.0,
              "calibration_gpu_ticks": 740167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 145208.0,
              "compute_span_ns": 145208.0,
              "command_buffer_ns": 145208.38158205152,
              "calibration_cpu_ns": 728708.0,
              "calibration_gpu_ticks": 728708.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            170.207578125,
            170.918015625,
            172.5156875,
            170.8066875,
            173.084609375,
            170.0169375,
            171.40365625,
            170.686171875,
            172.648390625
          ],
          "compute_throughput_us_p50": 170.918015625,
          "compute_throughput_us_p90": 172.735634375,
          "command_buffer_throughput_us": [
            186.76497438718798,
            186.71028738026507,
            188.7311182144913,
            188.64908906834899,
            189.40234258479904,
            185.79101379145868,
            188.79687559092417,
            186.53124971024226,
            188.94075674324995
          ],
          "command_buffer_throughput_us_p50": 188.64908906834899,
          "command_buffer_throughput_us_p90": 189.03307391155977,
          "compute_latency_us": [
            148.334,
            140.542,
            143.75,
            297.084,
            299.5,
            161.458,
            162.75,
            145.958,
            145.208
          ],
          "compute_latency_us_p50": 148.334,
          "compute_latency_us_p90": 297.5672,
          "command_buffer_latency_us": [
            148.37499475106597,
            140.58331726118922,
            143.7499886378646,
            297.1249632537365,
            299.5000104419887,
            161.45838890224695,
            162.7500168979168,
            145.95832908526063,
            145.20838158205152
          ],
          "command_buffer_latency_us_p50": 148.37499475106597,
          "command_buffer_latency_us_p90": 297.59997269138694,
          "counter_control_throughput_ratio": 1.1362187641234462,
          "counter_control_latency_ratio": 0.8991230646854678
        },
        "download_ms": 0.656167,
        "allocation_upload_ms": 2.753708,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 7.553236784944503e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-yts8etrn/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "5e168786762f89acc561194a669a3725da4fe77bed5800b1d573576d096bb3fc",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.744958,
        "capture_ms": 0.075166,
        "compile_ms": 54.888708,
        "allocation_upload_ms": 6.949,
        "cold_call_ms": 8.062583,
        "warmup_ms": 203.753833,
        "download_ms": 3.339541,
        "repetitions": 380,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          91.8474789474,
          90.5198447368,
          92.8248894737,
          95.4550447368,
          91.1878289474,
          91.2747815789,
          91.3796052632,
          98.3128289474,
          92.0017552632
        ],
        "latency_us": [
          354.125,
          325.458,
          291.75,
          324.709,
          291.875,
          297.125,
          532.583,
          1976.334,
          1041.208
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 384,
            "reduction_scalar_rounds": 120,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 24576,
              "global_write_bytes": 24576,
              "private_read_bytes": 147456,
              "private_write_bytes": 49152
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 96,
              "private_read_bytes": 576,
              "private_write_bytes": 192
            },
            "striped_storage_scalars_per_worker": 48,
            "reduction_operations": 2,
            "reduction_elements": 12288,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 160,
            "normalized_kernel_cost": 1920,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 18432,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4519791,
              "compute_span_ns": 4519791,
              "calibration_cpu_ns": 5475791,
              "calibration_gpu_ticks": 5475791,
              "compute_passes": 1,
              "command_buffer_ns": 4519791.69156,
              "command_buffers": 2
            },
            {
              "compute_ns": 4379959,
              "compute_span_ns": 4379959,
              "calibration_cpu_ns": 5638125,
              "calibration_gpu_ticks": 5638125,
              "compute_passes": 1,
              "command_buffer_ns": 4380125.02855,
              "command_buffers": 2
            },
            {
              "compute_ns": 4586583,
              "compute_span_ns": 4586583,
              "calibration_cpu_ns": 5791750,
              "calibration_gpu_ticks": 5791750,
              "compute_passes": 1,
              "command_buffer_ns": 4586708.26629,
              "command_buffers": 2
            },
            {
              "compute_ns": 4467875,
              "compute_span_ns": 4467875,
              "calibration_cpu_ns": 5572083,
              "calibration_gpu_ticks": 5572083,
              "compute_passes": 1,
              "command_buffer_ns": 4467916.67491,
              "command_buffers": 2
            },
            {
              "compute_ns": 4174291,
              "compute_span_ns": 4174291,
              "calibration_cpu_ns": 5702709,
              "calibration_gpu_ticks": 5702709,
              "compute_passes": 1,
              "command_buffer_ns": 4174250.00342,
              "command_buffers": 2
            },
            {
              "compute_ns": 4410500,
              "compute_span_ns": 4410500,
              "calibration_cpu_ns": 5586084,
              "calibration_gpu_ticks": 5586084,
              "compute_passes": 1,
              "command_buffer_ns": 4410541.67459,
              "command_buffers": 2
            },
            {
              "compute_ns": 4663250,
              "compute_span_ns": 4663250,
              "calibration_cpu_ns": 5746333,
              "calibration_gpu_ticks": 5746333,
              "compute_passes": 1,
              "command_buffer_ns": 4663291.62009,
              "command_buffers": 2
            },
            {
              "compute_ns": 4441250,
              "compute_span_ns": 4441250,
              "calibration_cpu_ns": 5617792,
              "calibration_gpu_ticks": 5617792,
              "compute_passes": 1,
              "command_buffer_ns": 4441249.99922,
              "command_buffers": 2
            },
            {
              "compute_ns": 4130041,
              "compute_span_ns": 4130041,
              "calibration_cpu_ns": 5419917,
              "calibration_gpu_ticks": 5419917,
              "compute_passes": 1,
              "command_buffer_ns": 4130041.6342,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 77792,
              "compute_span_ns": 77792,
              "calibration_cpu_ns": 671625,
              "calibration_gpu_ticks": 671625,
              "compute_passes": 1,
              "command_buffer_ns": 77916.7166911,
              "command_buffers": 2
            },
            {
              "compute_ns": 64667,
              "compute_span_ns": 64667,
              "calibration_cpu_ns": 570875,
              "calibration_gpu_ticks": 570875,
              "compute_passes": 1,
              "command_buffer_ns": 64791.7040624,
              "command_buffers": 2
            },
            {
              "compute_ns": 109584,
              "compute_span_ns": 109584,
              "calibration_cpu_ns": 821417,
              "calibration_gpu_ticks": 821417,
              "compute_passes": 1,
              "command_buffer_ns": 109624.990728,
              "command_buffers": 2
            },
            {
              "compute_ns": 76917,
              "compute_span_ns": 76917,
              "calibration_cpu_ns": 622292,
              "calibration_gpu_ticks": 622292,
              "compute_passes": 1,
              "command_buffer_ns": 77083.3576098,
              "command_buffers": 2
            },
            {
              "compute_ns": 63791,
              "compute_span_ns": 63791,
              "calibration_cpu_ns": 522500,
              "calibration_gpu_ticks": 522500,
              "compute_passes": 1,
              "command_buffer_ns": 63750.0197627,
              "command_buffers": 2
            },
            {
              "compute_ns": 113333,
              "compute_span_ns": 113333,
              "calibration_cpu_ns": 819584,
              "calibration_gpu_ticks": 819584,
              "compute_passes": 1,
              "command_buffer_ns": 113416.695967,
              "command_buffers": 2
            },
            {
              "compute_ns": 64250,
              "compute_span_ns": 64250,
              "calibration_cpu_ns": 579250,
              "calibration_gpu_ticks": 579250,
              "compute_passes": 1,
              "command_buffer_ns": 64250.0235699,
              "command_buffers": 2
            },
            {
              "compute_ns": 63375,
              "compute_span_ns": 63375,
              "calibration_cpu_ns": 553375,
              "calibration_gpu_ticks": 553375,
              "compute_passes": 1,
              "command_buffer_ns": 63458.2829662,
              "command_buffers": 2
            },
            {
              "compute_ns": 116416,
              "compute_span_ns": 116416,
              "calibration_cpu_ns": 814000,
              "calibration_gpu_ticks": 814000,
              "compute_passes": 1,
              "command_buffer_ns": 116416.660603,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4356291.66896,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4363666.69601,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4355125.01281,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4552291.648,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4579041.67706,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4463041.6669,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4442583.3039,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4449874.91937,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4494208.31865,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 93458.336778,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 113791.669719,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 74041.6580811,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 112291.658297,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 78833.3709352,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 346624.990925,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 102958.2927,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 107958.330773,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 64124.9935143,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              68.06705732750001,
              68.18229212515625,
              68.04882832515625,
              71.129557,
              71.54752620406249,
              69.7350260453125,
              69.4153641234375,
              69.52929561515626,
              70.22200497890624
            ],
            "command_buffer_throughput_us_p50": 69.52929561515626,
            "command_buffer_throughput_us_p90": 71.2131508408125,
            "command_buffer_latency_us": [
              93.458336778,
              113.791669719,
              74.0416580811,
              112.291658297,
              78.8333709352,
              346.62499092499996,
              102.9582927,
              107.958330773,
              64.1249935143
            ],
            "command_buffer_latency_us_p50": 102.9582927,
            "command_buffer_latency_us_p90": 160.35833396020004
          },
          "compute_throughput_us": [
            70.621734375,
            68.436859375,
            71.665359375,
            69.810546875,
            65.223296875,
            68.9140625,
            72.86328125,
            69.39453125,
            64.531890625
          ],
          "compute_throughput_us_p50": 69.39453125,
          "compute_throughput_us_p90": 71.90494375,
          "command_buffer_throughput_us": [
            70.621745180625,
            68.43945357109375,
            71.66731666078124,
            69.81119804546874,
            65.2226563034375,
            68.91471366546875,
            72.86393156390625,
            69.3945312378125,
            64.531900534375
          ],
          "command_buffer_throughput_us_p50": 69.3945312378125,
          "command_buffer_throughput_us_p90": 71.90663964140624,
          "compute_latency_us": [
            77.792,
            64.667,
            109.584,
            76.917,
            63.791,
            113.333,
            64.25,
            63.375,
            116.416
          ],
          "compute_latency_us_p50": 76.917,
          "compute_latency_us_p90": 113.9496,
          "command_buffer_latency_us": [
            77.91671669109999,
            64.7917040624,
            109.624990728,
            77.0833576098,
            63.7500197627,
            113.41669596700001,
            64.2500235699,
            63.4582829662,
            116.416660603
          ],
          "command_buffer_latency_us_p50": 77.0833576098,
          "command_buffer_latency_us_p90": 114.0166888942,
          "counter_control_throughput_ratio": 0.9980617612166003,
          "counter_control_latency_ratio": 0.6864566680983762
        },
        "correctness": {
          "max_abs_error": 7.116789644580346e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 91.8474789474,
        "throughput_us_p90": 96.02660157892001,
        "latency_us_p50": 325.458,
        "latency_us_p90": 1228.2332000000001
      },
      "slowdown": 0.5061169800718545,
      "device_compute_slowdown": 0.40601063027933804,
      "valid": true,
      "round": 0,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "softmax_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-2hns1lsn/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "117627a7b60c4cee3b9bce9542bc5f280f025f1ad809d52b8d22cb3cdd24d0d1",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.573416,
        "capture_ms": 0.085209,
        "compile_ms": 53.450542,
        "allocation_upload_ms": 6.961625,
        "cold_call_ms": 12.690541,
        "warmup_ms": 200.60225,
        "download_ms": 4.253542,
        "repetitions": 477,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          75.4270628931,
          81.5457714885,
          75.5539832285,
          80.9418238994,
          75.6175765199,
          81.3245974843,
          75.2530566038,
          82.3698469602,
          75.5038427673
        ],
        "latency_us": [
          2247.417,
          559.125,
          386.084,
          448.083,
          276.542,
          324.75,
          288.583,
          284.125,
          286.167
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 512,
            "reduction_scalar_rounds": 80,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 16384,
              "global_write_bytes": 16384,
              "private_read_bytes": 98304,
              "private_write_bytes": 32768
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 64,
              "global_write_bytes": 64,
              "private_read_bytes": 384,
              "private_write_bytes": 128
            },
            "striped_storage_scalars_per_worker": 32,
            "reduction_operations": 2,
            "reduction_elements": 8192,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 120,
            "normalized_kernel_cost": 1920,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3747917,
              "compute_span_ns": 3747917,
              "calibration_cpu_ns": 4740291,
              "calibration_gpu_ticks": 4740291,
              "compute_passes": 1,
              "command_buffer_ns": 3747958.34068,
              "command_buffers": 2
            },
            {
              "compute_ns": 4007125,
              "compute_span_ns": 4007125,
              "calibration_cpu_ns": 5025459,
              "calibration_gpu_ticks": 5025459,
              "compute_passes": 1,
              "command_buffer_ns": 4007333.36573,
              "command_buffers": 2
            },
            {
              "compute_ns": 3888834,
              "compute_span_ns": 3888834,
              "calibration_cpu_ns": 5092625,
              "calibration_gpu_ticks": 5092625,
              "compute_passes": 1,
              "command_buffer_ns": 3888916.57349,
              "command_buffers": 2
            },
            {
              "compute_ns": 3837209,
              "compute_span_ns": 3837209,
              "calibration_cpu_ns": 5011667,
              "calibration_gpu_ticks": 5011667,
              "compute_passes": 1,
              "command_buffer_ns": 3837333.35184,
              "command_buffers": 2
            },
            {
              "compute_ns": 3746375,
              "compute_span_ns": 3746375,
              "calibration_cpu_ns": 4977791,
              "calibration_gpu_ticks": 4977791,
              "compute_passes": 1,
              "command_buffer_ns": 3746416.65258,
              "command_buffers": 2
            },
            {
              "compute_ns": 3674541,
              "compute_span_ns": 3674541,
              "calibration_cpu_ns": 5110417,
              "calibration_gpu_ticks": 5110417,
              "compute_passes": 1,
              "command_buffer_ns": 3674500.03978,
              "command_buffers": 2
            },
            {
              "compute_ns": 3743666,
              "compute_span_ns": 3743666,
              "calibration_cpu_ns": 5029041,
              "calibration_gpu_ticks": 5029041,
              "compute_passes": 1,
              "command_buffer_ns": 3743750.04321,
              "command_buffers": 2
            },
            {
              "compute_ns": 3651333,
              "compute_span_ns": 3651333,
              "calibration_cpu_ns": 5000541,
              "calibration_gpu_ticks": 5000541,
              "compute_passes": 1,
              "command_buffer_ns": 3651333.33253,
              "command_buffers": 2
            },
            {
              "compute_ns": 6113416,
              "compute_span_ns": 6113416,
              "calibration_cpu_ns": 7158541,
              "calibration_gpu_ticks": 7158541,
              "compute_passes": 1,
              "command_buffer_ns": 6113458.3666,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 106958,
              "compute_span_ns": 106958,
              "calibration_cpu_ns": 917875,
              "calibration_gpu_ticks": 917875,
              "compute_passes": 1,
              "command_buffer_ns": 107041.676529,
              "command_buffers": 2
            },
            {
              "compute_ns": 55583,
              "compute_span_ns": 55583,
              "calibration_cpu_ns": 657791,
              "calibration_gpu_ticks": 657791,
              "compute_passes": 1,
              "command_buffer_ns": 55541.691836,
              "command_buffers": 2
            },
            {
              "compute_ns": 99333,
              "compute_span_ns": 99333,
              "calibration_cpu_ns": 780834,
              "calibration_gpu_ticks": 780834,
              "compute_passes": 1,
              "command_buffer_ns": 99375.0290945,
              "command_buffers": 2
            },
            {
              "compute_ns": 67125,
              "compute_span_ns": 67125,
              "calibration_cpu_ns": 493167,
              "calibration_gpu_ticks": 493167,
              "compute_passes": 1,
              "command_buffer_ns": 67125.0163577,
              "command_buffers": 2
            },
            {
              "compute_ns": 58042,
              "compute_span_ns": 58042,
              "calibration_cpu_ns": 474792,
              "calibration_gpu_ticks": 474792,
              "compute_passes": 1,
              "command_buffer_ns": 58125.006035,
              "command_buffers": 2
            },
            {
              "compute_ns": 268708,
              "compute_span_ns": 268708,
              "calibration_cpu_ns": 1677167,
              "calibration_gpu_ticks": 1677167,
              "compute_passes": 1,
              "command_buffer_ns": 268750.009127,
              "command_buffers": 2
            },
            {
              "compute_ns": 56083,
              "compute_span_ns": 56083,
              "calibration_cpu_ns": 568750,
              "calibration_gpu_ticks": 568750,
              "compute_passes": 1,
              "command_buffer_ns": 56124.9325983,
              "command_buffers": 2
            },
            {
              "compute_ns": 99417,
              "compute_span_ns": 99417,
              "calibration_cpu_ns": 2650292,
              "calibration_gpu_ticks": 2650292,
              "compute_passes": 1,
              "command_buffer_ns": 99458.3242573,
              "command_buffers": 2
            },
            {
              "compute_ns": 177375,
              "compute_span_ns": 177375,
              "calibration_cpu_ns": 1531292,
              "calibration_gpu_ticks": 1531292,
              "compute_passes": 1,
              "command_buffer_ns": 177458.336111,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 3777750.01107,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3955458.34908,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3882875.02574,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3815375.03563,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3918124.9449,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3894000.02314,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3814541.61834,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6079250.01532,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6087124.98797,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 58208.2429901,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 53666.5902473,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 100083.35812,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 99708.3261609,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 53750.0018254,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 98708.3185464,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 56958.3498873,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54541.7424291,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 104375.008959,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              59.02734392296875,
              61.804036704375,
              60.669922277187496,
              59.615234931718746,
              61.2207022640625,
              60.843750361562506,
              59.602212786562504,
              94.988281489375,
              95.11132793703125
            ],
            "command_buffer_throughput_us_p50": 60.843750361562506,
            "command_buffer_throughput_us_p90": 95.01289077890625,
            "command_buffer_latency_us": [
              58.2082429901,
              53.6665902473,
              100.08335812,
              99.7083261609,
              53.750001825400005,
              98.7083185464,
              56.9583498873,
              54.5417424291,
              104.375008959
            ],
            "command_buffer_latency_us_p50": 58.2082429901,
            "command_buffer_latency_us_p90": 100.9416882878
          },
          "compute_throughput_us": [
            58.561203125,
            62.611328125,
            60.76303125,
            59.956390625,
            58.537109375,
            57.414703125,
            58.49478125,
            57.052078125,
            95.522125
          ],
          "compute_throughput_us_p50": 58.561203125,
          "compute_throughput_us_p90": 69.1934875,
          "command_buffer_throughput_us": [
            58.561849073124996,
            62.61458383953125,
            60.76432146078125,
            59.9583336225,
            58.5377601965625,
            57.414063121562506,
            58.496094425156244,
            57.05208332078125,
            95.522786978125
          ],
          "command_buffer_throughput_us_p50": 58.561849073124996,
          "command_buffer_throughput_us_p90": 69.19622446725,
          "compute_latency_us": [
            106.958,
            55.583,
            99.333,
            67.125,
            58.042,
            268.708,
            56.083,
            99.417,
            177.375
          ],
          "compute_latency_us_p50": 99.333,
          "compute_latency_us_p90": 195.6416,
          "command_buffer_latency_us": [
            107.041676529,
            55.541691836,
            99.37502909450001,
            67.1250163577,
            58.125006035,
            268.750009127,
            56.1249325983,
            99.4583242573,
            177.45833611100002
          ],
          "command_buffer_latency_us_p50": 99.37502909450001,
          "command_buffer_latency_us_p90": 195.71667071420003,
          "counter_control_throughput_ratio": 0.9921139116398118,
          "counter_control_latency_ratio": 1.0813954243910844
        },
        "correctness": {
          "max_abs_error": 1.2343096304160556e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 75.6175765199,
        "throughput_us_p90": 81.71058658284,
        "latency_us_p50": 324.75,
        "latency_us_p90": 896.7834000000003
      },
      "torch": {
        "cold_call_ms": 7.836417,
        "warmup_ms": 201.091625,
        "repetitions": 182,
        "throughput_us": [
          165.75686813186815,
          163.7728901098901,
          174.13804945054946,
          172.15086813186815,
          161.07051648351649,
          171.9471153846154,
          172.9020164835165,
          163.4494065934066,
          169.35828571428573
        ],
        "latency_us": [
          576.4169999999999,
          438.583,
          441.29200000000003,
          2396.292,
          702.583,
          430.375,
          553.416,
          482.208,
          449.25
        ],
        "throughput_us_p50": 169.35828571428573,
        "throughput_us_p90": 173.14922307692308,
        "latency_us_p50": 482.208,
        "latency_us_p90": 1041.3248000000003,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 9757666.499353945,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9820166.742429137,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9801958.571188152,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9897291.602101177,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9672708.285506815,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9831416.711676866,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9788958.413992077,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9887499.909382313,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9827791.713178158,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 150333.33329483867,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 264499.9767653644,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 134125.00265985727,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 313291.61720350385,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 145583.35533365607,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 140624.99394640326,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 123250.00716373324,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 125249.96418505907,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 128791.66752099991,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              152.4635390524054,
              153.44010535045527,
              153.15560267481487,
              154.6451812828309,
              151.13606696104398,
              153.61588611995103,
              152.9524752186262,
              154.49218608409865,
              153.55924551840872
            ],
            "command_buffer_throughput_us_p50": 153.44010535045527,
            "command_buffer_throughput_us_p90": 154.5227851238451,
            "command_buffer_latency_us": [
              150.33333329483867,
              264.4999767653644,
              134.12500265985727,
              313.29161720350385,
              145.58335533365607,
              140.62499394640326,
              123.25000716373324,
              125.24996418505907,
              128.7916675209999
            ],
            "command_buffer_latency_us_p50": 140.62499394640326,
            "command_buffer_latency_us_p90": 274.2583048529923
          },
          "throughput": [
            {
              "compute_ns": 10161083.0,
              "compute_span_ns": 11378375.0,
              "command_buffer_ns": 11192875.099368393,
              "calibration_cpu_ns": 13124125.0,
              "calibration_gpu_ticks": 13124125.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9971376.0,
              "compute_span_ns": 11245875.0,
              "command_buffer_ns": 11029625.07866323,
              "calibration_cpu_ns": 12827000.0,
              "calibration_gpu_ticks": 12827000.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9993001.0,
              "compute_span_ns": 11306708.0,
              "command_buffer_ns": 11114708.380773664,
              "calibration_cpu_ns": 13088000.0,
              "calibration_gpu_ticks": 13088000.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10086043.0,
              "compute_span_ns": 11300375.0,
              "command_buffer_ns": 11120499.984826893,
              "calibration_cpu_ns": 12861750.0,
              "calibration_gpu_ticks": 12861750.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10085872.0,
              "compute_span_ns": 11347416.0,
              "command_buffer_ns": 11150708.36385712,
              "calibration_cpu_ns": 12846458.0,
              "calibration_gpu_ticks": 12846458.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10077209.0,
              "compute_span_ns": 11250167.0,
              "command_buffer_ns": 11064500.024076551,
              "calibration_cpu_ns": 12590916.0,
              "calibration_gpu_ticks": 12590916.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10263795.0,
              "compute_span_ns": 11427292.0,
              "command_buffer_ns": 11250791.721977293,
              "calibration_cpu_ns": 12698250.0,
              "calibration_gpu_ticks": 12698250.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9994502.0,
              "compute_span_ns": 11103750.0,
              "command_buffer_ns": 10929708.369076252,
              "calibration_cpu_ns": 12608083.0,
              "calibration_gpu_ticks": 12608083.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10043420.0,
              "compute_span_ns": 11239041.0,
              "command_buffer_ns": 11060124.84524399,
              "calibration_cpu_ns": 12939000.0,
              "calibration_gpu_ticks": 12939000.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 125292.0,
              "compute_span_ns": 125292.0,
              "command_buffer_ns": 125291.64087027311,
              "calibration_cpu_ns": 855917.0,
              "calibration_gpu_ticks": 855917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 125750.0,
              "compute_span_ns": 125750.0,
              "command_buffer_ns": 125750.02619996667,
              "calibration_cpu_ns": 651584.0,
              "calibration_gpu_ticks": 651584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 128709.0,
              "compute_span_ns": 128709.0,
              "command_buffer_ns": 128749.99083578587,
              "calibration_cpu_ns": 639084.0,
              "calibration_gpu_ticks": 639084.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 125709.0,
              "compute_span_ns": 125709.0,
              "command_buffer_ns": 125750.02619996667,
              "calibration_cpu_ns": 621792.0,
              "calibration_gpu_ticks": 621792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 129791.0,
              "compute_span_ns": 129791.0,
              "command_buffer_ns": 129749.99845027924,
              "calibration_cpu_ns": 821708.0,
              "calibration_gpu_ticks": 821708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 271833.0,
              "compute_span_ns": 271833.0,
              "command_buffer_ns": 271791.7086556554,
              "calibration_cpu_ns": 1819708.0,
              "calibration_gpu_ticks": 1819708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 159375.0,
              "compute_span_ns": 159375.0,
              "command_buffer_ns": 159374.9620951712,
              "calibration_cpu_ns": 1657208.0,
              "calibration_gpu_ticks": 1657208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 137125.0,
              "compute_span_ns": 137125.0,
              "command_buffer_ns": 137124.96729567647,
              "calibration_cpu_ns": 914084.0,
              "calibration_gpu_ticks": 914084.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 313958.0,
              "compute_span_ns": 313958.0,
              "command_buffer_ns": 313958.3277516067,
              "calibration_cpu_ns": 1559542.0,
              "calibration_gpu_ticks": 1559542.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            158.766921875,
            155.80275,
            156.140640625,
            157.594421875,
            157.59175,
            157.456390625,
            160.371796875,
            156.16409375,
            156.9284375
          ],
          "compute_throughput_us_p50": 157.456390625,
          "compute_throughput_us_p90": 159.087896875,
          "command_buffer_throughput_us": [
            174.88867342763115,
            172.33789185411297,
            173.6673184495885,
            173.7578122629202,
            174.2298181852675,
            172.8828128761961,
            175.7936206558952,
            170.77669326681644,
            172.81445070693735
          ],
          "command_buffer_throughput_us_p50": 173.6673184495885,
          "command_buffer_throughput_us_p90": 175.06966287328396,
          "compute_latency_us": [
            125.292,
            125.75,
            128.709,
            125.709,
            129.791,
            271.833,
            159.375,
            137.125,
            313.958
          ],
          "compute_latency_us_p50": 129.791,
          "compute_latency_us_p90": 280.25800000000004,
          "command_buffer_latency_us": [
            125.29164087027311,
            125.75002619996667,
            128.74999083578587,
            125.75002619996667,
            129.74999845027924,
            271.7917086556554,
            159.3749620951712,
            137.12496729567647,
            313.9583277516067
          ],
          "command_buffer_latency_us_p50": 129.74999845027924,
          "command_buffer_latency_us_p90": 280.22503247484565,
          "counter_control_throughput_ratio": 1.1254227491888469,
          "counter_control_latency_ratio": 0.9599253553216882
        },
        "download_ms": 0.890708,
        "allocation_upload_ms": 17.267334,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 9.74135319392784e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.44649469732747477,
      "device_compute_slowdown": 0.3719201417773512,
      "valid": true,
      "round": 0,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "softmax_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.56175,
        "warmup_ms": 200.174042,
        "repetitions": 165,
        "throughput_us": [
          165.99898787878786,
          183.69545454545454,
          164.73333333333332,
          169.83737575757576,
          172.15075757575758,
          164.7300484848485,
          173.41287878787878,
          168.14646666666667,
          165.61767272727275
        ],
        "latency_us": [
          592.875,
          2395.6659999999997,
          801.041,
          422.29100000000005,
          525.375,
          422.084,
          484.125,
          408.45799999999997,
          406.208
        ],
        "throughput_us_p50": 168.14646666666667,
        "throughput_us_p90": 175.46939393939394,
        "latency_us_p50": 484.125,
        "latency_us_p90": 1119.9660000000003,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 9737749.93326515,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9754208.498634398,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9818499.96605888,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9876625.030301511,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9699749.993160367,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9830291.67400673,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9903416.619636118,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9849666.617810726,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9902708.290610462,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 187625.00258162618,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 282208.37702974677,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 141958.35683494806,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 130874.97791275382,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 128458.31224694848,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 131666.6603088379,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 283291.6798070073,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 313666.6491627693,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 138791.62725061178,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              152.15234270726796,
              152.40950779116247,
              153.41406196967,
              154.3222660984611,
              151.55859364313073,
              153.59830740635516,
              154.74088468181435,
              153.9010409032926,
              154.72981704078848
            ],
            "command_buffer_throughput_us_p50": 153.59830740635516,
            "command_buffer_throughput_us_p90": 154.73203056899365,
            "command_buffer_latency_us": [
              187.62500258162618,
              282.20837702974677,
              141.95835683494806,
              130.87497791275382,
              128.45831224694848,
              131.6666603088379,
              283.2916798070073,
              313.6666491627693,
              138.79162725061178
            ],
            "command_buffer_latency_us_p50": 141.95835683494806,
            "command_buffer_latency_us_p90": 289.3666736781597
          },
          "throughput": [
            {
              "compute_ns": 10063496.0,
              "compute_span_ns": 11395584.0,
              "command_buffer_ns": 11200624.925550073,
              "calibration_cpu_ns": 13139500.0,
              "calibration_gpu_ticks": 13139500.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10019501.0,
              "compute_span_ns": 11229125.0,
              "command_buffer_ns": 11054916.714783758,
              "calibration_cpu_ns": 13217250.0,
              "calibration_gpu_ticks": 13217250.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10079081.0,
              "compute_span_ns": 11226208.0,
              "command_buffer_ns": 11017958.2843557,
              "calibration_cpu_ns": 12804084.0,
              "calibration_gpu_ticks": 12804084.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10115495.0,
              "compute_span_ns": 11244166.0,
              "command_buffer_ns": 11045916.588045657,
              "calibration_cpu_ns": 12771375.0,
              "calibration_gpu_ticks": 12771375.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10075545.0,
              "compute_span_ns": 11350459.0,
              "command_buffer_ns": 11146500.008180737,
              "calibration_cpu_ns": 12784833.0,
              "calibration_gpu_ticks": 12784833.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10154751.0,
              "compute_span_ns": 11346750.0,
              "command_buffer_ns": 11162708.39702338,
              "calibration_cpu_ns": 12955500.0,
              "calibration_gpu_ticks": 12955500.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10391995.0,
              "compute_span_ns": 11566791.0,
              "command_buffer_ns": 11395541.66002199,
              "calibration_cpu_ns": 13288917.0,
              "calibration_gpu_ticks": 13288917.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10017413.0,
              "compute_span_ns": 11190125.0,
              "command_buffer_ns": 11034916.56249389,
              "calibration_cpu_ns": 12355583.0,
              "calibration_gpu_ticks": 12355583.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10265796.0,
              "compute_span_ns": 11428042.0,
              "command_buffer_ns": 11269250.011537224,
              "calibration_cpu_ns": 12584166.0,
              "calibration_gpu_ticks": 12584166.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 137667.0,
              "compute_span_ns": 137667.0,
              "command_buffer_ns": 137708.32447335124,
              "calibration_cpu_ns": 946542.0,
              "calibration_gpu_ticks": 946542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 127667.0,
              "compute_span_ns": 127667.0,
              "command_buffer_ns": 127666.68805852532,
              "calibration_cpu_ns": 652875.0,
              "calibration_gpu_ticks": 652875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 124084.0,
              "compute_span_ns": 124084.0,
              "command_buffer_ns": 124125.0429302454,
              "calibration_cpu_ns": 623791.0,
              "calibration_gpu_ticks": 623791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 191916.0,
              "compute_span_ns": 191916.0,
              "command_buffer_ns": 191875.034943223,
              "calibration_cpu_ns": 1819833.0,
              "calibration_gpu_ticks": 1819833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 190625.0,
              "compute_span_ns": 190625.0,
              "command_buffer_ns": 190625.0254251063,
              "calibration_cpu_ns": 1687041.0,
              "calibration_gpu_ticks": 1687041.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 136666.0,
              "compute_span_ns": 136666.0,
              "command_buffer_ns": 136624.96348842978,
              "calibration_cpu_ns": 871666.0,
              "calibration_gpu_ticks": 871666.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 143333.0,
              "compute_span_ns": 143333.0,
              "command_buffer_ns": 143291.66151583195,
              "calibration_cpu_ns": 969542.0,
              "calibration_gpu_ticks": 969542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 123791.0,
              "compute_span_ns": 123791.0,
              "command_buffer_ns": 123750.01097097993,
              "calibration_cpu_ns": 673833.0,
              "calibration_gpu_ticks": 673833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 120667.0,
              "compute_span_ns": 120667.0,
              "command_buffer_ns": 120708.31144228578,
              "calibration_cpu_ns": 623416.0,
              "calibration_gpu_ticks": 623416.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            157.242125,
            156.554703125,
            157.485640625,
            158.054609375,
            157.430390625,
            158.667984375,
            162.374921875,
            156.522078125,
            160.4030625
          ],
          "compute_throughput_us_p50": 157.485640625,
          "compute_throughput_us_p90": 160.797434375,
          "command_buffer_throughput_us": [
            175.0097644617199,
            172.73307366849622,
            172.1555981930578,
            172.59244668821339,
            174.16406262782402,
            174.4173187034903,
            178.0553384378436,
            172.42057128896704,
            176.08203143026913
          ],
          "command_buffer_throughput_us_p50": 174.16406262782402,
          "command_buffer_throughput_us_p90": 176.47669283178402,
          "compute_latency_us": [
            137.667,
            127.667,
            124.084,
            191.916,
            190.625,
            136.666,
            143.333,
            123.791,
            120.667
          ],
          "compute_latency_us_p50": 136.666,
          "compute_latency_us_p90": 190.8832,
          "command_buffer_latency_us": [
            137.70832447335124,
            127.66668805852532,
            124.1250429302454,
            191.875034943223,
            190.6250254251063,
            136.62496348842978,
            143.29166151583195,
            123.75001097097993,
            120.70831144228578
          ],
          "command_buffer_latency_us_p50": 136.62496348842978,
          "command_buffer_latency_us_p90": 190.87502732872963,
          "counter_control_throughput_ratio": 1.1355419317353348,
          "counter_control_latency_ratio": 0.869708885423805
        },
        "download_ms": 0.666417,
        "allocation_upload_ms": 0.815625,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 9.74135319392784e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-3krqg6bl/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "8bb581805080afef9f47b3147742261e73f0bbc7f1d60afd69631b609a5e3f17",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.152667,
        "capture_ms": 0.065417,
        "compile_ms": 53.664625,
        "allocation_upload_ms": 6.232625,
        "cold_call_ms": 5.310833,
        "warmup_ms": 200.55125,
        "download_ms": 3.276958,
        "repetitions": 417,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          76.5001990408,
          72.9216618705,
          73.2910671463,
          76.5043980815,
          76.6988417266,
          71.7919664269,
          74.1902470024,
          76.5946235012,
          76.3673069544
        ],
        "latency_us": [
          406.917,
          386.666,
          335.875,
          337.167,
          285.667,
          295.667,
          286.75,
          292.125,
          321.25
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 1024,
            "reduction_scalar_rounds": 80,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 16384,
              "global_write_bytes": 16384,
              "private_read_bytes": 98304,
              "private_write_bytes": 32768
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 64,
              "global_write_bytes": 64,
              "private_read_bytes": 384,
              "private_write_bytes": 128
            },
            "striped_storage_scalars_per_worker": 32,
            "reduction_operations": 2,
            "reduction_elements": 8192,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 128,
            "normalized_kernel_cost": 2048,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3557833,
              "compute_span_ns": 3557833,
              "calibration_cpu_ns": 4135583,
              "calibration_gpu_ticks": 4135583,
              "compute_passes": 1,
              "command_buffer_ns": 3557916.61423,
              "command_buffers": 2
            },
            {
              "compute_ns": 3660166,
              "compute_span_ns": 3660166,
              "calibration_cpu_ns": 4774000,
              "calibration_gpu_ticks": 4774000,
              "compute_passes": 1,
              "command_buffer_ns": 3660208.31279,
              "command_buffers": 2
            },
            {
              "compute_ns": 3693917,
              "compute_span_ns": 3693917,
              "calibration_cpu_ns": 4810208,
              "calibration_gpu_ticks": 4810208,
              "compute_passes": 1,
              "command_buffer_ns": 3693958.33695,
              "command_buffers": 2
            },
            {
              "compute_ns": 3463083,
              "compute_span_ns": 3463083,
              "calibration_cpu_ns": 4711167,
              "calibration_gpu_ticks": 4711167,
              "compute_passes": 1,
              "command_buffer_ns": 3463166.70766,
              "command_buffers": 2
            },
            {
              "compute_ns": 3783834,
              "compute_span_ns": 3783834,
              "calibration_cpu_ns": 4882542,
              "calibration_gpu_ticks": 4882542,
              "compute_passes": 1,
              "command_buffer_ns": 3783958.32377,
              "command_buffers": 2
            },
            {
              "compute_ns": 3625958,
              "compute_span_ns": 3625958,
              "calibration_cpu_ns": 4750375,
              "calibration_gpu_ticks": 4750375,
              "compute_passes": 1,
              "command_buffer_ns": 3625958.34304,
              "command_buffers": 2
            },
            {
              "compute_ns": 3443833,
              "compute_span_ns": 3443833,
              "calibration_cpu_ns": 4966542,
              "calibration_gpu_ticks": 4966542,
              "compute_passes": 1,
              "command_buffer_ns": 3443916.7357,
              "command_buffers": 2
            },
            {
              "compute_ns": 3728209,
              "compute_span_ns": 3728209,
              "calibration_cpu_ns": 5030584,
              "calibration_gpu_ticks": 5030584,
              "compute_passes": 1,
              "command_buffer_ns": 3728333.39497,
              "command_buffers": 2
            },
            {
              "compute_ns": 6032042,
              "compute_span_ns": 6032042,
              "calibration_cpu_ns": 7237667,
              "calibration_gpu_ticks": 7237667,
              "compute_passes": 1,
              "command_buffer_ns": 6032166.65331,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 364167,
              "compute_span_ns": 364167,
              "calibration_cpu_ns": 1898709,
              "calibration_gpu_ticks": 1898709,
              "compute_passes": 1,
              "command_buffer_ns": 364208.302926,
              "command_buffers": 2
            },
            {
              "compute_ns": 101792,
              "compute_span_ns": 101792,
              "calibration_cpu_ns": 596958,
              "calibration_gpu_ticks": 596958,
              "compute_passes": 1,
              "command_buffer_ns": 101791.636553,
              "command_buffers": 2
            },
            {
              "compute_ns": 92208,
              "compute_span_ns": 92208,
              "calibration_cpu_ns": 1048000,
              "calibration_gpu_ticks": 1048000,
              "compute_passes": 1,
              "command_buffer_ns": 92166.6505747,
              "command_buffers": 2
            },
            {
              "compute_ns": 55250,
              "compute_span_ns": 55250,
              "calibration_cpu_ns": 559500,
              "calibration_gpu_ticks": 559500,
              "compute_passes": 1,
              "command_buffer_ns": 55333.3084099,
              "command_buffers": 2
            },
            {
              "compute_ns": 96333,
              "compute_span_ns": 96333,
              "calibration_cpu_ns": 701209,
              "calibration_gpu_ticks": 701209,
              "compute_passes": 1,
              "command_buffer_ns": 96374.9480434,
              "command_buffers": 2
            },
            {
              "compute_ns": 64833,
              "compute_span_ns": 64833,
              "calibration_cpu_ns": 595667,
              "calibration_gpu_ticks": 595667,
              "compute_passes": 1,
              "command_buffer_ns": 64874.9410175,
              "command_buffers": 2
            },
            {
              "compute_ns": 96208,
              "compute_span_ns": 96208,
              "calibration_cpu_ns": 948625,
              "calibration_gpu_ticks": 948625,
              "compute_passes": 1,
              "command_buffer_ns": 96291.7110883,
              "command_buffers": 2
            },
            {
              "compute_ns": 62291,
              "compute_span_ns": 62291,
              "calibration_cpu_ns": 485875,
              "calibration_gpu_ticks": 485875,
              "compute_passes": 1,
              "command_buffer_ns": 62250.0083409,
              "command_buffers": 2
            },
            {
              "compute_ns": 97500,
              "compute_span_ns": 97500,
              "calibration_cpu_ns": 3102708,
              "calibration_gpu_ticks": 3102708,
              "compute_passes": 1,
              "command_buffer_ns": 97541.6623987,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4556625.03373,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3630666.64431,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3748041.63584,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3767875.02319,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3647708.33403,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3609125.03675,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3715416.64936,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3422708.30879,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3402124.97395,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 70500.0129528,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 52791.6126885,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54583.302699,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 53416.704759,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 100750.010461,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 101500.016171,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 52999.9961145,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 52625.0223629,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 98374.9632724,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              71.19726615203125,
              56.72916631734375,
              58.563150560000004,
              58.873047237343755,
              56.99544271921875,
              56.39257869921875,
              58.05338514625,
              53.47981732484375,
              53.15820271796875
            ],
            "command_buffer_throughput_us_p50": 56.99544271921875,
            "command_buffer_throughput_us_p90": 61.337891020281255,
            "command_buffer_latency_us": [
              70.50001295279999,
              52.791612688499995,
              54.583302699,
              53.416704759,
              100.750010461,
              101.500016171,
              52.9999961145,
              52.625022362900005,
              98.3749632724
            ],
            "command_buffer_latency_us_p50": 54.583302699,
            "command_buffer_latency_us_p90": 100.900011603
          },
          "compute_throughput_us": [
            55.591140625,
            57.19009375,
            57.717453125,
            54.110671875,
            59.12240625,
            56.65559375,
            53.809890625,
            58.253265625,
            94.25065625
          ],
          "compute_throughput_us_p50": 57.19009375,
          "compute_throughput_us_p90": 66.14805625000001,
          "command_buffer_throughput_us": [
            55.59244709734375,
            57.19075488734375,
            57.71809901484375,
            54.111979807187495,
            59.124348808906255,
            56.655599110000004,
            53.8111989953125,
            58.25520929640625,
            94.25260395796874
          ],
          "command_buffer_throughput_us_p50": 57.19075488734375,
          "command_buffer_throughput_us_p90": 66.14999983871876,
          "compute_latency_us": [
            364.167,
            101.792,
            92.208,
            55.25,
            96.333,
            64.833,
            96.208,
            62.291,
            97.5
          ],
          "compute_latency_us_p50": 96.208,
          "compute_latency_us_p90": 154.26700000000005,
          "command_buffer_latency_us": [
            364.208302926,
            101.791636553,
            92.1666505747,
            55.3333084099,
            96.3749480434,
            64.8749410175,
            96.2917110883,
            62.2500083409,
            97.54166239870001
          ],
          "command_buffer_latency_us_p50": 96.2917110883,
          "command_buffer_latency_us_p90": 154.27496982760005,
          "counter_control_throughput_ratio": 1.0046640961780473,
          "counter_control_latency_ratio": 1.182897518059498
        },
        "correctness": {
          "max_abs_error": 1.2343096304160556e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 76.3673069544,
        "throughput_us_p90": 76.61546714628,
        "latency_us_p50": 321.25,
        "latency_us_p90": 390.7162
      },
      "slowdown": 0.4541713451867558,
      "device_compute_slowdown": 0.3631448144925118,
      "valid": true,
      "round": 0,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "rmsnorm_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-31vvscl6/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "54d99166e42515c7bfff1cac173ad46750fa91f9c92e668c07d69318fdc050bf",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.166292,
        "capture_ms": 0.075417,
        "compile_ms": 56.895292,
        "allocation_upload_ms": 1.60625,
        "cold_call_ms": 1.991291,
        "warmup_ms": 200.131625,
        "download_ms": 0.309084,
        "repetitions": 3641,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          5.57766849767,
          6.41448091184,
          6.15398709146,
          5.47525872013,
          6.3300032958,
          5.63788547102,
          5.84887393573,
          5.85574018127,
          5.56058308157
        ],
        "latency_us": [
          244.625,
          217.25,
          194.375,
          195.458,
          202.667,
          192.041,
          191.417,
          200.917,
          199.833
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 32,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 37,
            "reduction_scalar_rounds": 24,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 12296,
              "global_write_bytes": 6148,
              "private_read_bytes": 18444,
              "private_write_bytes": 6148
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 64,
              "global_write_bytes": 32,
              "private_read_bytes": 96,
              "private_write_bytes": 32
            },
            "striped_storage_scalars_per_worker": 8,
            "reduction_operations": 1,
            "reduction_elements": 1537,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 56,
            "normalized_kernel_cost": 56,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 3074,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 536584,
              "compute_span_ns": 536584,
              "calibration_cpu_ns": 2219541,
              "calibration_gpu_ticks": 2219541,
              "compute_passes": 1,
              "command_buffer_ns": 536666.71738,
              "command_buffers": 2
            },
            {
              "compute_ns": 291458,
              "compute_span_ns": 291458,
              "calibration_cpu_ns": 769500,
              "calibration_gpu_ticks": 769500,
              "compute_passes": 1,
              "command_buffer_ns": 291500.007734,
              "command_buffers": 2
            },
            {
              "compute_ns": 265000,
              "compute_span_ns": 265000,
              "calibration_cpu_ns": 698875,
              "calibration_gpu_ticks": 698875,
              "compute_passes": 1,
              "command_buffer_ns": 265041.715465,
              "command_buffers": 2
            },
            {
              "compute_ns": 247209,
              "compute_span_ns": 247209,
              "calibration_cpu_ns": 609250,
              "calibration_gpu_ticks": 609250,
              "compute_passes": 1,
              "command_buffer_ns": 247333.373409,
              "command_buffers": 2
            },
            {
              "compute_ns": 273417,
              "compute_span_ns": 273417,
              "calibration_cpu_ns": 781833,
              "calibration_gpu_ticks": 781833,
              "compute_passes": 1,
              "command_buffer_ns": 273541.663773,
              "command_buffers": 2
            },
            {
              "compute_ns": 741959,
              "compute_span_ns": 741959,
              "calibration_cpu_ns": 2384084,
              "calibration_gpu_ticks": 2384084,
              "compute_passes": 1,
              "command_buffer_ns": 742041.680496,
              "command_buffers": 2
            },
            {
              "compute_ns": 655500,
              "compute_span_ns": 655500,
              "calibration_cpu_ns": 2144125,
              "calibration_gpu_ticks": 2144125,
              "compute_passes": 1,
              "command_buffer_ns": 655541.662127,
              "command_buffers": 2
            },
            {
              "compute_ns": 266375,
              "compute_span_ns": 266375,
              "calibration_cpu_ns": 821750,
              "calibration_gpu_ticks": 821750,
              "compute_passes": 1,
              "command_buffer_ns": 266499.991994,
              "command_buffers": 2
            },
            {
              "compute_ns": 652292,
              "compute_span_ns": 652292,
              "calibration_cpu_ns": 1947792,
              "calibration_gpu_ticks": 1947792,
              "compute_passes": 1,
              "command_buffer_ns": 652333.372273,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 11708,
              "compute_span_ns": 11708,
              "calibration_cpu_ns": 325041,
              "calibration_gpu_ticks": 325041,
              "compute_passes": 1,
              "command_buffer_ns": 11791.7079479,
              "command_buffers": 2
            },
            {
              "compute_ns": 10291,
              "compute_span_ns": 10291,
              "calibration_cpu_ns": 420167,
              "calibration_gpu_ticks": 420167,
              "compute_passes": 1,
              "command_buffer_ns": 10250.0198409,
              "command_buffers": 2
            },
            {
              "compute_ns": 6500,
              "compute_span_ns": 6500,
              "calibration_cpu_ns": 394292,
              "calibration_gpu_ticks": 394292,
              "compute_passes": 1,
              "command_buffer_ns": 6583.34465697,
              "command_buffers": 2
            },
            {
              "compute_ns": 6916,
              "compute_span_ns": 6916,
              "calibration_cpu_ns": 418625,
              "calibration_gpu_ticks": 418625,
              "compute_passes": 1,
              "command_buffer_ns": 6999.99509379,
              "command_buffers": 2
            },
            {
              "compute_ns": 6792,
              "compute_span_ns": 6792,
              "calibration_cpu_ns": 439750,
              "calibration_gpu_ticks": 439750,
              "compute_passes": 1,
              "command_buffer_ns": 6916.69993103,
              "command_buffers": 2
            },
            {
              "compute_ns": 6875,
              "compute_span_ns": 6875,
              "calibration_cpu_ns": 439959,
              "calibration_gpu_ticks": 439959,
              "compute_passes": 1,
              "command_buffer_ns": 6916.64172336,
              "command_buffers": 2
            },
            {
              "compute_ns": 7333,
              "compute_span_ns": 7333,
              "calibration_cpu_ns": 831667,
              "calibration_gpu_ticks": 831667,
              "compute_passes": 1,
              "command_buffer_ns": 7416.64553061,
              "command_buffers": 2
            },
            {
              "compute_ns": 6500,
              "compute_span_ns": 6500,
              "calibration_cpu_ns": 402875,
              "calibration_gpu_ticks": 402875,
              "compute_passes": 1,
              "command_buffer_ns": 6499.99128655,
              "command_buffers": 2
            },
            {
              "compute_ns": 7666,
              "compute_span_ns": 7666,
              "calibration_cpu_ns": 723958,
              "calibration_gpu_ticks": 723958,
              "compute_passes": 1,
              "command_buffer_ns": 7708.26591179,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 548291.66038,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 293624.994811,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 679583.288729,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 618791.673332,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 291791.686323,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 290125.026368,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 244291.67388,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 649749.999866,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 258458.312601,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 6500.04949421,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 9083.30548555,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7750.00080466,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 17708.3420567,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7916.5911302,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6999.99509379,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 97500.0439212,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6874.96503815,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 69624.9771863,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              8.5670571934375,
              4.587890543921875,
              10.618488886390626,
              9.668619895812501,
              4.559245098796875,
              4.533203537,
              3.817057404375,
              10.15234374790625,
              4.038411134390625
            ],
            "command_buffer_throughput_us_p50": 4.587890543921875,
            "command_buffer_throughput_us_p90": 10.245572775603126,
            "command_buffer_latency_us": [
              6.500049494210001,
              9.08330548555,
              7.75000080466,
              17.708342056699998,
              7.9165911302000005,
              6.99999509379,
              97.5000439212,
              6.87496503815,
              69.6249771863
            ],
            "command_buffer_latency_us_p50": 7.9165911302000005,
            "command_buffer_latency_us_p90": 75.19999053328
          },
          "compute_throughput_us": [
            8.384125,
            4.55403125,
            4.140625,
            3.862640625,
            4.272140625,
            11.593109375,
            10.2421875,
            4.162109375,
            10.1920625
          ],
          "compute_throughput_us_p50": 4.55403125,
          "compute_throughput_us_p90": 10.512371875,
          "command_buffer_throughput_us": [
            8.3854174590625,
            4.55468762084375,
            4.141276804140626,
            3.864583959515625,
            4.274088496453125,
            11.59440125775,
            10.242838470734375,
            4.1640623749062495,
            10.192708941765625
          ],
          "command_buffer_throughput_us_p50": 4.55468762084375,
          "command_buffer_throughput_us_p90": 10.5131510281375,
          "compute_latency_us": [
            11.708,
            10.291,
            6.5,
            6.916,
            6.792,
            6.875,
            7.333,
            6.5,
            7.666
          ],
          "compute_latency_us_p50": 6.916,
          "compute_latency_us_p90": 10.5744,
          "command_buffer_latency_us": [
            11.7917079479,
            10.2500198409,
            6.58334465697,
            6.99999509379,
            6.91669993103,
            6.91664172336,
            7.416645530609999,
            6.49999128655,
            7.70826591179
          ],
          "command_buffer_latency_us_p50": 6.99999509379,
          "command_buffer_latency_us_p90": 10.5583574623,
          "counter_control_throughput_ratio": 0.9787978847025629,
          "counter_control_latency_ratio": 0.8736967486730947
        },
        "correctness": {
          "max_abs_error": 1.5564224420394623e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 5.84887393573,
        "throughput_us_p90": 6.3468988190080005,
        "latency_us_p50": 199.833,
        "latency_us_p90": 222.725
      },
      "torch": {
        "cold_call_ms": 70.993209,
        "warmup_ms": 200.707,
        "repetitions": 2832,
        "throughput_us": [
          9.487964689265537,
          10.844338276836158,
          9.165533898305085,
          9.23268290960452,
          9.429393008474577,
          9.923066737288135,
          9.373234463276836,
          9.081832627118644,
          9.13382768361582
        ],
        "latency_us": [
          352.333,
          373.542,
          253.5,
          247.417,
          255.166,
          256.334,
          239.042,
          232.333,
          1444.709
        ],
        "throughput_us_p50": 9.373234463276836,
        "throughput_us_p90": 10.10732104519774,
        "latency_us_p50": 255.166,
        "latency_us_p90": 587.7754000000002,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 493041.6471324861,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 631041.7084023356,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 293916.673399508,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 306333.3570025861,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 307000.00934302807,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 539875.0072345138,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 289250.0488087535,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 571999.9899156392,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 335249.9916218221,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 7958.326023072004,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 103166.61791875958,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 9250.01222640276,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 111875.00786036253,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 9291.688911616802,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8374.976459890604,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 9666.66266322136,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 7874.972652643919,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8374.976459890604,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              7.703775736445095,
              9.860026693786494,
              4.5924480218673125,
              4.786458703165408,
              4.796875145984814,
              8.435546988039277,
              4.519532012636773,
              8.937499842431862,
              5.238281119090971
            ],
            "command_buffer_throughput_us_p50": 5.238281119090971,
            "command_buffer_throughput_us_p90": 9.122005212702788,
            "command_buffer_latency_us": [
              7.958326023072004,
              103.16661791875958,
              9.25001222640276,
              111.87500786036253,
              9.291688911616802,
              8.374976459890604,
              9.66666266322136,
              7.874972652643919,
              8.374976459890604
            ],
            "command_buffer_latency_us_p50": 9.25001222640276,
            "command_buffer_latency_us_p90": 104.90829590708017
          },
          "throughput": [
            {
              "compute_ns": 330416.0,
              "compute_span_ns": 330416.0,
              "command_buffer_ns": 330374.9836049974,
              "calibration_cpu_ns": 925250.0,
              "calibration_gpu_ticks": 925250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 290417.0,
              "compute_span_ns": 290417.0,
              "command_buffer_ns": 290458.38164165616,
              "calibration_cpu_ns": 930792.0,
              "calibration_gpu_ticks": 930792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 289458.0,
              "compute_span_ns": 289458.0,
              "command_buffer_ns": 289458.3158195019,
              "calibration_cpu_ns": 994750.0,
              "calibration_gpu_ticks": 994750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 505792.0,
              "compute_span_ns": 505792.0,
              "command_buffer_ns": 505833.3626948297,
              "calibration_cpu_ns": 1602750.0,
              "calibration_gpu_ticks": 1602750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 458333.0,
              "compute_span_ns": 458333.0,
              "command_buffer_ns": 458333.3502523601,
              "calibration_cpu_ns": 2079166.0,
              "calibration_gpu_ticks": 2079166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 289167.0,
              "compute_span_ns": 289167.0,
              "command_buffer_ns": 289166.6372306645,
              "calibration_cpu_ns": 1005291.0,
              "calibration_gpu_ticks": 1005291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 507750.0,
              "compute_span_ns": 507750.0,
              "command_buffer_ns": 507750.02455338836,
              "calibration_cpu_ns": 2026458.0,
              "calibration_gpu_ticks": 2026458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 323417.0,
              "compute_span_ns": 323417.0,
              "command_buffer_ns": 323458.3418816328,
              "calibration_cpu_ns": 982583.0,
              "calibration_gpu_ticks": 982583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 526208.0,
              "compute_span_ns": 526208.0,
              "command_buffer_ns": 526166.6956357658,
              "calibration_cpu_ns": 2019208.0,
              "calibration_gpu_ticks": 2019208.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 8334.0,
              "compute_span_ns": 8334.0,
              "command_buffer_ns": 8374.976459890604,
              "calibration_cpu_ns": 466917.0,
              "calibration_gpu_ticks": 466917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8500.0,
              "compute_span_ns": 8500.0,
              "command_buffer_ns": 8500.006515532732,
              "calibration_cpu_ns": 410791.0,
              "calibration_gpu_ticks": 410791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8125.0,
              "compute_span_ns": 8125.0,
              "command_buffer_ns": 8124.9745562672615,
              "calibration_cpu_ns": 435709.0,
              "calibration_gpu_ticks": 435709.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8334.0,
              "compute_span_ns": 8334.0,
              "command_buffer_ns": 8375.034667551517,
              "calibration_cpu_ns": 425125.0,
              "calibration_gpu_ticks": 425125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 7875.0,
              "compute_span_ns": 7875.0,
              "command_buffer_ns": 7875.0308603048325,
              "calibration_cpu_ns": 446667.0,
              "calibration_gpu_ticks": 446667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8292.0,
              "compute_span_ns": 8292.0,
              "command_buffer_ns": 8291.681297123432,
              "calibration_cpu_ns": 469333.0,
              "calibration_gpu_ticks": 469333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10125.0,
              "compute_span_ns": 10125.0,
              "command_buffer_ns": 10124.989785254002,
              "calibration_cpu_ns": 783166.0,
              "calibration_gpu_ticks": 783166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8250.0,
              "compute_span_ns": 8250.0,
              "command_buffer_ns": 8250.00461190939,
              "calibration_cpu_ns": 435042.0,
              "calibration_gpu_ticks": 435042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 9666.0,
              "compute_span_ns": 9666.0,
              "command_buffer_ns": 9624.985978007317,
              "calibration_cpu_ns": 1883167.0,
              "calibration_gpu_ticks": 1883167.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            5.16275,
            4.537765625,
            4.52278125,
            7.903,
            7.161453125,
            4.518234375,
            7.93359375,
            5.053390625,
            8.222
          ],
          "compute_throughput_us_p50": 5.16275,
          "compute_throughput_us_p90": 7.991275,
          "command_buffer_throughput_us": [
            5.162109118828084,
            4.5384122131508775,
            4.522786184679717,
            7.903646292106714,
            7.161458597693127,
            4.518228706729133,
            7.933594133646693,
            5.054036591900513,
            8.22135461930884
          ],
          "command_buffer_throughput_us_p50": 5.162109118828084,
          "command_buffer_throughput_us_p90": 7.991146230779123,
          "compute_latency_us": [
            8.334,
            8.5,
            8.125,
            8.334,
            7.875,
            8.292,
            10.125,
            8.25,
            9.666
          ],
          "compute_latency_us_p50": 8.334,
          "compute_latency_us_p90": 9.7578,
          "command_buffer_latency_us": [
            8.374976459890604,
            8.500006515532732,
            8.124974556267262,
            8.375034667551517,
            7.8750308603048325,
            8.291681297123432,
            10.124989785254002,
            8.25000461190939,
            9.624985978007317
          ],
          "command_buffer_latency_us_p50": 8.374976459890604,
          "command_buffer_latency_us_p90": 9.724986739456654,
          "counter_control_throughput_ratio": 0.9848312192416996,
          "counter_control_latency_ratio": 0.9900542809683002
        },
        "download_ms": 0.518791,
        "allocation_upload_ms": 1.12425,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.807831588873654e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.6239974001125396,
      "device_compute_slowdown": 0.8820940874533921,
      "valid": true,
      "round": 0,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "rmsnorm_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.69325,
        "warmup_ms": 200.096042,
        "repetitions": 2547,
        "throughput_us": [
          11.132050647820966,
          10.131968983117392,
          9.574728307813112,
          9.681782489202984,
          9.808745583038869,
          9.873789556340792,
          9.772739301138593,
          9.834085983510011,
          10.145956026698077
        ],
        "latency_us": [
          362.458,
          220.125,
          201.792,
          212.167,
          221.167,
          227.458,
          219.958,
          212.375,
          224.375
        ],
        "throughput_us_p50": 9.834085983510011,
        "throughput_us_p90": 10.343174950922656,
        "latency_us_p50": 220.125,
        "latency_us_p90": 254.45800000000003,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 536833.3659134805,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 651124.9812319875,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 332875.0026412308,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 345041.6843406856,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 289999.9963119626,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 289916.6429415345,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 533291.6625775397,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 602125.0155754387,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 366375.02489611506,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 79249.96316432953,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 7833.295967429876,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 81541.65698215365,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8375.034667551517,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 73916.68623313308,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 11625.001206994057,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8250.00461190939,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8124.9745562672615,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8250.00461190939,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              8.388021342398133,
              10.173827831749804,
              5.201171916269232,
              5.391276317823213,
              4.531249942374416,
              4.529947545961477,
              8.332682227774058,
              9.40820336836623,
              5.724609764001798
            ],
            "command_buffer_throughput_us_p50": 5.724609764001798,
            "command_buffer_throughput_us_p90": 9.561328261042945,
            "command_buffer_latency_us": [
              79.24996316432953,
              7.833295967429876,
              81.54165698215365,
              8.375034667551517,
              73.91668623313308,
              11.625001206994057,
              8.25000461190939,
              8.124974556267262,
              8.25000461190939
            ],
            "command_buffer_latency_us_p50": 8.375034667551517,
            "command_buffer_latency_us_p90": 79.70830192789435
          },
          "throughput": [
            {
              "compute_ns": 337708.0,
              "compute_span_ns": 337708.0,
              "command_buffer_ns": 337666.65728762746,
              "calibration_cpu_ns": 937792.0,
              "calibration_gpu_ticks": 937792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 291000.0,
              "compute_span_ns": 291000.0,
              "command_buffer_ns": 291000.003926456,
              "calibration_cpu_ns": 910500.0,
              "calibration_gpu_ticks": 910500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 306833.0,
              "compute_span_ns": 306833.0,
              "command_buffer_ns": 306833.3608098328,
              "calibration_cpu_ns": 940083.0,
              "calibration_gpu_ticks": 940083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 507250.0,
              "compute_span_ns": 507250.0,
              "command_buffer_ns": 507249.96253848076,
              "calibration_cpu_ns": 1882750.0,
              "calibration_gpu_ticks": 1882750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 678375.0,
              "compute_span_ns": 678375.0,
              "command_buffer_ns": 678374.9558962882,
              "calibration_cpu_ns": 2006291.0,
              "calibration_gpu_ticks": 2006291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 303208.0,
              "compute_span_ns": 303208.0,
              "command_buffer_ns": 303208.3041034639,
              "calibration_cpu_ns": 1038250.0,
              "calibration_gpu_ticks": 1038250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 344459.0,
              "compute_span_ns": 344459.0,
              "command_buffer_ns": 344500.0038482249,
              "calibration_cpu_ns": 959958.0,
              "calibration_gpu_ticks": 959958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 306750.0,
              "compute_span_ns": 306750.0,
              "command_buffer_ns": 306750.0074394047,
              "calibration_cpu_ns": 908375.0,
              "calibration_gpu_ticks": 908375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 291292.0,
              "compute_span_ns": 291292.0,
              "command_buffer_ns": 291291.68251529336,
              "calibration_cpu_ns": 934458.0,
              "calibration_gpu_ticks": 934458.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 9250.0,
              "compute_span_ns": 9250.0,
              "command_buffer_ns": 9250.01222640276,
              "calibration_cpu_ns": 751291.0,
              "calibration_gpu_ticks": 751291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8625.0,
              "compute_span_ns": 8625.0,
              "command_buffer_ns": 8624.978363513947,
              "calibration_cpu_ns": 395042.0,
              "calibration_gpu_ticks": 395042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 9417.0,
              "compute_span_ns": 9417.0,
              "command_buffer_ns": 9416.660759598017,
              "calibration_cpu_ns": 1251792.0,
              "calibration_gpu_ticks": 1251792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8750.0,
              "compute_span_ns": 8750.0,
              "command_buffer_ns": 8750.008419156075,
              "calibration_cpu_ns": 505000.0,
              "calibration_gpu_ticks": 505000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 9417.0,
              "compute_span_ns": 9417.0,
              "command_buffer_ns": 9458.33744481206,
              "calibration_cpu_ns": 804917.0,
              "calibration_gpu_ticks": 804917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 7958.0,
              "compute_span_ns": 7958.0,
              "command_buffer_ns": 7916.649337857962,
              "calibration_cpu_ns": 391917.0,
              "calibration_gpu_ticks": 391917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 81708.0,
              "compute_span_ns": 81708.0,
              "command_buffer_ns": 81708.36372300982,
              "calibration_cpu_ns": 1231625.0,
              "calibration_gpu_ticks": 1231625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 9083.0,
              "compute_span_ns": 9083.0,
              "command_buffer_ns": 9041.628800332546,
              "calibration_cpu_ns": 849375.0,
              "calibration_gpu_ticks": 849375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8333.0,
              "compute_span_ns": 8333.0,
              "command_buffer_ns": 8291.681297123432,
              "calibration_cpu_ns": 402166.0,
              "calibration_gpu_ticks": 402166.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            5.2766875,
            4.546875,
            4.794265625,
            7.92578125,
            10.599609375,
            4.737625,
            5.382171875,
            4.79296875,
            4.5514375
          ],
          "compute_throughput_us_p50": 4.794265625,
          "compute_throughput_us_p90": 8.460546875,
          "command_buffer_throughput_us": [
            5.276041520119179,
            4.546875061350875,
            4.794271262653638,
            7.925780664663762,
            10.599608685879502,
            4.737629751616623,
            5.382812560128514,
            4.792968866240699,
            4.551432539301459
          ],
          "command_buffer_throughput_us_p50": 4.794271262653638,
          "command_buffer_throughput_us_p90": 8.46054626890691,
          "compute_latency_us": [
            9.25,
            8.625,
            9.417,
            8.75,
            9.417,
            7.958,
            81.708,
            9.083,
            8.333
          ],
          "compute_latency_us_p50": 9.083,
          "compute_latency_us_p90": 23.875200000000014,
          "command_buffer_latency_us": [
            9.25001222640276,
            8.624978363513947,
            9.416660759598017,
            8.750008419156075,
            9.45833744481206,
            7.916649337857962,
            81.70836372300982,
            9.041628800332546,
            8.291681297123432
          ],
          "command_buffer_latency_us_p50": 9.041628800332546,
          "command_buffer_latency_us_p90": 23.908342700451627,
          "counter_control_throughput_ratio": 0.7950642448892049,
          "counter_control_latency_ratio": 1.0050517165958768
        },
        "download_ms": 0.318458,
        "allocation_upload_ms": 3.046125,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.807831588873654e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-l514mu8x/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "e81310ef6c7b3435af776cabf332187d02f0ddc801b9234c32af8a56116f248b",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.158208,
        "capture_ms": 0.065625,
        "compile_ms": 64.572083,
        "allocation_upload_ms": 1.945833,
        "cold_call_ms": 2.045084,
        "warmup_ms": 200.131125,
        "download_ms": 0.366959,
        "repetitions": 3990,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          6.32488496241,
          6.29597969925,
          6.68083759398,
          6.1652358396,
          7.10079373434,
          6.36380526316,
          6.30533634085,
          6.31369047619,
          6.33959899749
        ],
        "latency_us": [
          252.25,
          244.042,
          224.667,
          214.375,
          196.292,
          229.792,
          287.708,
          230.708,
          233.167
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 19,
            "reduction_scalar_rounds": 24,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 12296,
              "global_write_bytes": 6148,
              "private_read_bytes": 18444,
              "private_write_bytes": 6148
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 64,
              "global_write_bytes": 32,
              "private_read_bytes": 96,
              "private_write_bytes": 32
            },
            "striped_storage_scalars_per_worker": 8,
            "reduction_operations": 1,
            "reduction_elements": 1537,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 48,
            "normalized_kernel_cost": 48,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 3074,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 265875,
              "compute_span_ns": 265875,
              "calibration_cpu_ns": 832542,
              "calibration_gpu_ticks": 832542,
              "compute_passes": 1,
              "command_buffer_ns": 265916.693024,
              "command_buffers": 2
            },
            {
              "compute_ns": 292667,
              "compute_span_ns": 292667,
              "calibration_cpu_ns": 776417,
              "calibration_gpu_ticks": 776417,
              "compute_passes": 1,
              "command_buffer_ns": 292791.635729,
              "command_buffers": 2
            },
            {
              "compute_ns": 600167,
              "compute_span_ns": 600167,
              "calibration_cpu_ns": 1833208,
              "calibration_gpu_ticks": 1833208,
              "compute_passes": 1,
              "command_buffer_ns": 600333.325565,
              "command_buffers": 2
            },
            {
              "compute_ns": 281625,
              "compute_span_ns": 281625,
              "calibration_cpu_ns": 759041,
              "calibration_gpu_ticks": 759041,
              "compute_passes": 1,
              "command_buffer_ns": 281708.373223,
              "command_buffers": 2
            },
            {
              "compute_ns": 265083,
              "compute_span_ns": 265083,
              "calibration_cpu_ns": 737875,
              "calibration_gpu_ticks": 737875,
              "compute_passes": 1,
              "command_buffer_ns": 265125.010628,
              "command_buffers": 2
            },
            {
              "compute_ns": 265667,
              "compute_span_ns": 265667,
              "calibration_cpu_ns": 731791,
              "calibration_gpu_ticks": 731791,
              "compute_passes": 1,
              "command_buffer_ns": 265750.044491,
              "command_buffers": 2
            },
            {
              "compute_ns": 292833,
              "compute_span_ns": 292833,
              "calibration_cpu_ns": 749584,
              "calibration_gpu_ticks": 749584,
              "compute_passes": 1,
              "command_buffer_ns": 292874.9891,
              "command_buffers": 2
            },
            {
              "compute_ns": 567250,
              "compute_span_ns": 567250,
              "calibration_cpu_ns": 2144000,
              "calibration_gpu_ticks": 2144000,
              "compute_passes": 1,
              "command_buffer_ns": 567333.365325,
              "command_buffers": 2
            },
            {
              "compute_ns": 292125,
              "compute_span_ns": 292125,
              "calibration_cpu_ns": 742958,
              "calibration_gpu_ticks": 742958,
              "compute_passes": 1,
              "command_buffer_ns": 292166.660074,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 11042,
              "compute_span_ns": 11042,
              "calibration_cpu_ns": 907417,
              "calibration_gpu_ticks": 907417,
              "compute_passes": 1,
              "command_buffer_ns": 11083.3207145,
              "command_buffers": 2
            },
            {
              "compute_ns": 7750,
              "compute_span_ns": 7750,
              "calibration_cpu_ns": 821750,
              "calibration_gpu_ticks": 821750,
              "compute_passes": 1,
              "command_buffer_ns": 7791.61928222,
              "command_buffers": 2
            },
            {
              "compute_ns": 7167,
              "compute_span_ns": 7167,
              "calibration_cpu_ns": 484167,
              "calibration_gpu_ticks": 484167,
              "compute_passes": 1,
              "command_buffer_ns": 7249.99699742,
              "command_buffers": 2
            },
            {
              "compute_ns": 92250,
              "compute_span_ns": 92250,
              "calibration_cpu_ns": 987333,
              "calibration_gpu_ticks": 987333,
              "compute_passes": 1,
              "command_buffer_ns": 92375.0340007,
              "command_buffers": 2
            },
            {
              "compute_ns": 6958,
              "compute_span_ns": 6958,
              "calibration_cpu_ns": 405125,
              "calibration_gpu_ticks": 405125,
              "compute_passes": 1,
              "command_buffer_ns": 7083.34846422,
              "command_buffers": 2
            },
            {
              "compute_ns": 124542,
              "compute_span_ns": 124542,
              "calibration_cpu_ns": 1544583,
              "calibration_gpu_ticks": 1544583,
              "compute_passes": 1,
              "command_buffer_ns": 124666.723423,
              "command_buffers": 2
            },
            {
              "compute_ns": 77000,
              "compute_span_ns": 77000,
              "calibration_cpu_ns": 1265250,
              "calibration_gpu_ticks": 1265250,
              "compute_passes": 1,
              "command_buffer_ns": 77000.0042394,
              "command_buffers": 2
            },
            {
              "compute_ns": 7250,
              "compute_span_ns": 7250,
              "calibration_cpu_ns": 794084,
              "calibration_gpu_ticks": 794084,
              "compute_passes": 1,
              "command_buffer_ns": 7541.67558625,
              "command_buffers": 2
            },
            {
              "compute_ns": 42375,
              "compute_span_ns": 42375,
              "calibration_cpu_ns": 1200625,
              "calibration_gpu_ticks": 1200625,
              "compute_passes": 1,
              "command_buffer_ns": 42499.97437,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 701958.255377,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 774083.309807,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 280416.687019,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 281541.666482,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 611249.997746,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 711500.062607,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 288458.308205,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 267166.702542,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 691374.938469,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 7124.96694177,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6624.96313453,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7249.99699742,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7708.32411945,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6666.6980274,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7166.64362699,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6541.66797176,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6458.31460133,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7083.34846422,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              10.968097740265625,
              12.095051715734375,
              4.381510734671875,
              4.39908853878125,
              9.55078121478125,
              11.117188478234375,
              4.507161065703125,
              4.17447972721875,
              10.802733413578126
            ],
            "command_buffer_throughput_us_p50": 9.55078121478125,
            "command_buffer_throughput_us_p90": 11.312761125734376,
            "command_buffer_latency_us": [
              7.1249669417699995,
              6.624963134530001,
              7.24999699742,
              7.708324119449999,
              6.6666980274,
              7.16664362699,
              6.54166797176,
              6.458314601330001,
              7.08334846422
            ],
            "command_buffer_latency_us_p50": 7.08334846422,
            "command_buffer_latency_us_p90": 7.341662421826
          },
          "compute_throughput_us": [
            4.154296875,
            4.572921875,
            9.377609375,
            4.400390625,
            4.141921875,
            4.151046875,
            4.575515625,
            8.86328125,
            4.564453125
          ],
          "compute_throughput_us_p50": 4.564453125,
          "compute_throughput_us_p90": 8.966146875,
          "command_buffer_throughput_us": [
            4.1549483285,
            4.574869308265624,
            9.380208211953125,
            4.4016933316093745,
            4.142578291062501,
            4.152344445171875,
            4.5761717046875,
            8.864583833203126,
            4.56510406365625
          ],
          "command_buffer_throughput_us_p50": 4.56510406365625,
          "command_buffer_throughput_us_p90": 8.967708708953126,
          "compute_latency_us": [
            11.042,
            7.75,
            7.167,
            92.25,
            6.958,
            124.542,
            77.0,
            7.25,
            42.375
          ],
          "compute_latency_us_p50": 11.042,
          "compute_latency_us_p90": 98.70840000000001,
          "command_buffer_latency_us": [
            11.0833207145,
            7.79161928222,
            7.24999699742,
            92.37503400070001,
            7.08334846422,
            124.66672342300001,
            77.0000042394,
            7.54167558625,
            42.499974370000004
          ],
          "command_buffer_latency_us_p50": 11.0833207145,
          "command_buffer_latency_us_p90": 98.83337188516002,
          "counter_control_throughput_ratio": 0.4337423502751007,
          "counter_control_latency_ratio": 1.5555610019078427
        },
        "correctness": {
          "max_abs_error": 1.5564224420394623e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 6.32488496241,
        "throughput_us_p90": 6.764828822051999,
        "latency_us_p50": 230.708,
        "latency_us_p90": 259.3416
      },
      "slowdown": 0.6431594123760653,
      "device_compute_slowdown": 0.9520651298915044,
      "valid": true,
      "round": 0,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "rmsnorm_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-5rlsc4dr/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "a8c235e7c8be57d97298e1f9f70525f6c2c7f4ae4cee9647b15503a46aab5ade",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.7505,
        "capture_ms": 0.079917,
        "compile_ms": 57.31525,
        "allocation_upload_ms": 2.289708,
        "cold_call_ms": 6.135417,
        "warmup_ms": 200.23625,
        "download_ms": 0.92775,
        "repetitions": 1577,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          23.3050095117,
          21.5134483196,
          21.5564629042,
          21.6622276474,
          22.4908319594,
          23.070439442,
          21.9010779962,
          21.7224426126,
          22.1561769182
        ],
        "latency_us": [
          300.417,
          252.5,
          263.833,
          256.291,
          245.458,
          445.916,
          2048.084,
          1005.834,
          541.416
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 128,
            "reduction_scalar_rounds": 36,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 24576,
              "global_write_bytes": 12288,
              "private_read_bytes": 36864,
              "private_write_bytes": 12288
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 48,
              "private_read_bytes": 144,
              "private_write_bytes": 48
            },
            "striped_storage_scalars_per_worker": 12,
            "reduction_operations": 1,
            "reduction_elements": 3072,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 60,
            "normalized_kernel_cost": 240,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 6144,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 985125,
              "compute_span_ns": 985125,
              "calibration_cpu_ns": 1448458,
              "calibration_gpu_ticks": 1448458,
              "compute_passes": 1,
              "command_buffer_ns": 985166.698229,
              "command_buffers": 2
            },
            {
              "compute_ns": 1313333,
              "compute_span_ns": 1313333,
              "calibration_cpu_ns": 2564875,
              "calibration_gpu_ticks": 2564875,
              "compute_passes": 1,
              "command_buffer_ns": 1313333.34139,
              "command_buffers": 2
            },
            {
              "compute_ns": 960833,
              "compute_span_ns": 960833,
              "calibration_cpu_ns": 1617708,
              "calibration_gpu_ticks": 1617708,
              "compute_passes": 1,
              "command_buffer_ns": 960875.011515,
              "command_buffers": 2
            },
            {
              "compute_ns": 1249458,
              "compute_span_ns": 1249458,
              "calibration_cpu_ns": 2470208,
              "calibration_gpu_ticks": 2470208,
              "compute_passes": 1,
              "command_buffer_ns": 1249541.64494,
              "command_buffers": 2
            },
            {
              "compute_ns": 957209,
              "compute_span_ns": 957209,
              "calibration_cpu_ns": 1464958,
              "calibration_gpu_ticks": 1464958,
              "compute_passes": 1,
              "command_buffer_ns": 957374.984864,
              "command_buffers": 2
            },
            {
              "compute_ns": 1462833,
              "compute_span_ns": 1462833,
              "calibration_cpu_ns": 2729875,
              "calibration_gpu_ticks": 2729875,
              "compute_passes": 1,
              "command_buffer_ns": 1462833.3156,
              "command_buffers": 2
            },
            {
              "compute_ns": 962041,
              "compute_span_ns": 962041,
              "calibration_cpu_ns": 1443042,
              "calibration_gpu_ticks": 1443042,
              "compute_passes": 1,
              "command_buffer_ns": 962083.344348,
              "command_buffers": 2
            },
            {
              "compute_ns": 1239833,
              "compute_span_ns": 1239833,
              "calibration_cpu_ns": 2553916,
              "calibration_gpu_ticks": 2553916,
              "compute_passes": 1,
              "command_buffer_ns": 1239791.68711,
              "command_buffers": 2
            },
            {
              "compute_ns": 970875,
              "compute_span_ns": 970875,
              "calibration_cpu_ns": 1454792,
              "calibration_gpu_ticks": 1454792,
              "compute_passes": 1,
              "command_buffer_ns": 970916.64793,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 20709,
              "compute_span_ns": 20709,
              "calibration_cpu_ns": 1112875,
              "calibration_gpu_ticks": 1112875,
              "compute_passes": 1,
              "command_buffer_ns": 20833.2785405,
              "command_buffers": 2
            },
            {
              "compute_ns": 19125,
              "compute_span_ns": 19125,
              "calibration_cpu_ns": 429209,
              "calibration_gpu_ticks": 429209,
              "compute_passes": 1,
              "command_buffer_ns": 19208.2952708,
              "command_buffers": 2
            },
            {
              "compute_ns": 22917,
              "compute_span_ns": 22917,
              "calibration_cpu_ns": 768292,
              "calibration_gpu_ticks": 768292,
              "compute_passes": 1,
              "command_buffer_ns": 23041.6771956,
              "command_buffers": 2
            },
            {
              "compute_ns": 16708,
              "compute_span_ns": 16708,
              "calibration_cpu_ns": 529583,
              "calibration_gpu_ticks": 529583,
              "compute_passes": 1,
              "command_buffer_ns": 16749.9529198,
              "command_buffers": 2
            },
            {
              "compute_ns": 21375,
              "compute_span_ns": 21375,
              "calibration_cpu_ns": 883000,
              "calibration_gpu_ticks": 883000,
              "compute_passes": 1,
              "command_buffer_ns": 21416.6357182,
              "command_buffers": 2
            },
            {
              "compute_ns": 17084,
              "compute_span_ns": 17084,
              "calibration_cpu_ns": 470125,
              "calibration_gpu_ticks": 470125,
              "compute_passes": 1,
              "command_buffer_ns": 17208.3382495,
              "command_buffers": 2
            },
            {
              "compute_ns": 20875,
              "compute_span_ns": 20875,
              "calibration_cpu_ns": 819959,
              "calibration_gpu_ticks": 819959,
              "compute_passes": 1,
              "command_buffer_ns": 21000.043489,
              "command_buffers": 2
            },
            {
              "compute_ns": 17125,
              "compute_span_ns": 17125,
              "calibration_cpu_ns": 452916,
              "calibration_gpu_ticks": 452916,
              "compute_passes": 1,
              "command_buffer_ns": 17250.0149347,
              "command_buffers": 2
            },
            {
              "compute_ns": 21334,
              "compute_span_ns": 21334,
              "calibration_cpu_ns": 840625,
              "calibration_gpu_ticks": 840625,
              "compute_passes": 1,
              "command_buffer_ns": 21416.6939259,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1395083.32359,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 953541.661147,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1486458.34997,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 962375.022937,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1343333.27878,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 962708.261795,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1448666.61867,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 967291.649431,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1497208.31541,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 50083.2684338,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19041.5885299,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63208.3974779,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19625.0039153,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 84708.2701512,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19333.3253264,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 42916.6830145,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19416.6786969,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 25875.0515059,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              21.79817693109375,
              14.899088455421875,
              23.225911718281253,
              15.037109733390626,
              20.9895824809375,
              15.042316590546877,
              22.63541591671875,
              15.113932022359375,
              23.39387992828125
            ],
            "command_buffer_throughput_us_p50": 20.9895824809375,
            "command_buffer_throughput_us_p90": 23.259505360281253,
            "command_buffer_latency_us": [
              50.0832684338,
              19.0415885299,
              63.2083974779,
              19.6250039153,
              84.70827015120001,
              19.3333253264,
              42.9166830145,
              19.4166786969,
              25.8750515059
            ],
            "command_buffer_latency_us_p50": 25.8750515059,
            "command_buffer_latency_us_p90": 67.50837201256
          },
          "compute_throughput_us": [
            15.392578125,
            20.520828125,
            15.013015625,
            19.52278125,
            14.956390625,
            22.856765625,
            15.031890625,
            19.372390625,
            15.169921875
          ],
          "compute_throughput_us_p50": 15.392578125,
          "compute_throughput_us_p90": 20.988015625000003,
          "command_buffer_throughput_us": [
            15.393229659828124,
            20.52083345921875,
            15.013672054921875,
            19.5240882021875,
            14.9589841385,
            22.85677055625,
            15.0325522554375,
            19.37174511109375,
            15.17057262390625
          ],
          "command_buffer_throughput_us_p50": 15.393229659828124,
          "command_buffer_throughput_us_p90": 20.988020878625,
          "compute_latency_us": [
            20.709,
            19.125,
            22.917,
            16.708,
            21.375,
            17.084,
            20.875,
            17.125,
            21.334
          ],
          "compute_latency_us_p50": 20.709,
          "compute_latency_us_p90": 21.6834,
          "command_buffer_latency_us": [
            20.8332785405,
            19.2082952708,
            23.041677195600002,
            16.7499529198,
            21.4166357182,
            17.2083382495,
            21.000043489,
            17.2500149347,
            21.4166939259
          ],
          "command_buffer_latency_us_p50": 20.8332785405,
          "command_buffer_latency_us_p90": 21.74169057984,
          "counter_control_throughput_ratio": 0.7126861218933525,
          "counter_control_latency_ratio": 0.8276966683918132
        },
        "correctness": {
          "max_abs_error": 1.6175184480715643e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 21.9010779962,
        "throughput_us_p90": 23.117353455940002,
        "latency_us_p50": 300.417,
        "latency_us_p90": 1214.284
      },
      "torch": {
        "cold_call_ms": 6.579667,
        "warmup_ms": 200.59575,
        "repetitions": 1109,
        "throughput_us": [
          25.963931469792605,
          26.937743913435526,
          26.67684851217313,
          26.27821641118124,
          27.00364472497746,
          26.58427321911632,
          28.52678809738503,
          26.493537421100093,
          27.777427412082957
        ],
        "latency_us": [
          365.209,
          284.125,
          286.95799999999997,
          263.958,
          260.541,
          261.084,
          279.792,
          3117.333,
          474.375
        ],
        "throughput_us_p50": 26.67684851217313,
        "throughput_us_p90": 27.92729954914337,
        "latency_us_p50": 284.125,
        "latency_us_p90": 1002.9666000000005,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1081124.9958351254,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1147250.0045783818,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1034749.9628551304,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1459416.642319411,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1061999.995727092,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1320499.9850131571,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1035333.3200328052,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1321083.342190832,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1038916.6418462992,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 18791.64483398199,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21291.663870215416,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 19125.00010803342,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 25124.987587332726,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 26124.995201826096,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 85624.9826028943,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21249.987185001373,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 28499.984182417393,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 25500.019546598196,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              16.892578059923835,
              17.925781321537215,
              16.167968169611413,
              22.803385036240797,
              16.593749933235813,
              20.63281226583058,
              16.17708312551258,
              20.64192722173175,
              16.233072528848425
            ],
            "command_buffer_throughput_us_p50": 16.892578059923835,
            "command_buffer_throughput_us_p90": 21.07421878463356,
            "command_buffer_latency_us": [
              18.79164483398199,
              21.291663870215416,
              19.12500010803342,
              25.124987587332726,
              26.124995201826096,
              85.6249826028943,
              21.249987185001373,
              28.499984182417393,
              25.500019546598196
            ],
            "command_buffer_latency_us_p50": 25.124987587332726,
            "command_buffer_latency_us_p90": 39.92498386651279
          },
          "throughput": [
            {
              "compute_ns": 1236541.0,
              "compute_span_ns": 1236541.0,
              "command_buffer_ns": 1236499.985679984,
              "calibration_cpu_ns": 2723458.0,
              "calibration_gpu_ticks": 2723458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1035541.0,
              "compute_span_ns": 1035541.0,
              "command_buffer_ns": 1035500.0267736614,
              "calibration_cpu_ns": 1847625.0,
              "calibration_gpu_ticks": 1847625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1185375.0,
              "compute_span_ns": 1185375.0,
              "command_buffer_ns": 1185374.9747388065,
              "calibration_cpu_ns": 2791250.0,
              "calibration_gpu_ticks": 2791250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1032792.0,
              "compute_span_ns": 1032792.0,
              "command_buffer_ns": 1032833.3592042327,
              "calibration_cpu_ns": 1761334.0,
              "calibration_gpu_ticks": 1761334.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1355084.0,
              "compute_span_ns": 1355084.0,
              "command_buffer_ns": 1355125.0449381769,
              "calibration_cpu_ns": 2642083.0,
              "calibration_gpu_ticks": 2642083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1031959.0,
              "compute_span_ns": 1031959.0,
              "command_buffer_ns": 1032000.0001229346,
              "calibration_cpu_ns": 1774458.0,
              "calibration_gpu_ticks": 1774458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1327542.0,
              "compute_span_ns": 1327542.0,
              "command_buffer_ns": 1327541.7149998248,
              "calibration_cpu_ns": 2642333.0,
              "calibration_gpu_ticks": 2642333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1078500.0,
              "compute_span_ns": 1078500.0,
              "command_buffer_ns": 1078500.0049509108,
              "calibration_cpu_ns": 1685500.0,
              "calibration_gpu_ticks": 1685500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1300833.0,
              "compute_span_ns": 1300833.0,
              "command_buffer_ns": 1300791.6859351099,
              "calibration_cpu_ns": 2741459.0,
              "calibration_gpu_ticks": 2741459.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 18666.0,
              "compute_span_ns": 18666.0,
              "command_buffer_ns": 18625.054508447647,
              "calibration_cpu_ns": 446417.0,
              "calibration_gpu_ticks": 446417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 95583.0,
              "compute_span_ns": 95583.0,
              "command_buffer_ns": 95541.70537739992,
              "calibration_cpu_ns": 1204834.0,
              "calibration_gpu_ticks": 1204834.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 20958.0,
              "compute_span_ns": 20958.0,
              "command_buffer_ns": 20958.308596163988,
              "calibration_cpu_ns": 500625.0,
              "calibration_gpu_ticks": 500625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 20875.0,
              "compute_span_ns": 20875.0,
              "command_buffer_ns": 20875.013433396816,
              "calibration_cpu_ns": 415042.0,
              "calibration_gpu_ticks": 415042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 18708.0,
              "compute_span_ns": 18708.0,
              "command_buffer_ns": 18666.672986000776,
              "calibration_cpu_ns": 444584.0,
              "calibration_gpu_ticks": 444584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21584.0,
              "compute_span_ns": 21584.0,
              "command_buffer_ns": 21625.019144266844,
              "calibration_cpu_ns": 566042.0,
              "calibration_gpu_ticks": 566042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 18667.0,
              "compute_span_ns": 18667.0,
              "command_buffer_ns": 18666.672986000776,
              "calibration_cpu_ns": 439417.0,
              "calibration_gpu_ticks": 439417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21167.0,
              "compute_span_ns": 21167.0,
              "command_buffer_ns": 21166.633814573288,
              "calibration_cpu_ns": 527625.0,
              "calibration_gpu_ticks": 527625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 18666.0,
              "compute_span_ns": 18666.0,
              "command_buffer_ns": 18624.996300786734,
              "calibration_cpu_ns": 446833.0,
              "calibration_gpu_ticks": 446833.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            19.320953125,
            16.180328125,
            18.521484375,
            16.137375,
            21.1731875,
            16.124359375,
            20.74284375,
            16.8515625,
            20.325515625
          ],
          "compute_throughput_us_p50": 18.521484375,
          "compute_throughput_us_p90": 20.828912499999998,
          "command_buffer_throughput_us": [
            19.32031227624975,
            16.17968791833846,
            18.52148398029385,
            16.138021237566136,
            21.173828827159014,
            16.125000001920853,
            20.742839296872262,
            16.85156257735798,
            20.32487009273609
          ],
          "command_buffer_throughput_us_p50": 18.52148398029385,
          "command_buffer_throughput_us_p90": 20.829037202929612,
          "compute_latency_us": [
            18.666,
            95.583,
            20.958,
            20.875,
            18.708,
            21.584,
            18.667,
            21.167,
            18.666
          ],
          "compute_latency_us_p50": 20.875,
          "compute_latency_us_p90": 36.38380000000001,
          "command_buffer_latency_us": [
            18.625054508447647,
            95.54170537739992,
            20.958308596163988,
            20.875013433396816,
            18.666672986000776,
            21.625019144266844,
            18.666672986000776,
            21.166633814573288,
            18.624996300786734
          ],
          "command_buffer_latency_us_p50": 20.875013433396816,
          "command_buffer_latency_us_p90": 36.40835639089347,
          "counter_control_throughput_ratio": 1.1437160276965364,
          "counter_control_latency_ratio": 0.8308467162754493
        },
        "download_ms": 0.352084,
        "allocation_upload_ms": 1.647417,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.8163651027691685e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.8209769600860515,
      "device_compute_slowdown": 0.8310661183169883,
      "valid": true,
      "round": 0,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "rmsnorm_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.368875,
        "warmup_ms": 200.097834,
        "repetitions": 1203,
        "throughput_us": [
          26.331463009143807,
          28.022789692435577,
          27.031275976724853,
          27.417463840399005,
          27.39314962593516,
          27.971356608478796,
          27.6880365752286,
          28.035224438902745,
          27.20109393183707
        ],
        "latency_us": [
          396.292,
          285.54200000000003,
          278.292,
          292.16700000000003,
          288.375,
          268.25,
          277.91700000000003,
          260.375,
          252.583
        ],
        "throughput_us_p50": 27.417463840399005,
        "throughput_us_p90": 28.02527664172901,
        "latency_us_p50": 278.292,
        "latency_us_p90": 312.992,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1362458.33709836,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1035541.6452512145,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1302124.9906159937,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1043249.9693706632,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1383124.9671056867,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1042083.3132229745,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1316500.0127628446,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1032750.0058338046,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1311249.9727867544,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 52999.99611452222,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21083.338651806116,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 155916.67033731937,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 20791.66006296873,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 99375.02909451723,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 20875.013433396816,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 20749.98337775469,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 18791.64483398199,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21125.01533702016,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              21.288411517161876,
              16.180338207050227,
              20.345702978374902,
              16.300780771416612,
              21.611327611026354,
              16.282551769108977,
              20.570312699419446,
              16.136718841153197,
              20.488280824793037
            ],
            "command_buffer_throughput_us_p50": 20.345702978374902,
            "command_buffer_throughput_us_p90": 21.35299473593477,
            "command_buffer_latency_us": [
              52.99999611452222,
              21.083338651806116,
              155.91667033731937,
              20.79166006296873,
              99.37502909451723,
              20.875013433396816,
              20.74998337775469,
              18.79164483398199,
              21.12501533702016
            ],
            "command_buffer_latency_us_p50": 21.083338651806116,
            "command_buffer_latency_us_p90": 110.68335734307767
          },
          "throughput": [
            {
              "compute_ns": 1059709.0,
              "compute_span_ns": 1059709.0,
              "command_buffer_ns": 1059750.0368021429,
              "calibration_cpu_ns": 1764250.0,
              "calibration_gpu_ticks": 1764250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1282875.0,
              "compute_span_ns": 1282875.0,
              "command_buffer_ns": 1282874.9604523182,
              "calibration_cpu_ns": 2678709.0,
              "calibration_gpu_ticks": 2678709.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1034667.0,
              "compute_span_ns": 1034667.0,
              "command_buffer_ns": 1034666.6676923633,
              "calibration_cpu_ns": 1749958.0,
              "calibration_gpu_ticks": 1749958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1410875.0,
              "compute_span_ns": 1410875.0,
              "command_buffer_ns": 1410875.003784895,
              "calibration_cpu_ns": 2765375.0,
              "calibration_gpu_ticks": 2765375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1033208.0,
              "compute_span_ns": 1033208.0,
              "command_buffer_ns": 1033166.6562706232,
              "calibration_cpu_ns": 1765083.0,
              "calibration_gpu_ticks": 1765083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1286833.0,
              "compute_span_ns": 1286833.0,
              "command_buffer_ns": 1286791.6957475245,
              "calibration_cpu_ns": 2644166.0,
              "calibration_gpu_ticks": 2644166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1044709.0,
              "compute_span_ns": 1044709.0,
              "command_buffer_ns": 1044749.9807924032,
              "calibration_cpu_ns": 1673750.0,
              "calibration_gpu_ticks": 1673750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1262833.0,
              "compute_span_ns": 1262833.0,
              "command_buffer_ns": 1262833.3643078804,
              "calibration_cpu_ns": 2790833.0,
              "calibration_gpu_ticks": 2790833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1044042.0,
              "compute_span_ns": 1044042.0,
              "command_buffer_ns": 1044083.3284519613,
              "calibration_cpu_ns": 1814792.0,
              "calibration_gpu_ticks": 1814792.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 22583.0,
              "compute_span_ns": 22583.0,
              "command_buffer_ns": 22583.35007354617,
              "calibration_cpu_ns": 965959.0,
              "calibration_gpu_ticks": 965959.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 18875.0,
              "compute_span_ns": 18875.0,
              "command_buffer_ns": 18874.998204410076,
              "calibration_cpu_ns": 439500.0,
              "calibration_gpu_ticks": 439500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 19708.0,
              "compute_span_ns": 19708.0,
              "command_buffer_ns": 19666.622392833233,
              "calibration_cpu_ns": 761125.0,
              "calibration_gpu_ticks": 761125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 18959.0,
              "compute_span_ns": 18959.0,
              "command_buffer_ns": 19000.028260052204,
              "calibration_cpu_ns": 638250.0,
              "calibration_gpu_ticks": 638250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 20333.0,
              "compute_span_ns": 20333.0,
              "command_buffer_ns": 20333.33294093609,
              "calibration_cpu_ns": 868834.0,
              "calibration_gpu_ticks": 868834.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 22417.0,
              "compute_span_ns": 22417.0,
              "command_buffer_ns": 22458.378225564957,
              "calibration_cpu_ns": 449750.0,
              "calibration_gpu_ticks": 449750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 93959.0,
              "compute_span_ns": 93959.0,
              "command_buffer_ns": 94000.01727044582,
              "calibration_cpu_ns": 1178417.0,
              "calibration_gpu_ticks": 1178417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21000.0,
              "compute_span_ns": 21000.0,
              "command_buffer_ns": 21000.043489038944,
              "calibration_cpu_ns": 843250.0,
              "calibration_gpu_ticks": 843250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21042.0,
              "compute_span_ns": 21042.0,
              "command_buffer_ns": 21083.338651806116,
              "calibration_cpu_ns": 457875.0,
              "calibration_gpu_ticks": 457875.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            16.557953125,
            20.044921875,
            16.166671875,
            22.044921875,
            16.143875,
            20.106765625,
            16.323578125,
            19.731765625,
            16.31315625
          ],
          "compute_throughput_us_p50": 16.557953125,
          "compute_throughput_us_p90": 20.494396875,
          "command_buffer_throughput_us": [
            16.558594325033482,
            20.04492125706747,
            16.166666682693176,
            22.044921934138983,
            16.143229004228488,
            20.10612024605507,
            16.3242184498813,
            19.73177131731063,
            16.313802007061895
          ],
          "command_buffer_throughput_us_p50": 16.558594325033482,
          "command_buffer_throughput_us_p90": 20.493880583671853,
          "compute_latency_us": [
            22.583,
            18.875,
            19.708,
            18.959,
            20.333,
            22.417,
            93.959,
            21.0,
            21.042
          ],
          "compute_latency_us_p50": 21.0,
          "compute_latency_us_p90": 36.85820000000001,
          "command_buffer_latency_us": [
            22.58335007354617,
            18.874998204410076,
            19.666622392833233,
            19.000028260052204,
            20.33333294093609,
            22.458378225564957,
            94.00001727044582,
            21.000043489038944,
            21.083338651806116
          ],
          "command_buffer_latency_us_p50": 21.000043489038944,
          "command_buffer_latency_us_p90": 36.866683512926116,
          "counter_control_throughput_ratio": 0.7962504100061157,
          "counter_control_latency_ratio": 0.9138293047553457
        },
        "download_ms": 0.364417,
        "allocation_upload_ms": 0.795334,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.8163651027691685e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-a62lcxbm/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "b5a32df1046f38df4f8e9cbba5277df92f4beb4d4f9631ac4564ef5cda134264",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.657,
        "capture_ms": 0.065416,
        "compile_ms": 56.442125,
        "allocation_upload_ms": 3.4105,
        "cold_call_ms": 3.979,
        "warmup_ms": 200.036209,
        "download_ms": 0.952083,
        "repetitions": 1181,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          21.3798687553,
          21.1161803556,
          21.2635478408,
          21.2423090601,
          21.8577832345,
          22.0325292125,
          21.3736951736,
          21.4887806943,
          21.4624961897
        ],
        "latency_us": [
          316.25,
          2343.375,
          569.292,
          337.583,
          482.167,
          253.666,
          305.375,
          260.833,
          282.042
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 32,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 256,
            "reduction_scalar_rounds": 36,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 24576,
              "global_write_bytes": 12288,
              "private_read_bytes": 36864,
              "private_write_bytes": 12288
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 48,
              "private_read_bytes": 144,
              "private_write_bytes": 48
            },
            "striped_storage_scalars_per_worker": 12,
            "reduction_operations": 1,
            "reduction_elements": 3072,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 68,
            "normalized_kernel_cost": 272,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 6144,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 1328083,
              "compute_span_ns": 1328083,
              "calibration_cpu_ns": 2678958,
              "calibration_gpu_ticks": 2678958,
              "compute_passes": 1,
              "command_buffer_ns": 1328125.01397,
              "command_buffers": 2
            },
            {
              "compute_ns": 935750,
              "compute_span_ns": 935750,
              "calibration_cpu_ns": 1585750,
              "calibration_gpu_ticks": 1585750,
              "compute_passes": 1,
              "command_buffer_ns": 935833.31909,
              "command_buffers": 2
            },
            {
              "compute_ns": 1356500,
              "compute_span_ns": 1356500,
              "calibration_cpu_ns": 2638292,
              "calibration_gpu_ticks": 2638292,
              "compute_passes": 1,
              "command_buffer_ns": 1356625.05636,
              "command_buffers": 2
            },
            {
              "compute_ns": 973583,
              "compute_span_ns": 973583,
              "calibration_cpu_ns": 1532750,
              "calibration_gpu_ticks": 1532750,
              "compute_passes": 1,
              "command_buffer_ns": 973666.66887,
              "command_buffers": 2
            },
            {
              "compute_ns": 1290917,
              "compute_span_ns": 1290917,
              "calibration_cpu_ns": 2643000,
              "calibration_gpu_ticks": 2643000,
              "compute_passes": 1,
              "command_buffer_ns": 1291083.28838,
              "command_buffers": 2
            },
            {
              "compute_ns": 935708,
              "compute_span_ns": 935708,
              "calibration_cpu_ns": 1521833,
              "calibration_gpu_ticks": 1521833,
              "compute_passes": 1,
              "command_buffer_ns": 935791.700613,
              "command_buffers": 2
            },
            {
              "compute_ns": 1362042,
              "compute_span_ns": 1362042,
              "calibration_cpu_ns": 2446584,
              "calibration_gpu_ticks": 2446584,
              "compute_passes": 1,
              "command_buffer_ns": 1362125.04003,
              "command_buffers": 2
            },
            {
              "compute_ns": 924750,
              "compute_span_ns": 924750,
              "calibration_cpu_ns": 1571500,
              "calibration_gpu_ticks": 1571500,
              "compute_passes": 1,
              "command_buffer_ns": 924791.675061,
              "command_buffers": 2
            },
            {
              "compute_ns": 1390625,
              "compute_span_ns": 1390625,
              "calibration_cpu_ns": 2729833,
              "calibration_gpu_ticks": 2729833,
              "compute_passes": 1,
              "command_buffer_ns": 1390749.93785,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 21542,
              "compute_span_ns": 21542,
              "calibration_cpu_ns": 1105458,
              "calibration_gpu_ticks": 1105458,
              "compute_passes": 1,
              "command_buffer_ns": 21708.3725147,
              "command_buffers": 2
            },
            {
              "compute_ns": 19125,
              "compute_span_ns": 19125,
              "calibration_cpu_ns": 789208,
              "calibration_gpu_ticks": 789208,
              "compute_passes": 1,
              "command_buffer_ns": 19166.6767932,
              "command_buffers": 2
            },
            {
              "compute_ns": 18500,
              "compute_span_ns": 18500,
              "calibration_cpu_ns": 422167,
              "calibration_gpu_ticks": 422167,
              "compute_passes": 1,
              "command_buffer_ns": 18500.0244528,
              "command_buffers": 2
            },
            {
              "compute_ns": 85083,
              "compute_span_ns": 85083,
              "calibration_cpu_ns": 1441917,
              "calibration_gpu_ticks": 1441917,
              "compute_passes": 1,
              "command_buffer_ns": 85083.3603181,
              "command_buffers": 2
            },
            {
              "compute_ns": 18416,
              "compute_span_ns": 18416,
              "calibration_cpu_ns": 541084,
              "calibration_gpu_ticks": 541084,
              "compute_passes": 1,
              "command_buffer_ns": 18374.9943972,
              "command_buffers": 2
            },
            {
              "compute_ns": 42292,
              "compute_span_ns": 42292,
              "calibration_cpu_ns": 1437750,
              "calibration_gpu_ticks": 1437750,
              "compute_passes": 1,
              "command_buffer_ns": 42333.3258368,
              "command_buffers": 2
            },
            {
              "compute_ns": 19042,
              "compute_span_ns": 19042,
              "calibration_cpu_ns": 479000,
              "calibration_gpu_ticks": 479000,
              "compute_passes": 1,
              "command_buffer_ns": 19125.000108,
              "command_buffers": 2
            },
            {
              "compute_ns": 63667,
              "compute_span_ns": 63667,
              "calibration_cpu_ns": 1593041,
              "calibration_gpu_ticks": 1593041,
              "compute_passes": 1,
              "command_buffer_ns": 63791.6964479,
              "command_buffers": 2
            },
            {
              "compute_ns": 18417,
              "compute_span_ns": 18417,
              "calibration_cpu_ns": 463459,
              "calibration_gpu_ticks": 463459,
              "compute_passes": 1,
              "command_buffer_ns": 18458.3477676,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1233333.3143,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1445916.65594,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 939041.72536,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1179624.93096,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 938250.042964,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1355958.34581,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 935750.023928,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1421249.99547,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 946750.049479,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 17291.6916199,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18666.6147783,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18250.0225492,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 31958.3341479,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 16875.0411831,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19708.299078,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 16583.2461789,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20208.361093,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 16250.0073202,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              19.2708330359375,
              22.5924477490625,
              14.67252695875,
              18.43163954625,
              14.6601569213125,
              21.18684915328125,
              14.621094123875,
              22.20703117921875,
              14.792969523109376
            ],
            "command_buffer_throughput_us_p50": 18.43163954625,
            "command_buffer_throughput_us_p90": 22.2841144931875,
            "command_buffer_latency_us": [
              17.2916916199,
              18.6666147783,
              18.2500225492,
              31.9583341479,
              16.8750411831,
              19.708299078,
              16.5832461789,
              20.208361093,
              16.2500073202
            ],
            "command_buffer_latency_us_p50": 18.2500225492,
            "command_buffer_latency_us_p90": 22.558355703980002
          },
          "compute_throughput_us": [
            20.751296875,
            14.62109375,
            21.1953125,
            15.212234375,
            20.170578125,
            14.6204375,
            21.28190625,
            14.44921875,
            21.728515625
          ],
          "compute_throughput_us_p50": 20.170578125,
          "compute_throughput_us_p90": 21.371228125,
          "command_buffer_throughput_us": [
            20.75195334328125,
            14.62239561078125,
            21.197266505625002,
            15.213541701093751,
            20.1731763809375,
            14.621745322078125,
            21.28320375046875,
            14.449869922828126,
            21.730467778906252
          ],
          "command_buffer_throughput_us_p50": 20.1731763809375,
          "command_buffer_throughput_us_p90": 21.37265655615625,
          "compute_latency_us": [
            21.542,
            19.125,
            18.5,
            85.083,
            18.416,
            42.292,
            19.042,
            63.667,
            18.417
          ],
          "compute_latency_us_p50": 19.125,
          "compute_latency_us_p90": 67.95020000000001,
          "command_buffer_latency_us": [
            21.7083725147,
            19.1666767932,
            18.5000244528,
            85.0833603181,
            18.374994397200002,
            42.3333258368,
            19.125000108000002,
            63.7916964479,
            18.4583477676
          ],
          "command_buffer_latency_us_p50": 19.1666767932,
          "command_buffer_latency_us_p90": 68.05002922194001,
          "counter_control_throughput_ratio": 1.0768581360536755,
          "counter_control_latency_ratio": 1.1532723992443679
        },
        "correctness": {
          "max_abs_error": 1.6175184480715643e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 21.3798687553,
        "throughput_us_p90": 21.8927324301,
        "latency_us_p50": 316.25,
        "latency_us_p90": 924.1086000000004
      },
      "slowdown": 0.7797901687681722,
      "device_compute_slowdown": 1.2181806514807365,
      "valid": true,
      "round": 0,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "rmsnorm_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-okhrvgao/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "232522da57db77c64b377ec2facdc2e1df33cd57064a4297b89d6582fd4ad63b",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.381041,
        "capture_ms": 0.063292,
        "compile_ms": 57.025917,
        "allocation_upload_ms": 7.65775,
        "cold_call_ms": 6.199125,
        "warmup_ms": 200.646916,
        "download_ms": 2.975083,
        "repetitions": 309,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          90.9286666667,
          86.7580906149,
          91.0985695793,
          90.1441456311,
          87.2042880259,
          90.8408867314,
          90.5246763754,
          87.0440938511,
          87.8704142395
        ],
        "latency_us": [
          2411.833,
          455.791,
          462.916,
          273.5,
          356.333,
          292.666,
          300.083,
          266.334,
          312.084
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 32,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 768,
            "reduction_scalar_rounds": 72,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 49152,
              "global_write_bytes": 24576,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 192,
              "global_write_bytes": 96,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 1,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 104,
            "normalized_kernel_cost": 1248,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4206167,
              "compute_span_ns": 4206167,
              "calibration_cpu_ns": 5313542,
              "calibration_gpu_ticks": 5313542,
              "compute_passes": 1,
              "command_buffer_ns": 4206250.01425,
              "command_buffers": 2
            },
            {
              "compute_ns": 4201666,
              "compute_span_ns": 4201666,
              "calibration_cpu_ns": 5403000,
              "calibration_gpu_ticks": 5403000,
              "compute_passes": 1,
              "command_buffer_ns": 4201666.62661,
              "command_buffers": 2
            },
            {
              "compute_ns": 6347583,
              "compute_span_ns": 6347583,
              "calibration_cpu_ns": 8921042,
              "calibration_gpu_ticks": 8921042,
              "compute_passes": 1,
              "command_buffer_ns": 6347708.28765,
              "command_buffers": 2
            },
            {
              "compute_ns": 3826792,
              "compute_span_ns": 3826792,
              "calibration_cpu_ns": 4698584,
              "calibration_gpu_ticks": 4698584,
              "compute_passes": 1,
              "command_buffer_ns": 3826875.00678,
              "command_buffers": 2
            },
            {
              "compute_ns": 4262042,
              "compute_span_ns": 4262042,
              "calibration_cpu_ns": 5335334,
              "calibration_gpu_ticks": 5335334,
              "compute_passes": 1,
              "command_buffer_ns": 4262125.00315,
              "command_buffers": 2
            },
            {
              "compute_ns": 4313125,
              "compute_span_ns": 4313125,
              "calibration_cpu_ns": 5557958,
              "calibration_gpu_ticks": 5557958,
              "compute_passes": 1,
              "command_buffer_ns": 4313250.01409,
              "command_buffers": 2
            },
            {
              "compute_ns": 4312459,
              "compute_span_ns": 4312459,
              "calibration_cpu_ns": 5276583,
              "calibration_gpu_ticks": 5276583,
              "compute_passes": 1,
              "command_buffer_ns": 4312500.00838,
              "command_buffers": 2
            },
            {
              "compute_ns": 4179958,
              "compute_span_ns": 4179958,
              "calibration_cpu_ns": 5376834,
              "calibration_gpu_ticks": 5376834,
              "compute_passes": 1,
              "command_buffer_ns": 4180041.66568,
              "command_buffers": 2
            },
            {
              "compute_ns": 6493667,
              "compute_span_ns": 6493667,
              "calibration_cpu_ns": 8285916,
              "calibration_gpu_ticks": 8285916,
              "compute_passes": 1,
              "command_buffer_ns": 6493750.02831,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 119583,
              "compute_span_ns": 119583,
              "calibration_cpu_ns": 1111959,
              "calibration_gpu_ticks": 1111959,
              "compute_passes": 1,
              "command_buffer_ns": 119541.655295,
              "command_buffers": 2
            },
            {
              "compute_ns": 62166,
              "compute_span_ns": 62166,
              "calibration_cpu_ns": 512375,
              "calibration_gpu_ticks": 512375,
              "compute_passes": 1,
              "command_buffer_ns": 62249.9501333,
              "command_buffers": 2
            },
            {
              "compute_ns": 61375,
              "compute_span_ns": 61375,
              "calibration_cpu_ns": 473417,
              "calibration_gpu_ticks": 473417,
              "compute_passes": 1,
              "command_buffer_ns": 61375.0307821,
              "command_buffers": 2
            },
            {
              "compute_ns": 63750,
              "compute_span_ns": 63750,
              "calibration_cpu_ns": 486333,
              "calibration_gpu_ticks": 486333,
              "compute_passes": 1,
              "command_buffer_ns": 63749.961555,
              "command_buffers": 2
            },
            {
              "compute_ns": 116666,
              "compute_span_ns": 116666,
              "calibration_cpu_ns": 1192125,
              "calibration_gpu_ticks": 1192125,
              "compute_passes": 1,
              "command_buffer_ns": 116666.720714,
              "command_buffers": 2
            },
            {
              "compute_ns": 111959,
              "compute_span_ns": 111959,
              "calibration_cpu_ns": 830458,
              "calibration_gpu_ticks": 830458,
              "compute_passes": 1,
              "command_buffer_ns": 111999.979708,
              "command_buffers": 2
            },
            {
              "compute_ns": 110750,
              "compute_span_ns": 110750,
              "calibration_cpu_ns": 1340958,
              "calibration_gpu_ticks": 1340958,
              "compute_passes": 1,
              "command_buffer_ns": 110833.323561,
              "command_buffers": 2
            },
            {
              "compute_ns": 80000,
              "compute_span_ns": 80000,
              "calibration_cpu_ns": 578042,
              "calibration_gpu_ticks": 578042,
              "compute_passes": 1,
              "command_buffer_ns": 80083.3222456,
              "command_buffers": 2
            },
            {
              "compute_ns": 68542,
              "compute_span_ns": 68542,
              "calibration_cpu_ns": 509750,
              "calibration_gpu_ticks": 509750,
              "compute_passes": 1,
              "command_buffer_ns": 68624.9695718,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4086416.68037,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4430249.97367,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4039124.99314,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3872708.35927,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4029916.6576,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4443541.63483,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4323541.65241,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6569916.61526,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3787249.96699,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 69833.3024047,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 252791.680396,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 108416.716103,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 117291.69637,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63374.9878034,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 120166.689157,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 88125.0016391,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 61791.6812189,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 131833.308842,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              63.85026063078125,
              69.22265583859375,
              63.1113280178125,
              60.51106811359375,
              62.967447775000004,
              69.43033804421874,
              67.55533831890624,
              102.6549471134375,
              59.175780734218755
            ],
            "command_buffer_throughput_us_p50": 63.85026063078125,
            "command_buffer_throughput_us_p90": 76.0752598580625,
            "command_buffer_latency_us": [
              69.8333024047,
              252.791680396,
              108.416716103,
              117.29169637000001,
              63.374987803399996,
              120.16668915700001,
              88.12500163909999,
              61.791681218899996,
              131.833308842
            ],
            "command_buffer_latency_us_p50": 108.416716103,
            "command_buffer_latency_us_p90": 156.02498315280002
          },
          "compute_throughput_us": [
            65.721359375,
            65.65103125,
            99.180984375,
            59.793625,
            66.59440625,
            67.392578125,
            67.382171875,
            65.31184375,
            101.463546875
          ],
          "compute_throughput_us_p50": 66.59440625,
          "compute_throughput_us_p90": 99.637496875,
          "command_buffer_throughput_us": [
            65.72265647265625,
            65.65104104078125,
            99.18294199453125,
            59.7949219809375,
            66.59570317421876,
            67.39453147015625,
            67.3828126309375,
            65.31315102625,
            101.46484419234375
          ],
          "command_buffer_throughput_us_p50": 66.59570317421876,
          "command_buffer_throughput_us_p90": 99.63932243409374,
          "compute_latency_us": [
            119.583,
            62.166,
            61.375,
            63.75,
            116.666,
            111.959,
            110.75,
            80.0,
            68.542
          ],
          "compute_latency_us_p50": 80.0,
          "compute_latency_us_p90": 117.2494,
          "command_buffer_latency_us": [
            119.541655295,
            62.249950133300004,
            61.375030782100005,
            63.749961555,
            116.666720714,
            111.999979708,
            110.833323561,
            80.08332224559999,
            68.6249695718
          ],
          "command_buffer_latency_us_p50": 80.08332224559999,
          "command_buffer_latency_us_p90": 117.2417076302,
          "counter_control_throughput_ratio": 0.9974461575907693,
          "counter_control_latency_ratio": 0.932038491646133
        },
        "correctness": {
          "max_abs_error": 1.4915225321487924e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 90.1441456311,
        "throughput_us_p90": 90.96264724922,
        "latency_us_p50": 312.084,
        "latency_us_p90": 852.6994000000004
      },
      "torch": {
        "cold_call_ms": 10.628375,
        "warmup_ms": 200.270625,
        "repetitions": 269,
        "throughput_us": [
          120.81707063197025,
          121.27663940520445,
          120.63150557620818,
          120.28609293680299,
          112.3130408921933,
          115.52571375464683,
          120.97877695167288,
          120.31505576208178,
          120.67224163568774
        ],
        "latency_us": [
          447.916,
          398.625,
          355.66700000000003,
          331.5,
          322.20799999999997,
          329.25,
          362.791,
          325.25,
          315.125
        ],
        "throughput_us_p50": 120.63150557620818,
        "throughput_us_p90": 121.0383494423792,
        "latency_us_p50": 331.5,
        "latency_us_p90": 408.4832,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 5195499.979890883,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5021708.377171308,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5403625.022154301,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5309833.330102265,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5122291.622683406,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5168750.009033829,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5075541.674159467,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5003666.679840535,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 7759624.975733459,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 219666.6900999844,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 216874.9924749136,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 95749.97238814831,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 173583.33570882678,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 87374.9959282577,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 243375.01963600516,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 110124.99453499913,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 113833.34640413523,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 122500.00145286322,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              81.17968718579505,
              78.46419339330168,
              84.43164097116096,
              82.96614578284789,
              80.03580660442822,
              80.76171889115358,
              79.30533865874168,
              78.18229187250836,
              121.2441402458353
            ],
            "command_buffer_throughput_us_p50": 80.76171889115358,
            "command_buffer_throughput_us_p90": 91.79414082609583,
            "command_buffer_latency_us": [
              219.6666900999844,
              216.8749924749136,
              95.74997238814831,
              173.58333570882678,
              87.3749959282577,
              243.37501963600516,
              110.12499453499913,
              113.83334640413523,
              122.50000145286322
            ],
            "command_buffer_latency_us_p50": 122.50000145286322,
            "command_buffer_latency_us_p90": 224.40835600718856
          },
          "throughput": [
            {
              "compute_ns": 5359875.0,
              "compute_span_ns": 5359875.0,
              "command_buffer_ns": 5359874.980058521,
              "calibration_cpu_ns": 6745500.0,
              "calibration_gpu_ticks": 6745500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5216208.0,
              "compute_span_ns": 5216208.0,
              "command_buffer_ns": 5216208.3447910845,
              "calibration_cpu_ns": 6532000.0,
              "calibration_gpu_ticks": 6532000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5219292.0,
              "compute_span_ns": 5219292.0,
              "command_buffer_ns": 5219333.339482546,
              "calibration_cpu_ns": 6600042.0,
              "calibration_gpu_ticks": 6600042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5035250.0,
              "compute_span_ns": 5035250.0,
              "command_buffer_ns": 5035249.9820292,
              "calibration_cpu_ns": 7864500.0,
              "calibration_gpu_ticks": 7864500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5015458.0,
              "compute_span_ns": 5015458.0,
              "command_buffer_ns": 5015416.65289551,
              "calibration_cpu_ns": 7042875.0,
              "calibration_gpu_ticks": 7042875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5077792.0,
              "compute_span_ns": 5077792.0,
              "command_buffer_ns": 5077791.691292077,
              "calibration_cpu_ns": 7366375.0,
              "calibration_gpu_ticks": 7366375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5060708.0,
              "compute_span_ns": 5060708.0,
              "command_buffer_ns": 5060666.64820537,
              "calibration_cpu_ns": 6855541.0,
              "calibration_gpu_ticks": 6855541.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5053000.0,
              "compute_span_ns": 5053000.0,
              "command_buffer_ns": 5053000.000771135,
              "calibration_cpu_ns": 7307708.0,
              "calibration_gpu_ticks": 7307708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5017625.0,
              "compute_span_ns": 5017625.0,
              "command_buffer_ns": 5017624.993342906,
              "calibration_cpu_ns": 7034875.0,
              "calibration_gpu_ticks": 7034875.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 84666.0,
              "compute_span_ns": 84666.0,
              "command_buffer_ns": 84625.03319606185,
              "calibration_cpu_ns": 707291.0,
              "calibration_gpu_ticks": 707291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80792.0,
              "compute_span_ns": 80792.0,
              "command_buffer_ns": 80833.32795649767,
              "calibration_cpu_ns": 667292.0,
              "calibration_gpu_ticks": 667292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 81083.0,
              "compute_span_ns": 81083.0,
              "command_buffer_ns": 81041.65317490697,
              "calibration_cpu_ns": 618334.0,
              "calibration_gpu_ticks": 618334.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 81583.0,
              "compute_span_ns": 81583.0,
              "command_buffer_ns": 81541.65698215365,
              "calibration_cpu_ns": 512208.0,
              "calibration_gpu_ticks": 512208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80125.0,
              "compute_span_ns": 80125.0,
              "command_buffer_ns": 80124.99893084168,
              "calibration_cpu_ns": 617500.0,
              "calibration_gpu_ticks": 617500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 120750.0,
              "compute_span_ns": 120750.0,
              "command_buffer_ns": 120749.98812749982,
              "calibration_cpu_ns": 950417.0,
              "calibration_gpu_ticks": 950417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 84542.0,
              "compute_span_ns": 84542.0,
              "command_buffer_ns": 84541.67982563376,
              "calibration_cpu_ns": 599625.0,
              "calibration_gpu_ticks": 599625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 81250.0,
              "compute_span_ns": 81250.0,
              "command_buffer_ns": 81250.03660097718,
              "calibration_cpu_ns": 1119875.0,
              "calibration_gpu_ticks": 1119875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 122959.0,
              "compute_span_ns": 122959.0,
              "command_buffer_ns": 123000.0052601099,
              "calibration_cpu_ns": 1306250.0,
              "calibration_gpu_ticks": 1306250.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            83.748046875,
            81.50325,
            81.5514375,
            78.67578125,
            78.36653125,
            79.3405,
            79.0735625,
            78.953125,
            78.400390625
          ],
          "compute_throughput_us_p50": 79.0735625,
          "compute_throughput_us_p90": 81.99075937500001,
          "command_buffer_throughput_us": [
            83.74804656341439,
            81.5032553873607,
            81.55208342941478,
            78.67578096920624,
            78.36588520149235,
            79.34049517643871,
            79.0729163782089,
            78.95312501204899,
            78.40039052098291
          ],
          "command_buffer_throughput_us_p50": 79.0729163782089,
          "command_buffer_throughput_us_p90": 81.9912760562147,
          "compute_latency_us": [
            84.666,
            80.792,
            81.083,
            81.583,
            80.125,
            120.75,
            84.542,
            81.25,
            122.959
          ],
          "compute_latency_us_p50": 81.583,
          "compute_latency_us_p90": 121.1918,
          "command_buffer_latency_us": [
            84.62503319606185,
            80.83332795649767,
            81.04165317490697,
            81.54165698215365,
            80.12499893084168,
            120.74998812749982,
            84.54167982563376,
            81.25003660097718,
            123.0000052601099
          ],
          "command_buffer_latency_us_p50": 81.54165698215365,
          "command_buffer_latency_us_p90": 121.19999155402184,
          "counter_control_throughput_ratio": 0.9824022602016393,
          "counter_control_latency_ratio": 0.7137630507015088
        },
        "download_ms": 0.686,
        "allocation_upload_ms": 18.107625,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.9755493041451189e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.7472686774530225,
      "device_compute_slowdown": 0.8421829514763548,
      "valid": true,
      "round": 0,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "rmsnorm_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.387458,
        "warmup_ms": 200.510709,
        "repetitions": 206,
        "throughput_us": [
          119.72128155339806,
          117.66100485436894,
          109.85133495145631,
          115.09466019417475,
          121.90533980582525,
          120.96783980582525,
          121.90473300970874,
          120.36084466019418,
          122.662213592233
        ],
        "latency_us": [
          424.58299999999997,
          2472.042,
          876.125,
          547.125,
          280.0,
          384.625,
          311.0,
          318.75,
          317.75
        ],
        "throughput_us_p50": 120.36084466019418,
        "throughput_us_p90": 122.0567145631068,
        "latency_us_p50": 384.625,
        "latency_us_p90": 1195.3084000000003,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 5284458.340611309,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5159749.99871105,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5325749.982148409,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5115000.007208437,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5245875.043328851,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5102499.9702349305,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5152666.650246829,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 7571333.320811391,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 7505499.990656972,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 117124.98962879181,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 83499.99552592635,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 118333.32246169448,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80375.00083446503,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 117291.69636964798,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80999.97648969293,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 116875.04593282938,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 117000.0177808106,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 114166.64347052574,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              82.5696615720517,
              80.62109372986015,
              83.21484347106889,
              79.92187511263182,
              81.9667975520133,
              79.72656203492079,
              80.5104164101067,
              118.30208313767798,
              117.27343735401519
            ],
            "command_buffer_throughput_us_p50": 81.9667975520133,
            "command_buffer_throughput_us_p90": 117.47916651074775,
            "command_buffer_latency_us": [
              117.12498962879181,
              83.49999552592635,
              118.33332246169448,
              80.37500083446503,
              117.29169636964798,
              80.99997648969293,
              116.87504593282938,
              117.0000177808106,
              114.16664347052574
            ],
            "command_buffer_latency_us_p50": 116.87504593282938,
            "command_buffer_latency_us_p90": 117.50002158805728
          },
          "throughput": [
            {
              "compute_ns": 5124125.0,
              "compute_span_ns": 5124125.0,
              "command_buffer_ns": 5124124.989379197,
              "calibration_cpu_ns": 6726625.0,
              "calibration_gpu_ticks": 6726625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5276500.0,
              "compute_span_ns": 5276500.0,
              "command_buffer_ns": 5276500.014588237,
              "calibration_cpu_ns": 7237250.0,
              "calibration_gpu_ticks": 7237250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5156250.0,
              "compute_span_ns": 5156250.0,
              "command_buffer_ns": 5156249.972060323,
              "calibration_cpu_ns": 7359916.0,
              "calibration_gpu_ticks": 7359916.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5035458.0,
              "compute_span_ns": 5035458.0,
              "command_buffer_ns": 5035416.688770056,
              "calibration_cpu_ns": 7196875.0,
              "calibration_gpu_ticks": 7196875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4970375.0,
              "compute_span_ns": 4970375.0,
              "command_buffer_ns": 4970374.98280406,
              "calibration_cpu_ns": 7754417.0,
              "calibration_gpu_ticks": 7754417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4983625.0,
              "compute_span_ns": 4983625.0,
              "command_buffer_ns": 4983624.967280775,
              "calibration_cpu_ns": 7323542.0,
              "calibration_gpu_ticks": 7323542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5246458.0,
              "compute_span_ns": 5246458.0,
              "command_buffer_ns": 5246458.342298865,
              "calibration_cpu_ns": 6513208.0,
              "calibration_gpu_ticks": 6513208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5159958.0,
              "compute_span_ns": 5159958.0,
              "command_buffer_ns": 5159916.647244245,
              "calibration_cpu_ns": 6634542.0,
              "calibration_gpu_ticks": 6634542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5203833.0,
              "compute_span_ns": 5203833.0,
              "command_buffer_ns": 5203833.33787322,
              "calibration_cpu_ns": 7333584.0,
              "calibration_gpu_ticks": 7333584.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 81791.0,
              "compute_span_ns": 81791.0,
              "command_buffer_ns": 81749.98220056295,
              "calibration_cpu_ns": 841417.0,
              "calibration_gpu_ticks": 841417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 114292.0,
              "compute_span_ns": 114292.0,
              "command_buffer_ns": 114291.67352616787,
              "calibration_cpu_ns": 803042.0,
              "calibration_gpu_ticks": 803042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80541.0,
              "compute_span_ns": 80541.0,
              "command_buffer_ns": 80499.97268244624,
              "calibration_cpu_ns": 633375.0,
              "calibration_gpu_ticks": 633375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 127750.0,
              "compute_span_ns": 127750.0,
              "command_buffer_ns": 127750.04142895341,
              "calibration_cpu_ns": 1204875.0,
              "calibration_gpu_ticks": 1204875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 119500.0,
              "compute_span_ns": 119500.0,
              "command_buffer_ns": 119500.03681704402,
              "calibration_cpu_ns": 1075333.0,
              "calibration_gpu_ticks": 1075333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 115083.0,
              "compute_span_ns": 115083.0,
              "command_buffer_ns": 115083.35592225194,
              "calibration_cpu_ns": 793667.0,
              "calibration_gpu_ticks": 793667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 81750.0,
              "compute_span_ns": 81750.0,
              "command_buffer_ns": 81749.98220056295,
              "calibration_cpu_ns": 518834.0,
              "calibration_gpu_ticks": 518834.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80917.0,
              "compute_span_ns": 80917.0,
              "command_buffer_ns": 80958.29980447888,
              "calibration_cpu_ns": 503000.0,
              "calibration_gpu_ticks": 503000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 81417.0,
              "compute_span_ns": 81417.0,
              "command_buffer_ns": 81416.68513417244,
              "calibration_cpu_ns": 665958.0,
              "calibration_gpu_ticks": 665958.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            80.064453125,
            82.4453125,
            80.56640625,
            78.67903125,
            77.662109375,
            77.869140625,
            81.97590625,
            80.62434375,
            81.309890625
          ],
          "compute_throughput_us_p50": 80.56640625,
          "compute_throughput_us_p90": 82.06978749999999,
          "command_buffer_throughput_us": [
            80.06445295904996,
            82.4453127279412,
            80.56640581344254,
            78.67838576203212,
            77.66210910631344,
            77.86914011376211,
            81.97591159841977,
            80.62369761319133,
            81.30989590426907
          ],
          "command_buffer_throughput_us_p50": 80.56640581344254,
          "command_buffer_throughput_us_p90": 82.06979182432406,
          "compute_latency_us": [
            81.791,
            114.292,
            80.541,
            127.75,
            119.5,
            115.083,
            81.75,
            80.917,
            81.417
          ],
          "compute_latency_us_p50": 81.791,
          "compute_latency_us_p90": 121.15,
          "command_buffer_latency_us": [
            81.74998220056295,
            114.29167352616787,
            80.49997268244624,
            127.75004142895341,
            119.50003681704402,
            115.08335592225194,
            81.74998220056295,
            80.95829980447888,
            81.41668513417244
          ],
          "command_buffer_latency_us_p50": 81.74998220056295,
          "command_buffer_latency_us_p90": 121.1500377394259,
          "counter_control_throughput_ratio": 0.9696594540257907,
          "counter_control_latency_ratio": 0.7131389927846279
        },
        "download_ms": 0.689208,
        "allocation_upload_ms": 1.197667,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.9755493041451189e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-mk1m5c_5/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "c9fd11968a5edbf77e894438f42e4bd465ad146629716a6f08eacac9ab82063f",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.440167,
        "capture_ms": 0.077708,
        "compile_ms": 56.398291,
        "allocation_upload_ms": 9.104458,
        "cold_call_ms": 9.796667,
        "warmup_ms": 200.538375,
        "download_ms": 3.475291,
        "repetitions": 309,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          93.998381877,
          100.738809061,
          91.8594919094,
          100.939184466,
          95.8807993528,
          95.8948220065,
          101.055957929,
          92.5316893204,
          100.418016181
        ],
        "latency_us": [
          417.541,
          318.5,
          279.042,
          288,
          278.625,
          342.167,
          2067.917,
          672.792,
          422.417
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 384,
            "reduction_scalar_rounds": 72,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 49152,
              "global_write_bytes": 24576,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 192,
              "global_write_bytes": 96,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 1,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 96,
            "normalized_kernel_cost": 1152,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4669542,
              "compute_span_ns": 4669542,
              "calibration_cpu_ns": 5805750,
              "calibration_gpu_ticks": 5805750,
              "compute_passes": 1,
              "command_buffer_ns": 4669708.31621,
              "command_buffers": 2
            },
            {
              "compute_ns": 4763375,
              "compute_span_ns": 4763375,
              "calibration_cpu_ns": 5744583,
              "calibration_gpu_ticks": 5744583,
              "compute_passes": 1,
              "command_buffer_ns": 4763500.00827,
              "command_buffers": 2
            },
            {
              "compute_ns": 4665208,
              "compute_span_ns": 4665208,
              "calibration_cpu_ns": 5749916,
              "calibration_gpu_ticks": 5749916,
              "compute_passes": 1,
              "command_buffer_ns": 4665291.69353,
              "command_buffers": 2
            },
            {
              "compute_ns": 4509917,
              "compute_span_ns": 4509917,
              "calibration_cpu_ns": 5644833,
              "calibration_gpu_ticks": 5644833,
              "compute_passes": 1,
              "command_buffer_ns": 4510000.05705,
              "command_buffers": 2
            },
            {
              "compute_ns": 4607500,
              "compute_span_ns": 4607500,
              "calibration_cpu_ns": 5698083,
              "calibration_gpu_ticks": 5698083,
              "compute_passes": 1,
              "command_buffer_ns": 4607541.60304,
              "command_buffers": 2
            },
            {
              "compute_ns": 4768333,
              "compute_span_ns": 4768333,
              "calibration_cpu_ns": 5774417,
              "calibration_gpu_ticks": 5774417,
              "compute_passes": 1,
              "command_buffer_ns": 4768458.36965,
              "command_buffers": 2
            },
            {
              "compute_ns": 4773250,
              "compute_span_ns": 4773250,
              "calibration_cpu_ns": 5726708,
              "calibration_gpu_ticks": 5726708,
              "compute_passes": 1,
              "command_buffer_ns": 4773333.31946,
              "command_buffers": 2
            },
            {
              "compute_ns": 4702041,
              "compute_span_ns": 4702041,
              "calibration_cpu_ns": 5711916,
              "calibration_gpu_ticks": 5711916,
              "compute_passes": 1,
              "command_buffer_ns": 4702125.03569,
              "command_buffers": 2
            },
            {
              "compute_ns": 4728875,
              "compute_span_ns": 4728875,
              "calibration_cpu_ns": 5813291,
              "calibration_gpu_ticks": 5813291,
              "compute_passes": 1,
              "command_buffer_ns": 4728958.30171,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 117625,
              "compute_span_ns": 117625,
              "calibration_cpu_ns": 857458,
              "calibration_gpu_ticks": 857458,
              "compute_passes": 1,
              "command_buffer_ns": 117666.670121,
              "command_buffers": 2
            },
            {
              "compute_ns": 66708,
              "compute_span_ns": 66708,
              "calibration_cpu_ns": 587500,
              "calibration_gpu_ticks": 587500,
              "compute_passes": 1,
              "command_buffer_ns": 66749.9843985,
              "command_buffers": 2
            },
            {
              "compute_ns": 112416,
              "compute_span_ns": 112416,
              "calibration_cpu_ns": 755667,
              "calibration_gpu_ticks": 755667,
              "compute_passes": 1,
              "command_buffer_ns": 112500.041723,
              "command_buffers": 2
            },
            {
              "compute_ns": 67000,
              "compute_span_ns": 67000,
              "calibration_cpu_ns": 599875,
              "calibration_gpu_ticks": 599875,
              "compute_passes": 1,
              "command_buffer_ns": 67124.9581501,
              "command_buffers": 2
            },
            {
              "compute_ns": 100667,
              "compute_span_ns": 100667,
              "calibration_cpu_ns": 768417,
              "calibration_gpu_ticks": 768417,
              "compute_passes": 1,
              "command_buffer_ns": 100666.65709,
              "command_buffers": 2
            },
            {
              "compute_ns": 67792,
              "compute_span_ns": 67792,
              "calibration_cpu_ns": 461458,
              "calibration_gpu_ticks": 461458,
              "compute_passes": 1,
              "command_buffer_ns": 67791.6686982,
              "command_buffers": 2
            },
            {
              "compute_ns": 118208,
              "compute_span_ns": 118208,
              "calibration_cpu_ns": 737584,
              "calibration_gpu_ticks": 737584,
              "compute_passes": 1,
              "command_buffer_ns": 118250.027299,
              "command_buffers": 2
            },
            {
              "compute_ns": 67042,
              "compute_span_ns": 67042,
              "calibration_cpu_ns": 576584,
              "calibration_gpu_ticks": 576584,
              "compute_passes": 1,
              "command_buffer_ns": 67124.9581501,
              "command_buffers": 2
            },
            {
              "compute_ns": 123000,
              "compute_span_ns": 123000,
              "calibration_cpu_ns": 656541,
              "calibration_gpu_ticks": 656541,
              "compute_passes": 1,
              "command_buffer_ns": 123000.00526,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4535374.98834,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4638500.04598,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4416625.01544,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4669833.28806,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4426833.35859,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4672958.28275,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4706374.95164,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4690291.70927,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4663708.32874,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 365958.316252,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 68791.618105,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 357416.633051,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 68333.290983,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 228833.348956,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 66666.6310281,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 320249.99382,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 66041.6553728,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 326166.686136,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              70.8652341928125,
              72.4765632184375,
              69.00976586625,
              72.9661451259375,
              69.16927122796875,
              73.01497316796875,
              73.53710861937499,
              73.28580795734376,
              72.8704426365625
            ],
            "command_buffer_throughput_us_p50": 72.8704426365625,
            "command_buffer_throughput_us_p90": 73.33606808975001,
            "command_buffer_latency_us": [
              365.958316252,
              68.791618105,
              357.416633051,
              68.333290983,
              228.833348956,
              66.6666310281,
              320.24999382,
              66.0416553728,
              326.16668613599995
            ],
            "command_buffer_latency_us_p50": 228.833348956,
            "command_buffer_latency_us_p90": 359.1249696912
          },
          "compute_throughput_us": [
            72.96159375,
            74.427734375,
            72.893875,
            70.467453125,
            71.9921875,
            74.505203125,
            74.58203125,
            73.469390625,
            73.888671875
          ],
          "compute_throughput_us_p50": 73.469390625,
          "compute_throughput_us_p90": 74.52056875,
          "command_buffer_throughput_us": [
            72.96419244078125,
            74.42968762921875,
            72.89518271140625,
            70.46875089140624,
            71.9928375475,
            74.50716202578124,
            74.5833331165625,
            73.47070368265625,
            73.88997346421876
          ],
          "command_buffer_throughput_us_p50": 73.47070368265625,
          "command_buffer_throughput_us_p90": 74.5223962439375,
          "compute_latency_us": [
            117.625,
            66.708,
            112.416,
            67.0,
            100.667,
            67.792,
            118.208,
            67.042,
            123.0
          ],
          "compute_latency_us_p50": 100.667,
          "compute_latency_us_p90": 119.1664,
          "command_buffer_latency_us": [
            117.66667012100001,
            66.7499843985,
            112.500041723,
            67.12495815010001,
            100.66665708999999,
            67.7916686982,
            118.250027299,
            67.12495815010001,
            123.00000526000001
          ],
          "command_buffer_latency_us_p50": 100.66665708999999,
          "command_buffer_latency_us_p90": 119.20002289119999,
          "counter_control_throughput_ratio": 1.020436751436993,
          "counter_control_latency_ratio": 0.4399125282624612
        },
        "correctness": {
          "max_abs_error": 1.4915225321487924e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 95.8948220065,
        "throughput_us_p90": 100.9625391586,
        "latency_us_p50": 342.167,
        "latency_us_p90": 951.8170000000002
      },
      "slowdown": 0.7967277255093441,
      "device_compute_slowdown": 0.9119109818181819,
      "valid": true,
      "round": 0,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "rmsnorm_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-wvwj9evp/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "310a51aaac34ebd04d17becc4645f77793138816428bb43603834b11ab1687fb",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.028667,
        "capture_ms": 0.079083,
        "compile_ms": 57.141292,
        "allocation_upload_ms": 7.5695,
        "cold_call_ms": 11.122917,
        "warmup_ms": 200.765333,
        "download_ms": 3.111333,
        "repetitions": 320,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          93.789975,
          89.51484375,
          96.25625,
          89.22513125,
          96.805859375,
          88.404296875,
          96.92200625,
          95.830078125,
          89.00546875
        ],
        "latency_us": [
          368.833,
          366.584,
          328.875,
          2279.042,
          763,
          544.5,
          265.542,
          290.709,
          360.667
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 512,
            "reduction_scalar_rounds": 48,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 32768,
              "global_write_bytes": 16384,
              "private_read_bytes": 49152,
              "private_write_bytes": 16384
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 128,
              "global_write_bytes": 64,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 1,
            "reduction_elements": 4096,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 72,
            "normalized_kernel_cost": 1152,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 8192,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4500834,
              "compute_span_ns": 4500834,
              "calibration_cpu_ns": 5482125,
              "calibration_gpu_ticks": 5482125,
              "compute_passes": 1,
              "command_buffer_ns": 4500916.63515,
              "command_buffers": 2
            },
            {
              "compute_ns": 4505209,
              "compute_span_ns": 4505209,
              "calibration_cpu_ns": 5698416,
              "calibration_gpu_ticks": 5698416,
              "compute_passes": 1,
              "command_buffer_ns": 4505291.69757,
              "command_buffers": 2
            },
            {
              "compute_ns": 4475000,
              "compute_span_ns": 4475000,
              "calibration_cpu_ns": 5454666,
              "calibration_gpu_ticks": 5454666,
              "compute_passes": 1,
              "command_buffer_ns": 4475083.31854,
              "command_buffers": 2
            },
            {
              "compute_ns": 4597167,
              "compute_span_ns": 4597167,
              "calibration_cpu_ns": 5758625,
              "calibration_gpu_ticks": 5758625,
              "compute_passes": 1,
              "command_buffer_ns": 4597208.34624,
              "command_buffers": 2
            },
            {
              "compute_ns": 4303416,
              "compute_span_ns": 4303416,
              "calibration_cpu_ns": 5667833,
              "calibration_gpu_ticks": 5667833,
              "compute_passes": 1,
              "command_buffer_ns": 4303416.7029,
              "command_buffers": 2
            },
            {
              "compute_ns": 4451958,
              "compute_span_ns": 4451958,
              "calibration_cpu_ns": 5688709,
              "calibration_gpu_ticks": 5688709,
              "compute_passes": 1,
              "command_buffer_ns": 4452000.02287,
              "command_buffers": 2
            },
            {
              "compute_ns": 4430000,
              "compute_span_ns": 4430000,
              "calibration_cpu_ns": 5655834,
              "calibration_gpu_ticks": 5655834,
              "compute_passes": 1,
              "command_buffer_ns": 4430083.32513,
              "command_buffers": 2
            },
            {
              "compute_ns": 6780000,
              "compute_span_ns": 6780000,
              "calibration_cpu_ns": 7648916,
              "calibration_gpu_ticks": 7648916,
              "compute_passes": 1,
              "command_buffer_ns": 6780124.96792,
              "command_buffers": 2
            },
            {
              "compute_ns": 6595041,
              "compute_span_ns": 6595041,
              "calibration_cpu_ns": 7313083,
              "calibration_gpu_ticks": 7313083,
              "compute_passes": 1,
              "command_buffer_ns": 6595041.71927,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 106250,
              "compute_span_ns": 106250,
              "calibration_cpu_ns": 845916,
              "calibration_gpu_ticks": 845916,
              "compute_passes": 1,
              "command_buffer_ns": 106291.61261,
              "command_buffers": 2
            },
            {
              "compute_ns": 66500,
              "compute_span_ns": 66500,
              "calibration_cpu_ns": 554875,
              "calibration_gpu_ticks": 554875,
              "compute_passes": 1,
              "command_buffer_ns": 66541.6591801,
              "command_buffers": 2
            },
            {
              "compute_ns": 104042,
              "compute_span_ns": 104042,
              "calibration_cpu_ns": 803667,
              "calibration_gpu_ticks": 803667,
              "compute_passes": 1,
              "command_buffer_ns": 104041.711893,
              "command_buffers": 2
            },
            {
              "compute_ns": 64958,
              "compute_span_ns": 64958,
              "calibration_cpu_ns": 504250,
              "calibration_gpu_ticks": 504250,
              "compute_passes": 1,
              "command_buffer_ns": 65041.6477583,
              "command_buffers": 2
            },
            {
              "compute_ns": 65000,
              "compute_span_ns": 65000,
              "calibration_cpu_ns": 478291,
              "calibration_gpu_ticks": 478291,
              "compute_passes": 1,
              "command_buffer_ns": 65125.0011288,
              "command_buffers": 2
            },
            {
              "compute_ns": 236083,
              "compute_span_ns": 236083,
              "calibration_cpu_ns": 1832833,
              "calibration_gpu_ticks": 1832833,
              "compute_passes": 1,
              "command_buffer_ns": 236166.699324,
              "command_buffers": 2
            },
            {
              "compute_ns": 65708,
              "compute_span_ns": 65708,
              "calibration_cpu_ns": 726417,
              "calibration_gpu_ticks": 726417,
              "compute_passes": 1,
              "command_buffer_ns": 65791.5952615,
              "command_buffers": 2
            },
            {
              "compute_ns": 108834,
              "compute_span_ns": 108834,
              "calibration_cpu_ns": 839500,
              "calibration_gpu_ticks": 839500,
              "compute_passes": 1,
              "command_buffer_ns": 108958.28018,
              "command_buffers": 2
            },
            {
              "compute_ns": 109583,
              "compute_span_ns": 109583,
              "calibration_cpu_ns": 1095000,
              "calibration_gpu_ticks": 1095000,
              "compute_passes": 1,
              "command_buffer_ns": 109583.314043,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4401958.3147,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4389624.98447,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4445500.03158,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4423500.03868,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4541916.5981,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4434749.94972,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4481416.66129,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6788708.38601,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6754583.32989,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 68208.3773427,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65750.0349917,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 106458.261143,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 108291.627839,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 64750.0273772,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 90499.9906197,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65083.3826512,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 103875.005152,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65041.705966,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              68.7805986671875,
              68.58789038234374,
              69.4609379934375,
              69.117188104375,
              70.9674468453125,
              69.292967964375,
              70.02213533265625,
              106.07356853140624,
              105.54036452953125
            ],
            "command_buffer_throughput_us_p50": 69.4609379934375,
            "command_buffer_throughput_us_p90": 105.64700532990625,
            "command_buffer_latency_us": [
              68.2083773427,
              65.7500349917,
              106.458261143,
              108.291627839,
              64.7500273772,
              90.4999906197,
              65.0833826512,
              103.875005152,
              65.041705966
            ],
            "command_buffer_latency_us_p50": 68.2083773427,
            "command_buffer_latency_us_p90": 106.8249344822
          },
          "compute_throughput_us": [
            70.32553125,
            70.393890625,
            69.921875,
            71.830734375,
            67.240875,
            69.56184375,
            69.21875,
            105.9375,
            103.047515625
          ],
          "compute_throughput_us_p50": 70.32553125,
          "compute_throughput_us_p90": 103.6255125,
          "command_buffer_throughput_us": [
            70.32682242421876,
            70.39518277453125,
            69.92317685218751,
            71.83138041,
            67.2408859828125,
            69.56250035734375,
            69.22005195515625,
            105.93945262375,
            103.04752686359376
          ],
          "command_buffer_throughput_us_p50": 70.32682242421876,
          "command_buffer_throughput_us_p90": 103.625912015625,
          "compute_latency_us": [
            106.25,
            66.5,
            104.042,
            64.958,
            65.0,
            236.083,
            65.708,
            108.834,
            109.583
          ],
          "compute_latency_us_p50": 104.042,
          "compute_latency_us_p90": 134.883,
          "command_buffer_latency_us": [
            106.29161261,
            66.5416591801,
            104.041711893,
            65.0416477583,
            65.12500112880001,
            236.16669932399998,
            65.79159526149999,
            108.95828018,
            109.583314043
          ],
          "command_buffer_latency_us_p50": 104.041711893,
          "command_buffer_latency_us_p90": 134.8999910992,
          "counter_control_throughput_ratio": 1.003889751022172,
          "counter_control_latency_ratio": 1.0120399052030913
        },
        "correctness": {
          "max_abs_error": 1.5832164268481108e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 93.789975,
        "throughput_us_p90": 96.82908875,
        "latency_us_p50": 366.584,
        "latency_us_p90": 1066.2084000000002
      },
      "torch": {
        "cold_call_ms": 8.187709,
        "warmup_ms": 201.766708,
        "repetitions": 383,
        "throughput_us": [
          106.63261618798956,
          106.73128720626632,
          105.0048955613577,
          104.13142036553525,
          98.8543315926893,
          105.02959007832898,
          104.95245953002612,
          103.08159268929504,
          97.96757963446476
        ],
        "latency_us": [
          429.417,
          312.125,
          306.25,
          299.75,
          2353.708,
          644.375,
          458.5,
          479.459,
          313.583
        ],
        "throughput_us_p50": 104.95245953002612,
        "throughput_us_p90": 106.6523503916449,
        "latency_us_p50": 429.417,
        "latency_us_p90": 986.2416000000003,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 6821000.017225742,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 6666958.332061768,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 6735333.299729973,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4659208.294469863,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4663250.001613051,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4585999.995470047,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4466624.988708645,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4559374.996460974,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4753666.697070003,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 289166.6372306645,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70124.9809935689,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 298666.6513606906,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 71166.66529327631,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 143791.66532307863,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70374.98289719224,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 195749.97713789344,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 306125.0317841768,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 89208.30441638827,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              106.57812526915222,
              104.17122393846512,
              105.23958280828083,
              72.80012960109161,
              72.86328127520392,
              71.65624992921948,
              69.79101544857258,
              71.24023431970272,
              74.27604214171879
            ],
            "command_buffer_throughput_us_p50": 72.86328127520392,
            "command_buffer_throughput_us_p90": 105.50729130045511,
            "command_buffer_latency_us": [
              289.1666372306645,
              70.1249809935689,
              298.6666513606906,
              71.16666529327631,
              143.79166532307863,
              70.37498289719224,
              195.74997713789344,
              306.1250317841768,
              89.20830441638827
            ],
            "command_buffer_latency_us_p50": 143.79166532307863,
            "command_buffer_latency_us_p90": 300.15832744538784
          },
          "throughput": [
            {
              "compute_ns": 6933333.0,
              "compute_span_ns": 6933333.0,
              "command_buffer_ns": 6933291.6755229235,
              "calibration_cpu_ns": 7918208.0,
              "calibration_gpu_ticks": 7918208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 7005333.0,
              "compute_span_ns": 7005333.0,
              "command_buffer_ns": 7005291.699897498,
              "calibration_cpu_ns": 7962084.0,
              "calibration_gpu_ticks": 7962084.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4343500.0,
              "compute_span_ns": 4343500.0,
              "command_buffer_ns": 4343500.011600554,
              "calibration_cpu_ns": 5412166.0,
              "calibration_gpu_ticks": 5412166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4616458.0,
              "compute_span_ns": 4616458.0,
              "command_buffer_ns": 4616458.318196237,
              "calibration_cpu_ns": 5876125.0,
              "calibration_gpu_ticks": 5876125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4521041.0,
              "compute_span_ns": 4521041.0,
              "command_buffer_ns": 4521000.024396926,
              "calibration_cpu_ns": 5888125.0,
              "calibration_gpu_ticks": 5888125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4687250.0,
              "compute_span_ns": 4687250.0,
              "command_buffer_ns": 4687249.951530248,
              "calibration_cpu_ns": 5991875.0,
              "calibration_gpu_ticks": 5991875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4561416.0,
              "compute_span_ns": 4561416.0,
              "command_buffer_ns": 4561375.011689961,
              "calibration_cpu_ns": 6109458.0,
              "calibration_gpu_ticks": 6109458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4539750.0,
              "compute_span_ns": 4539750.0,
              "command_buffer_ns": 4539749.992545694,
              "calibration_cpu_ns": 6035875.0,
              "calibration_gpu_ticks": 6035875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4471791.0,
              "compute_span_ns": 4471791.0,
              "command_buffer_ns": 4471749.998629093,
              "calibration_cpu_ns": 5898709.0,
              "calibration_gpu_ticks": 5898709.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 91250.0,
              "compute_span_ns": 91250.0,
              "command_buffer_ns": 91249.99633058906,
              "calibration_cpu_ns": 866791.0,
              "calibration_gpu_ticks": 866791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 71667.0,
              "compute_span_ns": 71667.0,
              "command_buffer_ns": 71666.669100523,
              "calibration_cpu_ns": 549334.0,
              "calibration_gpu_ticks": 549334.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 88791.0,
              "compute_span_ns": 88791.0,
              "command_buffer_ns": 88749.97729435563,
              "calibration_cpu_ns": 666458.0,
              "calibration_gpu_ticks": 666458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 72083.0,
              "compute_span_ns": 72083.0,
              "command_buffer_ns": 72041.70105978847,
              "calibration_cpu_ns": 496666.0,
              "calibration_gpu_ticks": 496666.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 79459.0,
              "compute_span_ns": 79459.0,
              "command_buffer_ns": 79499.96506795287,
              "calibration_cpu_ns": 830959.0,
              "calibration_gpu_ticks": 830959.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70334.0,
              "compute_span_ns": 70334.0,
              "command_buffer_ns": 70374.98289719224,
              "calibration_cpu_ns": 496292.0,
              "calibration_gpu_ticks": 496292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 73750.0,
              "compute_span_ns": 73750.0,
              "command_buffer_ns": 73749.9794922769,
              "calibration_cpu_ns": 627417.0,
              "calibration_gpu_ticks": 627417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 71792.0,
              "compute_span_ns": 71792.0,
              "command_buffer_ns": 71833.31763371825,
              "calibration_cpu_ns": 642500.0,
              "calibration_gpu_ticks": 642500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70333.0,
              "compute_span_ns": 70333.0,
              "command_buffer_ns": 70291.68773442507,
              "calibration_cpu_ns": 558000.0,
              "calibration_gpu_ticks": 558000.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            108.333328125,
            109.458328125,
            67.8671875,
            72.13215625,
            70.641265625,
            73.23828125,
            71.272125,
            70.93359375,
            69.871734375
          ],
          "compute_throughput_us_p50": 71.272125,
          "compute_throughput_us_p90": 108.55832812499999,
          "command_buffer_throughput_us": [
            108.33268243004568,
            109.4576828108984,
            67.86718768125866,
            72.1321612218162,
            70.64062538120197,
            73.23828049266012,
            71.27148455765564,
            70.93359363352647,
            69.87109372857958
          ],
          "command_buffer_throughput_us_p50": 71.27148455765564,
          "command_buffer_throughput_us_p90": 108.55768250621622,
          "compute_latency_us": [
            91.25,
            71.667,
            88.791,
            72.083,
            79.459,
            70.334,
            73.75,
            71.792,
            70.333
          ],
          "compute_latency_us_p50": 72.083,
          "compute_latency_us_p90": 89.2828,
          "command_buffer_latency_us": [
            91.24999633058906,
            71.666669100523,
            88.74997729435563,
            72.04170105978847,
            79.49996506795287,
            70.37498289719224,
            73.7499794922769,
            71.83331763371825,
            70.29168773442507
          ],
          "command_buffer_latency_us_p50": 72.04170105978847,
          "command_buffer_latency_us_p90": 89.24998110160232,
          "counter_control_throughput_ratio": 0.995695681111882,
          "counter_control_latency_ratio": 0.5528829844854234
        },
        "download_ms": 1.027042,
        "allocation_upload_ms": 19.1575,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.8339726848282112e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.8936424684089217,
      "device_compute_slowdown": 0.9867185979090702,
      "valid": true,
      "round": 0,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "rmsnorm_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.355708,
        "warmup_ms": 200.969333,
        "repetitions": 341,
        "throughput_us": [
          101.82673607038123,
          106.0294486803519,
          102.44085923753666,
          102.28274486803521,
          102.57685630498533,
          108.95246920821114,
          102.06365982404692,
          101.38526392961877,
          102.02981231671555
        ],
        "latency_us": [
          433.542,
          339.75,
          339.167,
          2345.2920000000004,
          678.375,
          451.042,
          427.416,
          319.75,
          317.125
        ],
        "throughput_us_p50": 102.28274486803521,
        "throughput_us_p90": 106.61405278592375,
        "latency_us_p50": 427.416,
        "latency_us_p90": 1011.7584000000004,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 6890249.962452799,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 6845000.0253506005,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 6682541.687041521,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4464041.6745096445,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4826583.317480981,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4576041.654217988,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4641416.657250375,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4460000.025574118,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4650666.669476777,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 310958.3049081266,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70500.01295283437,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 143708.31195265055,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70500.01295283437,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 192624.9824464321,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 76541.67711734772,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 337541.62723198533,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70291.68773442507,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 296291.6623800993,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              107.66015566332499,
              106.95312539610313,
              104.41471386002377,
              69.7506511642132,
              75.41536433564033,
              71.50065084715607,
              72.5221352695371,
              69.68750039959559,
              72.66666671057465
            ],
            "command_buffer_throughput_us_p50": 72.66666671057465,
            "command_buffer_throughput_us_p90": 107.0945314495475,
            "command_buffer_latency_us": [
              310.9583049081266,
              70.50001295283437,
              143.70831195265055,
              70.50001295283437,
              192.6249824464321,
              76.54167711734772,
              337.54162723198533,
              70.29168773442507,
              296.2916623800993
            ],
            "command_buffer_latency_us_p50": 143.70831195265055,
            "command_buffer_latency_us_p90": 316.27496937289834
          },
          "throughput": [
            {
              "compute_ns": 6881333.0,
              "compute_span_ns": 6881333.0,
              "command_buffer_ns": 6881291.687022895,
              "calibration_cpu_ns": 7996000.0,
              "calibration_gpu_ticks": 7996000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 6934458.0,
              "compute_span_ns": 6934458.0,
              "command_buffer_ns": 6934458.331670612,
              "calibration_cpu_ns": 7935625.0,
              "calibration_gpu_ticks": 7935625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 6199583.0,
              "compute_span_ns": 6199583.0,
              "command_buffer_ns": 6199583.294801414,
              "calibration_cpu_ns": 7573875.0,
              "calibration_gpu_ticks": 7573875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5341583.0,
              "compute_span_ns": 5341583.0,
              "command_buffer_ns": 5341583.339031786,
              "calibration_cpu_ns": 7033125.0,
              "calibration_gpu_ticks": 7033125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4737667.0,
              "compute_span_ns": 4737667.0,
              "command_buffer_ns": 4737666.69165343,
              "calibration_cpu_ns": 6176375.0,
              "calibration_gpu_ticks": 6176375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4722500.0,
              "compute_span_ns": 4722500.0,
              "command_buffer_ns": 4722499.987110496,
              "calibration_cpu_ns": 6077000.0,
              "calibration_gpu_ticks": 6077000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4552333.0,
              "compute_span_ns": 4552333.0,
              "command_buffer_ns": 4552291.647996753,
              "calibration_cpu_ns": 5980083.0,
              "calibration_gpu_ticks": 5980083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4585375.0,
              "compute_span_ns": 4585375.0,
              "command_buffer_ns": 4585375.019814819,
              "calibration_cpu_ns": 5953125.0,
              "calibration_gpu_ticks": 5953125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4732042.0,
              "compute_span_ns": 4732042.0,
              "command_buffer_ns": 4732083.35461095,
              "calibration_cpu_ns": 6114750.0,
              "calibration_gpu_ticks": 6114750.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 90833.0,
              "compute_span_ns": 90833.0,
              "command_buffer_ns": 90791.66920855641,
              "calibration_cpu_ns": 793125.0,
              "calibration_gpu_ticks": 793125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70875.0,
              "compute_span_ns": 70875.0,
              "command_buffer_ns": 70874.98670443892,
              "calibration_cpu_ns": 539875.0,
              "calibration_gpu_ticks": 539875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 90666.0,
              "compute_span_ns": 90666.0,
              "command_buffer_ns": 90625.02067536116,
              "calibration_cpu_ns": 1026125.0,
              "calibration_gpu_ticks": 1026125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 76125.0,
              "compute_span_ns": 76125.0,
              "command_buffer_ns": 76124.9684728682,
              "calibration_cpu_ns": 533792.0,
              "calibration_gpu_ticks": 533792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 85292.0,
              "compute_span_ns": 85292.0,
              "command_buffer_ns": 85291.68553650379,
              "calibration_cpu_ns": 814375.0,
              "calibration_gpu_ticks": 814375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70917.0,
              "compute_span_ns": 70917.0,
              "command_buffer_ns": 70958.34007486701,
              "calibration_cpu_ns": 495375.0,
              "calibration_gpu_ticks": 495375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 100625.0,
              "compute_span_ns": 100625.0,
              "command_buffer_ns": 100624.98040497303,
              "calibration_cpu_ns": 942792.0,
              "calibration_gpu_ticks": 942792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 71042.0,
              "compute_span_ns": 71042.0,
              "command_buffer_ns": 71083.31192284822,
              "calibration_cpu_ns": 548625.0,
              "calibration_gpu_ticks": 548625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 89875.0,
              "compute_span_ns": 89875.0,
              "command_buffer_ns": 89875.01496449113,
              "calibration_cpu_ns": 884916.0,
              "calibration_gpu_ticks": 884916.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            107.520828125,
            108.35090625,
            96.868484375,
            83.462234375,
            74.026046875,
            73.7890625,
            71.130203125,
            71.646484375,
            73.93815625
          ],
          "compute_throughput_us_p50": 74.026046875,
          "compute_throughput_us_p90": 107.68684375,
          "command_buffer_throughput_us": [
            107.52018260973273,
            108.35091143235331,
            96.8684889812721,
            83.46223967237165,
            74.02604205708485,
            73.7890622986015,
            71.12955699994927,
            71.64648468460655,
            73.93880241579609
          ],
          "command_buffer_throughput_us_p50": 74.02604205708485,
          "command_buffer_throughput_us_p90": 107.68632837425685,
          "compute_latency_us": [
            90.833,
            70.875,
            90.666,
            76.125,
            85.292,
            70.917,
            100.625,
            71.042,
            89.875
          ],
          "compute_latency_us_p50": 85.292,
          "compute_latency_us_p90": 92.7914,
          "command_buffer_latency_us": [
            90.79166920855641,
            70.87498670443892,
            90.62502067536116,
            76.1249684728682,
            85.29168553650379,
            70.95834007486701,
            100.62498040497303,
            71.08331192284822,
            89.87501496449113
          ],
          "command_buffer_latency_us_p50": 85.29168553650379,
          "command_buffer_latency_us_p90": 92.75833144783974,
          "counter_control_throughput_ratio": 1.0130691462364794,
          "counter_control_latency_ratio": 0.6306178080027867
        },
        "download_ms": 0.626875,
        "allocation_upload_ms": 1.227167,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.8339726848282112e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-uxrkr7y0/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "26111b6b600e52c50494ad3f783ef6fc3b02252d329d47ea39eb66749773d18e",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.078042,
        "capture_ms": 0.071,
        "compile_ms": 55.432375,
        "allocation_upload_ms": 5.77925,
        "cold_call_ms": 6.844584,
        "warmup_ms": 200.476167,
        "download_ms": 2.974417,
        "repetitions": 381,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          74.0955826772,
          81.4888451444,
          73.4520997375,
          80.687335958,
          73.6167979003,
          81.3713910761,
          75.1672152231,
          77.7517506562,
          73.8311469816
        ],
        "latency_us": [
          1658.375,
          2098.209,
          325.708,
          290.125,
          281.083,
          299.375,
          269.292,
          282.75,
          299.875
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 32,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 1024,
            "reduction_scalar_rounds": 48,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 32768,
              "global_write_bytes": 16384,
              "private_read_bytes": 49152,
              "private_write_bytes": 16384
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 128,
              "global_write_bytes": 64,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 1,
            "reduction_elements": 4096,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 80,
            "normalized_kernel_cost": 1280,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 8192,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3654791,
              "compute_span_ns": 3654791,
              "calibration_cpu_ns": 4909875,
              "calibration_gpu_ticks": 4909875,
              "compute_passes": 1,
              "command_buffer_ns": 3654874.91945,
              "command_buffers": 2
            },
            {
              "compute_ns": 3782458,
              "compute_span_ns": 3782458,
              "calibration_cpu_ns": 4804834,
              "calibration_gpu_ticks": 4804834,
              "compute_passes": 1,
              "command_buffer_ns": 3782499.98903,
              "command_buffers": 2
            },
            {
              "compute_ns": 3648166,
              "compute_span_ns": 3648166,
              "calibration_cpu_ns": 4901166,
              "calibration_gpu_ticks": 4901166,
              "compute_passes": 1,
              "command_buffer_ns": 3648208.33784,
              "command_buffers": 2
            },
            {
              "compute_ns": 3716916,
              "compute_span_ns": 3716916,
              "calibration_cpu_ns": 4801709,
              "calibration_gpu_ticks": 4801709,
              "compute_passes": 1,
              "command_buffer_ns": 3717000.07236,
              "command_buffers": 2
            },
            {
              "compute_ns": 3906500,
              "compute_span_ns": 3906500,
              "calibration_cpu_ns": 4864292,
              "calibration_gpu_ticks": 4864292,
              "compute_passes": 1,
              "command_buffer_ns": 3906625.03196,
              "command_buffers": 2
            },
            {
              "compute_ns": 3837083,
              "compute_span_ns": 3837083,
              "calibration_cpu_ns": 4850834,
              "calibration_gpu_ticks": 4850834,
              "compute_passes": 1,
              "command_buffer_ns": 3837208.32178,
              "command_buffers": 2
            },
            {
              "compute_ns": 3664000,
              "compute_span_ns": 3664000,
              "calibration_cpu_ns": 4864083,
              "calibration_gpu_ticks": 4864083,
              "compute_passes": 1,
              "command_buffer_ns": 3664125.04809,
              "command_buffers": 2
            },
            {
              "compute_ns": 3767375,
              "compute_span_ns": 3767375,
              "calibration_cpu_ns": 4962416,
              "calibration_gpu_ticks": 4962416,
              "compute_passes": 1,
              "command_buffer_ns": 3767458.25633,
              "command_buffers": 2
            },
            {
              "compute_ns": 3598292,
              "compute_span_ns": 3598292,
              "calibration_cpu_ns": 4801083,
              "calibration_gpu_ticks": 4801083,
              "compute_passes": 1,
              "command_buffer_ns": 3598458.30826,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 96166,
              "compute_span_ns": 96166,
              "calibration_cpu_ns": 790625,
              "calibration_gpu_ticks": 790625,
              "compute_passes": 1,
              "command_buffer_ns": 96125.0043474,
              "command_buffers": 2
            },
            {
              "compute_ns": 54250,
              "compute_span_ns": 54250,
              "calibration_cpu_ns": 566500,
              "calibration_gpu_ticks": 566500,
              "compute_passes": 1,
              "command_buffer_ns": 54333.3007954,
              "command_buffers": 2
            },
            {
              "compute_ns": 64334,
              "compute_span_ns": 64334,
              "calibration_cpu_ns": 490834,
              "calibration_gpu_ticks": 490834,
              "compute_passes": 1,
              "command_buffer_ns": 64374.9954179,
              "command_buffers": 2
            },
            {
              "compute_ns": 215875,
              "compute_span_ns": 215875,
              "calibration_cpu_ns": 1501250,
              "calibration_gpu_ticks": 1501250,
              "compute_passes": 1,
              "command_buffer_ns": 215916.661546,
              "command_buffers": 2
            },
            {
              "compute_ns": 271584,
              "compute_span_ns": 271584,
              "calibration_cpu_ns": 1624666,
              "calibration_gpu_ticks": 1624666,
              "compute_passes": 1,
              "command_buffer_ns": 271875.003818,
              "command_buffers": 2
            },
            {
              "compute_ns": 88541,
              "compute_span_ns": 88541,
              "calibration_cpu_ns": 738375,
              "calibration_gpu_ticks": 738375,
              "compute_passes": 1,
              "command_buffer_ns": 88541.6520759,
              "command_buffers": 2
            },
            {
              "compute_ns": 233042,
              "compute_span_ns": 233042,
              "calibration_cpu_ns": 1723125,
              "calibration_gpu_ticks": 1723125,
              "compute_passes": 1,
              "command_buffer_ns": 233124.999795,
              "command_buffers": 2
            },
            {
              "compute_ns": 96291,
              "compute_span_ns": 96291,
              "calibration_cpu_ns": 880750,
              "calibration_gpu_ticks": 880750,
              "compute_passes": 1,
              "command_buffer_ns": 96333.3295658,
              "command_buffers": 2
            },
            {
              "compute_ns": 195042,
              "compute_span_ns": 195042,
              "calibration_cpu_ns": 1668875,
              "calibration_gpu_ticks": 1668875,
              "compute_passes": 1,
              "command_buffer_ns": 195249.973331,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 3940916.62019,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3860249.99898,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3739624.9827,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3689083.32894,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3737125.02187,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3667416.69131,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3842208.30165,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3744375.01887,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3719166.67791,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 286291.586235,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 254583.253991,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 99000.0553429,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 95583.3820626,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 53541.676607,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 53583.3532922,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 53291.6747034,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54291.6823179,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54000.0619367,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              61.576822190468754,
              60.316406234062505,
              58.4316403546875,
              57.6419270146875,
              58.392578466718746,
              57.303385801718754,
              60.034504713281244,
              58.505859669843744,
              58.11197934234375
            ],
            "command_buffer_throughput_us_p50": 58.4316403546875,
            "command_buffer_throughput_us_p90": 60.568489425343756,
            "command_buffer_latency_us": [
              286.291586235,
              254.583253991,
              99.0000553429,
              95.5833820626,
              53.541676607,
              53.5833532922,
              53.2916747034,
              54.2916823179,
              54.00006193670001
            ],
            "command_buffer_latency_us_p50": 54.2916823179,
            "command_buffer_latency_us_p90": 260.9249204398
          },
          "compute_throughput_us": [
            57.106109375,
            59.10090625,
            57.00259375,
            58.0768125,
            61.0390625,
            59.954421875,
            57.25,
            58.865234375,
            56.2233125
          ],
          "compute_throughput_us_p50": 58.0768125,
          "compute_throughput_us_p90": 60.171350000000004,
          "command_buffer_throughput_us": [
            57.10742061640625,
            59.101562328593744,
            57.003255278750004,
            58.078126130625,
            61.041016124375,
            59.9563800278125,
            57.251953876406255,
            58.86653525515624,
            56.225911066562496
          ],
          "command_buffer_throughput_us_p50": 58.078126130625,
          "command_buffer_throughput_us_p90": 60.173307247125,
          "compute_latency_us": [
            96.166,
            54.25,
            64.334,
            215.875,
            271.584,
            88.541,
            233.042,
            96.291,
            195.042
          ],
          "compute_latency_us_p50": 96.291,
          "compute_latency_us_p90": 240.7504,
          "command_buffer_latency_us": [
            96.1250043474,
            54.3333007954,
            64.3749954179,
            215.916661546,
            271.87500381800004,
            88.5416520759,
            233.124999795,
            96.3333295658,
            195.24997333099998
          ],
          "command_buffer_latency_us_p50": 96.3333295658,
          "command_buffer_latency_us_p90": 240.8750005996,
          "counter_control_throughput_ratio": 0.9798588148512287,
          "counter_control_latency_ratio": 1.7743662648309362
        },
        "correctness": {
          "max_abs_error": 1.5832164268481108e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 75.1672152231,
        "throughput_us_p90": 81.39488188976,
        "latency_us_p50": 299.375,
        "latency_us_p90": 1746.3418000000001
      },
      "slowdown": 0.7348963436607069,
      "device_compute_slowdown": 0.7845456424016293,
      "valid": true,
      "round": 0,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "layernorm_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-2xjrlcfu/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "90a0b6dbe0fbfeffc09b3a8db9fc5856f9e2631dc2ce0e6d9ee991918165a6cf",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.942208,
        "capture_ms": 0.084125,
        "compile_ms": 65.268083,
        "allocation_upload_ms": 3.271375,
        "cold_call_ms": 2.172375,
        "warmup_ms": 200.039625,
        "download_ms": 0.301209,
        "repetitions": 3532,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          6.61332106455,
          5.88396545866,
          6.48290628539,
          7.10185447339,
          6.63008437146,
          6.27242582106,
          5.9587698188,
          6.62937684032,
          6.48703510759
        ],
        "latency_us": [
          309.292,
          454.458,
          209.25,
          200.542,
          192.167,
          200.334,
          288.292,
          205.125,
          201.291
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 37,
            "reduction_scalar_rounds": 40,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 18444,
              "global_write_bytes": 6148,
              "private_read_bytes": 36888,
              "private_write_bytes": 12296
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 32,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 2,
            "reduction_elements": 3074,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 88,
            "normalized_kernel_cost": 88,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 4611,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 313917,
              "compute_span_ns": 313917,
              "calibration_cpu_ns": 785750,
              "calibration_gpu_ticks": 785750,
              "compute_passes": 1,
              "command_buffer_ns": 313958.327752,
              "command_buffers": 2
            },
            {
              "compute_ns": 683166,
              "compute_span_ns": 683166,
              "calibration_cpu_ns": 2196584,
              "calibration_gpu_ticks": 2196584,
              "compute_passes": 1,
              "command_buffer_ns": 683249.963913,
              "command_buffers": 2
            },
            {
              "compute_ns": 726750,
              "compute_span_ns": 726750,
              "calibration_cpu_ns": 2359958,
              "calibration_gpu_ticks": 2359958,
              "compute_passes": 1,
              "command_buffer_ns": 726874.975953,
              "command_buffers": 2
            },
            {
              "compute_ns": 295375,
              "compute_span_ns": 295375,
              "calibration_cpu_ns": 861708,
              "calibration_gpu_ticks": 861708,
              "compute_passes": 1,
              "command_buffer_ns": 295416.626614,
              "command_buffers": 2
            },
            {
              "compute_ns": 292333,
              "compute_span_ns": 292333,
              "calibration_cpu_ns": 733917,
              "calibration_gpu_ticks": 733917,
              "compute_passes": 1,
              "command_buffer_ns": 292333.366815,
              "command_buffers": 2
            },
            {
              "compute_ns": 265375,
              "compute_span_ns": 265375,
              "calibration_cpu_ns": 1006500,
              "calibration_gpu_ticks": 1006500,
              "compute_passes": 1,
              "command_buffer_ns": 265499.926172,
              "command_buffers": 2
            },
            {
              "compute_ns": 279708,
              "compute_span_ns": 279708,
              "calibration_cpu_ns": 715916,
              "calibration_gpu_ticks": 715916,
              "compute_passes": 1,
              "command_buffer_ns": 279708.357994,
              "command_buffers": 2
            },
            {
              "compute_ns": 711959,
              "compute_span_ns": 711959,
              "calibration_cpu_ns": 2252083,
              "calibration_gpu_ticks": 2252083,
              "compute_passes": 1,
              "command_buffer_ns": 712041.684892,
              "command_buffers": 2
            },
            {
              "compute_ns": 776958,
              "compute_span_ns": 776958,
              "calibration_cpu_ns": 2311125,
              "calibration_gpu_ticks": 2311125,
              "compute_passes": 1,
              "command_buffer_ns": 777041.597757,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 6750,
              "compute_span_ns": 6750,
              "calibration_cpu_ns": 494459,
              "calibration_gpu_ticks": 494459,
              "compute_passes": 1,
              "command_buffer_ns": 6874.96503815,
              "command_buffers": 2
            },
            {
              "compute_ns": 7250,
              "compute_span_ns": 7250,
              "calibration_cpu_ns": 402375,
              "calibration_gpu_ticks": 402375,
              "compute_passes": 1,
              "command_buffer_ns": 7333.29216018,
              "command_buffers": 2
            },
            {
              "compute_ns": 6791,
              "compute_span_ns": 6791,
              "calibration_cpu_ns": 398416,
              "calibration_gpu_ticks": 398416,
              "compute_passes": 1,
              "command_buffer_ns": 6791.66987538,
              "command_buffers": 2
            },
            {
              "compute_ns": 7208,
              "compute_span_ns": 7208,
              "calibration_cpu_ns": 699750,
              "calibration_gpu_ticks": 699750,
              "compute_passes": 1,
              "command_buffer_ns": 7166.70183465,
              "command_buffers": 2
            },
            {
              "compute_ns": 6917,
              "compute_span_ns": 6917,
              "calibration_cpu_ns": 376000,
              "calibration_gpu_ticks": 376000,
              "compute_passes": 1,
              "command_buffer_ns": 7041.72998667,
              "command_buffers": 2
            },
            {
              "compute_ns": 7250,
              "compute_span_ns": 7250,
              "calibration_cpu_ns": 476583,
              "calibration_gpu_ticks": 476583,
              "compute_passes": 1,
              "command_buffer_ns": 7249.99699742,
              "command_buffers": 2
            },
            {
              "compute_ns": 7292,
              "compute_span_ns": 7292,
              "calibration_cpu_ns": 821500,
              "calibration_gpu_ticks": 821500,
              "compute_passes": 1,
              "command_buffer_ns": 7333.29216018,
              "command_buffers": 2
            },
            {
              "compute_ns": 6750,
              "compute_span_ns": 6750,
              "calibration_cpu_ns": 360291,
              "calibration_gpu_ticks": 360291,
              "compute_passes": 1,
              "command_buffer_ns": 6833.28835294,
              "command_buffers": 2
            },
            {
              "compute_ns": 8625,
              "compute_span_ns": 8625,
              "calibration_cpu_ns": 758500,
              "calibration_gpu_ticks": 758500,
              "compute_passes": 1,
              "command_buffer_ns": 8666.71325639,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 682083.365973,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 295875.011943,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 289416.639134,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 266916.642431,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 549916.643649,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 658416.654915,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 306500.005536,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 293333.316222,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 266333.285253,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 6958.37661624,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 9916.72277451,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7666.7056419,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 44958.316721,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7708.32411945,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7249.99699742,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76874.9741837,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 9458.33744481,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 42208.3539888,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              10.657552593328123,
              4.623047061609375,
              4.52213498646875,
              4.170572537984375,
              8.592447557015625,
              10.287760233046875,
              4.7890625865,
              4.58333306596875,
              4.161457582078125
            ],
            "command_buffer_throughput_us_p50": 4.623047061609375,
            "command_buffer_throughput_us_p90": 10.361718705103124,
            "command_buffer_latency_us": [
              6.95837661624,
              9.916722774510001,
              7.6667056419,
              44.958316721,
              7.708324119449999,
              7.24999699742,
              76.8749741837,
              9.45833744481,
              42.2083539888
            ],
            "command_buffer_latency_us_p50": 9.45833744481,
            "command_buffer_latency_us_p90": 51.34164821354001
          },
          "compute_throughput_us": [
            4.904953125,
            10.67446875,
            11.35546875,
            4.615234375,
            4.567703125,
            4.146484375,
            4.3704375,
            11.124359375,
            12.13996875
          ],
          "compute_throughput_us_p50": 4.904953125,
          "compute_throughput_us_p90": 11.51236875,
          "command_buffer_throughput_us": [
            4.905598871125,
            10.675780686140625,
            11.357421499265625,
            4.61588479084375,
            4.5677088564843755,
            4.1484363464375,
            4.370443093656251,
            11.1256513264375,
            12.141274964953125
          ],
          "command_buffer_throughput_us_p50": 4.905598871125,
          "command_buffer_throughput_us_p90": 11.514192192403124,
          "compute_latency_us": [
            6.75,
            7.25,
            6.791,
            7.208,
            6.917,
            7.25,
            7.292,
            6.75,
            8.625
          ],
          "compute_latency_us_p50": 7.208,
          "compute_latency_us_p90": 7.5586,
          "command_buffer_latency_us": [
            6.87496503815,
            7.33329216018,
            6.79166987538,
            7.1667018346499995,
            7.041729986669999,
            7.24999699742,
            7.33329216018,
            6.8332883529399995,
            8.666713256389999
          ],
          "command_buffer_latency_us_p50": 7.1667018346499995,
          "command_buffer_latency_us_p90": 7.599976379422,
          "counter_control_throughput_ratio": 1.1067748489694398,
          "counter_control_latency_ratio": 0.7394874624336111
        },
        "correctness": {
          "max_abs_error": 2.43199801630567e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 6.48703510759,
        "throughput_us_p90": 6.724438391846,
        "latency_us_p50": 205.125,
        "latency_us_p90": 338.3252
      },
      "torch": {
        "cold_call_ms": 11.251292,
        "warmup_ms": 200.164042,
        "repetitions": 1956,
        "throughput_us": [
          14.733895705521473,
          14.84270654396728,
          14.53007873210634,
          16.527053680981595,
          15.267574130879346,
          14.012461656441717,
          15.31639826175869,
          13.991244887525562,
          14.254367075664621
        ],
        "latency_us": [
          1366.792,
          389.625,
          418.792,
          514.75,
          244.834,
          218.95800000000003,
          217.209,
          217.917,
          291.25
        ],
        "throughput_us_p50": 14.733895705521473,
        "throughput_us_p90": 15.558529345603272,
        "latency_us_p50": 291.25,
        "latency_us_p90": 685.1584000000001,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 646833.3303928375,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 474250.0022985041,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 559958.3382718265,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 471124.9493993819,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 650416.6522063315,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 472624.96082112193,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 661416.6777580976,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 469749.96803328395,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 609958.3697505295,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 78875.04762038589,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 146874.9833293259,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10583.316907286644,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 116791.63435474038,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 11124.997399747372,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12374.94871020317,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10291.696526110172,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 9874.98788163066,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10000.017937272787,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              10.106770787388086,
              7.410156285914127,
              8.74934903549729,
              7.361327334365342,
              10.16276019072393,
              7.38476501283003,
              10.334635589970276,
              7.339843250520062,
              9.530599527352024
            ],
            "command_buffer_throughput_us_p50": 8.74934903549729,
            "command_buffer_throughput_us_p90": 10.197135270573199,
            "command_buffer_latency_us": [
              78.87504762038589,
              146.87498332932591,
              10.583316907286644,
              116.79163435474038,
              11.124997399747372,
              12.37494871020317,
              10.291696526110172,
              9.874987881630659,
              10.000017937272787
            ],
            "command_buffer_latency_us_p50": 11.124997399747372,
            "command_buffer_latency_us_p90": 122.80830414965749
          },
          "throughput": [
            {
              "compute_ns": 495875.0,
              "compute_span_ns": 495875.0,
              "command_buffer_ns": 495874.96323511004,
              "calibration_cpu_ns": 1356125.0,
              "calibration_gpu_ticks": 1356125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 687541.0,
              "compute_span_ns": 687541.0,
              "command_buffer_ns": 687499.9962747097,
              "calibration_cpu_ns": 2245833.0,
              "calibration_gpu_ticks": 2245833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 485166.0,
              "compute_span_ns": 485166.0,
              "command_buffer_ns": 485124.99779462814,
              "calibration_cpu_ns": 1261000.0,
              "calibration_gpu_ticks": 1261000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 546834.0,
              "compute_span_ns": 546834.0,
              "command_buffer_ns": 546875.0023283064,
              "calibration_cpu_ns": 1941791.0,
              "calibration_gpu_ticks": 1941791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 470958.0,
              "compute_span_ns": 470958.0,
              "command_buffer_ns": 470916.6823886335,
              "calibration_cpu_ns": 1237583.0,
              "calibration_gpu_ticks": 1237583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 656125.0,
              "compute_span_ns": 656125.0,
              "command_buffer_ns": 656125.0193044543,
              "calibration_cpu_ns": 2157625.0,
              "calibration_gpu_ticks": 2157625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 468000.0,
              "compute_span_ns": 468000.0,
              "command_buffer_ns": 468000.01291558146,
              "calibration_cpu_ns": 1219125.0,
              "calibration_gpu_ticks": 1219125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 717625.0,
              "compute_span_ns": 717625.0,
              "command_buffer_ns": 717625.0219345093,
              "calibration_cpu_ns": 2272542.0,
              "calibration_gpu_ticks": 2272542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 481292.0,
              "compute_span_ns": 481292.0,
              "command_buffer_ns": 481333.3507627249,
              "calibration_cpu_ns": 1315750.0,
              "calibration_gpu_ticks": 1315750.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 10958.0,
              "compute_span_ns": 10958.0,
              "command_buffer_ns": 10916.672181338072,
              "calibration_cpu_ns": 902125.0,
              "calibration_gpu_ticks": 902125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10292.0,
              "compute_span_ns": 10292.0,
              "command_buffer_ns": 10333.373211324215,
              "calibration_cpu_ns": 777208.0,
              "calibration_gpu_ticks": 777208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 9625.0,
              "compute_span_ns": 9625.0,
              "command_buffer_ns": 9624.985978007317,
              "calibration_cpu_ns": 515042.0,
              "calibration_gpu_ticks": 515042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10125.0,
              "compute_span_ns": 10125.0,
              "command_buffer_ns": 10124.989785254002,
              "calibration_cpu_ns": 446625.0,
              "calibration_gpu_ticks": 446625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10167.0,
              "compute_span_ns": 10167.0,
              "command_buffer_ns": 10208.343155682087,
              "calibration_cpu_ns": 578000.0,
              "calibration_gpu_ticks": 578000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 31542.0,
              "compute_span_ns": 31542.0,
              "command_buffer_ns": 31541.625503450632,
              "calibration_cpu_ns": 1334334.0,
              "calibration_gpu_ticks": 1334334.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 75250.0,
              "compute_span_ns": 75250.0,
              "command_buffer_ns": 75249.99091401696,
              "calibration_cpu_ns": 1594542.0,
              "calibration_gpu_ticks": 1594542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10792.0,
              "compute_span_ns": 10792.0,
              "command_buffer_ns": 10833.318810909986,
              "calibration_cpu_ns": 806500.0,
              "calibration_gpu_ticks": 806500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 115875.0,
              "compute_span_ns": 115875.0,
              "command_buffer_ns": 115874.9801106751,
              "calibration_cpu_ns": 1245166.0,
              "calibration_gpu_ticks": 1245166.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            7.748046875,
            10.742828125,
            7.58071875,
            8.54428125,
            7.35871875,
            10.251953125,
            7.3125,
            11.212890625,
            7.5201875
          ],
          "compute_throughput_us_p50": 7.748046875,
          "compute_throughput_us_p90": 10.836840625,
          "command_buffer_throughput_us": [
            7.7480463005485944,
            10.742187441792339,
            7.580078090541065,
            8.544921911379788,
            7.358073162322398,
            10.251953426632099,
            7.31250020180596,
            11.212890967726707,
            7.520833605667576
          ],
          "command_buffer_throughput_us_p50": 7.7480463005485944,
          "command_buffer_throughput_us_p90": 10.836328146979213,
          "compute_latency_us": [
            10.958,
            10.292,
            9.625,
            10.125,
            10.167,
            31.542,
            75.25,
            10.792,
            115.875
          ],
          "compute_latency_us_p50": 10.792,
          "compute_latency_us_p90": 83.375,
          "command_buffer_latency_us": [
            10.916672181338072,
            10.333373211324215,
            9.624985978007317,
            10.124989785254002,
            10.208343155682087,
            31.541625503450632,
            75.24999091401696,
            10.833318810909986,
            115.8749801106751
          ],
          "command_buffer_latency_us_p50": 10.833318810909986,
          "command_buffer_latency_us_p90": 83.37498875334859,
          "counter_control_throughput_ratio": 0.8663590925207881,
          "counter_control_latency_ratio": 0.9176040936345656
        },
        "download_ms": 0.767542,
        "allocation_upload_ms": 1.000416,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.0790672678193687e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.4402796950136553,
      "device_compute_slowdown": 0.6330567179228636,
      "valid": true,
      "round": 0,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "layernorm_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.330917,
        "warmup_ms": 200.2135,
        "repetitions": 1747,
        "throughput_us": [
          14.621708643388667,
          14.54531597023469,
          14.768603319977103,
          15.282889524899828,
          17.763189467658844,
          15.212841442472811,
          16.713199198626217,
          14.901378362907842,
          14.994132799084145
        ],
        "latency_us": [
          348.375,
          242.667,
          258.125,
          212.959,
          256.0,
          289.291,
          225.667,
          204.5,
          287.25
        ],
        "throughput_us_p50": 14.994132799084145,
        "throughput_us_p90": 16.923197252432743,
        "latency_us_p50": 256.0,
        "latency_us_p90": 301.1078,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 708750.041667372,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 477750.0289492309,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 643291.6852645576,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 468708.34194123745,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 572499.9937228858,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 767749.9670535326,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 495625.0195391476,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 610875.0239945948,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 508875.00401586294,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 17625.046893954277,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10041.69462248683,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 79500.02327561378,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10166.666470468044,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 13291.661161929369,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10250.01984089613,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10000.017937272787,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 9499.955922365189,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 25208.34095776081,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              11.074219401052687,
              7.464844202331733,
              10.051432582258713,
              7.323567842831835,
              8.945312401920091,
              11.996093235211447,
              7.7441409302991815,
              9.544922249915544,
              7.9511719377478585
            ],
            "command_buffer_throughput_us_p50": 8.945312401920091,
            "command_buffer_throughput_us_p90": 11.25859416788444,
            "command_buffer_latency_us": [
              17.625046893954277,
              10.04169462248683,
              79.50002327561378,
              10.166666470468044,
              13.291661161929369,
              10.25001984089613,
              10.000017937272787,
              9.499955922365189,
              25.20834095776081
            ],
            "command_buffer_latency_us_p50": 10.25001984089613,
            "command_buffer_latency_us_p90": 36.06667742133141
          },
          "throughput": [
            {
              "compute_ns": 494541.0,
              "compute_span_ns": 494541.0,
              "command_buffer_ns": 494499.9818690121,
              "calibration_cpu_ns": 1374250.0,
              "calibration_gpu_ticks": 1374250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 578708.0,
              "compute_span_ns": 578708.0,
              "command_buffer_ns": 578708.3064205945,
              "calibration_cpu_ns": 2381292.0,
              "calibration_gpu_ticks": 2381292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 487125.0,
              "compute_span_ns": 487125.0,
              "command_buffer_ns": 487125.0130236149,
              "calibration_cpu_ns": 1220833.0,
              "calibration_gpu_ticks": 1220833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 656667.0,
              "compute_span_ns": 656667.0,
              "command_buffer_ns": 656666.699796915,
              "calibration_cpu_ns": 2419917.0,
              "calibration_gpu_ticks": 2419917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 476167.0,
              "compute_span_ns": 476167.0,
              "command_buffer_ns": 476208.3408422768,
              "calibration_cpu_ns": 1172167.0,
              "calibration_gpu_ticks": 1172167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 524917.0,
              "compute_span_ns": 524917.0,
              "command_buffer_ns": 524958.3045952022,
              "calibration_cpu_ns": 1324375.0,
              "calibration_gpu_ticks": 1324375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 566250.0,
              "compute_span_ns": 566250.0,
              "command_buffer_ns": 566250.0043399632,
              "calibration_cpu_ns": 2382083.0,
              "calibration_gpu_ticks": 2382083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 493500.0,
              "compute_span_ns": 493500.0,
              "command_buffer_ns": 493499.97425451875,
              "calibration_cpu_ns": 1262458.0,
              "calibration_gpu_ticks": 1262458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 475208.0,
              "compute_span_ns": 475208.0,
              "command_buffer_ns": 475166.6565425694,
              "calibration_cpu_ns": 1148959.0,
              "calibration_gpu_ticks": 1148959.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 11125.0,
              "compute_span_ns": 11125.0,
              "command_buffer_ns": 11124.997399747372,
              "calibration_cpu_ns": 750167.0,
              "calibration_gpu_ticks": 750167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 9875.0,
              "compute_span_ns": 9875.0,
              "command_buffer_ns": 9875.046089291573,
              "calibration_cpu_ns": 404917.0,
              "calibration_gpu_ticks": 404917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10833.0,
              "compute_span_ns": 10833.0,
              "command_buffer_ns": 10833.3770185709,
              "calibration_cpu_ns": 859166.0,
              "calibration_gpu_ticks": 859166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 9917.0,
              "compute_span_ns": 9917.0,
              "command_buffer_ns": 9916.664566844702,
              "calibration_cpu_ns": 559542.0,
              "calibration_gpu_ticks": 559542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 11334.0,
              "compute_span_ns": 11334.0,
              "command_buffer_ns": 11374.999303370714,
              "calibration_cpu_ns": 769625.0,
              "calibration_gpu_ticks": 769625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 9584.0,
              "compute_span_ns": 9584.0,
              "command_buffer_ns": 9624.985978007317,
              "calibration_cpu_ns": 392666.0,
              "calibration_gpu_ticks": 392666.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 30375.0,
              "compute_span_ns": 30375.0,
              "command_buffer_ns": 30374.969355762005,
              "calibration_cpu_ns": 1551833.0,
              "calibration_gpu_ticks": 1551833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10708.0,
              "compute_span_ns": 10708.0,
              "command_buffer_ns": 10708.346962928772,
              "calibration_cpu_ns": 837708.0,
              "calibration_gpu_ticks": 837708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10084.0,
              "compute_span_ns": 10084.0,
              "command_buffer_ns": 10124.989785254002,
              "calibration_cpu_ns": 392375.0,
              "calibration_gpu_ticks": 392375.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            7.727203125,
            9.0423125,
            7.611328125,
            10.260421875,
            7.440109375,
            8.201828125,
            8.84765625,
            7.7109375,
            7.425125
          ],
          "compute_throughput_us_p50": 7.727203125,
          "compute_throughput_us_p90": 9.285934375,
          "command_buffer_throughput_us": [
            7.726562216703314,
            9.042317287821788,
            7.6113283284939826,
            10.260417184326798,
            7.440755325660575,
            8.202473509300034,
            8.847656317811925,
            7.710937097726855,
            7.424479008477647
          ],
          "command_buffer_throughput_us_p50": 7.726562216703314,
          "command_buffer_throughput_us_p90": 9.28593726712279,
          "compute_latency_us": [
            11.125,
            9.875,
            10.833,
            9.917,
            11.334,
            9.584,
            30.375,
            10.708,
            10.084
          ],
          "compute_latency_us_p50": 10.708,
          "compute_latency_us_p90": 15.142200000000003,
          "command_buffer_latency_us": [
            11.124997399747372,
            9.875046089291573,
            10.8333770185709,
            9.916664566844702,
            11.374999303370714,
            9.624985978007317,
            30.374969355762005,
            10.708346962928772,
            10.124989785254002
          ],
          "command_buffer_latency_us_p50": 10.708346962928772,
          "command_buffer_latency_us_p90": 15.174993313848976,
          "counter_control_throughput_ratio": 0.8318049712901512,
          "counter_control_latency_ratio": 0.9390212045839154
        },
        "download_ms": 0.334375,
        "allocation_upload_ms": 1.717792,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.0790672678193687e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-a0k85gr_/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "53243cd32bb4620fa91e934e3d500807c417fa70fc3ac1977dac07575a48be10",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.180833,
        "capture_ms": 0.083125,
        "compile_ms": 73.610292,
        "allocation_upload_ms": 1.534625,
        "cold_call_ms": 1.92825,
        "warmup_ms": 200.118625,
        "download_ms": 0.321292,
        "repetitions": 2716,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          6.74679381443,
          6.7874326215,
          7.83282731959,
          6.75455633284,
          6.80605670103,
          7.75024558174,
          6.78140353461,
          6.75541568483,
          7.64370103093
        ],
        "latency_us": [
          542.667,
          217.416,
          213.584,
          221.583,
          290.083,
          209.125,
          212.292,
          213.416,
          208.541
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 19,
            "reduction_scalar_rounds": 40,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 18444,
              "global_write_bytes": 6148,
              "private_read_bytes": 36888,
              "private_write_bytes": 12296
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 32,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 2,
            "reduction_elements": 3074,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 80,
            "normalized_kernel_cost": 80,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 4611,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 364042,
              "compute_span_ns": 364042,
              "calibration_cpu_ns": 2017417,
              "calibration_gpu_ticks": 2017417,
              "compute_passes": 1,
              "command_buffer_ns": 364125.007764,
              "command_buffers": 2
            },
            {
              "compute_ns": 294500,
              "compute_span_ns": 294500,
              "calibration_cpu_ns": 797500,
              "calibration_gpu_ticks": 797500,
              "compute_passes": 1,
              "command_buffer_ns": 294625.002425,
              "command_buffers": 2
            },
            {
              "compute_ns": 311667,
              "compute_span_ns": 311667,
              "calibration_cpu_ns": 743833,
              "calibration_gpu_ticks": 743833,
              "compute_passes": 1,
              "command_buffer_ns": 311791.722197,
              "command_buffers": 2
            },
            {
              "compute_ns": 295958,
              "compute_span_ns": 295958,
              "calibration_cpu_ns": 695959,
              "calibration_gpu_ticks": 695959,
              "compute_passes": 1,
              "command_buffer_ns": 295958.365314,
              "command_buffers": 2
            },
            {
              "compute_ns": 295208,
              "compute_span_ns": 295208,
              "calibration_cpu_ns": 785292,
              "calibration_gpu_ticks": 785292,
              "compute_passes": 1,
              "command_buffer_ns": 295208.359603,
              "command_buffers": 2
            },
            {
              "compute_ns": 761209,
              "compute_span_ns": 761209,
              "calibration_cpu_ns": 2224417,
              "calibration_gpu_ticks": 2224417,
              "compute_passes": 1,
              "command_buffer_ns": 761249.975767,
              "command_buffers": 2
            },
            {
              "compute_ns": 733375,
              "compute_span_ns": 733375,
              "calibration_cpu_ns": 2081750,
              "calibration_gpu_ticks": 2081750,
              "compute_passes": 1,
              "command_buffer_ns": 733458.32061,
              "command_buffers": 2
            },
            {
              "compute_ns": 289334,
              "compute_span_ns": 289334,
              "calibration_cpu_ns": 707708,
              "calibration_gpu_ticks": 707708,
              "compute_passes": 1,
              "command_buffer_ns": 289416.697342,
              "command_buffers": 2
            },
            {
              "compute_ns": 294625,
              "compute_span_ns": 294625,
              "calibration_cpu_ns": 802292,
              "calibration_gpu_ticks": 802292,
              "compute_passes": 1,
              "command_buffer_ns": 294625.002425,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 14083,
              "compute_span_ns": 14083,
              "calibration_cpu_ns": 672875,
              "calibration_gpu_ticks": 672875,
              "compute_passes": 1,
              "command_buffer_ns": 14083.2853504,
              "command_buffers": 2
            },
            {
              "compute_ns": 7042,
              "compute_span_ns": 7042,
              "calibration_cpu_ns": 439584,
              "calibration_gpu_ticks": 439584,
              "compute_passes": 1,
              "command_buffer_ns": 7083.29025656,
              "command_buffers": 2
            },
            {
              "compute_ns": 8500,
              "compute_span_ns": 8500,
              "calibration_cpu_ns": 833209,
              "calibration_gpu_ticks": 833209,
              "compute_passes": 1,
              "command_buffer_ns": 8583.35988596,
              "command_buffers": 2
            },
            {
              "compute_ns": 9833,
              "compute_span_ns": 9833,
              "calibration_cpu_ns": 447334,
              "calibration_gpu_ticks": 447334,
              "compute_passes": 1,
              "command_buffer_ns": 9874.98788163,
              "command_buffers": 2
            },
            {
              "compute_ns": 8666,
              "compute_span_ns": 8666,
              "calibration_cpu_ns": 863166,
              "calibration_gpu_ticks": 863166,
              "compute_passes": 1,
              "command_buffer_ns": 8666.65504873,
              "command_buffers": 2
            },
            {
              "compute_ns": 7042,
              "compute_span_ns": 7042,
              "calibration_cpu_ns": 484084,
              "calibration_gpu_ticks": 484084,
              "compute_passes": 1,
              "command_buffer_ns": 7083.34846422,
              "command_buffers": 2
            },
            {
              "compute_ns": 7875,
              "compute_span_ns": 7875,
              "calibration_cpu_ns": 631458,
              "calibration_gpu_ticks": 631458,
              "compute_passes": 1,
              "command_buffer_ns": 7875.0308603,
              "command_buffers": 2
            },
            {
              "compute_ns": 7625,
              "compute_span_ns": 7625,
              "calibration_cpu_ns": 484833,
              "calibration_gpu_ticks": 484833,
              "compute_passes": 1,
              "command_buffer_ns": 7708.32411945,
              "command_buffers": 2
            },
            {
              "compute_ns": 9209,
              "compute_span_ns": 9209,
              "calibration_cpu_ns": 825417,
              "calibration_gpu_ticks": 825417,
              "compute_passes": 1,
              "command_buffer_ns": 9250.0122264,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 671000.045259,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 296250.043903,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 703624.973539,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 687208.375894,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 295249.97808,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 295749.981888,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 319875.020068,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 312999.996822,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 684999.977238,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 87624.9978319,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7624.97074902,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 33083.2554027,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7583.35227147,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 114208.320156,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7625.02895668,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 59708.3126195,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7583.35227147,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 52166.6370332,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              10.484375707171875,
              4.628906935984375,
              10.994140211546876,
              10.73763087334375,
              4.613280907499999,
              4.621093467,
              4.9980471885625,
              4.89062495034375,
              10.70312464434375
            ],
            "command_buffer_throughput_us_p50": 4.9980471885625,
            "command_buffer_throughput_us_p90": 10.788932740984375,
            "command_buffer_latency_us": [
              87.62499783189999,
              7.62497074902,
              33.0832554027,
              7.58335227147,
              114.208320156,
              7.6250289566800005,
              59.7083126195,
              7.58335227147,
              52.1666370332
            ],
            "command_buffer_latency_us_p50": 33.0832554027,
            "command_buffer_latency_us_p90": 92.94166229672
          },
          "compute_throughput_us": [
            5.68815625,
            4.6015625,
            4.869796875,
            4.62434375,
            4.612625,
            11.893890625,
            11.458984375,
            4.52084375,
            4.603515625
          ],
          "compute_throughput_us_p50": 4.62434375,
          "compute_throughput_us_p90": 11.545965625,
          "command_buffer_throughput_us": [
            5.6894532463125,
            4.603515662890625,
            4.871745659328125,
            4.62434945803125,
            4.612630618796874,
            11.894530871359375,
            11.460286259531252,
            4.52213589596875,
            4.603515662890625
          ],
          "command_buffer_throughput_us_p50": 4.62434945803125,
          "command_buffer_throughput_us_p90": 11.547135181896877,
          "compute_latency_us": [
            14.083,
            7.042,
            8.5,
            9.833,
            8.666,
            7.042,
            7.875,
            7.625,
            9.209
          ],
          "compute_latency_us_p50": 8.5,
          "compute_latency_us_p90": 10.683000000000002,
          "command_buffer_latency_us": [
            14.083285350399999,
            7.08329025656,
            8.58335988596,
            9.87498788163,
            8.66665504873,
            7.08334846422,
            7.8750308603,
            7.708324119449999,
            9.2500122264
          ],
          "command_buffer_latency_us_p50": 8.58335988596,
          "command_buffer_latency_us_p90": 10.716647375384001,
          "counter_control_throughput_ratio": 0.9246539945065508,
          "counter_control_latency_ratio": 0.25944725757730275
        },
        "correctness": {
          "max_abs_error": 2.1991937471099732e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 6.7874326215,
        "throughput_us_p90": 7.766761929309999,
        "latency_us_p50": 213.584,
        "latency_us_p90": 340.5998000000001
      },
      "slowdown": 0.4526725694942879,
      "device_compute_slowdown": 0.5984498757433662,
      "valid": true,
      "round": 0,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "layernorm_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-mbsja9tt/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "b8d9d984f3aaf8bfdb9269f6e0efc630605362c6ff4e852cc112881dfaf5e97b",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.488875,
        "capture_ms": 0.089667,
        "compile_ms": 65.189875,
        "allocation_upload_ms": 4.017042,
        "cold_call_ms": 9.64275,
        "warmup_ms": 200.395791,
        "download_ms": 0.835458,
        "repetitions": 1510,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          23.5842993377,
          23.019894702,
          23.0344099338,
          23.5717437086,
          24.5195364238,
          23.0382172185,
          23.0501927152,
          24.7188741722,
          23.0996139073
        ],
        "latency_us": [
          269.75,
          256.75,
          252,
          224.125,
          234.209,
          225.333,
          221.417,
          247.5,
          246.167
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 128,
            "reduction_scalar_rounds": 60,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 36864,
              "global_write_bytes": 12288,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 144,
              "global_write_bytes": 48,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 2,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 100,
            "normalized_kernel_cost": 400,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 9216,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 1522083,
              "compute_span_ns": 1522083,
              "calibration_cpu_ns": 2702875,
              "calibration_gpu_ticks": 2702875,
              "compute_passes": 1,
              "command_buffer_ns": 1522124.97778,
              "command_buffers": 2
            },
            {
              "compute_ns": 1041625,
              "compute_span_ns": 1041625,
              "calibration_cpu_ns": 1572792,
              "calibration_gpu_ticks": 1572792,
              "compute_passes": 1,
              "command_buffer_ns": 1041708.33947,
              "command_buffers": 2
            },
            {
              "compute_ns": 1440500,
              "compute_span_ns": 1440500,
              "calibration_cpu_ns": 2676708,
              "calibration_gpu_ticks": 2676708,
              "compute_passes": 1,
              "command_buffer_ns": 1440625.05569,
              "command_buffers": 2
            },
            {
              "compute_ns": 1064750,
              "compute_span_ns": 1064750,
              "calibration_cpu_ns": 1557916,
              "calibration_gpu_ticks": 1557916,
              "compute_passes": 1,
              "command_buffer_ns": 1064791.69335,
              "command_buffers": 2
            },
            {
              "compute_ns": 1577125,
              "compute_span_ns": 1577125,
              "calibration_cpu_ns": 2904083,
              "calibration_gpu_ticks": 2904083,
              "compute_passes": 1,
              "command_buffer_ns": 1577124.98913,
              "command_buffers": 2
            },
            {
              "compute_ns": 1032584,
              "compute_span_ns": 1032584,
              "calibration_cpu_ns": 1583000,
              "calibration_gpu_ticks": 1583000,
              "compute_passes": 1,
              "command_buffer_ns": 1032666.71067,
              "command_buffers": 2
            },
            {
              "compute_ns": 1541792,
              "compute_span_ns": 1541792,
              "calibration_cpu_ns": 3072208,
              "calibration_gpu_ticks": 3072208,
              "compute_passes": 1,
              "command_buffer_ns": 1541833.33507,
              "command_buffers": 2
            },
            {
              "compute_ns": 1022042,
              "compute_span_ns": 1022042,
              "calibration_cpu_ns": 1465666,
              "calibration_gpu_ticks": 1465666,
              "compute_passes": 1,
              "command_buffer_ns": 1022124.95403,
              "command_buffers": 2
            },
            {
              "compute_ns": 1562250,
              "compute_span_ns": 1562250,
              "calibration_cpu_ns": 2769375,
              "calibration_gpu_ticks": 2769375,
              "compute_passes": 1,
              "command_buffer_ns": 1562374.99323,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 313125,
              "compute_span_ns": 313125,
              "calibration_cpu_ns": 1906708,
              "calibration_gpu_ticks": 1906708,
              "compute_passes": 1,
              "command_buffer_ns": 313250.056934,
              "command_buffers": 2
            },
            {
              "compute_ns": 22959,
              "compute_span_ns": 22959,
              "calibration_cpu_ns": 862625,
              "calibration_gpu_ticks": 862625,
              "compute_passes": 1,
              "command_buffer_ns": 23083.2956731,
              "command_buffers": 2
            },
            {
              "compute_ns": 99375,
              "compute_span_ns": 99375,
              "calibration_cpu_ns": 1657708,
              "calibration_gpu_ticks": 1657708,
              "compute_passes": 1,
              "command_buffer_ns": 99416.6475721,
              "command_buffers": 2
            },
            {
              "compute_ns": 20500,
              "compute_span_ns": 20500,
              "calibration_cpu_ns": 852916,
              "calibration_gpu_ticks": 852916,
              "compute_passes": 1,
              "command_buffer_ns": 20541.6581593,
              "command_buffers": 2
            },
            {
              "compute_ns": 86209,
              "compute_span_ns": 86209,
              "calibration_cpu_ns": 1485291,
              "calibration_gpu_ticks": 1485291,
              "compute_passes": 1,
              "command_buffer_ns": 86375.0465214,
              "command_buffers": 2
            },
            {
              "compute_ns": 20083,
              "compute_span_ns": 20083,
              "calibration_cpu_ns": 808292,
              "calibration_gpu_ticks": 808292,
              "compute_passes": 1,
              "command_buffer_ns": 20083.389245,
              "command_buffers": 2
            },
            {
              "compute_ns": 70875,
              "compute_span_ns": 70875,
              "calibration_cpu_ns": 1396500,
              "calibration_gpu_ticks": 1396500,
              "compute_passes": 1,
              "command_buffer_ns": 70874.9867044,
              "command_buffers": 2
            },
            {
              "compute_ns": 19416,
              "compute_span_ns": 19416,
              "calibration_cpu_ns": 677791,
              "calibration_gpu_ticks": 677791,
              "compute_passes": 1,
              "command_buffer_ns": 19458.2971744,
              "command_buffers": 2
            },
            {
              "compute_ns": 20167,
              "compute_span_ns": 20167,
              "calibration_cpu_ns": 437667,
              "calibration_gpu_ticks": 437667,
              "compute_passes": 1,
              "command_buffer_ns": 20291.7144634,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1242958.35849,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1468916.71466,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1030249.92859,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1579249.9762,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1031874.97007,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1489208.37091,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1036083.38395,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1423833.30967,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1053749.99112,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 17874.9323823,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 17708.3420567,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20041.7125598,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 21000.043489,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20041.6543521,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 17916.6672751,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20833.3367482,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18083.3158083,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20375.0096262,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              19.42122435140625,
              22.9518236665625,
              16.09765513421875,
              24.675780878125,
              16.12304640734375,
              23.26888079546875,
              16.18880287421875,
              22.24739546359375,
              16.46484361125
            ],
            "command_buffer_throughput_us_p50": 19.42122435140625,
            "command_buffer_throughput_us_p90": 23.550260812,
            "command_buffer_latency_us": [
              17.874932382300003,
              17.708342056699998,
              20.041712559799997,
              21.000043489,
              20.0416543521,
              17.9166672751,
              20.8333367482,
              18.0833158083,
              20.3750096262
            ],
            "command_buffer_latency_us_p50": 20.0416543521,
            "command_buffer_latency_us_p90": 20.86667809636
          },
          "compute_throughput_us": [
            23.782546875,
            16.275390625,
            22.5078125,
            16.63671875,
            24.642578125,
            16.134125,
            24.0905,
            15.96940625,
            24.41015625
          ],
          "compute_throughput_us_p50": 22.5078125,
          "compute_throughput_us_p90": 24.456640625,
          "command_buffer_throughput_us": [
            23.7832027778125,
            16.27669280421875,
            22.50976649515625,
            16.63737020859375,
            24.64257795515625,
            16.13541735421875,
            24.09114586046875,
            15.97070240671875,
            24.41210926921875
          ],
          "command_buffer_throughput_us_p50": 22.50976649515625,
          "command_buffer_throughput_us_p90": 24.45820300640625,
          "compute_latency_us": [
            313.125,
            22.959,
            99.375,
            20.5,
            86.209,
            20.083,
            70.875,
            19.416,
            20.167
          ],
          "compute_latency_us_p50": 22.959,
          "compute_latency_us_p90": 142.12500000000003,
          "command_buffer_latency_us": [
            313.250056934,
            23.0832956731,
            99.4166475721,
            20.541658159300002,
            86.3750465214,
            20.083389245,
            70.8749867044,
            19.4582971744,
            20.291714463399998
          ],
          "command_buffer_latency_us_p50": 23.0832956731,
          "command_buffer_latency_us_p90": 142.18332944448002,
          "counter_control_throughput_ratio": 1.2245985292935668,
          "counter_control_latency_ratio": 1.303526642934163
        },
        "correctness": {
          "max_abs_error": 2.9463829065434766e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 23.0996139073,
        "throughput_us_p90": 24.559403973480002,
        "latency_us_p50": 246.167,
        "latency_us_p90": 259.35
      },
      "torch": {
        "cold_call_ms": 7.703083,
        "warmup_ms": 200.013542,
        "repetitions": 398,
        "throughput_us": [
          63.876673366834176,
          62.52334673366833,
          63.26654020100502,
          62.388716080402006,
          63.093175879396995,
          62.76570351758794,
          62.87091708542714,
          62.65117085427135,
          62.82715577889447
        ],
        "latency_us": [
          415.375,
          307.417,
          296.125,
          2147.417,
          671.2909999999999,
          326.41700000000003,
          212.875,
          625.541,
          325.834
        ],
        "throughput_us_p50": 62.82715577889447,
        "throughput_us_p90": 63.38856683417085,
        "latency_us_p50": 326.41700000000003,
        "latency_us_p90": 966.5162000000003,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 2589749.9872371554,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2783125.0336021185,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2921416.6570454836,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2768874.983303249,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2729624.9754726887,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2806958.3349861205,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2755749.970674515,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2808208.344504237,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2630958.3336114883,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 107041.67652875185,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 217875.00008940697,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44125.01584738493,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 129000.05094707012,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43999.9857917428,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 73458.35911110044,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43583.3353549242,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 219708.30857753754,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44208.3110101521,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              40.464843550580554,
              43.4863286500331,
              45.64713526633568,
              43.263671614113264,
              42.65039024176076,
              43.85872398415813,
              43.05859329178929,
              43.878255382878706,
              41.108723962679505
            ],
            "command_buffer_throughput_us_p50": 43.263671614113264,
            "command_buffer_throughput_us_p90": 44.2320313595701,
            "command_buffer_latency_us": [
              107.04167652875185,
              217.87500008940697,
              44.12501584738493,
              129.00005094707012,
              43.9999857917428,
              73.45835911110044,
              43.5833353549242,
              219.70830857753754,
              44.2083110101521
            ],
            "command_buffer_latency_us_p50": 73.45835911110044,
            "command_buffer_latency_us_p90": 218.24166178703308
          },
          "throughput": [
            {
              "compute_ns": 2754500.0,
              "compute_span_ns": 2754500.0,
              "command_buffer_ns": 2754500.019364059,
              "calibration_cpu_ns": 3964708.0,
              "calibration_gpu_ticks": 3964708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2741667.0,
              "compute_span_ns": 2741667.0,
              "command_buffer_ns": 2741666.6853241622,
              "calibration_cpu_ns": 4163625.0,
              "calibration_gpu_ticks": 4163625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2759292.0,
              "compute_span_ns": 2759292.0,
              "command_buffer_ns": 2759333.3506956697,
              "calibration_cpu_ns": 4153459.0,
              "calibration_gpu_ticks": 4153459.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2653458.0,
              "compute_span_ns": 2653458.0,
              "command_buffer_ns": 2653416.7118370533,
              "calibration_cpu_ns": 4039416.0,
              "calibration_gpu_ticks": 4039416.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2704459.0,
              "compute_span_ns": 2704459.0,
              "command_buffer_ns": 2704499.987885356,
              "calibration_cpu_ns": 4322417.0,
              "calibration_gpu_ticks": 4322417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2746083.0,
              "compute_span_ns": 2746083.0,
              "command_buffer_ns": 2746041.6313260794,
              "calibration_cpu_ns": 4216333.0,
              "calibration_gpu_ticks": 4216333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2733917.0,
              "compute_span_ns": 2733917.0,
              "command_buffer_ns": 2733958.3029970527,
              "calibration_cpu_ns": 4249166.0,
              "calibration_gpu_ticks": 4249166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2776959.0,
              "compute_span_ns": 2776959.0,
              "command_buffer_ns": 2776999.957859516,
              "calibration_cpu_ns": 4287792.0,
              "calibration_gpu_ticks": 4287792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2696292.0,
              "compute_span_ns": 2696292.0,
              "command_buffer_ns": 2696291.6599586606,
              "calibration_cpu_ns": 4134667.0,
              "calibration_gpu_ticks": 4134667.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 43375.0,
              "compute_span_ns": 43375.0,
              "command_buffer_ns": 43375.0101365149,
              "calibration_cpu_ns": 936583.0,
              "calibration_gpu_ticks": 936583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 43250.0,
              "compute_span_ns": 43250.0,
              "command_buffer_ns": 43250.03828853369,
              "calibration_cpu_ns": 542250.0,
              "calibration_gpu_ticks": 542250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 43875.0,
              "compute_span_ns": 43875.0,
              "command_buffer_ns": 43875.01394376159,
              "calibration_cpu_ns": 621959.0,
              "calibration_gpu_ticks": 621959.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44250.0,
              "compute_span_ns": 44250.0,
              "command_buffer_ns": 44250.04590302706,
              "calibration_cpu_ns": 481958.0,
              "calibration_gpu_ticks": 481958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 42583.0,
              "compute_span_ns": 42583.0,
              "command_buffer_ns": 42541.65105521679,
              "calibration_cpu_ns": 560291.0,
              "calibration_gpu_ticks": 560291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 46708.0,
              "compute_span_ns": 46708.0,
              "command_buffer_ns": 46666.653361171484,
              "calibration_cpu_ns": 622000.0,
              "calibration_gpu_ticks": 622000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 42625.0,
              "compute_span_ns": 42625.0,
              "command_buffer_ns": 42625.004425644875,
              "calibration_cpu_ns": 574792.0,
              "calibration_gpu_ticks": 574792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44208.0,
              "compute_span_ns": 44208.0,
              "command_buffer_ns": 44166.69253259897,
              "calibration_cpu_ns": 494958.0,
              "calibration_gpu_ticks": 494958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44708.0,
              "compute_span_ns": 44708.0,
              "command_buffer_ns": 44666.69633984566,
              "calibration_cpu_ns": 661375.0,
              "calibration_gpu_ticks": 661375.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            43.0390625,
            42.838546875,
            43.1139375,
            41.46028125,
            42.257171875,
            42.907546875,
            42.717453125,
            43.389984375,
            42.1295625
          ],
          "compute_throughput_us_p50": 42.838546875,
          "compute_throughput_us_p90": 43.169146874999996,
          "command_buffer_throughput_us": [
            43.03906280256342,
            42.838541958190035,
            43.11458360461984,
            41.45963612245396,
            42.25781231070869,
            42.90690048946999,
            42.71809848432895,
            43.39062434155494,
            42.12955718685407
          ],
          "command_buffer_throughput_us_p50": 42.838541958190035,
          "command_buffer_throughput_us_p90": 43.16979175200686,
          "compute_latency_us": [
            43.375,
            43.25,
            43.875,
            44.25,
            42.583,
            46.708,
            42.625,
            44.208,
            44.708
          ],
          "compute_latency_us_p50": 43.875,
          "compute_latency_us_p90": 45.108,
          "command_buffer_latency_us": [
            43.3750101365149,
            43.25003828853369,
            43.87501394376159,
            44.25004590302706,
            42.54165105521679,
            46.666653361171484,
            42.625004425644875,
            44.16669253259897,
            44.66669633984566
          ],
          "command_buffer_latency_us_p50": 43.87501394376159,
          "command_buffer_latency_us_p90": 45.06668774411082,
          "counter_control_throughput_ratio": 0.9888867267609268,
          "counter_control_latency_ratio": 0.6352803673519518
        },
        "download_ms": 0.410458,
        "allocation_upload_ms": 1.439875,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 2.982737865586671e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.3676692605438595,
      "device_compute_slowdown": 0.5254102704668364,
      "valid": true,
      "round": 0,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "layernorm_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.433792,
        "warmup_ms": 200.612167,
        "repetitions": 543,
        "throughput_us": [
          58.96439594843463,
          61.14901841620625,
          59.85366850828729,
          59.81752670349908,
          60.776012891344386,
          61.00805709023941,
          61.617327808471444,
          61.37453959484346,
          61.03130755064457
        ],
        "latency_us": [
          417.084,
          302.70799999999997,
          296.08299999999997,
          306.625,
          2074.833,
          1234.5410000000002,
          726.6669999999999,
          555.292,
          357.666
        ],
        "throughput_us_p50": 61.00805709023941,
        "throughput_us_p90": 61.42309723756906,
        "latency_us_p50": 417.084,
        "latency_us_p90": 1402.5994000000003,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 3098374.989349395,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2713541.6748933494,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2707124.9787695706,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2758041.664492339,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2762125.0483207405,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2638958.3363197744,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2819499.99043718,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2683250.0007003546,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4864916.671067476,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 47000.00863522291,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43499.98198449612,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 42249.972466379404,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44125.01584738493,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44124.957639724016,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43875.01394376159,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 46666.653361171484,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 42625.004425644875,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43500.04019215703,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              48.4121092085843,
              42.399088670208585,
              42.29882779327454,
              43.094401007692795,
              43.15820388001157,
              41.233724004996475,
              44.054687350580934,
              41.92578126094304,
              76.01432298542932
            ],
            "command_buffer_throughput_us_p50": 43.094401007692795,
            "command_buffer_throughput_us_p90": 53.93255196395331,
            "command_buffer_latency_us": [
              47.00000863522291,
              43.49998198449612,
              42.249972466379404,
              44.12501584738493,
              44.124957639724016,
              43.87501394376159,
              46.666653361171484,
              42.625004425644875,
              43.50004019215703
            ],
            "command_buffer_latency_us_p50": 43.87501394376159,
            "command_buffer_latency_us_p90": 46.73332441598177
          },
          "throughput": [
            {
              "compute_ns": 2578209.0,
              "compute_span_ns": 2578209.0,
              "command_buffer_ns": 2578249.9578781426,
              "calibration_cpu_ns": 5266791.0,
              "calibration_gpu_ticks": 5266791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2759833.0,
              "compute_span_ns": 2759833.0,
              "command_buffer_ns": 2759791.6778177023,
              "calibration_cpu_ns": 4222042.0,
              "calibration_gpu_ticks": 4222042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2601500.0,
              "compute_span_ns": 2601500.0,
              "command_buffer_ns": 2601500.0184997916,
              "calibration_cpu_ns": 4579041.0,
              "calibration_gpu_ticks": 4579041.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2670000.0,
              "compute_span_ns": 2670000.0,
              "command_buffer_ns": 2670000.0162236392,
              "calibration_cpu_ns": 4320500.0,
              "calibration_gpu_ticks": 4320500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2727875.0,
              "compute_span_ns": 2727875.0,
              "command_buffer_ns": 2727875.020354986,
              "calibration_cpu_ns": 3981333.0,
              "calibration_gpu_ticks": 3981333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2663875.0,
              "compute_span_ns": 2663875.0,
              "command_buffer_ns": 2663874.998688698,
              "calibration_cpu_ns": 4470916.0,
              "calibration_gpu_ticks": 4470916.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2769500.0,
              "compute_span_ns": 2769500.0,
              "command_buffer_ns": 2769499.958958477,
              "calibration_cpu_ns": 4521625.0,
              "calibration_gpu_ticks": 4521625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2651083.0,
              "compute_span_ns": 2651083.0,
              "command_buffer_ns": 2651041.722856462,
              "calibration_cpu_ns": 5038209.0,
              "calibration_gpu_ticks": 5038209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2611041.0,
              "compute_span_ns": 2611041.0,
              "command_buffer_ns": 2611000.0326298177,
              "calibration_cpu_ns": 3588791.0,
              "calibration_gpu_ticks": 3588791.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 44292.0,
              "compute_span_ns": 44292.0,
              "command_buffer_ns": 44291.66438058019,
              "calibration_cpu_ns": 526291.0,
              "calibration_gpu_ticks": 526291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 113125.0,
              "compute_span_ns": 113125.0,
              "command_buffer_ns": 113125.01737847924,
              "calibration_cpu_ns": 1533292.0,
              "calibration_gpu_ticks": 1533292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44250.0,
              "compute_span_ns": 44250.0,
              "command_buffer_ns": 44249.987695366144,
              "calibration_cpu_ns": 469667.0,
              "calibration_gpu_ticks": 469667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 173625.0,
              "compute_span_ns": 173625.0,
              "command_buffer_ns": 173625.01239404082,
              "calibration_cpu_ns": 1830250.0,
              "calibration_gpu_ticks": 1830250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44084.0,
              "compute_span_ns": 44084.0,
              "command_buffer_ns": 44125.01584738493,
              "calibration_cpu_ns": 521292.0,
              "calibration_gpu_ticks": 521292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47167.0,
              "compute_span_ns": 47167.0,
              "command_buffer_ns": 47166.65716841817,
              "calibration_cpu_ns": 1047584.0,
              "calibration_gpu_ticks": 1047584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44375.0,
              "compute_span_ns": 44375.0,
              "command_buffer_ns": 44375.01775100827,
              "calibration_cpu_ns": 710500.0,
              "calibration_gpu_ticks": 710500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 45542.0,
              "compute_span_ns": 45542.0,
              "command_buffer_ns": 45541.6738986969,
              "calibration_cpu_ns": 954500.0,
              "calibration_gpu_ticks": 954500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44084.0,
              "compute_span_ns": 44084.0,
              "command_buffer_ns": 44125.01584738493,
              "calibration_cpu_ns": 482083.0,
              "calibration_gpu_ticks": 482083.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            40.284515625,
            43.122390625,
            40.6484375,
            41.71875,
            42.623046875,
            41.623046875,
            43.2734375,
            41.423171875,
            40.797515625
          ],
          "compute_throughput_us_p50": 41.623046875,
          "compute_throughput_us_p90": 43.1526,
          "command_buffer_throughput_us": [
            40.28515559184598,
            43.1217449659016,
            40.648437789059244,
            41.71875025349436,
            42.62304719304666,
            41.6230468545109,
            43.2734368587262,
            41.42252691963222,
            40.7968755098409
          ],
          "command_buffer_throughput_us_p50": 41.6230468545109,
          "command_buffer_throughput_us_p90": 43.15208334446652,
          "compute_latency_us": [
            44.292,
            113.125,
            44.25,
            173.625,
            44.084,
            47.167,
            44.375,
            45.542,
            44.084
          ],
          "compute_latency_us_p50": 44.375,
          "compute_latency_us_p90": 125.22500000000001,
          "command_buffer_latency_us": [
            44.29166438058019,
            113.12501737847924,
            44.249987695366144,
            173.62501239404082,
            44.12501584738493,
            47.16665716841817,
            44.37501775100827,
            45.5416738986969,
            44.12501584738493
          ],
          "command_buffer_latency_us_p50": 44.37501775100827,
          "command_buffer_latency_us_p90": 125.22501638159157,
          "counter_control_throughput_ratio": 0.982266348058774,
          "counter_control_latency_ratio": 1.0473376694050691
        },
        "download_ms": 0.398834,
        "allocation_upload_ms": 3.208333,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 2.982737865586671e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-zrf8seum/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "83a70ba0e4d1dab6671da56395e0b13750f8d9c9277e4782c4e371450de131c7",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 23.060125,
        "capture_ms": 0.075875,
        "compile_ms": 64.325333,
        "allocation_upload_ms": 2.165417,
        "cold_call_ms": 5.454792,
        "warmup_ms": 200.359709,
        "download_ms": 0.660667,
        "repetitions": 1462,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          23.4330827633,
          23.4223098495,
          23.3282318741,
          23.331880301,
          23.5616737346,
          24.0980677155,
          24.0816805746,
          23.3315943912,
          23.3718932969
        ],
        "latency_us": [
          283.625,
          263.584,
          250.833,
          225.625,
          234.875,
          235.583,
          237.834,
          268,
          2415.541
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 256,
            "reduction_scalar_rounds": 60,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 36864,
              "global_write_bytes": 12288,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 144,
              "global_write_bytes": 48,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 2,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 108,
            "normalized_kernel_cost": 432,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 9216,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 1038125,
              "compute_span_ns": 1038125,
              "calibration_cpu_ns": 1522667,
              "calibration_gpu_ticks": 1522667,
              "compute_passes": 1,
              "command_buffer_ns": 1038208.31282,
              "command_buffers": 2
            },
            {
              "compute_ns": 1494625,
              "compute_span_ns": 1494625,
              "calibration_cpu_ns": 2786500,
              "calibration_gpu_ticks": 2786500,
              "compute_passes": 1,
              "command_buffer_ns": 1494625.00121,
              "command_buffers": 2
            },
            {
              "compute_ns": 1066250,
              "compute_span_ns": 1066250,
              "calibration_cpu_ns": 1554833,
              "calibration_gpu_ticks": 1554833,
              "compute_passes": 1,
              "command_buffer_ns": 1066374.99994,
              "command_buffers": 2
            },
            {
              "compute_ns": 1507250,
              "compute_span_ns": 1507250,
              "calibration_cpu_ns": 2877083,
              "calibration_gpu_ticks": 2877083,
              "compute_passes": 1,
              "command_buffer_ns": 1507291.68672,
              "command_buffers": 2
            },
            {
              "compute_ns": 1033375,
              "compute_span_ns": 1033375,
              "calibration_cpu_ns": 1626250,
              "calibration_gpu_ticks": 1626250,
              "compute_passes": 1,
              "command_buffer_ns": 1033458.27665,
              "command_buffers": 2
            },
            {
              "compute_ns": 1480000,
              "compute_span_ns": 1480000,
              "calibration_cpu_ns": 2371250,
              "calibration_gpu_ticks": 2371250,
              "compute_passes": 1,
              "command_buffer_ns": 1480125.06543,
              "command_buffers": 2
            },
            {
              "compute_ns": 1044333,
              "compute_span_ns": 1044333,
              "calibration_cpu_ns": 1531291,
              "calibration_gpu_ticks": 1531291,
              "compute_passes": 1,
              "command_buffer_ns": 1044375.00704,
              "command_buffers": 2
            },
            {
              "compute_ns": 1107083,
              "compute_span_ns": 1107083,
              "calibration_cpu_ns": 3259042,
              "calibration_gpu_ticks": 3259042,
              "compute_passes": 1,
              "command_buffer_ns": 1107166.69587,
              "command_buffers": 2
            },
            {
              "compute_ns": 1027000,
              "compute_span_ns": 1027000,
              "calibration_cpu_ns": 1744000,
              "calibration_gpu_ticks": 1744000,
              "compute_passes": 1,
              "command_buffer_ns": 1027041.69694,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 22583,
              "compute_span_ns": 22583,
              "calibration_cpu_ns": 799083,
              "calibration_gpu_ticks": 799083,
              "compute_passes": 1,
              "command_buffer_ns": 22708.3801292,
              "command_buffers": 2
            },
            {
              "compute_ns": 17959,
              "compute_span_ns": 17959,
              "calibration_cpu_ns": 444334,
              "calibration_gpu_ticks": 444334,
              "compute_passes": 1,
              "command_buffer_ns": 18041.6973308,
              "command_buffers": 2
            },
            {
              "compute_ns": 22250,
              "compute_span_ns": 22250,
              "calibration_cpu_ns": 1136042,
              "calibration_gpu_ticks": 1136042,
              "compute_passes": 1,
              "command_buffer_ns": 22291.6714847,
              "command_buffers": 2
            },
            {
              "compute_ns": 17958,
              "compute_span_ns": 17958,
              "calibration_cpu_ns": 426416,
              "calibration_gpu_ticks": 426416,
              "compute_passes": 1,
              "command_buffer_ns": 17958.3439603,
              "command_buffers": 2
            },
            {
              "compute_ns": 20291,
              "compute_span_ns": 20291,
              "calibration_cpu_ns": 874625,
              "calibration_gpu_ticks": 874625,
              "compute_passes": 1,
              "command_buffer_ns": 20333.2747333,
              "command_buffers": 2
            },
            {
              "compute_ns": 18083,
              "compute_span_ns": 18083,
              "calibration_cpu_ns": 524250,
              "calibration_gpu_ticks": 524250,
              "compute_passes": 1,
              "command_buffer_ns": 18208.2876563,
              "command_buffers": 2
            },
            {
              "compute_ns": 21791,
              "compute_span_ns": 21791,
              "calibration_cpu_ns": 847792,
              "calibration_gpu_ticks": 847792,
              "compute_passes": 1,
              "command_buffer_ns": 21874.9628402,
              "command_buffers": 2
            },
            {
              "compute_ns": 18500,
              "compute_span_ns": 18500,
              "calibration_cpu_ns": 408250,
              "calibration_gpu_ticks": 408250,
              "compute_passes": 1,
              "command_buffer_ns": 18541.701138,
              "command_buffers": 2
            },
            {
              "compute_ns": 22375,
              "compute_span_ns": 22375,
              "calibration_cpu_ns": 904500,
              "calibration_gpu_ticks": 904500,
              "compute_passes": 1,
              "command_buffer_ns": 22458.2618102,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1266624.95313,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1036041.64906,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1424291.695,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1057874.93521,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1498208.26482,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1059208.2981,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1329791.67392,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1033374.98149,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1954208.30278,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 65000.0292808,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19999.9776669,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 94750.0229813,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20125.0077225,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 84458.2682475,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20249.9795705,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 102666.672319,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20041.6543521,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 127041.654196,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              19.79101489265625,
              16.1881507665625,
              22.254557734375002,
              16.52929586265625,
              23.409504137812498,
              16.5501296578125,
              20.777994905,
              16.14648408578125,
              30.5345047309375
            ],
            "command_buffer_throughput_us_p50": 19.79101489265625,
            "command_buffer_throughput_us_p90": 24.834504256437498,
            "command_buffer_latency_us": [
              65.0000292808,
              19.999977666899998,
              94.7500229813,
              20.125007722499998,
              84.4582682475,
              20.2499795705,
              102.666672319,
              20.0416543521,
              127.041654196
            ],
            "command_buffer_latency_us_p50": 65.0000292808,
            "command_buffer_latency_us_p90": 107.5416686944
          },
          "compute_throughput_us": [
            16.220703125,
            23.353515625,
            16.66015625,
            23.55078125,
            16.146484375,
            23.125,
            16.317703125,
            17.298171875,
            16.046875
          ],
          "compute_throughput_us_p50": 16.66015625,
          "compute_throughput_us_p90": 23.39296875,
          "command_buffer_throughput_us": [
            16.2220048878125,
            23.35351564390625,
            16.6621093740625,
            23.551432605000002,
            16.14778557265625,
            23.12695414734375,
            16.318359485000002,
            17.29947962296875,
            16.0475265146875
          ],
          "command_buffer_throughput_us_p50": 16.6621093740625,
          "command_buffer_throughput_us_p90": 23.393099036125,
          "compute_latency_us": [
            22.583,
            17.959,
            22.25,
            17.958,
            20.291,
            18.083,
            21.791,
            18.5,
            22.375
          ],
          "compute_latency_us_p50": 20.291,
          "compute_latency_us_p90": 22.4166,
          "command_buffer_latency_us": [
            22.7083801292,
            18.0416973308,
            22.2916714847,
            17.958343960300002,
            20.3332747333,
            18.208287656299998,
            21.874962840200002,
            18.541701138,
            22.4582618102
          ],
          "command_buffer_latency_us_p50": 20.3332747333,
          "command_buffer_latency_us_p90": 22.508285474,
          "counter_control_throughput_ratio": 0.8196651346986715,
          "counter_control_latency_ratio": 0.34935953691805033
        },
        "correctness": {
          "max_abs_error": 2.9463829065434766e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 23.4223098495,
        "throughput_us_p90": 24.08495800278,
        "latency_us_p50": 250.833,
        "latency_us_p90": 710.0082000000004
      },
      "slowdown": 0.3839215829288768,
      "device_compute_slowdown": 0.40026277509267516,
      "valid": true,
      "round": 0,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "layernorm_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-c0s8ojah/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "4f0c4e70dcb06b507aa82e242c8819acd3628d5f79a065eb6c2f6b73ec6f396c",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.353834,
        "capture_ms": 0.071,
        "compile_ms": 64.197666,
        "allocation_upload_ms": 7.027417,
        "cold_call_ms": 5.986875,
        "warmup_ms": 201.866666,
        "download_ms": 2.603208,
        "repetitions": 253,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          106.361166008,
          113.195652174,
          107.886857708,
          106.551715415,
          115.991928854,
          105.709814229,
          106.336790514,
          116.698288538,
          106.248683794
        ],
        "latency_us": [
          335.042,
          328.208,
          330.25,
          372.625,
          326.166,
          296.042,
          317.208,
          2065.083,
          949.083
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 768,
            "reduction_scalar_rounds": 120,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 73728,
              "global_write_bytes": 24576,
              "private_read_bytes": 147456,
              "private_write_bytes": 49152
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 288,
              "global_write_bytes": 96,
              "private_read_bytes": 576,
              "private_write_bytes": 192
            },
            "striped_storage_scalars_per_worker": 48,
            "reduction_operations": 2,
            "reduction_elements": 12288,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 168,
            "normalized_kernel_cost": 2016,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 18432,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 5153333,
              "compute_span_ns": 5153333,
              "calibration_cpu_ns": 6139333,
              "calibration_gpu_ticks": 6139333,
              "compute_passes": 1,
              "command_buffer_ns": 5153416.65596,
              "command_buffers": 2
            },
            {
              "compute_ns": 5079334,
              "compute_span_ns": 5079334,
              "calibration_cpu_ns": 6534958,
              "calibration_gpu_ticks": 6534958,
              "compute_passes": 1,
              "command_buffer_ns": 5079458.35125,
              "command_buffers": 2
            },
            {
              "compute_ns": 5138083,
              "compute_span_ns": 5138083,
              "calibration_cpu_ns": 6454167,
              "calibration_gpu_ticks": 6454167,
              "compute_passes": 1,
              "command_buffer_ns": 5138166.71446,
              "command_buffers": 2
            },
            {
              "compute_ns": 5025375,
              "compute_span_ns": 5025375,
              "calibration_cpu_ns": 6237750,
              "calibration_gpu_ticks": 6237750,
              "compute_passes": 1,
              "command_buffer_ns": 5025499.966,
              "command_buffers": 2
            },
            {
              "compute_ns": 4963208,
              "compute_span_ns": 4963208,
              "calibration_cpu_ns": 6534666,
              "calibration_gpu_ticks": 6534666,
              "compute_passes": 1,
              "command_buffer_ns": 4963208.28097,
              "command_buffers": 2
            },
            {
              "compute_ns": 5174167,
              "compute_span_ns": 5174167,
              "calibration_cpu_ns": 6250750,
              "calibration_gpu_ticks": 6250750,
              "compute_passes": 1,
              "command_buffer_ns": 5174249.99271,
              "command_buffers": 2
            },
            {
              "compute_ns": 5078416,
              "compute_span_ns": 5078416,
              "calibration_cpu_ns": 6469042,
              "calibration_gpu_ticks": 6469042,
              "compute_passes": 1,
              "command_buffer_ns": 5078458.34363,
              "command_buffers": 2
            },
            {
              "compute_ns": 7510542,
              "compute_span_ns": 7510542,
              "calibration_cpu_ns": 8841583,
              "calibration_gpu_ticks": 8841583,
              "compute_passes": 1,
              "command_buffer_ns": 7510625.00058,
              "command_buffers": 2
            },
            {
              "compute_ns": 7390541,
              "compute_span_ns": 7390541,
              "calibration_cpu_ns": 8586250,
              "calibration_gpu_ticks": 8586250,
              "compute_passes": 1,
              "command_buffer_ns": 7390541.723,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 104791,
              "compute_span_ns": 104791,
              "calibration_cpu_ns": 696416,
              "calibration_gpu_ticks": 696416,
              "compute_passes": 1,
              "command_buffer_ns": 104875.070974,
              "command_buffers": 2
            },
            {
              "compute_ns": 76333,
              "compute_span_ns": 76333,
              "calibration_cpu_ns": 584750,
              "calibration_gpu_ticks": 584750,
              "compute_passes": 1,
              "command_buffer_ns": 76416.7052694,
              "command_buffers": 2
            },
            {
              "compute_ns": 118125,
              "compute_span_ns": 118125,
              "calibration_cpu_ns": 1143459,
              "calibration_gpu_ticks": 1143459,
              "compute_passes": 1,
              "command_buffer_ns": 118166.732136,
              "command_buffers": 2
            },
            {
              "compute_ns": 119375,
              "compute_span_ns": 119375,
              "calibration_cpu_ns": 3082041,
              "calibration_gpu_ticks": 3082041,
              "compute_passes": 1,
              "command_buffer_ns": 119458.360132,
              "command_buffers": 2
            },
            {
              "compute_ns": 80291,
              "compute_span_ns": 80291,
              "calibration_cpu_ns": 523375,
              "calibration_gpu_ticks": 523375,
              "compute_passes": 1,
              "command_buffer_ns": 80250.0289865,
              "command_buffers": 2
            },
            {
              "compute_ns": 81625,
              "compute_span_ns": 81625,
              "calibration_cpu_ns": 635833,
              "calibration_gpu_ticks": 635833,
              "compute_passes": 1,
              "command_buffer_ns": 81666.6288301,
              "command_buffers": 2
            },
            {
              "compute_ns": 75875,
              "compute_span_ns": 75875,
              "calibration_cpu_ns": 635667,
              "calibration_gpu_ticks": 635667,
              "compute_passes": 1,
              "command_buffer_ns": 75916.6432545,
              "command_buffers": 2
            },
            {
              "compute_ns": 120458,
              "compute_span_ns": 120458,
              "calibration_cpu_ns": 2035250,
              "calibration_gpu_ticks": 2035250,
              "compute_passes": 1,
              "command_buffer_ns": 120541.604701,
              "command_buffers": 2
            },
            {
              "compute_ns": 114083,
              "compute_span_ns": 114083,
              "calibration_cpu_ns": 3121208,
              "calibration_gpu_ticks": 3121208,
              "compute_passes": 1,
              "command_buffer_ns": 114083.348308,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 5112625.01823,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5311999.99386,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4897291.59744,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5273333.285,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5319208.37238,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5231250.01928,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5097791.66896,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5321083.35756,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4874041.59503,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 226375.006605,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 74624.9570511,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 120125.012472,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 75541.6695029,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 118708.354421,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 149916.682858,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 99166.7038761,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76416.7052694,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 112625.013571,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              79.88476590984375,
              82.99999990406249,
              76.52018120999999,
              82.395832578125,
              83.1126308184375,
              81.73828155125,
              79.65299482750001,
              83.141927461875,
              76.15689992234375
            ],
            "command_buffer_throughput_us_p50": 81.73828155125,
            "command_buffer_throughput_us_p90": 83.118490147125,
            "command_buffer_latency_us": [
              226.375006605,
              74.62495705110001,
              120.12501247200001,
              75.54166950289999,
              118.708354421,
              149.91668285799997,
              99.1667038761,
              76.4167052694,
              112.625013571
            ],
            "command_buffer_latency_us_p50": 112.625013571,
            "command_buffer_latency_us_p90": 165.2083476074
          },
          "compute_throughput_us": [
            80.520828125,
            79.36459375,
            80.282546875,
            78.521484375,
            77.550125,
            80.846359375,
            79.35025,
            117.35221875,
            115.477203125
          ],
          "compute_throughput_us_p50": 80.282546875,
          "compute_throughput_us_p90": 115.85220625000001,
          "command_buffer_throughput_us": [
            80.522135249375,
            79.36653673828125,
            80.28385491343751,
            78.52343696875,
            77.55012939015624,
            80.84765613609375,
            79.35091161921875,
            117.35351563406249,
            115.477214421875
          ],
          "command_buffer_throughput_us_p50": 80.28385491343751,
          "command_buffer_throughput_us_p90": 115.8524746643125,
          "compute_latency_us": [
            104.791,
            76.333,
            118.125,
            119.375,
            80.291,
            81.625,
            75.875,
            120.458,
            114.083
          ],
          "compute_latency_us_p50": 104.791,
          "compute_latency_us_p90": 119.5916,
          "command_buffer_latency_us": [
            104.875070974,
            76.4167052694,
            118.16673213600001,
            119.458360132,
            80.2500289865,
            81.6666288301,
            75.9166432545,
            120.541604701,
            114.083348308
          ],
          "command_buffer_latency_us_p50": 104.875070974,
          "command_buffer_latency_us_p90": 119.6750090458,
          "counter_control_throughput_ratio": 0.9962075097247071,
          "counter_control_latency_ratio": 0.9836979801649848
        },
        "correctness": {
          "max_abs_error": 2.7199924934606656e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 106.551715415,
        "throughput_us_p90": 116.13320079079999,
        "latency_us_p50": 330.25,
        "latency_us_p90": 1172.2830000000001
      },
      "torch": {
        "cold_call_ms": 8.439333,
        "warmup_ms": 202.5675,
        "repetitions": 77,
        "throughput_us": [
          374.012987012987,
          412.92802597402596,
          402.9778181818182,
          373.82305194805195,
          412.93776623376624,
          406.0021558441559,
          371.4529220779221,
          408.0470779220779,
          406.5173116883117
        ],
        "latency_us": [
          1025.917,
          517.458,
          504.54200000000003,
          502.12500000000006,
          510.95900000000006,
          504.75000000000006,
          504.708,
          2737.917,
          828.917
        ],
        "throughput_us_p50": 406.0021558441559,
        "throughput_us_p90": 412.929974025974,
        "latency_us_p50": 510.95900000000006,
        "latency_us_p90": 1368.3170000000002,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 24847624.998074025,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22655749.98455122,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22365499.986335635,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21900416.642893106,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 23084166.692569852,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22942291.689105332,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22412833.35024491,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22184874.9788478,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22180625.004693866,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 278250.02325698733,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 579125.015065074,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 277208.3389572799,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 287749.9791793525,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 285250.01835078,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 284375.04079192877,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 271625.0019147992,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 286124.99590963125,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 271375.00001117587,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              388.24414059490664,
              353.99609350861283,
              349.4609372864943,
              342.1940100452048,
              360.69010457140394,
              358.4733076422708,
              350.2005210975767,
              346.6386715444969,
              346.57226569834165
            ],
            "command_buffer_throughput_us_p50": 350.2005210975767,
            "command_buffer_throughput_us_p90": 366.2009117761045,
            "command_buffer_latency_us": [
              278.25002325698733,
              579.125015065074,
              277.2083389572799,
              287.7499791793525,
              285.25001835078,
              284.37504079192877,
              271.6250019147992,
              286.12499590963125,
              271.37500001117587
            ],
            "command_buffer_latency_us_p50": 284.37504079192877,
            "command_buffer_latency_us_p90": 346.02498635649687
          },
          "throughput": [
            {
              "compute_ns": 22779250.0,
              "compute_span_ns": 22779250.0,
              "command_buffer_ns": 22779249.993618578,
              "calibration_cpu_ns": 25334625.0,
              "calibration_gpu_ticks": 25334625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 22036333.0,
              "compute_span_ns": 22036333.0,
              "command_buffer_ns": 22036291.658878326,
              "calibration_cpu_ns": 24205958.0,
              "calibration_gpu_ticks": 24205958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21867792.0,
              "compute_span_ns": 21867792.0,
              "command_buffer_ns": 21867833.333089948,
              "calibration_cpu_ns": 23042833.0,
              "calibration_gpu_ticks": 23042833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 22007417.0,
              "compute_span_ns": 22007417.0,
              "command_buffer_ns": 22007458.37762952,
              "calibration_cpu_ns": 23874083.0,
              "calibration_gpu_ticks": 23874083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 22046417.0,
              "compute_span_ns": 22046417.0,
              "command_buffer_ns": 22046416.70687124,
              "calibration_cpu_ns": 23586041.0,
              "calibration_gpu_ticks": 23586041.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 22156125.0,
              "compute_span_ns": 22156125.0,
              "command_buffer_ns": 22156124.99276176,
              "calibration_cpu_ns": 24132209.0,
              "calibration_gpu_ticks": 24132209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21994334.0,
              "compute_span_ns": 21994334.0,
              "command_buffer_ns": 21994374.983478338,
              "calibration_cpu_ns": 24122375.0,
              "calibration_gpu_ticks": 24122375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 25117583.0,
              "compute_span_ns": 25117583.0,
              "command_buffer_ns": 25117583.34003389,
              "calibration_cpu_ns": 28160792.0,
              "calibration_gpu_ticks": 28160792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21868250.0,
              "compute_span_ns": 21868250.0,
              "command_buffer_ns": 21868249.983526766,
              "calibration_cpu_ns": 23696166.0,
              "calibration_gpu_ticks": 23696166.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 271875.0,
              "compute_span_ns": 271875.0,
              "command_buffer_ns": 271875.00381842256,
              "calibration_cpu_ns": 1060291.0,
              "calibration_gpu_ticks": 1060291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 285416.0,
              "compute_span_ns": 285416.0,
              "command_buffer_ns": 285374.9901987612,
              "calibration_cpu_ns": 1599250.0,
              "calibration_gpu_ticks": 1599250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 275625.0,
              "compute_span_ns": 275625.0,
              "command_buffer_ns": 275625.0323727727,
              "calibration_cpu_ns": 761250.0,
              "calibration_gpu_ticks": 761250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 273208.0,
              "compute_span_ns": 273208.0,
              "command_buffer_ns": 273166.6318140924,
              "calibration_cpu_ns": 1179583.0,
              "calibration_gpu_ticks": 1179583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 304542.0,
              "compute_span_ns": 304542.0,
              "command_buffer_ns": 304583.3436772227,
              "calibration_cpu_ns": 1366042.0,
              "calibration_gpu_ticks": 1366042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 271083.0,
              "compute_span_ns": 271083.0,
              "command_buffer_ns": 271083.3214223385,
              "calibration_cpu_ns": 718291.0,
              "calibration_gpu_ticks": 718291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 271375.0,
              "compute_span_ns": 271375.0,
              "command_buffer_ns": 271375.00001117587,
              "calibration_cpu_ns": 730542.0,
              "calibration_gpu_ticks": 730542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 287917.0,
              "compute_span_ns": 287917.0,
              "command_buffer_ns": 287916.6277125478,
              "calibration_cpu_ns": 1512750.0,
              "calibration_gpu_ticks": 1512750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 284083.0,
              "compute_span_ns": 284083.0,
              "command_buffer_ns": 284083.30399543047,
              "calibration_cpu_ns": 3400458.0,
              "calibration_gpu_ticks": 3400458.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            355.92578125,
            344.317703125,
            341.68425,
            343.865890625,
            344.475265625,
            346.189453125,
            343.66146875,
            392.462234375,
            341.69140625
          ],
          "compute_throughput_us_p50": 344.317703125,
          "compute_throughput_us_p90": 363.233071875,
          "command_buffer_throughput_us": [
            355.9257811502903,
            344.31705716997385,
            341.68489582953043,
            343.8665371504612,
            344.47526104486315,
            346.1894530119025,
            343.662109116849,
            392.4622396880295,
            341.6914059926057
          ],
          "command_buffer_throughput_us_p50": 344.31705716997385,
          "command_buffer_throughput_us_p90": 363.2330728578381,
          "compute_latency_us": [
            271.875,
            285.416,
            275.625,
            273.208,
            304.542,
            271.083,
            271.375,
            287.917,
            284.083
          ],
          "compute_latency_us_p50": 275.625,
          "compute_latency_us_p90": 291.24199999999996,
          "command_buffer_latency_us": [
            271.87500381842256,
            285.3749901987612,
            275.6250323727727,
            273.1666318140924,
            304.58334367722273,
            271.0833214223385,
            271.37500001117587,
            287.9166277125478,
            284.08330399543047
          ],
          "command_buffer_latency_us_p50": 275.6250323727727,
          "command_buffer_latency_us_p90": 291.24997090548277,
          "counter_control_throughput_ratio": 0.9777484673470416,
          "counter_control_latency_ratio": 0.9942883876060048
        },
        "download_ms": 0.774583,
        "allocation_upload_ms": 20.225666,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.046016576746524e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.2624412552525951,
      "device_compute_slowdown": 0.2331641566679901,
      "valid": true,
      "round": 0,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "layernorm_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 1.129,
        "warmup_ms": 202.836583,
        "repetitions": 80,
        "throughput_us": [
          393.5421875,
          375.245825,
          386.39480000000003,
          398.1515625,
          383.9828125,
          370.0448,
          400.9416625,
          402.75103750000005,
          383.19635000000005
        ],
        "latency_us": [
          1115.583,
          787.875,
          479.08299999999997,
          524.6659999999999,
          529.792,
          513.0,
          513.417,
          510.375,
          512.3330000000001
        ],
        "throughput_us_p50": 386.39480000000003,
        "throughput_us_p90": 401.3035375,
        "latency_us_p50": 513.417,
        "latency_us_p90": 853.4166,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 22108124.976512045,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22197916.638106108,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22184958.33221823,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 23951500.01393631,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 24313000.0308156,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22383791.68557003,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22074625.01246482,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22569749.96998906,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22499000.013340265,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 309249.96826797724,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 274999.9985098839,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 271083.3796299994,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 271291.7048484087,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 514583.31290632486,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 454624.998383224,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 276624.98177960515,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 280208.36180076003,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 270583.3758227527,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              345.4394527580007,
              346.84244747040793,
              346.63997394090984,
              374.24218771775486,
              379.89062548149377,
              349.74674508703174,
              344.91601581976283,
              352.6523432810791,
              351.54687520844163
            ],
            "command_buffer_throughput_us_p50": 349.74674508703174,
            "command_buffer_throughput_us_p90": 375.37187527050264,
            "command_buffer_latency_us": [
              309.24996826797724,
              274.9999985098839,
              271.0833796299994,
              271.2917048484087,
              514.5833129063249,
              454.624998383224,
              276.62498177960515,
              280.20836180076003,
              270.5833758227527
            ],
            "command_buffer_latency_us_p50": 276.62498177960515,
            "command_buffer_latency_us_p90": 466.6166612878442
          },
          "throughput": [
            {
              "compute_ns": 22723083.0,
              "compute_span_ns": 22723083.0,
              "command_buffer_ns": 22723083.32612738,
              "calibration_cpu_ns": 24814500.0,
              "calibration_gpu_ticks": 24814500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 22308167.0,
              "compute_span_ns": 22308167.0,
              "command_buffer_ns": 22308208.339381963,
              "calibration_cpu_ns": 24842625.0,
              "calibration_gpu_ticks": 24842625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21870125.0,
              "compute_span_ns": 21870125.0,
              "command_buffer_ns": 21870124.96870011,
              "calibration_cpu_ns": 23445666.0,
              "calibration_gpu_ticks": 23445666.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21860375.0,
              "compute_span_ns": 21860375.0,
              "command_buffer_ns": 21860375.010874122,
              "calibration_cpu_ns": 24474708.0,
              "calibration_gpu_ticks": 24474708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 25396459.0,
              "compute_span_ns": 25396459.0,
              "command_buffer_ns": 25396500.015631318,
              "calibration_cpu_ns": 26543458.0,
              "calibration_gpu_ticks": 26543458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21933083.0,
              "compute_span_ns": 21933083.0,
              "command_buffer_ns": 21933041.68758914,
              "calibration_cpu_ns": 24401334.0,
              "calibration_gpu_ticks": 24401334.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21951458.0,
              "compute_span_ns": 21951458.0,
              "command_buffer_ns": 21951416.681986302,
              "calibration_cpu_ns": 24474875.0,
              "calibration_gpu_ticks": 24474875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 22666667.0,
              "compute_span_ns": 22666667.0,
              "command_buffer_ns": 22666708.333417773,
              "calibration_cpu_ns": 24946000.0,
              "calibration_gpu_ticks": 24946000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 22548208.0,
              "compute_span_ns": 22548208.0,
              "command_buffer_ns": 22548208.362422884,
              "calibration_cpu_ns": 24131584.0,
              "calibration_gpu_ticks": 24131584.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 271417.0,
              "compute_span_ns": 271417.0,
              "command_buffer_ns": 271416.6766963899,
              "calibration_cpu_ns": 744042.0,
              "calibration_gpu_ticks": 744042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 382333.0,
              "compute_span_ns": 382333.0,
              "command_buffer_ns": 382291.6769422591,
              "calibration_cpu_ns": 1988750.0,
              "calibration_gpu_ticks": 1988750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 595625.0,
              "compute_span_ns": 595625.0,
              "command_buffer_ns": 595624.9660812318,
              "calibration_cpu_ns": 1600167.0,
              "calibration_gpu_ticks": 1600167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 277084.0,
              "compute_span_ns": 277084.0,
              "command_buffer_ns": 277125.04379451275,
              "calibration_cpu_ns": 825667.0,
              "calibration_gpu_ticks": 825667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 278958.0,
              "compute_span_ns": 278958.0,
              "command_buffer_ns": 278916.6755974293,
              "calibration_cpu_ns": 989292.0,
              "calibration_gpu_ticks": 989292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 275834.0,
              "compute_span_ns": 275834.0,
              "command_buffer_ns": 275874.9760687351,
              "calibration_cpu_ns": 733417.0,
              "calibration_gpu_ticks": 733417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 270375.0,
              "compute_span_ns": 270375.0,
              "command_buffer_ns": 270375.0506043434,
              "calibration_cpu_ns": 714375.0,
              "calibration_gpu_ticks": 714375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 533375.0,
              "compute_span_ns": 533375.0,
              "command_buffer_ns": 533375.0159479678,
              "calibration_cpu_ns": 2089000.0,
              "calibration_gpu_ticks": 2089000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 564792.0,
              "compute_span_ns": 564792.0,
              "command_buffer_ns": 564833.3462886512,
              "calibration_cpu_ns": 1912959.0,
              "calibration_gpu_ticks": 1912959.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            355.048171875,
            348.565109375,
            341.720703125,
            341.568359375,
            396.819671875,
            342.704421875,
            342.99153125,
            354.166671875,
            352.31575
          ],
          "compute_throughput_us_p50": 348.565109375,
          "compute_throughput_us_p90": 363.402471875,
          "command_buffer_throughput_us": [
            355.0481769707403,
            348.5657553028432,
            341.72070263593923,
            341.56835954490816,
            396.82031274423935,
            342.7037763685803,
            342.99088565603597,
            354.1673177096527,
            352.31575566285755
          ],
          "command_buffer_throughput_us_p50": 348.5657553028432,
          "command_buffer_throughput_us_p90": 363.4026041254401,
          "compute_latency_us": [
            271.417,
            382.333,
            595.625,
            277.084,
            278.958,
            275.834,
            270.375,
            533.375,
            564.792
          ],
          "compute_latency_us_p50": 278.958,
          "compute_latency_us_p90": 570.9586,
          "command_buffer_latency_us": [
            271.4166766963899,
            382.2916769422591,
            595.6249660812318,
            277.12504379451275,
            278.9166755974293,
            275.8749760687351,
            270.3750506043434,
            533.3750159479678,
            564.8333462886512
          ],
          "command_buffer_latency_us_p50": 278.9166755974293,
          "command_buffer_latency_us_p90": 570.9916702471673,
          "counter_control_throughput_ratio": 1.002187134941706,
          "counter_control_latency_ratio": 1.0215020910770698
        },
        "download_ms": 0.918167,
        "allocation_upload_ms": 1.164791,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.046016576746524e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-2_m5sb4g/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "2f415c8199d32d1b999ce6dae6674c9bc8bbdf9b0adda641a6d2aee649468028",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.699959,
        "capture_ms": 0.085375,
        "compile_ms": 65.99675,
        "allocation_upload_ms": 8.148,
        "cold_call_ms": 8.441167,
        "warmup_ms": 200.409041,
        "download_ms": 3.406916,
        "repetitions": 287,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          105.319250871,
          103.50087108,
          109.627756098,
          102.355111498,
          107.041376307,
          110.604965157,
          102.472996516,
          107.923923345,
          110.986062718
        ],
        "latency_us": [
          346.708,
          317.375,
          296.375,
          321.875,
          274.625,
          298.333,
          2543.958,
          648.375,
          569.25
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 384,
            "reduction_scalar_rounds": 120,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 73728,
              "global_write_bytes": 24576,
              "private_read_bytes": 147456,
              "private_write_bytes": 49152
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 288,
              "global_write_bytes": 96,
              "private_read_bytes": 576,
              "private_write_bytes": 192
            },
            "striped_storage_scalars_per_worker": 48,
            "reduction_operations": 2,
            "reduction_elements": 12288,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 160,
            "normalized_kernel_cost": 1920,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 18432,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4587083,
              "compute_span_ns": 4587083,
              "calibration_cpu_ns": 6617958,
              "calibration_gpu_ticks": 6617958,
              "compute_passes": 1,
              "command_buffer_ns": 4587166.70983,
              "command_buffers": 2
            },
            {
              "compute_ns": 4721875,
              "compute_span_ns": 4721875,
              "calibration_cpu_ns": 6322958,
              "calibration_gpu_ticks": 6322958,
              "compute_passes": 1,
              "command_buffer_ns": 4721999.9833,
              "command_buffers": 2
            },
            {
              "compute_ns": 5300791,
              "compute_span_ns": 5300791,
              "calibration_cpu_ns": 6377792,
              "calibration_gpu_ticks": 6377792,
              "compute_passes": 1,
              "command_buffer_ns": 5300875.05467,
              "command_buffers": 2
            },
            {
              "compute_ns": 5017583,
              "compute_span_ns": 5017583,
              "calibration_cpu_ns": 6152084,
              "calibration_gpu_ticks": 6152084,
              "compute_passes": 1,
              "command_buffer_ns": 5017541.63997,
              "command_buffers": 2
            },
            {
              "compute_ns": 5072500,
              "compute_span_ns": 5072500,
              "calibration_cpu_ns": 6262875,
              "calibration_gpu_ticks": 6262875,
              "compute_passes": 1,
              "command_buffer_ns": 5072625.00469,
              "command_buffers": 2
            },
            {
              "compute_ns": 4857375,
              "compute_span_ns": 4857375,
              "calibration_cpu_ns": 6206417,
              "calibration_gpu_ticks": 6206417,
              "compute_passes": 1,
              "command_buffer_ns": 4857499.96733,
              "command_buffers": 2
            },
            {
              "compute_ns": 4935208,
              "compute_span_ns": 4935208,
              "calibration_cpu_ns": 6205542,
              "calibration_gpu_ticks": 6205542,
              "compute_passes": 1,
              "command_buffer_ns": 4935333.27244,
              "command_buffers": 2
            },
            {
              "compute_ns": 4853000,
              "compute_span_ns": 4853000,
              "calibration_cpu_ns": 6278834,
              "calibration_gpu_ticks": 6278834,
              "compute_passes": 1,
              "command_buffer_ns": 4853083.34464,
              "command_buffers": 2
            },
            {
              "compute_ns": 7296959,
              "compute_span_ns": 7296959,
              "calibration_cpu_ns": 8298250,
              "calibration_gpu_ticks": 8298250,
              "compute_passes": 1,
              "command_buffer_ns": 7297041.70953,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 118958,
              "compute_span_ns": 118958,
              "calibration_cpu_ns": 3865708,
              "calibration_gpu_ticks": 3865708,
              "compute_passes": 1,
              "command_buffer_ns": 118999.974802,
              "command_buffers": 2
            },
            {
              "compute_ns": 124042,
              "compute_span_ns": 124042,
              "calibration_cpu_ns": 1354041,
              "calibration_gpu_ticks": 1354041,
              "compute_passes": 1,
              "command_buffer_ns": 124083.308037,
              "command_buffers": 2
            },
            {
              "compute_ns": 121458,
              "compute_span_ns": 121458,
              "calibration_cpu_ns": 1214125,
              "calibration_gpu_ticks": 1214125,
              "compute_passes": 1,
              "command_buffer_ns": 121541.670524,
              "command_buffers": 2
            },
            {
              "compute_ns": 118458,
              "compute_span_ns": 118458,
              "calibration_cpu_ns": 1207417,
              "calibration_gpu_ticks": 1207417,
              "compute_passes": 1,
              "command_buffer_ns": 118541.705888,
              "command_buffers": 2
            },
            {
              "compute_ns": 95916,
              "compute_span_ns": 95916,
              "calibration_cpu_ns": 667708,
              "calibration_gpu_ticks": 667708,
              "compute_passes": 1,
              "command_buffer_ns": 95916.679129,
              "command_buffers": 2
            },
            {
              "compute_ns": 124583,
              "compute_span_ns": 124583,
              "calibration_cpu_ns": 2880542,
              "calibration_gpu_ticks": 2880542,
              "compute_passes": 1,
              "command_buffer_ns": 124583.311845,
              "command_buffers": 2
            },
            {
              "compute_ns": 125000,
              "compute_span_ns": 125000,
              "calibration_cpu_ns": 1461833,
              "calibration_gpu_ticks": 1461833,
              "compute_passes": 1,
              "command_buffer_ns": 125000.020489,
              "command_buffers": 2
            },
            {
              "compute_ns": 73541,
              "compute_span_ns": 73541,
              "calibration_cpu_ns": 531041,
              "calibration_gpu_ticks": 531041,
              "compute_passes": 1,
              "command_buffer_ns": 73500.0357963,
              "command_buffers": 2
            },
            {
              "compute_ns": 121959,
              "compute_span_ns": 121959,
              "calibration_cpu_ns": 1505417,
              "calibration_gpu_ticks": 1505417,
              "compute_passes": 1,
              "command_buffer_ns": 122041.674331,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 5538875.06248,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5134666.6296,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5096458.30607,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5021458.31706,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4937708.37784,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5048833.32178,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4816416.70922,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7481583.39411,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7273083.31989,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 118291.703984,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 121375.02199,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 73583.3309591,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 91083.3477974,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 123916.659504,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 73500.0357963,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 73166.6223146,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 125041.638967,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 73125.003837,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              86.54492285124999,
              80.22916608749999,
              79.63216103234375,
              78.4602862040625,
              77.15169340375,
              78.8880206528125,
              75.2565110815625,
              116.89974053296875,
              113.64192687328125
            ],
            "command_buffer_throughput_us_p50": 79.63216103234375,
            "command_buffer_throughput_us_p90": 114.29348960521875,
            "command_buffer_latency_us": [
              118.29170398400001,
              121.37502199,
              73.5833309591,
              91.0833477974,
              123.916659504,
              73.5000357963,
              73.16662231459999,
              125.04163896700001,
              73.125003837
            ],
            "command_buffer_latency_us_p50": 91.0833477974,
            "command_buffer_latency_us_p90": 124.1416553966
          },
          "compute_throughput_us": [
            71.673171875,
            73.779296875,
            82.824859375,
            78.399734375,
            79.2578125,
            75.896484375,
            77.112625,
            75.828125,
            114.014984375
          ],
          "compute_throughput_us_p50": 77.112625,
          "compute_throughput_us_p90": 89.06288437500001,
          "command_buffer_throughput_us": [
            71.67447984109376,
            73.7812497390625,
            82.82617272921875,
            78.39908812453125,
            79.25976569828124,
            75.89843698953126,
            77.11458238187501,
            75.82942726,
            114.01627671140625
          ],
          "command_buffer_throughput_us_p50": 77.11458238187501,
          "command_buffer_throughput_us_p90": 89.06419352565626,
          "compute_latency_us": [
            118.958,
            124.042,
            121.458,
            118.458,
            95.916,
            124.583,
            125.0,
            73.541,
            121.959
          ],
          "compute_latency_us_p50": 121.458,
          "compute_latency_us_p90": 124.6664,
          "command_buffer_latency_us": [
            118.999974802,
            124.083308037,
            121.541670524,
            118.541705888,
            95.916679129,
            124.583311845,
            125.00002048900001,
            73.5000357963,
            122.041674331
          ],
          "command_buffer_latency_us_p50": 121.541670524,
          "command_buffer_latency_us_p90": 124.6666535738,
          "counter_control_throughput_ratio": 0.9992200120278418,
          "counter_control_latency_ratio": 1.3014640848695704
        },
        "correctness": {
          "max_abs_error": 2.7199924934606656e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 107.041376307,
        "throughput_us_p90": 110.6811846692,
        "latency_us_p50": 321.875,
        "latency_us_p90": 1027.4916000000003
      },
      "slowdown": 0.27702592350362887,
      "device_compute_slowdown": 0.22122875447364185,
      "valid": true,
      "round": 0,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "layernorm_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-mab3qp4p/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "7d02ca31ce0660bb7b8cf2d4b7c68e2de8afe58cfb625e16c6aa93173ce8f234",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.664458,
        "capture_ms": 0.087708,
        "compile_ms": 64.953208,
        "allocation_upload_ms": 6.981958,
        "cold_call_ms": 9.456166,
        "warmup_ms": 200.152,
        "download_ms": 3.410166,
        "repetitions": 285,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          111.756726316,
          105.60877193,
          110.462719298,
          113.051315789,
          112.914768421,
          112.288157895,
          108.031873684,
          107.056578947,
          116.772512281
        ],
        "latency_us": [
          359.25,
          296.167,
          610.916,
          2236.584,
          426.459,
          530.834,
          304.042,
          286.041,
          365
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 512,
            "reduction_scalar_rounds": 80,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 49152,
              "global_write_bytes": 16384,
              "private_read_bytes": 98304,
              "private_write_bytes": 32768
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 192,
              "global_write_bytes": 64,
              "private_read_bytes": 384,
              "private_write_bytes": 128
            },
            "striped_storage_scalars_per_worker": 32,
            "reduction_operations": 2,
            "reduction_elements": 8192,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 120,
            "normalized_kernel_cost": 1920,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4993500,
              "compute_span_ns": 4993500,
              "calibration_cpu_ns": 6317542,
              "calibration_gpu_ticks": 6317542,
              "compute_passes": 1,
              "command_buffer_ns": 4993708.33859,
              "command_buffers": 2
            },
            {
              "compute_ns": 5120208,
              "compute_span_ns": 5120208,
              "calibration_cpu_ns": 6400875,
              "calibration_gpu_ticks": 6400875,
              "compute_passes": 1,
              "command_buffer_ns": 5120250.04718,
              "command_buffers": 2
            },
            {
              "compute_ns": 5117625,
              "compute_span_ns": 5117625,
              "calibration_cpu_ns": 6023417,
              "calibration_gpu_ticks": 6023417,
              "compute_passes": 1,
              "command_buffer_ns": 5117708.35146,
              "command_buffers": 2
            },
            {
              "compute_ns": 5093833,
              "compute_span_ns": 5093833,
              "calibration_cpu_ns": 6262167,
              "calibration_gpu_ticks": 6262167,
              "compute_passes": 1,
              "command_buffer_ns": 5093874.93366,
              "command_buffers": 2
            },
            {
              "compute_ns": 5278000,
              "compute_span_ns": 5278000,
              "calibration_cpu_ns": 6513208,
              "calibration_gpu_ticks": 6513208,
              "compute_passes": 1,
              "command_buffer_ns": 5278041.64449,
              "command_buffers": 2
            },
            {
              "compute_ns": 4946750,
              "compute_span_ns": 4946750,
              "calibration_cpu_ns": 6452708,
              "calibration_gpu_ticks": 6452708,
              "compute_passes": 1,
              "command_buffer_ns": 4946791.68332,
              "command_buffers": 2
            },
            {
              "compute_ns": 5110833,
              "compute_span_ns": 5110833,
              "calibration_cpu_ns": 6206292,
              "calibration_gpu_ticks": 6206292,
              "compute_passes": 1,
              "command_buffer_ns": 5110958.30007,
              "command_buffers": 2
            },
            {
              "compute_ns": 5240250,
              "compute_span_ns": 5240250,
              "calibration_cpu_ns": 6502667,
              "calibration_gpu_ticks": 6502667,
              "compute_passes": 1,
              "command_buffer_ns": 5240333.32476,
              "command_buffers": 2
            },
            {
              "compute_ns": 7375917,
              "compute_span_ns": 7375917,
              "calibration_cpu_ns": 8565334,
              "calibration_gpu_ticks": 8565334,
              "compute_passes": 1,
              "command_buffer_ns": 7375958.37563,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 113166,
              "compute_span_ns": 113166,
              "calibration_cpu_ns": 1360042,
              "calibration_gpu_ticks": 1360042,
              "compute_passes": 1,
              "command_buffer_ns": 113208.312541,
              "command_buffers": 2
            },
            {
              "compute_ns": 116166,
              "compute_span_ns": 116166,
              "calibration_cpu_ns": 1968333,
              "calibration_gpu_ticks": 1968333,
              "compute_passes": 1,
              "command_buffer_ns": 116124.982014,
              "command_buffers": 2
            },
            {
              "compute_ns": 76792,
              "compute_span_ns": 76792,
              "calibration_cpu_ns": 487333,
              "calibration_gpu_ticks": 487333,
              "compute_passes": 1,
              "command_buffer_ns": 76833.2974985,
              "command_buffers": 2
            },
            {
              "compute_ns": 343250,
              "compute_span_ns": 343250,
              "calibration_cpu_ns": 1766292,
              "calibration_gpu_ticks": 1766292,
              "compute_passes": 1,
              "command_buffer_ns": 343333.347701,
              "command_buffers": 2
            },
            {
              "compute_ns": 77125,
              "compute_span_ns": 77125,
              "calibration_cpu_ns": 530500,
              "calibration_gpu_ticks": 530500,
              "compute_passes": 1,
              "command_buffer_ns": 77166.6527726,
              "command_buffers": 2
            },
            {
              "compute_ns": 111500,
              "compute_span_ns": 111500,
              "calibration_cpu_ns": 736584,
              "calibration_gpu_ticks": 736584,
              "compute_passes": 1,
              "command_buffer_ns": 111583.329272,
              "command_buffers": 2
            },
            {
              "compute_ns": 75959,
              "compute_span_ns": 75959,
              "calibration_cpu_ns": 554875,
              "calibration_gpu_ticks": 554875,
              "compute_passes": 1,
              "command_buffer_ns": 76041.6733101,
              "command_buffers": 2
            },
            {
              "compute_ns": 77416,
              "compute_span_ns": 77416,
              "calibration_cpu_ns": 535708,
              "calibration_gpu_ticks": 535708,
              "compute_passes": 1,
              "command_buffer_ns": 77374.977991,
              "command_buffers": 2
            },
            {
              "compute_ns": 76250,
              "compute_span_ns": 76250,
              "calibration_cpu_ns": 523458,
              "calibration_gpu_ticks": 523458,
              "compute_passes": 1,
              "command_buffer_ns": 76333.2936913,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 5149583.33224,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5306708.33541,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5217000.02719,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5100624.98506,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5269958.34662,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5299166.60162,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5267458.26937,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7649999.98501,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7577624.98688,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 113458.256237,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 75999.9966249,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76375.0285842,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 106916.588265,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 78250.0137575,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76916.7090766,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 111916.684546,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 114333.350211,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 110666.61682,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              80.46223956625,
              82.91731774078124,
              81.51562542484375,
              79.6972653915625,
              82.3430991659375,
              82.7994781503125,
              82.30403545890624,
              119.53124976578125,
              118.40039042
            ],
            "command_buffer_throughput_us_p50": 82.3430991659375,
            "command_buffer_throughput_us_p90": 118.62656228915624,
            "command_buffer_latency_us": [
              113.458256237,
              75.99999662489999,
              76.3750285842,
              106.916588265,
              78.2500137575,
              76.9167090766,
              111.916684546,
              114.333350211,
              110.66661682
            ],
            "command_buffer_latency_us_p50": 106.916588265,
            "command_buffer_latency_us_p90": 113.6332750318
          },
          "compute_throughput_us": [
            78.0234375,
            80.00325,
            79.962890625,
            79.591140625,
            82.46875,
            77.29296875,
            79.856765625,
            81.87890625,
            115.248703125
          ],
          "compute_throughput_us_p50": 79.962890625,
          "compute_throughput_us_p90": 89.024740625,
          "command_buffer_throughput_us": [
            78.02669279046874,
            80.00390698718749,
            79.9641929915625,
            79.59179583843749,
            82.46940069515625,
            77.293620051875,
            79.85872343859374,
            81.880208199375,
            115.24934961921875
          ],
          "command_buffer_throughput_us_p50": 79.9641929915625,
          "command_buffer_throughput_us_p90": 89.02539047996876,
          "compute_latency_us": [
            113.166,
            116.166,
            76.792,
            343.25,
            77.125,
            111.5,
            75.959,
            77.416,
            76.25
          ],
          "compute_latency_us_p50": 77.416,
          "compute_latency_us_p90": 161.58280000000002,
          "command_buffer_latency_us": [
            113.20831254100001,
            116.124982014,
            76.8332974985,
            343.333347701,
            77.1666527726,
            111.583329272,
            76.0416733101,
            77.37497799100001,
            76.3332936913
          ],
          "command_buffer_latency_us_p50": 77.37497799100001,
          "command_buffer_latency_us_p90": 161.56665515140003,
          "counter_control_throughput_ratio": 0.9702892816047468,
          "counter_control_latency_ratio": 0.9977970426808086
        },
        "correctness": {
          "max_abs_error": 2.827305296548843e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 111.756726316,
        "throughput_us_p90": 113.7955550874,
        "latency_us_p50": 365.0,
        "latency_us_p90": 936.0496000000003
      },
      "torch": {
        "cold_call_ms": 7.145875,
        "warmup_ms": 201.251292,
        "repetitions": 102,
        "throughput_us": [
          296.2487745098039,
          305.07884313725486,
          280.6417549019608,
          305.0453431372549,
          305.09845098039216,
          277.302294117647,
          306.4313725490196,
          283.2957549019608,
          292.00407843137253
        ],
        "latency_us": [
          592.5409999999999,
          444.41700000000003,
          434.541,
          440.625,
          449.875,
          1207.333,
          2242.0,
          771.125,
          534.417
        ],
        "throughput_us_p50": 296.2487745098039,
        "throughput_us_p90": 305.36503529411766,
        "latency_us_p50": 534.417,
        "latency_us_p90": 1414.2664000000002,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 17762833.333108574,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17562749.970238656,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 18091916.6887179,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17521000.001579523,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17540708.35886523,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17501374.997664243,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17812374.979257584,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 15116541.646420956,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17681166.70427844,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 200666.6618399322,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 199124.9737329781,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 199041.62036255002,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 198416.70291498303,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200958.3404287696,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 201083.37048441172,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 395875.0166930258,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 199750.00759586692,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 373625.0218935311,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              277.54427082982147,
              274.417968284979,
              282.6861982612172,
              273.76562502468005,
              274.07356810726924,
              273.4589843385038,
              278.31835905089974,
              236.19596322532743,
              276.2682297543506
            ],
            "command_buffer_throughput_us_p50": 274.417968284979,
            "command_buffer_throughput_us_p90": 279.19192689296324,
            "command_buffer_latency_us": [
              200.6666618399322,
              199.1249737329781,
              199.04162036255002,
              198.41670291498303,
              200.9583404287696,
              201.08337048441172,
              395.8750166930258,
              199.75000759586692,
              373.6250218935311
            ],
            "command_buffer_latency_us_p50": 200.6666618399322,
            "command_buffer_latency_us_p90": 378.07502085343003
          },
          "throughput": [
            {
              "compute_ns": 17673792.0,
              "compute_span_ns": 17673792.0,
              "command_buffer_ns": 17673833.353910595,
              "calibration_cpu_ns": 18687666.0,
              "calibration_gpu_ticks": 18687666.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17827083.0,
              "compute_span_ns": 17827083.0,
              "command_buffer_ns": 17827041.67999327,
              "calibration_cpu_ns": 19528583.0,
              "calibration_gpu_ticks": 19528583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17583875.0,
              "compute_span_ns": 17583875.0,
              "command_buffer_ns": 17583875.043783337,
              "calibration_cpu_ns": 18668875.0,
              "calibration_gpu_ticks": 18668875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17810584.0,
              "compute_span_ns": 17810584.0,
              "command_buffer_ns": 17810624.96593222,
              "calibration_cpu_ns": 20294542.0,
              "calibration_gpu_ticks": 20294542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17512000.0,
              "compute_span_ns": 17512000.0,
              "command_buffer_ns": 17511999.991256744,
              "calibration_cpu_ns": 19644791.0,
              "calibration_gpu_ticks": 19644791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17584167.0,
              "compute_span_ns": 17584167.0,
              "command_buffer_ns": 17584208.340849727,
              "calibration_cpu_ns": 19866375.0,
              "calibration_gpu_ticks": 19866375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17540208.0,
              "compute_span_ns": 17540208.0,
              "command_buffer_ns": 17540166.67837277,
              "calibration_cpu_ns": 18714500.0,
              "calibration_gpu_ticks": 18714500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17667584.0,
              "compute_span_ns": 17667584.0,
              "command_buffer_ns": 17667625.041212887,
              "calibration_cpu_ns": 19906333.0,
              "calibration_gpu_ticks": 19906333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17597875.0,
              "compute_span_ns": 17597875.0,
              "command_buffer_ns": 17597874.97576326,
              "calibration_cpu_ns": 18843917.0,
              "calibration_gpu_ticks": 18843917.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 201000.0,
              "compute_span_ns": 201000.0,
              "command_buffer_ns": 200999.95890632272,
              "calibration_cpu_ns": 721292.0,
              "calibration_gpu_ticks": 721292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 199541.0,
              "compute_span_ns": 199541.0,
              "command_buffer_ns": 199500.00569224358,
              "calibration_cpu_ns": 817417.0,
              "calibration_gpu_ticks": 817417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 200750.0,
              "compute_span_ns": 200750.0,
              "command_buffer_ns": 200750.0152103603,
              "calibration_cpu_ns": 684375.0,
              "calibration_gpu_ticks": 684375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 346375.0,
              "compute_span_ns": 346375.0,
              "command_buffer_ns": 346374.9890215695,
              "calibration_cpu_ns": 1712541.0,
              "calibration_gpu_ticks": 1712541.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 314209.0,
              "compute_span_ns": 314209.0,
              "command_buffer_ns": 314250.0063404441,
              "calibration_cpu_ns": 1870000.0,
              "calibration_gpu_ticks": 1870000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 197500.0,
              "compute_span_ns": 197500.0,
              "command_buffer_ns": 197499.99046325684,
              "calibration_cpu_ns": 625292.0,
              "calibration_gpu_ticks": 625292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 197542.0,
              "compute_span_ns": 197542.0,
              "command_buffer_ns": 197541.66714847088,
              "calibration_cpu_ns": 886417.0,
              "calibration_gpu_ticks": 886417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201291.0,
              "compute_span_ns": 201291.0,
              "command_buffer_ns": 201250.01901760697,
              "calibration_cpu_ns": 723792.0,
              "calibration_gpu_ticks": 723792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 197334.0,
              "compute_span_ns": 197334.0,
              "command_buffer_ns": 197375.01861527562,
              "calibration_cpu_ns": 881000.0,
              "calibration_gpu_ticks": 881000.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            276.153,
            278.548171875,
            274.748046875,
            278.290375,
            273.625,
            274.752609375,
            274.06575,
            276.056,
            274.966796875
          ],
          "compute_throughput_us_p50": 274.966796875,
          "compute_throughput_us_p90": 278.341934375,
          "command_buffer_throughput_us": [
            276.15364615485305,
            278.5475262498949,
            274.74804755911464,
            278.29101509269094,
            273.6249998633866,
            274.753255325777,
            274.06510434957454,
            276.05664126895135,
            274.96679649630096
          ],
          "command_buffer_throughput_us_p50": 274.96679649630096,
          "command_buffer_throughput_us_p90": 278.3423173241317,
          "compute_latency_us": [
            201.0,
            199.541,
            200.75,
            346.375,
            314.209,
            197.5,
            197.542,
            201.291,
            197.334
          ],
          "compute_latency_us_p50": 200.75,
          "compute_latency_us_p90": 320.6422,
          "command_buffer_latency_us": [
            200.99995890632272,
            199.50000569224358,
            200.7500152103603,
            346.3749890215695,
            314.2500063404441,
            197.49999046325684,
            197.54166714847088,
            201.25001901760697,
            197.37501861527562
          ],
          "command_buffer_latency_us_p50": 200.7500152103603,
          "command_buffer_latency_us_p90": 320.67500287666917,
          "counter_control_throughput_ratio": 0.9983633290616808,
          "counter_control_latency_ratio": 1.0018833999183265
        },
        "download_ms": 0.717417,
        "allocation_upload_ms": 17.361042,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.1634934849478213e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.3772394552548658,
      "device_compute_slowdown": 0.2908092596407237,
      "valid": true,
      "round": 0,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "layernorm_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.540208,
        "warmup_ms": 201.694375,
        "repetitions": 115,
        "throughput_us": [
          293.6840608695652,
          291.1181217391304,
          292.370652173913,
          289.139852173913,
          292.1231913043478,
          290.91413043478263,
          301.2369565217391,
          293.94347826086954,
          279.6902173913044
        ],
        "latency_us": [
          3266.791,
          1013.833,
          499.667,
          468.125,
          447.625,
          442.083,
          435.459,
          445.0,
          452.667
        ],
        "throughput_us_p50": 292.1231913043478,
        "throughput_us_p90": 295.40217391304344,
        "latency_us_p50": 452.667,
        "latency_us_p90": 1464.4246000000003,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 17582541.680894792,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17721875.046845526,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17533000.034745783,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17618541.66397825,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17534375.01611188,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 15188291.668891907,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17569124.98967722,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17677875.00284612,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17616583.325434476,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 198916.6485145688,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200624.98515471816,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 199208.3271034062,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 199541.68237745762,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 413500.0053793192,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200500.01330673695,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 457500.0493787229,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200250.0114031136,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 343375.0243857503,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              274.7272137639811,
              276.90429760696134,
              273.95312554290285,
              275.28971349966014,
              273.97460962674813,
              237.31705732643604,
              274.5175779637066,
              276.21679691947065,
              275.2591144599137
            ],
            "command_buffer_throughput_us_p50": 274.7272137639811,
            "command_buffer_throughput_us_p90": 276.3542970569688,
            "command_buffer_latency_us": [
              198.9166485145688,
              200.62498515471816,
              199.2083271034062,
              199.54168237745762,
              413.5000053793192,
              200.50001330673695,
              457.5000493787229,
              200.2500114031136,
              343.3750243857503
            ],
            "command_buffer_latency_us_p50": 200.50001330673695,
            "command_buffer_latency_us_p90": 422.30001417919993
          },
          "throughput": [
            {
              "compute_ns": 17524417.0,
              "compute_span_ns": 17524417.0,
              "command_buffer_ns": 17524458.293337375,
              "calibration_cpu_ns": 18591792.0,
              "calibration_gpu_ticks": 18591792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 18474875.0,
              "compute_span_ns": 18474875.0,
              "command_buffer_ns": 18474875.018000603,
              "calibration_cpu_ns": 20139250.0,
              "calibration_gpu_ticks": 20139250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17773875.0,
              "compute_span_ns": 17773875.0,
              "command_buffer_ns": 17773875.035345554,
              "calibration_cpu_ns": 19491708.0,
              "calibration_gpu_ticks": 19491708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17619875.0,
              "compute_span_ns": 17619875.0,
              "command_buffer_ns": 17619874.968659133,
              "calibration_cpu_ns": 18622500.0,
              "calibration_gpu_ticks": 18622500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17686500.0,
              "compute_span_ns": 17686500.0,
              "command_buffer_ns": 17686499.981209636,
              "calibration_cpu_ns": 20070666.0,
              "calibration_gpu_ticks": 20070666.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17855208.0,
              "compute_span_ns": 17855208.0,
              "command_buffer_ns": 17855166.632216424,
              "calibration_cpu_ns": 19343458.0,
              "calibration_gpu_ticks": 19343458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 15110875.0,
              "compute_span_ns": 15110875.0,
              "command_buffer_ns": 15110874.956008047,
              "calibration_cpu_ns": 16267209.0,
              "calibration_gpu_ticks": 16267209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17610958.0,
              "compute_span_ns": 17610958.0,
              "command_buffer_ns": 17610916.635021567,
              "calibration_cpu_ns": 19057958.0,
              "calibration_gpu_ticks": 19057958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17523750.0,
              "compute_span_ns": 17523750.0,
              "command_buffer_ns": 17523750.02251938,
              "calibration_cpu_ns": 18520542.0,
              "calibration_gpu_ticks": 18520542.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 199750.0,
              "compute_span_ns": 199750.0,
              "command_buffer_ns": 199750.00759586692,
              "calibration_cpu_ns": 874750.0,
              "calibration_gpu_ticks": 874750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 199292.0,
              "compute_span_ns": 199292.0,
              "command_buffer_ns": 199291.68047383428,
              "calibration_cpu_ns": 1551917.0,
              "calibration_gpu_ticks": 1551917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 556833.0,
              "compute_span_ns": 556833.0,
              "command_buffer_ns": 556833.3435803652,
              "calibration_cpu_ns": 2006125.0,
              "calibration_gpu_ticks": 2006125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 196958.0,
              "compute_span_ns": 196958.0,
              "command_buffer_ns": 196958.3099707961,
              "calibration_cpu_ns": 1071792.0,
              "calibration_gpu_ticks": 1071792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 197875.0,
              "compute_span_ns": 197875.0,
              "command_buffer_ns": 197874.9642148614,
              "calibration_cpu_ns": 794583.0,
              "calibration_gpu_ticks": 794583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201916.0,
              "compute_span_ns": 201916.0,
              "command_buffer_ns": 201874.99467283487,
              "calibration_cpu_ns": 635417.0,
              "calibration_gpu_ticks": 635417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201375.0,
              "compute_span_ns": 201375.0,
              "command_buffer_ns": 201374.9908655882,
              "calibration_cpu_ns": 1067833.0,
              "calibration_gpu_ticks": 1067833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201291.0,
              "compute_span_ns": 201291.0,
              "command_buffer_ns": 201250.01901760697,
              "calibration_cpu_ns": 619542.0,
              "calibration_gpu_ticks": 619542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 196750.0,
              "compute_span_ns": 196750.0,
              "command_buffer_ns": 196749.9847523868,
              "calibration_cpu_ns": 836541.0,
              "calibration_gpu_ticks": 836541.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            273.819015625,
            288.669921875,
            277.716796875,
            275.310546875,
            276.3515625,
            278.987625,
            236.107421875,
            275.17121875,
            273.80859375
          ],
          "compute_throughput_us_p50": 275.310546875,
          "compute_throughput_us_p90": 280.924084375,
          "command_buffer_throughput_us": [
            273.8196608333965,
            288.6699221562594,
            277.7167974272743,
            275.31054638529895,
            276.35156220640056,
            278.9869786283816,
            236.10742118762573,
            275.170572422212,
            273.8085941018653
          ],
          "command_buffer_throughput_us_p50": 275.31054638529895,
          "command_buffer_throughput_us_p90": 280.9235673339572,
          "compute_latency_us": [
            199.75,
            199.292,
            556.833,
            196.958,
            197.875,
            201.916,
            201.375,
            201.291,
            196.75
          ],
          "compute_latency_us_p50": 199.75,
          "compute_latency_us_p90": 272.89940000000007,
          "command_buffer_latency_us": [
            199.75000759586692,
            199.29168047383428,
            556.8333435803652,
            196.9583099707961,
            197.8749642148614,
            201.87499467283487,
            201.3749908655882,
            201.25001901760697,
            196.7499847523868
          ],
          "command_buffer_latency_us_p50": 199.75000759586692,
          "command_buffer_latency_us_p90": 272.866664454341,
          "counter_control_throughput_ratio": 1.000075676222602,
          "counter_control_latency_ratio": 0.9933542440895102
        },
        "download_ms": 0.625125,
        "allocation_upload_ms": 1.305,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.1634934849478213e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-w7d01w4p/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "6caff4b0a979c3104fea9f23f0d5430392f6f4e7d743dab03e9fdb6de1428bf3",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.829,
        "capture_ms": 0.070042,
        "compile_ms": 64.294209,
        "allocation_upload_ms": 7.037167,
        "cold_call_ms": 5.921334,
        "warmup_ms": 200.98675,
        "download_ms": 3.187459,
        "repetitions": 432,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          92.3878287037,
          89.0200625,
          89.0532407407,
          92.7531828704,
          94.0170717593,
          87.2567523148,
          92.3525277778,
          93.4195601852,
          93.481
        ],
        "latency_us": [
          857.042,
          399.709,
          324.375,
          291.125,
          295.417,
          296.709,
          2829.667,
          716.542,
          397.708
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 1024,
            "reduction_scalar_rounds": 80,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 49152,
              "global_write_bytes": 16384,
              "private_read_bytes": 98304,
              "private_write_bytes": 32768
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 192,
              "global_write_bytes": 64,
              "private_read_bytes": 384,
              "private_write_bytes": 128
            },
            "striped_storage_scalars_per_worker": 32,
            "reduction_operations": 2,
            "reduction_elements": 8192,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 128,
            "normalized_kernel_cost": 2048,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4184625,
              "compute_span_ns": 4184625,
              "calibration_cpu_ns": 5537458,
              "calibration_gpu_ticks": 5537458,
              "compute_passes": 1,
              "command_buffer_ns": 4184708.29027,
              "command_buffers": 2
            },
            {
              "compute_ns": 4173875,
              "compute_span_ns": 4173875,
              "calibration_cpu_ns": 5350375,
              "calibration_gpu_ticks": 5350375,
              "compute_passes": 1,
              "command_buffer_ns": 4173874.97146,
              "command_buffers": 2
            },
            {
              "compute_ns": 4415959,
              "compute_span_ns": 4415959,
              "calibration_cpu_ns": 5641542,
              "calibration_gpu_ticks": 5641542,
              "compute_passes": 1,
              "command_buffer_ns": 4416124.95342,
              "command_buffers": 2
            },
            {
              "compute_ns": 4308209,
              "compute_span_ns": 4308209,
              "calibration_cpu_ns": 5645458,
              "calibration_gpu_ticks": 5645458,
              "compute_passes": 1,
              "command_buffer_ns": 4308291.65271,
              "command_buffers": 2
            },
            {
              "compute_ns": 4488250,
              "compute_span_ns": 4488250,
              "calibration_cpu_ns": 5641500,
              "calibration_gpu_ticks": 5641500,
              "compute_passes": 1,
              "command_buffer_ns": 4488291.68454,
              "command_buffers": 2
            },
            {
              "compute_ns": 4204959,
              "compute_span_ns": 4204959,
              "calibration_cpu_ns": 5820291,
              "calibration_gpu_ticks": 5820291,
              "compute_passes": 1,
              "command_buffer_ns": 4205124.97658,
              "command_buffers": 2
            },
            {
              "compute_ns": 4349042,
              "compute_span_ns": 4349042,
              "calibration_cpu_ns": 5471250,
              "calibration_gpu_ticks": 5471250,
              "compute_passes": 1,
              "command_buffer_ns": 4349083.34864,
              "command_buffers": 2
            },
            {
              "compute_ns": 6544334,
              "compute_span_ns": 6544334,
              "calibration_cpu_ns": 7371875,
              "calibration_gpu_ticks": 7371875,
              "compute_passes": 1,
              "command_buffer_ns": 6544416.65392,
              "command_buffers": 2
            },
            {
              "compute_ns": 6616750,
              "compute_span_ns": 6616750,
              "calibration_cpu_ns": 7778458,
              "calibration_gpu_ticks": 7778458,
              "compute_passes": 1,
              "command_buffer_ns": 6616791.71026,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 63167,
              "compute_span_ns": 63167,
              "calibration_cpu_ns": 481583,
              "calibration_gpu_ticks": 481583,
              "compute_passes": 1,
              "command_buffer_ns": 63250.0159554,
              "command_buffers": 2
            },
            {
              "compute_ns": 331166,
              "compute_span_ns": 331166,
              "calibration_cpu_ns": 1542709,
              "calibration_gpu_ticks": 1542709,
              "compute_passes": 1,
              "command_buffer_ns": 331208.400894,
              "command_buffers": 2
            },
            {
              "compute_ns": 63584,
              "compute_span_ns": 63584,
              "calibration_cpu_ns": 563625,
              "calibration_gpu_ticks": 563625,
              "compute_passes": 1,
              "command_buffer_ns": 63708.2848698,
              "command_buffers": 2
            },
            {
              "compute_ns": 104208,
              "compute_span_ns": 104208,
              "calibration_cpu_ns": 1717542,
              "calibration_gpu_ticks": 1717542,
              "compute_passes": 1,
              "command_buffer_ns": 104208.302218,
              "command_buffers": 2
            },
            {
              "compute_ns": 107417,
              "compute_span_ns": 107417,
              "calibration_cpu_ns": 1140292,
              "calibration_gpu_ticks": 1140292,
              "compute_passes": 1,
              "command_buffer_ns": 107458.326966,
              "command_buffers": 2
            },
            {
              "compute_ns": 63583,
              "compute_span_ns": 63583,
              "calibration_cpu_ns": 634459,
              "calibration_gpu_ticks": 634459,
              "compute_passes": 1,
              "command_buffer_ns": 63583.3130218,
              "command_buffers": 2
            },
            {
              "compute_ns": 63459,
              "compute_span_ns": 63459,
              "calibration_cpu_ns": 580209,
              "calibration_gpu_ticks": 580209,
              "compute_passes": 1,
              "command_buffer_ns": 63541.6363366,
              "command_buffers": 2
            },
            {
              "compute_ns": 103959,
              "compute_span_ns": 103959,
              "calibration_cpu_ns": 725167,
              "calibration_gpu_ticks": 725167,
              "compute_passes": 1,
              "command_buffer_ns": 104000.035208,
              "command_buffers": 2
            },
            {
              "compute_ns": 63250,
              "compute_span_ns": 63250,
              "calibration_cpu_ns": 483000,
              "calibration_gpu_ticks": 483000,
              "compute_passes": 1,
              "command_buffer_ns": 63291.634433,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4232624.95315,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4391500.02785,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4298874.93374,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4326958.26748,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4355541.60504,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4334249.99937,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4481874.98841,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3971083.38075,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4183999.96124,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 66791.6610837,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 97916.694358,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63624.9314994,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63333.3111182,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 104791.659396,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 243666.698225,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 102208.345197,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63541.6363366,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 108666.718006,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              66.13476489296876,
              68.61718793515625,
              67.1699208396875,
              67.608722929375,
              68.05533757875,
              67.72265624015625,
              70.02929669390625,
              62.04817782421875,
              65.374999394375
            ],
            "command_buffer_throughput_us_p50": 67.608722929375,
            "command_buffer_throughput_us_p90": 68.89960968690625,
            "command_buffer_latency_us": [
              66.7916610837,
              97.91669435799999,
              63.6249314994,
              63.3333111182,
              104.791659396,
              243.666698225,
              102.208345197,
              63.5416363366,
              108.666718006
            ],
            "command_buffer_latency_us_p50": 97.91669435799999,
            "command_buffer_latency_us_p90": 135.66671404980002
          },
          "compute_throughput_us": [
            65.384765625,
            65.216796875,
            68.999359375,
            67.315765625,
            70.12890625,
            65.702484375,
            67.95378125,
            102.25521875,
            103.38671875
          ],
          "compute_throughput_us_p50": 67.95378125,
          "compute_throughput_us_p90": 102.48151874999999,
          "command_buffer_throughput_us": [
            65.38606703546876,
            65.2167964290625,
            69.00195239718751,
            67.31705707359374,
            70.1295575709375,
            65.70507775906249,
            67.95442732250001,
            102.25651021750001,
            103.3873704728125
          ],
          "command_buffer_throughput_us_p50": 67.95442732250001,
          "command_buffer_throughput_us_p90": 102.48268226856251,
          "compute_latency_us": [
            63.167,
            331.166,
            63.584,
            104.208,
            107.417,
            63.583,
            63.459,
            103.959,
            63.25
          ],
          "compute_latency_us_p50": 63.584,
          "compute_latency_us_p90": 152.16680000000002,
          "command_buffer_latency_us": [
            63.250015955399995,
            331.208400894,
            63.7082848698,
            104.208302218,
            107.45832696599999,
            63.5833130218,
            63.5416363366,
            104.000035208,
            63.291634433
          ],
          "command_buffer_latency_us_p50": 63.7082848698,
          "command_buffer_latency_us_p90": 152.20834175160002,
          "counter_control_throughput_ratio": 0.9956859730054962,
          "counter_control_latency_ratio": 1.0013100740297187
        },
        "correctness": {
          "max_abs_error": 2.827305296548843e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 92.3878287037,
        "throughput_us_p90": 93.58821435185999,
        "latency_us_p50": 397.708,
        "latency_us_p90": 1251.5670000000005
      },
      "slowdown": 0.31626324596545285,
      "device_compute_slowdown": 0.24682592810675447,
      "valid": true,
      "round": 0,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "softmax_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-y15zbm72/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "0d8a82e58cbe97c43d15ef559cf043ad575745e0d3fb0f4f40ed60b2caca21c0",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.936708,
        "capture_ms": 0.07075,
        "compile_ms": 54.144209,
        "allocation_upload_ms": 1.759708,
        "cold_call_ms": 7.340375,
        "warmup_ms": 200.163125,
        "download_ms": 0.91275,
        "repetitions": 1238,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          19.8515751212,
          19.8722399031,
          19.8460549273,
          20.1513861066,
          20.7755783522,
          19.8934442649,
          19.8677641357,
          19.7341147011,
          19.3713311793
        ],
        "latency_us": [
          292.917,
          421.041,
          374.25,
          210.375,
          230,
          315.042,
          219.209,
          207.958,
          240.583
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 256,
            "reduction_scalar_rounds": 60,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 12288,
              "global_write_bytes": 12288,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 48,
              "global_write_bytes": 48,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 2,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 108,
            "normalized_kernel_cost": 432,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 9216,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 879958,
              "compute_span_ns": 879958,
              "calibration_cpu_ns": 1345000,
              "calibration_gpu_ticks": 1345000,
              "compute_passes": 1,
              "command_buffer_ns": 879999.948665,
              "command_buffers": 2
            },
            {
              "compute_ns": 1310000,
              "compute_span_ns": 1310000,
              "calibration_cpu_ns": 2382208,
              "calibration_gpu_ticks": 2382208,
              "compute_passes": 1,
              "command_buffer_ns": 1310083.37485,
              "command_buffers": 2
            },
            {
              "compute_ns": 859333,
              "compute_span_ns": 859333,
              "calibration_cpu_ns": 1425042,
              "calibration_gpu_ticks": 1425042,
              "compute_passes": 1,
              "command_buffer_ns": 859333.318658,
              "command_buffers": 2
            },
            {
              "compute_ns": 1314708,
              "compute_span_ns": 1314708,
              "calibration_cpu_ns": 2718916,
              "calibration_gpu_ticks": 2718916,
              "compute_passes": 1,
              "command_buffer_ns": 1314791.67612,
              "command_buffers": 2
            },
            {
              "compute_ns": 854792,
              "compute_span_ns": 854792,
              "calibration_cpu_ns": 1343208,
              "calibration_gpu_ticks": 1343208,
              "compute_passes": 1,
              "command_buffer_ns": 854875.019286,
              "command_buffers": 2
            },
            {
              "compute_ns": 1277167,
              "compute_span_ns": 1277167,
              "calibration_cpu_ns": 2244458,
              "calibration_gpu_ticks": 2244458,
              "compute_passes": 1,
              "command_buffer_ns": 1277250.00493,
              "command_buffers": 2
            },
            {
              "compute_ns": 863042,
              "compute_span_ns": 863042,
              "calibration_cpu_ns": 1379916,
              "calibration_gpu_ticks": 1379916,
              "compute_passes": 1,
              "command_buffer_ns": 863166.584168,
              "command_buffers": 2
            },
            {
              "compute_ns": 1378292,
              "compute_span_ns": 1378292,
              "calibration_cpu_ns": 2804334,
              "calibration_gpu_ticks": 2804334,
              "compute_passes": 1,
              "command_buffer_ns": 1378374.98914,
              "command_buffers": 2
            },
            {
              "compute_ns": 855041,
              "compute_span_ns": 855041,
              "calibration_cpu_ns": 1429000,
              "calibration_gpu_ticks": 1429000,
              "compute_passes": 1,
              "command_buffer_ns": 855083.344504,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 17916,
              "compute_span_ns": 17916,
              "calibration_cpu_ns": 835916,
              "calibration_gpu_ticks": 835916,
              "compute_passes": 1,
              "command_buffer_ns": 17999.9624379,
              "command_buffers": 2
            },
            {
              "compute_ns": 15917,
              "compute_span_ns": 15917,
              "calibration_cpu_ns": 411792,
              "calibration_gpu_ticks": 411792,
              "compute_passes": 1,
              "command_buffer_ns": 16041.6238941,
              "command_buffers": 2
            },
            {
              "compute_ns": 94000,
              "compute_span_ns": 94000,
              "calibration_cpu_ns": 1088083,
              "calibration_gpu_ticks": 1088083,
              "compute_passes": 1,
              "command_buffer_ns": 94041.6939557,
              "command_buffers": 2
            },
            {
              "compute_ns": 19500,
              "compute_span_ns": 19500,
              "calibration_cpu_ns": 890542,
              "calibration_gpu_ticks": 890542,
              "compute_passes": 1,
              "command_buffer_ns": 19625.0039153,
              "command_buffers": 2
            },
            {
              "compute_ns": 54500,
              "compute_span_ns": 54500,
              "calibration_cpu_ns": 1113041,
              "calibration_gpu_ticks": 1113041,
              "compute_passes": 1,
              "command_buffer_ns": 54541.6842215,
              "command_buffers": 2
            },
            {
              "compute_ns": 20000,
              "compute_span_ns": 20000,
              "calibration_cpu_ns": 834750,
              "calibration_gpu_ticks": 834750,
              "compute_passes": 1,
              "command_buffer_ns": 20000.0358745,
              "command_buffers": 2
            },
            {
              "compute_ns": 18042,
              "compute_span_ns": 18042,
              "calibration_cpu_ns": 419917,
              "calibration_gpu_ticks": 419917,
              "compute_passes": 1,
              "command_buffer_ns": 18166.6691788,
              "command_buffers": 2
            },
            {
              "compute_ns": 73791,
              "compute_span_ns": 73791,
              "calibration_cpu_ns": 992750,
              "calibration_gpu_ticks": 992750,
              "compute_passes": 1,
              "command_buffer_ns": 73833.274655,
              "command_buffers": 2
            },
            {
              "compute_ns": 16000,
              "compute_span_ns": 16000,
              "calibration_cpu_ns": 453208,
              "calibration_gpu_ticks": 453208,
              "compute_passes": 1,
              "command_buffer_ns": 16083.358787,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1101583.30062,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 873749.959283,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1229333.34205,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 858249.957673,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1257500.02917,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 870999.99655,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1375125.02261,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 863333.349116,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1325083.37265,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 110208.289698,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 17833.3139047,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18083.374016,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 15791.6801982,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18166.6691788,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 15791.6219905,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18374.9943972,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20125.0077225,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 16041.6821018,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              17.2122390721875,
              13.652343113796874,
              19.20833346953125,
              13.410155588640626,
              19.64843795578125,
              13.60937494609375,
              21.48632847828125,
              13.4895835799375,
              20.70442769765625
            ],
            "command_buffer_throughput_us_p50": 17.2122390721875,
            "command_buffer_throughput_us_p90": 20.86080785378125,
            "command_buffer_latency_us": [
              110.20828969799999,
              17.8333139047,
              18.083374016,
              15.7916801982,
              18.1666691788,
              15.7916219905,
              18.374994397200002,
              20.125007722499998,
              16.0416821018
            ],
            "command_buffer_latency_us_p50": 18.083374016,
            "command_buffer_latency_us_p90": 38.141664117600016
          },
          "compute_throughput_us": [
            13.74934375,
            20.46875,
            13.427078125,
            20.5423125,
            13.356125,
            19.955734375,
            13.48503125,
            21.5358125,
            13.360015625
          ],
          "compute_throughput_us_p50": 13.74934375,
          "compute_throughput_us_p90": 20.7410125,
          "command_buffer_throughput_us": [
            13.749999197890626,
            20.47005273203125,
            13.427083104031249,
            20.543619939375,
            13.35742217634375,
            19.957031327031252,
            13.486977877625,
            21.5371092053125,
            13.360677257875
          ],
          "command_buffer_throughput_us_p50": 13.749999197890626,
          "command_buffer_throughput_us_p90": 20.7423177925625,
          "compute_latency_us": [
            17.916,
            15.917,
            94.0,
            19.5,
            54.5,
            20.0,
            18.042,
            73.791,
            16.0
          ],
          "compute_latency_us_p50": 19.5,
          "compute_latency_us_p90": 77.8328,
          "command_buffer_latency_us": [
            17.999962437900002,
            16.0416238941,
            94.0416939557,
            19.6250039153,
            54.5416842215,
            20.0000358745,
            18.1666691788,
            73.833274655,
            16.083358786999998
          ],
          "command_buffer_latency_us_p50": 19.6250039153,
          "command_buffer_latency_us_p90": 77.87495851514,
          "counter_control_throughput_ratio": 0.7988501170721387,
          "counter_control_latency_ratio": 1.2427432463796309
        },
        "correctness": {
          "max_abs_error": 1.569250362249905e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 19.8677641357,
        "throughput_us_p90": 20.27622455572,
        "latency_us_p50": 240.583,
        "latency_us_p90": 383.6082
      },
      "torch": {
        "cold_call_ms": 10.773458,
        "warmup_ms": 200.732458,
        "repetitions": 711,
        "throughput_us": [
          46.06000984528832,
          46.20739662447258,
          45.99185372714487,
          42.79395218002813,
          44.67035864978903,
          45.772971870604785,
          46.16789732770746,
          46.016994374120955,
          46.00914205344585
        ],
        "latency_us": [
          520.792,
          311.209,
          361.917,
          318.583,
          354.5,
          344.167,
          323.542,
          337.583,
          347.45799999999997
        ],
        "throughput_us_p50": 46.00914205344585,
        "throughput_us_p90": 46.17579718706048,
        "latency_us_p50": 344.167,
        "latency_us_p90": 393.692,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1914208.2892358303,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1929333.2006782293,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 2029958.413913846,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1943208.2772254944,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1974333.48512277,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1944625.0516921282,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1880791.5039360523,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1934541.6221767664,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1933958.2649990916,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 65750.03499165177,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 47624.98429045081,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 154750.01418963075,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 48250.018153339624,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 84458.32645520568,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 57791.70896857977,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 47208.33385363221,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 46583.35819840431,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 47750.01434609294,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              29.90950451930985,
              30.145831260597333,
              31.718100217403844,
              30.36262933164835,
              30.848960705043282,
              30.384766432689503,
              29.387367249000818,
              30.227212846511975,
              30.218097890610807
            ],
            "command_buffer_throughput_us_p50": 30.227212846511975,
            "command_buffer_throughput_us_p90": 31.022788607515395,
            "command_buffer_latency_us": [
              65.75003499165177,
              47.62498429045081,
              154.75001418963075,
              48.250018153339624,
              84.45832645520568,
              57.79170896857977,
              47.20833385363221,
              46.58335819840431,
              47.75001434609294
            ],
            "command_buffer_latency_us_p50": 48.250018153339624,
            "command_buffer_latency_us_p90": 98.5166640020907
          },
          "throughput": [
            {
              "compute_ns": 3068088.0,
              "compute_span_ns": 4148250.0,
              "command_buffer_ns": 3774749.9882243574,
              "calibration_cpu_ns": 6008875.0,
              "calibration_gpu_ticks": 6008875.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3028083.0,
              "compute_span_ns": 4148583.0,
              "command_buffer_ns": 3996000.04311651,
              "calibration_cpu_ns": 5471167.0,
              "calibration_gpu_ticks": 5471167.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3001501.0,
              "compute_span_ns": 4355916.0,
              "command_buffer_ns": 4198333.3067037165,
              "calibration_cpu_ns": 6089875.0,
              "calibration_gpu_ticks": 6089875.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3039175.0,
              "compute_span_ns": 3931417.0,
              "command_buffer_ns": 3793208.335991949,
              "calibration_cpu_ns": 5295750.0,
              "calibration_gpu_ticks": 5295750.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3030503.0,
              "compute_span_ns": 4149458.0,
              "command_buffer_ns": 3994333.4413692355,
              "calibration_cpu_ns": 5907541.0,
              "calibration_gpu_ticks": 5907541.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2993081.0,
              "compute_span_ns": 4182125.0,
              "command_buffer_ns": 4013583.2969099283,
              "calibration_cpu_ns": 5883000.0,
              "calibration_gpu_ticks": 5883000.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3030000.0,
              "compute_span_ns": 4120667.0,
              "command_buffer_ns": 3955166.6122861207,
              "calibration_cpu_ns": 5730292.0,
              "calibration_gpu_ticks": 5730292.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3043084.0,
              "compute_span_ns": 4081417.0,
              "command_buffer_ns": 3824166.429694742,
              "calibration_cpu_ns": 5548042.0,
              "calibration_gpu_ticks": 5548042.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3075999.0,
              "compute_span_ns": 3943250.0,
              "command_buffer_ns": 3801333.368755877,
              "calibration_cpu_ns": 5822750.0,
              "calibration_gpu_ticks": 5822750.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 48416.0,
              "compute_span_ns": 48416.0,
              "command_buffer_ns": 48374.99000132084,
              "calibration_cpu_ns": 960125.0,
              "calibration_gpu_ticks": 960125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47334.0,
              "compute_span_ns": 47334.0,
              "command_buffer_ns": 47374.98238682747,
              "calibration_cpu_ns": 591208.0,
              "calibration_gpu_ticks": 591208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49583.0,
              "compute_span_ns": 49583.0,
              "command_buffer_ns": 49583.32283422351,
              "calibration_cpu_ns": 795084.0,
              "calibration_gpu_ticks": 795084.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 46958.0,
              "compute_span_ns": 46958.0,
              "command_buffer_ns": 46916.71347245574,
              "calibration_cpu_ns": 584209.0,
              "calibration_gpu_ticks": 584209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47500.0,
              "compute_span_ns": 47500.0,
              "command_buffer_ns": 47499.95423480868,
              "calibration_cpu_ns": 758125.0,
              "calibration_gpu_ticks": 758125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47708.0,
              "compute_span_ns": 47708.0,
              "command_buffer_ns": 47708.3376608789,
              "calibration_cpu_ns": 583667.0,
              "calibration_gpu_ticks": 583667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47875.0,
              "compute_span_ns": 47875.0,
              "command_buffer_ns": 47875.04440173507,
              "calibration_cpu_ns": 674791.0,
              "calibration_gpu_ticks": 674791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 51208.0,
              "compute_span_ns": 51208.0,
              "command_buffer_ns": 51166.68762639165,
              "calibration_cpu_ns": 2360875.0,
              "calibration_gpu_ticks": 2360875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 52375.0,
              "compute_span_ns": 52375.0,
              "command_buffer_ns": 52375.02045929432,
              "calibration_cpu_ns": 2462750.0,
              "calibration_gpu_ticks": 2462750.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            47.938875,
            47.313796875,
            46.898453125,
            47.487109375,
            47.351609375,
            46.766890625,
            47.34375,
            47.5481875,
            48.062484375
          ],
          "compute_throughput_us_p50": 47.351609375,
          "compute_throughput_us_p90": 47.963596875,
          "command_buffer_throughput_us": [
            58.980468566005584,
            62.43750067369547,
            65.59895791724557,
            59.2688802498742,
            62.411460021394305,
            62.71223901421763,
            61.799478316970635,
            59.75260046398034,
            59.39583388681058
          ],
          "command_buffer_throughput_us_p50": 61.799478316970635,
          "command_buffer_throughput_us_p90": 63.28958279482322,
          "compute_latency_us": [
            48.416,
            47.334,
            49.583,
            46.958,
            47.5,
            47.708,
            47.875,
            51.208,
            52.375
          ],
          "compute_latency_us_p50": 47.875,
          "compute_latency_us_p90": 51.4414,
          "command_buffer_latency_us": [
            48.37499000132084,
            47.37498238682747,
            49.58332283422351,
            46.91671347245574,
            47.49995423480868,
            47.7083376608789,
            47.87504440173507,
            51.16668762639165,
            52.37502045929432
          ],
          "command_buffer_latency_us_p50": 47.87504440173507,
          "command_buffer_latency_us_p90": 51.40835419297218,
          "counter_control_throughput_ratio": 2.0231300697008923,
          "counter_control_latency_ratio": 0.9723667527616962
        },
        "download_ms": 0.482583,
        "allocation_upload_ms": 1.167417,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 2.673614807194649e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.43182209554398776,
      "device_compute_slowdown": 0.2903669786830767,
      "valid": true,
      "round": 1,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "softmax_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.521167,
        "warmup_ms": 201.495167,
        "repetitions": 571,
        "throughput_us": [
          44.8206357267951,
          45.45913485113836,
          45.096322241681264,
          43.32100175131348,
          43.92155516637479,
          45.07428371278459,
          45.00211558669002,
          45.88178633975482,
          44.64674605954466
        ],
        "latency_us": [
          510.583,
          351.66700000000003,
          334.541,
          323.167,
          311.917,
          328.70899999999995,
          382.91700000000003,
          349.625,
          314.0
        ],
        "throughput_us_p50": 45.00211558669002,
        "throughput_us_p90": 45.54366514886165,
        "latency_us_p50": 334.541,
        "latency_us_p90": 408.45020000000005,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1900000.0320374966,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1933791.6164658964,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1967916.6143760085,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1930041.5879115462,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1937500.0847503543,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1908625.1850239933,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1928000.128827989,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1966999.9601319432,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1907833.3862125874,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 52249.99040365219,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 65375.0030323863,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 50249.97517466545,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 124500.01668184996,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 48291.63663089275,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 62624.982092529535,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 51791.66328161955,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 145166.70489683747,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49500.02767145634,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              29.687500500585884,
              30.21549400727963,
              30.748697099625133,
              30.15689981111791,
              30.273438824224286,
              29.822268515999895,
              30.12500201293733,
              30.734374377061613,
              29.809896659571677
            ],
            "command_buffer_throughput_us_p50": 30.15689981111791,
            "command_buffer_throughput_us_p90": 30.737238921574317,
            "command_buffer_latency_us": [
              52.24999040365219,
              65.3750030323863,
              50.24997517466545,
              124.50001668184996,
              48.291636630892754,
              62.624982092529535,
              51.79166328161955,
              145.16670489683747,
              49.50002767145634
            ],
            "command_buffer_latency_us_p50": 52.24999040365219,
            "command_buffer_latency_us_p90": 128.63335432484746
          },
          "throughput": [
            {
              "compute_ns": 2998250.0,
              "compute_span_ns": 4190417.0,
              "command_buffer_ns": 3906333.411578089,
              "calibration_cpu_ns": 5758042.0,
              "calibration_gpu_ticks": 5758042.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3038873.0,
              "compute_span_ns": 4012709.0,
              "command_buffer_ns": 3877833.3691880107,
              "calibration_cpu_ns": 5085083.0,
              "calibration_gpu_ticks": 5085083.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3017998.0,
              "compute_span_ns": 4334333.0,
              "command_buffer_ns": 3994458.355009556,
              "calibration_cpu_ns": 5457166.0,
              "calibration_gpu_ticks": 5457166.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2985168.0,
              "compute_span_ns": 4153042.0,
              "command_buffer_ns": 4006166.651379317,
              "calibration_cpu_ns": 5469792.0,
              "calibration_gpu_ticks": 5469792.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3011582.0,
              "compute_span_ns": 3829458.0,
              "command_buffer_ns": 3703374.939505011,
              "calibration_cpu_ns": 6347916.0,
              "calibration_gpu_ticks": 6347916.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3095584.0,
              "compute_span_ns": 4213250.0,
              "command_buffer_ns": 3954000.014346093,
              "calibration_cpu_ns": 5329834.0,
              "calibration_gpu_ticks": 5329834.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3037713.0,
              "compute_span_ns": 4059292.0,
              "command_buffer_ns": 3763166.6054949164,
              "calibration_cpu_ns": 5586875.0,
              "calibration_gpu_ticks": 5586875.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3014038.0,
              "compute_span_ns": 3975500.0,
              "command_buffer_ns": 3792666.713707149,
              "calibration_cpu_ns": 5271000.0,
              "calibration_gpu_ticks": 5271000.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3036746.0,
              "compute_span_ns": 4057625.0,
              "command_buffer_ns": 3878375.1078881323,
              "calibration_cpu_ns": 5208083.0,
              "calibration_gpu_ticks": 5208083.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 49417.0,
              "compute_span_ns": 49417.0,
              "command_buffer_ns": 49416.67430102825,
              "calibration_cpu_ns": 792916.0,
              "calibration_gpu_ticks": 792916.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49625.0,
              "compute_span_ns": 49625.0,
              "command_buffer_ns": 49624.99951943755,
              "calibration_cpu_ns": 650458.0,
              "calibration_gpu_ticks": 650458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 46250.0,
              "compute_span_ns": 46250.0,
              "command_buffer_ns": 46250.002924352884,
              "calibration_cpu_ns": 615667.0,
              "calibration_gpu_ticks": 615667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47791.0,
              "compute_span_ns": 47791.0,
              "command_buffer_ns": 47750.01434609294,
              "calibration_cpu_ns": 542166.0,
              "calibration_gpu_ticks": 542166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 46959.0,
              "compute_span_ns": 46959.0,
              "command_buffer_ns": 47000.00863522291,
              "calibration_cpu_ns": 496125.0,
              "calibration_gpu_ticks": 496125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47875.0,
              "compute_span_ns": 47875.0,
              "command_buffer_ns": 47875.04440173507,
              "calibration_cpu_ns": 655792.0,
              "calibration_gpu_ticks": 655792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49833.0,
              "compute_span_ns": 49833.0,
              "command_buffer_ns": 49791.64805263281,
              "calibration_cpu_ns": 617583.0,
              "calibration_gpu_ticks": 617583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47875.0,
              "compute_span_ns": 47875.0,
              "command_buffer_ns": 47874.986194074154,
              "calibration_cpu_ns": 514083.0,
              "calibration_gpu_ticks": 514083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49167.0,
              "compute_span_ns": 49167.0,
              "command_buffer_ns": 49166.67239740491,
              "calibration_cpu_ns": 525792.0,
              "calibration_gpu_ticks": 525792.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            46.84765625,
            47.482390625,
            47.15621875,
            46.64325,
            47.05596875,
            48.3685,
            47.464265625,
            47.09434375,
            47.44915625
          ],
          "compute_throughput_us_p50": 47.15621875,
          "compute_throughput_us_p90": 47.6596125,
          "command_buffer_throughput_us": [
            61.03645955590764,
            60.59114639356267,
            62.41341179702431,
            62.59635392780183,
            57.865233429765794,
            61.7812502241577,
            58.79947821085807,
            59.260417401674204,
            60.59961106075207
          ],
          "command_buffer_throughput_us_p50": 60.59961106075207,
          "command_buffer_throughput_us_p90": 62.45000022317981,
          "compute_latency_us": [
            49.417,
            49.625,
            46.25,
            47.791,
            46.959,
            47.875,
            49.833,
            47.875,
            49.167
          ],
          "compute_latency_us_p50": 47.875,
          "compute_latency_us_p90": 49.6666,
          "command_buffer_latency_us": [
            49.41667430102825,
            49.62499951943755,
            46.250002924352884,
            47.75001434609294,
            47.00000863522291,
            47.87504440173507,
            49.79164805263281,
            47.874986194074154,
            49.16667239740491
          ],
          "command_buffer_latency_us_p50": 47.87504440173507,
          "command_buffer_latency_us_p90": 49.6583292260766,
          "counter_control_throughput_ratio": 2.0297904524151433,
          "counter_control_latency_ratio": 0.9203985228567987
        },
        "download_ms": 0.379792,
        "allocation_upload_ms": 0.587209,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 2.673614807194649e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-umdlrqkp/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "7bac43b2781c01105a8e678fa6161555c0e89cbd2b0a98679fdc42bb1bd74b45",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.143959,
        "capture_ms": 0.075416,
        "compile_ms": 53.556459,
        "allocation_upload_ms": 4.087833,
        "cold_call_ms": 7.578791,
        "warmup_ms": 201.014459,
        "download_ms": 0.794333,
        "repetitions": 1226,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          22.5713026101,
          24.5214787928,
          22.2316476346,
          24.8756460033,
          22.5319127243,
          24.9492936378,
          22.4406264274,
          24.2394306688,
          22.4942903752
        ],
        "latency_us": [
          352.833,
          251.041,
          256.792,
          248.125,
          224.625,
          234.209,
          333.75,
          2125.666,
          942.459
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 128,
            "reduction_scalar_rounds": 60,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 12288,
              "global_write_bytes": 12288,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 48,
              "global_write_bytes": 48,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 2,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 100,
            "normalized_kernel_cost": 400,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 9216,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 948208,
              "compute_span_ns": 948208,
              "calibration_cpu_ns": 3000542,
              "calibration_gpu_ticks": 3000542,
              "compute_passes": 1,
              "command_buffer_ns": 948291.679379,
              "command_buffers": 2
            },
            {
              "compute_ns": 944875,
              "compute_span_ns": 944875,
              "calibration_cpu_ns": 3102709,
              "calibration_gpu_ticks": 3102709,
              "compute_passes": 1,
              "command_buffer_ns": 944958.301261,
              "command_buffers": 2
            },
            {
              "compute_ns": 937250,
              "compute_span_ns": 937250,
              "calibration_cpu_ns": 2854292,
              "calibration_gpu_ticks": 2854292,
              "compute_passes": 1,
              "command_buffer_ns": 937333.38872,
              "command_buffers": 2
            },
            {
              "compute_ns": 1271875,
              "compute_span_ns": 1271875,
              "calibration_cpu_ns": 2905917,
              "calibration_gpu_ticks": 2905917,
              "compute_passes": 1,
              "command_buffer_ns": 1271958.34648,
              "command_buffers": 2
            },
            {
              "compute_ns": 1301041,
              "compute_span_ns": 1301041,
              "calibration_cpu_ns": 2727958,
              "calibration_gpu_ticks": 2727958,
              "compute_passes": 1,
              "command_buffer_ns": 1301041.68784,
              "command_buffers": 2
            },
            {
              "compute_ns": 922250,
              "compute_span_ns": 922250,
              "calibration_cpu_ns": 1392333,
              "calibration_gpu_ticks": 1392333,
              "compute_passes": 1,
              "command_buffer_ns": 922333.33271,
              "command_buffers": 2
            },
            {
              "compute_ns": 909625,
              "compute_span_ns": 909625,
              "calibration_cpu_ns": 3164500,
              "calibration_gpu_ticks": 3164500,
              "compute_passes": 1,
              "command_buffer_ns": 909750.000574,
              "command_buffers": 2
            },
            {
              "compute_ns": 1448375,
              "compute_span_ns": 1448375,
              "calibration_cpu_ns": 2973292,
              "calibration_gpu_ticks": 2973292,
              "compute_passes": 1,
              "command_buffer_ns": 1448458.35166,
              "command_buffers": 2
            },
            {
              "compute_ns": 928416,
              "compute_span_ns": 928416,
              "calibration_cpu_ns": 1464875,
              "calibration_gpu_ticks": 1464875,
              "compute_passes": 1,
              "command_buffer_ns": 928458.350245,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 16334,
              "compute_span_ns": 16334,
              "calibration_cpu_ns": 601792,
              "calibration_gpu_ticks": 601792,
              "compute_passes": 1,
              "command_buffer_ns": 16458.3907463,
              "command_buffers": 2
            },
            {
              "compute_ns": 51708,
              "compute_span_ns": 51708,
              "calibration_cpu_ns": 2133916,
              "calibration_gpu_ticks": 2133916,
              "compute_passes": 1,
              "command_buffer_ns": 51749.9283887,
              "command_buffers": 2
            },
            {
              "compute_ns": 24708,
              "compute_span_ns": 24708,
              "calibration_cpu_ns": 2540083,
              "calibration_gpu_ticks": 2540083,
              "compute_passes": 1,
              "command_buffer_ns": 24874.9856837,
              "command_buffers": 2
            },
            {
              "compute_ns": 16292,
              "compute_span_ns": 16292,
              "calibration_cpu_ns": 543625,
              "calibration_gpu_ticks": 543625,
              "compute_passes": 1,
              "command_buffer_ns": 16333.3606906,
              "command_buffers": 2
            },
            {
              "compute_ns": 18500,
              "compute_span_ns": 18500,
              "calibration_cpu_ns": 461542,
              "calibration_gpu_ticks": 461542,
              "compute_passes": 1,
              "command_buffer_ns": 18541.6429304,
              "command_buffers": 2
            },
            {
              "compute_ns": 28416,
              "compute_span_ns": 28416,
              "calibration_cpu_ns": 2371750,
              "calibration_gpu_ticks": 2371750,
              "compute_passes": 1,
              "command_buffer_ns": 28458.3074972,
              "command_buffers": 2
            },
            {
              "compute_ns": 32834,
              "compute_span_ns": 32834,
              "calibration_cpu_ns": 1552417,
              "calibration_gpu_ticks": 1552417,
              "compute_passes": 1,
              "command_buffer_ns": 32958.2835548,
              "command_buffers": 2
            },
            {
              "compute_ns": 19291,
              "compute_span_ns": 19291,
              "calibration_cpu_ns": 725833,
              "calibration_gpu_ticks": 725833,
              "compute_passes": 1,
              "command_buffer_ns": 19333.3835341,
              "command_buffers": 2
            },
            {
              "compute_ns": 20083,
              "compute_span_ns": 20083,
              "calibration_cpu_ns": 607125,
              "calibration_gpu_ticks": 607125,
              "compute_passes": 1,
              "command_buffer_ns": 20083.3310373,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 971833.302174,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1407166.65192,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 928208.290134,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1498208.32303,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 912208.342925,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 930791.66254,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 932333.350647,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 920833.379496,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1290916.69805,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 23375.0324696,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18000.0206456,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18750.0263564,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 24166.7148657,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 16458.3325386,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 16208.330635,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18708.4078789,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18250.0225492,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 23124.9723583,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              15.18489534646875,
              21.98697893625,
              14.50325453334375,
              23.40950504734375,
              14.253255358203125,
              14.543619727187501,
              14.567708603859375,
              14.388021554625,
              20.17057340703125
            ],
            "command_buffer_throughput_us_p50": 14.567708603859375,
            "command_buffer_throughput_us_p90": 22.271484158468752,
            "command_buffer_latency_us": [
              23.375032469599997,
              18.000020645600003,
              18.750026356400003,
              24.1667148657,
              16.4583325386,
              16.208330635,
              18.7084078789,
              18.2500225492,
              23.124972358300003
            ],
            "command_buffer_latency_us_p50": 18.7084078789,
            "command_buffer_latency_us_p90": 23.533368948819998
          },
          "compute_throughput_us": [
            14.81575,
            14.763671875,
            14.64453125,
            19.873046875,
            20.328765625,
            14.41015625,
            14.212890625,
            22.630859375,
            14.5065
          ],
          "compute_throughput_us_p50": 14.763671875,
          "compute_throughput_us_p90": 20.789184375,
          "command_buffer_throughput_us": [
            14.817057490296875,
            14.764973457203125,
            14.64583419875,
            19.87434916375,
            20.328776372500002,
            14.41145832359375,
            14.21484375896875,
            22.632161744687497,
            14.507161722578125
          ],
          "command_buffer_throughput_us_p50": 14.764973457203125,
          "command_buffer_throughput_us_p90": 20.7894534469375,
          "compute_latency_us": [
            16.334,
            51.708,
            24.708,
            16.292,
            18.5,
            28.416,
            32.834,
            19.291,
            20.083
          ],
          "compute_latency_us_p50": 20.083,
          "compute_latency_us_p90": 36.6088,
          "command_buffer_latency_us": [
            16.4583907463,
            51.7499283887,
            24.8749856837,
            16.3333606906,
            18.541642930400002,
            28.4583074972,
            32.9582835548,
            19.3333835341,
            20.0833310373
          ],
          "command_buffer_latency_us_p50": 20.0833310373,
          "command_buffer_latency_us_p90": 36.716612521580004,
          "counter_control_throughput_ratio": 0.9757776013725904,
          "counter_control_latency_ratio": 1.1265808906773502
        },
        "correctness": {
          "max_abs_error": 1.569250362249905e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 22.5713026101,
        "throughput_us_p90": 24.8903755302,
        "latency_us_p50": 256.792,
        "latency_us_p90": 1179.1004000000003
      },
      "slowdown": 0.5015609225441785,
      "device_compute_slowdown": 0.3130800616832748,
      "valid": true,
      "round": 1,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "softmax_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-027p_2sf/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "5e168786762f89acc561194a669a3725da4fe77bed5800b1d573576d096bb3fc",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.600667,
        "capture_ms": 0.061167,
        "compile_ms": 55.937791,
        "allocation_upload_ms": 7.455958,
        "cold_call_ms": 9.51725,
        "warmup_ms": 200.720125,
        "download_ms": 3.194375,
        "repetitions": 311,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          103.428993569,
          99.5602893891,
          94.5379131833,
          100.797694534,
          102.509514469,
          102.57529582,
          94.5064308682,
          100.633173633,
          102.925376206
        ],
        "latency_us": [
          366.167,
          314.417,
          333.708,
          287.708,
          353.792,
          282.542,
          371.291,
          313.542,
          2451.375
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 384,
            "reduction_scalar_rounds": 120,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 24576,
              "global_write_bytes": 24576,
              "private_read_bytes": 147456,
              "private_write_bytes": 49152
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 96,
              "private_read_bytes": 576,
              "private_write_bytes": 192
            },
            "striped_storage_scalars_per_worker": 48,
            "reduction_operations": 2,
            "reduction_elements": 12288,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 160,
            "normalized_kernel_cost": 1920,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 18432,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4537958,
              "compute_span_ns": 4537958,
              "calibration_cpu_ns": 5230958,
              "calibration_gpu_ticks": 5230958,
              "compute_passes": 1,
              "command_buffer_ns": 4538041.65591,
              "command_buffers": 2
            },
            {
              "compute_ns": 4691084,
              "compute_span_ns": 4691084,
              "calibration_cpu_ns": 5338125,
              "calibration_gpu_ticks": 5338125,
              "compute_passes": 1,
              "command_buffer_ns": 4691125.01014,
              "command_buffers": 2
            },
            {
              "compute_ns": 4588416,
              "compute_span_ns": 4588416,
              "calibration_cpu_ns": 5241750,
              "calibration_gpu_ticks": 5241750,
              "compute_passes": 1,
              "command_buffer_ns": 4588458.33782,
              "command_buffers": 2
            },
            {
              "compute_ns": 4578917,
              "compute_span_ns": 4578917,
              "calibration_cpu_ns": 5324875,
              "calibration_gpu_ticks": 5324875,
              "compute_passes": 1,
              "command_buffer_ns": 4578958.26548,
              "command_buffers": 2
            },
            {
              "compute_ns": 4751334,
              "compute_span_ns": 4751334,
              "calibration_cpu_ns": 5360833,
              "calibration_gpu_ticks": 5360833,
              "compute_passes": 1,
              "command_buffer_ns": 4751500.03331,
              "command_buffers": 2
            },
            {
              "compute_ns": 4693542,
              "compute_span_ns": 4693542,
              "calibration_cpu_ns": 5366833,
              "calibration_gpu_ticks": 5366833,
              "compute_passes": 1,
              "command_buffer_ns": 4693624.97097,
              "command_buffers": 2
            },
            {
              "compute_ns": 4645833,
              "compute_span_ns": 4645833,
              "calibration_cpu_ns": 5389542,
              "calibration_gpu_ticks": 5389542,
              "compute_passes": 1,
              "command_buffer_ns": 4645833.33815,
              "command_buffers": 2
            },
            {
              "compute_ns": 4737542,
              "compute_span_ns": 4737542,
              "calibration_cpu_ns": 5391417,
              "calibration_gpu_ticks": 5391417,
              "compute_passes": 1,
              "command_buffer_ns": 4737583.33828,
              "command_buffers": 2
            },
            {
              "compute_ns": 4621708,
              "compute_span_ns": 4621708,
              "calibration_cpu_ns": 5751834,
              "calibration_gpu_ticks": 5751834,
              "compute_passes": 1,
              "command_buffer_ns": 4621708.24176,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 112458,
              "compute_span_ns": 112458,
              "calibration_cpu_ns": 816292,
              "calibration_gpu_ticks": 816292,
              "compute_passes": 1,
              "command_buffer_ns": 112499.983516,
              "command_buffers": 2
            },
            {
              "compute_ns": 81625,
              "compute_span_ns": 81625,
              "calibration_cpu_ns": 628666,
              "calibration_gpu_ticks": 628666,
              "compute_passes": 1,
              "command_buffer_ns": 81666.6870378,
              "command_buffers": 2
            },
            {
              "compute_ns": 111500,
              "compute_span_ns": 111500,
              "calibration_cpu_ns": 753209,
              "calibration_gpu_ticks": 753209,
              "compute_passes": 1,
              "command_buffer_ns": 111583.271064,
              "command_buffers": 2
            },
            {
              "compute_ns": 72916,
              "compute_span_ns": 72916,
              "calibration_cpu_ns": 562292,
              "calibration_gpu_ticks": 562292,
              "compute_passes": 1,
              "command_buffer_ns": 72916.6786186,
              "command_buffers": 2
            },
            {
              "compute_ns": 113041,
              "compute_span_ns": 113041,
              "calibration_cpu_ns": 712709,
              "calibration_gpu_ticks": 712709,
              "compute_passes": 1,
              "command_buffer_ns": 113000.04553,
              "command_buffers": 2
            },
            {
              "compute_ns": 72583,
              "compute_span_ns": 72583,
              "calibration_cpu_ns": 994250,
              "calibration_gpu_ticks": 994250,
              "compute_passes": 1,
              "command_buffer_ns": 72624.9418221,
              "command_buffers": 2
            },
            {
              "compute_ns": 71667,
              "compute_span_ns": 71667,
              "calibration_cpu_ns": 489292,
              "calibration_gpu_ticks": 489292,
              "compute_passes": 1,
              "command_buffer_ns": 71750.022471,
              "command_buffers": 2
            },
            {
              "compute_ns": 65333,
              "compute_span_ns": 65333,
              "calibration_cpu_ns": 535250,
              "calibration_gpu_ticks": 535250,
              "compute_passes": 1,
              "command_buffer_ns": 65333.3263472,
              "command_buffers": 2
            },
            {
              "compute_ns": 66291,
              "compute_span_ns": 66291,
              "calibration_cpu_ns": 495750,
              "calibration_gpu_ticks": 495750,
              "compute_passes": 1,
              "command_buffer_ns": 66375.0106469,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4427833.36621,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4747500.00285,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4471749.94042,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4765416.67012,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4740416.65439,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4727750.02709,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4662499.9959,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4840249.95239,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4812416.67876,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 403999.991249,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 64958.3525956,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 395291.659515,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 71833.3176337,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 485958.356876,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 375333.358534,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 112958.368845,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 482958.334032,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 111666.74085,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              69.18489634703124,
              74.17968754453125,
              69.8710928190625,
              74.459635470625,
              74.06901022484375,
              73.87109417328125,
              72.85156243593751,
              75.62890550609376,
              75.194010605625
            ],
            "command_buffer_throughput_us_p50": 74.06901022484375,
            "command_buffer_throughput_us_p90": 75.28098958571874,
            "command_buffer_latency_us": [
              403.999991249,
              64.9583525956,
              395.291659515,
              71.8333176337,
              485.958356876,
              375.333358534,
              112.95836884500001,
              482.958334032,
              111.66674085
            ],
            "command_buffer_latency_us_p50": 375.333358534,
            "command_buffer_latency_us_p90": 483.55833860079997
          },
          "compute_throughput_us": [
            70.90559375,
            73.2981875,
            71.694,
            71.545578125,
            74.23959375,
            73.33659375,
            72.591140625,
            74.02409375,
            72.2141875
          ],
          "compute_throughput_us_p50": 72.591140625,
          "compute_throughput_us_p90": 74.06719375,
          "command_buffer_throughput_us": [
            70.90690087359376,
            73.2988282834375,
            71.6946615284375,
            71.546222898125,
            74.24218802046875,
            73.33789017140626,
            72.59114590859376,
            74.024739660625,
            72.21419127749999
          ],
          "command_buffer_throughput_us_p50": 72.59114590859376,
          "command_buffer_throughput_us_p90": 74.06822933259375,
          "compute_latency_us": [
            112.458,
            81.625,
            111.5,
            72.916,
            113.041,
            72.583,
            71.667,
            65.333,
            66.291
          ],
          "compute_latency_us_p50": 72.916,
          "compute_latency_us_p90": 112.5746,
          "command_buffer_latency_us": [
            112.49998351599999,
            81.6666870378,
            111.583271064,
            72.9166786186,
            113.00004553000001,
            72.6249418221,
            71.75002247100001,
            65.3333263472,
            66.37501064690001
          ],
          "command_buffer_latency_us_p50": 72.9166786186,
          "command_buffer_latency_us_p90": 112.59999591879999,
          "counter_control_throughput_ratio": 0.9927819669135501,
          "counter_control_latency_ratio": 0.2822808637068291
        },
        "correctness": {
          "max_abs_error": 7.116789644580346e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 100.797694534,
        "throughput_us_p90": 103.02609967859999,
        "latency_us_p50": 333.708,
        "latency_us_p90": 787.3078000000003
      },
      "torch": {
        "cold_call_ms": 7.813084,
        "warmup_ms": 200.718333,
        "repetitions": 155,
        "throughput_us": [
          178.78306451612903,
          197.20053548387096,
          181.29730967741935,
          195.46370967741936,
          179.32392903225806,
          196.7766129032258,
          182.31290322580645,
          202.36209677419356,
          194.56586451612904
        ],
        "latency_us": [
          789.5,
          892.75,
          615.458,
          635.459,
          516.792,
          409.791,
          473.08299999999997,
          566.875,
          557.958
        ],
        "throughput_us_p50": 194.56586451612904,
        "throughput_us_p90": 198.23284774193547,
        "latency_us_p50": 566.875,
        "latency_us_p90": 810.15,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 10643958.172295243,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10704833.25747773,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10681624.873541296,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10623125.068377703,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10833708.33657682,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10583291.645161808,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10453041.58469662,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10734166.83357209,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10530375.002417713,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 358874.9677874148,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 444624.9804459512,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 163833.31967517734,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 384208.3388008177,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 167416.69969633222,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 459208.32781121135,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 164500.0302232802,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 318999.9843016267,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 166083.33680778742,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              166.31184644211316,
              167.26301964808954,
              166.90038864908274,
              165.98632919340162,
              169.2766927590128,
              165.36393195565324,
              163.3287747608847,
              167.7213567745639,
              164.53710941277677
            ],
            "command_buffer_throughput_us_p50": 166.31184644211316,
            "command_buffer_throughput_us_p90": 168.03242397145368,
            "command_buffer_latency_us": [
              358.8749677874148,
              444.6249804459512,
              163.83331967517734,
              384.2083388008177,
              167.41669969633222,
              459.20832781121135,
              164.5000302232802,
              318.9999843016267,
              166.08333680778742
            ],
            "command_buffer_latency_us_p50": 318.9999843016267,
            "command_buffer_latency_us_p90": 447.54164991900325
          },
          "throughput": [
            {
              "compute_ns": 11087335.0,
              "compute_span_ns": 12456541.0,
              "command_buffer_ns": 12211708.293762058,
              "calibration_cpu_ns": 13593208.0,
              "calibration_gpu_ticks": 13593208.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11008668.0,
              "compute_span_ns": 12329417.0,
              "command_buffer_ns": 12117208.272684366,
              "calibration_cpu_ns": 13492500.0,
              "calibration_gpu_ticks": 13492500.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11150046.0,
              "compute_span_ns": 12511542.0,
              "command_buffer_ns": 12276749.941520393,
              "calibration_cpu_ns": 13853167.0,
              "calibration_gpu_ticks": 13853167.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11032913.0,
              "compute_span_ns": 12361291.0,
              "command_buffer_ns": 12105625.122785568,
              "calibration_cpu_ns": 13508500.0,
              "calibration_gpu_ticks": 13508500.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11396124.0,
              "compute_span_ns": 12566416.0,
              "command_buffer_ns": 12379333.551507443,
              "calibration_cpu_ns": 13378250.0,
              "calibration_gpu_ticks": 13378250.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11194626.0,
              "compute_span_ns": 12650292.0,
              "command_buffer_ns": 12393374.985549599,
              "calibration_cpu_ns": 13650209.0,
              "calibration_gpu_ticks": 13650209.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11125329.0,
              "compute_span_ns": 12482125.0,
              "command_buffer_ns": 12237791.61227867,
              "calibration_cpu_ns": 13820333.0,
              "calibration_gpu_ticks": 13820333.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11027998.0,
              "compute_span_ns": 12399250.0,
              "command_buffer_ns": 12187458.45835656,
              "calibration_cpu_ns": 13660167.0,
              "calibration_gpu_ticks": 13660167.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11119450.0,
              "compute_span_ns": 12524000.0,
              "command_buffer_ns": 12287874.88071248,
              "calibration_cpu_ns": 13863167.0,
              "calibration_gpu_ticks": 13863167.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 162916.0,
              "compute_span_ns": 162916.0,
              "command_buffer_ns": 162874.988745898,
              "calibration_cpu_ns": 754667.0,
              "calibration_gpu_ticks": 754667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 140333.0,
              "compute_span_ns": 140333.0,
              "command_buffer_ns": 140291.69688001275,
              "calibration_cpu_ns": 676292.0,
              "calibration_gpu_ticks": 676292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 138667.0,
              "compute_span_ns": 138667.0,
              "command_buffer_ns": 138666.65540263057,
              "calibration_cpu_ns": 749042.0,
              "calibration_gpu_ticks": 749042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 139292.0,
              "compute_span_ns": 139292.0,
              "command_buffer_ns": 139291.68926551938,
              "calibration_cpu_ns": 721209.0,
              "calibration_gpu_ticks": 721209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 135459.0,
              "compute_span_ns": 135459.0,
              "command_buffer_ns": 135500.0422336161,
              "calibration_cpu_ns": 601208.0,
              "calibration_gpu_ticks": 601208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 138583.0,
              "compute_span_ns": 138583.0,
              "command_buffer_ns": 138583.30203220248,
              "calibration_cpu_ns": 634750.0,
              "calibration_gpu_ticks": 634750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 140042.0,
              "compute_span_ns": 140042.0,
              "command_buffer_ns": 140041.6367687285,
              "calibration_cpu_ns": 637166.0,
              "calibration_gpu_ticks": 637166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 138583.0,
              "compute_span_ns": 138583.0,
              "command_buffer_ns": 138583.30203220248,
              "calibration_cpu_ns": 593791.0,
              "calibration_gpu_ticks": 593791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 153292.0,
              "compute_span_ns": 153292.0,
              "command_buffer_ns": 153333.35613831878,
              "calibration_cpu_ns": 941708.0,
              "calibration_gpu_ticks": 941708.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            173.239609375,
            172.0104375,
            174.21946875,
            172.389265625,
            178.0644375,
            174.91603125,
            173.833265625,
            172.31246875,
            173.74140625
          ],
          "compute_throughput_us_p50": 173.74140625,
          "compute_throughput_us_p90": 175.5457125,
          "command_buffer_throughput_us": [
            190.80794209003216,
            189.3313792606932,
            191.82421783625614,
            189.1503925435245,
            193.4270867423038,
            193.64648414921248,
            191.21549394185422,
            190.42903841182124,
            191.9980450111325
          ],
          "command_buffer_throughput_us_p50": 191.21549394185422,
          "command_buffer_throughput_us_p90": 193.47096622368554,
          "compute_latency_us": [
            162.916,
            140.333,
            138.667,
            139.292,
            135.459,
            138.583,
            140.042,
            138.583,
            153.292
          ],
          "compute_latency_us_p50": 139.292,
          "compute_latency_us_p90": 155.2168,
          "command_buffer_latency_us": [
            162.874988745898,
            140.29169688001275,
            138.66665540263057,
            139.29168926551938,
            135.5000422336161,
            138.58330203220248,
            140.0416367687285,
            138.58330203220248,
            153.33335613831878
          ],
          "command_buffer_latency_us_p50": 139.29168926551938,
          "command_buffer_latency_us_p90": 155.24168265983462,
          "counter_control_throughput_ratio": 1.1472901430172333,
          "counter_control_latency_ratio": 0.4538488425372136
        },
        "download_ms": 0.670959,
        "allocation_upload_ms": 17.333208,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 7.553236784944503e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.5180646398826251,
      "device_compute_slowdown": 0.41781140254239185,
      "valid": true,
      "round": 1,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "softmax_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.511583,
        "warmup_ms": 201.402667,
        "repetitions": 147,
        "throughput_us": [
          182.43395918367347,
          201.14257142857144,
          182.32880272108844,
          181.0218231292517,
          200.64880952380952,
          179.9671156462585,
          181.1408707482993,
          187.4234693877551,
          192.07369387755102
        ],
        "latency_us": [
          629.084,
          495.75,
          437.625,
          429.625,
          434.041,
          400.709,
          370.916,
          402.834,
          696.2909999999999
        ],
        "throughput_us_p50": 182.43395918367347,
        "throughput_us_p90": 200.7475619047619,
        "latency_us_p50": 434.041,
        "latency_us_p90": 642.5254,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 10499291.645828635,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10429624.991957098,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10967166.628688574,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10525749.87988919,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10595208.44137296,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10343374.917283654,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10357833.351008594,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10696166.660636663,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10444333.252962679,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 363000.0283010304,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 384166.6621156037,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 164500.0302232802,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 310041.7088717222,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 161625.0374354422,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 302625.00513345003,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 162708.28200504184,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 293374.9929070473,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 149125.000461936,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              164.0514319660724,
              162.96289049932966,
              171.36197857325897,
              164.4648418732686,
              165.5501318964525,
              161.6152330825571,
              161.84114610950928,
              167.12760407244787,
              163.19270707754185
            ],
            "command_buffer_throughput_us_p50": 164.0514319660724,
            "command_buffer_throughput_us_p90": 167.9744789726101,
            "command_buffer_latency_us": [
              363.0000283010304,
              384.1666621156037,
              164.5000302232802,
              310.0417088717222,
              161.6250374354422,
              302.62500513345003,
              162.70828200504184,
              293.3749929070473,
              149.125000461936
            ],
            "command_buffer_latency_us_p50": 293.3749929070473,
            "command_buffer_latency_us_p90": 367.23335506394506
          },
          "throughput": [
            {
              "compute_ns": 11109002.0,
              "compute_span_ns": 12341625.0,
              "command_buffer_ns": 12149249.901995063,
              "calibration_cpu_ns": 13777334.0,
              "calibration_gpu_ticks": 13777334.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10834828.0,
              "compute_span_ns": 12120333.0,
              "command_buffer_ns": 11913416.674360633,
              "calibration_cpu_ns": 13958542.0,
              "calibration_gpu_ticks": 13958542.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10870754.0,
              "compute_span_ns": 12018167.0,
              "command_buffer_ns": 11843916.610814631,
              "calibration_cpu_ns": 13897083.0,
              "calibration_gpu_ticks": 13897083.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10984244.0,
              "compute_span_ns": 12247250.0,
              "command_buffer_ns": 12042541.580740362,
              "calibration_cpu_ns": 13608041.0,
              "calibration_gpu_ticks": 13608041.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10962084.0,
              "compute_span_ns": 12145750.0,
              "command_buffer_ns": 11973416.723776609,
              "calibration_cpu_ns": 13759000.0,
              "calibration_gpu_ticks": 13759000.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10989082.0,
              "compute_span_ns": 12207125.0,
              "command_buffer_ns": 12018624.984193593,
              "calibration_cpu_ns": 13685791.0,
              "calibration_gpu_ticks": 13685791.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11032752.0,
              "compute_span_ns": 12246708.0,
              "command_buffer_ns": 12069791.671819985,
              "calibration_cpu_ns": 13757584.0,
              "calibration_gpu_ticks": 13757584.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 11092456.0,
              "compute_span_ns": 12498667.0,
              "command_buffer_ns": 12262124.97567758,
              "calibration_cpu_ns": 13839250.0,
              "calibration_gpu_ticks": 13839250.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10905246.0,
              "compute_span_ns": 12344333.0,
              "command_buffer_ns": 12086500.006262213,
              "calibration_cpu_ns": 13575666.0,
              "calibration_gpu_ticks": 13575666.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 162625.0,
              "compute_span_ns": 162625.0,
              "command_buffer_ns": 162624.98684227467,
              "calibration_cpu_ns": 751958.0,
              "calibration_gpu_ticks": 751958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 139792.0,
              "compute_span_ns": 139792.0,
              "command_buffer_ns": 139833.3115503192,
              "calibration_cpu_ns": 707375.0,
              "calibration_gpu_ticks": 707375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 138333.0,
              "compute_span_ns": 138333.0,
              "command_buffer_ns": 138333.35833624005,
              "calibration_cpu_ns": 613250.0,
              "calibration_gpu_ticks": 613250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 136875.0,
              "compute_span_ns": 136875.0,
              "command_buffer_ns": 136875.02359971404,
              "calibration_cpu_ns": 647417.0,
              "calibration_gpu_ticks": 647417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 137125.0,
              "compute_span_ns": 137125.0,
              "command_buffer_ns": 137124.96729567647,
              "calibration_cpu_ns": 693167.0,
              "calibration_gpu_ticks": 693167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 137791.0,
              "compute_span_ns": 137791.0,
              "command_buffer_ns": 137750.00115856528,
              "calibration_cpu_ns": 611333.0,
              "calibration_gpu_ticks": 611333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 138875.0,
              "compute_span_ns": 138875.0,
              "command_buffer_ns": 138874.98062103987,
              "calibration_cpu_ns": 719584.0,
              "calibration_gpu_ticks": 719584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 146292.0,
              "compute_span_ns": 146292.0,
              "command_buffer_ns": 146333.3028368652,
              "calibration_cpu_ns": 590541.0,
              "calibration_gpu_ticks": 590541.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 144750.0,
              "compute_span_ns": 144750.0,
              "command_buffer_ns": 144749.99625235796,
              "calibration_cpu_ns": 644250.0,
              "calibration_gpu_ticks": 644250.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            173.57815625,
            169.2941875,
            169.85553125,
            171.6288125,
            171.2825625,
            171.70440625,
            172.38675,
            173.319625,
            170.39446875
          ],
          "compute_throughput_us_p50": 171.6288125,
          "compute_throughput_us_p90": 173.37133125,
          "command_buffer_throughput_us": [
            189.83202971867286,
            186.1471355368849,
            185.0611970439786,
            188.16471219906816,
            187.0846363090095,
            187.7910153780249,
            188.59049487218726,
            191.59570274496218,
            188.85156259784708
          ],
          "command_buffer_throughput_us_p50": 188.16471219906816,
          "command_buffer_throughput_us_p90": 190.18476432393072,
          "compute_latency_us": [
            162.625,
            139.792,
            138.333,
            136.875,
            137.125,
            137.791,
            138.875,
            146.292,
            144.75
          ],
          "compute_latency_us_p50": 138.875,
          "compute_latency_us_p90": 149.5586,
          "command_buffer_latency_us": [
            162.62498684227467,
            139.8333115503192,
            138.33335833624005,
            136.87502359971404,
            137.12496729567647,
            137.75000115856528,
            138.87498062103987,
            146.3333028368652,
            144.74999625235796
          ],
          "command_buffer_latency_us_p50": 138.87498062103987,
          "command_buffer_latency_us_p90": 149.59163963794708,
          "counter_control_throughput_ratio": 1.1464036944005234,
          "counter_control_latency_ratio": 0.49879269322463804
        },
        "download_ms": 0.644667,
        "allocation_upload_ms": 0.890041,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 7.553236784944503e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-2v6yqotm/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "7f45b1ebfa78e24e3d477c2ad514fde5c49a3d7b63ed786757d9dc5d1979c8b1",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.083208,
        "capture_ms": 0.07525,
        "compile_ms": 53.282542,
        "allocation_upload_ms": 5.515167,
        "cold_call_ms": 4.668416,
        "warmup_ms": 200.606625,
        "download_ms": 2.887875,
        "repetitions": 341,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          89.0986070381,
          73.2339912023,
          81.1180351906,
          86.0825982405,
          81.4710410557,
          82.9557683284,
          85.9490469208,
          80.2178651026,
          88.1317214076
        ],
        "latency_us": [
          365.959,
          292.375,
          310.834,
          270.042,
          281.875,
          295,
          331.416,
          306.583,
          276.458
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 768,
            "reduction_scalar_rounds": 120,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 24576,
              "global_write_bytes": 24576,
              "private_read_bytes": 147456,
              "private_write_bytes": 49152
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 96,
              "private_read_bytes": 576,
              "private_write_bytes": 192
            },
            "striped_storage_scalars_per_worker": 48,
            "reduction_operations": 2,
            "reduction_elements": 12288,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 168,
            "normalized_kernel_cost": 2016,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 18432,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3594541,
              "compute_span_ns": 3594541,
              "calibration_cpu_ns": 4376791,
              "calibration_gpu_ticks": 4376791,
              "compute_passes": 1,
              "command_buffer_ns": 3594541.57297,
              "command_buffers": 2
            },
            {
              "compute_ns": 4054750,
              "compute_span_ns": 4054750,
              "calibration_cpu_ns": 4972417,
              "calibration_gpu_ticks": 4972417,
              "compute_passes": 1,
              "command_buffer_ns": 4054791.70149,
              "command_buffers": 2
            },
            {
              "compute_ns": 4483166,
              "compute_span_ns": 4483166,
              "calibration_cpu_ns": 5454958,
              "calibration_gpu_ticks": 5454958,
              "compute_passes": 1,
              "command_buffer_ns": 4483166.67462,
              "command_buffers": 2
            },
            {
              "compute_ns": 4634958,
              "compute_span_ns": 4634958,
              "calibration_cpu_ns": 5552667,
              "calibration_gpu_ticks": 5552667,
              "compute_passes": 1,
              "command_buffer_ns": 4635083.3145,
              "command_buffers": 2
            },
            {
              "compute_ns": 4252333,
              "compute_span_ns": 4252333,
              "calibration_cpu_ns": 4950833,
              "calibration_gpu_ticks": 4950833,
              "compute_passes": 1,
              "command_buffer_ns": 4252416.6638,
              "command_buffers": 2
            },
            {
              "compute_ns": 4295417,
              "compute_span_ns": 4295417,
              "calibration_cpu_ns": 5185959,
              "calibration_gpu_ticks": 5185959,
              "compute_passes": 1,
              "command_buffer_ns": 4295458.37687,
              "command_buffers": 2
            },
            {
              "compute_ns": 4485667,
              "compute_span_ns": 4485667,
              "calibration_cpu_ns": 5352625,
              "calibration_gpu_ticks": 5352625,
              "compute_passes": 1,
              "command_buffer_ns": 4485791.6655,
              "command_buffers": 2
            },
            {
              "compute_ns": 4433583,
              "compute_span_ns": 4433583,
              "calibration_cpu_ns": 5380792,
              "calibration_gpu_ticks": 5380792,
              "compute_passes": 1,
              "command_buffer_ns": 4433666.70515,
              "command_buffers": 2
            },
            {
              "compute_ns": 4397292,
              "compute_span_ns": 4397292,
              "calibration_cpu_ns": 5200625,
              "calibration_gpu_ticks": 5200625,
              "compute_passes": 1,
              "command_buffer_ns": 4397333.3668,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 113083,
              "compute_span_ns": 113083,
              "calibration_cpu_ns": 916166,
              "calibration_gpu_ticks": 916166,
              "compute_passes": 1,
              "command_buffer_ns": 113083.340693,
              "command_buffers": 2
            },
            {
              "compute_ns": 62500,
              "compute_span_ns": 62500,
              "calibration_cpu_ns": 598084,
              "calibration_gpu_ticks": 598084,
              "compute_passes": 1,
              "command_buffer_ns": 62541.6869298,
              "command_buffers": 2
            },
            {
              "compute_ns": 58500,
              "compute_span_ns": 58500,
              "calibration_cpu_ns": 545958,
              "calibration_gpu_ticks": 545958,
              "compute_passes": 1,
              "command_buffer_ns": 58541.7146794,
              "command_buffers": 2
            },
            {
              "compute_ns": 58375,
              "compute_span_ns": 58375,
              "calibration_cpu_ns": 507167,
              "calibration_gpu_ticks": 507167,
              "compute_passes": 1,
              "command_buffer_ns": 58458.361309,
              "command_buffers": 2
            },
            {
              "compute_ns": 58292,
              "compute_span_ns": 58292,
              "calibration_cpu_ns": 526625,
              "calibration_gpu_ticks": 526625,
              "compute_passes": 1,
              "command_buffer_ns": 58375.0079386,
              "command_buffers": 2
            },
            {
              "compute_ns": 58208,
              "compute_span_ns": 58208,
              "calibration_cpu_ns": 511334,
              "calibration_gpu_ticks": 511334,
              "compute_passes": 1,
              "command_buffer_ns": 58250.0360906,
              "command_buffers": 2
            },
            {
              "compute_ns": 69708,
              "compute_span_ns": 69708,
              "calibration_cpu_ns": 464375,
              "calibration_gpu_ticks": 464375,
              "compute_passes": 1,
              "command_buffer_ns": 69666.6538715,
              "command_buffers": 2
            },
            {
              "compute_ns": 59583,
              "compute_span_ns": 59583,
              "calibration_cpu_ns": 469250,
              "calibration_gpu_ticks": 469250,
              "compute_passes": 1,
              "command_buffer_ns": 59625.0174567,
              "command_buffers": 2
            },
            {
              "compute_ns": 74584,
              "compute_span_ns": 74584,
              "calibration_cpu_ns": 555250,
              "calibration_gpu_ticks": 555250,
              "compute_passes": 1,
              "command_buffer_ns": 74666.691944,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 6433541.71189,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4338500.03174,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4239083.26775,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4631416.63931,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4432666.69754,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4412833.36841,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4244583.30963,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4265249.99784,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4390208.34165,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 377458.345611,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 468541.693408,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 107625.033706,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 557749.997824,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 105666.695163,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 377041.636966,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 109416.665509,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 419416.639488,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 109999.964479,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              100.52408924828124,
              67.7890629959375,
              66.23567605859374,
              72.36588498921876,
              69.2604171490625,
              68.95052138140625,
              66.32161421296875,
              66.64453121625,
              68.59700533828125
            ],
            "command_buffer_throughput_us_p50": 68.59700533828125,
            "command_buffer_throughput_us_p90": 77.99752584103126,
            "command_buffer_latency_us": [
              377.458345611,
              468.541693408,
              107.625033706,
              557.7499978239999,
              105.666695163,
              377.041636966,
              109.416665509,
              419.416639488,
              109.999964479
            ],
            "command_buffer_latency_us_p50": 377.041636966,
            "command_buffer_latency_us_p90": 486.3833542912
          },
          "compute_throughput_us": [
            56.164703125,
            63.35546875,
            70.04946875,
            72.42121875,
            66.442703125,
            67.115890625,
            70.088546875,
            69.274734375,
            68.7076875
          ],
          "compute_throughput_us_p50": 68.7076875,
          "compute_throughput_us_p90": 70.55508125,
          "command_buffer_throughput_us": [
            56.164712077656255,
            63.35612033578125,
            70.0494792909375,
            72.42317678906251,
            66.44401037187501,
            67.11653713859374,
            70.0904947734375,
            69.27604226796875,
            68.70833385625
          ],
          "command_buffer_throughput_us_p50": 68.70833385625,
          "command_buffer_throughput_us_p90": 70.5570311765625,
          "compute_latency_us": [
            113.083,
            62.5,
            58.5,
            58.375,
            58.292,
            58.208,
            69.708,
            59.583,
            74.584
          ],
          "compute_latency_us_p50": 59.583,
          "compute_latency_us_p90": 82.28380000000001,
          "command_buffer_latency_us": [
            113.08334069300001,
            62.5416869298,
            58.5417146794,
            58.458361309,
            58.3750079386,
            58.2500360906,
            69.6666538715,
            59.6250174567,
            74.66669194400001
          ],
          "command_buffer_latency_us_p50": 59.6250174567,
          "command_buffer_latency_us_p90": 82.35002169380002,
          "counter_control_throughput_ratio": 1.0007916962509653,
          "counter_control_latency_ratio": 0.29959157615113674
        },
        "correctness": {
          "max_abs_error": 7.116789644580346e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 82.9557683284,
        "throughput_us_p90": 88.3250985337,
        "latency_us_p50": 295.0,
        "latency_us_p90": 338.32460000000003
      },
      "slowdown": 0.4547167024143822,
      "device_compute_slowdown": 0.4003272323520854,
      "valid": true,
      "round": 1,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "softmax_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-578_ayd5/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "8bb581805080afef9f47b3147742261e73f0bbc7f1d60afd69631b609a5e3f17",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 23.137,
        "capture_ms": 0.070417,
        "compile_ms": 55.560083,
        "allocation_upload_ms": 6.638916,
        "cold_call_ms": 6.103792,
        "warmup_ms": 201.429959,
        "download_ms": 2.339375,
        "repetitions": 519,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          75.9142581888,
          71.2733622351,
          77.0770712909,
          75.8282755299,
          71.2273603083,
          77.1673082852,
          78.9249344894,
          75.5905587669,
          75.5735375723
        ],
        "latency_us": [
          1779.708,
          453.75,
          347.5,
          281.75,
          267.792,
          287.416,
          281.334,
          323,
          282.792
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 1024,
            "reduction_scalar_rounds": 80,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 16384,
              "global_write_bytes": 16384,
              "private_read_bytes": 98304,
              "private_write_bytes": 32768
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 64,
              "global_write_bytes": 64,
              "private_read_bytes": 384,
              "private_write_bytes": 128
            },
            "striped_storage_scalars_per_worker": 32,
            "reduction_operations": 2,
            "reduction_elements": 8192,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 128,
            "normalized_kernel_cost": 2048,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4127417,
              "compute_span_ns": 4127417,
              "calibration_cpu_ns": 4884667,
              "calibration_gpu_ticks": 4884667,
              "compute_passes": 1,
              "command_buffer_ns": 4127499.99668,
              "command_buffers": 2
            },
            {
              "compute_ns": 4273458,
              "compute_span_ns": 4273458,
              "calibration_cpu_ns": 5012083,
              "calibration_gpu_ticks": 5012083,
              "compute_passes": 1,
              "command_buffer_ns": 4273458.32577,
              "command_buffers": 2
            },
            {
              "compute_ns": 4020458,
              "compute_span_ns": 4020458,
              "calibration_cpu_ns": 5099208,
              "calibration_gpu_ticks": 5099208,
              "compute_passes": 1,
              "command_buffer_ns": 4020458.37836,
              "command_buffers": 2
            },
            {
              "compute_ns": 3990792,
              "compute_span_ns": 3990792,
              "calibration_cpu_ns": 4797292,
              "calibration_gpu_ticks": 4797292,
              "compute_passes": 1,
              "command_buffer_ns": 3990791.67983,
              "command_buffers": 2
            },
            {
              "compute_ns": 3877208,
              "compute_span_ns": 3877208,
              "calibration_cpu_ns": 4807833,
              "calibration_gpu_ticks": 4807833,
              "compute_passes": 1,
              "command_buffer_ns": 3877208.33533,
              "command_buffers": 2
            },
            {
              "compute_ns": 3784292,
              "compute_span_ns": 3784292,
              "calibration_cpu_ns": 4878625,
              "calibration_gpu_ticks": 4878625,
              "compute_passes": 1,
              "command_buffer_ns": 3784374.9742,
              "command_buffers": 2
            },
            {
              "compute_ns": 3719625,
              "compute_span_ns": 3719625,
              "calibration_cpu_ns": 4453125,
              "calibration_gpu_ticks": 4453125,
              "compute_passes": 1,
              "command_buffer_ns": 3719666.68172,
              "command_buffers": 2
            },
            {
              "compute_ns": 3948500,
              "compute_span_ns": 3948500,
              "calibration_cpu_ns": 4858000,
              "calibration_gpu_ticks": 4858000,
              "compute_passes": 1,
              "command_buffer_ns": 3948541.64915,
              "command_buffers": 2
            },
            {
              "compute_ns": 3984625,
              "compute_span_ns": 3984625,
              "calibration_cpu_ns": 4939459,
              "calibration_gpu_ticks": 4939459,
              "compute_passes": 1,
              "command_buffer_ns": 3984666.66229,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 97750,
              "compute_span_ns": 97750,
              "calibration_cpu_ns": 633333,
              "calibration_gpu_ticks": 633333,
              "compute_passes": 1,
              "command_buffer_ns": 97874.9594651,
              "command_buffers": 2
            },
            {
              "compute_ns": 52417,
              "compute_span_ns": 52417,
              "calibration_cpu_ns": 565584,
              "calibration_gpu_ticks": 565584,
              "compute_passes": 1,
              "command_buffer_ns": 52458.3156221,
              "command_buffers": 2
            },
            {
              "compute_ns": 100166,
              "compute_span_ns": 100166,
              "calibration_cpu_ns": 920042,
              "calibration_gpu_ticks": 920042,
              "compute_passes": 1,
              "command_buffer_ns": 100208.329968,
              "command_buffers": 2
            },
            {
              "compute_ns": 52542,
              "compute_span_ns": 52542,
              "calibration_cpu_ns": 555250,
              "calibration_gpu_ticks": 555250,
              "compute_passes": 1,
              "command_buffer_ns": 52666.6408405,
              "command_buffers": 2
            },
            {
              "compute_ns": 98250,
              "compute_span_ns": 98250,
              "calibration_cpu_ns": 871833,
              "calibration_gpu_ticks": 871833,
              "compute_passes": 1,
              "command_buffer_ns": 98333.3447948,
              "command_buffers": 2
            },
            {
              "compute_ns": 59459,
              "compute_span_ns": 59459,
              "calibration_cpu_ns": 498375,
              "calibration_gpu_ticks": 498375,
              "compute_passes": 1,
              "command_buffer_ns": 59624.959249,
              "command_buffers": 2
            },
            {
              "compute_ns": 94750,
              "compute_span_ns": 94750,
              "calibration_cpu_ns": 718166,
              "calibration_gpu_ticks": 718166,
              "compute_passes": 1,
              "command_buffer_ns": 94875.053037,
              "command_buffers": 2
            },
            {
              "compute_ns": 52292,
              "compute_span_ns": 52292,
              "calibration_cpu_ns": 730500,
              "calibration_gpu_ticks": 730500,
              "compute_passes": 1,
              "command_buffer_ns": 52291.6670889,
              "command_buffers": 2
            },
            {
              "compute_ns": 95458,
              "compute_span_ns": 95458,
              "calibration_cpu_ns": 681125,
              "calibration_gpu_ticks": 681125,
              "compute_passes": 1,
              "command_buffer_ns": 95541.7053774,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 3782916.69767,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4260291.63646,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3926833.33484,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3952624.97477,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3844666.644,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3919791.66307,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4098958.33582,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3897208.3712,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4013458.32506,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 447249.97133,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 53291.6747034,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 428791.68177,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 52874.9660589,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 437249.953393,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 52999.9379069,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 452958.338428,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 74958.3123252,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 427041.668445,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              59.108073401093755,
              66.5670568196875,
              61.356770856875,
              61.759765230781255,
              60.0729163125,
              61.246744735468745,
              64.0462239971875,
              60.8938808,
              62.710286329062505
            ],
            "command_buffer_throughput_us_p50": 61.356770856875,
            "command_buffer_throughput_us_p90": 64.5503905616875,
            "command_buffer_latency_us": [
              447.24997133,
              53.2916747034,
              428.79168177,
              52.8749660589,
              437.249953393,
              52.9999379069,
              452.958338428,
              74.9583123252,
              427.041668445
            ],
            "command_buffer_latency_us_p50": 427.041668445,
            "command_buffer_latency_us_p90": 448.3916447496
          },
          "compute_throughput_us": [
            64.490890625,
            66.77278125,
            62.81965625,
            62.356125,
            60.581375,
            59.1295625,
            58.119140625,
            61.6953125,
            62.259765625
          ],
          "compute_throughput_us_p50": 62.259765625,
          "compute_throughput_us_p90": 64.94726875,
          "command_buffer_throughput_us": [
            64.492187448125,
            66.77278634015624,
            62.819662161875,
            62.35611999734375,
            60.58138023953125,
            59.130858971875,
            58.119791901875004,
            61.69596326796875,
            62.260416598281246
          ],
          "command_buffer_throughput_us_p50": 62.260416598281246,
          "command_buffer_throughput_us_p90": 64.94830722653124,
          "compute_latency_us": [
            97.75,
            52.417,
            100.166,
            52.542,
            98.25,
            59.459,
            94.75,
            52.292,
            95.458
          ],
          "compute_latency_us_p50": 94.75,
          "compute_latency_us_p90": 98.6332,
          "command_buffer_latency_us": [
            97.87495946509999,
            52.4583156221,
            100.208329968,
            52.666640840499994,
            98.3333447948,
            59.624959249,
            94.875053037,
            52.2916670889,
            95.54170537739999
          ],
          "command_buffer_latency_us_p50": 94.875053037,
          "command_buffer_latency_us_p90": 98.70834182944,
          "counter_control_throughput_ratio": 1.0084641125858818,
          "counter_control_latency_ratio": 0.23369933286567543
        },
        "correctness": {
          "max_abs_error": 1.2343096304160556e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 75.8282755299,
        "throughput_us_p90": 77.51883352604,
        "latency_us_p50": 287.416,
        "latency_us_p90": 718.9416000000003
      },
      "torch": {
        "cold_call_ms": 4.503708,
        "warmup_ms": 200.498458,
        "repetitions": 171,
        "throughput_us": [
          178.14132748538012,
          164.69834502923976,
          174.73854970760235,
          164.8830409356725,
          174.4339649122807,
          162.02631578947367,
          177.70540935672514,
          164.03557309941522,
          180.18470175438597
        ],
        "latency_us": [
          594.333,
          427.66700000000003,
          455.0,
          427.29200000000003,
          415.875,
          406.709,
          2177.25,
          940.791,
          468.833
        ],
        "throughput_us_p50": 174.4339649122807,
        "throughput_us_p90": 178.55000233918128,
        "latency_us_p50": 455.0,
        "latency_us_p90": 1188.0828000000001,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 9759583.335835487,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9865875.006653368,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9799958.323128521,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9822583.349887282,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9825000.015553087,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9950666.630174965,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9678916.539996862,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10004083.334933966,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9855833.370238543,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 351624.97078999877,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 138250.00496581197,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 248791.6499376297,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 130500.00416114926,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 362583.3196565509,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 128499.98893216252,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 315416.6624881327,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 130500.00416114926,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 365874.96288120747,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              152.49348962242948,
              154.15429697895888,
              153.12434879888315,
              153.47786484198878,
              153.51562524301698,
              155.47916609648382,
              151.23307093745098,
              156.31380210834322,
              153.99739640997723
            ],
            "command_buffer_throughput_us_p50": 153.51562524301698,
            "command_buffer_throughput_us_p90": 155.6460932988557,
            "command_buffer_latency_us": [
              351.62497078999877,
              138.25000496581197,
              248.7916499376297,
              130.50000416114926,
              362.5833196565509,
              128.49998893216252,
              315.4166624881327,
              130.50000416114926,
              365.87496288120747
            ],
            "command_buffer_latency_us_p50": 248.7916499376297,
            "command_buffer_latency_us_p90": 363.2416483014822
          },
          "throughput": [
            {
              "compute_ns": 10256254.0,
              "compute_span_ns": 11465167.0,
              "command_buffer_ns": 11301041.638944298,
              "calibration_cpu_ns": 12865625.0,
              "calibration_gpu_ticks": 12865625.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10147580.0,
              "compute_span_ns": 11370167.0,
              "command_buffer_ns": 11192875.041160733,
              "calibration_cpu_ns": 12723959.0,
              "calibration_gpu_ticks": 12723959.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10126750.0,
              "compute_span_ns": 11404000.0,
              "command_buffer_ns": 11204249.982256442,
              "calibration_cpu_ns": 12513125.0,
              "calibration_gpu_ticks": 12513125.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10224954.0,
              "compute_span_ns": 11461834.0,
              "command_buffer_ns": 11284208.44906941,
              "calibration_cpu_ns": 12681500.0,
              "calibration_gpu_ticks": 12681500.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10247581.0,
              "compute_span_ns": 11449541.0,
              "command_buffer_ns": 11261958.279646933,
              "calibration_cpu_ns": 12830834.0,
              "calibration_gpu_ticks": 12830834.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10241206.0,
              "compute_span_ns": 11564667.0,
              "command_buffer_ns": 11357333.161868155,
              "calibration_cpu_ns": 12823416.0,
              "calibration_gpu_ticks": 12823416.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10161667.0,
              "compute_span_ns": 11463625.0,
              "command_buffer_ns": 11283374.97357279,
              "calibration_cpu_ns": 12761083.0,
              "calibration_gpu_ticks": 12761083.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10241751.0,
              "compute_span_ns": 11500625.0,
              "command_buffer_ns": 11294041.702058166,
              "calibration_cpu_ns": 13015708.0,
              "calibration_gpu_ticks": 13015708.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10422704.0,
              "compute_span_ns": 11600084.0,
              "command_buffer_ns": 11408749.90960583,
              "calibration_cpu_ns": 13665875.0,
              "calibration_gpu_ticks": 13665875.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 144542.0,
              "compute_span_ns": 144542.0,
              "command_buffer_ns": 144583.3477191627,
              "calibration_cpu_ns": 709250.0,
              "calibration_gpu_ticks": 709250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 121875.0,
              "compute_span_ns": 121875.0,
              "command_buffer_ns": 121874.9675899744,
              "calibration_cpu_ns": 672625.0,
              "calibration_gpu_ticks": 672625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 141667.0,
              "compute_span_ns": 141667.0,
              "command_buffer_ns": 141666.67824611068,
              "calibration_cpu_ns": 617750.0,
              "calibration_gpu_ticks": 617750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 136125.0,
              "compute_span_ns": 136125.0,
              "command_buffer_ns": 136125.017888844,
              "calibration_cpu_ns": 608875.0,
              "calibration_gpu_ticks": 608875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 148250.0,
              "compute_span_ns": 148250.0,
              "command_buffer_ns": 148250.02290308475,
              "calibration_cpu_ns": 803584.0,
              "calibration_gpu_ticks": 803584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 125584.0,
              "compute_span_ns": 125584.0,
              "command_buffer_ns": 125624.99614432454,
              "calibration_cpu_ns": 658958.0,
              "calibration_gpu_ticks": 658958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 150375.0,
              "compute_span_ns": 150375.0,
              "command_buffer_ns": 150375.0099800527,
              "calibration_cpu_ns": 792125.0,
              "calibration_gpu_ticks": 792125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 128292.0,
              "compute_span_ns": 128292.0,
              "command_buffer_ns": 128291.66371375322,
              "calibration_cpu_ns": 618667.0,
              "calibration_gpu_ticks": 618667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 143250.0,
              "compute_span_ns": 143250.0,
              "command_buffer_ns": 143250.04303827882,
              "calibration_cpu_ns": 794084.0,
              "calibration_gpu_ticks": 794084.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            160.25396875,
            158.5559375,
            158.23046875,
            159.76490625,
            160.118453125,
            160.01884375,
            158.776046875,
            160.027359375,
            162.85475
          ],
          "compute_throughput_us_p50": 160.01884375,
          "compute_throughput_us_p90": 160.774125,
          "command_buffer_throughput_us": [
            176.57877560850466,
            174.88867251813645,
            175.0664059727569,
            176.31575701670954,
            175.96809811948333,
            177.45833065418992,
            176.30273396207485,
            176.46940159465885,
            178.2617173375911
          ],
          "command_buffer_throughput_us_p50": 176.31575701670954,
          "command_buffer_throughput_us_p90": 177.61900799087016,
          "compute_latency_us": [
            144.542,
            121.875,
            141.667,
            136.125,
            148.25,
            125.584,
            150.375,
            128.292,
            143.25
          ],
          "compute_latency_us_p50": 141.667,
          "compute_latency_us_p90": 148.675,
          "command_buffer_latency_us": [
            144.5833477191627,
            121.8749675899744,
            141.66667824611068,
            136.125017888844,
            148.25002290308475,
            125.62499614432454,
            150.3750099800527,
            128.29166371375322,
            143.25004303827882
          ],
          "command_buffer_latency_us_p50": 141.66667824611068,
          "command_buffer_latency_us_p90": 148.67502031847835,
          "counter_control_throughput_ratio": 1.1462552938238295,
          "counter_control_latency_ratio": 0.5694189426438769
        },
        "download_ms": 0.606875,
        "allocation_upload_ms": 16.81025,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 9.74135319392784e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.4347104967087832,
      "device_compute_slowdown": 0.38907771213663706,
      "valid": true,
      "round": 1,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "softmax_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.574875,
        "warmup_ms": 200.787625,
        "repetitions": 160,
        "throughput_us": [
          164.17890625,
          172.68515625,
          171.10520625,
          167.38229375,
          168.1645875,
          182.10495,
          168.59791875000002,
          167.31744375,
          175.29635624999997
        ],
        "latency_us": [
          1035.875,
          596.4580000000001,
          415.334,
          417.541,
          398.208,
          391.875,
          413.58299999999997,
          402.91700000000003,
          382.875
        ],
        "throughput_us_p50": 168.59791875000002,
        "throughput_us_p90": 176.65807499999997,
        "latency_us_p50": 413.58299999999997,
        "latency_us_p90": 684.3414000000001,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 10274250.0603199,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9964666.56215489,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9737333.341035992,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9991041.617468,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9760958.258993924,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9996458.364184946,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9775624.78510663,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 10133375.006262213,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 9727166.674565524,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 472000.04337355494,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 125500.02429634333,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 267874.9733604491,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 252874.97555837035,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 149583.32758396864,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 142125.00536814332,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 132916.6698269546,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 125874.99804794788,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 128583.34230259061,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              160.53515719249845,
              155.69791503367014,
              152.14583345368737,
              156.1100252729375,
              152.51497279678006,
              156.19466194038978,
              152.74413726729108,
              158.33398447284708,
              151.9869792900863
            ],
            "command_buffer_throughput_us_p50": 155.69791503367014,
            "command_buffer_throughput_us_p90": 158.77421901677735,
            "command_buffer_latency_us": [
              472.00004337355494,
              125.50002429634333,
              267.8749733604491,
              252.87497555837035,
              149.58332758396864,
              142.12500536814332,
              132.9166698269546,
              125.87499804794788,
              128.5833423025906
            ],
            "command_buffer_latency_us_p50": 142.12500536814332,
            "command_buffer_latency_us_p90": 308.6999873630703
          },
          "throughput": [
            {
              "compute_ns": 10573834.0,
              "compute_span_ns": 11789584.0,
              "command_buffer_ns": 11583791.754674166,
              "calibration_cpu_ns": 13445792.0,
              "calibration_gpu_ticks": 13445792.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10185086.0,
              "compute_span_ns": 11393542.0,
              "command_buffer_ns": 11183041.671756655,
              "calibration_cpu_ns": 12922959.0,
              "calibration_gpu_ticks": 12922959.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10104210.0,
              "compute_span_ns": 11390166.0,
              "command_buffer_ns": 11195166.676770896,
              "calibration_cpu_ns": 12861208.0,
              "calibration_gpu_ticks": 12861208.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10442042.0,
              "compute_span_ns": 11696125.0,
              "command_buffer_ns": 11489708.151202649,
              "calibration_cpu_ns": 13002583.0,
              "calibration_gpu_ticks": 13002583.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10531841.0,
              "compute_span_ns": 11722291.0,
              "command_buffer_ns": 11518458.370119333,
              "calibration_cpu_ns": 13018625.0,
              "calibration_gpu_ticks": 13018625.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10241375.0,
              "compute_span_ns": 11471708.0,
              "command_buffer_ns": 11260583.240073174,
              "calibration_cpu_ns": 12839208.0,
              "calibration_gpu_ticks": 12839208.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10254543.0,
              "compute_span_ns": 11516417.0,
              "command_buffer_ns": 11316833.31914246,
              "calibration_cpu_ns": 12895083.0,
              "calibration_gpu_ticks": 12895083.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10248624.0,
              "compute_span_ns": 11591542.0,
              "command_buffer_ns": 11353666.719514877,
              "calibration_cpu_ns": 13303375.0,
              "calibration_gpu_ticks": 13303375.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 10300090.0,
              "compute_span_ns": 11492709.0,
              "command_buffer_ns": 11293166.724499315,
              "calibration_cpu_ns": 12475833.0,
              "calibration_gpu_ticks": 12475833.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 151583.0,
              "compute_span_ns": 151583.0,
              "command_buffer_ns": 151541.66612774134,
              "calibration_cpu_ns": 808333.0,
              "calibration_gpu_ticks": 808333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 131417.0,
              "compute_span_ns": 131417.0,
              "command_buffer_ns": 131416.65840521455,
              "calibration_cpu_ns": 725916.0,
              "calibration_gpu_ticks": 725916.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 141583.0,
              "compute_span_ns": 141583.0,
              "command_buffer_ns": 141541.64819046855,
              "calibration_cpu_ns": 634291.0,
              "calibration_gpu_ticks": 634291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 126209.0,
              "compute_span_ns": 126209.0,
              "command_buffer_ns": 126250.03000721335,
              "calibration_cpu_ns": 723500.0,
              "calibration_gpu_ticks": 723500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 125750.0,
              "compute_span_ns": 125750.0,
              "command_buffer_ns": 125750.02619996667,
              "calibration_cpu_ns": 696917.0,
              "calibration_gpu_ticks": 696917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 274583.0,
              "compute_span_ns": 274583.0,
              "command_buffer_ns": 274541.67138785124,
              "calibration_cpu_ns": 1527000.0,
              "calibration_gpu_ticks": 1527000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 399291.0,
              "compute_span_ns": 399291.0,
              "command_buffer_ns": 399250.0132881105,
              "calibration_cpu_ns": 1413792.0,
              "calibration_gpu_ticks": 1413792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 149792.0,
              "compute_span_ns": 149792.0,
              "command_buffer_ns": 149791.71101003885,
              "calibration_cpu_ns": 843375.0,
              "calibration_gpu_ticks": 843375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 343250.0,
              "compute_span_ns": 343250.0,
              "command_buffer_ns": 343249.99433010817,
              "calibration_cpu_ns": 1761250.0,
              "calibration_gpu_ticks": 1761250.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            165.21615625,
            159.14196875,
            157.87828125,
            163.15690625,
            164.560015625,
            160.021484375,
            160.227234375,
            160.13475,
            160.93890625
          ],
          "compute_throughput_us_p50": 160.227234375,
          "compute_throughput_us_p90": 164.69124375,
          "command_buffer_throughput_us": [
            180.99674616678385,
            174.73502612119773,
            174.92447932454525,
            179.5266898625414,
            179.97591203311458,
            175.94661312614335,
            176.82552061160095,
            177.40104249241995,
            176.4557300703018
          ],
          "command_buffer_throughput_us_p50": 176.82552061160095,
          "command_buffer_throughput_us_p90": 180.18007885984844,
          "compute_latency_us": [
            151.583,
            131.417,
            141.583,
            126.209,
            125.75,
            274.583,
            399.291,
            149.792,
            343.25
          ],
          "compute_latency_us_p50": 149.792,
          "compute_latency_us_p90": 354.45820000000003,
          "command_buffer_latency_us": [
            151.54166612774134,
            131.41665840521455,
            141.54164819046855,
            126.25003000721335,
            125.75002619996667,
            274.54167138785124,
            399.2500132881105,
            149.79171101003885,
            343.24999433010817
          ],
          "command_buffer_latency_us_p50": 149.79171101003885,
          "command_buffer_latency_us_p90": 354.44999812170863,
          "counter_control_throughput_ratio": 1.1497158703185362,
          "counter_control_latency_ratio": 1.0471444857644032
        },
        "download_ms": 0.614,
        "allocation_upload_ms": 0.806,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 9.74135319392784e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-89jka_sc/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "117627a7b60c4cee3b9bce9542bc5f280f025f1ad809d52b8d22cb3cdd24d0d1",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 18.886917,
        "capture_ms": 0.066167,
        "compile_ms": 53.942667,
        "allocation_upload_ms": 6.339333,
        "cold_call_ms": 9.40875,
        "warmup_ms": 200.130208,
        "download_ms": 2.806625,
        "repetitions": 364,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          83.6239697802,
          75.1686126374,
          82.6092032967,
          75.1127527473,
          76.96875,
          80.8721373626,
          75.1097747253,
          83.0869945055,
          75.3241758242
        ],
        "latency_us": [
          362.416,
          353,
          1882.708,
          656.417,
          392.25,
          312.917,
          280.541,
          297.584,
          279.375
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 512,
            "reduction_scalar_rounds": 80,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 16384,
              "global_write_bytes": 16384,
              "private_read_bytes": 98304,
              "private_write_bytes": 32768
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 64,
              "global_write_bytes": 64,
              "private_read_bytes": 384,
              "private_write_bytes": 128
            },
            "striped_storage_scalars_per_worker": 32,
            "reduction_operations": 2,
            "reduction_elements": 8192,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 120,
            "normalized_kernel_cost": 1920,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4130958,
              "compute_span_ns": 4130958,
              "calibration_cpu_ns": 5049000,
              "calibration_gpu_ticks": 5049000,
              "compute_passes": 1,
              "command_buffer_ns": 4131041.70002,
              "command_buffers": 2
            },
            {
              "compute_ns": 4092042,
              "compute_span_ns": 4092042,
              "calibration_cpu_ns": 5010041,
              "calibration_gpu_ticks": 5010041,
              "compute_passes": 1,
              "command_buffer_ns": 4092166.66594,
              "command_buffers": 2
            },
            {
              "compute_ns": 4044375,
              "compute_span_ns": 4044375,
              "calibration_cpu_ns": 5036167,
              "calibration_gpu_ticks": 5036167,
              "compute_passes": 1,
              "command_buffer_ns": 4044416.6516,
              "command_buffers": 2
            },
            {
              "compute_ns": 4114583,
              "compute_span_ns": 4114583,
              "calibration_cpu_ns": 5068458,
              "calibration_gpu_ticks": 5068458,
              "compute_passes": 1,
              "command_buffer_ns": 4114708.33933,
              "command_buffers": 2
            },
            {
              "compute_ns": 4090000,
              "compute_span_ns": 4090000,
              "calibration_cpu_ns": 5095875,
              "calibration_gpu_ticks": 5095875,
              "compute_passes": 1,
              "command_buffer_ns": 4090083.35555,
              "command_buffers": 2
            },
            {
              "compute_ns": 4083500,
              "compute_span_ns": 4083500,
              "calibration_cpu_ns": 5080708,
              "calibration_gpu_ticks": 5080708,
              "compute_passes": 1,
              "command_buffer_ns": 4083500.01089,
              "command_buffers": 2
            },
            {
              "compute_ns": 4180083,
              "compute_span_ns": 4180083,
              "calibration_cpu_ns": 5174292,
              "calibration_gpu_ticks": 5174292,
              "compute_passes": 1,
              "command_buffer_ns": 4180083.34236,
              "command_buffers": 2
            },
            {
              "compute_ns": 4004209,
              "compute_span_ns": 4004209,
              "calibration_cpu_ns": 4973250,
              "calibration_gpu_ticks": 4973250,
              "compute_passes": 1,
              "command_buffer_ns": 4004375.01958,
              "command_buffers": 2
            },
            {
              "compute_ns": 4119167,
              "compute_span_ns": 4119167,
              "calibration_cpu_ns": 4953208,
              "calibration_gpu_ticks": 4953208,
              "compute_passes": 1,
              "command_buffer_ns": 4119291.66876,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 96334,
              "compute_span_ns": 96334,
              "calibration_cpu_ns": 721541,
              "calibration_gpu_ticks": 721541,
              "compute_passes": 1,
              "command_buffer_ns": 96458.3014138,
              "command_buffers": 2
            },
            {
              "compute_ns": 54541,
              "compute_span_ns": 54541,
              "calibration_cpu_ns": 466542,
              "calibration_gpu_ticks": 466542,
              "compute_passes": 1,
              "command_buffer_ns": 54583.3609067,
              "command_buffers": 2
            },
            {
              "compute_ns": 98750,
              "compute_span_ns": 98750,
              "calibration_cpu_ns": 694625,
              "calibration_gpu_ticks": 694625,
              "compute_passes": 1,
              "command_buffer_ns": 98874.9670796,
              "command_buffers": 2
            },
            {
              "compute_ns": 54833,
              "compute_span_ns": 54833,
              "calibration_cpu_ns": 483583,
              "calibration_gpu_ticks": 483583,
              "compute_passes": 1,
              "command_buffer_ns": 54958.2764506,
              "command_buffers": 2
            },
            {
              "compute_ns": 318917,
              "compute_span_ns": 318917,
              "calibration_cpu_ns": 2041958,
              "calibration_gpu_ticks": 2041958,
              "compute_passes": 1,
              "command_buffer_ns": 318958.365824,
              "command_buffers": 2
            },
            {
              "compute_ns": 91666,
              "compute_span_ns": 91666,
              "calibration_cpu_ns": 602875,
              "calibration_gpu_ticks": 602875,
              "compute_passes": 1,
              "command_buffer_ns": 91708.265245,
              "command_buffers": 2
            },
            {
              "compute_ns": 184833,
              "compute_span_ns": 184833,
              "calibration_cpu_ns": 1924208,
              "calibration_gpu_ticks": 1924208,
              "compute_passes": 1,
              "command_buffer_ns": 184791.628271,
              "command_buffers": 2
            },
            {
              "compute_ns": 62791,
              "compute_span_ns": 62791,
              "calibration_cpu_ns": 489709,
              "calibration_gpu_ticks": 489709,
              "compute_passes": 1,
              "command_buffer_ns": 62791.6888334,
              "command_buffers": 2
            },
            {
              "compute_ns": 369333,
              "compute_span_ns": 369333,
              "calibration_cpu_ns": 1651167,
              "calibration_gpu_ticks": 1651167,
              "compute_passes": 1,
              "command_buffer_ns": 369333.254639,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4075583.30335,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4128499.94609,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3977124.9867,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4025333.38638,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3906874.97566,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4169041.64013,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4084000.0147,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3828666.63858,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4174500.06353,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 427166.640293,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54708.3909623,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 476874.993183,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54708.3327547,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54124.975577,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54458.330851,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 57208.3517909,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54750.0094399,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54375.0356883,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              63.680989114843754,
              64.50781165765625,
              62.142577917187495,
              62.8958341621875,
              61.0449214946875,
              65.14127562703125,
              63.8125002296875,
              59.8229162278125,
              65.22656349265625
            ],
            "command_buffer_throughput_us_p50": 63.680989114843754,
            "command_buffer_throughput_us_p90": 65.15833320015625,
            "command_buffer_latency_us": [
              427.166640293,
              54.7083909623,
              476.874993183,
              54.7083327547,
              54.124975577,
              54.458330851,
              57.208351790900004,
              54.7500094399,
              54.375035688299995
            ],
            "command_buffer_latency_us_p50": 54.7083909623,
            "command_buffer_latency_us_p90": 437.108310871
          },
          "compute_throughput_us": [
            64.54621875,
            63.93815625,
            63.193359375,
            64.290359375,
            63.90625,
            63.8046875,
            65.313796875,
            62.565765625,
            64.361984375
          ],
          "compute_throughput_us_p50": 63.93815625,
          "compute_throughput_us_p90": 64.69973437499999,
          "command_buffer_throughput_us": [
            64.5475265628125,
            63.940104155312504,
            63.19401018125,
            64.29231780203125,
            63.90755243046875,
            63.80468767015625,
            65.313802224375,
            62.568359680937505,
            64.363932324375
          ],
          "command_buffer_throughput_us_p50": 63.940104155312504,
          "command_buffer_throughput_us_p90": 64.70078169512499,
          "compute_latency_us": [
            96.334,
            54.541,
            98.75,
            54.833,
            318.917,
            91.666,
            184.833,
            62.791,
            369.333
          ],
          "compute_latency_us_p50": 96.334,
          "compute_latency_us_p90": 329.0002,
          "command_buffer_latency_us": [
            96.4583014138,
            54.5833609067,
            98.8749670796,
            54.9582764506,
            318.958365824,
            91.708265245,
            184.791628271,
            62.7916888334,
            369.333254639
          ],
          "command_buffer_latency_us_p50": 96.4583014138,
          "command_buffer_latency_us_p90": 329.033343587,
          "counter_control_throughput_ratio": 1.016919675676533,
          "counter_control_latency_ratio": 1.146879963597586
        },
        "correctness": {
          "max_abs_error": 1.2343096304160556e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 76.96875,
        "throughput_us_p90": 83.19438956044,
        "latency_us_p50": 353.0,
        "latency_us_p90": 901.6752000000002
      },
      "slowdown": 0.4565225393685353,
      "device_compute_slowdown": 0.3990467444526782,
      "valid": true,
      "round": 1,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "rmsnorm_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-32p7fmdu/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "e81310ef6c7b3435af776cabf332187d02f0ddc801b9234c32af8a56116f248b",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 23.083834,
        "capture_ms": 0.060792,
        "compile_ms": 65.397875,
        "allocation_upload_ms": 1.349708,
        "cold_call_ms": 2.85975,
        "warmup_ms": 200.239833,
        "download_ms": 0.343709,
        "repetitions": 3802,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          6.32251657023,
          5.8110973172,
          5.92162019989,
          6.3086094161,
          6.31095476065,
          6.31872488164,
          6.002783798,
          5.75233429774,
          6.31080142031
        ],
        "latency_us": [
          248.792,
          212.792,
          194.958,
          187.25,
          178.416,
          183.583,
          206.209,
          209.625,
          1288.25
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 19,
            "reduction_scalar_rounds": 24,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 12296,
              "global_write_bytes": 6148,
              "private_read_bytes": 18444,
              "private_write_bytes": 6148
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 64,
              "global_write_bytes": 32,
              "private_read_bytes": 96,
              "private_write_bytes": 32
            },
            "striped_storage_scalars_per_worker": 8,
            "reduction_operations": 1,
            "reduction_elements": 1537,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 48,
            "normalized_kernel_cost": 48,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 3074,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 298500,
              "compute_span_ns": 298500,
              "calibration_cpu_ns": 892375,
              "calibration_gpu_ticks": 892375,
              "compute_passes": 1,
              "command_buffer_ns": 298666.651361,
              "command_buffers": 2
            },
            {
              "compute_ns": 295042,
              "compute_span_ns": 295042,
              "calibration_cpu_ns": 751333,
              "calibration_gpu_ticks": 751333,
              "compute_passes": 1,
              "command_buffer_ns": 295125.006232,
              "command_buffers": 2
            },
            {
              "compute_ns": 943167,
              "compute_span_ns": 943167,
              "calibration_cpu_ns": 1892792,
              "calibration_gpu_ticks": 1892792,
              "compute_passes": 1,
              "command_buffer_ns": 943250.022829,
              "command_buffers": 2
            },
            {
              "compute_ns": 307792,
              "compute_span_ns": 307792,
              "calibration_cpu_ns": 728625,
              "calibration_gpu_ticks": 728625,
              "compute_passes": 1,
              "command_buffer_ns": 307833.368424,
              "command_buffers": 2
            },
            {
              "compute_ns": 980417,
              "compute_span_ns": 980417,
              "calibration_cpu_ns": 2041375,
              "calibration_gpu_ticks": 2041375,
              "compute_passes": 1,
              "command_buffer_ns": 980583.310593,
              "command_buffers": 2
            },
            {
              "compute_ns": 276042,
              "compute_span_ns": 276042,
              "calibration_cpu_ns": 710833,
              "calibration_gpu_ticks": 710833,
              "compute_passes": 1,
              "command_buffer_ns": 276208.273135,
              "command_buffers": 2
            },
            {
              "compute_ns": 307458,
              "compute_span_ns": 307458,
              "calibration_cpu_ns": 761958,
              "calibration_gpu_ticks": 761958,
              "compute_passes": 1,
              "command_buffer_ns": 307583.308313,
              "command_buffers": 2
            },
            {
              "compute_ns": 283250,
              "compute_span_ns": 283250,
              "calibration_cpu_ns": 735542,
              "calibration_gpu_ticks": 735542,
              "compute_passes": 1,
              "command_buffer_ns": 283333.298285,
              "command_buffers": 2
            },
            {
              "compute_ns": 305208,
              "compute_span_ns": 305208,
              "calibration_cpu_ns": 752792,
              "calibration_gpu_ticks": 752792,
              "compute_passes": 1,
              "command_buffer_ns": 305208.37754,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 6541,
              "compute_span_ns": 6541,
              "calibration_cpu_ns": 473667,
              "calibration_gpu_ticks": 473667,
              "compute_passes": 1,
              "command_buffer_ns": 6541.66797176,
              "command_buffers": 2
            },
            {
              "compute_ns": 31250,
              "compute_span_ns": 31250,
              "calibration_cpu_ns": 951375,
              "calibration_gpu_ticks": 951375,
              "compute_passes": 1,
              "command_buffer_ns": 31333.3584927,
              "command_buffers": 2
            },
            {
              "compute_ns": 7000,
              "compute_span_ns": 7000,
              "calibration_cpu_ns": 372000,
              "calibration_gpu_ticks": 372000,
              "compute_passes": 1,
              "command_buffer_ns": 7124.96694177,
              "command_buffers": 2
            },
            {
              "compute_ns": 6500,
              "compute_span_ns": 6500,
              "calibration_cpu_ns": 362708,
              "calibration_gpu_ticks": 362708,
              "compute_passes": 1,
              "command_buffer_ns": 6541.66797176,
              "command_buffers": 2
            },
            {
              "compute_ns": 40917,
              "compute_span_ns": 40917,
              "calibration_cpu_ns": 509375,
              "calibration_gpu_ticks": 509375,
              "compute_passes": 1,
              "command_buffer_ns": 40958.3444707,
              "command_buffers": 2
            },
            {
              "compute_ns": 7334,
              "compute_span_ns": 7334,
              "calibration_cpu_ns": 426708,
              "calibration_gpu_ticks": 426708,
              "compute_passes": 1,
              "command_buffer_ns": 7416.70373827,
              "command_buffers": 2
            },
            {
              "compute_ns": 8959,
              "compute_span_ns": 8959,
              "calibration_cpu_ns": 511000,
              "calibration_gpu_ticks": 511000,
              "compute_passes": 1,
              "command_buffer_ns": 9083.30548555,
              "command_buffers": 2
            },
            {
              "compute_ns": 228041,
              "compute_span_ns": 228041,
              "calibration_cpu_ns": 1607375,
              "calibration_gpu_ticks": 1607375,
              "compute_passes": 1,
              "command_buffer_ns": 228041.724768,
              "command_buffers": 2
            },
            {
              "compute_ns": 6500,
              "compute_span_ns": 6500,
              "calibration_cpu_ns": 344500,
              "calibration_gpu_ticks": 344500,
              "compute_passes": 1,
              "command_buffer_ns": 6583.28644931,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 684250.029735,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 276208.331343,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 263083.318714,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 270916.614681,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 261708.337348,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 307583.308313,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 905583.321583,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 262000.015937,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 975541.712251,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 6624.96313453,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 31083.3565891,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7208.37851986,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 10333.3732113,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7750.00080466,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6500.04949421,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6916.69993103,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7833.35417509,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6458.31460133,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              10.691406714609375,
              4.315755177234375,
              4.11067685490625,
              4.233072104390625,
              4.0891927710625,
              4.8059891923906255,
              14.149739399734376,
              4.093750249015625,
              15.242839253921876
            ],
            "command_buffer_throughput_us_p50": 4.315755177234375,
            "command_buffer_throughput_us_p90": 14.368359370571875,
            "command_buffer_latency_us": [
              6.624963134530001,
              31.0833565891,
              7.20837851986,
              10.333373211300001,
              7.75000080466,
              6.500049494210001,
              6.91669993103,
              7.83335417509,
              6.458314601330001
            ],
            "command_buffer_latency_us_p50": 7.20837851986,
            "command_buffer_latency_us_p90": 14.483369886860004
          },
          "compute_throughput_us": [
            4.6640625,
            4.61003125,
            14.736984375,
            4.80925,
            15.319015625,
            4.31315625,
            4.80403125,
            4.42578125,
            4.768875
          ],
          "compute_throughput_us_p50": 4.768875,
          "compute_throughput_us_p90": 14.853390625000001,
          "command_buffer_throughput_us": [
            4.666666427515626,
            4.6113282223750005,
            14.738281606703126,
            4.809896381625,
            15.321614228015626,
            4.315754267734375,
            4.8059891923906255,
            4.427082785703125,
            4.7688808990625
          ],
          "command_buffer_throughput_us_p50": 4.7688808990625,
          "command_buffer_throughput_us_p90": 14.854948130965626,
          "compute_latency_us": [
            6.541,
            31.25,
            7.0,
            6.5,
            40.917,
            7.334,
            8.959,
            228.041,
            6.5
          ],
          "compute_latency_us_p50": 7.334,
          "compute_latency_us_p90": 78.34180000000003,
          "command_buffer_latency_us": [
            6.54166797176,
            31.3333584927,
            7.1249669417699995,
            6.54166797176,
            40.9583444707,
            7.41670373827,
            9.08330548555,
            228.041724768,
            6.58328644931
          ],
          "command_buffer_latency_us_p50": 7.41670373827,
          "command_buffer_latency_us_p90": 78.37502053016003,
          "counter_control_throughput_ratio": 1.0684869815368059,
          "counter_control_latency_ratio": 1.0193505358122787
        },
        "correctness": {
          "max_abs_error": 1.5564224420394623e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 6.3086094161,
        "throughput_us_p90": 6.319483219358,
        "latency_us_p50": 206.209,
        "latency_us_p90": 456.6836000000002
      },
      "torch": {
        "cold_call_ms": 4.5105,
        "warmup_ms": 200.965792,
        "repetitions": 2662,
        "throughput_us": [
          9.245210368144253,
          9.221606686701728,
          9.367502253944403,
          9.375360255447031,
          10.190927873779113,
          10.248278362133734,
          9.565755822689706,
          9.344352742299023,
          9.645473328324568
        ],
        "latency_us": [
          347.792,
          309.08299999999997,
          234.08399999999997,
          259.458,
          289.625,
          247.709,
          235.083,
          218.417,
          229.25
        ],
        "throughput_us_p50": 9.375360255447031,
        "throughput_us_p90": 10.202397971450038,
        "latency_us_p50": 247.709,
        "latency_us_p90": 316.8248,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 829541.6482724249,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 816083.3385773003,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 291041.68061167,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 305291.6727028787,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 304416.6951440275,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1080750.022083521,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 290333.35158601403,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 728374.9873749912,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 288624.95673820376,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 234958.36649090052,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8083.356078714132,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 23250.002413988113,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8041.67939350009,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8000.002708286047,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 7708.324119448662,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8041.67939350009,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 9374.984074383974,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 7499.998901039362,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              12.96158825425664,
              12.751302165270317,
              4.547526259557344,
              4.77018238598248,
              4.756510861625429,
              16.886719095055014,
              4.536458618531469,
              11.380859177734237,
              4.509764949034434
            ],
            "command_buffer_throughput_us_p50": 4.77018238598248,
            "command_buffer_throughput_us_p90": 13.746614422416314,
            "command_buffer_latency_us": [
              234.95836649090052,
              8.083356078714132,
              23.250002413988113,
              8.04167939350009,
              8.000002708286047,
              7.708324119448662,
              8.04167939350009,
              9.374984074383974,
              7.499998901039362
            ],
            "command_buffer_latency_us_p50": 8.04167939350009,
            "command_buffer_latency_us_p90": 65.59167522937062
          },
          "throughput": [
            {
              "compute_ns": 306916.0,
              "compute_span_ns": 306916.0,
              "command_buffer_ns": 306874.97928738594,
              "calibration_cpu_ns": 937417.0,
              "calibration_gpu_ticks": 937417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 288959.0,
              "compute_span_ns": 288959.0,
              "command_buffer_ns": 288999.98869746923,
              "calibration_cpu_ns": 965209.0,
              "calibration_gpu_ticks": 965209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 289000.0,
              "compute_span_ns": 289000.0,
              "command_buffer_ns": 288999.98869746923,
              "calibration_cpu_ns": 927625.0,
              "calibration_gpu_ticks": 927625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 719042.0,
              "compute_span_ns": 719042.0,
              "command_buffer_ns": 719083.3566710353,
              "calibration_cpu_ns": 2019125.0,
              "calibration_gpu_ticks": 2019125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 784833.0,
              "compute_span_ns": 784833.0,
              "command_buffer_ns": 784833.3334550261,
              "calibration_cpu_ns": 1944250.0,
              "calibration_gpu_ticks": 1944250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 290750.0,
              "compute_span_ns": 290750.0,
              "command_buffer_ns": 290750.00202283263,
              "calibration_cpu_ns": 2011541.0,
              "calibration_gpu_ticks": 2011541.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 890834.0,
              "compute_span_ns": 890834.0,
              "command_buffer_ns": 890875.0023692846,
              "calibration_cpu_ns": 2857958.0,
              "calibration_gpu_ticks": 2857958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 288542.0,
              "compute_span_ns": 288542.0,
              "command_buffer_ns": 288541.6615754366,
              "calibration_cpu_ns": 920917.0,
              "calibration_gpu_ticks": 920917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 297250.0,
              "compute_span_ns": 297250.0,
              "command_buffer_ns": 297249.9933093786,
              "calibration_cpu_ns": 1022834.0,
              "calibration_gpu_ticks": 1022834.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 9000.0,
              "compute_span_ns": 9000.0,
              "command_buffer_ns": 9000.010322779417,
              "calibration_cpu_ns": 632125.0,
              "calibration_gpu_ticks": 632125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 7625.0,
              "compute_span_ns": 7625.0,
              "command_buffer_ns": 7625.02895668149,
              "calibration_cpu_ns": 427917.0,
              "calibration_gpu_ticks": 427917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 9250.0,
              "compute_span_ns": 9250.0,
              "command_buffer_ns": 9250.01222640276,
              "calibration_cpu_ns": 793125.0,
              "calibration_gpu_ticks": 793125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 7750.0,
              "compute_span_ns": 7750.0,
              "command_buffer_ns": 7750.0008046627045,
              "calibration_cpu_ns": 434125.0,
              "calibration_gpu_ticks": 434125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 248292.0,
              "compute_span_ns": 248292.0,
              "command_buffer_ns": 248291.70433804393,
              "calibration_cpu_ns": 1170334.0,
              "calibration_gpu_ticks": 1170334.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 9417.0,
              "compute_span_ns": 9417.0,
              "command_buffer_ns": 9416.660759598017,
              "calibration_cpu_ns": 645917.0,
              "calibration_gpu_ticks": 645917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8125.0,
              "compute_span_ns": 8125.0,
              "command_buffer_ns": 8124.9745562672615,
              "calibration_cpu_ns": 458042.0,
              "calibration_gpu_ticks": 458042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 213375.0,
              "compute_span_ns": 213375.0,
              "command_buffer_ns": 213374.9658241868,
              "calibration_cpu_ns": 1303625.0,
              "calibration_gpu_ticks": 1303625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 7750.0,
              "compute_span_ns": 7750.0,
              "command_buffer_ns": 7750.0008046627045,
              "calibration_cpu_ns": 445542.0,
              "calibration_gpu_ticks": 445542.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            4.7955625,
            4.514984375,
            4.515625,
            11.23503125,
            12.263015625,
            4.54296875,
            13.91928125,
            4.50846875,
            4.64453125
          ],
          "compute_throughput_us_p50": 4.64453125,
          "compute_throughput_us_p90": 12.59426875,
          "command_buffer_throughput_us": [
            4.794921551365405,
            4.515624823397957,
            4.515624823397957,
            11.235677447984926,
            12.263020835234784,
            4.54296878160676,
            13.919921912020072,
            4.508463462116197,
            4.644531145459041
          ],
          "command_buffer_throughput_us_p50": 4.644531145459041,
          "command_buffer_throughput_us_p90": 12.594401050591841,
          "compute_latency_us": [
            9.0,
            7.625,
            9.25,
            7.75,
            248.292,
            9.417,
            8.125,
            213.375,
            7.75
          ],
          "compute_latency_us_p50": 9.0,
          "compute_latency_us_p90": 220.35840000000002,
          "command_buffer_latency_us": [
            9.000010322779417,
            7.62502895668149,
            9.25001222640276,
            7.7500008046627045,
            248.29170433804393,
            9.416660759598017,
            8.124974556267262,
            213.3749658241868,
            7.7500008046627045
          ],
          "command_buffer_latency_us_p50": 9.000010322779417,
          "command_buffer_latency_us_p90": 220.35831352695823,
          "counter_control_throughput_ratio": 0.992984881375376,
          "counter_control_latency_ratio": 1.010357931309037
        },
        "download_ms": 0.319625,
        "allocation_upload_ms": 0.959833,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.807831588873654e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.6728924803113282,
      "device_compute_slowdown": 1.0267720773759463,
      "valid": true,
      "round": 1,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "rmsnorm_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.246542,
        "warmup_ms": 200.140416,
        "repetitions": 1536,
        "throughput_us": [
          9.698215494791667,
          9.211317057291668,
          10.530219401041665,
          10.545356119791666,
          10.935981119791668,
          10.906223307291667,
          10.874321614583332,
          10.836344401041666,
          10.670247395833334
        ],
        "latency_us": [
          308.334,
          222.167,
          300.666,
          248.458,
          260.291,
          260.416,
          219.125,
          236.33399999999997,
          223.208
        ],
        "throughput_us_p50": 10.670247395833334,
        "throughput_us_p90": 10.912174869791667,
        "latency_us_p50": 248.458,
        "latency_us_p90": 302.1996,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 753458.2982771099,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 628750.0145845115,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 288916.69353470206,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 302625.00513345003,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 287750.03738701344,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 303833.3379663527,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 829000.0259876251,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 839791.6681133211,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 284666.66117310524,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 264083.3263285458,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8000.002708286047,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8124.9745562672615,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 7750.0008046627045,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8124.9745562672615,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 9416.660759598017,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 7750.0008046627045,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 184874.9816417694,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 9124.982170760632,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              11.772785910579842,
              9.824218977882992,
              4.51432333647972,
              4.728515705210157,
              4.496094334172085,
              4.747395905724261,
              12.953125406056643,
              13.121744814270642,
              4.447916580829769
            ],
            "command_buffer_throughput_us_p50": 4.747395905724261,
            "command_buffer_throughput_us_p90": 12.986849287699442,
            "command_buffer_latency_us": [
              264.0833263285458,
              8.000002708286047,
              8.124974556267262,
              7.7500008046627045,
              8.124974556267262,
              9.416660759598017,
              7.7500008046627045,
              184.8749816417694,
              9.124982170760632
            ],
            "command_buffer_latency_us_p50": 8.124974556267262,
            "command_buffer_latency_us_p90": 200.7166505791247
          },
          "throughput": [
            {
              "compute_ns": 289584.0,
              "compute_span_ns": 289584.0,
              "command_buffer_ns": 289624.96435269713,
              "calibration_cpu_ns": 936417.0,
              "calibration_gpu_ticks": 936417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 288125.0,
              "compute_span_ns": 288125.0,
              "command_buffer_ns": 288125.011138618,
              "calibration_cpu_ns": 1124542.0,
              "calibration_gpu_ticks": 1124542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 297584.0,
              "compute_span_ns": 297584.0,
              "command_buffer_ns": 297625.0252686441,
              "calibration_cpu_ns": 1133584.0,
              "calibration_gpu_ticks": 1133584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 894042.0,
              "compute_span_ns": 894042.0,
              "command_buffer_ns": 894041.67374596,
              "calibration_cpu_ns": 1742625.0,
              "calibration_gpu_ticks": 1742625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 716667.0,
              "compute_span_ns": 716667.0,
              "command_buffer_ns": 716666.69100523,
              "calibration_cpu_ns": 2143333.0,
              "calibration_gpu_ticks": 2143333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 288417.0,
              "compute_span_ns": 288417.0,
              "command_buffer_ns": 288458.3082050085,
              "calibration_cpu_ns": 978375.0,
              "calibration_gpu_ticks": 978375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 289292.0,
              "compute_span_ns": 289292.0,
              "command_buffer_ns": 289291.6672863066,
              "calibration_cpu_ns": 953500.0,
              "calibration_gpu_ticks": 953500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 289167.0,
              "compute_span_ns": 289167.0,
              "command_buffer_ns": 289166.6954383254,
              "calibration_cpu_ns": 880708.0,
              "calibration_gpu_ticks": 880708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 288250.0,
              "compute_span_ns": 288250.0,
              "command_buffer_ns": 288249.9829865992,
              "calibration_cpu_ns": 872042.0,
              "calibration_gpu_ticks": 872042.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 9458.0,
              "compute_span_ns": 9458.0,
              "command_buffer_ns": 9416.660759598017,
              "calibration_cpu_ns": 627000.0,
              "calibration_gpu_ticks": 627000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 7625.0,
              "compute_span_ns": 7625.0,
              "command_buffer_ns": 7624.9707490205765,
              "calibration_cpu_ns": 674291.0,
              "calibration_gpu_ticks": 674291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 46042.0,
              "compute_span_ns": 46042.0,
              "command_buffer_ns": 46041.61949828267,
              "calibration_cpu_ns": 1356791.0,
              "calibration_gpu_ticks": 1356791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12916.0,
              "compute_span_ns": 12916.0,
              "command_buffer_ns": 12875.01072511077,
              "calibration_cpu_ns": 662000.0,
              "calibration_gpu_ticks": 662000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8209.0,
              "compute_span_ns": 8209.0,
              "command_buffer_ns": 8250.00461190939,
              "calibration_cpu_ns": 430541.0,
              "calibration_gpu_ticks": 430541.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 219708.0,
              "compute_span_ns": 219708.0,
              "command_buffer_ns": 219708.30857753754,
              "calibration_cpu_ns": 1427292.0,
              "calibration_gpu_ticks": 1427292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 7625.0,
              "compute_span_ns": 7625.0,
              "command_buffer_ns": 7625.02895668149,
              "calibration_cpu_ns": 444708.0,
              "calibration_gpu_ticks": 444708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 9833.0,
              "compute_span_ns": 9833.0,
              "command_buffer_ns": 9791.692718863487,
              "calibration_cpu_ns": 724083.0,
              "calibration_gpu_ticks": 724083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 7666.0,
              "compute_span_ns": 7666.0,
              "command_buffer_ns": 7624.9707490205765,
              "calibration_cpu_ns": 422250.0,
              "calibration_gpu_ticks": 422250.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            4.52475,
            4.501953125,
            4.64975,
            13.96940625,
            11.197921875,
            4.506515625,
            4.5201875,
            4.518234375,
            4.50390625
          ],
          "compute_throughput_us_p50": 4.5201875,
          "compute_throughput_us_p90": 11.75221875,
          "command_buffer_throughput_us": [
            4.525390068010893,
            4.501953299040906,
            4.650391019822564,
            13.969401152280625,
            11.197917046956718,
            4.507161065703258,
            4.520182301348541,
            4.5182296162238345,
            4.503905984165613
          ],
          "command_buffer_throughput_us_p50": 4.520182301348541,
          "command_buffer_throughput_us_p90": 11.7522138680215,
          "compute_latency_us": [
            9.458,
            7.625,
            46.042,
            12.916,
            8.209,
            219.708,
            7.625,
            9.833,
            7.666
          ],
          "compute_latency_us_p50": 9.458,
          "compute_latency_us_p90": 80.77520000000004,
          "command_buffer_latency_us": [
            9.416660759598017,
            7.6249707490205765,
            46.04161949828267,
            12.87501072511077,
            8.25000461190939,
            219.70830857753754,
            7.62502895668149,
            9.791692718863487,
            7.6249707490205765
          ],
          "command_buffer_latency_us_p50": 9.416660759598017,
          "command_buffer_latency_us_p90": 80.77495731413367,
          "counter_control_throughput_ratio": 0.949396501831386,
          "counter_control_latency_ratio": 0.9838746019347473
        },
        "download_ms": 0.335916,
        "allocation_upload_ms": 0.686292,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.807831588873654e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-s2w4dlh6/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "54d99166e42515c7bfff1cac173ad46750fa91f9c92e668c07d69318fdc050bf",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 24.571791,
        "capture_ms": 0.063375,
        "compile_ms": 56.550458,
        "allocation_upload_ms": 1.459625,
        "cold_call_ms": 2.034833,
        "warmup_ms": 200.218542,
        "download_ms": 0.436916,
        "repetitions": 4920,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          5.96831808943,
          5.51560813008,
          6.26676829268,
          5.53616199187,
          5.66664126016,
          5.92042682927,
          5.61729329268,
          6.02071463415,
          5.60112642276
        ],
        "latency_us": [
          287.375,
          251.625,
          242.5,
          207.791,
          202.417,
          195.625,
          2024.333,
          319.25,
          321.292
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 32,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 37,
            "reduction_scalar_rounds": 24,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 12296,
              "global_write_bytes": 6148,
              "private_read_bytes": 18444,
              "private_write_bytes": 6148
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 64,
              "global_write_bytes": 32,
              "private_read_bytes": 96,
              "private_write_bytes": 32
            },
            "striped_storage_scalars_per_worker": 8,
            "reduction_operations": 1,
            "reduction_elements": 1537,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 56,
            "normalized_kernel_cost": 56,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 3074,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 245209,
              "compute_span_ns": 245209,
              "calibration_cpu_ns": 782208,
              "calibration_gpu_ticks": 782208,
              "compute_passes": 1,
              "command_buffer_ns": 245291.623287,
              "command_buffers": 2
            },
            {
              "compute_ns": 243125,
              "compute_span_ns": 243125,
              "calibration_cpu_ns": 768459,
              "calibration_gpu_ticks": 768459,
              "compute_passes": 1,
              "command_buffer_ns": 243208.312895,
              "command_buffers": 2
            },
            {
              "compute_ns": 245209,
              "compute_span_ns": 245209,
              "calibration_cpu_ns": 705625,
              "calibration_gpu_ticks": 705625,
              "compute_passes": 1,
              "command_buffer_ns": 245375.034865,
              "command_buffers": 2
            },
            {
              "compute_ns": 252875,
              "compute_span_ns": 252875,
              "calibration_cpu_ns": 708708,
              "calibration_gpu_ticks": 708708,
              "compute_passes": 1,
              "command_buffer_ns": 253000.005614,
              "command_buffers": 2
            },
            {
              "compute_ns": 249750,
              "compute_span_ns": 249750,
              "calibration_cpu_ns": 793375,
              "calibration_gpu_ticks": 793375,
              "compute_passes": 1,
              "command_buffer_ns": 249833.27603,
              "command_buffers": 2
            },
            {
              "compute_ns": 593209,
              "compute_span_ns": 593209,
              "calibration_cpu_ns": 1859625,
              "calibration_gpu_ticks": 1859625,
              "compute_passes": 1,
              "command_buffer_ns": 593333.272263,
              "command_buffers": 2
            },
            {
              "compute_ns": 244125,
              "compute_span_ns": 244125,
              "calibration_cpu_ns": 666625,
              "calibration_gpu_ticks": 666625,
              "compute_passes": 1,
              "command_buffer_ns": 244250.055403,
              "command_buffers": 2
            },
            {
              "compute_ns": 882125,
              "compute_span_ns": 882125,
              "calibration_cpu_ns": 1731166,
              "calibration_gpu_ticks": 1731166,
              "compute_passes": 1,
              "command_buffer_ns": 882208.347321,
              "command_buffers": 2
            },
            {
              "compute_ns": 244875,
              "compute_span_ns": 244875,
              "calibration_cpu_ns": 694750,
              "calibration_gpu_ticks": 694750,
              "compute_passes": 1,
              "command_buffer_ns": 244874.97285,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 7667,
              "compute_span_ns": 7667,
              "calibration_cpu_ns": 611416,
              "calibration_gpu_ticks": 611416,
              "compute_passes": 1,
              "command_buffer_ns": 7708.32411945,
              "command_buffers": 2
            },
            {
              "compute_ns": 6500,
              "compute_span_ns": 6500,
              "calibration_cpu_ns": 399542,
              "calibration_gpu_ticks": 399542,
              "compute_passes": 1,
              "command_buffer_ns": 6500.04949421,
              "command_buffers": 2
            },
            {
              "compute_ns": 19208,
              "compute_span_ns": 19208,
              "calibration_cpu_ns": 853667,
              "calibration_gpu_ticks": 853667,
              "compute_passes": 1,
              "command_buffer_ns": 19333.3253264,
              "command_buffers": 2
            },
            {
              "compute_ns": 7459,
              "compute_span_ns": 7459,
              "calibration_cpu_ns": 503000,
              "calibration_gpu_ticks": 503000,
              "compute_passes": 1,
              "command_buffer_ns": 7583.29406381,
              "command_buffers": 2
            },
            {
              "compute_ns": 6750,
              "compute_span_ns": 6750,
              "calibration_cpu_ns": 400375,
              "calibration_gpu_ticks": 400375,
              "compute_passes": 1,
              "command_buffer_ns": 6791.61166772,
              "command_buffers": 2
            },
            {
              "compute_ns": 6708,
              "compute_span_ns": 6708,
              "calibration_cpu_ns": 391541,
              "calibration_gpu_ticks": 391541,
              "compute_passes": 1,
              "command_buffer_ns": 6958.31840858,
              "command_buffers": 2
            },
            {
              "compute_ns": 6708,
              "compute_span_ns": 6708,
              "calibration_cpu_ns": 474084,
              "calibration_gpu_ticks": 474084,
              "compute_passes": 1,
              "command_buffer_ns": 6791.66987538,
              "command_buffers": 2
            },
            {
              "compute_ns": 6334,
              "compute_span_ns": 6334,
              "calibration_cpu_ns": 389875,
              "calibration_gpu_ticks": 389875,
              "compute_passes": 1,
              "command_buffer_ns": 6416.69612378,
              "command_buffers": 2
            },
            {
              "compute_ns": 73625,
              "compute_span_ns": 73625,
              "calibration_cpu_ns": 1051250,
              "calibration_gpu_ticks": 1051250,
              "compute_passes": 1,
              "command_buffer_ns": 73749.9794923,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 848833.355121,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 243499.991484,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 840916.705783,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 821583.322249,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 283958.27394,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 265458.249487,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 269666.605163,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 262958.288658,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 261625.042185,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 228999.997489,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6999.93688613,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6875.02324581,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6749.99319017,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6458.37280899,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 69625.035394,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7624.97074902,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6833.3465606,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7333.29216018,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              13.263021173765624,
              3.8046873669375,
              13.139323527859375,
              12.837239410140624,
              4.4368480303124995,
              4.147785148234375,
              4.213540705671875,
              4.10872326028125,
              4.087891284140625
            ],
            "command_buffer_throughput_us_p50": 4.213540705671875,
            "command_buffer_throughput_us_p90": 13.164063057040625,
            "command_buffer_latency_us": [
              228.999997489,
              6.9999368861299995,
              6.87502324581,
              6.7499931901700005,
              6.458372808989999,
              69.62503539400001,
              7.62497074902,
              6.8333465606,
              7.33329216018
            ],
            "command_buffer_latency_us_p50": 6.9999368861299995,
            "command_buffer_latency_us_p90": 101.50002781300003
          },
          "compute_throughput_us": [
            3.831390625,
            3.798828125,
            3.831390625,
            3.951171875,
            3.90234375,
            9.268890625,
            3.814453125,
            13.783203125,
            3.826171875
          ],
          "compute_throughput_us_p50": 3.831390625,
          "compute_throughput_us_p90": 10.171753125,
          "command_buffer_throughput_us": [
            3.832681613859375,
            3.800129888984375,
            3.833984919765625,
            3.95312508771875,
            3.90364493796875,
            9.270832379109375,
            3.816407115671875,
            13.784505426890624,
            3.8261714507812496
          ],
          "command_buffer_throughput_us_p50": 3.833984919765625,
          "command_buffer_throughput_us_p90": 10.173566988665625,
          "compute_latency_us": [
            7.667,
            6.5,
            19.208,
            7.459,
            6.75,
            6.708,
            6.708,
            6.334,
            73.625
          ],
          "compute_latency_us_p50": 6.75,
          "compute_latency_us_p90": 30.091400000000007,
          "command_buffer_latency_us": [
            7.708324119449999,
            6.500049494210001,
            19.3333253264,
            7.5832940638099995,
            6.79161166772,
            6.95831840858,
            6.79166987538,
            6.4166961237799995,
            73.7499794923
          ],
          "command_buffer_latency_us_p50": 6.95831840858,
          "command_buffer_latency_us_p90": 30.216656159580012,
          "counter_control_throughput_ratio": 0.9057482488621943,
          "counter_control_latency_ratio": 0.9390268833689277
        },
        "correctness": {
          "max_abs_error": 1.5564224420394623e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 5.66664126016,
        "throughput_us_p90": 6.069925365856,
        "latency_us_p50": 251.625,
        "latency_us_p90": 661.9002000000003
      },
      "slowdown": 0.5310693416886274,
      "device_compute_slowdown": 0.8476176320119465,
      "valid": true,
      "round": 1,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "rmsnorm_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-5ijhkfzw/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "b5a32df1046f38df4f8e9cbba5277df92f4beb4d4f9631ac4564ef5cda134264",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.461541,
        "capture_ms": 0.062125,
        "compile_ms": 55.984791,
        "allocation_upload_ms": 2.120833,
        "cold_call_ms": 4.214041,
        "warmup_ms": 200.831333,
        "download_ms": 0.924209,
        "repetitions": 1254,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          21.581539075,
          20.8371212121,
          20.8992886762,
          20.8730398724,
          22.981891547,
          20.7150454545,
          20.6769665072,
          20.7233181818,
          20.8193452951
        ],
        "latency_us": [
          380.959,
          541.667,
          1640.25,
          452.291,
          306.167,
          359.833,
          240.917,
          240.25,
          208.292
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 32,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 256,
            "reduction_scalar_rounds": 36,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 24576,
              "global_write_bytes": 12288,
              "private_read_bytes": 36864,
              "private_write_bytes": 12288
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 48,
              "private_read_bytes": 144,
              "private_write_bytes": 48
            },
            "striped_storage_scalars_per_worker": 12,
            "reduction_operations": 1,
            "reduction_elements": 3072,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 68,
            "normalized_kernel_cost": 272,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 6144,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 974583,
              "compute_span_ns": 974583,
              "calibration_cpu_ns": 1525500,
              "calibration_gpu_ticks": 1525500,
              "compute_passes": 1,
              "command_buffer_ns": 974708.353169,
              "command_buffers": 2
            },
            {
              "compute_ns": 947500,
              "compute_span_ns": 947500,
              "calibration_cpu_ns": 1425459,
              "calibration_gpu_ticks": 1425459,
              "compute_passes": 1,
              "command_buffer_ns": 947624.96883,
              "command_buffers": 2
            },
            {
              "compute_ns": 1497292,
              "compute_span_ns": 1497292,
              "calibration_cpu_ns": 2902834,
              "calibration_gpu_ticks": 2902834,
              "compute_passes": 1,
              "command_buffer_ns": 1497416.69884,
              "command_buffers": 2
            },
            {
              "compute_ns": 910375,
              "compute_span_ns": 910375,
              "calibration_cpu_ns": 1474333,
              "calibration_gpu_ticks": 1474333,
              "compute_passes": 1,
              "command_buffer_ns": 910416.652914,
              "command_buffers": 2
            },
            {
              "compute_ns": 1607333,
              "compute_span_ns": 1607333,
              "calibration_cpu_ns": 2576041,
              "calibration_gpu_ticks": 2576041,
              "compute_passes": 1,
              "command_buffer_ns": 1607416.66332,
              "command_buffers": 2
            },
            {
              "compute_ns": 905958,
              "compute_span_ns": 905958,
              "calibration_cpu_ns": 1378500,
              "calibration_gpu_ticks": 1378500,
              "compute_passes": 1,
              "command_buffer_ns": 905999.972019,
              "command_buffers": 2
            },
            {
              "compute_ns": 1938375,
              "compute_span_ns": 1938375,
              "calibration_cpu_ns": 3054333,
              "calibration_gpu_ticks": 3054333,
              "compute_passes": 1,
              "command_buffer_ns": 1938416.68079,
              "command_buffers": 2
            },
            {
              "compute_ns": 946584,
              "compute_span_ns": 946584,
              "calibration_cpu_ns": 1499167,
              "calibration_gpu_ticks": 1499167,
              "compute_passes": 1,
              "command_buffer_ns": 946750.049479,
              "command_buffers": 2
            },
            {
              "compute_ns": 1528000,
              "compute_span_ns": 1528000,
              "calibration_cpu_ns": 2396625,
              "calibration_gpu_ticks": 2396625,
              "compute_passes": 1,
              "command_buffer_ns": 1528083.34678,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 16333,
              "compute_span_ns": 16333,
              "calibration_cpu_ns": 473542,
              "calibration_gpu_ticks": 473542,
              "compute_passes": 1,
              "command_buffer_ns": 16291.6257977,
              "command_buffers": 2
            },
            {
              "compute_ns": 24500,
              "compute_span_ns": 24500,
              "calibration_cpu_ns": 1015917,
              "calibration_gpu_ticks": 1015917,
              "compute_passes": 1,
              "command_buffer_ns": 24625.0419877,
              "command_buffers": 2
            },
            {
              "compute_ns": 16208,
              "compute_span_ns": 16208,
              "calibration_cpu_ns": 387750,
              "calibration_gpu_ticks": 387750,
              "compute_passes": 1,
              "command_buffer_ns": 16166.6539498,
              "command_buffers": 2
            },
            {
              "compute_ns": 18709,
              "compute_span_ns": 18709,
              "calibration_cpu_ns": 760209,
              "calibration_gpu_ticks": 760209,
              "compute_passes": 1,
              "command_buffer_ns": 18833.3215192,
              "command_buffers": 2
            },
            {
              "compute_ns": 20792,
              "compute_span_ns": 20792,
              "calibration_cpu_ns": 669959,
              "calibration_gpu_ticks": 669959,
              "compute_passes": 1,
              "command_buffer_ns": 20874.9552257,
              "command_buffers": 2
            },
            {
              "compute_ns": 16416,
              "compute_span_ns": 16416,
              "calibration_cpu_ns": 438000,
              "calibration_gpu_ticks": 438000,
              "compute_passes": 1,
              "command_buffer_ns": 16416.7140611,
              "command_buffers": 2
            },
            {
              "compute_ns": 126875,
              "compute_span_ns": 126875,
              "calibration_cpu_ns": 1300041,
              "calibration_gpu_ticks": 1300041,
              "compute_passes": 1,
              "command_buffer_ns": 126916.682348,
              "command_buffers": 2
            },
            {
              "compute_ns": 21625,
              "compute_span_ns": 21625,
              "calibration_cpu_ns": 654375,
              "calibration_gpu_ticks": 654375,
              "compute_passes": 1,
              "command_buffer_ns": 21666.6376218,
              "command_buffers": 2
            },
            {
              "compute_ns": 18500,
              "compute_span_ns": 18500,
              "calibration_cpu_ns": 493666,
              "calibration_gpu_ticks": 493666,
              "compute_passes": 1,
              "command_buffer_ns": 18541.6429304,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1436125.02143,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1679083.33242,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 913083.378691,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1607416.60511,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 915999.989957,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1515250.01274,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1176333.34594,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1836625.04423,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 919166.661333,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 16666.657757,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 48916.6122861,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 16374.9791682,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18541.6429304,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 214458.326809,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18833.3215192,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19041.6467376,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 16750.0693351,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18541.6429304,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              22.43945345984375,
              26.2356770690625,
              14.266927792046875,
              25.115884454843748,
              14.312499843078125,
              23.675781449062498,
              18.380208530312498,
              28.697266316093753,
              14.361979083328126
            ],
            "command_buffer_throughput_us_p50": 22.43945345984375,
            "command_buffer_throughput_us_p90": 26.727994918468752,
            "command_buffer_latency_us": [
              16.666657757,
              48.9166122861,
              16.3749791682,
              18.541642930400002,
              214.458326809,
              18.8333215192,
              19.0416467376,
              16.7500693351,
              18.541642930400002
            ],
            "command_buffer_latency_us_p50": 18.541642930400002,
            "command_buffer_latency_us_p90": 82.02495519068003
          },
          "compute_throughput_us": [
            15.227859375,
            14.8046875,
            23.3951875,
            14.224609375,
            25.114578125,
            14.15559375,
            30.287109375,
            14.790375,
            23.875
          ],
          "compute_throughput_us_p50": 15.227859375,
          "compute_throughput_us_p90": 26.149084375,
          "command_buffer_throughput_us": [
            15.229818018265625,
            14.806640137968751,
            23.397135919375,
            14.22526020178125,
            25.115885364375,
            14.156249562796875,
            30.287760637343748,
            14.792969523109376,
            23.8763022934375
          ],
          "command_buffer_throughput_us_p50": 15.229818018265625,
          "command_buffer_throughput_us_p90": 26.15026041896875,
          "compute_latency_us": [
            16.333,
            24.5,
            16.208,
            18.709,
            20.792,
            16.416,
            126.875,
            21.625,
            18.5
          ],
          "compute_latency_us_p50": 18.709,
          "compute_latency_us_p90": 44.97500000000002,
          "command_buffer_latency_us": [
            16.2916257977,
            24.625041987699998,
            16.1666539498,
            18.8333215192,
            20.8749552257,
            16.4167140611,
            126.91668234800001,
            21.6666376218,
            18.541642930400002
          ],
          "command_buffer_latency_us_p50": 18.8333215192,
          "command_buffer_latency_us_p90": 45.083370059760014,
          "counter_control_throughput_ratio": 0.6787071728605137,
          "counter_control_latency_ratio": 0.9872778330732436
        },
        "correctness": {
          "max_abs_error": 1.6175184480715643e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 20.8371212121,
        "throughput_us_p90": 21.8616095694,
        "latency_us_p50": 359.833,
        "latency_us_p90": 761.3836000000002
      },
      "torch": {
        "cold_call_ms": 5.270042,
        "warmup_ms": 200.291292,
        "repetitions": 1142,
        "throughput_us": [
          27.546264448336252,
          26.494964973730298,
          26.979093695271455,
          26.867703152364275,
          27.284442206654994,
          25.612558669001753,
          27.05385288966725,
          26.31819176882662,
          25.902364273204903
        ],
        "latency_us": [
          400.166,
          290.45799999999997,
          226.208,
          220.542,
          253.166,
          287.66700000000003,
          389.04100000000005,
          278.708,
          304.58299999999997
        ],
        "throughput_us_p50": 26.867703152364275,
        "throughput_us_p90": 27.336806654991246,
        "latency_us_p50": 287.66700000000003,
        "latency_us_p90": 391.266,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1402375.0554770231,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1036499.9761804938,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1384083.298034966,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1033750.013448298,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1457499.9804608524,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1035458.2918807864,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1500916.6672825813,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1044000.0332891941,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1442791.661247611,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 252666.70854762197,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 20999.98528137803,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 20916.631910949945,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 18874.998204410076,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21333.34055542946,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 23583.35768803954,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 18708.291463553905,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 30249.99750778079,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 180750.03754347563,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              21.912110241828486,
              16.195312127820216,
              21.626301531796344,
              16.152343960129656,
              22.77343719470082,
              16.179035810637288,
              23.451822926290333,
              16.312500520143658,
              22.543619706993923
            ],
            "command_buffer_throughput_us_p50": 21.626301531796344,
            "command_buffer_throughput_us_p90": 22.90911434101872,
            "command_buffer_latency_us": [
              252.66670854762197,
              20.99998528137803,
              20.916631910949945,
              18.874998204410076,
              21.33334055542946,
              23.58335768803954,
              18.708291463553905,
              30.24999750778079,
              180.75003754347563
            ],
            "command_buffer_latency_us_p50": 21.33334055542946,
            "command_buffer_latency_us_p90": 195.1333717443049
          },
          "throughput": [
            {
              "compute_ns": 1034250.0,
              "compute_span_ns": 1034250.0,
              "command_buffer_ns": 1034250.0172555447,
              "calibration_cpu_ns": 1811791.0,
              "calibration_gpu_ticks": 1811791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1429625.0,
              "compute_span_ns": 1429625.0,
              "command_buffer_ns": 1429624.971933663,
              "calibration_cpu_ns": 2374000.0,
              "calibration_gpu_ticks": 2374000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1035250.0,
              "compute_span_ns": 1035250.0,
              "command_buffer_ns": 1035249.9666623771,
              "calibration_cpu_ns": 1751417.0,
              "calibration_gpu_ticks": 1751417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1441292.0,
              "compute_span_ns": 1441292.0,
              "command_buffer_ns": 1441333.326511085,
              "calibration_cpu_ns": 2800875.0,
              "calibration_gpu_ticks": 2800875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1054250.0,
              "compute_span_ns": 1054250.0,
              "command_buffer_ns": 1054249.9949224293,
              "calibration_cpu_ns": 1870667.0,
              "calibration_gpu_ticks": 1870667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1266791.0,
              "compute_span_ns": 1266791.0,
              "command_buffer_ns": 1266750.0413954258,
              "calibration_cpu_ns": 2172917.0,
              "calibration_gpu_ticks": 2172917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1041125.0,
              "compute_span_ns": 1041125.0,
              "command_buffer_ns": 1041124.9822936952,
              "calibration_cpu_ns": 1904250.0,
              "calibration_gpu_ticks": 1904250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1551958.0,
              "compute_span_ns": 1551958.0,
              "command_buffer_ns": 1551958.3830609918,
              "calibration_cpu_ns": 2470291.0,
              "calibration_gpu_ticks": 2470291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1033125.0,
              "compute_span_ns": 1033125.0,
              "command_buffer_ns": 1033124.9795854092,
              "calibration_cpu_ns": 1773500.0,
              "calibration_gpu_ticks": 1773500.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 24084.0,
              "compute_span_ns": 24084.0,
              "command_buffer_ns": 24124.979972839355,
              "calibration_cpu_ns": 751250.0,
              "calibration_gpu_ticks": 751250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 19042.0,
              "compute_span_ns": 19042.0,
              "command_buffer_ns": 19083.38163048029,
              "calibration_cpu_ns": 465708.0,
              "calibration_gpu_ticks": 465708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 204125.0,
              "compute_span_ns": 204125.0,
              "command_buffer_ns": 204125.01180544496,
              "calibration_cpu_ns": 1604000.0,
              "calibration_gpu_ticks": 1604000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 25541.0,
              "compute_span_ns": 25541.0,
              "command_buffer_ns": 25500.019546598196,
              "calibration_cpu_ns": 797000.0,
              "calibration_gpu_ticks": 797000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21000.0,
              "compute_span_ns": 21000.0,
              "command_buffer_ns": 20999.98528137803,
              "calibration_cpu_ns": 436750.0,
              "calibration_gpu_ticks": 436750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 100791.0,
              "compute_span_ns": 100791.0,
              "command_buffer_ns": 100750.01046061516,
              "calibration_cpu_ns": 1155042.0,
              "calibration_gpu_ticks": 1155042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 20958.0,
              "compute_span_ns": 20958.0,
              "command_buffer_ns": 20958.308596163988,
              "calibration_cpu_ns": 426125.0,
              "calibration_gpu_ticks": 426125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 20917.0,
              "compute_span_ns": 20917.0,
              "command_buffer_ns": 20916.69011861086,
              "calibration_cpu_ns": 573458.0,
              "calibration_gpu_ticks": 573458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 24750.0,
              "compute_span_ns": 24750.0,
              "command_buffer_ns": 24750.01383572817,
              "calibration_cpu_ns": 753250.0,
              "calibration_gpu_ticks": 753250.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            16.16015625,
            22.337890625,
            16.17578125,
            22.5201875,
            16.47265625,
            19.793609375,
            16.267578125,
            24.24934375,
            16.142578125
          ],
          "compute_throughput_us_p50": 16.47265625,
          "compute_throughput_us_p90": 22.86601875,
          "command_buffer_throughput_us": [
            16.160156519617885,
            22.337890186463483,
            16.175780729099642,
            22.520833226735704,
            16.472656170662958,
            19.792969396803528,
            16.267577848338988,
            24.249349735327996,
            16.142577806022018
          ],
          "command_buffer_throughput_us_p50": 16.472656170662958,
          "command_buffer_throughput_us_p90": 22.866536528454162,
          "compute_latency_us": [
            24.084,
            19.042,
            204.125,
            25.541,
            21.0,
            100.791,
            20.958,
            20.917,
            24.75
          ],
          "compute_latency_us_p50": 24.084,
          "compute_latency_us_p90": 121.45780000000002,
          "command_buffer_latency_us": [
            24.124979972839355,
            19.08338163048029,
            204.12501180544496,
            25.500019546598196,
            20.99998528137803,
            100.75001046061516,
            20.958308596163988,
            20.91669011861086,
            24.75001383572817
          ],
          "command_buffer_latency_us_p50": 24.124979972839355,
          "command_buffer_latency_us_p90": 121.42501072958113,
          "counter_control_throughput_ratio": 0.7479679641623879,
          "counter_control_latency_ratio": 0.9843739768188069
        },
        "download_ms": 0.366834,
        "allocation_upload_ms": 1.552125,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.8163651027691685e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.7755453115562057,
      "device_compute_slowdown": 0.9244325349774721,
      "valid": true,
      "round": 1,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "rmsnorm_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.388375,
        "warmup_ms": 200.086791,
        "repetitions": 964,
        "throughput_us": [
          26.59664522821577,
          26.827196058091285,
          26.00553215767635,
          25.775414937759336,
          25.909060165975106,
          25.958290456431538,
          25.838692946058092,
          25.84936929460581,
          25.615534232365146
        ],
        "latency_us": [
          458.25,
          269.5,
          2129.25,
          386.084,
          309.417,
          336.95799999999997,
          324.20799999999997,
          247.25,
          232.958
        ],
        "throughput_us_p50": 25.909060165975106,
        "throughput_us_p90": 26.642755394190875,
        "latency_us_p50": 324.20799999999997,
        "latency_us_p90": 792.4500000000003,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1070458.3255574107,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1625875.011086464,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1032375.0320822,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1427374.9548010528,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1041333.3657197654,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1531374.9900087714,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1057874.9934211373,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1578083.320055157,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1035833.3820477128,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 18750.02635642886,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22791.67529195547,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 18874.998204410076,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 273666.6356213391,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 24500.011932104826,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 20999.98528137803,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 140333.3735652268,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21249.987185001373,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21041.661966592073,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              16.725911336834542,
              25.404297048226,
              16.130859876284376,
              22.30273366876645,
              16.270833839371335,
              23.927734218887053,
              16.52929677220527,
              24.657551875861827,
              16.184896594495513
            ],
            "command_buffer_throughput_us_p50": 16.725911336834542,
            "command_buffer_throughput_us_p90": 24.80690091033466,
            "command_buffer_latency_us": [
              18.75002635642886,
              22.79167529195547,
              18.874998204410076,
              273.6666356213391,
              24.500011932104826,
              20.99998528137803,
              140.3333735652268,
              21.249987185001373,
              21.041661966592073
            ],
            "command_buffer_latency_us_p50": 21.249987185001373,
            "command_buffer_latency_us_p90": 167.00002597644928
          },
          "throughput": [
            {
              "compute_ns": 1585250.0,
              "compute_span_ns": 1585250.0,
              "command_buffer_ns": 1585250.0218898058,
              "calibration_cpu_ns": 2910375.0,
              "calibration_gpu_ticks": 2910375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1036709.0,
              "compute_span_ns": 1036709.0,
              "command_buffer_ns": 1036750.0362917781,
              "calibration_cpu_ns": 1724875.0,
              "calibration_gpu_ticks": 1724875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1545042.0,
              "compute_span_ns": 1545042.0,
              "command_buffer_ns": 1545041.6831299663,
              "calibration_cpu_ns": 2658583.0,
              "calibration_gpu_ticks": 2658583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1050292.0,
              "compute_span_ns": 1050292.0,
              "command_buffer_ns": 1050291.64114967,
              "calibration_cpu_ns": 1815375.0,
              "calibration_gpu_ticks": 1815375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1434542.0,
              "compute_span_ns": 1434542.0,
              "command_buffer_ns": 1434541.6566357017,
              "calibration_cpu_ns": 2727666.0,
              "calibration_gpu_ticks": 2727666.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1030334.0,
              "compute_span_ns": 1030334.0,
              "command_buffer_ns": 1030374.9586455524,
              "calibration_cpu_ns": 1685333.0,
              "calibration_gpu_ticks": 1685333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1491084.0,
              "compute_span_ns": 1491084.0,
              "command_buffer_ns": 1491124.9745637178,
              "calibration_cpu_ns": 2686667.0,
              "calibration_gpu_ticks": 2686667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1140458.0,
              "compute_span_ns": 1140458.0,
              "command_buffer_ns": 1140416.6580177844,
              "calibration_cpu_ns": 1747500.0,
              "calibration_gpu_ticks": 1747500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1496583.0,
              "compute_span_ns": 1496583.0,
              "command_buffer_ns": 1496583.3397582173,
              "calibration_cpu_ns": 2677417.0,
              "calibration_gpu_ticks": 2677417.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 18959.0,
              "compute_span_ns": 18959.0,
              "command_buffer_ns": 19000.028260052204,
              "calibration_cpu_ns": 467542.0,
              "calibration_gpu_ticks": 467542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 214167.0,
              "compute_span_ns": 214167.0,
              "command_buffer_ns": 214208.32490548491,
              "calibration_cpu_ns": 1233750.0,
              "calibration_gpu_ticks": 1233750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 18791.0,
              "compute_span_ns": 18791.0,
              "command_buffer_ns": 18750.02635642886,
              "calibration_cpu_ns": 545208.0,
              "calibration_gpu_ticks": 545208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21083.0,
              "compute_span_ns": 21083.0,
              "command_buffer_ns": 21041.661966592073,
              "calibration_cpu_ns": 530208.0,
              "calibration_gpu_ticks": 530208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 18833.0,
              "compute_span_ns": 18833.0,
              "command_buffer_ns": 18833.321519196033,
              "calibration_cpu_ns": 502125.0,
              "calibration_gpu_ticks": 502125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 20917.0,
              "compute_span_ns": 20917.0,
              "command_buffer_ns": 20916.69011861086,
              "calibration_cpu_ns": 542458.0,
              "calibration_gpu_ticks": 542458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 22625.0,
              "compute_span_ns": 22625.0,
              "command_buffer_ns": 22624.9685510993,
              "calibration_cpu_ns": 740000.0,
              "calibration_gpu_ticks": 740000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 18584.0,
              "compute_span_ns": 18584.0,
              "command_buffer_ns": 18624.996300786734,
              "calibration_cpu_ns": 464833.0,
              "calibration_gpu_ticks": 464833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 125667.0,
              "compute_span_ns": 125667.0,
              "command_buffer_ns": 125708.34951475263,
              "calibration_cpu_ns": 1387917.0,
              "calibration_gpu_ticks": 1387917.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            24.76953125,
            16.198578125,
            24.14128125,
            16.4108125,
            22.41471875,
            16.09896875,
            23.2981875,
            17.81965625,
            23.384109375
          ],
          "compute_throughput_us_p50": 22.41471875,
          "compute_throughput_us_p90": 24.26693125,
          "command_buffer_throughput_us": [
            24.769531592028216,
            16.199219317059033,
            24.141276298905723,
            16.410806892963592,
            22.41471338493284,
            16.099608728836756,
            23.29882772755809,
            17.81901028152788,
            23.384114683722146
          ],
          "command_buffer_throughput_us_p50": 22.41471338493284,
          "command_buffer_throughput_us_p90": 24.26692735753022,
          "compute_latency_us": [
            18.959,
            214.167,
            18.791,
            21.083,
            18.833,
            20.917,
            22.625,
            18.584,
            125.667
          ],
          "compute_latency_us_p50": 20.917,
          "compute_latency_us_p90": 143.36700000000002,
          "command_buffer_latency_us": [
            19.000028260052204,
            214.20832490548491,
            18.75002635642886,
            21.041661966592073,
            18.833321519196033,
            20.91669011861086,
            22.6249685510993,
            18.624996300786734,
            125.70834951475263
          ],
          "command_buffer_latency_us_p50": 20.91669011861086,
          "command_buffer_latency_us_p90": 143.40834459289908,
          "counter_control_throughput_ratio": 1.377600779788855,
          "counter_control_latency_ratio": 0.9933789743115305
        },
        "download_ms": 0.503667,
        "allocation_upload_ms": 0.689917,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.8163651027691685e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-d_of3u67/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "a8c235e7c8be57d97298e1f9f70525f6c2c7f4ae4cee9647b15503a46aab5ade",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 21.521625,
        "capture_ms": 0.05975,
        "compile_ms": 60.368,
        "allocation_upload_ms": 3.7405,
        "cold_call_ms": 5.411958,
        "warmup_ms": 202.722375,
        "download_ms": 0.787166,
        "repetitions": 1202,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          21.2658760399,
          21.5927279534,
          21.4755266223,
          22.6314126456,
          21.0144201331,
          21.354687188,
          21.1523843594,
          21.5057545757,
          21.4260257903
        ],
        "latency_us": [
          325.916,
          256.542,
          243,
          269.458,
          250.833,
          303.5,
          313.625,
          262.333,
          296.459
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 128,
            "reduction_scalar_rounds": 36,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 24576,
              "global_write_bytes": 12288,
              "private_read_bytes": 36864,
              "private_write_bytes": 12288
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 48,
              "private_read_bytes": 144,
              "private_write_bytes": 48
            },
            "striped_storage_scalars_per_worker": 12,
            "reduction_operations": 1,
            "reduction_elements": 3072,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 60,
            "normalized_kernel_cost": 240,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 6144,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 974750,
              "compute_span_ns": 974750,
              "calibration_cpu_ns": 1476166,
              "calibration_gpu_ticks": 1476166,
              "compute_passes": 1,
              "command_buffer_ns": 974875.001702,
              "command_buffers": 2
            },
            {
              "compute_ns": 1567791,
              "compute_span_ns": 1567791,
              "calibration_cpu_ns": 2710584,
              "calibration_gpu_ticks": 2710584,
              "compute_passes": 1,
              "command_buffer_ns": 1567875.03511,
              "command_buffers": 2
            },
            {
              "compute_ns": 1007000,
              "compute_span_ns": 1007000,
              "calibration_cpu_ns": 2445958,
              "calibration_gpu_ticks": 2445958,
              "compute_passes": 1,
              "command_buffer_ns": 1007041.66107,
              "command_buffers": 2
            },
            {
              "compute_ns": 1949208,
              "compute_span_ns": 1949208,
              "calibration_cpu_ns": 2772333,
              "calibration_gpu_ticks": 2772333,
              "compute_passes": 1,
              "command_buffer_ns": 1949208.38112,
              "command_buffers": 2
            },
            {
              "compute_ns": 958583,
              "compute_span_ns": 958583,
              "calibration_cpu_ns": 1490750,
              "calibration_gpu_ticks": 1490750,
              "compute_passes": 1,
              "command_buffer_ns": 958624.994382,
              "command_buffers": 2
            },
            {
              "compute_ns": 1517500,
              "compute_span_ns": 1517500,
              "calibration_cpu_ns": 2349750,
              "calibration_gpu_ticks": 2349750,
              "compute_passes": 1,
              "command_buffer_ns": 1517625.05993,
              "command_buffers": 2
            },
            {
              "compute_ns": 970125,
              "compute_span_ns": 970125,
              "calibration_cpu_ns": 1663917,
              "calibration_gpu_ticks": 1663917,
              "compute_passes": 1,
              "command_buffer_ns": 970124.965534,
              "command_buffers": 2
            },
            {
              "compute_ns": 1591208,
              "compute_span_ns": 1591208,
              "calibration_cpu_ns": 2430958,
              "calibration_gpu_ticks": 2430958,
              "compute_passes": 1,
              "command_buffer_ns": 1591250.00937,
              "command_buffers": 2
            },
            {
              "compute_ns": 993917,
              "compute_span_ns": 993917,
              "calibration_cpu_ns": 1681292,
              "calibration_gpu_ticks": 1681292,
              "compute_passes": 1,
              "command_buffer_ns": 994041.678496,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 24541,
              "compute_span_ns": 24541,
              "calibration_cpu_ns": 426667,
              "calibration_gpu_ticks": 426667,
              "compute_passes": 1,
              "command_buffer_ns": 24541.6886173,
              "command_buffers": 2
            },
            {
              "compute_ns": 17750,
              "compute_span_ns": 17750,
              "calibration_cpu_ns": 421958,
              "calibration_gpu_ticks": 421958,
              "compute_passes": 1,
              "command_buffer_ns": 17874.9905899,
              "command_buffers": 2
            },
            {
              "compute_ns": 193750,
              "compute_span_ns": 193750,
              "calibration_cpu_ns": 1549125,
              "calibration_gpu_ticks": 1549125,
              "compute_passes": 1,
              "command_buffer_ns": 193875.050172,
              "command_buffers": 2
            },
            {
              "compute_ns": 21541,
              "compute_span_ns": 21541,
              "calibration_cpu_ns": 662375,
              "calibration_gpu_ticks": 662375,
              "compute_passes": 1,
              "command_buffer_ns": 21583.2842514,
              "command_buffers": 2
            },
            {
              "compute_ns": 60333,
              "compute_span_ns": 60333,
              "calibration_cpu_ns": 1305125,
              "calibration_gpu_ticks": 1305125,
              "compute_passes": 1,
              "command_buffer_ns": 60333.3464824,
              "command_buffers": 2
            },
            {
              "compute_ns": 22666,
              "compute_span_ns": 22666,
              "calibration_cpu_ns": 618750,
              "calibration_gpu_ticks": 618750,
              "compute_passes": 1,
              "command_buffer_ns": 22708.3219215,
              "command_buffers": 2
            },
            {
              "compute_ns": 19167,
              "compute_span_ns": 19167,
              "calibration_cpu_ns": 471500,
              "calibration_gpu_ticks": 471500,
              "compute_passes": 1,
              "command_buffer_ns": 19208.2952708,
              "command_buffers": 2
            },
            {
              "compute_ns": 267459,
              "compute_span_ns": 267459,
              "calibration_cpu_ns": 1046500,
              "calibration_gpu_ticks": 1046500,
              "compute_passes": 1,
              "command_buffer_ns": 267499.999609,
              "command_buffers": 2
            },
            {
              "compute_ns": 19125,
              "compute_span_ns": 19125,
              "calibration_cpu_ns": 582500,
              "calibration_gpu_ticks": 582500,
              "compute_passes": 1,
              "command_buffer_ns": 19208.3534785,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1476500.00872,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 949499.954004,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1667124.97594,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 952125.061303,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1711833.34896,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 978041.673079,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1452458.32391,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 936708.354857,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1614999.95738,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 60458.3183303,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 17374.9867827,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 16625.0392795,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 17166.6615643,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19083.3234228,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 17333.3100975,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19125.000108,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 22708.3219215,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 16458.3325386,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              23.07031263625,
              14.835936781312501,
              26.0488277490625,
              14.876954082859374,
              26.747396077500003,
              15.281901141859375,
              22.69466131109375,
              14.636068044640625,
              25.2343743340625
            ],
            "command_buffer_throughput_us_p50": 22.69466131109375,
            "command_buffer_throughput_us_p90": 26.188541414750002,
            "command_buffer_latency_us": [
              60.458318330299996,
              17.374986782700002,
              16.6250392795,
              17.1666615643,
              19.0833234228,
              17.3333100975,
              19.125000108000002,
              22.708321921499998,
              16.4583325386
            ],
            "command_buffer_latency_us_p50": 17.374986782700002,
            "command_buffer_latency_us_p90": 30.258321203260003
          },
          "compute_throughput_us": [
            15.23046875,
            24.496734375,
            15.734375,
            30.456375,
            14.977859375,
            23.7109375,
            15.158203125,
            24.862625,
            15.529953125
          ],
          "compute_throughput_us_p50": 15.734375,
          "compute_throughput_us_p90": 25.981375000000003,
          "command_buffer_throughput_us": [
            15.232421901593751,
            24.49804742359375,
            15.73502595421875,
            30.456380955,
            14.97851553721875,
            23.71289156140625,
            15.15820258646875,
            24.86328139640625,
            15.5319012265
          ],
          "command_buffer_throughput_us_p50": 15.73502595421875,
          "command_buffer_throughput_us_p90": 25.981901308125003,
          "compute_latency_us": [
            24.541,
            17.75,
            193.75,
            21.541,
            60.333,
            22.666,
            19.167,
            267.459,
            19.125
          ],
          "compute_latency_us_p50": 22.666,
          "compute_latency_us_p90": 208.4918,
          "command_buffer_latency_us": [
            24.5416886173,
            17.8749905899,
            193.875050172,
            21.583284251400002,
            60.333346482399996,
            22.708321921499998,
            19.2082952708,
            267.499999609,
            19.2083534785
          ],
          "command_buffer_latency_us_p50": 22.708321921499998,
          "command_buffer_latency_us_p90": 208.6000400594,
          "counter_control_throughput_ratio": 0.6679193127706655,
          "counter_control_latency_ratio": 1.2572790679513868
        },
        "correctness": {
          "max_abs_error": 1.6175184480715643e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 21.4260257903,
        "throughput_us_p90": 21.80046489184,
        "latency_us_p50": 269.458,
        "latency_us_p90": 316.0832
      },
      "slowdown": 0.8269703977312762,
      "device_compute_slowdown": 0.701966202453466,
      "valid": true,
      "round": 1,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "rmsnorm_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-ozwuei4e/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "c9fd11968a5edbf77e894438f42e4bd465ad146629716a6f08eacac9ab82063f",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 18.857875,
        "capture_ms": 0.068208,
        "compile_ms": 57.183875,
        "allocation_upload_ms": 7.983708,
        "cold_call_ms": 8.608084,
        "warmup_ms": 200.784875,
        "download_ms": 3.601875,
        "repetitions": 317,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          92.4931640379,
          98.6436624606,
          90.4108832808,
          98.9944794953,
          98.5059148265,
          90.7632776025,
          99.3600157729,
          99.5176119874,
          90.7359369085
        ],
        "latency_us": [
          2047.709,
          656.375,
          442.708,
          253.5,
          279.208,
          263.375,
          265.667,
          368.417,
          311.542
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 384,
            "reduction_scalar_rounds": 72,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 49152,
              "global_write_bytes": 24576,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 192,
              "global_write_bytes": 96,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 1,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 96,
            "normalized_kernel_cost": 1152,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4666458,
              "compute_span_ns": 4666458,
              "calibration_cpu_ns": 5326125,
              "calibration_gpu_ticks": 5326125,
              "compute_passes": 1,
              "command_buffer_ns": 4666500.02636,
              "command_buffers": 2
            },
            {
              "compute_ns": 4729667,
              "compute_span_ns": 4729667,
              "calibration_cpu_ns": 5870834,
              "calibration_gpu_ticks": 5870834,
              "compute_passes": 1,
              "command_buffer_ns": 4729791.66079,
              "command_buffers": 2
            },
            {
              "compute_ns": 4938000,
              "compute_span_ns": 4938000,
              "calibration_cpu_ns": 5913250,
              "calibration_gpu_ticks": 5913250,
              "compute_passes": 1,
              "command_buffer_ns": 4938041.6749,
              "command_buffers": 2
            },
            {
              "compute_ns": 4942458,
              "compute_span_ns": 4942458,
              "calibration_cpu_ns": 6002167,
              "calibration_gpu_ticks": 6002167,
              "compute_passes": 1,
              "command_buffer_ns": 4942458.29759,
              "command_buffers": 2
            },
            {
              "compute_ns": 4762083,
              "compute_span_ns": 4762083,
              "calibration_cpu_ns": 5610792,
              "calibration_gpu_ticks": 5610792,
              "compute_passes": 1,
              "command_buffer_ns": 4762124.96869,
              "command_buffers": 2
            },
            {
              "compute_ns": 4837000,
              "compute_span_ns": 4837000,
              "calibration_cpu_ns": 5737709,
              "calibration_gpu_ticks": 5737709,
              "compute_passes": 1,
              "command_buffer_ns": 4837083.28102,
              "command_buffers": 2
            },
            {
              "compute_ns": 4827292,
              "compute_span_ns": 4827292,
              "calibration_cpu_ns": 5827333,
              "calibration_gpu_ticks": 5827333,
              "compute_passes": 1,
              "command_buffer_ns": 4827333.32319,
              "command_buffers": 2
            },
            {
              "compute_ns": 6929292,
              "compute_span_ns": 6929292,
              "calibration_cpu_ns": 8340584,
              "calibration_gpu_ticks": 8340584,
              "compute_passes": 1,
              "command_buffer_ns": 6929333.32175,
              "command_buffers": 2
            },
            {
              "compute_ns": 6643167,
              "compute_span_ns": 6643167,
              "calibration_cpu_ns": 7673708,
              "calibration_gpu_ticks": 7673708,
              "compute_passes": 1,
              "command_buffer_ns": 6643208.32584,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 65417,
              "compute_span_ns": 65417,
              "calibration_cpu_ns": 487709,
              "calibration_gpu_ticks": 487709,
              "compute_passes": 1,
              "command_buffer_ns": 65416.6797176,
              "command_buffers": 2
            },
            {
              "compute_ns": 115833,
              "compute_span_ns": 115833,
              "calibration_cpu_ns": 2866459,
              "calibration_gpu_ticks": 2866459,
              "compute_passes": 1,
              "command_buffer_ns": 115833.361633,
              "command_buffers": 2
            },
            {
              "compute_ns": 117167,
              "compute_span_ns": 117167,
              "calibration_cpu_ns": 2167375,
              "calibration_gpu_ticks": 2167375,
              "compute_passes": 1,
              "command_buffer_ns": 117291.69637,
              "command_buffers": 2
            },
            {
              "compute_ns": 66875,
              "compute_span_ns": 66875,
              "calibration_cpu_ns": 473958,
              "calibration_gpu_ticks": 473958,
              "compute_passes": 1,
              "command_buffer_ns": 66916.6329317,
              "command_buffers": 2
            },
            {
              "compute_ns": 518917,
              "compute_span_ns": 518917,
              "calibration_cpu_ns": 1407125,
              "calibration_gpu_ticks": 1407125,
              "compute_passes": 1,
              "command_buffer_ns": 519041.670486,
              "command_buffers": 2
            },
            {
              "compute_ns": 115625,
              "compute_span_ns": 115625,
              "calibration_cpu_ns": 768917,
              "calibration_gpu_ticks": 768917,
              "compute_passes": 1,
              "command_buffer_ns": 115749.950055,
              "command_buffers": 2
            },
            {
              "compute_ns": 738458,
              "compute_span_ns": 738458,
              "calibration_cpu_ns": 2027875,
              "calibration_gpu_ticks": 2027875,
              "compute_passes": 1,
              "command_buffer_ns": 738541.653845,
              "command_buffers": 2
            },
            {
              "compute_ns": 99625,
              "compute_span_ns": 99625,
              "calibration_cpu_ns": 561791,
              "calibration_gpu_ticks": 561791,
              "compute_passes": 1,
              "command_buffer_ns": 99708.2679532,
              "command_buffers": 2
            },
            {
              "compute_ns": 317250,
              "compute_span_ns": 317250,
              "calibration_cpu_ns": 1477875,
              "calibration_gpu_ticks": 1477875,
              "compute_passes": 1,
              "command_buffer_ns": 317333.324347,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4585916.70031,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4830333.28783,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4987458.291,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4938999.94763,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4800874.97272,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4839750.00679,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4947083.2455,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4142875.02645,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4574208.28752,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 99916.6513793,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65958.3020024,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 66374.9524392,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 66958.3096169,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 69083.2966939,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 73458.3009034,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 106374.965981,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 66666.6310281,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 66083.332058,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              71.65494844234375,
              75.47395762234375,
              77.929035796875,
              77.17187418171875,
              75.01367144875,
              75.62109385609375,
              77.2981757109375,
              64.73242228828124,
              71.47200449249999
            ],
            "command_buffer_throughput_us_p50": 75.47395762234375,
            "command_buffer_throughput_us_p90": 77.424347728125,
            "command_buffer_latency_us": [
              99.9166513793,
              65.9583020024,
              66.3749524392,
              66.95830961690001,
              69.08329669390001,
              73.4583009034,
              106.374965981,
              66.6666310281,
              66.083332058
            ],
            "command_buffer_latency_us_p50": 66.95830961690001,
            "command_buffer_latency_us_p90": 101.20831429964001
          },
          "compute_throughput_us": [
            72.91340625,
            73.901046875,
            77.15625,
            77.22590625,
            74.407546875,
            75.578125,
            75.4264375,
            108.2701875,
            103.799484375
          ],
          "compute_throughput_us_p50": 75.578125,
          "compute_throughput_us_p90": 104.69362500000001,
          "command_buffer_throughput_us": [
            72.91406291187499,
            73.90299469984375,
            77.1569011703125,
            77.22591089984374,
            74.40820263578124,
            75.5794262659375,
            75.42708317484374,
            108.27083315234376,
            103.80013009125
          ],
          "command_buffer_throughput_us_p50": 75.5794262659375,
          "command_buffer_throughput_us_p90": 104.69427070346875,
          "compute_latency_us": [
            65.417,
            115.833,
            117.167,
            66.875,
            518.917,
            115.625,
            738.458,
            99.625,
            317.25
          ],
          "compute_latency_us_p50": 115.833,
          "compute_latency_us_p90": 562.8252000000001,
          "command_buffer_latency_us": [
            65.41667971759999,
            115.833361633,
            117.29169637000001,
            66.91663293170001,
            519.041670486,
            115.749950055,
            738.5416538449999,
            99.70826795320001,
            317.333324347
          ],
          "command_buffer_latency_us_p50": 115.833361633,
          "command_buffer_latency_us_p90": 562.9416671578001,
          "counter_control_throughput_ratio": 0.9994489951410178,
          "counter_control_latency_ratio": 1.7561604546579324
        },
        "correctness": {
          "max_abs_error": 1.4915225321487924e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 98.5059148265,
        "throughput_us_p90": 99.3915350158,
        "latency_us_p50": 311.542,
        "latency_us_p90": 934.6418000000003
      },
      "torch": {
        "cold_call_ms": 9.126416,
        "warmup_ms": 200.493958,
        "repetitions": 255,
        "throughput_us": [
          116.43415294117646,
          120.90963921568627,
          112.02075294117648,
          123.40506666666667,
          115.08398823529411,
          117.98071764705884,
          122.98610980392156,
          112.08022745098039,
          123.74101568627451
        ],
        "latency_us": [
          474.58299999999997,
          382.167,
          349.916,
          337.875,
          371.458,
          430.75,
          2027.584,
          972.0,
          408.58299999999997
        ],
        "throughput_us_p50": 117.98071764705884,
        "throughput_us_p90": 123.47225647058823,
        "latency_us_p50": 408.58299999999997,
        "latency_us_p90": 1183.1168000000002,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 7519749.982748181,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 7603125.006426126,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5579083.284828812,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5119458.364788443,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5227249.988820404,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5291374.982334673,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5328833.300154656,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5295291.71762988,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5205833.294894546,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 524375.0056251884,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 83999.99933317304,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 113833.34640413523,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 127583.33468809724,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 517125.00862777233,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80541.64936766028,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 519666.6461415589,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80124.99893084168,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 584624.9987371266,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              117.49609348044032,
              118.79882822540822,
              87.17317632545019,
              79.99153694981942,
              81.67578107531881,
              82.67773409897927,
              83.2630203149165,
              82.73893308796687,
              81.34114523272729
            ],
            "command_buffer_throughput_us_p50": 82.73893308796687,
            "command_buffer_throughput_us_p90": 117.7566404294339,
            "command_buffer_latency_us": [
              524.3750056251884,
              83.99999933317304,
              113.83334640413523,
              127.58333468809724,
              517.1250086277723,
              80.54164936766028,
              519.6666461415589,
              80.12499893084168,
              584.6249987371266
            ],
            "command_buffer_latency_us_p50": 127.58333468809724,
            "command_buffer_latency_us_p90": 536.425004247576
          },
          "throughput": [
            {
              "compute_ns": 7537584.0,
              "compute_span_ns": 7537584.0,
              "command_buffer_ns": 7537624.973338097,
              "calibration_cpu_ns": 8359917.0,
              "calibration_gpu_ticks": 8359917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 7575500.0,
              "compute_span_ns": 7575500.0,
              "command_buffer_ns": 7575499.99980256,
              "calibration_cpu_ns": 8500458.0,
              "calibration_gpu_ticks": 8500458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5255375.0,
              "compute_span_ns": 5255375.0,
              "command_buffer_ns": 5255374.999251217,
              "calibration_cpu_ns": 6602417.0,
              "calibration_gpu_ticks": 6602417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5339875.0,
              "compute_span_ns": 5339875.0,
              "command_buffer_ns": 5339874.944183975,
              "calibration_cpu_ns": 6776708.0,
              "calibration_gpu_ticks": 6776708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5550083.0,
              "compute_span_ns": 5550083.0,
              "command_buffer_ns": 5550041.678361595,
              "calibration_cpu_ns": 6862541.0,
              "calibration_gpu_ticks": 6862541.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5545791.0,
              "compute_span_ns": 5545791.0,
              "command_buffer_ns": 5545749.969314784,
              "calibration_cpu_ns": 7021333.0,
              "calibration_gpu_ticks": 7021333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5254792.0,
              "compute_span_ns": 5254792.0,
              "command_buffer_ns": 5254791.700281203,
              "calibration_cpu_ns": 6641792.0,
              "calibration_gpu_ticks": 6641792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5535083.0,
              "compute_span_ns": 5535083.0,
              "command_buffer_ns": 5535083.35724473,
              "calibration_cpu_ns": 7048500.0,
              "calibration_gpu_ticks": 7048500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5476667.0,
              "compute_span_ns": 5476667.0,
              "command_buffer_ns": 5476666.672620922,
              "calibration_cpu_ns": 6608000.0,
              "calibration_gpu_ticks": 6608000.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 107416.0,
              "compute_span_ns": 107416.0,
              "command_buffer_ns": 107375.03180280328,
              "calibration_cpu_ns": 585916.0,
              "calibration_gpu_ticks": 585916.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 81958.0,
              "compute_span_ns": 81958.0,
              "command_buffer_ns": 81916.63073375821,
              "calibration_cpu_ns": 696875.0,
              "calibration_gpu_ticks": 696875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80375.0,
              "compute_span_ns": 80375.0,
              "command_buffer_ns": 80375.00083446503,
              "calibration_cpu_ns": 608792.0,
              "calibration_gpu_ticks": 608792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80542.0,
              "compute_span_ns": 80542.0,
              "command_buffer_ns": 80541.64936766028,
              "calibration_cpu_ns": 529292.0,
              "calibration_gpu_ticks": 529292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 112625.0,
              "compute_span_ns": 112625.0,
              "command_buffer_ns": 112625.01357123256,
              "calibration_cpu_ns": 541083.0,
              "calibration_gpu_ticks": 541083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 96500.0,
              "compute_span_ns": 96500.0,
              "command_buffer_ns": 96500.03630667925,
              "calibration_cpu_ns": 657792.0,
              "calibration_gpu_ticks": 657792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 114791.0,
              "compute_span_ns": 114791.0,
              "command_buffer_ns": 114750.00064820051,
              "calibration_cpu_ns": 602000.0,
              "calibration_gpu_ticks": 602000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80541.0,
              "compute_span_ns": 80541.0,
              "command_buffer_ns": 80500.03089010715,
              "calibration_cpu_ns": 601250.0,
              "calibration_gpu_ticks": 601250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 114625.0,
              "compute_span_ns": 114625.0,
              "command_buffer_ns": 114625.0288002193,
              "calibration_cpu_ns": 644791.0,
              "calibration_gpu_ticks": 644791.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            117.77475,
            118.3671875,
            82.115234375,
            83.435546875,
            86.720046875,
            86.652984375,
            82.106125,
            86.485671875,
            85.572921875
          ],
          "compute_throughput_us_p50": 86.485671875,
          "compute_throughput_us_p90": 117.8932375,
          "command_buffer_throughput_us": [
            117.77539020840777,
            118.367187496915,
            82.11523436330026,
            83.43554600287462,
            86.71940122439992,
            86.6523432705435,
            82.1061203168938,
            86.4856774569489,
            85.57291675970191
          ],
          "command_buffer_throughput_us_p50": 86.4856774569489,
          "command_buffer_throughput_us_p90": 117.89374966610922,
          "compute_latency_us": [
            107.416,
            81.958,
            80.375,
            80.542,
            112.625,
            96.5,
            114.791,
            80.541,
            114.625
          ],
          "compute_latency_us_p50": 96.5,
          "compute_latency_us_p90": 114.6582,
          "command_buffer_latency_us": [
            107.37503180280328,
            81.91663073375821,
            80.37500083446503,
            80.54164936766028,
            112.62501357123256,
            96.50003630667925,
            114.75000064820051,
            80.50003089010715,
            114.6250288002193
          ],
          "command_buffer_latency_us_p50": 96.50003630667925,
          "command_buffer_latency_us_p90": 114.65002316981554,
          "counter_control_throughput_ratio": 1.04305466783587,
          "counter_control_latency_ratio": 0.6312865983990802
        },
        "download_ms": 0.951708,
        "allocation_upload_ms": 19.448041,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.9755493041451189e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.8349323244598493,
      "device_compute_slowdown": 0.8738803013432681,
      "valid": true,
      "round": 1,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "rmsnorm_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.502792,
        "warmup_ms": 200.274458,
        "repetitions": 265,
        "throughput_us": [
          120.74858490566038,
          120.44653962264151,
          111.2228,
          118.98412075471698,
          120.51509433962264,
          120.45487547169812,
          110.74402264150943,
          120.46981132075472,
          120.37625660377358
        ],
        "latency_us": [
          432.79200000000003,
          372.875,
          356.04200000000003,
          460.0,
          344.041,
          353.95799999999997,
          333.41700000000003,
          2358.833,
          1053.0
        ],
        "throughput_us_p50": 120.44653962264151,
        "throughput_us_p90": 120.56179245283019,
        "latency_us_p50": 372.875,
        "latency_us_p90": 1314.1666000000002,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 6302791.7058207095,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5706791.649572551,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5326458.311174065,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5464625.02097711,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5627541.686408222,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5723083.333577961,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5283000.005874783,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5510208.31335336,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5533249.9905489385,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 385250.02310052514,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80583.32605287433,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 576749.9678768218,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80208.29409360886,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 510250.0435896218,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 82541.66459664702,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80250.02898648381,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80333.32414925098,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 79916.61550477147,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              98.48112040344859,
              89.16861952457111,
              83.22591111209476,
              85.38476595276734,
              87.93033885012846,
              89.42317708715564,
              82.54687509179348,
              86.09700489614625,
              86.45703110232716
            ],
            "command_buffer_throughput_us_p50": 86.45703110232716,
            "command_buffer_throughput_us_p90": 91.23476575041423,
            "command_buffer_latency_us": [
              385.25002310052514,
              80.58332605287433,
              576.7499678768218,
              80.20829409360886,
              510.2500435896218,
              82.54166459664702,
              80.25002898648381,
              80.33332414925098,
              79.91661550477147
            ],
            "command_buffer_latency_us_p50": 80.58332605287433,
            "command_buffer_latency_us_p90": 523.5500284470618
          },
          "throughput": [
            {
              "compute_ns": 5129958.0,
              "compute_span_ns": 5129958.0,
              "command_buffer_ns": 5129916.651640087,
              "calibration_cpu_ns": 6531959.0,
              "calibration_gpu_ticks": 6531959.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5213167.0,
              "compute_span_ns": 5213167.0,
              "command_buffer_ns": 5213208.321947604,
              "calibration_cpu_ns": 6623167.0,
              "calibration_gpu_ticks": 6623167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5281000.0,
              "compute_span_ns": 5281000.0,
              "command_buffer_ns": 5280999.990645796,
              "calibration_cpu_ns": 6503125.0,
              "calibration_gpu_ticks": 6503125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5422917.0,
              "compute_span_ns": 5422917.0,
              "command_buffer_ns": 5422958.289273083,
              "calibration_cpu_ns": 6509833.0,
              "calibration_gpu_ticks": 6509833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5435459.0,
              "compute_span_ns": 5435459.0,
              "command_buffer_ns": 5435500.002931803,
              "calibration_cpu_ns": 7034250.0,
              "calibration_gpu_ticks": 7034250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5153250.0,
              "compute_span_ns": 5153250.0,
              "command_buffer_ns": 5153250.007424504,
              "calibration_cpu_ns": 6669500.0,
              "calibration_gpu_ticks": 6669500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5198750.0,
              "compute_span_ns": 5198750.0,
              "command_buffer_ns": 5198750.004637986,
              "calibration_cpu_ns": 6699250.0,
              "calibration_gpu_ticks": 6699250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5358375.0,
              "compute_span_ns": 5358375.0,
              "command_buffer_ns": 5358374.968636781,
              "calibration_cpu_ns": 6545292.0,
              "calibration_gpu_ticks": 6545292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5638250.0,
              "compute_span_ns": 5638250.0,
              "command_buffer_ns": 5638250.0333711505,
              "calibration_cpu_ns": 6600250.0,
              "calibration_gpu_ticks": 6600250.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 110125.0,
              "compute_span_ns": 110125.0,
              "command_buffer_ns": 110124.99453499913,
              "calibration_cpu_ns": 632417.0,
              "calibration_gpu_ticks": 632417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 81708.0,
              "compute_span_ns": 81708.0,
              "command_buffer_ns": 81708.30551534891,
              "calibration_cpu_ns": 654833.0,
              "calibration_gpu_ticks": 654833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 112167.0,
              "compute_span_ns": 112167.0,
              "command_buffer_ns": 112166.68644919991,
              "calibration_cpu_ns": 568541.0,
              "calibration_gpu_ticks": 568541.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 83250.0,
              "compute_span_ns": 83250.0,
              "command_buffer_ns": 83249.99362230301,
              "calibration_cpu_ns": 483542.0,
              "calibration_gpu_ticks": 483542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 108833.0,
              "compute_span_ns": 108833.0,
              "command_buffer_ns": 108833.30833166838,
              "calibration_cpu_ns": 573125.0,
              "calibration_gpu_ticks": 573125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80167.0,
              "compute_span_ns": 80167.0,
              "command_buffer_ns": 80166.67561605573,
              "calibration_cpu_ns": 496709.0,
              "calibration_gpu_ticks": 496709.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 516334.0,
              "compute_span_ns": 516334.0,
              "command_buffer_ns": 516375.0029169023,
              "calibration_cpu_ns": 1771666.0,
              "calibration_gpu_ticks": 1771666.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 112334.0,
              "compute_span_ns": 112334.0,
              "command_buffer_ns": 112375.01166760921,
              "calibration_cpu_ns": 556459.0,
              "calibration_gpu_ticks": 556459.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 577709.0,
              "compute_span_ns": 577709.0,
              "command_buffer_ns": 577749.9754913151,
              "calibration_cpu_ns": 1563958.0,
              "calibration_gpu_ticks": 1563958.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            80.15559375,
            81.455734375,
            82.515625,
            84.733078125,
            84.929046875,
            80.51953125,
            81.23046875,
            83.724609375,
            88.09765625
          ],
          "compute_throughput_us_p50": 82.515625,
          "compute_throughput_us_p90": 85.56276875,
          "command_buffer_throughput_us": [
            80.15494768187637,
            81.45638003043132,
            82.51562485384056,
            84.73372326989193,
            84.92968754580943,
            80.51953136600787,
            81.23046882246854,
            83.7246088849497,
            88.09765677142423
          ],
          "command_buffer_throughput_us_p50": 82.51562485384056,
          "command_buffer_throughput_us_p90": 85.56328139093239,
          "compute_latency_us": [
            110.125,
            81.708,
            112.167,
            83.25,
            108.833,
            80.167,
            516.334,
            112.334,
            577.709
          ],
          "compute_latency_us_p50": 110.125,
          "compute_latency_us_p90": 528.6089999999999,
          "command_buffer_latency_us": [
            110.12499453499913,
            81.70830551534891,
            112.16668644919991,
            83.24999362230301,
            108.83330833166838,
            80.16667561605573,
            516.3750029169023,
            112.37501166760921,
            577.7499754913151
          ],
          "command_buffer_latency_us_p50": 110.12499453499913,
          "command_buffer_latency_us_p90": 528.6499974317849,
          "counter_control_throughput_ratio": 0.9724450808241445,
          "counter_control_latency_ratio": 1.013960449606367
        },
        "download_ms": 0.777667,
        "allocation_upload_ms": 1.549875,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.9755493041451189e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-6vvbrr7s/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "232522da57db77c64b377ec2facdc2e1df33cd57064a4297b89d6582fd4ad63b",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 18.855291,
        "capture_ms": 0.064375,
        "compile_ms": 56.00125,
        "allocation_upload_ms": 6.3065,
        "cold_call_ms": 5.226417,
        "warmup_ms": 200.461542,
        "download_ms": 3.359208,
        "repetitions": 341,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          84.0399560117,
          95.175829912,
          91.6786422287,
          88.2385131965,
          86.7085777126,
          92.059627566,
          83.7453577713,
          92.4188651026,
          91.5876099707
        ],
        "latency_us": [
          387.875,
          319.334,
          283.833,
          267.708,
          343.875,
          338.458,
          297.25,
          276.208,
          278.125
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 32,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 768,
            "reduction_scalar_rounds": 72,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 49152,
              "global_write_bytes": 24576,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 192,
              "global_write_bytes": 96,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 1,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 104,
            "normalized_kernel_cost": 1248,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4289667,
              "compute_span_ns": 4289667,
              "calibration_cpu_ns": 5350250,
              "calibration_gpu_ticks": 5350250,
              "compute_passes": 1,
              "command_buffer_ns": 4289833.36315,
              "command_buffers": 2
            },
            {
              "compute_ns": 4626209,
              "compute_span_ns": 4626209,
              "calibration_cpu_ns": 5447084,
              "calibration_gpu_ticks": 5447084,
              "compute_passes": 1,
              "command_buffer_ns": 4626333.30608,
              "command_buffers": 2
            },
            {
              "compute_ns": 4564250,
              "compute_span_ns": 4564250,
              "calibration_cpu_ns": 5466250,
              "calibration_gpu_ticks": 5466250,
              "compute_passes": 1,
              "command_buffer_ns": 4564374.97633,
              "command_buffers": 2
            },
            {
              "compute_ns": 4732375,
              "compute_span_ns": 4732375,
              "calibration_cpu_ns": 5615041,
              "calibration_gpu_ticks": 5615041,
              "compute_passes": 1,
              "command_buffer_ns": 4732500.00505,
              "command_buffers": 2
            },
            {
              "compute_ns": 4409542,
              "compute_span_ns": 4409542,
              "calibration_cpu_ns": 5450875,
              "calibration_gpu_ticks": 5450875,
              "compute_passes": 1,
              "command_buffer_ns": 4409583.34366,
              "command_buffers": 2
            },
            {
              "compute_ns": 4518875,
              "compute_span_ns": 4518875,
              "calibration_cpu_ns": 5538875,
              "calibration_gpu_ticks": 5538875,
              "compute_passes": 1,
              "command_buffer_ns": 4518916.6558,
              "command_buffers": 2
            },
            {
              "compute_ns": 4454000,
              "compute_span_ns": 4454000,
              "calibration_cpu_ns": 5505792,
              "calibration_gpu_ticks": 5505792,
              "compute_passes": 1,
              "command_buffer_ns": 4454125.00994,
              "command_buffers": 2
            },
            {
              "compute_ns": 4477291,
              "compute_span_ns": 4477291,
              "calibration_cpu_ns": 5450625,
              "calibration_gpu_ticks": 5450625,
              "compute_passes": 1,
              "command_buffer_ns": 4477291.65899,
              "command_buffers": 2
            },
            {
              "compute_ns": 4337708,
              "compute_span_ns": 4337708,
              "calibration_cpu_ns": 5397791,
              "calibration_gpu_ticks": 5397791,
              "compute_passes": 1,
              "command_buffer_ns": 4337791.6445,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 108792,
              "compute_span_ns": 108792,
              "calibration_cpu_ns": 567750,
              "calibration_gpu_ticks": 567750,
              "compute_passes": 1,
              "command_buffer_ns": 108916.661702,
              "command_buffers": 2
            },
            {
              "compute_ns": 61833,
              "compute_span_ns": 61833,
              "calibration_cpu_ns": 551667,
              "calibration_gpu_ticks": 551667,
              "compute_passes": 1,
              "command_buffer_ns": 61875.0345893,
              "command_buffers": 2
            },
            {
              "compute_ns": 106834,
              "compute_span_ns": 106834,
              "calibration_cpu_ns": 697583,
              "calibration_gpu_ticks": 697583,
              "compute_passes": 1,
              "command_buffer_ns": 106916.646473,
              "command_buffers": 2
            },
            {
              "compute_ns": 61500,
              "compute_span_ns": 61500,
              "calibration_cpu_ns": 545250,
              "calibration_gpu_ticks": 545250,
              "compute_passes": 1,
              "command_buffer_ns": 61583.3560005,
              "command_buffers": 2
            },
            {
              "compute_ns": 112667,
              "compute_span_ns": 112667,
              "calibration_cpu_ns": 649417,
              "calibration_gpu_ticks": 649417,
              "compute_passes": 1,
              "command_buffer_ns": 112791.662104,
              "command_buffers": 2
            },
            {
              "compute_ns": 61208,
              "compute_span_ns": 61208,
              "calibration_cpu_ns": 501459,
              "calibration_gpu_ticks": 501459,
              "compute_passes": 1,
              "command_buffer_ns": 61291.6774116,
              "command_buffers": 2
            },
            {
              "compute_ns": 552834,
              "compute_span_ns": 552834,
              "calibration_cpu_ns": 2004459,
              "calibration_gpu_ticks": 2004459,
              "compute_passes": 1,
              "command_buffer_ns": 552999.961656,
              "command_buffers": 2
            },
            {
              "compute_ns": 107791,
              "compute_span_ns": 107791,
              "calibration_cpu_ns": 600583,
              "calibration_gpu_ticks": 600583,
              "compute_passes": 1,
              "command_buffer_ns": 107750.005554,
              "command_buffers": 2
            },
            {
              "compute_ns": 508208,
              "compute_span_ns": 508208,
              "calibration_cpu_ns": 2031208,
              "calibration_gpu_ticks": 2031208,
              "compute_passes": 1,
              "command_buffer_ns": 508249.970153,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4436958.34838,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4508000.04182,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4624958.32471,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4550958.34332,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4529166.61743,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4459125.04802,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4410749.99981,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4654750.05331,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4597916.67527,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 562375.003938,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 61500.0026301,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 475041.684695,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 70124.9809936,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 493416.679092,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 61083.2939856,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 61166.647356,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 61499.9444224,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 61083.3521932,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              69.32747419343751,
              70.4375006534375,
              72.26497382359375,
              71.108724114375,
              70.76822839734375,
              69.6738288753125,
              68.91796874703125,
              72.73046958296875,
              71.84244805109374
            ],
            "command_buffer_throughput_us_p50": 70.76822839734375,
            "command_buffer_throughput_us_p90": 72.35807297546874,
            "command_buffer_latency_us": [
              562.375003938,
              61.500002630100006,
              475.041684695,
              70.12498099359999,
              493.416679092,
              61.0832939856,
              61.166647356,
              61.499944422400006,
              61.0833521932
            ],
            "command_buffer_latency_us_p50": 61.500002630100006,
            "command_buffer_latency_us_p90": 507.20834406119997
          },
          "compute_throughput_us": [
            67.026046875,
            72.284515625,
            71.31640625,
            73.943359375,
            68.89909375,
            70.607421875,
            69.59375,
            69.957671875,
            67.7766875
          ],
          "compute_throughput_us_p50": 69.957671875,
          "compute_throughput_us_p90": 72.61628437499999,
          "command_buffer_throughput_us": [
            67.02864629921875,
            72.2864579075,
            71.31835900515625,
            73.94531257890625,
            68.8997397446875,
            70.608072746875,
            69.5957032803125,
            69.95768217171876,
            67.7779944453125
          ],
          "command_buffer_throughput_us_p50": 69.95768217171876,
          "command_buffer_throughput_us_p90": 72.61822884178125,
          "compute_latency_us": [
            108.792,
            61.833,
            106.834,
            61.5,
            112.667,
            61.208,
            552.834,
            107.791,
            508.208
          ],
          "compute_latency_us_p50": 107.791,
          "compute_latency_us_p90": 517.1332,
          "command_buffer_latency_us": [
            108.916661702,
            61.8750345893,
            106.916646473,
            61.583356000500004,
            112.791662104,
            61.2916774116,
            552.999961656,
            107.750005554,
            508.249970153
          ],
          "command_buffer_latency_us_p50": 107.750005554,
          "command_buffer_latency_us_p90": 517.1999684536,
          "counter_control_throughput_ratio": 0.9869007796121495,
          "counter_control_latency_ratio": 1.0034114634690317
        },
        "correctness": {
          "max_abs_error": 1.4915225321487924e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 91.5876099707,
        "throughput_us_p90": 92.97025806447999,
        "latency_us_p50": 297.25,
        "latency_us_p90": 352.675
      },
      "slowdown": 0.7604005084549842,
      "device_compute_slowdown": 0.8478112099981064,
      "valid": true,
      "round": 1,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "rmsnorm_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-7hux9_87/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "26111b6b600e52c50494ad3f783ef6fc3b02252d329d47ea39eb66749773d18e",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 18.894542,
        "capture_ms": 0.067166,
        "compile_ms": 55.425125,
        "allocation_upload_ms": 6.435166,
        "cold_call_ms": 6.343625,
        "warmup_ms": 200.021333,
        "download_ms": 2.405167,
        "repetitions": 398,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          72.5146582915,
          78.9267160804,
          79.3195150754,
          87.2384849246,
          79.3714396985,
          73.5617688442,
          76.4571809045,
          66.7130452261,
          74.4564497487
        ],
        "latency_us": [
          539.75,
          427.834,
          451.917,
          393.917,
          388.916,
          331.958,
          320,
          295.958,
          314.125
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 32,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 1024,
            "reduction_scalar_rounds": 48,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 32768,
              "global_write_bytes": 16384,
              "private_read_bytes": 49152,
              "private_write_bytes": 16384
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 128,
              "global_write_bytes": 64,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 1,
            "reduction_elements": 4096,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 80,
            "normalized_kernel_cost": 1280,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 8192,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3821958,
              "compute_span_ns": 3821958,
              "calibration_cpu_ns": 4774084,
              "calibration_gpu_ticks": 4774084,
              "compute_passes": 1,
              "command_buffer_ns": 3822041.67545,
              "command_buffers": 2
            },
            {
              "compute_ns": 4130208,
              "compute_span_ns": 4130208,
              "calibration_cpu_ns": 4995083,
              "calibration_gpu_ticks": 4995083,
              "compute_passes": 1,
              "command_buffer_ns": 4130166.66425,
              "command_buffers": 2
            },
            {
              "compute_ns": 3996083,
              "compute_span_ns": 3996083,
              "calibration_cpu_ns": 4925334,
              "calibration_gpu_ticks": 4925334,
              "compute_passes": 1,
              "command_buffer_ns": 3996083.28007,
              "command_buffers": 2
            },
            {
              "compute_ns": 4062375,
              "compute_span_ns": 4062375,
              "calibration_cpu_ns": 4973958,
              "calibration_gpu_ticks": 4973958,
              "compute_passes": 1,
              "command_buffer_ns": 4062374.99556,
              "command_buffers": 2
            },
            {
              "compute_ns": 4049834,
              "compute_span_ns": 4049834,
              "calibration_cpu_ns": 5098667,
              "calibration_gpu_ticks": 5098667,
              "compute_passes": 1,
              "command_buffer_ns": 4049916.63527,
              "command_buffers": 2
            },
            {
              "compute_ns": 3926750,
              "compute_span_ns": 3926750,
              "calibration_cpu_ns": 5135833,
              "calibration_gpu_ticks": 5135833,
              "compute_passes": 1,
              "command_buffer_ns": 3926749.98147,
              "command_buffers": 2
            },
            {
              "compute_ns": 4095875,
              "compute_span_ns": 4095875,
              "calibration_cpu_ns": 4927584,
              "calibration_gpu_ticks": 4927584,
              "compute_passes": 1,
              "command_buffer_ns": 4095875.01781,
              "command_buffers": 2
            },
            {
              "compute_ns": 3879708,
              "compute_span_ns": 3879708,
              "calibration_cpu_ns": 5217000,
              "calibration_gpu_ticks": 5217000,
              "compute_passes": 1,
              "command_buffer_ns": 3879791.59132,
              "command_buffers": 2
            },
            {
              "compute_ns": 3348917,
              "compute_span_ns": 3348917,
              "calibration_cpu_ns": 4188417,
              "calibration_gpu_ticks": 4188417,
              "compute_passes": 1,
              "command_buffer_ns": 3349041.62446,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 59750,
              "compute_span_ns": 59750,
              "calibration_cpu_ns": 562041,
              "calibration_gpu_ticks": 562041,
              "compute_passes": 1,
              "command_buffer_ns": 59749.9893047,
              "command_buffers": 2
            },
            {
              "compute_ns": 63542,
              "compute_span_ns": 63542,
              "calibration_cpu_ns": 512291,
              "calibration_gpu_ticks": 512291,
              "compute_passes": 1,
              "command_buffer_ns": 63624.9314994,
              "command_buffers": 2
            },
            {
              "compute_ns": 102916,
              "compute_span_ns": 102916,
              "calibration_cpu_ns": 607708,
              "calibration_gpu_ticks": 607708,
              "compute_passes": 1,
              "command_buffer_ns": 102916.674223,
              "command_buffers": 2
            },
            {
              "compute_ns": 60959,
              "compute_span_ns": 60959,
              "calibration_cpu_ns": 542708,
              "calibration_gpu_ticks": 542708,
              "compute_passes": 1,
              "command_buffer_ns": 61083.3521932,
              "command_buffers": 2
            },
            {
              "compute_ns": 54750,
              "compute_span_ns": 54750,
              "calibration_cpu_ns": 508792,
              "calibration_gpu_ticks": 508792,
              "compute_passes": 1,
              "command_buffer_ns": 54833.3046027,
              "command_buffers": 2
            },
            {
              "compute_ns": 413625,
              "compute_span_ns": 413625,
              "calibration_cpu_ns": 1548667,
              "calibration_gpu_ticks": 1548667,
              "compute_passes": 1,
              "command_buffer_ns": 413708.388805,
              "command_buffers": 2
            },
            {
              "compute_ns": 60542,
              "compute_span_ns": 60542,
              "calibration_cpu_ns": 741583,
              "calibration_gpu_ticks": 741583,
              "compute_passes": 1,
              "command_buffer_ns": 60583.2901783,
              "command_buffers": 2
            },
            {
              "compute_ns": 411750,
              "compute_span_ns": 411750,
              "calibration_cpu_ns": 1168750,
              "calibration_gpu_ticks": 1168750,
              "compute_passes": 1,
              "command_buffer_ns": 411875.02211,
              "command_buffers": 2
            },
            {
              "compute_ns": 53834,
              "compute_span_ns": 53834,
              "calibration_cpu_ns": 481708,
              "calibration_gpu_ticks": 481708,
              "compute_passes": 1,
              "command_buffer_ns": 54000.003729,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 3435375.0525,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4008999.96748,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4012499.99413,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4025208.35632,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4070583.38169,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4059916.6532,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4090500.00599,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6155958.28276,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3283083.32246,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 98208.3729468,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 101749.959867,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 69000.0015311,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 99249.9408312,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 66458.3058096,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 95875.0024438,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 53874.9736734,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 99541.6776277,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54000.003729,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              53.6777351953125,
              62.640624491875,
              62.695312408281254,
              62.8938805675,
              63.602865338906255,
              63.43619770625,
              63.91406259359375,
              96.186848168125,
              51.2981769134375
            ],
            "command_buffer_throughput_us_p50": 62.8938805675,
            "command_buffer_throughput_us_p90": 70.3686197085,
            "command_buffer_latency_us": [
              98.2083729468,
              101.749959867,
              69.0000015311,
              99.2499408312,
              66.4583058096,
              95.8750024438,
              53.87497367340001,
              99.5416776277,
              54.000003729
            ],
            "command_buffer_latency_us_p50": 95.8750024438,
            "command_buffer_latency_us_p90": 99.98333407556
          },
          "compute_throughput_us": [
            59.71809375,
            64.5345,
            62.438796875,
            63.474609375,
            63.27865625,
            61.35546875,
            63.998046875,
            60.6204375,
            52.326828125
          ],
          "compute_throughput_us_p50": 62.438796875,
          "compute_throughput_us_p90": 64.1053375,
          "command_buffer_throughput_us": [
            59.71940117890625,
            64.53385412890624,
            62.43880125109375,
            63.474609305625,
            63.27994742609375,
            61.35546846046875,
            63.99804715328125,
            60.621743614375,
            52.3287753821875
          ],
          "command_buffer_throughput_us_p50": 62.43880125109375,
          "command_buffer_throughput_us_p90": 64.10520854840625,
          "compute_latency_us": [
            59.75,
            63.542,
            102.916,
            60.959,
            54.75,
            413.625,
            60.542,
            411.75,
            53.834
          ],
          "compute_latency_us_p50": 60.959,
          "compute_latency_us_p90": 412.125,
          "command_buffer_latency_us": [
            59.7499893047,
            63.6249314994,
            102.916674223,
            61.0833521932,
            54.833304602700004,
            413.70838880499997,
            60.5832901783,
            411.87502211000003,
            54.000003729
          ],
          "command_buffer_latency_us_p50": 61.0833521932,
          "command_buffer_latency_us_p90": 412.241695449,
          "counter_control_throughput_ratio": 1.0013140231786162,
          "counter_control_latency_ratio": 1.0
        },
        "correctness": {
          "max_abs_error": 1.5832164268481108e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 76.4571809045,
        "throughput_us_p90": 80.94484874372,
        "latency_us_p50": 388.916,
        "latency_us_p90": 469.4836
      },
      "torch": {
        "cold_call_ms": 7.849125,
        "warmup_ms": 200.733625,
        "repetitions": 257,
        "throughput_us": [
          109.15840077821012,
          102.36624513618678,
          101.59192607003891,
          101.90061867704281,
          103.14834630350195,
          106.2057392996109,
          101.77836964980546,
          102.19049805447472,
          102.15515564202335
        ],
        "latency_us": [
          436.0,
          314.083,
          344.75,
          306.458,
          310.125,
          303.333,
          313.583,
          2170.292,
          1071.4170000000001
        ],
        "throughput_us_p50": 102.19049805447472,
        "throughput_us_p90": 106.79627159533074,
        "latency_us_p50": 314.083,
        "latency_us_p90": 1291.1920000000002,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4935791.715979576,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4690999.980084598,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5176291.684620082,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4916166.653856635,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4735500.027891248,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4833458.340726793,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4747500.002849847,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4882749.984972179,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4754166.700877249,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 421916.65852442384,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 71416.66719689965,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 416583.3233855665,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70750.01485645771,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 376874.9884329736,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 71541.69725254178,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 71416.66719689965,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 71124.98860806227,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70750.01485645771,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              77.12174556218088,
              73.29687468882184,
              80.87955757218879,
              76.81510396650992,
              73.99218793580076,
              75.52278657385614,
              74.17968754452886,
              76.2929685151903,
              74.28385470120702
            ],
            "command_buffer_throughput_us_p50": 75.52278657385614,
            "command_buffer_throughput_us_p90": 77.87330796418246,
            "command_buffer_latency_us": [
              421.91665852442384,
              71.41666719689965,
              416.5833233855665,
              70.75001485645771,
              376.8749884329736,
              71.54169725254178,
              71.41666719689965,
              71.12498860806227,
              70.75001485645771
            ],
            "command_buffer_latency_us_p50": 71.41666719689965,
            "command_buffer_latency_us_p90": 417.64999041333795
          },
          "throughput": [
            {
              "compute_ns": 4654084.0,
              "compute_span_ns": 4654084.0,
              "command_buffer_ns": 4654124.961234629,
              "calibration_cpu_ns": 5555250.0,
              "calibration_gpu_ticks": 5555250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4835458.0,
              "compute_span_ns": 4835458.0,
              "command_buffer_ns": 4835458.297748119,
              "calibration_cpu_ns": 5731042.0,
              "calibration_gpu_ticks": 5731042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4455291.0,
              "compute_span_ns": 4455291.0,
              "command_buffer_ns": 4455249.989405274,
              "calibration_cpu_ns": 6264666.0,
              "calibration_gpu_ticks": 6264666.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4688292.0,
              "compute_span_ns": 4688292.0,
              "command_buffer_ns": 4688333.312515169,
              "calibration_cpu_ns": 6102000.0,
              "calibration_gpu_ticks": 6102000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4518167.0,
              "compute_span_ns": 4518167.0,
              "command_buffer_ns": 4518208.326771855,
              "calibration_cpu_ns": 5845500.0,
              "calibration_gpu_ticks": 5845500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4736291.0,
              "compute_span_ns": 4736291.0,
              "command_buffer_ns": 4736250.0336021185,
              "calibration_cpu_ns": 5930875.0,
              "calibration_gpu_ticks": 5930875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4748625.0,
              "compute_span_ns": 4748625.0,
              "command_buffer_ns": 4748625.040519983,
              "calibration_cpu_ns": 5969083.0,
              "calibration_gpu_ticks": 5969083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4757333.0,
              "compute_span_ns": 4757333.0,
              "command_buffer_ns": 4757291.63736105,
              "calibration_cpu_ns": 5837542.0,
              "calibration_gpu_ticks": 5837542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4873708.0,
              "compute_span_ns": 4873708.0,
              "command_buffer_ns": 4873708.356171846,
              "calibration_cpu_ns": 5830291.0,
              "calibration_gpu_ticks": 5830291.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 97250.0,
              "compute_span_ns": 97250.0,
              "command_buffer_ns": 97249.98380988836,
              "calibration_cpu_ns": 619292.0,
              "calibration_gpu_ticks": 619292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70709.0,
              "compute_span_ns": 70709.0,
              "command_buffer_ns": 70749.9566487968,
              "calibration_cpu_ns": 564167.0,
              "calibration_gpu_ticks": 564167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 94875.0,
              "compute_span_ns": 94875.0,
              "command_buffer_ns": 94874.99482929707,
              "calibration_cpu_ns": 614042.0,
              "calibration_gpu_ticks": 614042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70792.0,
              "compute_span_ns": 70792.0,
              "command_buffer_ns": 70833.3682268858,
              "calibration_cpu_ns": 508167.0,
              "calibration_gpu_ticks": 508167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 92375.0,
              "compute_span_ns": 92375.0,
              "command_buffer_ns": 92375.03400072455,
              "calibration_cpu_ns": 541584.0,
              "calibration_gpu_ticks": 541584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70750.0,
              "compute_span_ns": 70750.0,
              "command_buffer_ns": 70750.01485645771,
              "calibration_cpu_ns": 595333.0,
              "calibration_gpu_ticks": 595333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 460625.0,
              "compute_span_ns": 460625.0,
              "command_buffer_ns": 460625.04407018423,
              "calibration_cpu_ns": 1674750.0,
              "calibration_gpu_ticks": 1674750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 91708.0,
              "compute_span_ns": 91708.0,
              "command_buffer_ns": 91708.3234526217,
              "calibration_cpu_ns": 536292.0,
              "calibration_gpu_ticks": 536292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 494708.0,
              "compute_span_ns": 494708.0,
              "command_buffer_ns": 494666.6886098683,
              "calibration_cpu_ns": 1638750.0,
              "calibration_gpu_ticks": 1638750.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            72.7200625,
            75.55403125,
            69.613921875,
            73.2545625,
            70.596359375,
            74.004546875,
            74.197265625,
            74.333328125,
            76.1516875
          ],
          "compute_throughput_us_p50": 74.004546875,
          "compute_throughput_us_p90": 75.67356249999999,
          "command_buffer_throughput_us": [
            72.72070251929108,
            75.55403590231435,
            69.61328108445741,
            73.25520800804952,
            70.59700510581024,
            74.0039067750331,
            74.19726625812473,
            74.3326818337664,
            76.1516930651851
          ],
          "command_buffer_throughput_us_p50": 74.0039067750331,
          "command_buffer_throughput_us_p90": 75.6735673348885,
          "compute_latency_us": [
            97.25,
            70.709,
            94.875,
            70.792,
            92.375,
            70.75,
            460.625,
            91.708,
            494.708
          ],
          "compute_latency_us_p50": 92.375,
          "compute_latency_us_p90": 467.4416,
          "command_buffer_latency_us": [
            97.24998380988836,
            70.7499566487968,
            94.87499482929707,
            70.8333682268858,
            92.37503400072455,
            70.75001485645771,
            460.62504407018423,
            91.7083234526217,
            494.6666886098683
          ],
          "command_buffer_latency_us_p50": 92.37503400072455,
          "command_buffer_latency_us_p90": 467.43337297812104,
          "counter_control_throughput_ratio": 0.9743058014444202,
          "counter_control_latency_ratio": 0.9906644964786063
        },
        "download_ms": 0.754666,
        "allocation_upload_ms": 17.526959,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.8339726848282112e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.7481828776658174,
      "device_compute_slowdown": 0.8437156838547294,
      "valid": true,
      "round": 1,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "rmsnorm_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.416,
        "warmup_ms": 200.107667,
        "repetitions": 288,
        "throughput_us": [
          106.42390277777777,
          107.72048611111111,
          99.8951111111111,
          111.92100694444444,
          111.39916319444445,
          115.40480208333335,
          111.81770833333333,
          117.55265972222222,
          90.18431597222222
        ],
        "latency_us": [
          405.791,
          343.625,
          566.667,
          336.5,
          367.375,
          4720.416,
          525.583,
          406.25,
          294.66700000000003
        ],
        "throughput_us_p50": 111.39916319444445,
        "throughput_us_p90": 115.83437361111112,
        "latency_us_p50": 405.791,
        "latency_us_p90": 1397.4168000000009,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4763583.303429186,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 6945624.947547913,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 6742750.003468245,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 6865874.980576336,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4476666.683331132,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4332249.9841451645,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4350958.333816379,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4633083.357475698,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4484666.686039418,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 414916.66343063116,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 431416.67265444994,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 104125.00705569983,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 295499.9799840152,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 211875.01261010766,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 344750.0057518482,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 302750.03518909216,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 478125.00270083547,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 586249.9820068479,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              74.43098911608104,
              108.52538980543613,
              105.35546880419133,
              107.27929657150526,
              69.94791692704894,
              67.6914060022682,
              67.98372396588093,
              72.39192746055778,
              70.0729169693659
            ],
            "command_buffer_throughput_us_p50": 72.39192746055778,
            "command_buffer_throughput_us_p90": 107.52851521829143,
            "command_buffer_latency_us": [
              414.91666343063116,
              431.41667265444994,
              104.12500705569983,
              295.4999799840152,
              211.87501261010766,
              344.7500057518482,
              302.75003518909216,
              478.12500270083547,
              586.2499820068479
            ],
            "command_buffer_latency_us_p50": 344.7500057518482,
            "command_buffer_latency_us_p90": 499.74999856203794
          },
          "throughput": [
            {
              "compute_ns": 4918000.0,
              "compute_span_ns": 4918000.0,
              "command_buffer_ns": 4918000.020552427,
              "calibration_cpu_ns": 6665750.0,
              "calibration_gpu_ticks": 6665750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5008917.0,
              "compute_span_ns": 5008917.0,
              "command_buffer_ns": 5008958.338294178,
              "calibration_cpu_ns": 7699208.0,
              "calibration_gpu_ticks": 7699208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4426416.0,
              "compute_span_ns": 4426416.0,
              "command_buffer_ns": 4426374.9732635915,
              "calibration_cpu_ns": 7657041.0,
              "calibration_gpu_ticks": 7657041.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 6985667.0,
              "compute_span_ns": 6985667.0,
              "command_buffer_ns": 6985666.695982218,
              "calibration_cpu_ns": 7959459.0,
              "calibration_gpu_ticks": 7959459.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4321083.0,
              "compute_span_ns": 4321083.0,
              "command_buffer_ns": 4321083.368267864,
              "calibration_cpu_ns": 5035416.0,
              "calibration_gpu_ticks": 5035416.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4634333.0,
              "compute_span_ns": 4634333.0,
              "command_buffer_ns": 4634291.690308601,
              "calibration_cpu_ns": 5595084.0,
              "calibration_gpu_ticks": 5595084.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4416958.0,
              "compute_span_ns": 4416958.0,
              "command_buffer_ns": 4416958.3125039935,
              "calibration_cpu_ns": 7574041.0,
              "calibration_gpu_ticks": 7574041.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4786917.0,
              "compute_span_ns": 4786917.0,
              "command_buffer_ns": 4786958.335898817,
              "calibration_cpu_ns": 7402583.0,
              "calibration_gpu_ticks": 7402583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4438167.0,
              "compute_span_ns": 4438167.0,
              "command_buffer_ns": 4438208.357896656,
              "calibration_cpu_ns": 7381667.0,
              "calibration_gpu_ticks": 7381667.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 279125.0,
              "compute_span_ns": 279125.0,
              "command_buffer_ns": 279125.0008158386,
              "calibration_cpu_ns": 2369000.0,
              "calibration_gpu_ticks": 2369000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 571709.0,
              "compute_span_ns": 571709.0,
              "command_buffer_ns": 571749.9880120158,
              "calibration_cpu_ns": 2570542.0,
              "calibration_gpu_ticks": 2570542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 297709.0,
              "compute_span_ns": 297709.0,
              "command_buffer_ns": 297749.9971166253,
              "calibration_cpu_ns": 2101667.0,
              "calibration_gpu_ticks": 2101667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 338209.0,
              "compute_span_ns": 338209.0,
              "command_buffer_ns": 338250.0144653022,
              "calibration_cpu_ns": 2456958.0,
              "calibration_gpu_ticks": 2456958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 300541.0,
              "compute_span_ns": 300541.0,
              "command_buffer_ns": 300499.95984882116,
              "calibration_cpu_ns": 2084208.0,
              "calibration_gpu_ticks": 2084208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 337209.0,
              "compute_span_ns": 337209.0,
              "command_buffer_ns": 337250.00685080886,
              "calibration_cpu_ns": 2296417.0,
              "calibration_gpu_ticks": 2296417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 402708.0,
              "compute_span_ns": 402708.0,
              "command_buffer_ns": 402708.36325362325,
              "calibration_cpu_ns": 2553583.0,
              "calibration_gpu_ticks": 2553583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 551083.0,
              "compute_span_ns": 551083.0,
              "command_buffer_ns": 551083.2997970283,
              "calibration_cpu_ns": 2506208.0,
              "calibration_gpu_ticks": 2506208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 113667.0,
              "compute_span_ns": 113667.0,
              "command_buffer_ns": 113708.3163484931,
              "calibration_cpu_ns": 2848791.0,
              "calibration_gpu_ticks": 2848791.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            76.84375,
            78.264328125,
            69.16275,
            109.151046875,
            67.516921875,
            72.411453125,
            69.01496875,
            74.795578125,
            69.346359375
          ],
          "compute_throughput_us_p50": 72.411453125,
          "compute_throughput_us_p90": 84.44167187500001,
          "command_buffer_throughput_us": [
            76.84375032113167,
            78.26497403584654,
            69.16210895724362,
            109.15104212472215,
            67.51692762918537,
            72.41080766107189,
            69.0149736328749,
            74.79622399841901,
            69.34700559213525
          ],
          "command_buffer_throughput_us_p50": 72.41080766107189,
          "command_buffer_throughput_us_p90": 84.44218765362166,
          "compute_latency_us": [
            279.125,
            571.709,
            297.709,
            338.209,
            300.541,
            337.209,
            402.708,
            551.083,
            113.667
          ],
          "compute_latency_us_p50": 337.209,
          "compute_latency_us_p90": 555.2081999999999,
          "command_buffer_latency_us": [
            279.1250008158386,
            571.7499880120158,
            297.7499971166253,
            338.25001446530223,
            300.49995984882116,
            337.25000685080886,
            402.70836325362325,
            551.0832997970283,
            113.7083163484931
          ],
          "command_buffer_latency_us_p50": 337.25000685080886,
          "command_buffer_latency_us_p90": 555.2166374400258,
          "counter_control_throughput_ratio": 1.015169067047747,
          "counter_control_latency_ratio": 1.1525925159405292
        },
        "download_ms": 0.881667,
        "allocation_upload_ms": 2.52975,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.8339726848282112e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-14fmthxw/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "310a51aaac34ebd04d17becc4645f77793138816428bb43603834b11ab1687fb",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 25.615417,
        "capture_ms": 0.061541,
        "compile_ms": 56.06375,
        "allocation_upload_ms": 6.61575,
        "cold_call_ms": 10.352541,
        "warmup_ms": 201.134625,
        "download_ms": 3.315541,
        "repetitions": 349,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          95.0686504298,
          94.27352149,
          94.8771489971,
          94.0703180516,
          94.5872722063,
          94.915713467,
          94.2836676218,
          94.5908567335,
          94.0756905444
        ],
        "latency_us": [
          416.375,
          264.042,
          323.666,
          381.042,
          352.333,
          2931.292,
          518.625,
          373.25,
          363.042
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 512,
            "reduction_scalar_rounds": 48,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 32768,
              "global_write_bytes": 16384,
              "private_read_bytes": 49152,
              "private_write_bytes": 16384
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 128,
              "global_write_bytes": 64,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 1,
            "reduction_elements": 4096,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 72,
            "normalized_kernel_cost": 1152,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 8192,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4434417,
              "compute_span_ns": 4434417,
              "calibration_cpu_ns": 5653750,
              "calibration_gpu_ticks": 5653750,
              "compute_passes": 1,
              "command_buffer_ns": 4434499.94782,
              "command_buffers": 2
            },
            {
              "compute_ns": 4522958,
              "compute_span_ns": 4522958,
              "calibration_cpu_ns": 5666666,
              "calibration_gpu_ticks": 5666666,
              "compute_passes": 1,
              "command_buffer_ns": 4523041.71631,
              "command_buffers": 2
            },
            {
              "compute_ns": 4552250,
              "compute_span_ns": 4552250,
              "calibration_cpu_ns": 5692500,
              "calibration_gpu_ticks": 5692500,
              "compute_passes": 1,
              "command_buffer_ns": 4552291.7062,
              "command_buffers": 2
            },
            {
              "compute_ns": 4568125,
              "compute_span_ns": 4568125,
              "calibration_cpu_ns": 5816250,
              "calibration_gpu_ticks": 5816250,
              "compute_passes": 1,
              "command_buffer_ns": 4568166.68157,
              "command_buffers": 2
            },
            {
              "compute_ns": 4485875,
              "compute_span_ns": 4485875,
              "calibration_cpu_ns": 5779375,
              "calibration_gpu_ticks": 5779375,
              "compute_passes": 1,
              "command_buffer_ns": 4485999.99072,
              "command_buffers": 2
            },
            {
              "compute_ns": 4588791,
              "compute_span_ns": 4588791,
              "calibration_cpu_ns": 5592250,
              "calibration_gpu_ticks": 5592250,
              "compute_passes": 1,
              "command_buffer_ns": 4588833.36978,
              "command_buffers": 2
            },
            {
              "compute_ns": 4492208,
              "compute_span_ns": 4492208,
              "calibration_cpu_ns": 5706458,
              "calibration_gpu_ticks": 5706458,
              "compute_passes": 1,
              "command_buffer_ns": 4492291.715,
              "command_buffers": 2
            },
            {
              "compute_ns": 4547959,
              "compute_span_ns": 4547959,
              "calibration_cpu_ns": 5812833,
              "calibration_gpu_ticks": 5812833,
              "compute_passes": 1,
              "command_buffer_ns": 4548083.29232,
              "command_buffers": 2
            },
            {
              "compute_ns": 4545166,
              "compute_span_ns": 4545166,
              "calibration_cpu_ns": 5697042,
              "calibration_gpu_ticks": 5697042,
              "compute_passes": 1,
              "command_buffer_ns": 4545208.35774,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 100333,
              "compute_span_ns": 100333,
              "calibration_cpu_ns": 752750,
              "calibration_gpu_ticks": 752750,
              "compute_passes": 1,
              "command_buffer_ns": 100416.596979,
              "command_buffers": 2
            },
            {
              "compute_ns": 64750,
              "compute_span_ns": 64750,
              "calibration_cpu_ns": 481833,
              "calibration_gpu_ticks": 481833,
              "compute_passes": 1,
              "command_buffer_ns": 64875.0574328,
              "command_buffers": 2
            },
            {
              "compute_ns": 65375,
              "compute_span_ns": 65375,
              "calibration_cpu_ns": 497542,
              "calibration_gpu_ticks": 497542,
              "compute_passes": 1,
              "command_buffer_ns": 65458.2981952,
              "command_buffers": 2
            },
            {
              "compute_ns": 253042,
              "compute_span_ns": 253042,
              "calibration_cpu_ns": 1619667,
              "calibration_gpu_ticks": 1619667,
              "compute_passes": 1,
              "command_buffer_ns": 253166.654147,
              "command_buffers": 2
            },
            {
              "compute_ns": 264750,
              "compute_span_ns": 264750,
              "calibration_cpu_ns": 1704834,
              "calibration_gpu_ticks": 1704834,
              "compute_passes": 1,
              "command_buffer_ns": 264791.655354,
              "command_buffers": 2
            },
            {
              "compute_ns": 101584,
              "compute_span_ns": 101584,
              "calibration_cpu_ns": 776625,
              "calibration_gpu_ticks": 776625,
              "compute_passes": 1,
              "command_buffer_ns": 101624.988019,
              "command_buffers": 2
            },
            {
              "compute_ns": 96459,
              "compute_span_ns": 96459,
              "calibration_cpu_ns": 542375,
              "calibration_gpu_ticks": 542375,
              "compute_passes": 1,
              "command_buffer_ns": 96583.3896771,
              "command_buffers": 2
            },
            {
              "compute_ns": 66209,
              "compute_span_ns": 66209,
              "calibration_cpu_ns": 539584,
              "calibration_gpu_ticks": 539584,
              "compute_passes": 1,
              "command_buffer_ns": 66333.275754,
              "command_buffers": 2
            },
            {
              "compute_ns": 65375,
              "compute_span_ns": 65375,
              "calibration_cpu_ns": 503791,
              "calibration_gpu_ticks": 503791,
              "compute_passes": 1,
              "command_buffer_ns": 65416.6797176,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4421333.31671,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4536374.99595,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4595625.03966,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4598666.62277,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4415458.35929,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4546083.3353,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4506625.00225,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4546999.98954,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4571541.67816,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 241208.355874,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 284874.986392,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 93791.6338444,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 102499.965578,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65500.033088,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65749.976784,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 319041.660987,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 349749.985617,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 98124.9613687,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              69.08333307359375,
              70.88085931171875,
              71.8066412446875,
              71.85416598078126,
              68.99153686390625,
              71.03255211406251,
              70.41601566015625,
              71.04687483656251,
              71.43033872125
            ],
            "command_buffer_throughput_us_p50": 71.03255211406251,
            "command_buffer_throughput_us_p90": 71.81614619190626,
            "command_buffer_latency_us": [
              241.208355874,
              284.874986392,
              93.7916338444,
              102.499965578,
              65.500033088,
              65.749976784,
              319.041660987,
              349.74998561700005,
              98.1249613687
            ],
            "command_buffer_latency_us_p50": 102.499965578,
            "command_buffer_latency_us_p90": 325.18332591300003
          },
          "compute_throughput_us": [
            69.287765625,
            70.67121875,
            71.12890625,
            71.376953125,
            70.091796875,
            71.699859375,
            70.19075,
            71.061859375,
            71.01821875
          ],
          "compute_throughput_us_p50": 71.01821875,
          "compute_throughput_us_p90": 71.441534375,
          "command_buffer_throughput_us": [
            69.28906168468751,
            70.67252681734375,
            71.129557909375,
            71.37760439953125,
            70.093749855,
            71.7005214028125,
            70.192058046875,
            71.0638014425,
            71.0188805896875
          ],
          "command_buffer_throughput_us_p50": 71.0188805896875,
          "command_buffer_throughput_us_p90": 71.4421878001875,
          "compute_latency_us": [
            100.333,
            64.75,
            65.375,
            253.042,
            264.75,
            101.584,
            96.459,
            66.209,
            65.375
          ],
          "compute_latency_us_p50": 96.459,
          "compute_latency_us_p90": 255.3836,
          "command_buffer_latency_us": [
            100.41659697899999,
            64.8750574328,
            65.4582981952,
            253.166654147,
            264.791655354,
            101.624988019,
            96.5833896771,
            66.333275754,
            65.41667971759999
          ],
          "command_buffer_latency_us_p50": 96.5833896771,
          "command_buffer_latency_us_p90": 255.4916543884,
          "counter_control_throughput_ratio": 0.9970608074394414,
          "counter_control_latency_ratio": 0.6666670621331493
        },
        "correctness": {
          "max_abs_error": 1.5832164268481108e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 94.5872722063,
        "throughput_us_p90": 94.94630085956,
        "latency_us_p50": 373.25,
        "latency_us_p90": 1001.1584000000005
      },
      "slowdown": 0.8490842255358801,
      "device_compute_slowdown": 0.9807594749880945,
      "valid": true,
      "round": 1,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "layernorm_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-90v38nhn/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "53243cd32bb4620fa91e934e3d500807c417fa70fc3ac1977dac07575a48be10",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 21.190917,
        "capture_ms": 0.079125,
        "compile_ms": 74.110209,
        "allocation_upload_ms": 2.097292,
        "cold_call_ms": 1.997291,
        "warmup_ms": 202.385542,
        "download_ms": 0.303584,
        "repetitions": 3938,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          6.93683341798,
          7.70551879126,
          7.06860512951,
          7.62419578466,
          6.83186262062,
          7.45464068055,
          6.82196952768,
          6.81271152869,
          7.50570289487
        ],
        "latency_us": [
          283.917,
          215.166,
          217.75,
          228.916,
          217.834,
          213.208,
          206.333,
          203.916,
          223.417
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 19,
            "reduction_scalar_rounds": 40,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 18444,
              "global_write_bytes": 6148,
              "private_read_bytes": 36888,
              "private_write_bytes": 12296
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 32,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 2,
            "reduction_elements": 3074,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 80,
            "normalized_kernel_cost": 80,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 4611,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 761834,
              "compute_span_ns": 761834,
              "calibration_cpu_ns": 2253875,
              "calibration_gpu_ticks": 2253875,
              "compute_passes": 1,
              "command_buffer_ns": 761874.951422,
              "command_buffers": 2
            },
            {
              "compute_ns": 295916,
              "compute_span_ns": 295916,
              "calibration_cpu_ns": 746667,
              "calibration_gpu_ticks": 746667,
              "compute_passes": 1,
              "command_buffer_ns": 295958.307106,
              "command_buffers": 2
            },
            {
              "compute_ns": 336125,
              "compute_span_ns": 336125,
              "calibration_cpu_ns": 776542,
              "calibration_gpu_ticks": 776542,
              "compute_passes": 1,
              "command_buffer_ns": 336166.704074,
              "command_buffers": 2
            },
            {
              "compute_ns": 319084,
              "compute_span_ns": 319084,
              "calibration_cpu_ns": 741459,
              "calibration_gpu_ticks": 741459,
              "compute_passes": 1,
              "command_buffer_ns": 319166.632835,
              "command_buffers": 2
            },
            {
              "compute_ns": 345875,
              "compute_span_ns": 345875,
              "calibration_cpu_ns": 821708,
              "calibration_gpu_ticks": 821708,
              "compute_passes": 1,
              "command_buffer_ns": 345958.280377,
              "command_buffers": 2
            },
            {
              "compute_ns": 355417,
              "compute_span_ns": 355417,
              "calibration_cpu_ns": 776750,
              "calibration_gpu_ticks": 776750,
              "compute_passes": 1,
              "command_buffer_ns": 355541.706085,
              "command_buffers": 2
            },
            {
              "compute_ns": 321542,
              "compute_span_ns": 321542,
              "calibration_cpu_ns": 826166,
              "calibration_gpu_ticks": 826166,
              "compute_passes": 1,
              "command_buffer_ns": 321625.033394,
              "command_buffers": 2
            },
            {
              "compute_ns": 798916,
              "compute_span_ns": 798916,
              "calibration_cpu_ns": 2305875,
              "calibration_gpu_ticks": 2305875,
              "compute_passes": 1,
              "command_buffer_ns": 798958.353698,
              "command_buffers": 2
            },
            {
              "compute_ns": 1027041,
              "compute_span_ns": 1027041,
              "calibration_cpu_ns": 2223000,
              "calibration_gpu_ticks": 2223000,
              "compute_passes": 1,
              "command_buffer_ns": 1027083.37363,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 7209,
              "compute_span_ns": 7209,
              "calibration_cpu_ns": 416000,
              "calibration_gpu_ticks": 416000,
              "compute_passes": 1,
              "command_buffer_ns": 7291.67368263,
              "command_buffers": 2
            },
            {
              "compute_ns": 7709,
              "compute_span_ns": 7709,
              "calibration_cpu_ns": 441542,
              "calibration_gpu_ticks": 441542,
              "compute_passes": 1,
              "command_buffer_ns": 7791.67748988,
              "command_buffers": 2
            },
            {
              "compute_ns": 7084,
              "compute_span_ns": 7084,
              "calibration_cpu_ns": 378375,
              "calibration_gpu_ticks": 378375,
              "compute_passes": 1,
              "command_buffer_ns": 7208.3203122,
              "command_buffers": 2
            },
            {
              "compute_ns": 7042,
              "compute_span_ns": 7042,
              "calibration_cpu_ns": 356958,
              "calibration_gpu_ticks": 356958,
              "compute_passes": 1,
              "command_buffer_ns": 7083.34846422,
              "command_buffers": 2
            },
            {
              "compute_ns": 31125,
              "compute_span_ns": 31125,
              "calibration_cpu_ns": 1176458,
              "calibration_gpu_ticks": 1176458,
              "compute_passes": 1,
              "command_buffer_ns": 31125.0332743,
              "command_buffers": 2
            },
            {
              "compute_ns": 8917,
              "compute_span_ns": 8917,
              "calibration_cpu_ns": 636250,
              "calibration_gpu_ticks": 636250,
              "compute_passes": 1,
              "command_buffer_ns": 9083.30548555,
              "command_buffers": 2
            },
            {
              "compute_ns": 7709,
              "compute_span_ns": 7709,
              "calibration_cpu_ns": 457875,
              "calibration_gpu_ticks": 457875,
              "compute_passes": 1,
              "command_buffer_ns": 7750.00080466,
              "command_buffers": 2
            },
            {
              "compute_ns": 7792,
              "compute_span_ns": 7792,
              "calibration_cpu_ns": 372791,
              "calibration_gpu_ticks": 372791,
              "compute_passes": 1,
              "command_buffer_ns": 7833.29596743,
              "command_buffers": 2
            },
            {
              "compute_ns": 7375,
              "compute_span_ns": 7375,
              "calibration_cpu_ns": 416708,
              "calibration_gpu_ticks": 416708,
              "compute_passes": 1,
              "command_buffer_ns": 7416.64553061,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 494708.365295,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 628375.040833,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 998875.009827,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 313125.026878,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1032958.33105,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 843624.99183,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 308624.992613,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 330333.30692,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 357666.693162,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 7333.35036784,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 61708.2696408,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 9124.98217076,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 11749.973055,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7499.94069338,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7124.96694177,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7416.64553061,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 116666.662507,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 9291.68891162,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              7.729818207734375,
              9.818360013015624,
              15.607422028546875,
              4.89257854496875,
              16.13997392265625,
              13.18164049734375,
              4.822265509578124,
              5.161457920625,
              5.58854208065625
            ],
            "command_buffer_throughput_us_p50": 7.729818207734375,
            "command_buffer_throughput_us_p90": 15.71393240736875,
            "command_buffer_latency_us": [
              7.3333503678400005,
              61.7082696408,
              9.124982170760001,
              11.749973055,
              7.49994069338,
              7.1249669417699995,
              7.416645530609999,
              116.666662507,
              9.29168891162
            ],
            "command_buffer_latency_us_p50": 9.124982170760001,
            "command_buffer_latency_us_p90": 72.69994821404
          },
          "compute_throughput_us": [
            11.90365625,
            4.6236875,
            5.251953125,
            4.9856875,
            5.404296875,
            5.553390625,
            5.02409375,
            12.4830625,
            16.047515625
          ],
          "compute_throughput_us_p50": 5.404296875,
          "compute_throughput_us_p90": 13.195953125,
          "command_buffer_throughput_us": [
            11.90429611596875,
            4.62434854853125,
            5.25260475115625,
            4.986978638046875,
            5.405598130890625,
            5.555339157578125,
            5.02539114678125,
            12.48372427653125,
            16.04817771296875
          ],
          "command_buffer_throughput_us_p50": 5.405598130890625,
          "command_buffer_throughput_us_p90": 13.196614963818751,
          "compute_latency_us": [
            7.209,
            7.709,
            7.084,
            7.042,
            31.125,
            8.917,
            7.709,
            7.792,
            7.375
          ],
          "compute_latency_us_p50": 7.709,
          "compute_latency_us_p90": 13.358600000000003,
          "command_buffer_latency_us": [
            7.29167368263,
            7.7916774898800005,
            7.208320312200001,
            7.08334846422,
            31.1250332743,
            9.08330548555,
            7.75000080466,
            7.83329596743,
            7.416645530609999
          ],
          "command_buffer_latency_us_p50": 7.75000080466,
          "command_buffer_latency_us_p90": 13.491651043300003,
          "counter_control_throughput_ratio": 1.019294548306271,
          "counter_control_latency_ratio": 0.7982020923381207
        },
        "correctness": {
          "max_abs_error": 2.1991937471099732e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 7.06860512951,
        "throughput_us_p90": 7.64046038598,
        "latency_us_p50": 217.75,
        "latency_us_p90": 239.9162
      },
      "torch": {
        "cold_call_ms": 9.236125,
        "warmup_ms": 200.471042,
        "repetitions": 1753,
        "throughput_us": [
          15.378969195664576,
          16.9092983456931,
          14.536413576725614,
          14.411936679977181,
          14.975066172276097,
          14.31600570450656,
          14.345003993154593,
          14.880205362236167,
          14.01851568739304
        ],
        "latency_us": [
          303.66700000000003,
          300.333,
          227.625,
          227.5,
          240.542,
          224.291,
          242.209,
          2148.459,
          619.958
        ],
        "throughput_us_p50": 14.536413576725614,
        "throughput_us_p90": 15.685035025670281,
        "latency_us_p50": 242.209,
        "latency_us_p90": 925.6582000000002,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 526499.9927021563,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 535124.9710656703,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 484624.99398738146,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 505083.3569839597,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 832375.0225827098,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 851958.3498127759,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 520250.00331923366,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 485499.9715462327,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 727708.3350345492,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 10250.01984089613,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10041.636414825916,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10583.316907286644,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 14166.638720780611,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 9874.98788163066,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 198708.3232961595,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 11374.999303370714,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10374.991688877344,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 195708.3586603403,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              8.226562385971192,
              8.361327672901098,
              7.572265531052835,
              7.89192745287437,
              13.00585972785484,
              13.311849215824623,
              8.128906301863026,
              7.585937055409886,
              11.370442734914832
            ],
            "command_buffer_throughput_us_p50": 8.226562385971192,
            "command_buffer_throughput_us_p90": 13.067057625448797,
            "command_buffer_latency_us": [
              10.25001984089613,
              10.041636414825916,
              10.583316907286644,
              14.166638720780611,
              9.874987881630659,
              198.7083232961595,
              11.374999303370714,
              10.374991688877344,
              195.7083586603403
            ],
            "command_buffer_latency_us_p50": 10.583316907286644,
            "command_buffer_latency_us_p90": 196.30835158750415
          },
          "throughput": [
            {
              "compute_ns": 506000.0,
              "compute_span_ns": 506000.0,
              "command_buffer_ns": 505999.95302036405,
              "calibration_cpu_ns": 1209833.0,
              "calibration_gpu_ticks": 1209833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 938916.0,
              "compute_span_ns": 938916.0,
              "command_buffer_ns": 938875.0186190009,
              "calibration_cpu_ns": 1918000.0,
              "calibration_gpu_ticks": 1918000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 941166.0,
              "compute_span_ns": 941166.0,
              "command_buffer_ns": 941124.9775439501,
              "calibration_cpu_ns": 2031167.0,
              "calibration_gpu_ticks": 2031167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 543209.0,
              "compute_span_ns": 543209.0,
              "command_buffer_ns": 543250.0038295984,
              "calibration_cpu_ns": 1286917.0,
              "calibration_gpu_ticks": 1286917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 505000.0,
              "compute_span_ns": 505000.0,
              "command_buffer_ns": 505000.0036135316,
              "calibration_cpu_ns": 1263042.0,
              "calibration_gpu_ticks": 1263042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 486708.0,
              "compute_span_ns": 486708.0,
              "command_buffer_ns": 486666.68590158224,
              "calibration_cpu_ns": 1179750.0,
              "calibration_gpu_ticks": 1179750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 820583.0,
              "compute_span_ns": 820583.0,
              "command_buffer_ns": 820541.6379496455,
              "calibration_cpu_ns": 2307125.0,
              "calibration_gpu_ticks": 2307125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 541291.0,
              "compute_span_ns": 541291.0,
              "command_buffer_ns": 541249.9886006117,
              "calibration_cpu_ns": 1226291.0,
              "calibration_gpu_ticks": 1226291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 527500.0,
              "compute_span_ns": 527500.0,
              "command_buffer_ns": 527500.0003166497,
              "calibration_cpu_ns": 1325416.0,
              "calibration_gpu_ticks": 1325416.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 202875.0,
              "compute_span_ns": 202875.0,
              "command_buffer_ns": 202875.00228732824,
              "calibration_cpu_ns": 1379375.0,
              "calibration_gpu_ticks": 1379375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 11125.0,
              "compute_span_ns": 11125.0,
              "command_buffer_ns": 11124.997399747372,
              "calibration_cpu_ns": 687875.0,
              "calibration_gpu_ticks": 687875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10459.0,
              "compute_span_ns": 10459.0,
              "command_buffer_ns": 10499.963536858559,
              "calibration_cpu_ns": 438375.0,
              "calibration_gpu_ticks": 438375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 189125.0,
              "compute_span_ns": 189125.0,
              "command_buffer_ns": 189125.01400336623,
              "calibration_cpu_ns": 1275042.0,
              "calibration_gpu_ticks": 1275042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10625.0,
              "compute_span_ns": 10625.0,
              "command_buffer_ns": 10624.993592500687,
              "calibration_cpu_ns": 437917.0,
              "calibration_gpu_ticks": 437917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10458.0,
              "compute_span_ns": 10458.0,
              "command_buffer_ns": 10416.668374091387,
              "calibration_cpu_ns": 479625.0,
              "calibration_gpu_ticks": 479625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10042.0,
              "compute_span_ns": 10042.0,
              "command_buffer_ns": 10041.69462248683,
              "calibration_cpu_ns": 452208.0,
              "calibration_gpu_ticks": 452208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 14583.0,
              "compute_span_ns": 14583.0,
              "command_buffer_ns": 14541.670680046082,
              "calibration_cpu_ns": 498250.0,
              "calibration_gpu_ticks": 498250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 11125.0,
              "compute_span_ns": 11125.0,
              "command_buffer_ns": 11124.997399747372,
              "calibration_cpu_ns": 649416.0,
              "calibration_gpu_ticks": 649416.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            7.90625,
            14.6705625,
            14.70571875,
            8.487640625,
            7.890625,
            7.6048125,
            12.821609375,
            8.457671875,
            8.2421875
          ],
          "compute_throughput_us_p50": 8.457671875,
          "compute_throughput_us_p90": 14.677593750000002,
          "command_buffer_throughput_us": [
            7.906249265943188,
            14.66992216592189,
            14.70507777412422,
            8.488281309837475,
            7.890625056461431,
            7.6041669672122225,
            12.820963092963211,
            8.457031071884558,
            8.242187504947651
          ],
          "command_buffer_throughput_us_p50": 8.457031071884558,
          "command_buffer_throughput_us_p90": 14.676953287562355,
          "compute_latency_us": [
            202.875,
            11.125,
            10.459,
            189.125,
            10.625,
            10.458,
            10.042,
            14.583,
            11.125
          ],
          "compute_latency_us_p50": 11.125,
          "compute_latency_us_p90": 191.875,
          "command_buffer_latency_us": [
            202.87500228732824,
            11.124997399747372,
            10.499963536858559,
            189.12501400336623,
            10.624993592500687,
            10.416668374091387,
            10.04169462248683,
            14.541670680046082,
            11.124997399747372
          ],
          "command_buffer_latency_us_p50": 11.124997399747372,
          "command_buffer_latency_us_p90": 191.87501166015863,
          "counter_control_throughput_ratio": 1.0755650454878298,
          "counter_control_latency_ratio": 1.075950038608673
        },
        "download_ms": 0.315042,
        "allocation_upload_ms": 0.851292,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.0790672678193687e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.48626885113035095,
      "device_compute_slowdown": 0.6389816198680561,
      "valid": true,
      "round": 1,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "layernorm_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.265542,
        "warmup_ms": 200.2815,
        "repetitions": 1383,
        "throughput_us": [
          14.016569775849604,
          16.589630513376715,
          16.35635068691251,
          14.916003615328995,
          14.813388286334057,
          15.167088937093276,
          14.996565437454809,
          15.210292118582792,
          15.666907447577728
        ],
        "latency_us": [
          318.5,
          267.666,
          446.83299999999997,
          302.08299999999997,
          239.958,
          220.208,
          231.333,
          232.917,
          166.042
        ],
        "throughput_us_p50": 15.167088937093276,
        "throughput_us_p90": 16.403006652205352,
        "latency_us_p50": 239.958,
        "latency_us_p90": 344.1666,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 763749.99480322,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 771666.644141078,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 570749.9803975224,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 517499.9823793769,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 491166.6619591415,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 485291.7045354843,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 931374.9615103006,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 854958.372656256,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 560166.6634902358,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 249041.65184125304,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 9833.311196416616,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10166.666470468044,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10791.700333356857,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 9874.98788163066,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 216624.99057129025,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 11291.704140603542,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10374.991688877344,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 221125.02483651042,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              11.933593668800313,
              12.057291314704344,
              8.917968443711288,
              8.085937224677764,
              7.674479093111586,
              7.582682883366942,
              14.552733773598447,
              13.358724572754,
              8.752604117034934
            ],
            "command_buffer_throughput_us_p50": 8.917968443711288,
            "command_buffer_throughput_us_p90": 13.597526412922889,
            "command_buffer_latency_us": [
              249.04165184125304,
              9.833311196416616,
              10.166666470468044,
              10.791700333356857,
              9.874987881630659,
              216.62499057129025,
              11.291704140603542,
              10.374991688877344,
              221.12502483651042
            ],
            "command_buffer_latency_us_p50": 10.791700333356857,
            "command_buffer_latency_us_p90": 226.70835023745894
          },
          "throughput": [
            {
              "compute_ns": 562250.0,
              "compute_span_ns": 562250.0,
              "command_buffer_ns": 562249.9738819897,
              "calibration_cpu_ns": 1296000.0,
              "calibration_gpu_ticks": 1296000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 483667.0,
              "compute_span_ns": 483667.0,
              "command_buffer_ns": 483666.66305810213,
              "calibration_cpu_ns": 1169500.0,
              "calibration_gpu_ticks": 1169500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 482959.0,
              "compute_span_ns": 482959.0,
              "command_buffer_ns": 483000.0107176602,
              "calibration_cpu_ns": 1169125.0,
              "calibration_gpu_ticks": 1169125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 865667.0,
              "compute_span_ns": 865667.0,
              "command_buffer_ns": 865708.3380967379,
              "calibration_cpu_ns": 2168792.0,
              "calibration_gpu_ticks": 2168792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 834916.0,
              "compute_span_ns": 834916.0,
              "command_buffer_ns": 834875.0416189432,
              "calibration_cpu_ns": 2127833.0,
              "calibration_gpu_ticks": 2127833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 557666.0,
              "compute_span_ns": 557666.0,
              "command_buffer_ns": 557625.0259764493,
              "calibration_cpu_ns": 1324541.0,
              "calibration_gpu_ticks": 1324541.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 548125.0,
              "compute_span_ns": 548125.0,
              "command_buffer_ns": 548125.0118464231,
              "calibration_cpu_ns": 1290500.0,
              "calibration_gpu_ticks": 1290500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 483458.0,
              "compute_span_ns": 483458.0,
              "command_buffer_ns": 483416.6611544788,
              "calibration_cpu_ns": 1158000.0,
              "calibration_gpu_ticks": 1158000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 483125.0,
              "compute_span_ns": 483125.0,
              "command_buffer_ns": 483124.9825656414,
              "calibration_cpu_ns": 1222417.0,
              "calibration_gpu_ticks": 1222417.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 11250.0,
              "compute_span_ns": 11250.0,
              "command_buffer_ns": 11250.0274553895,
              "calibration_cpu_ns": 506292.0,
              "calibration_gpu_ticks": 506292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 9792.0,
              "compute_span_ns": 9792.0,
              "command_buffer_ns": 9833.36940407753,
              "calibration_cpu_ns": 388458.0,
              "calibration_gpu_ticks": 388458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10125.0,
              "compute_span_ns": 10125.0,
              "command_buffer_ns": 10124.989785254002,
              "calibration_cpu_ns": 391625.0,
              "calibration_gpu_ticks": 391625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 169750.0,
              "compute_span_ns": 169750.0,
              "command_buffer_ns": 169750.01199170947,
              "calibration_cpu_ns": 1524250.0,
              "calibration_gpu_ticks": 1524250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10375.0,
              "compute_span_ns": 10375.0,
              "command_buffer_ns": 10375.049896538258,
              "calibration_cpu_ns": 414667.0,
              "calibration_gpu_ticks": 414667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10375.0,
              "compute_span_ns": 10375.0,
              "command_buffer_ns": 10374.991688877344,
              "calibration_cpu_ns": 412416.0,
              "calibration_gpu_ticks": 412416.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10042.0,
              "compute_span_ns": 10042.0,
              "command_buffer_ns": 10083.371307700872,
              "calibration_cpu_ns": 416125.0,
              "calibration_gpu_ticks": 416125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10167.0,
              "compute_span_ns": 10167.0,
              "command_buffer_ns": 10208.343155682087,
              "calibration_cpu_ns": 500667.0,
              "calibration_gpu_ticks": 500667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 11792.0,
              "compute_span_ns": 11792.0,
              "command_buffer_ns": 11833.326425403357,
              "calibration_cpu_ns": 687500.0,
              "calibration_gpu_ticks": 687500.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            8.78515625,
            7.557296875,
            7.546234375,
            13.526046875,
            13.0455625,
            8.71353125,
            8.564453125,
            7.55403125,
            7.548828125
          ],
          "compute_throughput_us_p50": 8.564453125,
          "compute_throughput_us_p90": 13.141659375000001,
          "command_buffer_throughput_us": [
            8.78515584190609,
            7.557291610282846,
            7.5468751674634404,
            13.526692782761529,
            13.044922525295988,
            8.71289103088202,
            8.564453310100362,
            7.553385330538731,
            7.548827852588147
          ],
          "command_buffer_throughput_us_p50": 8.564453310100362,
          "command_buffer_throughput_us_p90": 13.141276576789096,
          "compute_latency_us": [
            11.25,
            9.792,
            10.125,
            169.75,
            10.375,
            10.375,
            10.042,
            10.167,
            11.792
          ],
          "compute_latency_us_p50": 10.375,
          "compute_latency_us_p90": 43.38360000000003,
          "command_buffer_latency_us": [
            11.2500274553895,
            9.83336940407753,
            10.124989785254002,
            169.75001199170947,
            10.375049896538258,
            10.374991688877344,
            10.083371307700872,
            10.208343155682087,
            11.833326425403357
          ],
          "command_buffer_latency_us_p50": 10.374991688877344,
          "command_buffer_latency_us_p90": 43.41666353866461,
          "counter_control_throughput_ratio": 0.846254975569609,
          "counter_control_latency_ratio": 0.983937477909123
        },
        "download_ms": 0.3705,
        "allocation_upload_ms": 0.673083,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.0790672678193687e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-3w2f7p_7/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "90a0b6dbe0fbfeffc09b3a8db9fc5856f9e2631dc2ce0e6d9ee991918165a6cf",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.028625,
        "capture_ms": 0.089292,
        "compile_ms": 65.243416,
        "allocation_upload_ms": 2.088958,
        "cold_call_ms": 2.124916,
        "warmup_ms": 200.136333,
        "download_ms": 0.269958,
        "repetitions": 3229,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          6.0525447507,
          6.80435119232,
          6.01858160421,
          7.19554567978,
          6.09648250232,
          6.8158872716,
          6.09343728709,
          6.8267138433,
          6.06743573862
        ],
        "latency_us": [
          1870.333,
          840.583,
          274.125,
          244.875,
          204.708,
          191.167,
          178.583,
          185.458,
          188.084
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 37,
            "reduction_scalar_rounds": 40,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 18444,
              "global_write_bytes": 6148,
              "private_read_bytes": 36888,
              "private_write_bytes": 12296
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 32,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 2,
            "reduction_elements": 3074,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 88,
            "normalized_kernel_cost": 88,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 4611,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 293666,
              "compute_span_ns": 293666,
              "calibration_cpu_ns": 775416,
              "calibration_gpu_ticks": 775416,
              "compute_passes": 1,
              "command_buffer_ns": 293666.613288,
              "command_buffers": 2
            },
            {
              "compute_ns": 270583,
              "compute_span_ns": 270583,
              "calibration_cpu_ns": 785750,
              "calibration_gpu_ticks": 785750,
              "compute_passes": 1,
              "command_buffer_ns": 270583.317615,
              "command_buffers": 2
            },
            {
              "compute_ns": 297542,
              "compute_span_ns": 297542,
              "calibration_cpu_ns": 803500,
              "calibration_gpu_ticks": 803500,
              "compute_passes": 1,
              "command_buffer_ns": 297666.701954,
              "command_buffers": 2
            },
            {
              "compute_ns": 808125,
              "compute_span_ns": 808125,
              "calibration_cpu_ns": 2077667,
              "calibration_gpu_ticks": 2077667,
              "compute_passes": 1,
              "command_buffer_ns": 808208.365925,
              "command_buffers": 2
            },
            {
              "compute_ns": 941583,
              "compute_span_ns": 941583,
              "calibration_cpu_ns": 2365000,
              "calibration_gpu_ticks": 2365000,
              "compute_passes": 1,
              "command_buffer_ns": 941708.334722,
              "command_buffers": 2
            },
            {
              "compute_ns": 297458,
              "compute_span_ns": 297458,
              "calibration_cpu_ns": 785750,
              "calibration_gpu_ticks": 785750,
              "compute_passes": 1,
              "command_buffer_ns": 297541.730106,
              "command_buffers": 2
            },
            {
              "compute_ns": 270750,
              "compute_span_ns": 270750,
              "calibration_cpu_ns": 758583,
              "calibration_gpu_ticks": 758583,
              "compute_passes": 1,
              "command_buffer_ns": 270791.642834,
              "command_buffers": 2
            },
            {
              "compute_ns": 283459,
              "compute_span_ns": 283459,
              "calibration_cpu_ns": 814125,
              "calibration_gpu_ticks": 814125,
              "compute_passes": 1,
              "command_buffer_ns": 283583.300188,
              "command_buffers": 2
            },
            {
              "compute_ns": 271667,
              "compute_span_ns": 271667,
              "calibration_cpu_ns": 860792,
              "calibration_gpu_ticks": 860792,
              "compute_passes": 1,
              "command_buffer_ns": 271666.6786,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 7750,
              "compute_span_ns": 7750,
              "calibration_cpu_ns": 707333,
              "calibration_gpu_ticks": 707333,
              "compute_passes": 1,
              "command_buffer_ns": 7791.67748988,
              "command_buffers": 2
            },
            {
              "compute_ns": 6917,
              "compute_span_ns": 6917,
              "calibration_cpu_ns": 396333,
              "calibration_gpu_ticks": 396333,
              "compute_passes": 1,
              "command_buffer_ns": 6958.31840858,
              "command_buffers": 2
            },
            {
              "compute_ns": 139667,
              "compute_span_ns": 139667,
              "calibration_cpu_ns": 1456000,
              "calibration_gpu_ticks": 1456000,
              "compute_passes": 1,
              "command_buffer_ns": 139666.663017,
              "command_buffers": 2
            },
            {
              "compute_ns": 9083,
              "compute_span_ns": 9083,
              "calibration_cpu_ns": 584250,
              "calibration_gpu_ticks": 584250,
              "compute_passes": 1,
              "command_buffer_ns": 9208.27733353,
              "command_buffers": 2
            },
            {
              "compute_ns": 7958,
              "compute_span_ns": 7958,
              "calibration_cpu_ns": 531666,
              "calibration_gpu_ticks": 531666,
              "compute_passes": 1,
              "command_buffer_ns": 8000.00270829,
              "command_buffers": 2
            },
            {
              "compute_ns": 49958,
              "compute_span_ns": 49958,
              "calibration_cpu_ns": 966083,
              "calibration_gpu_ticks": 966083,
              "compute_passes": 1,
              "command_buffer_ns": 49958.3547935,
              "command_buffers": 2
            },
            {
              "compute_ns": 6917,
              "compute_span_ns": 6917,
              "calibration_cpu_ns": 390333,
              "calibration_gpu_ticks": 390333,
              "compute_passes": 1,
              "command_buffer_ns": 6916.64172336,
              "command_buffers": 2
            },
            {
              "compute_ns": 7292,
              "compute_span_ns": 7292,
              "calibration_cpu_ns": 470958,
              "calibration_gpu_ticks": 470958,
              "compute_passes": 1,
              "command_buffer_ns": 7333.35036784,
              "command_buffers": 2
            },
            {
              "compute_ns": 30792,
              "compute_span_ns": 30792,
              "calibration_cpu_ns": 550750,
              "calibration_gpu_ticks": 550750,
              "compute_passes": 1,
              "command_buffer_ns": 30791.6780002,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 957166.659646,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 975750.037469,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 295208.301395,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 292541.692033,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 281333.341263,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 287916.68592,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 921583.385207,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 934374.984354,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 270833.319519,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 97500.0439212,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 15458.3831318,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7458.32221583,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6874.96503815,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7374.9688454,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 42083.3239332,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7416.64553061,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 14458.3755173,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 10499.9635369,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              14.95572905696875,
              15.246094335453124,
              4.6126297092968755,
              4.570963938015625,
              4.3958334572343745,
              4.4986982175,
              14.399740393859375,
              14.599609130531249,
              4.231770617484375
            ],
            "command_buffer_throughput_us_p50": 4.6126297092968755,
            "command_buffer_throughput_us_p90": 15.013802112665624,
            "command_buffer_latency_us": [
              97.5000439212,
              15.4583831318,
              7.45832221583,
              6.87496503815,
              7.3749688454,
              42.0833239332,
              7.416645530609999,
              14.4583755173,
              10.499963536900001
            ],
            "command_buffer_latency_us_p50": 10.499963536900001,
            "command_buffer_latency_us_p90": 53.16666793080001
          },
          "compute_throughput_us": [
            4.58853125,
            4.227859375,
            4.64909375,
            12.626953125,
            14.712234375,
            4.64778125,
            4.23046875,
            4.429046875,
            4.244796875
          ],
          "compute_throughput_us_p50": 4.58853125,
          "compute_throughput_us_p90": 13.044009375,
          "command_buffer_throughput_us": [
            4.588540832625,
            4.227864337734375,
            4.651042218031249,
            12.628255717578126,
            14.71419273003125,
            4.64908953290625,
            4.23111941928125,
            4.4309890654375,
            4.244791853125
          ],
          "command_buffer_throughput_us_p50": 4.588540832625,
          "command_buffer_throughput_us_p90": 13.045443120068752,
          "compute_latency_us": [
            7.75,
            6.917,
            139.667,
            9.083,
            7.958,
            49.958,
            6.917,
            7.292,
            30.792
          ],
          "compute_latency_us_p50": 7.958,
          "compute_latency_us_p90": 67.89980000000001,
          "command_buffer_latency_us": [
            7.7916774898800005,
            6.95831840858,
            139.66666301700002,
            9.20827733353,
            8.00000270829,
            49.9583547935,
            6.91664172336,
            7.3333503678400005,
            30.7916780002
          ],
          "command_buffer_latency_us_p50": 8.00000270829,
          "command_buffer_latency_us_p90": 67.90001643820003,
          "counter_control_throughput_ratio": 1.0030770183021793,
          "counter_control_latency_ratio": 1.0847507123070566
        },
        "correctness": {
          "max_abs_error": 2.43199801630567e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 6.09648250232,
        "throughput_us_p90": 6.900480210596,
        "latency_us_p50": 204.708,
        "latency_us_p90": 1046.5330000000001
      },
      "slowdown": 0.40195468804894946,
      "device_compute_slowdown": 0.5357646522234891,
      "valid": true,
      "round": 1,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "layernorm_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-sj60hv22/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "83a70ba0e4d1dab6671da56395e0b13750f8d9c9277e4782c4e371450de131c7",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.439417,
        "capture_ms": 0.078334,
        "compile_ms": 64.86825,
        "allocation_upload_ms": 2.557667,
        "cold_call_ms": 6.29,
        "warmup_ms": 200.508917,
        "download_ms": 0.920708,
        "repetitions": 1213,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          22.9824130256,
          24.4746158285,
          22.6265111294,
          24.7409661995,
          22.6782423743,
          23.7644608409,
          23.1671813685,
          22.6654295136,
          24.8410626546
        ],
        "latency_us": [
          286.916,
          265.541,
          247.5,
          234.083,
          254.459,
          282,
          235.542,
          273.709,
          177.75
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 256,
            "reduction_scalar_rounds": 60,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 36864,
              "global_write_bytes": 12288,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 144,
              "global_write_bytes": 48,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 2,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 108,
            "normalized_kernel_cost": 432,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 9216,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 1695958,
              "compute_span_ns": 1695958,
              "calibration_cpu_ns": 2614042,
              "calibration_gpu_ticks": 2614042,
              "compute_passes": 1,
              "command_buffer_ns": 1696000.05029,
              "command_buffers": 2
            },
            {
              "compute_ns": 1037834,
              "compute_span_ns": 1037834,
              "calibration_cpu_ns": 1481875,
              "calibration_gpu_ticks": 1481875,
              "compute_passes": 1,
              "command_buffer_ns": 1037874.95755,
              "command_buffers": 2
            },
            {
              "compute_ns": 1771458,
              "compute_span_ns": 1771458,
              "calibration_cpu_ns": 2919333,
              "calibration_gpu_ticks": 2919333,
              "compute_passes": 1,
              "command_buffer_ns": 1771458.36642,
              "command_buffers": 2
            },
            {
              "compute_ns": 1063625,
              "compute_span_ns": 1063625,
              "calibration_cpu_ns": 1483000,
              "calibration_gpu_ticks": 1483000,
              "compute_passes": 1,
              "command_buffer_ns": 1063624.979,
              "command_buffers": 2
            },
            {
              "compute_ns": 1613125,
              "compute_span_ns": 1613125,
              "calibration_cpu_ns": 2532500,
              "calibration_gpu_ticks": 2532500,
              "compute_passes": 1,
              "command_buffer_ns": 1613166.64889,
              "command_buffers": 2
            },
            {
              "compute_ns": 1031750,
              "compute_span_ns": 1031750,
              "calibration_cpu_ns": 1490583,
              "calibration_gpu_ticks": 1490583,
              "compute_passes": 1,
              "command_buffer_ns": 1031833.29338,
              "command_buffers": 2
            },
            {
              "compute_ns": 1677625,
              "compute_span_ns": 1677625,
              "calibration_cpu_ns": 2662417,
              "calibration_gpu_ticks": 2662417,
              "compute_passes": 1,
              "command_buffer_ns": 1677625.05589,
              "command_buffers": 2
            },
            {
              "compute_ns": 1037458,
              "compute_span_ns": 1037458,
              "calibration_cpu_ns": 1615667,
              "calibration_gpu_ticks": 1615667,
              "compute_passes": 1,
              "command_buffer_ns": 1037458.30711,
              "command_buffers": 2
            },
            {
              "compute_ns": 1600292,
              "compute_span_ns": 1600292,
              "calibration_cpu_ns": 2450458,
              "calibration_gpu_ticks": 2450458,
              "compute_passes": 1,
              "command_buffer_ns": 1600333.37306,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 18000,
              "compute_span_ns": 18000,
              "calibration_cpu_ns": 417583,
              "calibration_gpu_ticks": 417583,
              "compute_passes": 1,
              "command_buffer_ns": 18083.3158083,
              "command_buffers": 2
            },
            {
              "compute_ns": 153458,
              "compute_span_ns": 153458,
              "calibration_cpu_ns": 1524541,
              "calibration_gpu_ticks": 1524541,
              "compute_passes": 1,
              "command_buffer_ns": 153541.623149,
              "command_buffers": 2
            },
            {
              "compute_ns": 19916,
              "compute_span_ns": 19916,
              "calibration_cpu_ns": 457959,
              "calibration_gpu_ticks": 457959,
              "compute_passes": 1,
              "command_buffer_ns": 19875.0058189,
              "command_buffers": 2
            },
            {
              "compute_ns": 20042,
              "compute_span_ns": 20042,
              "calibration_cpu_ns": 427209,
              "calibration_gpu_ticks": 427209,
              "compute_passes": 1,
              "command_buffer_ns": 20041.6543521,
              "command_buffers": 2
            },
            {
              "compute_ns": 18125,
              "compute_span_ns": 18125,
              "calibration_cpu_ns": 464542,
              "calibration_gpu_ticks": 464542,
              "compute_passes": 1,
              "command_buffer_ns": 18208.345864,
              "command_buffers": 2
            },
            {
              "compute_ns": 17916,
              "compute_span_ns": 17916,
              "calibration_cpu_ns": 512125,
              "calibration_gpu_ticks": 512125,
              "compute_passes": 1,
              "command_buffer_ns": 17958.2857527,
              "command_buffers": 2
            },
            {
              "compute_ns": 17916,
              "compute_span_ns": 17916,
              "calibration_cpu_ns": 485250,
              "calibration_gpu_ticks": 485250,
              "compute_passes": 1,
              "command_buffer_ns": 17958.2857527,
              "command_buffers": 2
            },
            {
              "compute_ns": 20000,
              "compute_span_ns": 20000,
              "calibration_cpu_ns": 476334,
              "calibration_gpu_ticks": 476334,
              "compute_passes": 1,
              "command_buffer_ns": 20041.6543521,
              "command_buffers": 2
            },
            {
              "compute_ns": 17958,
              "compute_span_ns": 17958,
              "calibration_cpu_ns": 564458,
              "calibration_gpu_ticks": 564458,
              "compute_passes": 1,
              "command_buffer_ns": 17958.3439603,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1683333.30657,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1731041.58603,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1066083.26314,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1701041.70684,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1029125.00734,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1545708.33547,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1033791.63193,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1627583.34773,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1038249.98951,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 23458.3858401,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 22833.2937695,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18208.4040716,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 213166.640606,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 25583.372917,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 258583.342656,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 24874.9856837,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 490208.389238,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 21916.6395254,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              26.30208291515625,
              27.04752478171875,
              16.657550986562498,
              26.578776669375003,
              16.0800782396875,
              24.15169274171875,
              16.15299424890625,
              25.43098980828125,
              16.22265608609375
            ],
            "command_buffer_throughput_us_p50": 24.15169274171875,
            "command_buffer_throughput_us_p90": 26.672526291843752,
            "command_buffer_latency_us": [
              23.4583858401,
              22.8332937695,
              18.2084040716,
              213.166640606,
              25.583372917000002,
              258.583342656,
              24.8749856837,
              490.208389238,
              21.9166395254
            ],
            "command_buffer_latency_us_p50": 24.8749856837,
            "command_buffer_latency_us_p90": 304.90835197240006
          },
          "compute_throughput_us": [
            26.49934375,
            16.21615625,
            27.67903125,
            16.619140625,
            25.205078125,
            16.12109375,
            26.212890625,
            16.21028125,
            25.0045625
          ],
          "compute_throughput_us_p50": 25.0045625,
          "compute_throughput_us_p90": 26.73528125,
          "command_buffer_throughput_us": [
            26.50000078578125,
            16.21679621171875,
            27.679036975312503,
            16.619140296875,
            25.20572888890625,
            16.1223952090625,
            26.212891498281248,
            16.21028604859375,
            25.0052089540625
          ],
          "command_buffer_throughput_us_p50": 25.0052089540625,
          "command_buffer_throughput_us_p90": 26.7358080236875,
          "compute_latency_us": [
            18.0,
            153.458,
            19.916,
            20.042,
            18.125,
            17.916,
            17.916,
            20.0,
            17.958
          ],
          "compute_latency_us_p50": 18.125,
          "compute_latency_us_p90": 46.72520000000003,
          "command_buffer_latency_us": [
            18.0833158083,
            153.541623149,
            19.8750058189,
            20.0416543521,
            18.208345863999998,
            17.9582857527,
            17.9582857527,
            20.0416543521,
            17.958343960300002
          ],
          "command_buffer_latency_us_p50": 18.208345863999998,
          "command_buffer_latency_us_p90": 46.74164811148002,
          "counter_control_throughput_ratio": 1.0075247983691418,
          "counter_control_latency_ratio": 0.7219415512857018
        },
        "correctness": {
          "max_abs_error": 2.9463829065434766e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 23.1671813685,
        "throughput_us_p90": 24.76098549052,
        "latency_us_p50": 254.459,
        "latency_us_p90": 282.9832
      },
      "torch": {
        "cold_call_ms": 6.175792,
        "warmup_ms": 200.391917,
        "repetitions": 471,
        "throughput_us": [
          66.00327388535031,
          59.99256900212314,
          64.075991507431,
          60.902866242038215,
          63.573602972399144,
          60.46487898089172,
          63.171798301486206,
          60.248318471337576,
          63.697452229299365
        ],
        "latency_us": [
          485.416,
          335.83299999999997,
          332.208,
          320.834,
          309.333,
          300.042,
          299.70799999999997,
          293.584,
          314.916
        ],
        "throughput_us_p50": 63.171798301486206,
        "throughput_us_p90": 64.46144798301486,
        "latency_us_p50": 314.916,
        "latency_us_p90": 365.7496,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 2975749.9578408897,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2691541.681997478,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2731833.3741277456,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 3022791.7013689876,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2917750.0400692225,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2925208.304077387,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2961416.670586914,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2879916.6320823133,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 3027708.3278633654,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 202916.6789725423,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 251249.99228864908,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44249.987695366144,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43124.950025230646,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 237416.7088419199,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 48583.31521973014,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43625.012040138245,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 42541.65105521679,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44291.66438058019,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              46.4960930912639,
              42.055338781210594,
              42.684896470746025,
              47.23112033389043,
              45.5898443760816,
              45.70637975120917,
              46.27213547792053,
              44.998697376286145,
              47.307942622865085
            ],
            "command_buffer_throughput_us_p50": 45.70637975120917,
            "command_buffer_throughput_us_p90": 47.24648479168536,
            "command_buffer_latency_us": [
              202.9166789725423,
              251.24999228864908,
              44.249987695366144,
              43.124950025230646,
              237.4167088419199,
              48.58331521973014,
              43.625012040138245,
              42.54165105521679,
              44.29166438058019
            ],
            "command_buffer_latency_us_p50": 44.29166438058019,
            "command_buffer_latency_us_p90": 240.18336553126574
          },
          "throughput": [
            {
              "compute_ns": 2876375.0,
              "compute_span_ns": 2876375.0,
              "command_buffer_ns": 2876375.0451616943,
              "calibration_cpu_ns": 4580459.0,
              "calibration_gpu_ticks": 4580459.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 3148291.0,
              "compute_span_ns": 3148291.0,
              "command_buffer_ns": 3148249.990772456,
              "calibration_cpu_ns": 4578666.0,
              "calibration_gpu_ticks": 4578666.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2923209.0,
              "compute_span_ns": 2923209.0,
              "command_buffer_ns": 2923250.023741275,
              "calibration_cpu_ns": 4080250.0,
              "calibration_gpu_ticks": 4080250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2966042.0,
              "compute_span_ns": 2966042.0,
              "command_buffer_ns": 2966041.676700115,
              "calibration_cpu_ns": 4608417.0,
              "calibration_gpu_ticks": 4608417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2978333.0,
              "compute_span_ns": 2978333.0,
              "command_buffer_ns": 2978291.653562337,
              "calibration_cpu_ns": 4112166.0,
              "calibration_gpu_ticks": 4112166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2904583.0,
              "compute_span_ns": 2904583.0,
              "command_buffer_ns": 2904583.3507552743,
              "calibration_cpu_ns": 4287125.0,
              "calibration_gpu_ticks": 4287125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2886291.0,
              "compute_span_ns": 2886291.0,
              "command_buffer_ns": 2886249.974835664,
              "calibration_cpu_ns": 4465208.0,
              "calibration_gpu_ticks": 4465208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2869375.0,
              "compute_span_ns": 2869375.0,
              "command_buffer_ns": 2869374.9918602407,
              "calibration_cpu_ns": 4232791.0,
              "calibration_gpu_ticks": 4232791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2787791.0,
              "compute_span_ns": 2787791.0,
              "command_buffer_ns": 2787749.981507659,
              "calibration_cpu_ns": 4005709.0,
              "calibration_gpu_ticks": 4005709.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 44916.0,
              "compute_span_ns": 44916.0,
              "command_buffer_ns": 44874.963350594044,
              "calibration_cpu_ns": 616708.0,
              "calibration_gpu_ticks": 616708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 42791.0,
              "compute_span_ns": 42791.0,
              "command_buffer_ns": 42749.97627362609,
              "calibration_cpu_ns": 609042.0,
              "calibration_gpu_ticks": 609042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 42875.0,
              "compute_span_ns": 42875.0,
              "command_buffer_ns": 42875.00632926822,
              "calibration_cpu_ns": 520875.0,
              "calibration_gpu_ticks": 520875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44625.0,
              "compute_span_ns": 44625.0,
              "command_buffer_ns": 44625.019654631615,
              "calibration_cpu_ns": 611917.0,
              "calibration_gpu_ticks": 611917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 45083.0,
              "compute_span_ns": 45083.0,
              "command_buffer_ns": 45041.670091450214,
              "calibration_cpu_ns": 772625.0,
              "calibration_gpu_ticks": 772625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 43417.0,
              "compute_span_ns": 43417.0,
              "command_buffer_ns": 43458.36350694299,
              "calibration_cpu_ns": 436000.0,
              "calibration_gpu_ticks": 436000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 208167.0,
              "compute_span_ns": 208167.0,
              "command_buffer_ns": 208166.66074097157,
              "calibration_cpu_ns": 1289584.0,
              "calibration_gpu_ticks": 1289584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44292.0,
              "compute_span_ns": 44292.0,
              "command_buffer_ns": 44333.34106579423,
              "calibration_cpu_ns": 614667.0,
              "calibration_gpu_ticks": 614667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 45250.0,
              "compute_span_ns": 45250.0,
              "command_buffer_ns": 45249.995309859514,
              "calibration_cpu_ns": 535125.0,
              "calibration_gpu_ticks": 535125.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            44.943359375,
            49.192046875,
            45.675140625,
            46.34440625,
            46.536453125,
            45.384109375,
            45.098296875,
            44.833984375,
            43.559234375
          ],
          "compute_throughput_us_p50": 45.384109375,
          "compute_throughput_us_p90": 47.067571875,
          "command_buffer_throughput_us": [
            44.94336008065147,
            49.191406105819624,
            45.67578162095742,
            46.3444011984393,
            46.53580708691152,
            45.38411485555116,
            45.09765585680725,
            44.83398424781626,
            43.55859346105717
          ],
          "command_buffer_throughput_us_p50": 45.38411485555116,
          "command_buffer_throughput_us_p90": 47.06692689069314,
          "compute_latency_us": [
            44.916,
            42.791,
            42.875,
            44.625,
            45.083,
            43.417,
            208.167,
            44.292,
            45.25
          ],
          "compute_latency_us_p50": 44.625,
          "compute_latency_us_p90": 77.83340000000003,
          "command_buffer_latency_us": [
            44.874963350594044,
            42.74997627362609,
            42.87500632926822,
            44.625019654631615,
            45.041670091450214,
            43.45836350694299,
            208.16666074097157,
            44.33334106579423,
            45.249995309859514
          ],
          "command_buffer_latency_us_p50": 44.625019654631615,
          "command_buffer_latency_us_p90": 77.83332839608195,
          "counter_control_throughput_ratio": 0.9929492360276142,
          "counter_control_latency_ratio": 0.968926966136944
        },
        "download_ms": 0.359,
        "allocation_upload_ms": 1.32075,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 2.982737865586671e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.36673297248773995,
      "device_compute_slowdown": 0.5509541300765032,
      "valid": true,
      "round": 1,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "layernorm_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.32225,
        "warmup_ms": 200.521458,
        "repetitions": 423,
        "throughput_us": [
          64.83884869976359,
          61.62105910165485,
          62.11327895981088,
          61.51231205673759,
          65.24891725768322,
          58.731973995271865,
          59.211879432624116,
          62.00138061465721,
          60.846827423167845
        ],
        "latency_us": [
          476.791,
          345.66700000000003,
          295.66700000000003,
          253.542,
          245.417,
          309.25,
          275.54200000000003,
          272.125,
          257.041
        ],
        "throughput_us_p50": 61.62105910165485,
        "throughput_us_p90": 64.92086241134751,
        "latency_us_p50": 275.54200000000003,
        "latency_us_p90": 371.89180000000005,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 2903291.6645519435,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2901625.004597008,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2866458.3223871887,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2959000.0049211085,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2839250.024408102,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2951624.9778680503,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2957416.6983366013,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5145166.70955345,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4867166.688200086,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 42791.65295884013,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43875.01394376159,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 45499.99721348286,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 48500.02005696297,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44125.01584738493,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 48083.311412483454,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43249.980080872774,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 216500.01872330904,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44291.66438058019,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              45.36393225862412,
              45.337890696828254,
              44.78841128729982,
              46.23437507689232,
              44.363281631376594,
              46.11914027918829,
              46.209635911509395,
              80.39322983677266,
              76.04947950312635
            ],
            "command_buffer_throughput_us_p50": 46.11914027918829,
            "command_buffer_throughput_us_p90": 76.91822956985561,
            "command_buffer_latency_us": [
              42.79165295884013,
              43.87501394376159,
              45.49999721348286,
              48.50002005696297,
              44.12501584738493,
              48.083311412483454,
              43.249980080872774,
              216.50001872330904,
              44.29166438058019
            ],
            "command_buffer_latency_us_p50": 44.29166438058019,
            "command_buffer_latency_us_p90": 82.10001979023221
          },
          "throughput": [
            {
              "compute_ns": 2989917.0,
              "compute_span_ns": 2989917.0,
              "command_buffer_ns": 2989916.654769331,
              "calibration_cpu_ns": 4455250.0,
              "calibration_gpu_ticks": 4455250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2957333.0,
              "compute_span_ns": 2957333.0,
              "command_buffer_ns": 2957291.668280959,
              "calibration_cpu_ns": 4288750.0,
              "calibration_gpu_ticks": 4288750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2964958.0,
              "compute_span_ns": 2964958.0,
              "command_buffer_ns": 2964916.6390299797,
              "calibration_cpu_ns": 4336125.0,
              "calibration_gpu_ticks": 4336125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2910583.0,
              "compute_span_ns": 2910583.0,
              "command_buffer_ns": 2910583.3382345736,
              "calibration_cpu_ns": 4558625.0,
              "calibration_gpu_ticks": 4558625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2889625.0,
              "compute_span_ns": 2889625.0,
              "command_buffer_ns": 2889625.0296384096,
              "calibration_cpu_ns": 4236584.0,
              "calibration_gpu_ticks": 4236584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2980250.0,
              "compute_span_ns": 2980250.0,
              "command_buffer_ns": 2980249.99210611,
              "calibration_cpu_ns": 4342709.0,
              "calibration_gpu_ticks": 4342709.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2676625.0,
              "compute_span_ns": 2676625.0,
              "command_buffer_ns": 2676625.0375658274,
              "calibration_cpu_ns": 4825542.0,
              "calibration_gpu_ticks": 4825542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2854625.0,
              "compute_span_ns": 2854625.0,
              "command_buffer_ns": 2854624.9959617853,
              "calibration_cpu_ns": 4298667.0,
              "calibration_gpu_ticks": 4298667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5010750.0,
              "compute_span_ns": 5010750.0,
              "command_buffer_ns": 5010750.028304756,
              "calibration_cpu_ns": 7282083.0,
              "calibration_gpu_ticks": 7282083.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 43667.0,
              "compute_span_ns": 43667.0,
              "command_buffer_ns": 43708.36541056633,
              "calibration_cpu_ns": 518167.0,
              "calibration_gpu_ticks": 518167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 50792.0,
              "compute_span_ns": 50792.0,
              "command_buffer_ns": 50791.65566712618,
              "calibration_cpu_ns": 3355667.0,
              "calibration_gpu_ticks": 3355667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47625.0,
              "compute_span_ns": 47625.0,
              "command_buffer_ns": 47624.98429045081,
              "calibration_cpu_ns": 475500.0,
              "calibration_gpu_ticks": 475500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 46417.0,
              "compute_span_ns": 46417.0,
              "command_buffer_ns": 46458.328142762184,
              "calibration_cpu_ns": 563833.0,
              "calibration_gpu_ticks": 563833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 43875.0,
              "compute_span_ns": 43875.0,
              "command_buffer_ns": 43875.01394376159,
              "calibration_cpu_ns": 567125.0,
              "calibration_gpu_ticks": 567125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 290709.0,
              "compute_span_ns": 290709.0,
              "command_buffer_ns": 290750.00202283263,
              "calibration_cpu_ns": 1599875.0,
              "calibration_gpu_ticks": 1599875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44583.0,
              "compute_span_ns": 44583.0,
              "command_buffer_ns": 44541.66628420353,
              "calibration_cpu_ns": 471083.0,
              "calibration_gpu_ticks": 471083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 43166.0,
              "compute_span_ns": 43166.0,
              "command_buffer_ns": 43125.00823289156,
              "calibration_cpu_ns": 580250.0,
              "calibration_gpu_ticks": 580250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 43125.0,
              "compute_span_ns": 43125.0,
              "command_buffer_ns": 43125.00823289156,
              "calibration_cpu_ns": 507167.0,
              "calibration_gpu_ticks": 507167.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            46.717453125,
            46.208328125,
            46.32746875,
            45.477859375,
            45.150390625,
            46.56640625,
            41.822265625,
            44.603515625,
            78.29296875
          ],
          "compute_throughput_us_p50": 46.208328125,
          "compute_throughput_us_p90": 53.032556250000006,
          "command_buffer_throughput_us": [
            46.7174477307708,
            46.207682316889986,
            46.32682248484343,
            45.47786465991521,
            45.15039108810015,
            46.56640612665797,
            41.82226621196605,
            44.603515561902896,
            78.29296919226181
          ],
          "command_buffer_throughput_us_p50": 46.207682316889986,
          "command_buffer_throughput_us_p90": 53.03255202306901,
          "compute_latency_us": [
            43.667,
            50.792,
            47.625,
            46.417,
            43.875,
            290.709,
            44.583,
            43.166,
            43.125
          ],
          "compute_latency_us_p50": 44.583,
          "compute_latency_us_p90": 98.77540000000005,
          "command_buffer_latency_us": [
            43.70836541056633,
            50.79165566712618,
            47.62498429045081,
            46.458328142762184,
            43.87501394376159,
            290.75000202283263,
            44.54166628420353,
            43.12500823289156,
            43.12500823289156
          ],
          "command_buffer_latency_us_p50": 44.54166628420353,
          "command_buffer_latency_us_p90": 98.78332493826751,
          "counter_control_throughput_ratio": 1.0177423632287577,
          "counter_control_latency_ratio": 1.0214226931735484
        },
        "download_ms": 0.377958,
        "allocation_upload_ms": 0.803583,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 2.982737865586671e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-avggotcn/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "b8d9d984f3aaf8bfdb9269f6e0efc630605362c6ff4e852cc112881dfaf5e97b",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.270916,
        "capture_ms": 0.078542,
        "compile_ms": 65.027542,
        "allocation_upload_ms": 3.869416,
        "cold_call_ms": 8.562291,
        "warmup_ms": 200.362292,
        "download_ms": 0.784458,
        "repetitions": 1484,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          24.9191374663,
          22.8936152291,
          23.0980175202,
          22.8821314016,
          23.0559858491,
          23.9276165768,
          22.5318113208,
          23.3479332884,
          22.8765161725
        ],
        "latency_us": [
          445.875,
          319,
          253.709,
          245.542,
          245.625,
          232,
          245.125,
          235.083,
          207.292
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 128,
            "reduction_scalar_rounds": 60,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 36864,
              "global_write_bytes": 12288,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 144,
              "global_write_bytes": 48,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 2,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 100,
            "normalized_kernel_cost": 400,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 9216,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 1405667,
              "compute_span_ns": 1405667,
              "calibration_cpu_ns": 2687792,
              "calibration_gpu_ticks": 2687792,
              "compute_passes": 1,
              "command_buffer_ns": 1405708.31718,
              "command_buffers": 2
            },
            {
              "compute_ns": 1067375,
              "compute_span_ns": 1067375,
              "calibration_cpu_ns": 1640875,
              "calibration_gpu_ticks": 1640875,
              "compute_passes": 1,
              "command_buffer_ns": 1067416.68424,
              "command_buffers": 2
            },
            {
              "compute_ns": 1688500,
              "compute_span_ns": 1688500,
              "calibration_cpu_ns": 3008084,
              "calibration_gpu_ticks": 3008084,
              "compute_passes": 1,
              "command_buffer_ns": 1688624.96503,
              "command_buffers": 2
            },
            {
              "compute_ns": 1024083,
              "compute_span_ns": 1024083,
              "calibration_cpu_ns": 1642292,
              "calibration_gpu_ticks": 1642292,
              "compute_passes": 1,
              "command_buffer_ns": 1024083.35079,
              "command_buffers": 2
            },
            {
              "compute_ns": 1672833,
              "compute_span_ns": 1672833,
              "calibration_cpu_ns": 2651292,
              "calibration_gpu_ticks": 2651292,
              "compute_passes": 1,
              "command_buffer_ns": 1672874.96151,
              "command_buffers": 2
            },
            {
              "compute_ns": 1052084,
              "compute_span_ns": 1052084,
              "calibration_cpu_ns": 1487875,
              "calibration_gpu_ticks": 1487875,
              "compute_passes": 1,
              "command_buffer_ns": 1052166.68453,
              "command_buffers": 2
            },
            {
              "compute_ns": 1606958,
              "compute_span_ns": 1606958,
              "calibration_cpu_ns": 2817292,
              "calibration_gpu_ticks": 2817292,
              "compute_passes": 1,
              "command_buffer_ns": 1607041.63136,
              "command_buffers": 2
            },
            {
              "compute_ns": 1046750,
              "compute_span_ns": 1046750,
              "calibration_cpu_ns": 1518167,
              "calibration_gpu_ticks": 1518167,
              "compute_passes": 1,
              "command_buffer_ns": 1046875.02608,
              "command_buffers": 2
            },
            {
              "compute_ns": 1653625,
              "compute_span_ns": 1653625,
              "calibration_cpu_ns": 2945375,
              "calibration_gpu_ticks": 2945375,
              "compute_passes": 1,
              "command_buffer_ns": 1653708.34293,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 17625,
              "compute_span_ns": 17625,
              "calibration_cpu_ns": 452459,
              "calibration_gpu_ticks": 452459,
              "compute_passes": 1,
              "command_buffer_ns": 17666.6653715,
              "command_buffers": 2
            },
            {
              "compute_ns": 272666,
              "compute_span_ns": 272666,
              "calibration_cpu_ns": 1597750,
              "calibration_gpu_ticks": 1597750,
              "compute_passes": 1,
              "command_buffer_ns": 272708.3629,
              "command_buffers": 2
            },
            {
              "compute_ns": 20042,
              "compute_span_ns": 20042,
              "calibration_cpu_ns": 634125,
              "calibration_gpu_ticks": 634125,
              "compute_passes": 1,
              "command_buffer_ns": 20125.0077225,
              "command_buffers": 2
            },
            {
              "compute_ns": 20333,
              "compute_span_ns": 20333,
              "calibration_cpu_ns": 418458,
              "calibration_gpu_ticks": 418458,
              "compute_passes": 1,
              "command_buffer_ns": 20416.6281037,
              "command_buffers": 2
            },
            {
              "compute_ns": 21875,
              "compute_span_ns": 21875,
              "calibration_cpu_ns": 422375,
              "calibration_gpu_ticks": 422375,
              "compute_passes": 1,
              "command_buffer_ns": 21916.6977331,
              "command_buffers": 2
            },
            {
              "compute_ns": 17708,
              "compute_span_ns": 17708,
              "calibration_cpu_ns": 533542,
              "calibration_gpu_ticks": 533542,
              "compute_passes": 1,
              "command_buffer_ns": 17791.6372195,
              "command_buffers": 2
            },
            {
              "compute_ns": 188875,
              "compute_span_ns": 188875,
              "calibration_cpu_ns": 1613666,
              "calibration_gpu_ticks": 1613666,
              "compute_passes": 1,
              "command_buffer_ns": 188999.983948,
              "command_buffers": 2
            },
            {
              "compute_ns": 22708,
              "compute_span_ns": 22708,
              "calibration_cpu_ns": 663583,
              "calibration_gpu_ticks": 663583,
              "compute_passes": 1,
              "command_buffer_ns": 22916.6471399,
              "command_buffers": 2
            },
            {
              "compute_ns": 20125,
              "compute_span_ns": 20125,
              "calibration_cpu_ns": 425250,
              "calibration_gpu_ticks": 425250,
              "compute_passes": 1,
              "command_buffer_ns": 20166.6844077,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1277708.33205,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1655458.35625,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1019041.69424,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1689791.67938,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1041541.69094,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1596000.04554,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1042750.02377,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1578124.99674,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1033041.68442,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 18541.701138,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 22958.2656175,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 17750.0187419,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 27125.0028163,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 40374.987293,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 35291.7122655,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20124.9495149,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 17875.0487976,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20208.361093,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              19.96419268828125,
              25.86653681640625,
              15.9225264725,
              26.4029949903125,
              16.2740889209375,
              24.9375007115625,
              16.292969121406248,
              24.658203074062502,
              16.1412763190625
            ],
            "command_buffer_throughput_us_p50": 19.96419268828125,
            "command_buffer_throughput_us_p90": 25.9738284511875,
            "command_buffer_latency_us": [
              18.541701138,
              22.9582656175,
              17.7500187419,
              27.1250028163,
              40.374987293,
              35.291712265499996,
              20.1249495149,
              17.8750487976,
              20.208361093
            ],
            "command_buffer_latency_us_p50": 20.208361093,
            "command_buffer_latency_us_p90": 36.308367270999995
          },
          "compute_throughput_us": [
            21.963546875,
            16.677734375,
            26.3828125,
            16.001296875,
            26.138015625,
            16.4388125,
            25.10871875,
            16.35546875,
            25.837890625
          ],
          "compute_throughput_us_p50": 21.963546875,
          "compute_throughput_us_p90": 26.186975,
          "command_buffer_throughput_us": [
            21.9641924559375,
            16.67838569125,
            26.38476507859375,
            16.001302356093753,
            26.13867127359375,
            16.440104445781248,
            25.11002549,
            16.3574222825,
            25.83919285828125
          ],
          "command_buffer_throughput_us_p50": 21.9641924559375,
          "command_buffer_throughput_us_p90": 26.18789003459375,
          "compute_latency_us": [
            17.625,
            272.666,
            20.042,
            20.333,
            21.875,
            17.708,
            188.875,
            22.708,
            20.125
          ],
          "compute_latency_us_p50": 20.333,
          "compute_latency_us_p90": 205.63320000000002,
          "command_buffer_latency_us": [
            17.6666653715,
            272.7083629,
            20.125007722499998,
            20.4166281037,
            21.9166977331,
            17.7916372195,
            188.99998394800002,
            22.9166471399,
            20.1666844077
          ],
          "command_buffer_latency_us_p50": 20.4166281037,
          "command_buffer_latency_us_p90": 205.74165973840002,
          "counter_control_throughput_ratio": 1.1001793460363778,
          "counter_control_latency_ratio": 0.9979376513954693
        },
        "correctness": {
          "max_abs_error": 2.9463829065434766e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 23.0559858491,
        "throughput_us_p90": 24.125920754699997,
        "latency_us_p50": 245.542,
        "latency_us_p90": 344.375
      },
      "slowdown": 0.3741575718629741,
      "device_compute_slowdown": 0.4753157659282874,
      "valid": true,
      "round": 1,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "layernorm_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-au_qh5zh/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "2f415c8199d32d1b999ce6dae6674c9bc8bbdf9b0adda641a6d2aee649468028",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.784584,
        "capture_ms": 0.078042,
        "compile_ms": 64.189917,
        "allocation_upload_ms": 8.408459,
        "cold_call_ms": 8.430583,
        "warmup_ms": 201.84075,
        "download_ms": 3.603625,
        "repetitions": 284,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          109.367221831,
          110.10578169,
          100.959651408,
          110.136295775,
          109.243542254,
          101.040785211,
          110.224911972,
          104.149207746,
          105.681193662
        ],
        "latency_us": [
          372.959,
          310.167,
          312.5,
          297.708,
          2091.125,
          949.5,
          498.875,
          393.875,
          315.042
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 384,
            "reduction_scalar_rounds": 120,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 73728,
              "global_write_bytes": 24576,
              "private_read_bytes": 147456,
              "private_write_bytes": 49152
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 288,
              "global_write_bytes": 96,
              "private_read_bytes": 576,
              "private_write_bytes": 192
            },
            "striped_storage_scalars_per_worker": 48,
            "reduction_operations": 2,
            "reduction_elements": 12288,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 160,
            "normalized_kernel_cost": 1920,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 18432,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 5286625,
              "compute_span_ns": 5286625,
              "calibration_cpu_ns": 6430250,
              "calibration_gpu_ticks": 6430250,
              "compute_passes": 1,
              "command_buffer_ns": 5286666.68106,
              "command_buffers": 2
            },
            {
              "compute_ns": 5261000,
              "compute_span_ns": 5261000,
              "calibration_cpu_ns": 6122500,
              "calibration_gpu_ticks": 6122500,
              "compute_passes": 1,
              "command_buffer_ns": 5261083.30814,
              "command_buffers": 2
            },
            {
              "compute_ns": 5219583,
              "compute_span_ns": 5219583,
              "calibration_cpu_ns": 6160916,
              "calibration_gpu_ticks": 6160916,
              "compute_passes": 1,
              "command_buffer_ns": 5219541.6647,
              "command_buffers": 2
            },
            {
              "compute_ns": 5361708,
              "compute_span_ns": 5361708,
              "calibration_cpu_ns": 6429417,
              "calibration_gpu_ticks": 6429417,
              "compute_passes": 1,
              "command_buffer_ns": 5361791.58371,
              "command_buffers": 2
            },
            {
              "compute_ns": 5234625,
              "compute_span_ns": 5234625,
              "calibration_cpu_ns": 6337833,
              "calibration_gpu_ticks": 6337833,
              "compute_passes": 1,
              "command_buffer_ns": 5234625.01587,
              "command_buffers": 2
            },
            {
              "compute_ns": 5242041,
              "compute_span_ns": 5242041,
              "calibration_cpu_ns": 6285875,
              "calibration_gpu_ticks": 6285875,
              "compute_passes": 1,
              "command_buffer_ns": 5242041.6614,
              "command_buffers": 2
            },
            {
              "compute_ns": 5295000,
              "compute_span_ns": 5295000,
              "calibration_cpu_ns": 6182417,
              "calibration_gpu_ticks": 6182417,
              "compute_passes": 1,
              "command_buffer_ns": 5295041.65752,
              "command_buffers": 2
            },
            {
              "compute_ns": 5333584,
              "compute_span_ns": 5333584,
              "calibration_cpu_ns": 9801416,
              "calibration_gpu_ticks": 9801416,
              "compute_passes": 1,
              "command_buffer_ns": 5333624.9548,
              "command_buffers": 2
            },
            {
              "compute_ns": 7268291,
              "compute_span_ns": 7268291,
              "calibration_cpu_ns": 8441750,
              "calibration_gpu_ticks": 8441750,
              "compute_passes": 1,
              "command_buffer_ns": 7268291.66524,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 116416,
              "compute_span_ns": 116416,
              "calibration_cpu_ns": 1080583,
              "calibration_gpu_ticks": 1080583,
              "compute_passes": 1,
              "command_buffer_ns": 116499.955766,
              "command_buffers": 2
            },
            {
              "compute_ns": 505333,
              "compute_span_ns": 505333,
              "calibration_cpu_ns": 1825083,
              "calibration_gpu_ticks": 1825083,
              "compute_passes": 1,
              "command_buffer_ns": 505416.65405,
              "command_buffers": 2
            },
            {
              "compute_ns": 73333,
              "compute_span_ns": 73333,
              "calibration_cpu_ns": 686042,
              "calibration_gpu_ticks": 686042,
              "compute_passes": 1,
              "command_buffer_ns": 73333.3290555,
              "command_buffers": 2
            },
            {
              "compute_ns": 112750,
              "compute_span_ns": 112750,
              "calibration_cpu_ns": 1050041,
              "calibration_gpu_ticks": 1050041,
              "compute_passes": 1,
              "command_buffer_ns": 112791.720312,
              "command_buffers": 2
            },
            {
              "compute_ns": 72833,
              "compute_span_ns": 72833,
              "calibration_cpu_ns": 640041,
              "calibration_gpu_ticks": 640041,
              "compute_passes": 1,
              "command_buffer_ns": 72916.6786186,
              "command_buffers": 2
            },
            {
              "compute_ns": 115083,
              "compute_span_ns": 115083,
              "calibration_cpu_ns": 1022375,
              "calibration_gpu_ticks": 1022375,
              "compute_passes": 1,
              "command_buffer_ns": 115125.032607,
              "command_buffers": 2
            },
            {
              "compute_ns": 114334,
              "compute_span_ns": 114334,
              "calibration_cpu_ns": 3102375,
              "calibration_gpu_ticks": 3102375,
              "compute_passes": 1,
              "command_buffer_ns": 114416.645374,
              "command_buffers": 2
            },
            {
              "compute_ns": 74666,
              "compute_span_ns": 74666,
              "calibration_cpu_ns": 515792,
              "calibration_gpu_ticks": 515792,
              "compute_passes": 1,
              "command_buffer_ns": 74625.0152588,
              "command_buffers": 2
            },
            {
              "compute_ns": 72916,
              "compute_span_ns": 72916,
              "calibration_cpu_ns": 499292,
              "calibration_gpu_ticks": 499292,
              "compute_passes": 1,
              "command_buffer_ns": 72916.6786186,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 5338125.04727,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5325666.68699,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5428333.3593,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5478083.33067,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5455791.65919,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5314291.68768,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5419000.05192,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7437625.0268,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7347416.71477,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 117541.640066,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 115458.329674,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 74541.6618884,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 75416.6394472,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 114916.649181,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 116625.044029,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 74958.3705328,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 72041.6428521,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 115208.32777,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              83.40820386359376,
              83.21354198421875,
              84.8177087390625,
              85.59505204171874,
              85.24674467484375,
              83.03580762,
              84.67187581125,
              116.21289104375,
              114.80338616828125
            ],
            "command_buffer_throughput_us_p50": 84.8177087390625,
            "command_buffer_throughput_us_p90": 115.085287143375,
            "command_buffer_latency_us": [
              117.54164006600001,
              115.458329674,
              74.5416618884,
              75.4166394472,
              114.916649181,
              116.625044029,
              74.9583705328,
              72.04164285210001,
              115.20832777
            ],
            "command_buffer_latency_us_p50": 114.916649181,
            "command_buffer_latency_us_p90": 116.8083632364
          },
          "compute_throughput_us": [
            82.603515625,
            82.203125,
            81.555984375,
            83.7766875,
            81.791015625,
            81.906890625,
            82.734375,
            83.33725,
            113.567046875
          ],
          "compute_throughput_us_p50": 82.603515625,
          "compute_throughput_us_p90": 89.734759375,
          "command_buffer_throughput_us": [
            82.6041668915625,
            82.20442668968751,
            81.55533851093749,
            83.77799349546875,
            81.79101587296876,
            81.906900959375,
            82.73502589875,
            83.33788991875001,
            113.567057269375
          ],
          "command_buffer_throughput_us_p50": 82.6041668915625,
          "command_buffer_throughput_us_p90": 89.73580625025001,
          "compute_latency_us": [
            116.416,
            505.333,
            73.333,
            112.75,
            72.833,
            115.083,
            114.334,
            74.666,
            72.916
          ],
          "compute_latency_us_p50": 112.75,
          "compute_latency_us_p90": 194.19940000000008,
          "command_buffer_latency_us": [
            116.499955766,
            505.41665405000003,
            73.3333290555,
            112.79172031200001,
            72.9166786186,
            115.125032607,
            114.416645374,
            74.6250152588,
            72.9166786186
          ],
          "command_buffer_latency_us_p50": 112.79172031200001,
          "command_buffer_latency_us_p90": 194.28329542280008,
          "counter_control_throughput_ratio": 0.9787714534554594,
          "counter_control_latency_ratio": 0.9911377423403732
        },
        "correctness": {
          "max_abs_error": 2.7199924934606656e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 109.243542254,
        "throughput_us_p90": 110.15401901439999,
        "latency_us_p50": 372.959,
        "latency_us_p90": 1177.8250000000003
      },
      "torch": {
        "cold_call_ms": 7.540958,
        "warmup_ms": 201.348125,
        "repetitions": 78,
        "throughput_us": [
          397.95780769230765,
          383.35096153846155,
          380.3296025641026,
          402.6768205128205,
          368.73878205128204,
          403.8039615384615,
          397.4097307692308,
          367.3947692307692,
          402.36110256410257
        ],
        "latency_us": [
          612.9590000000001,
          509.20799999999997,
          509.20799999999997,
          510.459,
          2751.7079999999996,
          718.7909999999999,
          599.292,
          499.29200000000003,
          581.542
        ],
        "throughput_us_p50": 397.4097307692308,
        "throughput_us_p90": 402.9022487179487,
        "latency_us_p50": 581.542,
        "latency_us_p90": 1125.3744000000002,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 21976458.374410868,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22432000.027038157,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 24229749.978985637,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 24209624.97126311,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22178583.312779665,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 24405375.006608665,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21802000.00293553,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 24177750.04869327,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 23202416.66166112,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 272041.6523516178,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 286041.7007468641,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 279125.0008158386,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 271875.00381842256,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 271375.00001117587,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 282499.99741092324,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 271250.02816319466,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 683041.6969023645,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 275875.03427639604,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              343.3821621001698,
              350.5000004224712,
              378.5898434216506,
              378.2753901759861,
              346.54036426218227,
              381.3339844782604,
              340.65625004586764,
              377.77734451083234,
              362.537760338455
            ],
            "command_buffer_throughput_us_p50": 362.537760338455,
            "command_buffer_throughput_us_p90": 379.13867163297255,
            "command_buffer_latency_us": [
              272.0416523516178,
              286.0417007468641,
              279.1250008158386,
              271.87500381842256,
              271.37500001117587,
              282.49999741092324,
              271.25002816319466,
              683.0416969023645,
              275.87503427639604
            ],
            "command_buffer_latency_us_p50": 275.87503427639604,
            "command_buffer_latency_us_p90": 365.4416999779642
          },
          "throughput": [
            {
              "compute_ns": 22117958.0,
              "compute_span_ns": 22117958.0,
              "command_buffer_ns": 22117916.66923091,
              "calibration_cpu_ns": 23761000.0,
              "calibration_gpu_ticks": 23761000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 24297375.0,
              "compute_span_ns": 24297375.0,
              "command_buffer_ns": 24297374.999150634,
              "calibration_cpu_ns": 25613917.0,
              "calibration_gpu_ticks": 25613917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 24386750.0,
              "compute_span_ns": 24386750.0,
              "command_buffer_ns": 24386750.01030788,
              "calibration_cpu_ns": 26445250.0,
              "calibration_gpu_ticks": 26445250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 24504708.0,
              "compute_span_ns": 24504708.0,
              "command_buffer_ns": 24504708.35901797,
              "calibration_cpu_ns": 25599167.0,
              "calibration_gpu_ticks": 25599167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 24283417.0,
              "compute_span_ns": 24283417.0,
              "command_buffer_ns": 24283416.685648263,
              "calibration_cpu_ns": 25900125.0,
              "calibration_gpu_ticks": 25900125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 24356458.0,
              "compute_span_ns": 24356458.0,
              "command_buffer_ns": 24356416.65942967,
              "calibration_cpu_ns": 26624167.0,
              "calibration_gpu_ticks": 26624167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21767583.0,
              "compute_span_ns": 21767583.0,
              "command_buffer_ns": 21767541.649751365,
              "calibration_cpu_ns": 22834584.0,
              "calibration_gpu_ticks": 22834584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 22122916.0,
              "compute_span_ns": 22122916.0,
              "command_buffer_ns": 22122874.9724105,
              "calibration_cpu_ns": 24104333.0,
              "calibration_gpu_ticks": 24104333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21995709.0,
              "compute_span_ns": 21995709.0,
              "command_buffer_ns": 21995749.964844435,
              "calibration_cpu_ns": 24189583.0,
              "calibration_gpu_ticks": 24189583.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 270375.0,
              "compute_span_ns": 270375.0,
              "command_buffer_ns": 270375.0506043434,
              "calibration_cpu_ns": 702458.0,
              "calibration_gpu_ticks": 702458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 284167.0,
              "compute_span_ns": 284167.0,
              "command_buffer_ns": 284166.65736585855,
              "calibration_cpu_ns": 1912958.0,
              "calibration_gpu_ticks": 1912958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 274250.0,
              "compute_span_ns": 274250.0,
              "command_buffer_ns": 274249.99279901385,
              "calibration_cpu_ns": 835209.0,
              "calibration_gpu_ticks": 835209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 282125.0,
              "compute_span_ns": 282125.0,
              "command_buffer_ns": 282125.0236593187,
              "calibration_cpu_ns": 1185709.0,
              "calibration_gpu_ticks": 1185709.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 287375.0,
              "compute_span_ns": 287375.0,
              "command_buffer_ns": 287375.00542774796,
              "calibration_cpu_ns": 966958.0,
              "calibration_gpu_ticks": 966958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 271375.0,
              "compute_span_ns": 271375.0,
              "command_buffer_ns": 271375.00001117587,
              "calibration_cpu_ns": 730125.0,
              "calibration_gpu_ticks": 730125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 284167.0,
              "compute_span_ns": 284167.0,
              "command_buffer_ns": 284208.3340510726,
              "calibration_cpu_ns": 1041000.0,
              "calibration_gpu_ticks": 1041000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 270958.0,
              "compute_span_ns": 270958.0,
              "command_buffer_ns": 270916.6728891432,
              "calibration_cpu_ns": 717084.0,
              "calibration_gpu_ticks": 717084.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 271666.0,
              "compute_span_ns": 271666.0,
              "command_buffer_ns": 271625.0019147992,
              "calibration_cpu_ns": 738333.0,
              "calibration_gpu_ticks": 738333.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            345.59309375,
            379.646484375,
            381.04296875,
            382.8860625,
            379.428390625,
            380.56965625,
            340.118484375,
            345.6705625,
            343.682953125
          ],
          "compute_throughput_us_p50": 379.428390625,
          "compute_throughput_us_p90": 381.4115875,
          "command_buffer_throughput_us": [
            345.59244795673294,
            379.64648436172865,
            381.0429689110606,
            382.88606810965575,
            379.4283857132541,
            380.5690103035886,
            340.1178382773651,
            345.66992144391406,
            343.6835932006943
          ],
          "command_buffer_throughput_us_p50": 379.4283857132541,
          "command_buffer_throughput_us_p90": 381.41158875077963,
          "compute_latency_us": [
            270.375,
            284.167,
            274.25,
            282.125,
            287.375,
            271.375,
            284.167,
            270.958,
            271.666
          ],
          "compute_latency_us_p50": 274.25,
          "compute_latency_us_p90": 284.80859999999996,
          "command_buffer_latency_us": [
            270.3750506043434,
            284.16665736585855,
            274.24999279901385,
            282.1250236593187,
            287.37500542774796,
            271.37500001117587,
            284.2083340510726,
            270.91667288914323,
            271.6250019147992
          ],
          "command_buffer_latency_us_p50": 274.24999279901385,
          "command_buffer_latency_us_p90": 284.84166832640767,
          "counter_control_throughput_ratio": 1.006436810354518,
          "counter_control_latency_ratio": 0.993444860046246
        },
        "download_ms": 0.707875,
        "allocation_upload_ms": 17.865875,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.046016576746524e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.27488894658554774,
      "device_compute_slowdown": 0.21770515245033265,
      "valid": true,
      "round": 1,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "layernorm_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.773166,
        "warmup_ms": 201.176834,
        "repetitions": 73,
        "throughput_us": [
          374.0285342465753,
          409.96690410958905,
          375.246,
          392.5428082191781,
          380.96690410958905,
          373.1095890410959,
          413.49087671232877,
          376.63298630136984,
          389.5473698630137
        ],
        "latency_us": [
          1454.625,
          596.4580000000001,
          704.125,
          552.333,
          544.792,
          511.375,
          500.54200000000003,
          495.70799999999997,
          496.875
        ],
        "throughput_us_p50": 380.96690410958905,
        "throughput_us_p90": 410.671698630137,
        "latency_us_p50": 544.792,
        "latency_us_p90": 854.2250000000001,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 22280125.005636364,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22147374.984342605,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22165958.303958178,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 24316374.969203025,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 24352291.715331376,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 22080624.99994412,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21863250.00366196,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 24247791.61810875,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 23893625.009804964,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 277499.9593384564,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 270874.9962039292,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 605958.339292556,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 270999.9680519104,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 738791.6557490826,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 672374.9684169888,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 278749.9688565731,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 274166.63942858577,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 270833.31951871514,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              348.1269532130682,
              346.0527341303532,
              346.3430984993465,
              379.94335889379727,
              380.50455805205274,
              345.0097656241269,
              341.61328130721813,
              378.8717440329492,
              373.33789077820256
            ],
            "command_buffer_throughput_us_p50": 348.1269532130682,
            "command_buffer_throughput_us_p90": 380.05559872544836,
            "command_buffer_latency_us": [
              277.4999593384564,
              270.8749962039292,
              605.958339292556,
              270.9999680519104,
              738.7916557490826,
              672.3749684169888,
              278.7499688565731,
              274.16663942858577,
              270.83331951871514
            ],
            "command_buffer_latency_us_p50": 277.4999593384564,
            "command_buffer_latency_us_p90": 685.6583058834076
          },
          "throughput": [
            {
              "compute_ns": 22071459.0,
              "compute_span_ns": 22071459.0,
              "command_buffer_ns": 22071500.01777336,
              "calibration_cpu_ns": 23560833.0,
              "calibration_gpu_ticks": 23560833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21984917.0,
              "compute_span_ns": 21984917.0,
              "command_buffer_ns": 21984916.646033525,
              "calibration_cpu_ns": 23849708.0,
              "calibration_gpu_ticks": 23849708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 22475333.0,
              "compute_span_ns": 22475333.0,
              "command_buffer_ns": 22475333.360489458,
              "calibration_cpu_ns": 25543209.0,
              "calibration_gpu_ticks": 25543209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 24280500.0,
              "compute_span_ns": 24280500.0,
              "command_buffer_ns": 24280500.01617521,
              "calibration_cpu_ns": 26630625.0,
              "calibration_gpu_ticks": 26630625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 24278292.0,
              "compute_span_ns": 24278292.0,
              "command_buffer_ns": 24278333.35241303,
              "calibration_cpu_ns": 28388792.0,
              "calibration_gpu_ticks": 28388792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 24236041.0,
              "compute_span_ns": 24236041.0,
              "command_buffer_ns": 24235999.96836856,
              "calibration_cpu_ns": 25308792.0,
              "calibration_gpu_ticks": 25308792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 22075209.0,
              "compute_span_ns": 22075209.0,
              "command_buffer_ns": 22075249.98812005,
              "calibration_cpu_ns": 24798125.0,
              "calibration_gpu_ticks": 24798125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 20084959.0,
              "compute_span_ns": 20084959.0,
              "command_buffer_ns": 20084999.967366457,
              "calibration_cpu_ns": 24915083.0,
              "calibration_gpu_ticks": 24915083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 24124208.0,
              "compute_span_ns": 24124208.0,
              "command_buffer_ns": 24124166.63719341,
              "calibration_cpu_ns": 25867792.0,
              "calibration_gpu_ticks": 25867792.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 643750.0,
              "compute_span_ns": 643750.0,
              "command_buffer_ns": 643750.0123865902,
              "calibration_cpu_ns": 1828042.0,
              "calibration_gpu_ticks": 1828042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 276208.0,
              "compute_span_ns": 276208.0,
              "command_buffer_ns": 276166.6546575725,
              "calibration_cpu_ns": 751459.0,
              "calibration_gpu_ticks": 751459.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 276500.0,
              "compute_span_ns": 276500.0,
              "command_buffer_ns": 276500.00993162394,
              "calibration_cpu_ns": 705500.0,
              "calibration_gpu_ticks": 705500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 272458.0,
              "compute_span_ns": 272458.0,
              "command_buffer_ns": 272416.62610322237,
              "calibration_cpu_ns": 736458.0,
              "calibration_gpu_ticks": 736458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 275958.0,
              "compute_span_ns": 275958.0,
              "command_buffer_ns": 275916.7109616101,
              "calibration_cpu_ns": 885041.0,
              "calibration_gpu_ticks": 885041.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 271042.0,
              "compute_span_ns": 271042.0,
              "command_buffer_ns": 271041.70294478536,
              "calibration_cpu_ns": 789792.0,
              "calibration_gpu_ticks": 789792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 277500.0,
              "compute_span_ns": 277500.0,
              "command_buffer_ns": 277500.0175461173,
              "calibration_cpu_ns": 1068333.0,
              "calibration_gpu_ticks": 1068333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 577792.0,
              "compute_span_ns": 577792.0,
              "command_buffer_ns": 577833.3288617432,
              "calibration_cpu_ns": 2189625.0,
              "calibration_gpu_ticks": 2189625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 952416.0,
              "compute_span_ns": 952416.0,
              "command_buffer_ns": 952375.0049993396,
              "calibration_cpu_ns": 2216500.0,
              "calibration_gpu_ticks": 2216500.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            344.866546875,
            343.514328125,
            351.177078125,
            379.3828125,
            379.3483125,
            378.688140625,
            344.925140625,
            313.827484375,
            376.94075
          ],
          "compute_throughput_us_p50": 351.177078125,
          "compute_throughput_us_p90": 379.3552125,
          "command_buffer_throughput_us": [
            344.86718777770875,
            343.51432259427384,
            351.1770837576478,
            379.38281275273766,
            379.34895863145357,
            378.68749950575875,
            344.92578106437577,
            313.8281244901009,
            376.94010370614706
          ],
          "command_buffer_throughput_us_p50": 351.1770837576478,
          "command_buffer_throughput_us_p90": 379.3557294557104,
          "compute_latency_us": [
            643.75,
            276.208,
            276.5,
            272.458,
            275.958,
            271.042,
            277.5,
            577.792,
            952.416
          ],
          "compute_latency_us_p50": 276.5,
          "compute_latency_us_p90": 705.4832000000001,
          "command_buffer_latency_us": [
            643.7500123865902,
            276.1666546575725,
            276.50000993162394,
            272.41662610322237,
            275.9167109616101,
            271.04170294478536,
            277.5000175461173,
            577.8333288617432,
            952.3750049993396
          ],
          "command_buffer_latency_us_p50": 276.50000993162394,
          "command_buffer_latency_us_p90": 705.4750109091401,
          "counter_control_throughput_ratio": 0.9985246586683562,
          "counter_control_latency_ratio": 1.0052275211008166
        },
        "download_ms": 0.968125,
        "allocation_upload_ms": 2.186333,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.046016576746524e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-rzpa3liq/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "4f0c4e70dcb06b507aa82e242c8819acd3628d5f79a065eb6c2f6b73ec6f396c",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.97,
        "capture_ms": 0.071958,
        "compile_ms": 63.756416,
        "allocation_upload_ms": 6.57175,
        "cold_call_ms": 4.841958,
        "warmup_ms": 203.134375,
        "download_ms": 3.253,
        "repetitions": 284,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          107.049880282,
          110.977257042,
          111.626908451,
          102.221098592,
          112.183242958,
          112.481514085,
          113.186327465,
          103.624855634,
          94.518193662
        ],
        "latency_us": [
          341.917,
          414.709,
          354.583,
          344.791,
          354.667,
          326.083,
          319.584,
          333.458,
          2937.709
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 768,
            "reduction_scalar_rounds": 120,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 73728,
              "global_write_bytes": 24576,
              "private_read_bytes": 147456,
              "private_write_bytes": 49152
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 288,
              "global_write_bytes": 96,
              "private_read_bytes": 576,
              "private_write_bytes": 192
            },
            "striped_storage_scalars_per_worker": 48,
            "reduction_operations": 2,
            "reduction_elements": 12288,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 168,
            "normalized_kernel_cost": 2016,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 18432,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 7298708,
              "compute_span_ns": 7298708,
              "calibration_cpu_ns": 8075292,
              "calibration_gpu_ticks": 8075292,
              "compute_passes": 1,
              "command_buffer_ns": 7298708.36949,
              "command_buffers": 2
            },
            {
              "compute_ns": 7313375,
              "compute_span_ns": 7313375,
              "calibration_cpu_ns": 8049792,
              "calibration_gpu_ticks": 8049792,
              "compute_passes": 1,
              "command_buffer_ns": 7313375.01202,
              "command_buffers": 2
            },
            {
              "compute_ns": 7271833,
              "compute_span_ns": 7271833,
              "calibration_cpu_ns": 8016500,
              "calibration_gpu_ticks": 8016500,
              "compute_passes": 1,
              "command_buffer_ns": 7271874.98705,
              "command_buffers": 2
            },
            {
              "compute_ns": 7323708,
              "compute_span_ns": 7323708,
              "calibration_cpu_ns": 8153166,
              "calibration_gpu_ticks": 8153166,
              "compute_passes": 1,
              "command_buffer_ns": 7323708.38523,
              "command_buffers": 2
            },
            {
              "compute_ns": 7281292,
              "compute_span_ns": 7281292,
              "calibration_cpu_ns": 8040459,
              "calibration_gpu_ticks": 8040459,
              "compute_passes": 1,
              "command_buffer_ns": 7281416.67787,
              "command_buffers": 2
            },
            {
              "compute_ns": 7338333,
              "compute_span_ns": 7338333,
              "calibration_cpu_ns": 8177791,
              "calibration_gpu_ticks": 8177791,
              "compute_passes": 1,
              "command_buffer_ns": 7338374.91134,
              "command_buffers": 2
            },
            {
              "compute_ns": 7366500,
              "compute_span_ns": 7366500,
              "calibration_cpu_ns": 8099042,
              "calibration_gpu_ticks": 8099042,
              "compute_passes": 1,
              "command_buffer_ns": 7366541.65667,
              "command_buffers": 2
            },
            {
              "compute_ns": 7317750,
              "compute_span_ns": 7317750,
              "calibration_cpu_ns": 8166041,
              "calibration_gpu_ticks": 8166041,
              "compute_passes": 1,
              "command_buffer_ns": 7317833.31139,
              "command_buffers": 2
            },
            {
              "compute_ns": 7490667,
              "compute_span_ns": 7490667,
              "calibration_cpu_ns": 8610042,
              "calibration_gpu_ticks": 8610042,
              "compute_passes": 1,
              "command_buffer_ns": 7490833.28992,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 76334,
              "compute_span_ns": 76334,
              "calibration_cpu_ns": 819917,
              "calibration_gpu_ticks": 819917,
              "compute_passes": 1,
              "command_buffer_ns": 76458.3819546,
              "command_buffers": 2
            },
            {
              "compute_ns": 119042,
              "compute_span_ns": 119042,
              "calibration_cpu_ns": 1001500,
              "calibration_gpu_ticks": 1001500,
              "compute_passes": 1,
              "command_buffer_ns": 119208.300021,
              "command_buffers": 2
            },
            {
              "compute_ns": 89875,
              "compute_span_ns": 89875,
              "calibration_cpu_ns": 499833,
              "calibration_gpu_ticks": 499833,
              "compute_passes": 1,
              "command_buffer_ns": 89958.3101273,
              "command_buffers": 2
            },
            {
              "compute_ns": 122750,
              "compute_span_ns": 122750,
              "calibration_cpu_ns": 1150000,
              "calibration_gpu_ticks": 1150000,
              "compute_passes": 1,
              "command_buffer_ns": 122833.298519,
              "command_buffers": 2
            },
            {
              "compute_ns": 118458,
              "compute_span_ns": 118458,
              "calibration_cpu_ns": 1206125,
              "calibration_gpu_ticks": 1206125,
              "compute_passes": 1,
              "command_buffer_ns": 118416.675832,
              "command_buffers": 2
            },
            {
              "compute_ns": 77375,
              "compute_span_ns": 77375,
              "calibration_cpu_ns": 526917,
              "calibration_gpu_ticks": 526917,
              "compute_passes": 1,
              "command_buffer_ns": 77375.0361986,
              "command_buffers": 2
            },
            {
              "compute_ns": 116084,
              "compute_span_ns": 116084,
              "calibration_cpu_ns": 696125,
              "calibration_gpu_ticks": 696125,
              "compute_passes": 1,
              "command_buffer_ns": 116166.600492,
              "command_buffers": 2
            },
            {
              "compute_ns": 91958,
              "compute_span_ns": 91958,
              "calibration_cpu_ns": 479917,
              "calibration_gpu_ticks": 479917,
              "compute_passes": 1,
              "command_buffer_ns": 91958.3253562,
              "command_buffers": 2
            },
            {
              "compute_ns": 116250,
              "compute_span_ns": 116250,
              "calibration_cpu_ns": 521000,
              "calibration_gpu_ticks": 521000,
              "compute_passes": 1,
              "command_buffer_ns": 116333.307233,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 7799916.60966,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7413500.04682,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7282708.36407,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7418999.97229,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7307541.61486,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7382416.63203,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7261375.02352,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7285374.97343,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7269291.67286,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 124416.663311,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 85375.0389069,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 121500.052046,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 120583.339594,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76250.0567362,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 78458.3389759,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 352500.006557,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76708.3256505,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 421291.682869,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              121.87369702593749,
              115.8359382315625,
              113.79231818859375,
              115.92187456703125,
              114.1803377321875,
              115.35025987546875,
              113.45898474250001,
              113.83398395984375,
              113.5826823884375
            ],
            "command_buffer_throughput_us_p50": 114.1803377321875,
            "command_buffer_throughput_us_p90": 117.1122390588125,
            "command_buffer_latency_us": [
              124.416663311,
              85.37503890689999,
              121.500052046,
              120.58333959400001,
              76.2500567362,
              78.45833897589999,
              352.500006557,
              76.7083256505,
              421.291682869
            ],
            "command_buffer_latency_us_p50": 120.58333959400001,
            "command_buffer_latency_us_p90": 366.2583418194
          },
          "compute_throughput_us": [
            114.0423125,
            114.271484375,
            113.622390625,
            114.4329375,
            113.7701875,
            114.661453125,
            115.1015625,
            114.33984375,
            117.041671875
          ],
          "compute_throughput_us_p50": 114.33984375,
          "compute_throughput_us_p90": 115.489584375,
          "command_buffer_throughput_us": [
            114.04231827328125,
            114.2714845628125,
            113.62304667265624,
            114.43294351921875,
            113.77213559171875,
            114.6621079896875,
            115.10221338546876,
            114.34114549046876,
            117.044270155
          ],
          "command_buffer_throughput_us_p50": 114.34114549046876,
          "command_buffer_throughput_us_p90": 115.49062473937501,
          "compute_latency_us": [
            76.334,
            119.042,
            89.875,
            122.75,
            118.458,
            77.375,
            116.084,
            91.958,
            116.25
          ],
          "compute_latency_us_p50": 116.084,
          "compute_latency_us_p90": 119.7836,
          "command_buffer_latency_us": [
            76.4583819546,
            119.208300021,
            89.95831012730001,
            122.83329851900001,
            118.416675832,
            77.3750361986,
            116.166600492,
            91.9583253562,
            116.333307233
          ],
          "command_buffer_latency_us_p50": 116.166600492,
          "command_buffer_latency_us_p90": 119.9332997206,
          "counter_control_throughput_ratio": 0.9964249349005588,
          "counter_control_latency_ratio": 0.9861926368638425
        },
        "correctness": {
          "max_abs_error": 2.7199924934606656e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 110.977257042,
        "throughput_us_p90": 112.622476761,
        "latency_us_p50": 344.791,
        "latency_us_p90": 919.3090000000004
      },
      "slowdown": 0.2913041942616523,
      "device_compute_slowdown": 0.3255902815766957,
      "valid": true,
      "round": 1,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "layernorm_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-yioj1q21/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "6caff4b0a979c3104fea9f23f0d5430392f6f4e7d743dab03e9fdb6de1428bf3",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.65,
        "capture_ms": 0.082,
        "compile_ms": 64.52825,
        "allocation_upload_ms": 5.922541,
        "cold_call_ms": 6.639167,
        "warmup_ms": 200.578708,
        "download_ms": 3.171291,
        "repetitions": 467,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          90.2193062099,
          90.4642226981,
          90.7099400428,
          91.182012848,
          97.9660963597,
          92.4304068522,
          80.0113319058,
          92.2483940043,
          80.1978051392
        ],
        "latency_us": [
          423.625,
          397.709,
          381,
          2401.291,
          600.042,
          710.75,
          409.792,
          350.542,
          340.833
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 1024,
            "reduction_scalar_rounds": 80,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 49152,
              "global_write_bytes": 16384,
              "private_read_bytes": 98304,
              "private_write_bytes": 32768
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 192,
              "global_write_bytes": 64,
              "private_read_bytes": 384,
              "private_write_bytes": 128
            },
            "striped_storage_scalars_per_worker": 32,
            "reduction_operations": 2,
            "reduction_elements": 8192,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 128,
            "normalized_kernel_cost": 2048,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3922583,
              "compute_span_ns": 3922583,
              "calibration_cpu_ns": 4621000,
              "calibration_gpu_ticks": 4621000,
              "compute_passes": 1,
              "command_buffer_ns": 3922666.71406,
              "command_buffers": 2
            },
            {
              "compute_ns": 4527625,
              "compute_span_ns": 4527625,
              "calibration_cpu_ns": 5474500,
              "calibration_gpu_ticks": 5474500,
              "compute_passes": 1,
              "command_buffer_ns": 4527624.98753,
              "command_buffers": 2
            },
            {
              "compute_ns": 4666625,
              "compute_span_ns": 4666625,
              "calibration_cpu_ns": 5453666,
              "calibration_gpu_ticks": 5453666,
              "compute_passes": 1,
              "command_buffer_ns": 4666708.35158,
              "command_buffers": 2
            },
            {
              "compute_ns": 4699834,
              "compute_span_ns": 4699834,
              "calibration_cpu_ns": 5665417,
              "calibration_gpu_ticks": 5665417,
              "compute_passes": 1,
              "command_buffer_ns": 4699916.69524,
              "command_buffers": 2
            },
            {
              "compute_ns": 4596583,
              "compute_span_ns": 4596583,
              "calibration_cpu_ns": 5676458,
              "calibration_gpu_ticks": 5676458,
              "compute_passes": 1,
              "command_buffer_ns": 4596666.72396,
              "command_buffers": 2
            },
            {
              "compute_ns": 4405541,
              "compute_span_ns": 4405541,
              "calibration_cpu_ns": 5451208,
              "calibration_gpu_ticks": 5451208,
              "compute_passes": 1,
              "command_buffer_ns": 4405500.01804,
              "command_buffers": 2
            },
            {
              "compute_ns": 4710042,
              "compute_span_ns": 4710042,
              "calibration_cpu_ns": 5316459,
              "calibration_gpu_ticks": 5316459,
              "compute_passes": 1,
              "command_buffer_ns": 4710083.36172,
              "command_buffers": 2
            },
            {
              "compute_ns": 4601584,
              "compute_span_ns": 4601584,
              "calibration_cpu_ns": 5525083,
              "calibration_gpu_ticks": 5525083,
              "compute_passes": 1,
              "command_buffer_ns": 4601708.3223,
              "command_buffers": 2
            },
            {
              "compute_ns": 4684375,
              "compute_span_ns": 4684375,
              "calibration_cpu_ns": 6070625,
              "calibration_gpu_ticks": 6070625,
              "compute_passes": 1,
              "command_buffer_ns": 4684458.31211,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 95167,
              "compute_span_ns": 95167,
              "calibration_cpu_ns": 752625,
              "calibration_gpu_ticks": 752625,
              "compute_passes": 1,
              "command_buffer_ns": 95250.0267886,
              "command_buffers": 2
            },
            {
              "compute_ns": 106875,
              "compute_span_ns": 106875,
              "calibration_cpu_ns": 1122500,
              "calibration_gpu_ticks": 1122500,
              "compute_passes": 1,
              "command_buffer_ns": 106874.969788,
              "command_buffers": 2
            },
            {
              "compute_ns": 102584,
              "compute_span_ns": 102584,
              "calibration_cpu_ns": 566417,
              "calibration_gpu_ticks": 566417,
              "compute_passes": 1,
              "command_buffer_ns": 102666.672319,
              "command_buffers": 2
            },
            {
              "compute_ns": 63583,
              "compute_span_ns": 63583,
              "calibration_cpu_ns": 508916,
              "calibration_gpu_ticks": 508916,
              "compute_passes": 1,
              "command_buffer_ns": 63624.989707,
              "command_buffers": 2
            },
            {
              "compute_ns": 386166,
              "compute_span_ns": 386166,
              "calibration_cpu_ns": 1744542,
              "calibration_gpu_ticks": 1744542,
              "compute_passes": 1,
              "command_buffer_ns": 386166.677345,
              "command_buffers": 2
            },
            {
              "compute_ns": 88375,
              "compute_span_ns": 88375,
              "calibration_cpu_ns": 513875,
              "calibration_gpu_ticks": 513875,
              "compute_passes": 1,
              "command_buffer_ns": 88458.3569132,
              "command_buffers": 2
            },
            {
              "compute_ns": 425459,
              "compute_span_ns": 425459,
              "calibration_cpu_ns": 1284333,
              "calibration_gpu_ticks": 1284333,
              "compute_passes": 1,
              "command_buffer_ns": 425625.010394,
              "command_buffers": 2
            },
            {
              "compute_ns": 101167,
              "compute_span_ns": 101167,
              "calibration_cpu_ns": 612083,
              "calibration_gpu_ticks": 612083,
              "compute_passes": 1,
              "command_buffer_ns": 101249.95606,
              "command_buffers": 2
            },
            {
              "compute_ns": 567083,
              "compute_span_ns": 567083,
              "calibration_cpu_ns": 1357167,
              "calibration_gpu_ticks": 1357167,
              "compute_passes": 1,
              "command_buffer_ns": 567166.658584,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4373416.65383,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4680208.33796,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4621333.32621,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4686541.68071,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4677749.99561,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4553791.65942,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4488708.27677,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4802250.01229,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4057333.2808,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 172375.002876,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 105583.283585,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 511791.615281,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63958.2867734,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 64583.3206363,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 66375.0106469,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63416.6644886,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63250.0159554,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63541.6945443,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              68.33463521609376,
              73.128255280625,
              72.20833322203124,
              73.22721376109375,
              73.08984368140625,
              71.15299467843751,
              70.13606682453126,
              75.03515644203125,
              63.3958325125
            ],
            "command_buffer_throughput_us_p50": 72.20833322203124,
            "command_buffer_throughput_us_p90": 73.58880229728125,
            "command_buffer_latency_us": [
              172.37500287600002,
              105.583283585,
              511.79161528099996,
              63.958286773400005,
              64.5833206363,
              66.37501064690001,
              63.4166644886,
              63.250015955399995,
              63.5416945443
            ],
            "command_buffer_latency_us_p50": 64.5833206363,
            "command_buffer_latency_us_p90": 240.25832535700005
          },
          "compute_throughput_us": [
            61.290359375,
            70.744140625,
            72.916015625,
            73.43490625,
            71.821609375,
            68.836578125,
            73.59440625,
            71.89975,
            73.193359375
          ],
          "compute_throughput_us_p50": 71.89975,
          "compute_throughput_us_p90": 73.46680625,
          "command_buffer_throughput_us": [
            61.291667407187504,
            70.74414043015625,
            72.91731799343749,
            73.436198363125,
            71.822917561875,
            68.835937781875,
            73.59505252687501,
            71.9016925359375,
            73.19466112671876
          ],
          "command_buffer_throughput_us_p50": 71.9016925359375,
          "command_buffer_throughput_us_p90": 73.467969195875,
          "compute_latency_us": [
            95.167,
            106.875,
            102.584,
            63.583,
            386.166,
            88.375,
            425.459,
            101.167,
            567.083
          ],
          "compute_latency_us_p50": 102.584,
          "compute_latency_us_p90": 453.78380000000004,
          "command_buffer_latency_us": [
            95.2500267886,
            106.874969788,
            102.666672319,
            63.624989707,
            386.166677345,
            88.45835691319999,
            425.62501039399996,
            101.24995606,
            567.166658584
          ],
          "command_buffer_latency_us_p50": 102.666672319,
          "command_buffer_latency_us_p90": 453.933340032,
          "counter_control_throughput_ratio": 0.9826661810216247,
          "counter_control_latency_ratio": 1.3327057284220771
        },
        "correctness": {
          "max_abs_error": 2.827305296548843e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 90.7099400428,
        "throughput_us_p90": 93.53754475369999,
        "latency_us_p50": 409.792,
        "latency_us_p90": 1048.8582000000004
      },
      "torch": {
        "cold_call_ms": 6.535917,
        "warmup_ms": 201.029667,
        "repetitions": 130,
        "throughput_us": [
          286.8666692307692,
          287.94807692307694,
          284.02819999999997,
          280.3310923076923,
          294.86474615384617,
          287.04776153846154,
          289.3285230769231,
          301.38237692307695,
          280.3884615384615
        ],
        "latency_us": [
          712.4580000000001,
          544.5,
          530.083,
          636.875,
          2786.5829999999996,
          1370.9579999999999,
          677.333,
          572.25,
          565.8330000000001
        ],
        "throughput_us_p50": 287.04776153846154,
        "throughput_us_p90": 296.16827230769235,
        "latency_us_p50": 636.875,
        "latency_us_p90": 1654.083,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 17503499.98474121,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 15165249.991696328,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17701750.039123,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17457624.955568463,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17493124.993052334,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 18403541.70417413,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17938749.98111278,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17510666.68657586,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17885499.983094633,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 198625.02813339233,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200500.01330673695,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 198625.02813339233,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 199625.0357478857,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200750.0152103603,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 199250.00378862023,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200625.04336237907,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 203666.6264757514,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200624.98515471816,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              273.4921872615814,
              236.95703112025512,
              276.58984436129685,
              272.7753899307572,
              273.3300780164427,
              287.5553391277208,
              280.29296845488716,
              273.6041669777478,
              279.46093723585363
            ],
            "command_buffer_throughput_us_p50": 273.6041669777478,
            "command_buffer_throughput_us_p90": 281.7454425894539,
            "command_buffer_latency_us": [
              198.62502813339233,
              200.50001330673695,
              198.62502813339233,
              199.6250357478857,
              200.7500152103603,
              199.25000378862023,
              200.62504336237907,
              203.6666264757514,
              200.62498515471816
            ],
            "command_buffer_latency_us_p50": 200.50001330673695,
            "command_buffer_latency_us_p90": 201.3333374634385
          },
          "throughput": [
            {
              "compute_ns": 17874458.0,
              "compute_span_ns": 17874458.0,
              "command_buffer_ns": 17874458.339065313,
              "calibration_cpu_ns": 20103125.0,
              "calibration_gpu_ticks": 20103125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17485042.0,
              "compute_span_ns": 17485042.0,
              "command_buffer_ns": 17485041.63697362,
              "calibration_cpu_ns": 18661000.0,
              "calibration_gpu_ticks": 18661000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17597833.0,
              "compute_span_ns": 17597833.0,
              "command_buffer_ns": 17597833.299078047,
              "calibration_cpu_ns": 20409292.0,
              "calibration_gpu_ticks": 20409292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17512958.0,
              "compute_span_ns": 17512958.0,
              "command_buffer_ns": 17512916.64550081,
              "calibration_cpu_ns": 20513584.0,
              "calibration_gpu_ticks": 20513584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17463209.0,
              "compute_span_ns": 17463209.0,
              "command_buffer_ns": 17463249.969296157,
              "calibration_cpu_ns": 19494417.0,
              "calibration_gpu_ticks": 19494417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17513958.0,
              "compute_span_ns": 17513958.0,
              "command_buffer_ns": 17513958.329800516,
              "calibration_cpu_ns": 19635375.0,
              "calibration_gpu_ticks": 19635375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17497458.0,
              "compute_span_ns": 17497458.0,
              "command_buffer_ns": 17497416.643891484,
              "calibration_cpu_ns": 18661333.0,
              "calibration_gpu_ticks": 18661333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17501875.0,
              "compute_span_ns": 17501875.0,
              "command_buffer_ns": 17501875.00147149,
              "calibration_cpu_ns": 18454083.0,
              "calibration_gpu_ticks": 18454083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17546500.0,
              "compute_span_ns": 17546500.0,
              "command_buffer_ns": 17546499.96291846,
              "calibration_cpu_ns": 19248167.0,
              "calibration_gpu_ticks": 19248167.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 201208.0,
              "compute_span_ns": 201208.0,
              "command_buffer_ns": 201166.6656471789,
              "calibration_cpu_ns": 893542.0,
              "calibration_gpu_ticks": 893542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 198625.0,
              "compute_span_ns": 198625.0,
              "command_buffer_ns": 198625.02813339233,
              "calibration_cpu_ns": 945500.0,
              "calibration_gpu_ticks": 945500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 199625.0,
              "compute_span_ns": 199625.0,
              "command_buffer_ns": 199625.0357478857,
              "calibration_cpu_ns": 794667.0,
              "calibration_gpu_ticks": 794667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 199000.0,
              "compute_span_ns": 199000.0,
              "command_buffer_ns": 199000.0018849969,
              "calibration_cpu_ns": 1053208.0,
              "calibration_gpu_ticks": 1053208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 202625.0,
              "compute_span_ns": 202625.0,
              "command_buffer_ns": 202625.0003837049,
              "calibration_cpu_ns": 714083.0,
              "calibration_gpu_ticks": 714083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 200584.0,
              "compute_span_ns": 200584.0,
              "command_buffer_ns": 200624.98515471816,
              "calibration_cpu_ns": 620250.0,
              "calibration_gpu_ticks": 620250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201875.0,
              "compute_span_ns": 201875.0,
              "command_buffer_ns": 201874.99467283487,
              "calibration_cpu_ns": 665917.0,
              "calibration_gpu_ticks": 665917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 200041.0,
              "compute_span_ns": 200041.0,
              "command_buffer_ns": 200000.00949949026,
              "calibration_cpu_ns": 804084.0,
              "calibration_gpu_ticks": 804084.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 199333.0,
              "compute_span_ns": 199333.0,
              "command_buffer_ns": 199333.35715904832,
              "calibration_cpu_ns": 706542.0,
              "calibration_gpu_ticks": 706542.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            279.28840625,
            273.20378125,
            274.966140625,
            273.63996875,
            272.862640625,
            273.65559375,
            273.39778125,
            273.466796875,
            274.1640625
          ],
          "compute_throughput_us_p50": 273.63996875,
          "compute_throughput_us_p90": 275.83059375,
          "command_buffer_throughput_us": [
            279.2884115478955,
            273.2037755777128,
            274.9661452980945,
            273.63932258595014,
            272.86328077025246,
            273.65559890313307,
            273.39713506080443,
            273.466796897992,
            274.16406192060094
          ],
          "command_buffer_throughput_us_p50": 273.63932258595014,
          "command_buffer_throughput_us_p90": 275.8305985480547,
          "compute_latency_us": [
            201.208,
            198.625,
            199.625,
            199.0,
            202.625,
            200.584,
            201.875,
            200.041,
            199.333
          ],
          "compute_latency_us_p50": 200.041,
          "compute_latency_us_p90": 202.025,
          "command_buffer_latency_us": [
            201.1666656471789,
            198.62502813339233,
            199.6250357478857,
            199.0000018849969,
            202.6250003837049,
            200.62498515471816,
            201.87499467283487,
            200.00000949949026,
            199.33335715904832
          ],
          "command_buffer_latency_us_p50": 200.00000949949026,
          "command_buffer_latency_us_p90": 202.02499581500888,
          "counter_control_throughput_ratio": 0.9982921848573059,
          "counter_control_latency_ratio": 1.0050346505870438
        },
        "download_ms": 1.043084,
        "allocation_upload_ms": 18.427333,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.1634934849478213e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.3160099195918857,
      "device_compute_slowdown": 0.26275309973335176,
      "valid": true,
      "round": 1,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "layernorm_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.574916,
        "warmup_ms": 200.799625,
        "repetitions": 105,
        "throughput_us": [
          299.4940476190476,
          275.1972285714286,
          297.4761904761905,
          298.14682857142856,
          274.4920666666667,
          299.5392857142857,
          251.61428571428573,
          298.2059523809524,
          284.66507619047616
        ],
        "latency_us": [
          1592.666,
          571.0,
          477.166,
          511.625,
          423.375,
          445.0,
          431.25,
          445.667,
          427.333
        ],
        "throughput_us_p50": 297.4761904761905,
        "throughput_us_p90": 299.5030952380952,
        "latency_us_p50": 445.667,
        "latency_us_p90": 775.3332000000001,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 17870250.04159659,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17490791.622549295,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17835083.359386772,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17739500.03553182,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17836166.662164032,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17772375.023923814,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17750833.299942315,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17552874.982357025,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17812874.98306483,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 198916.6485145688,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 221416.7034253478,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 199166.70862585306,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200749.95700269938,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200541.689991951,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 199999.95129182935,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 199875.03765150905,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 202249.96842443943,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200875.04526600242,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              279.22265689994674,
              273.29361910233274,
              278.6731774904183,
              277.17968805518467,
              278.690104096313,
              277.6933597488096,
              277.35677031159867,
              274.2636715993285,
              278.326171610388
            ],
            "command_buffer_throughput_us_p50": 277.6933597488096,
            "command_buffer_throughput_us_p90": 278.79661465703975,
            "command_buffer_latency_us": [
              198.9166485145688,
              221.4167034253478,
              199.16670862585306,
              200.74995700269938,
              200.541689991951,
              199.99995129182935,
              199.87503765150905,
              202.24996842443943,
              200.87504526600242
            ],
            "command_buffer_latency_us_p50": 200.541689991951,
            "command_buffer_latency_us_p90": 206.0833154246211
          },
          "throughput": [
            {
              "compute_ns": 17639292.0,
              "compute_span_ns": 17639292.0,
              "command_buffer_ns": 17639291.647356004,
              "calibration_cpu_ns": 18643959.0,
              "calibration_gpu_ticks": 18643959.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17916708.0,
              "compute_span_ns": 17916708.0,
              "command_buffer_ns": 17916708.369739354,
              "calibration_cpu_ns": 19137541.0,
              "calibration_gpu_ticks": 19137541.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17508291.0,
              "compute_span_ns": 17508291.0,
              "command_buffer_ns": 17508249.962702394,
              "calibration_cpu_ns": 18491375.0,
              "calibration_gpu_ticks": 18491375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17805834.0,
              "compute_span_ns": 17805834.0,
              "command_buffer_ns": 17805874.987971038,
              "calibration_cpu_ns": 18997416.0,
              "calibration_gpu_ticks": 18997416.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17595000.0,
              "compute_span_ns": 17595000.0,
              "command_buffer_ns": 17594999.982975423,
              "calibration_cpu_ns": 18713542.0,
              "calibration_gpu_ticks": 18713542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17764875.0,
              "compute_span_ns": 17764875.0,
              "command_buffer_ns": 17764874.966815114,
              "calibration_cpu_ns": 20528042.0,
              "calibration_gpu_ticks": 20528042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17794042.0,
              "compute_span_ns": 17794042.0,
              "command_buffer_ns": 17794041.661545634,
              "calibration_cpu_ns": 18813917.0,
              "calibration_gpu_ticks": 18813917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17620208.0,
              "compute_span_ns": 17620208.0,
              "command_buffer_ns": 17620208.323933184,
              "calibration_cpu_ns": 20600542.0,
              "calibration_gpu_ticks": 20600542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17539791.0,
              "compute_span_ns": 17539791.0,
              "command_buffer_ns": 17539750.027935952,
              "calibration_cpu_ns": 18600375.0,
              "calibration_gpu_ticks": 18600375.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 198833.0,
              "compute_span_ns": 198833.0,
              "command_buffer_ns": 198833.35335180163,
              "calibration_cpu_ns": 855500.0,
              "calibration_gpu_ticks": 855500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201500.0,
              "compute_span_ns": 201500.0,
              "command_buffer_ns": 201499.9627135694,
              "calibration_cpu_ns": 622750.0,
              "calibration_gpu_ticks": 622750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 200167.0,
              "compute_span_ns": 200167.0,
              "command_buffer_ns": 200208.33471789956,
              "calibration_cpu_ns": 874167.0,
              "calibration_gpu_ticks": 874167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 202291.0,
              "compute_span_ns": 202291.0,
              "command_buffer_ns": 202250.02663210034,
              "calibration_cpu_ns": 709750.0,
              "calibration_gpu_ticks": 709750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 200167.0,
              "compute_span_ns": 200167.0,
              "command_buffer_ns": 200208.33471789956,
              "calibration_cpu_ns": 594500.0,
              "calibration_gpu_ticks": 594500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 199792.0,
              "compute_span_ns": 199792.0,
              "command_buffer_ns": 199833.3027586341,
              "calibration_cpu_ns": 608333.0,
              "calibration_gpu_ticks": 608333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 200459.0,
              "compute_span_ns": 200459.0,
              "command_buffer_ns": 200500.01330673695,
              "calibration_cpu_ns": 635458.0,
              "calibration_gpu_ticks": 635458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 199875.0,
              "compute_span_ns": 199875.0,
              "command_buffer_ns": 199874.97944384813,
              "calibration_cpu_ns": 640000.0,
              "calibration_gpu_ticks": 640000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 199875.0,
              "compute_span_ns": 199875.0,
              "command_buffer_ns": 199874.97944384813,
              "calibration_cpu_ns": 683125.0,
              "calibration_gpu_ticks": 683125.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            275.6139375,
            279.9485625,
            273.567046875,
            278.21615625,
            274.921875,
            277.576171875,
            278.03190625,
            275.31575,
            274.059234375
          ],
          "compute_throughput_us_p50": 275.6139375,
          "compute_throughput_us_p90": 278.5626375,
          "command_buffer_throughput_us": [
            275.61393198993756,
            279.9485682771774,
            273.5664056672249,
            278.21679668704746,
            274.921874733991,
            277.57617135648616,
            278.03190096165054,
            275.315755061456,
            274.05859418649925
          ],
          "command_buffer_throughput_us_p50": 275.61393198993756,
          "command_buffer_throughput_us_p90": 278.56315100507345,
          "compute_latency_us": [
            198.833,
            201.5,
            200.167,
            202.291,
            200.167,
            199.792,
            200.459,
            199.875,
            199.875
          ],
          "compute_latency_us_p50": 200.167,
          "compute_latency_us_p90": 201.6582,
          "command_buffer_latency_us": [
            198.83335335180163,
            201.4999627135694,
            200.20833471789956,
            202.25002663210034,
            200.20833471789956,
            199.8333027586341,
            200.50001330673695,
            199.87497944384813,
            199.87497944384813
          ],
          "command_buffer_latency_us_p50": 200.20833471789956,
          "command_buffer_latency_us_p90": 201.6499754972756,
          "counter_control_throughput_ratio": 0.9995779935378021,
          "counter_control_latency_ratio": 0.999166757131095
        },
        "download_ms": 0.615,
        "allocation_upload_ms": 1.502708,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.1634934849478213e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-slo6ygjk/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "7d02ca31ce0660bb7b8cf2d4b7c68e2de8afe58cfb625e16c6aa93173ce8f234",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 21.589916,
        "capture_ms": 0.071125,
        "compile_ms": 64.6705,
        "allocation_upload_ms": 6.649583,
        "cold_call_ms": 8.393875,
        "warmup_ms": 200.208041,
        "download_ms": 2.846375,
        "repetitions": 289,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          112.089678201,
          111.709775087,
          112.100058824,
          107.431228374,
          106.21467474,
          111.553633218,
          112.399941176,
          113.427913495,
          112.930650519
        ],
        "latency_us": [
          442.125,
          376.458,
          353.667,
          467.583,
          377,
          365.5,
          465.292,
          4300.875,
          364.417
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 512,
            "reduction_scalar_rounds": 80,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 49152,
              "global_write_bytes": 16384,
              "private_read_bytes": 98304,
              "private_write_bytes": 32768
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 192,
              "global_write_bytes": 64,
              "private_read_bytes": 384,
              "private_write_bytes": 128
            },
            "striped_storage_scalars_per_worker": 32,
            "reduction_operations": 2,
            "reduction_elements": 8192,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 120,
            "normalized_kernel_cost": 1920,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 5533875,
              "compute_span_ns": 5533875,
              "calibration_cpu_ns": 6341625,
              "calibration_gpu_ticks": 6341625,
              "compute_passes": 1,
              "command_buffer_ns": 5533916.64289,
              "command_buffers": 2
            },
            {
              "compute_ns": 5209000,
              "compute_span_ns": 5209000,
              "calibration_cpu_ns": 6100209,
              "calibration_gpu_ticks": 6100209,
              "compute_passes": 1,
              "command_buffer_ns": 5209124.99633,
              "command_buffers": 2
            },
            {
              "compute_ns": 5817125,
              "compute_span_ns": 5817125,
              "calibration_cpu_ns": 6827584,
              "calibration_gpu_ticks": 6827584,
              "compute_passes": 1,
              "command_buffer_ns": 5817249.99938,
              "command_buffers": 2
            },
            {
              "compute_ns": 5349167,
              "compute_span_ns": 5349167,
              "calibration_cpu_ns": 6582042,
              "calibration_gpu_ticks": 6582042,
              "compute_passes": 1,
              "command_buffer_ns": 5349208.30978,
              "command_buffers": 2
            },
            {
              "compute_ns": 5491500,
              "compute_span_ns": 5491500,
              "calibration_cpu_ns": 6487916,
              "calibration_gpu_ticks": 6487916,
              "compute_passes": 1,
              "command_buffer_ns": 5491500.02189,
              "command_buffers": 2
            },
            {
              "compute_ns": 5591291,
              "compute_span_ns": 5591291,
              "calibration_cpu_ns": 6677500,
              "calibration_gpu_ticks": 6677500,
              "compute_passes": 1,
              "command_buffer_ns": 5591374.93838,
              "command_buffers": 2
            },
            {
              "compute_ns": 5637708,
              "compute_span_ns": 5637708,
              "calibration_cpu_ns": 6525083,
              "calibration_gpu_ticks": 6525083,
              "compute_passes": 1,
              "command_buffer_ns": 5637749.97136,
              "command_buffers": 2
            },
            {
              "compute_ns": 5435417,
              "compute_span_ns": 5435417,
              "calibration_cpu_ns": 6465541,
              "calibration_gpu_ticks": 6465541,
              "compute_passes": 1,
              "command_buffer_ns": 5435458.38445,
              "command_buffers": 2
            },
            {
              "compute_ns": 7526166,
              "compute_span_ns": 7526166,
              "calibration_cpu_ns": 8307917,
              "calibration_gpu_ticks": 8307917,
              "compute_passes": 1,
              "command_buffer_ns": 7526208.35556,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 104000,
              "compute_span_ns": 104000,
              "calibration_cpu_ns": 515125,
              "calibration_gpu_ticks": 515125,
              "compute_passes": 1,
              "command_buffer_ns": 104083.33037,
              "command_buffers": 2
            },
            {
              "compute_ns": 79500,
              "compute_span_ns": 79500,
              "calibration_cpu_ns": 483417,
              "calibration_gpu_ticks": 483417,
              "compute_passes": 1,
              "command_buffer_ns": 79583.3184384,
              "command_buffers": 2
            },
            {
              "compute_ns": 99000,
              "compute_span_ns": 99000,
              "calibration_cpu_ns": 586958,
              "calibration_gpu_ticks": 586958,
              "compute_passes": 1,
              "command_buffer_ns": 99083.3505057,
              "command_buffers": 2
            },
            {
              "compute_ns": 77375,
              "compute_span_ns": 77375,
              "calibration_cpu_ns": 533125,
              "calibration_gpu_ticks": 533125,
              "compute_passes": 1,
              "command_buffer_ns": 77458.2731538,
              "command_buffers": 2
            },
            {
              "compute_ns": 109917,
              "compute_span_ns": 109917,
              "calibration_cpu_ns": 746583,
              "calibration_gpu_ticks": 746583,
              "compute_passes": 1,
              "command_buffer_ns": 110000.022687,
              "command_buffers": 2
            },
            {
              "compute_ns": 78709,
              "compute_span_ns": 78709,
              "calibration_cpu_ns": 634625,
              "calibration_gpu_ticks": 634625,
              "compute_passes": 1,
              "command_buffer_ns": 78833.3709352,
              "command_buffers": 2
            },
            {
              "compute_ns": 79625,
              "compute_span_ns": 79625,
              "calibration_cpu_ns": 709125,
              "calibration_gpu_ticks": 709125,
              "compute_passes": 1,
              "command_buffer_ns": 79708.348494,
              "command_buffers": 2
            },
            {
              "compute_ns": 115584,
              "compute_span_ns": 115584,
              "calibration_cpu_ns": 745209,
              "calibration_gpu_ticks": 745209,
              "compute_passes": 1,
              "command_buffer_ns": 115624.978207,
              "command_buffers": 2
            },
            {
              "compute_ns": 75875,
              "compute_span_ns": 75875,
              "calibration_cpu_ns": 632042,
              "calibration_gpu_ticks": 632042,
              "compute_passes": 1,
              "command_buffer_ns": 76000.0548325,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 5406333.3082,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5815291.60263,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5514791.64278,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5582333.36778,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5354833.38172,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5265583.2842,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5801541.67255,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5566749.9546,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5287250.03824,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 509958.306793,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 85333.3040141,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 462041.702121,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76791.679021,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 115708.331577,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 115000.002552,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 77041.6227169,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 83499.9955259,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 112833.33879,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              84.473957940625,
              90.86393129109375,
              86.1686194184375,
              87.2239588715625,
              83.669271589375,
              82.274738815625,
              90.64908863359375,
              86.980468040625,
              82.61328184749999
            ],
            "command_buffer_throughput_us_p50": 86.1686194184375,
            "command_buffer_throughput_us_p90": 90.69205716509374,
            "command_buffer_latency_us": [
              509.958306793,
              85.3333040141,
              462.04170212099996,
              76.79167902100001,
              115.70833157700001,
              115.00000255200001,
              77.04162271690001,
              83.4999955259,
              112.83333879
            ],
            "command_buffer_latency_us_p50": 112.83333879,
            "command_buffer_latency_us_p90": 471.6250230554
          },
          "compute_throughput_us": [
            86.466796875,
            81.390625,
            90.892578125,
            83.580734375,
            85.8046875,
            87.363921875,
            88.0891875,
            84.928390625,
            117.59634375
          ],
          "compute_throughput_us_p50": 86.466796875,
          "compute_throughput_us_p90": 96.23333125,
          "command_buffer_throughput_us": [
            86.46744754515625,
            81.39257806765626,
            90.8945312403125,
            83.5813798403125,
            85.80468784203126,
            87.36523341218751,
            88.0898433025,
            84.92903725703124,
            117.597005555625
          ],
          "command_buffer_throughput_us_p50": 86.46744754515625,
          "command_buffer_throughput_us_p90": 96.235026103375,
          "compute_latency_us": [
            104.0,
            79.5,
            99.0,
            77.375,
            109.917,
            78.709,
            79.625,
            115.584,
            75.875
          ],
          "compute_latency_us_p50": 79.625,
          "compute_latency_us_p90": 111.0504,
          "command_buffer_latency_us": [
            104.08333037,
            79.5833184384,
            99.0833505057,
            77.4582731538,
            110.000022687,
            78.8333709352,
            79.708348494,
            115.624978207,
            76.0000548325
          ],
          "command_buffer_latency_us_p50": 79.708348494,
          "command_buffer_latency_us_p90": 111.125013791,
          "counter_control_throughput_ratio": 1.0235988658887327,
          "counter_control_latency_ratio": 0.932617333383108
        },
        "correctness": {
          "max_abs_error": 2.827305296548843e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 112.089678201,
        "throughput_us_p90": 113.0301031142,
        "latency_us_p50": 377.0,
        "latency_us_p90": 1234.2414000000008
      },
      "slowdown": 0.3768021838035857,
      "device_compute_slowdown": 0.31372432635051334,
      "valid": true,
      "round": 1,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "softmax_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-exi2gsii/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "d45e9e72129e03853b1c708b1bdcbc59be9547b0d0e02ca6bdc2eb0b41c08926",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 21.85825,
        "capture_ms": 0.06725,
        "compile_ms": 61.999125,
        "allocation_upload_ms": 1.004833,
        "cold_call_ms": 2.682792,
        "warmup_ms": 200.056917,
        "download_ms": 0.392458,
        "repetitions": 4029,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          6.38052246215,
          6.50328865724,
          6.78379647555,
          6.27798461157,
          6.24834524696,
          6.33249565649,
          6.93346140482,
          6.34030354927,
          6.17783569124
        ],
        "latency_us": [
          247.542,
          2301.667,
          572.125,
          425.334,
          213.875,
          196.333,
          211.291,
          194.584,
          284.333
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 19,
            "reduction_scalar_rounds": 40,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 6148,
              "global_write_bytes": 6148,
              "private_read_bytes": 36888,
              "private_write_bytes": 12296
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 32,
              "global_write_bytes": 32,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 2,
            "reduction_elements": 3074,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 80,
            "normalized_kernel_cost": 80,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 4611,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 263709,
              "compute_span_ns": 263709,
              "calibration_cpu_ns": 711416,
              "calibration_gpu_ticks": 711416,
              "compute_passes": 1,
              "command_buffer_ns": 263791.705947,
              "command_buffers": 2
            },
            {
              "compute_ns": 712458,
              "compute_span_ns": 712458,
              "calibration_cpu_ns": 1416667,
              "calibration_gpu_ticks": 1416667,
              "compute_passes": 1,
              "command_buffer_ns": 712541.688699,
              "command_buffers": 2
            },
            {
              "compute_ns": 715625,
              "compute_span_ns": 715625,
              "calibration_cpu_ns": 1280625,
              "calibration_gpu_ticks": 1280625,
              "compute_passes": 1,
              "command_buffer_ns": 715708.360076,
              "command_buffers": 2
            },
            {
              "compute_ns": 276250,
              "compute_span_ns": 276250,
              "calibration_cpu_ns": 741125,
              "calibration_gpu_ticks": 741125,
              "compute_passes": 1,
              "command_buffer_ns": 276250.008028,
              "command_buffers": 2
            },
            {
              "compute_ns": 298042,
              "compute_span_ns": 298042,
              "calibration_cpu_ns": 824333,
              "calibration_gpu_ticks": 824333,
              "compute_passes": 1,
              "command_buffer_ns": 298083.352391,
              "command_buffers": 2
            },
            {
              "compute_ns": 264834,
              "compute_span_ns": 264834,
              "calibration_cpu_ns": 678291,
              "calibration_gpu_ticks": 678291,
              "compute_passes": 1,
              "command_buffer_ns": 264875.008725,
              "command_buffers": 2
            },
            {
              "compute_ns": 297750,
              "compute_span_ns": 297750,
              "calibration_cpu_ns": 753166,
              "calibration_gpu_ticks": 753166,
              "compute_passes": 1,
              "command_buffer_ns": 297791.73201,
              "command_buffers": 2
            },
            {
              "compute_ns": 708542,
              "compute_span_ns": 708542,
              "calibration_cpu_ns": 1951167,
              "calibration_gpu_ticks": 1951167,
              "compute_passes": 1,
              "command_buffer_ns": 708666.688297,
              "command_buffers": 2
            },
            {
              "compute_ns": 818042,
              "compute_span_ns": 818042,
              "calibration_cpu_ns": 1583666,
              "calibration_gpu_ticks": 1583666,
              "compute_passes": 1,
              "command_buffer_ns": 818125.030492,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 6583,
              "compute_span_ns": 6583,
              "calibration_cpu_ns": 475750,
              "calibration_gpu_ticks": 475750,
              "compute_passes": 1,
              "command_buffer_ns": 6666.63981974,
              "command_buffers": 2
            },
            {
              "compute_ns": 197500,
              "compute_span_ns": 197500,
              "calibration_cpu_ns": 841584,
              "calibration_gpu_ticks": 841584,
              "compute_passes": 1,
              "command_buffer_ns": 197541.667148,
              "command_buffers": 2
            },
            {
              "compute_ns": 9833,
              "compute_span_ns": 9833,
              "calibration_cpu_ns": 473458,
              "calibration_gpu_ticks": 473458,
              "compute_passes": 1,
              "command_buffer_ns": 9833.31119642,
              "command_buffers": 2
            },
            {
              "compute_ns": 7000,
              "compute_span_ns": 7000,
              "calibration_cpu_ns": 374083,
              "calibration_gpu_ticks": 374083,
              "compute_passes": 1,
              "command_buffer_ns": 6999.99509379,
              "command_buffers": 2
            },
            {
              "compute_ns": 6500,
              "compute_span_ns": 6500,
              "calibration_cpu_ns": 374625,
              "calibration_gpu_ticks": 374625,
              "compute_passes": 1,
              "command_buffer_ns": 6583.34465697,
              "command_buffers": 2
            },
            {
              "compute_ns": 9375,
              "compute_span_ns": 9375,
              "calibration_cpu_ns": 365875,
              "calibration_gpu_ticks": 365875,
              "compute_passes": 1,
              "command_buffer_ns": 9374.98407438,
              "command_buffers": 2
            },
            {
              "compute_ns": 8083,
              "compute_span_ns": 8083,
              "calibration_cpu_ns": 675250,
              "calibration_gpu_ticks": 675250,
              "compute_passes": 1,
              "command_buffer_ns": 8083.23966339,
              "command_buffers": 2
            },
            {
              "compute_ns": 7625,
              "compute_span_ns": 7625,
              "calibration_cpu_ns": 1232583,
              "calibration_gpu_ticks": 1232583,
              "compute_passes": 1,
              "command_buffer_ns": 7666.64743423,
              "command_buffers": 2
            },
            {
              "compute_ns": 6916,
              "compute_span_ns": 6916,
              "calibration_cpu_ns": 534375,
              "calibration_gpu_ticks": 534375,
              "compute_passes": 1,
              "command_buffer_ns": 6916.64172336,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 447000.027634,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 279375.002719,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 263291.643932,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 295749.981888,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 832791.67302,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 592791.708186,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 266833.347268,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 265999.988187,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 263125.053607,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 6541.66797176,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 8208.26971903,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6625.02134219,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 88166.6783243,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7791.67748988,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7166.70183465,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 175166.642293,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 108500.011265,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7958.38423073,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              6.98437543178125,
              4.365234417484375,
              4.1139319364375,
              4.621093467,
              13.0123698909375,
              9.26237044040625,
              4.1692710510625,
              4.156249815421875,
              4.111328962609375
            ],
            "command_buffer_throughput_us_p50": 4.365234417484375,
            "command_buffer_throughput_us_p90": 10.012370330512502,
            "command_buffer_latency_us": [
              6.54166797176,
              8.208269719030001,
              6.62502134219,
              88.1666783243,
              7.7916774898800005,
              7.1667018346499995,
              175.16664229300002,
              108.500011265,
              7.95838423073
            ],
            "command_buffer_latency_us_p50": 7.95838423073,
            "command_buffer_latency_us_p90": 121.83333747060001
          },
          "compute_throughput_us": [
            4.120453125,
            11.13215625,
            11.181640625,
            4.31640625,
            4.65690625,
            4.13803125,
            4.65234375,
            11.07096875,
            12.78190625
          ],
          "compute_throughput_us_p50": 4.65690625,
          "compute_throughput_us_p90": 11.501693750000001,
          "command_buffer_throughput_us": [
            4.121745405421875,
            11.133463885921875,
            11.1829431261875,
            4.3164063754375,
            4.657552381109375,
            4.138672011328126,
            4.65299581265625,
            11.072917004640626,
            12.783203601437501
          ],
          "command_buffer_throughput_us_p50": 4.657552381109375,
          "command_buffer_throughput_us_p90": 11.5029952212375,
          "compute_latency_us": [
            6.583,
            197.5,
            9.833,
            7.0,
            6.5,
            9.375,
            8.083,
            7.625,
            6.916
          ],
          "compute_latency_us_p50": 7.625,
          "compute_latency_us_p90": 47.366400000000034,
          "command_buffer_latency_us": [
            6.66663981974,
            197.54166714800002,
            9.83331119642,
            6.99999509379,
            6.58334465697,
            9.37498407438,
            8.08323966339,
            7.66664743423,
            6.91664172336
          ],
          "command_buffer_latency_us_p50": 7.66664743423,
          "command_buffer_latency_us_p90": 47.374982386736036,
          "counter_control_throughput_ratio": 1.116021423330219,
          "counter_control_latency_ratio": 0.869101255082975
        },
        "correctness": {
          "max_abs_error": 2.6856349761260334e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 6.34030354927,
        "throughput_us_p90": 6.813729461404,
        "latency_us_p50": 247.542,
        "latency_us_p90": 918.0334000000003
      },
      "torch": {
        "cold_call_ms": 10.429584,
        "warmup_ms": 201.295584,
        "repetitions": 624,
        "throughput_us": [
          33.54360256410257,
          34.80488782051282,
          35.86885737179487,
          32.52670993589743,
          35.16386217948718,
          32.814770833333334,
          35.75801282051282,
          95.05221634615384,
          36.15344551282051
        ],
        "latency_us": [
          342.792,
          341.75,
          311.542,
          314.25,
          317.70799999999997,
          333.95799999999997,
          347.917,
          363.79200000000003,
          378.125
        ],
        "throughput_us_p50": 35.16386217948718,
        "throughput_us_p90": 47.93319967948719,
        "latency_us_p50": 341.75,
        "latency_us_p90": 366.65860000000004,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1043416.6179038584,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1032291.620504111,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1070000.056643039,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1030999.9925084412,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1274666.6907332838,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1025916.6592732072,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1311999.9784976244,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1122208.3121538162,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1187124.98806417,
                "command_buffers": 7
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 70250.01104921103,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 57250.02847611904,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 57208.293583244085,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70291.68773442507,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70125.03920122981,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 86041.69124737382,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 88458.29870551825,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 69750.00724196434,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 76875.03239139915,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              16.303384654747788,
              16.129556570376735,
              16.718750885047484,
              16.109374882944394,
              19.91666704270756,
              16.029947801143862,
              20.49999966402538,
              17.53450487740338,
              18.548827938502654
            ],
            "command_buffer_throughput_us_p50": 16.718750885047484,
            "command_buffer_throughput_us_p90": 20.033333566971123,
            "command_buffer_latency_us": [
              70.25001104921103,
              57.25002847611904,
              57.208293583244085,
              70.29168773442507,
              70.12503920122981,
              86.04169124737382,
              88.45829870551825,
              69.75000724196434,
              76.87503239139915
            ],
            "command_buffer_latency_us_p50": 70.25001104921103,
            "command_buffer_latency_us_p90": 86.5250127390027
          },
          "throughput": [
            {
              "compute_ns": 3846333.0,
              "compute_span_ns": 4740416.0,
              "command_buffer_ns": 4654874.908737838,
              "calibration_cpu_ns": 5543625.0,
              "calibration_gpu_ticks": 5543625.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3425875.0,
              "compute_span_ns": 4493792.0,
              "command_buffer_ns": 4412624.926771969,
              "calibration_cpu_ns": 6413000.0,
              "calibration_gpu_ticks": 6413000.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3668293.0,
              "compute_span_ns": 4897917.0,
              "command_buffer_ns": 4586625.145748258,
              "calibration_cpu_ns": 5875666.0,
              "calibration_gpu_ticks": 5875666.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3737958.0,
              "compute_span_ns": 5003875.0,
              "command_buffer_ns": 4736499.91909042,
              "calibration_cpu_ns": 6047792.0,
              "calibration_gpu_ticks": 6047792.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3628870.0,
              "compute_span_ns": 4703083.0,
              "command_buffer_ns": 4600875.137839466,
              "calibration_cpu_ns": 5625042.0,
              "calibration_gpu_ticks": 5625042.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3641455.0,
              "compute_span_ns": 4712708.0,
              "command_buffer_ns": 4601166.700012982,
              "calibration_cpu_ns": 5769917.0,
              "calibration_gpu_ticks": 5769917.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3813667.0,
              "compute_span_ns": 5257666.0,
              "command_buffer_ns": 4836583.335418254,
              "calibration_cpu_ns": 6305292.0,
              "calibration_gpu_ticks": 6305292.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3752453.0,
              "compute_span_ns": 4832250.0,
              "command_buffer_ns": 4706916.690338403,
              "calibration_cpu_ns": 5978584.0,
              "calibration_gpu_ticks": 5978584.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4030921.0,
              "compute_span_ns": 5176542.0,
              "command_buffer_ns": 5077416.659332812,
              "calibration_cpu_ns": 6130625.0,
              "calibration_gpu_ticks": 6130625.0,
              "compute_passes": 64,
              "command_buffers": 7
            }
          ],
          "latency": [
            {
              "compute_ns": 70250.0,
              "compute_span_ns": 70250.0,
              "command_buffer_ns": 70250.01104921103,
              "calibration_cpu_ns": 713750.0,
              "calibration_gpu_ticks": 713750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 57208.0,
              "compute_span_ns": 57208.0,
              "command_buffer_ns": 57166.675105690956,
              "calibration_cpu_ns": 524375.0,
              "calibration_gpu_ticks": 524375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70375.0,
              "compute_span_ns": 70375.0,
              "command_buffer_ns": 70374.98289719224,
              "calibration_cpu_ns": 696625.0,
              "calibration_gpu_ticks": 696625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 71750.0,
              "compute_span_ns": 71750.0,
              "command_buffer_ns": 71750.02247095108,
              "calibration_cpu_ns": 925583.0,
              "calibration_gpu_ticks": 925583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 69750.0,
              "compute_span_ns": 69750.0,
              "command_buffer_ns": 69749.94903430343,
              "calibration_cpu_ns": 472583.0,
              "calibration_gpu_ticks": 472583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 54209.0,
              "compute_span_ns": 54209.0,
              "command_buffer_ns": 54250.00563263893,
              "calibration_cpu_ns": 516542.0,
              "calibration_gpu_ticks": 516542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70167.0,
              "compute_span_ns": 70167.0,
              "command_buffer_ns": 70166.65767878294,
              "calibration_cpu_ns": 912208.0,
              "calibration_gpu_ticks": 912208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 69875.0,
              "compute_span_ns": 69875.0,
              "command_buffer_ns": 69874.97908994555,
              "calibration_cpu_ns": 612000.0,
              "calibration_gpu_ticks": 612000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 73542.0,
              "compute_span_ns": 73542.0,
              "command_buffer_ns": 73541.6542738676,
              "calibration_cpu_ns": 486375.0,
              "calibration_gpu_ticks": 486375.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            60.098953125,
            53.529296875,
            57.317078125,
            58.40559375,
            56.70109375,
            56.897734375,
            59.588546875,
            58.632078125,
            62.983140625
          ],
          "compute_throughput_us_p50": 58.40559375,
          "compute_throughput_us_p90": 60.675790625,
          "command_buffer_throughput_us": [
            72.73242044902872,
            68.94726448081201,
            71.66601790231653,
            74.00781123578781,
            71.88867402874166,
            71.89322968770284,
            75.57161461591022,
            73.54557328653755,
            79.33463530207518
          ],
          "command_buffer_throughput_us_p50": 72.73242044902872,
          "command_buffer_throughput_us_p90": 76.32421875314321,
          "compute_latency_us": [
            70.25,
            57.208,
            70.375,
            71.75,
            69.75,
            54.209,
            70.167,
            69.875,
            73.542
          ],
          "compute_latency_us_p50": 70.167,
          "compute_latency_us_p90": 72.1084,
          "command_buffer_latency_us": [
            70.25001104921103,
            57.166675105690956,
            70.37498289719224,
            71.75002247095108,
            69.74994903430343,
            54.25000563263893,
            70.16665767878294,
            69.87497908994555,
            73.54165427386761
          ],
          "command_buffer_latency_us_p50": 70.16665767878294,
          "command_buffer_latency_us_p90": 72.10834883153439,
          "counter_control_throughput_ratio": 4.277069988740186,
          "counter_control_latency_ratio": 0.9985440466555776
        },
        "download_ms": 0.334917,
        "allocation_upload_ms": 0.745458,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 4.2122453832241225e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.18030737115585135,
      "device_compute_slowdown": 0.07973390819265491,
      "valid": true,
      "round": 1,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "softmax_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.755208,
        "warmup_ms": 202.065292,
        "repetitions": 726,
        "throughput_us": [
          32.612546831955925,
          34.797462809917356,
          35.06887052341598,
          36.73829201101928,
          36.136307162534436,
          32.551768595041324,
          34.43864738292012,
          33.163107438016524,
          33.63756887052342
        ],
        "latency_us": [
          380.125,
          334.334,
          346.08299999999997,
          473.834,
          336.75,
          346.75,
          344.875,
          329.584,
          332.91700000000003
        ],
        "throughput_us_p50": 34.43864738292012,
        "throughput_us_p90": 36.256704132231405,
        "latency_us_p50": 344.875,
        "latency_us_p90": 398.8668,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1309124.8692944646,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1047125.0279806554,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1200833.3578705788,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1132875.0988468528,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1184499.8807646334,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1334708.4168344736,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1212291.6523367167,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1323000.0622570515,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1470833.3183079958,
                "command_buffers": 7
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 70458.33626762033,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 69416.71017557383,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70874.98670443892,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 73208.29899981618,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 74166.68813675642,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70000.00914558768,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70375.04110485315,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 91166.64296016097,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 77208.32945778966,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              20.45507608272601,
              16.36132856219774,
              18.763021216727793,
              17.701173419482075,
              18.507810636947397,
              20.85481901303865,
              18.942057067761198,
              20.67187597276643,
              22.981770598562434
            ],
            "command_buffer_throughput_us_p50": 18.942057067761198,
            "command_buffer_throughput_us_p90": 21.280209330143407,
            "command_buffer_latency_us": [
              70.45833626762033,
              69.41671017557383,
              70.87498670443892,
              73.20829899981618,
              74.16668813675642,
              70.00000914558768,
              70.37504110485315,
              91.16664296016097,
              77.20832945778966
            ],
            "command_buffer_latency_us_p50": 70.87498670443892,
            "command_buffer_latency_us_p90": 79.99999215826392
          },
          "throughput": [
            {
              "compute_ns": 3713370.0,
              "compute_span_ns": 4767042.0,
              "command_buffer_ns": 4669083.340559155,
              "calibration_cpu_ns": 5687209.0,
              "calibration_gpu_ticks": 5687209.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3704125.0,
              "compute_span_ns": 4764750.0,
              "command_buffer_ns": 4656833.421904594,
              "calibration_cpu_ns": 5762667.0,
              "calibration_gpu_ticks": 5762667.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3786502.0,
              "compute_span_ns": 4862917.0,
              "command_buffer_ns": 4775583.220180124,
              "calibration_cpu_ns": 5836583.0,
              "calibration_gpu_ticks": 5836583.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3816712.0,
              "compute_span_ns": 4781333.0,
              "command_buffer_ns": 4693124.967161566,
              "calibration_cpu_ns": 5802208.0,
              "calibration_gpu_ticks": 5802208.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4023252.0,
              "compute_span_ns": 5114791.0,
              "command_buffer_ns": 5001499.957870692,
              "calibration_cpu_ns": 6152833.0,
              "calibration_gpu_ticks": 6152833.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4119121.0,
              "compute_span_ns": 5276625.0,
              "command_buffer_ns": 5158041.6620709,
              "calibration_cpu_ns": 6347542.0,
              "calibration_gpu_ticks": 6347542.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4387496.0,
              "compute_span_ns": 5607708.0,
              "command_buffer_ns": 5490666.721016169,
              "calibration_cpu_ns": 6704125.0,
              "calibration_gpu_ticks": 6704125.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4446459.0,
              "compute_span_ns": 5533875.0,
              "command_buffer_ns": 5434124.963358045,
              "calibration_cpu_ns": 6592000.0,
              "calibration_gpu_ticks": 6592000.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4693250.0,
              "compute_span_ns": 5940792.0,
              "command_buffer_ns": 5812749.965116382,
              "calibration_cpu_ns": 7035000.0,
              "calibration_gpu_ticks": 7035000.0,
              "compute_passes": 64,
              "command_buffers": 7
            }
          ],
          "latency": [
            {
              "compute_ns": 69917.0,
              "compute_span_ns": 69917.0,
              "command_buffer_ns": 69916.6557751596,
              "calibration_cpu_ns": 571750.0,
              "calibration_gpu_ticks": 571750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 69875.0,
              "compute_span_ns": 69875.0,
              "command_buffer_ns": 69874.97908994555,
              "calibration_cpu_ns": 657791.0,
              "calibration_gpu_ticks": 657791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 86167.0,
              "compute_span_ns": 86167.0,
              "command_buffer_ns": 86208.33978056908,
              "calibration_cpu_ns": 769625.0,
              "calibration_gpu_ticks": 769625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 88875.0,
              "compute_span_ns": 88875.0,
              "command_buffer_ns": 88875.00734999776,
              "calibration_cpu_ns": 539958.0,
              "calibration_gpu_ticks": 539958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 69791.0,
              "compute_span_ns": 69791.0,
              "command_buffer_ns": 69750.00724196434,
              "calibration_cpu_ns": 722625.0,
              "calibration_gpu_ticks": 722625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 69917.0,
              "compute_span_ns": 69917.0,
              "command_buffer_ns": 69958.33246037364,
              "calibration_cpu_ns": 757250.0,
              "calibration_gpu_ticks": 757250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 88583.0,
              "compute_span_ns": 88583.0,
              "command_buffer_ns": 88541.65207594633,
              "calibration_cpu_ns": 549959.0,
              "calibration_gpu_ticks": 549959.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 86750.0,
              "compute_span_ns": 86750.0,
              "command_buffer_ns": 86750.0202730298,
              "calibration_cpu_ns": 798958.0,
              "calibration_gpu_ticks": 798958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70375.0,
              "compute_span_ns": 70375.0,
              "command_buffer_ns": 70374.98289719224,
              "calibration_cpu_ns": 587542.0,
              "calibration_gpu_ticks": 587542.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            58.02140625,
            57.876953125,
            59.16409375,
            59.636125,
            62.8633125,
            64.361265625,
            68.554625,
            69.475921875,
            73.33203125
          ],
          "compute_throughput_us_p50": 62.8633125,
          "compute_throughput_us_p90": 70.24714374999999,
          "command_buffer_throughput_us": [
            72.9544271962368,
            72.76302221725928,
            74.61848781531444,
            73.33007761189947,
            78.14843684172956,
            80.59440096985782,
            85.79166751587763,
            84.90820255246945,
            90.82421820494346
          ],
          "command_buffer_throughput_us_p50": 78.14843684172956,
          "command_buffer_throughput_us_p90": 86.7981776536908,
          "compute_latency_us": [
            69.917,
            69.875,
            86.167,
            88.875,
            69.791,
            69.917,
            88.583,
            86.75,
            70.375
          ],
          "compute_latency_us_p50": 70.375,
          "compute_latency_us_p90": 88.6414,
          "command_buffer_latency_us": [
            69.9166557751596,
            69.87497908994555,
            86.20833978056908,
            88.87500734999776,
            69.75000724196434,
            69.95833246037364,
            88.54165207594633,
            86.7500202730298,
            70.37498289719224
          ],
          "command_buffer_latency_us_p50": 70.37498289719224,
          "command_buffer_latency_us_p90": 88.60832313075662,
          "counter_control_throughput_ratio": 4.107426082873626,
          "counter_control_latency_ratio": 0.9994046188604438
        },
        "download_ms": 0.312709,
        "allocation_upload_ms": 0.703709,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 4.2122453832241225e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-1b7nt6ik/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "1ca0ba2a59ee9b200a0f7933e3c823dd873803c26a844f54bebdc043f01bc250",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.334334,
        "capture_ms": 0.067583,
        "compile_ms": 54.260042,
        "allocation_upload_ms": 1.075292,
        "cold_call_ms": 60.234625,
        "warmup_ms": 200.202625,
        "download_ms": 0.380917,
        "repetitions": 5101,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          4.41743462066,
          4.39870437169,
          4.78665947853,
          4.30409057048,
          4.37026230151,
          4.33314546167,
          4.26908939424,
          4.35920250931,
          4.2503838463
        ],
        "latency_us": [
          808.708,
          279.75,
          346.875,
          225.625,
          257.584,
          270.167,
          234.042,
          219,
          252.209
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 37,
            "reduction_scalar_rounds": 40,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 6148,
              "global_write_bytes": 6148,
              "private_read_bytes": 36888,
              "private_write_bytes": 12296
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 32,
              "global_write_bytes": 32,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 2,
            "reduction_elements": 3074,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 88,
            "normalized_kernel_cost": 88,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 4611,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 266375,
              "compute_span_ns": 266375,
              "calibration_cpu_ns": 1047167,
              "calibration_gpu_ticks": 1047167,
              "compute_passes": 1,
              "command_buffer_ns": 266458.373517,
              "command_buffers": 2
            },
            {
              "compute_ns": 245000,
              "compute_span_ns": 245000,
              "calibration_cpu_ns": 819000,
              "calibration_gpu_ticks": 819000,
              "compute_passes": 1,
              "command_buffer_ns": 245000.002906,
              "command_buffers": 2
            },
            {
              "compute_ns": 242333,
              "compute_span_ns": 242333,
              "calibration_cpu_ns": 692083,
              "calibration_gpu_ticks": 692083,
              "compute_passes": 1,
              "command_buffer_ns": 242416.688707,
              "command_buffers": 2
            },
            {
              "compute_ns": 243500,
              "compute_span_ns": 243500,
              "calibration_cpu_ns": 967500,
              "calibration_gpu_ticks": 967500,
              "compute_passes": 1,
              "command_buffer_ns": 243583.344854,
              "command_buffers": 2
            },
            {
              "compute_ns": 262375,
              "compute_span_ns": 262375,
              "calibration_cpu_ns": 769250,
              "calibration_gpu_ticks": 769250,
              "compute_passes": 1,
              "command_buffer_ns": 262416.666374,
              "command_buffers": 2
            },
            {
              "compute_ns": 262209,
              "compute_span_ns": 262209,
              "calibration_cpu_ns": 716750,
              "calibration_gpu_ticks": 716750,
              "compute_passes": 1,
              "command_buffer_ns": 262291.636318,
              "command_buffers": 2
            },
            {
              "compute_ns": 283875,
              "compute_span_ns": 283875,
              "calibration_cpu_ns": 766458,
              "calibration_gpu_ticks": 766458,
              "compute_passes": 1,
              "command_buffer_ns": 283875.036985,
              "command_buffers": 2
            },
            {
              "compute_ns": 277291,
              "compute_span_ns": 277291,
              "calibration_cpu_ns": 960125,
              "calibration_gpu_ticks": 960125,
              "compute_passes": 1,
              "command_buffer_ns": 277291.63412,
              "command_buffers": 2
            },
            {
              "compute_ns": 269709,
              "compute_span_ns": 269709,
              "calibration_cpu_ns": 713333,
              "calibration_gpu_ticks": 713333,
              "compute_passes": 1,
              "command_buffer_ns": 269749.958534,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 6709,
              "compute_span_ns": 6709,
              "calibration_cpu_ns": 423833,
              "calibration_gpu_ticks": 423833,
              "compute_passes": 1,
              "command_buffer_ns": 6833.3465606,
              "command_buffers": 2
            },
            {
              "compute_ns": 6834,
              "compute_span_ns": 6834,
              "calibration_cpu_ns": 781459,
              "calibration_gpu_ticks": 781459,
              "compute_passes": 1,
              "command_buffer_ns": 6958.31840858,
              "command_buffers": 2
            },
            {
              "compute_ns": 6708,
              "compute_span_ns": 6708,
              "calibration_cpu_ns": 2177834,
              "calibration_gpu_ticks": 2177834,
              "compute_passes": 1,
              "command_buffer_ns": 6833.3465606,
              "command_buffers": 2
            },
            {
              "compute_ns": 6583,
              "compute_span_ns": 6583,
              "calibration_cpu_ns": 436916,
              "calibration_gpu_ticks": 436916,
              "compute_passes": 1,
              "command_buffer_ns": 6666.63981974,
              "command_buffers": 2
            },
            {
              "compute_ns": 7708,
              "compute_span_ns": 7708,
              "calibration_cpu_ns": 706667,
              "calibration_gpu_ticks": 706667,
              "compute_passes": 1,
              "command_buffer_ns": 7708.38232711,
              "command_buffers": 2
            },
            {
              "compute_ns": 6708,
              "compute_span_ns": 6708,
              "calibration_cpu_ns": 503458,
              "calibration_gpu_ticks": 503458,
              "compute_passes": 1,
              "command_buffer_ns": 6749.93498251,
              "command_buffers": 2
            },
            {
              "compute_ns": 6667,
              "compute_span_ns": 6667,
              "calibration_cpu_ns": 413958,
              "calibration_gpu_ticks": 413958,
              "compute_passes": 1,
              "command_buffer_ns": 6708.31650496,
              "command_buffers": 2
            },
            {
              "compute_ns": 13625,
              "compute_span_ns": 13625,
              "calibration_cpu_ns": 480833,
              "calibration_gpu_ticks": 480833,
              "compute_passes": 1,
              "command_buffer_ns": 13708.3698064,
              "command_buffers": 2
            },
            {
              "compute_ns": 7542,
              "compute_span_ns": 7542,
              "calibration_cpu_ns": 600375,
              "calibration_gpu_ticks": 600375,
              "compute_passes": 1,
              "command_buffer_ns": 7625.02895668,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 266500.050202,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 249125.005212,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 242375.070229,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 242916.692514,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 296916.696243,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 269208.394457,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 242958.310992,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 244041.671976,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 263708.352577,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 6708.31650496,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6708.31650496,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7750.05901232,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 13749.9300763,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6749.99319017,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6999.99509379,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6791.66987538,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6624.96313453,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6791.66987538,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              4.16406328440625,
              3.8925782064375,
              3.7871104723281253,
              3.79557332053125,
              4.639323378796875,
              4.206381163390625,
              3.79622360925,
              3.813151124625,
              4.1204430090156245
            ],
            "command_buffer_throughput_us_p50": 3.8925782064375,
            "command_buffer_throughput_us_p90": 4.292969606471875,
            "command_buffer_latency_us": [
              6.70831650496,
              6.70831650496,
              7.75005901232,
              13.7499300763,
              6.7499931901700005,
              6.99999509379,
              6.79166987538,
              6.624963134530001,
              6.79166987538
            ],
            "command_buffer_latency_us_p50": 6.79166987538,
            "command_buffer_latency_us_p90": 8.950033225116002
          },
          "compute_throughput_us": [
            4.162109375,
            3.828125,
            3.786453125,
            3.8046875,
            4.099609375,
            4.097015625,
            4.435546875,
            4.332671875,
            4.214203125
          ],
          "compute_throughput_us_p50": 4.099609375,
          "compute_throughput_us_p90": 4.353246875,
          "command_buffer_throughput_us": [
            4.163412086203125,
            3.8281250454062503,
            3.787760761046875,
            3.80598976334375,
            4.100260412093751,
            4.09830681746875,
            4.435547452890625,
            4.332681783125,
            4.21484310209375
          ],
          "command_buffer_throughput_us_p50": 4.100260412093751,
          "command_buffer_throughput_us_p90": 4.353254917078125,
          "compute_latency_us": [
            6.709,
            6.834,
            6.708,
            6.583,
            7.708,
            6.708,
            6.667,
            13.625,
            7.542
          ],
          "compute_latency_us_p50": 6.709,
          "compute_latency_us_p90": 8.8914,
          "command_buffer_latency_us": [
            6.8333465606,
            6.95831840858,
            6.8333465606,
            6.66663981974,
            7.70838232711,
            6.74993498251,
            6.70831650496,
            13.7083698064,
            7.6250289566800005
          ],
          "command_buffer_latency_us_p50": 6.8333465606,
          "command_buffer_latency_us_p90": 8.908379822968001,
          "counter_control_throughput_ratio": 1.0001717110508135,
          "counter_control_latency_ratio": 1.0186380674715563
        },
        "correctness": {
          "max_abs_error": 2.6856349761260334e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 4.35920250931,
        "throughput_us_p90": 4.491279592234,
        "latency_us_p50": 257.584,
        "latency_us_p90": 439.24160000000006
      },
      "slowdown": 0.12657879564317476,
      "device_compute_slowdown": 0.06521465719770972,
      "valid": true,
      "round": 1,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "softmax_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-rsuyegtw/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "7f45b1ebfa78e24e3d477c2ad514fde5c49a3d7b63ed786757d9dc5d1979c8b1",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.797083,
        "capture_ms": 0.068792,
        "compile_ms": 56.075292,
        "allocation_upload_ms": 6.068708,
        "cold_call_ms": 5.129541,
        "warmup_ms": 200.321166,
        "download_ms": 3.284917,
        "repetitions": 448,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          61.5440848214,
          60.6872209821,
          61.3724888393,
          61.8630022321,
          62.6038861607,
          61.8493303571,
          62.9162946429,
          61.5627790179,
          62.326265625
        ],
        "latency_us": [
          665.042,
          1109.208,
          1732.083,
          730.125,
          388.25,
          459.833,
          531.084,
          318.125,
          291.583
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 768,
            "reduction_scalar_rounds": 120,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 24576,
              "global_write_bytes": 24576,
              "private_read_bytes": 147456,
              "private_write_bytes": 49152
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 96,
              "private_read_bytes": 576,
              "private_write_bytes": 192
            },
            "striped_storage_scalars_per_worker": 48,
            "reduction_operations": 2,
            "reduction_elements": 12288,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 168,
            "normalized_kernel_cost": 2016,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 18432,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3580833,
              "compute_span_ns": 3580833,
              "calibration_cpu_ns": 4385375,
              "calibration_gpu_ticks": 4385375,
              "compute_passes": 1,
              "command_buffer_ns": 3580958.34963,
              "command_buffers": 2
            },
            {
              "compute_ns": 3592250,
              "compute_span_ns": 3592250,
              "calibration_cpu_ns": 4261625,
              "calibration_gpu_ticks": 4261625,
              "compute_passes": 1,
              "command_buffer_ns": 3592374.96741,
              "command_buffers": 2
            },
            {
              "compute_ns": 3650166,
              "compute_span_ns": 3650166,
              "calibration_cpu_ns": 4476167,
              "calibration_gpu_ticks": 4476167,
              "compute_passes": 1,
              "command_buffer_ns": 3650250.02975,
              "command_buffers": 2
            },
            {
              "compute_ns": 3644708,
              "compute_span_ns": 3644708,
              "calibration_cpu_ns": 4288875,
              "calibration_gpu_ticks": 4288875,
              "compute_passes": 1,
              "command_buffer_ns": 3644708.31119,
              "command_buffers": 2
            },
            {
              "compute_ns": 3602500,
              "compute_span_ns": 3602500,
              "calibration_cpu_ns": 4419708,
              "calibration_gpu_ticks": 4419708,
              "compute_passes": 1,
              "command_buffer_ns": 3602541.69209,
              "command_buffers": 2
            },
            {
              "compute_ns": 4130500,
              "compute_span_ns": 4130500,
              "calibration_cpu_ns": 4703250,
              "calibration_gpu_ticks": 4703250,
              "compute_passes": 1,
              "command_buffer_ns": 4130541.69621,
              "command_buffers": 2
            },
            {
              "compute_ns": 4167250,
              "compute_span_ns": 4167250,
              "calibration_cpu_ns": 4699083,
              "calibration_gpu_ticks": 4699083,
              "compute_passes": 1,
              "command_buffer_ns": 4167291.56859,
              "command_buffers": 2
            },
            {
              "compute_ns": 3615625,
              "compute_span_ns": 3615625,
              "calibration_cpu_ns": 4436417,
              "calibration_gpu_ticks": 4436417,
              "compute_passes": 1,
              "command_buffer_ns": 3615749.94167,
              "command_buffers": 2
            },
            {
              "compute_ns": 3658583,
              "compute_span_ns": 3658583,
              "calibration_cpu_ns": 4544583,
              "calibration_gpu_ticks": 4544583,
              "compute_passes": 1,
              "command_buffer_ns": 3658708.35958,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 71834,
              "compute_span_ns": 71834,
              "calibration_cpu_ns": 640458,
              "calibration_gpu_ticks": 640458,
              "compute_passes": 1,
              "command_buffer_ns": 71874.9943189,
              "command_buffers": 2
            },
            {
              "compute_ns": 60000,
              "compute_span_ns": 60000,
              "calibration_cpu_ns": 513291,
              "calibration_gpu_ticks": 513291,
              "compute_passes": 1,
              "command_buffer_ns": 60083.3445787,
              "command_buffers": 2
            },
            {
              "compute_ns": 59583,
              "compute_span_ns": 59583,
              "calibration_cpu_ns": 486708,
              "calibration_gpu_ticks": 486708,
              "compute_passes": 1,
              "command_buffer_ns": 59583.3407715,
              "command_buffers": 2
            },
            {
              "compute_ns": 62958,
              "compute_span_ns": 62958,
              "calibration_cpu_ns": 482875,
              "calibration_gpu_ticks": 482875,
              "compute_passes": 1,
              "command_buffer_ns": 63000.0140518,
              "command_buffers": 2
            },
            {
              "compute_ns": 67625,
              "compute_span_ns": 67625,
              "calibration_cpu_ns": 600042,
              "calibration_gpu_ticks": 600042,
              "compute_passes": 1,
              "command_buffer_ns": 67749.992013,
              "command_buffers": 2
            },
            {
              "compute_ns": 63917,
              "compute_span_ns": 63917,
              "calibration_cpu_ns": 522917,
              "calibration_gpu_ticks": 522917,
              "compute_passes": 1,
              "command_buffer_ns": 64041.6983515,
              "command_buffers": 2
            },
            {
              "compute_ns": 58708,
              "compute_span_ns": 58708,
              "calibration_cpu_ns": 576834,
              "calibration_gpu_ticks": 576834,
              "compute_passes": 1,
              "command_buffer_ns": 58708.3632126,
              "command_buffers": 2
            },
            {
              "compute_ns": 92208,
              "compute_span_ns": 92208,
              "calibration_cpu_ns": 789333,
              "calibration_gpu_ticks": 789333,
              "compute_passes": 1,
              "command_buffer_ns": 92249.9457374,
              "command_buffers": 2
            },
            {
              "compute_ns": 62250,
              "compute_span_ns": 62250,
              "calibration_cpu_ns": 510917,
              "calibration_gpu_ticks": 510917,
              "compute_passes": 1,
              "command_buffer_ns": 62374.9801889,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 3740958.28738,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3724375.0412,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4131708.35236,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4206833.31322,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3607749.99717,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3669583.29687,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3600541.67686,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3657791.64713,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4121125.03545,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 95958.2976066,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 58625.0098422,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 58625.0098422,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 97041.7167991,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 61333.2958892,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 58999.9835938,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 59583.3407715,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 73041.6504666,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 66041.5971652,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              58.4524732403125,
              58.19336001875,
              64.55794300562499,
              65.7317705190625,
              56.37109370578125,
              57.33723901359375,
              56.2584637009375,
              57.15299448640625,
              64.39257867890625
            ],
            "command_buffer_throughput_us_p50": 58.19336001875,
            "command_buffer_throughput_us_p90": 64.79270850831249,
            "command_buffer_latency_us": [
              95.9582976066,
              58.6250098422,
              58.6250098422,
              97.0417167991,
              61.3332958892,
              58.9999835938,
              59.5833407715,
              73.04165046659999,
              66.0415971652
            ],
            "command_buffer_latency_us_p50": 61.3332958892,
            "command_buffer_latency_us_p90": 96.1749814451
          },
          "compute_throughput_us": [
            55.950515625,
            56.12890625,
            57.03384375,
            56.9485625,
            56.2890625,
            64.5390625,
            65.11328125,
            56.494140625,
            57.165359375
          ],
          "compute_throughput_us_p50": 56.9485625,
          "compute_throughput_us_p90": 64.65390625,
          "command_buffer_throughput_us": [
            55.95247421296875,
            56.13085886578125,
            57.03515671484375,
            56.948567362343745,
            56.28971393890625,
            64.53971400328125,
            65.11393075921875,
            56.49609283859375,
            57.1673181184375
          ],
          "command_buffer_throughput_us_p50": 56.948567362343745,
          "command_buffer_throughput_us_p90": 64.65455735446875,
          "compute_latency_us": [
            71.834,
            60.0,
            59.583,
            62.958,
            67.625,
            63.917,
            58.708,
            92.208,
            62.25
          ],
          "compute_latency_us_p50": 62.958,
          "compute_latency_us_p90": 75.9088,
          "command_buffer_latency_us": [
            71.87499431890001,
            60.0833445787,
            59.5833407715,
            63.000014051799994,
            67.749992013,
            64.0416983515,
            58.7083632126,
            92.24994573740001,
            62.374980188900004
          ],
          "command_buffer_latency_us_p50": 63.000014051799994,
          "command_buffer_latency_us_p90": 75.94998460260001,
          "counter_control_throughput_ratio": 0.9645577922927254,
          "counter_control_latency_ratio": 1.0163467934910293
        },
        "correctness": {
          "max_abs_error": 7.116789644580346e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 61.8493303571,
        "throughput_us_p90": 62.66636785714,
        "latency_us_p50": 531.084,
        "latency_us_p90": 1233.7830000000001
      },
      "torch": {
        "cold_call_ms": 7.372375,
        "warmup_ms": 201.292166,
        "repetitions": 174,
        "throughput_us": [
          140.53735632183907,
          140.58213218390804,
          142.03113218390806,
          140.37452298850573,
          143.16522988505747,
          141.76364942528735,
          143.26005747126436,
          142.41235632183907,
          140.5941091954023
        ],
        "latency_us": [
          775.625,
          837.8330000000001,
          607.333,
          488.666,
          431.458,
          410.25,
          402.33299999999997,
          400.875,
          430.375
        ],
        "throughput_us_p50": 141.76364942528735,
        "throughput_us_p90": 143.18419540229885,
        "latency_us_p50": 431.458,
        "latency_us_p90": 788.0666,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 8499166.753608733,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8451375.062577426,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8443083.323072642,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8399166.632443666,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8440166.65359959,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8530125.02193451,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8437125.070486218,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8406875.072978437,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8505875.011906028,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 154874.98603761196,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 135458.36554840207,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 138500.0068694353,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 136874.96539205313,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 137666.7059957981,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 152541.6737422347,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 137083.29061046243,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 140458.345413208,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 172041.64760187268,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              132.79948052513646,
              132.0527353527723,
              131.92317692301003,
              131.2369786319323,
              131.8776039624936,
              133.2832034677267,
              131.83007922634715,
              131.35742301528808,
              132.90429706103168
            ],
            "command_buffer_throughput_us_p50": 131.92317692301003,
            "command_buffer_throughput_us_p90": 132.9800783423707,
            "command_buffer_latency_us": [
              154.87498603761196,
              135.45836554840207,
              138.5000068694353,
              136.87496539205313,
              137.6667059957981,
              152.5416737422347,
              137.08329061046243,
              140.458345413208,
              172.04164760187268
            ],
            "command_buffer_latency_us_p50": 138.5000068694353,
            "command_buffer_latency_us_p90": 158.3083183504641
          },
          "throughput": [
            {
              "compute_ns": 9480960.0,
              "compute_span_ns": 10938625.0,
              "command_buffer_ns": 10686166.64269939,
              "calibration_cpu_ns": 12150583.0,
              "calibration_gpu_ticks": 12150583.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9208122.0,
              "compute_span_ns": 10690041.0,
              "command_buffer_ns": 10448000.044561923,
              "calibration_cpu_ns": 12057458.0,
              "calibration_gpu_ticks": 12057458.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9328543.0,
              "compute_span_ns": 10777792.0,
              "command_buffer_ns": 10527916.660066694,
              "calibration_cpu_ns": 11767667.0,
              "calibration_gpu_ticks": 11767667.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9337958.0,
              "compute_span_ns": 10880458.0,
              "command_buffer_ns": 10634999.896865338,
              "calibration_cpu_ns": 11871125.0,
              "calibration_gpu_ticks": 11871125.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9414372.0,
              "compute_span_ns": 10875334.0,
              "command_buffer_ns": 10636250.081006438,
              "calibration_cpu_ns": 12045584.0,
              "calibration_gpu_ticks": 12045584.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9325374.0,
              "compute_span_ns": 10914875.0,
              "command_buffer_ns": 10647833.172697574,
              "calibration_cpu_ns": 11969167.0,
              "calibration_gpu_ticks": 11969167.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9206459.0,
              "compute_span_ns": 10773667.0,
              "command_buffer_ns": 10539666.63312167,
              "calibration_cpu_ns": 11963459.0,
              "calibration_gpu_ticks": 11963459.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9404458.0,
              "compute_span_ns": 10910959.0,
              "command_buffer_ns": 10685624.96220693,
              "calibration_cpu_ns": 12035792.0,
              "calibration_gpu_ticks": 12035792.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9175123.0,
              "compute_span_ns": 10642416.0,
              "command_buffer_ns": 10377583.093941212,
              "calibration_cpu_ns": 11678500.0,
              "calibration_gpu_ticks": 11678500.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 147959.0,
              "compute_span_ns": 147959.0,
              "command_buffer_ns": 148000.0209994614,
              "calibration_cpu_ns": 781500.0,
              "calibration_gpu_ticks": 781500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 138209.0,
              "compute_span_ns": 138209.0,
              "command_buffer_ns": 138250.00496581197,
              "calibration_cpu_ns": 606667.0,
              "calibration_gpu_ticks": 606667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 159416.0,
              "compute_span_ns": 159416.0,
              "command_buffer_ns": 159375.02030283213,
              "calibration_cpu_ns": 1596834.0,
              "calibration_gpu_ticks": 1596834.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 148291.0,
              "compute_span_ns": 148291.0,
              "command_buffer_ns": 148250.02290308475,
              "calibration_cpu_ns": 618500.0,
              "calibration_gpu_ticks": 618500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 135542.0,
              "compute_span_ns": 135542.0,
              "command_buffer_ns": 135541.66071116924,
              "calibration_cpu_ns": 593083.0,
              "calibration_gpu_ticks": 593083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 157666.0,
              "compute_span_ns": 157666.0,
              "command_buffer_ns": 157625.00697746873,
              "calibration_cpu_ns": 1316458.0,
              "calibration_gpu_ticks": 1316458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 141167.0,
              "compute_span_ns": 141167.0,
              "command_buffer_ns": 141166.674438864,
              "calibration_cpu_ns": 650208.0,
              "calibration_gpu_ticks": 650208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 138166.0,
              "compute_span_ns": 138166.0,
              "command_buffer_ns": 138125.03311783075,
              "calibration_cpu_ns": 686833.0,
              "calibration_gpu_ticks": 686833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 151292.0,
              "compute_span_ns": 151292.0,
              "command_buffer_ns": 151291.664224118,
              "calibration_cpu_ns": 860625.0,
              "calibration_gpu_ticks": 860625.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            148.14,
            143.87690625,
            145.758484375,
            145.90559375,
            147.0995625,
            145.70896875,
            143.850921875,
            146.94465625,
            143.361296875
          ],
          "compute_throughput_us_p50": 145.758484375,
          "compute_throughput_us_p90": 147.30765,
          "command_buffer_throughput_us": [
            166.97135379217798,
            163.25000069628004,
            164.4986978135421,
            166.1718733885209,
            166.1914075157256,
            166.3723933233996,
            164.68229114252608,
            166.96289003448328,
            162.14973584283143
          ],
          "command_buffer_throughput_us_p50": 166.1718733885209,
          "command_buffer_throughput_us_p90": 166.96458278602222,
          "compute_latency_us": [
            147.959,
            138.209,
            159.416,
            148.291,
            135.542,
            157.666,
            141.167,
            138.166,
            151.292
          ],
          "compute_latency_us_p50": 147.959,
          "compute_latency_us_p90": 158.016,
          "command_buffer_latency_us": [
            148.0000209994614,
            138.25000496581197,
            159.37502030283213,
            148.25002290308475,
            135.54166071116924,
            157.62500697746873,
            141.166674438864,
            138.12503311783075,
            151.291664224118
          ],
          "command_buffer_latency_us_p50": 148.0000209994614,
          "command_buffer_latency_us_p90": 157.9750096425414,
          "counter_control_throughput_ratio": 1.2492011846535642,
          "counter_control_latency_ratio": 1.0206088373066364
        },
        "download_ms": 0.606625,
        "allocation_upload_ms": 16.667333,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 7.553236784944503e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.43628483470789875,
      "device_compute_slowdown": 0.3907049578910662,
      "valid": true,
      "round": 2,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "softmax_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.696209,
        "warmup_ms": 201.642833,
        "repetitions": 186,
        "throughput_us": [
          140.6478494623656,
          139.60035483870968,
          139.33870967741936,
          139.15098387096776,
          143.26187634408603,
          142.84050537634408,
          139.3709677419355,
          141.4540806451613,
          138.827064516129
        ],
        "latency_us": [
          564.2909999999999,
          420.25,
          401.125,
          430.084,
          421.58299999999997,
          416.708,
          539.625,
          668.5419999999999,
          935.5
        ],
        "throughput_us_p50": 139.60035483870968,
        "throughput_us_p90": 142.92477956989248,
        "latency_us_p50": 430.084,
        "latency_us_p90": 721.9336,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 8392624.964471906,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8385916.764382273,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8404375.053942204,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8422750.048339367,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8424624.97530505,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8413958.363234997,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8538291.67317599,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8396375.051233917,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8363166.7657755315,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 151625.01949816942,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 136499.99164044857,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 137874.9730065465,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 169250.0081844628,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 141499.9715052545,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 135916.6926704347,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 134749.97831508517,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 138916.6573062539,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 140124.99013915658,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              131.13476506987354,
              131.02994944347301,
              131.31836021784693,
              131.6054695053026,
              131.63476523914142,
              131.46809942554682,
              133.41080739337485,
              131.19336017552996,
              130.67448071524268
            ],
            "command_buffer_throughput_us_p50": 131.31836021784693,
            "command_buffer_throughput_us_p90": 131.9899736699881,
            "command_buffer_latency_us": [
              151.62501949816942,
              136.49999164044857,
              137.8749730065465,
              169.25000818446279,
              141.4999715052545,
              135.9166926704347,
              134.74997831508517,
              138.9166573062539,
              140.12499013915658
            ],
            "command_buffer_latency_us_p50": 138.9166573062539,
            "command_buffer_latency_us_p90": 155.1500172354281
          },
          "throughput": [
            {
              "compute_ns": 9106374.0,
              "compute_span_ns": 10602208.0,
              "command_buffer_ns": 10328458.389267325,
              "calibration_cpu_ns": 11537416.0,
              "calibration_gpu_ticks": 11537416.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9096831.0,
              "compute_span_ns": 10583583.0,
              "command_buffer_ns": 10337083.251215518,
              "calibration_cpu_ns": 11574083.0,
              "calibration_gpu_ticks": 11574083.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9097334.0,
              "compute_span_ns": 10556208.0,
              "command_buffer_ns": 10317916.807252914,
              "calibration_cpu_ns": 11491958.0,
              "calibration_gpu_ticks": 11491958.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9345548.0,
              "compute_span_ns": 10706583.0,
              "command_buffer_ns": 10502708.377316594,
              "calibration_cpu_ns": 11702583.0,
              "calibration_gpu_ticks": 11702583.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9059543.0,
              "compute_span_ns": 10564834.0,
              "command_buffer_ns": 10326291.725505143,
              "calibration_cpu_ns": 11576708.0,
              "calibration_gpu_ticks": 11576708.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9102460.0,
              "compute_span_ns": 10630500.0,
              "command_buffer_ns": 10419333.353638649,
              "calibration_cpu_ns": 11653041.0,
              "calibration_gpu_ticks": 11653041.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9096376.0,
              "compute_span_ns": 10604458.0,
              "command_buffer_ns": 10346916.678827256,
              "calibration_cpu_ns": 11848125.0,
              "calibration_gpu_ticks": 11848125.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9137995.0,
              "compute_span_ns": 10484666.0,
              "command_buffer_ns": 10255583.270918578,
              "calibration_cpu_ns": 11721916.0,
              "calibration_gpu_ticks": 11721916.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 9128920.0,
              "compute_span_ns": 10689167.0,
              "command_buffer_ns": 10423041.705507785,
              "calibration_cpu_ns": 11688250.0,
              "calibration_gpu_ticks": 11688250.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 140916.0,
              "compute_span_ns": 140916.0,
              "command_buffer_ns": 140874.9958500266,
              "calibration_cpu_ns": 682584.0,
              "calibration_gpu_ticks": 682584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 136583.0,
              "compute_span_ns": 136583.0,
              "command_buffer_ns": 136541.6683256626,
              "calibration_cpu_ns": 662583.0,
              "calibration_gpu_ticks": 662583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 135417.0,
              "compute_span_ns": 135417.0,
              "command_buffer_ns": 135458.36554840207,
              "calibration_cpu_ns": 636625.0,
              "calibration_gpu_ticks": 636625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 138875.0,
              "compute_span_ns": 138875.0,
              "command_buffer_ns": 138875.03882870078,
              "calibration_cpu_ns": 593458.0,
              "calibration_gpu_ticks": 593458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 142958.0,
              "compute_span_ns": 142958.0,
              "command_buffer_ns": 142958.30624178052,
              "calibration_cpu_ns": 671417.0,
              "calibration_gpu_ticks": 671417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 140334.0,
              "compute_span_ns": 140334.0,
              "command_buffer_ns": 140374.99204277992,
              "calibration_cpu_ns": 651334.0,
              "calibration_gpu_ticks": 651334.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 137833.0,
              "compute_span_ns": 137833.0,
              "command_buffer_ns": 137791.67784377933,
              "calibration_cpu_ns": 699417.0,
              "calibration_gpu_ticks": 699417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 153709.0,
              "compute_span_ns": 153709.0,
              "command_buffer_ns": 153749.94836747646,
              "calibration_cpu_ns": 899292.0,
              "calibration_gpu_ticks": 899292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 159291.0,
              "compute_span_ns": 159291.0,
              "command_buffer_ns": 159249.99024719,
              "calibration_cpu_ns": 709625.0,
              "calibration_gpu_ticks": 709625.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            142.28709375,
            142.137984375,
            142.14584375,
            146.0241875,
            141.555359375,
            142.2259375,
            142.130875,
            142.781171875,
            142.639375
          ],
          "compute_throughput_us_p50": 142.2259375,
          "compute_throughput_us_p90": 143.429775,
          "command_buffer_throughput_us": [
            161.38216233230196,
            161.51692580024246,
            161.21745011332678,
            164.10481839557178,
            161.34830821101787,
            162.8020836506039,
            161.67057310667587,
            160.24348860810278,
            162.86002664855914
          ],
          "command_buffer_throughput_us_p50": 161.51692580024246,
          "command_buffer_throughput_us_p90": 163.10898499796167,
          "compute_latency_us": [
            140.916,
            136.583,
            135.417,
            138.875,
            142.958,
            140.334,
            137.833,
            153.709,
            159.291
          ],
          "compute_latency_us_p50": 140.334,
          "compute_latency_us_p90": 154.8254,
          "command_buffer_latency_us": [
            140.8749958500266,
            136.5416683256626,
            135.45836554840207,
            138.87503882870078,
            142.95830624178052,
            140.37499204277992,
            137.79167784377933,
            153.74994836747646,
            159.24999024719
          ],
          "command_buffer_latency_us_p50": 140.37499204277992,
          "command_buffer_latency_us_p90": 154.84995674341917,
          "counter_control_throughput_ratio": 1.2306588740698279,
          "counter_control_latency_ratio": 1.0103062546303896
        },
        "download_ms": 0.649708,
        "allocation_upload_ms": 0.903458,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 7.553236784944503e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-_xq6v1uq/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "5e168786762f89acc561194a669a3725da4fe77bed5800b1d573576d096bb3fc",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.54575,
        "capture_ms": 0.087208,
        "compile_ms": 53.446041,
        "allocation_upload_ms": 6.457625,
        "cold_call_ms": 7.234041,
        "warmup_ms": 200.957542,
        "download_ms": 3.810375,
        "repetitions": 461,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          63.795824295,
          64.215472885,
          63.8664143167,
          63.9351952278,
          65.0895704989,
          63.4912342733,
          64.4941236443,
          65.1065618221,
          64.0964381779
        ],
        "latency_us": [
          362.416,
          293.25,
          301.166,
          353.625,
          295.375,
          297.833,
          301.791,
          303.542,
          288.792
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 384,
            "reduction_scalar_rounds": 120,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 24576,
              "global_write_bytes": 24576,
              "private_read_bytes": 147456,
              "private_write_bytes": 49152
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 96,
              "private_read_bytes": 576,
              "private_write_bytes": 192
            },
            "striped_storage_scalars_per_worker": 48,
            "reduction_operations": 2,
            "reduction_elements": 12288,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 160,
            "normalized_kernel_cost": 1920,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 18432,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4065500,
              "compute_span_ns": 4065500,
              "calibration_cpu_ns": 4692583,
              "calibration_gpu_ticks": 4692583,
              "compute_passes": 1,
              "command_buffer_ns": 4065625.0203,
              "command_buffers": 2
            },
            {
              "compute_ns": 4029167,
              "compute_span_ns": 4029167,
              "calibration_cpu_ns": 4616500,
              "calibration_gpu_ticks": 4616500,
              "compute_passes": 1,
              "command_buffer_ns": 4029250.06347,
              "command_buffers": 2
            },
            {
              "compute_ns": 4030541,
              "compute_span_ns": 4030541,
              "calibration_cpu_ns": 4697208,
              "calibration_gpu_ticks": 4697208,
              "compute_passes": 1,
              "command_buffer_ns": 4030541.63326,
              "command_buffers": 2
            },
            {
              "compute_ns": 4022917,
              "compute_span_ns": 4022917,
              "calibration_cpu_ns": 4737125,
              "calibration_gpu_ticks": 4737125,
              "compute_passes": 1,
              "command_buffer_ns": 4023041.69256,
              "command_buffers": 2
            },
            {
              "compute_ns": 4030583,
              "compute_span_ns": 4030583,
              "calibration_cpu_ns": 4628667,
              "calibration_gpu_ticks": 4628667,
              "compute_passes": 1,
              "command_buffer_ns": 4030708.34,
              "command_buffers": 2
            },
            {
              "compute_ns": 4053667,
              "compute_span_ns": 4053667,
              "calibration_cpu_ns": 4540667,
              "calibration_gpu_ticks": 4540667,
              "compute_passes": 1,
              "command_buffer_ns": 4053791.63567,
              "command_buffers": 2
            },
            {
              "compute_ns": 4022166,
              "compute_span_ns": 4022166,
              "calibration_cpu_ns": 4692792,
              "calibration_gpu_ticks": 4692792,
              "compute_passes": 1,
              "command_buffer_ns": 4022249.95196,
              "command_buffers": 2
            },
            {
              "compute_ns": 4084500,
              "compute_span_ns": 4084500,
              "calibration_cpu_ns": 4588208,
              "calibration_gpu_ticks": 4588208,
              "compute_passes": 1,
              "command_buffer_ns": 4084583.31367,
              "command_buffers": 2
            },
            {
              "compute_ns": 4037584,
              "compute_span_ns": 4037584,
              "calibration_cpu_ns": 4664208,
              "calibration_gpu_ticks": 4664208,
              "compute_passes": 1,
              "command_buffer_ns": 4037750.01178,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 98959,
              "compute_span_ns": 98959,
              "calibration_cpu_ns": 675666,
              "calibration_gpu_ticks": 675666,
              "compute_passes": 1,
              "command_buffer_ns": 99083.292298,
              "command_buffers": 2
            },
            {
              "compute_ns": 76792,
              "compute_span_ns": 76792,
              "calibration_cpu_ns": 551916,
              "calibration_gpu_ticks": 551916,
              "compute_passes": 1,
              "command_buffer_ns": 76958.3275542,
              "command_buffers": 2
            },
            {
              "compute_ns": 63917,
              "compute_span_ns": 63917,
              "calibration_cpu_ns": 481583,
              "calibration_gpu_ticks": 481583,
              "compute_passes": 1,
              "command_buffer_ns": 64000.0216663,
              "command_buffers": 2
            },
            {
              "compute_ns": 63084,
              "compute_span_ns": 63084,
              "calibration_cpu_ns": 479000,
              "calibration_gpu_ticks": 479000,
              "compute_passes": 1,
              "command_buffer_ns": 63166.662585,
              "command_buffers": 2
            },
            {
              "compute_ns": 76125,
              "compute_span_ns": 76125,
              "calibration_cpu_ns": 652792,
              "calibration_gpu_ticks": 652792,
              "compute_passes": 1,
              "command_buffer_ns": 76208.3218433,
              "command_buffers": 2
            },
            {
              "compute_ns": 65333,
              "compute_span_ns": 65333,
              "calibration_cpu_ns": 560500,
              "calibration_gpu_ticks": 560500,
              "compute_passes": 1,
              "command_buffer_ns": 65416.7379253,
              "command_buffers": 2
            },
            {
              "compute_ns": 63959,
              "compute_span_ns": 63959,
              "calibration_cpu_ns": 465542,
              "calibration_gpu_ticks": 465542,
              "compute_passes": 1,
              "command_buffer_ns": 64041.5819362,
              "command_buffers": 2
            },
            {
              "compute_ns": 67333,
              "compute_span_ns": 67333,
              "calibration_cpu_ns": 786959,
              "calibration_gpu_ticks": 786959,
              "compute_passes": 1,
              "command_buffer_ns": 67333.3415762,
              "command_buffers": 2
            },
            {
              "compute_ns": 66250,
              "compute_span_ns": 66250,
              "calibration_cpu_ns": 521500,
              "calibration_gpu_ticks": 521500,
              "compute_passes": 1,
              "command_buffer_ns": 66291.6572765,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4083125.03714,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4003374.95375,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4024375.05545,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4059291.67755,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4013958.32887,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4010125.00515,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4029375.03532,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3973624.96005,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4010625.00896,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 81999.9258965,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 68125.0239722,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63708.2848698,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 64166.6701995,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 112499.925308,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63291.6926406,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63166.7207927,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 113374.961074,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76166.7033657,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              63.7988287053125,
              62.55273365234375,
              62.880860241406246,
              63.426432461718754,
              62.71809888859375,
              62.65820320546875,
              62.958984926875,
              62.08789000078125,
              62.666015765
            ],
            "command_buffer_throughput_us_p50": 62.71809888859375,
            "command_buffer_throughput_us_p90": 63.50091171043751,
            "command_buffer_latency_us": [
              81.9999258965,
              68.1250239722,
              63.7082848698,
              64.16667019949999,
              112.499925308,
              63.2916926406,
              63.1667207927,
              113.37496107400001,
              76.1667033657
            ],
            "command_buffer_latency_us_p50": 68.1250239722,
            "command_buffer_latency_us_p90": 112.6749324612
          },
          "compute_throughput_us": [
            63.5234375,
            62.955734375,
            62.977203125,
            62.858078125,
            62.977859375,
            63.338546875,
            62.84634375,
            63.8203125,
            63.08725
          ],
          "compute_throughput_us_p50": 62.977859375,
          "compute_throughput_us_p90": 63.5828125,
          "command_buffer_throughput_us": [
            63.5253909421875,
            62.95703224171875,
            62.977213019687504,
            62.86002644625,
            62.979817812499995,
            63.340494307343754,
            62.847655499375,
            63.821614276093754,
            63.0898439340625
          ],
          "command_buffer_throughput_us_p50": 62.979817812499995,
          "command_buffer_throughput_us_p90": 63.58463560896875,
          "compute_latency_us": [
            98.959,
            76.792,
            63.917,
            63.084,
            76.125,
            65.333,
            63.959,
            67.333,
            66.25
          ],
          "compute_latency_us_p50": 66.25,
          "compute_latency_us_p90": 81.22540000000001,
          "command_buffer_latency_us": [
            99.083292298,
            76.95832755420001,
            64.0000216663,
            63.166662585,
            76.2083218433,
            65.4167379253,
            64.0415819362,
            67.3333415762,
            66.2916572765
          ],
          "command_buffer_latency_us_p50": 66.2916572765,
          "command_buffer_latency_us_p90": 81.38332050296002,
          "counter_control_throughput_ratio": 1.0041729409619247,
          "counter_control_latency_ratio": 1.00457925993607
        },
        "correctness": {
          "max_abs_error": 7.116789644580346e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 64.0964381779,
        "throughput_us_p90": 65.09296876354,
        "latency_us_p50": 301.166,
        "latency_us_p90": 355.3832
      },
      "slowdown": 0.4591423728969401,
      "device_compute_slowdown": 0.44280150640596067,
      "valid": true,
      "round": 2,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "softmax_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-lb7f0n_f/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "117627a7b60c4cee3b9bce9542bc5f280f025f1ad809d52b8d22cb3cdd24d0d1",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 21.10725,
        "capture_ms": 0.0675,
        "compile_ms": 55.151792,
        "allocation_upload_ms": 6.456583,
        "cold_call_ms": 10.113167,
        "warmup_ms": 200.103166,
        "download_ms": 3.350833,
        "repetitions": 508,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          54.2895354331,
          54.4155177165,
          54.2079232283,
          54.3626141732,
          54.1437814961,
          54.9537401575,
          54.4613681102,
          55.1861870079,
          54.7763287402
        ],
        "latency_us": [
          403.834,
          281.667,
          300.125,
          358.25,
          365.417,
          366.167,
          303.416,
          360,
          272.959
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 512,
            "reduction_scalar_rounds": 80,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 16384,
              "global_write_bytes": 16384,
              "private_read_bytes": 98304,
              "private_write_bytes": 32768
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 64,
              "global_write_bytes": 64,
              "private_read_bytes": 384,
              "private_write_bytes": 128
            },
            "striped_storage_scalars_per_worker": 32,
            "reduction_operations": 2,
            "reduction_elements": 8192,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 120,
            "normalized_kernel_cost": 1920,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3569834,
              "compute_span_ns": 3569834,
              "calibration_cpu_ns": 4317541,
              "calibration_gpu_ticks": 4317541,
              "compute_passes": 1,
              "command_buffer_ns": 3569916.64739,
              "command_buffers": 2
            },
            {
              "compute_ns": 3599417,
              "compute_span_ns": 3599417,
              "calibration_cpu_ns": 4433833,
              "calibration_gpu_ticks": 4433833,
              "compute_passes": 1,
              "command_buffer_ns": 3599541.72745,
              "command_buffers": 2
            },
            {
              "compute_ns": 3432667,
              "compute_span_ns": 3432667,
              "calibration_cpu_ns": 4092375,
              "calibration_gpu_ticks": 4092375,
              "compute_passes": 1,
              "command_buffer_ns": 3432708.32673,
              "command_buffers": 2
            },
            {
              "compute_ns": 3460459,
              "compute_span_ns": 3460459,
              "calibration_cpu_ns": 3985000,
              "calibration_gpu_ticks": 3985000,
              "compute_passes": 1,
              "command_buffer_ns": 3460583.27705,
              "command_buffers": 2
            },
            {
              "compute_ns": 3500042,
              "compute_span_ns": 3500042,
              "calibration_cpu_ns": 3992208,
              "calibration_gpu_ticks": 3992208,
              "compute_passes": 1,
              "command_buffer_ns": 3500125.02167,
              "command_buffers": 2
            },
            {
              "compute_ns": 3394750,
              "compute_span_ns": 3394750,
              "calibration_cpu_ns": 4179250,
              "calibration_gpu_ticks": 4179250,
              "compute_passes": 1,
              "command_buffer_ns": 3394750.0051,
              "command_buffers": 2
            },
            {
              "compute_ns": 3537459,
              "compute_span_ns": 3537459,
              "calibration_cpu_ns": 4071375,
              "calibration_gpu_ticks": 4071375,
              "compute_passes": 1,
              "command_buffer_ns": 3537583.33949,
              "command_buffers": 2
            },
            {
              "compute_ns": 3364333,
              "compute_span_ns": 3364333,
              "calibration_cpu_ns": 4092250,
              "calibration_gpu_ticks": 4092250,
              "compute_passes": 1,
              "command_buffer_ns": 3364333.30085,
              "command_buffers": 2
            },
            {
              "compute_ns": 3374916,
              "compute_span_ns": 3374916,
              "calibration_cpu_ns": 3983500,
              "calibration_gpu_ticks": 3983500,
              "compute_passes": 1,
              "command_buffer_ns": 3374916.67597,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 105167,
              "compute_span_ns": 105167,
              "calibration_cpu_ns": 757667,
              "calibration_gpu_ticks": 757667,
              "compute_passes": 1,
              "command_buffer_ns": 105250.044726,
              "command_buffers": 2
            },
            {
              "compute_ns": 69500,
              "compute_span_ns": 69500,
              "calibration_cpu_ns": 588459,
              "calibration_gpu_ticks": 588459,
              "compute_passes": 1,
              "command_buffer_ns": 69541.6820236,
              "command_buffers": 2
            },
            {
              "compute_ns": 54666,
              "compute_span_ns": 54666,
              "calibration_cpu_ns": 581458,
              "calibration_gpu_ticks": 581458,
              "compute_passes": 1,
              "command_buffer_ns": 54624.9793842,
              "command_buffers": 2
            },
            {
              "compute_ns": 62958,
              "compute_span_ns": 62958,
              "calibration_cpu_ns": 569417,
              "calibration_gpu_ticks": 569417,
              "compute_passes": 1,
              "command_buffer_ns": 63041.6325293,
              "command_buffers": 2
            },
            {
              "compute_ns": 59292,
              "compute_span_ns": 59292,
              "calibration_cpu_ns": 528208,
              "calibration_gpu_ticks": 528208,
              "compute_passes": 1,
              "command_buffer_ns": 59416.750446,
              "command_buffers": 2
            },
            {
              "compute_ns": 54042,
              "compute_span_ns": 54042,
              "calibration_cpu_ns": 487292,
              "calibration_gpu_ticks": 487292,
              "compute_passes": 1,
              "command_buffer_ns": 54166.7104699,
              "command_buffers": 2
            },
            {
              "compute_ns": 94958,
              "compute_span_ns": 94958,
              "calibration_cpu_ns": 719625,
              "calibration_gpu_ticks": 719625,
              "compute_passes": 1,
              "command_buffer_ns": 94916.6715145,
              "command_buffers": 2
            },
            {
              "compute_ns": 66250,
              "compute_span_ns": 66250,
              "calibration_cpu_ns": 553833,
              "calibration_gpu_ticks": 553833,
              "compute_passes": 1,
              "command_buffer_ns": 66291.6572765,
              "command_buffers": 2
            },
            {
              "compute_ns": 54208,
              "compute_span_ns": 54208,
              "calibration_cpu_ns": 445625,
              "calibration_gpu_ticks": 445625,
              "compute_passes": 1,
              "command_buffer_ns": 54208.2707398,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 3597499.91912,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3458499.90845,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3483166.68533,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3377666.69691,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3478083.35209,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3442374.98939,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3438250.04529,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3427666.67018,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3454499.9944,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 61208.3240412,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 62375.0383966,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 58958.3069086,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 101291.690953,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 66875.0144541,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 53791.6785106,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 56375.0509173,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54333.3590031,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 58041.6526645,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              56.21093623625001,
              54.03906106953125,
              54.42447945828125,
              52.77604213921875,
              54.34505237640625,
              53.78710920921875,
              53.72265695765625,
              53.5572917215625,
              53.9765624125
            ],
            "command_buffer_throughput_us_p50": 53.9765624125,
            "command_buffer_throughput_us_p90": 54.781770813875,
            "command_buffer_latency_us": [
              61.2083240412,
              62.375038396600004,
              58.9583069086,
              101.291690953,
              66.87501445410001,
              53.7916785106,
              56.3750509173,
              54.3333590031,
              58.0416526645
            ],
            "command_buffer_latency_us_p50": 58.9583069086,
            "command_buffer_latency_us_p90": 73.75834975388001
          },
          "compute_throughput_us": [
            55.77865625,
            56.240890625,
            53.635421875,
            54.069671875,
            54.68815625,
            53.04296875,
            55.272796875,
            52.567703125,
            52.7330625
          ],
          "compute_throughput_us_p50": 54.069671875,
          "compute_throughput_us_p90": 55.871103125,
          "command_buffer_throughput_us": [
            55.77994761546875,
            56.24283949140625,
            53.63606760515625,
            54.07161370390625,
            54.68945346359375,
            53.0429688296875,
            55.27473967953125,
            52.567707825781255,
            52.73307306203125
          ],
          "command_buffer_throughput_us_p50": 54.07161370390625,
          "command_buffer_throughput_us_p90": 55.87252599065625,
          "compute_latency_us": [
            105.167,
            69.5,
            54.666,
            62.958,
            59.292,
            54.042,
            94.958,
            66.25,
            54.208
          ],
          "compute_latency_us_p50": 62.958,
          "compute_latency_us_p90": 96.99980000000001,
          "command_buffer_latency_us": [
            105.250044726,
            69.5416820236,
            54.6249793842,
            63.041632529299996,
            59.416750445999995,
            54.166710469899996,
            94.91667151450001,
            66.2916572765,
            54.2082707398
          ],
          "command_buffer_latency_us_p50": 63.041632529299996,
          "command_buffer_latency_us_p90": 96.98334615680001,
          "counter_control_throughput_ratio": 0.9923326553578499,
          "counter_control_latency_ratio": 1.0069719326424456
        },
        "correctness": {
          "max_abs_error": 1.2343096304160556e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 54.4155177165,
        "throughput_us_p90": 55.000229527580004,
        "latency_us_p50": 358.25,
        "latency_us_p90": 373.7004
      },
      "torch": {
        "cold_call_ms": 4.421083,
        "warmup_ms": 200.681416,
        "repetitions": 229,
        "throughput_us": [
          129.05603930131005,
          128.94832751091704,
          130.61626637554585,
          129.27292576419214,
          128.63773362445414,
          128.40393013100436,
          128.44196069868994,
          128.74963318777293,
          129.65829694323145
        ],
        "latency_us": [
          602.333,
          407.292,
          446.125,
          735.709,
          393.83299999999997,
          392.95799999999997,
          419.79100000000005,
          494.875,
          491.833
        ],
        "throughput_us_p50": 128.94832751091704,
        "throughput_us_p90": 129.84989082969432,
        "latency_us_p50": 446.125,
        "latency_us_p90": 629.0082,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 8049916.70884192,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7798208.273015916,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7725041.708908975,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7892041.525337845,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7878458.360210061,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7807749.9638311565,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7738874.992355704,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7806166.599038988,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7758958.323393017,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 141249.96960163116,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 125916.61652550101,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 130000.00035390258,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 133583.32216739655,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 135875.01598522067,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 129333.34801346064,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 148041.63947701454,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 128749.99083578587,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 133083.31836014986,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              125.779948575655,
              121.8470042658737,
              120.70377670170274,
              123.31314883340383,
              123.1009118782822,
              121.99609318486182,
              120.91992175555788,
              121.9713531099842,
              121.23372380301589
            ],
            "command_buffer_throughput_us_p50": 121.9713531099842,
            "command_buffer_throughput_us_p90": 123.80650878185406,
            "command_buffer_latency_us": [
              141.24996960163116,
              125.91661652550101,
              130.00000035390258,
              133.58332216739655,
              135.87501598522067,
              129.33334801346064,
              148.04163947701454,
              128.74999083578587,
              133.08331836014986
            ],
            "command_buffer_latency_us_p50": 133.08331836014986,
            "command_buffer_latency_us_p90": 142.60830357670784
          },
          "throughput": [
            {
              "compute_ns": 8106544.0,
              "compute_span_ns": 9315750.0,
              "command_buffer_ns": 9133041.60349071,
              "calibration_cpu_ns": 10625917.0,
              "calibration_gpu_ticks": 10625917.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8222204.0,
              "compute_span_ns": 9632333.0,
              "command_buffer_ns": 9397749.963682145,
              "calibration_cpu_ns": 10643209.0,
              "calibration_gpu_ticks": 10643209.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 7960045.0,
              "compute_span_ns": 9299250.0,
              "command_buffer_ns": 9069208.230357617,
              "calibration_cpu_ns": 10547250.0,
              "calibration_gpu_ticks": 10547250.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8080499.0,
              "compute_span_ns": 9492833.0,
              "command_buffer_ns": 9261541.650630534,
              "calibration_cpu_ns": 10449625.0,
              "calibration_gpu_ticks": 10449625.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8198452.0,
              "compute_span_ns": 9465875.0,
              "command_buffer_ns": 9277791.65795073,
              "calibration_cpu_ns": 10610875.0,
              "calibration_gpu_ticks": 10610875.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8129457.0,
              "compute_span_ns": 9551042.0,
              "command_buffer_ns": 9330124.885309488,
              "calibration_cpu_ns": 10589500.0,
              "calibration_gpu_ticks": 10589500.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8051253.0,
              "compute_span_ns": 9454292.0,
              "command_buffer_ns": 9235041.739884764,
              "calibration_cpu_ns": 10459667.0,
              "calibration_gpu_ticks": 10459667.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8319540.0,
              "compute_span_ns": 9562250.0,
              "command_buffer_ns": 9378208.254929632,
              "calibration_cpu_ns": 10583458.0,
              "calibration_gpu_ticks": 10583458.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8057336.0,
              "compute_span_ns": 9562791.0,
              "command_buffer_ns": 9337083.43654871,
              "calibration_cpu_ns": 10592750.0,
              "calibration_gpu_ticks": 10592750.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 145125.0,
              "compute_span_ns": 145125.0,
              "command_buffer_ns": 145125.02821162343,
              "calibration_cpu_ns": 682167.0,
              "calibration_gpu_ticks": 682167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 126000.0,
              "compute_span_ns": 126000.0,
              "command_buffer_ns": 125999.9698959291,
              "calibration_cpu_ns": 596375.0,
              "calibration_gpu_ticks": 596375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 134833.0,
              "compute_span_ns": 134833.0,
              "command_buffer_ns": 134833.33168551326,
              "calibration_cpu_ns": 606584.0,
              "calibration_gpu_ticks": 606584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 148250.0,
              "compute_span_ns": 148250.0,
              "command_buffer_ns": 148250.02290308475,
              "calibration_cpu_ns": 897875.0,
              "calibration_gpu_ticks": 897875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 130625.0,
              "compute_span_ns": 130625.0,
              "command_buffer_ns": 130624.97600913048,
              "calibration_cpu_ns": 733542.0,
              "calibration_gpu_ticks": 733542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 131958.0,
              "compute_span_ns": 131958.0,
              "command_buffer_ns": 131916.66221246123,
              "calibration_cpu_ns": 636583.0,
              "calibration_gpu_ticks": 636583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 137625.0,
              "compute_span_ns": 137625.0,
              "command_buffer_ns": 137624.97110292315,
              "calibration_cpu_ns": 891500.0,
              "calibration_gpu_ticks": 891500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 131250.0,
              "compute_span_ns": 131250.0,
              "command_buffer_ns": 131250.0098720193,
              "calibration_cpu_ns": 620458.0,
              "calibration_gpu_ticks": 620458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 156375.0,
              "compute_span_ns": 156375.0,
              "command_buffer_ns": 156374.99745935202,
              "calibration_cpu_ns": 943042.0,
              "calibration_gpu_ticks": 943042.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            126.66475,
            128.4719375,
            124.375703125,
            126.257796875,
            128.1008125,
            127.022765625,
            125.800828125,
            129.9928125,
            125.895875
          ],
          "compute_throughput_us_p50": 126.66475,
          "compute_throughput_us_p90": 128.7761125,
          "command_buffer_throughput_us": [
            142.70377505454235,
            146.8398431825335,
            141.70637859933777,
            144.7115882911021,
            144.96549465548014,
            145.78320133296074,
            144.29752718569944,
            146.5345039832755,
            145.8919286960736
          ],
          "command_buffer_throughput_us_p50": 144.96549465548014,
          "command_buffer_throughput_us_p90": 146.5955718231271,
          "compute_latency_us": [
            145.125,
            126.0,
            134.833,
            148.25,
            130.625,
            131.958,
            137.625,
            131.25,
            156.375
          ],
          "compute_latency_us_p50": 134.833,
          "compute_latency_us_p90": 149.875,
          "command_buffer_latency_us": [
            145.12502821162343,
            125.9999698959291,
            134.83333168551326,
            148.25002290308475,
            130.62497600913048,
            131.91666221246123,
            137.62497110292315,
            131.2500098720193,
            156.37499745935202
          ],
          "command_buffer_latency_us_p50": 134.83333168551326,
          "command_buffer_latency_us_p90": 149.8750178143382,
          "counter_control_throughput_ratio": 1.1933312980254807,
          "counter_control_latency_ratio": 1.0199740765910716
        },
        "download_ms": 0.732958,
        "allocation_upload_ms": 16.758875,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 9.74135319392784e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.42199475376594603,
      "device_compute_slowdown": 0.42687228984385944,
      "valid": true,
      "round": 2,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "softmax_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.599916,
        "warmup_ms": 200.290208,
        "repetitions": 219,
        "throughput_us": [
          143.07058447488583,
          134.08504566210047,
          129.9554794520548,
          132.45947488584474,
          139.86948401826484,
          200.77435159817352,
          140.40753424657535,
          133.46936529680363,
          130.656200913242
        ],
        "latency_us": [
          535.0,
          439.25,
          426.042,
          407.58299999999997,
          439.083,
          757.042,
          410.08299999999997,
          398.54200000000003,
          492.041
        ],
        "throughput_us_p50": 134.08504566210047,
        "throughput_us_p90": 154.61133789954337,
        "latency_us_p50": 439.083,
        "latency_us_p90": 579.4084,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 7909083.2732617855,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7918625.022284687,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7896833.354607224,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7887249.98710677,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 8007916.621863842,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7857833.348680288,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7818750.047590584,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7815208.402462304,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7919916.5920726955,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 128666.63746535778,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 142749.98102337122,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 142499.97911974788,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 151458.37096497416,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 124083.36624503136,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 130041.67703911662,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 129541.67323186994,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 131624.98362362385,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 173999.98614564538,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              123.5794261447154,
              123.72851597319823,
              123.38802116573788,
              123.23828104854329,
              125.12369721662253,
              122.7786460731295,
              122.16796949360287,
              122.1126312884735,
              123.74869675113587
            ],
            "command_buffer_throughput_us_p50": 123.38802116573788,
            "command_buffer_throughput_us_p90": 124.0236968442332,
            "command_buffer_latency_us": [
              128.66663746535778,
              142.74998102337122,
              142.49997911974788,
              151.45837096497416,
              124.08336624503136,
              130.04167703911662,
              129.54167323186994,
              131.62498362362385,
              173.99998614564538
            ],
            "command_buffer_latency_us_p50": 131.62498362362385,
            "command_buffer_latency_us_p90": 155.9666940011084
          },
          "throughput": [
            {
              "compute_ns": 8174786.0,
              "compute_span_ns": 9540875.0,
              "command_buffer_ns": 9317125.13556704,
              "calibration_cpu_ns": 10645000.0,
              "calibration_gpu_ticks": 10645000.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8377501.0,
              "compute_span_ns": 9789583.0,
              "command_buffer_ns": 9578916.7098701,
              "calibration_cpu_ns": 11129708.0,
              "calibration_gpu_ticks": 11129708.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8339668.0,
              "compute_span_ns": 9398625.0,
              "command_buffer_ns": 9243791.573680937,
              "calibration_cpu_ns": 10431833.0,
              "calibration_gpu_ticks": 10431833.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8380496.0,
              "compute_span_ns": 9798584.0,
              "command_buffer_ns": 9572041.570208967,
              "calibration_cpu_ns": 11032042.0,
              "calibration_gpu_ticks": 11032042.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8243876.0,
              "compute_span_ns": 9574583.0,
              "command_buffer_ns": 9345166.676212102,
              "calibration_cpu_ns": 11002792.0,
              "calibration_gpu_ticks": 11002792.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8191917.0,
              "compute_span_ns": 9693334.0,
              "command_buffer_ns": 9436791.588086635,
              "calibration_cpu_ns": 10720375.0,
              "calibration_gpu_ticks": 10720375.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8222878.0,
              "compute_span_ns": 9691083.0,
              "command_buffer_ns": 9458291.63538292,
              "calibration_cpu_ns": 10811416.0,
              "calibration_gpu_ticks": 10811416.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8263500.0,
              "compute_span_ns": 9731917.0,
              "command_buffer_ns": 9487416.653428227,
              "calibration_cpu_ns": 10903542.0,
              "calibration_gpu_ticks": 10903542.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8257914.0,
              "compute_span_ns": 9709000.0,
              "command_buffer_ns": 9477416.810113937,
              "calibration_cpu_ns": 10865584.0,
              "calibration_gpu_ticks": 10865584.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 130042.0,
              "compute_span_ns": 130042.0,
              "command_buffer_ns": 130083.35372433066,
              "calibration_cpu_ns": 677542.0,
              "calibration_gpu_ticks": 677542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 154500.0,
              "compute_span_ns": 154500.0,
              "command_buffer_ns": 154500.0122860074,
              "calibration_cpu_ns": 952584.0,
              "calibration_gpu_ticks": 952584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 130333.0,
              "compute_span_ns": 130333.0,
              "command_buffer_ns": 130291.62073507905,
              "calibration_cpu_ns": 614833.0,
              "calibration_gpu_ticks": 614833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 136167.0,
              "compute_span_ns": 136167.0,
              "command_buffer_ns": 136208.31305161119,
              "calibration_cpu_ns": 712458.0,
              "calibration_gpu_ticks": 712458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 133083.0,
              "compute_span_ns": 133083.0,
              "command_buffer_ns": 133083.31836014986,
              "calibration_cpu_ns": 680709.0,
              "calibration_gpu_ticks": 680709.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 132792.0,
              "compute_span_ns": 132792.0,
              "command_buffer_ns": 132833.31645652652,
              "calibration_cpu_ns": 793375.0,
              "calibration_gpu_ticks": 793375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 130042.0,
              "compute_span_ns": 130042.0,
              "command_buffer_ns": 130083.35372433066,
              "calibration_cpu_ns": 639459.0,
              "calibration_gpu_ticks": 639459.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 135625.0,
              "compute_span_ns": 135625.0,
              "command_buffer_ns": 135624.95587393641,
              "calibration_cpu_ns": 769834.0,
              "calibration_gpu_ticks": 769834.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 129958.0,
              "compute_span_ns": 129958.0,
              "command_buffer_ns": 129916.6469834745,
              "calibration_cpu_ns": 636209.0,
              "calibration_gpu_ticks": 636209.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            127.73103125,
            130.898453125,
            130.3073125,
            130.94525,
            128.8105625,
            127.998703125,
            128.48246875,
            129.1171875,
            129.02990625
          ],
          "compute_throughput_us_p50": 129.02990625,
          "compute_throughput_us_p90": 130.9078125,
          "command_buffer_throughput_us": [
            145.580080243235,
            149.6705735917203,
            144.43424333876465,
            149.5631495345151,
            146.0182293158141,
            147.44986856385367,
            147.78580680285813,
            148.24088520981604,
            148.08463765803026
          ],
          "command_buffer_throughput_us_p50": 147.78580680285813,
          "command_buffer_throughput_us_p90": 149.58463434595615,
          "compute_latency_us": [
            130.042,
            154.5,
            130.333,
            136.167,
            133.083,
            132.792,
            130.042,
            135.625,
            129.958
          ],
          "compute_latency_us_p50": 132.792,
          "compute_latency_us_p90": 139.83360000000002,
          "command_buffer_latency_us": [
            130.08335372433066,
            154.5000122860074,
            130.29162073507905,
            136.20831305161119,
            133.08331836014986,
            132.83331645652652,
            130.08335372433066,
            135.62495587393641,
            129.9166469834745
          ],
          "command_buffer_latency_us_p50": 132.83331645652652,
          "command_buffer_latency_us_p90": 139.86665289849043,
          "counter_control_throughput_ratio": 1.2009406625646408,
          "counter_control_latency_ratio": 1.0110107506256572
        },
        "download_ms": 0.573208,
        "allocation_upload_ms": 2.187875,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 9.74135319392784e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-200xztc0/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "8bb581805080afef9f47b3147742261e73f0bbc7f1d60afd69631b609a5e3f17",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.562458,
        "capture_ms": 0.061917,
        "compile_ms": 51.379958,
        "allocation_upload_ms": 5.428833,
        "cold_call_ms": 6.403875,
        "warmup_ms": 200.498416,
        "download_ms": 3.385083,
        "repetitions": 494,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          51.933451417,
          52.0436072874,
          52.6306497976,
          52.145917004,
          52.3895910931,
          52.0655364372,
          53.3641194332,
          52.1474352227,
          51.8901821862
        ],
        "latency_us": [
          409.916,
          360,
          384.834,
          313,
          298.542,
          320.75,
          297.208,
          303.291,
          340.666
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 1024,
            "reduction_scalar_rounds": 80,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 16384,
              "global_write_bytes": 16384,
              "private_read_bytes": 98304,
              "private_write_bytes": 32768
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 64,
              "global_write_bytes": 64,
              "private_read_bytes": 384,
              "private_write_bytes": 128
            },
            "striped_storage_scalars_per_worker": 32,
            "reduction_operations": 2,
            "reduction_elements": 8192,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 128,
            "normalized_kernel_cost": 2048,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3334292,
              "compute_span_ns": 3334292,
              "calibration_cpu_ns": 3874667,
              "calibration_gpu_ticks": 3874667,
              "compute_passes": 1,
              "command_buffer_ns": 3334374.98193,
              "command_buffers": 2
            },
            {
              "compute_ns": 3325875,
              "compute_span_ns": 3325875,
              "calibration_cpu_ns": 4365375,
              "calibration_gpu_ticks": 4365375,
              "compute_passes": 1,
              "command_buffer_ns": 3325958.32879,
              "command_buffers": 2
            },
            {
              "compute_ns": 3313708,
              "compute_span_ns": 3313708,
              "calibration_cpu_ns": 4114166,
              "calibration_gpu_ticks": 4114166,
              "compute_passes": 1,
              "command_buffer_ns": 3313791.64709,
              "command_buffers": 2
            },
            {
              "compute_ns": 3344792,
              "compute_span_ns": 3344792,
              "calibration_cpu_ns": 4081250,
              "calibration_gpu_ticks": 4081250,
              "compute_passes": 1,
              "command_buffer_ns": 3344916.68036,
              "command_buffers": 2
            },
            {
              "compute_ns": 3188292,
              "compute_span_ns": 3188292,
              "calibration_cpu_ns": 3746500,
              "calibration_gpu_ticks": 3746500,
              "compute_passes": 1,
              "command_buffer_ns": 3188333.35768,
              "command_buffers": 2
            },
            {
              "compute_ns": 3305083,
              "compute_span_ns": 3305083,
              "calibration_cpu_ns": 4077792,
              "calibration_gpu_ticks": 4077792,
              "compute_passes": 1,
              "command_buffer_ns": 3305083.31535,
              "command_buffers": 2
            },
            {
              "compute_ns": 3254709,
              "compute_span_ns": 3254709,
              "calibration_cpu_ns": 4013042,
              "calibration_gpu_ticks": 4013042,
              "compute_passes": 1,
              "command_buffer_ns": 3254791.66349,
              "command_buffers": 2
            },
            {
              "compute_ns": 3318333,
              "compute_span_ns": 3318333,
              "calibration_cpu_ns": 3940541,
              "calibration_gpu_ticks": 3940541,
              "compute_passes": 1,
              "command_buffer_ns": 3318374.97652,
              "command_buffers": 2
            },
            {
              "compute_ns": 3322458,
              "compute_span_ns": 3322458,
              "calibration_cpu_ns": 4076333,
              "calibration_gpu_ticks": 4076333,
              "compute_passes": 1,
              "command_buffer_ns": 3322500.03703,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 65416,
              "compute_span_ns": 65416,
              "calibration_cpu_ns": 547500,
              "calibration_gpu_ticks": 547500,
              "compute_passes": 1,
              "command_buffer_ns": 65458.3564028,
              "command_buffers": 2
            },
            {
              "compute_ns": 75625,
              "compute_span_ns": 75625,
              "calibration_cpu_ns": 746750,
              "calibration_gpu_ticks": 746750,
              "compute_passes": 1,
              "command_buffer_ns": 75666.6995585,
              "command_buffers": 2
            },
            {
              "compute_ns": 98334,
              "compute_span_ns": 98334,
              "calibration_cpu_ns": 811750,
              "calibration_gpu_ticks": 811750,
              "compute_passes": 1,
              "command_buffer_ns": 98375.02148,
              "command_buffers": 2
            },
            {
              "compute_ns": 65125,
              "compute_span_ns": 65125,
              "calibration_cpu_ns": 520042,
              "calibration_gpu_ticks": 520042,
              "compute_passes": 1,
              "command_buffer_ns": 65125.0011288,
              "command_buffers": 2
            },
            {
              "compute_ns": 63333,
              "compute_span_ns": 63333,
              "calibration_cpu_ns": 559833,
              "calibration_gpu_ticks": 559833,
              "compute_passes": 1,
              "command_buffer_ns": 63291.634433,
              "command_buffers": 2
            },
            {
              "compute_ns": 107417,
              "compute_span_ns": 107417,
              "calibration_cpu_ns": 681584,
              "calibration_gpu_ticks": 681584,
              "compute_passes": 1,
              "command_buffer_ns": 107500.003651,
              "command_buffers": 2
            },
            {
              "compute_ns": 82709,
              "compute_span_ns": 82709,
              "calibration_cpu_ns": 783084,
              "calibration_gpu_ticks": 783084,
              "compute_passes": 1,
              "command_buffer_ns": 82791.7247079,
              "command_buffers": 2
            },
            {
              "compute_ns": 111250,
              "compute_span_ns": 111250,
              "calibration_cpu_ns": 1539417,
              "calibration_gpu_ticks": 1539417,
              "compute_passes": 1,
              "command_buffer_ns": 111291.650683,
              "command_buffers": 2
            },
            {
              "compute_ns": 100959,
              "compute_span_ns": 100959,
              "calibration_cpu_ns": 1232125,
              "calibration_gpu_ticks": 1232125,
              "compute_passes": 1,
              "command_buffer_ns": 101000.012364,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 3503000.01446,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3400916.64111,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3334708.279,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3296166.6584,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3316666.69808,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3339916.64229,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3281124.98391,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3214208.35098,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3247250.04612,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 103833.328467,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 64458.2905807,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 72000.0243746,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 66666.6892357,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 77416.6546762,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 93291.6300371,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 82999.9917187,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 90416.695457,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 70250.0110492,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              54.734375225937505,
              53.13932251734375,
              52.104816859375,
              51.5026040375,
              51.8229171575,
              52.18619753578125,
              51.26757787359375,
              50.2220054840625,
              50.738281970625
            ],
            "command_buffer_throughput_us_p50": 51.8229171575,
            "command_buffer_throughput_us_p90": 53.458333059062504,
            "command_buffer_latency_us": [
              103.833328467,
              64.4582905807,
              72.0000243746,
              66.6666892357,
              77.41665467620001,
              93.29163003709999,
              82.9999917187,
              90.41669545699999,
              70.2500110492
            ],
            "command_buffer_latency_us_p50": 77.41665467620001,
            "command_buffer_latency_us_p90": 95.39996972307999
          },
          "compute_throughput_us": [
            52.0983125,
            51.966796875,
            51.7766875,
            52.262375,
            49.8170625,
            51.641921875,
            50.854828125,
            51.848953125,
            51.91340625
          ],
          "compute_throughput_us_p50": 51.848953125,
          "compute_throughput_us_p90": 52.131125,
          "command_buffer_throughput_us": [
            52.099609092656245,
            51.96809888734375,
            51.77799448578125,
            52.264323130625,
            49.81770871375,
            51.64192680234375,
            50.856119742031254,
            51.849609008125,
            51.91406307859375
          ],
          "command_buffer_throughput_us_p50": 51.849609008125,
          "command_buffer_throughput_us_p90": 52.13255190024999,
          "compute_latency_us": [
            65.416,
            75.625,
            98.334,
            65.125,
            63.333,
            107.417,
            82.709,
            111.25,
            100.959
          ],
          "compute_latency_us_p50": 82.709,
          "compute_latency_us_p90": 108.1836,
          "command_buffer_latency_us": [
            65.45835640279999,
            75.6666995585,
            98.37502148,
            65.12500112880001,
            63.291634433,
            107.500003651,
            82.79172470789999,
            111.291650683,
            101.000012364
          ],
          "command_buffer_latency_us_p50": 82.79172470789999,
          "command_buffer_latency_us_p90": 108.2583330574,
          "counter_control_throughput_ratio": 0.9919743013298387,
          "counter_control_latency_ratio": 1.1523006255571873
        },
        "correctness": {
          "max_abs_error": 1.2343096304160556e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 52.145917004,
        "throughput_us_p90": 52.77734372472,
        "latency_us_p50": 320.75,
        "latency_us_p90": 389.8504
      },
      "slowdown": 0.38890181038838395,
      "device_compute_slowdown": 0.40183671082067457,
      "valid": true,
      "round": 2,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "rmsnorm_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-ccv1qgpr/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "54d99166e42515c7bfff1cac173ad46750fa91f9c92e668c07d69318fdc050bf",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.539916,
        "capture_ms": 0.060959,
        "compile_ms": 58.582041,
        "allocation_upload_ms": 1.73675,
        "cold_call_ms": 2.27025,
        "warmup_ms": 200.263667,
        "download_ms": 0.367459,
        "repetitions": 7067,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          4.12976991651,
          4.1687126079,
          4.12219937739,
          4.16416669025,
          4.26096066223,
          4.15354223857,
          4.20431229659,
          4.30201415028,
          4.16065289373
        ],
        "latency_us": [
          286.166,
          201.083,
          189.542,
          171.75,
          176,
          202.041,
          187.334,
          312.916,
          202.5
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 32,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 37,
            "reduction_scalar_rounds": 24,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 12296,
              "global_write_bytes": 6148,
              "private_read_bytes": 18444,
              "private_write_bytes": 6148
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 64,
              "global_write_bytes": 32,
              "private_read_bytes": 96,
              "private_write_bytes": 32
            },
            "striped_storage_scalars_per_worker": 8,
            "reduction_operations": 1,
            "reduction_elements": 1537,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 56,
            "normalized_kernel_cost": 56,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 3074,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 260541,
              "compute_span_ns": 260541,
              "calibration_cpu_ns": 915542,
              "calibration_gpu_ticks": 915542,
              "compute_passes": 1,
              "command_buffer_ns": 260583.357885,
              "command_buffers": 2
            },
            {
              "compute_ns": 245167,
              "compute_span_ns": 245167,
              "calibration_cpu_ns": 891875,
              "calibration_gpu_ticks": 891875,
              "compute_passes": 1,
              "command_buffer_ns": 245250.004809,
              "command_buffers": 2
            },
            {
              "compute_ns": 254875,
              "compute_span_ns": 254875,
              "calibration_cpu_ns": 685708,
              "calibration_gpu_ticks": 685708,
              "compute_passes": 1,
              "command_buffer_ns": 255000.020843,
              "command_buffers": 2
            },
            {
              "compute_ns": 260750,
              "compute_span_ns": 260750,
              "calibration_cpu_ns": 896500,
              "calibration_gpu_ticks": 896500,
              "compute_passes": 1,
              "command_buffer_ns": 260833.359789,
              "command_buffers": 2
            },
            {
              "compute_ns": 244709,
              "compute_span_ns": 244709,
              "calibration_cpu_ns": 637458,
              "calibration_gpu_ticks": 637458,
              "compute_passes": 1,
              "command_buffer_ns": 244833.354373,
              "command_buffers": 2
            },
            {
              "compute_ns": 245000,
              "compute_span_ns": 245000,
              "calibration_cpu_ns": 676709,
              "calibration_gpu_ticks": 676709,
              "compute_passes": 1,
              "command_buffer_ns": 245041.679591,
              "command_buffers": 2
            },
            {
              "compute_ns": 244250,
              "compute_span_ns": 244250,
              "calibration_cpu_ns": 806542,
              "calibration_gpu_ticks": 806542,
              "compute_passes": 1,
              "command_buffer_ns": 244249.997195,
              "command_buffers": 2
            },
            {
              "compute_ns": 244958,
              "compute_span_ns": 244958,
              "calibration_cpu_ns": 804083,
              "calibration_gpu_ticks": 804083,
              "compute_passes": 1,
              "command_buffer_ns": 244958.326221,
              "command_buffers": 2
            },
            {
              "compute_ns": 259084,
              "compute_span_ns": 259084,
              "calibration_cpu_ns": 875875,
              "calibration_gpu_ticks": 875875,
              "compute_passes": 1,
              "command_buffer_ns": 259124.964941,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 6333,
              "compute_span_ns": 6333,
              "calibration_cpu_ns": 449292,
              "calibration_gpu_ticks": 449292,
              "compute_passes": 1,
              "command_buffer_ns": 6291.66606814,
              "command_buffers": 2
            },
            {
              "compute_ns": 6959,
              "compute_span_ns": 6959,
              "calibration_cpu_ns": 468459,
              "calibration_gpu_ticks": 468459,
              "compute_passes": 1,
              "command_buffer_ns": 7041.67177901,
              "command_buffers": 2
            },
            {
              "compute_ns": 6833,
              "compute_span_ns": 6833,
              "calibration_cpu_ns": 421084,
              "calibration_gpu_ticks": 421084,
              "compute_passes": 1,
              "command_buffer_ns": 6875.02324581,
              "command_buffers": 2
            },
            {
              "compute_ns": 6834,
              "compute_span_ns": 6834,
              "calibration_cpu_ns": 434959,
              "calibration_gpu_ticks": 434959,
              "compute_passes": 1,
              "command_buffer_ns": 6916.64172336,
              "command_buffers": 2
            },
            {
              "compute_ns": 8209,
              "compute_span_ns": 8209,
              "calibration_cpu_ns": 676208,
              "calibration_gpu_ticks": 676208,
              "compute_passes": 1,
              "command_buffer_ns": 8333.35798234,
              "command_buffers": 2
            },
            {
              "compute_ns": 7334,
              "compute_span_ns": 7334,
              "calibration_cpu_ns": 805958,
              "calibration_gpu_ticks": 805958,
              "compute_passes": 1,
              "command_buffer_ns": 7416.64553061,
              "command_buffers": 2
            },
            {
              "compute_ns": 6916,
              "compute_span_ns": 6916,
              "calibration_cpu_ns": 574917,
              "calibration_gpu_ticks": 574917,
              "compute_passes": 1,
              "command_buffer_ns": 6875.02324581,
              "command_buffers": 2
            },
            {
              "compute_ns": 6875,
              "compute_span_ns": 6875,
              "calibration_cpu_ns": 458542,
              "calibration_gpu_ticks": 458542,
              "compute_passes": 1,
              "command_buffer_ns": 6916.69993103,
              "command_buffers": 2
            },
            {
              "compute_ns": 7250,
              "compute_span_ns": 7250,
              "calibration_cpu_ns": 511125,
              "calibration_gpu_ticks": 511125,
              "compute_passes": 1,
              "command_buffer_ns": 7375.02705306,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 263875.00111,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 252125.028055,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 244125.025347,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 245083.356276,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 244666.647632,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 260791.624896,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 260166.707449,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 246958.283242,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 261958.339252,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 6249.98938292,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 8791.68510437,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6958.37661624,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7375.02705306,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6666.6980274,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6875.02324581,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6875.02324581,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 8124.97455627,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7250.05520508,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              4.12304689234375,
              3.939453563359375,
              3.814453521046875,
              3.8294274418125003,
              3.82291636925,
              4.074869139,
              4.065104803890625,
              3.85872317565625,
              4.0930990508125005
            ],
            "command_buffer_throughput_us_p50": 3.939453563359375,
            "command_buffer_throughput_us_p90": 4.09908861911875,
            "command_buffer_latency_us": [
              6.24998938292,
              8.79168510437,
              6.95837661624,
              7.37502705306,
              6.6666980274,
              6.87502324581,
              6.87502324581,
              8.12497455627,
              7.25005520508
            ],
            "command_buffer_latency_us_p50": 6.95837661624,
            "command_buffer_latency_us_p90": 8.25831666589
          },
          "compute_throughput_us": [
            4.070953125,
            3.830734375,
            3.982421875,
            4.07421875,
            3.823578125,
            3.828125,
            3.81640625,
            3.82746875,
            4.0481875
          ],
          "compute_throughput_us_p50": 3.830734375,
          "compute_throughput_us_p90": 4.07160625,
          "command_buffer_throughput_us": [
            4.071614966953125,
            3.8320313251406253,
            3.9843753256718752,
            4.075521246703125,
            3.8255211620781253,
            3.828776243609375,
            3.816406206171875,
            3.827473847203125,
            4.0488275772031255
          ],
          "command_buffer_throughput_us_p50": 3.8320313251406253,
          "command_buffer_throughput_us_p90": 4.072396222903125,
          "compute_latency_us": [
            6.333,
            6.959,
            6.833,
            6.834,
            8.209,
            7.334,
            6.916,
            6.875,
            7.25
          ],
          "compute_latency_us_p50": 6.916,
          "compute_latency_us_p90": 7.5089999999999995,
          "command_buffer_latency_us": [
            6.29166606814,
            7.0416717790100005,
            6.87502324581,
            6.91664172336,
            8.333357982339999,
            7.416645530609999,
            6.87502324581,
            6.91669993103,
            7.37502705306
          ],
          "command_buffer_latency_us_p50": 6.91669993103,
          "command_buffer_latency_us_p90": 7.599988020955999,
          "counter_control_throughput_ratio": 0.9891838743554014,
          "counter_control_latency_ratio": 1.0
        },
        "correctness": {
          "max_abs_error": 1.5564224420394623e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 4.16416669025,
        "throughput_us_p90": 4.26917135984,
        "latency_us_p50": 201.083,
        "latency_us_p90": 291.516
      },
      "torch": {
        "cold_call_ms": 3.180167,
        "warmup_ms": 200.23225,
        "repetitions": 3328,
        "throughput_us": [
          7.661157752403846,
          7.639060096153846,
          7.570074519230769,
          7.614107572115385,
          7.713429086538462,
          7.519831730769231,
          7.487067007211538,
          7.510091045673078,
          7.546850060096154
        ],
        "latency_us": [
          413.08299999999997,
          275.959,
          277.041,
          282.292,
          249.417,
          236.667,
          305.25,
          234.16600000000003,
          227.167
        ],
        "throughput_us_p50": 7.570074519230769,
        "throughput_us_p90": 7.6716120192307695,
        "latency_us_p50": 275.959,
        "latency_us_p90": 326.8166,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 306458.32885056734,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 342125.0148676336,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 289875.0244639814,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 304999.9941140413,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 321124.97137859464,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 304624.96215477586,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 305125.02416968346,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 305708.3231396973,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 307874.9869018793,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 7666.705641895533,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8333.299774676561,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8833.36178958416,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8166.651241481304,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 9375.042282044888,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8041.67939350009,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 9750.016033649445,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8583.359885960817,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8333.357982337475,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              4.788411388290115,
              5.345703357306775,
              4.529297257249709,
              4.765624908031896,
              5.017577677790541,
              4.759765033668373,
              4.767578502651304,
              4.7766925490577705,
              4.810546670341864
            ],
            "command_buffer_throughput_us_p50": 4.7766925490577705,
            "command_buffer_throughput_us_p90": 5.083202813693788,
            "command_buffer_latency_us": [
              7.666705641895533,
              8.333299774676561,
              8.83336178958416,
              8.166651241481304,
              9.375042282044888,
              8.04167939350009,
              9.750016033649445,
              8.583359885960817,
              8.333357982337475
            ],
            "command_buffer_latency_us_p50": 8.333357982337475,
            "command_buffer_latency_us_p90": 9.450037032365799
          },
          "throughput": [
            {
              "compute_ns": 289750.0,
              "compute_span_ns": 289750.0,
              "command_buffer_ns": 289749.99440833926,
              "calibration_cpu_ns": 972250.0,
              "calibration_gpu_ticks": 972250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 288375.0,
              "compute_span_ns": 288375.0,
              "command_buffer_ns": 288375.01304224133,
              "calibration_cpu_ns": 928334.0,
              "calibration_gpu_ticks": 928334.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 288916.0,
              "compute_span_ns": 288916.0,
              "command_buffer_ns": 288874.9586418271,
              "calibration_cpu_ns": 909125.0,
              "calibration_gpu_ticks": 909125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 304875.0,
              "compute_span_ns": 304875.0,
              "command_buffer_ns": 304875.0222660601,
              "calibration_cpu_ns": 939583.0,
              "calibration_gpu_ticks": 939583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 308916.0,
              "compute_span_ns": 308916.0,
              "command_buffer_ns": 308875.0527240336,
              "calibration_cpu_ns": 978000.0,
              "calibration_gpu_ticks": 978000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 289583.0,
              "compute_span_ns": 289583.0,
              "command_buffer_ns": 289583.345875144,
              "calibration_cpu_ns": 973167.0,
              "calibration_gpu_ticks": 973167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 289792.0,
              "compute_span_ns": 289792.0,
              "command_buffer_ns": 289833.28957110643,
              "calibration_cpu_ns": 907292.0,
              "calibration_gpu_ticks": 907292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 290375.0,
              "compute_span_ns": 290375.0,
              "command_buffer_ns": 290374.97006356716,
              "calibration_cpu_ns": 962667.0,
              "calibration_gpu_ticks": 962667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 314500.0,
              "compute_span_ns": 314500.0,
              "command_buffer_ns": 314500.00824406743,
              "calibration_cpu_ns": 927459.0,
              "calibration_gpu_ticks": 927459.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 7791.0,
              "compute_span_ns": 7791.0,
              "command_buffer_ns": 7750.0008046627045,
              "calibration_cpu_ns": 589791.0,
              "calibration_gpu_ticks": 589791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 9166.0,
              "compute_span_ns": 9166.0,
              "command_buffer_ns": 9124.982170760632,
              "calibration_cpu_ns": 697833.0,
              "calibration_gpu_ticks": 697833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8167.0,
              "compute_span_ns": 8167.0,
              "command_buffer_ns": 8166.651241481304,
              "calibration_cpu_ns": 485500.0,
              "calibration_gpu_ticks": 485500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8208.0,
              "compute_span_ns": 8208.0,
              "command_buffer_ns": 8166.651241481304,
              "calibration_cpu_ns": 476709.0,
              "calibration_gpu_ticks": 476709.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8500.0,
              "compute_span_ns": 8500.0,
              "command_buffer_ns": 8499.948307871819,
              "calibration_cpu_ns": 715667.0,
              "calibration_gpu_ticks": 715667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8042.0,
              "compute_span_ns": 8042.0,
              "command_buffer_ns": 8041.67939350009,
              "calibration_cpu_ns": 449167.0,
              "calibration_gpu_ticks": 449167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 9833.0,
              "compute_span_ns": 9833.0,
              "command_buffer_ns": 9791.692718863487,
              "calibration_cpu_ns": 556750.0,
              "calibration_gpu_ticks": 556750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 9292.0,
              "compute_span_ns": 9292.0,
              "command_buffer_ns": 9333.365596830845,
              "calibration_cpu_ns": 612125.0,
              "calibration_gpu_ticks": 612125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8167.0,
              "compute_span_ns": 8167.0,
              "command_buffer_ns": 8208.38613435626,
              "calibration_cpu_ns": 466500.0,
              "calibration_gpu_ticks": 466500.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            4.52734375,
            4.505859375,
            4.5143125,
            4.763671875,
            4.8268125,
            4.524734375,
            4.528,
            4.537109375,
            4.9140625
          ],
          "compute_throughput_us_p50": 4.528,
          "compute_throughput_us_p90": 4.8442625,
          "command_buffer_throughput_us": [
            4.527343662630301,
            4.505859578785021,
            4.5136712287785485,
            4.763672222907189,
            4.826172698813025,
            4.524739779299125,
            4.528645149548538,
            4.537108907243237,
            4.914062628813554
          ],
          "command_buffer_throughput_us_p50": 4.528645149548538,
          "command_buffer_throughput_us_p90": 4.843750684813131,
          "compute_latency_us": [
            7.791,
            9.166,
            8.167,
            8.208,
            8.5,
            8.042,
            9.833,
            9.292,
            8.167
          ],
          "compute_latency_us_p50": 8.208,
          "compute_latency_us_p90": 9.4002,
          "command_buffer_latency_us": [
            7.7500008046627045,
            9.124982170760632,
            8.166651241481304,
            8.166651241481304,
            8.499948307871819,
            8.04167939350009,
            9.791692718863487,
            9.333365596830845,
            8.20838613435626
          ],
          "command_buffer_latency_us_p50": 8.20838613435626,
          "command_buffer_latency_us_p90": 9.425031021237373,
          "counter_control_throughput_ratio": 0.9506225091560638,
          "counter_control_latency_ratio": 1.0
        },
        "download_ms": 0.27125,
        "allocation_upload_ms": 0.860792,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.807831588873654e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.5500826550216233,
      "device_compute_slowdown": 0.846010241828622,
      "valid": true,
      "round": 2,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "rmsnorm_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.280959,
        "warmup_ms": 200.029583,
        "repetitions": 2984,
        "throughput_us": [
          8.007735589812333,
          7.913915884718499,
          7.841711461126006,
          8.238717493297587,
          8.070277815013405,
          8.267342493297587,
          8.265792560321715,
          8.465678284182305,
          7.812806970509384
        ],
        "latency_us": [
          393.125,
          259.25,
          210.916,
          192.125,
          988.458,
          228.542,
          191.625,
          202.459,
          194.542
        ],
        "throughput_us_p50": 8.070277815013405,
        "throughput_us_p90": 8.30700965147453,
        "latency_us_p50": 210.916,
        "latency_us_p90": 512.1916000000001,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 374583.3528228104,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 366708.3219625056,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 341750.041116029,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 289583.345875144,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 365124.95717033744,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 289541.66918992996,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 328666.64696484804,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 336624.97298792005,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 311249.983496964,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 8124.9745562672615,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8000.002708286047,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8000.002708286047,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 9500.014130026102,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8000.002708286047,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8041.67939350009,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8333.299774676561,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8999.952115118504,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8124.9745562672615,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              5.852864887856413,
              5.72981753066415,
              5.3398443924379535,
              4.524739779299125,
              5.7050774557865225,
              4.524088581092656,
              5.135416358825751,
              5.259765202936251,
              4.863280992140062
            ],
            "command_buffer_throughput_us_p50": 5.259765202936251,
            "command_buffer_throughput_us_p90": 5.754427002102602,
            "command_buffer_latency_us": [
              8.124974556267262,
              8.000002708286047,
              8.000002708286047,
              9.500014130026102,
              8.000002708286047,
              8.04167939350009,
              8.333299774676561,
              8.999952115118504,
              8.124974556267262
            ],
            "command_buffer_latency_us_p50": 8.124974556267262,
            "command_buffer_latency_us_p90": 9.099964518100023
          },
          "throughput": [
            {
              "compute_ns": 343625.0,
              "compute_span_ns": 343625.0,
              "command_buffer_ns": 343625.02628937364,
              "calibration_cpu_ns": 1009166.0,
              "calibration_gpu_ticks": 1009166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 289958.0,
              "compute_span_ns": 289958.0,
              "command_buffer_ns": 289916.70114919543,
              "calibration_cpu_ns": 863292.0,
              "calibration_gpu_ticks": 863292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 361041.0,
              "compute_span_ns": 361041.0,
              "command_buffer_ns": 360999.95486438274,
              "calibration_cpu_ns": 970333.0,
              "calibration_gpu_ticks": 970333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 338417.0,
              "compute_span_ns": 338417.0,
              "command_buffer_ns": 338416.6629984975,
              "calibration_cpu_ns": 932792.0,
              "calibration_gpu_ticks": 932792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 362084.0,
              "compute_span_ns": 362084.0,
              "command_buffer_ns": 362124.99253451824,
              "calibration_cpu_ns": 1244792.0,
              "calibration_gpu_ticks": 1244792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 339542.0,
              "compute_span_ns": 339542.0,
              "command_buffer_ns": 339583.3191461861,
              "calibration_cpu_ns": 972417.0,
              "calibration_gpu_ticks": 972417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 350625.0,
              "compute_span_ns": 350625.0,
              "command_buffer_ns": 350625.0213831663,
              "calibration_cpu_ns": 1334916.0,
              "calibration_gpu_ticks": 1334916.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 337000.0,
              "compute_span_ns": 337000.0,
              "command_buffer_ns": 337000.0049471855,
              "calibration_cpu_ns": 1047958.0,
              "calibration_gpu_ticks": 1047958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 363250.0,
              "compute_span_ns": 363250.0,
              "command_buffer_ns": 363250.03020465374,
              "calibration_cpu_ns": 1116708.0,
              "calibration_gpu_ticks": 1116708.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 8333.0,
              "compute_span_ns": 8333.0,
              "command_buffer_ns": 8291.681297123432,
              "calibration_cpu_ns": 434500.0,
              "calibration_gpu_ticks": 434500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8000.0,
              "compute_span_ns": 8000.0,
              "command_buffer_ns": 8000.002708286047,
              "calibration_cpu_ns": 443792.0,
              "calibration_gpu_ticks": 443792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8000.0,
              "compute_span_ns": 8000.0,
              "command_buffer_ns": 8000.002708286047,
              "calibration_cpu_ns": 463292.0,
              "calibration_gpu_ticks": 463292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8042.0,
              "compute_span_ns": 8042.0,
              "command_buffer_ns": 8041.621185839176,
              "calibration_cpu_ns": 418958.0,
              "calibration_gpu_ticks": 418958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8542.0,
              "compute_span_ns": 8542.0,
              "command_buffer_ns": 8583.301678299904,
              "calibration_cpu_ns": 657459.0,
              "calibration_gpu_ticks": 657459.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8125.0,
              "compute_span_ns": 8125.0,
              "command_buffer_ns": 8125.032763928175,
              "calibration_cpu_ns": 444959.0,
              "calibration_gpu_ticks": 444959.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8083.0,
              "compute_span_ns": 8083.0,
              "command_buffer_ns": 8083.356078714132,
              "calibration_cpu_ns": 480292.0,
              "calibration_gpu_ticks": 480292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8000.0,
              "compute_span_ns": 8000.0,
              "command_buffer_ns": 8000.002708286047,
              "calibration_cpu_ns": 435417.0,
              "calibration_gpu_ticks": 435417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8042.0,
              "compute_span_ns": 8042.0,
              "command_buffer_ns": 8083.356078714132,
              "calibration_cpu_ns": 437583.0,
              "calibration_gpu_ticks": 437583.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            5.369140625,
            4.53059375,
            5.641265625,
            5.287765625,
            5.6575625,
            5.30534375,
            5.478515625,
            5.265625,
            5.67578125
          ],
          "compute_throughput_us_p50": 5.369140625,
          "compute_throughput_us_p90": 5.66120625,
          "command_buffer_throughput_us": [
            5.369141035771463,
            4.529948455456179,
            5.64062429475598,
            5.287760359351523,
            5.6582030083518475,
            5.305989361659158,
            5.478515959111974,
            5.265625077299774,
            5.675781721947715
          ],
          "command_buffer_throughput_us_p50": 5.369141035771463,
          "command_buffer_throughput_us_p90": 5.661718751071021,
          "compute_latency_us": [
            8.333,
            8.0,
            8.0,
            8.042,
            8.542,
            8.125,
            8.083,
            8.0,
            8.042
          ],
          "compute_latency_us_p50": 8.042,
          "compute_latency_us_p90": 8.3748,
          "command_buffer_latency_us": [
            8.291681297123432,
            8.000002708286047,
            8.000002708286047,
            8.041621185839176,
            8.583301678299904,
            8.125032763928175,
            8.083356078714132,
            8.000002708286047,
            8.083356078714132
          ],
          "command_buffer_latency_us_p50": 8.083356078714132,
          "command_buffer_latency_us_p90": 8.350005373358727,
          "counter_control_throughput_ratio": 1.0563274657860775,
          "counter_control_latency_ratio": 1.0
        },
        "download_ms": 0.475875,
        "allocation_upload_ms": 1.829792,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.807831588873654e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-o6cz5rxq/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "e81310ef6c7b3435af776cabf332187d02f0ddc801b9234c32af8a56116f248b",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 28.9585,
        "capture_ms": 0.065792,
        "compile_ms": 65.94525,
        "allocation_upload_ms": 1.700709,
        "cold_call_ms": 2.088375,
        "warmup_ms": 200.137834,
        "download_ms": 0.291709,
        "repetitions": 6029,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          4.44813252612,
          4.52011112954,
          4.77045662631,
          4.51434732128,
          4.53763061868,
          4.48615724001,
          4.40817299718,
          4.46916984575,
          4.48080792835
        ],
        "latency_us": [
          625.041,
          244.417,
          213.833,
          208.75,
          207.833,
          236.208,
          391.333,
          255.584,
          200.75
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 19,
            "reduction_scalar_rounds": 24,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 12296,
              "global_write_bytes": 6148,
              "private_read_bytes": 18444,
              "private_write_bytes": 6148
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 64,
              "global_write_bytes": 32,
              "private_read_bytes": 96,
              "private_write_bytes": 32
            },
            "striped_storage_scalars_per_worker": 8,
            "reduction_operations": 1,
            "reduction_elements": 1537,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 48,
            "normalized_kernel_cost": 48,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 3074,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 300083,
              "compute_span_ns": 300083,
              "calibration_cpu_ns": 1008500,
              "calibration_gpu_ticks": 1008500,
              "compute_passes": 1,
              "command_buffer_ns": 300124.986097,
              "command_buffers": 2
            },
            {
              "compute_ns": 265333,
              "compute_span_ns": 265333,
              "calibration_cpu_ns": 728875,
              "calibration_gpu_ticks": 728875,
              "compute_passes": 1,
              "command_buffer_ns": 265416.689217,
              "command_buffers": 2
            },
            {
              "compute_ns": 292500,
              "compute_span_ns": 292500,
              "calibration_cpu_ns": 1022541,
              "calibration_gpu_ticks": 1022541,
              "compute_passes": 1,
              "command_buffer_ns": 292541.633826,
              "command_buffers": 2
            },
            {
              "compute_ns": 292875,
              "compute_span_ns": 292875,
              "calibration_cpu_ns": 1062292,
              "calibration_gpu_ticks": 1062292,
              "compute_passes": 1,
              "command_buffer_ns": 292958.34247,
              "command_buffers": 2
            },
            {
              "compute_ns": 293375,
              "compute_span_ns": 293375,
              "calibration_cpu_ns": 1280208,
              "calibration_gpu_ticks": 1280208,
              "compute_passes": 1,
              "command_buffer_ns": 293416.669592,
              "command_buffers": 2
            },
            {
              "compute_ns": 290458,
              "compute_span_ns": 290458,
              "calibration_cpu_ns": 755458,
              "calibration_gpu_ticks": 755458,
              "compute_passes": 1,
              "command_buffer_ns": 290500.058327,
              "command_buffers": 2
            },
            {
              "compute_ns": 295042,
              "compute_span_ns": 295042,
              "calibration_cpu_ns": 982875,
              "calibration_gpu_ticks": 982875,
              "compute_passes": 1,
              "command_buffer_ns": 295083.329547,
              "command_buffers": 2
            },
            {
              "compute_ns": 291750,
              "compute_span_ns": 291750,
              "calibration_cpu_ns": 1171209,
              "calibration_gpu_ticks": 1171209,
              "compute_passes": 1,
              "command_buffer_ns": 291750.009637,
              "command_buffers": 2
            },
            {
              "compute_ns": 294583,
              "compute_span_ns": 294583,
              "calibration_cpu_ns": 1293125,
              "calibration_gpu_ticks": 1293125,
              "compute_passes": 1,
              "command_buffer_ns": 294583.32574,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 7041,
              "compute_span_ns": 7041,
              "calibration_cpu_ns": 457416,
              "calibration_gpu_ticks": 457416,
              "compute_passes": 1,
              "command_buffer_ns": 7041.67177901,
              "command_buffers": 2
            },
            {
              "compute_ns": 7166,
              "compute_span_ns": 7166,
              "calibration_cpu_ns": 452333,
              "calibration_gpu_ticks": 452333,
              "compute_passes": 1,
              "command_buffer_ns": 7125.02514943,
              "command_buffers": 2
            },
            {
              "compute_ns": 7583,
              "compute_span_ns": 7583,
              "calibration_cpu_ns": 754834,
              "calibration_gpu_ticks": 754834,
              "compute_passes": 1,
              "command_buffer_ns": 7625.02895668,
              "command_buffers": 2
            },
            {
              "compute_ns": 7166,
              "compute_span_ns": 7166,
              "calibration_cpu_ns": 430416,
              "calibration_gpu_ticks": 430416,
              "compute_passes": 1,
              "command_buffer_ns": 7208.37851986,
              "command_buffers": 2
            },
            {
              "compute_ns": 7125,
              "compute_span_ns": 7125,
              "calibration_cpu_ns": 472000,
              "calibration_gpu_ticks": 472000,
              "compute_passes": 1,
              "command_buffer_ns": 7124.96694177,
              "command_buffers": 2
            },
            {
              "compute_ns": 7083,
              "compute_span_ns": 7083,
              "calibration_cpu_ns": 406708,
              "calibration_gpu_ticks": 406708,
              "compute_passes": 1,
              "command_buffer_ns": 7124.96694177,
              "command_buffers": 2
            },
            {
              "compute_ns": 7292,
              "compute_span_ns": 7292,
              "calibration_cpu_ns": 718458,
              "calibration_gpu_ticks": 718458,
              "compute_passes": 1,
              "command_buffer_ns": 7458.32221583,
              "command_buffers": 2
            },
            {
              "compute_ns": 7125,
              "compute_span_ns": 7125,
              "calibration_cpu_ns": 433709,
              "calibration_gpu_ticks": 433709,
              "compute_passes": 1,
              "command_buffer_ns": 7166.70183465,
              "command_buffers": 2
            },
            {
              "compute_ns": 7083,
              "compute_span_ns": 7083,
              "calibration_cpu_ns": 389875,
              "calibration_gpu_ticks": 389875,
              "compute_passes": 1,
              "command_buffer_ns": 7041.67177901,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 295333.331451,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 265208.363999,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 292583.310511,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 266083.341558,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 290750.002023,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 291624.979582,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 291500.007734,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 291916.658171,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 265666.691121,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 6958.37661624,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7208.3203122,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 8791.62689671,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7124.96694177,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6958.37661624,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 8249.94640425,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7125.02514943,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7208.26210454,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7124.96694177,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              4.614583303921875,
              4.1438806874843745,
              4.571614226734375,
              4.1575522118437505,
              4.542968781609375,
              4.55664030596875,
              4.55468762084375,
              4.561197783921875,
              4.151042048765625
            ],
            "command_buffer_throughput_us_p50": 4.55468762084375,
            "command_buffer_throughput_us_p90": 4.580208042171875,
            "command_buffer_latency_us": [
              6.95837661624,
              7.208320312200001,
              8.791626896710001,
              7.1249669417699995,
              6.95837661624,
              8.24994640425,
              7.12502514943,
              7.20826210454,
              7.1249669417699995
            ],
            "command_buffer_latency_us_p50": 7.12502514943,
            "command_buffer_latency_us_p90": 8.358282502742
          },
          "compute_throughput_us": [
            4.688796875,
            4.145828125,
            4.5703125,
            4.576171875,
            4.583984375,
            4.53840625,
            4.61003125,
            4.55859375,
            4.602859375
          ],
          "compute_throughput_us_p50": 4.576171875,
          "compute_throughput_us_p90": 4.625784374999999,
          "command_buffer_throughput_us": [
            4.6894529077656255,
            4.147135769015625,
            4.570963028531249,
            4.57747410109375,
            4.584635462375,
            4.5390634113593755,
            4.610677024171875,
            4.558593900578125,
            4.6028644646875
          ],
          "command_buffer_throughput_us_p50": 4.57747410109375,
          "command_buffer_throughput_us_p90": 4.626432200890625,
          "compute_latency_us": [
            7.041,
            7.166,
            7.583,
            7.166,
            7.125,
            7.083,
            7.292,
            7.125,
            7.083
          ],
          "compute_latency_us_p50": 7.125,
          "compute_latency_us_p90": 7.3502,
          "command_buffer_latency_us": [
            7.0416717790100005,
            7.12502514943,
            7.6250289566800005,
            7.20837851986,
            7.1249669417699995,
            7.1249669417699995,
            7.45832221583,
            7.1667018346499995,
            7.0416717790100005
          ],
          "command_buffer_latency_us_p50": 7.12502514943,
          "command_buffer_latency_us_p90": 7.491663564,
          "counter_control_throughput_ratio": 1.0091716854701487,
          "counter_control_latency_ratio": 0.9942343564527399
        },
        "correctness": {
          "max_abs_error": 1.5564224420394623e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 4.48615724001,
        "throughput_us_p90": 4.584195820205999,
        "latency_us_p50": 236.208,
        "latency_us_p90": 438.0746000000001
      },
      "slowdown": 0.5558863452834614,
      "device_compute_slowdown": 0.8523099308839578,
      "valid": true,
      "round": 2,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "rmsnorm_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-5u5hceat/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "a8c235e7c8be57d97298e1f9f70525f6c2c7f4ae4cee9647b15503a46aab5ade",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.249333,
        "capture_ms": 0.060042,
        "compile_ms": 57.388917,
        "allocation_upload_ms": 2.102791,
        "cold_call_ms": 6.208125,
        "warmup_ms": 200.039875,
        "download_ms": 0.963708,
        "repetitions": 1801,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          16.2803764575,
          16.0008789561,
          16.1503098279,
          15.6584069961,
          16.3587362576,
          15.6237273737,
          16.1908200999,
          17.3316213215,
          16.5526329817
        ],
        "latency_us": [
          289.667,
          302.5,
          258.833,
          217.333,
          254.666,
          275.583,
          244.542,
          301.125,
          235.208
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 128,
            "reduction_scalar_rounds": 36,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 24576,
              "global_write_bytes": 12288,
              "private_read_bytes": 36864,
              "private_write_bytes": 12288
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 48,
              "private_read_bytes": 144,
              "private_write_bytes": 48
            },
            "striped_storage_scalars_per_worker": 12,
            "reduction_operations": 1,
            "reduction_elements": 3072,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 60,
            "normalized_kernel_cost": 240,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 6144,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 988500,
              "compute_span_ns": 988500,
              "calibration_cpu_ns": 1461875,
              "calibration_gpu_ticks": 1461875,
              "compute_passes": 1,
              "command_buffer_ns": 988583.371509,
              "command_buffers": 2
            },
            {
              "compute_ns": 1006792,
              "compute_span_ns": 1006792,
              "calibration_cpu_ns": 1695875,
              "calibration_gpu_ticks": 1695875,
              "compute_passes": 1,
              "command_buffer_ns": 1006833.33585,
              "command_buffers": 2
            },
            {
              "compute_ns": 1016917,
              "compute_span_ns": 1016917,
              "calibration_cpu_ns": 1718417,
              "calibration_gpu_ticks": 1718417,
              "compute_passes": 1,
              "command_buffer_ns": 1016958.38384,
              "command_buffers": 2
            },
            {
              "compute_ns": 957041,
              "compute_span_ns": 957041,
              "calibration_cpu_ns": 1473833,
              "calibration_gpu_ticks": 1473833,
              "compute_passes": 1,
              "command_buffer_ns": 957000.011113,
              "command_buffers": 2
            },
            {
              "compute_ns": 940292,
              "compute_span_ns": 940292,
              "calibration_cpu_ns": 1393541,
              "calibration_gpu_ticks": 1393541,
              "compute_passes": 1,
              "command_buffer_ns": 940374.971833,
              "command_buffers": 2
            },
            {
              "compute_ns": 953125,
              "compute_span_ns": 953125,
              "calibration_cpu_ns": 1464834,
              "calibration_gpu_ticks": 1464834,
              "compute_passes": 1,
              "command_buffer_ns": 953166.687395,
              "command_buffers": 2
            },
            {
              "compute_ns": 960625,
              "compute_span_ns": 960625,
              "calibration_cpu_ns": 1424708,
              "calibration_gpu_ticks": 1424708,
              "compute_passes": 1,
              "command_buffer_ns": 960708.304774,
              "command_buffers": 2
            },
            {
              "compute_ns": 1013375,
              "compute_span_ns": 1013375,
              "calibration_cpu_ns": 1833208,
              "calibration_gpu_ticks": 1833208,
              "compute_passes": 1,
              "command_buffer_ns": 1013499.97567,
              "command_buffers": 2
            },
            {
              "compute_ns": 986750,
              "compute_span_ns": 986750,
              "calibration_cpu_ns": 1749834,
              "calibration_gpu_ticks": 1749834,
              "compute_passes": 1,
              "command_buffer_ns": 986749.946605,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 17000,
              "compute_span_ns": 17000,
              "calibration_cpu_ns": 555125,
              "calibration_gpu_ticks": 555125,
              "compute_passes": 1,
              "command_buffer_ns": 17000.0130311,
              "command_buffers": 2
            },
            {
              "compute_ns": 19334,
              "compute_span_ns": 19334,
              "calibration_cpu_ns": 443042,
              "calibration_gpu_ticks": 443042,
              "compute_passes": 1,
              "command_buffer_ns": 19458.3553821,
              "command_buffers": 2
            },
            {
              "compute_ns": 19333,
              "compute_span_ns": 19333,
              "calibration_cpu_ns": 460500,
              "calibration_gpu_ticks": 460500,
              "compute_passes": 1,
              "command_buffer_ns": 19333.3253264,
              "command_buffers": 2
            },
            {
              "compute_ns": 19292,
              "compute_span_ns": 19292,
              "calibration_cpu_ns": 497625,
              "calibration_gpu_ticks": 497625,
              "compute_passes": 1,
              "command_buffer_ns": 19375.0020117,
              "command_buffers": 2
            },
            {
              "compute_ns": 24625,
              "compute_span_ns": 24625,
              "calibration_cpu_ns": 700458,
              "calibration_gpu_ticks": 700458,
              "compute_passes": 1,
              "command_buffer_ns": 24666.718673,
              "command_buffers": 2
            },
            {
              "compute_ns": 17583,
              "compute_span_ns": 17583,
              "calibration_cpu_ns": 622834,
              "calibration_gpu_ticks": 622834,
              "compute_passes": 1,
              "command_buffer_ns": 17583.3120011,
              "command_buffers": 2
            },
            {
              "compute_ns": 19458,
              "compute_span_ns": 19458,
              "calibration_cpu_ns": 445542,
              "calibration_gpu_ticks": 445542,
              "compute_passes": 1,
              "command_buffer_ns": 19499.9738596,
              "command_buffers": 2
            },
            {
              "compute_ns": 26375,
              "compute_span_ns": 26375,
              "calibration_cpu_ns": 913458,
              "calibration_gpu_ticks": 913458,
              "compute_passes": 1,
              "command_buffer_ns": 26374.9971054,
              "command_buffers": 2
            },
            {
              "compute_ns": 19792,
              "compute_span_ns": 19792,
              "calibration_cpu_ns": 409375,
              "calibration_gpu_ticks": 409375,
              "compute_passes": 1,
              "command_buffer_ns": 19791.6524485,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 991000.037175,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 957624.986768,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 957000.011113,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 961875.019129,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1017583.30129,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 940125.028137,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 974916.678388,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 963458.383922,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 957291.689701,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 16791.629605,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 24666.718673,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19583.3272301,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19250.0301637,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19500.0320673,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18958.3515748,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19250.0301637,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19208.3534785,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19208.3534785,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              15.484375580859375,
              14.96289041825,
              14.953125173640625,
              15.029297173890626,
              15.89973908265625,
              14.689453564640626,
              15.2330730998125,
              15.054037248781249,
              14.957682651578125
            ],
            "command_buffer_throughput_us_p50": 15.029297173890626,
            "command_buffer_throughput_us_p90": 15.56744828121875,
            "command_buffer_latency_us": [
              16.791629604999997,
              24.666718673,
              19.5833272301,
              19.2500301637,
              19.500032067299998,
              18.958351574799998,
              19.2500301637,
              19.2083534785,
              19.2083534785
            ],
            "command_buffer_latency_us_p50": 19.2500301637,
            "command_buffer_latency_us_p90": 20.60000551868
          },
          "compute_throughput_us": [
            15.4453125,
            15.731125,
            15.889328125,
            14.953765625,
            14.6920625,
            14.892578125,
            15.009765625,
            15.833984375,
            15.41796875
          ],
          "compute_throughput_us_p50": 15.41796875,
          "compute_throughput_us_p90": 15.845053125,
          "command_buffer_throughput_us": [
            15.446615179828125,
            15.73177087265625,
            15.8899747475,
            14.953125173640625,
            14.693358934890624,
            14.893229490546876,
            15.01106726209375,
            15.835937119843749,
            15.417967915703125
          ],
          "command_buffer_throughput_us_p50": 15.417967915703125,
          "command_buffer_throughput_us_p90": 15.846744645374999,
          "compute_latency_us": [
            17.0,
            19.334,
            19.333,
            19.292,
            24.625,
            17.583,
            19.458,
            26.375,
            19.792
          ],
          "compute_latency_us_p50": 19.334,
          "compute_latency_us_p90": 24.975,
          "command_buffer_latency_us": [
            17.0000130311,
            19.4583553821,
            19.3333253264,
            19.375002011699998,
            24.666718673,
            17.583312001099998,
            19.499973859599997,
            26.374997105400002,
            19.791652448500002
          ],
          "command_buffer_latency_us_p50": 19.4583553821,
          "command_buffer_latency_us_p90": 25.008374359479998,
          "counter_control_throughput_ratio": 1.0138722604628918,
          "counter_control_latency_ratio": 1.0124099584734736
        },
        "correctness": {
          "max_abs_error": 1.6175184480715643e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 16.1908200999,
        "throughput_us_p90": 16.70843064966,
        "latency_us_p50": 258.833,
        "latency_us_p90": 301.4
      },
      "torch": {
        "cold_call_ms": 5.662,
        "warmup_ms": 200.049125,
        "repetitions": 1432,
        "throughput_us": [
          20.356639664804472,
          20.388617318435752,
          20.570879888268156,
          20.325418994413408,
          20.291986731843576,
          19.886900139664807,
          20.035323324022347,
          21.34843505586592,
          20.309648743016762
        ],
        "latency_us": [
          413.792,
          247.958,
          264.75,
          264.875,
          282.58299999999997,
          225.5,
          241.875,
          237.625,
          256.666
        ],
        "throughput_us_p50": 20.325418994413408,
        "throughput_us_p90": 20.726390921787708,
        "latency_us_p50": 256.666,
        "latency_us_p90": 308.8248,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1152458.309661597,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1134791.6442900896,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1033874.9852962792,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1033458.3348594606,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1053416.6358411312,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1093833.358027041,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1030624.9605491757,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1110750.0176876783,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1147250.0045783818,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 18624.996300786734,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 20999.98528137803,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 24000.00812485814,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 26791.647542268038,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21083.338651806116,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 20999.98528137803,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 20874.955225735903,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21625.019144266844,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21124.957129359245,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              18.007161088462453,
              17.73111944203265,
              16.154296645254362,
              16.14778648217907,
              16.459634935017675,
              17.091146219172515,
              16.10351500858087,
              17.355469026369974,
              17.925781321537215
            ],
            "command_buffer_throughput_us_p50": 17.091146219172515,
            "command_buffer_throughput_us_p90": 17.942057274922263,
            "command_buffer_latency_us": [
              18.624996300786734,
              20.99998528137803,
              24.00000812485814,
              26.791647542268038,
              21.083338651806116,
              20.99998528137803,
              20.874955225735903,
              21.625019144266844,
              21.124957129359245
            ],
            "command_buffer_latency_us_p50": 21.083338651806116,
            "command_buffer_latency_us_p90": 24.55833600834012
          },
          "throughput": [
            {
              "compute_ns": 1115542.0,
              "compute_span_ns": 1115542.0,
              "command_buffer_ns": 1115583.349019289,
              "calibration_cpu_ns": 1815917.0,
              "calibration_gpu_ticks": 1815917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1111334.0,
              "compute_span_ns": 1111334.0,
              "command_buffer_ns": 1111374.9933429062,
              "calibration_cpu_ns": 2359834.0,
              "calibration_gpu_ticks": 2359834.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1057125.0,
              "compute_span_ns": 1057125.0,
              "command_buffer_ns": 1057125.0459179282,
              "calibration_cpu_ns": 2053667.0,
              "calibration_gpu_ticks": 2053667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1129916.0,
              "compute_span_ns": 1129916.0,
              "command_buffer_ns": 1129875.0177957118,
              "calibration_cpu_ns": 1814875.0,
              "calibration_gpu_ticks": 1814875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1031625.0,
              "compute_span_ns": 1031625.0,
              "command_buffer_ns": 1031625.02637133,
              "calibration_cpu_ns": 1677500.0,
              "calibration_gpu_ticks": 1677500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1036917.0,
              "compute_span_ns": 1036917.0,
              "command_buffer_ns": 1036958.3033025265,
              "calibration_cpu_ns": 1716500.0,
              "calibration_gpu_ticks": 1716500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1038709.0,
              "compute_span_ns": 1038709.0,
              "command_buffer_ns": 1038749.9933131039,
              "calibration_cpu_ns": 1947708.0,
              "calibration_gpu_ticks": 1947708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1153042.0,
              "compute_span_ns": 1153042.0,
              "command_buffer_ns": 1153041.6668392718,
              "calibration_cpu_ns": 2111041.0,
              "calibration_gpu_ticks": 2111041.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1147625.0,
              "compute_span_ns": 1147625.0,
              "command_buffer_ns": 1147624.9783299863,
              "calibration_cpu_ns": 1767875.0,
              "calibration_gpu_ticks": 1767875.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 28917.0,
              "compute_span_ns": 28917.0,
              "command_buffer_ns": 28916.634619235992,
              "calibration_cpu_ns": 935917.0,
              "calibration_gpu_ticks": 935917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21541.0,
              "compute_span_ns": 21541.0,
              "command_buffer_ns": 21499.989088624716,
              "calibration_cpu_ns": 721875.0,
              "calibration_gpu_ticks": 721875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21041.0,
              "compute_span_ns": 21041.0,
              "command_buffer_ns": 21000.043489038944,
              "calibration_cpu_ns": 520875.0,
              "calibration_gpu_ticks": 520875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 20958.0,
              "compute_span_ns": 20958.0,
              "command_buffer_ns": 20958.3668038249,
              "calibration_cpu_ns": 435458.0,
              "calibration_gpu_ticks": 435458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 20916.0,
              "compute_span_ns": 20916.0,
              "command_buffer_ns": 20874.955225735903,
              "calibration_cpu_ns": 497417.0,
              "calibration_gpu_ticks": 497417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21125.0,
              "compute_span_ns": 21125.0,
              "command_buffer_ns": 21124.957129359245,
              "calibration_cpu_ns": 441917.0,
              "calibration_gpu_ticks": 441917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 30416.0,
              "compute_span_ns": 30416.0,
              "command_buffer_ns": 30375.02756342292,
              "calibration_cpu_ns": 811792.0,
              "calibration_gpu_ticks": 811792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21416.0,
              "compute_span_ns": 21416.0,
              "command_buffer_ns": 21374.959032982588,
              "calibration_cpu_ns": 638542.0,
              "calibration_gpu_ticks": 638542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21167.0,
              "compute_span_ns": 21167.0,
              "command_buffer_ns": 21166.633814573288,
              "calibration_cpu_ns": 429167.0,
              "calibration_gpu_ticks": 429167.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            17.43034375,
            17.36459375,
            16.517578125,
            17.6549375,
            16.119140625,
            16.201828125,
            16.229828125,
            18.01628125,
            17.931640625
          ],
          "compute_throughput_us_p50": 17.36459375,
          "compute_throughput_us_p90": 17.94856875,
          "command_buffer_throughput_us": [
            17.43098982842639,
            17.36523427098291,
            16.51757884246763,
            17.654297153057996,
            16.11914103705203,
            16.202473489101976,
            16.23046864551725,
            18.01627604436362,
            17.931640286406036
          ],
          "command_buffer_throughput_us_p50": 17.36523427098291,
          "command_buffer_throughput_us_p90": 17.948567437997553,
          "compute_latency_us": [
            28.917,
            21.541,
            21.041,
            20.958,
            20.916,
            21.125,
            30.416,
            21.416,
            21.167
          ],
          "compute_latency_us_p50": 21.167,
          "compute_latency_us_p90": 29.216800000000003,
          "command_buffer_latency_us": [
            28.916634619235992,
            21.499989088624716,
            21.000043489038944,
            20.9583668038249,
            20.874955225735903,
            21.124957129359245,
            30.37502756342292,
            21.374959032982588,
            21.166633814573288
          ],
          "command_buffer_latency_us_p50": 21.166633814573288,
          "command_buffer_latency_us_p90": 29.208313208073378,
          "counter_control_throughput_ratio": 1.0003268457181156,
          "counter_control_latency_ratio": 1.001972864842226
        },
        "download_ms": 0.341709,
        "allocation_upload_ms": 1.4285,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.8163651027691685e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.7965798935977737,
      "device_compute_slowdown": 0.8878968878842903,
      "valid": true,
      "round": 2,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "rmsnorm_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.288875,
        "warmup_ms": 200.420584,
        "repetitions": 736,
        "throughput_us": [
          23.136323369565222,
          20.81148097826087,
          22.582258152173914,
          20.159703804347824,
          20.377661684782606,
          20.809158967391305,
          20.808990489130434,
          20.504642663043477,
          21.930254076086957
        ],
        "latency_us": [
          310.75,
          373.584,
          461.625,
          335.583,
          273.084,
          260.29200000000003,
          242.125,
          255.91599999999997,
          257.084
        ],
        "throughput_us_p50": 20.809158967391305,
        "throughput_us_p90": 22.693071195652177,
        "latency_us_p50": 273.084,
        "latency_us_p90": 391.1922,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1095124.9860227108,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1033583.3649151027,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1071624.9817050993,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1049583.3703316748,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1028999.9772794545,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1146750.000771135,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1035749.9704696238,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1059208.3563096821,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1095500.0179819763,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 19125.00010803342,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 23874.978069216013,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21375.0172406435,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21000.043489038944,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 20916.69011861086,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21333.282347768545,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 20999.98528137803,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 28625.01423805952,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 25541.69623181224,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              17.111327906604856,
              16.14974007679848,
              16.744140339142177,
              16.39974016143242,
              16.078124644991476,
              17.917968762048986,
              16.183593288587872,
              16.550130567338783,
              17.11718778096838
            ],
            "command_buffer_throughput_us_p50": 16.550130567338783,
            "command_buffer_throughput_us_p90": 17.2773439771845,
            "command_buffer_latency_us": [
              19.12500010803342,
              23.874978069216013,
              21.3750172406435,
              21.000043489038944,
              20.91669011861086,
              21.333282347768545,
              20.99998528137803,
              28.62501423805952,
              25.54169623181224
            ],
            "command_buffer_latency_us_p50": 21.333282347768545,
            "command_buffer_latency_us_p90": 26.158359833061695
          },
          "throughput": [
            {
              "compute_ns": 1049417.0,
              "compute_span_ns": 1049417.0,
              "command_buffer_ns": 1049458.3402760327,
              "calibration_cpu_ns": 1986292.0,
              "calibration_gpu_ticks": 1986292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1105500.0,
              "compute_span_ns": 1105500.0,
              "command_buffer_ns": 1105499.9777115881,
              "calibration_cpu_ns": 1820500.0,
              "calibration_gpu_ticks": 1820500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1120125.0,
              "compute_span_ns": 1120125.0,
              "command_buffer_ns": 1120125.0017620623,
              "calibration_cpu_ns": 1876458.0,
              "calibration_gpu_ticks": 1876458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1054625.0,
              "compute_span_ns": 1054625.0,
              "command_buffer_ns": 1054625.0268816948,
              "calibration_cpu_ns": 1681459.0,
              "calibration_gpu_ticks": 1681459.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1037334.0,
              "compute_span_ns": 1037334.0,
              "command_buffer_ns": 1037375.011947006,
              "calibration_cpu_ns": 1712667.0,
              "calibration_gpu_ticks": 1712667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1072500.0,
              "compute_span_ns": 1072500.0,
              "command_buffer_ns": 1072499.9592639506,
              "calibration_cpu_ns": 2074209.0,
              "calibration_gpu_ticks": 2074209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1042166.0,
              "compute_span_ns": 1042166.0,
              "command_buffer_ns": 1042124.9899081886,
              "calibration_cpu_ns": 1901125.0,
              "calibration_gpu_ticks": 1901125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1163500.0,
              "compute_span_ns": 1163500.0,
              "command_buffer_ns": 1163500.0118985772,
              "calibration_cpu_ns": 1837375.0,
              "calibration_gpu_ticks": 1837375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1156833.0,
              "compute_span_ns": 1156833.0,
              "command_buffer_ns": 1156791.695393622,
              "calibration_cpu_ns": 1789375.0,
              "calibration_gpu_ticks": 1789375.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 25209.0,
              "compute_span_ns": 25209.0,
              "command_buffer_ns": 25250.017642974854,
              "calibration_cpu_ns": 684875.0,
              "calibration_gpu_ticks": 684875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 19083.0,
              "compute_span_ns": 19083.0,
              "command_buffer_ns": 19041.646737605333,
              "calibration_cpu_ns": 448750.0,
              "calibration_gpu_ticks": 448750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21333.0,
              "compute_span_ns": 21333.0,
              "command_buffer_ns": 21333.34055542946,
              "calibration_cpu_ns": 448458.0,
              "calibration_gpu_ticks": 448458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21125.0,
              "compute_span_ns": 21125.0,
              "command_buffer_ns": 21125.01533702016,
              "calibration_cpu_ns": 387750.0,
              "calibration_gpu_ticks": 387750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 30083.0,
              "compute_span_ns": 30083.0,
              "command_buffer_ns": 30083.29076692462,
              "calibration_cpu_ns": 1019583.0,
              "calibration_gpu_ticks": 1019583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21125.0,
              "compute_span_ns": 21125.0,
              "command_buffer_ns": 21125.01533702016,
              "calibration_cpu_ns": 623792.0,
              "calibration_gpu_ticks": 623792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21166.0,
              "compute_span_ns": 21166.0,
              "command_buffer_ns": 21125.01533702016,
              "calibration_cpu_ns": 559125.0,
              "calibration_gpu_ticks": 559125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 20917.0,
              "compute_span_ns": 20917.0,
              "command_buffer_ns": 20916.69011861086,
              "calibration_cpu_ns": 479542.0,
              "calibration_gpu_ticks": 479542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21167.0,
              "compute_span_ns": 21167.0,
              "command_buffer_ns": 21166.6920222342,
              "calibration_cpu_ns": 447083.0,
              "calibration_gpu_ticks": 447083.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            16.397140625,
            17.2734375,
            17.501953125,
            16.478515625,
            16.20834375,
            16.7578125,
            16.28384375,
            18.1796875,
            18.075515625
          ],
          "compute_throughput_us_p50": 16.7578125,
          "compute_throughput_us_p90": 18.09635,
          "command_buffer_throughput_us": [
            16.39778656681301,
            17.273437151743565,
            17.501953152532224,
            16.47851604502648,
            16.20898456167197,
            16.757811863499228,
            16.283202967315447,
            18.17968768591527,
            18.074870240525343
          ],
          "command_buffer_throughput_us_p50": 16.757811863499228,
          "command_buffer_throughput_us_p90": 18.095833729603328,
          "compute_latency_us": [
            25.209,
            19.083,
            21.333,
            21.125,
            30.083,
            21.125,
            21.166,
            20.917,
            21.167
          ],
          "compute_latency_us_p50": 21.166,
          "compute_latency_us_p90": 26.1838,
          "command_buffer_latency_us": [
            25.250017642974854,
            19.041646737605333,
            21.33334055542946,
            21.12501533702016,
            30.08329076692462,
            21.12501533702016,
            21.12501533702016,
            20.91669011861086,
            21.1666920222342
          ],
          "command_buffer_latency_us_p50": 21.12501533702016,
          "command_buffer_latency_us_p90": 26.216672267764807,
          "counter_control_throughput_ratio": 1.008139003743901,
          "counter_control_latency_ratio": 0.9980502151298949
        },
        "download_ms": 0.370959,
        "allocation_upload_ms": 0.711208,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.8163651027691685e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-74duy0bz/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "b5a32df1046f38df4f8e9cbba5277df92f4beb4d4f9631ac4564ef5cda134264",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.079166,
        "capture_ms": 0.062542,
        "compile_ms": 111.46075,
        "allocation_upload_ms": 2.113042,
        "cold_call_ms": 3.600333,
        "warmup_ms": 200.200166,
        "download_ms": 0.8845,
        "repetitions": 1898,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          15.8496227608,
          15.7541053741,
          15.7845321391,
          15.7956622761,
          16.113869863,
          16.0145985248,
          15.7049083246,
          15.9046364594,
          15.7390458377
        ],
        "latency_us": [
          1404.292,
          631.375,
          319.667,
          258.041,
          244.792,
          238.834,
          256.333,
          228.834,
          225.584
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 32,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 256,
            "reduction_scalar_rounds": 36,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 24576,
              "global_write_bytes": 12288,
              "private_read_bytes": 36864,
              "private_write_bytes": 12288
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 48,
              "private_read_bytes": 144,
              "private_write_bytes": 48
            },
            "striped_storage_scalars_per_worker": 12,
            "reduction_operations": 1,
            "reduction_elements": 3072,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 68,
            "normalized_kernel_cost": 272,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 6144,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 937708,
              "compute_span_ns": 937708,
              "calibration_cpu_ns": 1874292,
              "calibration_gpu_ticks": 1874292,
              "compute_passes": 1,
              "command_buffer_ns": 937750.039157,
              "command_buffers": 2
            },
            {
              "compute_ns": 926583,
              "compute_span_ns": 926583,
              "calibration_cpu_ns": 1474875,
              "calibration_gpu_ticks": 1474875,
              "compute_passes": 1,
              "command_buffer_ns": 926625.041757,
              "command_buffers": 2
            },
            {
              "compute_ns": 952791,
              "compute_span_ns": 952791,
              "calibration_cpu_ns": 2122250,
              "calibration_gpu_ticks": 2122250,
              "compute_passes": 1,
              "command_buffer_ns": 952791.655436,
              "command_buffers": 2
            },
            {
              "compute_ns": 934750,
              "compute_span_ns": 934750,
              "calibration_cpu_ns": 1461334,
              "calibration_gpu_ticks": 1461334,
              "compute_passes": 1,
              "command_buffer_ns": 934874.988161,
              "command_buffers": 2
            },
            {
              "compute_ns": 1234042,
              "compute_span_ns": 1234042,
              "calibration_cpu_ns": 2200958,
              "calibration_gpu_ticks": 2200958,
              "compute_passes": 1,
              "command_buffer_ns": 1234166.67338,
              "command_buffers": 2
            },
            {
              "compute_ns": 940166,
              "compute_span_ns": 940166,
              "calibration_cpu_ns": 1429875,
              "calibration_gpu_ticks": 1429875,
              "compute_passes": 1,
              "command_buffer_ns": 940208.3233,
              "command_buffers": 2
            },
            {
              "compute_ns": 983708,
              "compute_span_ns": 983708,
              "calibration_cpu_ns": 1924875,
              "calibration_gpu_ticks": 1924875,
              "compute_passes": 1,
              "command_buffer_ns": 983750.040177,
              "command_buffers": 2
            },
            {
              "compute_ns": 917625,
              "compute_span_ns": 917625,
              "calibration_cpu_ns": 1397375,
              "calibration_gpu_ticks": 1397375,
              "compute_passes": 1,
              "command_buffer_ns": 917708.384804,
              "command_buffers": 2
            },
            {
              "compute_ns": 933125,
              "compute_span_ns": 933125,
              "calibration_cpu_ns": 1434500,
              "calibration_gpu_ticks": 1434500,
              "compute_passes": 1,
              "command_buffer_ns": 933208.269998,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 18417,
              "compute_span_ns": 18417,
              "calibration_cpu_ns": 520500,
              "calibration_gpu_ticks": 520500,
              "compute_passes": 1,
              "command_buffer_ns": 18500.0244528,
              "command_buffers": 2
            },
            {
              "compute_ns": 18417,
              "compute_span_ns": 18417,
              "calibration_cpu_ns": 451542,
              "calibration_gpu_ticks": 451542,
              "compute_passes": 1,
              "command_buffer_ns": 18416.6710824,
              "command_buffers": 2
            },
            {
              "compute_ns": 25208,
              "compute_span_ns": 25208,
              "calibration_cpu_ns": 1125750,
              "calibration_gpu_ticks": 1125750,
              "compute_passes": 1,
              "command_buffer_ns": 25166.6642725,
              "command_buffers": 2
            },
            {
              "compute_ns": 18417,
              "compute_span_ns": 18417,
              "calibration_cpu_ns": 456125,
              "calibration_gpu_ticks": 456125,
              "compute_passes": 1,
              "command_buffer_ns": 18416.6710824,
              "command_buffers": 2
            },
            {
              "compute_ns": 18542,
              "compute_span_ns": 18542,
              "calibration_cpu_ns": 527500,
              "calibration_gpu_ticks": 527500,
              "compute_passes": 1,
              "command_buffer_ns": 18541.6429304,
              "command_buffers": 2
            },
            {
              "compute_ns": 25083,
              "compute_span_ns": 25083,
              "calibration_cpu_ns": 1381417,
              "calibration_gpu_ticks": 1381417,
              "compute_passes": 1,
              "command_buffer_ns": 25125.045795,
              "command_buffers": 2
            },
            {
              "compute_ns": 18458,
              "compute_span_ns": 18458,
              "calibration_cpu_ns": 462167,
              "calibration_gpu_ticks": 462167,
              "compute_passes": 1,
              "command_buffer_ns": 18458.3477676,
              "command_buffers": 2
            },
            {
              "compute_ns": 20458,
              "compute_span_ns": 20458,
              "calibration_cpu_ns": 540375,
              "calibration_gpu_ticks": 540375,
              "compute_passes": 1,
              "command_buffer_ns": 20500.0396818,
              "command_buffers": 2
            },
            {
              "compute_ns": 18500,
              "compute_span_ns": 18500,
              "calibration_cpu_ns": 409625,
              "calibration_gpu_ticks": 409625,
              "compute_passes": 1,
              "command_buffer_ns": 18541.701138,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 987166.713458,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 990125.001408,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 934666.662943,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 923749.990761,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 937083.386816,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1208791.68389,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 912333.256565,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 958624.994382,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 908958.376385,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 23833.3595917,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18374.9943972,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18291.6410267,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18416.6710824,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18583.3196156,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18249.9643415,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18624.9963008,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 26750.0290647,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18583.3778232,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              15.42447989778125,
              15.470703147,
              14.604166608484375,
              14.433593605640624,
              14.641927919,
              18.88737006078125,
              14.255207133828126,
              14.97851553721875,
              14.202474631015624
            ],
            "command_buffer_throughput_us_p50": 14.641927919,
            "command_buffer_throughput_us_p90": 16.15403652975625,
            "command_buffer_latency_us": [
              23.8333595917,
              18.374994397200002,
              18.2916410267,
              18.4166710824,
              18.5833196156,
              18.2499643415,
              18.624996300800003,
              26.7500290647,
              18.5833778232
            ],
            "command_buffer_latency_us_p50": 18.5833196156,
            "command_buffer_latency_us_p90": 24.4166934863
          },
          "compute_throughput_us": [
            14.6516875,
            14.477859375,
            14.887359375,
            14.60546875,
            19.28190625,
            14.69009375,
            15.3704375,
            14.337890625,
            14.580078125
          ],
          "compute_throughput_us_p50": 14.6516875,
          "compute_throughput_us_p90": 16.15273125,
          "command_buffer_throughput_us": [
            14.652344361828126,
            14.478516277453126,
            14.887369616187499,
            14.607421690015624,
            19.2838542715625,
            14.690755051562501,
            15.371094377765624,
            14.3391935125625,
            14.58137921871875
          ],
          "command_buffer_throughput_us_p50": 14.652344361828126,
          "command_buffer_throughput_us_p90": 16.153646356525,
          "compute_latency_us": [
            18.417,
            18.417,
            25.208,
            18.417,
            18.542,
            25.083,
            18.458,
            20.458,
            18.5
          ],
          "compute_latency_us_p50": 18.5,
          "compute_latency_us_p90": 25.107999999999997,
          "command_buffer_latency_us": [
            18.5000244528,
            18.4166710824,
            25.166664272499997,
            18.4166710824,
            18.541642930400002,
            25.125045795000002,
            18.4583477676,
            20.5000396818,
            18.541701138
          ],
          "command_buffer_latency_us_p50": 18.541642930400002,
          "command_buffer_latency_us_p90": 25.1333694905,
          "counter_control_throughput_ratio": 1.0120432990649721,
          "counter_control_latency_ratio": 0.9977573137888867
        },
        "correctness": {
          "max_abs_error": 1.6175184480715643e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 15.7956622761,
        "throughput_us_p90": 16.03445279244,
        "latency_us_p50": 256.333,
        "latency_us_p90": 785.9584000000001
      },
      "slowdown": 0.759072593988655,
      "device_compute_slowdown": 0.8743198135198135,
      "valid": true,
      "round": 2,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "rmsnorm_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-enih3m58/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "232522da57db77c64b377ec2facdc2e1df33cd57064a4297b89d6582fd4ad63b",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.710333,
        "capture_ms": 0.065,
        "compile_ms": 54.660542,
        "allocation_upload_ms": 6.511833,
        "cold_call_ms": 5.2535,
        "warmup_ms": 200.612917,
        "download_ms": 3.524833,
        "repetitions": 332,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          64.2206325301,
          65.8682228916,
          65.9956084337,
          64.7289156627,
          65.6670421687,
          63.7980692771,
          65.7510060241,
          66.4986204819,
          65.5813253012
        ],
        "latency_us": [
          376.583,
          338.375,
          309.042,
          300.875,
          280.667,
          1484.125,
          329.458,
          300.584,
          266.833
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 32,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 768,
            "reduction_scalar_rounds": 72,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 49152,
              "global_write_bytes": 24576,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 192,
              "global_write_bytes": 96,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 1,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 104,
            "normalized_kernel_cost": 1248,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3838084,
              "compute_span_ns": 3838084,
              "calibration_cpu_ns": 4576042,
              "calibration_gpu_ticks": 4576042,
              "compute_passes": 1,
              "command_buffer_ns": 3838166.65271,
              "command_buffers": 2
            },
            {
              "compute_ns": 3904709,
              "compute_span_ns": 3904709,
              "calibration_cpu_ns": 4587666,
              "calibration_gpu_ticks": 4587666,
              "compute_passes": 1,
              "command_buffer_ns": 3904833.28374,
              "command_buffers": 2
            },
            {
              "compute_ns": 3932125,
              "compute_span_ns": 3932125,
              "calibration_cpu_ns": 4830250,
              "calibration_gpu_ticks": 4830250,
              "compute_passes": 1,
              "command_buffer_ns": 3932249.96515,
              "command_buffers": 2
            },
            {
              "compute_ns": 3886541,
              "compute_span_ns": 3886541,
              "calibration_cpu_ns": 4707167,
              "calibration_gpu_ticks": 4707167,
              "compute_passes": 1,
              "command_buffer_ns": 3886541.70092,
              "command_buffers": 2
            },
            {
              "compute_ns": 3896500,
              "compute_span_ns": 3896500,
              "calibration_cpu_ns": 4751208,
              "calibration_gpu_ticks": 4751208,
              "compute_passes": 1,
              "command_buffer_ns": 3896541.66065,
              "command_buffers": 2
            },
            {
              "compute_ns": 3948500,
              "compute_span_ns": 3948500,
              "calibration_cpu_ns": 4774208,
              "calibration_gpu_ticks": 4774208,
              "compute_passes": 1,
              "command_buffer_ns": 3948541.70736,
              "command_buffers": 2
            },
            {
              "compute_ns": 3911709,
              "compute_span_ns": 3911709,
              "calibration_cpu_ns": 4588625,
              "calibration_gpu_ticks": 4588625,
              "compute_passes": 1,
              "command_buffer_ns": 3911749.98367,
              "command_buffers": 2
            },
            {
              "compute_ns": 3845541,
              "compute_span_ns": 3845541,
              "calibration_cpu_ns": 4579500,
              "calibration_gpu_ticks": 4579500,
              "compute_passes": 1,
              "command_buffer_ns": 3845541.67977,
              "command_buffers": 2
            },
            {
              "compute_ns": 3864125,
              "compute_span_ns": 3864125,
              "calibration_cpu_ns": 4589375,
              "calibration_gpu_ticks": 4589375,
              "compute_passes": 1,
              "command_buffer_ns": 3864124.99938,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 88416,
              "compute_span_ns": 88416,
              "calibration_cpu_ns": 697458,
              "calibration_gpu_ticks": 697458,
              "compute_passes": 1,
              "command_buffer_ns": 88416.680228,
              "command_buffers": 2
            },
            {
              "compute_ns": 120125,
              "compute_span_ns": 120125,
              "calibration_cpu_ns": 718000,
              "calibration_gpu_ticks": 718000,
              "compute_passes": 1,
              "command_buffer_ns": 120166.689157,
              "command_buffers": 2
            },
            {
              "compute_ns": 114709,
              "compute_span_ns": 114709,
              "calibration_cpu_ns": 986709,
              "calibration_gpu_ticks": 986709,
              "compute_passes": 1,
              "command_buffer_ns": 114750.000648,
              "command_buffers": 2
            },
            {
              "compute_ns": 69000,
              "compute_span_ns": 69000,
              "calibration_cpu_ns": 461208,
              "calibration_gpu_ticks": 461208,
              "compute_passes": 1,
              "command_buffer_ns": 69083.3549015,
              "command_buffers": 2
            },
            {
              "compute_ns": 75917,
              "compute_span_ns": 75917,
              "calibration_cpu_ns": 681792,
              "calibration_gpu_ticks": 681792,
              "compute_passes": 1,
              "command_buffer_ns": 76041.6733101,
              "command_buffers": 2
            },
            {
              "compute_ns": 113584,
              "compute_span_ns": 113584,
              "calibration_cpu_ns": 1041458,
              "calibration_gpu_ticks": 1041458,
              "compute_passes": 1,
              "command_buffer_ns": 113708.258141,
              "command_buffers": 2
            },
            {
              "compute_ns": 80625,
              "compute_span_ns": 80625,
              "calibration_cpu_ns": 486333,
              "calibration_gpu_ticks": 486333,
              "compute_passes": 1,
              "command_buffer_ns": 80666.6794233,
              "command_buffers": 2
            },
            {
              "compute_ns": 71583,
              "compute_span_ns": 71583,
              "calibration_cpu_ns": 495208,
              "calibration_gpu_ticks": 495208,
              "compute_passes": 1,
              "command_buffer_ns": 71541.6972525,
              "command_buffers": 2
            },
            {
              "compute_ns": 92917,
              "compute_span_ns": 92917,
              "calibration_cpu_ns": 625750,
              "calibration_gpu_ticks": 625750,
              "compute_passes": 1,
              "command_buffer_ns": 93000.009656,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 3936249.9956,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3842958.30736,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3915916.66266,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3875916.59091,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3834875.00949,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3866999.99217,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3875416.70352,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3838833.30505,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3878708.28854,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 112624.955364,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 92958.3329707,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 69166.6500643,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 67499.9901094,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 64874.9992251,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 112875.015475,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 78291.6904427,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 109208.340291,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 98083.3428912,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              61.50390618125,
              60.0462235525,
              61.1861978540625,
              60.561196732968746,
              59.91992202328125,
              60.42187487765625,
              60.5533859925,
              59.98177039140625,
              60.6048170084375
            ],
            "command_buffer_throughput_us_p50": 60.5533859925,
            "command_buffer_throughput_us_p90": 61.2497395195,
            "command_buffer_latency_us": [
              112.62495536399999,
              92.9583329707,
              69.1666500643,
              67.49999010939999,
              64.8749992251,
              112.875015475,
              78.2916904427,
              109.208340291,
              98.08334289119999
            ],
            "command_buffer_latency_us_p50": 92.9583329707,
            "command_buffer_latency_us_p90": 112.6749673862
          },
          "compute_throughput_us": [
            59.9700625,
            61.011078125,
            61.439453125,
            60.727203125,
            60.8828125,
            61.6953125,
            61.120453125,
            60.086578125,
            60.376953125
          ],
          "compute_throughput_us_p50": 60.8828125,
          "compute_throughput_us_p90": 61.490625,
          "command_buffer_throughput_us": [
            59.97135394859375,
            61.0130200584375,
            61.44140570546875,
            60.727214076875,
            60.88346344765625,
            61.695964177499995,
            61.12109349484375,
            60.08658874640625,
            60.376953115312496
          ],
          "command_buffer_throughput_us_p50": 60.88346344765625,
          "command_buffer_throughput_us_p90": 61.492317399875,
          "compute_latency_us": [
            88.416,
            120.125,
            114.709,
            69.0,
            75.917,
            113.584,
            80.625,
            71.583,
            92.917
          ],
          "compute_latency_us_p50": 88.416,
          "compute_latency_us_p90": 115.79220000000001,
          "command_buffer_latency_us": [
            88.416680228,
            120.16668915700001,
            114.750000648,
            69.0833549015,
            76.0416733101,
            113.708258141,
            80.6666794233,
            71.5416972525,
            93.00000965599999
          ],
          "command_buffer_latency_us_p50": 88.416680228,
          "command_buffer_latency_us_p90": 115.8333383498,
          "counter_control_throughput_ratio": 1.0041710035981475,
          "counter_control_latency_ratio": 1.023457259616391
        },
        "correctness": {
          "max_abs_error": 1.4915225321487924e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 65.6670421687,
        "throughput_us_p90": 66.09621084334,
        "latency_us_p50": 309.042,
        "latency_us_p90": 598.0914000000002
      },
      "torch": {
        "cold_call_ms": 9.931084,
        "warmup_ms": 200.189583,
        "repetitions": 255,
        "throughput_us": [
          84.40588235294118,
          84.81421568627451,
          86.78104705882352,
          86.2581725490196,
          85.10865882352942,
          83.29166666666667,
          86.48055294117647,
          85.73006666666666,
          84.2063725490196
        ],
        "latency_us": [
          496.917,
          335.334,
          323.709,
          911.25,
          410.04200000000003,
          337.084,
          402.75,
          612.917,
          474.917
        ],
        "throughput_us_p50": 85.10865882352942,
        "throughput_us_p90": 86.54065176470588,
        "latency_us_p50": 410.04200000000003,
        "latency_us_p90": 672.5836,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 5012416.688259691,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4981875.012163073,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4970499.954652041,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5020458.367653191,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4970041.6857376695,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5102625.000290573,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5033958.35403353,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5030791.6244491935,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5075083.347037435,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 114625.0288002193,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 84583.3565108478,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 86374.98831376433,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 106874.96978789568,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 85124.97879564762,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 117374.99153241515,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 85874.98450651765,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 79624.995123595,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 84125.02938881516,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              78.31901075405767,
              77.84179706504801,
              77.66406179143814,
              78.44466199458111,
              77.65690133965109,
              79.7285156295402,
              78.6555992817739,
              78.60611913201865,
              79.29817729745992
            ],
            "command_buffer_throughput_us_p50": 78.44466199458111,
            "command_buffer_throughput_us_p90": 79.38424496387597,
            "command_buffer_latency_us": [
              114.6250288002193,
              84.5833565108478,
              86.37498831376433,
              106.87496978789568,
              85.12497879564762,
              117.37499153241515,
              85.87498450651765,
              79.624995123595,
              84.12502938881516
            ],
            "command_buffer_latency_us_p50": 85.87498450651765,
            "command_buffer_latency_us_p90": 115.17502134665847
          },
          "throughput": [
            {
              "compute_ns": 4922000.0,
              "compute_span_ns": 4922000.0,
              "command_buffer_ns": 4921999.992802739,
              "calibration_cpu_ns": 6332166.0,
              "calibration_gpu_ticks": 6332166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5048750.0,
              "compute_span_ns": 5048750.0,
              "command_buffer_ns": 5048749.968409538,
              "calibration_cpu_ns": 6150250.0,
              "calibration_gpu_ticks": 6150250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5078083.0,
              "compute_span_ns": 5078083.0,
              "command_buffer_ns": 5078041.693195701,
              "calibration_cpu_ns": 6240417.0,
              "calibration_gpu_ticks": 6240417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5012791.0,
              "compute_span_ns": 5012791.0,
              "command_buffer_ns": 5012749.9853260815,
              "calibration_cpu_ns": 6031125.0,
              "calibration_gpu_ticks": 6031125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5108208.0,
              "compute_span_ns": 5108208.0,
              "command_buffer_ns": 5108166.7188555,
              "calibration_cpu_ns": 6021834.0,
              "calibration_gpu_ticks": 6021834.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5155958.0,
              "compute_span_ns": 5155958.0,
              "command_buffer_ns": 5155958.351679146,
              "calibration_cpu_ns": 6234875.0,
              "calibration_gpu_ticks": 6234875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5072584.0,
              "compute_span_ns": 5072584.0,
              "command_buffer_ns": 5072625.004686415,
              "calibration_cpu_ns": 6010667.0,
              "calibration_gpu_ticks": 6010667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5002000.0,
              "compute_span_ns": 5002000.0,
              "command_buffer_ns": 5002000.0198856,
              "calibration_cpu_ns": 6047334.0,
              "calibration_gpu_ticks": 6047334.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5103666.0,
              "compute_span_ns": 5103666.0,
              "command_buffer_ns": 5103625.007905066,
              "calibration_cpu_ns": 6037250.0,
              "calibration_gpu_ticks": 6037250.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 113417.0,
              "compute_span_ns": 113417.0,
              "command_buffer_ns": 113458.31444486976,
              "calibration_cpu_ns": 880459.0,
              "calibration_gpu_ticks": 880459.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 93584.0,
              "compute_span_ns": 93584.0,
              "command_buffer_ns": 93624.98531118035,
              "calibration_cpu_ns": 522958.0,
              "calibration_gpu_ticks": 522958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 114250.0,
              "compute_span_ns": 114250.0,
              "command_buffer_ns": 114249.99684095383,
              "calibration_cpu_ns": 788417.0,
              "calibration_gpu_ticks": 788417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 87333.0,
              "compute_span_ns": 87333.0,
              "command_buffer_ns": 87333.37745070457,
              "calibration_cpu_ns": 616833.0,
              "calibration_gpu_ticks": 616833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 82083.0,
              "compute_span_ns": 82083.0,
              "command_buffer_ns": 82083.33747461438,
              "calibration_cpu_ns": 488417.0,
              "calibration_gpu_ticks": 488417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 79834.0,
              "compute_span_ns": 79834.0,
              "command_buffer_ns": 79874.99702721834,
              "calibration_cpu_ns": 615708.0,
              "calibration_gpu_ticks": 615708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 111291.0,
              "compute_span_ns": 111291.0,
              "command_buffer_ns": 111249.97399747372,
              "calibration_cpu_ns": 1285125.0,
              "calibration_gpu_ticks": 1285125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 99750.0,
              "compute_span_ns": 99750.0,
              "command_buffer_ns": 99750.00284612179,
              "calibration_cpu_ns": 569000.0,
              "calibration_gpu_ticks": 569000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 109542.0,
              "compute_span_ns": 109542.0,
              "command_buffer_ns": 109583.37225019932,
              "calibration_cpu_ns": 738459.0,
              "calibration_gpu_ticks": 738459.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            76.90625,
            78.88671875,
            79.345046875,
            78.324859375,
            79.81575,
            80.56184375,
            79.259125,
            78.15625,
            79.74478125
          ],
          "compute_throughput_us_p50": 79.259125,
          "compute_throughput_us_p90": 79.96496875,
          "command_buffer_throughput_us": [
            76.9062498875428,
            78.88671825639904,
            79.34440145618282,
            78.32421852072002,
            79.81510498211719,
            80.56184924498666,
            79.25976569822524,
            78.1562503107125,
            79.74414074851666
          ],
          "command_buffer_throughput_us_p50": 79.25976569822524,
          "command_buffer_throughput_us_p90": 79.96445383469108,
          "compute_latency_us": [
            113.417,
            93.584,
            114.25,
            87.333,
            82.083,
            79.834,
            111.291,
            99.75,
            109.542
          ],
          "compute_latency_us_p50": 99.75,
          "compute_latency_us_p90": 113.5836,
          "command_buffer_latency_us": [
            113.45831444486976,
            93.62498531118035,
            114.24999684095383,
            87.33337745070457,
            82.08333747461438,
            79.87499702721834,
            111.24997399747372,
            99.75000284612179,
            109.58337225019932
          ],
          "command_buffer_latency_us_p50": 99.75000284612179,
          "command_buffer_latency_us_p90": 113.61665092408657,
          "counter_control_throughput_ratio": 1.0076811622054647,
          "counter_control_latency_ratio": 1.1068960747516914
        },
        "download_ms": 0.788666,
        "allocation_upload_ms": 18.001209,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.9755493041451189e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.7715671128699007,
      "device_compute_slowdown": 0.7681489355326595,
      "valid": true,
      "round": 2,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "rmsnorm_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.424541,
        "warmup_ms": 201.688375,
        "repetitions": 346,
        "throughput_us": [
          83.2047225433526,
          83.58140751445087,
          83.32526589595376,
          82.82128901734103,
          86.45351445086705,
          82.66642485549133,
          83.6447514450867,
          82.55419075144509,
          83.84344797687861
        ],
        "latency_us": [
          437.20799999999997,
          403.375,
          324.33299999999997,
          363.458,
          321.79200000000003,
          303.542,
          307.125,
          318.083,
          300.25
        ],
        "throughput_us_p50": 83.32526589595376,
        "throughput_us_p90": 84.3654612716763,
        "latency_us_p50": 321.79200000000003,
        "latency_us_p90": 410.1416,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4982500.046025962,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4944708.3147242665,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5029041.669331491,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5024624.988436699,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5016541.632357985,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5086374.992970377,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5047833.372373134,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4986166.663002223,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5050041.654612869,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 104458.30412209034,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80750.0327937305,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 83749.9974295497,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 123375.03721937537,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80875.00464171171,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 82083.33747461438,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 81125.00654533505,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 85124.97879564762,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80375.00083446503,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              77.85156321915565,
              77.26106741756666,
              78.57877608330455,
              78.50976544432342,
              78.38346300559351,
              79.47460926516214,
              78.87239644333022,
              77.90885410940973,
              78.90690085332608
            ],
            "command_buffer_throughput_us_p50": 78.50976544432342,
            "command_buffer_throughput_us_p90": 79.02044253569329,
            "command_buffer_latency_us": [
              104.45830412209034,
              80.7500327937305,
              83.7499974295497,
              123.37503721937537,
              80.87500464171171,
              82.08333747461438,
              81.12500654533505,
              85.12497879564762,
              80.37500083446503
            ],
            "command_buffer_latency_us_p50": 82.08333747461438,
            "command_buffer_latency_us_p90": 108.24165074154735
          },
          "throughput": [
            {
              "compute_ns": 5009625.0,
              "compute_span_ns": 5009625.0,
              "command_buffer_ns": 5009624.99063462,
              "calibration_cpu_ns": 6121500.0,
              "calibration_gpu_ticks": 6121500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4986916.0,
              "compute_span_ns": 4986916.0,
              "command_buffer_ns": 4986875.05023554,
              "calibration_cpu_ns": 5863375.0,
              "calibration_gpu_ticks": 5863375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4988459.0,
              "compute_span_ns": 4988459.0,
              "command_buffer_ns": 4988499.9752976,
              "calibration_cpu_ns": 5983750.0,
              "calibration_gpu_ticks": 5983750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5023583.0,
              "compute_span_ns": 5023583.0,
              "command_buffer_ns": 5023583.362344652,
              "calibration_cpu_ns": 5989791.0,
              "calibration_gpu_ticks": 5989791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5063500.0,
              "compute_span_ns": 5063500.0,
              "command_buffer_ns": 5063499.964307994,
              "calibration_cpu_ns": 6213416.0,
              "calibration_gpu_ticks": 6213416.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5061667.0,
              "compute_span_ns": 5061667.0,
              "command_buffer_ns": 5061666.655819863,
              "calibration_cpu_ns": 6089584.0,
              "calibration_gpu_ticks": 6089584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4989666.0,
              "compute_span_ns": 4989666.0,
              "command_buffer_ns": 4989624.954760075,
              "calibration_cpu_ns": 5970500.0,
              "calibration_gpu_ticks": 5970500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4950791.0,
              "compute_span_ns": 4950791.0,
              "command_buffer_ns": 4950749.97888878,
              "calibration_cpu_ns": 5830750.0,
              "calibration_gpu_ticks": 5830750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4955291.0,
              "compute_span_ns": 4955291.0,
              "command_buffer_ns": 4955250.013154,
              "calibration_cpu_ns": 5949250.0,
              "calibration_gpu_ticks": 5949250.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 96750.0,
              "compute_span_ns": 96750.0,
              "command_buffer_ns": 96749.98000264168,
              "calibration_cpu_ns": 584000.0,
              "calibration_gpu_ticks": 584000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 84709.0,
              "compute_span_ns": 84709.0,
              "command_buffer_ns": 84750.00504404306,
              "calibration_cpu_ns": 621542.0,
              "calibration_gpu_ticks": 621542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 82500.0,
              "compute_span_ns": 82500.0,
              "command_buffer_ns": 82499.98791143298,
              "calibration_cpu_ns": 569375.0,
              "calibration_gpu_ticks": 569375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80417.0,
              "compute_span_ns": 80417.0,
              "command_buffer_ns": 80416.61931201816,
              "calibration_cpu_ns": 519042.0,
              "calibration_gpu_ticks": 519042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 90834.0,
              "compute_span_ns": 90834.0,
              "command_buffer_ns": 90875.0225789845,
              "calibration_cpu_ns": 926708.0,
              "calibration_gpu_ticks": 926708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80875.0,
              "compute_span_ns": 80875.0,
              "command_buffer_ns": 80874.9464340508,
              "calibration_cpu_ns": 486750.0,
              "calibration_gpu_ticks": 486750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 114291.0,
              "compute_span_ns": 114291.0,
              "command_buffer_ns": 114249.99684095383,
              "calibration_cpu_ns": 769458.0,
              "calibration_gpu_ticks": 769458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 86125.0,
              "compute_span_ns": 86125.0,
              "command_buffer_ns": 86124.98641014099,
              "calibration_cpu_ns": 847667.0,
              "calibration_gpu_ticks": 847667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 82209.0,
              "compute_span_ns": 82209.0,
              "command_buffer_ns": 82249.98600780964,
              "calibration_cpu_ns": 525417.0,
              "calibration_gpu_ticks": 525417.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            78.275390625,
            77.9205625,
            77.944671875,
            78.493484375,
            79.1171875,
            79.088546875,
            77.96353125,
            77.356109375,
            77.426421875
          ],
          "compute_throughput_us_p50": 77.96353125,
          "compute_throughput_us_p90": 79.09427500000001,
          "command_buffer_throughput_us": [
            78.27539047866594,
            77.91992265993031,
            77.945312114025,
            78.4934900366352,
            79.1171869423124,
            79.08854149718536,
            77.96288991812617,
            77.35546842013719,
            77.42578145553125
          ],
          "command_buffer_throughput_us_p50": 77.96288991812617,
          "command_buffer_throughput_us_p90": 79.09427058621077,
          "compute_latency_us": [
            96.75,
            84.709,
            82.5,
            80.417,
            90.834,
            80.875,
            114.291,
            86.125,
            82.209
          ],
          "compute_latency_us_p50": 84.709,
          "compute_latency_us_p90": 100.2582,
          "command_buffer_latency_us": [
            96.74998000264168,
            84.75000504404306,
            82.49998791143298,
            80.41661931201816,
            90.8750225789845,
            80.8749464340508,
            114.24999684095383,
            86.12498641014099,
            82.24998600780964
          ],
          "command_buffer_latency_us_p50": 84.75000504404306,
          "command_buffer_latency_us_p90": 100.24998337030411,
          "counter_control_throughput_ratio": 0.9951422501910178,
          "counter_control_latency_ratio": 1.0117475226266313
        },
        "download_ms": 0.576416,
        "allocation_upload_ms": 1.694042,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.9755493041451189e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-q3nq4zzf/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "c9fd11968a5edbf77e894438f42e4bd465ad146629716a6f08eacac9ab82063f",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 23.226458,
        "capture_ms": 0.059875,
        "compile_ms": 90.198584,
        "allocation_upload_ms": 5.911041,
        "cold_call_ms": 10.272375,
        "warmup_ms": 200.627292,
        "download_ms": 2.558125,
        "repetitions": 425,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          66.88392,
          66.7468635294,
          66.6147058824,
          68.0657835294,
          67.1752941176,
          67.1581364706,
          66.78804,
          66.7577435294,
          68.38
        ],
        "latency_us": [
          391.667,
          312.75,
          279.625,
          292.042,
          290.334,
          740.334,
          510.208,
          1212.333,
          332.75
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 384,
            "reduction_scalar_rounds": 72,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 49152,
              "global_write_bytes": 24576,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 192,
              "global_write_bytes": 96,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 1,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 96,
            "normalized_kernel_cost": 1152,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4199667,
              "compute_span_ns": 4199667,
              "calibration_cpu_ns": 4922584,
              "calibration_gpu_ticks": 4922584,
              "compute_passes": 1,
              "command_buffer_ns": 4199833.31813,
              "command_buffers": 2
            },
            {
              "compute_ns": 4221666,
              "compute_span_ns": 4221666,
              "calibration_cpu_ns": 4939625,
              "calibration_gpu_ticks": 4939625,
              "compute_passes": 1,
              "command_buffer_ns": 4221624.9858,
              "command_buffers": 2
            },
            {
              "compute_ns": 4217750,
              "compute_span_ns": 4217750,
              "calibration_cpu_ns": 5011416,
              "calibration_gpu_ticks": 5011416,
              "compute_passes": 1,
              "command_buffer_ns": 4217833.28056,
              "command_buffers": 2
            },
            {
              "compute_ns": 4206792,
              "compute_span_ns": 4206792,
              "calibration_cpu_ns": 4843666,
              "calibration_gpu_ticks": 4843666,
              "compute_passes": 1,
              "command_buffer_ns": 4206874.9899,
              "command_buffers": 2
            },
            {
              "compute_ns": 4199542,
              "compute_span_ns": 4199542,
              "calibration_cpu_ns": 4841625,
              "calibration_gpu_ticks": 4841625,
              "compute_passes": 1,
              "command_buffer_ns": 4199583.31622,
              "command_buffers": 2
            },
            {
              "compute_ns": 4222958,
              "compute_span_ns": 4222958,
              "calibration_cpu_ns": 5015458,
              "calibration_gpu_ticks": 5015458,
              "compute_passes": 1,
              "command_buffer_ns": 4223000.02538,
              "command_buffers": 2
            },
            {
              "compute_ns": 4175041,
              "compute_span_ns": 4175041,
              "calibration_cpu_ns": 4946708,
              "calibration_gpu_ticks": 4946708,
              "compute_passes": 1,
              "command_buffer_ns": 4175000.00913,
              "command_buffers": 2
            },
            {
              "compute_ns": 4242333,
              "compute_span_ns": 4242333,
              "calibration_cpu_ns": 5129917,
              "calibration_gpu_ticks": 5129917,
              "compute_passes": 1,
              "command_buffer_ns": 4242333.3507,
              "command_buffers": 2
            },
            {
              "compute_ns": 4284750,
              "compute_span_ns": 4284750,
              "calibration_cpu_ns": 4882584,
              "calibration_gpu_ticks": 4882584,
              "compute_passes": 1,
              "command_buffer_ns": 4284791.7066,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 68792,
              "compute_span_ns": 68792,
              "calibration_cpu_ns": 515250,
              "calibration_gpu_ticks": 515250,
              "compute_passes": 1,
              "command_buffer_ns": 68833.2947902,
              "command_buffers": 2
            },
            {
              "compute_ns": 81958,
              "compute_span_ns": 81958,
              "calibration_cpu_ns": 685625,
              "calibration_gpu_ticks": 685625,
              "compute_passes": 1,
              "command_buffer_ns": 81958.3656266,
              "command_buffers": 2
            },
            {
              "compute_ns": 65958,
              "compute_span_ns": 65958,
              "calibration_cpu_ns": 487792,
              "calibration_gpu_ticks": 487792,
              "compute_passes": 1,
              "command_buffer_ns": 66000.0368953,
              "command_buffers": 2
            },
            {
              "compute_ns": 78792,
              "compute_span_ns": 78792,
              "calibration_cpu_ns": 576584,
              "calibration_gpu_ticks": 576584,
              "compute_passes": 1,
              "command_buffer_ns": 78916.7243056,
              "command_buffers": 2
            },
            {
              "compute_ns": 66125,
              "compute_span_ns": 66125,
              "calibration_cpu_ns": 495000,
              "calibration_gpu_ticks": 495000,
              "compute_passes": 1,
              "command_buffer_ns": 66124.9505356,
              "command_buffers": 2
            },
            {
              "compute_ns": 74583,
              "compute_span_ns": 74583,
              "calibration_cpu_ns": 462542,
              "calibration_gpu_ticks": 462542,
              "compute_passes": 1,
              "command_buffer_ns": 74583.3385736,
              "command_buffers": 2
            },
            {
              "compute_ns": 108292,
              "compute_span_ns": 108292,
              "calibration_cpu_ns": 783166,
              "calibration_gpu_ticks": 783166,
              "compute_passes": 1,
              "command_buffer_ns": 108333.304524,
              "command_buffers": 2
            },
            {
              "compute_ns": 67667,
              "compute_span_ns": 67667,
              "calibration_cpu_ns": 497834,
              "calibration_gpu_ticks": 497834,
              "compute_passes": 1,
              "command_buffer_ns": 67791.6686982,
              "command_buffers": 2
            },
            {
              "compute_ns": 74583,
              "compute_span_ns": 74583,
              "calibration_cpu_ns": 492583,
              "calibration_gpu_ticks": 492583,
              "compute_passes": 1,
              "command_buffer_ns": 74625.0152588,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4218375.01926,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4254875.06436,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4202749.9876,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4225499.98621,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4213666.65978,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4316375.00878,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4203083.34287,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4249791.67292,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4223916.67962,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 112500.041723,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 74791.663792,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65374.9448247,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 112916.69216,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 87083.3173394,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 67124.9581501,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 77416.5964685,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 77000.062447,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 66625.0707582,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              65.9121096759375,
              66.48242288062501,
              65.66796855624999,
              66.02343728453124,
              65.8385415590625,
              67.4433595121875,
              65.67317723234375,
              66.402994889375,
              65.9986981190625
            ],
            "command_buffer_throughput_us_p50": 65.9986981190625,
            "command_buffer_throughput_us_p90": 66.6746102069375,
            "command_buffer_latency_us": [
              112.500041723,
              74.79166379200001,
              65.3749448247,
              112.91669216000001,
              87.0833173394,
              67.12495815010001,
              77.4165964685,
              77.000062447,
              66.6250707582
            ],
            "command_buffer_latency_us_p50": 77.000062447,
            "command_buffer_latency_us_p90": 112.5833718104
          },
          "compute_throughput_us": [
            65.619796875,
            65.96353125,
            65.90234375,
            65.731125,
            65.61784375,
            65.98371875,
            65.235015625,
            66.286453125,
            66.94921875
          ],
          "compute_throughput_us_p50": 65.90234375,
          "compute_throughput_us_p90": 66.41900625,
          "command_buffer_throughput_us": [
            65.62239559578124,
            65.962890403125,
            65.90364500875,
            65.7324217171875,
            65.61848931593751,
            65.9843753965625,
            65.23437514265625,
            66.28645860468751,
            66.949870415625
          ],
          "command_buffer_throughput_us_p50": 65.90364500875,
          "command_buffer_throughput_us_p90": 66.419140966875,
          "compute_latency_us": [
            68.792,
            81.958,
            65.958,
            78.792,
            66.125,
            74.583,
            108.292,
            67.667,
            74.583
          ],
          "compute_latency_us_p50": 74.583,
          "compute_latency_us_p90": 87.2248,
          "command_buffer_latency_us": [
            68.8332947902,
            81.9583656266,
            66.0000368953,
            78.9167243056,
            66.12495053560001,
            74.5833385736,
            108.33330452400001,
            67.7916686982,
            74.6250152588
          ],
          "command_buffer_latency_us_p50": 74.5833385736,
          "command_buffer_latency_us_p90": 87.23335340608,
          "counter_control_throughput_ratio": 0.9956045394149775,
          "counter_control_latency_ratio": 1.0095616458610583
        },
        "correctness": {
          "max_abs_error": 1.4915225321487924e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 66.88392,
        "throughput_us_p90": 68.12862682352,
        "latency_us_p50": 332.75,
        "latency_us_p90": 834.7338000000001
      },
      "slowdown": 0.8026847473072133,
      "device_compute_slowdown": 0.8452970599635325,
      "valid": true,
      "round": 2,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "rmsnorm_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-1lz36snh/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "310a51aaac34ebd04d17becc4645f77793138816428bb43603834b11ab1687fb",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.170167,
        "capture_ms": 0.058333,
        "compile_ms": 55.299958,
        "allocation_upload_ms": 5.730208,
        "cold_call_ms": 10.17775,
        "warmup_ms": 200.58575,
        "download_ms": 2.375792,
        "repetitions": 457,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          64.8975207877,
          66.1202603939,
          64.8800131291,
          64.9897877462,
          65.2636761488,
          64.7026805252,
          65.0599934354,
          65.2401531729,
          65.9793019694
        ],
        "latency_us": [
          347.667,
          280.042,
          279,
          277.708,
          275.042,
          288.125,
          291.333,
          268.333,
          1016.625
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 512,
            "reduction_scalar_rounds": 48,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 32768,
              "global_write_bytes": 16384,
              "private_read_bytes": 49152,
              "private_write_bytes": 16384
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 128,
              "global_write_bytes": 64,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 1,
            "reduction_elements": 4096,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 72,
            "normalized_kernel_cost": 1152,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 8192,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4108334,
              "compute_span_ns": 4108334,
              "calibration_cpu_ns": 5414042,
              "calibration_gpu_ticks": 5414042,
              "compute_passes": 1,
              "command_buffer_ns": 4108458.29174,
              "command_buffers": 2
            },
            {
              "compute_ns": 4110708,
              "compute_span_ns": 4110708,
              "calibration_cpu_ns": 5171083,
              "calibration_gpu_ticks": 5171083,
              "compute_passes": 1,
              "command_buffer_ns": 4110708.36708,
              "command_buffers": 2
            },
            {
              "compute_ns": 4087459,
              "compute_span_ns": 4087459,
              "calibration_cpu_ns": 4764000,
              "calibration_gpu_ticks": 4764000,
              "compute_passes": 1,
              "command_buffer_ns": 4087541.65983,
              "command_buffers": 2
            },
            {
              "compute_ns": 4092209,
              "compute_span_ns": 4092209,
              "calibration_cpu_ns": 4906834,
              "calibration_gpu_ticks": 4906834,
              "compute_passes": 1,
              "command_buffer_ns": 4092291.63779,
              "command_buffers": 2
            },
            {
              "compute_ns": 4068375,
              "compute_span_ns": 4068375,
              "calibration_cpu_ns": 4931666,
              "calibration_gpu_ticks": 4931666,
              "compute_passes": 1,
              "command_buffer_ns": 4068375.04124,
              "command_buffers": 2
            },
            {
              "compute_ns": 4073125,
              "compute_span_ns": 4073125,
              "calibration_cpu_ns": 4819083,
              "calibration_gpu_ticks": 4819083,
              "compute_passes": 1,
              "command_buffer_ns": 4073166.69589,
              "command_buffers": 2
            },
            {
              "compute_ns": 4105042,
              "compute_span_ns": 4105042,
              "calibration_cpu_ns": 6255542,
              "calibration_gpu_ticks": 6255542,
              "compute_passes": 1,
              "command_buffer_ns": 4105083.35335,
              "command_buffers": 2
            },
            {
              "compute_ns": 4107416,
              "compute_span_ns": 4107416,
              "calibration_cpu_ns": 4967458,
              "calibration_gpu_ticks": 4967458,
              "compute_passes": 1,
              "command_buffer_ns": 4107374.98896,
              "command_buffers": 2
            },
            {
              "compute_ns": 4076583,
              "compute_span_ns": 4076583,
              "calibration_cpu_ns": 4741792,
              "calibration_gpu_ticks": 4741792,
              "compute_passes": 1,
              "command_buffer_ns": 4076583.31096,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 90458,
              "compute_span_ns": 90458,
              "calibration_cpu_ns": 618250,
              "calibration_gpu_ticks": 618250,
              "compute_passes": 1,
              "command_buffer_ns": 90500.0488274,
              "command_buffers": 2
            },
            {
              "compute_ns": 84875,
              "compute_span_ns": 84875,
              "calibration_cpu_ns": 774250,
              "calibration_gpu_ticks": 774250,
              "compute_passes": 1,
              "command_buffer_ns": 84999.94874,
              "command_buffers": 2
            },
            {
              "compute_ns": 77292,
              "compute_span_ns": 77292,
              "calibration_cpu_ns": 494125,
              "calibration_gpu_ticks": 494125,
              "compute_passes": 1,
              "command_buffer_ns": 77375.0361986,
              "command_buffers": 2
            },
            {
              "compute_ns": 107042,
              "compute_span_ns": 107042,
              "calibration_cpu_ns": 769416,
              "calibration_gpu_ticks": 769416,
              "compute_passes": 1,
              "command_buffer_ns": 107166.648377,
              "command_buffers": 2
            },
            {
              "compute_ns": 64959,
              "compute_span_ns": 64959,
              "calibration_cpu_ns": 503167,
              "calibration_gpu_ticks": 503167,
              "compute_passes": 1,
              "command_buffer_ns": 65083.3244435,
              "command_buffers": 2
            },
            {
              "compute_ns": 64666,
              "compute_span_ns": 64666,
              "calibration_cpu_ns": 446250,
              "calibration_gpu_ticks": 446250,
              "compute_passes": 1,
              "command_buffer_ns": 64666.6740067,
              "command_buffers": 2
            },
            {
              "compute_ns": 98583,
              "compute_span_ns": 98583,
              "calibration_cpu_ns": 752750,
              "calibration_gpu_ticks": 752750,
              "compute_passes": 1,
              "command_buffer_ns": 98625.0233836,
              "command_buffers": 2
            },
            {
              "compute_ns": 66375,
              "compute_span_ns": 66375,
              "calibration_cpu_ns": 558333,
              "calibration_gpu_ticks": 558333,
              "compute_passes": 1,
              "command_buffer_ns": 66458.3640173,
              "command_buffers": 2
            },
            {
              "compute_ns": 64917,
              "compute_span_ns": 64917,
              "calibration_cpu_ns": 465667,
              "calibration_gpu_ticks": 465667,
              "compute_passes": 1,
              "command_buffer_ns": 64958.3525956,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4099458.33962,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4066791.67645,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4110083.39142,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4080666.63658,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4084458.34182,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4103166.63329,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4095333.39553,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4110749.92735,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4068583.36646,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 97458.367236,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 64958.2943879,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65083.3826512,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 73958.246503,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 87124.9940246,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 67833.3453834,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 77583.3614171,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65208.3544992,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65708.3583064,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              64.05403655656251,
              63.54361994453125,
              64.2200529909375,
              63.7604161965625,
              63.8196615909375,
              64.11197864515624,
              63.98958430515625,
              64.23046761484375,
              63.5716151009375
            ],
            "command_buffer_throughput_us_p50": 63.98958430515625,
            "command_buffer_throughput_us_p90": 64.22213591571875,
            "command_buffer_latency_us": [
              97.458367236,
              64.9582943879,
              65.0833826512,
              73.958246503,
              87.1249940246,
              67.8333453834,
              77.5833614171,
              65.2083544992,
              65.7083583064
            ],
            "command_buffer_latency_us_p50": 67.8333453834,
            "command_buffer_latency_us_p90": 89.19166866688
          },
          "compute_throughput_us": [
            64.19271875,
            64.2298125,
            63.866546875,
            63.940765625,
            63.568359375,
            63.642578125,
            64.14128125,
            64.178375,
            63.696609375
          ],
          "compute_throughput_us_p50": 63.940765625,
          "compute_throughput_us_p90": 64.2001375,
          "command_buffer_throughput_us": [
            64.19466080843749,
            64.229818235625,
            63.86783843484375,
            63.94205684046875,
            63.568360019375,
            63.64322962328125,
            64.14192739609375,
            64.1777342025,
            63.69661423375
          ],
          "command_buffer_throughput_us_p50": 63.94205684046875,
          "command_buffer_throughput_us_p90": 64.201692293875,
          "compute_latency_us": [
            90.458,
            84.875,
            77.292,
            107.042,
            64.959,
            64.666,
            98.583,
            66.375,
            64.917
          ],
          "compute_latency_us_p50": 77.292,
          "compute_latency_us_p90": 100.2748,
          "command_buffer_latency_us": [
            90.5000488274,
            84.99994874000001,
            77.3750361986,
            107.166648377,
            65.0833244435,
            64.66667400669999,
            98.6250233836,
            66.4583640173,
            64.9583525956
          ],
          "command_buffer_latency_us_p50": 77.3750361986,
          "command_buffer_latency_us_p90": 100.33334838228,
          "counter_control_throughput_ratio": 1.0019662727242973,
          "counter_control_latency_ratio": 1.0191694688157686
        },
        "correctness": {
          "max_abs_error": 1.5832164268481108e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 65.0599934354,
        "throughput_us_p90": 66.00749365429999,
        "latency_us_p50": 280.042,
        "latency_us_p90": 481.4586000000001
      },
      "torch": {
        "cold_call_ms": 7.482667,
        "warmup_ms": 200.356166,
        "repetitions": 377,
        "throughput_us": [
          72.30349071618038,
          72.82681432360742,
          74.63063660477454,
          72.16810344827586,
          72.64666047745358,
          73.01215649867375,
          72.41246684350132,
          72.26204774535809,
          72.85046419098143
        ],
        "latency_us": [
          414.417,
          364.20799999999997,
          439.125,
          396.25,
          322.75,
          310.58299999999997,
          331.083,
          294.917,
          327.41700000000003
        ],
        "throughput_us_p50": 72.64666047745358,
        "throughput_us_p90": 73.3358525198939,
        "latency_us_p50": 331.083,
        "latency_us_p90": 419.35859999999997,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4495500.004850328,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4420125.04208833,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4422666.679602116,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4422375.001013279,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4399208.293762058,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4410000.052303076,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4401958.372909576,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4422375.001013279,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4418625.03066659,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 104416.6274368763,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 71374.99051168561,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70458.33626762033,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 94958.34819972515,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 83499.99552592635,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70333.36441963911,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70041.68583080173,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 76249.99852851033,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 71874.9943189323,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              70.24218757578637,
              69.06445378263015,
              69.10416686878307,
              69.09960939083248,
              68.73762959003216,
              68.90625081723556,
              68.78059957671212,
              69.09960939083248,
              69.04101610416546
            ],
            "command_buffer_throughput_us_p50": 69.06445378263015,
            "command_buffer_throughput_us_p90": 69.33177101018373,
            "command_buffer_latency_us": [
              104.4166274368763,
              71.37499051168561,
              70.45833626762033,
              94.95834819972515,
              83.49999552592635,
              70.33336441963911,
              70.04168583080173,
              76.24999852851033,
              71.8749943189323
            ],
            "command_buffer_latency_us_p50": 71.8749943189323,
            "command_buffer_latency_us_p90": 96.85000404715538
          },
          "throughput": [
            {
              "compute_ns": 4308125.0,
              "compute_span_ns": 4308125.0,
              "command_buffer_ns": 4308125.004172325,
              "calibration_cpu_ns": 5326584.0,
              "calibration_gpu_ticks": 5326584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4411833.0,
              "compute_span_ns": 4411833.0,
              "command_buffer_ns": 4411791.684105992,
              "calibration_cpu_ns": 5267125.0,
              "calibration_gpu_ticks": 5267125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4426666.0,
              "compute_span_ns": 4426666.0,
              "command_buffer_ns": 4426625.033374876,
              "calibration_cpu_ns": 5276500.0,
              "calibration_gpu_ticks": 5276500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4461209.0,
              "compute_span_ns": 4461209.0,
              "command_buffer_ns": 4461249.976884574,
              "calibration_cpu_ns": 5334459.0,
              "calibration_gpu_ticks": 5334459.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4452500.0,
              "compute_span_ns": 4452500.0,
              "command_buffer_ns": 4452500.026673079,
              "calibration_cpu_ns": 5384375.0,
              "calibration_gpu_ticks": 5384375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4401583.0,
              "compute_span_ns": 4401583.0,
              "command_buffer_ns": 4401541.664265096,
              "calibration_cpu_ns": 5251625.0,
              "calibration_gpu_ticks": 5251625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4460417.0,
              "compute_span_ns": 4460417.0,
              "command_buffer_ns": 4460458.352696151,
              "calibration_cpu_ns": 5362125.0,
              "calibration_gpu_ticks": 5362125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4397209.0,
              "compute_span_ns": 4397209.0,
              "command_buffer_ns": 4397250.013425946,
              "calibration_cpu_ns": 5353167.0,
              "calibration_gpu_ticks": 5353167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4452083.0,
              "compute_span_ns": 4452083.0,
              "command_buffer_ns": 4452083.318028599,
              "calibration_cpu_ns": 5486667.0,
              "calibration_gpu_ticks": 5486667.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 71583.0,
              "compute_span_ns": 71583.0,
              "command_buffer_ns": 71541.69725254178,
              "calibration_cpu_ns": 658333.0,
              "calibration_gpu_ticks": 658333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 71792.0,
              "compute_span_ns": 71792.0,
              "command_buffer_ns": 71833.37584137917,
              "calibration_cpu_ns": 595750.0,
              "calibration_gpu_ticks": 595750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70625.0,
              "compute_span_ns": 70625.0,
              "command_buffer_ns": 70625.0430084765,
              "calibration_cpu_ns": 520916.0,
              "calibration_gpu_ticks": 520916.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 71083.0,
              "compute_span_ns": 71083.0,
              "command_buffer_ns": 71083.37013050914,
              "calibration_cpu_ns": 608875.0,
              "calibration_gpu_ticks": 608875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 72500.0,
              "compute_span_ns": 72500.0,
              "command_buffer_ns": 72499.9699741602,
              "calibration_cpu_ns": 606833.0,
              "calibration_gpu_ticks": 606833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 72709.0,
              "compute_span_ns": 72709.0,
              "command_buffer_ns": 72749.97187778354,
              "calibration_cpu_ns": 534583.0,
              "calibration_gpu_ticks": 534583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 105000.0,
              "compute_span_ns": 105000.0,
              "command_buffer_ns": 104999.98461455107,
              "calibration_cpu_ns": 793375.0,
              "calibration_gpu_ticks": 793375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 82542.0,
              "compute_span_ns": 82542.0,
              "command_buffer_ns": 82583.34128186107,
              "calibration_cpu_ns": 543625.0,
              "calibration_gpu_ticks": 543625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70125.0,
              "compute_span_ns": 70125.0,
              "command_buffer_ns": 70124.9809935689,
              "calibration_cpu_ns": 568708.0,
              "calibration_gpu_ticks": 568708.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            67.314453125,
            68.934890625,
            69.16665625,
            69.706390625,
            69.5703125,
            68.774734375,
            69.694015625,
            68.706390625,
            69.563796875
          ],
          "compute_throughput_us_p50": 69.16665625,
          "compute_throughput_us_p90": 69.69649062500001,
          "command_buffer_throughput_us": [
            67.31445319019258,
            68.93424506415613,
            69.16601614648243,
            69.70703088882146,
            69.57031291676685,
            68.77408850414213,
            69.69466176087735,
            68.70703145978041,
            69.56380184419686
          ],
          "command_buffer_throughput_us_p50": 69.16601614648243,
          "command_buffer_throughput_us_p90": 69.69713558646617,
          "compute_latency_us": [
            71.583,
            71.792,
            70.625,
            71.083,
            72.5,
            72.709,
            105.0,
            82.542,
            70.125
          ],
          "compute_latency_us_p50": 71.792,
          "compute_latency_us_p90": 87.0336,
          "command_buffer_latency_us": [
            71.54169725254178,
            71.83337584137917,
            70.6250430084765,
            71.08337013050914,
            72.4999699741602,
            72.74997187778354,
            104.99998461455107,
            82.58334128186107,
            70.1249809935689
          ],
          "command_buffer_latency_us_p50": 71.83337584137917,
          "command_buffer_latency_us_p90": 87.06666994839907,
          "counter_control_throughput_ratio": 1.000895015170602,
          "counter_control_latency_ratio": 1.0023660328881876
        },
        "download_ms": 0.634375,
        "allocation_upload_ms": 16.872291,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.8339726848282112e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.8955675733448456,
      "device_compute_slowdown": 0.9244449434405034,
      "valid": true,
      "round": 2,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "rmsnorm_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.382291,
        "warmup_ms": 200.756916,
        "repetitions": 393,
        "throughput_us": [
          72.55290585241731,
          72.91963613231552,
          72.26378117048347,
          73.72731043256996,
          72.94582188295166,
          72.33449872773538,
          72.87606106870228,
          72.54675572519083,
          74.5637175572519
        ],
        "latency_us": [
          412.209,
          305.25,
          297.292,
          309.79200000000003,
          287.292,
          369.542,
          290.416,
          287.0,
          298.125
        ],
        "throughput_us_p50": 72.87606106870228,
        "throughput_us_p90": 73.89459185750636,
        "latency_us_p50": 298.125,
        "latency_us_p90": 378.0754,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4524791.671428829,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4486749.996431172,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4594250.000081956,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4409499.990288168,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4488624.9816045165,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4481124.982703477,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4490333.376452327,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4473249.951843172,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4471874.970477074,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 97750.04582479596,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 69874.97908994555,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70041.62762314081,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 84458.32645520568,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 90166.6353456676,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 71749.96426329017,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70541.6314303875,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 71166.66529327631,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 86041.6330397129,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              70.69986986607546,
              70.10546869423706,
              71.78515625128057,
              68.89843734825263,
              70.13476533757057,
              70.01757785474183,
              70.1614590070676,
              69.89453049754957,
              69.87304641370429
            ],
            "command_buffer_throughput_us_p50": 70.10546869423706,
            "command_buffer_throughput_us_p90": 70.91692714311648,
            "command_buffer_latency_us": [
              97.75004582479596,
              69.87497908994555,
              70.04162762314081,
              84.45832645520568,
              90.1666353456676,
              71.74996426329017,
              70.5416314303875,
              71.16666529327631,
              86.0416330397129
            ],
            "command_buffer_latency_us_p50": 71.74996426329017,
            "command_buffer_latency_us_p90": 91.68331744149327
          },
          "throughput": [
            {
              "compute_ns": 4451458.0,
              "compute_span_ns": 4451458.0,
              "command_buffer_ns": 4451416.665688157,
              "calibration_cpu_ns": 5308083.0,
              "calibration_gpu_ticks": 5308083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4455375.0,
              "compute_span_ns": 4455375.0,
              "command_buffer_ns": 4455375.0194609165,
              "calibration_cpu_ns": 6089042.0,
              "calibration_gpu_ticks": 6089042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4479458.0,
              "compute_span_ns": 4479458.0,
              "command_buffer_ns": 4479416.646063328,
              "calibration_cpu_ns": 5339583.0,
              "calibration_gpu_ticks": 5339583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4468042.0,
              "compute_span_ns": 4468042.0,
              "command_buffer_ns": 4468041.704967618,
              "calibration_cpu_ns": 5324417.0,
              "calibration_gpu_ticks": 5324417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4488000.0,
              "compute_span_ns": 4488000.0,
              "command_buffer_ns": 4488000.005949289,
              "calibration_cpu_ns": 5326083.0,
              "calibration_gpu_ticks": 5326083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4496667.0,
              "compute_span_ns": 4496667.0,
              "command_buffer_ns": 4496666.660998017,
              "calibration_cpu_ns": 5343125.0,
              "calibration_gpu_ticks": 5343125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4425791.0,
              "compute_span_ns": 4425791.0,
              "command_buffer_ns": 4425749.997608364,
              "calibration_cpu_ns": 5570583.0,
              "calibration_gpu_ticks": 5570583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4440333.0,
              "compute_span_ns": 4440333.0,
              "command_buffer_ns": 4440291.66828841,
              "calibration_cpu_ns": 5290291.0,
              "calibration_gpu_ticks": 5290291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4498250.0,
              "compute_span_ns": 4498250.0,
              "command_buffer_ns": 4498249.967582524,
              "calibration_cpu_ns": 5505750.0,
              "calibration_gpu_ticks": 5505750.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 83917.0,
              "compute_span_ns": 83917.0,
              "command_buffer_ns": 83916.64596274495,
              "calibration_cpu_ns": 558667.0,
              "calibration_gpu_ticks": 558667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 92375.0,
              "compute_span_ns": 92375.0,
              "command_buffer_ns": 92375.03400072455,
              "calibration_cpu_ns": 902542.0,
              "calibration_gpu_ticks": 902542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 71125.0,
              "compute_span_ns": 71125.0,
              "command_buffer_ns": 71125.04681572318,
              "calibration_cpu_ns": 529584.0,
              "calibration_gpu_ticks": 529584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 104708.0,
              "compute_span_ns": 104708.0,
              "command_buffer_ns": 104708.3642333746,
              "calibration_cpu_ns": 725417.0,
              "calibration_gpu_ticks": 725417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70792.0,
              "compute_span_ns": 70792.0,
              "command_buffer_ns": 70791.63333401084,
              "calibration_cpu_ns": 483708.0,
              "calibration_gpu_ticks": 483708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70084.0,
              "compute_span_ns": 70084.0,
              "command_buffer_ns": 70125.03920122981,
              "calibration_cpu_ns": 487000.0,
              "calibration_gpu_ticks": 487000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 96750.0,
              "compute_span_ns": 96750.0,
              "command_buffer_ns": 96749.98000264168,
              "calibration_cpu_ns": 709500.0,
              "calibration_gpu_ticks": 709500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70791.0,
              "compute_span_ns": 70791.0,
              "command_buffer_ns": 70750.01485645771,
              "calibration_cpu_ns": 508875.0,
              "calibration_gpu_ticks": 508875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 71583.0,
              "compute_span_ns": 71583.0,
              "command_buffer_ns": 71541.63904488087,
              "calibration_cpu_ns": 501166.0,
              "calibration_gpu_ticks": 501166.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            69.55403125,
            69.615234375,
            69.99153125,
            69.81315625,
            70.125,
            70.260421875,
            69.152984375,
            69.380203125,
            70.28515625
          ],
          "compute_throughput_us_p50": 69.81315625,
          "compute_throughput_us_p90": 70.26536875000001,
          "command_buffer_throughput_us": [
            69.55338540137745,
            69.61523467907682,
            69.9908850947395,
            69.81315164011903,
            70.12500009295763,
            70.26041657809401,
            69.15234371263068,
            69.3795573170064,
            70.28515574347693
          ],
          "command_buffer_throughput_us_p50": 69.81315164011903,
          "command_buffer_throughput_us_p90": 70.2653644111706,
          "compute_latency_us": [
            83.917,
            92.375,
            71.125,
            104.708,
            70.792,
            70.084,
            96.75,
            70.791,
            71.583
          ],
          "compute_latency_us_p50": 71.583,
          "compute_latency_us_p90": 98.3416,
          "command_buffer_latency_us": [
            83.91664596274495,
            92.37503400072455,
            71.12504681572318,
            104.7083642333746,
            70.79163333401084,
            70.12503920122981,
            96.74998000264168,
            70.75001485645771,
            71.54163904488087
          ],
          "command_buffer_latency_us_p50": 71.54163904488087,
          "command_buffer_latency_us_p90": 98.34165684878826,
          "counter_control_throughput_ratio": 0.9930071929581075,
          "counter_control_latency_ratio": 0.9941454270043202
        },
        "download_ms": 0.630916,
        "allocation_upload_ms": 1.213708,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.8339726848282112e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-aho14kb1/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "26111b6b600e52c50494ad3f783ef6fc3b02252d329d47ea39eb66749773d18e",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 52.696167,
        "capture_ms": 0.070583,
        "compile_ms": 79.577791,
        "allocation_upload_ms": 6.842333,
        "cold_call_ms": 6.029417,
        "warmup_ms": 200.879791,
        "download_ms": 3.095959,
        "repetitions": 503,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          53.3310139165,
          53.4697654076,
          53.4648787276,
          53.4254473161,
          54.0708250497,
          54.4107017893,
          53.7355029821,
          53.7854532803,
          53.2118946322
        ],
        "latency_us": [
          361.208,
          322.459,
          692.667,
          289.166,
          318.958,
          387.875,
          320.167,
          398.541,
          335.334
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 32,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 1024,
            "reduction_scalar_rounds": 48,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 32768,
              "global_write_bytes": 16384,
              "private_read_bytes": 49152,
              "private_write_bytes": 16384
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 128,
              "global_write_bytes": 64,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 1,
            "reduction_elements": 4096,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 80,
            "normalized_kernel_cost": 1280,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 8192,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3368458,
              "compute_span_ns": 3368458,
              "calibration_cpu_ns": 3862084,
              "calibration_gpu_ticks": 3862084,
              "compute_passes": 1,
              "command_buffer_ns": 3368458.36136,
              "command_buffers": 2
            },
            {
              "compute_ns": 3476959,
              "compute_span_ns": 3476959,
              "calibration_cpu_ns": 4149000,
              "calibration_gpu_ticks": 4149000,
              "compute_passes": 1,
              "command_buffer_ns": 3477041.66779,
              "command_buffers": 2
            },
            {
              "compute_ns": 3323750,
              "compute_span_ns": 3323750,
              "calibration_cpu_ns": 3983042,
              "calibration_gpu_ticks": 3983042,
              "compute_passes": 1,
              "command_buffer_ns": 3323791.66503,
              "command_buffers": 2
            },
            {
              "compute_ns": 3443375,
              "compute_span_ns": 3443375,
              "calibration_cpu_ns": 3950541,
              "calibration_gpu_ticks": 3950541,
              "compute_passes": 1,
              "command_buffer_ns": 3443374.997,
              "command_buffers": 2
            },
            {
              "compute_ns": 3371459,
              "compute_span_ns": 3371459,
              "calibration_cpu_ns": 4536875,
              "calibration_gpu_ticks": 4536875,
              "compute_passes": 1,
              "command_buffer_ns": 3371500.00269,
              "command_buffers": 2
            },
            {
              "compute_ns": 3352500,
              "compute_span_ns": 3352500,
              "calibration_cpu_ns": 4175959,
              "calibration_gpu_ticks": 4175959,
              "compute_passes": 1,
              "command_buffer_ns": 3352541.5929,
              "command_buffers": 2
            },
            {
              "compute_ns": 3363333,
              "compute_span_ns": 3363333,
              "calibration_cpu_ns": 4146666,
              "calibration_gpu_ticks": 4146666,
              "compute_passes": 1,
              "command_buffer_ns": 3363374.96992,
              "command_buffers": 2
            },
            {
              "compute_ns": 3460292,
              "compute_span_ns": 3460292,
              "calibration_cpu_ns": 4145208,
              "calibration_gpu_ticks": 4145208,
              "compute_passes": 1,
              "command_buffer_ns": 3460416.68672,
              "command_buffers": 2
            },
            {
              "compute_ns": 3484250,
              "compute_span_ns": 3484250,
              "calibration_cpu_ns": 4424458,
              "calibration_gpu_ticks": 4424458,
              "compute_passes": 1,
              "command_buffer_ns": 3484333.34148,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 83666,
              "compute_span_ns": 83666,
              "calibration_cpu_ns": 726667,
              "calibration_gpu_ticks": 726667,
              "compute_passes": 1,
              "command_buffer_ns": 83666.7022668,
              "command_buffers": 2
            },
            {
              "compute_ns": 54250,
              "compute_span_ns": 54250,
              "calibration_cpu_ns": 460375,
              "calibration_gpu_ticks": 460375,
              "compute_passes": 1,
              "command_buffer_ns": 54333.3007954,
              "command_buffers": 2
            },
            {
              "compute_ns": 54291,
              "compute_span_ns": 54291,
              "calibration_cpu_ns": 497333,
              "calibration_gpu_ticks": 497333,
              "compute_passes": 1,
              "command_buffer_ns": 54250.0056326,
              "command_buffers": 2
            },
            {
              "compute_ns": 91250,
              "compute_span_ns": 91250,
              "calibration_cpu_ns": 722833,
              "calibration_gpu_ticks": 722833,
              "compute_passes": 1,
              "command_buffer_ns": 91291.6730158,
              "command_buffers": 2
            },
            {
              "compute_ns": 66000,
              "compute_span_ns": 66000,
              "calibration_cpu_ns": 530833,
              "calibration_gpu_ticks": 530833,
              "compute_passes": 1,
              "command_buffer_ns": 66083.2738504,
              "command_buffers": 2
            },
            {
              "compute_ns": 54458,
              "compute_span_ns": 54458,
              "calibration_cpu_ns": 456958,
              "calibration_gpu_ticks": 456958,
              "compute_passes": 1,
              "command_buffer_ns": 54458.330851,
              "command_buffers": 2
            },
            {
              "compute_ns": 88500,
              "compute_span_ns": 88500,
              "calibration_cpu_ns": 666208,
              "calibration_gpu_ticks": 666208,
              "compute_passes": 1,
              "command_buffer_ns": 88583.3287612,
              "command_buffers": 2
            },
            {
              "compute_ns": 65375,
              "compute_span_ns": 65375,
              "calibration_cpu_ns": 858458,
              "calibration_gpu_ticks": 858458,
              "compute_passes": 1,
              "command_buffer_ns": 65375.0030324,
              "command_buffers": 2
            },
            {
              "compute_ns": 54500,
              "compute_span_ns": 54500,
              "calibration_cpu_ns": 478416,
              "calibration_gpu_ticks": 478416,
              "compute_passes": 1,
              "command_buffer_ns": 54624.9793842,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 3498833.33547,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3444250.03277,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3437874.95513,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3346708.31216,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3407916.578,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3412500.02384,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3323375.01459,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3421916.6846,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3342916.72334,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 98083.4010988,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54625.0375919,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 55708.2821615,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 53958.3852515,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 67541.6667946,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54666.6560695,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54874.9812879,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 79833.320342,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 53500.0581294,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              54.66927086671875,
              53.81640676203125,
              53.71679617390625,
              52.292317377500005,
              53.248696531250005,
              53.3203128725,
              51.92773460296875,
              53.467448196875,
              52.2330738021875
            ],
            "command_buffer_throughput_us_p50": 53.3203128725,
            "command_buffer_throughput_us_p90": 53.98697958296875,
            "command_buffer_latency_us": [
              98.0834010988,
              54.625037591899996,
              55.7082821615,
              53.958385251500005,
              67.54166679459999,
              54.6666560695,
              54.8749812879,
              79.83332034200001,
              53.5000581294
            ],
            "command_buffer_latency_us_p50": 54.8749812879,
            "command_buffer_latency_us_p90": 83.48333649336001
          },
          "compute_throughput_us": [
            52.63215625,
            54.327484375,
            51.93359375,
            53.802734375,
            52.679046875,
            52.3828125,
            52.552078125,
            54.0670625,
            54.44140625
          ],
          "compute_throughput_us_p50": 52.679046875,
          "compute_throughput_us_p90": 54.35026875,
          "command_buffer_throughput_us": [
            52.63216189625,
            54.32877605921875,
            51.93424476609375,
            53.802734328125,
            52.67968754203125,
            52.3834623890625,
            52.552733905000004,
            54.06901073,
            54.442708460625
          ],
          "command_buffer_throughput_us_p50": 52.67968754203125,
          "command_buffer_throughput_us_p90": 54.351562539499994,
          "compute_latency_us": [
            83.666,
            54.25,
            54.291,
            91.25,
            66.0,
            54.458,
            88.5,
            65.375,
            54.5
          ],
          "compute_latency_us_p50": 65.375,
          "compute_latency_us_p90": 89.05,
          "command_buffer_latency_us": [
            83.6667022668,
            54.3333007954,
            54.2500056326,
            91.2916730158,
            66.0832738504,
            54.458330851,
            88.5833287612,
            65.3750030324,
            54.6249793842
          ],
          "command_buffer_latency_us_p50": 65.3750030324,
          "command_buffer_latency_us_p90": 89.12499761212,
          "counter_control_throughput_ratio": 1.0095206894702786,
          "counter_control_latency_ratio": 0.9946592842886526
        },
        "correctness": {
          "max_abs_error": 1.5832164268481108e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 53.4697654076,
        "throughput_us_p90": 54.138800397619995,
        "latency_us_p50": 335.334,
        "latency_us_p90": 457.36620000000005
      },
      "slowdown": 0.7337082249436144,
      "device_compute_slowdown": 0.7545719131556954,
      "valid": true,
      "round": 2,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "layernorm_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-2b9qrg_v/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "90a0b6dbe0fbfeffc09b3a8db9fc5856f9e2631dc2ce0e6d9ee991918165a6cf",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.603958,
        "capture_ms": 0.078875,
        "compile_ms": 63.395792,
        "allocation_upload_ms": 1.286833,
        "cold_call_ms": 2.052083,
        "warmup_ms": 200.057,
        "download_ms": 0.288042,
        "repetitions": 5733,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          4.59467852782,
          4.49701290773,
          4.55004639805,
          4.5567402756,
          4.48808791209,
          4.80329530787,
          4.51412140241,
          4.46876992848,
          4.51284964242
        ],
        "latency_us": [
          265.541,
          206.083,
          217.25,
          238.125,
          248.25,
          221.541,
          214.208,
          972.667,
          222.584
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 37,
            "reduction_scalar_rounds": 40,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 18444,
              "global_write_bytes": 6148,
              "private_read_bytes": 36888,
              "private_write_bytes": 12296
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 32,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 2,
            "reduction_elements": 3074,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 88,
            "normalized_kernel_cost": 88,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 4611,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 281709,
              "compute_span_ns": 281709,
              "calibration_cpu_ns": 734084,
              "calibration_gpu_ticks": 734084,
              "compute_passes": 1,
              "command_buffer_ns": 281749.9917,
              "command_buffers": 2
            },
            {
              "compute_ns": 304250,
              "compute_span_ns": 304250,
              "calibration_cpu_ns": 868083,
              "calibration_gpu_ticks": 868083,
              "compute_passes": 1,
              "command_buffer_ns": 304374.960251,
              "command_buffers": 2
            },
            {
              "compute_ns": 284791,
              "compute_span_ns": 284791,
              "calibration_cpu_ns": 850708,
              "calibration_gpu_ticks": 850708,
              "compute_passes": 1,
              "command_buffer_ns": 284833.309706,
              "command_buffers": 2
            },
            {
              "compute_ns": 262292,
              "compute_span_ns": 262292,
              "calibration_cpu_ns": 757000,
              "calibration_gpu_ticks": 757000,
              "compute_passes": 1,
              "command_buffer_ns": 262416.724581,
              "command_buffers": 2
            },
            {
              "compute_ns": 263625,
              "compute_span_ns": 263625,
              "calibration_cpu_ns": 732958,
              "calibration_gpu_ticks": 732958,
              "compute_passes": 1,
              "command_buffer_ns": 263708.294369,
              "command_buffers": 2
            },
            {
              "compute_ns": 282833,
              "compute_span_ns": 282833,
              "calibration_cpu_ns": 680666,
              "calibration_gpu_ticks": 680666,
              "compute_passes": 1,
              "command_buffer_ns": 282791.676,
              "command_buffers": 2
            },
            {
              "compute_ns": 310042,
              "compute_span_ns": 310042,
              "calibration_cpu_ns": 846416,
              "calibration_gpu_ticks": 846416,
              "compute_passes": 1,
              "command_buffer_ns": 310125.004034,
              "command_buffers": 2
            },
            {
              "compute_ns": 264334,
              "compute_span_ns": 264334,
              "calibration_cpu_ns": 717584,
              "calibration_gpu_ticks": 717584,
              "compute_passes": 1,
              "command_buffer_ns": 264416.681603,
              "command_buffers": 2
            },
            {
              "compute_ns": 282541,
              "compute_span_ns": 282541,
              "calibration_cpu_ns": 964875,
              "calibration_gpu_ticks": 964875,
              "compute_passes": 1,
              "command_buffer_ns": 282583.292574,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 7042,
              "compute_span_ns": 7042,
              "calibration_cpu_ns": 468000,
              "calibration_gpu_ticks": 468000,
              "compute_passes": 1,
              "command_buffer_ns": 7208.3203122,
              "command_buffers": 2
            },
            {
              "compute_ns": 7542,
              "compute_span_ns": 7542,
              "calibration_cpu_ns": 394917,
              "calibration_gpu_ticks": 394917,
              "compute_passes": 1,
              "command_buffer_ns": 7541.67558625,
              "command_buffers": 2
            },
            {
              "compute_ns": 7250,
              "compute_span_ns": 7250,
              "calibration_cpu_ns": 416667,
              "calibration_gpu_ticks": 416667,
              "compute_passes": 1,
              "command_buffer_ns": 7291.61547497,
              "command_buffers": 2
            },
            {
              "compute_ns": 10375,
              "compute_span_ns": 10375,
              "calibration_cpu_ns": 787208,
              "calibration_gpu_ticks": 787208,
              "compute_passes": 1,
              "command_buffer_ns": 10499.9635369,
              "command_buffers": 2
            },
            {
              "compute_ns": 7250,
              "compute_span_ns": 7250,
              "calibration_cpu_ns": 403875,
              "calibration_gpu_ticks": 403875,
              "compute_passes": 1,
              "command_buffer_ns": 7333.29216018,
              "command_buffers": 2
            },
            {
              "compute_ns": 10500,
              "compute_span_ns": 10500,
              "calibration_cpu_ns": 357708,
              "calibration_gpu_ticks": 357708,
              "compute_passes": 1,
              "command_buffer_ns": 10625.0518002,
              "command_buffers": 2
            },
            {
              "compute_ns": 7125,
              "compute_span_ns": 7125,
              "calibration_cpu_ns": 440500,
              "calibration_gpu_ticks": 440500,
              "compute_passes": 1,
              "command_buffer_ns": 7208.37851986,
              "command_buffers": 2
            },
            {
              "compute_ns": 8416,
              "compute_span_ns": 8416,
              "calibration_cpu_ns": 833125,
              "calibration_gpu_ticks": 833125,
              "compute_passes": 1,
              "command_buffer_ns": 8458.27162266,
              "command_buffers": 2
            },
            {
              "compute_ns": 7333,
              "compute_span_ns": 7333,
              "calibration_cpu_ns": 384584,
              "calibration_gpu_ticks": 384584,
              "compute_passes": 1,
              "command_buffer_ns": 7375.02705306,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 309625.000227,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 285583.373625,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 262541.638222,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 285166.66498,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 298791.681416,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 264916.627202,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 284708.337858,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 280125.00843,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 262833.31681,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 7666.64743423,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7083.29025656,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7375.02705306,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7041.67177901,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7541.67558625,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 11666.6778922,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7208.26210454,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7166.70183465,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7125.02514943,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              4.837890628546875,
              4.462240212890626,
              4.102213097218749,
              4.4557291403125,
              4.668620022125,
              4.1393223000312505,
              4.44856777903125,
              4.37695325671875,
              4.10677057515625
            ],
            "command_buffer_throughput_us_p50": 4.44856777903125,
            "command_buffer_throughput_us_p90": 4.7024741434093755,
            "command_buffer_latency_us": [
              7.66664743423,
              7.08329025656,
              7.37502705306,
              7.0416717790100005,
              7.54167558625,
              11.6666778922,
              7.20826210454,
              7.1667018346499995,
              7.12502514943
            ],
            "command_buffer_latency_us_p50": 7.20826210454,
            "command_buffer_latency_us_p90": 8.466653525824
          },
          "compute_throughput_us": [
            4.401703125,
            4.75390625,
            4.449859375,
            4.0983125,
            4.119140625,
            4.419265625,
            4.84440625,
            4.13021875,
            4.414703125
          ],
          "compute_throughput_us_p50": 4.414703125,
          "compute_throughput_us_p90": 4.77200625,
          "command_buffer_throughput_us": [
            4.4023436203125,
            4.755858753921875,
            4.4505204641562495,
            4.100261321578125,
            4.120442099515625,
            4.4186199375,
            4.8457031880312496,
            4.131510650046875,
            4.415363946468751
          ],
          "command_buffer_throughput_us_p50": 4.415363946468751,
          "command_buffer_throughput_us_p90": 4.77382764074375,
          "compute_latency_us": [
            7.042,
            7.542,
            7.25,
            10.375,
            7.25,
            10.5,
            7.125,
            8.416,
            7.333
          ],
          "compute_latency_us_p50": 7.333,
          "compute_latency_us_p90": 10.4,
          "command_buffer_latency_us": [
            7.208320312200001,
            7.54167558625,
            7.2916154749699995,
            10.499963536900001,
            7.33329216018,
            10.625051800200001,
            7.20837851986,
            8.45827162266,
            7.37502705306
          ],
          "command_buffer_latency_us_p50": 7.37502705306,
          "command_buffer_latency_us_p90": 10.524981189560002,
          "counter_control_throughput_ratio": 1.0658007025670733,
          "counter_control_latency_ratio": 1.0000161502617846
        },
        "correctness": {
          "max_abs_error": 2.43199801630567e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 4.51412140241,
        "throughput_us_p90": 4.6364018838300005,
        "latency_us_p50": 222.584,
        "latency_us_p90": 406.96620000000013
      },
      "torch": {
        "cold_call_ms": 9.819,
        "warmup_ms": 200.125458,
        "repetitions": 2385,
        "throughput_us": [
          12.548934171907757,
          12.025576519916143,
          12.49112536687631,
          12.013259958071279,
          12.010656603773585,
          11.802795387840671,
          11.88539496855346,
          11.686705241090147,
          11.790129140461216
        ],
        "latency_us": [
          401.375,
          273.625,
          216.084,
          209.625,
          565.75,
          494.833,
          234.042,
          230.584,
          394.375
        ],
        "throughput_us_p50": 12.010656603773585,
        "throughput_us_p90": 12.502687127882599,
        "latency_us_p50": 273.625,
        "latency_us_p90": 509.01640000000003,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 486250.03546476364,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 486791.65774956346,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 497208.32612365484,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 499000.0161342323,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 485000.0259466469,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 518625.0200495124,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 506041.687913239,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 487499.98677521944,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 566000.0024363399,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 11666.6778922081,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10499.963536858559,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10750.023648142815,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10500.021744519472,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10458.34505930543,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10375.049896538258,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10874.99549612403,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10249.961633235216,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10499.963536858559,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              7.597656804136932,
              7.606119652336929,
              7.768880095682107,
              7.796875252097379,
              7.578125405416358,
              8.103515938273631,
              7.906901373644359,
              7.617187293362804,
              8.84375003806781
            ],
            "command_buffer_throughput_us_p50": 7.768880095682107,
            "command_buffer_throughput_us_p90": 8.251562758232467,
            "command_buffer_latency_us": [
              11.6666778922081,
              10.499963536858559,
              10.750023648142815,
              10.500021744519472,
              10.45834505930543,
              10.375049896538258,
              10.87499549612403,
              10.249961633235216,
              10.499963536858559
            ],
            "command_buffer_latency_us_p50": 10.499963536858559,
            "command_buffer_latency_us_p90": 11.033331975340843
          },
          "throughput": [
            {
              "compute_ns": 483542.0,
              "compute_span_ns": 483542.0,
              "command_buffer_ns": 483583.30968767405,
              "calibration_cpu_ns": 1200250.0,
              "calibration_gpu_ticks": 1200250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 485125.0,
              "compute_span_ns": 485125.0,
              "command_buffer_ns": 485124.99779462814,
              "calibration_cpu_ns": 1241750.0,
              "calibration_gpu_ticks": 1241750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 484125.0,
              "compute_span_ns": 484125.0,
              "command_buffer_ns": 484124.9901801348,
              "calibration_cpu_ns": 1173584.0,
              "calibration_gpu_ticks": 1173584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 485166.0,
              "compute_span_ns": 485166.0,
              "command_buffer_ns": 485124.99779462814,
              "calibration_cpu_ns": 1219750.0,
              "calibration_gpu_ticks": 1219750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 485625.0,
              "compute_span_ns": 485625.0,
              "command_buffer_ns": 485625.0016018748,
              "calibration_cpu_ns": 1260000.0,
              "calibration_gpu_ticks": 1260000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 484792.0,
              "compute_span_ns": 484792.0,
              "command_buffer_ns": 484833.31920579076,
              "calibration_cpu_ns": 1200667.0,
              "calibration_gpu_ticks": 1200667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 485417.0,
              "compute_span_ns": 485417.0,
              "command_buffer_ns": 485416.6181758046,
              "calibration_cpu_ns": 1166875.0,
              "calibration_gpu_ticks": 1166875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 484542.0,
              "compute_span_ns": 484542.0,
              "command_buffer_ns": 484541.6988246143,
              "calibration_cpu_ns": 1278292.0,
              "calibration_gpu_ticks": 1278292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 543250.0,
              "compute_span_ns": 543250.0,
              "command_buffer_ns": 543250.0038295984,
              "calibration_cpu_ns": 1282792.0,
              "calibration_gpu_ticks": 1282792.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 10625.0,
              "compute_span_ns": 10625.0,
              "command_buffer_ns": 10624.993592500687,
              "calibration_cpu_ns": 431041.0,
              "calibration_gpu_ticks": 431041.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10458.0,
              "compute_span_ns": 10458.0,
              "command_buffer_ns": 10416.668374091387,
              "calibration_cpu_ns": 475916.0,
              "calibration_gpu_ticks": 475916.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10542.0,
              "compute_span_ns": 10542.0,
              "command_buffer_ns": 10541.640222072601,
              "calibration_cpu_ns": 390209.0,
              "calibration_gpu_ticks": 390209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10583.0,
              "compute_span_ns": 10583.0,
              "command_buffer_ns": 10583.316907286644,
              "calibration_cpu_ns": 389000.0,
              "calibration_gpu_ticks": 389000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12000.0,
              "compute_span_ns": 12000.0,
              "command_buffer_ns": 11999.974958598614,
              "calibration_cpu_ns": 677000.0,
              "calibration_gpu_ticks": 677000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10417.0,
              "compute_span_ns": 10417.0,
              "command_buffer_ns": 10416.668374091387,
              "calibration_cpu_ns": 400750.0,
              "calibration_gpu_ticks": 400750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10500.0,
              "compute_span_ns": 10500.0,
              "command_buffer_ns": 10499.963536858559,
              "calibration_cpu_ns": 406167.0,
              "calibration_gpu_ticks": 406167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10333.0,
              "compute_span_ns": 10333.0,
              "command_buffer_ns": 10291.638318449259,
              "calibration_cpu_ns": 381958.0,
              "calibration_gpu_ticks": 381958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10375.0,
              "compute_span_ns": 10375.0,
              "command_buffer_ns": 10374.991688877344,
              "calibration_cpu_ns": 592708.0,
              "calibration_gpu_ticks": 592708.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            7.55534375,
            7.580078125,
            7.564453125,
            7.58071875,
            7.587890625,
            7.574875,
            7.584640625,
            7.57096875,
            8.48828125
          ],
          "compute_throughput_us_p50": 7.580078125,
          "compute_throughput_us_p90": 7.7679687500000005,
          "command_buffer_throughput_us": [
            7.555989213869907,
            7.580078090541065,
            7.564452971564606,
            7.580078090541065,
            7.587890650029294,
            7.575520612590481,
            7.584634658996947,
            7.570964044134598,
            8.488281309837475
          ],
          "command_buffer_throughput_us_p50": 7.580078090541065,
          "command_buffer_throughput_us_p90": 7.76796878199093,
          "compute_latency_us": [
            10.625,
            10.458,
            10.542,
            10.583,
            12.0,
            10.417,
            10.5,
            10.333,
            10.375
          ],
          "compute_latency_us_p50": 10.5,
          "compute_latency_us_p90": 10.9,
          "command_buffer_latency_us": [
            10.624993592500687,
            10.416668374091387,
            10.541640222072601,
            10.583316907286644,
            11.999974958598614,
            10.416668374091387,
            10.499963536858559,
            10.291638318449259,
            10.374991688877344
          ],
          "command_buffer_latency_us_p50": 10.499963536858559,
          "command_buffer_latency_us_p90": 10.899989865720272,
          "counter_control_throughput_ratio": 0.9736864101904313,
          "counter_control_latency_ratio": 0.9920670998070825
        },
        "download_ms": 0.330458,
        "allocation_upload_ms": 0.847834,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.0790672678193687e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.3758430160255955,
      "device_compute_slowdown": 0.582408657562484,
      "valid": true,
      "round": 2,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "layernorm_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.260833,
        "warmup_ms": 200.257083,
        "repetitions": 2428,
        "throughput_us": [
          12.892590197693574,
          12.175521828665568,
          12.523149917627677,
          12.280288714991762,
          12.517538714991762,
          12.326588962108733,
          12.351626853377265,
          12.73616845140033,
          12.434805601317956
        ],
        "latency_us": [
          327.083,
          223.125,
          202.75,
          230.708,
          336.83299999999997,
          451.459,
          312.833,
          221.584,
          211.375
        ],
        "throughput_us_p50": 12.434805601317956,
        "throughput_us_p90": 12.767452800658978,
        "latency_us_p50": 230.708,
        "latency_us_p90": 359.7582,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 563875.0153593719,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 540416.6877269745,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 557625.0259764493,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 571999.9899156392,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 571499.9861083925,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 545500.0209622085,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 502000.0389777124,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 518499.98999387026,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 518625.0200495124,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 10750.023648142815,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 15708.326827734709,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10250.01984089613,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10583.316907286644,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10374.991688877344,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12875.01072511077,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 13291.661161929369,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10250.01984089613,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10416.668374091387,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              8.810547114990186,
              8.444010745733976,
              8.71289103088202,
              8.937499842431862,
              8.929687282943632,
              8.523437827534508,
              7.843750609026756,
              8.101562343654223,
              8.103515938273631
            ],
            "command_buffer_throughput_us_p50": 8.523437827534508,
            "command_buffer_throughput_us_p90": 8.931249794841278,
            "command_buffer_latency_us": [
              10.750023648142815,
              15.708326827734709,
              10.25001984089613,
              10.583316907286644,
              10.374991688877344,
              12.87501072511077,
              13.291661161929369,
              10.25001984089613,
              10.416668374091387
            ],
            "command_buffer_latency_us_p50": 10.583316907286644,
            "command_buffer_latency_us_p90": 13.774994295090437
          },
          "throughput": [
            {
              "compute_ns": 485250.0,
              "compute_span_ns": 485250.0,
              "command_buffer_ns": 485250.0278502703,
              "calibration_cpu_ns": 1190583.0,
              "calibration_gpu_ticks": 1190583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 555708.0,
              "compute_span_ns": 555708.0,
              "command_buffer_ns": 555708.3059102297,
              "calibration_cpu_ns": 1431458.0,
              "calibration_gpu_ticks": 1431458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 484625.0,
              "compute_span_ns": 484625.0,
              "command_buffer_ns": 484624.99398738146,
              "calibration_cpu_ns": 1177375.0,
              "calibration_gpu_ticks": 1177375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 497417.0,
              "compute_span_ns": 497417.0,
              "command_buffer_ns": 497416.65134206414,
              "calibration_cpu_ns": 1392833.0,
              "calibration_gpu_ticks": 1392833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 485083.0,
              "compute_span_ns": 485083.0,
              "command_buffer_ns": 485083.3211094141,
              "calibration_cpu_ns": 1164875.0,
              "calibration_gpu_ticks": 1164875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 487084.0,
              "compute_span_ns": 487084.0,
              "command_buffer_ns": 487125.0130236149,
              "calibration_cpu_ns": 1815459.0,
              "calibration_gpu_ticks": 1815459.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 491167.0,
              "compute_span_ns": 491167.0,
              "command_buffer_ns": 491166.6619591415,
              "calibration_cpu_ns": 1522542.0,
              "calibration_gpu_ticks": 1522542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 564500.0,
              "compute_span_ns": 564500.0,
              "command_buffer_ns": 564499.9910145998,
              "calibration_cpu_ns": 1331333.0,
              "calibration_gpu_ticks": 1331333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 563292.0,
              "compute_span_ns": 563292.0,
              "command_buffer_ns": 563333.3348669112,
              "calibration_cpu_ns": 1509708.0,
              "calibration_gpu_ticks": 1509708.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 10375.0,
              "compute_span_ns": 10375.0,
              "command_buffer_ns": 10374.991688877344,
              "calibration_cpu_ns": 403375.0,
              "calibration_gpu_ticks": 403375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12375.0,
              "compute_span_ns": 12375.0,
              "command_buffer_ns": 12374.94871020317,
              "calibration_cpu_ns": 475625.0,
              "calibration_gpu_ticks": 475625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10250.0,
              "compute_span_ns": 10250.0,
              "command_buffer_ns": 10250.01984089613,
              "calibration_cpu_ns": 802708.0,
              "calibration_gpu_ticks": 802708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 11167.0,
              "compute_span_ns": 11167.0,
              "command_buffer_ns": 11208.350770175457,
              "calibration_cpu_ns": 520041.0,
              "calibration_gpu_ticks": 520041.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10583.0,
              "compute_span_ns": 10583.0,
              "command_buffer_ns": 10583.316907286644,
              "calibration_cpu_ns": 416042.0,
              "calibration_gpu_ticks": 416042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10541.0,
              "compute_span_ns": 10541.0,
              "command_buffer_ns": 10500.021744519472,
              "calibration_cpu_ns": 409333.0,
              "calibration_gpu_ticks": 409333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10291.0,
              "compute_span_ns": 10291.0,
              "command_buffer_ns": 10250.01984089613,
              "calibration_cpu_ns": 925375.0,
              "calibration_gpu_ticks": 925375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10750.0,
              "compute_span_ns": 10750.0,
              "command_buffer_ns": 10749.965440481901,
              "calibration_cpu_ns": 438708.0,
              "calibration_gpu_ticks": 438708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10417.0,
              "compute_span_ns": 10417.0,
              "command_buffer_ns": 10416.668374091387,
              "calibration_cpu_ns": 400792.0,
              "calibration_gpu_ticks": 400792.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            7.58203125,
            8.6829375,
            7.572265625,
            7.772140625,
            7.579421875,
            7.6106875,
            7.674484375,
            8.8203125,
            8.8014375
          ],
          "compute_throughput_us_p50": 7.674484375,
          "compute_throughput_us_p90": 8.8052125,
          "command_buffer_throughput_us": [
            7.582031685160473,
            8.682942279847339,
            7.572265531052835,
            7.772135177219752,
            7.579426892334595,
            7.6113283284939826,
            7.674479093111586,
            8.820312359603122,
            8.802083357295487
          ],
          "command_buffer_throughput_us_p50": 7.674479093111586,
          "command_buffer_throughput_us_p90": 8.805729157757014,
          "compute_latency_us": [
            10.375,
            12.375,
            10.25,
            11.167,
            10.583,
            10.541,
            10.291,
            10.75,
            10.417
          ],
          "compute_latency_us_p50": 10.541,
          "compute_latency_us_p90": 11.4086,
          "command_buffer_latency_us": [
            10.374991688877344,
            12.37494871020317,
            10.25001984089613,
            11.208350770175457,
            10.583316907286644,
            10.500021744519472,
            10.25001984089613,
            10.749965440481901,
            10.416668374091387
          ],
          "command_buffer_latency_us_p50": 10.500021744519472,
          "command_buffer_latency_us_p90": 11.441670358181,
          "counter_control_throughput_ratio": 0.8929880738856327,
          "counter_control_latency_ratio": 1.0
        },
        "download_ms": 0.323916,
        "allocation_upload_ms": 0.718125,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.0790672678193687e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-y1v585ee/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "53243cd32bb4620fa91e934e3d500807c417fa70fc3ac1977dac07575a48be10",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 18.649459,
        "capture_ms": 0.075709,
        "compile_ms": 72.99925,
        "allocation_upload_ms": 1.588625,
        "cold_call_ms": 2.082875,
        "warmup_ms": 200.401542,
        "download_ms": 0.314791,
        "repetitions": 6026,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          5.055772154,
          4.98113043478,
          4.87037421175,
          4.93115947561,
          4.92723896449,
          4.93671174909,
          4.92855260538,
          4.93774195154,
          4.93675323598
        ],
        "latency_us": [
          249.459,
          213.083,
          198.791,
          197.958,
          181.75,
          213.416,
          311.541,
          217.375,
          210.875
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 19,
            "reduction_scalar_rounds": 40,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 18444,
              "global_write_bytes": 6148,
              "private_read_bytes": 36888,
              "private_write_bytes": 12296
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 32,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 2,
            "reduction_elements": 3074,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 80,
            "normalized_kernel_cost": 80,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 4611,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 312583,
              "compute_span_ns": 312583,
              "calibration_cpu_ns": 743167,
              "calibration_gpu_ticks": 743167,
              "compute_passes": 1,
              "command_buffer_ns": 312666.699756,
              "command_buffers": 2
            },
            {
              "compute_ns": 315708,
              "compute_span_ns": 315708,
              "calibration_cpu_ns": 728791,
              "calibration_gpu_ticks": 728791,
              "compute_passes": 1,
              "command_buffer_ns": 315750.017762,
              "command_buffers": 2
            },
            {
              "compute_ns": 293459,
              "compute_span_ns": 293459,
              "calibration_cpu_ns": 836875,
              "calibration_gpu_ticks": 836875,
              "compute_passes": 1,
              "command_buffer_ns": 293541.699648,
              "command_buffers": 2
            },
            {
              "compute_ns": 325083,
              "compute_span_ns": 325083,
              "calibration_cpu_ns": 837208,
              "calibration_gpu_ticks": 837208,
              "compute_passes": 1,
              "command_buffer_ns": 325083.325151,
              "command_buffers": 2
            },
            {
              "compute_ns": 294209,
              "compute_span_ns": 294209,
              "calibration_cpu_ns": 686709,
              "calibration_gpu_ticks": 686709,
              "compute_passes": 1,
              "command_buffer_ns": 294333.323836,
              "command_buffers": 2
            },
            {
              "compute_ns": 321292,
              "compute_span_ns": 321292,
              "calibration_cpu_ns": 1031708,
              "calibration_gpu_ticks": 1031708,
              "compute_passes": 1,
              "command_buffer_ns": 321375.03149,
              "command_buffers": 2
            },
            {
              "compute_ns": 299375,
              "compute_span_ns": 299375,
              "calibration_cpu_ns": 954375,
              "calibration_gpu_ticks": 954375,
              "compute_passes": 1,
              "command_buffer_ns": 299375.038594,
              "command_buffers": 2
            },
            {
              "compute_ns": 308209,
              "compute_span_ns": 308209,
              "calibration_cpu_ns": 742959,
              "calibration_gpu_ticks": 742959,
              "compute_passes": 1,
              "command_buffer_ns": 308333.372232,
              "command_buffers": 2
            },
            {
              "compute_ns": 338667,
              "compute_span_ns": 338667,
              "calibration_cpu_ns": 773375,
              "calibration_gpu_ticks": 773375,
              "compute_passes": 1,
              "command_buffer_ns": 338666.664902,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 7416,
              "compute_span_ns": 7416,
              "calibration_cpu_ns": 357708,
              "calibration_gpu_ticks": 357708,
              "compute_passes": 1,
              "command_buffer_ns": 7458.26400816,
              "command_buffers": 2
            },
            {
              "compute_ns": 7125,
              "compute_span_ns": 7125,
              "calibration_cpu_ns": 391333,
              "calibration_gpu_ticks": 391333,
              "compute_passes": 1,
              "command_buffer_ns": 7166.64362699,
              "command_buffers": 2
            },
            {
              "compute_ns": 7917,
              "compute_span_ns": 7917,
              "calibration_cpu_ns": 373167,
              "calibration_gpu_ticks": 373167,
              "compute_passes": 1,
              "command_buffer_ns": 8000.00270829,
              "command_buffers": 2
            },
            {
              "compute_ns": 8250,
              "compute_span_ns": 8250,
              "calibration_cpu_ns": 740833,
              "calibration_gpu_ticks": 740833,
              "compute_passes": 1,
              "command_buffer_ns": 8291.68129712,
              "command_buffers": 2
            },
            {
              "compute_ns": 8292,
              "compute_span_ns": 8292,
              "calibration_cpu_ns": 357875,
              "calibration_gpu_ticks": 357875,
              "compute_passes": 1,
              "command_buffer_ns": 8291.68129712,
              "command_buffers": 2
            },
            {
              "compute_ns": 7708,
              "compute_span_ns": 7708,
              "calibration_cpu_ns": 442167,
              "calibration_gpu_ticks": 442167,
              "compute_passes": 1,
              "command_buffer_ns": 7708.32411945,
              "command_buffers": 2
            },
            {
              "compute_ns": 7625,
              "compute_span_ns": 7625,
              "calibration_cpu_ns": 386083,
              "calibration_gpu_ticks": 386083,
              "compute_passes": 1,
              "command_buffer_ns": 7708.32411945,
              "command_buffers": 2
            },
            {
              "compute_ns": 7709,
              "compute_span_ns": 7709,
              "calibration_cpu_ns": 368250,
              "calibration_gpu_ticks": 368250,
              "compute_passes": 1,
              "command_buffer_ns": 7833.35417509,
              "command_buffers": 2
            },
            {
              "compute_ns": 7625,
              "compute_span_ns": 7625,
              "calibration_cpu_ns": 382500,
              "calibration_gpu_ticks": 382500,
              "compute_passes": 1,
              "command_buffer_ns": 7666.64743423,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 328458.379954,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 312874.966767,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 320041.668601,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 306624.977384,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 305083.289277,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 318000.034895,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 331458.34459,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 336624.972988,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 327916.699462,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 7249.99699742,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 10125.0479929,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7833.29596743,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7875.0308603,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7499.99890104,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7583.35227147,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 8833.30358192,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 11041.6440293,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7791.67748988,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              5.13216218678125,
              4.888671355734375,
              5.0006510718906245,
              4.7910152716250005,
              4.766926394953125,
              4.968750545234375,
              5.1790366342187495,
              5.2597652029375,
              5.12369842909375
            ],
            "command_buffer_throughput_us_p50": 5.0006510718906245,
            "command_buffer_throughput_us_p90": 5.1951823479625,
            "command_buffer_latency_us": [
              7.24999699742,
              10.125047992899999,
              7.83329596743,
              7.8750308603,
              7.4999989010400006,
              7.58335227147,
              8.83330358192,
              11.0416440293,
              7.7916774898800005
            ],
            "command_buffer_latency_us_p50": 7.83329596743,
            "command_buffer_latency_us_p90": 10.30836720018
          },
          "compute_throughput_us": [
            4.884109375,
            4.9329375,
            4.585296875,
            5.079421875,
            4.597015625,
            5.0201875,
            4.677734375,
            4.815765625,
            5.291671875
          ],
          "compute_throughput_us_p50": 4.884109375,
          "compute_throughput_us_p90": 5.121871875,
          "command_buffer_throughput_us": [
            4.8854171836875,
            4.9335940275312495,
            4.586589057,
            5.079426955484375,
            4.5989581849375,
            5.02148486703125,
            4.67773497803125,
            4.817708941125,
            5.29166663909375
          ],
          "command_buffer_throughput_us_p50": 4.8854171836875,
          "command_buffer_throughput_us_p90": 5.12187489220625,
          "compute_latency_us": [
            7.416,
            7.125,
            7.917,
            8.25,
            8.292,
            7.708,
            7.625,
            7.709,
            7.625
          ],
          "compute_latency_us_p50": 7.708,
          "compute_latency_us_p90": 8.2584,
          "command_buffer_latency_us": [
            7.4582640081600005,
            7.16664362699,
            8.00000270829,
            8.29168129712,
            8.29168129712,
            7.708324119449999,
            7.708324119449999,
            7.83335417509,
            7.66664743423
          ],
          "command_buffer_latency_us_p50": 7.708324119449999,
          "command_buffer_latency_us_p90": 8.29168129712,
          "counter_control_throughput_ratio": 0.9647638339468683,
          "counter_control_latency_ratio": 1.0164797629737137
        },
        "correctness": {
          "max_abs_error": 2.1991937471099732e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 4.93671174909,
        "throughput_us_p90": 4.996058778624,
        "latency_us_p50": 213.083,
        "latency_us_p90": 261.8754
      },
      "slowdown": 0.3970075534246198,
      "device_compute_slowdown": 0.6364087978223294,
      "valid": true,
      "round": 2,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "layernorm_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-hy2qqgbf/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "b8d9d984f3aaf8bfdb9269f6e0efc630605362c6ff4e852cc112881dfaf5e97b",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.105834,
        "capture_ms": 0.077458,
        "compile_ms": 64.523333,
        "allocation_upload_ms": 2.034625,
        "cold_call_ms": 9.394042,
        "warmup_ms": 200.249459,
        "download_ms": 0.897209,
        "repetitions": 1764,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          18.0781366213,
          16.9538219955,
          16.6022772109,
          16.5927579365,
          16.553760771,
          16.7946196145,
          16.8110119048,
          17.7272534014,
          17.4821660998
        ],
        "latency_us": [
          282.208,
          421.542,
          277.5,
          248.917,
          282.834,
          247.917,
          263.416,
          255.834,
          216.125
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 128,
            "reduction_scalar_rounds": 60,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 36864,
              "global_write_bytes": 12288,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 144,
              "global_write_bytes": 48,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 2,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 100,
            "normalized_kernel_cost": 400,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 9216,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 1052375,
              "compute_span_ns": 1052375,
              "calibration_cpu_ns": 1791625,
              "calibration_gpu_ticks": 1791625,
              "compute_passes": 1,
              "command_buffer_ns": 1052416.68643,
              "command_buffers": 2
            },
            {
              "compute_ns": 1038042,
              "compute_span_ns": 1038042,
              "calibration_cpu_ns": 1474541,
              "calibration_gpu_ticks": 1474541,
              "compute_passes": 1,
              "command_buffer_ns": 1038124.95945,
              "command_buffers": 2
            },
            {
              "compute_ns": 1075167,
              "compute_span_ns": 1075167,
              "calibration_cpu_ns": 1577208,
              "calibration_gpu_ticks": 1577208,
              "compute_passes": 1,
              "command_buffer_ns": 1075333.33357,
              "command_buffers": 2
            },
            {
              "compute_ns": 1047875,
              "compute_span_ns": 1047875,
              "calibration_cpu_ns": 1499209,
              "calibration_gpu_ticks": 1499209,
              "compute_passes": 1,
              "command_buffer_ns": 1047958.32885,
              "command_buffers": 2
            },
            {
              "compute_ns": 1030583,
              "compute_span_ns": 1030583,
              "calibration_cpu_ns": 1583042,
              "calibration_gpu_ticks": 1583042,
              "compute_passes": 1,
              "command_buffer_ns": 1030541.66539,
              "command_buffers": 2
            },
            {
              "compute_ns": 1088000,
              "compute_span_ns": 1088000,
              "calibration_cpu_ns": 1853833,
              "calibration_gpu_ticks": 1853833,
              "compute_passes": 1,
              "command_buffer_ns": 1088124.99093,
              "command_buffers": 2
            },
            {
              "compute_ns": 1102250,
              "compute_span_ns": 1102250,
              "calibration_cpu_ns": 1726625,
              "calibration_gpu_ticks": 1726625,
              "compute_passes": 1,
              "command_buffer_ns": 1102291.68786,
              "command_buffers": 2
            },
            {
              "compute_ns": 1045667,
              "compute_span_ns": 1045667,
              "calibration_cpu_ns": 1504792,
              "calibration_gpu_ticks": 1504792,
              "compute_passes": 1,
              "command_buffer_ns": 1045791.66509,
              "command_buffers": 2
            },
            {
              "compute_ns": 1083833,
              "compute_span_ns": 1083833,
              "calibration_cpu_ns": 1830000,
              "calibration_gpu_ticks": 1830000,
              "compute_passes": 1,
              "command_buffer_ns": 1083958.37015,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 26083,
              "compute_span_ns": 26083,
              "calibration_cpu_ns": 651375,
              "calibration_gpu_ticks": 651375,
              "compute_passes": 1,
              "command_buffer_ns": 26041.7000391,
              "command_buffers": 2
            },
            {
              "compute_ns": 20000,
              "compute_span_ns": 20000,
              "calibration_cpu_ns": 585167,
              "calibration_gpu_ticks": 585167,
              "compute_passes": 1,
              "command_buffer_ns": 20083.3310373,
              "command_buffers": 2
            },
            {
              "compute_ns": 20084,
              "compute_span_ns": 20084,
              "calibration_cpu_ns": 426209,
              "calibration_gpu_ticks": 426209,
              "compute_passes": 1,
              "command_buffer_ns": 20208.361093,
              "command_buffers": 2
            },
            {
              "compute_ns": 21084,
              "compute_span_ns": 21084,
              "calibration_cpu_ns": 614459,
              "calibration_gpu_ticks": 614459,
              "compute_passes": 1,
              "command_buffer_ns": 21166.6338146,
              "command_buffers": 2
            },
            {
              "compute_ns": 24750,
              "compute_span_ns": 24750,
              "calibration_cpu_ns": 1139042,
              "calibration_gpu_ticks": 1139042,
              "compute_passes": 1,
              "command_buffer_ns": 24833.3672062,
              "command_buffers": 2
            },
            {
              "compute_ns": 17750,
              "compute_span_ns": 17750,
              "calibration_cpu_ns": 435542,
              "calibration_gpu_ticks": 435542,
              "compute_passes": 1,
              "command_buffer_ns": 17833.3139047,
              "command_buffers": 2
            },
            {
              "compute_ns": 21875,
              "compute_span_ns": 21875,
              "calibration_cpu_ns": 613667,
              "calibration_gpu_ticks": 613667,
              "compute_passes": 1,
              "command_buffer_ns": 21874.9628402,
              "command_buffers": 2
            },
            {
              "compute_ns": 19708,
              "compute_span_ns": 19708,
              "calibration_cpu_ns": 472042,
              "calibration_gpu_ticks": 472042,
              "compute_passes": 1,
              "command_buffer_ns": 19791.6524485,
              "command_buffers": 2
            },
            {
              "compute_ns": 17542,
              "compute_span_ns": 17542,
              "calibration_cpu_ns": 430584,
              "calibration_gpu_ticks": 430584,
              "compute_passes": 1,
              "command_buffer_ns": 17666.6653715,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1144375.01179,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1027375.05222,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1105749.97962,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1091791.72432,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1017916.71477,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1041749.95795,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1034708.34438,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1043958.2984,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1104291.70309,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 18541.701138,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 26458.2922682,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20250.0377782,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20333.3329409,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20458.3047889,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20499.9814741,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20125.0077225,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 40874.9911003,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19375.0020117,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              17.88085955921875,
              16.0527351909375,
              17.277343431562503,
              17.0592456925,
              15.90494866828125,
              16.27734309296875,
              16.1673178809375,
              16.311848412499998,
              17.25455786078125
            ],
            "command_buffer_throughput_us_p50": 16.311848412499998,
            "command_buffer_throughput_us_p90": 17.39804665709375,
            "command_buffer_latency_us": [
              18.541701138,
              26.458292268199997,
              20.2500377782,
              20.3333329409,
              20.4583047889,
              20.4999814741,
              20.125007722499998,
              40.8749911003,
              19.375002011699998
            ],
            "command_buffer_latency_us_p50": 20.3333329409,
            "command_buffer_latency_us_p90": 29.34163203462
          },
          "compute_throughput_us": [
            16.443359375,
            16.21940625,
            16.799484375,
            16.373046875,
            16.102859375,
            17.0,
            17.22265625,
            16.338546875,
            16.934890625
          ],
          "compute_throughput_us_p50": 16.443359375,
          "compute_throughput_us_p90": 17.04453125,
          "command_buffer_throughput_us": [
            16.44401072546875,
            16.22070249140625,
            16.80208333703125,
            16.37434888828125,
            16.10221352171875,
            17.001952983281253,
            17.2233076228125,
            16.34049476703125,
            16.93684953359375
          ],
          "command_buffer_throughput_us_p50": 16.44401072546875,
          "command_buffer_throughput_us_p90": 17.046223911187504,
          "compute_latency_us": [
            26.083,
            20.0,
            20.084,
            21.084,
            24.75,
            17.75,
            21.875,
            19.708,
            17.542
          ],
          "compute_latency_us_p50": 20.084,
          "compute_latency_us_p90": 25.0166,
          "command_buffer_latency_us": [
            26.0417000391,
            20.0833310373,
            20.208361093,
            21.1666338146,
            24.833367206200002,
            17.8333139047,
            21.874962840200002,
            19.791652448500002,
            17.6666653715
          ],
          "command_buffer_latency_us_p50": 20.208361093,
          "command_buffer_latency_us_p90": 25.075033772780003,
          "counter_control_throughput_ratio": 1.0017561685105716,
          "counter_control_latency_ratio": 0.9979418959284675
        },
        "correctness": {
          "max_abs_error": 2.9463829065434766e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 16.8110119048,
        "throughput_us_p90": 17.79743004538,
        "latency_us_p50": 263.416,
        "latency_us_p90": 310.5756
      },
      "torch": {
        "cold_call_ms": 7.455625,
        "warmup_ms": 200.719834,
        "repetitions": 628,
        "throughput_us": [
          45.47744108280255,
          46.05812101910828,
          46.99648248407644,
          45.765127388535035,
          45.8454076433121,
          45.79889808917197,
          47.35496178343949,
          45.960654458598725,
          47.64052547770701
        ],
        "latency_us": [
          377.79200000000003,
          347.625,
          410.833,
          497.16700000000003,
          1015.2080000000001,
          338.625,
          291.25,
          468.333,
          285.125
        ],
        "throughput_us_p50": 45.960654458598725,
        "throughput_us_p90": 47.412074522293,
        "latency_us_p50": 377.79200000000003,
        "latency_us_p90": 600.7752000000002,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 2787166.682537645,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2929500.0131241977,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2635375.0145062804,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2846333.372872323,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2644416.701514274,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2802625.0074617565,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2656833.326909691,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 3051250.008866191,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2574499.9875314534,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 42791.65295884013,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49916.67810827494,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 41958.35208520293,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 50249.97517466545,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44666.69633984566,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44749.99150261283,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 46500.00482797623,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 45041.670091450214,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43041.654862463474,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              43.549479414650705,
              45.77343770506559,
              41.17773460166063,
              44.473958951130044,
              41.31901096116053,
              43.791015741589945,
              41.51302073296392,
              47.67578138853423,
              40.22656230517896
            ],
            "command_buffer_throughput_us_p50": 43.549479414650705,
            "command_buffer_throughput_us_p90": 46.15390644175932,
            "command_buffer_latency_us": [
              42.79165295884013,
              49.91667810827494,
              41.95835208520293,
              50.24997517466545,
              44.66669633984566,
              44.74999150261283,
              46.50000482797623,
              45.041670091450214,
              43.041654862463474
            ],
            "command_buffer_latency_us_p50": 44.74999150261283,
            "command_buffer_latency_us_p90": 49.98333752155304
          },
          "throughput": [
            {
              "compute_ns": 2568916.0,
              "compute_span_ns": 2568916.0,
              "command_buffer_ns": 2568874.9738037586,
              "calibration_cpu_ns": 3429792.0,
              "calibration_gpu_ticks": 3429792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 3025167.0,
              "compute_span_ns": 3025167.0,
              "command_buffer_ns": 3025166.690349579,
              "calibration_cpu_ns": 4241375.0,
              "calibration_gpu_ticks": 4241375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2872459.0,
              "compute_span_ns": 2872459.0,
              "command_buffer_ns": 2872499.986551702,
              "calibration_cpu_ns": 3749292.0,
              "calibration_gpu_ticks": 3749292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2559208.0,
              "compute_span_ns": 2559208.0,
              "command_buffer_ns": 2559208.3111405373,
              "calibration_cpu_ns": 3517792.0,
              "calibration_gpu_ticks": 3517792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2812917.0,
              "compute_span_ns": 2812917.0,
              "command_buffer_ns": 2812916.6457802057,
              "calibration_cpu_ns": 3721083.0,
              "calibration_gpu_ticks": 3721083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2636125.0,
              "compute_span_ns": 2636125.0,
              "command_buffer_ns": 2636125.0202171504,
              "calibration_cpu_ns": 3837583.0,
              "calibration_gpu_ticks": 3837583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2815500.0,
              "compute_span_ns": 2815500.0,
              "command_buffer_ns": 2815500.0181868672,
              "calibration_cpu_ns": 3749917.0,
              "calibration_gpu_ticks": 3749917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2569166.0,
              "compute_span_ns": 2569166.0,
              "command_buffer_ns": 2569125.033915043,
              "calibration_cpu_ns": 3972209.0,
              "calibration_gpu_ticks": 3972209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2836875.0,
              "compute_span_ns": 2836875.0,
              "command_buffer_ns": 2836874.97721985,
              "calibration_cpu_ns": 3958542.0,
              "calibration_gpu_ticks": 3958542.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 49334.0,
              "compute_span_ns": 49334.0,
              "command_buffer_ns": 49374.99761581421,
              "calibration_cpu_ns": 825292.0,
              "calibration_gpu_ticks": 825292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44167.0,
              "compute_span_ns": 44167.0,
              "command_buffer_ns": 44208.3110101521,
              "calibration_cpu_ns": 573417.0,
              "calibration_gpu_ticks": 573417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 43459.0,
              "compute_span_ns": 43459.0,
              "command_buffer_ns": 43500.04019215703,
              "calibration_cpu_ns": 520000.0,
              "calibration_gpu_ticks": 520000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44583.0,
              "compute_span_ns": 44583.0,
              "command_buffer_ns": 44541.66628420353,
              "calibration_cpu_ns": 449917.0,
              "calibration_gpu_ticks": 449917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 45166.0,
              "compute_span_ns": 45166.0,
              "command_buffer_ns": 45125.0234618783,
              "calibration_cpu_ns": 1347333.0,
              "calibration_gpu_ticks": 1347333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44083.0,
              "compute_span_ns": 44083.0,
              "command_buffer_ns": 44083.33916217089,
              "calibration_cpu_ns": 478583.0,
              "calibration_gpu_ticks": 478583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 43750.0,
              "compute_span_ns": 43750.0,
              "command_buffer_ns": 43750.04209578037,
              "calibration_cpu_ns": 463375.0,
              "calibration_gpu_ticks": 463375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44584.0,
              "compute_span_ns": 44584.0,
              "command_buffer_ns": 44625.019654631615,
              "calibration_cpu_ns": 944792.0,
              "calibration_gpu_ticks": 944792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 43458.0,
              "compute_span_ns": 43458.0,
              "command_buffer_ns": 43416.686821728945,
              "calibration_cpu_ns": 522208.0,
              "calibration_gpu_ticks": 522208.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            40.1393125,
            47.268234375,
            44.882171875,
            39.987625,
            43.951828125,
            41.189453125,
            43.9921875,
            40.14321875,
            44.326171875
          ],
          "compute_throughput_us_p50": 43.951828125,
          "compute_throughput_us_p90": 45.359384375,
          "command_buffer_throughput_us": [
            40.13867146568373,
            47.26822953671217,
            44.882812289870344,
            39.987629861570895,
            43.951822590315714,
            41.189453440892976,
            43.9921877841698,
            40.142578654922545,
            44.32617151906015
          ],
          "command_buffer_throughput_us_p50": 43.951822590315714,
          "command_buffer_throughput_us_p90": 45.35989573923871,
          "compute_latency_us": [
            49.334,
            44.167,
            43.459,
            44.583,
            45.166,
            44.083,
            43.75,
            44.584,
            43.458
          ],
          "compute_latency_us_p50": 44.167,
          "compute_latency_us_p90": 45.9996,
          "command_buffer_latency_us": [
            49.37499761581421,
            44.2083110101521,
            43.50004019215703,
            44.54166628420353,
            45.1250234618783,
            44.08333916217089,
            43.75004209578037,
            44.625019654631615,
            43.416686821728945
          ],
          "command_buffer_latency_us_p50": 44.2083110101521,
          "command_buffer_latency_us_p90": 45.97501829266548,
          "counter_control_throughput_ratio": 1.032656315684176,
          "counter_control_latency_ratio": 0.990749667230974
        },
        "download_ms": 0.36875,
        "allocation_upload_ms": 1.827208,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 2.982737865586671e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.3657696371565668,
      "device_compute_slowdown": 0.3741223079102583,
      "valid": true,
      "round": 2,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "layernorm_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.298667,
        "warmup_ms": 201.122125,
        "repetitions": 589,
        "throughput_us": [
          46.27801358234296,
          46.16843463497453,
          46.534592529711375,
          47.36622750424448,
          45.89643463497453,
          46.57130730050934,
          46.24455348047538,
          45.935483870967744,
          45.84939049235994
        ],
        "latency_us": [
          412.292,
          263.291,
          273.0,
          260.91700000000003,
          273.083,
          282.375,
          288.95799999999997,
          327.29200000000003,
          304.666
        ],
        "throughput_us_p50": 46.24455348047538,
        "throughput_us_p90": 46.73029134125637,
        "latency_us_p50": 282.375,
        "latency_us_p90": 344.29200000000003,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 2659999.9982863665,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2581249.9807216227,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2617000.020109117,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2591541.677247733,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2605749.9926537275,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2587124.996352941,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2599874.9770224094,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2626791.7128279805,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2596708.363853395,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 43999.9857917428,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44249.987695366144,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44249.987695366144,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 46583.35819840431,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49541.646149009466,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43791.6605733335,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43625.012040138245,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44333.34106579423,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 45041.670091450214,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              41.562499973224476,
              40.332030948775355,
              40.890625314204954,
              40.492838706995826,
              40.71484363521449,
              40.4238280680147,
              40.62304651597515,
              41.043620512937196,
              40.5735681852093
            ],
            "command_buffer_throughput_us_p50": 40.62304651597515,
            "command_buffer_throughput_us_p90": 41.14739640499465,
            "command_buffer_latency_us": [
              43.9999857917428,
              44.249987695366144,
              44.249987695366144,
              46.58335819840431,
              49.541646149009466,
              43.7916605733335,
              43.625012040138245,
              44.33334106579423,
              45.041670091450214
            ],
            "command_buffer_latency_us_p50": 44.249987695366144,
            "command_buffer_latency_us_p90": 47.17501578852534
          },
          "throughput": [
            {
              "compute_ns": 2653250.0,
              "compute_span_ns": 2653250.0,
              "command_buffer_ns": 2653250.005096197,
              "calibration_cpu_ns": 3696042.0,
              "calibration_gpu_ticks": 3696042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2593667.0,
              "compute_span_ns": 2593667.0,
              "command_buffer_ns": 2593666.664324701,
              "calibration_cpu_ns": 3523000.0,
              "calibration_gpu_ticks": 3523000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2613000.0,
              "compute_span_ns": 2613000.0,
              "command_buffer_ns": 2612999.9896511436,
              "calibration_cpu_ns": 3680625.0,
              "calibration_gpu_ticks": 3680625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2597125.0,
              "compute_span_ns": 2597125.0,
              "command_buffer_ns": 2597125.0142902136,
              "calibration_cpu_ns": 3552084.0,
              "calibration_gpu_ticks": 3552084.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2653750.0,
              "compute_span_ns": 2653750.0,
              "command_buffer_ns": 2653749.950695783,
              "calibration_cpu_ns": 3631750.0,
              "calibration_gpu_ticks": 3631750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2613584.0,
              "compute_span_ns": 2613584.0,
              "command_buffer_ns": 2613625.0235140324,
              "calibration_cpu_ns": 3491417.0,
              "calibration_gpu_ticks": 3491417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2602000.0,
              "compute_span_ns": 2602000.0,
              "command_buffer_ns": 2601999.9640993774,
              "calibration_cpu_ns": 3584166.0,
              "calibration_gpu_ticks": 3584166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2604166.0,
              "compute_span_ns": 2604166.0,
              "command_buffer_ns": 2604125.0093840063,
              "calibration_cpu_ns": 3622417.0,
              "calibration_gpu_ticks": 3622417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2566542.0,
              "compute_span_ns": 2566542.0,
              "command_buffer_ns": 2566541.6615083814,
              "calibration_cpu_ns": 3506416.0,
              "calibration_gpu_ticks": 3506416.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 44042.0,
              "compute_span_ns": 44042.0,
              "command_buffer_ns": 44083.33916217089,
              "calibration_cpu_ns": 602500.0,
              "calibration_gpu_ticks": 602500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44958.0,
              "compute_span_ns": 44958.0,
              "command_buffer_ns": 44916.64003580809,
              "calibration_cpu_ns": 528750.0,
              "calibration_gpu_ticks": 528750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44750.0,
              "compute_span_ns": 44750.0,
              "command_buffer_ns": 44749.99150261283,
              "calibration_cpu_ns": 455375.0,
              "calibration_gpu_ticks": 455375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44958.0,
              "compute_span_ns": 44958.0,
              "command_buffer_ns": 44958.37492868304,
              "calibration_cpu_ns": 455291.0,
              "calibration_gpu_ticks": 455291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44042.0,
              "compute_span_ns": 44042.0,
              "command_buffer_ns": 44041.662476956844,
              "calibration_cpu_ns": 488000.0,
              "calibration_gpu_ticks": 488000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44041.0,
              "compute_span_ns": 44041.0,
              "command_buffer_ns": 43999.9857917428,
              "calibration_cpu_ns": 445375.0,
              "calibration_gpu_ticks": 445375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47709.0,
              "compute_span_ns": 47709.0,
              "command_buffer_ns": 47750.01434609294,
              "calibration_cpu_ns": 757250.0,
              "calibration_gpu_ticks": 757250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44916.0,
              "compute_span_ns": 44916.0,
              "command_buffer_ns": 44875.02155825496,
              "calibration_cpu_ns": 724292.0,
              "calibration_gpu_ticks": 724292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44792.0,
              "compute_span_ns": 44792.0,
              "command_buffer_ns": 44833.344873040915,
              "calibration_cpu_ns": 481917.0,
              "calibration_gpu_ticks": 481917.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            41.45703125,
            40.526046875,
            40.828125,
            40.580078125,
            41.46484375,
            40.83725,
            40.65625,
            40.69009375,
            40.10221875
          ],
          "compute_throughput_us_p50": 40.69009375,
          "compute_throughput_us_p90": 41.45859375,
          "command_buffer_throughput_us": [
            41.45703132962808,
            40.52604163007345,
            40.82812483829912,
            40.58007834828459,
            41.46484297962161,
            40.837890992406756,
            40.65624943905277,
            40.6894532716251,
            40.10221346106846
          ],
          "command_buffer_throughput_us_p50": 40.6894532716251,
          "command_buffer_throughput_us_p90": 41.458593659626786,
          "compute_latency_us": [
            44.042,
            44.958,
            44.75,
            44.958,
            44.042,
            44.041,
            47.709,
            44.916,
            44.792
          ],
          "compute_latency_us_p50": 44.792,
          "compute_latency_us_p90": 45.5082,
          "command_buffer_latency_us": [
            44.08333916217089,
            44.916640035808086,
            44.74999150261283,
            44.95837492868304,
            44.041662476956844,
            43.9999857917428,
            47.75001434609294,
            44.87502155825496,
            44.833344873040915
          ],
          "command_buffer_latency_us_p50": 44.833344873040915,
          "command_buffer_latency_us_p90": 45.51670281216502,
          "counter_control_throughput_ratio": 1.0008173420244237,
          "counter_control_latency_ratio": 1.004757189284029
        },
        "download_ms": 0.426125,
        "allocation_upload_ms": 1.010417,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 2.982737865586671e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-bwz4kdxy/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "83a70ba0e4d1dab6671da56395e0b13750f8d9c9277e4782c4e371450de131c7",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 18.900041,
        "capture_ms": 0.073584,
        "compile_ms": 63.341333,
        "allocation_upload_ms": 2.824459,
        "cold_call_ms": 4.789875,
        "warmup_ms": 200.381583,
        "download_ms": 1.467,
        "repetitions": 1758,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          17.084447099,
          17.1710989761,
          17.081129124,
          20.9672690557,
          18.7458993174,
          18.0834755404,
          20.8624385666,
          17.5531382253,
          17.4598737201
        ],
        "latency_us": [
          316,
          214.875,
          210.041,
          228.208,
          245.834,
          248.25,
          279.708,
          238.667,
          313.792
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 256,
            "reduction_scalar_rounds": 60,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 36864,
              "global_write_bytes": 12288,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 144,
              "global_write_bytes": 48,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 2,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 108,
            "normalized_kernel_cost": 432,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 9216,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 1034083,
              "compute_span_ns": 1034083,
              "calibration_cpu_ns": 1613209,
              "calibration_gpu_ticks": 1613209,
              "compute_passes": 1,
              "command_buffer_ns": 1034166.72209,
              "command_buffers": 2
            },
            {
              "compute_ns": 1046500,
              "compute_span_ns": 1046500,
              "calibration_cpu_ns": 1498041,
              "calibration_gpu_ticks": 1498041,
              "compute_passes": 1,
              "command_buffer_ns": 1046583.34749,
              "command_buffers": 2
            },
            {
              "compute_ns": 1043583,
              "compute_span_ns": 1043583,
              "calibration_cpu_ns": 1603750,
              "calibration_gpu_ticks": 1603750,
              "compute_passes": 1,
              "command_buffer_ns": 1043583.32464,
              "command_buffers": 2
            },
            {
              "compute_ns": 1122334,
              "compute_span_ns": 1122334,
              "calibration_cpu_ns": 1885625,
              "calibration_gpu_ticks": 1885625,
              "compute_passes": 1,
              "command_buffer_ns": 1122375.01889,
              "command_buffers": 2
            },
            {
              "compute_ns": 1135875,
              "compute_span_ns": 1135875,
              "calibration_cpu_ns": 2138125,
              "calibration_gpu_ticks": 2138125,
              "compute_passes": 1,
              "command_buffer_ns": 1135916.68196,
              "command_buffers": 2
            },
            {
              "compute_ns": 1067583,
              "compute_span_ns": 1067583,
              "calibration_cpu_ns": 1514667,
              "calibration_gpu_ticks": 1514667,
              "compute_passes": 1,
              "command_buffer_ns": 1067541.65608,
              "command_buffers": 2
            },
            {
              "compute_ns": 1057875,
              "compute_span_ns": 1057875,
              "calibration_cpu_ns": 1537834,
              "calibration_gpu_ticks": 1537834,
              "compute_passes": 1,
              "command_buffer_ns": 1057874.99342,
              "command_buffers": 2
            },
            {
              "compute_ns": 1114709,
              "compute_span_ns": 1114709,
              "calibration_cpu_ns": 1904375,
              "calibration_gpu_ticks": 1904375,
              "compute_passes": 1,
              "command_buffer_ns": 1114833.34331,
              "command_buffers": 2
            },
            {
              "compute_ns": 1080042,
              "compute_span_ns": 1080042,
              "calibration_cpu_ns": 1796208,
              "calibration_gpu_ticks": 1796208,
              "compute_passes": 1,
              "command_buffer_ns": 1080124.98822,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 18750,
              "compute_span_ns": 18750,
              "calibration_cpu_ns": 442833,
              "calibration_gpu_ticks": 442833,
              "compute_passes": 1,
              "command_buffer_ns": 18750.0263564,
              "command_buffers": 2
            },
            {
              "compute_ns": 25541,
              "compute_span_ns": 25541,
              "calibration_cpu_ns": 778542,
              "calibration_gpu_ticks": 778542,
              "compute_passes": 1,
              "command_buffer_ns": 25583.372917,
              "command_buffers": 2
            },
            {
              "compute_ns": 20084,
              "compute_span_ns": 20084,
              "calibration_cpu_ns": 427875,
              "calibration_gpu_ticks": 427875,
              "compute_passes": 1,
              "command_buffer_ns": 20208.361093,
              "command_buffers": 2
            },
            {
              "compute_ns": 19917,
              "compute_span_ns": 19917,
              "calibration_cpu_ns": 547125,
              "calibration_gpu_ticks": 547125,
              "compute_passes": 1,
              "command_buffer_ns": 20041.6543521,
              "command_buffers": 2
            },
            {
              "compute_ns": 20042,
              "compute_span_ns": 20042,
              "calibration_cpu_ns": 457458,
              "calibration_gpu_ticks": 457458,
              "compute_passes": 1,
              "command_buffer_ns": 20125.0077225,
              "command_buffers": 2
            },
            {
              "compute_ns": 19792,
              "compute_span_ns": 19792,
              "calibration_cpu_ns": 411625,
              "calibration_gpu_ticks": 411625,
              "compute_passes": 1,
              "command_buffer_ns": 19916.6825041,
              "command_buffers": 2
            },
            {
              "compute_ns": 26542,
              "compute_span_ns": 26542,
              "calibration_cpu_ns": 658542,
              "calibration_gpu_ticks": 658542,
              "compute_passes": 1,
              "command_buffer_ns": 26541.6456386,
              "command_buffers": 2
            },
            {
              "compute_ns": 20458,
              "compute_span_ns": 20458,
              "calibration_cpu_ns": 437625,
              "calibration_gpu_ticks": 437625,
              "compute_passes": 1,
              "command_buffer_ns": 20499.9814741,
              "command_buffers": 2
            },
            {
              "compute_ns": 28833,
              "compute_span_ns": 28833,
              "calibration_cpu_ns": 2102167,
              "calibration_gpu_ticks": 2102167,
              "compute_passes": 1,
              "command_buffer_ns": 28916.6346192,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1148083.36366,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1125958.34071,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1033416.65817,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1046791.67271,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1059791.65528,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1139124.97181,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1064583.36813,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1089124.99854,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1069791.67322,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 17958.3439603,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 22374.9666475,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20125.0077225,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 27875.0085272,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20916.6901186,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 23541.6810028,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 27374.9465123,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19916.6825041,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20083.3310373,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              17.9388025571875,
              17.59309907359375,
              16.14713528390625,
              16.35611988609375,
              16.55924461375,
              17.79882768453125,
              16.63411512703125,
              17.0175781021875,
              16.7154948940625
            ],
            "command_buffer_throughput_us_p50": 16.7154948940625,
            "command_buffer_throughput_us_p90": 17.8268226590625,
            "command_buffer_latency_us": [
              17.958343960300002,
              22.374966647500003,
              20.125007722499998,
              27.8750085272,
              20.9166901186,
              23.5416810028,
              27.3749465123,
              19.916682504100002,
              20.0833310373
            ],
            "command_buffer_latency_us_p50": 20.9166901186,
            "command_buffer_latency_us_p90": 27.47495891528
          },
          "compute_throughput_us": [
            16.157546875,
            16.3515625,
            16.305984375,
            17.53646875,
            17.748046875,
            16.680984375,
            16.529296875,
            17.417328125,
            16.87565625
          ],
          "compute_throughput_us_p50": 16.680984375,
          "compute_throughput_us_p90": 17.578784375,
          "command_buffer_throughput_us": [
            16.158855032656252,
            16.35286480453125,
            16.3059894475,
            17.537109670156248,
            17.748698155625,
            16.680338376250003,
            16.5292967721875,
            17.41927098921875,
            16.8769529409375
          ],
          "command_buffer_throughput_us_p50": 16.680338376250003,
          "command_buffer_throughput_us_p90": 17.57942736725,
          "compute_latency_us": [
            18.75,
            25.541,
            20.084,
            19.917,
            20.042,
            19.792,
            26.542,
            20.458,
            28.833
          ],
          "compute_latency_us_p50": 20.084,
          "compute_latency_us_p90": 27.000200000000003,
          "command_buffer_latency_us": [
            18.750026356400003,
            25.583372917000002,
            20.208361093,
            20.0416543521,
            20.125007722499998,
            19.916682504100002,
            26.5416456386,
            20.4999814741,
            28.9166346192
          ],
          "command_buffer_latency_us_p50": 20.208361093,
          "command_buffer_latency_us_p90": 27.01664343472,
          "counter_control_throughput_ratio": 1.0096591843614724,
          "counter_control_latency_ratio": 1.0041417807957813
        },
        "correctness": {
          "max_abs_error": 2.9463829065434766e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 17.5531382253,
        "throughput_us_p90": 20.883404664420002,
        "latency_us_p50": 245.834,
        "latency_us_p90": 314.23359999999997
      },
      "slowdown": 0.3795720123605691,
      "device_compute_slowdown": 0.40995197694770613,
      "valid": true,
      "round": 2,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "layernorm_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-7cd1zkdh/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "4f0c4e70dcb06b507aa82e242c8819acd3628d5f79a065eb6c2f6b73ec6f396c",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 22.831958,
        "capture_ms": 0.079083,
        "compile_ms": 63.027209,
        "allocation_upload_ms": 7.086292,
        "cold_call_ms": 5.285792,
        "warmup_ms": 200.539875,
        "download_ms": 3.696125,
        "repetitions": 382,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          77.163065445,
          77.0316308901,
          76.5173429319,
          77.457460733,
          76.7750863874,
          77.0513743455,
          77.0297774869,
          77.1367801047,
          76.7993010471
        ],
        "latency_us": [
          1080.333,
          326.792,
          299.666,
          338.625,
          338.666,
          312.625,
          360.667,
          309,
          333.958
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 768,
            "reduction_scalar_rounds": 120,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 73728,
              "global_write_bytes": 24576,
              "private_read_bytes": 147456,
              "private_write_bytes": 49152
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 288,
              "global_write_bytes": 96,
              "private_read_bytes": 576,
              "private_write_bytes": 192
            },
            "striped_storage_scalars_per_worker": 48,
            "reduction_operations": 2,
            "reduction_elements": 12288,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 168,
            "normalized_kernel_cost": 2016,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 18432,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4781416,
              "compute_span_ns": 4781416,
              "calibration_cpu_ns": 5484625,
              "calibration_gpu_ticks": 5484625,
              "compute_passes": 1,
              "command_buffer_ns": 4781374.99886,
              "command_buffers": 2
            },
            {
              "compute_ns": 4754958,
              "compute_span_ns": 4754958,
              "calibration_cpu_ns": 5397458,
              "calibration_gpu_ticks": 5397458,
              "compute_passes": 1,
              "command_buffer_ns": 4754916.70659,
              "command_buffers": 2
            },
            {
              "compute_ns": 4784250,
              "compute_span_ns": 4784250,
              "calibration_cpu_ns": 5471667,
              "calibration_gpu_ticks": 5471667,
              "compute_passes": 1,
              "command_buffer_ns": 4784375.0217,
              "command_buffers": 2
            },
            {
              "compute_ns": 4773166,
              "compute_span_ns": 4773166,
              "calibration_cpu_ns": 5435166,
              "calibration_gpu_ticks": 5435166,
              "compute_passes": 1,
              "command_buffer_ns": 4773208.34761,
              "command_buffers": 2
            },
            {
              "compute_ns": 4891167,
              "compute_span_ns": 4891167,
              "calibration_cpu_ns": 5687250,
              "calibration_gpu_ticks": 5687250,
              "compute_passes": 1,
              "command_buffer_ns": 4891249.99149,
              "command_buffers": 2
            },
            {
              "compute_ns": 4873958,
              "compute_span_ns": 4873958,
              "calibration_cpu_ns": 5609750,
              "calibration_gpu_ticks": 5609750,
              "compute_passes": 1,
              "command_buffer_ns": 4873958.29987,
              "command_buffers": 2
            },
            {
              "compute_ns": 4854750,
              "compute_span_ns": 4854750,
              "calibration_cpu_ns": 5483875,
              "calibration_gpu_ticks": 5483875,
              "compute_passes": 1,
              "command_buffer_ns": 4854750.0046,
              "command_buffers": 2
            },
            {
              "compute_ns": 4907500,
              "compute_span_ns": 4907500,
              "calibration_cpu_ns": 5632958,
              "calibration_gpu_ticks": 5632958,
              "compute_passes": 1,
              "command_buffer_ns": 4907541.67549,
              "command_buffers": 2
            },
            {
              "compute_ns": 4831917,
              "compute_span_ns": 4831917,
              "calibration_cpu_ns": 5561084,
              "calibration_gpu_ticks": 5561084,
              "compute_passes": 1,
              "command_buffer_ns": 4831958.32931,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 111833,
              "compute_span_ns": 111833,
              "calibration_cpu_ns": 1004958,
              "calibration_gpu_ticks": 1004958,
              "compute_passes": 1,
              "command_buffer_ns": 111791.65449,
              "command_buffers": 2
            },
            {
              "compute_ns": 75459,
              "compute_span_ns": 75459,
              "calibration_cpu_ns": 483125,
              "calibration_gpu_ticks": 483125,
              "compute_passes": 1,
              "command_buffer_ns": 75499.9928176,
              "command_buffers": 2
            },
            {
              "compute_ns": 117583,
              "compute_span_ns": 117583,
              "calibration_cpu_ns": 1029000,
              "calibration_gpu_ticks": 1029000,
              "compute_passes": 1,
              "command_buffer_ns": 117541.640066,
              "command_buffers": 2
            },
            {
              "compute_ns": 99625,
              "compute_span_ns": 99625,
              "calibration_cpu_ns": 651917,
              "calibration_gpu_ticks": 651917,
              "compute_passes": 1,
              "command_buffer_ns": 99750.0028461,
              "command_buffers": 2
            },
            {
              "compute_ns": 76500,
              "compute_span_ns": 76500,
              "calibration_cpu_ns": 497375,
              "calibration_gpu_ticks": 497375,
              "compute_passes": 1,
              "command_buffer_ns": 76541.6189097,
              "command_buffers": 2
            },
            {
              "compute_ns": 92458,
              "compute_span_ns": 92458,
              "calibration_cpu_ns": 568042,
              "calibration_gpu_ticks": 568042,
              "compute_passes": 1,
              "command_buffer_ns": 92541.6825339,
              "command_buffers": 2
            },
            {
              "compute_ns": 103958,
              "compute_span_ns": 103958,
              "calibration_cpu_ns": 831333,
              "calibration_gpu_ticks": 831333,
              "compute_passes": 1,
              "command_buffer_ns": 103958.358523,
              "command_buffers": 2
            },
            {
              "compute_ns": 76875,
              "compute_span_ns": 76875,
              "calibration_cpu_ns": 493542,
              "calibration_gpu_ticks": 493542,
              "compute_passes": 1,
              "command_buffer_ns": 76875.0323914,
              "command_buffers": 2
            },
            {
              "compute_ns": 113125,
              "compute_span_ns": 113125,
              "calibration_cpu_ns": 706334,
              "calibration_gpu_ticks": 706334,
              "compute_passes": 1,
              "command_buffer_ns": 113166.694064,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4841624.99197,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4808583.29684,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4830208.37419,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4745624.95947,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4974583.33848,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4906541.72609,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4869750.06061,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4826291.63889,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4808125.02792,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 121666.642372,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 75458.3743401,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 93000.009656,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 100958.393887,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76124.9684729,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 117874.937132,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 79833.3785497,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 75625.0228733,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 94124.9891184,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              75.65039049953124,
              75.134114013125,
              75.47200584671874,
              74.15038999171875,
              77.72786466375001,
              76.66471447015626,
              76.08984469703125,
              75.41080685765625,
              75.12695356125
            ],
            "command_buffer_throughput_us_p50": 75.47200584671874,
            "command_buffer_throughput_us_p90": 76.87734450887501,
            "command_buffer_latency_us": [
              121.666642372,
              75.4583743401,
              93.00000965599999,
              100.958393887,
              76.1249684729,
              117.874937132,
              79.8333785497,
              75.6250228733,
              94.1249891184
            ],
            "command_buffer_latency_us_p50": 93.00000965599999,
            "command_buffer_latency_us_p90": 118.63327818
          },
          "compute_throughput_us": [
            74.709625,
            74.29621875,
            74.75390625,
            74.58071875,
            76.424484375,
            76.15559375,
            75.85546875,
            76.6796875,
            75.498703125
          ],
          "compute_throughput_us_p50": 75.498703125,
          "compute_throughput_us_p90": 76.475525,
          "command_buffer_throughput_us": [
            74.70898435718749,
            74.29557354046874,
            74.75585971406251,
            74.58138043140625,
            76.42578111703125,
            76.15559843546875,
            75.855468821875,
            76.68033867953126,
            75.49934889546874
          ],
          "command_buffer_throughput_us_p50": 75.49934889546874,
          "command_buffer_throughput_us_p90": 76.47669262953126,
          "compute_latency_us": [
            111.833,
            75.459,
            117.583,
            99.625,
            76.5,
            92.458,
            103.958,
            76.875,
            113.125
          ],
          "compute_latency_us_p50": 99.625,
          "compute_latency_us_p90": 114.0166,
          "command_buffer_latency_us": [
            111.79165449,
            75.4999928176,
            117.54164006600001,
            99.7500028461,
            76.5416189097,
            92.5416825339,
            103.958358523,
            76.8750323914,
            113.166694064
          ],
          "command_buffer_latency_us_p50": 99.7500028461,
          "command_buffer_latency_us_p90": 114.0416832644,
          "counter_control_throughput_ratio": 0.9933591869713976,
          "counter_control_latency_ratio": 1.0054732428158355
        },
        "correctness": {
          "max_abs_error": 2.7199924934606656e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 77.0316308901,
        "throughput_us_p90": 77.2219445026,
        "latency_us_p50": 333.958,
        "latency_us_p90": 504.60020000000014
      },
      "torch": {
        "cold_call_ms": 8.562375,
        "warmup_ms": 202.171417,
        "repetitions": 106,
        "throughput_us": [
          282.64740566037733,
          279.89700943396224,
          281.07390566037736,
          282.1057358490566,
          279.44654716981137,
          282.1619433962264,
          277.89111320754716,
          278.04245283018867,
          278.77240566037733
        ],
        "latency_us": [
          1229.917,
          525.208,
          551.2909999999999,
          812.2080000000001,
          505.12500000000006,
          501.25,
          508.87499999999994,
          517.667,
          536.1669999999999
        ],
        "throughput_us_p50": 279.89700943396224,
        "throughput_us_p90": 282.25903584905654,
        "latency_us_p50": 525.208,
        "latency_us_p90": 895.7498000000002,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 17369874.9858886,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17438083.36323127,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17252749.99625981,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17332000.0176318,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17193250.00885874,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17304749.984759837,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17340250.02224371,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17311166.680883616,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17311375.006102026,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 273166.6900217533,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 283333.35649222136,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 279875.0065267086,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 283125.03127381206,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 283250.00312179327,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 282749.9993145466,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 281458.31311121583,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 271124.9981075525,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 287208.3568945527,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              271.4042966545094,
              272.4700525504886,
              269.5742186915595,
              270.81250027549686,
              268.6445313884178,
              270.38671851187246,
              270.94140659755794,
              270.4869793888065,
              270.49023447034415
            ],
            "command_buffer_throughput_us_p50": 270.49023447034415,
            "command_buffer_throughput_us_p90": 271.6174478337052,
            "command_buffer_latency_us": [
              273.1666900217533,
              283.33335649222136,
              279.8750065267086,
              283.12503127381206,
              283.25000312179327,
              282.7499993145466,
              281.45831311121583,
              271.1249981075525,
              287.2083568945527
            ],
            "command_buffer_latency_us_p50": 282.7499993145466,
            "command_buffer_latency_us_p90": 284.1083565726876
          },
          "throughput": [
            {
              "compute_ns": 17251708.0,
              "compute_span_ns": 17251708.0,
              "command_buffer_ns": 17251666.635274887,
              "calibration_cpu_ns": 18290250.0,
              "calibration_gpu_ticks": 18290250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17283917.0,
              "compute_span_ns": 17283917.0,
              "command_buffer_ns": 17283958.32469687,
              "calibration_cpu_ns": 18291583.0,
              "calibration_gpu_ticks": 18291583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17414917.0,
              "compute_span_ns": 17414917.0,
              "command_buffer_ns": 17414958.332665265,
              "calibration_cpu_ns": 18414917.0,
              "calibration_gpu_ticks": 18414917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17271334.0,
              "compute_span_ns": 17271334.0,
              "command_buffer_ns": 17271374.992560595,
              "calibration_cpu_ns": 18560667.0,
              "calibration_gpu_ticks": 18560667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17618583.0,
              "compute_span_ns": 17618583.0,
              "command_buffer_ns": 17618541.66397825,
              "calibration_cpu_ns": 19067958.0,
              "calibration_gpu_ticks": 19067958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17635875.0,
              "compute_span_ns": 17635875.0,
              "command_buffer_ns": 17635875.032283366,
              "calibration_cpu_ns": 18876292.0,
              "calibration_gpu_ticks": 18876292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17440833.0,
              "compute_span_ns": 17440833.0,
              "command_buffer_ns": 17440833.325963467,
              "calibration_cpu_ns": 18493708.0,
              "calibration_gpu_ticks": 18493708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17347083.0,
              "compute_span_ns": 17347083.0,
              "command_buffer_ns": 17347041.69211909,
              "calibration_cpu_ns": 18683167.0,
              "calibration_gpu_ticks": 18683167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17339875.0,
              "compute_span_ns": 17339875.0,
              "command_buffer_ns": 17339874.990284443,
              "calibration_cpu_ns": 18588750.0,
              "calibration_gpu_ticks": 18588750.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 272958.0,
              "compute_span_ns": 272958.0,
              "command_buffer_ns": 272916.68811812997,
              "calibration_cpu_ns": 713250.0,
              "calibration_gpu_ticks": 713250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 271875.0,
              "compute_span_ns": 271875.0,
              "command_buffer_ns": 271875.00381842256,
              "calibration_cpu_ns": 715750.0,
              "calibration_gpu_ticks": 715750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 272625.0,
              "compute_span_ns": 272625.0,
              "command_buffer_ns": 272625.0095292926,
              "calibration_cpu_ns": 774750.0,
              "calibration_gpu_ticks": 774750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 271250.0,
              "compute_span_ns": 271250.0,
              "command_buffer_ns": 271249.96995553374,
              "calibration_cpu_ns": 775041.0,
              "calibration_gpu_ticks": 775041.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 275416.0,
              "compute_span_ns": 275416.0,
              "command_buffer_ns": 275374.97226148844,
              "calibration_cpu_ns": 840583.0,
              "calibration_gpu_ticks": 840583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 272584.0,
              "compute_span_ns": 272584.0,
              "command_buffer_ns": 272625.0095292926,
              "calibration_cpu_ns": 914209.0,
              "calibration_gpu_ticks": 914209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 271750.0,
              "compute_span_ns": 271750.0,
              "command_buffer_ns": 271750.03197044134,
              "calibration_cpu_ns": 726292.0,
              "calibration_gpu_ticks": 726292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 273541.0,
              "compute_span_ns": 273541.0,
              "command_buffer_ns": 273499.9870881438,
              "calibration_cpu_ns": 679000.0,
              "calibration_gpu_ticks": 679000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 309958.0,
              "compute_span_ns": 309958.0,
              "command_buffer_ns": 309958.35550129414,
              "calibration_cpu_ns": 911000.0,
              "calibration_gpu_ticks": 911000.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            269.5579375,
            270.061203125,
            272.108078125,
            269.86459375,
            275.290359375,
            275.560546875,
            272.513015625,
            271.048171875,
            270.935546875
          ],
          "compute_throughput_us_p50": 271.048171875,
          "compute_throughput_us_p90": 275.344396875,
          "command_buffer_throughput_us": [
            269.5572911761701,
            270.0618488233886,
            272.10872394789476,
            269.8652342587593,
            275.28971349966014,
            275.5605473794276,
            272.5130207181792,
            271.0475264393608,
            270.9355467231944
          ],
          "command_buffer_throughput_us_p50": 271.0475264393608,
          "command_buffer_throughput_us_p90": 275.3438802756136,
          "compute_latency_us": [
            272.958,
            271.875,
            272.625,
            271.25,
            275.416,
            272.584,
            271.75,
            273.541,
            309.958
          ],
          "compute_latency_us_p50": 272.625,
          "compute_latency_us_p90": 282.3244,
          "command_buffer_latency_us": [
            272.91668811812997,
            271.87500381842256,
            272.6250095292926,
            271.24996995553374,
            275.37497226148844,
            272.6250095292926,
            271.75003197044134,
            273.4999870881438,
            309.95835550129414
          ],
          "command_buffer_latency_us_p50": 272.6250095292926,
          "command_buffer_latency_us_p90": 282.2916489094496,
          "counter_control_throughput_ratio": 1.0020723624176695,
          "counter_control_latency_ratio": 0.9721975965630663
        },
        "download_ms": 0.824542,
        "allocation_upload_ms": 18.441458,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.046016576746524e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.2752141977003671,
      "device_compute_slowdown": 0.2785434876860854,
      "valid": true,
      "round": 2,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "layernorm_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.580542,
        "warmup_ms": 201.323,
        "repetitions": 107,
        "throughput_us": [
          292.3107476635514,
          343.63862616822433,
          313.7102803738318,
          308.5093457943925,
          314.5525700934579,
          301.86292523364483,
          304.62733644859816,
          305.05412149532714,
          313.8446261682243
        ],
        "latency_us": [
          1103.9170000000001,
          1464.875,
          544.0409999999999,
          579.75,
          587.584,
          1610.6670000000001,
          622.5,
          648.916,
          541.5
        ],
        "throughput_us_p50": 308.5093457943925,
        "throughput_us_p90": 320.3697813084112,
        "latency_us_p50": 622.5,
        "latency_us_p90": 1494.0334,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 18876833.375543356,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 19422125.013079494,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 19006874.99437481,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 18274541.65322706,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 19303000.008221716,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17346375.03977865,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17388333.33365619,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17210749.96748939,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17350083.333440125,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 281541.6664816439,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 272874.953225255,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 283708.3302438259,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 272916.62991046906,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 277291.692327708,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 272833.3347477019,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 284833.3097063005,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 274958.38003233075,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 278000.021353364,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              294.95052149286494,
              303.4707033293671,
              296.98242178710643,
              285.53971333167283,
              301.6093751284643,
              271.0371099965414,
              271.692708338378,
              268.91796824202174,
              271.09505208500195
            ],
            "command_buffer_throughput_us_p50": 285.53971333167283,
            "command_buffer_throughput_us_p90": 301.98164076864487,
            "command_buffer_latency_us": [
              281.5416664816439,
              272.874953225255,
              283.7083302438259,
              272.91662991046906,
              277.291692327708,
              272.8333347477019,
              284.8333097063005,
              274.95838003233075,
              278.000021353364
            ],
            "command_buffer_latency_us_p50": 277.291692327708,
            "command_buffer_latency_us_p90": 283.93332613632083
          },
          "throughput": [
            {
              "compute_ns": 19036375.0,
              "compute_span_ns": 19036375.0,
              "command_buffer_ns": 19036375.044379383,
              "calibration_cpu_ns": 20733875.0,
              "calibration_gpu_ticks": 20733875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 19037667.0,
              "compute_span_ns": 19037667.0,
              "command_buffer_ns": 19037666.672375053,
              "calibration_cpu_ns": 20272833.0,
              "calibration_gpu_ticks": 20272833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21647333.0,
              "compute_span_ns": 21647333.0,
              "command_buffer_ns": 21647333.342116326,
              "calibration_cpu_ns": 23218500.0,
              "calibration_gpu_ticks": 23218500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21324000.0,
              "compute_span_ns": 21324000.0,
              "command_buffer_ns": 21324000.030290335,
              "calibration_cpu_ns": 24339083.0,
              "calibration_gpu_ticks": 24339083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 19720791.0,
              "compute_span_ns": 19720791.0,
              "command_buffer_ns": 19720750.04596263,
              "calibration_cpu_ns": 20794458.0,
              "calibration_gpu_ticks": 20794458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17314542.0,
              "compute_span_ns": 17314542.0,
              "command_buffer_ns": 17314583.354163915,
              "calibration_cpu_ns": 22541166.0,
              "calibration_gpu_ticks": 22541166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17275375.0,
              "compute_span_ns": 17275375.0,
              "command_buffer_ns": 17275375.02301857,
              "calibration_cpu_ns": 19394875.0,
              "calibration_gpu_ticks": 19394875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17365417.0,
              "compute_span_ns": 17365417.0,
              "command_buffer_ns": 17365458.36320147,
              "calibration_cpu_ns": 18474625.0,
              "calibration_gpu_ticks": 18474625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17227042.0,
              "compute_span_ns": 17227042.0,
              "command_buffer_ns": 17227041.6514948,
              "calibration_cpu_ns": 18917625.0,
              "calibration_gpu_ticks": 18917625.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 273333.0,
              "compute_span_ns": 273333.0,
              "command_buffer_ns": 273333.33855494857,
              "calibration_cpu_ns": 1103167.0,
              "calibration_gpu_ticks": 1103167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 276666.0,
              "compute_span_ns": 276666.0,
              "command_buffer_ns": 276624.98177960515,
              "calibration_cpu_ns": 881417.0,
              "calibration_gpu_ticks": 881417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 273209.0,
              "compute_span_ns": 273209.0,
              "command_buffer_ns": 273250.0433921814,
              "calibration_cpu_ns": 803375.0,
              "calibration_gpu_ticks": 803375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 275333.0,
              "compute_span_ns": 275333.0,
              "command_buffer_ns": 275333.3537839353,
              "calibration_cpu_ns": 1282209.0,
              "calibration_gpu_ticks": 1282209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 273708.0,
              "compute_span_ns": 273708.0,
              "command_buffer_ns": 273666.693829,
              "calibration_cpu_ns": 836750.0,
              "calibration_gpu_ticks": 836750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 278417.0,
              "compute_span_ns": 278417.0,
              "command_buffer_ns": 278416.6717901826,
              "calibration_cpu_ns": 842791.0,
              "calibration_gpu_ticks": 842791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 273334.0,
              "compute_span_ns": 273334.0,
              "command_buffer_ns": 273375.0152401626,
              "calibration_cpu_ns": 1152125.0,
              "calibration_gpu_ticks": 1152125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 277708.0,
              "compute_span_ns": 277708.0,
              "command_buffer_ns": 277708.3427645266,
              "calibration_cpu_ns": 1132667.0,
              "calibration_gpu_ticks": 1132667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 275417.0,
              "compute_span_ns": 275417.0,
              "command_buffer_ns": 275416.7071543634,
              "calibration_cpu_ns": 954542.0,
              "calibration_gpu_ticks": 954542.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            297.443359375,
            297.463546875,
            338.239578125,
            333.1875,
            308.137359375,
            270.53971875,
            269.927734375,
            271.334640625,
            269.17253125
          ],
          "compute_throughput_us_p50": 297.443359375,
          "compute_throughput_us_p90": 334.197915625,
          "command_buffer_throughput_us": [
            297.44336006842786,
            297.4635417558602,
            338.2395834705676,
            333.1875004732865,
            308.1367194681661,
            270.54036490881117,
            269.92773473466514,
            271.33528692502296,
            269.17252580460627
          ],
          "command_buffer_throughput_us_p50": 297.44336006842786,
          "command_buffer_throughput_us_p90": 334.1979170727427,
          "compute_latency_us": [
            273.333,
            276.666,
            273.209,
            275.333,
            273.708,
            278.417,
            273.334,
            277.708,
            275.417
          ],
          "compute_latency_us_p50": 275.333,
          "compute_latency_us_p90": 277.8498,
          "command_buffer_latency_us": [
            273.33333855494857,
            276.62498177960515,
            273.2500433921814,
            275.3333537839353,
            273.666693829,
            278.4166717901826,
            273.3750152401626,
            277.7083427645266,
            275.4167071543634
          ],
          "command_buffer_latency_us_p50": 275.3333537839353,
          "command_buffer_latency_us_p90": 277.8500085696578,
          "counter_control_throughput_ratio": 1.008451717809976,
          "counter_control_latency_ratio": 0.9907075035950557
        },
        "download_ms": 1.572875,
        "allocation_upload_ms": 1.228208,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.046016576746524e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-0h981zxf/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "2f415c8199d32d1b999ce6dae6674c9bc8bbdf9b0adda641a6d2aee649468028",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 24.174,
        "capture_ms": 0.099083,
        "compile_ms": 92.074042,
        "allocation_upload_ms": 7.653125,
        "cold_call_ms": 8.604333,
        "warmup_ms": 201.008458,
        "download_ms": 6.641916,
        "repetitions": 412,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          72.1553398058,
          72.8881456311,
          72.1030533981,
          72.1999393204,
          72.0639174757,
          72.256776699,
          71.7942961165,
          72.1989296117,
          72.9864490291
        ],
        "latency_us": [
          384.208,
          380.209,
          330.167,
          335.041,
          326.916,
          284.958,
          292.667,
          300.625,
          291.541
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 384,
            "reduction_scalar_rounds": 120,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 73728,
              "global_write_bytes": 24576,
              "private_read_bytes": 147456,
              "private_write_bytes": 49152
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 288,
              "global_write_bytes": 96,
              "private_read_bytes": 576,
              "private_write_bytes": 192
            },
            "striped_storage_scalars_per_worker": 48,
            "reduction_operations": 2,
            "reduction_elements": 12288,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 160,
            "normalized_kernel_cost": 1920,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 18432,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4603958,
              "compute_span_ns": 4603958,
              "calibration_cpu_ns": 5187875,
              "calibration_gpu_ticks": 5187875,
              "compute_passes": 1,
              "command_buffer_ns": 4603958.33943,
              "command_buffers": 2
            },
            {
              "compute_ns": 4641000,
              "compute_span_ns": 4641000,
              "calibration_cpu_ns": 5466500,
              "calibration_gpu_ticks": 5466500,
              "compute_passes": 1,
              "command_buffer_ns": 4641041.6835,
              "command_buffers": 2
            },
            {
              "compute_ns": 4690625,
              "compute_span_ns": 4690625,
              "calibration_cpu_ns": 5460958,
              "calibration_gpu_ticks": 5460958,
              "compute_passes": 1,
              "command_buffer_ns": 4690708.3015,
              "command_buffers": 2
            },
            {
              "compute_ns": 4661000,
              "compute_span_ns": 4661000,
              "calibration_cpu_ns": 5579125,
              "calibration_gpu_ticks": 5579125,
              "compute_passes": 1,
              "command_buffer_ns": 4661083.33785,
              "command_buffers": 2
            },
            {
              "compute_ns": 4687208,
              "compute_span_ns": 4687208,
              "calibration_cpu_ns": 5179083,
              "calibration_gpu_ticks": 5179083,
              "compute_passes": 1,
              "command_buffer_ns": 4687166.65637,
              "command_buffers": 2
            },
            {
              "compute_ns": 4641791,
              "compute_span_ns": 4641791,
              "calibration_cpu_ns": 5181667,
              "calibration_gpu_ticks": 5181667,
              "compute_passes": 1,
              "command_buffer_ns": 4641791.68921,
              "command_buffers": 2
            },
            {
              "compute_ns": 4765792,
              "compute_span_ns": 4765792,
              "calibration_cpu_ns": 5790250,
              "calibration_gpu_ticks": 5790250,
              "compute_passes": 1,
              "command_buffer_ns": 4765875.05545,
              "command_buffers": 2
            },
            {
              "compute_ns": 4880500,
              "compute_span_ns": 4880500,
              "calibration_cpu_ns": 5745959,
              "calibration_gpu_ticks": 5745959,
              "compute_passes": 1,
              "command_buffer_ns": 4880499.96784,
              "command_buffers": 2
            },
            {
              "compute_ns": 4973209,
              "compute_span_ns": 4973209,
              "calibration_cpu_ns": 5850583,
              "calibration_gpu_ticks": 5850583,
              "compute_passes": 1,
              "command_buffer_ns": 4973291.65228,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 84291,
              "compute_span_ns": 84291,
              "calibration_cpu_ns": 600333,
              "calibration_gpu_ticks": 600333,
              "compute_passes": 1,
              "command_buffer_ns": 84250.0012368,
              "command_buffers": 2
            },
            {
              "compute_ns": 73875,
              "compute_span_ns": 73875,
              "calibration_cpu_ns": 547125,
              "calibration_gpu_ticks": 547125,
              "compute_passes": 1,
              "command_buffer_ns": 73999.9813959,
              "command_buffers": 2
            },
            {
              "compute_ns": 183750,
              "compute_span_ns": 183750,
              "calibration_cpu_ns": 1063584,
              "calibration_gpu_ticks": 1063584,
              "compute_passes": 1,
              "command_buffer_ns": 183833.35555,
              "command_buffers": 2
            },
            {
              "compute_ns": 73167,
              "compute_span_ns": 73167,
              "calibration_cpu_ns": 568458,
              "calibration_gpu_ticks": 568458,
              "compute_passes": 1,
              "command_buffer_ns": 73208.2989998,
              "command_buffers": 2
            },
            {
              "compute_ns": 94042,
              "compute_span_ns": 94042,
              "calibration_cpu_ns": 610041,
              "calibration_gpu_ticks": 610041,
              "compute_passes": 1,
              "command_buffer_ns": 94166.7240113,
              "command_buffers": 2
            },
            {
              "compute_ns": 78459,
              "compute_span_ns": 78459,
              "calibration_cpu_ns": 641625,
              "calibration_gpu_ticks": 641625,
              "compute_passes": 1,
              "command_buffer_ns": 78541.6923463,
              "command_buffers": 2
            },
            {
              "compute_ns": 86083,
              "compute_span_ns": 86083,
              "calibration_cpu_ns": 609625,
              "calibration_gpu_ticks": 609625,
              "compute_passes": 1,
              "command_buffer_ns": 86166.6630954,
              "command_buffers": 2
            },
            {
              "compute_ns": 104875,
              "compute_span_ns": 104875,
              "calibration_cpu_ns": 713667,
              "calibration_gpu_ticks": 713667,
              "compute_passes": 1,
              "command_buffer_ns": 104958.366137,
              "command_buffers": 2
            },
            {
              "compute_ns": 124375,
              "compute_span_ns": 124375,
              "calibration_cpu_ns": 678625,
              "calibration_gpu_ticks": 678625,
              "compute_passes": 1,
              "command_buffer_ns": 124500.016682,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4748166.65519,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4613833.2691,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4640458.32632,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4641500.01062,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4623916.64041,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4666541.64484,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4624208.37721,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 5079291.5863,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4909458.27914,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 243708.316702,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 73583.3309591,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 85291.6273288,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 86083.3097249,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 124333.309941,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 105625.018477,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 133416.673634,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 158625.014592,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 94083.3706409,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              74.19010398734376,
              72.09114482968751,
              72.50716134875,
              72.5234376659375,
              72.24869750640626,
              72.914713200625,
              72.25325589390624,
              79.3639310359375,
              76.7102856115625
            ],
            "command_buffer_throughput_us_p50": 72.5234376659375,
            "command_buffer_throughput_us_p90": 77.2410146964375,
            "command_buffer_latency_us": [
              243.70831670200002,
              73.5833309591,
              85.29162732879999,
              86.0833097249,
              124.333309941,
              105.62501847700001,
              133.416673634,
              158.62501459199999,
              94.0833706409
            ],
            "command_buffer_latency_us_p50": 105.62501847700001,
            "command_buffer_latency_us_p90": 175.64167501400001
          },
          "compute_throughput_us": [
            71.93684375,
            72.515625,
            73.291015625,
            72.828125,
            73.237625,
            72.527984375,
            74.4655,
            76.2578125,
            77.706390625
          ],
          "compute_throughput_us_p50": 73.237625,
          "compute_throughput_us_p90": 76.547528125,
          "command_buffer_throughput_us": [
            71.93684905359375,
            72.51627630468751,
            73.2923172109375,
            72.82942715390625,
            73.23697900578125,
            72.52799514390625,
            74.46679774140625,
            76.2578119975,
            77.707682066875
          ],
          "command_buffer_throughput_us_p50": 73.23697900578125,
          "command_buffer_throughput_us_p90": 76.547786011375,
          "compute_latency_us": [
            84.291,
            73.875,
            183.75,
            73.167,
            94.042,
            78.459,
            86.083,
            104.875,
            124.375
          ],
          "compute_latency_us_p50": 86.083,
          "compute_latency_us_p90": 136.25,
          "command_buffer_latency_us": [
            84.2500012368,
            73.99998139590001,
            183.83335555000002,
            73.2082989998,
            94.1667240113,
            78.5416923463,
            86.1666630954,
            104.958366137,
            124.500016682
          ],
          "command_buffer_latency_us_p50": 86.1666630954,
          "command_buffer_latency_us_p90": 136.36668445560002,
          "counter_control_throughput_ratio": 1.0058971386292221,
          "counter_control_latency_ratio": 0.7573732578661746
        },
        "correctness": {
          "max_abs_error": 2.7199924934606656e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 72.1989296117,
        "throughput_us_p90": 72.9078063107,
        "latency_us_p50": 326.916,
        "latency_us_p90": 381.0088
      },
      "slowdown": 0.234025097119156,
      "device_compute_slowdown": 0.24622376896861928,
      "valid": true,
      "round": 2,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "layernorm_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-nqtx6xpj/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "7d02ca31ce0660bb7b8cf2d4b7c68e2de8afe58cfb625e16c6aa93173ce8f234",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 28.719792,
        "capture_ms": 0.070666,
        "compile_ms": 63.486916,
        "allocation_upload_ms": 6.097458,
        "cold_call_ms": 7.724958,
        "warmup_ms": 200.068292,
        "download_ms": 3.268417,
        "repetitions": 346,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          76.5275780347,
          75.9432803468,
          76.1164508671,
          76.1728092486,
          76.4517109827,
          77.1145231214,
          76.7337427746,
          76.7862485549,
          76.8697023121
        ],
        "latency_us": [
          324.5,
          293.625,
          286.75,
          306.917,
          294.75,
          271.75,
          279.042,
          297.042,
          277.042
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 512,
            "reduction_scalar_rounds": 80,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 49152,
              "global_write_bytes": 16384,
              "private_read_bytes": 98304,
              "private_write_bytes": 32768
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 192,
              "global_write_bytes": 64,
              "private_read_bytes": 384,
              "private_write_bytes": 128
            },
            "striped_storage_scalars_per_worker": 32,
            "reduction_operations": 2,
            "reduction_elements": 8192,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 120,
            "normalized_kernel_cost": 1920,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4826833,
              "compute_span_ns": 4826833,
              "calibration_cpu_ns": 5470958,
              "calibration_gpu_ticks": 5470958,
              "compute_passes": 1,
              "command_buffer_ns": 4826833.31938,
              "command_buffers": 2
            },
            {
              "compute_ns": 4815583,
              "compute_span_ns": 4815583,
              "calibration_cpu_ns": 5288416,
              "calibration_gpu_ticks": 5288416,
              "compute_passes": 1,
              "command_buffer_ns": 4815625.02682,
              "command_buffers": 2
            },
            {
              "compute_ns": 4848083,
              "compute_span_ns": 4848083,
              "calibration_cpu_ns": 5318167,
              "calibration_gpu_ticks": 5318167,
              "compute_passes": 1,
              "command_buffer_ns": 4848083.30657,
              "command_buffers": 2
            },
            {
              "compute_ns": 4871417,
              "compute_span_ns": 4871417,
              "calibration_cpu_ns": 5368834,
              "calibration_gpu_ticks": 5368834,
              "compute_passes": 1,
              "command_buffer_ns": 4871458.33904,
              "command_buffers": 2
            },
            {
              "compute_ns": 4897875,
              "compute_span_ns": 4897875,
              "calibration_cpu_ns": 5393416,
              "calibration_gpu_ticks": 5393416,
              "compute_passes": 1,
              "command_buffer_ns": 4897916.68952,
              "command_buffers": 2
            },
            {
              "compute_ns": 4829959,
              "compute_span_ns": 4829959,
              "calibration_cpu_ns": 5341167,
              "calibration_gpu_ticks": 5341167,
              "compute_passes": 1,
              "command_buffer_ns": 4830083.28592,
              "command_buffers": 2
            },
            {
              "compute_ns": 4841792,
              "compute_span_ns": 4841792,
              "calibration_cpu_ns": 5380875,
              "calibration_gpu_ticks": 5380875,
              "compute_passes": 1,
              "command_buffer_ns": 4841791.6405,
              "command_buffers": 2
            },
            {
              "compute_ns": 4846417,
              "compute_span_ns": 4846417,
              "calibration_cpu_ns": 5365875,
              "calibration_gpu_ticks": 5365875,
              "compute_passes": 1,
              "command_buffer_ns": 4846541.67667,
              "command_buffers": 2
            },
            {
              "compute_ns": 4858625,
              "compute_span_ns": 4858625,
              "calibration_cpu_ns": 5298083,
              "calibration_gpu_ticks": 5298083,
              "compute_passes": 1,
              "command_buffer_ns": 4858666.68168,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 75875,
              "compute_span_ns": 75875,
              "calibration_cpu_ns": 473792,
              "calibration_gpu_ticks": 473792,
              "compute_passes": 1,
              "command_buffer_ns": 75916.7014621,
              "command_buffers": 2
            },
            {
              "compute_ns": 76584,
              "compute_span_ns": 76584,
              "calibration_cpu_ns": 497292,
              "calibration_gpu_ticks": 497292,
              "compute_passes": 1,
              "command_buffer_ns": 76708.3256505,
              "command_buffers": 2
            },
            {
              "compute_ns": 76500,
              "compute_span_ns": 76500,
              "calibration_cpu_ns": 453250,
              "calibration_gpu_ticks": 453250,
              "compute_passes": 1,
              "command_buffer_ns": 76583.3538026,
              "command_buffers": 2
            },
            {
              "compute_ns": 77875,
              "compute_span_ns": 77875,
              "calibration_cpu_ns": 457709,
              "calibration_gpu_ticks": 457709,
              "compute_passes": 1,
              "command_buffer_ns": 77958.3351687,
              "command_buffers": 2
            },
            {
              "compute_ns": 83959,
              "compute_span_ns": 83959,
              "calibration_cpu_ns": 471917,
              "calibration_gpu_ticks": 471917,
              "compute_passes": 1,
              "command_buffer_ns": 83999.9993332,
              "command_buffers": 2
            },
            {
              "compute_ns": 76791,
              "compute_span_ns": 76791,
              "calibration_cpu_ns": 430625,
              "calibration_gpu_ticks": 430625,
              "compute_passes": 1,
              "command_buffer_ns": 76791.679021,
              "command_buffers": 2
            },
            {
              "compute_ns": 77916,
              "compute_span_ns": 77916,
              "calibration_cpu_ns": 446417,
              "calibration_gpu_ticks": 446417,
              "compute_passes": 1,
              "command_buffer_ns": 77958.3351687,
              "command_buffers": 2
            },
            {
              "compute_ns": 75625,
              "compute_span_ns": 75625,
              "calibration_cpu_ns": 453375,
              "calibration_gpu_ticks": 453375,
              "compute_passes": 1,
              "command_buffer_ns": 75625.0228733,
              "command_buffers": 2
            },
            {
              "compute_ns": 90042,
              "compute_span_ns": 90042,
              "calibration_cpu_ns": 504917,
              "calibration_gpu_ticks": 504917,
              "compute_passes": 1,
              "command_buffer_ns": 90208.3702385,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4923833.3595,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4867208.36489,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4843750.03725,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4900083.35328,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4843791.65573,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4880708.35127,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4849124.99087,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4844249.98285,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4877208.26641,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 77958.3933763,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76208.380051,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76708.3256505,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 77833.305113,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 83499.9955259,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 77874.9817982,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76750.0023358,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 77250.0643507,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 77249.9479353,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              76.9348962421875,
              76.05013070140625,
              75.68359433203125,
              76.56380239500001,
              75.68424462078124,
              76.26106798859375,
              75.76757798234375,
              75.69140598203126,
              76.20637916265625
            ],
            "command_buffer_throughput_us_p50": 76.05013070140625,
            "command_buffer_throughput_us_p90": 76.6380211644375,
            "command_buffer_latency_us": [
              77.9583933763,
              76.208380051,
              76.7083256505,
              77.833305113,
              83.4999955259,
              77.8749817982,
              76.7500023358,
              77.2500643507,
              77.2499479353
            ],
            "command_buffer_latency_us_p50": 77.2500643507,
            "command_buffer_latency_us_p90": 79.06671380622
          },
          "compute_throughput_us": [
            75.419265625,
            75.243484375,
            75.751296875,
            76.115890625,
            76.529296875,
            75.468109375,
            75.653,
            75.725265625,
            75.916015625
          ],
          "compute_throughput_us_p50": 75.725265625,
          "compute_throughput_us_p90": 76.198571875,
          "command_buffer_throughput_us": [
            75.4192706153125,
            75.2441410440625,
            75.75130166515625,
            76.1165365475,
            76.52994827375,
            75.4700513425,
            75.6529943828125,
            75.72721369796875,
            75.91666690125001
          ],
          "command_buffer_throughput_us_p50": 75.72721369796875,
          "command_buffer_throughput_us_p90": 76.19921889275,
          "compute_latency_us": [
            75.875,
            76.584,
            76.5,
            77.875,
            83.959,
            76.791,
            77.916,
            75.625,
            90.042
          ],
          "compute_latency_us_p50": 76.791,
          "compute_latency_us_p90": 85.1756,
          "command_buffer_latency_us": [
            75.9167014621,
            76.7083256505,
            76.5833538026,
            77.95833516869999,
            83.9999993332,
            76.79167902100001,
            77.95833516869999,
            75.6250228733,
            90.20837023850001
          ],
          "command_buffer_latency_us_p50": 76.79167902100001,
          "command_buffer_latency_us_p90": 85.24167351426,
          "counter_control_throughput_ratio": 0.9961983200804243,
          "counter_control_latency_ratio": 1.0016063824543808
        },
        "correctness": {
          "max_abs_error": 2.827305296548843e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 76.5275780347,
        "throughput_us_p90": 76.91866647396,
        "latency_us_p50": 293.625,
        "latency_us_p90": 310.43359999999996
      },
      "torch": {
        "cold_call_ms": 6.961834,
        "warmup_ms": 200.727375,
        "repetitions": 143,
        "throughput_us": [
          208.99388111888112,
          205.15239160839158,
          206.05506993006992,
          205.66753846153844,
          204.58041958041957,
          204.93094405594405,
          204.78904195804196,
          206.36072027972028,
          207.05565734265733
        ],
        "latency_us": [
          599.7909999999999,
          466.04200000000003,
          466.667,
          456.875,
          442.125,
          446.20799999999997,
          453.542,
          462.541,
          447.375
        ],
        "throughput_us_p50": 205.66753846153844,
        "throughput_us_p90": 207.4433020979021,
        "latency_us_p50": 456.875,
        "latency_us_p90": 493.29179999999997,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 12826874.968595803,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12782083.300407976,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12963666.69703275,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12928500.01482293,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12893583.334516734,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12869958.358351141,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12958916.660863906,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 13095083.297230303,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 13199749.984778464,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 200249.9531954527,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 202249.96842443943,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200583.30846950412,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 199750.00759586692,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200500.01330673695,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 201541.63939878345,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 238749.95531514287,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 201291.6374951601,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 199541.68237745762,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              200.41992138430942,
              199.72005156887462,
              202.5572921411367,
              202.00781273160828,
              201.46223960182397,
              201.09309934923658,
              202.48307282599853,
              204.6106765192235,
              206.2460935121635
            ],
            "command_buffer_throughput_us_p50": 202.00781273160828,
            "command_buffer_throughput_us_p90": 204.9377599178115,
            "command_buffer_latency_us": [
              200.2499531954527,
              202.24996842443943,
              200.58330846950412,
              199.75000759586692,
              200.50001330673695,
              201.54163939878345,
              238.74995531514287,
              201.2916374951601,
              199.54168237745762
            ],
            "command_buffer_latency_us_p50": 200.58330846950412,
            "command_buffer_latency_us_p90": 209.54996580258012
          },
          "throughput": [
            {
              "compute_ns": 12919875.0,
              "compute_span_ns": 12919875.0,
              "command_buffer_ns": 12919875.036459416,
              "calibration_cpu_ns": 13759333.0,
              "calibration_gpu_ticks": 13759333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12866667.0,
              "compute_span_ns": 12866667.0,
              "command_buffer_ns": 12866666.656918824,
              "calibration_cpu_ns": 13742042.0,
              "calibration_gpu_ticks": 13742042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12885833.0,
              "compute_span_ns": 12885833.0,
              "command_buffer_ns": 12885833.333712071,
              "calibration_cpu_ns": 14035416.0,
              "calibration_gpu_ticks": 14035416.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12894708.0,
              "compute_span_ns": 12894708.0,
              "command_buffer_ns": 12894666.637293994,
              "calibration_cpu_ns": 14023500.0,
              "calibration_gpu_ticks": 14023500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12965250.0,
              "compute_span_ns": 12965250.0,
              "command_buffer_ns": 12965250.003617257,
              "calibration_cpu_ns": 13865292.0,
              "calibration_gpu_ticks": 13865292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12979625.0,
              "compute_span_ns": 12979625.0,
              "command_buffer_ns": 12979625.025764108,
              "calibration_cpu_ns": 13753709.0,
              "calibration_gpu_ticks": 13753709.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 13010417.0,
              "compute_span_ns": 13010417.0,
              "command_buffer_ns": 13010458.322241902,
              "calibration_cpu_ns": 14960125.0,
              "calibration_gpu_ticks": 14960125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12970583.0,
              "compute_span_ns": 12970583.0,
              "command_buffer_ns": 12970583.338756114,
              "calibration_cpu_ns": 15067708.0,
              "calibration_gpu_ticks": 15067708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 13159208.0,
              "compute_span_ns": 13159208.0,
              "command_buffer_ns": 13159208.348952234,
              "calibration_cpu_ns": 14408000.0,
              "calibration_gpu_ticks": 14408000.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 204208.0,
              "compute_span_ns": 204208.0,
              "command_buffer_ns": 204166.688490659,
              "calibration_cpu_ns": 805750.0,
              "calibration_gpu_ticks": 805750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 200500.0,
              "compute_span_ns": 200500.0,
              "command_buffer_ns": 200500.01330673695,
              "calibration_cpu_ns": 755333.0,
              "calibration_gpu_ticks": 755333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201792.0,
              "compute_span_ns": 201792.0,
              "command_buffer_ns": 201833.37619528174,
              "calibration_cpu_ns": 736292.0,
              "calibration_gpu_ticks": 736292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 200500.0,
              "compute_span_ns": 200500.0,
              "command_buffer_ns": 200500.01330673695,
              "calibration_cpu_ns": 637125.0,
              "calibration_gpu_ticks": 637125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 202750.0,
              "compute_span_ns": 202750.0,
              "command_buffer_ns": 202750.03043934703,
              "calibration_cpu_ns": 808250.0,
              "calibration_gpu_ticks": 808250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 202208.0,
              "compute_span_ns": 202208.0,
              "command_buffer_ns": 202166.67326167226,
              "calibration_cpu_ns": 755667.0,
              "calibration_gpu_ticks": 755667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 200833.0,
              "compute_span_ns": 200833.0,
              "command_buffer_ns": 200833.31037312746,
              "calibration_cpu_ns": 881708.0,
              "calibration_gpu_ticks": 881708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 200875.0,
              "compute_span_ns": 200875.0,
              "command_buffer_ns": 200875.04526600242,
              "calibration_cpu_ns": 972416.0,
              "calibration_gpu_ticks": 972416.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 238209.0,
              "compute_span_ns": 238209.0,
              "command_buffer_ns": 238249.95150789618,
              "calibration_cpu_ns": 871417.0,
              "calibration_gpu_ticks": 871417.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            201.873046875,
            201.041671875,
            201.341140625,
            201.4798125,
            202.58203125,
            202.806640625,
            203.287765625,
            202.665359375,
            205.612625
          ],
          "compute_throughput_us_p50": 202.58203125,
          "compute_throughput_us_p90": 203.7527375,
          "command_buffer_throughput_us": [
            201.87304744467838,
            201.04166651435662,
            201.3411458392511,
            201.47916620771866,
            202.58203130651964,
            202.80664102756418,
            203.28841128502972,
            202.66536466806429,
            205.61263045237865
          ],
          "command_buffer_throughput_us_p50": 202.58203130651964,
          "command_buffer_throughput_us_p90": 203.7532551184995,
          "compute_latency_us": [
            204.208,
            200.5,
            201.792,
            200.5,
            202.75,
            202.208,
            200.833,
            200.875,
            238.209
          ],
          "compute_latency_us_p50": 201.792,
          "compute_latency_us_p90": 211.00820000000002,
          "command_buffer_latency_us": [
            204.166688490659,
            200.50001330673695,
            201.83337619528174,
            200.50001330673695,
            202.75003043934703,
            202.16667326167226,
            200.83331037312746,
            200.87504526600242,
            238.24995150789618
          ],
          "command_buffer_latency_us_p50": 201.83337619528174,
          "command_buffer_latency_us_p90": 210.98334109410644,
          "counter_control_throughput_ratio": 1.00397731251206,
          "counter_control_latency_ratio": 1.003754721814016
        },
        "download_ms": 1.02575,
        "allocation_upload_ms": 16.877208,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.1634934849478213e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.3720936157798733,
      "device_compute_slowdown": 0.3738005051965832,
      "valid": true,
      "round": 2,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "layernorm_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.52175,
        "warmup_ms": 201.355,
        "repetitions": 142,
        "throughput_us": [
          210.97241549295774,
          213.58128169014083,
          213.89935915492958,
          211.152,
          212.60064084507044,
          211.5314014084507,
          212.23973239436617,
          214.64260563380282,
          213.11326056338027
        ],
        "latency_us": [
          502.45799999999997,
          415.95799999999997,
          472.04200000000003,
          486.584,
          535.75,
          558.5,
          600.125,
          491.70799999999997,
          417.5
        ],
        "throughput_us_p50": 212.60064084507044,
        "throughput_us_p90": 214.04800845070423,
        "latency_us_p50": 491.70799999999997,
        "latency_us_p90": 566.825,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 13414458.313491195,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 13362166.70460999,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 13298375.008162111,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 13136083.318386227,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 13188874.98928234,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 13443375.006318092,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 13572333.380579948,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 13164458.330720663,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 13351458.2994394,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 200791.69189557433,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200208.33471789956,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200999.95890632272,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 201333.37238803506,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200374.98325109482,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 201458.34423601627,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 264833.33203941584,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 201500.02092123032,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 201375.0490732491,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              209.60091114829993,
              208.7838547595311,
              207.78710950253299,
              205.2513018497848,
              206.07617170753656,
              210.0527344737202,
              212.0677090715617,
              205.69466141751036,
              208.61653592874063
            ],
            "command_buffer_throughput_us_p50": 208.61653592874063,
            "command_buffer_throughput_us_p90": 210.4557293932885,
            "command_buffer_latency_us": [
              200.79169189557433,
              200.20833471789956,
              200.99995890632272,
              201.33337238803506,
              200.37498325109482,
              201.45834423601627,
              264.83333203941584,
              201.50002092123032,
              201.3750490732491
            ],
            "command_buffer_latency_us_p50": 201.33337238803506,
            "command_buffer_latency_us_p90": 214.16668314486742
          },
          "throughput": [
            {
              "compute_ns": 13328709.0,
              "compute_span_ns": 13328709.0,
              "command_buffer_ns": 13328749.977517873,
              "calibration_cpu_ns": 14195917.0,
              "calibration_gpu_ticks": 14195917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 13207916.0,
              "compute_span_ns": 13207916.0,
              "command_buffer_ns": 13207875.017542392,
              "calibration_cpu_ns": 14015583.0,
              "calibration_gpu_ticks": 14015583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 13260500.0,
              "compute_span_ns": 13260500.0,
              "command_buffer_ns": 13260500.03990531,
              "calibration_cpu_ns": 14188875.0,
              "calibration_gpu_ticks": 14188875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 13158916.0,
              "compute_span_ns": 13158916.0,
              "command_buffer_ns": 13158874.993678182,
              "calibration_cpu_ns": 14040375.0,
              "calibration_gpu_ticks": 14040375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 13618541.0,
              "compute_span_ns": 13618541.0,
              "command_buffer_ns": 13618500.030133873,
              "calibration_cpu_ns": 14688792.0,
              "calibration_gpu_ticks": 14688792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 13544750.0,
              "compute_span_ns": 13544750.0,
              "command_buffer_ns": 13544749.992433935,
              "calibration_cpu_ns": 14674792.0,
              "calibration_gpu_ticks": 14674792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 13481375.0,
              "compute_span_ns": 13481375.0,
              "command_buffer_ns": 13481375.004630536,
              "calibration_cpu_ns": 14515333.0,
              "calibration_gpu_ticks": 14515333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 13491666.0,
              "compute_span_ns": 13491666.0,
              "command_buffer_ns": 13491625.024471432,
              "calibration_cpu_ns": 14503208.0,
              "calibration_gpu_ticks": 14503208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 13198250.0,
              "compute_span_ns": 13198250.0,
              "command_buffer_ns": 13198250.031564385,
              "calibration_cpu_ns": 14118333.0,
              "calibration_gpu_ticks": 14118333.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 200792.0,
              "compute_span_ns": 200792.0,
              "command_buffer_ns": 200833.31037312746,
              "calibration_cpu_ns": 615250.0,
              "calibration_gpu_ticks": 615250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201500.0,
              "compute_span_ns": 201500.0,
              "command_buffer_ns": 201500.02092123032,
              "calibration_cpu_ns": 620417.0,
              "calibration_gpu_ticks": 620417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 264584.0,
              "compute_span_ns": 264584.0,
              "command_buffer_ns": 264625.00682100654,
              "calibration_cpu_ns": 806125.0,
              "calibration_gpu_ticks": 806125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201291.0,
              "compute_span_ns": 201291.0,
              "command_buffer_ns": 201250.01901760697,
              "calibration_cpu_ns": 640750.0,
              "calibration_gpu_ticks": 640750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201417.0,
              "compute_span_ns": 201417.0,
              "command_buffer_ns": 201458.34423601627,
              "calibration_cpu_ns": 639125.0,
              "calibration_gpu_ticks": 639125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 199500.0,
              "compute_span_ns": 199500.0,
              "command_buffer_ns": 199500.00569224358,
              "calibration_cpu_ns": 625833.0,
              "calibration_gpu_ticks": 625833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201958.0,
              "compute_span_ns": 201958.0,
              "command_buffer_ns": 201916.67135804892,
              "calibration_cpu_ns": 642834.0,
              "calibration_gpu_ticks": 642834.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 199917.0,
              "compute_span_ns": 199917.0,
              "command_buffer_ns": 199958.33281427622,
              "calibration_cpu_ns": 614209.0,
              "calibration_gpu_ticks": 614209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 199917.0,
              "compute_span_ns": 199917.0,
              "command_buffer_ns": 199958.33281427622,
              "calibration_cpu_ns": 602084.0,
              "calibration_gpu_ticks": 602084.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            208.261078125,
            206.3736875,
            207.1953125,
            205.6080625,
            212.789703125,
            211.63671875,
            210.646484375,
            210.80728125,
            206.22265625
          ],
          "compute_throughput_us_p50": 208.261078125,
          "compute_throughput_us_p90": 211.867315625,
          "command_buffer_throughput_us": [
            208.26171839871677,
            206.37304714909988,
            207.19531312352046,
            205.6074217762216,
            212.78906297084177,
            211.63671863178024,
            210.64648444735212,
            210.80664100736612,
            206.2226567431935
          ],
          "command_buffer_throughput_us_p50": 208.26171839871677,
          "command_buffer_throughput_us_p90": 211.86718749959255,
          "compute_latency_us": [
            200.792,
            201.5,
            264.584,
            201.291,
            201.417,
            199.5,
            201.958,
            199.917,
            199.917
          ],
          "compute_latency_us_p50": 201.291,
          "compute_latency_us_p90": 214.4832,
          "command_buffer_latency_us": [
            200.83331037312746,
            201.50002092123032,
            264.62500682100654,
            201.25001901760697,
            201.45834423601627,
            199.50000569224358,
            201.91667135804892,
            199.95833281427622,
            199.95833281427622
          ],
          "command_buffer_latency_us_p50": 201.25001901760697,
          "command_buffer_latency_us_p90": 214.45833845064044,
          "counter_control_throughput_ratio": 0.997151910046637,
          "counter_control_latency_ratio": 0.9995859932735471
        },
        "download_ms": 0.623292,
        "allocation_upload_ms": 1.505625,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.1634934849478213e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-vb4ocm41/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "6caff4b0a979c3104fea9f23f0d5430392f6f4e7d743dab03e9fdb6de1428bf3",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 18.162875,
        "capture_ms": 0.078834,
        "compile_ms": 58.465959,
        "allocation_upload_ms": 5.82275,
        "cold_call_ms": 5.007292,
        "warmup_ms": 201.097459,
        "download_ms": 3.583083,
        "repetitions": 454,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          61.8627026432,
          62.4648502203,
          61.5594713656,
          62.7985484581,
          62.7799185022,
          63.0672731278,
          62.016154185,
          62.1504207048,
          62.158222467
        ],
        "latency_us": [
          351.958,
          291.25,
          304.375,
          400.959,
          302.375,
          298.917,
          286.083,
          300,
          308.834
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 1024,
            "reduction_scalar_rounds": 80,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 49152,
              "global_write_bytes": 16384,
              "private_read_bytes": 98304,
              "private_write_bytes": 32768
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 192,
              "global_write_bytes": 64,
              "private_read_bytes": 384,
              "private_write_bytes": 128
            },
            "striped_storage_scalars_per_worker": 32,
            "reduction_operations": 2,
            "reduction_elements": 8192,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 128,
            "normalized_kernel_cost": 2048,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3939042,
              "compute_span_ns": 3939042,
              "calibration_cpu_ns": 4454458,
              "calibration_gpu_ticks": 4454458,
              "compute_passes": 1,
              "command_buffer_ns": 3939166.72328,
              "command_buffers": 2
            },
            {
              "compute_ns": 3964917,
              "compute_span_ns": 3964917,
              "calibration_cpu_ns": 4458084,
              "calibration_gpu_ticks": 4458084,
              "compute_passes": 1,
              "command_buffer_ns": 3965083.33506,
              "command_buffers": 2
            },
            {
              "compute_ns": 3927625,
              "compute_span_ns": 3927625,
              "calibration_cpu_ns": 4442375,
              "calibration_gpu_ticks": 4442375,
              "compute_passes": 1,
              "command_buffer_ns": 3927708.37061,
              "command_buffers": 2
            },
            {
              "compute_ns": 4018125,
              "compute_span_ns": 4018125,
              "calibration_cpu_ns": 4456041,
              "calibration_gpu_ticks": 4456041,
              "compute_passes": 1,
              "command_buffer_ns": 4018249.97971,
              "command_buffers": 2
            },
            {
              "compute_ns": 3965958,
              "compute_span_ns": 3965958,
              "calibration_cpu_ns": 4439125,
              "calibration_gpu_ticks": 4439125,
              "compute_passes": 1,
              "command_buffer_ns": 3965958.31262,
              "command_buffers": 2
            },
            {
              "compute_ns": 3994541,
              "compute_span_ns": 3994541,
              "calibration_cpu_ns": 4475334,
              "calibration_gpu_ticks": 4475334,
              "compute_passes": 1,
              "command_buffer_ns": 3994500.03169,
              "command_buffers": 2
            },
            {
              "compute_ns": 3971750,
              "compute_span_ns": 3971750,
              "calibration_cpu_ns": 4426250,
              "calibration_gpu_ticks": 4426250,
              "compute_passes": 1,
              "command_buffer_ns": 3971750.03309,
              "command_buffers": 2
            },
            {
              "compute_ns": 3962417,
              "compute_span_ns": 3962417,
              "calibration_cpu_ns": 4437417,
              "calibration_gpu_ticks": 4437417,
              "compute_passes": 1,
              "command_buffer_ns": 3962458.28597,
              "command_buffers": 2
            },
            {
              "compute_ns": 3967792,
              "compute_span_ns": 3967792,
              "calibration_cpu_ns": 4520709,
              "calibration_gpu_ticks": 4520709,
              "compute_passes": 1,
              "command_buffer_ns": 3967875.03269,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 75583,
              "compute_span_ns": 75583,
              "calibration_cpu_ns": 568458,
              "calibration_gpu_ticks": 568458,
              "compute_passes": 1,
              "command_buffer_ns": 75583.3461881,
              "command_buffers": 2
            },
            {
              "compute_ns": 71334,
              "compute_span_ns": 71334,
              "calibration_cpu_ns": 544958,
              "calibration_gpu_ticks": 544958,
              "compute_passes": 1,
              "command_buffer_ns": 71374.9905117,
              "command_buffers": 2
            },
            {
              "compute_ns": 63584,
              "compute_span_ns": 63584,
              "calibration_cpu_ns": 472208,
              "calibration_gpu_ticks": 472208,
              "compute_passes": 1,
              "command_buffer_ns": 63666.6663922,
              "command_buffers": 2
            },
            {
              "compute_ns": 64459,
              "compute_span_ns": 64459,
              "calibration_cpu_ns": 489625,
              "calibration_gpu_ticks": 489625,
              "compute_passes": 1,
              "command_buffer_ns": 64500.0254735,
              "command_buffers": 2
            },
            {
              "compute_ns": 78458,
              "compute_span_ns": 78458,
              "calibration_cpu_ns": 578459,
              "calibration_gpu_ticks": 578459,
              "compute_passes": 1,
              "command_buffer_ns": 78458.3389759,
              "command_buffers": 2
            },
            {
              "compute_ns": 70125,
              "compute_span_ns": 70125,
              "calibration_cpu_ns": 564000,
              "calibration_gpu_ticks": 564000,
              "compute_passes": 1,
              "command_buffer_ns": 70208.334364,
              "command_buffers": 2
            },
            {
              "compute_ns": 63375,
              "compute_span_ns": 63375,
              "calibration_cpu_ns": 480917,
              "calibration_gpu_ticks": 480917,
              "compute_passes": 1,
              "command_buffer_ns": 63458.3411738,
              "command_buffers": 2
            },
            {
              "compute_ns": 64416,
              "compute_span_ns": 64416,
              "calibration_cpu_ns": 583500,
              "calibration_gpu_ticks": 583500,
              "compute_passes": 1,
              "command_buffer_ns": 64416.6721031,
              "command_buffers": 2
            },
            {
              "compute_ns": 63834,
              "compute_span_ns": 63834,
              "calibration_cpu_ns": 521042,
              "calibration_gpu_ticks": 521042,
              "compute_passes": 1,
              "command_buffer_ns": 63874.9916106,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4005083.3486,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3969791.69454,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3947583.31822,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3935333.34136,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3955041.69865,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3907083.30087,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3985291.63795,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3976791.63143,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3966124.96115,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 66541.7173877,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 64374.9372102,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 64291.7002551,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63333.3693258,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 64666.6740067,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 67083.2814649,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63708.2848698,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 71375.0487193,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 70708.3381712,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              62.579427321875,
              62.0279952271875,
              61.6809893471875,
              61.48958345875,
              61.79752654140625,
              61.04817657609375,
              62.270181842968746,
              62.13736924109375,
              61.970702517968746
            ],
            "command_buffer_throughput_us_p50": 61.970702517968746,
            "command_buffer_throughput_us_p90": 62.332030938749995,
            "command_buffer_latency_us": [
              66.5417173877,
              64.3749372102,
              64.2917002551,
              63.3333693258,
              64.66667400669999,
              67.08328146490001,
              63.7082848698,
              71.3750487193,
              70.7083381712
            ],
            "command_buffer_latency_us_p50": 64.66667400669999,
            "command_buffer_latency_us_p90": 70.84168028082
          },
          "compute_throughput_us": [
            61.54753125,
            61.951828125,
            61.369140625,
            62.783203125,
            61.96809375,
            62.414703125,
            62.05859375,
            61.912765625,
            61.99675
          ],
          "compute_throughput_us_p50": 61.96809375,
          "compute_throughput_us_p90": 62.488403125000005,
          "command_buffer_throughput_us": [
            61.54948005125,
            61.9544271103125,
            61.37044329078125,
            62.78515593296875,
            61.9680986346875,
            62.41406299515625,
            62.05859426703125,
            61.91341071828125,
            61.99804738578125
          ],
          "command_buffer_throughput_us_p50": 61.9680986346875,
          "command_buffer_throughput_us_p90": 62.48828158271875,
          "compute_latency_us": [
            75.583,
            71.334,
            63.584,
            64.459,
            78.458,
            70.125,
            63.375,
            64.416,
            63.834
          ],
          "compute_latency_us_p50": 64.459,
          "compute_latency_us_p90": 76.158,
          "command_buffer_latency_us": [
            75.5833461881,
            71.3749905117,
            63.6666663922,
            64.5000254735,
            78.45833897589999,
            70.208334364,
            63.4583411738,
            64.4166721031,
            63.8749916106
          ],
          "command_buffer_latency_us_p50": 64.5000254735,
          "command_buffer_latency_us_p90": 76.15834474565999,
          "counter_control_throughput_ratio": 0.9988139530125785,
          "counter_control_latency_ratio": 1.018420876074009
        },
        "correctness": {
          "max_abs_error": 2.827305296548843e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 62.158222467,
        "throughput_us_p90": 62.85229339204,
        "latency_us_p50": 302.375,
        "latency_us_p90": 361.75820000000004
      },
      "slowdown": 0.29237081421733285,
      "device_compute_slowdown": 0.2975500477953266,
      "valid": true,
      "round": 2,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "softmax_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-3ib98y42/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "1ca0ba2a59ee9b200a0f7933e3c823dd873803c26a844f54bebdc043f01bc250",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.890709,
        "capture_ms": 0.070083,
        "compile_ms": 53.727041,
        "allocation_upload_ms": 1.24225,
        "cold_call_ms": 2.130833,
        "warmup_ms": 200.157125,
        "download_ms": 0.309292,
        "repetitions": 6238,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          4.69091311318,
          4.65025521,
          4.27934963129,
          4.57462984931,
          4.45017778134,
          4.00270519397,
          4.02319781981,
          4.72508560436,
          3.98897883937
        ],
        "latency_us": [
          261.209,
          220.875,
          244.625,
          220.125,
          215.25,
          215.833,
          197.375,
          208.166,
          204.125
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 37,
            "reduction_scalar_rounds": 40,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 6148,
              "global_write_bytes": 6148,
              "private_read_bytes": 36888,
              "private_write_bytes": 12296
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 32,
              "global_write_bytes": 32,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 2,
            "reduction_elements": 3074,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 88,
            "normalized_kernel_cost": 88,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 4611,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 273417,
              "compute_span_ns": 273417,
              "calibration_cpu_ns": 759209,
              "calibration_gpu_ticks": 759209,
              "compute_passes": 1,
              "command_buffer_ns": 273499.987088,
              "command_buffers": 2
            },
            {
              "compute_ns": 262167,
              "compute_span_ns": 262167,
              "calibration_cpu_ns": 718875,
              "calibration_gpu_ticks": 718875,
              "compute_passes": 1,
              "command_buffer_ns": 262291.694526,
              "command_buffers": 2
            },
            {
              "compute_ns": 242666,
              "compute_span_ns": 242666,
              "calibration_cpu_ns": 672625,
              "calibration_gpu_ticks": 672625,
              "compute_passes": 1,
              "command_buffer_ns": 242708.309088,
              "command_buffers": 2
            },
            {
              "compute_ns": 242958,
              "compute_span_ns": 242958,
              "calibration_cpu_ns": 693458,
              "calibration_gpu_ticks": 693458,
              "compute_passes": 1,
              "command_buffer_ns": 242999.987677,
              "command_buffers": 2
            },
            {
              "compute_ns": 261750,
              "compute_span_ns": 261750,
              "calibration_cpu_ns": 677875,
              "calibration_gpu_ticks": 677875,
              "compute_passes": 1,
              "command_buffer_ns": 261874.985881,
              "command_buffers": 2
            },
            {
              "compute_ns": 261792,
              "compute_span_ns": 261792,
              "calibration_cpu_ns": 669250,
              "calibration_gpu_ticks": 669250,
              "compute_passes": 1,
              "command_buffer_ns": 261916.720774,
              "command_buffers": 2
            },
            {
              "compute_ns": 268917,
              "compute_span_ns": 268917,
              "calibration_cpu_ns": 692292,
              "calibration_gpu_ticks": 692292,
              "compute_passes": 1,
              "command_buffer_ns": 269041.629508,
              "command_buffers": 2
            },
            {
              "compute_ns": 295834,
              "compute_span_ns": 295834,
              "calibration_cpu_ns": 702750,
              "calibration_gpu_ticks": 702750,
              "compute_passes": 1,
              "command_buffer_ns": 295916.688628,
              "command_buffers": 2
            },
            {
              "compute_ns": 276083,
              "compute_span_ns": 276083,
              "calibration_cpu_ns": 670042,
              "calibration_gpu_ticks": 670042,
              "compute_passes": 1,
              "command_buffer_ns": 276124.977972,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 6000,
              "compute_span_ns": 6000,
              "calibration_cpu_ns": 382375,
              "calibration_gpu_ticks": 382375,
              "compute_passes": 1,
              "command_buffer_ns": 6083.34084973,
              "command_buffers": 2
            },
            {
              "compute_ns": 6458,
              "compute_span_ns": 6458,
              "calibration_cpu_ns": 478625,
              "calibration_gpu_ticks": 478625,
              "compute_passes": 1,
              "command_buffer_ns": 6541.72617942,
              "command_buffers": 2
            },
            {
              "compute_ns": 6500,
              "compute_span_ns": 6500,
              "calibration_cpu_ns": 378667,
              "calibration_gpu_ticks": 378667,
              "compute_passes": 1,
              "command_buffer_ns": 6583.34465697,
              "command_buffers": 2
            },
            {
              "compute_ns": 6458,
              "compute_span_ns": 6458,
              "calibration_cpu_ns": 366500,
              "calibration_gpu_ticks": 366500,
              "compute_passes": 1,
              "command_buffer_ns": 6416.63791612,
              "command_buffers": 2
            },
            {
              "compute_ns": 6459,
              "compute_span_ns": 6459,
              "calibration_cpu_ns": 380125,
              "calibration_gpu_ticks": 380125,
              "compute_passes": 1,
              "command_buffer_ns": 6499.99128655,
              "command_buffers": 2
            },
            {
              "compute_ns": 6500,
              "compute_span_ns": 6500,
              "calibration_cpu_ns": 444875,
              "calibration_gpu_ticks": 444875,
              "compute_passes": 1,
              "command_buffer_ns": 6541.6097641,
              "command_buffers": 2
            },
            {
              "compute_ns": 6416,
              "compute_span_ns": 6416,
              "calibration_cpu_ns": 369542,
              "calibration_gpu_ticks": 369542,
              "compute_passes": 1,
              "command_buffer_ns": 6375.01943856,
              "command_buffers": 2
            },
            {
              "compute_ns": 6458,
              "compute_span_ns": 6458,
              "calibration_cpu_ns": 376542,
              "calibration_gpu_ticks": 376542,
              "compute_passes": 1,
              "command_buffer_ns": 6500.04949421,
              "command_buffers": 2
            },
            {
              "compute_ns": 6416,
              "compute_span_ns": 6416,
              "calibration_cpu_ns": 396000,
              "calibration_gpu_ticks": 396000,
              "compute_passes": 1,
              "command_buffer_ns": 6500.04949421,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 288791.663479,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 261666.660663,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 262791.640125,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 262416.666374,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 261333.305389,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 241875.008214,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 262041.692622,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 299166.655168,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 306333.357003,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 6291.7242758,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6624.96313453,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6375.01943856,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6499.99128655,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6499.99128655,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6499.99128655,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6375.01943856,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6541.66797176,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6541.6097641,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              4.5123697418593745,
              4.088541572859375,
              4.106119376953124,
              4.100260412093751,
              4.083332896703125,
              3.77929700334375,
              4.09440144721875,
              4.6744789870000005,
              4.786458703171875
            ],
            "command_buffer_throughput_us_p50": 4.100260412093751,
            "command_buffer_throughput_us_p90": 4.696874930234375,
            "command_buffer_latency_us": [
              6.2917242758,
              6.624963134530001,
              6.37501943856,
              6.49999128655,
              6.49999128655,
              6.49999128655,
              6.37501943856,
              6.54166797176,
              6.5416097640999995
            ],
            "command_buffer_latency_us_p50": 6.49999128655,
            "command_buffer_latency_us_p90": 6.558327004314
          },
          "compute_throughput_us": [
            4.272140625,
            4.096359375,
            3.79165625,
            3.79621875,
            4.08984375,
            4.0905,
            4.201828125,
            4.62240625,
            4.313796875
          ],
          "compute_throughput_us_p50": 4.096359375,
          "compute_throughput_us_p90": 4.37551875,
          "command_buffer_throughput_us": [
            4.27343729825,
            4.09830772696875,
            3.7923173295000003,
            3.796874807453125,
            4.091796654390625,
            4.09244876209375,
            4.2037754610624996,
            4.6236982598125,
            4.3144527808125
          ],
          "command_buffer_throughput_us_p50": 4.09830772696875,
          "command_buffer_throughput_us_p90": 4.3763018766125,
          "compute_latency_us": [
            6.0,
            6.458,
            6.5,
            6.458,
            6.459,
            6.5,
            6.416,
            6.458,
            6.416
          ],
          "compute_latency_us_p50": 6.458,
          "compute_latency_us_p90": 6.5,
          "command_buffer_latency_us": [
            6.08334084973,
            6.54172617942,
            6.58334465697,
            6.41663791612,
            6.49999128655,
            6.5416097640999995,
            6.37501943856,
            6.500049494210001,
            6.500049494210001
          ],
          "command_buffer_latency_us_p50": 6.500049494210001,
          "command_buffer_latency_us_p90": 6.55004987493,
          "counter_control_throughput_ratio": 0.9891366016771656,
          "counter_control_latency_ratio": 0.9936467824604763
        },
        "correctness": {
          "max_abs_error": 2.6856349761260334e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 4.45017778134,
        "throughput_us_p90": 4.697747611416,
        "latency_us_p50": 215.833,
        "latency_us_p90": 247.9418
      },
      "torch": {
        "cold_call_ms": 9.033666,
        "warmup_ms": 200.174083,
        "repetitions": 893,
        "throughput_us": [
          30.636851063829784,
          29.451754759238522,
          29.757978723404257,
          30.225596864501682,
          38.37826539753639,
          33.076987681970884,
          31.272910414333705,
          31.269036954087348,
          30.35288353863382
        ],
        "latency_us": [
          370.584,
          324.458,
          395.583,
          316.33299999999997,
          298.75,
          303.875,
          391.542,
          343.708,
          477.16700000000003
        ],
        "throughput_us_p50": 30.636851063829784,
        "throughput_us_p90": 34.137243225083985,
        "latency_us_p50": 343.708,
        "latency_us_p90": 411.8998,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1110833.3128504455,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1234208.3500698209,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1138416.7009964585,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1327250.0364109874,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1317791.6407585144,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1322624.972090125,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1485791.6394248605,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1325541.699770838,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1486416.6732877493,
                "command_buffers": 7
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 89000.03740563989,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 91291.6730158031,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70041.62762314081,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 85874.98450651765,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 88499.97539073229,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 89541.6596904397,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 88625.00544637442,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 106666.64456948638,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 107250.00174716115,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              17.35677051328821,
              19.28450546984095,
              17.787760953069665,
              20.738281818921678,
              20.590494386851788,
              20.666015188908204,
              23.215494366013445,
              20.711589058919344,
              23.225260520121083
            ],
            "command_buffer_throughput_us_p50": 20.666015188908204,
            "command_buffer_throughput_us_p90": 23.217447596834973,
            "command_buffer_latency_us": [
              89.00003740563989,
              91.2916730158031,
              70.04162762314081,
              85.87498450651765,
              88.49997539073229,
              89.5416596904397,
              88.62500544637442,
              106.66664456948638,
              107.25000174716115
            ],
            "command_buffer_latency_us_p50": 89.00003740563989,
            "command_buffer_latency_us_p90": 106.78331600502133
          },
          "throughput": [
            {
              "compute_ns": 3575506.0,
              "compute_span_ns": 4644584.0,
              "command_buffer_ns": 4530041.653197259,
              "calibration_cpu_ns": 5539083.0,
              "calibration_gpu_ticks": 5539083.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4125667.0,
              "compute_span_ns": 5294333.0,
              "command_buffer_ns": 5180458.247195929,
              "calibration_cpu_ns": 6054083.0,
              "calibration_gpu_ticks": 6054083.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4280503.0,
              "compute_span_ns": 5345417.0,
              "command_buffer_ns": 5262208.345811814,
              "calibration_cpu_ns": 6025084.0,
              "calibration_gpu_ticks": 6025084.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4078792.0,
              "compute_span_ns": 5211875.0,
              "command_buffer_ns": 5090083.344839513,
              "calibration_cpu_ns": 5938625.0,
              "calibration_gpu_ticks": 5938625.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4260039.0,
              "compute_span_ns": 5446458.0,
              "command_buffer_ns": 5298041.563946754,
              "calibration_cpu_ns": 6220792.0,
              "calibration_gpu_ticks": 6220792.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4617794.0,
              "compute_span_ns": 5842584.0,
              "command_buffer_ns": 5727083.480451256,
              "calibration_cpu_ns": 6606792.0,
              "calibration_gpu_ticks": 6606792.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4691332.0,
              "compute_span_ns": 5809333.0,
              "command_buffer_ns": 5691333.382856101,
              "calibration_cpu_ns": 6575625.0,
              "calibration_gpu_ticks": 6575625.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4796500.0,
              "compute_span_ns": 6064917.0,
              "command_buffer_ns": 5933791.806455702,
              "calibration_cpu_ns": 6933875.0,
              "calibration_gpu_ticks": 6933875.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 5138001.0,
              "compute_span_ns": 6431958.0,
              "command_buffer_ns": 6305333.401542157,
              "calibration_cpu_ns": 7137834.0,
              "calibration_gpu_ticks": 7137834.0,
              "compute_passes": 64,
              "command_buffers": 7
            }
          ],
          "latency": [
            {
              "compute_ns": 88542.0,
              "compute_span_ns": 88542.0,
              "command_buffer_ns": 88583.32876116037,
              "calibration_cpu_ns": 567458.0,
              "calibration_gpu_ticks": 567458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 88458.0,
              "compute_span_ns": 88458.0,
              "command_buffer_ns": 88416.68022796512,
              "calibration_cpu_ns": 522208.0,
              "calibration_gpu_ticks": 522208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70042.0,
              "compute_span_ns": 70042.0,
              "command_buffer_ns": 70083.30430835485,
              "calibration_cpu_ns": 534125.0,
              "calibration_gpu_ticks": 534125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 104291.0,
              "compute_span_ns": 104291.0,
              "command_buffer_ns": 104249.97890368104,
              "calibration_cpu_ns": 781375.0,
              "calibration_gpu_ticks": 781375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 88292.0,
              "compute_span_ns": 88292.0,
              "command_buffer_ns": 88333.32685753703,
              "calibration_cpu_ns": 590250.0,
              "calibration_gpu_ticks": 590250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 88375.0,
              "compute_span_ns": 88375.0,
              "command_buffer_ns": 88375.00354275107,
              "calibration_cpu_ns": 530875.0,
              "calibration_gpu_ticks": 530875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 104542.0,
              "compute_span_ns": 104542.0,
              "command_buffer_ns": 104583.33417773247,
              "calibration_cpu_ns": 547250.0,
              "calibration_gpu_ticks": 547250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 104333.0,
              "compute_span_ns": 104333.0,
              "command_buffer_ns": 104333.33227410913,
              "calibration_cpu_ns": 666292.0,
              "calibration_gpu_ticks": 666292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 104375.0,
              "compute_span_ns": 104375.0,
              "command_buffer_ns": 104375.00895932317,
              "calibration_cpu_ns": 646042.0,
              "calibration_gpu_ticks": 646042.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            55.86728125,
            64.463546875,
            66.882859375,
            63.731125,
            66.563109375,
            72.15303125,
            73.3020625,
            74.9453125,
            80.281265625
          ],
          "compute_throughput_us_p50": 66.882859375,
          "compute_throughput_us_p90": 76.012503125,
          "command_buffer_throughput_us": [
            70.78190083120717,
            80.9446601124364,
            82.2220054033096,
            79.5325522631174,
            82.78189943666803,
            89.48567938205088,
            88.92708410712657,
            92.71549697587034,
            98.5208343990962
          ],
          "command_buffer_throughput_us_p50": 82.78189943666803,
          "command_buffer_throughput_us_p90": 93.87656446051551,
          "compute_latency_us": [
            88.542,
            88.458,
            70.042,
            104.291,
            88.292,
            88.375,
            104.542,
            104.333,
            104.375
          ],
          "compute_latency_us_p50": 88.542,
          "compute_latency_us_p90": 104.4084,
          "command_buffer_latency_us": [
            88.58332876116037,
            88.41668022796512,
            70.08330430835485,
            104.24997890368104,
            88.33332685753703,
            88.37500354275107,
            104.58333417773247,
            104.33333227410913,
            104.37500895932317
          ],
          "command_buffer_latency_us_p50": 88.58332876116037,
          "command_buffer_latency_us_p90": 104.41667400300503,
          "counter_control_throughput_ratio": 4.197393614216646,
          "counter_control_latency_ratio": 0.9953178823669449
        },
        "download_ms": 0.264458,
        "allocation_upload_ms": 0.699833,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 4.2122453832241225e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.14525571743872628,
      "device_compute_slowdown": 0.061246774035668236,
      "valid": true,
      "round": 2,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "softmax_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.57425,
        "warmup_ms": 200.530584,
        "repetitions": 905,
        "throughput_us": [
          29.71358232044199,
          29.238903867403312,
          29.154097237569058,
          28.827807734806626,
          29.494428729281765,
          29.107643093922654,
          29.481077348066297,
          29.052624309392264,
          29.096040883977903
        ],
        "latency_us": [
          313.042,
          307.08299999999997,
          344.875,
          247.416,
          325.042,
          324.791,
          322.958,
          313.416,
          344.625
        ],
        "throughput_us_p50": 29.154097237569058,
        "throughput_us_p90": 29.53825944751381,
        "latency_us_p50": 322.958,
        "latency_us_p90": 344.675,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1050625.1128390431,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1048416.6559763253,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1054083.346389234,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1047500.00173226,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1072833.1981226802,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1045583.3398737013,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1147749.9501779675,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1144708.3088569343,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1052666.688337922,
                "command_buffers": 7
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 60499.99501556158,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 62875.04220381379,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 65375.0030323863,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 74499.98520314693,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 61166.64735600352,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80208.29409360886,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 77124.97608736157,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70124.9809935689,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 77000.00423938036,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              16.41601738811005,
              16.381510249630082,
              16.470052287331782,
              16.367187527066562,
              16.763018720666878,
              16.337239685526583,
              17.933592971530743,
              17.8860673258896,
              16.447917005280033
            ],
            "command_buffer_throughput_us_p50": 16.447917005280033,
            "command_buffer_throughput_us_p90": 17.895572455017827,
            "command_buffer_latency_us": [
              60.49999501556158,
              62.87504220381379,
              65.3750030323863,
              74.49998520314693,
              61.16664735600352,
              80.20829409360886,
              77.12497608736157,
              70.1249809935689,
              77.00000423938036
            ],
            "command_buffer_latency_us_p50": 70.1249809935689,
            "command_buffer_latency_us_p90": 77.74163968861103
          },
          "throughput": [
            {
              "compute_ns": 3824667.0,
              "compute_span_ns": 4925083.0,
              "command_buffer_ns": 4803708.288818598,
              "calibration_cpu_ns": 5588208.0,
              "calibration_gpu_ticks": 5588208.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3583957.0,
              "compute_span_ns": 4679333.0,
              "command_buffer_ns": 4558875.109069049,
              "calibration_cpu_ns": 5359625.0,
              "calibration_gpu_ticks": 5359625.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3685380.0,
              "compute_span_ns": 4870000.0,
              "command_buffer_ns": 4755958.332680166,
              "calibration_cpu_ns": 5625125.0,
              "calibration_gpu_ticks": 5625125.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3801419.0,
              "compute_span_ns": 4899417.0,
              "command_buffer_ns": 4797791.596502066,
              "calibration_cpu_ns": 5580875.0,
              "calibration_gpu_ticks": 5580875.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3736039.0,
              "compute_span_ns": 4919000.0,
              "command_buffer_ns": 4796499.968506396,
              "calibration_cpu_ns": 5573166.0,
              "calibration_gpu_ticks": 5573166.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3883582.0,
              "compute_span_ns": 5033875.0,
              "command_buffer_ns": 4916249.949019402,
              "calibration_cpu_ns": 5662125.0,
              "calibration_gpu_ticks": 5662125.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3622456.0,
              "compute_span_ns": 4646000.0,
              "command_buffer_ns": 4541375.092230737,
              "calibration_cpu_ns": 5249875.0,
              "calibration_gpu_ticks": 5249875.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3976536.0,
              "compute_span_ns": 5219291.0,
              "command_buffer_ns": 5095666.740089655,
              "calibration_cpu_ns": 5960500.0,
              "calibration_gpu_ticks": 5960500.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4215167.0,
              "compute_span_ns": 5354500.0,
              "command_buffer_ns": 5252291.681244969,
              "calibration_cpu_ns": 6021166.0,
              "calibration_gpu_ticks": 6021166.0,
              "compute_passes": 64,
              "command_buffers": 7
            }
          ],
          "latency": [
            {
              "compute_ns": 54000.0,
              "compute_span_ns": 54000.0,
              "command_buffer_ns": 54000.00372901559,
              "calibration_cpu_ns": 524083.0,
              "calibration_gpu_ticks": 524083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 58125.0,
              "compute_span_ns": 58125.0,
              "command_buffer_ns": 58125.00603497028,
              "calibration_cpu_ns": 515917.0,
              "calibration_gpu_ticks": 515917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 73625.0,
              "compute_span_ns": 73625.0,
              "command_buffer_ns": 73625.00764429569,
              "calibration_cpu_ns": 698333.0,
              "calibration_gpu_ticks": 698333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 77917.0,
              "compute_span_ns": 77917.0,
              "command_buffer_ns": 77916.65848344564,
              "calibration_cpu_ns": 581875.0,
              "calibration_gpu_ticks": 581875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 58084.0,
              "compute_span_ns": 58084.0,
              "command_buffer_ns": 58125.00603497028,
              "calibration_cpu_ns": 512333.0,
              "calibration_gpu_ticks": 512333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 56709.0,
              "compute_span_ns": 56709.0,
              "command_buffer_ns": 56749.96646121144,
              "calibration_cpu_ns": 492833.0,
              "calibration_gpu_ticks": 492833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 71459.0,
              "compute_span_ns": 71459.0,
              "command_buffer_ns": 71499.96235966682,
              "calibration_cpu_ns": 517333.0,
              "calibration_gpu_ticks": 517333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 76750.0,
              "compute_span_ns": 76750.0,
              "command_buffer_ns": 76750.00233575702,
              "calibration_cpu_ns": 513042.0,
              "calibration_gpu_ticks": 513042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 76708.0,
              "compute_span_ns": 76708.0,
              "command_buffer_ns": 76666.64896532893,
              "calibration_cpu_ns": 523458.0,
              "calibration_gpu_ticks": 523458.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            59.760421875,
            55.999328125,
            57.5840625,
            59.397171875,
            58.375609375,
            60.68096875,
            56.600875,
            62.133375,
            65.861984375
          ],
          "compute_throughput_us_p50": 59.397171875,
          "compute_throughput_us_p90": 62.879096875,
          "command_buffer_throughput_us": [
            75.05794201279059,
            71.2324235792039,
            74.31184894812759,
            74.96549369534478,
            74.94531200791243,
            76.81640545342816,
            70.95898581610527,
            79.61979281390086,
            82.06705751945265
          ],
          "command_buffer_throughput_us_p50": 74.96549369534478,
          "command_buffer_throughput_us_p90": 80.10924575501122,
          "compute_latency_us": [
            54.0,
            58.125,
            73.625,
            77.917,
            58.084,
            56.709,
            71.459,
            76.75,
            76.708
          ],
          "compute_latency_us_p50": 71.459,
          "compute_latency_us_p90": 76.9834,
          "command_buffer_latency_us": [
            54.00000372901559,
            58.12500603497028,
            73.62500764429569,
            77.91665848344564,
            58.12500603497028,
            56.74996646121144,
            71.49996235966682,
            76.75000233575702,
            76.66664896532893
          ],
          "command_buffer_latency_us_p50": 71.49996235966682,
          "command_buffer_latency_us_p90": 76.98333356529474,
          "counter_control_throughput_ratio": 4.511937646080566,
          "counter_control_latency_ratio": 0.9502728782349275
        },
        "download_ms": 0.2885,
        "allocation_upload_ms": 0.804458,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 4.2122453832241225e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-m2o1qkif/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "d45e9e72129e03853b1c708b1bdcbc59be9547b0d0e02ca6bdc2eb0b41c08926",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.812166,
        "capture_ms": 0.070209,
        "compile_ms": 56.524292,
        "allocation_upload_ms": 1.227167,
        "cold_call_ms": 1.812917,
        "warmup_ms": 200.144542,
        "download_ms": 0.290958,
        "repetitions": 6146,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          4.88454555809,
          4.50686088513,
          4.88386755613,
          4.89137910836,
          4.99125447445,
          4.68943898471,
          4.37012561015,
          4.37208493329,
          4.34260087862
        ],
        "latency_us": [
          250.791,
          311.542,
          206.334,
          212.25,
          230.875,
          206.75,
          207.542,
          207.292,
          205.084
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 19,
            "reduction_scalar_rounds": 40,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 6148,
              "global_write_bytes": 6148,
              "private_read_bytes": 36888,
              "private_write_bytes": 12296
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 32,
              "global_write_bytes": 32,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 2,
            "reduction_elements": 3074,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 80,
            "normalized_kernel_cost": 80,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 4611,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 268875,
              "compute_span_ns": 268875,
              "calibration_cpu_ns": 721875,
              "calibration_gpu_ticks": 721875,
              "compute_passes": 1,
              "command_buffer_ns": 268958.392553,
              "command_buffers": 2
            },
            {
              "compute_ns": 268333,
              "compute_span_ns": 268333,
              "calibration_cpu_ns": 698917,
              "calibration_gpu_ticks": 698917,
              "compute_passes": 1,
              "command_buffer_ns": 268375.035375,
              "command_buffers": 2
            },
            {
              "compute_ns": 291292,
              "compute_span_ns": 291292,
              "calibration_cpu_ns": 790334,
              "calibration_gpu_ticks": 790334,
              "compute_passes": 1,
              "command_buffer_ns": 291458.331048,
              "command_buffers": 2
            },
            {
              "compute_ns": 291917,
              "compute_span_ns": 291917,
              "calibration_cpu_ns": 751292,
              "calibration_gpu_ticks": 751292,
              "compute_passes": 1,
              "command_buffer_ns": 292083.306704,
              "command_buffers": 2
            },
            {
              "compute_ns": 294417,
              "compute_span_ns": 294417,
              "calibration_cpu_ns": 766542,
              "calibration_gpu_ticks": 766542,
              "compute_passes": 1,
              "command_buffer_ns": 294500.030577,
              "command_buffers": 2
            },
            {
              "compute_ns": 291250,
              "compute_span_ns": 291250,
              "calibration_cpu_ns": 703292,
              "calibration_gpu_ticks": 703292,
              "compute_passes": 1,
              "command_buffer_ns": 291291.682515,
              "command_buffers": 2
            },
            {
              "compute_ns": 266833,
              "compute_span_ns": 266833,
              "calibration_cpu_ns": 759500,
              "calibration_gpu_ticks": 759500,
              "compute_passes": 1,
              "command_buffer_ns": 266874.965746,
              "command_buffers": 2
            },
            {
              "compute_ns": 266583,
              "compute_span_ns": 266583,
              "calibration_cpu_ns": 775250,
              "calibration_gpu_ticks": 775250,
              "compute_passes": 1,
              "command_buffer_ns": 266708.317213,
              "command_buffers": 2
            },
            {
              "compute_ns": 312875,
              "compute_span_ns": 312875,
              "calibration_cpu_ns": 726916,
              "calibration_gpu_ticks": 726916,
              "compute_passes": 1,
              "command_buffer_ns": 312999.938615,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 6708,
              "compute_span_ns": 6708,
              "calibration_cpu_ns": 382250,
              "calibration_gpu_ticks": 382250,
              "compute_passes": 1,
              "command_buffer_ns": 6749.93498251,
              "command_buffers": 2
            },
            {
              "compute_ns": 7083,
              "compute_span_ns": 7083,
              "calibration_cpu_ns": 467208,
              "calibration_gpu_ticks": 467208,
              "compute_passes": 1,
              "command_buffer_ns": 7166.64362699,
              "command_buffers": 2
            },
            {
              "compute_ns": 7041,
              "compute_span_ns": 7041,
              "calibration_cpu_ns": 372750,
              "calibration_gpu_ticks": 372750,
              "compute_passes": 1,
              "command_buffer_ns": 6999.99509379,
              "command_buffers": 2
            },
            {
              "compute_ns": 7125,
              "compute_span_ns": 7125,
              "calibration_cpu_ns": 381125,
              "calibration_gpu_ticks": 381125,
              "compute_passes": 1,
              "command_buffer_ns": 7208.3203122,
              "command_buffers": 2
            },
            {
              "compute_ns": 10042,
              "compute_span_ns": 10042,
              "calibration_cpu_ns": 380667,
              "calibration_gpu_ticks": 380667,
              "compute_passes": 1,
              "command_buffer_ns": 10083.3713077,
              "command_buffers": 2
            },
            {
              "compute_ns": 6958,
              "compute_span_ns": 6958,
              "calibration_cpu_ns": 373458,
              "calibration_gpu_ticks": 373458,
              "compute_passes": 1,
              "command_buffer_ns": 6958.26020092,
              "command_buffers": 2
            },
            {
              "compute_ns": 7083,
              "compute_span_ns": 7083,
              "calibration_cpu_ns": 399542,
              "calibration_gpu_ticks": 399542,
              "compute_passes": 1,
              "command_buffer_ns": 7125.02514943,
              "command_buffers": 2
            },
            {
              "compute_ns": 7500,
              "compute_span_ns": 7500,
              "calibration_cpu_ns": 457042,
              "calibration_gpu_ticks": 457042,
              "compute_passes": 1,
              "command_buffer_ns": 7541.67558625,
              "command_buffers": 2
            },
            {
              "compute_ns": 7042,
              "compute_span_ns": 7042,
              "calibration_cpu_ns": 421375,
              "calibration_gpu_ticks": 421375,
              "compute_passes": 1,
              "command_buffer_ns": 7083.34846422,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 295750.040095,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 269000.011031,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 291791.686323,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 292208.394967,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 292291.69013,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 293499.964755,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 296291.66238,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 292124.983389,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 292125.041597,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 6874.96503815,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6999.99509379,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7624.97074902,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7125.02514943,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7166.70183465,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7166.64362699,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7333.29216018,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7208.3203122,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7083.34846422,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              4.6210943764843755,
              4.203125172359375,
              4.559245098796875,
              4.565756171359375,
              4.56705765828125,
              4.585936949296875,
              4.6295572246875,
              4.564452865453125,
              4.564453774953124
            ],
            "command_buffer_throughput_us_p50": 4.565756171359375,
            "command_buffer_throughput_us_p90": 4.622786946125,
            "command_buffer_latency_us": [
              6.87496503815,
              6.99999509379,
              7.62497074902,
              7.12502514943,
              7.1667018346499995,
              7.16664362699,
              7.33329216018,
              7.208320312200001,
              7.08334846422
            ],
            "command_buffer_latency_us_p50": 7.16664362699,
            "command_buffer_latency_us_p90": 7.391627877948
          },
          "compute_throughput_us": [
            4.201171875,
            4.192703125,
            4.5514375,
            4.561203125,
            4.600265625,
            4.55078125,
            4.169265625,
            4.165359375,
            4.888671875
          ],
          "compute_throughput_us_p50": 4.55078125,
          "compute_throughput_us_p90": 4.6579468749999995,
          "command_buffer_throughput_us": [
            4.202474883640625,
            4.193359927734375,
            4.554036422625001,
            4.56380166725,
            4.601562977765625,
            4.551432539296875,
            4.16992133978125,
            4.167317456453125,
            4.890624040859375
          ],
          "command_buffer_throughput_us_p50": 4.551432539296875,
          "command_buffer_throughput_us_p90": 4.659375190384376,
          "compute_latency_us": [
            6.708,
            7.083,
            7.041,
            7.125,
            10.042,
            6.958,
            7.083,
            7.5,
            7.042
          ],
          "compute_latency_us_p50": 7.083,
          "compute_latency_us_p90": 8.0084,
          "command_buffer_latency_us": [
            6.74993498251,
            7.16664362699,
            6.99999509379,
            7.208320312200001,
            10.083371307699998,
            6.95826020092,
            7.12502514943,
            7.54167558625,
            7.08334846422
          ],
          "command_buffer_latency_us_p50": 7.12502514943,
          "command_buffer_latency_us_p90": 8.050014730540001,
          "counter_control_throughput_ratio": 0.9976766705190655,
          "counter_control_latency_ratio": 1.0
        },
        "correctness": {
          "max_abs_error": 2.6856349761260334e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 4.68943898471,
        "throughput_us_p90": 4.911354181578,
        "latency_us_p50": 207.542,
        "latency_us_p90": 262.9412
      },
      "slowdown": 0.1608500838320253,
      "device_compute_slowdown": 0.07661612676739923,
      "valid": true,
      "round": 2,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "softmax_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-n2ljwpkp/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "7bac43b2781c01105a8e678fa6161555c0e89cbd2b0a98679fdc42bb1bd74b45",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 25.622375,
        "capture_ms": 0.07975,
        "compile_ms": 48.585083,
        "allocation_upload_ms": 1.662042,
        "cold_call_ms": 5.955792,
        "warmup_ms": 200.561125,
        "download_ms": 1.015708,
        "repetitions": 1995,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          15.2578320802,
          14.7851924812,
          15.2214285714,
          14.7811192982,
          14.3850250627,
          14.3116330827,
          14.5105057644,
          15.1328947368,
          14.9272972431
        ],
        "latency_us": [
          272.208,
          232.75,
          284.833,
          233.666,
          228,
          235.833,
          254.625,
          244.333,
          233.125
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 128,
            "reduction_scalar_rounds": 60,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 12288,
              "global_write_bytes": 12288,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 48,
              "global_write_bytes": 48,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 2,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 100,
            "normalized_kernel_cost": 400,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 9216,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 946542,
              "compute_span_ns": 946542,
              "calibration_cpu_ns": 1387666,
              "calibration_gpu_ticks": 1387666,
              "compute_passes": 1,
              "command_buffer_ns": 946625.019424,
              "command_buffers": 2
            },
            {
              "compute_ns": 958416,
              "compute_span_ns": 958416,
              "calibration_cpu_ns": 1396958,
              "calibration_gpu_ticks": 1396958,
              "compute_passes": 1,
              "command_buffer_ns": 958416.669164,
              "command_buffers": 2
            },
            {
              "compute_ns": 928791,
              "compute_span_ns": 928791,
              "calibration_cpu_ns": 1499208,
              "calibration_gpu_ticks": 1499208,
              "compute_passes": 1,
              "command_buffer_ns": 928875.058889,
              "command_buffers": 2
            },
            {
              "compute_ns": 935125,
              "compute_span_ns": 935125,
              "calibration_cpu_ns": 1386459,
              "calibration_gpu_ticks": 1386459,
              "compute_passes": 1,
              "command_buffer_ns": 935249.961913,
              "command_buffers": 2
            },
            {
              "compute_ns": 953375,
              "compute_span_ns": 953375,
              "calibration_cpu_ns": 1396333,
              "calibration_gpu_ticks": 1396333,
              "compute_passes": 1,
              "command_buffer_ns": 953375.012614,
              "command_buffers": 2
            },
            {
              "compute_ns": 959959,
              "compute_span_ns": 959959,
              "calibration_cpu_ns": 1409208,
              "calibration_gpu_ticks": 1409208,
              "compute_passes": 1,
              "command_buffer_ns": 960041.652434,
              "command_buffers": 2
            },
            {
              "compute_ns": 972375,
              "compute_span_ns": 972375,
              "calibration_cpu_ns": 1395875,
              "calibration_gpu_ticks": 1395875,
              "compute_passes": 1,
              "command_buffer_ns": 972374.982666,
              "command_buffers": 2
            },
            {
              "compute_ns": 975958,
              "compute_span_ns": 975958,
              "calibration_cpu_ns": 1360292,
              "calibration_gpu_ticks": 1360292,
              "compute_passes": 1,
              "command_buffer_ns": 975958.362687,
              "command_buffers": 2
            },
            {
              "compute_ns": 960584,
              "compute_span_ns": 960584,
              "calibration_cpu_ns": 1391416,
              "calibration_gpu_ticks": 1391416,
              "compute_passes": 1,
              "command_buffer_ns": 960708.362982,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 18667,
              "compute_span_ns": 18667,
              "calibration_cpu_ns": 433042,
              "calibration_gpu_ticks": 433042,
              "compute_passes": 1,
              "command_buffer_ns": 18749.9681488,
              "command_buffers": 2
            },
            {
              "compute_ns": 18667,
              "compute_span_ns": 18667,
              "calibration_cpu_ns": 386333,
              "calibration_gpu_ticks": 386333,
              "compute_passes": 1,
              "command_buffer_ns": 18708.3496712,
              "command_buffers": 2
            },
            {
              "compute_ns": 18666,
              "compute_span_ns": 18666,
              "calibration_cpu_ns": 434625,
              "calibration_gpu_ticks": 434625,
              "compute_passes": 1,
              "command_buffer_ns": 18708.3496712,
              "command_buffers": 2
            },
            {
              "compute_ns": 18541,
              "compute_span_ns": 18541,
              "calibration_cpu_ns": 411417,
              "calibration_gpu_ticks": 411417,
              "compute_passes": 1,
              "command_buffer_ns": 18583.3778232,
              "command_buffers": 2
            },
            {
              "compute_ns": 19042,
              "compute_span_ns": 19042,
              "calibration_cpu_ns": 454000,
              "calibration_gpu_ticks": 454000,
              "compute_passes": 1,
              "command_buffer_ns": 19124.9419004,
              "command_buffers": 2
            },
            {
              "compute_ns": 18709,
              "compute_span_ns": 18709,
              "calibration_cpu_ns": 445125,
              "calibration_gpu_ticks": 445125,
              "compute_passes": 1,
              "command_buffer_ns": 18874.9399967,
              "command_buffers": 2
            },
            {
              "compute_ns": 18833,
              "compute_span_ns": 18833,
              "calibration_cpu_ns": 421167,
              "calibration_gpu_ticks": 421167,
              "compute_passes": 1,
              "command_buffer_ns": 18916.6748896,
              "command_buffers": 2
            },
            {
              "compute_ns": 18583,
              "compute_span_ns": 18583,
              "calibration_cpu_ns": 416625,
              "calibration_gpu_ticks": 416625,
              "compute_passes": 1,
              "command_buffer_ns": 18666.672986,
              "command_buffers": 2
            },
            {
              "compute_ns": 18584,
              "compute_span_ns": 18584,
              "calibration_cpu_ns": 457333,
              "calibration_gpu_ticks": 457333,
              "compute_passes": 1,
              "command_buffer_ns": 18624.9963008,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1033166.71448,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 924791.675061,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1003833.31301,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 943124.992773,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1004583.31872,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 944416.620769,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1027874.99782,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 940875.033848,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1059708.2437,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 16833.3062902,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18749.9681488,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18666.672986,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18999.9700524,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18666.672986,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18708.2914636,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 21583.3424591,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18624.9963008,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18500.0244528,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              16.143229913749998,
              14.449869922828126,
              15.684895515781252,
              14.736328012078125,
              15.696614355,
              14.756509699515625,
              16.0605468409375,
              14.701172403875,
              16.5579413078125
            ],
            "command_buffer_throughput_us_p50": 15.684895515781252,
            "command_buffer_throughput_us_p90": 16.226172192562498,
            "command_buffer_latency_us": [
              16.8333062902,
              18.749968148799997,
              18.666672986000002,
              18.9999700524,
              18.666672986000002,
              18.7082914636,
              21.5833424591,
              18.624996300800003,
              18.5000244528
            ],
            "command_buffer_latency_us_p50": 18.666672986000002,
            "command_buffer_latency_us_p90": 19.51664453374
          },
          "compute_throughput_us": [
            14.78971875,
            14.97525,
            14.512359375,
            14.611328125,
            14.896484375,
            14.999359375,
            15.193359375,
            15.24934375,
            15.009125
          ],
          "compute_throughput_us_p50": 14.97525,
          "compute_throughput_us_p90": 15.20455625,
          "command_buffer_throughput_us": [
            14.7910159285,
            14.975260455687499,
            14.513672795140625,
            14.613280654890625,
            14.89648457209375,
            15.000650819281251,
            15.19335910415625,
            15.249349416984375,
            15.01106817159375
          ],
          "command_buffer_throughput_us_p50": 14.975260455687499,
          "command_buffer_throughput_us_p90": 15.204557166721875,
          "compute_latency_us": [
            18.667,
            18.667,
            18.666,
            18.541,
            19.042,
            18.709,
            18.833,
            18.583,
            18.584
          ],
          "compute_latency_us_p50": 18.667,
          "compute_latency_us_p90": 18.8748,
          "command_buffer_latency_us": [
            18.749968148799997,
            18.7083496712,
            18.7083496712,
            18.5833778232,
            19.1249419004,
            18.8749399967,
            18.9166748896,
            18.666672986000002,
            18.624996300800003
          ],
          "command_buffer_latency_us_p50": 18.7083496712,
          "command_buffer_latency_us_p90": 18.95832829176,
          "counter_control_throughput_ratio": 0.9490253270667011,
          "counter_control_latency_ratio": 1.0022376748175896
        },
        "correctness": {
          "max_abs_error": 1.569250362249905e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 14.7851924812,
        "throughput_us_p90": 15.22870927316,
        "latency_us_p50": 235.833,
        "latency_us_p90": 274.733
      },
      "torch": {
        "cold_call_ms": 6.748459,
        "warmup_ms": 200.821667,
        "repetitions": 884,
        "throughput_us": [
          33.64691742081448,
          33.46460294117647,
          33.36420588235294,
          33.525263574660634,
          35.169354072398185,
          33.73274886877828,
          33.779742081447964,
          33.57866628959276,
          33.62697963800905
        ],
        "latency_us": [
          356.66700000000003,
          333.625,
          319.125,
          543.584,
          353.292,
          331.25,
          405.875,
          327.084,
          341.0
        ],
        "throughput_us_p50": 33.62697963800905,
        "throughput_us_p90": 34.05766447963801,
        "latency_us_p50": 341.0,
        "latency_us_p90": 433.4168,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1861874.8290464282,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1886708.3126679063,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1839458.1857137382,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1837375.049944967,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1841083.4018141031,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1838083.437178284,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1858250.0633783638,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1870500.0402405858,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1838750.1477263868,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 48249.95994567871,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49291.644245386124,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49291.644245386124,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49999.97327104211,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49541.646149009466,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49333.320930600166,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49499.96946379542,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 48541.69674217701,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49374.99761581421,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              29.09179420385044,
              29.479817385436036,
              28.74153415177716,
              28.70898515539011,
              28.76692815334536,
              28.720053705910686,
              29.035157240286935,
              29.226563128759153,
              28.730471058224794
            ],
            "command_buffer_throughput_us_p50": 28.76692815334536,
            "command_buffer_throughput_us_p90": 29.27721398009453,
            "command_buffer_latency_us": [
              48.24995994567871,
              49.291644245386124,
              49.291644245386124,
              49.99997327104211,
              49.541646149009466,
              49.333320930600166,
              49.49996946379542,
              48.54169674217701,
              49.37499761581421
            ],
            "command_buffer_latency_us_p50": 49.333320930600166,
            "command_buffer_latency_us_p90": 49.633311573415995
          },
          "throughput": [
            {
              "compute_ns": 2967579.0,
              "compute_span_ns": 3864417.0,
              "command_buffer_ns": 3713333.3389647305,
              "calibration_cpu_ns": 4804500.0,
              "calibration_gpu_ticks": 4804500.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2983498.0,
              "compute_span_ns": 3867917.0,
              "command_buffer_ns": 3736083.22115615,
              "calibration_cpu_ns": 4660292.0,
              "calibration_gpu_ticks": 4660292.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2981835.0,
              "compute_span_ns": 3934333.0,
              "command_buffer_ns": 3783458.2617506385,
              "calibration_cpu_ns": 4712375.0,
              "calibration_gpu_ticks": 4712375.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2971666.0,
              "compute_span_ns": 3875916.0,
              "command_buffer_ns": 3733958.350494504,
              "calibration_cpu_ns": 4718417.0,
              "calibration_gpu_ticks": 4718417.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2985539.0,
              "compute_span_ns": 3915792.0,
              "command_buffer_ns": 3730249.940417707,
              "calibration_cpu_ns": 4592833.0,
              "calibration_gpu_ticks": 4592833.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2954252.0,
              "compute_span_ns": 3829750.0,
              "command_buffer_ns": 3690125.0132359564,
              "calibration_cpu_ns": 4617333.0,
              "calibration_gpu_ticks": 4617333.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3007250.0,
              "compute_span_ns": 3816500.0,
              "command_buffer_ns": 3694291.6340194643,
              "calibration_cpu_ns": 4472875.0,
              "calibration_gpu_ticks": 4472875.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3015038.0,
              "compute_span_ns": 3930625.0,
              "command_buffer_ns": 3772374.999243766,
              "calibration_cpu_ns": 4686792.0,
              "calibration_gpu_ticks": 4686792.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3005666.0,
              "compute_span_ns": 3874166.0,
              "command_buffer_ns": 3724458.5109874606,
              "calibration_cpu_ns": 4489458.0,
              "calibration_gpu_ticks": 4489458.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 48417.0,
              "compute_span_ns": 48417.0,
              "command_buffer_ns": 48458.343371748924,
              "calibration_cpu_ns": 569584.0,
              "calibration_gpu_ticks": 569584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47416.0,
              "compute_span_ns": 47416.0,
              "command_buffer_ns": 47374.98238682747,
              "calibration_cpu_ns": 482625.0,
              "calibration_gpu_ticks": 482625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49500.0,
              "compute_span_ns": 49500.0,
              "command_buffer_ns": 49500.02767145634,
              "calibration_cpu_ns": 480166.0,
              "calibration_gpu_ticks": 480166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49666.0,
              "compute_span_ns": 49666.0,
              "command_buffer_ns": 49624.99951943755,
              "calibration_cpu_ns": 488083.0,
              "calibration_gpu_ticks": 488083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47792.0,
              "compute_span_ns": 47792.0,
              "command_buffer_ns": 47791.63282364607,
              "calibration_cpu_ns": 457542.0,
              "calibration_gpu_ticks": 457542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 48042.0,
              "compute_span_ns": 48042.0,
              "command_buffer_ns": 48083.311412483454,
              "calibration_cpu_ns": 476208.0,
              "calibration_gpu_ticks": 476208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 50125.0,
              "compute_span_ns": 50125.0,
              "command_buffer_ns": 50125.00332668424,
              "calibration_cpu_ns": 520000.0,
              "calibration_gpu_ticks": 520000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 50333.0,
              "compute_span_ns": 50333.0,
              "command_buffer_ns": 50291.651859879494,
              "calibration_cpu_ns": 472000.0,
              "calibration_gpu_ticks": 472000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47917.0,
              "compute_span_ns": 47917.0,
              "command_buffer_ns": 47958.281356841326,
              "calibration_cpu_ns": 509583.0,
              "calibration_gpu_ticks": 509583.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            46.368421875,
            46.61715625,
            46.591171875,
            46.43228125,
            46.649046875,
            46.1601875,
            46.98828125,
            47.10996875,
            46.96353125
          ],
          "compute_throughput_us_p50": 46.61715625,
          "compute_throughput_us_p90": 47.01261875,
          "command_buffer_throughput_us": [
            58.020833421323914,
            58.376300330564845,
            59.116535339853726,
            58.343099226476625,
            58.28515531902667,
            57.65820333181182,
            57.72330678155413,
            58.943359363183845,
            58.19466423417907
          ],
          "command_buffer_throughput_us_p50": 58.28515531902667,
          "command_buffer_throughput_us_p90": 58.97799455851782,
          "compute_latency_us": [
            48.417,
            47.416,
            49.5,
            49.666,
            47.792,
            48.042,
            50.125,
            50.333,
            47.917
          ],
          "compute_latency_us_p50": 48.417,
          "compute_latency_us_p90": 50.1666,
          "command_buffer_latency_us": [
            48.458343371748924,
            47.37498238682747,
            49.50002767145634,
            49.62499951943755,
            47.79163282364607,
            48.083311412483454,
            50.12500332668424,
            50.291651859879494,
            47.958281356841326
          ],
          "command_buffer_latency_us_p50": 48.458343371748924,
          "command_buffer_latency_us_p90": 50.15833303332329,
          "counter_control_throughput_ratio": 2.0167735461575074,
          "counter_control_latency_ratio": 0.9925005209588437
        },
        "download_ms": 0.600625,
        "allocation_upload_ms": 1.101291,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 2.673614807194649e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.43968244071757456,
      "device_compute_slowdown": 0.3212390288178507,
      "valid": true,
      "round": 2,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "softmax_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.438167,
        "warmup_ms": 200.422875,
        "repetitions": 713,
        "throughput_us": [
          45.885227208976154,
          33.455995792426364,
          33.56358064516129,
          33.28342636746143,
          33.53266760168302,
          33.272206171107996,
          33.33058625525946,
          33.27062973352034,
          33.58216549789621
        ],
        "latency_us": [
          333.5,
          305.291,
          299.417,
          290.66700000000003,
          285.625,
          306.291,
          295.833,
          287.292,
          281.583
        ],
        "throughput_us_p50": 33.455995792426364,
        "throughput_us_p90": 36.0427778401122,
        "latency_us_p50": 295.833,
        "latency_us_p90": 311.7328,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1866458.2748897374,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1864583.2315087318,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1836708.3393968642,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1851416.7750254273,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1867791.695985943,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1851833.2508392632,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1846791.5360815823,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1827000.0582560897,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1841833.2911096513,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 49041.700549423695,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49875.001423060894,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 53083.29127728939,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 50291.71006754041,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 47708.3376608789,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49624.99951943755,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49333.320930600166,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 48666.668590158224,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 47958.281356841326,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              29.163410545152146,
              29.134112992323935,
              28.698567803076003,
              28.928387109772302,
              29.18424524978036,
              28.934894544363488,
              28.856117751274724,
              28.5468759102514,
              28.778645173588302
            ],
            "command_buffer_throughput_us_p50": 28.928387109772302,
            "command_buffer_throughput_us_p90": 29.16757748607779,
            "command_buffer_latency_us": [
              49.041700549423695,
              49.875001423060894,
              53.08329127728939,
              50.29171006754041,
              47.7083376608789,
              49.62499951943755,
              49.333320930600166,
              48.666668590158224,
              47.958281356841326
            ],
            "command_buffer_latency_us_p50": 49.333320930600166,
            "command_buffer_latency_us_p90": 50.850026309490204
          },
          "throughput": [
            {
              "compute_ns": 3020500.0,
              "compute_span_ns": 4004375.0,
              "command_buffer_ns": 3838374.9779313803,
              "calibration_cpu_ns": 4633125.0,
              "calibration_gpu_ticks": 4633125.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3021414.0,
              "compute_span_ns": 3978250.0,
              "command_buffer_ns": 3811250.0915303826,
              "calibration_cpu_ns": 4572041.0,
              "calibration_gpu_ticks": 4572041.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2993667.0,
              "compute_span_ns": 3853250.0,
              "command_buffer_ns": 3714583.464898169,
              "calibration_cpu_ns": 4468666.0,
              "calibration_gpu_ticks": 4468666.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3031956.0,
              "compute_span_ns": 3976875.0,
              "command_buffer_ns": 3826916.6834652424,
              "calibration_cpu_ns": 4589666.0,
              "calibration_gpu_ticks": 4589666.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3031748.0,
              "compute_span_ns": 3937875.0,
              "command_buffer_ns": 3797249.9849274755,
              "calibration_cpu_ns": 4572375.0,
              "calibration_gpu_ticks": 4572375.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3013544.0,
              "compute_span_ns": 4005667.0,
              "command_buffer_ns": 3855916.7296625674,
              "calibration_cpu_ns": 4639083.0,
              "calibration_gpu_ticks": 4639083.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3018250.0,
              "compute_span_ns": 3901500.0,
              "command_buffer_ns": 3760458.3194479346,
              "calibration_cpu_ns": 4502542.0,
              "calibration_gpu_ticks": 4502542.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3073831.0,
              "compute_span_ns": 4049792.0,
              "command_buffer_ns": 3876624.919939786,
              "calibration_cpu_ns": 4852708.0,
              "calibration_gpu_ticks": 4852708.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3025377.0,
              "compute_span_ns": 3977667.0,
              "command_buffer_ns": 3790041.6646152735,
              "calibration_cpu_ns": 4591416.0,
              "calibration_gpu_ticks": 4591416.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 46834.0,
              "compute_span_ns": 46834.0,
              "command_buffer_ns": 46875.0367872417,
              "calibration_cpu_ns": 505291.0,
              "calibration_gpu_ticks": 505291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49916.0,
              "compute_span_ns": 49916.0,
              "command_buffer_ns": 49875.001423060894,
              "calibration_cpu_ns": 496417.0,
              "calibration_gpu_ticks": 496417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49666.0,
              "compute_span_ns": 49666.0,
              "command_buffer_ns": 49624.99951943755,
              "calibration_cpu_ns": 499834.0,
              "calibration_gpu_ticks": 499834.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 50666.0,
              "compute_span_ns": 50666.0,
              "command_buffer_ns": 50625.00713393092,
              "calibration_cpu_ns": 502791.0,
              "calibration_gpu_ticks": 502791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49708.0,
              "compute_span_ns": 49708.0,
              "command_buffer_ns": 49666.676204651594,
              "calibration_cpu_ns": 494750.0,
              "calibration_gpu_ticks": 494750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49750.0,
              "compute_span_ns": 49750.0,
              "command_buffer_ns": 49749.971367418766,
              "calibration_cpu_ns": 497791.0,
              "calibration_gpu_ticks": 497791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 50542.0,
              "compute_span_ns": 50542.0,
              "command_buffer_ns": 50541.653763502836,
              "calibration_cpu_ns": 565334.0,
              "calibration_gpu_ticks": 565334.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47916.0,
              "compute_span_ns": 47916.0,
              "command_buffer_ns": 47874.986194074154,
              "calibration_cpu_ns": 494208.0,
              "calibration_gpu_ticks": 494208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49750.0,
              "compute_span_ns": 49750.0,
              "command_buffer_ns": 49749.971367418766,
              "calibration_cpu_ns": 506291.0,
              "calibration_gpu_ticks": 506291.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            47.1953125,
            47.20959375,
            46.776046875,
            47.3743125,
            47.3710625,
            47.086625,
            47.16015625,
            48.028609375,
            47.271515625
          ],
          "compute_throughput_us_p50": 47.20959375,
          "compute_throughput_us_p90": 47.505171875,
          "command_buffer_throughput_us": [
            59.97460903017782,
            59.55078268016223,
            58.04036663903389,
            59.79557317914441,
            59.332031014491804,
            60.248698900977615,
            58.75716124137398,
            60.57226437405916,
            59.21940100961365
          ],
          "command_buffer_throughput_us_p50": 59.55078268016223,
          "command_buffer_throughput_us_p90": 60.313411995593924,
          "compute_latency_us": [
            46.834,
            49.916,
            49.666,
            50.666,
            49.708,
            49.75,
            50.542,
            47.916,
            49.75
          ],
          "compute_latency_us_p50": 49.75,
          "compute_latency_us_p90": 50.5668,
          "command_buffer_latency_us": [
            46.8750367872417,
            49.875001423060894,
            49.62499951943755,
            50.62500713393092,
            49.666676204651594,
            49.749971367418766,
            50.541653763502836,
            47.874986194074154,
            49.749971367418766
          ],
          "command_buffer_latency_us_p50": 49.749971367418766,
          "command_buffer_latency_us_p90": 50.55832443758845,
          "counter_control_throughput_ratio": 2.0565018942939592,
          "counter_control_latency_ratio": 1.0025183244169558
        },
        "download_ms": 0.70925,
        "allocation_upload_ms": 0.632542,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 2.673614807194649e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-0unc_rxv/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "0d8a82e58cbe97c43d15ef559cf043ad575745e0d3fb0f4f40ed60b2caca21c0",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.945167,
        "capture_ms": 0.071791,
        "compile_ms": 49.942375,
        "allocation_upload_ms": 1.797208,
        "cold_call_ms": 5.191333,
        "warmup_ms": 200.139917,
        "download_ms": 0.924,
        "repetitions": 1712,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          14.4546098131,
          14.140478972,
          14.3589853972,
          13.9424649533,
          13.9541962617,
          13.9026483645,
          14.1664228972,
          14.3365946262,
          14.1946553738
        ],
        "latency_us": [
          320.917,
          248.541,
          253.375,
          263.958,
          258.75,
          261.25,
          325.542,
          280.042,
          251
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 256,
            "reduction_scalar_rounds": 60,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 12288,
              "global_write_bytes": 12288,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 48,
              "global_write_bytes": 48,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 2,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 108,
            "normalized_kernel_cost": 432,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 9216,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 880666,
              "compute_span_ns": 880666,
              "calibration_cpu_ns": 1333459,
              "calibration_gpu_ticks": 1333459,
              "compute_passes": 1,
              "command_buffer_ns": 880708.335899,
              "command_buffers": 2
            },
            {
              "compute_ns": 887541,
              "compute_span_ns": 887541,
              "calibration_cpu_ns": 1337917,
              "calibration_gpu_ticks": 1337917,
              "compute_passes": 1,
              "command_buffer_ns": 887541.682459,
              "command_buffers": 2
            },
            {
              "compute_ns": 936250,
              "compute_span_ns": 936250,
              "calibration_cpu_ns": 1424917,
              "calibration_gpu_ticks": 1424917,
              "compute_passes": 1,
              "command_buffer_ns": 936250.027735,
              "command_buffers": 2
            },
            {
              "compute_ns": 866417,
              "compute_span_ns": 866417,
              "calibration_cpu_ns": 1365125,
              "calibration_gpu_ticks": 1365125,
              "compute_passes": 1,
              "command_buffer_ns": 866541.580763,
              "command_buffers": 2
            },
            {
              "compute_ns": 851250,
              "compute_span_ns": 851250,
              "calibration_cpu_ns": 1383125,
              "calibration_gpu_ticks": 1383125,
              "compute_passes": 1,
              "command_buffer_ns": 851250.020787,
              "command_buffers": 2
            },
            {
              "compute_ns": 917334,
              "compute_span_ns": 917334,
              "calibration_cpu_ns": 1390834,
              "calibration_gpu_ticks": 1390834,
              "compute_passes": 1,
              "command_buffer_ns": 917416.648008,
              "command_buffers": 2
            },
            {
              "compute_ns": 844667,
              "compute_span_ns": 844667,
              "calibration_cpu_ns": 1334459,
              "calibration_gpu_ticks": 1334459,
              "compute_passes": 1,
              "command_buffer_ns": 844791.706186,
              "command_buffers": 2
            },
            {
              "compute_ns": 914708,
              "compute_span_ns": 914708,
              "calibration_cpu_ns": 1408708,
              "calibration_gpu_ticks": 1408708,
              "compute_passes": 1,
              "command_buffer_ns": 914708.361961,
              "command_buffers": 2
            },
            {
              "compute_ns": 868583,
              "compute_span_ns": 868583,
              "calibration_cpu_ns": 1347208,
              "calibration_gpu_ticks": 1347208,
              "compute_passes": 1,
              "command_buffer_ns": 868666.684255,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 18000,
              "compute_span_ns": 18000,
              "calibration_cpu_ns": 454750,
              "calibration_gpu_ticks": 454750,
              "compute_passes": 1,
              "command_buffer_ns": 18000.0206456,
              "command_buffers": 2
            },
            {
              "compute_ns": 18250,
              "compute_span_ns": 18250,
              "calibration_cpu_ns": 453417,
              "calibration_gpu_ticks": 453417,
              "compute_passes": 1,
              "command_buffer_ns": 18291.6992344,
              "command_buffers": 2
            },
            {
              "compute_ns": 17917,
              "compute_span_ns": 17917,
              "calibration_cpu_ns": 455708,
              "calibration_gpu_ticks": 455708,
              "compute_passes": 1,
              "command_buffer_ns": 18000.0206456,
              "command_buffers": 2
            },
            {
              "compute_ns": 17875,
              "compute_span_ns": 17875,
              "calibration_cpu_ns": 405625,
              "calibration_gpu_ticks": 405625,
              "compute_passes": 1,
              "command_buffer_ns": 17875.0487976,
              "command_buffers": 2
            },
            {
              "compute_ns": 18041,
              "compute_span_ns": 18041,
              "calibration_cpu_ns": 424459,
              "calibration_gpu_ticks": 424459,
              "compute_passes": 1,
              "command_buffer_ns": 18083.374016,
              "command_buffers": 2
            },
            {
              "compute_ns": 18083,
              "compute_span_ns": 18083,
              "calibration_cpu_ns": 448042,
              "calibration_gpu_ticks": 448042,
              "compute_passes": 1,
              "command_buffer_ns": 18124.9342859,
              "command_buffers": 2
            },
            {
              "compute_ns": 18000,
              "compute_span_ns": 18000,
              "calibration_cpu_ns": 457042,
              "calibration_gpu_ticks": 457042,
              "compute_passes": 1,
              "command_buffer_ns": 18083.2576007,
              "command_buffers": 2
            },
            {
              "compute_ns": 19042,
              "compute_span_ns": 19042,
              "calibration_cpu_ns": 593875,
              "calibration_gpu_ticks": 593875,
              "compute_passes": 1,
              "command_buffer_ns": 19041.7049453,
              "command_buffers": 2
            },
            {
              "compute_ns": 18000,
              "compute_span_ns": 18000,
              "calibration_cpu_ns": 411209,
              "calibration_gpu_ticks": 411209,
              "compute_passes": 1,
              "command_buffer_ns": 18000.0206456,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 891458.359547,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 855416.699778,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 933625.036851,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 843583.373353,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 865750.014782,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 861416.62905,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 849624.97931,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 864458.270371,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 857249.950059,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 15583.3549798,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18208.2876563,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18208.345864,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 17958.3439603,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18458.3477676,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18041.6391231,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18458.3477676,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18291.6410267,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18041.6973308,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              13.929036867921875,
              13.365885934031251,
              14.587891200796875,
              13.180990208640626,
              13.52734398096875,
              13.45963482890625,
              13.27539030171875,
              13.507160474546875,
              13.394530469671876
            ],
            "command_buffer_throughput_us_p50": 13.45963482890625,
            "command_buffer_throughput_us_p90": 14.060807734496874,
            "command_buffer_latency_us": [
              15.583354979800001,
              18.208287656299998,
              18.208345863999998,
              17.958343960300002,
              18.4583477676,
              18.0416391231,
              18.4583477676,
              18.2916410267,
              18.0416973308
            ],
            "command_buffer_latency_us_p50": 18.208287656299998,
            "command_buffer_latency_us_p90": 18.4583477676
          },
          "compute_throughput_us": [
            13.76040625,
            13.867828125,
            14.62890625,
            13.537765625,
            13.30078125,
            14.33334375,
            13.197921875,
            14.2923125,
            13.571609375
          ],
          "compute_throughput_us_p50": 13.76040625,
          "compute_throughput_us_p90": 14.392456249999999,
          "command_buffer_throughput_us": [
            13.761067748421876,
            13.867838788421876,
            14.628906683359375,
            13.539712199421874,
            13.300781574796874,
            14.334635125125,
            13.19987040915625,
            14.292318155640626,
            13.572916941484374
          ],
          "command_buffer_throughput_us_p50": 13.761067748421876,
          "command_buffer_throughput_us_p90": 14.393489436771876,
          "compute_latency_us": [
            18.0,
            18.25,
            17.917,
            17.875,
            18.041,
            18.083,
            18.0,
            19.042,
            18.0
          ],
          "compute_latency_us_p50": 18.0,
          "compute_latency_us_p90": 18.4084,
          "command_buffer_latency_us": [
            18.000020645600003,
            18.2916992344,
            18.000020645600003,
            17.8750487976,
            18.083374016,
            18.1249342859,
            18.083257600699998,
            19.0417049453,
            18.000020645600003
          ],
          "command_buffer_latency_us_p50": 18.083257600699998,
          "command_buffer_latency_us_p90": 18.44170037658,
          "counter_control_throughput_ratio": 1.0133178592722158,
          "counter_control_latency_ratio": 0.9976899798042366
        },
        "correctness": {
          "max_abs_error": 1.569250362249905e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 14.1664228972,
        "throughput_us_p90": 14.37811028038,
        "latency_us_p50": 261.25,
        "latency_us_p90": 321.842
      },
      "slowdown": 0.42343450139980404,
      "device_compute_slowdown": 0.2914747863086621,
      "valid": true,
      "round": 2,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "softmax_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-jjutyw6u/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "8bb581805080afef9f47b3147742261e73f0bbc7f1d60afd69631b609a5e3f17",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.314667,
        "capture_ms": 0.078333,
        "compile_ms": 50.031875,
        "allocation_upload_ms": 5.362875,
        "cold_call_ms": 5.581625,
        "warmup_ms": 200.21175,
        "download_ms": 3.084541,
        "repetitions": 568,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          51.7043714789,
          51.3982552817,
          51.7032711268,
          51.7450123239,
          52.0239137324,
          51.8711848592,
          51.9851091549,
          51.6663732394,
          52.0373380282
        ],
        "latency_us": [
          297.958,
          271.125,
          257.333,
          275.625,
          287.542,
          263.458,
          261.459,
          284.5,
          247.584
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 1024,
            "reduction_scalar_rounds": 80,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 16384,
              "global_write_bytes": 16384,
              "private_read_bytes": 98304,
              "private_write_bytes": 32768
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 64,
              "global_write_bytes": 64,
              "private_read_bytes": 384,
              "private_write_bytes": 128
            },
            "striped_storage_scalars_per_worker": 32,
            "reduction_operations": 2,
            "reduction_elements": 8192,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 128,
            "normalized_kernel_cost": 2048,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3366375,
              "compute_span_ns": 3366375,
              "calibration_cpu_ns": 3802375,
              "calibration_gpu_ticks": 3802375,
              "compute_passes": 1,
              "command_buffer_ns": 3366458.34614,
              "command_buffers": 2
            },
            {
              "compute_ns": 3294459,
              "compute_span_ns": 3294459,
              "calibration_cpu_ns": 3750666,
              "calibration_gpu_ticks": 3750666,
              "compute_passes": 1,
              "command_buffer_ns": 3294541.67513,
              "command_buffers": 2
            },
            {
              "compute_ns": 3238041,
              "compute_span_ns": 3238041,
              "calibration_cpu_ns": 3693666,
              "calibration_gpu_ticks": 3693666,
              "compute_passes": 1,
              "command_buffer_ns": 3238083.38726,
              "command_buffers": 2
            },
            {
              "compute_ns": 3282083,
              "compute_span_ns": 3282083,
              "calibration_cpu_ns": 3645250,
              "calibration_gpu_ticks": 3645250,
              "compute_passes": 1,
              "command_buffer_ns": 3282208.28669,
              "command_buffers": 2
            },
            {
              "compute_ns": 3247041,
              "compute_span_ns": 3247041,
              "calibration_cpu_ns": 3683500,
              "calibration_gpu_ticks": 3683500,
              "compute_passes": 1,
              "command_buffer_ns": 3247083.33937,
              "command_buffers": 2
            },
            {
              "compute_ns": 3271500,
              "compute_span_ns": 3271500,
              "calibration_cpu_ns": 3684125,
              "calibration_gpu_ticks": 3684125,
              "compute_passes": 1,
              "command_buffer_ns": 3271583.35131,
              "command_buffers": 2
            },
            {
              "compute_ns": 3218292,
              "compute_span_ns": 3218292,
              "calibration_cpu_ns": 3671167,
              "calibration_gpu_ticks": 3671167,
              "compute_passes": 1,
              "command_buffer_ns": 3218333.35329,
              "command_buffers": 2
            },
            {
              "compute_ns": 3217333,
              "compute_span_ns": 3217333,
              "calibration_cpu_ns": 3638417,
              "calibration_gpu_ticks": 3638417,
              "compute_passes": 1,
              "command_buffer_ns": 3217375.02236,
              "command_buffers": 2
            },
            {
              "compute_ns": 3325167,
              "compute_span_ns": 3325167,
              "calibration_cpu_ns": 3742875,
              "calibration_gpu_ticks": 3742875,
              "compute_passes": 1,
              "command_buffer_ns": 3325166.7046,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 56708,
              "compute_span_ns": 56708,
              "calibration_cpu_ns": 511334,
              "calibration_gpu_ticks": 511334,
              "compute_passes": 1,
              "command_buffer_ns": 56791.6431464,
              "command_buffers": 2
            },
            {
              "compute_ns": 64375,
              "compute_span_ns": 64375,
              "calibration_cpu_ns": 509500,
              "calibration_gpu_ticks": 509500,
              "compute_passes": 1,
              "command_buffer_ns": 64416.6138954,
              "command_buffers": 2
            },
            {
              "compute_ns": 58833,
              "compute_span_ns": 58833,
              "calibration_cpu_ns": 466125,
              "calibration_gpu_ticks": 466125,
              "compute_passes": 1,
              "command_buffer_ns": 58916.6884311,
              "command_buffers": 2
            },
            {
              "compute_ns": 56208,
              "compute_span_ns": 56208,
              "calibration_cpu_ns": 419958,
              "calibration_gpu_ticks": 419958,
              "compute_passes": 1,
              "command_buffer_ns": 56249.962654,
              "command_buffers": 2
            },
            {
              "compute_ns": 57208,
              "compute_span_ns": 57208,
              "calibration_cpu_ns": 516334,
              "calibration_gpu_ticks": 516334,
              "compute_passes": 1,
              "command_buffer_ns": 57250.0284761,
              "command_buffers": 2
            },
            {
              "compute_ns": 65750,
              "compute_span_ns": 65750,
              "calibration_cpu_ns": 541000,
              "calibration_gpu_ticks": 541000,
              "compute_passes": 1,
              "command_buffer_ns": 65791.5952615,
              "command_buffers": 2
            },
            {
              "compute_ns": 52542,
              "compute_span_ns": 52542,
              "calibration_cpu_ns": 457125,
              "calibration_gpu_ticks": 457125,
              "compute_passes": 1,
              "command_buffer_ns": 52666.6408405,
              "command_buffers": 2
            },
            {
              "compute_ns": 56708,
              "compute_span_ns": 56708,
              "calibration_cpu_ns": 446833,
              "calibration_gpu_ticks": 446833,
              "compute_passes": 1,
              "command_buffer_ns": 56749.9082536,
              "command_buffers": 2
            },
            {
              "compute_ns": 61792,
              "compute_span_ns": 61792,
              "calibration_cpu_ns": 617084,
              "calibration_gpu_ticks": 617084,
              "compute_passes": 1,
              "command_buffer_ns": 61875.0345893,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 3390583.38432,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3272958.33267,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3260583.38396,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3277666.63395,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3272291.62212,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3252291.64446,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3228500.01976,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3276708.36123,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3274374.99072,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 52875.0242665,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 64874.9992251,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 56166.7256989,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 57249.9702685,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 53125.0261702,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 64583.378844,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 55166.6598767,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 53083.349485,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 52499.9923073,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              52.977865380000004,
              51.13997394796875,
              50.946615374375,
              51.213541155468754,
              51.129556595625,
              50.8170569446875,
              50.44531280875,
              51.19856814421875,
              51.16210923
            ],
            "command_buffer_throughput_us_p50": 51.13997394796875,
            "command_buffer_throughput_us_p90": 51.566406000375004,
            "command_buffer_latency_us": [
              52.8750242665,
              64.8749992251,
              56.1667256989,
              57.2499702685,
              53.1250261702,
              64.583378844,
              55.1666598767,
              53.083349485,
              52.499992307300005
            ],
            "command_buffer_latency_us_p50": 55.1666598767,
            "command_buffer_latency_us_p90": 64.64170292022
          },
          "compute_throughput_us": [
            52.599609375,
            51.475921875,
            50.594390625,
            51.282546875,
            50.735015625,
            51.1171875,
            50.2858125,
            50.270828125,
            51.955734375
          ],
          "compute_throughput_us_p50": 51.1171875,
          "compute_throughput_us_p90": 52.084509374999996,
          "command_buffer_throughput_us": [
            52.600911658437504,
            51.47721367390625,
            50.5950529259375,
            51.28450447953125,
            50.73567717765625,
            51.118489864218745,
            50.28645864515625,
            50.271484724375,
            51.955729759375
          ],
          "command_buffer_throughput_us_p50": 51.118489864218745,
          "command_buffer_throughput_us_p90": 52.0847661391875,
          "compute_latency_us": [
            56.708,
            64.375,
            58.833,
            56.208,
            57.208,
            65.75,
            52.542,
            56.708,
            61.792
          ],
          "compute_latency_us_p50": 57.208,
          "compute_latency_us_p90": 64.65,
          "command_buffer_latency_us": [
            56.7916431464,
            64.4166138954,
            58.916688431100006,
            56.249962654,
            57.2500284761,
            65.79159526149999,
            52.666640840499994,
            56.7499082536,
            61.8750345893
          ],
          "command_buffer_latency_us_p50": 57.2500284761,
          "command_buffer_latency_us_p90": 64.69161016862,
          "counter_control_throughput_ratio": 0.996850962859602,
          "counter_control_latency_ratio": 1.0489607093520472
        },
        "correctness": {
          "max_abs_error": 1.2343096304160556e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 51.7450123239,
        "throughput_us_p90": 52.026598591559996,
        "latency_us_p50": 271.125,
        "latency_us_p90": 289.6252
      },
      "torch": {
        "cold_call_ms": 4.712459,
        "warmup_ms": 200.893875,
        "repetitions": 188,
        "throughput_us": [
          129.2247340425532,
          128.2165319148936,
          129.0159574468085,
          128.6626755319149,
          128.60172872340425,
          128.82602127659572,
          128.9033670212766,
          128.81449468085106,
          128.86059574468084
        ],
        "latency_us": [
          610.875,
          559.5830000000001,
          408.08299999999997,
          400.666,
          376.25,
          378.20799999999997,
          380.083,
          363.625,
          392.792
        ],
        "throughput_us_p50": 128.82602127659572,
        "throughput_us_p90": 129.05771276595743,
        "latency_us_p50": 392.792,
        "latency_us_p90": 569.8414,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 7700416.666921228,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7753416.779451072,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7733583.392109722,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7748999.982140958,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7683083.473239094,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7756708.306260407,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7706499.833147973,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7637499.9480322,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7661291.630938649,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 139374.98442828655,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 121749.99574199319,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 123208.3304785192,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 138041.67974740267,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 125208.34570750594,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 128041.66181012988,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 136000.0460408628,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 121833.34911242127,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 128666.63746535778,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              120.31901042064419,
              121.147137178923,
              120.8372405017144,
              121.07812472095247,
              120.04817926936084,
              121.19856728531886,
              120.41405989293708,
              119.33593668800313,
              119.7076817334164
            ],
            "command_buffer_throughput_us_p50": 120.41405989293708,
            "command_buffer_throughput_us_p90": 121.15742320020217,
            "command_buffer_latency_us": [
              139.37498442828655,
              121.74999574199319,
              123.2083304785192,
              138.04167974740267,
              125.20834570750594,
              128.04166181012988,
              136.0000460408628,
              121.83334911242127,
              128.66663746535778
            ],
            "command_buffer_latency_us_p50": 128.04166181012988,
            "command_buffer_latency_us_p90": 138.30834068357944
          },
          "throughput": [
            {
              "compute_ns": 8080335.0,
              "compute_span_ns": 9630042.0,
              "command_buffer_ns": 9386083.402205259,
              "calibration_cpu_ns": 10474583.0,
              "calibration_gpu_ticks": 10474583.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8160040.0,
              "compute_span_ns": 9751500.0,
              "command_buffer_ns": 9502083.237748593,
              "calibration_cpu_ns": 10481292.0,
              "calibration_gpu_ticks": 10481292.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8180372.0,
              "compute_span_ns": 9674083.0,
              "command_buffer_ns": 9430083.27158168,
              "calibration_cpu_ns": 10332292.0,
              "calibration_gpu_ticks": 10332292.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8165379.0,
              "compute_span_ns": 9715125.0,
              "command_buffer_ns": 9450250.130612403,
              "calibration_cpu_ns": 10442667.0,
              "calibration_gpu_ticks": 10442667.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8111957.0,
              "compute_span_ns": 9652375.0,
              "command_buffer_ns": 9399166.621733457,
              "calibration_cpu_ns": 11048209.0,
              "calibration_gpu_ticks": 11048209.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8129953.0,
              "compute_span_ns": 9637958.0,
              "command_buffer_ns": 9390041.755978018,
              "calibration_cpu_ns": 10307834.0,
              "calibration_gpu_ticks": 10307834.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8133249.0,
              "compute_span_ns": 9685000.0,
              "command_buffer_ns": 9435166.663024575,
              "calibration_cpu_ns": 10446500.0,
              "calibration_gpu_ticks": 10446500.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8257953.0,
              "compute_span_ns": 9799667.0,
              "command_buffer_ns": 9554749.995004386,
              "calibration_cpu_ns": 10518042.0,
              "calibration_gpu_ticks": 10518042.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8186126.0,
              "compute_span_ns": 9754333.0,
              "command_buffer_ns": 9510708.44894275,
              "calibration_cpu_ns": 10583708.0,
              "calibration_gpu_ticks": 10583708.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 135708.0,
              "compute_span_ns": 135708.0,
              "command_buffer_ns": 135708.3092443645,
              "calibration_cpu_ns": 683916.0,
              "calibration_gpu_ticks": 683916.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 128292.0,
              "compute_span_ns": 128292.0,
              "command_buffer_ns": 128291.66371375322,
              "calibration_cpu_ns": 606583.0,
              "calibration_gpu_ticks": 606583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 130959.0,
              "compute_span_ns": 130959.0,
              "command_buffer_ns": 131000.00796839595,
              "calibration_cpu_ns": 674541.0,
              "calibration_gpu_ticks": 674541.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 138292.0,
              "compute_span_ns": 138292.0,
              "command_buffer_ns": 138333.35833624005,
              "calibration_cpu_ns": 625166.0,
              "calibration_gpu_ticks": 625166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 132042.0,
              "compute_span_ns": 132042.0,
              "command_buffer_ns": 132083.3107456565,
              "calibration_cpu_ns": 652125.0,
              "calibration_gpu_ticks": 652125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 119625.0,
              "compute_span_ns": 119625.0,
              "command_buffer_ns": 119625.00866502523,
              "calibration_cpu_ns": 598625.0,
              "calibration_gpu_ticks": 598625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 144875.0,
              "compute_span_ns": 144875.0,
              "command_buffer_ns": 144875.0263080001,
              "calibration_cpu_ns": 677750.0,
              "calibration_gpu_ticks": 677750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 133375.0,
              "compute_span_ns": 133375.0,
              "command_buffer_ns": 133374.99694898725,
              "calibration_cpu_ns": 588167.0,
              "calibration_gpu_ticks": 588167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 133209.0,
              "compute_span_ns": 133209.0,
              "command_buffer_ns": 133250.02510100603,
              "calibration_cpu_ns": 629375.0,
              "calibration_gpu_ticks": 629375.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            126.255234375,
            127.500625,
            127.8183125,
            127.584046875,
            126.749328125,
            127.030515625,
            127.082015625,
            129.030515625,
            127.90821875
          ],
          "compute_throughput_us_p50": 127.500625,
          "compute_throughput_us_p90": 128.132678125,
          "command_buffer_throughput_us": [
            146.65755315945717,
            148.47005058982177,
            147.34505111846374,
            147.6601582908188,
            146.86197846458526,
            146.71940243715653,
            147.42447910975898,
            149.29296867194353,
            148.60481951473048
          ],
          "command_buffer_throughput_us_p50": 147.42447910975898,
          "command_buffer_throughput_us_p90": 148.7424493461731,
          "compute_latency_us": [
            135.708,
            128.292,
            130.959,
            138.292,
            132.042,
            119.625,
            144.875,
            133.375,
            133.209
          ],
          "compute_latency_us_p50": 133.209,
          "compute_latency_us_p90": 139.6086,
          "command_buffer_latency_us": [
            135.7083092443645,
            128.29166371375322,
            131.00000796839595,
            138.33335833624005,
            132.0833107456565,
            119.62500866502523,
            144.8750263080001,
            133.37499694898725,
            133.25002510100603
          ],
          "command_buffer_latency_us_p50": 133.25002510100603,
          "command_buffer_latency_us_p90": 139.64169193059206,
          "counter_control_throughput_ratio": 1.2233586494890551,
          "counter_control_latency_ratio": 1.0537303342960507
        },
        "download_ms": 0.6765,
        "allocation_upload_ms": 16.453292,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 9.74135319392784e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.4016658421267311,
      "device_compute_slowdown": 0.4009171523669002,
      "valid": true,
      "round": 3,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "softmax_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.544458,
        "warmup_ms": 201.32975,
        "repetitions": 227,
        "throughput_us": [
          127.64959911894273,
          128.267436123348,
          129.652718061674,
          127.11711013215859,
          127.72742290748899,
          128.22118502202642,
          128.2628502202643,
          127.7778986784141,
          128.52220704845814
        ],
        "latency_us": [
          416.208,
          372.25,
          413.584,
          398.584,
          474.208,
          388.5,
          368.16700000000003,
          370.333,
          372.625
        ],
        "throughput_us_p50": 128.22118502202642,
        "throughput_us_p90": 128.74830925110132,
        "latency_us_p50": 388.5,
        "latency_us_p90": 427.80800000000005,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 7649916.806258261,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7725500.036031008,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7768791.692797095,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7664000.033400953,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7729083.416052163,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7681875.08219853,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7794916.62979126,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7813124.975655228,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7752625.038847327,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 131833.30884203315,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 123000.0052601099,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 126500.0319108367,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 123750.01097097993,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 145833.2990296185,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 137166.6439808905,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 142583.33249017596,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 127000.03571808338,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 125125.05054473877,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              119.52995009778533,
              120.7109380629845,
              121.3873701999546,
              119.75000052188989,
              120.76692837581504,
              120.02929815935204,
              121.79557234048843,
              122.08007774461294,
              121.13476623198949
            ],
            "command_buffer_throughput_us_p50": 120.76692837581504,
            "command_buffer_throughput_us_p90": 121.85247342131333,
            "command_buffer_latency_us": [
              131.83330884203315,
              123.0000052601099,
              126.5000319108367,
              123.75001097097993,
              145.8332990296185,
              137.1666439808905,
              142.58333249017596,
              127.00003571808338,
              125.12505054473877
            ],
            "command_buffer_latency_us_p50": 127.00003571808338,
            "command_buffer_latency_us_p90": 143.23332579806447
          },
          "throughput": [
            {
              "compute_ns": 8008166.0,
              "compute_span_ns": 9495541.0,
              "command_buffer_ns": 9262624.836992472,
              "calibration_cpu_ns": 10368375.0,
              "calibration_gpu_ticks": 10368375.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8043290.0,
              "compute_span_ns": 9473417.0,
              "command_buffer_ns": 9237583.260983229,
              "calibration_cpu_ns": 10382292.0,
              "calibration_gpu_ticks": 10382292.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 7949253.0,
              "compute_span_ns": 9464334.0,
              "command_buffer_ns": 9235083.358362317,
              "calibration_cpu_ns": 10233875.0,
              "calibration_gpu_ticks": 10233875.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8067002.0,
              "compute_span_ns": 9513333.0,
              "command_buffer_ns": 9297041.804529727,
              "calibration_cpu_ns": 10325708.0,
              "calibration_gpu_ticks": 10325708.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8156919.0,
              "compute_span_ns": 9619125.0,
              "command_buffer_ns": 9414666.623342782,
              "calibration_cpu_ns": 10359375.0,
              "calibration_gpu_ticks": 10359375.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8118873.0,
              "compute_span_ns": 9698875.0,
              "command_buffer_ns": 9451624.937355518,
              "calibration_cpu_ns": 10370375.0,
              "calibration_gpu_ticks": 10370375.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8111586.0,
              "compute_span_ns": 9679459.0,
              "command_buffer_ns": 9433916.711714119,
              "calibration_cpu_ns": 10509625.0,
              "calibration_gpu_ticks": 10509625.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8136915.0,
              "compute_span_ns": 9589166.0,
              "command_buffer_ns": 9375541.529152542,
              "calibration_cpu_ns": 10467125.0,
              "calibration_gpu_ticks": 10467125.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8055881.0,
              "compute_span_ns": 9589250.0,
              "command_buffer_ns": 9342833.422124386,
              "calibration_cpu_ns": 10417750.0,
              "calibration_gpu_ticks": 10417750.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 128208.0,
              "compute_span_ns": 128208.0,
              "command_buffer_ns": 128208.31034332514,
              "calibration_cpu_ns": 739291.0,
              "calibration_gpu_ticks": 739291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 118125.0,
              "compute_span_ns": 118125.0,
              "command_buffer_ns": 118124.99724328518,
              "calibration_cpu_ns": 629333.0,
              "calibration_gpu_ticks": 629333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 118708.0,
              "compute_span_ns": 118708.0,
              "command_buffer_ns": 118666.6777357459,
              "calibration_cpu_ns": 636833.0,
              "calibration_gpu_ticks": 636833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 125542.0,
              "compute_span_ns": 125542.0,
              "command_buffer_ns": 125541.70098155737,
              "calibration_cpu_ns": 723417.0,
              "calibration_gpu_ticks": 723417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 122333.0,
              "compute_span_ns": 122333.0,
              "command_buffer_ns": 122333.35291966796,
              "calibration_cpu_ns": 673333.0,
              "calibration_gpu_ticks": 673333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 125834.0,
              "compute_span_ns": 125834.0,
              "command_buffer_ns": 125874.99804794788,
              "calibration_cpu_ns": 659292.0,
              "calibration_gpu_ticks": 659292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 131708.0,
              "compute_span_ns": 131708.0,
              "command_buffer_ns": 131708.33699405193,
              "calibration_cpu_ns": 672625.0,
              "calibration_gpu_ticks": 672625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 125375.0,
              "compute_span_ns": 125375.0,
              "command_buffer_ns": 125374.9942407012,
              "calibration_cpu_ns": 642584.0,
              "calibration_gpu_ticks": 642584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 125167.0,
              "compute_span_ns": 125167.0,
              "command_buffer_ns": 125166.6690222919,
              "calibration_cpu_ns": 1253583.0,
              "calibration_gpu_ticks": 1253583.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            125.12759375,
            125.67640625,
            124.207078125,
            126.04690625,
            127.451859375,
            126.857390625,
            126.74353125,
            127.139296875,
            125.873140625
          ],
          "compute_throughput_us_p50": 126.04690625,
          "compute_throughput_us_p90": 127.201809375,
          "command_buffer_throughput_us": [
            144.72851307800738,
            144.33723845286295,
            144.2981774744112,
            145.26627819577698,
            147.10416598973097,
            147.68163964617997,
            147.4049486205331,
            146.49283639300847,
            145.98177222069353
          ],
          "command_buffer_throughput_us_p50": 145.98177222069353,
          "command_buffer_throughput_us_p90": 147.46028682566248,
          "compute_latency_us": [
            128.208,
            118.125,
            118.708,
            125.542,
            122.333,
            125.834,
            131.708,
            125.375,
            125.167
          ],
          "compute_latency_us_p50": 125.375,
          "compute_latency_us_p90": 128.908,
          "command_buffer_latency_us": [
            128.20831034332514,
            118.12499724328518,
            118.6666777357459,
            125.54170098155737,
            122.33335291966796,
            125.87499804794788,
            131.70833699405193,
            125.3749942407012,
            125.1666690222919
          ],
          "command_buffer_latency_us_p50": 125.3749942407012,
          "command_buffer_latency_us_p90": 128.9083156734705,
          "counter_control_throughput_ratio": 1.2102652484644663,
          "counter_control_latency_ratio": 0.9603657901760616
        },
        "download_ms": 0.761667,
        "allocation_upload_ms": 0.768042,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 9.74135319392784e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-_t_r10cx/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "117627a7b60c4cee3b9bce9542bc5f280f025f1ad809d52b8d22cb3cdd24d0d1",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.267834,
        "capture_ms": 0.068666,
        "compile_ms": 49.338375,
        "allocation_upload_ms": 5.51575,
        "cold_call_ms": 8.873334,
        "warmup_ms": 200.266292,
        "download_ms": 4.445292,
        "repetitions": 523,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          54.4209674952,
          53.9874130019,
          54.1991720841,
          53.8557992352,
          53.6169521989,
          53.8590669216,
          54.1387017208,
          53.7034741874,
          54.3857552581
        ],
        "latency_us": [
          406.125,
          319.667,
          318.375,
          286.667,
          404.625,
          314.584,
          290.417,
          308.667,
          298.583
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 512,
            "reduction_scalar_rounds": 80,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 16384,
              "global_write_bytes": 16384,
              "private_read_bytes": 98304,
              "private_write_bytes": 32768
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 64,
              "global_write_bytes": 64,
              "private_read_bytes": 384,
              "private_write_bytes": 128
            },
            "striped_storage_scalars_per_worker": 32,
            "reduction_operations": 2,
            "reduction_elements": 8192,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 120,
            "normalized_kernel_cost": 1920,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3556375,
              "compute_span_ns": 3556375,
              "calibration_cpu_ns": 4044292,
              "calibration_gpu_ticks": 4044292,
              "compute_passes": 1,
              "command_buffer_ns": 3556499.95618,
              "command_buffers": 2
            },
            {
              "compute_ns": 3556292,
              "compute_span_ns": 3556292,
              "calibration_cpu_ns": 4053292,
              "calibration_gpu_ticks": 4053292,
              "compute_passes": 1,
              "command_buffer_ns": 3556333.36585,
              "command_buffers": 2
            },
            {
              "compute_ns": 3455042,
              "compute_span_ns": 3455042,
              "calibration_cpu_ns": 3925625,
              "calibration_gpu_ticks": 3925625,
              "compute_passes": 1,
              "command_buffer_ns": 3455083.29337,
              "command_buffers": 2
            },
            {
              "compute_ns": 3454292,
              "compute_span_ns": 3454292,
              "calibration_cpu_ns": 3971958,
              "calibration_gpu_ticks": 3971958,
              "compute_passes": 1,
              "command_buffer_ns": 3454416.64103,
              "command_buffers": 2
            },
            {
              "compute_ns": 3524625,
              "compute_span_ns": 3524625,
              "calibration_cpu_ns": 3992209,
              "calibration_gpu_ticks": 3992209,
              "compute_passes": 1,
              "command_buffer_ns": 3524666.71029,
              "command_buffers": 2
            },
            {
              "compute_ns": 3503417,
              "compute_span_ns": 3503417,
              "calibration_cpu_ns": 4104417,
              "calibration_gpu_ticks": 4104417,
              "compute_passes": 1,
              "command_buffer_ns": 3503458.34158,
              "command_buffers": 2
            },
            {
              "compute_ns": 3467500,
              "compute_span_ns": 3467500,
              "calibration_cpu_ns": 3910083,
              "calibration_gpu_ticks": 3910083,
              "compute_passes": 1,
              "command_buffer_ns": 3467583.33035,
              "command_buffers": 2
            },
            {
              "compute_ns": 3525000,
              "compute_span_ns": 3525000,
              "calibration_cpu_ns": 4001834,
              "calibration_gpu_ticks": 4001834,
              "compute_passes": 1,
              "command_buffer_ns": 3525000.00736,
              "command_buffers": 2
            },
            {
              "compute_ns": 3430833,
              "compute_span_ns": 3430833,
              "calibration_cpu_ns": 3917875,
              "calibration_gpu_ticks": 3917875,
              "compute_passes": 1,
              "command_buffer_ns": 3430958.3134,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 68208,
              "compute_span_ns": 68208,
              "calibration_cpu_ns": 513875,
              "calibration_gpu_ticks": 513875,
              "compute_passes": 1,
              "command_buffer_ns": 68208.319135,
              "command_buffers": 2
            },
            {
              "compute_ns": 56500,
              "compute_span_ns": 56500,
              "calibration_cpu_ns": 471875,
              "calibration_gpu_ticks": 471875,
              "compute_passes": 1,
              "command_buffer_ns": 56500.0227652,
              "command_buffers": 2
            },
            {
              "compute_ns": 55833,
              "compute_span_ns": 55833,
              "calibration_cpu_ns": 455584,
              "calibration_gpu_ticks": 455584,
              "compute_passes": 1,
              "command_buffer_ns": 55916.6073799,
              "command_buffers": 2
            },
            {
              "compute_ns": 54708,
              "compute_span_ns": 54708,
              "calibration_cpu_ns": 512209,
              "calibration_gpu_ticks": 512209,
              "compute_passes": 1,
              "command_buffer_ns": 54708.3327547,
              "command_buffers": 2
            },
            {
              "compute_ns": 66208,
              "compute_span_ns": 66208,
              "calibration_cpu_ns": 530750,
              "calibration_gpu_ticks": 530750,
              "compute_passes": 1,
              "command_buffer_ns": 66250.0387989,
              "command_buffers": 2
            },
            {
              "compute_ns": 54959,
              "compute_span_ns": 54959,
              "calibration_cpu_ns": 443833,
              "calibration_gpu_ticks": 443833,
              "compute_passes": 1,
              "command_buffer_ns": 55083.3647139,
              "command_buffers": 2
            },
            {
              "compute_ns": 57292,
              "compute_span_ns": 57292,
              "calibration_cpu_ns": 476500,
              "calibration_gpu_ticks": 476500,
              "compute_passes": 1,
              "command_buffer_ns": 57333.3818465,
              "command_buffers": 2
            },
            {
              "compute_ns": 56500,
              "compute_span_ns": 56500,
              "calibration_cpu_ns": 514125,
              "calibration_gpu_ticks": 514125,
              "compute_passes": 1,
              "command_buffer_ns": 56500.0227652,
              "command_buffers": 2
            },
            {
              "compute_ns": 70042,
              "compute_span_ns": 70042,
              "calibration_cpu_ns": 453833,
              "calibration_gpu_ticks": 453833,
              "compute_passes": 1,
              "command_buffer_ns": 70166.6576788,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 3545250.04514,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3444541.71136,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3487291.68763,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3442583.37282,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3416875.02805,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3499833.34308,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3457541.69393,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3450999.96775,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3464333.36381,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 66958.3096169,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 55000.0113435,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 55833.3704248,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54916.6579731,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 70458.27806,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 56624.9946132,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 55499.9569431,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 55458.3384655,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 74666.691944,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              55.3945319553125,
              53.82096424,
              54.48893261921875,
              53.7903652003125,
              53.38867231328125,
              54.684895985625,
              54.024088967656255,
              53.92187449609375,
              54.13020880953125
            ],
            "command_buffer_throughput_us_p50": 54.024088967656255,
            "command_buffer_throughput_us_p90": 54.826823179562496,
            "command_buffer_latency_us": [
              66.95830961690001,
              55.0000113435,
              55.8333704248,
              54.9166579731,
              70.45827806,
              56.624994613199995,
              55.499956943099995,
              55.4583384655,
              74.66669194400001
            ],
            "command_buffer_latency_us_p50": 55.8333704248,
            "command_buffer_latency_us_p90": 71.2999608368
          },
          "compute_throughput_us": [
            55.568359375,
            55.5670625,
            53.98503125,
            53.9733125,
            55.072265625,
            54.740890625,
            54.1796875,
            55.078125,
            53.606765625
          ],
          "compute_throughput_us_p50": 54.740890625,
          "compute_throughput_us_p90": 55.567321875,
          "command_buffer_throughput_us": [
            55.5703118153125,
            55.56770884140625,
            53.98567645890625,
            53.97526001609375,
            55.072917348281244,
            54.7415365871875,
            54.18098953671875,
            55.078125115000006,
            53.608723646875
          ],
          "command_buffer_throughput_us_p50": 54.7415365871875,
          "command_buffer_throughput_us_p90": 55.5682294361875,
          "compute_latency_us": [
            68.208,
            56.5,
            55.833,
            54.708,
            66.208,
            54.959,
            57.292,
            56.5,
            70.042
          ],
          "compute_latency_us_p50": 56.5,
          "compute_latency_us_p90": 68.5748,
          "command_buffer_latency_us": [
            68.208319135,
            56.5000227652,
            55.9166073799,
            54.7083327547,
            66.25003879890001,
            55.0833647139,
            57.333381846500004,
            56.5000227652,
            70.1666576788
          ],
          "command_buffer_latency_us_p50": 56.5000227652,
          "command_buffer_latency_us_p90": 68.59998684375999,
          "counter_control_throughput_ratio": 1.003173234862636,
          "counter_control_latency_ratio": 1.0014908101457374
        },
        "correctness": {
          "max_abs_error": 1.2343096304160556e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 53.9874130019,
        "throughput_us_p90": 54.392797705519996,
        "latency_us_p50": 314.584,
        "latency_us_p90": 404.925
      },
      "slowdown": 0.4210490878915664,
      "device_compute_slowdown": 0.43428983902569995,
      "valid": true,
      "round": 3,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "rmsnorm_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-csh4nehf/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "e81310ef6c7b3435af776cabf332187d02f0ddc801b9234c32af8a56116f248b",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 18.890791,
        "capture_ms": 0.064667,
        "compile_ms": 59.162875,
        "allocation_upload_ms": 1.5205,
        "cold_call_ms": 2.111292,
        "warmup_ms": 200.187709,
        "download_ms": 0.28575,
        "repetitions": 5680,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          4.84880422535,
          4.74087429577,
          4.43324524648,
          4.71177376761,
          4.58392024648,
          4.79148327465,
          4.64053697183,
          4.72248380282,
          4.75699823944
        ],
        "latency_us": [
          411.125,
          198.542,
          178,
          175.042,
          184.666,
          356.167,
          214.875,
          209.833,
          203.333
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 19,
            "reduction_scalar_rounds": 24,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 12296,
              "global_write_bytes": 6148,
              "private_read_bytes": 18444,
              "private_write_bytes": 6148
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 64,
              "global_write_bytes": 32,
              "private_read_bytes": 96,
              "private_write_bytes": 32
            },
            "striped_storage_scalars_per_worker": 8,
            "reduction_operations": 1,
            "reduction_elements": 1537,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 48,
            "normalized_kernel_cost": 48,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 3074,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 321333,
              "compute_span_ns": 321333,
              "calibration_cpu_ns": 750792,
              "calibration_gpu_ticks": 750792,
              "compute_passes": 1,
              "command_buffer_ns": 321375.03149,
              "command_buffers": 2
            },
            {
              "compute_ns": 336333,
              "compute_span_ns": 336333,
              "calibration_cpu_ns": 799709,
              "calibration_gpu_ticks": 799709,
              "compute_passes": 1,
              "command_buffer_ns": 336374.971084,
              "command_buffers": 2
            },
            {
              "compute_ns": 343250,
              "compute_span_ns": 343250,
              "calibration_cpu_ns": 775250,
              "calibration_gpu_ticks": 775250,
              "compute_passes": 1,
              "command_buffer_ns": 343249.99433,
              "command_buffers": 2
            },
            {
              "compute_ns": 330625,
              "compute_span_ns": 330625,
              "calibration_cpu_ns": 703833,
              "calibration_gpu_ticks": 703833,
              "compute_passes": 1,
              "command_buffer_ns": 330624.985509,
              "command_buffers": 2
            },
            {
              "compute_ns": 265791,
              "compute_span_ns": 265791,
              "calibration_cpu_ns": 745708,
              "calibration_gpu_ticks": 745708,
              "compute_passes": 1,
              "command_buffer_ns": 265833.339654,
              "command_buffers": 2
            },
            {
              "compute_ns": 290167,
              "compute_span_ns": 290167,
              "calibration_cpu_ns": 779167,
              "calibration_gpu_ticks": 779167,
              "compute_passes": 1,
              "command_buffer_ns": 290291.674901,
              "command_buffers": 2
            },
            {
              "compute_ns": 290958,
              "compute_span_ns": 290958,
              "calibration_cpu_ns": 700667,
              "calibration_gpu_ticks": 700667,
              "compute_passes": 1,
              "command_buffer_ns": 290999.945719,
              "command_buffers": 2
            },
            {
              "compute_ns": 339084,
              "compute_span_ns": 339084,
              "calibration_cpu_ns": 732333,
              "calibration_gpu_ticks": 732333,
              "compute_passes": 1,
              "command_buffer_ns": 339208.345395,
              "command_buffers": 2
            },
            {
              "compute_ns": 298334,
              "compute_span_ns": 298334,
              "calibration_cpu_ns": 711750,
              "calibration_gpu_ticks": 711750,
              "compute_passes": 1,
              "command_buffer_ns": 298416.707665,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 6792,
              "compute_span_ns": 6792,
              "calibration_cpu_ns": 388209,
              "calibration_gpu_ticks": 388209,
              "compute_passes": 1,
              "command_buffer_ns": 6958.26020092,
              "command_buffers": 2
            },
            {
              "compute_ns": 7042,
              "compute_span_ns": 7042,
              "calibration_cpu_ns": 441125,
              "calibration_gpu_ticks": 441125,
              "compute_passes": 1,
              "command_buffer_ns": 7208.3203122,
              "command_buffers": 2
            },
            {
              "compute_ns": 7000,
              "compute_span_ns": 7000,
              "calibration_cpu_ns": 360458,
              "calibration_gpu_ticks": 360458,
              "compute_passes": 1,
              "command_buffer_ns": 6999.99509379,
              "command_buffers": 2
            },
            {
              "compute_ns": 7042,
              "compute_span_ns": 7042,
              "calibration_cpu_ns": 464708,
              "calibration_gpu_ticks": 464708,
              "compute_passes": 1,
              "command_buffer_ns": 7166.70183465,
              "command_buffers": 2
            },
            {
              "compute_ns": 7125,
              "compute_span_ns": 7125,
              "calibration_cpu_ns": 369917,
              "calibration_gpu_ticks": 369917,
              "compute_passes": 1,
              "command_buffer_ns": 7125.02514943,
              "command_buffers": 2
            },
            {
              "compute_ns": 6917,
              "compute_span_ns": 6917,
              "calibration_cpu_ns": 343292,
              "calibration_gpu_ticks": 343292,
              "compute_passes": 1,
              "command_buffer_ns": 6916.64172336,
              "command_buffers": 2
            },
            {
              "compute_ns": 14333,
              "compute_span_ns": 14333,
              "calibration_cpu_ns": 340208,
              "calibration_gpu_ticks": 340208,
              "compute_passes": 1,
              "command_buffer_ns": 14374.9639392,
              "command_buffers": 2
            },
            {
              "compute_ns": 7042,
              "compute_span_ns": 7042,
              "calibration_cpu_ns": 359125,
              "calibration_gpu_ticks": 359125,
              "compute_passes": 1,
              "command_buffer_ns": 7124.96694177,
              "command_buffers": 2
            },
            {
              "compute_ns": 7083,
              "compute_span_ns": 7083,
              "calibration_cpu_ns": 364042,
              "calibration_gpu_ticks": 364042,
              "compute_passes": 1,
              "command_buffer_ns": 7125.02514943,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 308624.992613,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 318083.330058,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 324999.971781,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 264458.30008,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 273250.043392,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 282291.613985,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 293458.346277,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 310083.385557,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 290875.032078,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 7166.64362699,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7124.96694177,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7083.34846422,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7208.3203122,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7208.3203122,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7458.26400816,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7333.35036784,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7333.29216018,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7375.02705306,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              4.822265509578124,
              4.97005203215625,
              5.078124559078125,
              4.13216093875,
              4.269531928,
              4.410806468515625,
              4.585286660578125,
              4.845052899328125,
              4.54492237621875
            ],
            "command_buffer_throughput_us_p50": 4.585286660578125,
            "command_buffer_throughput_us_p90": 4.991666537540625,
            "command_buffer_latency_us": [
              7.16664362699,
              7.1249669417699995,
              7.08334846422,
              7.208320312200001,
              7.208320312200001,
              7.4582640081600005,
              7.3333503678400005,
              7.33329216018,
              7.37502705306
            ],
            "command_buffer_latency_us_p50": 7.208320312200001,
            "command_buffer_latency_us_p90": 7.39167444408
          },
          "compute_throughput_us": [
            5.020828125,
            5.255203125,
            5.36328125,
            5.166015625,
            4.152984375,
            4.533859375,
            4.54621875,
            5.2981875,
            4.66146875
          ],
          "compute_throughput_us_p50": 5.020828125,
          "compute_throughput_us_p90": 5.31120625,
          "command_buffer_throughput_us": [
            5.02148486703125,
            5.255858923187501,
            5.36328116140625,
            5.166015398578125,
            4.15364593209375,
            4.5358074203281245,
            4.546874151859376,
            5.300130396796876,
            4.662761057265625
          ],
          "command_buffer_throughput_us_p50": 5.02148486703125,
          "command_buffer_throughput_us_p90": 5.3127605497187504,
          "compute_latency_us": [
            6.792,
            7.042,
            7.0,
            7.042,
            7.125,
            6.917,
            14.333,
            7.042,
            7.083
          ],
          "compute_latency_us_p50": 7.042,
          "compute_latency_us_p90": 8.566600000000001,
          "command_buffer_latency_us": [
            6.95826020092,
            7.208320312200001,
            6.99999509379,
            7.1667018346499995,
            7.12502514943,
            6.91664172336,
            14.3749639392,
            7.1249669417699995,
            7.12502514943
          ],
          "command_buffer_latency_us_p50": 7.12502514943,
          "command_buffer_latency_us_p90": 8.641649037600002,
          "counter_control_throughput_ratio": 1.0413123991321984,
          "counter_control_latency_ratio": 0.9882324904879322
        },
        "correctness": {
          "max_abs_error": 1.5564224420394623e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 4.72248380282,
        "throughput_us_p90": 4.80294746479,
        "latency_us_p50": 203.333,
        "latency_us_p90": 367.1586
      },
      "torch": {
        "cold_call_ms": 3.192167,
        "warmup_ms": 200.080583,
        "repetitions": 3099,
        "throughput_us": [
          8.063958696353662,
          8.14070409809616,
          7.5278315585672795,
          7.219613746369796,
          7.3845730880929334,
          7.489579864472411,
          7.422878347854146,
          7.743290738948048,
          7.851565020974508
        ],
        "latency_us": [
          280.917,
          216.667,
          220.458,
          228.959,
          217.625,
          224.5,
          229.042,
          235.875,
          229.333
        ],
        "throughput_us_p50": 7.5278315585672795,
        "throughput_us_p90": 8.079307776702162,
        "latency_us_p50": 228.959,
        "latency_us_p90": 244.8834,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 364208.30292627215,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 347083.3180472255,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 354416.6684150696,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 289708.37593078613,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 303041.71377792954,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 290166.64484515786,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 289625.02256035805,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 304333.3417735994,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 289625.02256035805,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 8041.621185839176,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8125.032763928175,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8000.002708286047,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8124.9745562672615,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8000.002708286047,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8166.651241481304,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8083.297871053219,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8124.9745562672615,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8125.032763928175,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              5.690754733223002,
              5.423176844487898,
              5.537760443985462,
              4.526693373918533,
              4.735026777780149,
              4.533853825705592,
              4.5253909775055945,
              4.75520846521249,
              4.5253909775055945
            ],
            "command_buffer_throughput_us_p50": 4.735026777780149,
            "command_buffer_throughput_us_p90": 5.56835930183297,
            "command_buffer_latency_us": [
              8.041621185839176,
              8.125032763928175,
              8.000002708286047,
              8.124974556267262,
              8.000002708286047,
              8.166651241481304,
              8.083297871053219,
              8.124974556267262,
              8.125032763928175
            ],
            "command_buffer_latency_us_p50": 8.124974556267262,
            "command_buffer_latency_us_p90": 8.1333564594388
          },
          "throughput": [
            {
              "compute_ns": 356042.0,
              "compute_span_ns": 356042.0,
              "command_buffer_ns": 356083.3283700049,
              "calibration_cpu_ns": 960042.0,
              "calibration_gpu_ticks": 960042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 368667.0,
              "compute_span_ns": 368667.0,
              "command_buffer_ns": 368708.3371914923,
              "calibration_cpu_ns": 963042.0,
              "calibration_gpu_ticks": 963042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 368167.0,
              "compute_span_ns": 368167.0,
              "command_buffer_ns": 368166.6566990316,
              "calibration_cpu_ns": 956000.0,
              "calibration_gpu_ticks": 956000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 304000.0,
              "compute_span_ns": 304000.0,
              "command_buffer_ns": 303999.98649954796,
              "calibration_cpu_ns": 862500.0,
              "calibration_gpu_ticks": 862500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 304333.0,
              "compute_span_ns": 304333.0,
              "command_buffer_ns": 304333.2835659385,
              "calibration_cpu_ns": 873833.0,
              "calibration_gpu_ticks": 873833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 310166.0,
              "compute_span_ns": 310166.0,
              "command_buffer_ns": 310125.0040344894,
              "calibration_cpu_ns": 942708.0,
              "calibration_gpu_ticks": 942708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 288708.0,
              "compute_span_ns": 288708.0,
              "command_buffer_ns": 288666.6334234178,
              "calibration_cpu_ns": 881459.0,
              "calibration_gpu_ticks": 881459.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 303584.0,
              "compute_span_ns": 303584.0,
              "command_buffer_ns": 303625.0127479434,
              "calibration_cpu_ns": 904708.0,
              "calibration_gpu_ticks": 904708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 290667.0,
              "compute_span_ns": 290667.0,
              "command_buffer_ns": 290666.64865240455,
              "calibration_cpu_ns": 942084.0,
              "calibration_gpu_ticks": 942084.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 8125.0,
              "compute_span_ns": 8125.0,
              "command_buffer_ns": 8124.9745562672615,
              "calibration_cpu_ns": 407583.0,
              "calibration_gpu_ticks": 407583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8250.0,
              "compute_span_ns": 8250.0,
              "command_buffer_ns": 8250.00461190939,
              "calibration_cpu_ns": 493875.0,
              "calibration_gpu_ticks": 493875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8292.0,
              "compute_span_ns": 8292.0,
              "command_buffer_ns": 8333.357982337475,
              "calibration_cpu_ns": 428833.0,
              "calibration_gpu_ticks": 428833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8125.0,
              "compute_span_ns": 8125.0,
              "command_buffer_ns": 8124.9745562672615,
              "calibration_cpu_ns": 393208.0,
              "calibration_gpu_ticks": 393208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8125.0,
              "compute_span_ns": 8125.0,
              "command_buffer_ns": 8124.9745562672615,
              "calibration_cpu_ns": 385833.0,
              "calibration_gpu_ticks": 385833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8250.0,
              "compute_span_ns": 8250.0,
              "command_buffer_ns": 8250.00461190939,
              "calibration_cpu_ns": 488750.0,
              "calibration_gpu_ticks": 488750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8666.0,
              "compute_span_ns": 8666.0,
              "command_buffer_ns": 8624.978363513947,
              "calibration_cpu_ns": 442625.0,
              "calibration_gpu_ticks": 442625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8167.0,
              "compute_span_ns": 8167.0,
              "command_buffer_ns": 8166.651241481304,
              "calibration_cpu_ns": 387792.0,
              "calibration_gpu_ticks": 387792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8209.0,
              "compute_span_ns": 8209.0,
              "command_buffer_ns": 8250.00461190939,
              "calibration_cpu_ns": 387917.0,
              "calibration_gpu_ticks": 387917.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            5.56315625,
            5.760421875,
            5.752609375,
            4.75,
            4.755203125,
            4.84634375,
            4.5110625,
            4.7435,
            4.541671875
          ],
          "compute_throughput_us_p50": 4.755203125,
          "compute_throughput_us_p90": 5.754171875,
          "command_buffer_throughput_us": [
            5.563802005781326,
            5.7610677686170675,
            5.752604010922369,
            4.749999789055437,
            4.755207555717789,
            4.845703188038897,
            4.510416147240903,
            4.744140824186616,
            4.541666385193821
          ],
          "command_buffer_throughput_us_p50": 4.755207555717789,
          "command_buffer_throughput_us_p90": 5.754296762461308,
          "compute_latency_us": [
            8.125,
            8.25,
            8.292,
            8.125,
            8.125,
            8.25,
            8.666,
            8.167,
            8.209
          ],
          "compute_latency_us_p50": 8.209,
          "compute_latency_us_p90": 8.3668,
          "command_buffer_latency_us": [
            8.124974556267262,
            8.25000461190939,
            8.333357982337475,
            8.124974556267262,
            8.124974556267262,
            8.25000461190939,
            8.624978363513947,
            8.166651241481304,
            8.25000461190939
          ],
          "command_buffer_latency_us_p50": 8.25000461190939,
          "command_buffer_latency_us_p90": 8.39168205857277,
          "counter_control_throughput_ratio": 1.0042620198120824,
          "counter_control_latency_ratio": 1.0153810884967798
        },
        "download_ms": 1.072875,
        "allocation_upload_ms": 0.934792,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.807831588873654e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.6273365398891574,
      "device_compute_slowdown": 1.0558598640305192,
      "valid": true,
      "round": 3,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "rmsnorm_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.26,
        "warmup_ms": 200.290542,
        "repetitions": 3452,
        "throughput_us": [
          8.332319235225956,
          8.013458285052144,
          8.499746523754345,
          8.147861239860951,
          8.139303302433373,
          8.034701911935109,
          7.650540845886442,
          7.330388180764774,
          8.011997682502898
        ],
        "latency_us": [
          233.084,
          244.708,
          180.333,
          226.458,
          222.75,
          217.333,
          226.5,
          240.209,
          306.5
        ],
        "throughput_us_p50": 8.034701911935109,
        "throughput_us_p90": 8.365804692931635,
        "latency_us_p50": 226.5,
        "latency_us_p90": 257.0664,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 369458.34290236235,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 360874.98301640153,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 371833.33188295364,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 367166.6490845382,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 370249.9670907855,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 374249.997548759,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 366874.97049570084,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 368541.63045063615,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 370124.9952428043,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 8041.621185839176,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8208.327926695347,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8250.00461190939,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8624.978363513947,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8083.356078714132,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8375.034667551517,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8250.00461190939,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 11041.644029319286,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 8750.008419156075,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              5.772786607849412,
              5.638671609631274,
              5.809895810671151,
              5.73697889194591,
              5.7851557357935235,
              5.847656211699359,
              5.732421413995326,
              5.75846297579119,
              5.783203050668817
            ],
            "command_buffer_throughput_us_p50": 5.772786607849412,
            "command_buffer_throughput_us_p90": 5.817447890876792,
            "command_buffer_latency_us": [
              8.041621185839176,
              8.208327926695347,
              8.25000461190939,
              8.624978363513947,
              8.083356078714132,
              8.375034667551517,
              8.25000461190939,
              11.041644029319286,
              8.750008419156075
            ],
            "command_buffer_latency_us_p50": 8.25000461190939,
            "command_buffer_latency_us_p90": 9.208335541188717
          },
          "throughput": [
            {
              "compute_ns": 366291.0,
              "compute_span_ns": 366291.0,
              "command_buffer_ns": 366249.99484047294,
              "calibration_cpu_ns": 938333.0,
              "calibration_gpu_ticks": 938333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 371584.0,
              "compute_span_ns": 371584.0,
              "command_buffer_ns": 371625.00666454434,
              "calibration_cpu_ns": 873459.0,
              "calibration_gpu_ticks": 873459.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 363125.0,
              "compute_span_ns": 363125.0,
              "command_buffer_ns": 363125.0001490116,
              "calibration_cpu_ns": 951750.0,
              "calibration_gpu_ticks": 951750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 375250.0,
              "compute_span_ns": 375250.0,
              "command_buffer_ns": 375250.00516325235,
              "calibration_cpu_ns": 944375.0,
              "calibration_gpu_ticks": 944375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 376625.0,
              "compute_span_ns": 376625.0,
              "command_buffer_ns": 376624.9865293503,
              "calibration_cpu_ns": 953041.0,
              "calibration_gpu_ticks": 953041.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 366708.0,
              "compute_span_ns": 366708.0,
              "command_buffer_ns": 366708.3219625056,
              "calibration_cpu_ns": 929542.0,
              "calibration_gpu_ticks": 929542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 374834.0,
              "compute_span_ns": 374834.0,
              "command_buffer_ns": 374875.0314116478,
              "calibration_cpu_ns": 917792.0,
              "calibration_gpu_ticks": 917792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 348042.0,
              "compute_span_ns": 348042.0,
              "command_buffer_ns": 348041.6489765048,
              "calibration_cpu_ns": 922834.0,
              "calibration_gpu_ticks": 922834.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 363125.0,
              "compute_span_ns": 363125.0,
              "command_buffer_ns": 363125.0001490116,
              "calibration_cpu_ns": 915875.0,
              "calibration_gpu_ticks": 915875.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 8292.0,
              "compute_span_ns": 8292.0,
              "command_buffer_ns": 8333.357982337475,
              "calibration_cpu_ns": 391791.0,
              "calibration_gpu_ticks": 391791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8292.0,
              "compute_span_ns": 8292.0,
              "command_buffer_ns": 8291.681297123432,
              "calibration_cpu_ns": 367375.0,
              "calibration_gpu_ticks": 367375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8167.0,
              "compute_span_ns": 8167.0,
              "command_buffer_ns": 8208.327926695347,
              "calibration_cpu_ns": 390541.0,
              "calibration_gpu_ticks": 390541.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8125.0,
              "compute_span_ns": 8125.0,
              "command_buffer_ns": 8125.032763928175,
              "calibration_cpu_ns": 408334.0,
              "calibration_gpu_ticks": 408334.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8333.0,
              "compute_span_ns": 8333.0,
              "command_buffer_ns": 8333.299774676561,
              "calibration_cpu_ns": 373291.0,
              "calibration_gpu_ticks": 373291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8208.0,
              "compute_span_ns": 8208.0,
              "command_buffer_ns": 8166.651241481304,
              "calibration_cpu_ns": 374791.0,
              "calibration_gpu_ticks": 374791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8041.0,
              "compute_span_ns": 8041.0,
              "command_buffer_ns": 8000.002708286047,
              "calibration_cpu_ns": 406458.0,
              "calibration_gpu_ticks": 406458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 11416.0,
              "compute_span_ns": 11416.0,
              "command_buffer_ns": 11374.999303370714,
              "calibration_cpu_ns": 456375.0,
              "calibration_gpu_ticks": 456375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 8291.0,
              "compute_span_ns": 8291.0,
              "command_buffer_ns": 8250.00461190939,
              "calibration_cpu_ns": 439959.0,
              "calibration_gpu_ticks": 439959.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            5.723296875,
            5.806,
            5.673828125,
            5.86328125,
            5.884765625,
            5.7298125,
            5.85678125,
            5.43815625,
            5.673828125
          ],
          "compute_throughput_us_p50": 5.7298125,
          "compute_throughput_us_p90": 5.867578125,
          "command_buffer_throughput_us": [
            5.72265616938239,
            5.806640729133505,
            5.673828127328306,
            5.863281330675818,
            5.884765414521098,
            5.72981753066415,
            5.857422365806997,
            5.4381507652578875,
            5.673828127328306
          ],
          "command_buffer_throughput_us_p50": 5.72981753066415,
          "command_buffer_throughput_us_p90": 5.867578147444874,
          "compute_latency_us": [
            8.292,
            8.292,
            8.167,
            8.125,
            8.333,
            8.208,
            8.041,
            11.416,
            8.291
          ],
          "compute_latency_us_p50": 8.291,
          "compute_latency_us_p90": 8.9496,
          "command_buffer_latency_us": [
            8.333357982337475,
            8.291681297123432,
            8.208327926695347,
            8.125032763928175,
            8.333299774676561,
            8.166651241481304,
            8.000002708286047,
            11.374999303370714,
            8.25000461190939
          ],
          "command_buffer_latency_us_p50": 8.25000461190939,
          "command_buffer_latency_us_p90": 8.941686246544123,
          "counter_control_throughput_ratio": 0.9913160762951365,
          "counter_control_latency_ratio": 0.9949482834041232
        },
        "download_ms": 0.283875,
        "allocation_upload_ms": 0.612333,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.807831588873654e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-myi3cl_k/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "54d99166e42515c7bfff1cac173ad46750fa91f9c92e668c07d69318fdc050bf",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.1195,
        "capture_ms": 0.063084,
        "compile_ms": 50.7655,
        "allocation_upload_ms": 1.194083,
        "cold_call_ms": 1.773,
        "warmup_ms": 200.020166,
        "download_ms": 0.287041,
        "repetitions": 7183,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          4.13619532229,
          4.17425978004,
          4.17098245858,
          4.09549157734,
          4.34383266045,
          4.81698106641,
          4.47215063344,
          4.08889043575,
          4.12546415147
        ],
        "latency_us": [
          281.333,
          223.417,
          197.625,
          206.917,
          209.625,
          205.458,
          201.25,
          218.959,
          239.208
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 32,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 37,
            "reduction_scalar_rounds": 24,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 12296,
              "global_write_bytes": 6148,
              "private_read_bytes": 18444,
              "private_write_bytes": 6148
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 64,
              "global_write_bytes": 32,
              "private_read_bytes": 96,
              "private_write_bytes": 32
            },
            "striped_storage_scalars_per_worker": 8,
            "reduction_operations": 1,
            "reduction_elements": 1537,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 56,
            "normalized_kernel_cost": 56,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 3074,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 246208,
              "compute_span_ns": 246208,
              "calibration_cpu_ns": 1049917,
              "calibration_gpu_ticks": 1049917,
              "compute_passes": 1,
              "command_buffer_ns": 246249.954216,
              "command_buffers": 2
            },
            {
              "compute_ns": 267792,
              "compute_span_ns": 267792,
              "calibration_cpu_ns": 718250,
              "calibration_gpu_ticks": 718250,
              "compute_passes": 1,
              "command_buffer_ns": 267833.354883,
              "command_buffers": 2
            },
            {
              "compute_ns": 257875,
              "compute_span_ns": 257875,
              "calibration_cpu_ns": 660666,
              "calibration_gpu_ticks": 660666,
              "compute_passes": 1,
              "command_buffer_ns": 257916.632108,
              "command_buffers": 2
            },
            {
              "compute_ns": 246166,
              "compute_span_ns": 246166,
              "calibration_cpu_ns": 665167,
              "calibration_gpu_ticks": 665167,
              "compute_passes": 1,
              "command_buffer_ns": 246208.335739,
              "command_buffers": 2
            },
            {
              "compute_ns": 265666,
              "compute_span_ns": 265666,
              "calibration_cpu_ns": 696292,
              "calibration_gpu_ticks": 696292,
              "compute_passes": 1,
              "command_buffer_ns": 265666.632913,
              "command_buffers": 2
            },
            {
              "compute_ns": 261875,
              "compute_span_ns": 261875,
              "calibration_cpu_ns": 711875,
              "calibration_gpu_ticks": 711875,
              "compute_passes": 1,
              "command_buffer_ns": 261999.957729,
              "command_buffers": 2
            },
            {
              "compute_ns": 246333,
              "compute_span_ns": 246333,
              "calibration_cpu_ns": 687083,
              "calibration_gpu_ticks": 687083,
              "compute_passes": 1,
              "command_buffer_ns": 246291.689109,
              "command_buffers": 2
            },
            {
              "compute_ns": 267583,
              "compute_span_ns": 267583,
              "calibration_cpu_ns": 717000,
              "calibration_gpu_ticks": 717000,
              "compute_passes": 1,
              "command_buffer_ns": 267624.971457,
              "command_buffers": 2
            },
            {
              "compute_ns": 267334,
              "compute_span_ns": 267334,
              "calibration_cpu_ns": 670417,
              "calibration_gpu_ticks": 670417,
              "compute_passes": 1,
              "command_buffer_ns": 267458.322924,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 6750,
              "compute_span_ns": 6750,
              "calibration_cpu_ns": 432458,
              "calibration_gpu_ticks": 432458,
              "compute_passes": 1,
              "command_buffer_ns": 6833.28835294,
              "command_buffers": 2
            },
            {
              "compute_ns": 6708,
              "compute_span_ns": 6708,
              "calibration_cpu_ns": 379250,
              "calibration_gpu_ticks": 379250,
              "compute_passes": 1,
              "command_buffer_ns": 6791.61166772,
              "command_buffers": 2
            },
            {
              "compute_ns": 7166,
              "compute_span_ns": 7166,
              "calibration_cpu_ns": 524291,
              "calibration_gpu_ticks": 524291,
              "compute_passes": 1,
              "command_buffer_ns": 7208.3203122,
              "command_buffers": 2
            },
            {
              "compute_ns": 6791,
              "compute_span_ns": 6791,
              "calibration_cpu_ns": 460500,
              "calibration_gpu_ticks": 460500,
              "compute_passes": 1,
              "command_buffer_ns": 6791.66987538,
              "command_buffers": 2
            },
            {
              "compute_ns": 6750,
              "compute_span_ns": 6750,
              "calibration_cpu_ns": 413083,
              "calibration_gpu_ticks": 413083,
              "compute_passes": 1,
              "command_buffer_ns": 6749.99319017,
              "command_buffers": 2
            },
            {
              "compute_ns": 6792,
              "compute_span_ns": 6792,
              "calibration_cpu_ns": 396000,
              "calibration_gpu_ticks": 396000,
              "compute_passes": 1,
              "command_buffer_ns": 6916.69993103,
              "command_buffers": 2
            },
            {
              "compute_ns": 6958,
              "compute_span_ns": 6958,
              "calibration_cpu_ns": 401333,
              "calibration_gpu_ticks": 401333,
              "compute_passes": 1,
              "command_buffer_ns": 7041.67177901,
              "command_buffers": 2
            },
            {
              "compute_ns": 6958,
              "compute_span_ns": 6958,
              "calibration_cpu_ns": 374792,
              "calibration_gpu_ticks": 374792,
              "compute_passes": 1,
              "command_buffer_ns": 7041.67177901,
              "command_buffers": 2
            },
            {
              "compute_ns": 7125,
              "compute_span_ns": 7125,
              "calibration_cpu_ns": 365750,
              "calibration_gpu_ticks": 365750,
              "compute_passes": 1,
              "command_buffer_ns": 7208.37851986,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 271583.383437,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 287958.362605,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 246499.95612,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 246666.662861,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 267458.322924,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 245541.625191,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 246666.662861,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 267208.379228,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 246250.012424,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 6916.64172336,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6791.61166772,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6499.93307889,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6749.99319017,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6791.66987538,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6666.63981974,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6874.96503815,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6999.99509379,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6875.08145347,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              4.243490366203125,
              4.499349415703125,
              3.8515618143749997,
              3.854166607203125,
              4.179036295687499,
              3.836587893609375,
              3.854166607203125,
              4.1751309254375,
              3.8476564441249996
            ],
            "command_buffer_throughput_us_p50": 3.854166607203125,
            "command_buffer_throughput_us_p90": 4.294662176103125,
            "command_buffer_latency_us": [
              6.91664172336,
              6.79161166772,
              6.49993307889,
              6.7499931901700005,
              6.79166987538,
              6.66663981974,
              6.87496503815,
              6.99999509379,
              6.87508145347
            ],
            "command_buffer_latency_us_p50": 6.79166987538,
            "command_buffer_latency_us_p90": 6.933312397446
          },
          "compute_throughput_us": [
            3.847,
            4.18425,
            4.029296875,
            3.84634375,
            4.15103125,
            4.091796875,
            3.848953125,
            4.180984375,
            4.17709375
          ],
          "compute_throughput_us_p50": 4.091796875,
          "compute_throughput_us_p90": 4.1816375,
          "command_buffer_throughput_us": [
            3.8476555346250003,
            4.184896170046875,
            4.0299473766875,
            3.847005245921875,
            4.151041139265625,
            4.093749339515624,
            3.848307642328125,
            4.181640179015625,
            4.179036295687499
          ],
          "command_buffer_throughput_us_p50": 4.093749339515624,
          "command_buffer_throughput_us_p90": 4.182291377221875,
          "compute_latency_us": [
            6.75,
            6.708,
            7.166,
            6.791,
            6.75,
            6.792,
            6.958,
            6.958,
            7.125
          ],
          "compute_latency_us_p50": 6.792,
          "compute_latency_us_p90": 7.1332,
          "command_buffer_latency_us": [
            6.8332883529399995,
            6.79161166772,
            7.208320312200001,
            6.79166987538,
            6.7499931901700005,
            6.91669993103,
            7.0416717790100005,
            7.0416717790100005,
            7.20837851986
          ],
          "command_buffer_latency_us_p50": 6.91669993103,
          "command_buffer_latency_us_p90": 7.208331953732,
          "counter_control_throughput_ratio": 0.9984798361170869,
          "counter_control_latency_ratio": 1.0061743299638721
        },
        "correctness": {
          "max_abs_error": 1.5564224420394623e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 4.17098245858,
        "throughput_us_p90": 4.541116720034,
        "latency_us_p50": 209.625,
        "latency_us_p90": 247.633
      },
      "slowdown": 0.5191209959369164,
      "device_compute_slowdown": 0.7141240441986539,
      "valid": true,
      "round": 3,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "rmsnorm_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-kx8a32he/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "b5a32df1046f38df4f8e9cbba5277df92f4beb4d4f9631ac4564ef5cda134264",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 19.112875,
        "capture_ms": 0.067417,
        "compile_ms": 51.234791,
        "allocation_upload_ms": 2.005166,
        "cold_call_ms": 3.434416,
        "warmup_ms": 200.529792,
        "download_ms": 0.701125,
        "repetitions": 1850,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          15.2615313514,
          15.4653151351,
          15.4971848649,
          15.6561259459,
          15.5602475676,
          15.4712610811,
          14.7658108108,
          15.2043243243,
          14.5684459459
        ],
        "latency_us": [
          310.458,
          249.291,
          237.958,
          252.791,
          259.958,
          236.583,
          248.958,
          241.542,
          259.292
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 32,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 256,
            "reduction_scalar_rounds": 36,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 24576,
              "global_write_bytes": 12288,
              "private_read_bytes": 36864,
              "private_write_bytes": 12288
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 48,
              "private_read_bytes": 144,
              "private_write_bytes": 48
            },
            "striped_storage_scalars_per_worker": 12,
            "reduction_operations": 1,
            "reduction_elements": 3072,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 68,
            "normalized_kernel_cost": 272,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 6144,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 911667,
              "compute_span_ns": 911667,
              "calibration_cpu_ns": 1339042,
              "calibration_gpu_ticks": 1339042,
              "compute_passes": 1,
              "command_buffer_ns": 911750.015803,
              "command_buffers": 2
            },
            {
              "compute_ns": 910542,
              "compute_span_ns": 910542,
              "calibration_cpu_ns": 1329333,
              "calibration_gpu_ticks": 1329333,
              "compute_passes": 1,
              "command_buffer_ns": 910583.301447,
              "command_buffers": 2
            },
            {
              "compute_ns": 915834,
              "compute_span_ns": 915834,
              "calibration_cpu_ns": 1357000,
              "calibration_gpu_ticks": 1357000,
              "compute_passes": 1,
              "command_buffer_ns": 916000.048164,
              "command_buffers": 2
            },
            {
              "compute_ns": 921292,
              "compute_span_ns": 921292,
              "calibration_cpu_ns": 1399708,
              "calibration_gpu_ticks": 1399708,
              "compute_passes": 1,
              "command_buffer_ns": 921416.678466,
              "command_buffers": 2
            },
            {
              "compute_ns": 897833,
              "compute_span_ns": 897833,
              "calibration_cpu_ns": 1333250,
              "calibration_gpu_ticks": 1333250,
              "compute_passes": 1,
              "command_buffer_ns": 897833.378986,
              "command_buffers": 2
            },
            {
              "compute_ns": 931417,
              "compute_span_ns": 931417,
              "calibration_cpu_ns": 1341375,
              "calibration_gpu_ticks": 1341375,
              "compute_passes": 1,
              "command_buffer_ns": 931416.638196,
              "command_buffers": 2
            },
            {
              "compute_ns": 906375,
              "compute_span_ns": 906375,
              "calibration_cpu_ns": 1357083,
              "calibration_gpu_ticks": 1357083,
              "compute_passes": 1,
              "command_buffer_ns": 906375.003979,
              "command_buffers": 2
            },
            {
              "compute_ns": 902750,
              "compute_span_ns": 902750,
              "calibration_cpu_ns": 1319250,
              "calibration_gpu_ticks": 1319250,
              "compute_passes": 1,
              "command_buffer_ns": 902833.35885,
              "command_buffers": 2
            },
            {
              "compute_ns": 972959,
              "compute_span_ns": 972959,
              "calibration_cpu_ns": 1385583,
              "calibration_gpu_ticks": 1385583,
              "compute_passes": 1,
              "command_buffer_ns": 973041.635007,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 18750,
              "compute_span_ns": 18750,
              "calibration_cpu_ns": 440166,
              "calibration_gpu_ticks": 440166,
              "compute_passes": 1,
              "command_buffer_ns": 18750.0263564,
              "command_buffers": 2
            },
            {
              "compute_ns": 18458,
              "compute_span_ns": 18458,
              "calibration_cpu_ns": 441625,
              "calibration_gpu_ticks": 441625,
              "compute_passes": 1,
              "command_buffer_ns": 18458.3477676,
              "command_buffers": 2
            },
            {
              "compute_ns": 18500,
              "compute_span_ns": 18500,
              "calibration_cpu_ns": 441916,
              "calibration_gpu_ticks": 441916,
              "compute_passes": 1,
              "command_buffer_ns": 18583.3778232,
              "command_buffers": 2
            },
            {
              "compute_ns": 18667,
              "compute_span_ns": 18667,
              "calibration_cpu_ns": 432708,
              "calibration_gpu_ticks": 432708,
              "compute_passes": 1,
              "command_buffer_ns": 18666.672986,
              "command_buffers": 2
            },
            {
              "compute_ns": 18708,
              "compute_span_ns": 18708,
              "calibration_cpu_ns": 427083,
              "calibration_gpu_ticks": 427083,
              "compute_passes": 1,
              "command_buffer_ns": 18791.644834,
              "command_buffers": 2
            },
            {
              "compute_ns": 18708,
              "compute_span_ns": 18708,
              "calibration_cpu_ns": 416041,
              "calibration_gpu_ticks": 416041,
              "compute_passes": 1,
              "command_buffer_ns": 18749.9099411,
              "command_buffers": 2
            },
            {
              "compute_ns": 18500,
              "compute_span_ns": 18500,
              "calibration_cpu_ns": 404625,
              "calibration_gpu_ticks": 404625,
              "compute_passes": 1,
              "command_buffer_ns": 18500.0244528,
              "command_buffers": 2
            },
            {
              "compute_ns": 18375,
              "compute_span_ns": 18375,
              "calibration_cpu_ns": 603875,
              "calibration_gpu_ticks": 603875,
              "compute_passes": 1,
              "command_buffer_ns": 18416.6128747,
              "command_buffers": 2
            },
            {
              "compute_ns": 18542,
              "compute_span_ns": 18542,
              "calibration_cpu_ns": 420500,
              "calibration_gpu_ticks": 420500,
              "compute_passes": 1,
              "command_buffer_ns": 18541.6429304,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 982249.970548,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 933333.300054,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 950333.313085,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 964250.00811,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 908749.992959,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 904874.992557,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 908833.288122,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 975666.684099,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 967166.677583,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 16416.6558534,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18791.644834,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18791.7030416,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18708.3496712,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18833.3797269,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18624.9963008,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18416.6128747,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18541.701138,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19208.3534785,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              15.3476557898125,
              14.58333281334375,
              14.848958016953125,
              15.066406376718751,
              14.199218639984377,
              14.138671758703126,
              14.200520126906252,
              15.244791939046873,
              15.111979337234375
            ],
            "command_buffer_throughput_us_p50": 14.848958016953125,
            "command_buffer_throughput_us_p90": 15.265364709199998,
            "command_buffer_latency_us": [
              16.416655853399998,
              18.791644834,
              18.7917030416,
              18.7083496712,
              18.8333797269,
              18.624996300800003,
              18.4166128747,
              18.541701138,
              19.2083534785
            ],
            "command_buffer_latency_us_p50": 18.7083496712,
            "command_buffer_latency_us_p90": 18.90837447722
          },
          "compute_throughput_us": [
            14.244796875,
            14.22721875,
            14.30990625,
            14.3951875,
            14.028640625,
            14.553390625,
            14.162109375,
            14.10546875,
            15.202484375
          ],
          "compute_throughput_us_p50": 14.244796875,
          "compute_throughput_us_p90": 14.683209375,
          "command_buffer_throughput_us": [
            14.246093996921875,
            14.227864085109376,
            14.312500752562501,
            14.39713560103125,
            14.02864654665625,
            14.5533849718125,
            14.162109437171875,
            14.10677123203125,
            15.203775546984375
          ],
          "command_buffer_throughput_us_p50": 14.246093996921875,
          "command_buffer_throughput_us_p90": 14.683463086846874,
          "compute_latency_us": [
            18.75,
            18.458,
            18.5,
            18.667,
            18.708,
            18.708,
            18.5,
            18.375,
            18.542
          ],
          "compute_latency_us_p50": 18.542,
          "compute_latency_us_p90": 18.7164,
          "command_buffer_latency_us": [
            18.750026356400003,
            18.4583477676,
            18.5833778232,
            18.666672986000002,
            18.791644834,
            18.7499099411,
            18.5000244528,
            18.4166128747,
            18.541642930400002
          ],
          "command_buffer_latency_us_p50": 18.5833778232,
          "command_buffer_latency_us_p90": 18.75835005192,
          "counter_control_throughput_ratio": 0.9756250006233745,
          "counter_control_latency_ratio": 0.9977722949414315
        },
        "correctness": {
          "max_abs_error": 1.6175184480715643e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 15.4653151351,
        "throughput_us_p90": 15.579423243259999,
        "latency_us_p50": 249.291,
        "latency_us_p90": 270.05800000000005
      },
      "torch": {
        "cold_call_ms": 4.956625,
        "warmup_ms": 200.338083,
        "repetitions": 1325,
        "throughput_us": [
          19.827924528301885,
          18.88364754716981,
          20.668301886792452,
          20.940503396226415,
          20.95210716981132,
          20.723993962264153,
          19.04776679245283,
          19.58877358490566,
          19.56380528301887
        ],
        "latency_us": [
          432.375,
          249.583,
          254.91599999999997,
          247.25,
          260.041,
          269.875,
          272.792,
          257.41700000000003,
          244.875
        ],
        "throughput_us_p50": 19.827924528301885,
        "throughput_us_p90": 20.942824150943395,
        "latency_us_p50": 257.41700000000003,
        "latency_us_p90": 304.7086,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1069749.9965317547,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1028458.3549946547,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1093291.6775345802,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1083125.0110641122,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1029541.6577719152,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1087791.6938625276,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1031249.9944120646,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1027916.6745021939,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1082375.0053532422,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 18666.672986000776,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21041.661966592073,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21000.043489038944,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 20875.013433396816,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 20958.308596163988,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 20916.69011861086,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21166.633814573288,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21166.6920222342,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21125.01533702016,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              16.714843695808668,
              16.06966179679148,
              17.082682461477816,
              16.923828297876753,
              16.086588402686175,
              16.996745216601994,
              16.11328116268851,
              16.06119803909678,
              16.91210945864441
            ],
            "command_buffer_throughput_us_p50": 16.714843695808668,
            "command_buffer_throughput_us_p90": 17.01393266557716,
            "command_buffer_latency_us": [
              18.666672986000776,
              21.041661966592073,
              21.000043489038944,
              20.875013433396816,
              20.958308596163988,
              20.91669011861086,
              21.166633814573288,
              21.1666920222342,
              21.12501533702016
            ],
            "command_buffer_latency_us_p50": 21.000043489038944,
            "command_buffer_latency_us_p90": 21.16664545610547
          },
          "throughput": [
            {
              "compute_ns": 1086167.0,
              "compute_span_ns": 1086167.0,
              "command_buffer_ns": 1086208.3290703595,
              "calibration_cpu_ns": 1775084.0,
              "calibration_gpu_ticks": 1775084.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1030750.0,
              "compute_span_ns": 1030750.0,
              "command_buffer_ns": 1030749.9906048179,
              "calibration_cpu_ns": 1761083.0,
              "calibration_gpu_ticks": 1761083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1029209.0,
              "compute_span_ns": 1029209.0,
              "command_buffer_ns": 1029249.9791830778,
              "calibration_cpu_ns": 1676041.0,
              "calibration_gpu_ticks": 1676041.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1029292.0,
              "compute_span_ns": 1029292.0,
              "command_buffer_ns": 1029333.3325535059,
              "calibration_cpu_ns": 1629250.0,
              "calibration_gpu_ticks": 1629250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1030667.0,
              "compute_span_ns": 1030667.0,
              "command_buffer_ns": 1030708.3139196038,
              "calibration_cpu_ns": 1701708.0,
              "calibration_gpu_ticks": 1701708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1076833.0,
              "compute_span_ns": 1076833.0,
              "command_buffer_ns": 1076791.6683107615,
              "calibration_cpu_ns": 1791458.0,
              "calibration_gpu_ticks": 1791458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1067292.0,
              "compute_span_ns": 1067292.0,
              "command_buffer_ns": 1067291.6541807353,
              "calibration_cpu_ns": 1705916.0,
              "calibration_gpu_ticks": 1705916.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1056625.0,
              "compute_span_ns": 1056625.0,
              "command_buffer_ns": 1056625.0421106815,
              "calibration_cpu_ns": 1716041.0,
              "calibration_gpu_ticks": 1716041.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1030416.0,
              "compute_span_ns": 1030416.0,
              "command_buffer_ns": 1030374.9586455524,
              "calibration_cpu_ns": 1689458.0,
              "calibration_gpu_ticks": 1689458.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 20916.0,
              "compute_span_ns": 20916.0,
              "command_buffer_ns": 20875.013433396816,
              "calibration_cpu_ns": 492208.0,
              "calibration_gpu_ticks": 492208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21208.0,
              "compute_span_ns": 21208.0,
              "command_buffer_ns": 21166.633814573288,
              "calibration_cpu_ns": 503000.0,
              "calibration_gpu_ticks": 503000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 20792.0,
              "compute_span_ns": 20792.0,
              "command_buffer_ns": 20833.27854052186,
              "calibration_cpu_ns": 443458.0,
              "calibration_gpu_ticks": 443458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21250.0,
              "compute_span_ns": 21250.0,
              "command_buffer_ns": 21249.987185001373,
              "calibration_cpu_ns": 432084.0,
              "calibration_gpu_ticks": 432084.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 20875.0,
              "compute_span_ns": 20875.0,
              "command_buffer_ns": 20875.013433396816,
              "calibration_cpu_ns": 495209.0,
              "calibration_gpu_ticks": 495209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21208.0,
              "compute_span_ns": 21208.0,
              "command_buffer_ns": 21166.6920222342,
              "calibration_cpu_ns": 434500.0,
              "calibration_gpu_ticks": 434500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21000.0,
              "compute_span_ns": 21000.0,
              "command_buffer_ns": 20999.98528137803,
              "calibration_cpu_ns": 472500.0,
              "calibration_gpu_ticks": 472500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21041.0,
              "compute_span_ns": 21041.0,
              "command_buffer_ns": 21000.043489038944,
              "calibration_cpu_ns": 538000.0,
              "calibration_gpu_ticks": 538000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21125.0,
              "compute_span_ns": 21125.0,
              "command_buffer_ns": 21125.01533702016,
              "calibration_cpu_ns": 461042.0,
              "calibration_gpu_ticks": 461042.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            16.971359375,
            16.10546875,
            16.081390625,
            16.0826875,
            16.104171875,
            16.825515625,
            16.6764375,
            16.509765625,
            16.10025
          ],
          "compute_throughput_us_p50": 16.10546875,
          "compute_throughput_us_p90": 16.854684375,
          "command_buffer_throughput_us": [
            16.972005141724367,
            16.10546860320028,
            16.08203092473559,
            16.08333332114853,
            16.10481740499381,
            16.824869817355648,
            16.67643209657399,
            16.5097662829794,
            16.099608728836756
          ],
          "command_buffer_throughput_us_p50": 16.10546860320028,
          "command_buffer_throughput_us_p90": 16.85429688222939,
          "compute_latency_us": [
            20.916,
            21.208,
            20.792,
            21.25,
            20.875,
            21.208,
            21.0,
            21.041,
            21.125
          ],
          "compute_latency_us_p50": 21.041,
          "compute_latency_us_p90": 21.2164,
          "command_buffer_latency_us": [
            20.875013433396816,
            21.166633814573288,
            20.83327854052186,
            21.249987185001373,
            20.875013433396816,
            21.1666920222342,
            20.99998528137803,
            21.000043489038944,
            21.12501533702016
          ],
          "command_buffer_latency_us_p50": 21.000043489038944,
          "command_buffer_latency_us_p90": 21.183351054787636,
          "counter_control_throughput_ratio": 1.001133180128149,
          "counter_control_latency_ratio": 1.0
        },
        "download_ms": 0.354167,
        "allocation_upload_ms": 1.399667,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.8163651027691685e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.7799764979448653,
      "device_compute_slowdown": 0.8844695609992724,
      "valid": true,
      "round": 3,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "rmsnorm_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.306,
        "warmup_ms": 200.320042,
        "repetitions": 1253,
        "throughput_us": [
          20.716114924181962,
          20.990090183559456,
          20.869247406225057,
          21.308592976855547,
          20.321328810853952,
          19.553870710295293,
          19.007083000798083,
          19.181597765363126,
          19.477122106943337
        ],
        "latency_us": [
          363.70799999999997,
          294.333,
          250.04199999999997,
          247.875,
          256.834,
          282.5,
          255.042,
          245.458,
          240.125
        ],
        "throughput_us_p50": 20.321328810853952,
        "throughput_us_p90": 21.053790742218673,
        "latency_us_p50": 255.042,
        "latency_us_p90": 308.208,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1193666.7142435908,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1049624.988809228,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1124124.9740123749,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1181583.3277069032,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1168624.9636113644,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1186958.3395309746,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1256916.6719913483,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1288999.9779872596,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 1218624.9950900674,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 18916.67488962412,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 20999.98528137803,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21166.6920222342,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21291.663870215416,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21124.957129359245,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21083.338651806116,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21499.989088624716,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21791.6676774621,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 21666.695829480886,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              18.651042410056107,
              16.400390450144187,
              17.564452718943357,
              18.462239495420363,
              18.259765056427568,
              18.54622405517148,
              19.639322999864817,
              20.14062465605093,
              19.041015548282303
            ],
            "command_buffer_throughput_us_p50": 18.54622405517148,
            "command_buffer_throughput_us_p90": 19.73958333110204,
            "command_buffer_latency_us": [
              18.91667488962412,
              20.99998528137803,
              21.1666920222342,
              21.291663870215416,
              21.124957129359245,
              21.083338651806116,
              21.499989088624716,
              21.7916676774621,
              21.666695829480886
            ],
            "command_buffer_latency_us_p50": 21.1666920222342,
            "command_buffer_latency_us_p90": 21.69169019907713
          },
          "throughput": [
            {
              "compute_ns": 1111333.0,
              "compute_span_ns": 1111333.0,
              "command_buffer_ns": 1111291.698180139,
              "calibration_cpu_ns": 1698042.0,
              "calibration_gpu_ticks": 1698042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1028500.0,
              "compute_span_ns": 1028500.0,
              "command_buffer_ns": 1028499.9734722078,
              "calibration_cpu_ns": 1660959.0,
              "calibration_gpu_ticks": 1660959.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1125667.0,
              "compute_span_ns": 1125667.0,
              "command_buffer_ns": 1125666.662119329,
              "calibration_cpu_ns": 1842291.0,
              "calibration_gpu_ticks": 1842291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1191583.0,
              "compute_span_ns": 1191583.0,
              "command_buffer_ns": 1191583.345644176,
              "calibration_cpu_ns": 1778042.0,
              "calibration_gpu_ticks": 1778042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1212208.0,
              "compute_span_ns": 1212208.0,
              "command_buffer_ns": 1212208.2989662886,
              "calibration_cpu_ns": 1799750.0,
              "calibration_gpu_ticks": 1799750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1210417.0,
              "compute_span_ns": 1210417.0,
              "command_buffer_ns": 1210458.343848586,
              "calibration_cpu_ns": 1833458.0,
              "calibration_gpu_ticks": 1833458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1197917.0,
              "compute_span_ns": 1197917.0,
              "command_buffer_ns": 1197916.6883975267,
              "calibration_cpu_ns": 1792792.0,
              "calibration_gpu_ticks": 1792792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1198000.0,
              "compute_span_ns": 1198000.0,
              "command_buffer_ns": 1198000.0417679548,
              "calibration_cpu_ns": 1811875.0,
              "calibration_gpu_ticks": 1811875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 1177417.0,
              "compute_span_ns": 1177417.0,
              "command_buffer_ns": 1177458.3254009485,
              "calibration_cpu_ns": 1786750.0,
              "calibration_gpu_ticks": 1786750.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 18708.0,
              "compute_span_ns": 18708.0,
              "command_buffer_ns": 18666.614778339863,
              "calibration_cpu_ns": 447917.0,
              "calibration_gpu_ticks": 447917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 23708.0,
              "compute_span_ns": 23708.0,
              "command_buffer_ns": 23708.329536020756,
              "calibration_cpu_ns": 456125.0,
              "calibration_gpu_ticks": 456125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21000.0,
              "compute_span_ns": 21000.0,
              "command_buffer_ns": 20999.98528137803,
              "calibration_cpu_ns": 482167.0,
              "calibration_gpu_ticks": 482167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 20958.0,
              "compute_span_ns": 20958.0,
              "command_buffer_ns": 20916.631910949945,
              "calibration_cpu_ns": 396500.0,
              "calibration_gpu_ticks": 396500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21333.0,
              "compute_span_ns": 21333.0,
              "command_buffer_ns": 21291.663870215416,
              "calibration_cpu_ns": 480041.0,
              "calibration_gpu_ticks": 480041.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21250.0,
              "compute_span_ns": 21250.0,
              "command_buffer_ns": 21249.987185001373,
              "calibration_cpu_ns": 587875.0,
              "calibration_gpu_ticks": 587875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 25167.0,
              "compute_span_ns": 25167.0,
              "command_buffer_ns": 25208.34095776081,
              "calibration_cpu_ns": 518292.0,
              "calibration_gpu_ticks": 518292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21125.0,
              "compute_span_ns": 21125.0,
              "command_buffer_ns": 21125.01533702016,
              "calibration_cpu_ns": 418084.0,
              "calibration_gpu_ticks": 418084.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 21541.0,
              "compute_span_ns": 21541.0,
              "command_buffer_ns": 21500.04729628563,
              "calibration_cpu_ns": 449292.0,
              "calibration_gpu_ticks": 449292.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            17.364578125,
            16.0703125,
            17.588546875,
            18.618484375,
            18.94075,
            18.912765625,
            18.717453125,
            18.71875,
            18.397140625
          ],
          "compute_throughput_us_p50": 18.618484375,
          "compute_throughput_us_p90": 18.9183625,
          "command_buffer_throughput_us": [
            17.363932784064673,
            16.070312085503247,
            17.588541595614515,
            18.61848977569025,
            18.94075467134826,
            18.913411622634158,
            18.717448256211355,
            18.718750652624294,
            18.39778633438982
          ],
          "command_buffer_throughput_us_p50": 18.61848977569025,
          "command_buffer_throughput_us_p90": 18.918880232376978,
          "compute_latency_us": [
            18.708,
            23.708,
            21.0,
            20.958,
            21.333,
            21.25,
            25.167,
            21.125,
            21.541
          ],
          "compute_latency_us_p50": 21.25,
          "compute_latency_us_p90": 23.9998,
          "command_buffer_latency_us": [
            18.666614778339863,
            23.708329536020756,
            20.99998528137803,
            20.916631910949945,
            21.291663870215416,
            21.249987185001373,
            25.20834095776081,
            21.12501533702016,
            21.50004729628563
          ],
          "command_buffer_latency_us_p50": 21.249987185001373,
          "command_buffer_latency_us_p90": 24.008331820368767,
          "counter_control_throughput_ratio": 0.9798737496131966,
          "counter_control_latency_ratio": 0.9923085395896634
        },
        "download_ms": 0.352,
        "allocation_upload_ms": 1.170583,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.8163651027691685e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-avjol3jh/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "a8c235e7c8be57d97298e1f9f70525f6c2c7f4ae4cee9647b15503a46aab5ade",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 16.5885,
        "capture_ms": 0.060166,
        "compile_ms": 53.663458,
        "allocation_upload_ms": 1.926417,
        "cold_call_ms": 5.3145,
        "warmup_ms": 200.368,
        "download_ms": 0.886542,
        "repetitions": 1715,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          15.6179055394,
          15.6171527697,
          15.7204326531,
          15.6710396501,
          15.4741498542,
          14.8060979592,
          14.9465014577,
          15.8655492711,
          14.906535277
        ],
        "latency_us": [
          322.417,
          295.5,
          278.042,
          270.875,
          239.875,
          280.167,
          292,
          257.208,
          337.334
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 128,
            "reduction_scalar_rounds": 36,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 24576,
              "global_write_bytes": 12288,
              "private_read_bytes": 36864,
              "private_write_bytes": 12288
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 48,
              "private_read_bytes": 144,
              "private_write_bytes": 48
            },
            "striped_storage_scalars_per_worker": 12,
            "reduction_operations": 1,
            "reduction_elements": 3072,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 60,
            "normalized_kernel_cost": 240,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 6144,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 973125,
              "compute_span_ns": 973125,
              "calibration_cpu_ns": 1460250,
              "calibration_gpu_ticks": 1460250,
              "compute_passes": 1,
              "command_buffer_ns": 973166.606855,
              "command_buffers": 2
            },
            {
              "compute_ns": 921542,
              "compute_span_ns": 921542,
              "calibration_cpu_ns": 1394208,
              "calibration_gpu_ticks": 1394208,
              "compute_passes": 1,
              "command_buffer_ns": 921666.68037,
              "command_buffers": 2
            },
            {
              "compute_ns": 993583,
              "compute_span_ns": 993583,
              "calibration_cpu_ns": 1468292,
              "calibration_gpu_ticks": 1468292,
              "compute_passes": 1,
              "command_buffer_ns": 993624.969851,
              "command_buffers": 2
            },
            {
              "compute_ns": 950125,
              "compute_span_ns": 950125,
              "calibration_cpu_ns": 1419208,
              "calibration_gpu_ticks": 1419208,
              "compute_passes": 1,
              "command_buffer_ns": 950208.399445,
              "command_buffers": 2
            },
            {
              "compute_ns": 918500,
              "compute_span_ns": 918500,
              "calibration_cpu_ns": 1382875,
              "calibration_gpu_ticks": 1382875,
              "compute_passes": 1,
              "command_buffer_ns": 918624.980841,
              "command_buffers": 2
            },
            {
              "compute_ns": 941375,
              "compute_span_ns": 941375,
              "calibration_cpu_ns": 1364333,
              "calibration_gpu_ticks": 1364333,
              "compute_passes": 1,
              "command_buffer_ns": 941499.951296,
              "command_buffers": 2
            },
            {
              "compute_ns": 908375,
              "compute_span_ns": 908375,
              "calibration_cpu_ns": 1363500,
              "calibration_gpu_ticks": 1363500,
              "compute_passes": 1,
              "command_buffer_ns": 908458.372578,
              "command_buffers": 2
            },
            {
              "compute_ns": 977917,
              "compute_span_ns": 977917,
              "calibration_cpu_ns": 1438250,
              "calibration_gpu_ticks": 1438250,
              "compute_passes": 1,
              "command_buffer_ns": 978041.673079,
              "command_buffers": 2
            },
            {
              "compute_ns": 1001041,
              "compute_span_ns": 1001041,
              "calibration_cpu_ns": 1441917,
              "calibration_gpu_ticks": 1441917,
              "compute_passes": 1,
              "command_buffer_ns": 1001083.29207,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 16500,
              "compute_span_ns": 16500,
              "calibration_cpu_ns": 417000,
              "calibration_gpu_ticks": 417000,
              "compute_passes": 1,
              "command_buffer_ns": 16500.0092238,
              "command_buffers": 2
            },
            {
              "compute_ns": 19083,
              "compute_span_ns": 19083,
              "calibration_cpu_ns": 416791,
              "calibration_gpu_ticks": 416791,
              "compute_passes": 1,
              "command_buffer_ns": 19041.6467376,
              "command_buffers": 2
            },
            {
              "compute_ns": 18875,
              "compute_span_ns": 18875,
              "calibration_cpu_ns": 423334,
              "calibration_gpu_ticks": 423334,
              "compute_passes": 1,
              "command_buffer_ns": 19000.0282601,
              "command_buffers": 2
            },
            {
              "compute_ns": 18875,
              "compute_span_ns": 18875,
              "calibration_cpu_ns": 437333,
              "calibration_gpu_ticks": 437333,
              "compute_passes": 1,
              "command_buffer_ns": 18958.2933672,
              "command_buffers": 2
            },
            {
              "compute_ns": 19125,
              "compute_span_ns": 19125,
              "calibration_cpu_ns": 459458,
              "calibration_gpu_ticks": 459458,
              "compute_passes": 1,
              "command_buffer_ns": 19208.3534785,
              "command_buffers": 2
            },
            {
              "compute_ns": 20667,
              "compute_span_ns": 20667,
              "calibration_cpu_ns": 617792,
              "calibration_gpu_ticks": 617792,
              "compute_passes": 1,
              "command_buffer_ns": 20666.688215,
              "command_buffers": 2
            },
            {
              "compute_ns": 19041,
              "compute_span_ns": 19041,
              "calibration_cpu_ns": 484667,
              "calibration_gpu_ticks": 484667,
              "compute_passes": 1,
              "command_buffer_ns": 19083.3234228,
              "command_buffers": 2
            },
            {
              "compute_ns": 19041,
              "compute_span_ns": 19041,
              "calibration_cpu_ns": 456125,
              "calibration_gpu_ticks": 456125,
              "compute_passes": 1,
              "command_buffer_ns": 18999.9700524,
              "command_buffers": 2
            },
            {
              "compute_ns": 20542,
              "compute_span_ns": 20542,
              "calibration_cpu_ns": 672959,
              "calibration_gpu_ticks": 672959,
              "compute_passes": 1,
              "command_buffer_ns": 20583.3348446,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 994999.951217,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 918666.657526,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 925708.271097,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 948125.030845,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 962541.67147,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 919583.253562,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1023916.70225,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 914416.683372,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 925499.945879,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 17208.2800418,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19458.2971744,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19208.3534785,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19166.6767932,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19083.3816305,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19291.6486412,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18958.3515748,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19249.971956,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19208.3534785,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              15.546874237765625,
              14.354166523843752,
              14.464191735890624,
              14.814453606953125,
              15.03971361671875,
              14.36848833690625,
              15.99869847265625,
              14.2877606776875,
              14.460936654359376
            ],
            "command_buffer_throughput_us_p50": 14.464191735890624,
            "command_buffer_throughput_us_p90": 15.637239084743749,
            "command_buffer_latency_us": [
              17.208280041800002,
              19.4582971744,
              19.2083534785,
              19.1666767932,
              19.0833816305,
              19.291648641200002,
              18.958351574799998,
              19.249971956,
              19.2083534785
            ],
            "command_buffer_latency_us_p50": 19.2083534785,
            "command_buffer_latency_us_p90": 19.324978347840002
          },
          "compute_throughput_us": [
            15.205078125,
            14.39909375,
            15.524734375,
            14.845703125,
            14.3515625,
            14.708984375,
            14.193359375,
            15.279953125,
            15.641265625
          ],
          "compute_throughput_us_p50": 14.845703125,
          "compute_throughput_us_p90": 15.548040625,
          "command_buffer_throughput_us": [
            15.205728232109376,
            14.40104188078125,
            15.525390153921874,
            14.847006241328124,
            14.353515325640625,
            14.710936739000001,
            14.19466207153125,
            15.281901141859375,
            15.64192643859375
          ],
          "command_buffer_throughput_us_p50": 14.847006241328124,
          "command_buffer_throughput_us_p90": 15.54869741085625,
          "compute_latency_us": [
            16.5,
            19.083,
            18.875,
            18.875,
            19.125,
            20.667,
            19.041,
            19.041,
            20.542
          ],
          "compute_latency_us_p50": 19.041,
          "compute_latency_us_p90": 20.567,
          "command_buffer_latency_us": [
            16.5000092238,
            19.0416467376,
            19.0000282601,
            18.9582933672,
            19.2083534785,
            20.666688214999997,
            19.0833234228,
            18.9999700524,
            20.583334844600003
          ],
          "command_buffer_latency_us_p50": 19.0416467376,
          "command_buffer_latency_us_p90": 20.600005518680003,
          "counter_control_throughput_ratio": 1.003265627221172,
          "counter_control_latency_ratio": 0.9891544468591137
        },
        "correctness": {
          "max_abs_error": 1.6175184480715643e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 15.6171527697,
        "throughput_us_p90": 15.7494559767,
        "latency_us_p50": 280.167,
        "latency_us_p90": 325.4004
      },
      "slowdown": 0.7685104116497847,
      "device_compute_slowdown": 0.797363675044038,
      "valid": true,
      "round": 3,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "rmsnorm_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-891z74qm/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "c9fd11968a5edbf77e894438f42e4bd465ad146629716a6f08eacac9ab82063f",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.600625,
        "capture_ms": 0.06775,
        "compile_ms": 52.604917,
        "allocation_upload_ms": 6.212416,
        "cold_call_ms": 7.81725,
        "warmup_ms": 200.188458,
        "download_ms": 3.879541,
        "repetitions": 443,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          65.9753589165,
          65.3704853273,
          66.6272573363,
          66.5381873589,
          66.600261851,
          66.2517878104,
          66.1798329571,
          65.6318668172,
          65.6719322799
        ],
        "latency_us": [
          313.083,
          287.625,
          278,
          274.041,
          281.417,
          268.083,
          268.708,
          271.084,
          302.5
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 384,
            "reduction_scalar_rounds": 72,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 49152,
              "global_write_bytes": 24576,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 192,
              "global_write_bytes": 96,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 1,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 96,
            "normalized_kernel_cost": 1152,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4175500,
              "compute_span_ns": 4175500,
              "calibration_cpu_ns": 4668667,
              "calibration_gpu_ticks": 4668667,
              "compute_passes": 1,
              "command_buffer_ns": 4175625.04299,
              "command_buffers": 2
            },
            {
              "compute_ns": 4154625,
              "compute_span_ns": 4154625,
              "calibration_cpu_ns": 4659042,
              "calibration_gpu_ticks": 4659042,
              "compute_passes": 1,
              "command_buffer_ns": 4154666.67619,
              "command_buffers": 2
            },
            {
              "compute_ns": 4236709,
              "compute_span_ns": 4236709,
              "calibration_cpu_ns": 4715209,
              "calibration_gpu_ticks": 4715209,
              "compute_passes": 1,
              "command_buffer_ns": 4236875.04372,
              "command_buffers": 2
            },
            {
              "compute_ns": 4169875,
              "compute_span_ns": 4169875,
              "calibration_cpu_ns": 4741167,
              "calibration_gpu_ticks": 4741167,
              "compute_passes": 1,
              "command_buffer_ns": 4169999.97105,
              "command_buffers": 2
            },
            {
              "compute_ns": 4278083,
              "compute_span_ns": 4278083,
              "calibration_cpu_ns": 4811167,
              "calibration_gpu_ticks": 4811167,
              "compute_passes": 1,
              "command_buffer_ns": 4278166.68525,
              "command_buffers": 2
            },
            {
              "compute_ns": 4182833,
              "compute_span_ns": 4182833,
              "calibration_cpu_ns": 4611417,
              "calibration_gpu_ticks": 4611417,
              "compute_passes": 1,
              "command_buffer_ns": 4182958.39336,
              "command_buffers": 2
            },
            {
              "compute_ns": 4199000,
              "compute_span_ns": 4199000,
              "calibration_cpu_ns": 4631875,
              "calibration_gpu_ticks": 4631875,
              "compute_passes": 1,
              "command_buffer_ns": 4199041.69394,
              "command_buffers": 2
            },
            {
              "compute_ns": 4192208,
              "compute_span_ns": 4192208,
              "calibration_cpu_ns": 4699458,
              "calibration_gpu_ticks": 4699458,
              "compute_passes": 1,
              "command_buffer_ns": 4192208.34738,
              "command_buffers": 2
            },
            {
              "compute_ns": 4200417,
              "compute_span_ns": 4200417,
              "calibration_cpu_ns": 4880084,
              "calibration_gpu_ticks": 4880084,
              "compute_passes": 1,
              "command_buffer_ns": 4200541.70536,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 78458,
              "compute_span_ns": 78458,
              "calibration_cpu_ns": 566583,
              "calibration_gpu_ticks": 566583,
              "compute_passes": 1,
              "command_buffer_ns": 78500.0156611,
              "command_buffers": 2
            },
            {
              "compute_ns": 65250,
              "compute_span_ns": 65250,
              "calibration_cpu_ns": 465708,
              "calibration_gpu_ticks": 465708,
              "compute_passes": 1,
              "command_buffer_ns": 65250.0311844,
              "command_buffers": 2
            },
            {
              "compute_ns": 66416,
              "compute_span_ns": 66416,
              "calibration_cpu_ns": 439292,
              "calibration_gpu_ticks": 439292,
              "compute_passes": 1,
              "command_buffer_ns": 66375.0106469,
              "command_buffers": 2
            },
            {
              "compute_ns": 66459,
              "compute_span_ns": 66459,
              "calibration_cpu_ns": 424750,
              "calibration_gpu_ticks": 424750,
              "compute_passes": 1,
              "command_buffer_ns": 66541.6009724,
              "command_buffers": 2
            },
            {
              "compute_ns": 77875,
              "compute_span_ns": 77875,
              "calibration_cpu_ns": 527291,
              "calibration_gpu_ticks": 527291,
              "compute_passes": 1,
              "command_buffer_ns": 77874.9817982,
              "command_buffers": 2
            },
            {
              "compute_ns": 74834,
              "compute_span_ns": 74834,
              "calibration_cpu_ns": 582125,
              "calibration_gpu_ticks": 582125,
              "compute_passes": 1,
              "command_buffer_ns": 74958.3705328,
              "command_buffers": 2
            },
            {
              "compute_ns": 68333,
              "compute_span_ns": 68333,
              "calibration_cpu_ns": 487958,
              "calibration_gpu_ticks": 487958,
              "compute_passes": 1,
              "command_buffer_ns": 68416.7025611,
              "command_buffers": 2
            },
            {
              "compute_ns": 67542,
              "compute_span_ns": 67542,
              "calibration_cpu_ns": 495917,
              "calibration_gpu_ticks": 495917,
              "compute_passes": 1,
              "command_buffer_ns": 67583.3434798,
              "command_buffers": 2
            },
            {
              "compute_ns": 66875,
              "compute_span_ns": 66875,
              "calibration_cpu_ns": 465166,
              "calibration_gpu_ticks": 465166,
              "compute_passes": 1,
              "command_buffer_ns": 66875.0144541,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4252958.3443,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4218000.04551,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4286874.95878,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4188125.02176,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4297458.33389,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4172833.28716,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4235791.62452,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4217916.69214,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4190416.65737,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 75374.962762,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65208.2962915,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 69374.9752827,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 66208.3621137,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76833.2974985,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 67666.6386425,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 67083.3396725,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 77166.6527726,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 80375.0008345,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              66.4524741296875,
              65.90625071109375,
              66.9824212309375,
              65.439453465,
              67.14778646703125,
              65.200520111875,
              66.184244133125,
              65.9049483146875,
              65.47526027140626
            ],
            "command_buffer_throughput_us_p50": 65.90625071109375,
            "command_buffer_throughput_us_p90": 67.01549427815625,
            "command_buffer_latency_us": [
              75.374962762,
              65.2082962915,
              69.3749752827,
              66.2083621137,
              76.8332974985,
              67.66663864249999,
              67.0833396725,
              77.1666527726,
              80.3750008345
            ],
            "command_buffer_latency_us_p50": 69.3749752827,
            "command_buffer_latency_us_p90": 77.80832238498
          },
          "compute_throughput_us": [
            65.2421875,
            64.916015625,
            66.198578125,
            65.154296875,
            66.845046875,
            65.356765625,
            65.609375,
            65.50325,
            65.631515625
          ],
          "compute_throughput_us_p50": 65.50325,
          "compute_throughput_us_p90": 66.327871875,
          "command_buffer_throughput_us": [
            65.24414129671875,
            64.91666681546874,
            66.20117255812501,
            65.15624954765624,
            66.84635445703125,
            65.35872489625,
            65.61002646781249,
            65.5032554278125,
            65.63346414625
          ],
          "command_buffer_throughput_us_p50": 65.5032554278125,
          "command_buffer_throughput_us_p90": 66.33020893790626,
          "compute_latency_us": [
            78.458,
            65.25,
            66.416,
            66.459,
            77.875,
            74.834,
            68.333,
            67.542,
            66.875
          ],
          "compute_latency_us_p50": 67.542,
          "compute_latency_us_p90": 77.9916,
          "command_buffer_latency_us": [
            78.5000156611,
            65.2500311844,
            66.37501064690001,
            66.5416009724,
            77.8749817982,
            74.9583705328,
            68.4167025611,
            67.58334347979999,
            66.87501445410001
          ],
          "command_buffer_latency_us_p50": 67.58334347979999,
          "command_buffer_latency_us_p90": 77.99998857077999,
          "counter_control_throughput_ratio": 0.9939049662104739,
          "counter_control_latency_ratio": 1.0050331838465922
        },
        "correctness": {
          "max_abs_error": 1.4915225321487924e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 66.1798329571,
        "throughput_us_p90": 66.60566094806,
        "latency_us_p50": 278.0,
        "latency_us_p90": 304.6166
      },
      "torch": {
        "cold_call_ms": 9.732625,
        "warmup_ms": 200.142709,
        "repetitions": 291,
        "throughput_us": [
          82.07087628865979,
          80.68800343642611,
          81.40864948453608,
          80.88860137457044,
          83.66752577319588,
          81.1877147766323,
          81.05970790378007,
          80.93427835051547,
          80.96535051546392
        ],
        "latency_us": [
          438.625,
          339.83299999999997,
          301.417,
          404.75,
          315.416,
          313.625,
          295.04200000000003,
          335.334,
          321.79200000000003
        ],
        "throughput_us_p50": 81.05970790378007,
        "throughput_us_p90": 82.39020618556701,
        "latency_us_p50": 321.79200000000003,
        "latency_us_p90": 411.525,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 5029666.70319438,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5069500.009994954,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5097749.992273748,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5099833.360873163,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5202125.001233071,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5056291.643995792,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4915374.971460551,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4976583.3537094295,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5059083.3416208625,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 98499.99332800508,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80499.97268244624,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80916.68132692575,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 97499.9857135117,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80124.99893084168,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80499.97268244624,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80000.02708286047,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 96791.71489551663,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 81041.65317490697,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              78.58854223741218,
              79.21093765617115,
              79.65234362927731,
              79.68489626364317,
              81.28320314426674,
              79.00455693743424,
              76.80273392907111,
              77.75911490170984,
              79.04817721282598
            ],
            "command_buffer_throughput_us_p50": 79.04817721282598,
            "command_buffer_throughput_us_p90": 80.00455763976788,
            "command_buffer_latency_us": [
              98.49999332800508,
              80.49997268244624,
              80.91668132692575,
              97.4999857135117,
              80.12499893084168,
              80.49997268244624,
              80.00002708286047,
              96.79171489551663,
              81.04165317490697
            ],
            "command_buffer_latency_us_p50": 80.91668132692575,
            "command_buffer_latency_us_p90": 97.69998723641038
          },
          "throughput": [
            {
              "compute_ns": 5031459.0,
              "compute_span_ns": 5031459.0,
              "command_buffer_ns": 5031499.953474849,
              "calibration_cpu_ns": 5672250.0,
              "calibration_gpu_ticks": 5672250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5102334.0,
              "compute_span_ns": 5102334.0,
              "command_buffer_ns": 5102374.998386949,
              "calibration_cpu_ns": 5757250.0,
              "calibration_gpu_ticks": 5757250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5055833.0,
              "compute_span_ns": 5055833.0,
              "command_buffer_ns": 5055791.640188545,
              "calibration_cpu_ns": 5735834.0,
              "calibration_gpu_ticks": 5735834.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5098000.0,
              "compute_span_ns": 5098000.0,
              "command_buffer_ns": 5097999.994177371,
              "calibration_cpu_ns": 5796333.0,
              "calibration_gpu_ticks": 5796333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5138750.0,
              "compute_span_ns": 5138750.0,
              "command_buffer_ns": 5138749.955222011,
              "calibration_cpu_ns": 5868167.0,
              "calibration_gpu_ticks": 5868167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5108458.0,
              "compute_span_ns": 5108458.0,
              "command_buffer_ns": 5108416.662551463,
              "calibration_cpu_ns": 5848500.0,
              "calibration_gpu_ticks": 5848500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4944792.0,
              "compute_span_ns": 4944792.0,
              "command_buffer_ns": 4944791.668094695,
              "calibration_cpu_ns": 5781458.0,
              "calibration_gpu_ticks": 5781458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4949459.0,
              "compute_span_ns": 4949459.0,
              "command_buffer_ns": 4949500.027578324,
              "calibration_cpu_ns": 5756875.0,
              "calibration_gpu_ticks": 5756875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5026250.0,
              "compute_span_ns": 5026250.0,
              "command_buffer_ns": 5026250.029914081,
              "calibration_cpu_ns": 5826167.0,
              "calibration_gpu_ticks": 5826167.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 96458.0,
              "compute_span_ns": 96458.0,
              "command_buffer_ns": 96458.30141380429,
              "calibration_cpu_ns": 600167.0,
              "calibration_gpu_ticks": 600167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 79709.0,
              "compute_span_ns": 79709.0,
              "command_buffer_ns": 79749.96697157621,
              "calibration_cpu_ns": 527750.0,
              "calibration_gpu_ticks": 527750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80375.0,
              "compute_span_ns": 80375.0,
              "command_buffer_ns": 80375.00083446503,
              "calibration_cpu_ns": 573584.0,
              "calibration_gpu_ticks": 573584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80708.0,
              "compute_span_ns": 80708.0,
              "command_buffer_ns": 80666.67942330241,
              "calibration_cpu_ns": 578333.0,
              "calibration_gpu_ticks": 578333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80375.0,
              "compute_span_ns": 80375.0,
              "command_buffer_ns": 80375.00083446503,
              "calibration_cpu_ns": 585333.0,
              "calibration_gpu_ticks": 585333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 81375.0,
              "compute_span_ns": 81375.0,
              "command_buffer_ns": 81374.95024129748,
              "calibration_cpu_ns": 590500.0,
              "calibration_gpu_ticks": 590500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80208.0,
              "compute_span_ns": 80208.0,
              "command_buffer_ns": 80166.67561605573,
              "calibration_cpu_ns": 575584.0,
              "calibration_gpu_ticks": 575584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 99625.0,
              "compute_span_ns": 99625.0,
              "command_buffer_ns": 99625.03099814057,
              "calibration_cpu_ns": 664792.0,
              "calibration_gpu_ticks": 664792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80042.0,
              "compute_span_ns": 80042.0,
              "command_buffer_ns": 80083.32224562764,
              "calibration_cpu_ns": 563875.0,
              "calibration_gpu_ticks": 563875.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            78.616546875,
            79.72396875,
            78.997390625,
            79.65625,
            80.29296875,
            79.81965625,
            77.262375,
            77.335296875,
            78.53515625
          ],
          "compute_throughput_us_p50": 78.997390625,
          "compute_throughput_us_p90": 79.91431874999999,
          "command_buffer_throughput_us": [
            78.61718677304452,
            79.72460934979608,
            78.99674437794602,
            79.65624990902143,
            80.29296805034392,
            79.8190103523666,
            77.2623698139796,
            77.33593793091131,
            78.53515671740752
          ],
          "command_buffer_throughput_us_p50": 78.99674437794602,
          "command_buffer_throughput_us_p90": 79.91380189196207,
          "compute_latency_us": [
            96.458,
            79.709,
            80.375,
            80.708,
            80.375,
            81.375,
            80.208,
            99.625,
            80.042
          ],
          "compute_latency_us_p50": 80.375,
          "compute_latency_us_p90": 97.0914,
          "command_buffer_latency_us": [
            96.45830141380429,
            79.74996697157621,
            80.37500083446503,
            80.66667942330241,
            80.37500083446503,
            81.37495024129748,
            80.16667561605573,
            99.62503099814057,
            80.08332224562764
          ],
          "command_buffer_latency_us_p50": 80.37500083446503,
          "command_buffer_latency_us_p90": 97.09164733067155,
          "counter_control_throughput_ratio": 0.9996405045878837,
          "counter_control_latency_ratio": 0.9933057005851218
        },
        "download_ms": 0.809208,
        "allocation_upload_ms": 15.88125,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.9755493041451189e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.8164331536409821,
      "device_compute_slowdown": 0.8291824512399836,
      "valid": true,
      "round": 3,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "rmsnorm_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.4265,
        "warmup_ms": 200.446792,
        "repetitions": 356,
        "throughput_us": [
          81.57560955056181,
          81.70446910112359,
          81.76650280898876,
          81.83040730337079,
          81.87371067415731,
          81.9170168539326,
          81.8560393258427,
          82.0886011235955,
          81.68094662921348
        ],
        "latency_us": [
          336.5,
          312.042,
          327.33299999999997,
          325.083,
          370.70799999999997,
          263.5,
          462.083,
          377.70799999999997,
          341.75
        ],
        "throughput_us_p50": 81.83040730337079,
        "throughput_us_p90": 81.95133370786517,
        "latency_us_p50": 336.5,
        "latency_us_p90": 394.58299999999997,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 5048999.970313162,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5083750.002086163,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5074916.640296578,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5114208.383020014,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5114166.648127139,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5111625.010613352,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5103250.034153461,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5100124.981254339,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 5057791.655417532,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 94458.34439247847,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 81708.30551534891,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 81125.00654533505,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 79874.99702721834,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 81083.32986012101,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 80124.99893084168,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 81875.01225620508,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 114374.96868893504,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 128749.99083578587,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              78.89062453614315,
              79.43359378259629,
              79.29557250463404,
              79.90950598468771,
              79.90885387698654,
              79.86914079083363,
              79.73828178364784,
              79.68945283209905,
              79.02799461589893
            ],
            "command_buffer_throughput_us_p50": 79.68945283209905,
            "command_buffer_throughput_us_p90": 79.90898429852678,
            "command_buffer_latency_us": [
              94.45834439247847,
              81.70830551534891,
              81.12500654533505,
              79.87499702721834,
              81.08332986012101,
              80.12499893084168,
              81.87501225620508,
              114.37496868893504,
              128.74999083578587
            ],
            "command_buffer_latency_us_p50": 81.70830551534891,
            "command_buffer_latency_us_p90": 117.2499731183052
          },
          "throughput": [
            {
              "compute_ns": 5076959.0,
              "compute_span_ns": 5076959.0,
              "command_buffer_ns": 5077000.008895993,
              "calibration_cpu_ns": 5851500.0,
              "calibration_gpu_ticks": 5851500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5118625.0,
              "compute_span_ns": 5118625.0,
              "command_buffer_ns": 5118625.005707145,
              "calibration_cpu_ns": 5756666.0,
              "calibration_gpu_ticks": 5756666.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5099334.0,
              "compute_span_ns": 5099334.0,
              "command_buffer_ns": 5099374.975543469,
              "calibration_cpu_ns": 5797167.0,
              "calibration_gpu_ticks": 5797167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5069292.0,
              "compute_span_ns": 5069292.0,
              "command_buffer_ns": 5069333.361461759,
              "calibration_cpu_ns": 5734375.0,
              "calibration_gpu_ticks": 5734375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5072458.0,
              "compute_span_ns": 5072458.0,
              "command_buffer_ns": 5072416.679468006,
              "calibration_cpu_ns": 5858208.0,
              "calibration_gpu_ticks": 5858208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5077833.0,
              "compute_span_ns": 5077833.0,
              "command_buffer_ns": 5077833.367977291,
              "calibration_cpu_ns": 5815250.0,
              "calibration_gpu_ticks": 5815250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5053084.0,
              "compute_span_ns": 5053084.0,
              "command_buffer_ns": 5053124.972619116,
              "calibration_cpu_ns": 5840875.0,
              "calibration_gpu_ticks": 5840875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5095292.0,
              "compute_span_ns": 5095292.0,
              "command_buffer_ns": 5095291.649922729,
              "calibration_cpu_ns": 5921583.0,
              "calibration_gpu_ticks": 5921583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 5097833.0,
              "compute_span_ns": 5097833.0,
              "command_buffer_ns": 5097791.668958962,
              "calibration_cpu_ns": 6213334.0,
              "calibration_gpu_ticks": 6213334.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 81042.0,
              "compute_span_ns": 81042.0,
              "command_buffer_ns": 81083.32986012101,
              "calibration_cpu_ns": 671750.0,
              "calibration_gpu_ticks": 671750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80333.0,
              "compute_span_ns": 80333.0,
              "command_buffer_ns": 80291.70567169785,
              "calibration_cpu_ns": 557125.0,
              "calibration_gpu_ticks": 557125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 100042.0,
              "compute_span_ns": 100042.0,
              "command_buffer_ns": 100041.62322729826,
              "calibration_cpu_ns": 701125.0,
              "calibration_gpu_ticks": 701125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 80542.0,
              "compute_span_ns": 80542.0,
              "command_buffer_ns": 80583.32605287433,
              "calibration_cpu_ns": 672125.0,
              "calibration_gpu_ticks": 672125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 82750.0,
              "compute_span_ns": 82750.0,
              "command_buffer_ns": 82749.98981505632,
              "calibration_cpu_ns": 1736625.0,
              "calibration_gpu_ticks": 1736625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 99667.0,
              "compute_span_ns": 99667.0,
              "command_buffer_ns": 99666.6494756937,
              "calibration_cpu_ns": 1541167.0,
              "calibration_gpu_ticks": 1541167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 98166.0,
              "compute_span_ns": 98166.0,
              "command_buffer_ns": 98125.01957640052,
              "calibration_cpu_ns": 1393583.0,
              "calibration_gpu_ticks": 1393583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 97125.0,
              "compute_span_ns": 97125.0,
              "command_buffer_ns": 97125.01196190715,
              "calibration_cpu_ns": 880041.0,
              "calibration_gpu_ticks": 880041.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 101416.0,
              "compute_span_ns": 101416.0,
              "command_buffer_ns": 101375.04432350397,
              "calibration_cpu_ns": 912417.0,
              "calibration_gpu_ticks": 912417.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            79.327484375,
            79.978515625,
            79.67709375,
            79.2076875,
            79.25715625,
            79.341140625,
            78.9544375,
            79.6139375,
            79.653640625
          ],
          "compute_throughput_us_p50": 79.341140625,
          "compute_throughput_us_p90": 79.73737812499999,
          "command_buffer_throughput_us": [
            79.3281251389999,
            79.97851571417414,
            79.6777339928667,
            79.20833377283998,
            79.25651061668759,
            79.34114637464518,
            78.95507769717369,
            79.61393203004263,
            79.65299482748378
          ],
          "command_buffer_throughput_us_p50": 79.34114637464518,
          "command_buffer_throughput_us_p90": 79.73789033712819,
          "compute_latency_us": [
            81.042,
            80.333,
            100.042,
            80.542,
            82.75,
            99.667,
            98.166,
            97.125,
            101.416
          ],
          "compute_latency_us_p50": 97.125,
          "compute_latency_us_p90": 100.3168,
          "command_buffer_latency_us": [
            81.08332986012101,
            80.29170567169785,
            100.04162322729826,
            80.58332605287433,
            82.74998981505632,
            99.6666494756937,
            98.12501957640052,
            97.12501196190715,
            101.37504432350397
          ],
          "command_buffer_latency_us_p50": 97.12501196190715,
          "command_buffer_latency_us_p90": 100.3083074465394,
          "counter_control_throughput_ratio": 0.9990523111983773,
          "counter_control_latency_ratio": 1.0088679693523446
        },
        "download_ms": 1.468958,
        "allocation_upload_ms": 1.402417,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.9755493041451189e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-6bg4vbyf/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "232522da57db77c64b377ec2facdc2e1df33cd57064a4297b89d6582fd4ad63b",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 18.033,
        "capture_ms": 0.061,
        "compile_ms": 52.230208,
        "allocation_upload_ms": 6.765084,
        "cold_call_ms": 5.148917,
        "warmup_ms": 200.736958,
        "download_ms": 4.271375,
        "repetitions": 482,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          60.7901099585,
          60.1186037344,
          60.0372572614,
          59.9086286307,
          59.8719751037,
          60.0632780083,
          60.0026784232,
          60.174186722,
          60.2247593361
        ],
        "latency_us": [
          362.958,
          321.042,
          362,
          305.666,
          293.584,
          304.208,
          296.542,
          309.083,
          278.667
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 32,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 768,
            "reduction_scalar_rounds": 72,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 49152,
              "global_write_bytes": 24576,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 192,
              "global_write_bytes": 96,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 1,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 104,
            "normalized_kernel_cost": 1248,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3895625,
              "compute_span_ns": 3895625,
              "calibration_cpu_ns": 4394167,
              "calibration_gpu_ticks": 4394167,
              "compute_passes": 1,
              "command_buffer_ns": 3895625.00641,
              "command_buffers": 2
            },
            {
              "compute_ns": 3833042,
              "compute_span_ns": 3833042,
              "calibration_cpu_ns": 4339583,
              "calibration_gpu_ticks": 4339583,
              "compute_passes": 1,
              "command_buffer_ns": 3833083.31948,
              "command_buffers": 2
            },
            {
              "compute_ns": 3924625,
              "compute_span_ns": 3924625,
              "calibration_cpu_ns": 4418459,
              "calibration_gpu_ticks": 4418459,
              "compute_passes": 1,
              "command_buffer_ns": 3924708.34777,
              "command_buffers": 2
            },
            {
              "compute_ns": 3938958,
              "compute_span_ns": 3938958,
              "calibration_cpu_ns": 4386042,
              "calibration_gpu_ticks": 4386042,
              "compute_passes": 1,
              "command_buffer_ns": 3938999.95834,
              "command_buffers": 2
            },
            {
              "compute_ns": 3923959,
              "compute_span_ns": 3923959,
              "calibration_cpu_ns": 4382834,
              "calibration_gpu_ticks": 4382834,
              "compute_passes": 1,
              "command_buffer_ns": 3924041.69543,
              "command_buffers": 2
            },
            {
              "compute_ns": 3948041,
              "compute_span_ns": 3948041,
              "calibration_cpu_ns": 4407667,
              "calibration_gpu_ticks": 4407667,
              "compute_passes": 1,
              "command_buffer_ns": 3948041.70355,
              "command_buffers": 2
            },
            {
              "compute_ns": 3894250,
              "compute_span_ns": 3894250,
              "calibration_cpu_ns": 4371500,
              "calibration_gpu_ticks": 4371500,
              "compute_passes": 1,
              "command_buffer_ns": 3894374.93868,
              "command_buffers": 2
            },
            {
              "compute_ns": 3902500,
              "compute_span_ns": 3902500,
              "calibration_cpu_ns": 4606000,
              "calibration_gpu_ticks": 4606000,
              "compute_passes": 1,
              "command_buffer_ns": 3902541.64813,
              "command_buffers": 2
            },
            {
              "compute_ns": 3898292,
              "compute_span_ns": 3898292,
              "calibration_cpu_ns": 4393833,
              "calibration_gpu_ticks": 4393833,
              "compute_passes": 1,
              "command_buffer_ns": 3898458.32251,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 60875,
              "compute_span_ns": 60875,
              "calibration_cpu_ns": 507375,
              "calibration_gpu_ticks": 507375,
              "compute_passes": 1,
              "command_buffer_ns": 60958.2639299,
              "command_buffers": 2
            },
            {
              "compute_ns": 78167,
              "compute_span_ns": 78167,
              "calibration_cpu_ns": 614875,
              "calibration_gpu_ticks": 614875,
              "compute_passes": 1,
              "command_buffer_ns": 78208.3370723,
              "command_buffers": 2
            },
            {
              "compute_ns": 65167,
              "compute_span_ns": 65167,
              "calibration_cpu_ns": 537958,
              "calibration_gpu_ticks": 537958,
              "compute_passes": 1,
              "command_buffer_ns": 65208.2962915,
              "command_buffers": 2
            },
            {
              "compute_ns": 60750,
              "compute_span_ns": 60750,
              "calibration_cpu_ns": 566000,
              "calibration_gpu_ticks": 566000,
              "compute_passes": 1,
              "command_buffer_ns": 60791.6736044,
              "command_buffers": 2
            },
            {
              "compute_ns": 75417,
              "compute_span_ns": 75417,
              "calibration_cpu_ns": 539833,
              "calibration_gpu_ticks": 539833,
              "compute_passes": 1,
              "command_buffer_ns": 75458.3161324,
              "command_buffers": 2
            },
            {
              "compute_ns": 70250,
              "compute_span_ns": 70250,
              "calibration_cpu_ns": 514416,
              "calibration_gpu_ticks": 514416,
              "compute_passes": 1,
              "command_buffer_ns": 70291.6295268,
              "command_buffers": 2
            },
            {
              "compute_ns": 61458,
              "compute_span_ns": 61458,
              "calibration_cpu_ns": 491292,
              "calibration_gpu_ticks": 491292,
              "compute_passes": 1,
              "command_buffer_ns": 61458.3259448,
              "command_buffers": 2
            },
            {
              "compute_ns": 75667,
              "compute_span_ns": 75667,
              "calibration_cpu_ns": 568125,
              "calibration_gpu_ticks": 568125,
              "compute_passes": 1,
              "command_buffer_ns": 75749.9947213,
              "command_buffers": 2
            },
            {
              "compute_ns": 61583,
              "compute_span_ns": 61583,
              "calibration_cpu_ns": 504041,
              "calibration_gpu_ticks": 504041,
              "compute_passes": 1,
              "command_buffer_ns": 61583.3560005,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 3923458.39646,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3859874.90881,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3892125.03796,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3841958.29974,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3973083.33777,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4015666.66551,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3914791.6832,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3846708.33591,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3870000.01501,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 71208.3419785,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 73833.3328627,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 72708.3534002,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 75249.990914,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 75999.9966249,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 62083.3598077,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 62083.3598077,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 73500.0357963,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 61416.6492596,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              61.3040374446875,
              60.31054545015625,
              60.814453718124994,
              60.030598433437504,
              62.07942715265625,
              62.744791648593754,
              61.16862005,
              60.104817748593746,
              60.46875023453126
            ],
            "command_buffer_throughput_us_p50": 60.814453718124994,
            "command_buffer_throughput_us_p90": 62.212500051843755,
            "command_buffer_latency_us": [
              71.2083419785,
              73.8333328627,
              72.7083534002,
              75.249990914,
              75.99999662489999,
              62.0833598077,
              62.0833598077,
              73.5000357963,
              61.4166492596
            ],
            "command_buffer_latency_us_p50": 72.7083534002,
            "command_buffer_latency_us_p90": 75.39999205618
          },
          "compute_throughput_us": [
            60.869140625,
            59.89128125,
            61.322265625,
            61.54621875,
            61.311859375,
            61.688140625,
            60.84765625,
            60.9765625,
            60.9108125
          ],
          "compute_throughput_us_p50": 60.9765625,
          "compute_throughput_us_p90": 61.574603125,
          "command_buffer_throughput_us": [
            60.86914072515625,
            59.891926866875004,
            61.32356793390625,
            61.5468743490625,
            61.31315149109375,
            61.68815161796876,
            60.849608416875,
            60.97721325203125,
            60.91341128921875
          ],
          "command_buffer_throughput_us_p50": 60.97721325203125,
          "command_buffer_throughput_us_p90": 61.57512980284375,
          "compute_latency_us": [
            60.875,
            78.167,
            65.167,
            60.75,
            75.417,
            70.25,
            61.458,
            75.667,
            61.583
          ],
          "compute_latency_us_p50": 65.167,
          "compute_latency_us_p90": 76.167,
          "command_buffer_latency_us": [
            60.9582639299,
            78.2083370723,
            65.2082962915,
            60.791673604399996,
            75.4583161324,
            70.2916295268,
            61.4583259448,
            75.7499947213,
            61.583356000500004
          ],
          "command_buffer_latency_us_p50": 65.2082962915,
          "command_buffer_latency_us_p90": 76.2416631915,
          "counter_control_throughput_ratio": 0.9947847175093335,
          "counter_control_latency_ratio": 0.9928726247821631
        },
        "correctness": {
          "max_abs_error": 1.4915225321487924e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 60.0632780083,
        "throughput_us_p90": 60.33782946058,
        "latency_us_p50": 305.666,
        "latency_us_p90": 362.1916
      },
      "slowdown": 0.733997055466518,
      "device_compute_slowdown": 0.7685364997234057,
      "valid": true,
      "round": 3,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "rmsnorm_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-ug2g8zog/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "26111b6b600e52c50494ad3f783ef6fc3b02252d329d47ea39eb66749773d18e",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.699875,
        "capture_ms": 0.06925,
        "compile_ms": 52.188,
        "allocation_upload_ms": 5.665542,
        "cold_call_ms": 5.79275,
        "warmup_ms": 200.559125,
        "download_ms": 3.072125,
        "repetitions": 524,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          53.2163645038,
          53.5580458015,
          52.3951965649,
          53.4526889313,
          53.3884370229,
          52.8742843511,
          53.6303263359,
          53.4328874046,
          53.4165877863
        ],
        "latency_us": [
          286.083,
          264.708,
          347.209,
          261.25,
          265.334,
          277.833,
          258.917,
          259.208,
          266.917
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 32,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 1024,
            "reduction_scalar_rounds": 48,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 32768,
              "global_write_bytes": 16384,
              "private_read_bytes": 49152,
              "private_write_bytes": 16384
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 128,
              "global_write_bytes": 64,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 1,
            "reduction_elements": 4096,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 80,
            "normalized_kernel_cost": 1280,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 8192,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3353625,
              "compute_span_ns": 3353625,
              "calibration_cpu_ns": 3833417,
              "calibration_gpu_ticks": 3833417,
              "compute_passes": 1,
              "command_buffer_ns": 3353666.68878,
              "command_buffers": 2
            },
            {
              "compute_ns": 3426292,
              "compute_span_ns": 3426292,
              "calibration_cpu_ns": 3893583,
              "calibration_gpu_ticks": 3893583,
              "compute_passes": 1,
              "command_buffer_ns": 3426333.3655,
              "command_buffers": 2
            },
            {
              "compute_ns": 3312125,
              "compute_span_ns": 3312125,
              "calibration_cpu_ns": 3840666,
              "calibration_gpu_ticks": 3840666,
              "compute_passes": 1,
              "command_buffer_ns": 3312124.98713,
              "command_buffers": 2
            },
            {
              "compute_ns": 3366750,
              "compute_span_ns": 3366750,
              "calibration_cpu_ns": 3978750,
              "calibration_gpu_ticks": 3978750,
              "compute_passes": 1,
              "command_buffer_ns": 3366791.6432,
              "command_buffers": 2
            },
            {
              "compute_ns": 3387833,
              "compute_span_ns": 3387833,
              "calibration_cpu_ns": 3893542,
              "calibration_gpu_ticks": 3893542,
              "compute_passes": 1,
              "command_buffer_ns": 3387875.04006,
              "command_buffers": 2
            },
            {
              "compute_ns": 3352708,
              "compute_span_ns": 3352708,
              "calibration_cpu_ns": 3789458,
              "calibration_gpu_ticks": 3789458,
              "compute_passes": 1,
              "command_buffer_ns": 3352791.71122,
              "command_buffers": 2
            },
            {
              "compute_ns": 3404875,
              "compute_span_ns": 3404875,
              "calibration_cpu_ns": 3836958,
              "calibration_gpu_ticks": 3836958,
              "compute_passes": 1,
              "command_buffer_ns": 3404958.40646,
              "command_buffers": 2
            },
            {
              "compute_ns": 3367584,
              "compute_span_ns": 3367584,
              "calibration_cpu_ns": 3840291,
              "calibration_gpu_ticks": 3840291,
              "compute_passes": 1,
              "command_buffer_ns": 3367708.35565,
              "command_buffers": 2
            },
            {
              "compute_ns": 3391166,
              "compute_span_ns": 3391166,
              "calibration_cpu_ns": 3829542,
              "calibration_gpu_ticks": 3829542,
              "compute_passes": 1,
              "command_buffer_ns": 3391166.68329,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 54125,
              "compute_span_ns": 54125,
              "calibration_cpu_ns": 475834,
              "calibration_gpu_ticks": 475834,
              "compute_passes": 1,
              "command_buffer_ns": 54166.7104699,
              "command_buffers": 2
            },
            {
              "compute_ns": 64500,
              "compute_span_ns": 64500,
              "calibration_cpu_ns": 492500,
              "calibration_gpu_ticks": 492500,
              "compute_passes": 1,
              "command_buffer_ns": 64541.7021587,
              "command_buffers": 2
            },
            {
              "compute_ns": 53833,
              "compute_span_ns": 53833,
              "calibration_cpu_ns": 443167,
              "calibration_gpu_ticks": 443167,
              "compute_passes": 1,
              "command_buffer_ns": 53874.9736734,
              "command_buffers": 2
            },
            {
              "compute_ns": 54834,
              "compute_span_ns": 54834,
              "calibration_cpu_ns": 422833,
              "calibration_gpu_ticks": 422833,
              "compute_passes": 1,
              "command_buffer_ns": 54999.9531358,
              "command_buffers": 2
            },
            {
              "compute_ns": 58125,
              "compute_span_ns": 58125,
              "calibration_cpu_ns": 532875,
              "calibration_gpu_ticks": 532875,
              "compute_passes": 1,
              "command_buffer_ns": 58166.6245125,
              "command_buffers": 2
            },
            {
              "compute_ns": 64292,
              "compute_span_ns": 64292,
              "calibration_cpu_ns": 488417,
              "calibration_gpu_ticks": 488417,
              "compute_passes": 1,
              "command_buffer_ns": 64333.3769403,
              "command_buffers": 2
            },
            {
              "compute_ns": 57833,
              "compute_span_ns": 57833,
              "calibration_cpu_ns": 461458,
              "calibration_gpu_ticks": 461458,
              "compute_passes": 1,
              "command_buffer_ns": 57833.3274461,
              "command_buffers": 2
            },
            {
              "compute_ns": 54334,
              "compute_span_ns": 54334,
              "calibration_cpu_ns": 451041,
              "calibration_gpu_ticks": 451041,
              "compute_passes": 1,
              "command_buffer_ns": 54416.6541658,
              "command_buffers": 2
            },
            {
              "compute_ns": 53916,
              "compute_span_ns": 53916,
              "calibration_cpu_ns": 510166,
              "calibration_gpu_ticks": 510166,
              "compute_passes": 1,
              "command_buffer_ns": 53875.031881,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 3479333.36161,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3448875.03888,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3331083.39692,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3468916.63503,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3350958.34453,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3365875.04717,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3455666.65055,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3395666.65934,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3374916.61776,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 54333.3590031,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 66333.3339617,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54500.0075363,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54374.9774806,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 54333.3590031,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65708.3000988,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 57250.0284761,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 55000.0113435,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 56833.3198316,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              54.36458377515625,
              53.8886724825,
              52.048178076875004,
              54.20182242234375,
              52.35872413328125,
              52.59179761203125,
              53.994791414843746,
              53.05729155218749,
              52.7330721525
            ],
            "command_buffer_throughput_us_p50": 53.05729155218749,
            "command_buffer_throughput_us_p90": 54.23437469290625,
            "command_buffer_latency_us": [
              54.3333590031,
              66.3333339617,
              54.5000075363,
              54.3749774806,
              54.3333590031,
              65.7083000988,
              57.2500284761,
              55.0000113435,
              56.8333198316
            ],
            "command_buffer_latency_us_p50": 55.0000113435,
            "command_buffer_latency_us_p90": 65.83330687138
          },
          "compute_throughput_us": [
            52.400390625,
            53.5358125,
            51.751953125,
            52.60546875,
            52.934890625,
            52.3860625,
            53.201171875,
            52.6185,
            52.98696875
          ],
          "compute_throughput_us_p50": 52.6185,
          "compute_throughput_us_p90": 53.2681,
          "command_buffer_throughput_us": [
            52.4010420121875,
            53.536458835937495,
            51.751952923906245,
            52.606119424999996,
            52.9355475009375,
            52.3873704878125,
            53.202475100937505,
            52.620443057031245,
            52.98697942640625
          ],
          "command_buffer_throughput_us_p50": 52.620443057031245,
          "command_buffer_throughput_us_p90": 53.2692718479375,
          "compute_latency_us": [
            54.125,
            64.5,
            53.833,
            54.834,
            58.125,
            64.292,
            57.833,
            54.334,
            53.916
          ],
          "compute_latency_us_p50": 54.834,
          "compute_latency_us_p90": 64.3336,
          "command_buffer_latency_us": [
            54.166710469899996,
            64.5417021587,
            53.87497367340001,
            54.999953135800006,
            58.1666245125,
            64.3333769403,
            57.83332744609999,
            54.4166541658,
            53.875031881000005
          ],
          "command_buffer_latency_us_p50": 54.999953135800006,
          "command_buffer_latency_us_p90": 64.37504198398,
          "counter_control_throughput_ratio": 0.9934640504147345,
          "counter_control_latency_ratio": 0.9893935080475407
        },
        "correctness": {
          "max_abs_error": 1.5832164268481108e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 53.4165877863,
        "throughput_us_p90": 53.57250190838,
        "latency_us_p50": 265.334,
        "latency_us_p90": 298.30820000000006
      },
      "torch": {
        "cold_call_ms": 7.380666,
        "warmup_ms": 200.218166,
        "repetitions": 406,
        "throughput_us": [
          73.30326354679804,
          74.56999261083745,
          72.35601231527095,
          71.3527315270936,
          72.05726600985221,
          71.25082266009852,
          71.2481527093596,
          71.99404926108375,
          71.02575862068966
        ],
        "latency_us": [
          456.166,
          369.334,
          309.70799999999997,
          317.541,
          320.083,
          453.125,
          367.042,
          331.45799999999997,
          314.458
        ],
        "throughput_us_p50": 71.99404926108375,
        "throughput_us_p90": 73.55660935960591,
        "latency_us_p50": 331.45799999999997,
        "latency_us_p90": 453.7332,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4405875.049997121,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4391374.997794628,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4398416.669573635,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4506708.355620503,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4379583.289846778,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4515541.659202427,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4476833.331864327,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4424791.666679084,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4383624.996989965,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 86625.04842504859,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 69999.95093792677,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 71249.96045604348,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 84125.02938881516,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70666.71969369054,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70791.63333401084,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 73333.3290554583,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 85458.33406969905,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 87249.96587261558,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              68.84179765620502,
              68.61523434054106,
              68.72526046208804,
              70.41731805657037,
              68.4309889038559,
              70.55533842503792,
              69.95052081038011,
              69.13736979186069,
              68.49414057796821
            ],
            "command_buffer_throughput_us_p50": 68.84179765620502,
            "command_buffer_throughput_us_p90": 70.44492213026388,
            "command_buffer_latency_us": [
              86.62504842504859,
              69.99995093792677,
              71.24996045604348,
              84.12502938881516,
              70.66671969369054,
              70.79163333401084,
              73.33332905545831,
              85.45833406969905,
              87.24996587261558
            ],
            "command_buffer_latency_us_p50": 73.33332905545831,
            "command_buffer_latency_us_p90": 86.75003191456199
          },
          "throughput": [
            {
              "compute_ns": 4493125.0,
              "compute_span_ns": 4493125.0,
              "command_buffer_ns": 4493125.015869737,
              "calibration_cpu_ns": 5212041.0,
              "calibration_gpu_ticks": 5212041.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4533375.0,
              "compute_span_ns": 4533375.0,
              "command_buffer_ns": 4533375.03131479,
              "calibration_cpu_ns": 5244292.0,
              "calibration_gpu_ticks": 5244292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4491375.0,
              "compute_span_ns": 4491375.0,
              "command_buffer_ns": 4491375.002544373,
              "calibration_cpu_ns": 5122708.0,
              "calibration_gpu_ticks": 5122708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4486000.0,
              "compute_span_ns": 4486000.0,
              "command_buffer_ns": 4485999.990720302,
              "calibration_cpu_ns": 5164625.0,
              "calibration_gpu_ticks": 5164625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4444167.0,
              "compute_span_ns": 4444167.0,
              "command_buffer_ns": 4444208.345375955,
              "calibration_cpu_ns": 5154250.0,
              "calibration_gpu_ticks": 5154250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4480416.0,
              "compute_span_ns": 4480416.0,
              "command_buffer_ns": 4480374.976992607,
              "calibration_cpu_ns": 5176000.0,
              "calibration_gpu_ticks": 5176000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4478916.0,
              "compute_span_ns": 4478916.0,
              "command_buffer_ns": 4478875.023778528,
              "calibration_cpu_ns": 5248833.0,
              "calibration_gpu_ticks": 5248833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4507958.0,
              "compute_span_ns": 4507958.0,
              "command_buffer_ns": 4507916.630245745,
              "calibration_cpu_ns": 5338292.0,
              "calibration_gpu_ticks": 5338292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4530042.0,
              "compute_span_ns": 4530042.0,
              "command_buffer_ns": 4530041.71140492,
              "calibration_cpu_ns": 5340791.0,
              "calibration_gpu_ticks": 5340791.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 70667.0,
              "compute_span_ns": 70667.0,
              "command_buffer_ns": 70666.66148602962,
              "calibration_cpu_ns": 517666.0,
              "calibration_gpu_ticks": 517666.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70417.0,
              "compute_span_ns": 70417.0,
              "command_buffer_ns": 70458.33626762033,
              "calibration_cpu_ns": 558583.0,
              "calibration_gpu_ticks": 558583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70250.0,
              "compute_span_ns": 70250.0,
              "command_buffer_ns": 70250.01104921103,
              "calibration_cpu_ns": 551542.0,
              "calibration_gpu_ticks": 551542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 84417.0,
              "compute_span_ns": 84417.0,
              "command_buffer_ns": 84458.32645520568,
              "calibration_cpu_ns": 617458.0,
              "calibration_gpu_ticks": 617458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 71125.0,
              "compute_span_ns": 71125.0,
              "command_buffer_ns": 71125.04681572318,
              "calibration_cpu_ns": 540959.0,
              "calibration_gpu_ticks": 540959.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70250.0,
              "compute_span_ns": 70250.0,
              "command_buffer_ns": 70250.01104921103,
              "calibration_cpu_ns": 537625.0,
              "calibration_gpu_ticks": 537625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70666.0,
              "compute_span_ns": 70666.0,
              "command_buffer_ns": 70624.98480081558,
              "calibration_cpu_ns": 527959.0,
              "calibration_gpu_ticks": 527959.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 73792.0,
              "compute_span_ns": 73792.0,
              "command_buffer_ns": 73791.65617749095,
              "calibration_cpu_ns": 628125.0,
              "calibration_gpu_ticks": 628125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 72292.0,
              "compute_span_ns": 72292.0,
              "command_buffer_ns": 72291.6447557509,
              "calibration_cpu_ns": 585959.0,
              "calibration_gpu_ticks": 585959.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            70.205078125,
            70.833984375,
            70.177734375,
            70.09375,
            69.440109375,
            70.0065,
            69.9830625,
            70.43684375,
            70.78190625
          ],
          "compute_throughput_us_p50": 70.177734375,
          "compute_throughput_us_p90": 70.792321875,
          "command_buffer_throughput_us": [
            70.20507837296464,
            70.8339848642936,
            70.17773441475583,
            70.09374985500472,
            69.4407553964993,
            70.00585901550949,
            69.9824222465395,
            70.43619734758977,
            70.78190174070187
          ],
          "command_buffer_throughput_us_p50": 70.17773441475583,
          "command_buffer_throughput_us_p90": 70.79231836542021,
          "compute_latency_us": [
            70.667,
            70.417,
            70.25,
            84.417,
            71.125,
            70.25,
            70.666,
            73.792,
            72.292
          ],
          "compute_latency_us_p50": 70.667,
          "compute_latency_us_p90": 75.917,
          "command_buffer_latency_us": [
            70.66666148602962,
            70.45833626762033,
            70.25001104921103,
            84.45832645520568,
            71.12504681572318,
            70.25001104921103,
            70.62498480081558,
            73.79165617749095,
            72.2916447557509
          ],
          "command_buffer_latency_us_p50": 70.66666148602962,
          "command_buffer_latency_us_p90": 75.9249902330339,
          "counter_control_throughput_ratio": 1.0187861869729673,
          "counter_control_latency_ratio": 0.9859656145711216
        },
        "download_ms": 0.847208,
        "allocation_upload_ms": 15.727125,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.8339726848282112e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.7419583748177121,
      "device_compute_slowdown": 0.7497890957668865,
      "valid": true,
      "round": 3,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "rmsnorm",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "rmsnorm_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.4065,
        "warmup_ms": 200.006375,
        "repetitions": 405,
        "throughput_us": [
          72.13672839506172,
          72.40195308641975,
          72.37798518518518,
          73.11337283950617,
          72.75514320987654,
          72.92602716049383,
          73.29310864197531,
          73.57942469135803,
          74.4751012345679
        ],
        "latency_us": [
          356.66700000000003,
          316.541,
          316.5,
          448.458,
          316.459,
          287.917,
          296.58299999999997,
          292.208,
          297.916
        ],
        "throughput_us_p50": 72.92602716049383,
        "throughput_us_p90": 73.75856,
        "latency_us_p50": 316.459,
        "latency_us_p90": 375.02520000000004,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4532875.027507544,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4602874.97844547,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4599333.33331719,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4448624.968063086,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4574291.640892625,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4599083.331413567,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4686500.004027039,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4503458.3308734,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 4539625.020697713,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 72041.70105978847,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 86708.34358781576,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 84750.00504404306,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 72583.3815522492,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70291.62952676415,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 84208.32455158234,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70250.01104921103,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 71083.31192284822,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 83666.64405912161,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              70.82617230480537,
              71.91992153821047,
              71.8645833330811,
              69.50976512598572,
              71.47330688894726,
              71.86067705333699,
              73.22656256292248,
              70.36653641989687,
              70.93164094840176
            ],
            "command_buffer_throughput_us_p50": 71.47330688894726,
            "command_buffer_throughput_us_p90": 72.18124974315288,
            "command_buffer_latency_us": [
              72.04170105978847,
              86.70834358781576,
              84.75000504404306,
              72.5833815522492,
              70.29162952676415,
              84.20832455158234,
              70.25001104921103,
              71.08331192284822,
              83.66664405912161
            ],
            "command_buffer_latency_us_p50": 72.5833815522492,
            "command_buffer_latency_us_p90": 85.1416727527976
          },
          "throughput": [
            {
              "compute_ns": 4388458.0,
              "compute_span_ns": 4388458.0,
              "command_buffer_ns": 4388458.328321576,
              "calibration_cpu_ns": 5052667.0,
              "calibration_gpu_ticks": 5052667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4620625.0,
              "compute_span_ns": 4620625.0,
              "command_buffer_ns": 4620624.997187406,
              "calibration_cpu_ns": 5415375.0,
              "calibration_gpu_ticks": 5415375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4515125.0,
              "compute_span_ns": 4515125.0,
              "command_buffer_ns": 4515125.008765608,
              "calibration_cpu_ns": 5155333.0,
              "calibration_gpu_ticks": 5155333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4719417.0,
              "compute_span_ns": 4719417.0,
              "command_buffer_ns": 4719458.345789462,
              "calibration_cpu_ns": 5552791.0,
              "calibration_gpu_ticks": 5552791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4626708.0,
              "compute_span_ns": 4626708.0,
              "command_buffer_ns": 4626666.661351919,
              "calibration_cpu_ns": 5281958.0,
              "calibration_gpu_ticks": 5281958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4385208.0,
              "compute_span_ns": 4385208.0,
              "command_buffer_ns": 4385166.626889259,
              "calibration_cpu_ns": 5213833.0,
              "calibration_gpu_ticks": 5213833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4364250.0,
              "compute_span_ns": 4364250.0,
              "command_buffer_ns": 4364249.994978309,
              "calibration_cpu_ns": 5000292.0,
              "calibration_gpu_ticks": 5000292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4542958.0,
              "compute_span_ns": 4542958.0,
              "command_buffer_ns": 4542916.6639223695,
              "calibration_cpu_ns": 5400416.0,
              "calibration_gpu_ticks": 5400416.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 4445250.0,
              "compute_span_ns": 4445250.0,
              "command_buffer_ns": 4445249.971468002,
              "calibration_cpu_ns": 5098459.0,
              "calibration_gpu_ticks": 5098459.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 71458.0,
              "compute_span_ns": 71458.0,
              "command_buffer_ns": 71458.3438821137,
              "calibration_cpu_ns": 535125.0,
              "calibration_gpu_ticks": 535125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70917.0,
              "compute_span_ns": 70917.0,
              "command_buffer_ns": 70916.66338965297,
              "calibration_cpu_ns": 516667.0,
              "calibration_gpu_ticks": 516667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 84500.0,
              "compute_span_ns": 84500.0,
              "command_buffer_ns": 84500.00314041972,
              "calibration_cpu_ns": 581209.0,
              "calibration_gpu_ticks": 581209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 71209.0,
              "compute_span_ns": 71209.0,
              "command_buffer_ns": 71250.0186637044,
              "calibration_cpu_ns": 502833.0,
              "calibration_gpu_ticks": 502833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 82875.0,
              "compute_span_ns": 82875.0,
              "command_buffer_ns": 82874.96166303754,
              "calibration_cpu_ns": 644959.0,
              "calibration_gpu_ticks": 644959.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 84375.0,
              "compute_span_ns": 84375.0,
              "command_buffer_ns": 84374.9730847776,
              "calibration_cpu_ns": 540667.0,
              "calibration_gpu_ticks": 540667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 72000.0,
              "compute_span_ns": 72000.0,
              "command_buffer_ns": 72000.02437457442,
              "calibration_cpu_ns": 481667.0,
              "calibration_gpu_ticks": 481667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 83000.0,
              "compute_span_ns": 83000.0,
              "command_buffer_ns": 82999.99171867967,
              "calibration_cpu_ns": 555875.0,
              "calibration_gpu_ticks": 555875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 84500.0,
              "compute_span_ns": 84500.0,
              "command_buffer_ns": 84500.00314041972,
              "calibration_cpu_ns": 594292.0,
              "calibration_gpu_ticks": 594292.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            68.56965625,
            72.197265625,
            70.548828125,
            73.740890625,
            72.2923125,
            68.518875,
            68.19140625,
            70.98371875,
            69.45703125
          ],
          "compute_throughput_us_p50": 70.548828125,
          "compute_throughput_us_p90": 72.582028125,
          "command_buffer_throughput_us": [
            68.56966138002463,
            72.19726558105322,
            70.54882826196263,
            73.74153665296035,
            72.29166658362374,
            68.51822854514467,
            68.19140617153607,
            70.98307287378702,
            69.45703080418753
          ],
          "command_buffer_throughput_us_p50": 70.54882826196263,
          "command_buffer_throughput_us_p90": 72.58164059749106,
          "compute_latency_us": [
            71.458,
            70.917,
            84.5,
            71.209,
            82.875,
            84.375,
            72.0,
            83.0,
            84.5
          ],
          "compute_latency_us_p50": 82.875,
          "compute_latency_us_p90": 84.5,
          "command_buffer_latency_us": [
            71.4583438821137,
            70.91666338965297,
            84.50000314041972,
            71.2500186637044,
            82.87496166303754,
            84.3749730847776,
            72.00002437457442,
            82.99999171867967,
            84.50000314041972
          ],
          "command_buffer_latency_us_p50": 82.87496166303754,
          "command_buffer_latency_us_p90": 84.50000314041972,
          "counter_control_throughput_ratio": 0.9816911890378581,
          "counter_control_latency_ratio": 1.001979003074609
        },
        "download_ms": 0.663625,
        "allocation_upload_ms": 1.380083,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 1.8339726848282112e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "rmsnorm",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-5t6amlpr/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "310a51aaac34ebd04d17becc4645f77793138816428bb43603834b11ab1687fb",
      "native": {
        "backend": "metal",
        "operation": "rmsnorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.64675,
        "capture_ms": 0.060708,
        "compile_ms": 51.233667,
        "allocation_upload_ms": 6.664417,
        "cold_call_ms": 11.7555,
        "warmup_ms": 200.317458,
        "download_ms": 2.959917,
        "repetitions": 449,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          64.7372873051,
          65.3519844098,
          65.832311804,
          65.6242583519,
          65.1752026726,
          65.295935412,
          65.6803073497,
          64.8470668151,
          65.5991091314
        ],
        "latency_us": [
          339.75,
          274.042,
          283.833,
          283.042,
          270.041,
          268.292,
          275.458,
          326.625,
          289.083
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 512,
            "reduction_scalar_rounds": 48,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 32768,
              "global_write_bytes": 16384,
              "private_read_bytes": 49152,
              "private_write_bytes": 16384
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 128,
              "global_write_bytes": 64,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 1,
            "reduction_elements": 4096,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 1,
            "group_barrier_sites_after": 1,
            "independent_subgroups": false,
            "normalized_cost": 72,
            "normalized_kernel_cost": 1152,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 8192,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4159792,
              "compute_span_ns": 4159792,
              "calibration_cpu_ns": 4655083,
              "calibration_gpu_ticks": 4655083,
              "compute_passes": 1,
              "command_buffer_ns": 4159875.03948,
              "command_buffers": 2
            },
            {
              "compute_ns": 4222500,
              "compute_span_ns": 4222500,
              "calibration_cpu_ns": 4728084,
              "calibration_gpu_ticks": 4728084,
              "compute_passes": 1,
              "command_buffer_ns": 4222625.05163,
              "command_buffers": 2
            },
            {
              "compute_ns": 4283625,
              "compute_span_ns": 4283625,
              "calibration_cpu_ns": 4775209,
              "calibration_gpu_ticks": 4775209,
              "compute_passes": 1,
              "command_buffer_ns": 4283750.0223,
              "command_buffers": 2
            },
            {
              "compute_ns": 4312167,
              "compute_span_ns": 4312167,
              "calibration_cpu_ns": 4694333,
              "calibration_gpu_ticks": 4694333,
              "compute_passes": 1,
              "command_buffer_ns": 4312250.00648,
              "command_buffers": 2
            },
            {
              "compute_ns": 4107459,
              "compute_span_ns": 4107459,
              "calibration_cpu_ns": 4624208,
              "calibration_gpu_ticks": 4624208,
              "compute_passes": 1,
              "command_buffer_ns": 4107625.04907,
              "command_buffers": 2
            },
            {
              "compute_ns": 4303417,
              "compute_span_ns": 4303417,
              "calibration_cpu_ns": 4841500,
              "calibration_gpu_ticks": 4841500,
              "compute_passes": 1,
              "command_buffer_ns": 4303458.37958,
              "command_buffers": 2
            },
            {
              "compute_ns": 4236167,
              "compute_span_ns": 4236167,
              "calibration_cpu_ns": 4765833,
              "calibration_gpu_ticks": 4765833,
              "compute_passes": 1,
              "command_buffer_ns": 4236333.30502,
              "command_buffers": 2
            },
            {
              "compute_ns": 4239458,
              "compute_span_ns": 4239458,
              "calibration_cpu_ns": 4729125,
              "calibration_gpu_ticks": 4729125,
              "compute_passes": 1,
              "command_buffer_ns": 4239541.71129,
              "command_buffers": 2
            },
            {
              "compute_ns": 4197042,
              "compute_span_ns": 4197042,
              "calibration_cpu_ns": 4701667,
              "calibration_gpu_ticks": 4701667,
              "compute_passes": 1,
              "command_buffer_ns": 4197208.32724,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 65291,
              "compute_span_ns": 65291,
              "calibration_cpu_ns": 556292,
              "calibration_gpu_ticks": 556292,
              "compute_passes": 1,
              "command_buffer_ns": 65333.3263472,
              "command_buffers": 2
            },
            {
              "compute_ns": 67459,
              "compute_span_ns": 67459,
              "calibration_cpu_ns": 539250,
              "calibration_gpu_ticks": 539250,
              "compute_passes": 1,
              "command_buffer_ns": 67541.6667946,
              "command_buffers": 2
            },
            {
              "compute_ns": 76750,
              "compute_span_ns": 76750,
              "calibration_cpu_ns": 840333,
              "calibration_gpu_ticks": 840333,
              "compute_passes": 1,
              "command_buffer_ns": 76791.679021,
              "command_buffers": 2
            },
            {
              "compute_ns": 65042,
              "compute_span_ns": 65042,
              "calibration_cpu_ns": 472917,
              "calibration_gpu_ticks": 472917,
              "compute_passes": 1,
              "command_buffer_ns": 65041.705966,
              "command_buffers": 2
            },
            {
              "compute_ns": 76667,
              "compute_span_ns": 76667,
              "calibration_cpu_ns": 564167,
              "calibration_gpu_ticks": 564167,
              "compute_passes": 1,
              "command_buffer_ns": 76791.679021,
              "command_buffers": 2
            },
            {
              "compute_ns": 82584,
              "compute_span_ns": 82584,
              "calibration_cpu_ns": 567750,
              "calibration_gpu_ticks": 567750,
              "compute_passes": 1,
              "command_buffer_ns": 82624.9597594,
              "command_buffers": 2
            },
            {
              "compute_ns": 65250,
              "compute_span_ns": 65250,
              "calibration_cpu_ns": 510167,
              "calibration_gpu_ticks": 510167,
              "compute_passes": 1,
              "command_buffer_ns": 65291.649662,
              "command_buffers": 2
            },
            {
              "compute_ns": 66500,
              "compute_span_ns": 66500,
              "calibration_cpu_ns": 505833,
              "calibration_gpu_ticks": 505833,
              "compute_passes": 1,
              "command_buffer_ns": 66499.9824949,
              "command_buffers": 2
            },
            {
              "compute_ns": 73291,
              "compute_span_ns": 73291,
              "calibration_cpu_ns": 554667,
              "calibration_gpu_ticks": 554667,
              "compute_passes": 1,
              "command_buffer_ns": 73375.0057407,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4294166.69067,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4142374.96443,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4181916.65085,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4090374.97593,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4251333.36103,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4205958.27745,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4158125.02615,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4340791.66735,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4092416.66785,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 65333.3263472,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65249.9729767,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76583.3538026,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65125.0011288,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65083.3826512,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 65208.3544992,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 64749.9691695,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 77208.3294578,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 86791.6969582,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              67.09635454171875,
              64.72460881921876,
              65.34244766953125,
              63.912108998906255,
              66.42708376609376,
              65.71809808515624,
              64.97070353359375,
              67.82486980234374,
              63.94401043515625
            ],
            "command_buffer_throughput_us_p50": 65.34244766953125,
            "command_buffer_throughput_us_p90": 67.24205759384375,
            "command_buffer_latency_us": [
              65.3333263472,
              65.24997297670001,
              76.5833538026,
              65.12500112880001,
              65.0833826512,
              65.2083544992,
              64.7499691695,
              77.2083294578,
              86.7916969582
            ],
            "command_buffer_latency_us_p50": 65.24997297670001,
            "command_buffer_latency_us_p90": 79.12500295788001
          },
          "compute_throughput_us": [
            64.99675,
            65.9765625,
            66.931640625,
            67.377609375,
            64.179046875,
            67.240890625,
            66.190109375,
            66.24153125,
            65.57878125
          ],
          "compute_throughput_us_p50": 66.190109375,
          "compute_throughput_us_p90": 67.268234375,
          "command_buffer_throughput_us": [
            64.998047491875,
            65.97851643171875,
            66.93359409843751,
            67.37890635125,
            64.18164139171876,
            67.24153718093751,
            66.1927078909375,
            66.24283923890626,
            65.581380113125
          ],
          "command_buffer_throughput_us_p50": 66.1927078909375,
          "command_buffer_throughput_us_p90": 67.269011015,
          "compute_latency_us": [
            65.291,
            67.459,
            76.75,
            65.042,
            76.667,
            82.584,
            65.25,
            66.5,
            73.291
          ],
          "compute_latency_us_p50": 67.459,
          "compute_latency_us_p90": 77.9168,
          "command_buffer_latency_us": [
            65.3333263472,
            67.54166679459999,
            76.79167902100001,
            65.041705966,
            76.79167902100001,
            82.6249597594,
            65.291649662,
            66.49998249490001,
            73.3750057407
          ],
          "command_buffer_latency_us_p50": 67.54166679459999,
          "command_buffer_latency_us_p90": 77.95833516868001,
          "counter_control_throughput_ratio": 1.019372965482144,
          "counter_control_latency_ratio": 1.0027202415153686
        },
        "correctness": {
          "max_abs_error": 1.5832164268481108e-07,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 65.3519844098,
        "throughput_us_p90": 65.71070824056,
        "latency_us_p50": 283.042,
        "latency_us_p90": 329.25
      },
      "slowdown": 0.8961407463754324,
      "device_compute_slowdown": 0.9382169928850255,
      "valid": true,
      "round": 3,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "layernorm_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-l0wa9snq/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "53243cd32bb4620fa91e934e3d500807c417fa70fc3ac1977dac07575a48be10",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.844459,
        "capture_ms": 0.072667,
        "compile_ms": 65.388167,
        "allocation_upload_ms": 1.298375,
        "cold_call_ms": 1.973709,
        "warmup_ms": 200.16875,
        "download_ms": 0.272125,
        "repetitions": 4965,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          4.92341389728,
          5.42305317221,
          5.37196213494,
          5.39360523666,
          5.4710306143,
          5.45047834844,
          5.35303786506,
          5.36133756294,
          5.40877804632
        ],
        "latency_us": [
          227.541,
          230.708,
          248,
          230.166,
          240.125,
          224.583,
          257.291,
          261.917,
          238.709
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 19,
            "reduction_scalar_rounds": 40,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 18444,
              "global_write_bytes": 6148,
              "private_read_bytes": 36888,
              "private_write_bytes": 12296
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 32,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 2,
            "reduction_elements": 3074,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 80,
            "normalized_kernel_cost": 80,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 4611,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 338750,
              "compute_span_ns": 338750,
              "calibration_cpu_ns": 771083,
              "calibration_gpu_ticks": 771083,
              "compute_passes": 1,
              "command_buffer_ns": 338874.990121,
              "command_buffers": 2
            },
            {
              "compute_ns": 339583,
              "compute_span_ns": 339583,
              "calibration_cpu_ns": 792000,
              "calibration_gpu_ticks": 792000,
              "compute_passes": 1,
              "command_buffer_ns": 339583.319146,
              "command_buffers": 2
            },
            {
              "compute_ns": 351292,
              "compute_span_ns": 351292,
              "calibration_cpu_ns": 763125,
              "calibration_gpu_ticks": 763125,
              "compute_passes": 1,
              "command_buffer_ns": 351374.910679,
              "command_buffers": 2
            },
            {
              "compute_ns": 362292,
              "compute_span_ns": 362292,
              "calibration_cpu_ns": 751292,
              "calibration_gpu_ticks": 751292,
              "compute_passes": 1,
              "command_buffer_ns": 362416.671123,
              "command_buffers": 2
            },
            {
              "compute_ns": 374666,
              "compute_span_ns": 374666,
              "calibration_cpu_ns": 781917,
              "calibration_gpu_ticks": 781917,
              "compute_passes": 1,
              "command_buffer_ns": 374708.324671,
              "command_buffers": 2
            },
            {
              "compute_ns": 364541,
              "compute_span_ns": 364541,
              "calibration_cpu_ns": 767833,
              "calibration_gpu_ticks": 767833,
              "compute_passes": 1,
              "command_buffer_ns": 364583.276678,
              "command_buffers": 2
            },
            {
              "compute_ns": 355042,
              "compute_span_ns": 355042,
              "calibration_cpu_ns": 768834,
              "calibration_gpu_ticks": 768834,
              "compute_passes": 1,
              "command_buffer_ns": 355124.997441,
              "command_buffers": 2
            },
            {
              "compute_ns": 380708,
              "compute_span_ns": 380708,
              "calibration_cpu_ns": 811458,
              "calibration_gpu_ticks": 811458,
              "compute_passes": 1,
              "command_buffer_ns": 380708.370358,
              "command_buffers": 2
            },
            {
              "compute_ns": 377333,
              "compute_span_ns": 377333,
              "calibration_cpu_ns": 785542,
              "calibration_gpu_ticks": 785542,
              "compute_passes": 1,
              "command_buffer_ns": 377333.315555,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 7583,
              "compute_span_ns": 7583,
              "calibration_cpu_ns": 469542,
              "calibration_gpu_ticks": 469542,
              "compute_passes": 1,
              "command_buffer_ns": 7625.02895668,
              "command_buffers": 2
            },
            {
              "compute_ns": 7750,
              "compute_span_ns": 7750,
              "calibration_cpu_ns": 378583,
              "calibration_gpu_ticks": 378583,
              "compute_passes": 1,
              "command_buffer_ns": 7916.64933786,
              "command_buffers": 2
            },
            {
              "compute_ns": 7584,
              "compute_span_ns": 7584,
              "calibration_cpu_ns": 369750,
              "calibration_gpu_ticks": 369750,
              "compute_passes": 1,
              "command_buffer_ns": 7708.38232711,
              "command_buffers": 2
            },
            {
              "compute_ns": 7708,
              "compute_span_ns": 7708,
              "calibration_cpu_ns": 369291,
              "calibration_gpu_ticks": 369291,
              "compute_passes": 1,
              "command_buffer_ns": 7666.64743423,
              "command_buffers": 2
            },
            {
              "compute_ns": 7708,
              "compute_span_ns": 7708,
              "calibration_cpu_ns": 375417,
              "calibration_gpu_ticks": 375417,
              "compute_passes": 1,
              "command_buffer_ns": 7708.38232711,
              "command_buffers": 2
            },
            {
              "compute_ns": 7708,
              "compute_span_ns": 7708,
              "calibration_cpu_ns": 372500,
              "calibration_gpu_ticks": 372500,
              "compute_passes": 1,
              "command_buffer_ns": 7666.64743423,
              "command_buffers": 2
            },
            {
              "compute_ns": 7791,
              "compute_span_ns": 7791,
              "calibration_cpu_ns": 408542,
              "calibration_gpu_ticks": 408542,
              "compute_passes": 1,
              "command_buffer_ns": 7791.67748988,
              "command_buffers": 2
            },
            {
              "compute_ns": 7708,
              "compute_span_ns": 7708,
              "calibration_cpu_ns": 407250,
              "calibration_gpu_ticks": 407250,
              "compute_passes": 1,
              "command_buffer_ns": 7666.64743423,
              "command_buffers": 2
            },
            {
              "compute_ns": 7500,
              "compute_span_ns": 7500,
              "calibration_cpu_ns": 370500,
              "calibration_gpu_ticks": 370500,
              "compute_passes": 1,
              "command_buffer_ns": 7583.35227147,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 365791.667718,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 356958.364137,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 363541.708793,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 382874.975912,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 344208.325259,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 364999.985322,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 356416.683644,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 366291.613318,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 387500.040233,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 7666.58922657,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7458.32221583,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7374.9688454,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7791.67748988,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7541.67558625,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 8250.00461191,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7583.35227147,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7833.35417509,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7833.29596743,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              5.71549480809375,
              5.577474439640625,
              5.680339199890626,
              5.9824214986249995,
              5.378255082171875,
              5.70312477065625,
              5.5690106819375,
              5.72330645809375,
              6.054688128640625
            ],
            "command_buffer_throughput_us_p50": 5.70312477065625,
            "command_buffer_throughput_us_p90": 5.996874824628125,
            "command_buffer_latency_us": [
              7.66658922657,
              7.45832221583,
              7.3749688454,
              7.7916774898800005,
              7.54167558625,
              8.25000461191,
              7.58335227147,
              7.83335417509,
              7.83329596743
            ],
            "command_buffer_latency_us_p50": 7.66658922657,
            "command_buffer_latency_us_p90": 7.916684262454
          },
          "compute_throughput_us": [
            5.29296875,
            5.305984375,
            5.4889375,
            5.6608125,
            5.85415625,
            5.695953125,
            5.54753125,
            5.9485625,
            5.895828125
          ],
          "compute_throughput_us_p50": 5.6608125,
          "compute_throughput_us_p90": 5.906375000000001,
          "command_buffer_throughput_us": [
            5.294921720640625,
            5.30598936165625,
            5.490232979359376,
            5.662760486296874,
            5.854817572984375,
            5.69661369809375,
            5.548828085015625,
            5.9485682868437495,
            5.895833055546874
          ],
          "command_buffer_throughput_us_p50": 5.662760486296874,
          "command_buffer_throughput_us_p90": 5.906380101806249,
          "compute_latency_us": [
            7.583,
            7.75,
            7.584,
            7.708,
            7.708,
            7.708,
            7.791,
            7.708,
            7.5
          ],
          "compute_latency_us_p50": 7.708,
          "compute_latency_us_p90": 7.7582,
          "command_buffer_latency_us": [
            7.6250289566800005,
            7.91664933786,
            7.70838232711,
            7.66664743423,
            7.70838232711,
            7.66664743423,
            7.7916774898800005,
            7.66664743423,
            7.58335227147
          ],
          "command_buffer_latency_us_p50": 7.66664743423,
          "command_buffer_latency_us_p90": 7.816671859476,
          "counter_control_throughput_ratio": 0.9737632938776294,
          "counter_control_latency_ratio": 0.9945790404752657
        },
        "correctness": {
          "max_abs_error": 2.1991937471099732e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 5.39360523666,
        "throughput_us_p90": 5.454588801612,
        "latency_us_p50": 238.709,
        "latency_us_p90": 258.2162
      },
      "torch": {
        "cold_call_ms": 8.971042,
        "warmup_ms": 200.333334,
        "repetitions": 1992,
        "throughput_us": [
          12.71607279116466,
          12.650769578313252,
          12.318963353413654,
          11.373472891566264,
          11.806559738955825,
          12.750606425702813,
          12.740629016064256,
          12.696390060240963,
          11.489478915662652
        ],
        "latency_us": [
          339.209,
          217.70800000000003,
          212.25,
          207.334,
          198.5,
          196.209,
          207.083,
          291.041,
          200.167
        ],
        "throughput_us_p50": 12.650769578313252,
        "throughput_us_p90": 12.742624497991967,
        "latency_us_p50": 207.334,
        "latency_us_p90": 300.6746,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 484333.3153985441,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 491666.6657663882,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 483999.96012449265,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 502458.36609974504,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 485000.0259466469,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 483666.66305810213,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 482250.00500679016,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 504874.97355788946,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 483750.0164285302,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 10250.01984089613,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10041.636414825916,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10333.373211324215,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10124.989785254002,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10208.343155682087,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10250.01984089613,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10250.01984089613,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10250.01984089613,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10250.01984089613,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              7.567708053102251,
              7.682291652599815,
              7.562499376945198,
              7.850911970308516,
              7.578125405416358,
              7.557291610282846,
              7.535156328231096,
              7.888671461842023,
              7.558594006695785
            ],
            "command_buffer_throughput_us_p50": 7.567708053102251,
            "command_buffer_throughput_us_p90": 7.8584638686152175,
            "command_buffer_latency_us": [
              10.25001984089613,
              10.041636414825916,
              10.333373211324215,
              10.124989785254002,
              10.208343155682087,
              10.25001984089613,
              10.25001984089613,
              10.25001984089613,
              10.25001984089613
            ],
            "command_buffer_latency_us_p50": 10.25001984089613,
            "command_buffer_latency_us_p90": 10.266690514981747
          },
          "throughput": [
            {
              "compute_ns": 543917.0,
              "compute_span_ns": 543917.0,
              "command_buffer_ns": 543958.3328552544,
              "calibration_cpu_ns": 1323083.0,
              "calibration_gpu_ticks": 1323083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 482167.0,
              "compute_span_ns": 482167.0,
              "command_buffer_ns": 482166.6516363621,
              "calibration_cpu_ns": 1247167.0,
              "calibration_gpu_ticks": 1247167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 483125.0,
              "compute_span_ns": 483125.0,
              "command_buffer_ns": 483125.0407733023,
              "calibration_cpu_ns": 1186791.0,
              "calibration_gpu_ticks": 1186791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 483875.0,
              "compute_span_ns": 483875.0,
              "command_buffer_ns": 483874.98827651143,
              "calibration_cpu_ns": 1179583.0,
              "calibration_gpu_ticks": 1179583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 529083.0,
              "compute_span_ns": 529083.0,
              "command_buffer_ns": 529041.6302159429,
              "calibration_cpu_ns": 1698667.0,
              "calibration_gpu_ticks": 1698667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 508291.0,
              "compute_span_ns": 508291.0,
              "command_buffer_ns": 508249.9701529741,
              "calibration_cpu_ns": 1331542.0,
              "calibration_gpu_ticks": 1331542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 484417.0,
              "compute_span_ns": 484417.0,
              "command_buffer_ns": 484416.66876897216,
              "calibration_cpu_ns": 1203792.0,
              "calibration_gpu_ticks": 1203792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 483125.0,
              "compute_span_ns": 483125.0,
              "command_buffer_ns": 483124.9825656414,
              "calibration_cpu_ns": 1164375.0,
              "calibration_gpu_ticks": 1164375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 481667.0,
              "compute_span_ns": 481667.0,
              "command_buffer_ns": 481708.32451432943,
              "calibration_cpu_ns": 1187750.0,
              "calibration_gpu_ticks": 1187750.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 10166.0,
              "compute_span_ns": 10166.0,
              "command_buffer_ns": 10124.989785254002,
              "calibration_cpu_ns": 404500.0,
              "calibration_gpu_ticks": 404500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10750.0,
              "compute_span_ns": 10750.0,
              "command_buffer_ns": 10750.023648142815,
              "calibration_cpu_ns": 500000.0,
              "calibration_gpu_ticks": 500000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10250.0,
              "compute_span_ns": 10250.0,
              "command_buffer_ns": 10250.01984089613,
              "calibration_cpu_ns": 428583.0,
              "calibration_gpu_ticks": 428583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10333.0,
              "compute_span_ns": 10333.0,
              "command_buffer_ns": 10333.373211324215,
              "calibration_cpu_ns": 463250.0,
              "calibration_gpu_ticks": 463250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10333.0,
              "compute_span_ns": 10333.0,
              "command_buffer_ns": 10291.638318449259,
              "calibration_cpu_ns": 412667.0,
              "calibration_gpu_ticks": 412667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10250.0,
              "compute_span_ns": 10250.0,
              "command_buffer_ns": 10250.01984089613,
              "calibration_cpu_ns": 459792.0,
              "calibration_gpu_ticks": 459792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10292.0,
              "compute_span_ns": 10292.0,
              "command_buffer_ns": 10291.696526110172,
              "calibration_cpu_ns": 424584.0,
              "calibration_gpu_ticks": 424584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10167.0,
              "compute_span_ns": 10167.0,
              "command_buffer_ns": 10208.343155682087,
              "calibration_cpu_ns": 424125.0,
              "calibration_gpu_ticks": 424125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10625.0,
              "compute_span_ns": 10625.0,
              "command_buffer_ns": 10624.993592500687,
              "calibration_cpu_ns": 436167.0,
              "calibration_gpu_ticks": 436167.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            8.498703125,
            7.533859375,
            7.548828125,
            7.560546875,
            8.266921875,
            7.942046875,
            7.569015625,
            7.548828125,
            7.526046875
          ],
          "compute_throughput_us_p50": 7.560546875,
          "compute_throughput_us_p90": 8.313278125,
          "command_buffer_throughput_us": [
            8.49934895086335,
            7.533853931818157,
            7.548828762082849,
            7.560546691820491,
            8.266275472124107,
            7.941405783640221,
            7.56901044951519,
            7.548827852588147,
            7.526692570536397
          ],
          "command_buffer_throughput_us_p50": 7.560546691820491,
          "command_buffer_throughput_us_p90": 8.312890167871956,
          "compute_latency_us": [
            10.166,
            10.75,
            10.25,
            10.333,
            10.333,
            10.25,
            10.292,
            10.167,
            10.625
          ],
          "compute_latency_us_p50": 10.292,
          "compute_latency_us_p90": 10.65,
          "command_buffer_latency_us": [
            10.124989785254002,
            10.750023648142815,
            10.25001984089613,
            10.333373211324215,
            10.291638318449259,
            10.25001984089613,
            10.291696526110172,
            10.208343155682087,
            10.624993592500687
          ],
          "command_buffer_latency_us_p50": 10.291638318449259,
          "command_buffer_latency_us_p90": 10.649999603629112,
          "counter_control_throughput_ratio": 0.9981923152411721,
          "counter_control_latency_ratio": 1.0040660101990981
        },
        "download_ms": 0.33425,
        "allocation_upload_ms": 1.003,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.0790672678193687e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.4263460181826455,
      "device_compute_slowdown": 0.7487305605786618,
      "valid": true,
      "round": 3,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "layernorm_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.270542,
        "warmup_ms": 200.28225,
        "repetitions": 1910,
        "throughput_us": [
          12.940117801047121,
          12.821029319371728,
          12.809751308900523,
          12.79622617801047,
          13.167517277486912,
          12.552464921465967,
          11.999389528795811,
          12.721421989528796,
          12.680388481675394
        ],
        "latency_us": [
          259.083,
          295.625,
          224.916,
          244.125,
          237.875,
          240.667,
          227.375,
          228.834,
          223.583
        ],
        "throughput_us_p50": 12.79622617801047,
        "throughput_us_p90": 12.985597696335079,
        "latency_us_p50": 237.875,
        "latency_us_p90": 266.39140000000003,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 560833.3158306777,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 582999.9572597444,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 568999.967072159,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 564999.9948218465,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 575583.3117291331,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 560750.0206679106,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 559041.6840277612,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 565875.0305883586,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 564249.9891109765,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 10499.963536858559,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10250.01984089613,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10708.346962928772,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10249.961633235216,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10374.991688877344,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10416.668374091387,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10500.021744519472,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10500.021744519472,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 10166.666470468044,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              8.76302055985434,
              9.109374332183506,
              8.890624485502485,
              8.828124919091351,
              8.993489245767705,
              8.761719072936103,
              8.735026312933769,
              8.841797352943104,
              8.816406079859007
            ],
            "command_buffer_throughput_us_p50": 8.828124919091351,
            "command_buffer_throughput_us_p90": 9.016666263050865,
            "command_buffer_latency_us": [
              10.499963536858559,
              10.25001984089613,
              10.708346962928772,
              10.249961633235216,
              10.374991688877344,
              10.416668374091387,
              10.500021744519472,
              10.500021744519472,
              10.166666470468044
            ],
            "command_buffer_latency_us_p50": 10.416668374091387,
            "command_buffer_latency_us_p90": 10.541686788201332
          },
          "throughput": [
            {
              "compute_ns": 533584.0,
              "compute_span_ns": 533584.0,
              "command_buffer_ns": 533625.0178515911,
              "calibration_cpu_ns": 1302209.0,
              "calibration_gpu_ticks": 1302209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 566083.0,
              "compute_span_ns": 566083.0,
              "command_buffer_ns": 566083.3558067679,
              "calibration_cpu_ns": 1263750.0,
              "calibration_gpu_ticks": 1263750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 557166.0,
              "compute_span_ns": 557166.0,
              "command_buffer_ns": 557124.9639615417,
              "calibration_cpu_ns": 1198542.0,
              "calibration_gpu_ticks": 1198542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 566458.0,
              "compute_span_ns": 566458.0,
              "command_buffer_ns": 566458.3295583725,
              "calibration_cpu_ns": 1173459.0,
              "calibration_gpu_ticks": 1173459.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 543583.0,
              "compute_span_ns": 543583.0,
              "command_buffer_ns": 543541.6824184358,
              "calibration_cpu_ns": 1168500.0,
              "calibration_gpu_ticks": 1168500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 545500.0,
              "compute_span_ns": 545500.0,
              "command_buffer_ns": 545500.0209622085,
              "calibration_cpu_ns": 1214375.0,
              "calibration_gpu_ticks": 1214375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 551375.0,
              "compute_span_ns": 551375.0,
              "command_buffer_ns": 551375.0365935266,
              "calibration_cpu_ns": 1191459.0,
              "calibration_gpu_ticks": 1191459.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 555250.0,
              "compute_span_ns": 555250.0,
              "command_buffer_ns": 555249.978788197,
              "calibration_cpu_ns": 1204000.0,
              "calibration_gpu_ticks": 1204000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 564500.0,
              "compute_span_ns": 564500.0,
              "command_buffer_ns": 564499.9910145998,
              "calibration_cpu_ns": 1200375.0,
              "calibration_gpu_ticks": 1200375.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 10250.0,
              "compute_span_ns": 10250.0,
              "command_buffer_ns": 10249.961633235216,
              "calibration_cpu_ns": 377500.0,
              "calibration_gpu_ticks": 377500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 13875.0,
              "compute_span_ns": 13875.0,
              "command_buffer_ns": 13875.01833960414,
              "calibration_cpu_ns": 410125.0,
              "calibration_gpu_ticks": 410125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10125.0,
              "compute_span_ns": 10125.0,
              "command_buffer_ns": 10125.047992914915,
              "calibration_cpu_ns": 390542.0,
              "calibration_gpu_ticks": 390542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10334.0,
              "compute_span_ns": 10334.0,
              "command_buffer_ns": 10375.049896538258,
              "calibration_cpu_ns": 389042.0,
              "calibration_gpu_ticks": 389042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10416.0,
              "compute_span_ns": 10416.0,
              "command_buffer_ns": 10374.991688877344,
              "calibration_cpu_ns": 414625.0,
              "calibration_gpu_ticks": 414625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10167.0,
              "compute_span_ns": 10167.0,
              "command_buffer_ns": 10208.343155682087,
              "calibration_cpu_ns": 369541.0,
              "calibration_gpu_ticks": 369541.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10208.0,
              "compute_span_ns": 10208.0,
              "command_buffer_ns": 10166.666470468044,
              "calibration_cpu_ns": 443583.0,
              "calibration_gpu_ticks": 443583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10541.0,
              "compute_span_ns": 10541.0,
              "command_buffer_ns": 10499.963536858559,
              "calibration_cpu_ns": 406542.0,
              "calibration_gpu_ticks": 406542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 10375.0,
              "compute_span_ns": 10375.0,
              "command_buffer_ns": 10374.991688877344,
              "calibration_cpu_ns": 462375.0,
              "calibration_gpu_ticks": 462375.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            8.33725,
            8.845046875,
            8.70571875,
            8.85090625,
            8.493484375,
            8.5234375,
            8.615234375,
            8.67578125,
            8.8203125
          ],
          "compute_throughput_us_p50": 8.67578125,
          "compute_throughput_us_p90": 8.84621875,
          "command_buffer_throughput_us": [
            8.337890903931111,
            8.845052434480749,
            8.705077561899088,
            8.85091139934957,
            8.49283878778806,
            8.523437827534508,
            8.615234946773853,
            8.675780918565579,
            8.820312359603122
          ],
          "command_buffer_throughput_us_p50": 8.675780918565579,
          "command_buffer_throughput_us_p90": 8.846224227454513,
          "compute_latency_us": [
            10.25,
            13.875,
            10.125,
            10.334,
            10.416,
            10.167,
            10.208,
            10.541,
            10.375
          ],
          "compute_latency_us_p50": 10.334,
          "compute_latency_us_p90": 11.2078,
          "command_buffer_latency_us": [
            10.249961633235216,
            13.87501833960414,
            10.125047992914915,
            10.375049896538258,
            10.374991688877344,
            10.208343155682087,
            10.166666470468044,
            10.499963536858559,
            10.374991688877344
          ],
          "command_buffer_latency_us_p50": 10.374991688877344,
          "command_buffer_latency_us_p90": 11.174974497407675,
          "counter_control_throughput_ratio": 0.9791300460495256,
          "counter_control_latency_ratio": 0.9999944564247265
        },
        "download_ms": 0.310375,
        "allocation_upload_ms": 0.895958,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.0790672678193687e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-l8_xg6hw/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "90a0b6dbe0fbfeffc09b3a8db9fc5856f9e2631dc2ce0e6d9ee991918165a6cf",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 16.773667,
        "capture_ms": 0.076292,
        "compile_ms": 58.336542,
        "allocation_upload_ms": 1.496791,
        "cold_call_ms": 1.928833,
        "warmup_ms": 200.028,
        "download_ms": 0.294375,
        "repetitions": 5847,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          4.95989396272,
          4.96321463999,
          4.7014638276,
          4.3789122627,
          4.53208910552,
          4.85045749957,
          5.01221429793,
          4.77215512228,
          4.37186454592
        ],
        "latency_us": [
          286.084,
          204.166,
          197.791,
          216.208,
          347.5,
          235.667,
          228.167,
          248.375,
          236.834
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 37,
            "reduction_scalar_rounds": 40,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 18444,
              "global_write_bytes": 6148,
              "private_read_bytes": 36888,
              "private_write_bytes": 12296
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 32,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 2,
            "reduction_elements": 3074,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 88,
            "normalized_kernel_cost": 88,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 4611,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 283833,
              "compute_span_ns": 283833,
              "calibration_cpu_ns": 737583,
              "calibration_gpu_ticks": 737583,
              "compute_passes": 1,
              "command_buffer_ns": 283874.978777,
              "command_buffers": 2
            },
            {
              "compute_ns": 283375,
              "compute_span_ns": 283375,
              "calibration_cpu_ns": 724208,
              "calibration_gpu_ticks": 724208,
              "compute_passes": 1,
              "command_buffer_ns": 283416.651655,
              "command_buffers": 2
            },
            {
              "compute_ns": 283208,
              "compute_span_ns": 283208,
              "calibration_cpu_ns": 712959,
              "calibration_gpu_ticks": 712959,
              "compute_passes": 1,
              "command_buffer_ns": 283291.679807,
              "command_buffers": 2
            },
            {
              "compute_ns": 283833,
              "compute_span_ns": 283833,
              "calibration_cpu_ns": 821750,
              "calibration_gpu_ticks": 821750,
              "compute_passes": 1,
              "command_buffer_ns": 283833.360299,
              "command_buffers": 2
            },
            {
              "compute_ns": 269375,
              "compute_span_ns": 269375,
              "calibration_cpu_ns": 713459,
              "calibration_gpu_ticks": 713459,
              "compute_passes": 1,
              "command_buffer_ns": 269374.984782,
              "command_buffers": 2
            },
            {
              "compute_ns": 282792,
              "compute_span_ns": 282792,
              "calibration_cpu_ns": 727584,
              "calibration_gpu_ticks": 727584,
              "compute_passes": 1,
              "command_buffer_ns": 282958.382741,
              "command_buffers": 2
            },
            {
              "compute_ns": 267833,
              "compute_span_ns": 267833,
              "calibration_cpu_ns": 745375,
              "calibration_gpu_ticks": 745375,
              "compute_passes": 1,
              "command_buffer_ns": 267874.97336,
              "command_buffers": 2
            },
            {
              "compute_ns": 285250,
              "compute_span_ns": 285250,
              "calibration_cpu_ns": 707167,
              "calibration_gpu_ticks": 707167,
              "compute_passes": 1,
              "command_buffer_ns": 285291.695036,
              "command_buffers": 2
            },
            {
              "compute_ns": 283667,
              "compute_span_ns": 283667,
              "calibration_cpu_ns": 773083,
              "calibration_gpu_ticks": 773083,
              "compute_passes": 1,
              "command_buffer_ns": 283666.711766,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 7250,
              "compute_span_ns": 7250,
              "calibration_cpu_ns": 419333,
              "calibration_gpu_ticks": 419333,
              "compute_passes": 1,
              "command_buffer_ns": 7333.35036784,
              "command_buffers": 2
            },
            {
              "compute_ns": 7417,
              "compute_span_ns": 7417,
              "calibration_cpu_ns": 459041,
              "calibration_gpu_ticks": 459041,
              "compute_passes": 1,
              "command_buffer_ns": 7541.67558625,
              "command_buffers": 2
            },
            {
              "compute_ns": 7208,
              "compute_span_ns": 7208,
              "calibration_cpu_ns": 422250,
              "calibration_gpu_ticks": 422250,
              "compute_passes": 1,
              "command_buffer_ns": 7208.3203122,
              "command_buffers": 2
            },
            {
              "compute_ns": 7625,
              "compute_span_ns": 7625,
              "calibration_cpu_ns": 459708,
              "calibration_gpu_ticks": 459708,
              "compute_passes": 1,
              "command_buffer_ns": 7625.02895668,
              "command_buffers": 2
            },
            {
              "compute_ns": 7417,
              "compute_span_ns": 7417,
              "calibration_cpu_ns": 449584,
              "calibration_gpu_ticks": 449584,
              "compute_passes": 1,
              "command_buffer_ns": 7458.32221583,
              "command_buffers": 2
            },
            {
              "compute_ns": 6959,
              "compute_span_ns": 6959,
              "calibration_cpu_ns": 428791,
              "calibration_gpu_ticks": 428791,
              "compute_passes": 1,
              "command_buffer_ns": 7041.67177901,
              "command_buffers": 2
            },
            {
              "compute_ns": 7125,
              "compute_span_ns": 7125,
              "calibration_cpu_ns": 445292,
              "calibration_gpu_ticks": 445292,
              "compute_passes": 1,
              "command_buffer_ns": 7208.37851986,
              "command_buffers": 2
            },
            {
              "compute_ns": 7375,
              "compute_span_ns": 7375,
              "calibration_cpu_ns": 410416,
              "calibration_gpu_ticks": 410416,
              "compute_passes": 1,
              "command_buffer_ns": 7375.02705306,
              "command_buffers": 2
            },
            {
              "compute_ns": 7375,
              "compute_span_ns": 7375,
              "calibration_cpu_ns": 439459,
              "calibration_gpu_ticks": 439459,
              "compute_passes": 1,
              "command_buffer_ns": 7499.99890104,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 288124.952931,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 268500.007223,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 267666.648142,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 269666.663371,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 269249.954727,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 283458.32834,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 283291.679807,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 268458.27233,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 267833.354883,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 7541.67558625,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7458.32221583,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7416.58732295,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7458.32221583,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7541.61737859,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7374.9688454,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7416.64553061,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7833.29596743,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7375.02705306,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              4.501952389546875,
              4.195312612859375,
              4.182291377218751,
              4.2135416151718745,
              4.207030542609375,
              4.4290363803125,
              4.426432496984375,
              4.19466050515625,
              4.184896170046875
            ],
            "command_buffer_throughput_us_p50": 4.207030542609375,
            "command_buffer_throughput_us_p90": 4.443619582159375,
            "command_buffer_latency_us": [
              7.54167558625,
              7.45832221583,
              7.41658732295,
              7.45832221583,
              7.54161737859,
              7.3749688454,
              7.416645530609999,
              7.83329596743,
              7.37502705306
            ],
            "command_buffer_latency_us_p50": 7.45832221583,
            "command_buffer_latency_us_p90": 7.599999662486
          },
          "compute_throughput_us": [
            4.434890625,
            4.427734375,
            4.425125,
            4.434890625,
            4.208984375,
            4.418625,
            4.184890625,
            4.45703125,
            4.432296875
          ],
          "compute_throughput_us_p50": 4.427734375,
          "compute_throughput_us_p90": 4.43931875,
          "command_buffer_throughput_us": [
            4.4355465433906245,
            4.428385182109375,
            4.426432496984375,
            4.434896254671875,
            4.20898413721875,
            4.421224730328125,
            4.18554645875,
            4.4576827349374994,
            4.432292371343751
          ],
          "command_buffer_throughput_us_p50": 4.428385182109375,
          "command_buffer_throughput_us_p90": 4.439973781699999,
          "compute_latency_us": [
            7.25,
            7.417,
            7.208,
            7.625,
            7.417,
            6.959,
            7.125,
            7.375,
            7.375
          ],
          "compute_latency_us_p50": 7.375,
          "compute_latency_us_p90": 7.4586,
          "command_buffer_latency_us": [
            7.3333503678400005,
            7.54167558625,
            7.208320312200001,
            7.6250289566800005,
            7.45832221583,
            7.0416717790100005,
            7.20837851986,
            7.37502705306,
            7.4999989010400006
          ],
          "command_buffer_latency_us_p50": 7.37502705306,
          "command_buffer_latency_us_p90": 7.558346260336,
          "counter_control_throughput_ratio": 1.052534105443022,
          "counter_control_latency_ratio": 0.9723767992898263
        },
        "correctness": {
          "max_abs_error": 2.43199801630567e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 4.77215512228,
        "throughput_us_p90": 4.973014571578,
        "latency_us_p50": 235.667,
        "latency_us_p90": 298.3672
      },
      "slowdown": 0.37293457116916673,
      "device_compute_slowdown": 0.510355695632598,
      "valid": true,
      "round": 3,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "layernorm_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-_yj0cu2t/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "83a70ba0e4d1dab6671da56395e0b13750f8d9c9277e4782c4e371450de131c7",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 18.204833,
        "capture_ms": 0.078584,
        "compile_ms": 59.039292,
        "allocation_upload_ms": 1.977791,
        "cold_call_ms": 5.1485,
        "warmup_ms": 200.084,
        "download_ms": 0.957083,
        "repetitions": 1305,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          17.1000957854,
          17.3565770115,
          17.4242015326,
          17.3332375479,
          16.5785762452,
          17.7555877395,
          16.8129310345,
          17.300702682,
          17.3558429119
        ],
        "latency_us": [
          269.292,
          332.25,
          314,
          225.375,
          238.208,
          290.417,
          272.916,
          275.209,
          264
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 256,
            "reduction_scalar_rounds": 60,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 36864,
              "global_write_bytes": 12288,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 144,
              "global_write_bytes": 48,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 2,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 108,
            "normalized_kernel_cost": 432,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 9216,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 1090125,
              "compute_span_ns": 1090125,
              "calibration_cpu_ns": 1508542,
              "calibration_gpu_ticks": 1508542,
              "compute_passes": 1,
              "command_buffer_ns": 1090125.00616,
              "command_buffers": 2
            },
            {
              "compute_ns": 1068541,
              "compute_span_ns": 1068541,
              "calibration_cpu_ns": 1581375,
              "calibration_gpu_ticks": 1581375,
              "compute_passes": 1,
              "command_buffer_ns": 1068500.04522,
              "command_buffers": 2
            },
            {
              "compute_ns": 1109791,
              "compute_span_ns": 1109791,
              "calibration_cpu_ns": 1507250,
              "calibration_gpu_ticks": 1507250,
              "compute_passes": 1,
              "command_buffer_ns": 1109833.36344,
              "command_buffers": 2
            },
            {
              "compute_ns": 1119000,
              "compute_span_ns": 1119000,
              "calibration_cpu_ns": 1552750,
              "calibration_gpu_ticks": 1552750,
              "compute_passes": 1,
              "command_buffer_ns": 1119083.31746,
              "command_buffers": 2
            },
            {
              "compute_ns": 1100458,
              "compute_span_ns": 1100458,
              "calibration_cpu_ns": 1585792,
              "calibration_gpu_ticks": 1585792,
              "compute_passes": 1,
              "command_buffer_ns": 1100499.93964,
              "command_buffers": 2
            },
            {
              "compute_ns": 1094709,
              "compute_span_ns": 1094709,
              "calibration_cpu_ns": 1548667,
              "calibration_gpu_ticks": 1548667,
              "compute_passes": 1,
              "command_buffer_ns": 1094833.36564,
              "command_buffers": 2
            },
            {
              "compute_ns": 1116500,
              "compute_span_ns": 1116500,
              "calibration_cpu_ns": 1533458,
              "calibration_gpu_ticks": 1533458,
              "compute_passes": 1,
              "command_buffer_ns": 1116541.67995,
              "command_buffers": 2
            },
            {
              "compute_ns": 1101709,
              "compute_span_ns": 1101709,
              "calibration_cpu_ns": 1592334,
              "calibration_gpu_ticks": 1592334,
              "compute_passes": 1,
              "command_buffer_ns": 1101750.00736,
              "command_buffers": 2
            },
            {
              "compute_ns": 1085875,
              "compute_span_ns": 1085875,
              "calibration_cpu_ns": 1490708,
              "calibration_gpu_ticks": 1490708,
              "compute_passes": 1,
              "command_buffer_ns": 1085916.65048,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 20500,
              "compute_span_ns": 20500,
              "calibration_cpu_ns": 482209,
              "calibration_gpu_ticks": 482209,
              "compute_passes": 1,
              "command_buffer_ns": 20499.9814741,
              "command_buffers": 2
            },
            {
              "compute_ns": 23167,
              "compute_span_ns": 23167,
              "calibration_cpu_ns": 485709,
              "calibration_gpu_ticks": 485709,
              "compute_passes": 1,
              "command_buffer_ns": 23166.6490436,
              "command_buffers": 2
            },
            {
              "compute_ns": 20458,
              "compute_span_ns": 20458,
              "calibration_cpu_ns": 422042,
              "calibration_gpu_ticks": 422042,
              "compute_passes": 1,
              "command_buffer_ns": 20541.5999517,
              "command_buffers": 2
            },
            {
              "compute_ns": 20750,
              "compute_span_ns": 20750,
              "calibration_cpu_ns": 462833,
              "calibration_gpu_ticks": 462833,
              "compute_passes": 1,
              "command_buffer_ns": 20791.660063,
              "command_buffers": 2
            },
            {
              "compute_ns": 20667,
              "compute_span_ns": 20667,
              "calibration_cpu_ns": 475209,
              "calibration_gpu_ticks": 475209,
              "compute_passes": 1,
              "command_buffer_ns": 20708.3066925,
              "command_buffers": 2
            },
            {
              "compute_ns": 20333,
              "compute_span_ns": 20333,
              "calibration_cpu_ns": 433250,
              "calibration_gpu_ticks": 433250,
              "compute_passes": 1,
              "command_buffer_ns": 20375.0096262,
              "command_buffers": 2
            },
            {
              "compute_ns": 20291,
              "compute_span_ns": 20291,
              "calibration_cpu_ns": 509458,
              "calibration_gpu_ticks": 509458,
              "compute_passes": 1,
              "command_buffer_ns": 20374.9514185,
              "command_buffers": 2
            },
            {
              "compute_ns": 20375,
              "compute_span_ns": 20375,
              "calibration_cpu_ns": 454750,
              "calibration_gpu_ticks": 454750,
              "compute_passes": 1,
              "command_buffer_ns": 20416.6863114,
              "command_buffers": 2
            },
            {
              "compute_ns": 20542,
              "compute_span_ns": 20542,
              "calibration_cpu_ns": 446625,
              "calibration_gpu_ticks": 446625,
              "compute_passes": 1,
              "command_buffer_ns": 20666.7464226,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1120125.00176,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1119916.67654,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1108291.61713,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1113250.03672,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1118708.34371,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1075916.63254,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1128583.27338,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1096083.31695,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1035666.67531,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 20458.3629966,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 23333.3557844,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20749.9833778,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20750.0415854,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20666.688215,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20541.6581593,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20458.3047889,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20083.3310373,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19999.9776669,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              17.501953152499997,
              17.498698070937497,
              17.317056517656248,
              17.39453182375,
              17.47981787046875,
              16.8111973834375,
              17.634113646562497,
              17.12630182734375,
              16.18229180171875
            ],
            "command_buffer_throughput_us_p50": 17.39453182375,
            "command_buffer_throughput_us_p90": 17.528385251312496,
            "command_buffer_latency_us": [
              20.458362996600002,
              23.3333557844,
              20.7499833778,
              20.750041585399998,
              20.666688214999997,
              20.541658159300002,
              20.4583047889,
              20.0833310373,
              19.999977666899998
            ],
            "command_buffer_latency_us_p50": 20.541658159300002,
            "command_buffer_latency_us_p90": 21.2667044252
          },
          "compute_throughput_us": [
            17.033203125,
            16.695953125,
            17.340484375,
            17.484375,
            17.19465625,
            17.104828125,
            17.4453125,
            17.214203125,
            16.966796875
          ],
          "compute_throughput_us_p50": 17.19465625,
          "compute_throughput_us_p90": 17.453125,
          "command_buffer_throughput_us": [
            17.03320322125,
            16.6953132065625,
            17.34114630375,
            17.4856768353125,
            17.195311556875,
            17.106771338125,
            17.445963749218752,
            17.214843865,
            16.96744766375
          ],
          "command_buffer_throughput_us_p50": 17.195311556875,
          "command_buffer_throughput_us_p90": 17.4539063664375,
          "compute_latency_us": [
            20.5,
            23.167,
            20.458,
            20.75,
            20.667,
            20.333,
            20.291,
            20.375,
            20.542
          ],
          "compute_latency_us_p50": 20.5,
          "compute_latency_us_p90": 21.2334,
          "command_buffer_latency_us": [
            20.4999814741,
            23.1666490436,
            20.541599951699997,
            20.791660063,
            20.7083066925,
            20.3750096262,
            20.3749514185,
            20.4166863114,
            20.666746422600003
          ],
          "command_buffer_latency_us_p50": 20.541599951699997,
          "command_buffer_latency_us_p90": 21.26665785912,
          "counter_control_throughput_ratio": 1.0013911016614856,
          "counter_control_latency_ratio": 1.0020057057441891
        },
        "correctness": {
          "max_abs_error": 2.9463829065434766e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 17.3332375479,
        "throughput_us_p90": 17.49047877398,
        "latency_us_p50": 272.916,
        "latency_us_p90": 317.65
      },
      "torch": {
        "cold_call_ms": 7.595208,
        "warmup_ms": 200.178959,
        "repetitions": 632,
        "throughput_us": [
          46.17820411392405,
          46.90783227848101,
          45.585443037974684,
          46.40816139240507,
          45.44244462025316,
          45.83056487341772,
          46.38699841772152,
          45.06006012658228,
          44.561773734177216
        ],
        "latency_us": [
          418.292,
          276.83299999999997,
          289.875,
          251.292,
          265.75,
          262.33299999999997,
          268.75,
          282.5,
          301.66700000000003
        ],
        "throughput_us_p50": 45.83056487341772,
        "throughput_us_p90": 46.50809556962026,
        "latency_us_p50": 276.83299999999997,
        "latency_us_p90": 324.992,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 2573833.3351910114,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2568958.3271741867,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2584958.332590759,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2578124.9860301614,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2580208.2964219153,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2599208.3246819675,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2581125.0088736415,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2581499.982625246,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2578750.01989305,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 44541.66628420353,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44499.98959898949,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44291.66438058019,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44375.01775100827,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43625.012040138245,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43750.04209578037,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43875.01394376159,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44041.662476956844,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43875.01394376159,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              40.216145862359554,
              40.13997386209667,
              40.389973946730606,
              40.28320290672127,
              40.31575463159243,
              40.61263007315574,
              40.33007826365065,
              40.33593722851947,
              40.29296906082891
            ],
            "command_buffer_throughput_us_p50": 40.31575463159243,
            "command_buffer_throughput_us_p90": 40.43450517201563,
            "command_buffer_latency_us": [
              44.54166628420353,
              44.49998959898949,
              44.29166438058019,
              44.37501775100827,
              43.625012040138245,
              43.75004209578037,
              43.87501394376159,
              44.041662476956844,
              43.87501394376159
            ],
            "command_buffer_latency_us_p50": 44.041662476956844,
            "command_buffer_latency_us_p90": 44.508324936032295
          },
          "throughput": [
            {
              "compute_ns": 2638709.0,
              "compute_span_ns": 2638709.0,
              "command_buffer_ns": 2638750.011101365,
              "calibration_cpu_ns": 3609208.0,
              "calibration_gpu_ticks": 3609208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2594125.0,
              "compute_span_ns": 2594125.0,
              "command_buffer_ns": 2594125.0496543944,
              "calibration_cpu_ns": 3413750.0,
              "calibration_gpu_ticks": 3413750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2580958.0,
              "compute_span_ns": 2580958.0,
              "command_buffer_ns": 2580958.3603404462,
              "calibration_cpu_ns": 3375500.0,
              "calibration_gpu_ticks": 3375500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2583041.0,
              "compute_span_ns": 2583041.0,
              "command_buffer_ns": 2582999.994046986,
              "calibration_cpu_ns": 3370083.0,
              "calibration_gpu_ticks": 3370083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2607834.0,
              "compute_span_ns": 2607834.0,
              "command_buffer_ns": 2607875.0379383564,
              "calibration_cpu_ns": 3399916.0,
              "calibration_gpu_ticks": 3399916.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2567125.0,
              "compute_span_ns": 2567125.0,
              "command_buffer_ns": 2567125.018686056,
              "calibration_cpu_ns": 3410167.0,
              "calibration_gpu_ticks": 3410167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2563542.0,
              "compute_span_ns": 2563542.0,
              "command_buffer_ns": 2563541.696872562,
              "calibration_cpu_ns": 3350542.0,
              "calibration_gpu_ticks": 3350542.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2578000.0,
              "compute_span_ns": 2578000.0,
              "command_buffer_ns": 2578000.01418218,
              "calibration_cpu_ns": 3365083.0,
              "calibration_gpu_ticks": 3365083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2573833.0,
              "compute_span_ns": 2573833.0,
              "command_buffer_ns": 2573833.3351910114,
              "calibration_cpu_ns": 3396583.0,
              "calibration_gpu_ticks": 3396583.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 44709.0,
              "compute_span_ns": 44709.0,
              "command_buffer_ns": 44749.99150261283,
              "calibration_cpu_ns": 501292.0,
              "calibration_gpu_ticks": 501292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44333.0,
              "compute_span_ns": 44333.0,
              "command_buffer_ns": 44333.34106579423,
              "calibration_cpu_ns": 489334.0,
              "calibration_gpu_ticks": 489334.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44584.0,
              "compute_span_ns": 44584.0,
              "command_buffer_ns": 44625.019654631615,
              "calibration_cpu_ns": 438708.0,
              "calibration_gpu_ticks": 438708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44500.0,
              "compute_span_ns": 44500.0,
              "command_buffer_ns": 44499.98959898949,
              "calibration_cpu_ns": 446500.0,
              "calibration_gpu_ticks": 446500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44042.0,
              "compute_span_ns": 44042.0,
              "command_buffer_ns": 44041.662476956844,
              "calibration_cpu_ns": 462709.0,
              "calibration_gpu_ticks": 462709.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44125.0,
              "compute_span_ns": 44125.0,
              "command_buffer_ns": 44124.957639724016,
              "calibration_cpu_ns": 491834.0,
              "calibration_gpu_ticks": 491834.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44459.0,
              "compute_span_ns": 44459.0,
              "command_buffer_ns": 44499.98959898949,
              "calibration_cpu_ns": 432917.0,
              "calibration_gpu_ticks": 432917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 43709.0,
              "compute_span_ns": 43709.0,
              "command_buffer_ns": 43750.04209578037,
              "calibration_cpu_ns": 429625.0,
              "calibration_gpu_ticks": 429625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44250.0,
              "compute_span_ns": 44250.0,
              "command_buffer_ns": 44250.04590302706,
              "calibration_cpu_ns": 469000.0,
              "calibration_gpu_ticks": 469000.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            41.229828125,
            40.533203125,
            40.32746875,
            40.360015625,
            40.74740625,
            40.111328125,
            40.05534375,
            40.28125,
            40.216140625
          ],
          "compute_throughput_us_p50": 40.32746875,
          "compute_throughput_us_p90": 40.843890625,
          "command_buffer_throughput_us": [
            41.23046892345883,
            40.53320390084991,
            40.32747438031947,
            40.35937490698416,
            40.74804746778682,
            40.11132841696963,
            40.055339013633784,
            40.281250221596565,
            40.216145862359554
          ],
          "command_buffer_throughput_us_p50": 40.32747438031947,
          "command_buffer_throughput_us_p90": 40.84453175892122,
          "compute_latency_us": [
            44.709,
            44.333,
            44.584,
            44.5,
            44.042,
            44.125,
            44.459,
            43.709,
            44.25
          ],
          "compute_latency_us_p50": 44.333,
          "compute_latency_us_p90": 44.609,
          "command_buffer_latency_us": [
            44.74999150261283,
            44.33334106579423,
            44.625019654631615,
            44.49998959898949,
            44.041662476956844,
            44.124957639724016,
            44.49998959898949,
            43.75004209578037,
            44.25004590302706
          ],
          "command_buffer_latency_us_p50": 44.33334106579423,
          "command_buffer_latency_us_p90": 44.65001402422786,
          "counter_control_throughput_ratio": 0.9986442113241827,
          "counter_control_latency_ratio": 1.0075263659361962
        },
        "download_ms": 0.364958,
        "allocation_upload_ms": 1.76825,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 2.982737865586671e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.3782025727977332,
      "device_compute_slowdown": 0.42637578759514877,
      "valid": true,
      "round": 3,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "layernorm_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.301667,
        "warmup_ms": 200.489167,
        "repetitions": 574,
        "throughput_us": [
          45.01604181184669,
          45.97546515679443,
          45.90933449477352,
          45.72568118466899,
          45.46907665505226,
          45.3185243902439,
          45.827670731707315,
          46.02409930313589,
          46.42936933797909
        ],
        "latency_us": [
          398.83299999999997,
          314.542,
          367.66700000000003,
          226.792,
          274.667,
          276.0,
          271.0,
          264.25,
          251.16699999999997
        ],
        "throughput_us_p50": 45.827670731707315,
        "throughput_us_p90": 46.105153310104534,
        "latency_us_p50": 274.667,
        "latency_us_p90": 373.90020000000004,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 2697416.639421135,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2683500.002603978,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2635041.659232229,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2670874.9937824905,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2712375.018745661,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2632333.373185247,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2688125.0087171793,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2642125.0076964498,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 2682291.6697710752,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 44499.98959898949,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49166.67239740491,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 43958.36731418967,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44749.99150261283,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44875.02155825496,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 45416.64384305477,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44958.37492868304,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 44875.02155825496,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 45124.965254217386,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              42.14713499095524,
              41.929687540687155,
              41.17252592550358,
              41.732421777851414,
              42.38085966790095,
              41.13020895601949,
              42.00195326120593,
              41.28320324525703,
              41.91080734017305
            ],
            "command_buffer_throughput_us_p50": 41.91080734017305,
            "command_buffer_throughput_us_p90": 42.19387992634438,
            "command_buffer_latency_us": [
              44.49998959898949,
              49.16667239740491,
              43.95836731418967,
              44.74999150261283,
              44.87502155825496,
              45.41664384305477,
              44.95837492868304,
              44.87502155825496,
              45.124965254217386
            ],
            "command_buffer_latency_us_p50": 44.87502155825496,
            "command_buffer_latency_us_p90": 46.1666495539248
          },
          "throughput": [
            {
              "compute_ns": 2687375.0,
              "compute_span_ns": 2687375.0,
              "command_buffer_ns": 2687375.0030063093,
              "calibration_cpu_ns": 3393291.0,
              "calibration_gpu_ticks": 3393291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2689500.0,
              "compute_span_ns": 2689500.0,
              "command_buffer_ns": 2689499.990083277,
              "calibration_cpu_ns": 3411125.0,
              "calibration_gpu_ticks": 3411125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2624708.0,
              "compute_span_ns": 2624708.0,
              "command_buffer_ns": 2624708.3442285657,
              "calibration_cpu_ns": 3384875.0,
              "calibration_gpu_ticks": 3384875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2680417.0,
              "compute_span_ns": 2680417.0,
              "command_buffer_ns": 2680458.3612829447,
              "calibration_cpu_ns": 3369500.0,
              "calibration_gpu_ticks": 3369500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2634583.0,
              "compute_span_ns": 2634583.0,
              "command_buffer_ns": 2634541.6554249823,
              "calibration_cpu_ns": 3302875.0,
              "calibration_gpu_ticks": 3302875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2650250.0,
              "compute_span_ns": 2650250.0,
              "command_buffer_ns": 2650249.982252717,
              "calibration_cpu_ns": 3312916.0,
              "calibration_gpu_ticks": 3312916.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2662583.0,
              "compute_span_ns": 2662583.0,
              "command_buffer_ns": 2662583.312485367,
              "calibration_cpu_ns": 3411834.0,
              "calibration_gpu_ticks": 3411834.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2650000.0,
              "compute_span_ns": 2650000.0,
              "command_buffer_ns": 2649999.9803490937,
              "calibration_cpu_ns": 3324917.0,
              "calibration_gpu_ticks": 3324917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 2686708.0,
              "compute_span_ns": 2686708.0,
              "command_buffer_ns": 2686708.3506658673,
              "calibration_cpu_ns": 3353667.0,
              "calibration_gpu_ticks": 3353667.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 45291.0,
              "compute_span_ns": 45291.0,
              "command_buffer_ns": 45249.995309859514,
              "calibration_cpu_ns": 510667.0,
              "calibration_gpu_ticks": 510667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 45042.0,
              "compute_span_ns": 45042.0,
              "command_buffer_ns": 45083.34677666426,
              "calibration_cpu_ns": 478708.0,
              "calibration_gpu_ticks": 478708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 45333.0,
              "compute_span_ns": 45333.0,
              "command_buffer_ns": 45291.67199507356,
              "calibration_cpu_ns": 529875.0,
              "calibration_gpu_ticks": 529875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44875.0,
              "compute_span_ns": 44875.0,
              "command_buffer_ns": 44875.02155825496,
              "calibration_cpu_ns": 502708.0,
              "calibration_gpu_ticks": 502708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44875.0,
              "compute_span_ns": 44875.0,
              "command_buffer_ns": 44875.02155825496,
              "calibration_cpu_ns": 532792.0,
              "calibration_gpu_ticks": 532792.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 45292.0,
              "compute_span_ns": 45292.0,
              "command_buffer_ns": 45333.3486802876,
              "calibration_cpu_ns": 486291.0,
              "calibration_gpu_ticks": 486291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 45209.0,
              "compute_span_ns": 45209.0,
              "command_buffer_ns": 45249.995309859514,
              "calibration_cpu_ns": 487625.0,
              "calibration_gpu_ticks": 487625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44708.0,
              "compute_span_ns": 44708.0,
              "command_buffer_ns": 44666.638132184744,
              "calibration_cpu_ns": 511584.0,
              "calibration_gpu_ticks": 511584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 44500.0,
              "compute_span_ns": 44500.0,
              "command_buffer_ns": 44499.98959898949,
              "calibration_cpu_ns": 453667.0,
              "calibration_gpu_ticks": 453667.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            41.990234375,
            42.0234375,
            41.0110625,
            41.881515625,
            41.165359375,
            41.41015625,
            41.602859375,
            41.40625,
            41.9798125
          ],
          "compute_throughput_us_p50": 41.602859375,
          "compute_throughput_us_p90": 41.996875,
          "command_buffer_throughput_us": [
            41.99023442197358,
            42.02343734505121,
            41.01106787857134,
            41.88216189504601,
            41.16471336601535,
            41.4101559726987,
            41.60286425758386,
            41.40624969295459,
            41.97981797915418
          ],
          "command_buffer_throughput_us_p50": 41.60286425758386,
          "command_buffer_throughput_us_p90": 41.99687500658911,
          "compute_latency_us": [
            45.291,
            45.042,
            45.333,
            44.875,
            44.875,
            45.292,
            45.209,
            44.708,
            44.5
          ],
          "compute_latency_us_p50": 45.042,
          "compute_latency_us_p90": 45.300200000000004,
          "command_buffer_latency_us": [
            45.249995309859514,
            45.08334677666426,
            45.29167199507356,
            44.87502155825496,
            44.87502155825496,
            45.3333486802876,
            45.249995309859514,
            44.666638132184744,
            44.49998959898949
          ],
          "command_buffer_latency_us_p50": 45.08334677666426,
          "command_buffer_latency_us_p90": 45.300007332116365,
          "counter_control_throughput_ratio": 1.001646607244308,
          "counter_control_latency_ratio": 1.0
        },
        "download_ms": 0.366041,
        "allocation_upload_ms": 0.8755,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 2.982737865586671e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-4hr0_ipv/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "b8d9d984f3aaf8bfdb9269f6e0efc630605362c6ff4e852cc112881dfaf5e97b",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.752416,
        "capture_ms": 0.068708,
        "compile_ms": 59.838084,
        "allocation_upload_ms": 1.891833,
        "cold_call_ms": 9.35025,
        "warmup_ms": 200.21425,
        "download_ms": 0.627792,
        "repetitions": 1558,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          17.3721116816,
          17.4114781772,
          16.746817715,
          16.3635269576,
          16.3456623877,
          17.2869062901,
          17.5697207959,
          17.5682233633,
          16.3413831836
        ],
        "latency_us": [
          309.75,
          244.125,
          249.75,
          259.541,
          216.125,
          218.375,
          226.167,
          233.5,
          232.958
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 128,
            "reduction_scalar_rounds": 60,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 36864,
              "global_write_bytes": 12288,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 144,
              "global_write_bytes": 48,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 2,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 100,
            "normalized_kernel_cost": 400,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 9216,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 1013667,
              "compute_span_ns": 1013667,
              "calibration_cpu_ns": 1517291,
              "calibration_gpu_ticks": 1517291,
              "compute_passes": 1,
              "command_buffer_ns": 1013791.59605,
              "command_buffers": 2
            },
            {
              "compute_ns": 1065375,
              "compute_span_ns": 1065375,
              "calibration_cpu_ns": 1569958,
              "calibration_gpu_ticks": 1569958,
              "compute_passes": 1,
              "command_buffer_ns": 1065416.66901,
              "command_buffers": 2
            },
            {
              "compute_ns": 1069042,
              "compute_span_ns": 1069042,
              "calibration_cpu_ns": 1529542,
              "calibration_gpu_ticks": 1529542,
              "compute_passes": 1,
              "command_buffer_ns": 1069208.31604,
              "command_buffers": 2
            },
            {
              "compute_ns": 1015709,
              "compute_span_ns": 1015709,
              "calibration_cpu_ns": 1443292,
              "calibration_gpu_ticks": 1443292,
              "compute_passes": 1,
              "command_buffer_ns": 1015833.28797,
              "command_buffers": 2
            },
            {
              "compute_ns": 1018167,
              "compute_span_ns": 1018167,
              "calibration_cpu_ns": 1466250,
              "calibration_gpu_ticks": 1466250,
              "compute_passes": 1,
              "command_buffer_ns": 1018250.01184,
              "command_buffers": 2
            },
            {
              "compute_ns": 1016958,
              "compute_span_ns": 1016958,
              "calibration_cpu_ns": 1449666,
              "calibration_gpu_ticks": 1449666,
              "compute_passes": 1,
              "command_buffer_ns": 1017041.67901,
              "command_buffers": 2
            },
            {
              "compute_ns": 1014458,
              "compute_span_ns": 1014458,
              "calibration_cpu_ns": 1564625,
              "calibration_gpu_ticks": 1564625,
              "compute_passes": 1,
              "command_buffer_ns": 1014541.65997,
              "command_buffers": 2
            },
            {
              "compute_ns": 1111125,
              "compute_span_ns": 1111125,
              "calibration_cpu_ns": 1595708,
              "calibration_gpu_ticks": 1595708,
              "compute_passes": 1,
              "command_buffer_ns": 1111208.34481,
              "command_buffers": 2
            },
            {
              "compute_ns": 1088084,
              "compute_span_ns": 1088084,
              "calibration_cpu_ns": 1632000,
              "calibration_gpu_ticks": 1632000,
              "compute_passes": 1,
              "command_buffer_ns": 1088249.96278,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 17625,
              "compute_span_ns": 17625,
              "calibration_cpu_ns": 390791,
              "calibration_gpu_ticks": 390791,
              "compute_passes": 1,
              "command_buffer_ns": 17624.9886863,
              "command_buffers": 2
            },
            {
              "compute_ns": 20208,
              "compute_span_ns": 20208,
              "calibration_cpu_ns": 396875,
              "calibration_gpu_ticks": 396875,
              "compute_passes": 1,
              "command_buffer_ns": 20291.5980481,
              "command_buffers": 2
            },
            {
              "compute_ns": 20708,
              "compute_span_ns": 20708,
              "calibration_cpu_ns": 475667,
              "calibration_gpu_ticks": 475667,
              "compute_passes": 1,
              "command_buffer_ns": 20708.3066925,
              "command_buffers": 2
            },
            {
              "compute_ns": 19916,
              "compute_span_ns": 19916,
              "calibration_cpu_ns": 447875,
              "calibration_gpu_ticks": 447875,
              "compute_passes": 1,
              "command_buffer_ns": 19999.9776669,
              "command_buffers": 2
            },
            {
              "compute_ns": 19875,
              "compute_span_ns": 19875,
              "calibration_cpu_ns": 438584,
              "calibration_gpu_ticks": 438584,
              "compute_passes": 1,
              "command_buffer_ns": 19958.3009817,
              "command_buffers": 2
            },
            {
              "compute_ns": 20083,
              "compute_span_ns": 20083,
              "calibration_cpu_ns": 433542,
              "calibration_gpu_ticks": 433542,
              "compute_passes": 1,
              "command_buffer_ns": 20083.3310373,
              "command_buffers": 2
            },
            {
              "compute_ns": 20333,
              "compute_span_ns": 20333,
              "calibration_cpu_ns": 438083,
              "calibration_gpu_ticks": 438083,
              "compute_passes": 1,
              "command_buffer_ns": 20333.3329409,
              "command_buffers": 2
            },
            {
              "compute_ns": 20459,
              "compute_span_ns": 20459,
              "calibration_cpu_ns": 458167,
              "calibration_gpu_ticks": 458167,
              "compute_passes": 1,
              "command_buffer_ns": 20500.0396818,
              "command_buffers": 2
            },
            {
              "compute_ns": 20166,
              "compute_span_ns": 20166,
              "calibration_cpu_ns": 416083,
              "calibration_gpu_ticks": 416083,
              "compute_passes": 1,
              "command_buffer_ns": 20208.361093,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1073916.67552,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1017541.68281,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1014708.3085,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1020875.00272,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1071000.00605,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1060791.66289,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1018374.98369,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1027458.34738,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 1015333.34237,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 17791.6372195,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19833.270926,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20583.3348446,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20208.3028853,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20083.3310373,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 19916.6825041,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20208.3028853,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20166.6844077,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 20375.0096262,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              16.779948055000002,
              15.89908879390625,
              15.8548173203125,
              15.9511719175,
              16.73437509453125,
              16.57486973265625,
              15.91210912015625,
              16.0540366778125,
              15.86458347453125
            ],
            "command_buffer_throughput_us_p50": 15.9511719175,
            "command_buffer_throughput_us_p90": 16.743489686625,
            "command_buffer_latency_us": [
              17.7916372195,
              19.833270926,
              20.583334844600003,
              20.2083028853,
              20.0833310373,
              19.916682504100002,
              20.2083028853,
              20.1666844077,
              20.3750096262
            ],
            "command_buffer_latency_us_p50": 20.1666844077,
            "command_buffer_latency_us_p90": 20.416674669880003
          },
          "compute_throughput_us": [
            15.838546875,
            16.646484375,
            16.70378125,
            15.870453125,
            15.908859375,
            15.88996875,
            15.85090625,
            17.361328125,
            17.0013125
          ],
          "compute_throughput_us_p50": 15.908859375,
          "compute_throughput_us_p90": 17.073315625,
          "command_buffer_throughput_us": [
            15.840493688281251,
            16.64713545328125,
            16.706379938125,
            15.872395124531248,
            15.910156435000001,
            15.891276234531249,
            15.852213437031251,
            17.36263038765625,
            17.0039056684375
          ],
          "command_buffer_throughput_us_p50": 15.910156435000001,
          "command_buffer_throughput_us_p90": 17.07565061228125,
          "compute_latency_us": [
            17.625,
            20.208,
            20.708,
            19.916,
            19.875,
            20.083,
            20.333,
            20.459,
            20.166
          ],
          "compute_latency_us_p50": 20.166,
          "compute_latency_us_p90": 20.5088,
          "command_buffer_latency_us": [
            17.6249886863,
            20.2915980481,
            20.7083066925,
            19.999977666899998,
            19.9583009817,
            20.0833310373,
            20.3333329409,
            20.5000396818,
            20.208361093
          ],
          "command_buffer_latency_us_p50": 20.208361093,
          "command_buffer_latency_us_p90": 20.54169308394,
          "counter_control_throughput_ratio": 0.9962358426106361,
          "counter_control_latency_ratio": 1.0060715063347854
        },
        "correctness": {
          "max_abs_error": 2.9463829065434766e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 17.2869062901,
        "throughput_us_p90": 17.56852284982,
        "latency_us_p50": 233.5,
        "latency_us_p90": 269.5828
      },
      "slowdown": 0.37721546860420097,
      "device_compute_slowdown": 0.3823982200742662,
      "valid": true,
      "round": 3,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "layernorm_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-drj5pej8/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "2f415c8199d32d1b999ce6dae6674c9bc8bbdf9b0adda641a6d2aee649468028",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.964166,
        "capture_ms": 0.075875,
        "compile_ms": 60.223125,
        "allocation_upload_ms": 6.116125,
        "cold_call_ms": 7.143917,
        "warmup_ms": 200.442334,
        "download_ms": 3.759459,
        "repetitions": 353,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          72.7245042493,
          72.6961756374,
          72.6851983003,
          72.7221444759,
          72.0030708215,
          71.931776204,
          71.9809943343,
          72.3622521246,
          72.085101983
        ],
        "latency_us": [
          371.708,
          326.25,
          271.541,
          273.375,
          282.833,
          295.25,
          379.709,
          283.583,
          296.625
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 384,
            "reduction_scalar_rounds": 120,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 73728,
              "global_write_bytes": 24576,
              "private_read_bytes": 147456,
              "private_write_bytes": 49152
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 288,
              "global_write_bytes": 96,
              "private_read_bytes": 576,
              "private_write_bytes": 192
            },
            "striped_storage_scalars_per_worker": 48,
            "reduction_operations": 2,
            "reduction_elements": 12288,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 160,
            "normalized_kernel_cost": 1920,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 18432,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4563625,
              "compute_span_ns": 4563625,
              "calibration_cpu_ns": 5242000,
              "calibration_gpu_ticks": 5242000,
              "compute_passes": 1,
              "command_buffer_ns": 4563708.38219,
              "command_buffers": 2
            },
            {
              "compute_ns": 4619667,
              "compute_span_ns": 4619667,
              "calibration_cpu_ns": 5159833,
              "calibration_gpu_ticks": 5159833,
              "compute_passes": 1,
              "command_buffer_ns": 4619791.63811,
              "command_buffers": 2
            },
            {
              "compute_ns": 4586750,
              "compute_span_ns": 4586750,
              "calibration_cpu_ns": 5037750,
              "calibration_gpu_ticks": 5037750,
              "compute_passes": 1,
              "command_buffer_ns": 4586875.03124,
              "command_buffers": 2
            },
            {
              "compute_ns": 4607959,
              "compute_span_ns": 4607959,
              "calibration_cpu_ns": 5075083,
              "calibration_gpu_ticks": 5075083,
              "compute_passes": 1,
              "command_buffer_ns": 4608124.96021,
              "command_buffers": 2
            },
            {
              "compute_ns": 4544375,
              "compute_span_ns": 4544375,
              "calibration_cpu_ns": 5008167,
              "calibration_gpu_ticks": 5008167,
              "compute_passes": 1,
              "command_buffer_ns": 4544458.35203,
              "command_buffers": 2
            },
            {
              "compute_ns": 4664875,
              "compute_span_ns": 4664875,
              "calibration_cpu_ns": 5125584,
              "calibration_gpu_ticks": 5125584,
              "compute_passes": 1,
              "command_buffer_ns": 4664916.66157,
              "command_buffers": 2
            },
            {
              "compute_ns": 4659583,
              "compute_span_ns": 4659583,
              "calibration_cpu_ns": 5110709,
              "calibration_gpu_ticks": 5110709,
              "compute_passes": 1,
              "command_buffer_ns": 4659583.32643,
              "command_buffers": 2
            },
            {
              "compute_ns": 4652333,
              "compute_span_ns": 4652333,
              "calibration_cpu_ns": 5090708,
              "calibration_gpu_ticks": 5090708,
              "compute_passes": 1,
              "command_buffer_ns": 4652416.6828,
              "command_buffers": 2
            },
            {
              "compute_ns": 4519250,
              "compute_span_ns": 4519250,
              "calibration_cpu_ns": 5055541,
              "calibration_gpu_ticks": 5055541,
              "compute_passes": 1,
              "command_buffer_ns": 4519375.04113,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 74292,
              "compute_span_ns": 74292,
              "calibration_cpu_ns": 564291,
              "calibration_gpu_ticks": 564291,
              "compute_passes": 1,
              "command_buffer_ns": 74416.6900404,
              "command_buffers": 2
            },
            {
              "compute_ns": 73250,
              "compute_span_ns": 73250,
              "calibration_cpu_ns": 610458,
              "calibration_gpu_ticks": 610458,
              "compute_passes": 1,
              "command_buffer_ns": 73291.5941626,
              "command_buffers": 2
            },
            {
              "compute_ns": 74250,
              "compute_span_ns": 74250,
              "calibration_cpu_ns": 500667,
              "calibration_gpu_ticks": 500667,
              "compute_passes": 1,
              "command_buffer_ns": 74333.33667,
              "command_buffers": 2
            },
            {
              "compute_ns": 73083,
              "compute_span_ns": 73083,
              "calibration_cpu_ns": 484125,
              "calibration_gpu_ticks": 484125,
              "compute_passes": 1,
              "command_buffer_ns": 73083.3853595,
              "command_buffers": 2
            },
            {
              "compute_ns": 86167,
              "compute_span_ns": 86167,
              "calibration_cpu_ns": 578583,
              "calibration_gpu_ticks": 578583,
              "compute_passes": 1,
              "command_buffer_ns": 86249.9582581,
              "command_buffers": 2
            },
            {
              "compute_ns": 82834,
              "compute_span_ns": 82834,
              "calibration_cpu_ns": 531583,
              "calibration_gpu_ticks": 531583,
              "compute_passes": 1,
              "command_buffer_ns": 82916.6965559,
              "command_buffers": 2
            },
            {
              "compute_ns": 72292,
              "compute_span_ns": 72292,
              "calibration_cpu_ns": 488583,
              "calibration_gpu_ticks": 488583,
              "compute_passes": 1,
              "command_buffer_ns": 72374.9399185,
              "command_buffers": 2
            },
            {
              "compute_ns": 87458,
              "compute_span_ns": 87458,
              "calibration_cpu_ns": 644500,
              "calibration_gpu_ticks": 644500,
              "compute_passes": 1,
              "command_buffer_ns": 87500.0259839,
              "command_buffers": 2
            },
            {
              "compute_ns": 73542,
              "compute_span_ns": 73542,
              "calibration_cpu_ns": 527083,
              "calibration_gpu_ticks": 527083,
              "compute_passes": 1,
              "command_buffer_ns": 73583.3309591,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4643041.69873,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4609708.38321,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4611541.6917,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4653291.66036,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4552124.99946,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4637166.62489,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4700625.02427,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4592083.33632,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4604250.07623,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 86791.6387506,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 75666.6413508,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 73708.3610147,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 73874.9513403,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 73666.6261218,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 74291.7181924,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76541.6771173,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 83000.0499263,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 77541.6847318,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              72.54752654265626,
              72.02669348765625,
              72.0553389328125,
              72.707682193125,
              71.12695311656249,
              72.45572851390624,
              73.44726600421875,
              71.75130213,
              71.94140744109374
            ],
            "command_buffer_throughput_us_p50": 72.0553389328125,
            "command_buffer_throughput_us_p90": 72.85559895534375,
            "command_buffer_latency_us": [
              86.7916387506,
              75.66664135079999,
              73.7083610147,
              73.8749513403,
              73.6666261218,
              74.2917181924,
              76.5416771173,
              83.0000499263,
              77.54168473179999
            ],
            "command_buffer_latency_us_p50": 75.66664135079999,
            "command_buffer_latency_us_p90": 83.75836769115999
          },
          "compute_throughput_us": [
            71.306640625,
            72.182296875,
            71.66796875,
            71.999359375,
            71.005859375,
            72.888671875,
            72.805984375,
            72.692703125,
            70.61328125
          ],
          "compute_throughput_us_p50": 71.999359375,
          "compute_throughput_us_p90": 72.82252187499999,
          "command_buffer_throughput_us": [
            71.30794347171876,
            72.18424434546874,
            71.66992236312501,
            72.00195250328126,
            71.00716175046874,
            72.88932283703126,
            72.80598947546875,
            72.69401066875,
            70.61523501765625
          ],
          "command_buffer_throughput_us_p50": 72.00195250328126,
          "command_buffer_throughput_us_p90": 72.82265614778125,
          "compute_latency_us": [
            74.292,
            73.25,
            74.25,
            73.083,
            86.167,
            82.834,
            72.292,
            87.458,
            73.542
          ],
          "compute_latency_us_p50": 74.25,
          "compute_latency_us_p90": 86.4252,
          "command_buffer_latency_us": [
            74.4166900404,
            73.29159416259999,
            74.33333667000001,
            73.0833853595,
            86.2499582581,
            82.91669655589999,
            72.37493991849999,
            87.5000259839,
            73.5833309591
          ],
          "command_buffer_latency_us_p50": 74.33333667000001,
          "command_buffer_latency_us_p90": 86.49997180326,
          "counter_control_throughput_ratio": 0.9946511032298817,
          "counter_control_latency_ratio": 0.9892850558079732
        },
        "correctness": {
          "max_abs_error": 2.7199924934606656e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 72.3622521246,
        "throughput_us_p90": 72.72261643057999,
        "latency_us_p50": 295.25,
        "latency_us_p90": 373.3082
      },
      "torch": {
        "cold_call_ms": 7.400333,
        "warmup_ms": 201.620792,
        "repetitions": 108,
        "throughput_us": [
          278.22608333333335,
          276.49923148148144,
          275.2341759259259,
          274.9961388888889,
          274.72106481481484,
          275.4050925925926,
          274.6982962962963,
          274.4706759259259,
          274.8734537037037
        ],
        "latency_us": [
          668.4590000000001,
          525.208,
          518.9169999999999,
          565.584,
          542.083,
          518.417,
          506.791,
          522.5840000000001,
          579.375
        ],
        "throughput_us_p50": 274.9961388888889,
        "throughput_us_p90": 276.84460185185185,
        "latency_us_p50": 525.208,
        "latency_us_p90": 597.1918000000001,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 17067333.33412558,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17044333.333615214,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17051916.68588668,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17215916.654095054,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17128749.983385205,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17075916.63580388,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17095875.05320087,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17102750.01823902,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17112333.327531815,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 271999.97566640377,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 284124.9806806445,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 270916.6146814823,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 273833.34236219525,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 279375.0027194619,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 270874.9962039292,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 271000.0262595713,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 270999.9680519104,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 270499.9642446637,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              266.67708334571216,
              266.3177083377377,
              266.4361982169794,
              268.9986977202352,
              267.63671849039383,
              266.8111974344356,
              267.1230477062636,
              267.2304690349847,
              267.3802082426846
            ],
            "command_buffer_throughput_us_p50": 267.1230477062636,
            "command_buffer_throughput_us_p90": 267.9091143363621,
            "command_buffer_latency_us": [
              271.99997566640377,
              284.1249806806445,
              270.9166146814823,
              273.83334236219525,
              279.3750027194619,
              270.8749962039292,
              271.0000262595713,
              270.9999680519104,
              270.4999642446637
            ],
            "command_buffer_latency_us_p50": 271.0000262595713,
            "command_buffer_latency_us_p90": 280.32499831169844
          },
          "throughput": [
            {
              "compute_ns": 17087083.0,
              "compute_span_ns": 17087083.0,
              "command_buffer_ns": 17087041.691411287,
              "calibration_cpu_ns": 18038292.0,
              "calibration_gpu_ticks": 18038292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17103458.0,
              "compute_span_ns": 17103458.0,
              "command_buffer_ns": 17103458.289057016,
              "calibration_cpu_ns": 18102834.0,
              "calibration_gpu_ticks": 18102834.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17184917.0,
              "compute_span_ns": 17184917.0,
              "command_buffer_ns": 17184916.709084064,
              "calibration_cpu_ns": 18154334.0,
              "calibration_gpu_ticks": 18154334.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17086250.0,
              "compute_span_ns": 17086250.0,
              "command_buffer_ns": 17086249.95080754,
              "calibration_cpu_ns": 18082834.0,
              "calibration_gpu_ticks": 18082834.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17162583.0,
              "compute_span_ns": 17162583.0,
              "command_buffer_ns": 17162541.626021266,
              "calibration_cpu_ns": 18025167.0,
              "calibration_gpu_ticks": 18025167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17110708.0,
              "compute_span_ns": 17110708.0,
              "command_buffer_ns": 17110708.286054432,
              "calibration_cpu_ns": 18027416.0,
              "calibration_gpu_ticks": 18027416.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17147291.0,
              "compute_span_ns": 17147291.0,
              "command_buffer_ns": 17147250.00783801,
              "calibration_cpu_ns": 17915750.0,
              "calibration_gpu_ticks": 17915750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17096209.0,
              "compute_span_ns": 17096209.0,
              "command_buffer_ns": 17096249.968744814,
              "calibration_cpu_ns": 17865708.0,
              "calibration_gpu_ticks": 17865708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17181292.0,
              "compute_span_ns": 17181292.0,
              "command_buffer_ns": 17181333.32906291,
              "calibration_cpu_ns": 17921208.0,
              "calibration_gpu_ticks": 17921208.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 270334.0,
              "compute_span_ns": 270334.0,
              "command_buffer_ns": 270374.9923966825,
              "calibration_cpu_ns": 700875.0,
              "calibration_gpu_ticks": 700875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 271291.0,
              "compute_span_ns": 271291.0,
              "command_buffer_ns": 271249.96995553374,
              "calibration_cpu_ns": 731500.0,
              "calibration_gpu_ticks": 731500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 270750.0,
              "compute_span_ns": 270750.0,
              "command_buffer_ns": 270750.024355948,
              "calibration_cpu_ns": 684000.0,
              "calibration_gpu_ticks": 684000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 274958.0,
              "compute_span_ns": 274958.0,
              "command_buffer_ns": 274958.38003233075,
              "calibration_cpu_ns": 810375.0,
              "calibration_gpu_ticks": 810375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 270958.0,
              "compute_span_ns": 270958.0,
              "command_buffer_ns": 270958.3495743573,
              "calibration_cpu_ns": 655709.0,
              "calibration_gpu_ticks": 655709.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 270958.0,
              "compute_span_ns": 270958.0,
              "command_buffer_ns": 270958.3495743573,
              "calibration_cpu_ns": 714458.0,
              "calibration_gpu_ticks": 714458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 271916.0,
              "compute_span_ns": 271916.0,
              "command_buffer_ns": 271875.00381842256,
              "calibration_cpu_ns": 711084.0,
              "calibration_gpu_ticks": 711084.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 279875.0,
              "compute_span_ns": 279875.0,
              "command_buffer_ns": 279875.0065267086,
              "calibration_cpu_ns": 737084.0,
              "calibration_gpu_ticks": 737084.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 270917.0,
              "compute_span_ns": 270917.0,
              "command_buffer_ns": 270916.6728891432,
              "calibration_cpu_ns": 689125.0,
              "calibration_gpu_ticks": 689125.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            266.985671875,
            267.24153125,
            268.514328125,
            266.97265625,
            268.165359375,
            267.3548125,
            267.926421875,
            267.128265625,
            268.4576875
          ],
          "compute_throughput_us_p50": 267.3548125,
          "compute_throughput_us_p90": 268.469015625,
          "command_buffer_throughput_us": [
            266.98502642830135,
            267.2415357665159,
            268.5143235794385,
            266.97265548136784,
            268.1647129065823,
            267.3548169696005,
            267.9257813724689,
            267.1289057616377,
            268.45833326660795
          ],
          "command_buffer_throughput_us_p50": 267.3548169696005,
          "command_buffer_throughput_us_p90": 268.46953132917406,
          "compute_latency_us": [
            270.334,
            271.291,
            270.75,
            274.958,
            270.958,
            270.958,
            271.916,
            279.875,
            270.917
          ],
          "compute_latency_us_p50": 270.958,
          "compute_latency_us_p90": 275.94140000000004,
          "command_buffer_latency_us": [
            270.3749923966825,
            271.24996995553374,
            270.75002435594797,
            274.95838003233075,
            270.95834957435727,
            270.95834957435727,
            271.87500381842256,
            279.8750065267086,
            270.91667288914323
          ],
          "command_buffer_latency_us_p50": 270.95834957435727,
          "command_buffer_latency_us_p90": 275.9417053312063,
          "counter_control_throughput_ratio": 1.0020374689682898,
          "counter_control_latency_ratio": 1.000307718953747
        },
        "download_ms": 0.669834,
        "allocation_upload_ms": 16.375209,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.046016576746524e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.2631391568513538,
      "device_compute_slowdown": 0.26930264954553607,
      "valid": true,
      "round": 3,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "layernorm_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.543,
        "warmup_ms": 201.596708,
        "repetitions": 107,
        "throughput_us": [
          275.63006542056075,
          275.20054205607477,
          275.3247663551402,
          275.90147663551403,
          275.41160747663554,
          275.55918691588784,
          276.85670093457946,
          274.54594392523364,
          275.7527289719626
        ],
        "latency_us": [
          656.584,
          567.5409999999999,
          508.042,
          512.8340000000001,
          503.95900000000006,
          502.75000000000006,
          646.8340000000001,
          475.75,
          503.87499999999994
        ],
        "throughput_us_p50": 275.55918691588784,
        "throughput_us_p90": 276.0925214953271,
        "latency_us_p50": 508.042,
        "latency_us_p90": 648.784,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 17183166.6957587,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17231666.657608002,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17204416.6829437,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17196125.00164658,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17179500.02057478,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17198916.64106399,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17201250.011567026,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17082583.33383128,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 17138874.97317046,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 281000.0441968441,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 271124.9981075525,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 270750.024355948,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 271333.38153362274,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 271124.9981075525,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 279583.3279378712,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 276750.0118352473,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 270583.3758227527,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 270500.0224523246,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              268.4869796212297,
              269.24479152512504,
              268.81901067099534,
              268.6894531507278,
              268.4296878214809,
              268.7330725166248,
              268.7695314307348,
              266.91536459111376,
              267.7949214557884
            ],
            "command_buffer_throughput_us_p50": 268.6894531507278,
            "command_buffer_throughput_us_p90": 268.9041668418213,
            "command_buffer_latency_us": [
              281.0000441968441,
              271.1249981075525,
              270.75002435594797,
              271.33338153362274,
              271.1249981075525,
              279.5833279378712,
              276.7500118352473,
              270.5833758227527,
              270.50002245232463
            ],
            "command_buffer_latency_us_p50": 271.1249981075525,
            "command_buffer_latency_us_p90": 279.8666711896658
          },
          "throughput": [
            {
              "compute_ns": 17157541.0,
              "compute_span_ns": 17157541.0,
              "command_buffer_ns": 17157500.027678907,
              "calibration_cpu_ns": 17984042.0,
              "calibration_gpu_ticks": 17984042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17217000.0,
              "compute_span_ns": 17217000.0,
              "command_buffer_ns": 17217000.015079975,
              "calibration_cpu_ns": 18052458.0,
              "calibration_gpu_ticks": 18052458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17159291.0,
              "compute_span_ns": 17159291.0,
              "command_buffer_ns": 17159250.04100427,
              "calibration_cpu_ns": 17860958.0,
              "calibration_gpu_ticks": 17860958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17218334.0,
              "compute_span_ns": 17218334.0,
              "command_buffer_ns": 17218374.996446073,
              "calibration_cpu_ns": 17946333.0,
              "calibration_gpu_ticks": 17946333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17222375.0,
              "compute_span_ns": 17222375.0,
              "command_buffer_ns": 17222375.026904047,
              "calibration_cpu_ns": 18024291.0,
              "calibration_gpu_ticks": 18024291.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17168500.0,
              "compute_span_ns": 17168500.0,
              "command_buffer_ns": 17168499.995023012,
              "calibration_cpu_ns": 17963958.0,
              "calibration_gpu_ticks": 17963958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17279292.0,
              "compute_span_ns": 17279292.0,
              "command_buffer_ns": 17279333.318583667,
              "calibration_cpu_ns": 18102625.0,
              "calibration_gpu_ticks": 18102625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17221125.0,
              "compute_span_ns": 17221125.0,
              "command_buffer_ns": 17221125.01738593,
              "calibration_cpu_ns": 18022084.0,
              "calibration_gpu_ticks": 18022084.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 17111834.0,
              "compute_span_ns": 17111834.0,
              "command_buffer_ns": 17111875.000409782,
              "calibration_cpu_ns": 17955417.0,
              "calibration_gpu_ticks": 17955417.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 272708.0,
              "compute_span_ns": 272708.0,
              "command_buffer_ns": 272666.6862145066,
              "calibration_cpu_ns": 782916.0,
              "calibration_gpu_ticks": 782916.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 270875.0,
              "compute_span_ns": 270875.0,
              "command_buffer_ns": 270874.9962039292,
              "calibration_cpu_ns": 732917.0,
              "calibration_gpu_ticks": 732917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 271542.0,
              "compute_span_ns": 271542.0,
              "command_buffer_ns": 271541.70675203204,
              "calibration_cpu_ns": 695292.0,
              "calibration_gpu_ticks": 695292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 280291.0,
              "compute_span_ns": 280291.0,
              "command_buffer_ns": 280249.98027831316,
              "calibration_cpu_ns": 750875.0,
              "calibration_gpu_ticks": 750875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 271209.0,
              "compute_span_ns": 271209.0,
              "command_buffer_ns": 271250.02816319466,
              "calibration_cpu_ns": 694125.0,
              "calibration_gpu_ticks": 694125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 271208.0,
              "compute_span_ns": 271208.0,
              "command_buffer_ns": 271208.3514779806,
              "calibration_cpu_ns": 696208.0,
              "calibration_gpu_ticks": 696208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 274750.0,
              "compute_span_ns": 274750.0,
              "command_buffer_ns": 274749.99660626054,
              "calibration_cpu_ns": 752833.0,
              "calibration_gpu_ticks": 752833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 271125.0,
              "compute_span_ns": 271125.0,
              "command_buffer_ns": 271124.9981075525,
              "calibration_cpu_ns": 693500.0,
              "calibration_gpu_ticks": 693500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 270208.0,
              "compute_span_ns": 270208.0,
              "command_buffer_ns": 270208.34386348724,
              "calibration_cpu_ns": 714958.0,
              "calibration_gpu_ticks": 714958.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            268.086578125,
            269.015625,
            268.113921875,
            269.03646875,
            269.099609375,
            268.2578125,
            269.9889375,
            269.080078125,
            267.37240625
          ],
          "compute_throughput_us_p50": 269.015625,
          "compute_throughput_us_p90": 269.277475,
          "command_buffer_throughput_us": [
            268.0859379324829,
            269.0156252356246,
            268.1132818906917,
            269.0371093194699,
            269.0996097953757,
            268.25781242223457,
            269.9895831028698,
            269.08007839665515,
            267.37304688140284
          ],
          "command_buffer_throughput_us_p50": 269.0156252356246,
          "command_buffer_throughput_us_p90": 269.27760445687454,
          "compute_latency_us": [
            272.708,
            270.875,
            271.542,
            280.291,
            271.209,
            271.208,
            274.75,
            271.125,
            270.208
          ],
          "compute_latency_us_p50": 271.209,
          "compute_latency_us_p90": 275.8582,
          "command_buffer_latency_us": [
            272.6666862145066,
            270.8749962039292,
            271.54170675203204,
            280.24998027831316,
            271.25002816319466,
            271.2083514779806,
            274.74999660626054,
            271.1249981075525,
            270.20834386348724
          ],
          "command_buffer_latency_us_p50": 271.25002816319466,
          "command_buffer_latency_us_p90": 275.84999334067106,
          "counter_control_throughput_ratio": 0.9991488552547207,
          "counter_control_latency_ratio": 0.9990779090627262
        },
        "download_ms": 0.709916,
        "allocation_upload_ms": 1.249875,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.046016576746524e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-ohnmqt4k/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "4f0c4e70dcb06b507aa82e242c8819acd3628d5f79a065eb6c2f6b73ec6f396c",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.248667,
        "capture_ms": 0.079625,
        "compile_ms": 58.5775,
        "allocation_upload_ms": 6.45075,
        "cold_call_ms": 4.180167,
        "warmup_ms": 200.984542,
        "download_ms": 15.021875,
        "repetitions": 394,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          74.7832055838,
          75.2687182741,
          74.8873730964,
          76.0171319797,
          75.7707284264,
          75.6498502538,
          75.9854060914,
          75.124786802,
          75.7137258883
        ],
        "latency_us": [
          349.5,
          322.583,
          303.542,
          311.208,
          360.208,
          286.292,
          300.291,
          283.167,
          307.083
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 768,
            "reduction_scalar_rounds": 120,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 73728,
              "global_write_bytes": 24576,
              "private_read_bytes": 147456,
              "private_write_bytes": 49152
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 288,
              "global_write_bytes": 96,
              "private_read_bytes": 576,
              "private_write_bytes": 192
            },
            "striped_storage_scalars_per_worker": 48,
            "reduction_operations": 2,
            "reduction_elements": 12288,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 168,
            "normalized_kernel_cost": 2016,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 18432,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4789083,
              "compute_span_ns": 4789083,
              "calibration_cpu_ns": 5369209,
              "calibration_gpu_ticks": 5369209,
              "compute_passes": 1,
              "command_buffer_ns": 4789124.94145,
              "command_buffers": 2
            },
            {
              "compute_ns": 4845208,
              "compute_span_ns": 4845208,
              "calibration_cpu_ns": 5290167,
              "calibration_gpu_ticks": 5290167,
              "compute_passes": 1,
              "command_buffer_ns": 4845333.34384,
              "command_buffers": 2
            },
            {
              "compute_ns": 4812583,
              "compute_span_ns": 4812583,
              "calibration_cpu_ns": 5282583,
              "calibration_gpu_ticks": 5282583,
              "compute_passes": 1,
              "command_buffer_ns": 4812750.03403,
              "command_buffers": 2
            },
            {
              "compute_ns": 4897333,
              "compute_span_ns": 4897333,
              "calibration_cpu_ns": 5331416,
              "calibration_gpu_ticks": 5331416,
              "compute_passes": 1,
              "command_buffer_ns": 4897416.6275,
              "command_buffers": 2
            },
            {
              "compute_ns": 4819292,
              "compute_span_ns": 4819292,
              "calibration_cpu_ns": 5364459,
              "calibration_gpu_ticks": 5364459,
              "compute_passes": 1,
              "command_buffer_ns": 4819291.70201,
              "command_buffers": 2
            },
            {
              "compute_ns": 4812458,
              "compute_span_ns": 4812458,
              "calibration_cpu_ns": 5324416,
              "calibration_gpu_ticks": 5324416,
              "compute_passes": 1,
              "command_buffer_ns": 4812583.3855,
              "command_buffers": 2
            },
            {
              "compute_ns": 4824333,
              "compute_span_ns": 4824333,
              "calibration_cpu_ns": 5369375,
              "calibration_gpu_ticks": 5369375,
              "compute_passes": 1,
              "command_buffer_ns": 4824375.03524,
              "command_buffers": 2
            },
            {
              "compute_ns": 4778292,
              "compute_span_ns": 4778292,
              "calibration_cpu_ns": 5372542,
              "calibration_gpu_ticks": 5372542,
              "compute_passes": 1,
              "command_buffer_ns": 4778291.62264,
              "command_buffers": 2
            },
            {
              "compute_ns": 5063250,
              "compute_span_ns": 5063250,
              "calibration_cpu_ns": 7234541,
              "calibration_gpu_ticks": 7234541,
              "compute_passes": 1,
              "command_buffer_ns": 5063291.58088,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 84750,
              "compute_span_ns": 84750,
              "calibration_cpu_ns": 704250,
              "calibration_gpu_ticks": 704250,
              "compute_passes": 1,
              "command_buffer_ns": 84791.6817293,
              "command_buffers": 2
            },
            {
              "compute_ns": 91166,
              "compute_span_ns": 91166,
              "calibration_cpu_ns": 837916,
              "calibration_gpu_ticks": 837916,
              "compute_passes": 1,
              "command_buffer_ns": 91124.9662749,
              "command_buffers": 2
            },
            {
              "compute_ns": 112792,
              "compute_span_ns": 112792,
              "calibration_cpu_ns": 791958,
              "calibration_gpu_ticks": 791958,
              "compute_passes": 1,
              "command_buffer_ns": 112791.662104,
              "command_buffers": 2
            },
            {
              "compute_ns": 101709,
              "compute_span_ns": 101709,
              "calibration_cpu_ns": 698959,
              "calibration_gpu_ticks": 698959,
              "compute_passes": 1,
              "command_buffer_ns": 101833.313238,
              "command_buffers": 2
            },
            {
              "compute_ns": 75583,
              "compute_span_ns": 75583,
              "calibration_cpu_ns": 609000,
              "calibration_gpu_ticks": 609000,
              "compute_passes": 1,
              "command_buffer_ns": 75624.9646656,
              "command_buffers": 2
            },
            {
              "compute_ns": 91208,
              "compute_span_ns": 91208,
              "calibration_cpu_ns": 799542,
              "calibration_gpu_ticks": 799542,
              "compute_passes": 1,
              "command_buffer_ns": 91250.0545382,
              "command_buffers": 2
            },
            {
              "compute_ns": 76292,
              "compute_span_ns": 76292,
              "calibration_cpu_ns": 650958,
              "calibration_gpu_ticks": 650958,
              "compute_passes": 1,
              "command_buffer_ns": 76291.6170061,
              "command_buffers": 2
            },
            {
              "compute_ns": 116792,
              "compute_span_ns": 116792,
              "calibration_cpu_ns": 693709,
              "calibration_gpu_ticks": 693709,
              "compute_passes": 1,
              "command_buffer_ns": 116833.31104,
              "command_buffers": 2
            },
            {
              "compute_ns": 149916,
              "compute_span_ns": 149916,
              "calibration_cpu_ns": 1987208,
              "calibration_gpu_ticks": 1987208,
              "compute_passes": 1,
              "command_buffer_ns": 149875.006173,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4856166.66265,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4811458.34783,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4842333.32099,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4840666.66104,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4874000.03476,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4823083.34904,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4833833.31448,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4810958.28582,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4772583.25554,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 87750.0278875,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 84749.9468364,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 75375.0209697,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 97666.6342467,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76291.6752137,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 135041.656904,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 98499.993328,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 147583.370563,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 148624.996655,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              75.87760410390625,
              75.17903668484375,
              75.66145814046875,
              75.63541657875,
              76.156250543125,
              75.36067732875,
              75.52864553875,
              75.1712232159375,
              74.5716133678125
            ],
            "command_buffer_throughput_us_p50": 75.52864553875,
            "command_buffer_throughput_us_p90": 75.93333339175,
            "command_buffer_latency_us": [
              87.7500278875,
              84.7499468364,
              75.37502096969999,
              97.66663424670001,
              76.2916752137,
              135.041656904,
              98.499993328,
              147.58337056300002,
              148.624996655
            ],
            "command_buffer_latency_us_p50": 97.66663424670001,
            "command_buffer_latency_us_p90": 147.79169578140002
          },
          "compute_throughput_us": [
            74.829421875,
            75.706375,
            75.196609375,
            76.520828125,
            75.3014375,
            75.19465625,
            75.380203125,
            74.6608125,
            79.11328125
          ],
          "compute_throughput_us_p50": 75.3014375,
          "compute_throughput_us_p90": 77.03931874999999,
          "command_buffer_throughput_us": [
            74.83007721015625,
            75.7083334975,
            75.19921928171874,
            76.52213480468751,
            75.30143284390626,
            75.1966153984375,
            75.380859925625,
            74.66080660374999,
            79.11393095125
          ],
          "command_buffer_throughput_us_p50": 75.30143284390626,
          "command_buffer_throughput_us_p90": 77.040494034,
          "compute_latency_us": [
            84.75,
            91.166,
            112.792,
            101.709,
            75.583,
            91.208,
            76.292,
            116.792,
            149.916
          ],
          "compute_latency_us_p50": 91.208,
          "compute_latency_us_p90": 123.41680000000001,
          "command_buffer_latency_us": [
            84.79168172930001,
            91.1249662749,
            112.791662104,
            101.833313238,
            75.6249646656,
            91.2500545382,
            76.2916170061,
            116.83331104,
            149.875006173
          ],
          "command_buffer_latency_us_p50": 91.2500545382,
          "command_buffer_latency_us_p90": 123.4416500666,
          "counter_control_throughput_ratio": 0.9978229769671947,
          "counter_control_latency_ratio": 0.9912610314790902
        },
        "correctness": {
          "max_abs_error": 2.7199924934606656e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 75.6498502538,
        "throughput_us_p90": 75.99175126905999,
        "latency_us_p50": 307.083,
        "latency_us_p90": 351.6416
      },
      "slowdown": 0.27453212901550433,
      "device_compute_slowdown": 0.2799147354359064,
      "valid": true,
      "round": 3,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "layernorm_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-uz5r1khm/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "6caff4b0a979c3104fea9f23f0d5430392f6f4e7d743dab03e9fdb6de1428bf3",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.736708,
        "capture_ms": 0.072834,
        "compile_ms": 59.92075,
        "allocation_upload_ms": 5.8585,
        "cold_call_ms": 5.628958,
        "warmup_ms": 200.684916,
        "download_ms": 3.683833,
        "repetitions": 468,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          62.8797200855,
          62.0661495726,
          62.2834764957,
          62.5283119658,
          62.3776709402,
          62.3291495726,
          62.0056965812,
          62.4007307692,
          62.4147970085
        ],
        "latency_us": [
          388.375,
          288.458,
          280.333,
          354.5,
          287.584,
          262.167,
          272.875,
          286.833,
          262.166
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 1024,
            "reduction_scalar_rounds": 80,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 49152,
              "global_write_bytes": 16384,
              "private_read_bytes": 98304,
              "private_write_bytes": 32768
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 192,
              "global_write_bytes": 64,
              "private_read_bytes": 384,
              "private_write_bytes": 128
            },
            "striped_storage_scalars_per_worker": 32,
            "reduction_operations": 2,
            "reduction_elements": 8192,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 128,
            "normalized_kernel_cost": 2048,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3919042,
              "compute_span_ns": 3919042,
              "calibration_cpu_ns": 4376958,
              "calibration_gpu_ticks": 4376958,
              "compute_passes": 1,
              "command_buffer_ns": 3919083.33404,
              "command_buffers": 2
            },
            {
              "compute_ns": 3921750,
              "compute_span_ns": 3921750,
              "calibration_cpu_ns": 4424708,
              "calibration_gpu_ticks": 4424708,
              "compute_passes": 1,
              "command_buffer_ns": 3921791.62009,
              "command_buffers": 2
            },
            {
              "compute_ns": 3949542,
              "compute_span_ns": 3949542,
              "calibration_cpu_ns": 4417916,
              "calibration_gpu_ticks": 4417916,
              "compute_passes": 1,
              "command_buffer_ns": 3949624.95193,
              "command_buffers": 2
            },
            {
              "compute_ns": 3933167,
              "compute_span_ns": 3933167,
              "calibration_cpu_ns": 4397167,
              "calibration_gpu_ticks": 4397167,
              "compute_passes": 1,
              "command_buffer_ns": 3933291.64945,
              "command_buffers": 2
            },
            {
              "compute_ns": 3975917,
              "compute_span_ns": 3975917,
              "calibration_cpu_ns": 4428917,
              "calibration_gpu_ticks": 4428917,
              "compute_passes": 1,
              "command_buffer_ns": 3976083.36061,
              "command_buffers": 2
            },
            {
              "compute_ns": 3932125,
              "compute_span_ns": 3932125,
              "calibration_cpu_ns": 4389458,
              "calibration_gpu_ticks": 4389458,
              "compute_passes": 1,
              "command_buffer_ns": 3932166.66998,
              "command_buffers": 2
            },
            {
              "compute_ns": 3984583,
              "compute_span_ns": 3984583,
              "calibration_cpu_ns": 4469709,
              "calibration_gpu_ticks": 4469709,
              "compute_passes": 1,
              "command_buffer_ns": 3984624.98561,
              "command_buffers": 2
            },
            {
              "compute_ns": 3975625,
              "compute_span_ns": 3975625,
              "calibration_cpu_ns": 4424875,
              "calibration_gpu_ticks": 4424875,
              "compute_passes": 1,
              "command_buffer_ns": 3975624.97528,
              "command_buffers": 2
            },
            {
              "compute_ns": 3956375,
              "compute_span_ns": 3956375,
              "calibration_cpu_ns": 4594541,
              "calibration_gpu_ticks": 4594541,
              "compute_passes": 1,
              "command_buffer_ns": 3956375.00333,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 71167,
              "compute_span_ns": 71167,
              "calibration_cpu_ns": 557834,
              "calibration_gpu_ticks": 557834,
              "compute_passes": 1,
              "command_buffer_ns": 71291.6371413,
              "command_buffers": 2
            },
            {
              "compute_ns": 70375,
              "compute_span_ns": 70375,
              "calibration_cpu_ns": 500125,
              "calibration_gpu_ticks": 500125,
              "compute_passes": 1,
              "command_buffer_ns": 70416.6595824,
              "command_buffers": 2
            },
            {
              "compute_ns": 63250,
              "compute_span_ns": 63250,
              "calibration_cpu_ns": 504000,
              "calibration_gpu_ticks": 504000,
              "compute_passes": 1,
              "command_buffer_ns": 63333.3111182,
              "command_buffers": 2
            },
            {
              "compute_ns": 63084,
              "compute_span_ns": 63084,
              "calibration_cpu_ns": 459083,
              "calibration_gpu_ticks": 459083,
              "compute_passes": 1,
              "command_buffer_ns": 63208.3392702,
              "command_buffers": 2
            },
            {
              "compute_ns": 70084,
              "compute_span_ns": 70084,
              "calibration_cpu_ns": 504625,
              "calibration_gpu_ticks": 504625,
              "compute_passes": 1,
              "command_buffer_ns": 70124.9809936,
              "command_buffers": 2
            },
            {
              "compute_ns": 63583,
              "compute_span_ns": 63583,
              "calibration_cpu_ns": 464000,
              "calibration_gpu_ticks": 464000,
              "compute_passes": 1,
              "command_buffer_ns": 63541.6363366,
              "command_buffers": 2
            },
            {
              "compute_ns": 63500,
              "compute_span_ns": 63500,
              "calibration_cpu_ns": 431042,
              "calibration_gpu_ticks": 431042,
              "compute_passes": 1,
              "command_buffer_ns": 63500.017859,
              "command_buffers": 2
            },
            {
              "compute_ns": 63917,
              "compute_span_ns": 63917,
              "calibration_cpu_ns": 497375,
              "calibration_gpu_ticks": 497375,
              "compute_passes": 1,
              "command_buffer_ns": 63958.3449811,
              "command_buffers": 2
            },
            {
              "compute_ns": 63625,
              "compute_span_ns": 63625,
              "calibration_cpu_ns": 436750,
              "calibration_gpu_ticks": 436750,
              "compute_passes": 1,
              "command_buffer_ns": 63666.7245999,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4000375.04733,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3933375.06102,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3945374.97777,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3943999.99641,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4001291.64336,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3968958.33546,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3996541.72361,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3958875.02236,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3941624.94922,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 63124.9858998,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 69833.3024047,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63208.3974779,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63833.3149254,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 70666.661486,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63583.3130218,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63791.6382402,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 70375.0411049,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 70999.9585524,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              62.50586011453125,
              61.4589853284375,
              61.64648402765626,
              61.62499994390625,
              62.520181927500005,
              62.014973991562506,
              62.44596443140625,
              61.857422224375,
              61.5878898315625
            ],
            "command_buffer_throughput_us_p50": 61.857422224375,
            "command_buffer_throughput_us_p90": 62.508724477125,
            "command_buffer_latency_us": [
              63.124985899799995,
              69.8333024047,
              63.2083974779,
              63.8333149254,
              70.666661486,
              63.5833130218,
              63.7916382402,
              70.37504110489999,
              70.9999585524
            ],
            "command_buffer_latency_us_p50": 63.8333149254,
            "command_buffer_latency_us_p90": 70.73332089928
          },
          "compute_throughput_us": [
            61.23503125,
            61.27734375,
            61.71159375,
            61.455734375,
            62.123703125,
            61.439453125,
            62.259109375,
            62.119140625,
            61.818359375
          ],
          "compute_throughput_us_p50": 61.71159375,
          "compute_throughput_us_p90": 62.150784375,
          "command_buffer_throughput_us": [
            61.235677094375,
            61.27799406390625,
            61.71288987390625,
            61.45768202265625,
            62.12630250953125,
            61.4401042184375,
            62.25976540015625,
            62.11914023875,
            61.818359427031254
          ],
          "command_buffer_throughput_us_p50": 61.71288987390625,
          "command_buffer_throughput_us_p90": 62.15299508765625,
          "compute_latency_us": [
            71.167,
            70.375,
            63.25,
            63.084,
            70.084,
            63.583,
            63.5,
            63.917,
            63.625
          ],
          "compute_latency_us_p50": 63.625,
          "compute_latency_us_p90": 70.5334,
          "command_buffer_latency_us": [
            71.2916371413,
            70.4166595824,
            63.3333111182,
            63.2083392702,
            70.12498099359999,
            63.5416363366,
            63.500017858999996,
            63.958344981100005,
            63.6667245999
          ],
          "command_buffer_latency_us_p50": 63.6667245999,
          "command_buffer_latency_us_p90": 70.59165509418,
          "counter_control_throughput_ratio": 0.9970550886324591,
          "counter_control_latency_ratio": 0.9954285484862148
        },
        "correctness": {
          "max_abs_error": 2.827305296548843e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 62.3776709402,
        "throughput_us_p90": 62.59859358974,
        "latency_us_p50": 286.833,
        "latency_us_p90": 361.275
      },
      "torch": {
        "cold_call_ms": 6.67925,
        "warmup_ms": 201.0485,
        "repetitions": 143,
        "throughput_us": [
          207.24358741258743,
          206.42045454545453,
          205.96008391608393,
          205.30274125874124,
          205.03117482517484,
          205.22814685314685,
          205.17307692307693,
          205.94872027972028,
          207.32837762237762
        ],
        "latency_us": [
          511.667,
          604.0,
          472.292,
          450.625,
          450.209,
          490.125,
          463.583,
          484.458,
          384.625
        ],
        "throughput_us_p50": 205.94872027972028,
        "throughput_us_p90": 207.26054545454548,
        "latency_us_p50": 472.292,
        "latency_us_p90": 530.1336,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 12797624.978702515,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12826124.962884933,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12802541.663404554,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12772999.99492243,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12801999.982912093,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12684875.051490963,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12709333.328530192,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12804500.001948327,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12712583.295069635,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 203583.33131298423,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 201374.9908655882,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200291.68808832765,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200541.689991951,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200874.9870583415,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 201208.34233239293,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 201750.02282485366,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 202875.00228732824,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 202499.97032806277,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              199.9628902922268,
              200.40820254507707,
              200.03971349069616,
              199.57812492066296,
              200.03124973300146,
              198.2011726795463,
              198.58333325828426,
              200.0703125304426,
              198.63411398546305
            ],
            "command_buffer_throughput_us_p50": 199.9628902922268,
            "command_buffer_throughput_us_p90": 200.1378905333695,
            "command_buffer_latency_us": [
              203.58333131298423,
              201.3749908655882,
              200.29168808832765,
              200.541689991951,
              200.8749870583415,
              201.20834233239293,
              201.75002282485366,
              202.87500228732824,
              202.49997032806277
            ],
            "command_buffer_latency_us_p50": 201.3749908655882,
            "command_buffer_latency_us_p90": 203.01666809245944
          },
          "throughput": [
            {
              "compute_ns": 12797958.0,
              "compute_span_ns": 12797958.0,
              "command_buffer_ns": 12797916.657291353,
              "calibration_cpu_ns": 13762625.0,
              "calibration_gpu_ticks": 13762625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12824167.0,
              "compute_span_ns": 12824167.0,
              "command_buffer_ns": 12824166.62434116,
              "calibration_cpu_ns": 13717292.0,
              "calibration_gpu_ticks": 13717292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12800500.0,
              "compute_span_ns": 12800500.0,
              "command_buffer_ns": 12800500.029698014,
              "calibration_cpu_ns": 13891958.0,
              "calibration_gpu_ticks": 13891958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12805750.0,
              "compute_span_ns": 12805750.0,
              "command_buffer_ns": 12805750.011466444,
              "calibration_cpu_ns": 13851208.0,
              "calibration_gpu_ticks": 13851208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12784625.0,
              "compute_span_ns": 12784625.0,
              "command_buffer_ns": 12784624.996129423,
              "calibration_cpu_ns": 14925709.0,
              "calibration_gpu_ticks": 14925709.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12660625.0,
              "compute_span_ns": 12660625.0,
              "command_buffer_ns": 12660625.041462481,
              "calibration_cpu_ns": 14644917.0,
              "calibration_gpu_ticks": 14644917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12733625.0,
              "compute_span_ns": 12733625.0,
              "command_buffer_ns": 12733625.015243888,
              "calibration_cpu_ns": 15298375.0,
              "calibration_gpu_ticks": 15298375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12700500.0,
              "compute_span_ns": 12700500.0,
              "command_buffer_ns": 12700500.02494827,
              "calibration_cpu_ns": 19169375.0,
              "calibration_gpu_ticks": 19169375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12704042.0,
              "compute_span_ns": 12704042.0,
              "command_buffer_ns": 12704083.346761763,
              "calibration_cpu_ns": 15133625.0,
              "calibration_gpu_ticks": 15133625.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 201167.0,
              "compute_span_ns": 201167.0,
              "command_buffer_ns": 201166.6656471789,
              "calibration_cpu_ns": 1072875.0,
              "calibration_gpu_ticks": 1072875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 200375.0,
              "compute_span_ns": 200375.0,
              "command_buffer_ns": 200374.98325109482,
              "calibration_cpu_ns": 824750.0,
              "calibration_gpu_ticks": 824750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201500.0,
              "compute_span_ns": 201500.0,
              "command_buffer_ns": 201500.02092123032,
              "calibration_cpu_ns": 751875.0,
              "calibration_gpu_ticks": 751875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 200833.0,
              "compute_span_ns": 200833.0,
              "command_buffer_ns": 200833.31037312746,
              "calibration_cpu_ns": 819167.0,
              "calibration_gpu_ticks": 819167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 202209.0,
              "compute_span_ns": 202209.0,
              "command_buffer_ns": 202249.96842443943,
              "calibration_cpu_ns": 1055167.0,
              "calibration_gpu_ticks": 1055167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 212958.0,
              "compute_span_ns": 212958.0,
              "command_buffer_ns": 212958.37359502912,
              "calibration_cpu_ns": 1369583.0,
              "calibration_gpu_ticks": 1369583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 200750.0,
              "compute_span_ns": 200750.0,
              "command_buffer_ns": 200749.95700269938,
              "calibration_cpu_ns": 719167.0,
              "calibration_gpu_ticks": 719167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 199792.0,
              "compute_span_ns": 199792.0,
              "command_buffer_ns": 199791.68428108096,
              "calibration_cpu_ns": 753625.0,
              "calibration_gpu_ticks": 753625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201417.0,
              "compute_span_ns": 201417.0,
              "command_buffer_ns": 201458.34423601627,
              "calibration_cpu_ns": 1010666.0,
              "calibration_gpu_ticks": 1010666.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            199.96809375,
            200.377609375,
            200.0078125,
            200.08984375,
            199.759765625,
            197.822265625,
            198.962890625,
            198.4453125,
            198.50065625
          ],
          "compute_throughput_us_p50": 199.759765625,
          "compute_throughput_us_p90": 200.147396875,
          "command_buffer_throughput_us": [
            199.9674477701774,
            200.37760350533063,
            200.00781296403147,
            200.08984392916318,
            199.75976556452224,
            197.82226627285127,
            198.96289086318575,
            198.4453128898167,
            198.50130229315255
          ],
          "command_buffer_throughput_us_p50": 199.75976556452224,
          "command_buffer_throughput_us_p90": 200.14739584439667,
          "compute_latency_us": [
            201.167,
            200.375,
            201.5,
            200.833,
            202.209,
            212.958,
            200.75,
            199.792,
            201.417
          ],
          "compute_latency_us_p50": 201.167,
          "compute_latency_us_p90": 204.3588,
          "command_buffer_latency_us": [
            201.1666656471789,
            200.37498325109482,
            201.50002092123032,
            200.83331037312746,
            202.24996842443943,
            212.95837359502912,
            200.74995700269938,
            199.79168428108096,
            201.45834423601627
          ],
          "command_buffer_latency_us_p50": 201.1666656471789,
          "command_buffer_latency_us_p90": 204.39164945855737,
          "counter_control_throughput_ratio": 0.9998405290324205,
          "counter_control_latency_ratio": 0.9950430448128252
        },
        "download_ms": 1.170791,
        "allocation_upload_ms": 14.867834,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.1634934849478213e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.30287962389607676,
      "device_compute_slowdown": 0.3089290456309825,
      "valid": true,
      "round": 3,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "layernorm",
        "m": 1024,
        "n": 4096,
        "k": 1
      },
      "name": "layernorm_1024x4096",
      "backend": "metal",
      "block": [
        1,
        4096,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.599583,
        "warmup_ms": 201.600833,
        "repetitions": 143,
        "throughput_us": [
          204.95658741258742,
          205.32167832167832,
          205.4018111888112,
          205.88024475524475,
          206.30594405594405,
          207.50758041958045,
          206.7695174825175,
          205.93851748251748,
          206.26427972027972
        ],
        "latency_us": [
          445.666,
          432.66700000000003,
          502.87499999999994,
          438.167,
          410.916,
          408.08299999999997,
          417.084,
          459.041,
          408.333
        ],
        "throughput_us_p50": 205.93851748251748,
        "throughput_us_p90": 206.9171300699301,
        "latency_us_p50": 432.66700000000003,
        "latency_us_p90": 467.8078,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 12783875.048626214,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12699625.047389418,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12789916.654583067,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12828958.337195218,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12797000.003047287,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12821500.014979392,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12812625.034712255,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12819624.971598387,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 12876541.644800454,
                "command_buffers": 1
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 204749.98746067286,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200750.0152103603,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 202041.64320603013,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200958.3404287696,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200166.65803268552,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 199958.33281427622,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 201083.3122767508,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 200333.30656588078,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 201208.34233239293,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              199.7480476347846,
              198.43164136545965,
              199.84244772786042,
              200.45247401867528,
              199.95312504761387,
              200.335937734053,
              200.19726616737898,
              200.3066401812248,
              201.1959632000071
            ],
            "command_buffer_throughput_us_p50": 200.19726616737898,
            "command_buffer_throughput_us_p90": 200.60117185494164,
            "command_buffer_latency_us": [
              204.74998746067286,
              200.7500152103603,
              202.04164320603013,
              200.9583404287696,
              200.16665803268552,
              199.95833281427622,
              201.0833122767508,
              200.33330656588078,
              201.20834233239293
            ],
            "command_buffer_latency_us_p50": 200.9583404287696,
            "command_buffer_latency_us_p90": 202.58331205695868
          },
          "throughput": [
            {
              "compute_ns": 12818792.0,
              "compute_span_ns": 12818792.0,
              "command_buffer_ns": 12818791.67072475,
              "calibration_cpu_ns": 13653875.0,
              "calibration_gpu_ticks": 13653875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12791000.0,
              "compute_span_ns": 12791000.0,
              "command_buffer_ns": 12791000.015567988,
              "calibration_cpu_ns": 13576209.0,
              "calibration_gpu_ticks": 13576209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12721791.0,
              "compute_span_ns": 12721791.0,
              "command_buffer_ns": 12721750.01213327,
              "calibration_cpu_ns": 13536541.0,
              "calibration_gpu_ticks": 13536541.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12765500.0,
              "compute_span_ns": 12765500.0,
              "command_buffer_ns": 12765499.99602139,
              "calibration_cpu_ns": 13535000.0,
              "calibration_gpu_ticks": 13535000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12818666.0,
              "compute_span_ns": 12818666.0,
              "command_buffer_ns": 12818624.963983893,
              "calibration_cpu_ns": 13536208.0,
              "calibration_gpu_ticks": 13536208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12750667.0,
              "compute_span_ns": 12750667.0,
              "command_buffer_ns": 12750666.646752506,
              "calibration_cpu_ns": 13540083.0,
              "calibration_gpu_ticks": 13540083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12794125.0,
              "compute_span_ns": 12794125.0,
              "command_buffer_ns": 12794125.01025945,
              "calibration_cpu_ns": 13516958.0,
              "calibration_gpu_ticks": 13516958.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12833625.0,
              "compute_span_ns": 12833625.0,
              "command_buffer_ns": 12833625.019993633,
              "calibration_cpu_ns": 13561166.0,
              "calibration_gpu_ticks": 13561166.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 12751916.0,
              "compute_span_ns": 12751916.0,
              "command_buffer_ns": 12751875.03779307,
              "calibration_cpu_ns": 13593000.0,
              "calibration_gpu_ticks": 13593000.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "latency": [
            {
              "compute_ns": 200958.0,
              "compute_span_ns": 200958.0,
              "command_buffer_ns": 200958.3404287696,
              "calibration_cpu_ns": 634250.0,
              "calibration_gpu_ticks": 634250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 200334.0,
              "compute_span_ns": 200334.0,
              "command_buffer_ns": 200374.98325109482,
              "calibration_cpu_ns": 623209.0,
              "calibration_gpu_ticks": 623209.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 202083.0,
              "compute_span_ns": 202083.0,
              "command_buffer_ns": 202041.70141369104,
              "calibration_cpu_ns": 634416.0,
              "calibration_gpu_ticks": 634416.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201041.0,
              "compute_span_ns": 201041.0,
              "command_buffer_ns": 201000.01711398363,
              "calibration_cpu_ns": 670458.0,
              "calibration_gpu_ticks": 670458.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201500.0,
              "compute_span_ns": 201500.0,
              "command_buffer_ns": 201499.9627135694,
              "calibration_cpu_ns": 638208.0,
              "calibration_gpu_ticks": 638208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 200916.0,
              "compute_span_ns": 200916.0,
              "command_buffer_ns": 200874.9870583415,
              "calibration_cpu_ns": 611917.0,
              "calibration_gpu_ticks": 611917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201417.0,
              "compute_span_ns": 201417.0,
              "command_buffer_ns": 201458.34423601627,
              "calibration_cpu_ns": 657208.0,
              "calibration_gpu_ticks": 657208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 201208.0,
              "compute_span_ns": 201208.0,
              "command_buffer_ns": 201208.34233239293,
              "calibration_cpu_ns": 640750.0,
              "calibration_gpu_ticks": 640750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 199917.0,
              "compute_span_ns": 199917.0,
              "command_buffer_ns": 199916.65612906218,
              "calibration_cpu_ns": 629083.0,
              "calibration_gpu_ticks": 629083.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            200.293625,
            199.859375,
            198.777984375,
            199.4609375,
            200.29165625,
            199.229171875,
            199.908203125,
            200.525390625,
            199.2486875
          ],
          "compute_throughput_us_p50": 199.859375,
          "compute_throughput_us_p90": 200.339978125,
          "command_buffer_throughput_us": [
            200.2936198550742,
            199.85937524324981,
            198.77734393958235,
            199.46093743783422,
            200.29101506224833,
            199.2291663555079,
            199.9082032853039,
            200.52539093740052,
            199.24804746551672
          ],
          "command_buffer_throughput_us_p50": 199.85937524324981,
          "command_buffer_throughput_us_p90": 200.33997407153947,
          "compute_latency_us": [
            200.958,
            200.334,
            202.083,
            201.041,
            201.5,
            200.916,
            201.417,
            201.208,
            199.917
          ],
          "compute_latency_us_p50": 201.041,
          "compute_latency_us_p90": 201.6166,
          "command_buffer_latency_us": [
            200.9583404287696,
            200.37498325109482,
            202.04170141369104,
            201.00001711398363,
            201.4999627135694,
            200.8749870583415,
            201.45834423601627,
            201.20834233239293,
            199.91665612906218
          ],
          "command_buffer_latency_us_p50": 201.00001711398363,
          "command_buffer_latency_us_p90": 201.60831045359373,
          "counter_control_throughput_ratio": 0.99855610974311,
          "counter_control_latency_ratio": 1.0002073896765127
        },
        "download_ms": 0.591417,
        "allocation_upload_ms": 1.235458,
        "device": "mps:0",
        "output_policy": "framework_return_value",
        "correctness": {
          "max_abs_error": 3.1634934849478213e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "layernorm",
        "1024",
        "4096",
        "1",
        "1",
        "4096",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-qzwaf1gn/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "7d02ca31ce0660bb7b8cf2d4b7c68e2de8afe58cfb625e16c6aa93173ce8f234",
      "native": {
        "backend": "metal",
        "operation": "layernorm",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.390375,
        "capture_ms": 0.0685,
        "compile_ms": 58.161375,
        "allocation_upload_ms": 5.536084,
        "cold_call_ms": 7.166917,
        "warmup_ms": 200.777917,
        "download_ms": 3.901667,
        "repetitions": 323,
        "output_elements": 4194304,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          76.0839783282,
          76.6306749226,
          76.2769628483,
          76.1715665635,
          76.1585417957,
          76.3707430341,
          76.7202012384,
          76.7969566563,
          76.5735294118
        ],
        "latency_us": [
          399.541,
          316.916,
          311.458,
          336.042,
          330.792,
          318.625,
          319.958,
          318.125,
          353.25
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 1024,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 512,
            "reduction_scalar_rounds": 80,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 49152,
              "global_write_bytes": 16384,
              "private_read_bytes": 98304,
              "private_write_bytes": 32768
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 192,
              "global_write_bytes": 64,
              "private_read_bytes": 384,
              "private_write_bytes": 128
            },
            "striped_storage_scalars_per_worker": 32,
            "reduction_operations": 2,
            "reduction_elements": 8192,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 120,
            "normalized_kernel_cost": 1920,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 12288,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 16,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4914750,
              "compute_span_ns": 4914750,
              "calibration_cpu_ns": 5385750,
              "calibration_gpu_ticks": 5385750,
              "compute_passes": 1,
              "command_buffer_ns": 4914749.99581,
              "command_buffers": 2
            },
            {
              "compute_ns": 4861083,
              "compute_span_ns": 4861083,
              "calibration_cpu_ns": 5380333,
              "calibration_gpu_ticks": 5380333,
              "compute_passes": 1,
              "command_buffer_ns": 4861083.34735,
              "command_buffers": 2
            },
            {
              "compute_ns": 4883417,
              "compute_span_ns": 4883417,
              "calibration_cpu_ns": 5386500,
              "calibration_gpu_ticks": 5386500,
              "compute_passes": 1,
              "command_buffer_ns": 4883500.04889,
              "command_buffers": 2
            },
            {
              "compute_ns": 4921583,
              "compute_span_ns": 4921583,
              "calibration_cpu_ns": 5384834,
              "calibration_gpu_ticks": 5384834,
              "compute_passes": 1,
              "command_buffer_ns": 4921624.96084,
              "command_buffers": 2
            },
            {
              "compute_ns": 4886083,
              "compute_span_ns": 4886083,
              "calibration_cpu_ns": 5502167,
              "calibration_gpu_ticks": 5502167,
              "compute_passes": 1,
              "command_buffer_ns": 4886208.33494,
              "command_buffers": 2
            },
            {
              "compute_ns": 4913666,
              "compute_span_ns": 4913666,
              "calibration_cpu_ns": 5430125,
              "calibration_gpu_ticks": 5430125,
              "compute_passes": 1,
              "command_buffer_ns": 4913749.92998,
              "command_buffers": 2
            },
            {
              "compute_ns": 4847459,
              "compute_span_ns": 4847459,
              "calibration_cpu_ns": 5387333,
              "calibration_gpu_ticks": 5387333,
              "compute_passes": 1,
              "command_buffer_ns": 4847583.36097,
              "command_buffers": 2
            },
            {
              "compute_ns": 4879333,
              "compute_span_ns": 4879333,
              "calibration_cpu_ns": 5348375,
              "calibration_gpu_ticks": 5348375,
              "compute_passes": 1,
              "command_buffer_ns": 4879333.3699,
              "command_buffers": 2
            },
            {
              "compute_ns": 4886625,
              "compute_span_ns": 4886625,
              "calibration_cpu_ns": 5506500,
              "calibration_gpu_ticks": 5506500,
              "compute_passes": 1,
              "command_buffer_ns": 4886666.66206,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 77209,
              "compute_span_ns": 77209,
              "calibration_cpu_ns": 519000,
              "calibration_gpu_ticks": 519000,
              "compute_passes": 1,
              "command_buffer_ns": 77291.6828282,
              "command_buffers": 2
            },
            {
              "compute_ns": 76667,
              "compute_span_ns": 76667,
              "calibration_cpu_ns": 504292,
              "calibration_gpu_ticks": 504292,
              "compute_passes": 1,
              "command_buffer_ns": 76666.707173,
              "command_buffers": 2
            },
            {
              "compute_ns": 77250,
              "compute_span_ns": 77250,
              "calibration_cpu_ns": 534667,
              "calibration_gpu_ticks": 534667,
              "compute_passes": 1,
              "command_buffer_ns": 77333.3595134,
              "command_buffers": 2
            },
            {
              "compute_ns": 77208,
              "compute_span_ns": 77208,
              "calibration_cpu_ns": 502792,
              "calibration_gpu_ticks": 502792,
              "compute_passes": 1,
              "command_buffer_ns": 77291.6828282,
              "command_buffers": 2
            },
            {
              "compute_ns": 85375,
              "compute_span_ns": 85375,
              "calibration_cpu_ns": 644375,
              "calibration_gpu_ticks": 644375,
              "compute_passes": 1,
              "command_buffer_ns": 85458.3340697,
              "command_buffers": 2
            },
            {
              "compute_ns": 85750,
              "compute_span_ns": 85750,
              "calibration_cpu_ns": 571375,
              "calibration_gpu_ticks": 571375,
              "compute_passes": 1,
              "command_buffer_ns": 85791.6311361,
              "command_buffers": 2
            },
            {
              "compute_ns": 77625,
              "compute_span_ns": 77625,
              "calibration_cpu_ns": 526250,
              "calibration_gpu_ticks": 526250,
              "compute_passes": 1,
              "command_buffer_ns": 77625.0381023,
              "command_buffers": 2
            },
            {
              "compute_ns": 76916,
              "compute_span_ns": 76916,
              "calibration_cpu_ns": 526250,
              "calibration_gpu_ticks": 526250,
              "compute_passes": 1,
              "command_buffer_ns": 76999.9460317,
              "command_buffers": 2
            },
            {
              "compute_ns": 75750,
              "compute_span_ns": 75750,
              "calibration_cpu_ns": 504792,
              "calibration_gpu_ticks": 504792,
              "compute_passes": 1,
              "command_buffer_ns": 75750.0529289,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4891625.02345,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4881541.65214,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4938333.35349,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4914874.96765,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4872041.63801,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4898875.02044,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4869500.0005,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4788874.99776,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4848416.66184,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 77624.9798946,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76583.2955949,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76249.9985285,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 78375.0438131,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 94208.3424889,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 83583.2906887,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 77291.6828282,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 79708.348494,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 75666.6413508,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              76.43164099140625,
              76.2740883146875,
              77.16145864828124,
              76.79492136953125,
              76.12565059390624,
              76.544922194375,
              76.0859375078125,
              74.82617184,
              75.75651034125
            ],
            "command_buffer_throughput_us_p50": 76.2740883146875,
            "command_buffer_throughput_us_p90": 76.86822882528125,
            "command_buffer_latency_us": [
              77.62497989459999,
              76.58329559490001,
              76.2499985285,
              78.37504381309999,
              94.2083424889,
              83.5832906887,
              77.29168282820001,
              79.708348494,
              75.66664135079999
            ],
            "command_buffer_latency_us_p50": 77.62497989459999,
            "command_buffer_latency_us_p90": 85.70830104874001
          },
          "compute_throughput_us": [
            76.79296875,
            75.954421875,
            76.303390625,
            76.899734375,
            76.345046875,
            76.77603125,
            75.741546875,
            76.239578125,
            76.353515625
          ],
          "compute_throughput_us_p50": 76.345046875,
          "compute_throughput_us_p90": 76.814321875,
          "command_buffer_throughput_us": [
            76.79296868453126,
            75.95442730234376,
            76.30468826390626,
            76.90039001312499,
            76.3470052334375,
            76.77734265593749,
            75.74349001515624,
            76.2395839046875,
            76.3541665946875
          ],
          "command_buffer_throughput_us_p50": 76.3470052334375,
          "command_buffer_throughput_us_p90": 76.81445295025,
          "compute_latency_us": [
            77.209,
            76.667,
            77.25,
            77.208,
            85.375,
            85.75,
            77.625,
            76.916,
            75.75
          ],
          "compute_latency_us_p50": 77.209,
          "compute_latency_us_p90": 85.45,
          "command_buffer_latency_us": [
            77.29168282820001,
            76.666707173,
            77.33335951340001,
            77.29168282820001,
            85.4583340697,
            85.7916311361,
            77.6250381023,
            76.9999460317,
            75.75005292889999
          ],
          "command_buffer_latency_us_p50": 77.29168282820001,
          "command_buffer_latency_us_p90": 85.52499348298,
          "counter_control_throughput_ratio": 1.0029077536652145,
          "counter_control_latency_ratio": 1.0010891615129913
        },
        "correctness": {
          "max_abs_error": 2.827305296548843e-07,
          "atol": 1e-05,
          "rtol": 2e-05
        },
        "throughput_us_p50": 76.3707430341,
        "throughput_us_p90": 76.73555232198,
        "latency_us_p50": 319.958,
        "latency_us_p90": 362.5082
      },
      "slowdown": 0.37084244350056206,
      "device_compute_slowdown": 0.3819938237823469,
      "valid": true,
      "round": 3,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "softmax_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-0pg3lc43/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "d45e9e72129e03853b1c708b1bdcbc59be9547b0d0e02ca6bdc2eb0b41c08926",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.135458,
        "capture_ms": 0.07175,
        "compile_ms": 57.137,
        "allocation_upload_ms": 1.172167,
        "cold_call_ms": 1.9875,
        "warmup_ms": 200.03175,
        "download_ms": 0.288916,
        "repetitions": 4859,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          4.75689442272,
          4.60747252521,
          4.85538876312,
          4.92922062153,
          4.91179604857,
          4.97688145709,
          4.66138444124,
          4.76925108047,
          4.99569520477
        ],
        "latency_us": [
          221.125,
          184.792,
          197.167,
          201.417,
          312.208,
          198.667,
          224.917,
          222.792,
          207.166
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 19,
            "reduction_scalar_rounds": 40,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 6148,
              "global_write_bytes": 6148,
              "private_read_bytes": 36888,
              "private_write_bytes": 12296
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 32,
              "global_write_bytes": 32,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 2,
            "reduction_elements": 3074,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 80,
            "normalized_kernel_cost": 80,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 4611,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 313541,
              "compute_span_ns": 313541,
              "calibration_cpu_ns": 757125,
              "calibration_gpu_ticks": 757125,
              "compute_passes": 1,
              "command_buffer_ns": 313541.677315,
              "command_buffers": 2
            },
            {
              "compute_ns": 311208,
              "compute_span_ns": 311208,
              "calibration_cpu_ns": 752875,
              "calibration_gpu_ticks": 752875,
              "compute_passes": 1,
              "command_buffer_ns": 311249.925289,
              "command_buffers": 2
            },
            {
              "compute_ns": 317208,
              "compute_span_ns": 317208,
              "calibration_cpu_ns": 814750,
              "calibration_gpu_ticks": 814750,
              "compute_passes": 1,
              "command_buffer_ns": 317250.029184,
              "command_buffers": 2
            },
            {
              "compute_ns": 315583,
              "compute_span_ns": 315583,
              "calibration_cpu_ns": 702250,
              "calibration_gpu_ticks": 702250,
              "compute_passes": 1,
              "command_buffer_ns": 315583.311021,
              "command_buffers": 2
            },
            {
              "compute_ns": 319583,
              "compute_span_ns": 319583,
              "calibration_cpu_ns": 735500,
              "calibration_gpu_ticks": 735500,
              "compute_passes": 1,
              "command_buffer_ns": 319583.283272,
              "command_buffers": 2
            },
            {
              "compute_ns": 323125,
              "compute_span_ns": 323125,
              "calibration_cpu_ns": 754667,
              "calibration_gpu_ticks": 754667,
              "compute_passes": 1,
              "command_buffer_ns": 323166.663293,
              "command_buffers": 2
            },
            {
              "compute_ns": 300750,
              "compute_span_ns": 300750,
              "calibration_cpu_ns": 714792,
              "calibration_gpu_ticks": 714792,
              "compute_passes": 1,
              "command_buffer_ns": 300749.961752,
              "command_buffers": 2
            },
            {
              "compute_ns": 284917,
              "compute_span_ns": 284917,
              "calibration_cpu_ns": 674292,
              "calibration_gpu_ticks": 674292,
              "compute_passes": 1,
              "command_buffer_ns": 284958.339762,
              "command_buffers": 2
            },
            {
              "compute_ns": 312041,
              "compute_span_ns": 312041,
              "calibration_cpu_ns": 806583,
              "calibration_gpu_ticks": 806583,
              "compute_passes": 1,
              "command_buffer_ns": 312041.665893,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 7000,
              "compute_span_ns": 7000,
              "calibration_cpu_ns": 356458,
              "calibration_gpu_ticks": 356458,
              "compute_passes": 1,
              "command_buffer_ns": 6999.99509379,
              "command_buffers": 2
            },
            {
              "compute_ns": 6958,
              "compute_span_ns": 6958,
              "calibration_cpu_ns": 457083,
              "calibration_gpu_ticks": 457083,
              "compute_passes": 1,
              "command_buffer_ns": 6958.26020092,
              "command_buffers": 2
            },
            {
              "compute_ns": 7000,
              "compute_span_ns": 7000,
              "calibration_cpu_ns": 397667,
              "calibration_gpu_ticks": 397667,
              "compute_passes": 1,
              "command_buffer_ns": 7083.29025656,
              "command_buffers": 2
            },
            {
              "compute_ns": 6916,
              "compute_span_ns": 6916,
              "calibration_cpu_ns": 351250,
              "calibration_gpu_ticks": 351250,
              "compute_passes": 1,
              "command_buffer_ns": 6999.99509379,
              "command_buffers": 2
            },
            {
              "compute_ns": 6958,
              "compute_span_ns": 6958,
              "calibration_cpu_ns": 445416,
              "calibration_gpu_ticks": 445416,
              "compute_passes": 1,
              "command_buffer_ns": 6999.99509379,
              "command_buffers": 2
            },
            {
              "compute_ns": 9417,
              "compute_span_ns": 9417,
              "calibration_cpu_ns": 430333,
              "calibration_gpu_ticks": 430333,
              "compute_passes": 1,
              "command_buffer_ns": 9458.33744481,
              "command_buffers": 2
            },
            {
              "compute_ns": 7334,
              "compute_span_ns": 7334,
              "calibration_cpu_ns": 396208,
              "calibration_gpu_ticks": 396208,
              "compute_passes": 1,
              "command_buffer_ns": 7375.02705306,
              "command_buffers": 2
            },
            {
              "compute_ns": 10000,
              "compute_span_ns": 10000,
              "calibration_cpu_ns": 401834,
              "calibration_gpu_ticks": 401834,
              "compute_passes": 1,
              "command_buffer_ns": 10041.6364148,
              "command_buffers": 2
            },
            {
              "compute_ns": 7167,
              "compute_span_ns": 7167,
              "calibration_cpu_ns": 399375,
              "calibration_gpu_ticks": 399375,
              "compute_passes": 1,
              "command_buffer_ns": 7249.99699742,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 331500.021275,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 291666.656267,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 316250.021569,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 322124.978993,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 317208.352499,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 329333.299305,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 303166.685626,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 323124.986608,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 313624.972478,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 13416.6912176,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7125.02514943,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 9500.07233769,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7374.9688454,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7041.67177901,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6958.31840858,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7208.3203122,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7124.96694177,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7458.26400816,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              5.179687832421875,
              4.557291504171875,
              4.941406587015624,
              5.033202796765625,
              4.956380507796874,
              5.1458328016406245,
              4.73697946290625,
              5.0488279157500005,
              4.900390194968749
            ],
            "command_buffer_throughput_us_p50": 4.956380507796874,
            "command_buffer_throughput_us_p90": 5.152603807796875,
            "command_buffer_latency_us": [
              13.4166912176,
              7.12502514943,
              9.50007233769,
              7.3749688454,
              7.0416717790100005,
              6.95831840858,
              7.208320312200001,
              7.1249669417699995,
              7.4582640081600005
            ],
            "command_buffer_latency_us_p50": 7.208320312200001,
            "command_buffer_latency_us_p90": 10.283396113672001
          },
          "compute_throughput_us": [
            4.899078125,
            4.862625,
            4.956375,
            4.930984375,
            4.993484375,
            5.048828125,
            4.69921875,
            4.451828125,
            4.875640625
          ],
          "compute_throughput_us_p50": 4.899078125,
          "compute_throughput_us_p90": 5.004553125,
          "command_buffer_throughput_us": [
            4.899088708046874,
            4.863280082640625,
            4.9570317059999995,
            4.930989234703125,
            4.993488801124999,
            5.049479113953125,
            4.699218152375,
            4.452474058781251,
            4.875651029578125
          ],
          "command_buffer_throughput_us_p50": 4.899088708046874,
          "command_buffer_throughput_us_p90": 5.004686863690624,
          "compute_latency_us": [
            7.0,
            6.958,
            7.0,
            6.916,
            6.958,
            9.417,
            7.334,
            10.0,
            7.167
          ],
          "compute_latency_us_p50": 7.0,
          "compute_latency_us_p90": 9.5336,
          "command_buffer_latency_us": [
            6.99999509379,
            6.95826020092,
            7.08329025656,
            6.99999509379,
            6.99999509379,
            9.45833744481,
            7.37502705306,
            10.0416364148,
            7.24999699742
          ],
          "command_buffer_latency_us_p50": 7.08329025656,
          "command_buffer_latency_us_p90": 9.574997238808,
          "counter_control_throughput_ratio": 0.9920283989350288,
          "counter_control_latency_ratio": 0.9765944758071007
        },
        "correctness": {
          "max_abs_error": 2.6856349761260334e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 4.85538876312,
        "throughput_us_p90": 4.980644206626,
        "latency_us_p50": 207.166,
        "latency_us_p90": 242.3752
      },
      "torch": {
        "cold_call_ms": 7.507833,
        "warmup_ms": 200.306541,
        "repetitions": 832,
        "throughput_us": [
          32.42483052884615,
          30.351712740384617,
          29.26307091346154,
          29.647436298076926,
          30.316856971153847,
          29.60792307692308,
          29.903496394230768,
          29.372346153846156,
          29.801983173076923
        ],
        "latency_us": [
          309.042,
          296.5,
          310.333,
          297.125,
          301.16700000000003,
          316.458,
          306.66700000000003,
          362.25,
          317.83299999999997
        ],
        "throughput_us_p50": 29.801983173076923,
        "throughput_us_p90": 30.766336298076922,
        "latency_us_p50": 309.042,
        "latency_us_p90": 326.71639999999996,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1192583.2950510085,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1174583.3908207715,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1064624.9284036458,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1339583.3084359765,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1314041.786827147,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1346749.8938553035,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1345541.6192300618,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1341333.32176134,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1370041.7475774884,
                "command_buffers": 7
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 89499.98300522566,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 88875.00734999776,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70125.03920122981,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 93124.98150393367,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 88000.02979114652,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 104625.01086294651,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 103541.70808568597,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 89416.62963479757,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 88041.64826869965,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              18.634113985172007,
              18.352865481574554,
              16.634764506306965,
              20.930989194312133,
              20.531902919174172,
              21.042967091489118,
              21.024087800469715,
              20.958333152520936,
              21.406902305898257
            ],
            "command_buffer_throughput_us_p50": 20.930989194312133,
            "command_buffer_throughput_us_p90": 21.115754134370945,
            "command_buffer_latency_us": [
              89.49998300522566,
              88.87500734999776,
              70.12503920122981,
              93.12498150393367,
              88.00002979114652,
              104.62501086294651,
              103.54170808568597,
              89.41662963479757,
              88.04164826869965
            ],
            "command_buffer_latency_us_p50": 89.41662963479757,
            "command_buffer_latency_us_p90": 103.75836864113808
          },
          "throughput": [
            {
              "compute_ns": 3607287.0,
              "compute_span_ns": 4673875.0,
              "command_buffer_ns": 4577083.222102374,
              "calibration_cpu_ns": 5505500.0,
              "calibration_gpu_ticks": 5505500.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3940045.0,
              "compute_span_ns": 5059000.0,
              "command_buffer_ns": 4932874.930091202,
              "calibration_cpu_ns": 5727625.0,
              "calibration_gpu_ticks": 5727625.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4017045.0,
              "compute_span_ns": 5116000.0,
              "command_buffer_ns": 5005416.751373559,
              "calibration_cpu_ns": 5828791.0,
              "calibration_gpu_ticks": 5828791.0,
              "compute_passes": 64,
              "command_buffers": 8
            },
            {
              "compute_ns": 3576166.0,
              "compute_span_ns": 4686667.0,
              "command_buffer_ns": 4579625.034239143,
              "calibration_cpu_ns": 5540959.0,
              "calibration_gpu_ticks": 5540959.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4483878.0,
              "compute_span_ns": 5666125.0,
              "command_buffer_ns": 5566541.629377753,
              "calibration_cpu_ns": 8416250.0,
              "calibration_gpu_ticks": 8416250.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4190826.0,
              "compute_span_ns": 5395542.0,
              "command_buffer_ns": 5266208.318062127,
              "calibration_cpu_ns": 6231333.0,
              "calibration_gpu_ticks": 6231333.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4658835.0,
              "compute_span_ns": 5893000.0,
              "command_buffer_ns": 5774666.729848832,
              "calibration_cpu_ns": 6653708.0,
              "calibration_gpu_ticks": 6653708.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4819708.0,
              "compute_span_ns": 6036459.0,
              "command_buffer_ns": 5905166.67580232,
              "calibration_cpu_ns": 6843250.0,
              "calibration_gpu_ticks": 6843250.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 5093416.0,
              "compute_span_ns": 6284084.0,
              "command_buffer_ns": 6187375.052832067,
              "calibration_cpu_ns": 7177875.0,
              "calibration_gpu_ticks": 7177875.0,
              "compute_passes": 64,
              "command_buffers": 7
            }
          ],
          "latency": [
            {
              "compute_ns": 89500.0,
              "compute_span_ns": 89500.0,
              "command_buffer_ns": 89499.98300522566,
              "calibration_cpu_ns": 618625.0,
              "calibration_gpu_ticks": 618625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 88875.0,
              "compute_span_ns": 88875.0,
              "command_buffer_ns": 88875.00734999776,
              "calibration_cpu_ns": 555375.0,
              "calibration_gpu_ticks": 555375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 70459.0,
              "compute_span_ns": 70459.0,
              "command_buffer_ns": 70499.95474517345,
              "calibration_cpu_ns": 578417.0,
              "calibration_gpu_ticks": 578417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 103042.0,
              "compute_span_ns": 103042.0,
              "command_buffer_ns": 103083.32275599241,
              "calibration_cpu_ns": 2568292.0,
              "calibration_gpu_ticks": 2568292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 89375.0,
              "compute_span_ns": 89375.0,
              "command_buffer_ns": 89375.01115724444,
              "calibration_cpu_ns": 565875.0,
              "calibration_gpu_ticks": 565875.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 104792.0,
              "compute_span_ns": 104792.0,
              "command_buffer_ns": 104791.65939614177,
              "calibration_cpu_ns": 605708.0,
              "calibration_gpu_ticks": 605708.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 105250.0,
              "compute_span_ns": 105250.0,
              "command_buffer_ns": 105249.98651817441,
              "calibration_cpu_ns": 612416.0,
              "calibration_gpu_ticks": 612416.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 103459.0,
              "compute_span_ns": 103459.0,
              "command_buffer_ns": 103499.97319281101,
              "calibration_cpu_ns": 576917.0,
              "calibration_gpu_ticks": 576917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 137000.0,
              "compute_span_ns": 137000.0,
              "command_buffer_ns": 136999.99544769526,
              "calibration_cpu_ns": 703416.0,
              "calibration_gpu_ticks": 703416.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            56.363859375,
            61.563203125,
            62.766328125,
            55.87759375,
            70.06059375,
            65.48165625,
            72.794296875,
            75.3079375,
            79.584625
          ],
          "compute_throughput_us_p50": 65.48165625,
          "compute_throughput_us_p90": 76.163275,
          "command_buffer_throughput_us": [
            71.51692534534959,
            77.07617078267504,
            78.20963674021186,
            71.55664115998661,
            86.97721295902738,
            82.28450496972073,
            90.229167653888,
            92.26822930941125,
            96.67773520050105
          ],
          "command_buffer_throughput_us_p50": 82.28450496972073,
          "command_buffer_throughput_us_p90": 93.15013048762921,
          "compute_latency_us": [
            89.5,
            88.875,
            70.459,
            103.042,
            89.375,
            104.792,
            105.25,
            103.459,
            137.0
          ],
          "compute_latency_us_p50": 103.042,
          "compute_latency_us_p90": 111.60000000000001,
          "command_buffer_latency_us": [
            89.49998300522566,
            88.87500734999776,
            70.49995474517345,
            103.08332275599241,
            89.37501115724444,
            104.79165939614177,
            105.24998651817441,
            103.49997319281101,
            136.99999544769526
          ],
          "command_buffer_latency_us_p50": 103.08332275599241,
          "command_buffer_latency_us_p90": 111.59998830407858,
          "counter_control_throughput_ratio": 4.236198334924026,
          "counter_control_latency_ratio": 1.015624782961565
        },
        "download_ms": 0.299666,
        "allocation_upload_ms": 0.682583,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 4.2122453832241225e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.16292166648514694,
      "device_compute_slowdown": 0.07481603865204615,
      "valid": true,
      "round": 3,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 37,
        "n": 1537,
        "k": 1
      },
      "name": "softmax_37x1537",
      "backend": "metal",
      "block": [
        1,
        1537,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.563708,
        "warmup_ms": 200.454708,
        "repetitions": 880,
        "throughput_us": [
          29.753361363636362,
          29.053503409090908,
          29.437073863636364,
          29.496401136363634,
          29.49715909090909,
          29.2725375,
          29.26998068181818,
          29.29962159090909,
          29.204877272727273
        ],
        "latency_us": [
          327.0,
          299.833,
          310.333,
          313.58399999999995,
          292.292,
          309.125,
          307.70799999999997,
          298.79200000000003,
          332.792
        ],
        "throughput_us_p50": 29.29962159090909,
        "throughput_us_p90": 29.548399545454544,
        "latency_us_p50": 309.125,
        "latency_us_p90": 328.1584,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1062624.97138232,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1064333.3080224693,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1074541.7093858123,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1061791.7287163436,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1104583.265259862,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1060541.660990566,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1217458.3389423788,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1214666.641317308,
                "command_buffers": 7
              },
              {
                "command_buffer_ns": 1063541.6838340461,
                "command_buffers": 7
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 62583.30540731549,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 71500.02056732774,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 76125.02668052912,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 74749.98710677028,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 58458.36130902171,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 70750.01485645771,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 75791.67140647769,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 72875.00193342566,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 76500.00043213367,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              16.60351517784875,
              16.630207937851083,
              16.789714209153317,
              16.59049576119287,
              17.259113519685343,
              16.570963452977594,
              19.022786545974668,
              18.979166270582937,
              16.61783880990697
            ],
            "command_buffer_throughput_us_p50": 16.630207937851083,
            "command_buffer_throughput_us_p90": 18.987890325661283,
            "command_buffer_latency_us": [
              62.58330540731549,
              71.50002056732774,
              76.12502668052912,
              74.74998710677028,
              58.45836130902171,
              70.75001485645771,
              75.79167140647769,
              72.87500193342566,
              76.50000043213367
            ],
            "command_buffer_latency_us_p50": 72.87500193342566,
            "command_buffer_latency_us_p90": 76.20002143085003
          },
          "throughput": [
            {
              "compute_ns": 3482877.0,
              "compute_span_ns": 4522458.0,
              "command_buffer_ns": 4414041.701238602,
              "calibration_cpu_ns": 5237666.0,
              "calibration_gpu_ticks": 5237666.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3475879.0,
              "compute_span_ns": 4564000.0,
              "command_buffer_ns": 4464083.234779537,
              "calibration_cpu_ns": 5317250.0,
              "calibration_gpu_ticks": 5317250.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3520330.0,
              "compute_span_ns": 4448916.0,
              "command_buffer_ns": 4379791.673272848,
              "calibration_cpu_ns": 5119666.0,
              "calibration_gpu_ticks": 5119666.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3738581.0,
              "compute_span_ns": 4820750.0,
              "command_buffer_ns": 4727750.027086586,
              "calibration_cpu_ns": 5496500.0,
              "calibration_gpu_ticks": 5496500.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3696453.0,
              "compute_span_ns": 4849833.0,
              "command_buffer_ns": 4731833.2944996655,
              "calibration_cpu_ns": 5521916.0,
              "calibration_gpu_ticks": 5521916.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3920833.0,
              "compute_span_ns": 5001167.0,
              "command_buffer_ns": 4903833.323623985,
              "calibration_cpu_ns": 5689959.0,
              "calibration_gpu_ticks": 5689959.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3513374.0,
              "compute_span_ns": 4621917.0,
              "command_buffer_ns": 4458833.253011107,
              "calibration_cpu_ns": 5555042.0,
              "calibration_gpu_ticks": 5555042.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 3826292.0,
              "compute_span_ns": 4984834.0,
              "command_buffer_ns": 4876291.612163186,
              "calibration_cpu_ns": 5797833.0,
              "calibration_gpu_ticks": 5797833.0,
              "compute_passes": 64,
              "command_buffers": 7
            },
            {
              "compute_ns": 4423457.0,
              "compute_span_ns": 5552833.0,
              "command_buffer_ns": 5446708.470117301,
              "calibration_cpu_ns": 6247167.0,
              "calibration_gpu_ticks": 6247167.0,
              "compute_passes": 64,
              "command_buffers": 7
            }
          ],
          "latency": [
            {
              "compute_ns": 57875.0,
              "compute_span_ns": 57875.0,
              "command_buffer_ns": 57874.94592368603,
              "calibration_cpu_ns": 531917.0,
              "calibration_gpu_ticks": 531917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 62667.0,
              "compute_span_ns": 62667.0,
              "command_buffer_ns": 62666.65877774358,
              "calibration_cpu_ns": 533167.0,
              "calibration_gpu_ticks": 533167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 76625.0,
              "compute_span_ns": 76625.0,
              "command_buffer_ns": 76624.97228011489,
              "calibration_cpu_ns": 532500.0,
              "calibration_gpu_ticks": 532500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 77333.0,
              "compute_span_ns": 77333.0,
              "command_buffer_ns": 77333.30130577087,
              "calibration_cpu_ns": 531333.0,
              "calibration_gpu_ticks": 531333.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 56875.0,
              "compute_span_ns": 56875.0,
              "command_buffer_ns": 56874.99651685357,
              "calibration_cpu_ns": 473000.0,
              "calibration_gpu_ticks": 473000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 63417.0,
              "compute_span_ns": 63417.0,
              "command_buffer_ns": 63416.664488613605,
              "calibration_cpu_ns": 508583.0,
              "calibration_gpu_ticks": 508583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 76416.0,
              "compute_span_ns": 76416.0,
              "command_buffer_ns": 76374.97037649155,
              "calibration_cpu_ns": 550416.0,
              "calibration_gpu_ticks": 550416.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 78458.0,
              "compute_span_ns": 78458.0,
              "command_buffer_ns": 78458.28076824546,
              "calibration_cpu_ns": 539000.0,
              "calibration_gpu_ticks": 539000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 74333.0,
              "compute_span_ns": 74333.0,
              "command_buffer_ns": 74291.65998473763,
              "calibration_cpu_ns": 520500.0,
              "calibration_gpu_ticks": 520500.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            54.419953125,
            54.310609375,
            55.00515625,
            58.415328125,
            57.757078125,
            61.263015625,
            54.89646875,
            59.7858125,
            69.116515625
          ],
          "compute_throughput_us_p50": 57.757078125,
          "compute_throughput_us_p90": 62.833715625,
          "command_buffer_throughput_us": [
            68.96940158185316,
            69.75130054343026,
            68.43424489488825,
            73.8710941732279,
            73.93489522655727,
            76.62239568162477,
            69.66926957829855,
            76.19205644004978,
            85.10481984558282
          ],
          "command_buffer_throughput_us_p50": 73.8710941732279,
          "command_buffer_throughput_us_p90": 78.31888051441638,
          "compute_latency_us": [
            57.875,
            62.667,
            76.625,
            77.333,
            56.875,
            63.417,
            76.416,
            78.458,
            74.333
          ],
          "compute_latency_us_p50": 74.333,
          "compute_latency_us_p90": 77.55799999999999,
          "command_buffer_latency_us": [
            57.87494592368603,
            62.66665877774358,
            76.62497228011489,
            77.33330130577087,
            56.87499651685357,
            63.416664488613605,
            76.37497037649155,
            78.45828076824546,
            74.29165998473763
          ],
          "command_buffer_latency_us_p50": 74.29165998473763,
          "command_buffer_latency_us_p90": 77.55829719826579,
          "counter_control_throughput_ratio": 4.194253060701258,
          "counter_control_latency_ratio": 0.9729146565741352
        },
        "download_ms": 0.281833,
        "allocation_upload_ms": 0.711125,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 4.2122453832241225e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "37",
        "1537",
        "1",
        "1",
        "1537",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-4ti1myrn/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "1ca0ba2a59ee9b200a0f7933e3c823dd873803c26a844f54bebdc043f01bc250",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.376417,
        "capture_ms": 0.0685,
        "compile_ms": 49.153916,
        "allocation_upload_ms": 1.255417,
        "cold_call_ms": 1.855125,
        "warmup_ms": 200.28775,
        "download_ms": 0.309125,
        "repetitions": 5566,
        "output_elements": 56869,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          4.78623194395,
          3.98021469637,
          3.95511444484,
          4.07747933884,
          4.73653287819,
          4.76630434783,
          4.52477093065,
          3.97143370464,
          3.98489346029
        ],
        "latency_us": [
          312.417,
          216.083,
          241.333,
          270.083,
          185.5,
          190.833,
          208.5,
          186.75,
          188.542
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 37,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 37,
            "reduction_scalar_rounds": 40,
            "reduction_lane_utilization": 0.75048828125,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 6148,
              "global_write_bytes": 6148,
              "private_read_bytes": 36888,
              "private_write_bytes": 12296
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 32,
              "global_write_bytes": 32,
              "private_read_bytes": 192,
              "private_write_bytes": 64
            },
            "striped_storage_scalars_per_worker": 16,
            "reduction_operations": 2,
            "reduction_elements": 3074,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 88,
            "normalized_kernel_cost": 88,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 4611,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 1,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 261459,
              "compute_span_ns": 261459,
              "calibration_cpu_ns": 813541,
              "calibration_gpu_ticks": 813541,
              "compute_passes": 1,
              "command_buffer_ns": 261583.307292,
              "command_buffers": 2
            },
            {
              "compute_ns": 243250,
              "compute_span_ns": 243250,
              "calibration_cpu_ns": 707208,
              "calibration_gpu_ticks": 707208,
              "compute_passes": 1,
              "command_buffer_ns": 243333.284743,
              "command_buffers": 2
            },
            {
              "compute_ns": 262250,
              "compute_span_ns": 262250,
              "calibration_cpu_ns": 759958,
              "calibration_gpu_ticks": 759958,
              "compute_passes": 1,
              "command_buffer_ns": 262374.989688,
              "command_buffers": 2
            },
            {
              "compute_ns": 281500,
              "compute_span_ns": 281500,
              "calibration_cpu_ns": 789417,
              "calibration_gpu_ticks": 789417,
              "compute_passes": 1,
              "command_buffer_ns": 281624.961644,
              "command_buffers": 2
            },
            {
              "compute_ns": 261750,
              "compute_span_ns": 261750,
              "calibration_cpu_ns": 685667,
              "calibration_gpu_ticks": 685667,
              "compute_passes": 1,
              "command_buffer_ns": 261791.690718,
              "command_buffers": 2
            },
            {
              "compute_ns": 261625,
              "compute_span_ns": 261625,
              "calibration_cpu_ns": 683084,
              "calibration_gpu_ticks": 683084,
              "compute_passes": 1,
              "command_buffer_ns": 261666.71887,
              "command_buffers": 2
            },
            {
              "compute_ns": 243959,
              "compute_span_ns": 243959,
              "calibration_cpu_ns": 703291,
              "calibration_gpu_ticks": 703291,
              "compute_passes": 1,
              "command_buffer_ns": 243999.995291,
              "command_buffers": 2
            },
            {
              "compute_ns": 261750,
              "compute_span_ns": 261750,
              "calibration_cpu_ns": 681041,
              "calibration_gpu_ticks": 681041,
              "compute_passes": 1,
              "command_buffer_ns": 261916.662566,
              "command_buffers": 2
            },
            {
              "compute_ns": 262416,
              "compute_span_ns": 262416,
              "calibration_cpu_ns": 704375,
              "calibration_gpu_ticks": 704375,
              "compute_passes": 1,
              "command_buffer_ns": 262458.343059,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 6000,
              "compute_span_ns": 6000,
              "calibration_cpu_ns": 386500,
              "calibration_gpu_ticks": 386500,
              "compute_passes": 1,
              "command_buffer_ns": 6083.39905739,
              "command_buffers": 2
            },
            {
              "compute_ns": 6500,
              "compute_span_ns": 6500,
              "calibration_cpu_ns": 389416,
              "calibration_gpu_ticks": 389416,
              "compute_passes": 1,
              "command_buffer_ns": 6583.28644931,
              "command_buffers": 2
            },
            {
              "compute_ns": 6375,
              "compute_span_ns": 6375,
              "calibration_cpu_ns": 553750,
              "calibration_gpu_ticks": 553750,
              "compute_passes": 1,
              "command_buffer_ns": 6458.31460133,
              "command_buffers": 2
            },
            {
              "compute_ns": 6459,
              "compute_span_ns": 6459,
              "calibration_cpu_ns": 415667,
              "calibration_gpu_ticks": 415667,
              "compute_passes": 1,
              "command_buffer_ns": 6500.04949421,
              "command_buffers": 2
            },
            {
              "compute_ns": 6459,
              "compute_span_ns": 6459,
              "calibration_cpu_ns": 391333,
              "calibration_gpu_ticks": 391333,
              "compute_passes": 1,
              "command_buffer_ns": 6625.02134219,
              "command_buffers": 2
            },
            {
              "compute_ns": 6417,
              "compute_span_ns": 6417,
              "calibration_cpu_ns": 410000,
              "calibration_gpu_ticks": 410000,
              "compute_passes": 1,
              "command_buffer_ns": 6500.04949421,
              "command_buffers": 2
            },
            {
              "compute_ns": 6458,
              "compute_span_ns": 6458,
              "calibration_cpu_ns": 401833,
              "calibration_gpu_ticks": 401833,
              "compute_passes": 1,
              "command_buffer_ns": 6500.04949421,
              "command_buffers": 2
            },
            {
              "compute_ns": 6459,
              "compute_span_ns": 6459,
              "calibration_cpu_ns": 401250,
              "calibration_gpu_ticks": 401250,
              "compute_passes": 1,
              "command_buffer_ns": 6499.99128655,
              "command_buffers": 2
            },
            {
              "compute_ns": 6458,
              "compute_span_ns": 6458,
              "calibration_cpu_ns": 404333,
              "calibration_gpu_ticks": 404333,
              "compute_passes": 1,
              "command_buffer_ns": 6500.04949421,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 265458.307695,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 242625.013925,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 261708.337348,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 243375.019636,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 242875.015829,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 262625.0498,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 244125.025347,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 261500.01213,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 262000.015937,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 6083.39905739,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6625.02134219,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6541.66797176,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6375.01943856,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6583.34465697,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 7083.34846422,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6499.99128655,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6458.31460133,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 6541.66797176,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              4.1477860577343755,
              3.7910158425781253,
              4.0891927710625,
              3.8027346818125003,
              3.794922122328125,
              4.103516403125,
              3.814453521046875,
              4.08593768953125,
              4.093750249015625
            ],
            "command_buffer_throughput_us_p50": 4.08593768953125,
            "command_buffer_throughput_us_p90": 4.112370334046875,
            "command_buffer_latency_us": [
              6.08339905739,
              6.62502134219,
              6.54166797176,
              6.37501943856,
              6.58334465697,
              7.08334846422,
              6.49999128655,
              6.458314601330001,
              6.54166797176
            ],
            "command_buffer_latency_us_p50": 6.54166797176,
            "command_buffer_latency_us_p90": 6.716686766596
          },
          "compute_throughput_us": [
            4.085296875,
            3.80078125,
            4.09765625,
            4.3984375,
            4.08984375,
            4.087890625,
            3.811859375,
            4.08984375,
            4.10025
          ],
          "compute_throughput_us_p50": 4.08984375,
          "compute_throughput_us_p90": 4.1598875,
          "command_buffer_throughput_us": [
            4.0872391764375005,
            3.802082574109375,
            4.099609213875,
            4.400390025687501,
            4.09049516746875,
            4.08854248234375,
            3.812499926421875,
            4.09244785259375,
            4.100911610296874
          ],
          "command_buffer_throughput_us_p50": 4.09049516746875,
          "command_buffer_throughput_us_p90": 4.160807293375,
          "compute_latency_us": [
            6.0,
            6.5,
            6.375,
            6.459,
            6.459,
            6.417,
            6.458,
            6.459,
            6.458
          ],
          "compute_latency_us_p50": 6.458,
          "compute_latency_us_p90": 6.4672,
          "command_buffer_latency_us": [
            6.08339905739,
            6.58328644931,
            6.458314601330001,
            6.500049494210001,
            6.62502134219,
            6.500049494210001,
            6.500049494210001,
            6.49999128655,
            6.500049494210001
          ],
          "command_buffer_latency_us_p50": 6.500049494210001,
          "command_buffer_latency_us_p90": 6.591633427886,
          "counter_control_throughput_ratio": 1.001749340053896,
          "counter_control_latency_ratio": 1.0
        },
        "correctness": {
          "max_abs_error": 2.6856349761260334e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 4.07747933884,
        "throughput_us_p90": 4.770289867054,
        "latency_us_p50": 208.5,
        "latency_us_p90": 278.5498
      },
      "slowdown": 0.13916491467948294,
      "device_compute_slowdown": 0.07081112623371648,
      "valid": true,
      "round": 3,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "softmax_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-72qcqc8b/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "0d8a82e58cbe97c43d15ef559cf043ad575745e0d3fb0f4f40ed60b2caca21c0",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.096375,
        "capture_ms": 0.070958,
        "compile_ms": 49.412584,
        "allocation_upload_ms": 1.847041,
        "cold_call_ms": 5.445458,
        "warmup_ms": 200.025292,
        "download_ms": 0.905875,
        "repetitions": 1518,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          13.8873247694,
          14.2367147563,
          14.2653162055,
          13.6109466403,
          14.3277068511,
          14.6312312253,
          13.9863306983,
          13.5355184453,
          14.3244400527
        ],
        "latency_us": [
          268.334,
          276.5,
          246.958,
          245.708,
          284.792,
          257.75,
          254.917,
          251.25,
          263.166
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 256,
            "reduction_scalar_rounds": 60,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 12288,
              "global_write_bytes": 12288,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 48,
              "global_write_bytes": 48,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 2,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 108,
            "normalized_kernel_cost": 432,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 9216,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 899583,
              "compute_span_ns": 899583,
              "calibration_cpu_ns": 1367625,
              "calibration_gpu_ticks": 1367625,
              "compute_passes": 1,
              "command_buffer_ns": 899583.334103,
              "command_buffers": 2
            },
            {
              "compute_ns": 890667,
              "compute_span_ns": 890667,
              "calibration_cpu_ns": 1327833,
              "calibration_gpu_ticks": 1327833,
              "compute_passes": 1,
              "command_buffer_ns": 890833.325684,
              "command_buffers": 2
            },
            {
              "compute_ns": 924542,
              "compute_span_ns": 924542,
              "calibration_cpu_ns": 1387708,
              "calibration_gpu_ticks": 1387708,
              "compute_passes": 1,
              "command_buffer_ns": 924624.96832,
              "command_buffers": 2
            },
            {
              "compute_ns": 861167,
              "compute_span_ns": 861167,
              "calibration_cpu_ns": 1287709,
              "calibration_gpu_ticks": 1287709,
              "compute_passes": 1,
              "command_buffer_ns": 861291.71541,
              "command_buffers": 2
            },
            {
              "compute_ns": 884375,
              "compute_span_ns": 884375,
              "calibration_cpu_ns": 1356208,
              "calibration_gpu_ticks": 1356208,
              "compute_passes": 1,
              "command_buffer_ns": 884416.62956,
              "command_buffers": 2
            },
            {
              "compute_ns": 879041,
              "compute_span_ns": 879041,
              "calibration_cpu_ns": 1336084,
              "calibration_gpu_ticks": 1336084,
              "compute_passes": 1,
              "command_buffer_ns": 879083.352629,
              "command_buffers": 2
            },
            {
              "compute_ns": 918334,
              "compute_span_ns": 918334,
              "calibration_cpu_ns": 1312083,
              "calibration_gpu_ticks": 1312083,
              "compute_passes": 1,
              "command_buffer_ns": 918416.655622,
              "command_buffers": 2
            },
            {
              "compute_ns": 893916,
              "compute_span_ns": 893916,
              "calibration_cpu_ns": 1352417,
              "calibration_gpu_ticks": 1352417,
              "compute_passes": 1,
              "command_buffer_ns": 893999.997061,
              "command_buffers": 2
            },
            {
              "compute_ns": 882834,
              "compute_span_ns": 882834,
              "calibration_cpu_ns": 1290375,
              "calibration_gpu_ticks": 1290375,
              "compute_passes": 1,
              "command_buffer_ns": 882999.971509,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 15709,
              "compute_span_ns": 15709,
              "calibration_cpu_ns": 401208,
              "calibration_gpu_ticks": 401208,
              "compute_passes": 1,
              "command_buffer_ns": 15750.0035129,
              "command_buffers": 2
            },
            {
              "compute_ns": 18250,
              "compute_span_ns": 18250,
              "calibration_cpu_ns": 444250,
              "calibration_gpu_ticks": 444250,
              "compute_passes": 1,
              "command_buffer_ns": 18333.2595043,
              "command_buffers": 2
            },
            {
              "compute_ns": 18375,
              "compute_span_ns": 18375,
              "calibration_cpu_ns": 446667,
              "calibration_gpu_ticks": 446667,
              "compute_passes": 1,
              "command_buffer_ns": 18416.72929,
              "command_buffers": 2
            },
            {
              "compute_ns": 18292,
              "compute_span_ns": 18292,
              "calibration_cpu_ns": 423000,
              "calibration_gpu_ticks": 423000,
              "compute_passes": 1,
              "command_buffer_ns": 18375.0526048,
              "command_buffers": 2
            },
            {
              "compute_ns": 18500,
              "compute_span_ns": 18500,
              "calibration_cpu_ns": 508750,
              "calibration_gpu_ticks": 508750,
              "compute_passes": 1,
              "command_buffer_ns": 18583.3778232,
              "command_buffers": 2
            },
            {
              "compute_ns": 17917,
              "compute_span_ns": 17917,
              "calibration_cpu_ns": 426959,
              "calibration_gpu_ticks": 426959,
              "compute_passes": 1,
              "command_buffer_ns": 17999.9624379,
              "command_buffers": 2
            },
            {
              "compute_ns": 18250,
              "compute_span_ns": 18250,
              "calibration_cpu_ns": 424375,
              "calibration_gpu_ticks": 424375,
              "compute_passes": 1,
              "command_buffer_ns": 18375.0526048,
              "command_buffers": 2
            },
            {
              "compute_ns": 18541,
              "compute_span_ns": 18541,
              "calibration_cpu_ns": 429458,
              "calibration_gpu_ticks": 429458,
              "compute_passes": 1,
              "command_buffer_ns": 18541.6429304,
              "command_buffers": 2
            },
            {
              "compute_ns": 18167,
              "compute_span_ns": 18167,
              "calibration_cpu_ns": 492792,
              "calibration_gpu_ticks": 492792,
              "compute_passes": 1,
              "command_buffer_ns": 18166.6691788,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 950458.343141,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 986124.97095,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 926833.308768,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 931749.99347,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 879750.00497,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 943000.020925,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 910791.684873,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 957250.071224,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 908958.318178,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 15583.3549798,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18458.2895599,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18416.6710824,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18499.9662451,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 21125.015337,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18083.3158083,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18208.345864,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18291.6410267,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18333.3759196,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              14.850911611578125,
              15.40820267109375,
              14.481770449499999,
              14.558593647968749,
              13.74609382765625,
              14.734375326953124,
              14.231120076140625,
              14.957032362875,
              14.20247372153125
            ],
            "command_buffer_throughput_us_p50": 14.558593647968749,
            "command_buffer_throughput_us_p90": 15.04726642451875,
            "command_buffer_latency_us": [
              15.583354979800001,
              18.4582895599,
              18.4166710824,
              18.4999662451,
              21.125015337,
              18.0833158083,
              18.208345863999998,
              18.2916410267,
              18.333375919599998
            ],
            "command_buffer_latency_us_p50": 18.333375919599998,
            "command_buffer_latency_us_p90": 19.02497606348
          },
          "compute_throughput_us": [
            14.055984375,
            13.916671875,
            14.44596875,
            13.455734375,
            13.818359375,
            13.735015625,
            14.34896875,
            13.9674375,
            13.79428125
          ],
          "compute_throughput_us_p50": 13.916671875,
          "compute_throughput_us_p90": 14.36836875,
          "command_buffer_throughput_us": [
            14.055989595359375,
            13.9192707138125,
            14.44726513,
            13.457683053281249,
            13.819009836875,
            13.735677384828124,
            14.35026024409375,
            13.968749954078124,
            13.796874554828126
          ],
          "command_buffer_throughput_us_p50": 13.9192707138125,
          "command_buffer_throughput_us_p90": 14.369661221275,
          "compute_latency_us": [
            15.709,
            18.25,
            18.375,
            18.292,
            18.5,
            17.917,
            18.25,
            18.541,
            18.167
          ],
          "compute_latency_us_p50": 18.25,
          "compute_latency_us_p90": 18.5082,
          "command_buffer_latency_us": [
            15.750003512900001,
            18.3332595043,
            18.41672929,
            18.3750526048,
            18.5833778232,
            17.999962437900002,
            18.3750526048,
            18.541642930400002,
            18.1666691788
          ],
          "command_buffer_latency_us_p50": 18.3750526048,
          "command_buffer_latency_us_p90": 18.54998990896,
          "counter_control_throughput_ratio": 0.9464731838011203,
          "counter_control_latency_ratio": 0.9953905925614736
        },
        "correctness": {
          "max_abs_error": 1.569250362249905e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 14.2367147563,
        "throughput_us_p90": 14.388411725940001,
        "latency_us_p50": 257.75,
        "latency_us_p90": 278.1584
      },
      "torch": {
        "cold_call_ms": 6.833083,
        "warmup_ms": 200.488042,
        "repetitions": 812,
        "throughput_us": [
          33.462182266009854,
          33.57953694581281,
          33.50307881773399,
          33.37843842364532,
          35.06291009852217,
          33.71177093596059,
          33.23286083743842,
          33.16666748768473,
          33.44227216748769
        ],
        "latency_us": [
          312.042,
          277.95799999999997,
          278.916,
          292.208,
          276.625,
          279.667,
          282.58299999999997,
          294.75,
          273.084
        ],
        "throughput_us_p50": 33.462182266009854,
        "throughput_us_p90": 33.981998768472906,
        "latency_us_p50": 279.667,
        "latency_us_p90": 298.2084,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1868250.0813156366,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1879333.40203017,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1824749.9829158187,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1831666.6828468442,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1820666.8901257217,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1851374.9819248915,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1830374.8802281916,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1835583.4763497114,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1824791.6596010327,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 46458.328142762184,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 47874.986194074154,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 48291.69483855367,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 48250.018153339624,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 47874.986194074154,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 47624.98429045081,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 47666.660975664854,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 47708.3376608789,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 47750.01434609294,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              29.191407520556822,
              29.364584406721406,
              28.511718483059667,
              28.61979191948194,
              28.4479201582144,
              28.92773409257643,
              28.599607503565494,
              28.68099181796424,
              28.512369681266136
            ],
            "command_buffer_throughput_us_p50": 28.61979191948194,
            "command_buffer_throughput_us_p90": 29.22604289778974,
            "command_buffer_latency_us": [
              46.458328142762184,
              47.874986194074154,
              48.29169483855367,
              48.250018153339624,
              47.874986194074154,
              47.62498429045081,
              47.666660975664854,
              47.7083376608789,
              47.75001434609294
            ],
            "command_buffer_latency_us_p50": 47.75001434609294,
            "command_buffer_latency_us_p90": 48.25835349038243
          },
          "throughput": [
            {
              "compute_ns": 3017500.0,
              "compute_span_ns": 4019875.0,
              "command_buffer_ns": 3831708.221696317,
              "calibration_cpu_ns": 4639625.0,
              "calibration_gpu_ticks": 4639625.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3107375.0,
              "compute_span_ns": 3919250.0,
              "command_buffer_ns": 3786083.310842514,
              "calibration_cpu_ns": 4550000.0,
              "calibration_gpu_ticks": 4550000.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2989836.0,
              "compute_span_ns": 3880250.0,
              "command_buffer_ns": 3737249.9937191606,
              "calibration_cpu_ns": 4561708.0,
              "calibration_gpu_ticks": 4561708.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2978620.0,
              "compute_span_ns": 3866792.0,
              "command_buffer_ns": 3724833.3101160824,
              "calibration_cpu_ns": 4572917.0,
              "calibration_gpu_ticks": 4572917.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2993585.0,
              "compute_span_ns": 3949083.0,
              "command_buffer_ns": 3819624.9515749514,
              "calibration_cpu_ns": 4623875.0,
              "calibration_gpu_ticks": 4623875.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2985003.0,
              "compute_span_ns": 4101834.0,
              "command_buffer_ns": 3976499.95284155,
              "calibration_cpu_ns": 4801375.0,
              "calibration_gpu_ticks": 4801375.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2990624.0,
              "compute_span_ns": 3892042.0,
              "command_buffer_ns": 3726791.706867516,
              "calibration_cpu_ns": 4573625.0,
              "calibration_gpu_ticks": 4573625.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2981289.0,
              "compute_span_ns": 3846542.0,
              "command_buffer_ns": 3693916.7184755206,
              "calibration_cpu_ns": 4600209.0,
              "calibration_gpu_ticks": 4600209.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2994085.0,
              "compute_span_ns": 3907792.0,
              "command_buffer_ns": 3745458.205230534,
              "calibration_cpu_ns": 4591792.0,
              "calibration_gpu_ticks": 4591792.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 47667.0,
              "compute_span_ns": 47667.0,
              "command_buffer_ns": 47666.660975664854,
              "calibration_cpu_ns": 558791.0,
              "calibration_gpu_ticks": 558791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47625.0,
              "compute_span_ns": 47625.0,
              "command_buffer_ns": 47624.98429045081,
              "calibration_cpu_ns": 493584.0,
              "calibration_gpu_ticks": 493584.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47917.0,
              "compute_span_ns": 47917.0,
              "command_buffer_ns": 47958.33956450224,
              "calibration_cpu_ns": 538833.0,
              "calibration_gpu_ticks": 538833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47542.0,
              "compute_span_ns": 47542.0,
              "command_buffer_ns": 47541.68912768364,
              "calibration_cpu_ns": 487833.0,
              "calibration_gpu_ticks": 487833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 48250.0,
              "compute_span_ns": 48250.0,
              "command_buffer_ns": 48250.018153339624,
              "calibration_cpu_ns": 580541.0,
              "calibration_gpu_ticks": 580541.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47916.0,
              "compute_span_ns": 47916.0,
              "command_buffer_ns": 47874.986194074154,
              "calibration_cpu_ns": 508750.0,
              "calibration_gpu_ticks": 508750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47917.0,
              "compute_span_ns": 47917.0,
              "command_buffer_ns": 47958.33956450224,
              "calibration_cpu_ns": 568375.0,
              "calibration_gpu_ticks": 568375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47750.0,
              "compute_span_ns": 47750.0,
              "command_buffer_ns": 47749.956138432026,
              "calibration_cpu_ns": 514833.0,
              "calibration_gpu_ticks": 514833.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 48125.0,
              "compute_span_ns": 48125.0,
              "command_buffer_ns": 48124.9880976975,
              "calibration_cpu_ns": 601125.0,
              "calibration_gpu_ticks": 601125.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            47.1484375,
            48.552734375,
            46.7161875,
            46.5409375,
            46.774765625,
            46.640671875,
            46.7285,
            46.582640625,
            46.782578125
          ],
          "compute_throughput_us_p50": 46.7285,
          "compute_throughput_us_p90": 47.429296875,
          "command_buffer_throughput_us": [
            59.870440964004956,
            59.15755173191428,
            58.394531151861884,
            58.20052047056379,
            59.681639868358616,
            62.13281176314922,
            58.23112041980494,
            57.71744872618001,
            58.522784456727095
          ],
          "command_buffer_throughput_us_p50": 58.522784456727095,
          "command_buffer_throughput_us_p90": 60.32291512383381,
          "compute_latency_us": [
            47.667,
            47.625,
            47.917,
            47.542,
            48.25,
            47.916,
            47.917,
            47.75,
            48.125
          ],
          "compute_latency_us_p50": 47.916,
          "compute_latency_us_p90": 48.15,
          "command_buffer_latency_us": [
            47.666660975664854,
            47.62498429045081,
            47.95833956450224,
            47.54168912768364,
            48.250018153339624,
            47.874986194074154,
            47.95833956450224,
            47.749956138432026,
            48.124988097697496
          ],
          "command_buffer_latency_us_p50": 47.874986194074154,
          "command_buffer_latency_us_p90": 48.14999410882592,
          "counter_control_throughput_ratio": 2.048088795017992,
          "counter_control_latency_ratio": 1.0052493855346734
        },
        "download_ms": 0.35975,
        "allocation_upload_ms": 1.091875,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 2.673614807194649e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.4254568528473213,
      "device_compute_slowdown": 0.29781978610483967,
      "valid": true,
      "round": 3,
      "variant": "reference"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 256,
        "n": 3072,
        "k": 1
      },
      "name": "softmax_256x3072",
      "backend": "metal",
      "block": [
        1,
        3072,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.550875,
        "warmup_ms": 200.003292,
        "repetitions": 630,
        "throughput_us": [
          42.45615079365079,
          33.654695238095236,
          33.79358412698412,
          33.505290476190474,
          33.494509523809526,
          33.4365746031746,
          33.51488095238095,
          37.88439206349206,
          33.851257142857136
        ],
        "latency_us": [
          318.20799999999997,
          413.208,
          314.167,
          297.08299999999997,
          296.916,
          440.70799999999997,
          296.125,
          310.66700000000003,
          273.666
        ],
        "throughput_us_p50": 33.654695238095236,
        "throughput_us_p90": 38.798743809523806,
        "latency_us_p50": 310.66700000000003,
        "latency_us_p90": 418.708,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 1892208.4709629416,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1859083.2478366792,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1825416.6352562606,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1829291.5192432702,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1817416.7489632964,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1826374.907977879,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1827208.383474499,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1830208.3481103182,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 1850124.9724067748,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 48874.993808567524,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 47416.65907204151,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 48666.668590158224,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 47833.367716521025,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 51333.33615958691,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 49624.99951943755,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 48749.963752925396,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 47374.98238682747,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 48249.95994567871,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              29.565757358795963,
              29.048175747448113,
              28.522134925879072,
              28.582679988176096,
              28.397136702551506,
              28.53710793715436,
              28.550130991789047,
              28.59700543922372,
              28.908202693855856
            ],
            "command_buffer_throughput_us_p50": 28.582679988176096,
            "command_buffer_throughput_us_p90": 29.151692069717683,
            "command_buffer_latency_us": [
              48.874993808567524,
              47.41665907204151,
              48.666668590158224,
              47.833367716521025,
              51.333336159586906,
              49.62499951943755,
              48.749963752925396,
              47.37498238682747,
              48.24995994567871
            ],
            "command_buffer_latency_us_p50": 48.666668590158224,
            "command_buffer_latency_us_p90": 49.96666684746742
          },
          "throughput": [
            {
              "compute_ns": 3026755.0,
              "compute_span_ns": 3954250.0,
              "command_buffer_ns": 3806750.0572651625,
              "calibration_cpu_ns": 4799375.0,
              "calibration_gpu_ticks": 4799375.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 2969373.0,
              "compute_span_ns": 3883666.0,
              "command_buffer_ns": 3706583.345774561,
              "calibration_cpu_ns": 4547167.0,
              "calibration_gpu_ticks": 4547167.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3024752.0,
              "compute_span_ns": 3995458.0,
              "command_buffer_ns": 3817375.109065324,
              "calibration_cpu_ns": 4627042.0,
              "calibration_gpu_ticks": 4627042.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3037080.0,
              "compute_span_ns": 3923833.0,
              "command_buffer_ns": 3785375.156439841,
              "calibration_cpu_ns": 4541667.0,
              "calibration_gpu_ticks": 4541667.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3022798.0,
              "compute_span_ns": 4038875.0,
              "command_buffer_ns": 3866249.9864585698,
              "calibration_cpu_ns": 4820667.0,
              "calibration_gpu_ticks": 4820667.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3017253.0,
              "compute_span_ns": 3969083.0,
              "command_buffer_ns": 3809541.6966825724,
              "calibration_cpu_ns": 4709500.0,
              "calibration_gpu_ticks": 4709500.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3014622.0,
              "compute_span_ns": 4034250.0,
              "command_buffer_ns": 3856624.9422729015,
              "calibration_cpu_ns": 4673916.0,
              "calibration_gpu_ticks": 4673916.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3048335.0,
              "compute_span_ns": 4034792.0,
              "command_buffer_ns": 3873125.009704381,
              "calibration_cpu_ns": 4735666.0,
              "calibration_gpu_ticks": 4735666.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 3020214.0,
              "compute_span_ns": 3972292.0,
              "command_buffer_ns": 3816374.985035509,
              "calibration_cpu_ns": 4662125.0,
              "calibration_gpu_ticks": 4662125.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 47083.0,
              "compute_span_ns": 47083.0,
              "command_buffer_ns": 47083.303797990084,
              "calibration_cpu_ns": 502208.0,
              "calibration_gpu_ticks": 502208.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49917.0,
              "compute_span_ns": 49917.0,
              "command_buffer_ns": 49958.296585828066,
              "calibration_cpu_ns": 565791.0,
              "calibration_gpu_ticks": 565791.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 52209.0,
              "compute_span_ns": 52209.0,
              "command_buffer_ns": 52249.99040365219,
              "calibration_cpu_ns": 503667.0,
              "calibration_gpu_ticks": 503667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 48291.0,
              "compute_span_ns": 48291.0,
              "command_buffer_ns": 48250.018153339624,
              "calibration_cpu_ns": 558292.0,
              "calibration_gpu_ticks": 558292.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49750.0,
              "compute_span_ns": 49750.0,
              "command_buffer_ns": 49749.971367418766,
              "calibration_cpu_ns": 492917.0,
              "calibration_gpu_ticks": 492917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 50334.0,
              "compute_span_ns": 50334.0,
              "command_buffer_ns": 50375.00523030758,
              "calibration_cpu_ns": 523666.0,
              "calibration_gpu_ticks": 523666.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 51625.0,
              "compute_span_ns": 51625.0,
              "command_buffer_ns": 51625.01474842429,
              "calibration_cpu_ns": 465834.0,
              "calibration_gpu_ticks": 465834.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 47583.0,
              "compute_span_ns": 47583.0,
              "command_buffer_ns": 47541.68912768364,
              "calibration_cpu_ns": 574500.0,
              "calibration_gpu_ticks": 574500.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 49792.0,
              "compute_span_ns": 49792.0,
              "command_buffer_ns": 49833.32473784685,
              "calibration_cpu_ns": 474917.0,
              "calibration_gpu_ticks": 474917.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            47.293046875,
            46.396453125,
            47.26175,
            47.454375,
            47.23121875,
            47.144578125,
            47.10346875,
            47.630234375,
            47.19084375
          ],
          "compute_throughput_us_p50": 47.23121875,
          "compute_throughput_us_p90": 47.489546875,
          "command_buffer_throughput_us": [
            59.480469644768164,
            57.91536477772752,
            59.64648607914569,
            59.14648681937251,
            60.41015603841515,
            59.52408901066519,
            60.25976472301409,
            60.51757827663096,
            59.630859141179826
          ],
          "command_buffer_throughput_us_p50": 59.630859141179826,
          "command_buffer_throughput_us_p90": 60.43164048605831,
          "compute_latency_us": [
            47.083,
            49.917,
            52.209,
            48.291,
            49.75,
            50.334,
            51.625,
            47.583,
            49.792
          ],
          "compute_latency_us_p50": 49.792,
          "compute_latency_us_p90": 51.7418,
          "command_buffer_latency_us": [
            47.083303797990084,
            49.958296585828066,
            52.24999040365219,
            48.250018153339624,
            49.749971367418766,
            50.37500523030758,
            51.62501474842429,
            47.54168912768364,
            49.83332473784685
          ],
          "command_buffer_latency_us_p50": 49.83332473784685,
          "command_buffer_latency_us_p90": 51.75000987946987,
          "counter_control_throughput_ratio": 2.0858486831164424,
          "counter_control_latency_ratio": 1.0151134653528058
        },
        "download_ms": 0.367,
        "allocation_upload_ms": 0.885875,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 2.673614807194649e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "256",
        "3072",
        "1",
        "1",
        "3072",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-uta2q8a3/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "7bac43b2781c01105a8e678fa6161555c0e89cbd2b0a98679fdc42bb1bd74b45",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.28925,
        "capture_ms": 0.064209,
        "compile_ms": 51.185625,
        "allocation_upload_ms": 1.866833,
        "cold_call_ms": 5.118792,
        "warmup_ms": 200.396542,
        "download_ms": 0.860125,
        "repetitions": 1553,
        "output_elements": 786432,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          15.4144932389,
          15.3742491951,
          15.5245492595,
          14.6984063104,
          14.6168705731,
          14.9611500322,
          14.505499678,
          14.7282949131,
          14.5037025113
        ],
        "latency_us": [
          267.208,
          239.708,
          212,
          215.459,
          220.334,
          222.833,
          238.375,
          234.458,
          250.708
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 256,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 128,
            "reduction_scalar_rounds": 60,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 12288,
              "global_write_bytes": 12288,
              "private_read_bytes": 73728,
              "private_write_bytes": 24576
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 48,
              "global_write_bytes": 48,
              "private_read_bytes": 288,
              "private_write_bytes": 96
            },
            "striped_storage_scalars_per_worker": 24,
            "reduction_operations": 2,
            "reduction_elements": 6144,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 100,
            "normalized_kernel_cost": 400,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 9216,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 4,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 911334,
              "compute_span_ns": 911334,
              "calibration_cpu_ns": 1393958,
              "calibration_gpu_ticks": 1393958,
              "compute_passes": 1,
              "command_buffer_ns": 911416.660529,
              "command_buffers": 2
            },
            {
              "compute_ns": 892333,
              "compute_span_ns": 892333,
              "calibration_cpu_ns": 1319666,
              "calibration_gpu_ticks": 1319666,
              "compute_passes": 1,
              "command_buffer_ns": 892416.690476,
              "command_buffers": 2
            },
            {
              "compute_ns": 933083,
              "compute_span_ns": 933083,
              "calibration_cpu_ns": 1507500,
              "calibration_gpu_ticks": 1507500,
              "compute_passes": 1,
              "command_buffer_ns": 933124.916628,
              "command_buffers": 2
            },
            {
              "compute_ns": 958750,
              "compute_span_ns": 958750,
              "calibration_cpu_ns": 1441917,
              "calibration_gpu_ticks": 1441917,
              "compute_passes": 1,
              "command_buffer_ns": 958874.996286,
              "command_buffers": 2
            },
            {
              "compute_ns": 909583,
              "compute_span_ns": 909583,
              "calibration_cpu_ns": 1392834,
              "calibration_gpu_ticks": 1392834,
              "compute_passes": 1,
              "command_buffer_ns": 909624.970518,
              "command_buffers": 2
            },
            {
              "compute_ns": 970458,
              "compute_span_ns": 970458,
              "calibration_cpu_ns": 1448917,
              "calibration_gpu_ticks": 1448917,
              "compute_passes": 1,
              "command_buffer_ns": 970499.997493,
              "command_buffers": 2
            },
            {
              "compute_ns": 908041,
              "compute_span_ns": 908041,
              "calibration_cpu_ns": 1417958,
              "calibration_gpu_ticks": 1417958,
              "compute_passes": 1,
              "command_buffer_ns": 908041.605726,
              "command_buffers": 2
            },
            {
              "compute_ns": 971959,
              "compute_span_ns": 971959,
              "calibration_cpu_ns": 1459250,
              "calibration_gpu_ticks": 1459250,
              "compute_passes": 1,
              "command_buffer_ns": 972083.362285,
              "command_buffers": 2
            },
            {
              "compute_ns": 910250,
              "compute_span_ns": 910250,
              "calibration_cpu_ns": 1333208,
              "calibration_gpu_ticks": 1333208,
              "compute_passes": 1,
              "command_buffer_ns": 910250.004381,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 18750,
              "compute_span_ns": 18750,
              "calibration_cpu_ns": 410083,
              "calibration_gpu_ticks": 410083,
              "compute_passes": 1,
              "command_buffer_ns": 18750.0263564,
              "command_buffers": 2
            },
            {
              "compute_ns": 18583,
              "compute_span_ns": 18583,
              "calibration_cpu_ns": 426875,
              "calibration_gpu_ticks": 426875,
              "compute_passes": 1,
              "command_buffer_ns": 18666.6147783,
              "command_buffers": 2
            },
            {
              "compute_ns": 18500,
              "compute_span_ns": 18500,
              "calibration_cpu_ns": 407584,
              "calibration_gpu_ticks": 407584,
              "compute_passes": 1,
              "command_buffer_ns": 18541.6429304,
              "command_buffers": 2
            },
            {
              "compute_ns": 18541,
              "compute_span_ns": 18541,
              "calibration_cpu_ns": 401417,
              "calibration_gpu_ticks": 401417,
              "compute_passes": 1,
              "command_buffer_ns": 18541.6429304,
              "command_buffers": 2
            },
            {
              "compute_ns": 18584,
              "compute_span_ns": 18584,
              "calibration_cpu_ns": 451292,
              "calibration_gpu_ticks": 451292,
              "compute_passes": 1,
              "command_buffer_ns": 18624.9963008,
              "command_buffers": 2
            },
            {
              "compute_ns": 18584,
              "compute_span_ns": 18584,
              "calibration_cpu_ns": 411083,
              "calibration_gpu_ticks": 411083,
              "compute_passes": 1,
              "command_buffer_ns": 18624.9963008,
              "command_buffers": 2
            },
            {
              "compute_ns": 18500,
              "compute_span_ns": 18500,
              "calibration_cpu_ns": 436208,
              "calibration_gpu_ticks": 436208,
              "compute_passes": 1,
              "command_buffer_ns": 18499.9662451,
              "command_buffers": 2
            },
            {
              "compute_ns": 18500,
              "compute_span_ns": 18500,
              "calibration_cpu_ns": 401167,
              "calibration_gpu_ticks": 401167,
              "compute_passes": 1,
              "command_buffer_ns": 18500.0244528,
              "command_buffers": 2
            },
            {
              "compute_ns": 18541,
              "compute_span_ns": 18541,
              "calibration_cpu_ns": 478708,
              "calibration_gpu_ticks": 478708,
              "compute_passes": 1,
              "command_buffer_ns": 18583.3196156,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 978791.67879,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 890791.707207,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 961208.308581,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 895541.685168,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 903541.687876,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 930041.715037,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 920749.967918,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 931791.670155,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 903125.037439,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 18875.0564121,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18541.5847227,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18750.0263564,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18624.9963008,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18833.3215192,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18541.6429304,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18541.6429304,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18583.3196156,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 18666.672986,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              15.29361998109375,
              13.918620425109374,
              15.018879821578125,
              13.992838830750001,
              14.1178388730625,
              14.531901797453125,
              14.386718248718749,
              14.559244846171875,
              14.111328709984374
            ],
            "command_buffer_throughput_us_p50": 14.386718248718749,
            "command_buffer_throughput_us_p90": 15.07382785348125,
            "command_buffer_latency_us": [
              18.875056412099998,
              18.5415847227,
              18.750026356400003,
              18.624996300800003,
              18.8333215192,
              18.541642930400002,
              18.541642930400002,
              18.5833196156,
              18.666672986000002
            ],
            "command_buffer_latency_us_p50": 18.624996300800003,
            "command_buffer_latency_us_p90": 18.84166849778
          },
          "compute_throughput_us": [
            14.23959375,
            13.942703125,
            14.579421875,
            14.98046875,
            14.212234375,
            15.16340625,
            14.188140625,
            15.186859375,
            14.22265625
          ],
          "compute_throughput_us_p50": 14.23959375,
          "compute_throughput_us_p90": 15.168096875,
          "command_buffer_throughput_us": [
            14.240885320765624,
            13.9440107886875,
            14.580076822312499,
            14.98242181696875,
            14.212890164343749,
            15.164062460828125,
            14.188150089468751,
            15.188802535703125,
            14.222656318453124
          ],
          "command_buffer_throughput_us_p50": 14.240885320765624,
          "command_buffer_throughput_us_p90": 15.169010475803125,
          "compute_latency_us": [
            18.75,
            18.583,
            18.5,
            18.541,
            18.584,
            18.584,
            18.5,
            18.5,
            18.541
          ],
          "compute_latency_us_p50": 18.541,
          "compute_latency_us_p90": 18.6172,
          "command_buffer_latency_us": [
            18.750026356400003,
            18.6666147783,
            18.541642930400002,
            18.541642930400002,
            18.624996300800003,
            18.624996300800003,
            18.4999662451,
            18.5000244528,
            18.5833196156
          ],
          "command_buffer_latency_us_p50": 18.5833196156,
          "command_buffer_latency_us_p90": 18.68329709392,
          "counter_control_throughput_ratio": 1.0067327083228448,
          "counter_control_latency_ratio": 0.9955246503648207
        },
        "correctness": {
          "max_abs_error": 1.569250362249905e-10,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 14.7282949131,
        "throughput_us_p90": 15.43650444302,
        "latency_us_p50": 234.458,
        "latency_us_p90": 254.008
      },
      "slowdown": 0.4376297217640049,
      "device_compute_slowdown": 0.301486900547109,
      "valid": true,
      "round": 3,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "softmax_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "native_first",
      "implementation_order": [
        "native",
        "torch"
      ],
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-v4w_vpk0/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "512",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "2",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "5e168786762f89acc561194a669a3725da4fe77bed5800b1d573576d096bb3fc",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 2,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 512,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 17.817417,
        "capture_ms": 0.069833,
        "compile_ms": 48.831583,
        "allocation_upload_ms": 5.784166,
        "cold_call_ms": 6.662791,
        "warmup_ms": 200.51025,
        "download_ms": 3.605375,
        "repetitions": 454,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          63.3773854626,
          63.5185396476,
          63.513123348,
          62.9382356828,
          63.293685022,
          63.3784867841,
          63.0260660793,
          63.127660793,
          62.9814603524
        ],
        "latency_us": [
          349.125,
          308.708,
          272.5,
          278.916,
          280.291,
          270.458,
          289.791,
          279.5,
          261.792
        ],
        "execution_plans": [
          {
            "threads": 512,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 128,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 2,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 384,
            "reduction_scalar_rounds": 120,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 24576,
              "global_write_bytes": 24576,
              "private_read_bytes": 147456,
              "private_write_bytes": 49152
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 96,
              "private_read_bytes": 576,
              "private_write_bytes": 192
            },
            "striped_storage_scalars_per_worker": 48,
            "reduction_operations": 2,
            "reduction_elements": 12288,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 160,
            "normalized_kernel_cost": 1920,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 18432,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 4031709,
              "compute_span_ns": 4031709,
              "calibration_cpu_ns": 4542167,
              "calibration_gpu_ticks": 4542167,
              "compute_passes": 1,
              "command_buffer_ns": 4031874.99614,
              "command_buffers": 2
            },
            {
              "compute_ns": 3993167,
              "compute_span_ns": 3993167,
              "calibration_cpu_ns": 4444541,
              "calibration_gpu_ticks": 4444541,
              "compute_passes": 1,
              "command_buffer_ns": 3993250.02218,
              "command_buffers": 2
            },
            {
              "compute_ns": 3993708,
              "compute_span_ns": 3993708,
              "calibration_cpu_ns": 4469000,
              "calibration_gpu_ticks": 4469000,
              "compute_passes": 1,
              "command_buffer_ns": 3993708.29109,
              "command_buffers": 2
            },
            {
              "compute_ns": 3994375,
              "compute_span_ns": 3994375,
              "calibration_cpu_ns": 4454250,
              "calibration_gpu_ticks": 4454250,
              "compute_passes": 1,
              "command_buffer_ns": 3994416.62012,
              "command_buffers": 2
            },
            {
              "compute_ns": 3974333,
              "compute_span_ns": 3974333,
              "calibration_cpu_ns": 4417500,
              "calibration_gpu_ticks": 4417500,
              "compute_passes": 1,
              "command_buffer_ns": 3974333.34729,
              "command_buffers": 2
            },
            {
              "compute_ns": 4087209,
              "compute_span_ns": 4087209,
              "calibration_cpu_ns": 4542209,
              "calibration_gpu_ticks": 4542209,
              "compute_passes": 1,
              "command_buffer_ns": 4087249.98124,
              "command_buffers": 2
            },
            {
              "compute_ns": 4094000,
              "compute_span_ns": 4094000,
              "calibration_cpu_ns": 4553333,
              "calibration_gpu_ticks": 4553333,
              "compute_passes": 1,
              "command_buffer_ns": 4093999.97443,
              "command_buffers": 2
            },
            {
              "compute_ns": 4010459,
              "compute_span_ns": 4010459,
              "calibration_cpu_ns": 4438750,
              "calibration_gpu_ticks": 4438750,
              "compute_passes": 1,
              "command_buffer_ns": 4010541.7138,
              "command_buffers": 2
            },
            {
              "compute_ns": 3971917,
              "compute_span_ns": 3971917,
              "calibration_cpu_ns": 4448375,
              "calibration_gpu_ticks": 4448375,
              "compute_passes": 1,
              "command_buffer_ns": 3972000.03499,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 63291,
              "compute_span_ns": 63291,
              "calibration_cpu_ns": 468125,
              "calibration_gpu_ticks": 468125,
              "compute_passes": 1,
              "command_buffer_ns": 63333.3693258,
              "command_buffers": 2
            },
            {
              "compute_ns": 63333,
              "compute_span_ns": 63333,
              "calibration_cpu_ns": 435833,
              "calibration_gpu_ticks": 435833,
              "compute_passes": 1,
              "command_buffer_ns": 63375.0460111,
              "command_buffers": 2
            },
            {
              "compute_ns": 63292,
              "compute_span_ns": 63292,
              "calibration_cpu_ns": 487375,
              "calibration_gpu_ticks": 487375,
              "compute_passes": 1,
              "command_buffer_ns": 63374.9878034,
              "command_buffers": 2
            },
            {
              "compute_ns": 63167,
              "compute_span_ns": 63167,
              "calibration_cpu_ns": 537875,
              "calibration_gpu_ticks": 537875,
              "compute_passes": 1,
              "command_buffer_ns": 63249.9577478,
              "command_buffers": 2
            },
            {
              "compute_ns": 63500,
              "compute_span_ns": 63500,
              "calibration_cpu_ns": 441833,
              "calibration_gpu_ticks": 441833,
              "compute_passes": 1,
              "command_buffer_ns": 63541.5781289,
              "command_buffers": 2
            },
            {
              "compute_ns": 62875,
              "compute_span_ns": 62875,
              "calibration_cpu_ns": 449917,
              "calibration_gpu_ticks": 449917,
              "compute_passes": 1,
              "command_buffer_ns": 62916.718889,
              "command_buffers": 2
            },
            {
              "compute_ns": 63125,
              "compute_span_ns": 63125,
              "calibration_cpu_ns": 499250,
              "calibration_gpu_ticks": 499250,
              "compute_passes": 1,
              "command_buffer_ns": 63208.3392702,
              "command_buffers": 2
            },
            {
              "compute_ns": 63458,
              "compute_span_ns": 63458,
              "calibration_cpu_ns": 451792,
              "calibration_gpu_ticks": 451792,
              "compute_passes": 1,
              "command_buffer_ns": 63500.017859,
              "command_buffers": 2
            },
            {
              "compute_ns": 63583,
              "compute_span_ns": 63583,
              "calibration_cpu_ns": 428125,
              "calibration_gpu_ticks": 428125,
              "compute_passes": 1,
              "command_buffer_ns": 63583.3130218,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 4082166.648,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4026875.01628,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4064249.98073,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4097583.35445,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4011791.66511,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4007875.04623,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3986041.64366,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4015499.95877,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 4090791.68458,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 63874.9916106,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63416.6644886,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63458.3411738,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 82333.3975859,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 77083.3576098,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 66833.2795613,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 63416.6644886,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 77583.3614171,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 76833.3557062,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              63.783853875,
              62.919922129375,
              63.50390594890625,
              64.02473991328125,
              62.68424476734375,
              62.62304759734375,
              62.2819006821875,
              62.74218685578125,
              63.9186200715625
            ],
            "command_buffer_throughput_us_p50": 62.919922129375,
            "command_buffer_throughput_us_p90": 63.93984403990625,
            "command_buffer_latency_us": [
              63.8749916106,
              63.4166644886,
              63.4583411738,
              82.3333975859,
              77.0833576098,
              66.8332795613,
              63.4166644886,
              77.5833614171,
              76.83335570620001
            ],
            "command_buffer_latency_us_p50": 66.8332795613,
            "command_buffer_latency_us_p90": 78.53336865086
          },
          "compute_throughput_us": [
            62.995453125,
            62.393234375,
            62.4016875,
            62.412109375,
            62.098953125,
            63.862640625,
            63.96875,
            62.663421875,
            62.061203125
          ],
          "compute_throughput_us_p50": 62.412109375,
          "compute_throughput_us_p90": 63.8838625,
          "command_buffer_throughput_us": [
            62.9980468146875,
            62.394531596562494,
            62.40169204828125,
            62.412759689375,
            62.09895855140625,
            63.863280956875,
            63.96874960046875,
            62.664714278125004,
            62.06250054671875
          ],
          "command_buffer_throughput_us_p50": 62.412759689375,
          "command_buffer_throughput_us_p90": 63.88437468559375,
          "compute_latency_us": [
            63.291,
            63.333,
            63.292,
            63.167,
            63.5,
            62.875,
            63.125,
            63.458,
            63.583
          ],
          "compute_latency_us_p50": 63.292,
          "compute_latency_us_p90": 63.5166,
          "command_buffer_latency_us": [
            63.3333693258,
            63.3750460111,
            63.374987803399996,
            63.249957747799996,
            63.5415781289,
            62.916718889,
            63.2083392702,
            63.500017858999996,
            63.5833130218
          ],
          "command_buffer_latency_us_p50": 63.374987803399996,
          "command_buffer_latency_us_p90": 63.54992510748,
          "counter_control_throughput_ratio": 0.9906629454002386,
          "counter_control_latency_ratio": 0.9413980475294842
        },
        "correctness": {
          "max_abs_error": 7.116789644580346e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 63.293685022,
        "throughput_us_p90": 63.51420660792,
        "latency_us_p50": 279.5,
        "latency_us_p90": 316.7914
      },
      "torch": {
        "cold_call_ms": 7.0635,
        "warmup_ms": 201.478709,
        "repetitions": 188,
        "throughput_us": [
          131.59308510638297,
          131.78745744680853,
          132.57358510638298,
          132.04232978723405,
          131.84995212765955,
          132.0474308510638,
          132.47340425531914,
          132.71476063829786,
          131.1852819148936
        ],
        "latency_us": [
          630.6659999999999,
          440.458,
          426.70799999999997,
          410.95799999999997,
          409.125,
          425.25,
          424.375,
          420.459,
          546.417
        ],
        "throughput_us_p50": 132.04232978723405,
        "throughput_us_p90": 132.60182021276594,
        "latency_us_p50": 425.25,
        "latency_us_p90": 563.2668,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 7877125.055529177,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7911916.64757207,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7885916.6242182255,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7870875.007938594,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7804416.643921286,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7922750.141005963,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7887999.99281764,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7910583.3428911865,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7881833.298597485,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 134166.67934507132,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 152666.64559021592,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 137375.02740696073,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 135749.98592957854,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 141250.02780929208,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 150875.0137872994,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 134250.0327154994,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 137125.02550333738,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 158499.98453631997,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              123.0800789926434,
              123.6236976183136,
              123.21744725340977,
              122.98242199904053,
              121.94401006127009,
              123.79297095321817,
              123.24999988777563,
              123.60286473267479,
              123.1536452905857
            ],
            "command_buffer_throughput_us_p50": 123.21744725340977,
            "command_buffer_throughput_us_p90": 123.65755228529451,
            "command_buffer_latency_us": [
              134.16667934507132,
              152.66664559021592,
              137.37502740696073,
              135.74998592957854,
              141.25002780929208,
              150.8750137872994,
              134.2500327154994,
              137.12502550333738,
              158.49998453631997
            ],
            "command_buffer_latency_us_p50": 137.37502740696073,
            "command_buffer_latency_us_p90": 153.83331337943673
          },
          "throughput": [
            {
              "compute_ns": 8785209.0,
              "compute_span_ns": 10422041.0,
              "command_buffer_ns": 10154625.051654875,
              "calibration_cpu_ns": 11426792.0,
              "calibration_gpu_ticks": 11426792.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8719958.0,
              "compute_span_ns": 10393500.0,
              "command_buffer_ns": 10132124.996744096,
              "calibration_cpu_ns": 11248291.0,
              "calibration_gpu_ticks": 11248291.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8658247.0,
              "compute_span_ns": 10281166.0,
              "command_buffer_ns": 10000458.278227597,
              "calibration_cpu_ns": 10964750.0,
              "calibration_gpu_ticks": 10964750.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8653376.0,
              "compute_span_ns": 10267584.0,
              "command_buffer_ns": 10030333.301983774,
              "calibration_cpu_ns": 10994833.0,
              "calibration_gpu_ticks": 10994833.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8694837.0,
              "compute_span_ns": 10307709.0,
              "command_buffer_ns": 10052166.471723467,
              "calibration_cpu_ns": 11249417.0,
              "calibration_gpu_ticks": 11249417.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8609673.0,
              "compute_span_ns": 10212500.0,
              "command_buffer_ns": 9953458.386007696,
              "calibration_cpu_ns": 11059208.0,
              "calibration_gpu_ticks": 11059208.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8656463.0,
              "compute_span_ns": 10291125.0,
              "command_buffer_ns": 10016083.426307887,
              "calibration_cpu_ns": 11367291.0,
              "calibration_gpu_ticks": 11367291.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8689796.0,
              "compute_span_ns": 10386375.0,
              "command_buffer_ns": 10143166.524358094,
              "calibration_cpu_ns": 11209250.0,
              "calibration_gpu_ticks": 11209250.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8673872.0,
              "compute_span_ns": 10325959.0,
              "command_buffer_ns": 10060166.707262397,
              "calibration_cpu_ns": 11013375.0,
              "calibration_gpu_ticks": 11013375.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 133625.0,
              "compute_span_ns": 133625.0,
              "command_buffer_ns": 133624.9988526106,
              "calibration_cpu_ns": 671375.0,
              "calibration_gpu_ticks": 671375.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 135291.0,
              "compute_span_ns": 135291.0,
              "command_buffer_ns": 135250.04032999277,
              "calibration_cpu_ns": 593000.0,
              "calibration_gpu_ticks": 593000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 134250.0,
              "compute_span_ns": 134250.0,
              "command_buffer_ns": 134249.9745078385,
              "calibration_cpu_ns": 639917.0,
              "calibration_gpu_ticks": 639917.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 143083.0,
              "compute_span_ns": 143083.0,
              "command_buffer_ns": 143083.33629742265,
              "calibration_cpu_ns": 624042.0,
              "calibration_gpu_ticks": 624042.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 136417.0,
              "compute_span_ns": 136417.0,
              "command_buffer_ns": 136416.6964776814,
              "calibration_cpu_ns": 618125.0,
              "calibration_gpu_ticks": 618125.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 150208.0,
              "compute_span_ns": 150208.0,
              "command_buffer_ns": 150166.6847616434,
              "calibration_cpu_ns": 630083.0,
              "calibration_gpu_ticks": 630083.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 138084.0,
              "compute_span_ns": 138084.0,
              "command_buffer_ns": 138124.97491016984,
              "calibration_cpu_ns": 607000.0,
              "calibration_gpu_ticks": 607000.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 143334.0,
              "compute_span_ns": 143334.0,
              "command_buffer_ns": 143375.01488626003,
              "calibration_cpu_ns": 603667.0,
              "calibration_gpu_ticks": 603667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 133750.0,
              "compute_span_ns": 133750.0,
              "command_buffer_ns": 133749.9707005918,
              "calibration_cpu_ns": 609000.0,
              "calibration_gpu_ticks": 609000.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            137.268890625,
            136.24934375,
            135.285109375,
            135.209,
            135.856828125,
            134.526140625,
            135.257234375,
            135.7780625,
            135.52925
          ],
          "compute_throughput_us_p50": 135.52925,
          "compute_throughput_us_p90": 136.453253125,
          "command_buffer_throughput_us": [
            158.66601643210743,
            158.3144530741265,
            156.2571605973062,
            156.72395784349646,
            157.06510112067917,
            155.52278728137026,
            156.50130353606073,
            158.48697694309521,
            157.19010480097495
          ],
          "command_buffer_throughput_us_p50": 157.06510112067917,
          "command_buffer_throughput_us_p90": 158.52278484089766,
          "compute_latency_us": [
            133.625,
            135.291,
            134.25,
            143.083,
            136.417,
            150.208,
            138.084,
            143.334,
            133.75
          ],
          "compute_latency_us_p50": 136.417,
          "compute_latency_us_p90": 144.7088,
          "command_buffer_latency_us": [
            133.6249988526106,
            135.25004032999277,
            134.2499745078385,
            143.08333629742265,
            136.4166964776814,
            150.1666847616434,
            138.12497491016984,
            143.37501488626003,
            133.7499707005918
          ],
          "command_buffer_latency_us_p50": 136.4166964776814,
          "command_buffer_latency_us_p90": 144.7333488613367,
          "counter_control_throughput_ratio": 1.2763739508487868,
          "counter_control_latency_ratio": 0.9953051932995707
        },
        "download_ms": 0.62675,
        "allocation_upload_ms": 16.66,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 7.553236784944503e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "slowdown": 0.47934389770301733,
      "device_compute_slowdown": 0.46050656500349557,
      "valid": true,
      "round": 3,
      "variant": "candidate"
    },
    {
      "case": {
        "operation": "softmax",
        "m": 768,
        "n": 6144,
        "k": 1
      },
      "name": "softmax_768x6144",
      "backend": "metal",
      "block": [
        1,
        6144,
        1
      ],
      "timing_order": "torch_first",
      "implementation_order": [
        "torch",
        "native"
      ],
      "torch": {
        "cold_call_ms": 0.637208,
        "warmup_ms": 201.140709,
        "repetitions": 225,
        "throughput_us": [
          130.52944444444444,
          129.88870222222224,
          130.23036888888888,
          129.77777777777777,
          131.4407422222222,
          129.17388888888888,
          130.185,
          130.87129777777776,
          130.67759555555554
        ],
        "latency_us": [
          436.541,
          563.7919999999999,
          402.45799999999997,
          388.291,
          380.33299999999997,
          369.958,
          401.875,
          449.5,
          385.542
        ],
        "throughput_us_p50": 130.23036888888888,
        "throughput_us_p90": 130.98518666666664,
        "latency_us_p50": 401.875,
        "latency_us_p90": 472.3584,
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 7796083.40235427,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7875874.871388078,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7862958.300393075,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7830041.751731187,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7825499.982573092,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7861041.638534516,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7893333.4443718195,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7860208.512283862,
                "command_buffers": 11
              },
              {
                "command_buffer_ns": 7837833.371013403,
                "command_buffers": 11
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 138000.00306218863,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 136666.64017364383,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 134375.00456348062,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 147124.98523294926,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 134125.00265985727,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 136999.99544769526,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 151749.99134615064,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 136291.66642203927,
                "command_buffers": 1
              },
              {
                "command_buffer_ns": 133624.9988526106,
                "command_buffers": 1
              }
            ],
            "command_buffer_throughput_us": [
              121.81380316178547,
              123.06054486543871,
              122.8587234436418,
              122.3444023707998,
              122.27343722770456,
              122.82877560210181,
              123.33333506830968,
              122.81575800443534,
              122.46614642208442
            ],
            "command_buffer_throughput_us_p50": 122.81575800443534,
            "command_buffer_throughput_us_p90": 123.11510290601291,
            "command_buffer_latency_us": [
              138.00000306218863,
              136.66664017364383,
              134.37500456348062,
              147.12498523294926,
              134.12500265985727,
              136.99999544769526,
              151.74999134615064,
              136.29166642203927,
              133.6249988526106
            ],
            "command_buffer_latency_us_p50": 136.66664017364383,
            "command_buffer_latency_us_p90": 148.04998645558953
          },
          "throughput": [
            {
              "compute_ns": 8630670.0,
              "compute_span_ns": 10273584.0,
              "command_buffer_ns": 10019666.74812138,
              "calibration_cpu_ns": 11067875.0,
              "calibration_gpu_ticks": 11067875.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8664540.0,
              "compute_span_ns": 10344667.0,
              "command_buffer_ns": 10055041.639134288,
              "calibration_cpu_ns": 11009125.0,
              "calibration_gpu_ticks": 11009125.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8581745.0,
              "compute_span_ns": 10201833.0,
              "command_buffer_ns": 9948333.259671926,
              "calibration_cpu_ns": 11050958.0,
              "calibration_gpu_ticks": 11050958.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8653543.0,
              "compute_span_ns": 10209250.0,
              "command_buffer_ns": 9957500.034943223,
              "calibration_cpu_ns": 10876958.0,
              "calibration_gpu_ticks": 10876958.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8629210.0,
              "compute_span_ns": 10184917.0,
              "command_buffer_ns": 9920416.574459523,
              "calibration_cpu_ns": 10935875.0,
              "calibration_gpu_ticks": 10935875.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8692504.0,
              "compute_span_ns": 10372334.0,
              "command_buffer_ns": 10095416.859257966,
              "calibration_cpu_ns": 11063791.0,
              "calibration_gpu_ticks": 11063791.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8659083.0,
              "compute_span_ns": 10250458.0,
              "command_buffer_ns": 10023791.692219675,
              "calibration_cpu_ns": 11131083.0,
              "calibration_gpu_ticks": 11131083.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8702043.0,
              "compute_span_ns": 10319958.0,
              "command_buffer_ns": 10047833.260614425,
              "calibration_cpu_ns": 11102167.0,
              "calibration_gpu_ticks": 11102167.0,
              "compute_passes": 64,
              "command_buffers": 11
            },
            {
              "compute_ns": 8616577.0,
              "compute_span_ns": 10210916.0,
              "command_buffer_ns": 9956333.320587873,
              "calibration_cpu_ns": 10996958.0,
              "calibration_gpu_ticks": 10996958.0,
              "compute_passes": 64,
              "command_buffers": 11
            }
          ],
          "latency": [
            {
              "compute_ns": 137125.0,
              "compute_span_ns": 137125.0,
              "command_buffer_ns": 137124.96729567647,
              "calibration_cpu_ns": 642625.0,
              "calibration_gpu_ticks": 642625.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 141125.0,
              "compute_span_ns": 141125.0,
              "command_buffer_ns": 141124.99775364995,
              "calibration_cpu_ns": 643667.0,
              "calibration_gpu_ticks": 643667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 137750.0,
              "compute_span_ns": 137750.0,
              "command_buffer_ns": 137750.00115856528,
              "calibration_cpu_ns": 714250.0,
              "calibration_gpu_ticks": 714250.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 154042.0,
              "compute_span_ns": 154042.0,
              "command_buffer_ns": 154041.68516397476,
              "calibration_cpu_ns": 627583.0,
              "calibration_gpu_ticks": 627583.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 142333.0,
              "compute_span_ns": 142333.0,
              "command_buffer_ns": 142333.33058655262,
              "calibration_cpu_ns": 900667.0,
              "calibration_gpu_ticks": 900667.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 134792.0,
              "compute_span_ns": 134792.0,
              "command_buffer_ns": 134791.71320796013,
              "calibration_cpu_ns": 550167.0,
              "calibration_gpu_ticks": 550167.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 145833.0,
              "compute_span_ns": 145833.0,
              "command_buffer_ns": 145791.62234440446,
              "calibration_cpu_ns": 620417.0,
              "calibration_gpu_ticks": 620417.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 135375.0,
              "compute_span_ns": 135375.0,
              "command_buffer_ns": 135375.012177974,
              "calibration_cpu_ns": 743750.0,
              "calibration_gpu_ticks": 743750.0,
              "compute_passes": 1,
              "command_buffers": 1
            },
            {
              "compute_ns": 136917.0,
              "compute_span_ns": 136917.0,
              "command_buffer_ns": 136916.64207726717,
              "calibration_cpu_ns": 687208.0,
              "calibration_gpu_ticks": 687208.0,
              "compute_passes": 1,
              "command_buffers": 1
            }
          ],
          "compute_throughput_us": [
            134.85421875,
            135.3834375,
            134.089765625,
            135.211609375,
            134.83140625,
            135.820375,
            135.298171875,
            135.969421875,
            134.634015625
          ],
          "compute_throughput_us_p50": 135.211609375,
          "compute_throughput_us_p90": 135.850184375,
          "command_buffer_throughput_us": [
            156.55729293939658,
            157.11002561147325,
            155.44270718237385,
            155.58593804598786,
            155.00650897593005,
            157.74088842590572,
            156.62174519093242,
            156.9973946971004,
            155.56770813418552
          ],
          "command_buffer_throughput_us_p50": 156.55729293939658,
          "command_buffer_throughput_us_p90": 157.23619817435974,
          "compute_latency_us": [
            137.125,
            141.125,
            137.75,
            154.042,
            142.333,
            134.792,
            145.833,
            135.375,
            136.917
          ],
          "compute_latency_us_p50": 137.75,
          "compute_latency_us_p90": 147.4748,
          "command_buffer_latency_us": [
            137.12496729567647,
            141.12499775364995,
            137.75000115856528,
            154.04168516397476,
            142.33333058655262,
            134.79171320796013,
            145.79162234440446,
            135.37501217797399,
            136.91664207726717
          ],
          "command_buffer_latency_us_p50": 137.75000115856528,
          "command_buffer_latency_us_p90": 147.44163490831852,
          "counter_control_throughput_ratio": 1.2717045899201833,
          "counter_control_latency_ratio": 1.0246334387496405
        },
        "download_ms": 0.632666,
        "allocation_upload_ms": 0.953791,
        "device": "mps:0",
        "output_policy": "preallocated_out",
        "correctness": {
          "max_abs_error": 7.553236784944503e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        }
      },
      "native_command": [
        "/Users/mike/CLionProjects/luisa/cmake-build-tirx/bin/benchmark_tile_tirx",
        "metal",
        "softmax",
        "768",
        "6144",
        "1",
        "1",
        "6144",
        "1",
        "9",
        "30",
        "200",
        "/var/folders/2_/k4txnts916qb_hd5bnch_jfw0000gn/T/luisa-tile-benchmark-iwukpe9t/output.f32",
        "auto",
        "2",
        "subgroup-reduce",
        "vectorize",
        "256",
        "1",
        "tvm",
        "retain-subgroup-fences",
        "0",
        "16",
        "retain-input-snapshots",
        "generic",
        "reference",
        "reference",
        "preserve",
        "1",
        "auto",
        "1",
        "4",
        "cache"
      ],
      "native_stderr": "",
      "native_source_sha256": "7f45b1ebfa78e24e3d477c2ad514fde5c49a3d7b63ed786757d9dc5d1979c8b1",
      "native": {
        "backend": "metal",
        "operation": "softmax",
        "execution_scope": "auto",
        "pipeline_window": 2,
        "cooperative_matrix": false,
        "metal_mpp": false,
        "metal_subgroup_reductions": true,
        "metal_max_threads": 1024,
        "reduction_programs_per_group": 1,
        "reduction_unroll_factor": 1,
        "reduction_lane_elements": 4,
        "cache_reduction_inputs": true,
        "reduction_cost_profile": "analytic",
        "fuse_gpu_elementwise": true,
        "shared_tile_materialization": "preserve",
        "forward_readonly_tile_loads": true,
        "elide_independent_subgroup_barriers": false,
        "vectorize": true,
        "auto_vectorize": false,
        "cpu_stack_bytes": 0,
        "cpu_parallel_task_threshold": 64,
        "cpu_vector_lanes": 16,
        "max_reduction_striped_scalars_per_worker": 64,
        "cpu_target_policy": "generic",
        "cpu_model": "generic",
        "cpu_matrix_backend": "reference",
        "cpu_math_backend": "reference",
        "planner_threads": 256,
        "copy_batch": 1,
        "matrix_intrinsics": 0,
        "external_matrix_calls": 0,
        "external_vector_math_calls": 0,
        "simdgroup_intrinsics": 0,
        "mpp_intrinsics": 0,
        "runtime_init_ms": 20.595042,
        "capture_ms": 0.061916,
        "compile_ms": 48.080709,
        "allocation_upload_ms": 6.005708,
        "cold_call_ms": 4.810209,
        "warmup_ms": 200.030833,
        "download_ms": 3.210458,
        "repetitions": 446,
        "output_elements": 4718592,
        "mma_operations": 0,
        "compiler": "21.0.0 (clang-2100.1.1.101)",
        "throughput_us": [
          57.7531748879,
          57.7215067265,
          56.7023542601,
          57.9460941704,
          56.7273901345,
          56.6120156951,
          56.71132287,
          57.5671704036,
          56.5948251121
        ],
        "latency_us": [
          353.667,
          287.583,
          277.584,
          285.667,
          323.625,
          276.208,
          337.458,
          374.125,
          284.625
        ],
        "execution_plans": [
          {
            "threads": 256,
            "metal_mpp": false,
            "cost_basis": "metal_subgroup_reduction_v1",
            "programs": 768,
            "shared_memory_bytes": 64,
            "optimized": true,
            "candidates_considered": 1,
            "candidates_rejected": 0,
            "max_copy_batch": 1,
            "batched_copy_operations": 0,
            "prefetched_pipeline_loops": 0,
            "prefetch_storage_scalars_per_lane": 0,
            "reduction_subgroups_per_program": 8,
            "reduction_programs_per_group": 1,
            "reduction_unroll_factor": 1,
            "reduction_lane_elements": 4,
            "reduction_threadgroups": 768,
            "reduction_scalar_rounds": 120,
            "reduction_lane_utilization": 1,
            "reduction_payload_accesses_known": true,
            "reduction_payload_accesses_per_program": {
              "global_read_bytes": 24576,
              "global_write_bytes": 24576,
              "private_read_bytes": 147456,
              "private_write_bytes": 49152
            },
            "reduction_payload_accesses_per_worker": {
              "global_read_bytes": 96,
              "global_write_bytes": 96,
              "private_read_bytes": 576,
              "private_write_bytes": 192
            },
            "striped_storage_scalars_per_worker": 48,
            "reduction_operations": 2,
            "reduction_elements": 12288,
            "elementwise_elements_per_program": 0,
            "elementwise_scalar_temporaries": 0,
            "group_barrier_sites_before": 2,
            "group_barrier_sites_after": 2,
            "independent_subgroups": false,
            "normalized_cost": 168,
            "normalized_kernel_cost": 2016,
            "matrix_issues": 0,
            "shared_fragment_transfers": 0,
            "direct_fragment_stores": 0,
            "metal_mpp_operations": 0,
            "memory_fragment_reads": 0,
            "lhs_footprint_fragments": 0,
            "rhs_footprint_fragments": 0,
            "accumulator_initializations": 0,
            "tile_aspect_fragments": 0,
            "local_row_aspect_issues": 0,
            "local_column_aspect_issues": 0,
            "independent_elements": 18432,
            "fragment_scalars_per_lane": 0,
            "concurrent_waves": 12,
            "matrices": []
          }
        ],
        "device_timing": {
          "method": "metal_compute_pass_timestamps_v1",
          "scope": "sum_of_compute_encoder_gpu_intervals",
          "host_samples_instrumented": false,
          "repetitions": 64,
          "throughput": [
            {
              "compute_ns": 3742208,
              "compute_span_ns": 3742208,
              "calibration_cpu_ns": 4243709,
              "calibration_gpu_ticks": 4243709,
              "compute_passes": 1,
              "command_buffer_ns": 3742250.03179,
              "command_buffers": 2
            },
            {
              "compute_ns": 3684875,
              "compute_span_ns": 3684875,
              "calibration_cpu_ns": 4226416,
              "calibration_gpu_ticks": 4226416,
              "compute_passes": 1,
              "command_buffer_ns": 3684875.03147,
              "command_buffers": 2
            },
            {
              "compute_ns": 3664917,
              "compute_span_ns": 3664917,
              "calibration_cpu_ns": 4120583,
              "calibration_gpu_ticks": 4120583,
              "compute_passes": 1,
              "command_buffer_ns": 3664958.29076,
              "command_buffers": 2
            },
            {
              "compute_ns": 3595084,
              "compute_span_ns": 3595084,
              "calibration_cpu_ns": 4104791,
              "calibration_gpu_ticks": 4104791,
              "compute_passes": 1,
              "command_buffer_ns": 3595124.98835,
              "command_buffers": 2
            },
            {
              "compute_ns": 3543000,
              "compute_span_ns": 3543000,
              "calibration_cpu_ns": 4065125,
              "calibration_gpu_ticks": 4065125,
              "compute_passes": 1,
              "command_buffer_ns": 3542999.9698,
              "command_buffers": 2
            },
            {
              "compute_ns": 3717666,
              "compute_span_ns": 3717666,
              "calibration_cpu_ns": 4261209,
              "calibration_gpu_ticks": 4261209,
              "compute_passes": 1,
              "command_buffer_ns": 3717708.28497,
              "command_buffers": 2
            },
            {
              "compute_ns": 3563084,
              "compute_span_ns": 3563084,
              "calibration_cpu_ns": 4099750,
              "calibration_gpu_ticks": 4099750,
              "compute_passes": 1,
              "command_buffer_ns": 3563208.33089,
              "command_buffers": 2
            },
            {
              "compute_ns": 3567125,
              "compute_span_ns": 3567125,
              "calibration_cpu_ns": 4052167,
              "calibration_gpu_ticks": 4052167,
              "compute_passes": 1,
              "command_buffer_ns": 3567208.30314,
              "command_buffers": 2
            },
            {
              "compute_ns": 3577125,
              "compute_span_ns": 3577125,
              "calibration_cpu_ns": 4072500,
              "calibration_gpu_ticks": 4072500,
              "compute_passes": 1,
              "command_buffer_ns": 3577124.96771,
              "command_buffers": 2
            }
          ],
          "latency": [
            {
              "compute_ns": 87500,
              "compute_span_ns": 87500,
              "calibration_cpu_ns": 530959,
              "calibration_gpu_ticks": 530959,
              "compute_passes": 1,
              "command_buffer_ns": 87541.6444615,
              "command_buffers": 2
            },
            {
              "compute_ns": 71583,
              "compute_span_ns": 71583,
              "calibration_cpu_ns": 487250,
              "calibration_gpu_ticks": 487250,
              "compute_passes": 1,
              "command_buffer_ns": 71624.9924153,
              "command_buffers": 2
            },
            {
              "compute_ns": 70458,
              "compute_span_ns": 70458,
              "calibration_cpu_ns": 507792,
              "calibration_gpu_ticks": 507792,
              "compute_passes": 1,
              "command_buffer_ns": 70458.3362676,
              "command_buffers": 2
            },
            {
              "compute_ns": 59250,
              "compute_span_ns": 59250,
              "calibration_cpu_ns": 576625,
              "calibration_gpu_ticks": 576625,
              "compute_passes": 1,
              "command_buffer_ns": 59333.3388679,
              "command_buffers": 2
            },
            {
              "compute_ns": 72625,
              "compute_span_ns": 72625,
              "calibration_cpu_ns": 488125,
              "calibration_gpu_ticks": 488125,
              "compute_passes": 1,
              "command_buffer_ns": 72625.0000298,
              "command_buffers": 2
            },
            {
              "compute_ns": 58209,
              "compute_span_ns": 58209,
              "calibration_cpu_ns": 465041,
              "calibration_gpu_ticks": 465041,
              "compute_passes": 1,
              "command_buffer_ns": 58291.6545682,
              "command_buffers": 2
            },
            {
              "compute_ns": 58417,
              "compute_span_ns": 58417,
              "calibration_cpu_ns": 488125,
              "calibration_gpu_ticks": 488125,
              "compute_passes": 1,
              "command_buffer_ns": 58416.6846238,
              "command_buffers": 2
            },
            {
              "compute_ns": 58417,
              "compute_span_ns": 58417,
              "calibration_cpu_ns": 524875,
              "calibration_gpu_ticks": 524875,
              "compute_passes": 1,
              "command_buffer_ns": 58541.7146794,
              "command_buffers": 2
            },
            {
              "compute_ns": 71208,
              "compute_span_ns": 71208,
              "calibration_cpu_ns": 488375,
              "calibration_gpu_ticks": 488375,
              "compute_passes": 1,
              "command_buffer_ns": 71166.6652933,
              "command_buffers": 2
            }
          ],
          "control": {
            "method": "metal_command_buffer_timestamps_v1",
            "scope": "sum_of_command_buffer_gpu_intervals",
            "encoder_instrumentation": false,
            "repetitions": 64,
            "throughput": [
              {
                "command_buffer_ns": 3753458.32435,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3542124.99224,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3672958.35167,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3623208.3221,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3605791.65863,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3555666.6553,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3602833.31247,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3636708.36668,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 3542208.34561,
                "command_buffers": 2
              }
            ],
            "latency": [
              {
                "command_buffer_ns": 60333.2882747,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 58666.6865274,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 69125.0315867,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 57541.6488573,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 71374.9905117,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 58458.3031014,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 74124.9532439,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 58833.276853,
                "command_buffers": 2
              },
              {
                "command_buffer_ns": 70750.0148565,
                "command_buffers": 2
              }
            ],
            "command_buffer_throughput_us": [
              58.64778631796875,
              55.34570300375,
              57.38997424484375,
              56.6126300328125,
              56.34049466609375,
              55.5572914890625,
              56.29427050734375,
              56.823568229375,
              55.347005400156256
            ],
            "command_buffer_throughput_us_p50": 56.34049466609375,
            "command_buffer_throughput_us_p90": 57.64153665946875,
            "command_buffer_latency_us": [
              60.333288274699996,
              58.6666865274,
              69.1250315867,
              57.541648857300004,
              71.3749905117,
              58.4583031014,
              74.12495324390001,
              58.833276853,
              70.7500148565
            ],
            "command_buffer_latency_us_p50": 60.333288274699996,
            "command_buffer_latency_us_p90": 71.92498305814
          },
          "compute_throughput_us": [
            58.472,
            57.576171875,
            57.264328125,
            56.1731875,
            55.359375,
            58.08853125,
            55.6731875,
            55.736328125,
            55.892578125
          ],
          "compute_throughput_us_p50": 56.1731875,
          "compute_throughput_us_p90": 58.165225,
          "command_buffer_throughput_us": [
            58.47265674671875,
            57.576172366718744,
            57.264973293124996,
            56.17382794296876,
            55.359374528125,
            58.08919195265625,
            55.67513017015625,
            55.737629736562496,
            55.89257762046875
          ],
          "command_buffer_throughput_us_p50": 56.17382794296876,
          "command_buffer_throughput_us_p90": 58.16588491146875,
          "compute_latency_us": [
            87.5,
            71.583,
            70.458,
            59.25,
            72.625,
            58.209,
            58.417,
            58.417,
            71.208
          ],
          "compute_latency_us_p50": 70.458,
          "compute_latency_us_p90": 75.60000000000001,
          "command_buffer_latency_us": [
            87.5416444615,
            71.6249924153,
            70.45833626759999,
            59.3333388679,
            72.62500002979999,
            58.2916545682,
            58.4166846238,
            58.5417146794,
            71.1666652933
          ],
          "command_buffer_latency_us_p50": 70.45833626759999,
          "command_buffer_latency_us_p90": 75.60832891614,
          "counter_control_throughput_ratio": 0.9970138758468989,
          "counter_control_latency_ratio": 1.0175132705326957
        },
        "correctness": {
          "max_abs_error": 7.116789644580346e-11,
          "atol": 2e-06,
          "rtol": 2e-05
        },
        "throughput_us_p50": 56.7273901345,
        "throughput_us_p90": 57.7917587444,
        "latency_us_p50": 287.583,
        "latency_us_p90": 357.7586
      },
      "slowdown": 0.43559263955475075,
      "device_compute_slowdown": 0.415446482440776,
      "valid": true,
      "round": 3,
      "variant": "reference"
    }
  ]
}
