From d329454179930fa4606c2c35bc17749e9d2809be Mon Sep 17 00:00:00 2001 From: Mark Saroufim Date: Sun, 15 Nov 2026 14:00:45 -0800 Subject: [PATCH 1/4] Add opt-in NVCC framework PCH cache and release validation --- .github/workflows/modal-deploy.yml | 12 + docs/benchmarks/nvcc-pch-profiles.json | 1461 +++++ docs/benchmarks/nvcc-pch.json | 6957 ++++++++++++++++++++++++ docs/modal-pch-cache.md | 5 +- docs/nvcc-pch-benchmark.md | 214 + docs/nvcc-pch-cache.md | 205 + scripts/nvcc_pch_benchmark.py | 656 +++ scripts/upload_nvcc_toolchain.py | 52 + scripts/validate_nvcc_pch.py | 612 +++ src/libkernelbot/inline_artifacts.py | 97 +- src/libkernelbot/python_precompile.py | 5 +- src/runners/modal_runner.py | 113 +- src/runners/modal_runner_archs.py | 14 +- src/runners/nvcc_toolchain.py | 114 + src/runners/pch/nvcc.py | 128 + src/runners/pch/nvcc_common.py | 189 + src/runners/pch/nvcc_warm.py | 137 + tests/test_inline_artifacts.py | 48 +- tests/test_nvcc_pch.py | 321 ++ tests/test_nvcc_release_validation.py | 212 + tests/test_nvcc_toolchain.py | 241 + tests/test_python_precompile.py | 131 +- 22 files changed, 11902 insertions(+), 22 deletions(-) create mode 100644 docs/benchmarks/nvcc-pch-profiles.json create mode 100644 docs/benchmarks/nvcc-pch.json create mode 100644 docs/nvcc-pch-benchmark.md create mode 100644 docs/nvcc-pch-cache.md create mode 100644 scripts/nvcc_pch_benchmark.py create mode 100644 scripts/upload_nvcc_toolchain.py create mode 100644 scripts/validate_nvcc_pch.py create mode 100644 src/runners/nvcc_toolchain.py create mode 100644 src/runners/pch/nvcc.py create mode 100644 src/runners/pch/nvcc_common.py create mode 100644 src/runners/pch/nvcc_warm.py create mode 100644 tests/test_nvcc_pch.py create mode 100644 tests/test_nvcc_release_validation.py create mode 100644 tests/test_nvcc_toolchain.py diff --git a/.github/workflows/modal-deploy.yml b/.github/workflows/modal-deploy.yml index ad8bb19d..3851790c 100644 --- a/.github/workflows/modal-deploy.yml +++ b/.github/workflows/modal-deploy.yml @@ -14,6 +14,13 @@ jobs: env: MODAL_TOKEN_ID: ${{ secrets.MODAL_TOKEN_ID }} MODAL_TOKEN_SECRET: ${{ secrets.MODAL_TOKEN_SECRET }} + # Enable only after validating this toolchain in the dev leaderboard. + KERNELBOT_NVCC_TOOLCHAIN: ${{ vars.KERNELBOT_NVCC_TOOLCHAIN || '0' }} + KERNELBOT_NVCC_PCH: ${{ vars.KERNELBOT_NVCC_PCH || '0' }} + KERNELBOT_NVCC_PCH_PROFILES: ${{ vars.KERNELBOT_NVCC_PCH_PROFILES }} + KERNELBOT_NVCC_TOOLCHAIN_VOLUME: ${{ vars.KERNELBOT_NVCC_TOOLCHAIN_VOLUME || 'kernelbot-nvcc-toolchains' }} + KERNELBOT_NVCC_PCH_VOLUME: ${{ vars.KERNELBOT_NVCC_PCH_VOLUME || 'kernelbot-nvcc-pch' }} + KERNELBOT_PCH_VOLUME: ${{ vars.KERNELBOT_PCH_VOLUME || 'kernelbot-torch-pch' }} steps: - name: Checkout Repository @@ -33,6 +40,11 @@ jobs: run: | PYTHONPATH=${{ github.workspace }}/src:${{ github.workspace }}/src/runners modal run src/runners/modal_runner.py::warm_pch + - name: Warm NVCC framework header cache + if: env.KERNELBOT_NVCC_TOOLCHAIN == '1' + run: | + PYTHONPATH=${{ github.workspace }}/src:${{ github.workspace }}/src/runners modal run src/runners/modal_runner.py::warm_nvcc_pch --profiles "$KERNELBOT_NVCC_PCH_PROFILES" + - name: Deploy job run: | # TODO: Remove pythonpath here diff --git a/docs/benchmarks/nvcc-pch-profiles.json b/docs/benchmarks/nvcc-pch-profiles.json new file mode 100644 index 00000000..fc7188dc --- /dev/null +++ b/docs/benchmarks/nvcc-pch-profiles.json @@ -0,0 +1,1461 @@ +{ + "scope": "2026-11-13 private-diff snapshot only (expected 366803 unique submissions); public baseline excluded. Static candidate inventory of calls with explicit cuda_sources keyword; submission code was never executed. Counts do not represent actual NVCC argv, guaranteed cache hits, or all KernelBot submissions.", + "method": { + "flag_order": "preserved exactly", + "name_resolution": "single unconditional assignment/import in lexical scope before call; ambiguous, reassigned, mutated, aliased-to-mutated, and escaped variables unresolved", + "eligibility": "proven torch.utils.cpp_extension.load_inline origin; no kwargs unpacking; resolved nonempty CUDA source; implicit headers enabled; no custom include paths; flags entirely in listed initial allowlist", + "defaults": "missing or literal None extra_cuda_cflags means empty tuple; **kwargs never assumed default", + "runner": "known NVIDIA separated from missing or unrecognized runner; explicit AMD/CPU excluded from analysis", + "limitations": [ + "Conservative static analysis is not Python execution or a proof of runtime eligibility.", + "Only explicit cuda_sources keyword calls are inventoried; positional CUDA arguments are outside scope.", + "Dynamic sources, flags, monkeypatches, environment changes, and runtime branches can change executed behavior.", + "Top-N candidate coverage uses per-submission set union; row frequencies must not be summed as unique coverage.", + "Custom include paths, arbitrary backend/compiler flags, response files, and unknown flags are outside the initial warm-catalog allowlist. Fixed recognized NVIDIA architecture tokens and literal relocatable-device-code=false are included.", + "Source prefilter and analysis failures are reported rather than treated as default flags." + ], + "initial_allowlist": [ + "--expt-extended-lambda", + "--expt-relaxed-constexpr", + "--extended-lambda", + "--generate-code=arch=compute_100,code=sm_100", + "--generate-code=arch=compute_100a,code=sm_100a", + "--generate-code=arch=compute_103,code=sm_103", + "--generate-code=arch=compute_103a,code=sm_103a", + "--generate-code=arch=compute_120,code=sm_120", + "--generate-code=arch=compute_120a,code=sm_120a", + "--generate-code=arch=compute_75,code=sm_75", + "--generate-code=arch=compute_80,code=sm_80", + "--generate-code=arch=compute_86,code=sm_86", + "--generate-code=arch=compute_89,code=sm_89", + "--generate-code=arch=compute_90,code=sm_90", + "--generate-code=arch=compute_90a,code=sm_90a", + "--generate-line-info", + "--gpu-architecture=sm_100", + "--gpu-architecture=sm_100a", + "--gpu-architecture=sm_103", + "--gpu-architecture=sm_103a", + "--gpu-architecture=sm_120", + "--gpu-architecture=sm_120a", + "--gpu-architecture=sm_75", + "--gpu-architecture=sm_80", + "--gpu-architecture=sm_86", + "--gpu-architecture=sm_89", + "--gpu-architecture=sm_90", + "--gpu-architecture=sm_90a", + "--relocatable-device-code=false", + "--std=c++17", + "--std=c++20", + "--use_fast_math", + "-O0", + "-O1", + "-O2", + "-O3", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF_OPERATORS__", + "-arch=sm_100", + "-arch=sm_100a", + "-arch=sm_103", + "-arch=sm_103a", + "-arch=sm_120", + "-arch=sm_120a", + "-arch=sm_75", + "-arch=sm_80", + "-arch=sm_86", + "-arch=sm_89", + "-arch=sm_90", + "-arch=sm_90a", + "-gencode=arch=compute_100,code=sm_100", + "-gencode=arch=compute_100a,code=sm_100a", + "-gencode=arch=compute_103,code=sm_103", + "-gencode=arch=compute_103a,code=sm_103a", + "-gencode=arch=compute_120,code=sm_120", + "-gencode=arch=compute_120a,code=sm_120a", + "-gencode=arch=compute_75,code=sm_75", + "-gencode=arch=compute_80,code=sm_80", + "-gencode=arch=compute_86,code=sm_86", + "-gencode=arch=compute_89,code=sm_89", + "-gencode=arch=compute_90,code=sm_90", + "-gencode=arch=compute_90a,code=sm_90a", + "-lineinfo", + "-rdc=false", + "-std=c++17", + "-std=c++20" + ] + }, + "counts": { + "rows": 367118, + "unique_submissions": 366803, + "duplicate_submission_rows": 315 + }, + "cohorts": { + "unknown": { + "all_submissions": 242065, + "source_prefilter_candidates": 118885, + "static_cuda_load_inline_submissions": 112520, + "static_cuda_load_inline_calls": 133869, + "statically_resolved_flag_submissions": 70311, + "eligible_candidate_calls": 33082, + "eligible_candidate_submissions": 25996, + "parse_or_analysis_failure": 5322 + }, + "nvidia": { + "all_submissions": 124736, + "source_prefilter_candidates": 70854, + "static_cuda_load_inline_submissions": 70461, + "static_cuda_load_inline_calls": 74676, + "statically_resolved_flag_submissions": 64735, + "eligible_candidate_calls": 20717, + "eligible_candidate_submissions": 20411, + "parse_or_analysis_failure": 16 + }, + "other": { + "all_submissions": 2 + }, + "combined_analyzed": { + "all_submissions": 366801, + "source_prefilter_candidates": 189739, + "static_cuda_load_inline_submissions": 182981, + "static_cuda_load_inline_calls": 208545, + "statically_resolved_flag_submissions": 135046, + "eligible_candidate_calls": 53799, + "eligible_candidate_submissions": 46407, + "parse_or_analysis_failure": 5338 + } + }, + "coverage_warning": "Flag-tuple candidate coverage only. Exact framework prefix, generated architecture flags, include order, compiler generation, and environment must still match at runtime; these figures are not measured PCH hit rates. Known NVIDIA and unknown runner cohorts remain separate.", + "selected_default_profile_count": 18, + "selected_distinct_ordered_flag_tuples": 6, + "selection": "Six most frequent eligible ordered flag tuples in the known-NVIDIA runner cohort. Three neutral tuples are warmed across five runner architectures; the three explicit-architecture tuples each add one profile.", + "selected_default_profiles": [ + { + "name": "sm75-default", + "architecture": "75", + "extra_cuda_cflags": [] + }, + { + "name": "sm75-O3", + "architecture": "75", + "extra_cuda_cflags": [ + "-O3" + ] + }, + { + "name": "sm75-O3-fast-math", + "architecture": "75", + "extra_cuda_cflags": [ + "-O3", + "--use_fast_math" + ] + }, + { + "name": "sm80-default", + "architecture": "80", + "extra_cuda_cflags": [] + }, + { + "name": "sm80-O3", + "architecture": "80", + "extra_cuda_cflags": [ + "-O3" + ] + }, + { + "name": "sm80-O3-fast-math", + "architecture": "80", + "extra_cuda_cflags": [ + "-O3", + "--use_fast_math" + ] + }, + { + "name": "sm89-default", + "architecture": "89", + "extra_cuda_cflags": [] + }, + { + "name": "sm89-O3", + "architecture": "89", + "extra_cuda_cflags": [ + "-O3" + ] + }, + { + "name": "sm89-O3-fast-math", + "architecture": "89", + "extra_cuda_cflags": [ + "-O3", + "--use_fast_math" + ] + }, + { + "name": "sm90a-default", + "architecture": "90a", + "extra_cuda_cflags": [] + }, + { + "name": "sm90a-O3", + "architecture": "90a", + "extra_cuda_cflags": [ + "-O3" + ] + }, + { + "name": "sm90a-O3-fast-math", + "architecture": "90a", + "extra_cuda_cflags": [ + "-O3", + "--use_fast_math" + ] + }, + { + "name": "sm100-default", + "architecture": "100", + "extra_cuda_cflags": [] + }, + { + "name": "sm100-O3", + "architecture": "100", + "extra_cuda_cflags": [ + "-O3" + ] + }, + { + "name": "sm100-O3-fast-math", + "architecture": "100", + "extra_cuda_cflags": [ + "-O3", + "--use_fast_math" + ] + }, + { + "name": "sm100a-O3-fast-math-arch", + "architecture": "100a", + "extra_cuda_cflags": [ + "-O3", + "--use_fast_math", + "-arch=sm_100a" + ] + }, + { + "name": "sm100-O3-arch-fast-math", + "architecture": "100", + "extra_cuda_cflags": [ + "-O3", + "-arch=sm_100", + "--use_fast_math" + ] + }, + { + "name": "sm80-O3-cxx17-arch-fast-math", + "architecture": "80", + "extra_cuda_cflags": [ + "-O3", + "-std=c++17", + "-arch=sm_80", + "--use_fast_math" + ] + } + ], + "master_header_direct_includes": [ + "torch/types.h", + "cuda.h", + "cuda_runtime.h" + ], + "selected_catalog_flags": [ + { + "flags": [ + "-O3" + ], + "known_nvidia_unique_submissions": 5419, + "unknown_runner_unique_submissions": 10457, + "combined_unique_submissions": 15876 + }, + { + "flags": [ + "-O3", + "--use_fast_math" + ], + "known_nvidia_unique_submissions": 3358, + "unknown_runner_unique_submissions": 6695, + "combined_unique_submissions": 10053 + }, + { + "flags": [], + "known_nvidia_unique_submissions": 2670, + "unknown_runner_unique_submissions": 2556, + "combined_unique_submissions": 5226 + }, + { + "flags": [ + "-O3", + "--use_fast_math", + "-arch=sm_100a" + ], + "known_nvidia_unique_submissions": 2609, + "unknown_runner_unique_submissions": 113, + "combined_unique_submissions": 2722 + }, + { + "flags": [ + "-O3", + "-arch=sm_100", + "--use_fast_math" + ], + "known_nvidia_unique_submissions": 771, + "unknown_runner_unique_submissions": 99, + "combined_unique_submissions": 870 + }, + { + "flags": [ + "-O3", + "-std=c++17", + "-arch=sm_80", + "--use_fast_math" + ], + "known_nvidia_unique_submissions": 628, + "unknown_runner_unique_submissions": 174, + "combined_unique_submissions": 802 + } + ], + "selected_catalog_candidate_coverage": { + "unknown": { + "unique_submissions_with_any_candidate_call_covered": 18283, + "unique_submissions_with_all_observed_calls_eligible_and_covered": 16356, + "static_cuda_load_inline_submissions": 112520, + "eligible_candidate_submissions": 25996, + "fraction_of_static_cuda_load_inline_submissions": 0.1624866690366157 + }, + "nvidia": { + "unique_submissions_with_any_candidate_call_covered": 15421, + "unique_submissions_with_all_observed_calls_eligible_and_covered": 15176, + "static_cuda_load_inline_submissions": 70461, + "eligible_candidate_submissions": 20411, + "fraction_of_static_cuda_load_inline_submissions": 0.21885865940023558 + }, + "combined_analyzed": { + "unique_submissions_with_any_candidate_call_covered": 33704, + "unique_submissions_with_all_observed_calls_eligible_and_covered": 31532, + "static_cuda_load_inline_submissions": 182981, + "eligible_candidate_submissions": 46407, + "fraction_of_static_cuda_load_inline_submissions": 0.18419398735387826 + } + }, + "top_eligible_flags": { + "unknown": [ + { + "flags": [ + "-O3" + ], + "unique_submissions": 10457 + }, + { + "flags": [ + "-O3", + "--use_fast_math" + ], + "unique_submissions": 6695 + }, + { + "flags": [], + "unique_submissions": 2556 + }, + { + "flags": [ + "-O3", + "--use_fast_math", + "-gencode=arch=compute_100a,code=sm_100a", + "--expt-relaxed-constexpr" + ], + "unique_submissions": 770 + }, + { + "flags": [ + "-O3", + "--use_fast_math", + "-lineinfo", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-gencode=arch=compute_100a,code=sm_100a" + ], + "unique_submissions": 593 + }, + { + "flags": [ + "-O2" + ], + "unique_submissions": 491 + }, + { + "flags": [ + "-O2", + "--use_fast_math" + ], + "unique_submissions": 488 + }, + { + "flags": [ + "-O3", + "-gencode=arch=compute_100a,code=sm_100a", + "--use_fast_math", + "--expt-relaxed-constexpr", + "--relocatable-device-code=false" + ], + "unique_submissions": 392 + }, + { + "flags": [ + "-O3", + "--use_fast_math", + "-gencode=arch=compute_100a,code=sm_100a", + "-lineinfo", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__" + ], + "unique_submissions": 310 + }, + { + "flags": [ + "-O3", + "--use_fast_math", + "-gencode=arch=compute_100a,code=sm_100a", + "--expt-relaxed-constexpr", + "--relocatable-device-code=false" + ], + "unique_submissions": 298 + }, + { + "flags": [ + "-O3", + "-lineinfo" + ], + "unique_submissions": 248 + }, + { + "flags": [ + "-O3", + "-gencode=arch=compute_100a,code=sm_100a", + "--use_fast_math", + "--expt-relaxed-constexpr", + "--relocatable-device-code=false", + "-lineinfo" + ], + "unique_submissions": 247 + } + ], + "nvidia": [ + { + "flags": [ + "-O3" + ], + "unique_submissions": 5419 + }, + { + "flags": [ + "-O3", + "--use_fast_math" + ], + "unique_submissions": 3358 + }, + { + "flags": [], + "unique_submissions": 2670 + }, + { + "flags": [ + "-O3", + "--use_fast_math", + "-arch=sm_100a" + ], + "unique_submissions": 2609 + }, + { + "flags": [ + "-O3", + "-arch=sm_100", + "--use_fast_math" + ], + "unique_submissions": 771 + }, + { + "flags": [ + "-O3", + "-std=c++17", + "-arch=sm_80", + "--use_fast_math" + ], + "unique_submissions": 628 + }, + { + "flags": [ + "-O3", + "-gencode=arch=compute_100a,code=sm_100a", + "--use_fast_math", + "--expt-relaxed-constexpr", + "--relocatable-device-code=false", + "-lineinfo" + ], + "unique_submissions": 540 + }, + { + "flags": [ + "-O3", + "--use_fast_math", + "-arch=sm_100a", + "-std=c++17" + ], + "unique_submissions": 434 + }, + { + "flags": [ + "-O3", + "--use_fast_math", + "-lineinfo", + "-gencode=arch=compute_100,code=sm_100", + "-arch=sm_100" + ], + "unique_submissions": 362 + }, + { + "flags": [ + "-O3", + "--use_fast_math", + "-gencode=arch=compute_100a,code=sm_100a" + ], + "unique_submissions": 285 + }, + { + "flags": [ + "-O3", + "--use_fast_math", + "-gencode=arch=compute_100a,code=sm_100a", + "--expt-relaxed-constexpr", + "--relocatable-device-code=false" + ], + "unique_submissions": 282 + }, + { + "flags": [ + "-O3", + "--use_fast_math", + "-lineinfo", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-gencode=arch=compute_100a,code=sm_100a" + ], + "unique_submissions": 234 + } + ], + "combined_analyzed": [ + { + "flags": [ + "-O3" + ], + "unique_submissions": 15876 + }, + { + "flags": [ + "-O3", + "--use_fast_math" + ], + "unique_submissions": 10053 + }, + { + "flags": [], + "unique_submissions": 5226 + }, + { + "flags": [ + "-O3", + "--use_fast_math", + "-arch=sm_100a" + ], + "unique_submissions": 2722 + }, + { + "flags": [ + "-O3", + "-arch=sm_100", + "--use_fast_math" + ], + "unique_submissions": 870 + }, + { + "flags": [ + "-O3", + "--use_fast_math", + "-lineinfo", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-gencode=arch=compute_100a,code=sm_100a" + ], + "unique_submissions": 827 + }, + { + "flags": [ + "-O3", + "--use_fast_math", + "-gencode=arch=compute_100a,code=sm_100a", + "--expt-relaxed-constexpr" + ], + "unique_submissions": 826 + }, + { + "flags": [ + "-O3", + "-std=c++17", + "-arch=sm_80", + "--use_fast_math" + ], + "unique_submissions": 802 + }, + { + "flags": [ + "-O3", + "-gencode=arch=compute_100a,code=sm_100a", + "--use_fast_math", + "--expt-relaxed-constexpr", + "--relocatable-device-code=false", + "-lineinfo" + ], + "unique_submissions": 787 + }, + { + "flags": [ + "-O3", + "--use_fast_math", + "-gencode=arch=compute_100a,code=sm_100a", + "--expt-relaxed-constexpr", + "--relocatable-device-code=false" + ], + "unique_submissions": 580 + }, + { + "flags": [ + "-O2" + ], + "unique_submissions": 561 + }, + { + "flags": [ + "-O2", + "--use_fast_math" + ], + "unique_submissions": 522 + } + ] + }, + "ineligibility_reasons_unique_submissions_overlapping": { + "unknown": { + "custom_include_paths": 900, + "flags_outside_initial_allowlist": 24563, + "cuda_sources_unresolved": 21583, + "unverified_load_inline_origin": 10897, + "cuda_sources_empty": 576, + "flags_unresolved_or_ambiguous": 41236, + "no_implicit_headers_true": 52839, + "include_paths_unresolved": 3437, + "kwargs_unpacking": 1400 + }, + "nvidia": { + "custom_include_paths": 146, + "flags_outside_initial_allowlist": 7091, + "cuda_sources_unresolved": 13229, + "unverified_load_inline_origin": 6229, + "flags_unresolved_or_ambiguous": 5842, + "no_implicit_headers_true": 35586, + "include_paths_unresolved": 625, + "cuda_sources_empty": 88, + "kwargs_unpacking": 6 + }, + "combined_analyzed": { + "custom_include_paths": 1046, + "flags_outside_initial_allowlist": 31654, + "cuda_sources_unresolved": 34812, + "unverified_load_inline_origin": 17126, + "flags_unresolved_or_ambiguous": 47078, + "no_implicit_headers_true": 88425, + "include_paths_unresolved": 4062, + "cuda_sources_empty": 664, + "kwargs_unpacking": 1406 + } + }, + "source_header_count_scope": "Distinct known-NVIDIA static CUDA load_inline submissions with each explicit include in their resolved C++ or CUDA source strings; PyTorch-generated implicit includes are not counted.", + "known_nvidia_framework_headers": [ + { + "header": "cuda_runtime.h", + "unique_submissions": 50015, + "direct_master_include": true, + "cutlass_or_cute_excluded": false + }, + { + "header": "cuda_fp16.h", + "unique_submissions": 38863, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "cuda.h", + "unique_submissions": 35932, + "direct_master_include": true, + "cutlass_or_cute_excluded": false + }, + { + "header": "cudaTypedefs.h", + "unique_submissions": 29600, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "torch/library.h", + "unique_submissions": 29011, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "ATen/core/Tensor.h", + "unique_submissions": 28375, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "torch/extension.h", + "unique_submissions": 24752, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "ATen/cuda/CUDAContext.h", + "unique_submissions": 6201, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "cuda_fp4.h", + "unique_submissions": 3846, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "cuda_fp8.h", + "unique_submissions": 3768, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "cublasLt.h", + "unique_submissions": 3753, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "cublas_v2.h", + "unique_submissions": 3597, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "ATen/ATen.h", + "unique_submissions": 3448, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "cuda_bf16.h", + "unique_submissions": 2550, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "c10/cuda/CUDAGuard.h", + "unique_submissions": 1525, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "ATen/cuda/CUDABlas.h", + "unique_submissions": 1170, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "torch/csrc/autograd/python_variable.h", + "unique_submissions": 933, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "cuda/atomic", + "unique_submissions": 648, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "ATen/cuda/CUDAUtils.h", + "unique_submissions": 491, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "c10/cuda/CUDAException.h", + "unique_submissions": 418, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "cuda/ptx", + "unique_submissions": 320, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "c10/cuda/CUDACachingAllocator.h", + "unique_submissions": 318, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "cuda_pipeline.h", + "unique_submissions": 312, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "torch/torch.h", + "unique_submissions": 277, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "cutlass/cutlass.h", + "unique_submissions": 263, + "direct_master_include": false, + "cutlass_or_cute_excluded": true + }, + { + "header": "cutlass/gemm/kernel/gemm_universal.hpp", + "unique_submissions": 244, + "direct_master_include": false, + "cutlass_or_cute_excluded": true + }, + { + "header": "cutlass/gemm/collective/collective_builder.hpp", + "unique_submissions": 244, + "direct_master_include": false, + "cutlass_or_cute_excluded": true + }, + { + "header": "cutlass/epilogue/collective/collective_builder.hpp", + "unique_submissions": 244, + "direct_master_include": false, + "cutlass_or_cute_excluded": true + }, + { + "header": "cutlass/gemm/device/gemm_universal_adapter.h", + "unique_submissions": 244, + "direct_master_include": false, + "cutlass_or_cute_excluded": true + }, + { + "header": "cutlass/util/packed_stride.hpp", + "unique_submissions": 243, + "direct_master_include": false, + "cutlass_or_cute_excluded": true + }, + { + "header": "cute/tensor.hpp", + "unique_submissions": 232, + "direct_master_include": false, + "cutlass_or_cute_excluded": true + }, + { + "header": "cuda/pipeline", + "unique_submissions": 195, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "cuda/barrier", + "unique_submissions": 159, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "torch/types.h", + "unique_submissions": 159, + "direct_master_include": true, + "cutlass_or_cute_excluded": false + }, + { + "header": "cutlass/gemm/dispatch_policy.hpp", + "unique_submissions": 157, + "direct_master_include": false, + "cutlass_or_cute_excluded": true + }, + { + "header": "ATen/ops/cudnn_convolution.h", + "unique_submissions": 140, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "cutlass/detail/sm100_blockscaled_layout.hpp", + "unique_submissions": 137, + "direct_master_include": false, + "cutlass_or_cute_excluded": true + }, + { + "header": "c10/cuda/CUDAStream.h", + "unique_submissions": 74, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "ATen/cuda/CUDAContextLight.h", + "unique_submissions": 71, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + }, + { + "header": "cuda_runtime_api.h", + "unique_submissions": 64, + "direct_master_include": false, + "cutlass_or_cute_excluded": false + } + ], + "runner_eligible_flags": [ + { + "runner_class": "unknown", + "flags": [ + "-O3" + ], + "unique_submissions": 10457 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "--use_fast_math" + ], + "unique_submissions": 6695 + }, + { + "runner_class": "B200", + "flags": [ + "-O3" + ], + "unique_submissions": 2613 + }, + { + "runner_class": "B200", + "flags": [ + "-O3", + "--use_fast_math", + "-arch=sm_100a" + ], + "unique_submissions": 2608 + }, + { + "runner_class": "unknown", + "flags": [], + "unique_submissions": 2556 + }, + { + "runner_class": "A100", + "flags": [ + "-O3" + ], + "unique_submissions": 1862 + }, + { + "runner_class": "B200", + "flags": [], + "unique_submissions": 1054 + }, + { + "runner_class": "B200", + "flags": [ + "-O3", + "--use_fast_math" + ], + "unique_submissions": 922 + }, + { + "runner_class": "A100", + "flags": [ + "-O3", + "--use_fast_math" + ], + "unique_submissions": 794 + }, + { + "runner_class": "B200", + "flags": [ + "-O3", + "-arch=sm_100", + "--use_fast_math" + ], + "unique_submissions": 771 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "--use_fast_math", + "-gencode=arch=compute_100a,code=sm_100a", + "--expt-relaxed-constexpr" + ], + "unique_submissions": 770 + }, + { + "runner_class": "NVIDIA_OTHER", + "flags": [ + "-O3", + "--use_fast_math" + ], + "unique_submissions": 737 + }, + { + "runner_class": "A100", + "flags": [], + "unique_submissions": 688 + }, + { + "runner_class": "A100", + "flags": [ + "-O3", + "-std=c++17", + "-arch=sm_80", + "--use_fast_math" + ], + "unique_submissions": 628 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "--use_fast_math", + "-lineinfo", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-gencode=arch=compute_100a,code=sm_100a" + ], + "unique_submissions": 593 + }, + { + "runner_class": "H100", + "flags": [], + "unique_submissions": 563 + }, + { + "runner_class": "NVIDIA_OTHER", + "flags": [ + "-O3", + "-gencode=arch=compute_100a,code=sm_100a", + "--use_fast_math", + "--expt-relaxed-constexpr", + "--relocatable-device-code=false", + "-lineinfo" + ], + "unique_submissions": 540 + }, + { + "runner_class": "L4", + "flags": [ + "-O3" + ], + "unique_submissions": 507 + }, + { + "runner_class": "unknown", + "flags": [ + "-O2" + ], + "unique_submissions": 491 + }, + { + "runner_class": "unknown", + "flags": [ + "-O2", + "--use_fast_math" + ], + "unique_submissions": 488 + }, + { + "runner_class": "L4", + "flags": [ + "-O3", + "--use_fast_math" + ], + "unique_submissions": 469 + }, + { + "runner_class": "NVIDIA_OTHER", + "flags": [ + "-O3", + "--use_fast_math", + "-arch=sm_100a", + "-std=c++17" + ], + "unique_submissions": 434 + }, + { + "runner_class": "H100", + "flags": [ + "-O3", + "--use_fast_math" + ], + "unique_submissions": 431 + }, + { + "runner_class": "H100", + "flags": [ + "-O3" + ], + "unique_submissions": 414 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "-gencode=arch=compute_100a,code=sm_100a", + "--use_fast_math", + "--expt-relaxed-constexpr", + "--relocatable-device-code=false" + ], + "unique_submissions": 392 + }, + { + "runner_class": "NVIDIA_OTHER", + "flags": [ + "-O3", + "--use_fast_math", + "-lineinfo", + "-gencode=arch=compute_100,code=sm_100", + "-arch=sm_100" + ], + "unique_submissions": 362 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "--use_fast_math", + "-gencode=arch=compute_100a,code=sm_100a", + "-lineinfo", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__" + ], + "unique_submissions": 310 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "--use_fast_math", + "-gencode=arch=compute_100a,code=sm_100a", + "--expt-relaxed-constexpr", + "--relocatable-device-code=false" + ], + "unique_submissions": 298 + }, + { + "runner_class": "T4", + "flags": [], + "unique_submissions": 286 + }, + { + "runner_class": "B200", + "flags": [ + "-O3", + "--use_fast_math", + "-gencode=arch=compute_100a,code=sm_100a" + ], + "unique_submissions": 285 + }, + { + "runner_class": "B200", + "flags": [ + "-O3", + "--use_fast_math", + "-gencode=arch=compute_100a,code=sm_100a", + "--expt-relaxed-constexpr", + "--relocatable-device-code=false" + ], + "unique_submissions": 265 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "-lineinfo" + ], + "unique_submissions": 248 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "-gencode=arch=compute_100a,code=sm_100a", + "--use_fast_math", + "--expt-relaxed-constexpr", + "--relocatable-device-code=false", + "-lineinfo" + ], + "unique_submissions": 247 + }, + { + "runner_class": "NVIDIA_OTHER", + "flags": [ + "-O3", + "--use_fast_math", + "-lineinfo", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-gencode=arch=compute_100a,code=sm_100a" + ], + "unique_submissions": 234 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "--use_fast_math", + "-lineinfo" + ], + "unique_submissions": 227 + }, + { + "runner_class": "B200", + "flags": [ + "-O3", + "-std=c++17", + "-gencode=arch=compute_100a,code=sm_100a", + "--expt-relaxed-constexpr" + ], + "unique_submissions": 223 + }, + { + "runner_class": "unknown", + "flags": [ + "-O2", + "--use_fast_math", + "-gencode=arch=compute_100a,code=sm_100a" + ], + "unique_submissions": 217 + }, + { + "runner_class": "B200", + "flags": [ + "-O3", + "-lineinfo" + ], + "unique_submissions": 198 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "-std=c++17" + ], + "unique_submissions": 192 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "-gencode=arch=compute_100a,code=sm_100a" + ], + "unique_submissions": 186 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "-arch=sm_100a", + "--use_fast_math" + ], + "unique_submissions": 185 + }, + { + "runner_class": "B200", + "flags": [ + "-O3", + "-arch=sm_100" + ], + "unique_submissions": 179 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "-std=c++17", + "-arch=sm_80", + "--use_fast_math" + ], + "unique_submissions": 174 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "--use_fast_math", + "-std=c++17", + "-gencode=arch=compute_100a,code=sm_100a" + ], + "unique_submissions": 141 + }, + { + "runner_class": "NVIDIA_OTHER", + "flags": [ + "-O3", + "-std=c++17", + "--use_fast_math", + "-lineinfo", + "-arch=sm_100a" + ], + "unique_submissions": 141 + }, + { + "runner_class": "NVIDIA_OTHER", + "flags": [ + "-O3", + "--use_fast_math", + "-arch=sm_100a", + "-std=c++17", + "-lineinfo", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__" + ], + "unique_submissions": 138 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "--use_fast_math", + "-gencode=arch=compute_100a,code=sm_100a", + "--expt-relaxed-constexpr", + "--std=c++20" + ], + "unique_submissions": 133 + }, + { + "runner_class": "A100", + "flags": [ + "-O3", + "-arch=sm_80" + ], + "unique_submissions": 131 + }, + { + "runner_class": "A100", + "flags": [ + "-arch=sm_80" + ], + "unique_submissions": 113 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "--use_fast_math", + "-arch=sm_100a" + ], + "unique_submissions": 113 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "--std=c++17" + ], + "unique_submissions": 113 + }, + { + "runner_class": "unknown", + "flags": [ + "-arch=sm_80" + ], + "unique_submissions": 112 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "--use_fast_math", + "-std=c++17" + ], + "unique_submissions": 109 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "-std=c++17", + "-gencode=arch=compute_100a,code=sm_100a", + "--expt-relaxed-constexpr" + ], + "unique_submissions": 107 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "--use_fast_math", + "-arch=sm_100" + ], + "unique_submissions": 104 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "-arch=sm_100", + "--use_fast_math" + ], + "unique_submissions": 99 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "--use_fast_math", + "-gencode=arch=compute_100a,code=sm_100a" + ], + "unique_submissions": 95 + }, + { + "runner_class": "unknown", + "flags": [ + "-O3", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_100a,code=sm_100a" + ], + "unique_submissions": 93 + }, + { + "runner_class": "NVIDIA_OTHER", + "flags": [ + "-O3", + "--use_fast_math", + "-gencode=arch=compute_100a,code=sm_100a", + "-lineinfo", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__" + ], + "unique_submissions": 93 + }, + { + "runner_class": "unknown", + "flags": [ + "-arch=sm_89", + "-O3", + "--use_fast_math" + ], + "unique_submissions": 86 + } + ] +} diff --git a/docs/benchmarks/nvcc-pch.json b/docs/benchmarks/nvcc-pch.json new file mode 100644 index 00000000..2dc7192c --- /dev/null +++ b/docs/benchmarks/nvcc-pch.json @@ -0,0 +1,6957 @@ +{ + "summary": { + "primary": [ + { + "case": "cutlass", + "mode": "patched_all_pch", + "n": 4, + "median_seconds": 34.026361068499995, + "min_seconds": 28.440589126, + "max_seconds": 43.662042164, + "seconds": [ + 43.662042164, + 39.165971053, + 28.886751083999997, + 28.440589126 + ] + }, + { + "case": "cutlass", + "mode": "patched_host_pch", + "n": 4, + "median_seconds": 45.7583980275, + "min_seconds": 41.732421651, + "max_seconds": 51.061138988, + "seconds": [ + 51.061138988, + 42.159665512000004, + 41.732421651, + 49.357130543 + ] + }, + { + "case": "cutlass", + "mode": "stock_host_pch", + "n": 4, + "median_seconds": 43.724847137, + "min_seconds": 41.497131416, + "max_seconds": 48.826632251999996, + "seconds": [ + 48.826632251999996, + 41.497131416, + 43.52756049, + 43.922133784 + ] + }, + { + "case": "cutlass", + "mode": "stock_no_pch", + "n": 4, + "median_seconds": 48.817288012, + "min_seconds": 42.949985459, + "max_seconds": 54.309780212999996, + "seconds": [ + 54.136435518, + 42.949985459, + 54.309780212999996, + 43.498140506 + ] + }, + { + "case": "simple", + "mode": "patched_all_pch", + "n": 4, + "median_seconds": 24.172627273, + "min_seconds": 23.934274335, + "max_seconds": 29.137589784, + "seconds": [ + 24.183967813, + 24.161286733, + 23.934274335, + 29.137589784 + ] + }, + { + "case": "simple", + "mode": "patched_host_pch", + "n": 4, + "median_seconds": 36.013780205, + "min_seconds": 34.736912296, + "max_seconds": 38.467215558, + "seconds": [ + 34.736912296, + 35.908356786, + 36.119203624, + 38.467215558 + ] + }, + { + "case": "simple", + "mode": "stock_host_pch", + "n": 4, + "median_seconds": 35.3712442095, + "min_seconds": 35.025937535000004, + "max_seconds": 36.431860271, + "seconds": [ + 36.431860271, + 35.256968721999996, + 35.485519697, + 35.025937535000004 + ] + }, + { + "case": "simple", + "mode": "stock_no_pch", + "n": 4, + "median_seconds": 37.2270242755, + "min_seconds": 35.746856564, + "max_seconds": 42.885632301, + "seconds": [ + 36.414967365, + 42.885632301, + 38.039081186, + 35.746856564 + ] + } + ], + "supplemental": [ + { + "case": "cutlass", + "mode": "patched_all_pch", + "n": 2, + "median_seconds": 32.899585346500004, + "min_seconds": 32.528592897, + "max_seconds": 33.270577796, + "seconds": [ + 32.528592897, + 33.270577796 + ] + }, + { + "case": "cutlass", + "mode": "patched_all_pch_prefetch", + "n": 2, + "median_seconds": 35.439373548499994, + "min_seconds": 30.342368112999996, + "max_seconds": 40.536378983999995, + "seconds": [ + 30.342368112999996, + 40.536378983999995 + ] + }, + { + "case": "cutlass", + "mode": "patched_host_pch", + "n": 2, + "median_seconds": 50.212558552000004, + "min_seconds": 49.549131939, + "max_seconds": 50.875985165, + "seconds": [ + 50.875985165, + 49.549131939 + ] + }, + { + "case": "simple", + "mode": "patched_all_pch", + "n": 2, + "median_seconds": 25.238124405, + "min_seconds": 22.962655791, + "max_seconds": 27.513593019, + "seconds": [ + 27.513593019, + 22.962655791 + ] + }, + { + "case": "simple", + "mode": "patched_all_pch_prefetch", + "n": 2, + "median_seconds": 25.7022750095, + "min_seconds": 25.672850840000002, + "max_seconds": 25.731699179, + "seconds": [ + 25.731699179, + 25.672850840000002 + ] + }, + { + "case": "simple", + "mode": "patched_host_pch", + "n": 2, + "median_seconds": 40.63829330600001, + "min_seconds": 37.205667172000005, + "max_seconds": 44.070919440000004, + "seconds": [ + 37.205667172000005, + 44.070919440000004 + ] + } + ], + "production_cuda133": [ + { + "case": "cutlass_gemm_implicit", + "enabled": false, + "n": 4, + "median_seconds": 48.7533205525, + "min_seconds": 46.312597678, + "max_seconds": 54.753927797, + "seconds": [ + 54.753927797, + 46.312597678, + 51.057156396, + 46.449484709 + ] + }, + { + "case": "cutlass_gemm_implicit", + "enabled": true, + "n": 4, + "median_seconds": 45.322798237, + "min_seconds": 43.304864159, + "max_seconds": 55.171705347, + "seconds": [ + 55.171705347, + 46.287869813, + 43.304864159, + 44.357726661 + ] + }, + { + "case": "inline_simple", + "enabled": false, + "n": 4, + "median_seconds": 20.7435791125, + "min_seconds": 19.752127795, + "max_seconds": 21.796966691, + "seconds": [ + 21.796966691, + 19.861022294999998, + 19.752127795, + 21.62613593 + ] + }, + { + "case": "inline_simple", + "enabled": true, + "n": 4, + "median_seconds": 10.365594108499998, + "min_seconds": 10.092071702999998, + "max_seconds": 11.21638811, + "seconds": [ + 10.092071702999998, + 10.122365914999998, + 11.21638811, + 10.608822302 + ] + } + ], + "validated_timed_pch_builds": 18, + "fresh_gpu_containers": 62, + "correctness_checks": 492, + "max_absolute_error": 8.529314072802663e-06 + }, + "validation": { + "timed_pch_proof": [ + { + "container_id": "ta-01M7GG9SJ3GHSRNQV0Z0ACJC3R", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ] + }, + { + "container_id": "ta-01M7GGBPJYFMHYHKCZZFP9G16R", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ] + }, + { + "container_id": "ta-01M7GGJYZRTAM0NXNGM9XQ7FTR", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" + ] + }, + { + "container_id": "ta-01M7GGKGST0M04EKKWTZSZAZMR", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" + ] + }, + { + "container_id": "ta-01M7GGP9SHQZVXA5GZ2ZE3HXKR", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ] + }, + { + "container_id": "ta-01M7GGSSKT9DDF905Z9S6J9SER", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ] + }, + { + "container_id": "ta-01M7GGT09WZH9JV14JHDV0YQVR", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" + ] + }, + { + "container_id": "ta-01M7GGY1T72T9C66SB78QVCPJR", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" + ] + }, + { + "container_id": "ta-01M7GGED50E6R503HVDSM8MGRR", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ] + }, + { + "container_id": "ta-01M7GGED5AZC6SWFBFCVMC7C5R", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ] + }, + { + "container_id": "ta-01M7GGED5AR5589PP71ECYZT5R", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ] + }, + { + "container_id": "ta-01M7GGGJTBKKMAZG8E43P3FYXR", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ] + }, + { + "container_id": "ta-01M7GGH8YZHK6M1B0FWN50S9BR", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" + ] + }, + { + "container_id": "ta-01M7GGJEA1CBKA78JQGA7X6YTR", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" + ] + }, + { + "container_id": "ta-01M7GGKN4QWSM62X7YPQY4B47R", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" + ] + }, + { + "container_id": "ta-01M7GGM20MGZZK7K4NM6A452NR", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" + ] + }, + { + "container_id": "ta-01M7GG5X2J0T5Z7DF200Z9AWXR", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ] + }, + { + "container_id": "ta-01M7GG5X2JWF1E5D985MF82VDR", + "timed_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ] + } + ] + }, + "primary": { + "probe": { + "stock_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patched_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "paths": "/usr/local/cuda/bin/nvcc.profile\n/usr/local/cuda/nvvm/bin/cicc\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "nvcc_profiles": { + "/usr/local/cuda/bin/nvcc.profile": "\nTOP = $(_HERE_)/..\n\nCICC_PATH = $(TOP)/nvvm-next/bin\nNVVMIR_LIBRARY_DIR = $(TOP)/nvvm/libdevice\n\nLD_LIBRARY_PATH += $(TOP)/lib:\nPATH += $(CICC_PATH):$(_HERE_):\n\nINCLUDES += \"-I$(TOP)/$(_TARGET_DIR_)/include\" $(_SPACE_)\nSYSTEM_INCLUDES += \"-isystem\" \"$(TOP)/$(_TARGET_DIR_)/include/cccl\" $(_SPACE_)\n\nLIBRARIES =+ $(_SPACE_) \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)/stubs\" \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)\"\n\nCUDAFE_FLAGS +=\nPTXAS_FLAGS +=\n" + }, + "pch_help": " Specify options directly to library manager.\n\n--pch --pch \n Enable precompiled header (PCH) processing. When specified, each compilation\n searches the PCH directory (see --pch-dir) for a reusable precompiled header,\n and creates a new one when none is found.\n\n--pch-dir --pch-dir \n Specify the directory in which precompiled header files are read from and\n written to. By default this is the current working directory. Ignored unless\n --pch, --create-pch, or --use-pch is enabled.\n\n--pch-verbose --pch-verbose \n Print (to stderr) messages related to precompiled header processing. Ignored\n unless --pch, --create-pch, or --use-pch is enabled.\n\n--create-pch --create-pch \n Manually create a precompiled header (PCH) at the first PCH terminal stop-point\n (i.e., #pragma hdrstop) and save it under the specified name. Intermediate\n stop points (#pragma hdrstop++) are ignored. The directory into which the\n PCH file is written is determined by --pch-dir (or is the current working\n directory if not specified). This option forces a PCH file to be created\n (any existing PCH file with the same name is overwritten). This option is\n mutually exclusive with '--use-pch' and '--pch'.\n\n--use-pch --use-pch \n If a precompiled header (PCH) with the specified name exists in the directory\n determined by '--pch-dir' (or the current working directory if not specified),\n try to use it in compilation.This option is mutually exclusive with '--create-pch'\n and '--pch'.\n\n--ptxas-options ,... (-Xptxas) \n Specify options directly to ptxas, the PTX optimizing assembler.\n\n--nvlink-options ,... (-Xnvlink) \n Specify options directly to nvlink.\n\n--nvprune-options ,... (-Xnvprune) \n Specify options directly to nvprune (used with -prune).\n\n\nMiscellaneous options for guiding the compiler driver.\n======================================================", + "strace": null, + "smoke": [ + { + "arch": "sm_75", + "command": [ + "/usr/local/cuda/bin/nvcc", + "-arch=sm_75", + "-c", + "/tmp/tmpw1rin_68/smoke.cu", + "-o", + "/tmp/tmpw1rin_68/sm_75.o" + ], + "dryrun_returncode": 0, + "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpw1rin_68/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw1rin_68/smoke.cu\" --orig_src_path_name \"/tmp/tmpw1rin_68/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpw1rin_68/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw1rin_68/smoke.cu\" --orig_src_path_name \"/tmp/tmpw1rin_68/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_0000001e_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_0000001e_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_0000001e_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpw1rin_68/sm_75.o\" \n", + "next_backend_selected": true, + "returncode": 0, + "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpw1rin_68/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw1rin_68/smoke.cu\" --orig_src_path_name \"/tmp/tmpw1rin_68/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpw1rin_68/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw1rin_68/smoke.cu\" --orig_src_path_name \"/tmp/tmpw1rin_68/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000021_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000021_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000021_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpw1rin_68/sm_75.o\" \n", + "object_bytes": 10200 + }, + { + "arch": "sm_100", + "command": [ + "/usr/local/cuda/bin/nvcc", + "-arch=sm_100", + "-c", + "/tmp/tmpw1rin_68/smoke.cu", + "-o", + "/tmp/tmpw1rin_68/sm_100.o" + ], + "dryrun_returncode": 0, + "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpw1rin_68/smoke.cu\" -o \"/tmp/tmpxft_00000036_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw1rin_68/smoke.cu\" --orig_src_path_name \"/tmp/tmpw1rin_68/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000036_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000036_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000036_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpw1rin_68/smoke.cu\" -o \"/tmp/tmpxft_00000036_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw1rin_68/smoke.cu\" --orig_src_path_name \"/tmp/tmpw1rin_68/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000036_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000036_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000036_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000036_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000036_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000036_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000036_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000036_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000036_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000036_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpw1rin_68/sm_100.o\" \n", + "next_backend_selected": true, + "returncode": 0, + "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpw1rin_68/smoke.cu\" -o \"/tmp/tmpxft_00000039_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw1rin_68/smoke.cu\" --orig_src_path_name \"/tmp/tmpw1rin_68/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000039_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000039_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000039_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpw1rin_68/smoke.cu\" -o \"/tmp/tmpxft_00000039_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw1rin_68/smoke.cu\" --orig_src_path_name \"/tmp/tmpw1rin_68/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000039_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000039_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000039_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000039_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000039_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000039_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000039_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000039_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000039_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000039_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpw1rin_68/sm_100.o\" \n", + "object_bytes": 12504 + } + ], + "compatible": true + }, + "rows": [ + { + "case": "simple", + "variant": 0, + "mode": "stock_no_pch", + "repeat": 0, + "extension_name": "kb_nvcc_simple_0", + "returncode": 0, + "log": "[1/3] /tmp/tmp_wiyksu7/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp_wiyksu7/main.cpp -o main.o \n[2/3] /tmp/tmp_wiyksu7/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmp_wiyksu7/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmp_wiyksu7/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", + "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", + "container_id": "ta-01M7GG9R0KZAS84KFWFH0WTG6R", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patch": null, + "pch_entry": null, + "process_seconds": 40.325069411, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp_wiyksu7/main.cpp", + "-o", + "main.o" + ], + "seconds": 19.397623978, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp_wiyksu7/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 35.671998917, + "returncode": 0 + } + ], + "seconds": 36.414967365, + "outputs": [ + 2.0, + 2.0, + 2.0, + 2.0, + 2.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": false, + "host_hit_keys": [], + "remote_seconds": 58.20478887 + }, + { + "case": "simple", + "variant": 0, + "mode": "stock_host_pch", + "repeat": 0, + "extension_name": "kb_nvcc_simple_0", + "returncode": 0, + "log": "[1/3] /tmp/tmpz0wlsxxa/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpz0wlsxxa/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpz0wlsxxa/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpz0wlsxxa/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpz0wlsxxa/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", + "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", + "container_id": "ta-01M7GG9SJ3NSEV91JE6RJNCHTR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patch": null, + "pch_entry": null, + "process_seconds": 40.810135701, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpz0wlsxxa/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.101947039, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpz0wlsxxa/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 35.389258092, + "returncode": 0 + } + ], + "seconds": 36.431860271, + "outputs": [ + 2.0, + 2.0, + 2.0, + 2.0, + 2.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "remote_seconds": 48.346788907000004 + }, + { + "case": "simple", + "variant": 0, + "mode": "patched_host_pch", + "repeat": 0, + "extension_name": "kb_nvcc_simple_0", + "returncode": 0, + "log": "[1/3] /tmp/tmpqhcvxumg/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpqhcvxumg/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpqhcvxumg/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpqhcvxumg/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpqhcvxumg/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", + "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", + "container_id": "ta-01M7GG9SJE0C3GB9N1TYQPQWNR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": null, + "process_seconds": 38.605369446, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpqhcvxumg/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.167082585, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpqhcvxumg/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 33.756942244, + "returncode": 0 + } + ], + "seconds": 34.736912296, + "outputs": [ + 2.0, + 2.0, + 2.0, + 2.0, + 2.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "remote_seconds": 47.10341623 + }, + { + "case": "simple", + "variant": 0, + "mode": "patched_all_pch", + "repeat": 0, + "extension_name": "kb_nvcc_simple_0", + "returncode": 0, + "log": "[1/3] /tmp/tmpxpe57igv/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpxpe57igv/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpxpe57igv/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpxpe57igv/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000031_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000031_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmpxpe57igv/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", + "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", + "container_id": "ta-01M7GG9SJ3GHSRNQV0Z0ACJC3R", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", + "process_seconds": 28.188046862, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpxpe57igv/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.079369766, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpxpe57igv/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" + ], + "seconds": 23.198122242, + "returncode": 0 + } + ], + "seconds": 24.183967813, + "outputs": [ + 2.0, + 2.0, + 2.0, + 2.0, + 2.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmpxpe57igv/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmpxpe57igv/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxpe57igv/cuda.cu\" --orig_src_path_name \"/tmp/tmpxpe57igv/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxpe57igv/cuda.cu\" --orig_src_path_name \"/tmp/tmpxpe57igv/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpxpe57igv/diagnostic.cuda.o\" \n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 58.327971263 + }, + { + "case": "simple", + "variant": 1, + "mode": "patched_all_pch", + "repeat": 0, + "extension_name": "kb_nvcc_simple_1", + "returncode": 0, + "log": "[1/3] /tmp/tmp9723ol1n/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp9723ol1n/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp9723ol1n/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmp9723ol1n/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000031_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000031_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmp9723ol1n/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", + "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", + "container_id": "ta-01M7GGBPJYFMHYHKCZZFP9G16R", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", + "process_seconds": 27.839054124999997, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp9723ol1n/main.cpp", + "-o", + "main.o" + ], + "seconds": 7.848992351999999, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp9723ol1n/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" + ], + "seconds": 23.174968698, + "returncode": 0 + } + ], + "seconds": 24.161286733, + "outputs": [ + 3.0, + 3.0, + 3.0, + 3.0, + 3.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp9723ol1n/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp9723ol1n/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp9723ol1n/cuda.cu\" --orig_src_path_name \"/tmp/tmp9723ol1n/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp9723ol1n/cuda.cu\" --orig_src_path_name \"/tmp/tmp9723ol1n/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmp9723ol1n/diagnostic.cuda.o\" \n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 56.157136401 + }, + { + "case": "simple", + "variant": 1, + "mode": "patched_host_pch", + "repeat": 0, + "extension_name": "kb_nvcc_simple_1", + "returncode": 0, + "log": "[1/3] /tmp/tmp4eloe17n/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp4eloe17n/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp4eloe17n/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmp4eloe17n/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmp4eloe17n/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", + "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", + "container_id": "ta-01M7GGGG38BJT0QQKXKPVXKHBR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": null, + "process_seconds": 39.721209833, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp4eloe17n/main.cpp", + "-o", + "main.o" + ], + "seconds": 7.6305212120000006, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp4eloe17n/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 34.912277029, + "returncode": 0 + } + ], + "seconds": 35.908356786, + "outputs": [ + 3.0, + 3.0, + 3.0, + 3.0, + 3.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "remote_seconds": 49.137765439000006 + }, + { + "case": "simple", + "variant": 1, + "mode": "stock_host_pch", + "repeat": 0, + "extension_name": "kb_nvcc_simple_1", + "returncode": 0, + "log": "[1/3] /tmp/tmpsrn85hw5/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpsrn85hw5/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpsrn85hw5/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpsrn85hw5/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpsrn85hw5/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", + "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", + "container_id": "ta-01M7GGGHRFMCBN61Z5SVD1QG4R", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patch": null, + "pch_entry": null, + "process_seconds": 38.687573193, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpsrn85hw5/main.cpp", + "-o", + "main.o" + ], + "seconds": 7.539243877, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpsrn85hw5/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 34.245309142, + "returncode": 0 + } + ], + "seconds": 35.256968721999996, + "outputs": [ + 3.0, + 3.0, + 3.0, + 3.0, + 3.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "remote_seconds": 45.988208858 + }, + { + "case": "simple", + "variant": 1, + "mode": "stock_no_pch", + "repeat": 0, + "extension_name": "kb_nvcc_simple_1", + "returncode": 0, + "log": "[1/3] /tmp/tmpjbfcorzj/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpjbfcorzj/main.cpp -o main.o \n[2/3] /tmp/tmpjbfcorzj/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpjbfcorzj/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpjbfcorzj/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", + "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", + "container_id": "ta-01M7GGGKF3C7V0H9KR3G5ESY1R", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patch": null, + "pch_entry": null, + "process_seconds": 47.066089143999996, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpjbfcorzj/main.cpp", + "-o", + "main.o" + ], + "seconds": 23.122218314, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpjbfcorzj/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 41.880367325, + "returncode": 0 + } + ], + "seconds": 42.885632301, + "outputs": [ + 3.0, + 3.0, + 3.0, + 3.0, + 3.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": false, + "host_hit_keys": [], + "remote_seconds": 67.5727541 + }, + { + "case": "cutlass", + "variant": 0, + "mode": "stock_no_pch", + "repeat": 0, + "extension_name": "kb_nvcc_cutlass_0", + "returncode": 0, + "log": "[1/3] /tmp/tmp11nt69um/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp11nt69um/main.cpp -o main.o \n[2/3] /tmp/tmp11nt69um/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmp11nt69um/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmp11nt69um/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", + "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", + "container_id": "ta-01M7GGGKV69BR167MB9GHB1FWR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patch": null, + "pch_entry": null, + "process_seconds": 60.260855773, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp11nt69um/main.cpp", + "-o", + "main.o" + ], + "seconds": 24.870952887999998, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmp11nt69um/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 53.1093533, + "returncode": 0 + } + ], + "seconds": 54.136435518, + "outputs": [ + 8.0 + ], + "max_absolute_error": 4.2646570364013314e-06, + "correctness_checks": 10, + "host_pch_consumed": false, + "host_hit_keys": [], + "remote_seconds": 82.230074779 + }, + { + "case": "cutlass", + "variant": 0, + "mode": "stock_host_pch", + "repeat": 0, + "extension_name": "kb_nvcc_cutlass_0", + "returncode": 0, + "log": "[1/3] /tmp/tmpqd3a7m0i/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpqd3a7m0i/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpqd3a7m0i/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpqd3a7m0i/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpqd3a7m0i/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", + "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", + "container_id": "ta-01M7GGJ4Y2SED4Z9ZS48M6HFPR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patch": null, + "pch_entry": null, + "process_seconds": 53.059010687, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpqd3a7m0i/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.62824865, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmpqd3a7m0i/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 47.631198614, + "returncode": 0 + } + ], + "seconds": 48.826632251999996, + "outputs": [ + 8.0 + ], + "max_absolute_error": 4.2646570364013314e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "remote_seconds": 61.13100136399999 + }, + { + "case": "cutlass", + "variant": 0, + "mode": "patched_host_pch", + "repeat": 0, + "extension_name": "kb_nvcc_cutlass_0", + "returncode": 0, + "log": "[1/3] /tmp/tmppx4mtayi/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmppx4mtayi/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmppx4mtayi/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmppx4mtayi/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmppx4mtayi/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", + "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", + "container_id": "ta-01M7GGJ6GEPN98RD1T5H1HNMTR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": null, + "process_seconds": 56.536430032999995, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmppx4mtayi/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.464728404999999, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmppx4mtayi/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 49.374024531, + "returncode": 0 + } + ], + "seconds": 51.061138988, + "outputs": [ + 8.0 + ], + "max_absolute_error": 4.2646570364013314e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "remote_seconds": 66.024979734 + }, + { + "case": "cutlass", + "variant": 0, + "mode": "patched_all_pch", + "repeat": 0, + "extension_name": "kb_nvcc_cutlass_0", + "returncode": 0, + "log": "[1/3] /tmp/tmpvj9dykih/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpvj9dykih/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpvj9dykih/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpvj9dykih/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n[3/3] /tmp/tmpvj9dykih/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", + "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", + "container_id": "ta-01M7GGJYZRTAM0NXNGM9XQ7FTR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", + "process_seconds": 48.695631634, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpvj9dykih/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.893383368000002, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmpvj9dykih/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch" + ], + "seconds": 42.391933361999996, + "returncode": 0 + } + ], + "seconds": 43.662042164, + "outputs": [ + 8.0 + ], + "max_absolute_error": 4.2646570364013314e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpvj9dykih/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpvj9dykih/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpvj9dykih/cuda.cu\" --orig_src_path_name \"/tmp/tmpvj9dykih/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpvj9dykih/cuda.cu\" --orig_src_path_name \"/tmp/tmpvj9dykih/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpvj9dykih/diagnostic.cuda.o\" \n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 88.982580958 + }, + { + "case": "cutlass", + "variant": 1, + "mode": "patched_all_pch", + "repeat": 0, + "extension_name": "kb_nvcc_cutlass_1", + "returncode": 0, + "log": "[1/3] /tmp/tmp17x8fwhm/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp17x8fwhm/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp17x8fwhm/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmp17x8fwhm/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000031_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n\"/tmp/tmpxft_00000031_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n[3/3] /tmp/tmp17x8fwhm/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", + "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", + "container_id": "ta-01M7GGKGST0M04EKKWTZSZAZMR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", + "process_seconds": 43.799417716, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp17x8fwhm/main.cpp", + "-o", + "main.o" + ], + "seconds": 9.442032742999999, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmp17x8fwhm/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch" + ], + "seconds": 37.932805739, + "returncode": 0 + } + ], + "seconds": 39.165971053, + "outputs": [ + 16.0 + ], + "max_absolute_error": 8.529314072802663e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmp17x8fwhm/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmp17x8fwhm/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp17x8fwhm/cuda.cu\" --orig_src_path_name \"/tmp/tmp17x8fwhm/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp17x8fwhm/cuda.cu\" --orig_src_path_name \"/tmp/tmp17x8fwhm/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmp17x8fwhm/diagnostic.cuda.o\" \n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 82.974530772 + }, + { + "case": "cutlass", + "variant": 1, + "mode": "patched_host_pch", + "repeat": 0, + "extension_name": "kb_nvcc_cutlass_1", + "returncode": 0, + "log": "[1/3] /tmp/tmpqxtqfy1o/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpqxtqfy1o/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpqxtqfy1o/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpqxtqfy1o/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpqxtqfy1o/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", + "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", + "container_id": "ta-01M7GGMAPNEZDRDHFYXWX097RR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": null, + "process_seconds": 46.039022967, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpqxtqfy1o/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.110150218, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmpqxtqfy1o/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 41.130776678, + "returncode": 0 + } + ], + "seconds": 42.159665512000004, + "outputs": [ + 16.0 + ], + "max_absolute_error": 8.529314072802663e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "remote_seconds": 54.831509720999996 + }, + { + "case": "cutlass", + "variant": 1, + "mode": "stock_host_pch", + "repeat": 0, + "extension_name": "kb_nvcc_cutlass_1", + "returncode": 0, + "log": "[1/3] /tmp/tmpi_7ydo9h/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpi_7ydo9h/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpi_7ydo9h/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpi_7ydo9h/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpi_7ydo9h/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", + "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", + "container_id": "ta-01M7GGMJPMNY2K94DA3P3SCGPR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patch": null, + "pch_entry": null, + "process_seconds": 45.051432037000005, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpi_7ydo9h/main.cpp", + "-o", + "main.o" + ], + "seconds": 7.547227215, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmpi_7ydo9h/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 40.564446911, + "returncode": 0 + } + ], + "seconds": 41.497131416, + "outputs": [ + 16.0 + ], + "max_absolute_error": 8.529314072802663e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "remote_seconds": 52.125646299 + }, + { + "case": "cutlass", + "variant": 1, + "mode": "stock_no_pch", + "repeat": 0, + "extension_name": "kb_nvcc_cutlass_1", + "returncode": 0, + "log": "[1/3] /tmp/tmpcu5fn3dj/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpcu5fn3dj/main.cpp -o main.o \n[2/3] /tmp/tmpcu5fn3dj/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpcu5fn3dj/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpcu5fn3dj/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", + "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", + "container_id": "ta-01M7GGP2866GKDMKQ7CB7P4DWR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patch": null, + "pch_entry": null, + "process_seconds": 46.651000024999995, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpcu5fn3dj/main.cpp", + "-o", + "main.o" + ], + "seconds": 18.774265045, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmpcu5fn3dj/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 42.147979744, + "returncode": 0 + } + ], + "seconds": 42.949985459, + "outputs": [ + 16.0 + ], + "max_absolute_error": 8.529314072802663e-06, + "correctness_checks": 10, + "host_pch_consumed": false, + "host_hit_keys": [], + "remote_seconds": 64.96884930200001 + }, + { + "case": "simple", + "variant": 0, + "mode": "patched_all_pch", + "repeat": 1, + "extension_name": "kb_nvcc_simple_0", + "returncode": 0, + "log": "[1/3] /tmp/tmpjwxy1kiy/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpjwxy1kiy/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpjwxy1kiy/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpjwxy1kiy/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmpjwxy1kiy/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", + "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", + "container_id": "ta-01M7GGP9SHQZVXA5GZ2ZE3HXKR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", + "process_seconds": 27.58442731, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpjwxy1kiy/main.cpp", + "-o", + "main.o" + ], + "seconds": 7.728120469, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpjwxy1kiy/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" + ], + "seconds": 23.012643256, + "returncode": 0 + } + ], + "seconds": 23.934274335, + "outputs": [ + 2.0, + 2.0, + 2.0, + 2.0, + 2.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmpjwxy1kiy/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmpjwxy1kiy/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpjwxy1kiy/cuda.cu\" --orig_src_path_name \"/tmp/tmpjwxy1kiy/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpjwxy1kiy/cuda.cu\" --orig_src_path_name \"/tmp/tmpjwxy1kiy/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpjwxy1kiy/diagnostic.cuda.o\" \n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 55.646309089000006 + }, + { + "case": "simple", + "variant": 0, + "mode": "patched_host_pch", + "repeat": 1, + "extension_name": "kb_nvcc_simple_0", + "returncode": 0, + "log": "[1/3] /tmp/tmponu3k1v4/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmponu3k1v4/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmponu3k1v4/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmponu3k1v4/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmponu3k1v4/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", + "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", + "container_id": "ta-01M7GGPAVJ1N65VC6G7VXNW35R", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": null, + "process_seconds": 39.487048599000005, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmponu3k1v4/main.cpp", + "-o", + "main.o" + ], + "seconds": 7.554983864, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmponu3k1v4/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 35.113119433, + "returncode": 0 + } + ], + "seconds": 36.119203624, + "outputs": [ + 2.0, + 2.0, + 2.0, + 2.0, + 2.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "remote_seconds": 47.540994509 + }, + { + "case": "simple", + "variant": 0, + "mode": "stock_host_pch", + "repeat": 1, + "extension_name": "kb_nvcc_simple_0", + "returncode": 0, + "log": "[1/3] /tmp/tmp40_sti7n/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp40_sti7n/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp40_sti7n/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmp40_sti7n/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmp40_sti7n/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", + "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", + "container_id": "ta-01M7GGPCR34H6B7QV57HMT3TFR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patch": null, + "pch_entry": null, + "process_seconds": 39.490719059, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp40_sti7n/main.cpp", + "-o", + "main.o" + ], + "seconds": 7.914581139, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp40_sti7n/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 34.503693313, + "returncode": 0 + } + ], + "seconds": 35.485519697, + "outputs": [ + 2.0, + 2.0, + 2.0, + 2.0, + 2.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "remote_seconds": 47.474009400999996 + }, + { + "case": "simple", + "variant": 0, + "mode": "stock_no_pch", + "repeat": 1, + "extension_name": "kb_nvcc_simple_0", + "returncode": 0, + "log": "[1/3] /tmp/tmplhkbmlgw/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmplhkbmlgw/main.cpp -o main.o \n[2/3] /tmp/tmplhkbmlgw/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmplhkbmlgw/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmplhkbmlgw/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", + "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", + "container_id": "ta-01M7GGR0BSXMZF6601Q2HH4GDR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patch": null, + "pch_entry": null, + "process_seconds": 41.56644511099999, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmplhkbmlgw/main.cpp", + "-o", + "main.o" + ], + "seconds": 19.695273711, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmplhkbmlgw/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 37.226831019, + "returncode": 0 + } + ], + "seconds": 38.039081186, + "outputs": [ + 2.0, + 2.0, + 2.0, + 2.0, + 2.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": false, + "host_hit_keys": [], + "remote_seconds": 59.653186801 + }, + { + "case": "simple", + "variant": 1, + "mode": "stock_no_pch", + "repeat": 1, + "extension_name": "kb_nvcc_simple_1", + "returncode": 0, + "log": "[1/3] /tmp/tmpjq7sj4w6/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpjq7sj4w6/main.cpp -o main.o \n[2/3] /tmp/tmpjq7sj4w6/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpjq7sj4w6/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpjq7sj4w6/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", + "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", + "container_id": "ta-01M7GGR1YGMM34BW5HTWRC9WQR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patch": null, + "pch_entry": null, + "process_seconds": 39.52563073, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpjq7sj4w6/main.cpp", + "-o", + "main.o" + ], + "seconds": 19.178143019, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpjq7sj4w6/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 35.017214447, + "returncode": 0 + } + ], + "seconds": 35.746856564, + "outputs": [ + 3.0, + 3.0, + 3.0, + 3.0, + 3.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": false, + "host_hit_keys": [], + "remote_seconds": 57.896333988 + }, + { + "case": "simple", + "variant": 1, + "mode": "stock_host_pch", + "repeat": 1, + "extension_name": "kb_nvcc_simple_1", + "returncode": 0, + "log": "[1/3] /tmp/tmptwzsi6e7/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmptwzsi6e7/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmptwzsi6e7/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmptwzsi6e7/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmptwzsi6e7/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", + "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", + "container_id": "ta-01M7GGR6WK131M0DZZXP133CPR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patch": null, + "pch_entry": null, + "process_seconds": 38.652557074, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmptwzsi6e7/main.cpp", + "-o", + "main.o" + ], + "seconds": 7.423724915000001, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmptwzsi6e7/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 34.067612735, + "returncode": 0 + } + ], + "seconds": 35.025937535000004, + "outputs": [ + 3.0, + 3.0, + 3.0, + 3.0, + 3.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "remote_seconds": 45.924620759 + }, + { + "case": "simple", + "variant": 1, + "mode": "patched_host_pch", + "repeat": 1, + "extension_name": "kb_nvcc_simple_1", + "returncode": 0, + "log": "[1/3] /tmp/tmpihl2xip2/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpihl2xip2/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpihl2xip2/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpihl2xip2/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpihl2xip2/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", + "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", + "container_id": "ta-01M7GGR8JJ7XAQGGNEN85CYRER", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": null, + "process_seconds": 42.065910112, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpihl2xip2/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.606916758, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpihl2xip2/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 37.376972288000005, + "returncode": 0 + } + ], + "seconds": 38.467215558, + "outputs": [ + 3.0, + 3.0, + 3.0, + 3.0, + 3.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "remote_seconds": 50.520850166 + }, + { + "case": "simple", + "variant": 1, + "mode": "patched_all_pch", + "repeat": 1, + "extension_name": "kb_nvcc_simple_1", + "returncode": 0, + "log": "[1/3] /tmp/tmp858f0ai0/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp858f0ai0/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp858f0ai0/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmp858f0ai0/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000031_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000031_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmp858f0ai0/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", + "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", + "container_id": "ta-01M7GGSSKT9DDF905Z9S6J9SER", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", + "process_seconds": 33.745011223999995, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp858f0ai0/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.940535134, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp858f0ai0/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" + ], + "seconds": 27.845098587000003, + "returncode": 0 + } + ], + "seconds": 29.137589784, + "outputs": [ + 3.0, + 3.0, + 3.0, + 3.0, + 3.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp858f0ai0/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp858f0ai0/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp858f0ai0/cuda.cu\" --orig_src_path_name \"/tmp/tmp858f0ai0/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp858f0ai0/cuda.cu\" --orig_src_path_name \"/tmp/tmp858f0ai0/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmp858f0ai0/diagnostic.cuda.o\" \n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 72.54307103100001 + }, + { + "case": "cutlass", + "variant": 0, + "mode": "patched_all_pch", + "repeat": 1, + "extension_name": "kb_nvcc_cutlass_0", + "returncode": 0, + "log": "[1/3] /tmp/tmp6b6mdyz9/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp6b6mdyz9/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp6b6mdyz9/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmp6b6mdyz9/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n[3/3] /tmp/tmp6b6mdyz9/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", + "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", + "container_id": "ta-01M7GGT09WZH9JV14JHDV0YQVR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", + "process_seconds": 32.546973025, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp6b6mdyz9/main.cpp", + "-o", + "main.o" + ], + "seconds": 7.5847706829999995, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmp6b6mdyz9/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch" + ], + "seconds": 27.954299207, + "returncode": 0 + } + ], + "seconds": 28.886751083999997, + "outputs": [ + 8.0 + ], + "max_absolute_error": 4.2646570364013314e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmp6b6mdyz9/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmp6b6mdyz9/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp6b6mdyz9/cuda.cu\" --orig_src_path_name \"/tmp/tmp6b6mdyz9/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp6b6mdyz9/cuda.cu\" --orig_src_path_name \"/tmp/tmp6b6mdyz9/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmp6b6mdyz9/diagnostic.cuda.o\" \n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 65.119184206 + }, + { + "case": "cutlass", + "variant": 0, + "mode": "patched_host_pch", + "repeat": 1, + "extension_name": "kb_nvcc_cutlass_0", + "returncode": 0, + "log": "[1/3] /tmp/tmpx6o9xcvw/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpx6o9xcvw/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpx6o9xcvw/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpx6o9xcvw/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpx6o9xcvw/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", + "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", + "container_id": "ta-01M7GGT1KA1EXM9Z1NHPNBF6FR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": null, + "process_seconds": 45.630522881, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpx6o9xcvw/main.cpp", + "-o", + "main.o" + ], + "seconds": 7.483017078999999, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmpx6o9xcvw/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 40.705319495, + "returncode": 0 + } + ], + "seconds": 41.732421651, + "outputs": [ + 8.0 + ], + "max_absolute_error": 4.2646570364013314e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "remote_seconds": 53.773416114 + }, + { + "case": "cutlass", + "variant": 0, + "mode": "stock_host_pch", + "repeat": 1, + "extension_name": "kb_nvcc_cutlass_0", + "returncode": 0, + "log": "[1/3] /tmp/tmpli06qb2k/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpli06qb2k/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpli06qb2k/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpli06qb2k/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpli06qb2k/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", + "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", + "container_id": "ta-01M7GGT2A6N9TPEAVPNNM3K8TR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patch": null, + "pch_entry": null, + "process_seconds": 47.514512463, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpli06qb2k/main.cpp", + "-o", + "main.o" + ], + "seconds": 7.684575478, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmpli06qb2k/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 42.481320051, + "returncode": 0 + } + ], + "seconds": 43.52756049, + "outputs": [ + 8.0 + ], + "max_absolute_error": 4.2646570364013314e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "remote_seconds": 54.892269899999995 + }, + { + "case": "cutlass", + "variant": 0, + "mode": "stock_no_pch", + "repeat": 1, + "extension_name": "kb_nvcc_cutlass_0", + "returncode": 0, + "log": "[1/3] /tmp/tmppeys7t2i/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmppeys7t2i/main.cpp -o main.o \n[2/3] /tmp/tmppeys7t2i/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmppeys7t2i/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmppeys7t2i/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", + "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", + "container_id": "ta-01M7GGVWMY5WT8TDPYNZFRJY0R", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patch": null, + "pch_entry": null, + "process_seconds": 59.069312495, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmppeys7t2i/main.cpp", + "-o", + "main.o" + ], + "seconds": 22.843029878, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmppeys7t2i/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 53.267142170999996, + "returncode": 0 + } + ], + "seconds": 54.309780212999996, + "outputs": [ + 8.0 + ], + "max_absolute_error": 4.2646570364013314e-06, + "correctness_checks": 10, + "host_pch_consumed": false, + "host_hit_keys": [], + "remote_seconds": 82.28517314700001 + }, + { + "case": "cutlass", + "variant": 1, + "mode": "stock_no_pch", + "repeat": 1, + "extension_name": "kb_nvcc_cutlass_1", + "returncode": 0, + "log": "[1/3] /tmp/tmphrakn43k/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmphrakn43k/main.cpp -o main.o \n[2/3] /tmp/tmphrakn43k/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmphrakn43k/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmphrakn43k/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", + "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", + "container_id": "ta-01M7GGVZYE7ZXED0JB12EWFP6R", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patch": null, + "pch_entry": null, + "process_seconds": 47.475423516999996, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmphrakn43k/main.cpp", + "-o", + "main.o" + ], + "seconds": 18.90647545, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmphrakn43k/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 42.640482535, + "returncode": 0 + } + ], + "seconds": 43.498140506, + "outputs": [ + 16.0 + ], + "max_absolute_error": 8.529314072802663e-06, + "correctness_checks": 10, + "host_pch_consumed": false, + "host_hit_keys": [], + "remote_seconds": 65.75153526199999 + }, + { + "case": "cutlass", + "variant": 1, + "mode": "stock_host_pch", + "repeat": 1, + "extension_name": "kb_nvcc_cutlass_1", + "returncode": 0, + "log": "[1/3] /tmp/tmp1hzlz2oz/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp1hzlz2oz/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp1hzlz2oz/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmp1hzlz2oz/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmp1hzlz2oz/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", + "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", + "container_id": "ta-01M7GGW6GV9V5H8EYB13ANYRBR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patch": null, + "pch_entry": null, + "process_seconds": 47.481454979000006, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp1hzlz2oz/main.cpp", + "-o", + "main.o" + ], + "seconds": 7.983763389000001, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmp1hzlz2oz/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 42.884354589999994, + "returncode": 0 + } + ], + "seconds": 43.922133784, + "outputs": [ + 16.0 + ], + "max_absolute_error": 8.529314072802663e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "remote_seconds": 54.927763373 + }, + { + "case": "cutlass", + "variant": 1, + "mode": "patched_host_pch", + "repeat": 1, + "extension_name": "kb_nvcc_cutlass_1", + "returncode": 0, + "log": "[1/3] /tmp/tmpimrh22t3/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpimrh22t3/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpimrh22t3/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpimrh22t3/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpimrh22t3/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", + "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", + "container_id": "ta-01M7GGW978S61BB6WXTVDCE8GR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": null, + "process_seconds": 54.218881587999995, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpimrh22t3/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.671524286999999, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmpimrh22t3/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 48.066379684000005, + "returncode": 0 + } + ], + "seconds": 49.357130543, + "outputs": [ + 16.0 + ], + "max_absolute_error": 8.529314072802663e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "remote_seconds": 63.463944837999996 + }, + { + "case": "cutlass", + "variant": 1, + "mode": "patched_all_pch", + "repeat": 1, + "extension_name": "kb_nvcc_cutlass_1", + "returncode": 0, + "log": "[1/3] /tmp/tmpxzjxiynb/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpxzjxiynb/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpxzjxiynb/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpxzjxiynb/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n[3/3] /tmp/tmpxzjxiynb/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", + "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", + "container_id": "ta-01M7GGY1T72T9C66SB78QVCPJR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", + "process_seconds": 32.155385849999995, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpxzjxiynb/main.cpp", + "-o", + "main.o" + ], + "seconds": 7.430922593, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmpxzjxiynb/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch" + ], + "seconds": 27.405767641000004, + "returncode": 0 + } + ], + "seconds": 28.440589126, + "outputs": [ + 16.0 + ], + "max_absolute_error": 8.529314072802663e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpxzjxiynb/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpxzjxiynb/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzjxiynb/cuda.cu\" --orig_src_path_name \"/tmp/tmpxzjxiynb/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzjxiynb/cuda.cu\" --orig_src_path_name \"/tmp/tmpxzjxiynb/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpxzjxiynb/diagnostic.cuda.o\" \n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 65.992117256 + } + ] + }, + "supplemental": { + "probe": { + "stock_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patched_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "paths": "/usr/local/cuda/bin/nvcc.profile\n/usr/local/cuda/nvvm/bin/cicc\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "nvcc_profiles": { + "/usr/local/cuda/bin/nvcc.profile": "\nTOP = $(_HERE_)/..\n\nCICC_PATH = $(TOP)/nvvm-next/bin\nNVVMIR_LIBRARY_DIR = $(TOP)/nvvm/libdevice\n\nLD_LIBRARY_PATH += $(TOP)/lib:\nPATH += $(CICC_PATH):$(_HERE_):\n\nINCLUDES += \"-I$(TOP)/$(_TARGET_DIR_)/include\" $(_SPACE_)\nSYSTEM_INCLUDES += \"-isystem\" \"$(TOP)/$(_TARGET_DIR_)/include/cccl\" $(_SPACE_)\n\nLIBRARIES =+ $(_SPACE_) \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)/stubs\" \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)\"\n\nCUDAFE_FLAGS +=\nPTXAS_FLAGS +=\n" + }, + "pch_help": " Specify options directly to library manager.\n\n--pch --pch \n Enable precompiled header (PCH) processing. When specified, each compilation\n searches the PCH directory (see --pch-dir) for a reusable precompiled header,\n and creates a new one when none is found.\n\n--pch-dir --pch-dir \n Specify the directory in which precompiled header files are read from and\n written to. By default this is the current working directory. Ignored unless\n --pch, --create-pch, or --use-pch is enabled.\n\n--pch-verbose --pch-verbose \n Print (to stderr) messages related to precompiled header processing. Ignored\n unless --pch, --create-pch, or --use-pch is enabled.\n\n--create-pch --create-pch \n Manually create a precompiled header (PCH) at the first PCH terminal stop-point\n (i.e., #pragma hdrstop) and save it under the specified name. Intermediate\n stop points (#pragma hdrstop++) are ignored. The directory into which the\n PCH file is written is determined by --pch-dir (or is the current working\n directory if not specified). This option forces a PCH file to be created\n (any existing PCH file with the same name is overwritten). This option is\n mutually exclusive with '--use-pch' and '--pch'.\n\n--use-pch --use-pch \n If a precompiled header (PCH) with the specified name exists in the directory\n determined by '--pch-dir' (or the current working directory if not specified),\n try to use it in compilation.This option is mutually exclusive with '--create-pch'\n and '--pch'.\n\n--ptxas-options ,... (-Xptxas) \n Specify options directly to ptxas, the PTX optimizing assembler.\n\n--nvlink-options ,... (-Xnvlink) \n Specify options directly to nvlink.\n\n--nvprune-options ,... (-Xnvprune) \n Specify options directly to nvprune (used with -prune).\n\n\nMiscellaneous options for guiding the compiler driver.\n======================================================", + "strace": null, + "smoke": [ + { + "arch": "sm_75", + "command": [ + "/usr/local/cuda/bin/nvcc", + "-arch=sm_75", + "-c", + "/tmp/tmpxzw56gh8/smoke.cu", + "-o", + "/tmp/tmpxzw56gh8/sm_75.o" + ], + "dryrun_returncode": 0, + "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpxzw56gh8/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzw56gh8/smoke.cu\" --orig_src_path_name \"/tmp/tmpxzw56gh8/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpxzw56gh8/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzw56gh8/smoke.cu\" --orig_src_path_name \"/tmp/tmpxzw56gh8/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_0000001e_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_0000001e_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_0000001e_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpxzw56gh8/sm_75.o\" \n", + "next_backend_selected": true, + "returncode": 0, + "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpxzw56gh8/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzw56gh8/smoke.cu\" --orig_src_path_name \"/tmp/tmpxzw56gh8/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpxzw56gh8/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzw56gh8/smoke.cu\" --orig_src_path_name \"/tmp/tmpxzw56gh8/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000021_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000021_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000021_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpxzw56gh8/sm_75.o\" \n", + "object_bytes": 10200 + }, + { + "arch": "sm_100", + "command": [ + "/usr/local/cuda/bin/nvcc", + "-arch=sm_100", + "-c", + "/tmp/tmpxzw56gh8/smoke.cu", + "-o", + "/tmp/tmpxzw56gh8/sm_100.o" + ], + "dryrun_returncode": 0, + "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpxzw56gh8/smoke.cu\" -o \"/tmp/tmpxft_00000036_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzw56gh8/smoke.cu\" --orig_src_path_name \"/tmp/tmpxzw56gh8/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000036_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000036_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000036_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpxzw56gh8/smoke.cu\" -o \"/tmp/tmpxft_00000036_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzw56gh8/smoke.cu\" --orig_src_path_name \"/tmp/tmpxzw56gh8/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000036_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000036_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000036_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000036_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000036_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000036_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000036_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000036_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000036_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000036_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpxzw56gh8/sm_100.o\" \n", + "next_backend_selected": true, + "returncode": 0, + "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpxzw56gh8/smoke.cu\" -o \"/tmp/tmpxft_00000039_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzw56gh8/smoke.cu\" --orig_src_path_name \"/tmp/tmpxzw56gh8/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000039_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000039_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000039_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpxzw56gh8/smoke.cu\" -o \"/tmp/tmpxft_00000039_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzw56gh8/smoke.cu\" --orig_src_path_name \"/tmp/tmpxzw56gh8/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000039_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000039_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000039_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000039_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000039_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000039_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000039_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000039_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000039_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000039_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpxzw56gh8/sm_100.o\" \n", + "object_bytes": 12504 + } + ], + "compatible": true + }, + "rows": [ + { + "case": "simple", + "variant": 0, + "mode": "patched_host_pch", + "repeat": 0, + "extension_name": "kb_nvcc_simple_0", + "returncode": 0, + "log": "[1/3] /tmp/tmpix5k2j49/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpix5k2j49/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpix5k2j49/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpix5k2j49/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpix5k2j49/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", + "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", + "container_id": "ta-01M7GGED50386E0QEMXCQGKJAR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": null, + "process_seconds": 41.332697233000005, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpix5k2j49/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.343771207, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpix5k2j49/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 36.214367632000005, + "returncode": 0 + } + ], + "seconds": 37.205667172000005, + "outputs": [ + 2.0, + 2.0, + 2.0, + 2.0, + 2.0 + ], + "load_inline_seconds": 37.205656627, + "prefetch_seconds": 0, + "prefetch_bytes": 0, + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpix5k2j49/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpix5k2j49/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpix5k2j49/cuda.cu\" --orig_src_path_name \"/tmp/tmpix5k2j49/cuda.cu\" --allow_managed --relaxed_constexpr --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ \"$CICC_PATH/cicc\" --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpix5k2j49/cuda.cu\" --orig_src_path_name \"/tmp/tmpix5k2j49/cuda.cu\" --allow_managed --relaxed_constexpr -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpix5k2j49/diagnostic.cuda.o\" \n", + "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmpix5k2j49/cuda.cu , gcc (preprocessing 1) , /tmp/tmpix5k2j49/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 2824.1750 , ms\n/tmp/tmpix5k2j49/cuda.cu , gcc (preprocessing 4) , /tmp/tmpix5k2j49/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 2755.5020 , ms\n/tmp/tmpix5k2j49/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 12239.0156 , ms\n/tmp/tmpix5k2j49/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 8393.6445 , ms\n/tmp/tmpix5k2j49/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 25.3120 , ms\n/tmp/tmpix5k2j49/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 4.2280 , ms\n/tmp/tmpix5k2j49/cuda.cu , gcc (compiling) , , /tmp/tmpix5k2j49/diagnostic.cuda.o , compute_75 , nvcc , 7372.6270 , ms\n , nvcc (driver) , , /tmp/tmpix5k2j49/diagnostic.cuda.o , , nvcc , 65.8086 , ms\n" + }, + "device_pch_files_used": [], + "device_pch_files_created": [], + "device_pch_consumed": false, + "remote_seconds": 83.449806407 + }, + { + "case": "simple", + "variant": 0, + "mode": "patched_all_pch", + "repeat": 0, + "extension_name": "kb_nvcc_simple_0", + "returncode": 0, + "log": "[1/3] /tmp/tmp279h84nf/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp279h84nf/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp279h84nf/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmp279h84nf/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmp279h84nf/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", + "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", + "container_id": "ta-01M7GGED50E6R503HVDSM8MGRR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", + "process_seconds": 31.842433575999998, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp279h84nf/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.092909477, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp279h84nf/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" + ], + "seconds": 26.555807986, + "returncode": 0 + } + ], + "seconds": 27.513593019, + "outputs": [ + 2.0, + 2.0, + 2.0, + 2.0, + 2.0 + ], + "load_inline_seconds": 27.513590667000003, + "prefetch_seconds": 0, + "prefetch_bytes": 0, + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp279h84nf/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp279h84nf/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp279h84nf/cuda.cu\" --orig_src_path_name \"/tmp/tmp279h84nf/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp279h84nf/cuda.cu\" --orig_src_path_name \"/tmp/tmp279h84nf/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmp279h84nf/diagnostic.cuda.o\" \n", + "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmp279h84nf/cuda.cu , gcc (preprocessing 1) , /tmp/tmp279h84nf/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 2803.0601 , ms\n/tmp/tmp279h84nf/cuda.cu , gcc (preprocessing 4) , /tmp/tmp279h84nf/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 2761.7849 , ms\n/tmp/tmp279h84nf/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 3816.4561 , ms\n/tmp/tmp279h84nf/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 4133.8110 , ms\n/tmp/tmp279h84nf/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 31.0950 , ms\n/tmp/tmp279h84nf/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 25.4210 , ms\n/tmp/tmp279h84nf/cuda.cu , gcc (compiling) , , /tmp/tmp279h84nf/diagnostic.cuda.o , compute_75 , nvcc , 7562.0918 , ms\n , nvcc (driver) , , /tmp/tmp279h84nf/diagnostic.cuda.o , , nvcc , 65.3496 , ms\n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 61.302708119 + }, + { + "case": "simple", + "variant": 0, + "mode": "patched_all_pch_prefetch", + "repeat": 0, + "extension_name": "kb_nvcc_simple_0", + "returncode": 0, + "log": "[1/3] /tmp/tmpcucf0lnk/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpcucf0lnk/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpcucf0lnk/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpcucf0lnk/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmpcucf0lnk/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", + "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", + "container_id": "ta-01M7GGED5AZC6SWFBFCVMC7C5R", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", + "process_seconds": 30.038420162, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpcucf0lnk/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.254390696, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpcucf0lnk/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" + ], + "seconds": 23.209440734000005, + "returncode": 0 + } + ], + "seconds": 25.731699179, + "outputs": [ + 2.0, + 2.0, + 2.0, + 2.0, + 2.0 + ], + "load_inline_seconds": 24.222713579, + "prefetch_seconds": 1.5089852529999996, + "prefetch_bytes": 2524303360, + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmpcucf0lnk/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmpcucf0lnk/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpcucf0lnk/cuda.cu\" --orig_src_path_name \"/tmp/tmpcucf0lnk/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpcucf0lnk/cuda.cu\" --orig_src_path_name \"/tmp/tmpcucf0lnk/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpcucf0lnk/diagnostic.cuda.o\" \n", + "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmpcucf0lnk/cuda.cu , gcc (preprocessing 1) , /tmp/tmpcucf0lnk/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 2736.3269 , ms\n/tmp/tmpcucf0lnk/cuda.cu , gcc (preprocessing 4) , /tmp/tmpcucf0lnk/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 2780.0220 , ms\n/tmp/tmpcucf0lnk/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 3874.4480 , ms\n/tmp/tmpcucf0lnk/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 4437.3799 , ms\n/tmp/tmpcucf0lnk/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 26.7620 , ms\n/tmp/tmpcucf0lnk/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 4.7330 , ms\n/tmp/tmpcucf0lnk/cuda.cu , gcc (compiling) , , /tmp/tmpcucf0lnk/diagnostic.cuda.o , compute_75 , nvcc , 7862.6460 , ms\n , nvcc (driver) , , /tmp/tmpcucf0lnk/diagnostic.cuda.o , , nvcc , 67.0508 , ms\n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 60.145700806 + }, + { + "case": "simple", + "variant": 1, + "mode": "patched_all_pch_prefetch", + "repeat": 0, + "extension_name": "kb_nvcc_simple_1", + "returncode": 0, + "log": "[1/3] /tmp/tmp__6x16w0/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp__6x16w0/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp__6x16w0/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmp__6x16w0/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000031_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000031_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmp__6x16w0/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", + "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", + "container_id": "ta-01M7GGED5AR5589PP71ECYZT5R", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", + "process_seconds": 30.059914977, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp__6x16w0/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.255181534999998, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp__6x16w0/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" + ], + "seconds": 23.040583837, + "returncode": 0 + } + ], + "seconds": 25.672850840000002, + "outputs": [ + 3.0, + 3.0, + 3.0, + 3.0, + 3.0 + ], + "load_inline_seconds": 24.072621637, + "prefetch_seconds": 1.6002290010000007, + "prefetch_bytes": 2524303360, + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp__6x16w0/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp__6x16w0/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp__6x16w0/cuda.cu\" --orig_src_path_name \"/tmp/tmp__6x16w0/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp__6x16w0/cuda.cu\" --orig_src_path_name \"/tmp/tmp__6x16w0/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmp__6x16w0/diagnostic.cuda.o\" \n", + "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmp__6x16w0/cuda.cu , gcc (preprocessing 1) , /tmp/tmp__6x16w0/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 2714.1531 , ms\n/tmp/tmp__6x16w0/cuda.cu , gcc (preprocessing 4) , /tmp/tmp__6x16w0/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 2679.6211 , ms\n/tmp/tmp__6x16w0/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 3953.6370 , ms\n/tmp/tmp__6x16w0/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 4338.0518 , ms\n/tmp/tmp__6x16w0/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 32.9190 , ms\n/tmp/tmp__6x16w0/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 4.1430 , ms\n/tmp/tmp__6x16w0/cuda.cu , gcc (compiling) , , /tmp/tmp__6x16w0/diagnostic.cuda.o , compute_75 , nvcc , 7706.5312 , ms\n , nvcc (driver) , , /tmp/tmp__6x16w0/diagnostic.cuda.o , , nvcc , 66.0625 , ms\n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 60.086438656999995 + }, + { + "case": "simple", + "variant": 1, + "mode": "patched_all_pch", + "repeat": 0, + "extension_name": "kb_nvcc_simple_1", + "returncode": 0, + "log": "[1/3] /tmp/tmppd82nejq/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmppd82nejq/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmppd82nejq/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmppd82nejq/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmppd82nejq/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", + "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", + "container_id": "ta-01M7GGGJTBKKMAZG8E43P3FYXR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", + "process_seconds": 26.685320548, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmppd82nejq/main.cpp", + "-o", + "main.o" + ], + "seconds": 7.648132214000001, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmppd82nejq/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" + ], + "seconds": 22.029563485, + "returncode": 0 + } + ], + "seconds": 22.962655791, + "outputs": [ + 3.0, + 3.0, + 3.0, + 3.0, + 3.0 + ], + "load_inline_seconds": 22.962653443, + "prefetch_seconds": 0, + "prefetch_bytes": 0, + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmppd82nejq/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmppd82nejq/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmppd82nejq/cuda.cu\" --orig_src_path_name \"/tmp/tmppd82nejq/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmppd82nejq/cuda.cu\" --orig_src_path_name \"/tmp/tmppd82nejq/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmppd82nejq/diagnostic.cuda.o\" \n", + "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmppd82nejq/cuda.cu , gcc (preprocessing 1) , /tmp/tmppd82nejq/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 2792.3230 , ms\n/tmp/tmppd82nejq/cuda.cu , gcc (preprocessing 4) , /tmp/tmppd82nejq/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 2763.0911 , ms\n/tmp/tmppd82nejq/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 3405.5300 , ms\n/tmp/tmppd82nejq/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 3858.2410 , ms\n/tmp/tmppd82nejq/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 23.1240 , ms\n/tmp/tmppd82nejq/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 4.2590 , ms\n/tmp/tmppd82nejq/cuda.cu , gcc (compiling) , , /tmp/tmppd82nejq/diagnostic.cuda.o , compute_75 , nvcc , 7438.9702 , ms\n , nvcc (driver) , , /tmp/tmppd82nejq/diagnostic.cuda.o , , nvcc , 65.9883 , ms\n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 55.045973556 + }, + { + "case": "simple", + "variant": 1, + "mode": "patched_host_pch", + "repeat": 0, + "extension_name": "kb_nvcc_simple_1", + "returncode": 0, + "log": "[1/3] /tmp/tmpdogjj3qw/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpdogjj3qw/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpdogjj3qw/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpdogjj3qw/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpdogjj3qw/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", + "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", + "container_id": "ta-01M7GGGJTBSQFTS3HEY1Y4K4GR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": null, + "process_seconds": 49.306690329, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpdogjj3qw/main.cpp", + "-o", + "main.o" + ], + "seconds": 12.670156114000001, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpdogjj3qw/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 42.628599909, + "returncode": 0 + } + ], + "seconds": 44.070919440000004, + "outputs": [ + 3.0, + 3.0, + 3.0, + 3.0, + 3.0 + ], + "load_inline_seconds": 44.070915819, + "prefetch_seconds": 0, + "prefetch_bytes": 0, + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpdogjj3qw/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpdogjj3qw/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpdogjj3qw/cuda.cu\" --orig_src_path_name \"/tmp/tmpdogjj3qw/cuda.cu\" --allow_managed --relaxed_constexpr --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ \"$CICC_PATH/cicc\" --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpdogjj3qw/cuda.cu\" --orig_src_path_name \"/tmp/tmpdogjj3qw/cuda.cu\" --allow_managed --relaxed_constexpr -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpdogjj3qw/diagnostic.cuda.o\" \n", + "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmpdogjj3qw/cuda.cu , gcc (preprocessing 1) , /tmp/tmpdogjj3qw/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 3497.3870 , ms\n/tmp/tmpdogjj3qw/cuda.cu , gcc (preprocessing 4) , /tmp/tmpdogjj3qw/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 3515.5879 , ms\n/tmp/tmpdogjj3qw/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 15524.1338 , ms\n/tmp/tmpdogjj3qw/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 11493.0625 , ms\n/tmp/tmpdogjj3qw/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 33.9710 , ms\n/tmp/tmpdogjj3qw/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 5.8130 , ms\n/tmp/tmpdogjj3qw/cuda.cu , gcc (compiling) , , /tmp/tmpdogjj3qw/diagnostic.cuda.o , compute_75 , nvcc , 9197.6523 , ms\n , nvcc (driver) , , /tmp/tmpdogjj3qw/diagnostic.cuda.o , , nvcc , 84.5664 , ms\n" + }, + "device_pch_files_used": [], + "device_pch_files_created": [], + "device_pch_consumed": false, + "remote_seconds": 101.697851061 + }, + { + "case": "cutlass", + "variant": 0, + "mode": "patched_host_pch", + "repeat": 0, + "extension_name": "kb_nvcc_cutlass_0", + "returncode": 0, + "log": "[1/3] /tmp/tmpd36naltj/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpd36naltj/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpd36naltj/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpd36naltj/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpd36naltj/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", + "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", + "container_id": "ta-01M7GGGZXAYNTYNS7JADW3KA9R", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": null, + "process_seconds": 55.738259471999996, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpd36naltj/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.778172018, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmpd36naltj/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 49.517795254999996, + "returncode": 0 + } + ], + "seconds": 50.875985165, + "outputs": [ + 8.0 + ], + "load_inline_seconds": 50.875981628, + "prefetch_seconds": 0, + "prefetch_bytes": 0, + "max_absolute_error": 4.2646570364013314e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpd36naltj/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpd36naltj/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpd36naltj/cuda.cu\" --orig_src_path_name \"/tmp/tmpd36naltj/cuda.cu\" --allow_managed --relaxed_constexpr --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ \"$CICC_PATH/cicc\" --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpd36naltj/cuda.cu\" --orig_src_path_name \"/tmp/tmpd36naltj/cuda.cu\" --allow_managed --relaxed_constexpr -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpd36naltj/diagnostic.cuda.o\" \n", + "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmpd36naltj/cuda.cu , gcc (preprocessing 1) , /tmp/tmpd36naltj/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 4387.8960 , ms\n/tmp/tmpd36naltj/cuda.cu , gcc (preprocessing 4) , /tmp/tmpd36naltj/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 4520.6260 , ms\n/tmp/tmpd36naltj/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 19127.1426 , ms\n/tmp/tmpd36naltj/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 11011.8525 , ms\n/tmp/tmpd36naltj/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 413.6620 , ms\n/tmp/tmpd36naltj/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 7.7100 , ms\n/tmp/tmpd36naltj/cuda.cu , gcc (compiling) , , /tmp/tmpd36naltj/diagnostic.cuda.o , compute_75 , nvcc , 9960.2764 , ms\n , nvcc (driver) , , /tmp/tmpd36naltj/diagnostic.cuda.o , , nvcc , 110.8203 , ms\n" + }, + "device_pch_files_used": [], + "device_pch_files_created": [], + "device_pch_consumed": false, + "remote_seconds": 114.717228026 + }, + { + "case": "cutlass", + "variant": 0, + "mode": "patched_all_pch", + "repeat": 0, + "extension_name": "kb_nvcc_cutlass_0", + "returncode": 0, + "log": "[1/3] /tmp/tmp8ber6ya5/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp8ber6ya5/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp8ber6ya5/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmp8ber6ya5/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n[3/3] /tmp/tmp8ber6ya5/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", + "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", + "container_id": "ta-01M7GGH8YZHK6M1B0FWN50S9BR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", + "process_seconds": 36.453414843000004, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp8ber6ya5/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.150167573, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmp8ber6ya5/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch" + ], + "seconds": 31.500741393, + "returncode": 0 + } + ], + "seconds": 32.528592897, + "outputs": [ + 8.0 + ], + "load_inline_seconds": 32.528590723, + "prefetch_seconds": 0, + "prefetch_bytes": 0, + "max_absolute_error": 4.2646570364013314e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmp8ber6ya5/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmp8ber6ya5/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp8ber6ya5/cuda.cu\" --orig_src_path_name \"/tmp/tmp8ber6ya5/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp8ber6ya5/cuda.cu\" --orig_src_path_name \"/tmp/tmp8ber6ya5/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmp8ber6ya5/diagnostic.cuda.o\" \n", + "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmp8ber6ya5/cuda.cu , gcc (preprocessing 1) , /tmp/tmp8ber6ya5/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 3192.6299 , ms\n/tmp/tmp8ber6ya5/cuda.cu , gcc (preprocessing 4) , /tmp/tmp8ber6ya5/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 3075.6389 , ms\n/tmp/tmp8ber6ya5/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 5036.9600 , ms\n/tmp/tmp8ber6ya5/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 5079.5298 , ms\n/tmp/tmp8ber6ya5/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 320.1950 , ms\n/tmp/tmp8ber6ya5/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 5.3060 , ms\n/tmp/tmp8ber6ya5/cuda.cu , gcc (compiling) , , /tmp/tmp8ber6ya5/diagnostic.cuda.o , compute_75 , nvcc , 9192.4521 , ms\n , nvcc (driver) , , /tmp/tmp8ber6ya5/diagnostic.cuda.o , , nvcc , 96.7441 , ms\n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 71.035927235 + }, + { + "case": "cutlass", + "variant": 0, + "mode": "patched_all_pch_prefetch", + "repeat": 0, + "extension_name": "kb_nvcc_cutlass_0", + "returncode": 0, + "log": "[1/3] /tmp/tmpz4s1y_ge/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpz4s1y_ge/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpz4s1y_ge/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpz4s1y_ge/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n[3/3] /tmp/tmpz4s1y_ge/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", + "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", + "container_id": "ta-01M7GGJEA1CBKA78JQGA7X6YTR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", + "process_seconds": 34.245046614, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpz4s1y_ge/main.cpp", + "-o", + "main.o" + ], + "seconds": 7.674453068000002, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmpz4s1y_ge/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch" + ], + "seconds": 26.635627733, + "returncode": 0 + } + ], + "seconds": 30.342368112999996, + "outputs": [ + 8.0 + ], + "load_inline_seconds": 27.733861292, + "prefetch_seconds": 2.6085065270000003, + "prefetch_bytes": 2999697408, + "max_absolute_error": 4.2646570364013314e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpz4s1y_ge/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpz4s1y_ge/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpz4s1y_ge/cuda.cu\" --orig_src_path_name \"/tmp/tmpz4s1y_ge/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpz4s1y_ge/cuda.cu\" --orig_src_path_name \"/tmp/tmpz4s1y_ge/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpz4s1y_ge/diagnostic.cuda.o\" \n", + "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmpz4s1y_ge/cuda.cu , gcc (preprocessing 1) , /tmp/tmpz4s1y_ge/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 3394.4260 , ms\n/tmp/tmpz4s1y_ge/cuda.cu , gcc (preprocessing 4) , /tmp/tmpz4s1y_ge/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 3355.1160 , ms\n/tmp/tmpz4s1y_ge/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 4308.4219 , ms\n/tmp/tmpz4s1y_ge/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 5538.6201 , ms\n/tmp/tmpz4s1y_ge/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 329.7930 , ms\n/tmp/tmpz4s1y_ge/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 5.9670 , ms\n/tmp/tmpz4s1y_ge/cuda.cu , gcc (compiling) , , /tmp/tmpz4s1y_ge/diagnostic.cuda.o , compute_75 , nvcc , 9146.8916 , ms\n , nvcc (driver) , , /tmp/tmpz4s1y_ge/diagnostic.cuda.o , , nvcc , 92.5684 , ms\n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 68.732929376 + }, + { + "case": "cutlass", + "variant": 1, + "mode": "patched_all_pch_prefetch", + "repeat": 0, + "extension_name": "kb_nvcc_cutlass_1", + "returncode": 0, + "log": "[1/3] /tmp/tmpw_st2x9m/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpw_st2x9m/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpw_st2x9m/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpw_st2x9m/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n[3/3] /tmp/tmpw_st2x9m/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", + "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", + "container_id": "ta-01M7GGKN4QWSM62X7YPQY4B47R", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", + "process_seconds": 44.962129579, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpw_st2x9m/main.cpp", + "-o", + "main.o" + ], + "seconds": 9.114189858, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmpw_st2x9m/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch" + ], + "seconds": 31.421957751, + "returncode": 0 + } + ], + "seconds": 40.536378983999995, + "outputs": [ + 16.0 + ], + "load_inline_seconds": 32.801416029, + "prefetch_seconds": 7.734961381, + "prefetch_bytes": 2999697408, + "max_absolute_error": 8.529314072802663e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpw_st2x9m/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpw_st2x9m/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw_st2x9m/cuda.cu\" --orig_src_path_name \"/tmp/tmpw_st2x9m/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw_st2x9m/cuda.cu\" --orig_src_path_name \"/tmp/tmpw_st2x9m/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpw_st2x9m/diagnostic.cuda.o\" \n", + "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmpw_st2x9m/cuda.cu , gcc (preprocessing 1) , /tmp/tmpw_st2x9m/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 4109.4219 , ms\n/tmp/tmpw_st2x9m/cuda.cu , gcc (preprocessing 4) , /tmp/tmpw_st2x9m/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 4190.4790 , ms\n/tmp/tmpw_st2x9m/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 4666.5859 , ms\n/tmp/tmpw_st2x9m/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 6280.6641 , ms\n/tmp/tmpw_st2x9m/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 381.9780 , ms\n/tmp/tmpw_st2x9m/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 7.4310 , ms\n/tmp/tmpw_st2x9m/cuda.cu , gcc (compiling) , , /tmp/tmpw_st2x9m/diagnostic.cuda.o , compute_75 , nvcc , 9935.6123 , ms\n , nvcc (driver) , , /tmp/tmpw_st2x9m/diagnostic.cuda.o , , nvcc , 106.5527 , ms\n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 83.664074216 + }, + { + "case": "cutlass", + "variant": 1, + "mode": "patched_all_pch", + "repeat": 0, + "extension_name": "kb_nvcc_cutlass_1", + "returncode": 0, + "log": "[1/3] /tmp/tmpcylmnezm/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpcylmnezm/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpcylmnezm/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpcylmnezm/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000031_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n\"/tmp/tmpxft_00000031_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n[3/3] /tmp/tmpcylmnezm/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", + "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", + "container_id": "ta-01M7GGM20MGZZK7K4NM6A452NR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", + "process_seconds": 37.669464049, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpcylmnezm/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.499430686, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmpcylmnezm/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch" + ], + "seconds": 32.01834994, + "returncode": 0 + } + ], + "seconds": 33.270577796, + "outputs": [ + 16.0 + ], + "load_inline_seconds": 33.270573987, + "prefetch_seconds": 0, + "prefetch_bytes": 0, + "max_absolute_error": 8.529314072802663e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpcylmnezm/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpcylmnezm/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpcylmnezm/cuda.cu\" --orig_src_path_name \"/tmp/tmpcylmnezm/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpcylmnezm/cuda.cu\" --orig_src_path_name \"/tmp/tmpcylmnezm/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpcylmnezm/diagnostic.cuda.o\" \n", + "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmpcylmnezm/cuda.cu , gcc (preprocessing 1) , /tmp/tmpcylmnezm/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 4109.8291 , ms\n/tmp/tmpcylmnezm/cuda.cu , gcc (preprocessing 4) , /tmp/tmpcylmnezm/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 4238.2021 , ms\n/tmp/tmpcylmnezm/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 4641.4351 , ms\n/tmp/tmpcylmnezm/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 6262.5859 , ms\n/tmp/tmpcylmnezm/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 390.5320 , ms\n/tmp/tmpcylmnezm/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 7.3630 , ms\n/tmp/tmpcylmnezm/cuda.cu , gcc (compiling) , , /tmp/tmpcylmnezm/diagnostic.cuda.o , compute_75 , nvcc , 9785.2178 , ms\n , nvcc (driver) , , /tmp/tmpcylmnezm/diagnostic.cuda.o , , nvcc , 108.5156 , ms\n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 77.071578788 + }, + { + "case": "cutlass", + "variant": 1, + "mode": "patched_host_pch", + "repeat": 0, + "extension_name": "kb_nvcc_cutlass_1", + "returncode": 0, + "log": "[1/3] /tmp/tmpanf9brnl/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpanf9brnl/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpanf9brnl/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpanf9brnl/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpanf9brnl/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", + "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", + "container_id": "ta-01M7GGMR8RYJMZVXGZ6MZKK5ER", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": null, + "process_seconds": 53.932866476, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpanf9brnl/main.cpp", + "-o", + "main.o" + ], + "seconds": 8.41651323, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "-c", + "/tmp/tmpanf9brnl/cuda.cu", + "-o", + "cuda.cuda.o" + ], + "seconds": 48.115408051, + "returncode": 0 + } + ], + "seconds": 49.549131939, + "outputs": [ + 16.0 + ], + "load_inline_seconds": 49.549129623, + "prefetch_seconds": 0, + "prefetch_bytes": 0, + "max_absolute_error": 8.529314072802663e-06, + "correctness_checks": 10, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpanf9brnl/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpanf9brnl/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpanf9brnl/cuda.cu\" --orig_src_path_name \"/tmp/tmpanf9brnl/cuda.cu\" --allow_managed --relaxed_constexpr --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ \"$CICC_PATH/cicc\" --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpanf9brnl/cuda.cu\" --orig_src_path_name \"/tmp/tmpanf9brnl/cuda.cu\" --allow_managed --relaxed_constexpr -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpanf9brnl/diagnostic.cuda.o\" \n", + "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmpanf9brnl/cuda.cu , gcc (preprocessing 1) , /tmp/tmpanf9brnl/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 4388.6689 , ms\n/tmp/tmpanf9brnl/cuda.cu , gcc (preprocessing 4) , /tmp/tmpanf9brnl/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 4251.5449 , ms\n/tmp/tmpanf9brnl/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 17644.5254 , ms\n/tmp/tmpanf9brnl/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 10723.0273 , ms\n/tmp/tmpanf9brnl/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 391.4860 , ms\n/tmp/tmpanf9brnl/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 7.5860 , ms\n/tmp/tmpanf9brnl/cuda.cu , gcc (compiling) , , /tmp/tmpanf9brnl/diagnostic.cuda.o , compute_75 , nvcc , 9803.3232 , ms\n , nvcc (driver) , , /tmp/tmpanf9brnl/diagnostic.cuda.o , , nvcc , 102.2070 , ms\n" + }, + "device_pch_files_used": [], + "device_pch_files_created": [], + "device_pch_consumed": false, + "remote_seconds": 112.023578118 + } + ] + }, + "production_cuda133": [ + { + "case": "inline_simple", + "variant": 0, + "enabled": false, + "repeat": 0, + "extension_name": "pch_matrix_inline_simple_0_0", + "seconds": 21.796966691, + "timing_samples": [ + 21.796966691 + ], + "host_compile_seconds": 20.979295318000002, + "pch_consumed": false, + "hit_keys": [], + "cache_miss": false, + "correctness_checks": 6, + "outputs": [ + 2.0, + 2.0, + 2.0, + 2.0, + 2.0 + ], + "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", + "container_id": "ta-01M7GFSN13SKZYRAX01JEAZB2R", + "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_seconds": [ + 3.0216056989999984 + ], + "max_absolute_error": 0, + "remote_seconds": 45.022717201000006 + }, + { + "case": "inline_simple", + "variant": 0, + "enabled": true, + "repeat": 0, + "extension_name": "pch_matrix_inline_simple_0_1", + "seconds": 10.092071702999998, + "timing_samples": [ + 10.092071702999998 + ], + "host_compile_seconds": 8.931718365, + "pch_consumed": true, + "hit_keys": [ + "cd04740b9ab36ac9a2a8d533e50fc30250ab89f341186829da169aefb0865f38" + ], + "cache_miss": false, + "correctness_checks": 6, + "outputs": [ + 2.0, + 2.0, + 2.0, + 2.0, + 2.0 + ], + "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", + "container_id": "ta-01M7GFSN1CJ9VQMYM83YJZ6YCR", + "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_seconds": [ + 2.8047221360000005 + ], + "max_absolute_error": 0, + "remote_seconds": 22.803371427000002 + }, + { + "case": "inline_simple", + "variant": 1, + "enabled": true, + "repeat": 0, + "extension_name": "pch_matrix_inline_simple_1_1", + "seconds": 10.122365914999998, + "timing_samples": [ + 10.122365914999998 + ], + "host_compile_seconds": 8.959541608, + "pch_consumed": true, + "hit_keys": [ + "cd04740b9ab36ac9a2a8d533e50fc30250ab89f341186829da169aefb0865f38" + ], + "cache_miss": false, + "correctness_checks": 6, + "outputs": [ + 3.0, + 3.0, + 3.0, + 3.0, + 3.0 + ], + "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", + "container_id": "ta-01M7GFSN13X2GVZ36HQVSN6S2R", + "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_seconds": [ + 2.7680481429999997 + ], + "max_absolute_error": 0, + "remote_seconds": 22.595921241 + }, + { + "case": "inline_simple", + "variant": 1, + "enabled": false, + "repeat": 0, + "extension_name": "pch_matrix_inline_simple_1_0", + "seconds": 19.861022294999998, + "timing_samples": [ + 19.861022294999998 + ], + "host_compile_seconds": 19.03845912, + "pch_consumed": false, + "hit_keys": [], + "cache_miss": false, + "correctness_checks": 6, + "outputs": [ + 3.0, + 3.0, + 3.0, + 3.0, + 3.0 + ], + "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", + "container_id": "ta-01M7GFSPSNZV6D195BH3PEWKYR", + "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_seconds": [ + 2.0814953679999997 + ], + "max_absolute_error": 0, + "remote_seconds": 41.715831361 + }, + { + "case": "cutlass_gemm_implicit", + "variant": 0, + "enabled": false, + "repeat": 0, + "extension_name": "pch_matrix_cutlass_gemm_implicit_0_0", + "seconds": 54.753927797, + "timing_samples": [ + 54.753927797 + ], + "host_compile_seconds": 23.725823061, + "pch_consumed": false, + "hit_keys": [], + "cache_miss": false, + "correctness_checks": 10, + "outputs": [ + 8.0 + ], + "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", + "container_id": "ta-01M7GFTQD7VSWT2AGTQ9041GCR", + "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_seconds": [ + 53.95127715 + ], + "max_absolute_error": 4.2646570364013314e-06, + "remote_seconds": 77.437517225 + }, + { + "case": "cutlass_gemm_implicit", + "variant": 0, + "enabled": true, + "repeat": 0, + "extension_name": "pch_matrix_cutlass_gemm_implicit_0_1", + "seconds": 55.171705347, + "timing_samples": [ + 55.171705347 + ], + "host_compile_seconds": 7.705170879000001, + "pch_consumed": true, + "hit_keys": [ + "cd04740b9ab36ac9a2a8d533e50fc30250ab89f341186829da169aefb0865f38" + ], + "cache_miss": false, + "correctness_checks": 10, + "outputs": [ + 8.0 + ], + "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", + "container_id": "ta-01M7GFTQJPYCBMYG758BQ09FXR", + "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_seconds": [ + 54.050255391 + ], + "max_absolute_error": 4.2646570364013314e-06, + "remote_seconds": 66.164975964 + }, + { + "case": "cutlass_gemm_implicit", + "variant": 1, + "enabled": true, + "repeat": 0, + "extension_name": "pch_matrix_cutlass_gemm_implicit_1_1", + "seconds": 46.287869813, + "timing_samples": [ + 46.287869813 + ], + "host_compile_seconds": 7.971731351, + "pch_consumed": true, + "hit_keys": [ + "cd04740b9ab36ac9a2a8d533e50fc30250ab89f341186829da169aefb0865f38" + ], + "cache_miss": false, + "correctness_checks": 10, + "outputs": [ + 16.0 + ], + "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", + "container_id": "ta-01M7GFVD2473KRK7WSX8NNEP2R", + "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_seconds": [ + 45.135409772 + ], + "max_absolute_error": 8.529314072802663e-06, + "remote_seconds": 58.033208515 + }, + { + "case": "cutlass_gemm_implicit", + "variant": 1, + "enabled": false, + "repeat": 0, + "extension_name": "pch_matrix_cutlass_gemm_implicit_1_0", + "seconds": 46.312597678, + "timing_samples": [ + 46.312597678 + ], + "host_compile_seconds": 19.034647205000002, + "pch_consumed": false, + "hit_keys": [], + "cache_miss": false, + "correctness_checks": 10, + "outputs": [ + 16.0 + ], + "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", + "container_id": "ta-01M7GFVEKV0YSM0R3DR3S4RAER", + "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_seconds": [ + 45.396892644999994 + ], + "max_absolute_error": 8.529314072802663e-06, + "remote_seconds": 68.968569437 + }, + { + "case": "inline_simple", + "variant": 0, + "enabled": false, + "repeat": 1, + "extension_name": "pch_matrix_inline_simple_0_0", + "seconds": 19.752127795, + "timing_samples": [ + 19.752127795 + ], + "host_compile_seconds": 19.002610604, + "pch_consumed": false, + "hit_keys": [], + "cache_miss": false, + "correctness_checks": 6, + "outputs": [ + 2.0, + 2.0, + 2.0, + 2.0, + 2.0 + ], + "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", + "container_id": "ta-01M7GFWZZJ4FJB7NPRQJTQ8JSR", + "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_seconds": [ + 1.7609109730000005 + ], + "max_absolute_error": 0, + "remote_seconds": 42.332975843999996 + }, + { + "case": "inline_simple", + "variant": 0, + "enabled": true, + "repeat": 1, + "extension_name": "pch_matrix_inline_simple_0_1", + "seconds": 11.21638811, + "timing_samples": [ + 11.21638811 + ], + "host_compile_seconds": 10.272402451000001, + "pch_consumed": true, + "hit_keys": [ + "cd04740b9ab36ac9a2a8d533e50fc30250ab89f341186829da169aefb0865f38" + ], + "cache_miss": false, + "correctness_checks": 6, + "outputs": [ + 2.0, + 2.0, + 2.0, + 2.0, + 2.0 + ], + "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", + "container_id": "ta-01M7GG0KV2AHWV84NN3ECQX0QR", + "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_seconds": [ + 1.6058505829999996 + ], + "max_absolute_error": 0, + "remote_seconds": 22.183825709 + }, + { + "case": "inline_simple", + "variant": 1, + "enabled": true, + "repeat": 1, + "extension_name": "pch_matrix_inline_simple_1_1", + "seconds": 10.608822302, + "timing_samples": [ + 10.608822302 + ], + "host_compile_seconds": 9.567511885, + "pch_consumed": true, + "hit_keys": [ + "cd04740b9ab36ac9a2a8d533e50fc30250ab89f341186829da169aefb0865f38" + ], + "cache_miss": false, + "correctness_checks": 6, + "outputs": [ + 3.0, + 3.0, + 3.0, + 3.0, + 3.0 + ], + "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", + "container_id": "ta-01M7GG0KV2QXH009EK3QFRJ3KR", + "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_seconds": [ + 1.722032843 + ], + "max_absolute_error": 0, + "remote_seconds": 21.330559701 + }, + { + "case": "inline_simple", + "variant": 1, + "enabled": false, + "repeat": 1, + "extension_name": "pch_matrix_inline_simple_1_0", + "seconds": 21.62613593, + "timing_samples": [ + 21.62613593 + ], + "host_compile_seconds": 20.728185403, + "pch_consumed": false, + "hit_keys": [], + "cache_miss": false, + "correctness_checks": 6, + "outputs": [ + 3.0, + 3.0, + 3.0, + 3.0, + 3.0 + ], + "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", + "container_id": "ta-01M7GG0KP789FE1B50S50BZD8R", + "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_seconds": [ + 1.7904973860000002 + ], + "max_absolute_error": 0, + "remote_seconds": 45.89947318 + }, + { + "case": "cutlass_gemm_implicit", + "variant": 0, + "enabled": false, + "repeat": 1, + "extension_name": "pch_matrix_cutlass_gemm_implicit_0_0", + "seconds": 51.057156396, + "timing_samples": [ + 51.057156396 + ], + "host_compile_seconds": 20.795003843, + "pch_consumed": false, + "hit_keys": [], + "cache_miss": false, + "correctness_checks": 10, + "outputs": [ + 8.0 + ], + "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", + "container_id": "ta-01M7GG0KP7MXTNR2TTPF0PZM5R", + "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_seconds": [ + 50.191617069 + ], + "max_absolute_error": 4.2646570364013314e-06, + "remote_seconds": 76.48378077699999 + }, + { + "case": "cutlass_gemm_implicit", + "variant": 0, + "enabled": true, + "repeat": 1, + "extension_name": "pch_matrix_cutlass_gemm_implicit_0_1", + "seconds": 43.304864159, + "timing_samples": [ + 43.304864159 + ], + "host_compile_seconds": 7.688989933999999, + "pch_consumed": true, + "hit_keys": [ + "cd04740b9ab36ac9a2a8d533e50fc30250ab89f341186829da169aefb0865f38" + ], + "cache_miss": false, + "correctness_checks": 10, + "outputs": [ + 8.0 + ], + "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", + "container_id": "ta-01M7GG1H2G9JENDFKF2E6J86FR", + "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_seconds": [ + 42.357097376 + ], + "max_absolute_error": 4.2646570364013314e-06, + "remote_seconds": 55.385258896 + }, + { + "case": "cutlass_gemm_implicit", + "variant": 1, + "enabled": true, + "repeat": 1, + "extension_name": "pch_matrix_cutlass_gemm_implicit_1_1", + "seconds": 44.357726661, + "timing_samples": [ + 44.357726661 + ], + "host_compile_seconds": 7.677523335, + "pch_consumed": true, + "hit_keys": [ + "cd04740b9ab36ac9a2a8d533e50fc30250ab89f341186829da169aefb0865f38" + ], + "cache_miss": false, + "correctness_checks": 10, + "outputs": [ + 16.0 + ], + "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", + "container_id": "ta-01M7GG1H2G5ZZTD7T5RB3RJ71R", + "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_seconds": [ + 43.322019976 + ], + "max_absolute_error": 8.529314072802663e-06, + "remote_seconds": 55.589907571000005 + }, + { + "case": "cutlass_gemm_implicit", + "variant": 1, + "enabled": false, + "repeat": 1, + "extension_name": "pch_matrix_cutlass_gemm_implicit_1_0", + "seconds": 46.449484709, + "timing_samples": [ + 46.449484709 + ], + "host_compile_seconds": 19.092076731, + "pch_consumed": false, + "hit_keys": [], + "cache_miss": false, + "correctness_checks": 10, + "outputs": [ + 16.0 + ], + "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", + "container_id": "ta-01M7GG27HD6MS78NKS8C24741R", + "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_seconds": [ + 45.656016197 + ], + "max_absolute_error": 8.529314072802663e-06, + "remote_seconds": 69.458907044 + } + ], + "reuse_smoke": { + "probe": { + "stock_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patched_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "paths": "/usr/local/cuda/nvvm/bin/cicc\n/usr/local/cuda/bin/nvcc.profile\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "nvcc_profiles": { + "/usr/local/cuda/bin/nvcc.profile": "\nTOP = $(_HERE_)/..\n\nCICC_PATH = $(TOP)/nvvm-next/bin\nNVVMIR_LIBRARY_DIR = $(TOP)/nvvm/libdevice\n\nLD_LIBRARY_PATH += $(TOP)/lib:\nPATH += $(CICC_PATH):$(_HERE_):\n\nINCLUDES += \"-I$(TOP)/$(_TARGET_DIR_)/include\" $(_SPACE_)\nSYSTEM_INCLUDES += \"-isystem\" \"$(TOP)/$(_TARGET_DIR_)/include/cccl\" $(_SPACE_)\n\nLIBRARIES =+ $(_SPACE_) \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)/stubs\" \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)\"\n\nCUDAFE_FLAGS +=\nPTXAS_FLAGS +=\n" + }, + "pch_help": " Specify options directly to library manager.\n\n--pch --pch \n Enable precompiled header (PCH) processing. When specified, each compilation\n searches the PCH directory (see --pch-dir) for a reusable precompiled header,\n and creates a new one when none is found.\n\n--pch-dir --pch-dir \n Specify the directory in which precompiled header files are read from and\n written to. By default this is the current working directory. Ignored unless\n --pch, --create-pch, or --use-pch is enabled.\n\n--pch-verbose --pch-verbose \n Print (to stderr) messages related to precompiled header processing. Ignored\n unless --pch, --create-pch, or --use-pch is enabled.\n\n--create-pch --create-pch \n Manually create a precompiled header (PCH) at the first PCH terminal stop-point\n (i.e., #pragma hdrstop) and save it under the specified name. Intermediate\n stop points (#pragma hdrstop++) are ignored. The directory into which the\n PCH file is written is determined by --pch-dir (or is the current working\n directory if not specified). This option forces a PCH file to be created\n (any existing PCH file with the same name is overwritten). This option is\n mutually exclusive with '--use-pch' and '--pch'.\n\n--use-pch --use-pch \n If a precompiled header (PCH) with the specified name exists in the directory\n determined by '--pch-dir' (or the current working directory if not specified),\n try to use it in compilation.This option is mutually exclusive with '--create-pch'\n and '--pch'.\n\n--ptxas-options ,... (-Xptxas) \n Specify options directly to ptxas, the PTX optimizing assembler.\n\n--nvlink-options ,... (-Xnvlink) \n Specify options directly to nvlink.\n\n--nvprune-options ,... (-Xnvprune) \n Specify options directly to nvprune (used with -prune).\n\n\nMiscellaneous options for guiding the compiler driver.\n======================================================", + "strace": null, + "smoke": [ + { + "arch": "sm_75", + "command": [ + "/usr/local/cuda/bin/nvcc", + "-arch=sm_75", + "-c", + "/tmp/tmp0a2ctqns/smoke.cu", + "-o", + "/tmp/tmp0a2ctqns/sm_75.o" + ], + "dryrun_returncode": 0, + "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmp0a2ctqns/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp0a2ctqns/smoke.cu\" --orig_src_path_name \"/tmp/tmp0a2ctqns/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmp0a2ctqns/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp0a2ctqns/smoke.cu\" --orig_src_path_name \"/tmp/tmp0a2ctqns/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_0000001e_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_0000001e_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_0000001e_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmp0a2ctqns/sm_75.o\" \n", + "next_backend_selected": true, + "returncode": 0, + "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmp0a2ctqns/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp0a2ctqns/smoke.cu\" --orig_src_path_name \"/tmp/tmp0a2ctqns/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmp0a2ctqns/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp0a2ctqns/smoke.cu\" --orig_src_path_name \"/tmp/tmp0a2ctqns/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000021_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000021_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000021_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmp0a2ctqns/sm_75.o\" \n", + "object_bytes": 10200 + }, + { + "arch": "sm_100", + "command": [ + "/usr/local/cuda/bin/nvcc", + "-arch=sm_100", + "-c", + "/tmp/tmp0a2ctqns/smoke.cu", + "-o", + "/tmp/tmp0a2ctqns/sm_100.o" + ], + "dryrun_returncode": 0, + "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmp0a2ctqns/smoke.cu\" -o \"/tmp/tmpxft_00000036_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp0a2ctqns/smoke.cu\" --orig_src_path_name \"/tmp/tmp0a2ctqns/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000036_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000036_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000036_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmp0a2ctqns/smoke.cu\" -o \"/tmp/tmpxft_00000036_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp0a2ctqns/smoke.cu\" --orig_src_path_name \"/tmp/tmp0a2ctqns/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000036_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000036_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000036_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000036_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000036_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000036_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000036_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000036_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000036_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000036_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmp0a2ctqns/sm_100.o\" \n", + "next_backend_selected": true, + "returncode": 0, + "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmp0a2ctqns/smoke.cu\" -o \"/tmp/tmpxft_00000039_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp0a2ctqns/smoke.cu\" --orig_src_path_name \"/tmp/tmp0a2ctqns/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000039_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000039_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000039_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmp0a2ctqns/smoke.cu\" -o \"/tmp/tmpxft_00000039_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp0a2ctqns/smoke.cu\" --orig_src_path_name \"/tmp/tmp0a2ctqns/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000039_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000039_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000039_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000039_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000039_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000039_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000039_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000039_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000039_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000039_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmp0a2ctqns/sm_100.o\" \n", + "object_bytes": 12504 + } + ], + "compatible": true + }, + "rows": [ + { + "case": "simple", + "variant": 0, + "mode": "patched_all_pch", + "repeat": 0, + "extension_name": "kb_nvcc_simple_0", + "returncode": 0, + "log": "[1/3] /tmp/tmp8ze6n0e1/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp8ze6n0e1/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp8ze6n0e1/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmp8ze6n0e1/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmp8ze6n0e1/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", + "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", + "container_id": "ta-01M7GG5X2J0T5Z7DF200Z9AWXR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", + "process_seconds": 33.416754872, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp8ze6n0e1/main.cpp", + "-o", + "main.o" + ], + "seconds": 10.638259281000002, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmp8ze6n0e1/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" + ], + "seconds": 27.723024821, + "returncode": 0 + } + ], + "seconds": 28.746708839999997, + "outputs": [ + 2.0, + 2.0, + 2.0, + 2.0, + 2.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp8ze6n0e1/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp8ze6n0e1/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp8ze6n0e1/cuda.cu\" --orig_src_path_name \"/tmp/tmp8ze6n0e1/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp8ze6n0e1/cuda.cu\" --orig_src_path_name \"/tmp/tmp8ze6n0e1/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmp8ze6n0e1/diagnostic.cuda.o\" \n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 65.051380577 + }, + { + "case": "simple", + "variant": 1, + "mode": "patched_all_pch", + "repeat": 0, + "extension_name": "kb_nvcc_simple_1", + "returncode": 0, + "log": "[1/3] /tmp/tmpzsqcjep0/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpzsqcjep0/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpzsqcjep0/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpzsqcjep0/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmpzsqcjep0/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", + "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", + "container_id": "ta-01M7GG5X2JWF1E5D985MF82VDR", + "image_id": "im-f00RGIxc0I83yIrG8DurE1", + "gpu": "Tesla T4", + "torch": "2.12.0+cu130", + "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", + "process_seconds": 33.46370826, + "cxx_commands": [ + { + "args": [ + "-MMD", + "-MF", + "main.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpzsqcjep0/main.cpp", + "-o", + "main.o" + ], + "seconds": 10.227091994, + "returncode": 0 + } + ], + "nvcc_commands": [ + { + "args": [ + "-MD", + "-MF", + "cuda.cuda.o.d", + "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "-c", + "/tmp/tmpzsqcjep0/cuda.cu", + "-o", + "cuda.cuda.o", + "-include", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", + "--pch", + "--pch-dir", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" + ], + "seconds": 27.899412197000004, + "returncode": 0 + } + ], + "seconds": 28.910585837000003, + "outputs": [ + 3.0, + 3.0, + 3.0, + 3.0, + 3.0 + ], + "max_absolute_error": 0, + "correctness_checks": 6, + "host_pch_consumed": true, + "host_hit_keys": [ + "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" + ], + "nvcc_diagnostic": { + "returncode": 0, + "stdout": "", + "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmpzsqcjep0/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmpzsqcjep0/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpzsqcjep0/cuda.cu\" --orig_src_path_name \"/tmp/tmpzsqcjep0/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpzsqcjep0/cuda.cu\" --orig_src_path_name \"/tmp/tmpzsqcjep0/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpzsqcjep0/diagnostic.cuda.o\" \n" + }, + "device_pch_files_used": [ + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" + ], + "device_pch_files_created": [], + "device_pch_consumed": true, + "remote_seconds": 64.866187693 + } + ] + }, + "warmup": { + "probe": { + "stock_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", + "patched_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "paths": "/usr/local/cuda/nvvm/bin/cicc\n/usr/local/cuda/bin/nvcc.profile\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" + } + ], + "nvcc_profiles": { + "/usr/local/cuda/bin/nvcc.profile": "\nTOP = $(_HERE_)/..\n\nCICC_PATH = $(TOP)/nvvm-next/bin\nNVVMIR_LIBRARY_DIR = $(TOP)/nvvm/libdevice\n\nLD_LIBRARY_PATH += $(TOP)/lib:\nPATH += $(CICC_PATH):$(_HERE_):\n\nINCLUDES += \"-I$(TOP)/$(_TARGET_DIR_)/include\" $(_SPACE_)\nSYSTEM_INCLUDES += \"-isystem\" \"$(TOP)/$(_TARGET_DIR_)/include/cccl\" $(_SPACE_)\n\nLIBRARIES =+ $(_SPACE_) \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)/stubs\" \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)\"\n\nCUDAFE_FLAGS +=\nPTXAS_FLAGS +=\n" + }, + "pch_help": " Specify options directly to library manager.\n\n--pch --pch \n Enable precompiled header (PCH) processing. When specified, each compilation\n searches the PCH directory (see --pch-dir) for a reusable precompiled header,\n and creates a new one when none is found.\n\n--pch-dir --pch-dir \n Specify the directory in which precompiled header files are read from and\n written to. By default this is the current working directory. Ignored unless\n --pch, --create-pch, or --use-pch is enabled.\n\n--pch-verbose --pch-verbose \n Print (to stderr) messages related to precompiled header processing. Ignored\n unless --pch, --create-pch, or --use-pch is enabled.\n\n--create-pch --create-pch \n Manually create a precompiled header (PCH) at the first PCH terminal stop-point\n (i.e., #pragma hdrstop) and save it under the specified name. Intermediate\n stop points (#pragma hdrstop++) are ignored. The directory into which the\n PCH file is written is determined by --pch-dir (or is the current working\n directory if not specified). This option forces a PCH file to be created\n (any existing PCH file with the same name is overwritten). This option is\n mutually exclusive with '--use-pch' and '--pch'.\n\n--use-pch --use-pch \n If a precompiled header (PCH) with the specified name exists in the directory\n determined by '--pch-dir' (or the current working directory if not specified),\n try to use it in compilation.This option is mutually exclusive with '--create-pch'\n and '--pch'.\n\n--ptxas-options ,... (-Xptxas) \n Specify options directly to ptxas, the PTX optimizing assembler.\n\n--nvlink-options ,... (-Xnvlink) \n Specify options directly to nvlink.\n\n--nvprune-options ,... (-Xnvprune) \n Specify options directly to nvprune (used with -prune).\n\n\nMiscellaneous options for guiding the compiler driver.\n======================================================", + "strace": null, + "smoke": [ + { + "arch": "sm_75", + "command": [ + "/usr/local/cuda/bin/nvcc", + "-arch=sm_75", + "-c", + "/tmp/tmps_9b9md0/smoke.cu", + "-o", + "/tmp/tmps_9b9md0/sm_75.o" + ], + "dryrun_returncode": 0, + "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmps_9b9md0/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmps_9b9md0/smoke.cu\" --orig_src_path_name \"/tmp/tmps_9b9md0/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmps_9b9md0/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmps_9b9md0/smoke.cu\" --orig_src_path_name \"/tmp/tmps_9b9md0/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_0000001e_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_0000001e_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_0000001e_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmps_9b9md0/sm_75.o\" \n", + "next_backend_selected": true, + "returncode": 0, + "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmps_9b9md0/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmps_9b9md0/smoke.cu\" --orig_src_path_name \"/tmp/tmps_9b9md0/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmps_9b9md0/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmps_9b9md0/smoke.cu\" --orig_src_path_name \"/tmp/tmps_9b9md0/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000021_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000021_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000021_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmps_9b9md0/sm_75.o\" \n", + "object_bytes": 10200 + }, + { + "arch": "sm_100", + "command": [ + "/usr/local/cuda/bin/nvcc", + "-arch=sm_100", + "-c", + "/tmp/tmps_9b9md0/smoke.cu", + "-o", + "/tmp/tmps_9b9md0/sm_100.o" + ], + "dryrun_returncode": 0, + "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmps_9b9md0/smoke.cu\" -o \"/tmp/tmpxft_00000036_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmps_9b9md0/smoke.cu\" --orig_src_path_name \"/tmp/tmps_9b9md0/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000036_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000036_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000036_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmps_9b9md0/smoke.cu\" -o \"/tmp/tmpxft_00000036_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmps_9b9md0/smoke.cu\" --orig_src_path_name \"/tmp/tmps_9b9md0/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000036_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000036_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000036_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000036_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000036_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000036_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000036_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000036_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000036_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000036_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmps_9b9md0/sm_100.o\" \n", + "next_backend_selected": true, + "returncode": 0, + "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmps_9b9md0/smoke.cu\" -o \"/tmp/tmpxft_00000039_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmps_9b9md0/smoke.cu\" --orig_src_path_name \"/tmp/tmps_9b9md0/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000039_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000039_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000039_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmps_9b9md0/smoke.cu\" -o \"/tmp/tmpxft_00000039_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmps_9b9md0/smoke.cu\" --orig_src_path_name \"/tmp/tmps_9b9md0/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000039_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000039_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000039_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000039_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000039_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000039_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000039_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000039_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000039_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000039_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmps_9b9md0/sm_100.o\" \n", + "object_bytes": 12504 + } + ], + "compatible": true + }, + "rows": [], + "warmup": { + "host_seconds": 72.258458109, + "host_log": "[1/2] /opt/kernelbot-pch/compiler.py -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kernelbot_pch_warm_1_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpnw106ee9/main.cpp -o main.o \n[kernelbot-pch] built 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/2] /opt/kernelbot-pch/compiler.py main.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kernelbot_pch_warm_1_0.so\nW1115 20:57:29.119000 26 site-packages/torch/utils/cpp_extension.py:140] No CUDA runtime is found, using CUDA_HOME='/usr/local/cuda'\n", + "device": [ + { + "case": "simple", + "key": "8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", + "command": [ + "/usr/local/cuda/bin/nvcc", + "-DTORCH_EXTENSION_NAME=kb_nvcc_pch_warmer", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + "--pch-dir", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch", + "--pch", + "--pch-verbose", + "-x", + "cu", + "-c", + "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", + "-o", + "/tmp/tmp_hj5fj_n/warm.o" + ], + "seconds": 44.738656711999994, + "stdout": "", + "stderr": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh:1:9: warning: #pragma once in main file\n 1 | #pragma once\n | ^~~~\n\"/tmp/tmpxft_00000046_00000000-5_gpumode.cpp4.ii\": creating precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch.tmp.77\"\n/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh:1:9: warning: #pragma once in main file\n 1 | #pragma once\n | ^~~~\n\"/tmp/tmpxft_00000046_00000000-7_gpumode.cpp1.ii\": creating precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch.tmp.82\"\n", + "returncode": 0, + "files": [ + { + "path": "pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch", + "bytes": 1229062144 + }, + { + "path": "pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", + "bytes": 1295241216 + } + ] + }, + { + "case": "cutlass", + "key": "9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", + "command": [ + "/usr/local/cuda/bin/nvcc", + "-DTORCH_EXTENSION_NAME=kb_nvcc_pch_warmer", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include", + "-isystem", + "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", + "-isystem", + "/usr/local/cuda/include", + "-isystem", + "/usr/local/include/python3.13", + "-D__CUDA_NO_HALF_OPERATORS__", + "-D__CUDA_NO_HALF_CONVERSIONS__", + "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", + "-D__CUDA_NO_HALF2_OPERATORS__", + "--expt-relaxed-constexpr", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", + "-std=c++20", + "--pch-dir", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch", + "--pch", + "--pch-verbose", + "-x", + "cu", + "-c", + "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", + "-o", + "/tmp/tmpw_lkzwtz/warm.o" + ], + "seconds": 51.79863217900001, + "stdout": "", + "stderr": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh:1:9: warning: #pragma once in main file\n 1 | #pragma once\n | ^~~~\n\"/tmp/tmpxft_00000070_00000000-5_gpumode.cpp4.ii\": creating precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch.tmp.119\"\n/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh:1:9: warning: #pragma once in main file\n 1 | #pragma once\n | ^~~~\n\"/tmp/tmpxft_00000070_00000000-7_gpumode.cpp1.ii\": creating precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch.tmp.124\"\n", + "returncode": 0, + "files": [ + { + "path": "pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch", + "bytes": 1443643392 + }, + { + "path": "pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", + "bytes": 1556054016 + } + ] + } + ] + } + }, + "initial_cuda133_incompatibility": { + "probe": { + "stock_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Fri_Apr_24_07:22:02_PM_PDT_2026\nCuda compilation tools, release 13.3, V13.3.33\nBuild cuda_13.3.r13.3/compiler.37862127_0\n", + "patched_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", + "paths": "/usr/local/cuda/bin/nvcc.profile\n/usr/local/cuda/nvvm/bin/cicc\n", + "patch": [ + { + "file": "/usr/local/cuda/bin/nvcc", + "replaced_existing": true, + "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" + }, + { + "file": "/usr/local/cuda/bin/cudafe++", + "replaced_existing": true, + "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" + }, + { + "file": "/usr/local/cuda/nvvm/bin/cicc", + "replaced_existing": true, + "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" + }, + { + "file": "/usr/local/cuda/nvvm-next/bin/cicc", + "replaced_existing": false, + "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + { + "file": "/usr/local/cuda/bin/nvcc.profile", + "replaced_existing": true, + "sha256": "2cb235ee19bab06e9a280fa888d489c4ac964b40279db1f050e8fda2e9fe9f60" + } + ], + "nvcc_profiles": { + "/usr/local/cuda/bin/nvcc.profile": "\nTOP = $(_HERE_)/..\n\nCICC_PATH = $(TOP)/$(_NVVM_BRANCH_)/bin\nNVVMIR_LIBRARY_DIR = $(TOP)/nvvm/libdevice\n\nLD_LIBRARY_PATH += $(TOP)/lib:\nPATH += $(CICC_PATH):$(_HERE_):\n\nINCLUDES += \"-I$(TOP)/$(_TARGET_DIR_)/include\" $(_SPACE_)\nSYSTEM_INCLUDES += \"-isystem\" \"$(TOP)/$(_TARGET_DIR_)/include/cccl\" $(_SPACE_)\n\nLIBRARIES =+ $(_SPACE_) \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)/stubs\" \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)\"\n\nCUDAFE_FLAGS +=\nPTXAS_FLAGS +=\n" + }, + "pch_help": " Specify options directly to library manager.\n\n--pch --pch \n Enable precompiled header (PCH) processing. When specified, each compilation\n searches the PCH directory (see --pch-dir) for a reusable precompiled header,\n and creates a new one when none is found.\n\n--pch-dir --pch-dir \n Specify the directory in which precompiled header files are read from and\n written to. By default this is the current working directory. Ignored unless\n --pch, --create-pch, or --use-pch is enabled.\n\n--pch-verbose --pch-verbose \n Print (to stderr) messages related to precompiled header processing. Ignored\n unless --pch, --create-pch, or --use-pch is enabled.\n\n--create-pch --create-pch \n Manually create a precompiled header (PCH) at the first PCH terminal stop-point\n (i.e., #pragma hdrstop) and save it under the specified name. Intermediate\n stop points (#pragma hdrstop++) are ignored. The directory into which the\n PCH file is written is determined by --pch-dir (or is the current working\n directory if not specified). This option forces a PCH file to be created\n (any existing PCH file with the same name is overwritten). This option is\n mutually exclusive with '--use-pch' and '--pch'.\n\n--use-pch --use-pch \n If a precompiled header (PCH) with the specified name exists in the directory\n determined by '--pch-dir' (or the current working directory if not specified),\n try to use it in compilation.This option is mutually exclusive with '--create-pch'\n and '--pch'.\n\n--ptxas-options ,... (-Xptxas) \n Specify options directly to ptxas, the PTX optimizing assembler.\n\n--nvlink-options ,... (-Xnvlink) \n Specify options directly to nvlink.\n\n--nvprune-options ,... (-Xnvprune) \n Specify options directly to nvprune (used with -prune).\n\n\nMiscellaneous options for guiding the compiler driver.\n======================================================", + "strace": null, + "smoke": [ + { + "arch": "sm_75", + "command": [ + "/usr/local/cuda/bin/nvcc", + "-arch=sm_75", + "-c", + "/tmp/tmpyfjiygy_/smoke.cu", + "-o", + "/tmp/tmpyfjiygy_/sm_75.o" + ], + "dryrun_returncode": 0, + "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpyfjiygy_/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpyfjiygy_/smoke.cu\" --orig_src_path_name \"/tmp/tmpyfjiygy_/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpyfjiygy_/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpyfjiygy_/smoke.cu\" --orig_src_path_name \"/tmp/tmpyfjiygy_/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_0000001e_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_0000001e_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_0000001e_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpyfjiygy_/sm_75.o\" \n", + "next_backend_selected": false, + "returncode": 255, + "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpyfjiygy_/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpyfjiygy_/smoke.cu\" --orig_src_path_name \"/tmp/tmpyfjiygy_/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpyfjiygy_/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpyfjiygy_/smoke.cu\" --orig_src_path_name \"/tmp/tmpyfjiygy_/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000021_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \nptxas /tmp/tmpxft_00000021_00000000-6_smoke.ptx, line 9; fatal : Unsupported .version 9.4; current version is '9.3'\nptxas fatal : Ptx assembly aborted due to errors\n# --error 0xff --\n", + "object_bytes": 0 + }, + { + "arch": "sm_100", + "command": [ + "/usr/local/cuda/bin/nvcc", + "-arch=sm_100", + "-c", + "/tmp/tmpyfjiygy_/smoke.cu", + "-o", + "/tmp/tmpyfjiygy_/sm_100.o" + ], + "dryrun_returncode": 0, + "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpyfjiygy_/smoke.cu\" -o \"/tmp/tmpxft_00000030_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpyfjiygy_/smoke.cu\" --orig_src_path_name \"/tmp/tmpyfjiygy_/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000030_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000030_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000030_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000030_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpyfjiygy_/smoke.cu\" -o \"/tmp/tmpxft_00000030_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpyfjiygy_/smoke.cu\" --orig_src_path_name \"/tmp/tmpyfjiygy_/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000030_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000030_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000030_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000030_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000030_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000030_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000030_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000030_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000030_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000030_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000030_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000030_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000030_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000030_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpyfjiygy_/sm_100.o\" \n", + "next_backend_selected": false, + "returncode": 255, + "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpyfjiygy_/smoke.cu\" -o \"/tmp/tmpxft_00000033_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpyfjiygy_/smoke.cu\" --orig_src_path_name \"/tmp/tmpyfjiygy_/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000033_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000033_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000033_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000033_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpyfjiygy_/smoke.cu\" -o \"/tmp/tmpxft_00000033_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpyfjiygy_/smoke.cu\" --orig_src_path_name \"/tmp/tmpyfjiygy_/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000033_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000033_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000033_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000033_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000033_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000033_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000033_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000033_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000033_00000000-8_smoke.sm_100.cubin\" \nptxas /tmp/tmpxft_00000033_00000000-6_smoke.ptx, line 9; fatal : Unsupported .version 9.4; current version is '9.3'\nptxas fatal : Ptx assembly aborted due to errors\n# --error 0xff --\n", + "object_bytes": 0 + } + ], + "compatible": false + }, + "rows": [] + }, + "binary_sha256": { + "pid-drop/cudart/bin/x86_64_Linux_release/nvcc": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298", + "pid-drop/cudart/bin/x86_64_Linux_release/cudafe++": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442", + "pid-drop/cudart/bin/x86_64_Linux_release/nvvm/bin/cicc": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050", + "pid-drop/cudart/bin/x86_64_Linux_release/nvvm-next/bin/cicc": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" + }, + "modal_runs": { + "primary": "https://modal.com/apps/gpu-mode/main/ap-sCbGWCCsGHByigNhMrPUYA", + "supplemental": "https://modal.com/apps/gpu-mode/main/ap-P9owg8O1R3RlDGy0ozOSE1", + "warmup": "https://modal.com/apps/gpu-mode/main/ap-ypnw7T8Xntoe9WGjwycFyh", + "reuse_smoke": "https://modal.com/apps/gpu-mode/main/ap-ya4TMM2KIT440BrHtTc77E", + "production_cuda133": "https://modal.com/apps/gpu-mode/main/ap-42e4FtDMIK2OGSSHHcAg51" + } +} diff --git a/docs/modal-pch-cache.md b/docs/modal-pch-cache.md index 845900cb..4fabd3b0 100644 --- a/docs/modal-pch-cache.md +++ b/docs/modal-pch-cache.md @@ -5,7 +5,10 @@ The runner sets `CXX` to a GCC wrapper that loads a precompiled changes. Leave `use_pch=False`; PyTorch's separate native PCH option is unnecessary. The submission path assumes `load_inline()`. -Only the Torch host header is cached. CUDA/CUTLASS templates, user headers, kernel +An additional, opt-in [NVCC framework header cache](nvcc-pch-cache.md) can cache +the CUDA-side Torch/runtime includes with NVIDIA's private compiler drop. + +This host cache only precompiles the Torch host header. CUDA/CUTLASS templates, user headers, kernel objects, and shared libraries still compile in each submission container. Different sources and module names share a PCH when their host flags match; changing a user header after the Torch include still changes the compiled code. diff --git a/docs/nvcc-pch-benchmark.md b/docs/nvcc-pch-benchmark.md new file mode 100644 index 00000000..8bb8e20d --- /dev/null +++ b/docs/nvcc-pch-benchmark.md @@ -0,0 +1,214 @@ +# Experimental NVCC PCH benchmark + +`scripts/nvcc_pch_benchmark.py` tests NVIDIA's supplied PCH compiler against +ordinary `torch.utils.cpp_extension.load_inline()` builds. The existing +[Modal PCH cache](modal-pch-cache.md) precompiles the host-side Torch header with +GCC. This experiment adds NVCC header reuse from a separate shared Modal Volume. +It measures compilation latency; it does not optimize kernel execution. + +Measured on 2026-11-15: adding the supplied compiler and shared NVCC PCH to +an existing host-PCH setup reduced median build time **31.7% for the simple +implicit-header case (35.37s → 24.17s)** and **22.2% for CUTLASS +(43.72s → 34.03s)** on the same CUDA 13.4 toolkit. This is a useful incremental +improvement, approximately **1.46× and 1.29×**, with substantial CUTLASS timing +variation. Replacing the compiler without enabling its PCH did not improve +these medians. The experiment did not modify production. + +These are results for the original workload-specific forwarding headers. The +simple profile also precompiled `cuda_fp16.h`; the CUTLASS profile precompiled +`cutlass/gemm/device/gemm.h`. The [shipping runtime](nvcc-pch-cache.md) caches +only PyTorch's three implicit CUDA headers and leaves CUTLASS headers in each +submission. These POC speedups do not establish the final runtime's speedup. +Use the [release validation results](benchmarks/nvcc-pch-release.json) for that +separate implementation and execution path. + +## Toolkit compatibility + +The supplied compiler reports CUDA **13.4.99**. The initial attempt to overlay +its four binaries on KernelBot's CUDA 13.3 image failed: the compiler emitted +PTX 9.4, while the installed assembler supported PTX 9.3. Those failures are +compatibility results, not PCH performance measurements. + +The benchmark therefore installs the official CUDA 13.4 compiler/toolkit +dependencies, which resolved to compiler **13.4.92**, before applying the +experimental binaries. The stock and patched timing arms use this same 13.4 +environment. The official `cuda-nvcc-13-4` package supplies matching `ptxas`, +`nvlink`, and `fatbinary`; `libnvvm-13-4` supplies `cicc` and `libdevice`. +The JSON probe records the actual stock and experimental compiler versions. + +Both the official 13.3 and 13.4 `nvcc.profile` files select `nvvm/bin/cicc`. +Copying the supplied second backend to `nvvm-next/bin/cicc` does not select it. +For this bounded SM75/SM100 experiment, patched containers explicitly set: + +```text +CICC_PATH = $(TOP)/nvvm-next/bin +``` + +The probe checks the dispatched path and compiles a minimal kernel for both +`sm_75` and `sm_100`. SM100 is a **compile-only Blackwell smoke test**; it is +not a Blackwell runtime or performance measurement. The separate +[runtime installer](nvcc-pch-cache.md) checks all supported runner architectures; +this experiment makes no claim about older targets or their backend dispatch. + +## Comparison + +| Mode | Compiler | GCC host PCH | NVCC PCH | +| --- | --- | --- | --- | +| `stock_no_pch` | Official 13.4 | Off | Off | +| `stock_host_pch` | Official 13.4 | On | Off | +| `patched_host_pch` | Experimental 13.4.99 | On | Off | +| `patched_all_pch` | Experimental 13.4.99 | On | On | + +Compare `patched_all_pch` with `patched_host_pch` to isolate the added NVCC PCH +path. Compare it with `stock_host_pch` to estimate the combined effect of the +compiler replacement and NVCC PCH on the existing host-cache setup. + +Each timed build uses a fresh T4 container with four CPU cores, 16 GiB RAM, a +fresh build directory, and normal Ninja parallelism. Both workloads retain +PyTorch's default implicit CUDA headers: + +- `simple`: FP16 vector addition, with two launch/source variants. +- `cutlass`: an actual SM75 tensor-core GEMM, with two threadblock tile variants. + +Both variants change a user header and have distinct expected outputs. The +GEMM checks random and zero inputs, tile tails, and a non-default CUDA stream +against a float64 PyTorch reference. Each trial records `load_inline()` wall +time, individual C++/NVCC command times, correctness results, compiler versions, +binary hashes, and container/image identities. Warmup and diagnostic builds are +outside the reported build time. The command order alternates between variants +and repetitions to reduce ordering bias. + +Untimed diagnostic compilations verify actual cache consumption: GCC `-H` +must report a valid PCH, and NVCC `--pch-verbose` must report using both +`cudafe_` and `cicc_` PCH files from the expected read-only directory without +creating one. The actual timed-build logs independently confirm both hits. A successful build by itself is insufficient evidence of reuse. + +## Run + +Use an authenticated Modal environment. Extract the supplied package privately; +the directory below must contain `nvcc`, `cudafe++`, `nvvm/bin/cicc`, and +`nvvm-next/bin/cicc`. Do not commit the supplied binaries or credentials. + +```sh +export NVIDIA_PCH_BINARIES=/path/to/extracted/x86_64_Linux_release +export PYTHONPATH=src:src/runners:scripts + +# Optional: choose fresh, dedicated benchmark volumes. +export KERNELBOT_HOST_PCH_BENCHMARK_VOLUME=kernelbot-nvcc-host-pch-my-test +export KERNELBOT_NVCC_PCH_VOLUME=kernelbot-nvcc-pch-my-test + +# Check compiler versions, backend dispatch, and SM75/SM100 compilation. +uv run modal run scripts/nvcc_pch_benchmark.py \ + --probe-only --output /tmp/nvcc-pch-probe.json + +# Warm shared headers, then run all four modes and both source variants. +uv run modal run scripts/nvcc_pch_benchmark.py \ + --cases simple,cutlass --repeats 2 --output /tmp/nvcc-pch-results.json +``` + +`--warm-only` prepares and commits headers without timing trials. A subsequent +`--skip-warm` run reuses those entries and validates their identity. The harness +uses the temporary `kernelbot-nvcc-pch-benchmark` app and dedicated volumes; it +does not deploy over the production runner. Only the trusted warmer writes the +volumes. Trials mount both caches read-only and restrict Modal API access. + +## Integration requirements and limits + +In this benchmark, the warmer captures PyTorch's NVCC flags, then compiles a +forwarding `gpumode.cuh` containing the common Torch/CUDA headers and the workload's CUDA +or CUTLASS headers. The header ends with `#pragma hdrstop`. The warmer must specify +`-x cu -c gpumode.cuh -o /tmp/header.o` along with the matched flags and PCH +options; the bare `.cuh` command in the high-level setup note was rejected by +this NVCC build. Trials pass +`-include gpumode.cuh --pch --pch-dir ...`. Cache identity includes supplied +binary/profile hashes, the installed-header fingerprint, forwarding-header +contents, workload flags, and architecture. Exact commands are retained in the +results. The warmup and trials use different extension names, so successful +reuse also tests that per-module naming does not invalidate these profiles. + +The separate [runtime](nvcc-pch-cache.md) uses curated submission flag sets, +immutable compiler/header identities, and fallback on a miss. Its release +validator exercises the CPU precompile worker's 8 GiB memory budget and GPU +artifact replay; these original GPU trials used 16 GiB and did not exercise +that path. The runtime also narrows the forwarding header to preserve the +implicit include sequence. User source, kernel objects, and shared libraries +remain outside the shared PCH cache. + +This matrix covers two fixtures and one runtime GPU, not the submission +population. It excludes scheduling, image startup, PCH preparation, and the +existing [CPU-to-GPU artifact transfer](python-cpu-compilation.md) from the +`load_inline()` metric. Compiler subprocess times overlap under Ninja and must +not be summed to estimate wall time. The controlled CUDA 13.4 matrix isolates the cache/compiler changes. A +separate CUDA 13.3 baseline below measures the combined toolkit migration and +PCH effect for the same CUTLASS fixture. Fresh containers do not guarantee +cold underlying host or Modal storage caches. Two repetitions +per variant provide a small sample; retain individual measurements alongside +medians and avoid generalizing small differences. + +## Results + +Main matrix: four builds per cell, two source/tile variants × two repetitions. +Values are median `load_inline()` seconds, with the full min–max range. + +| CUDA 13.4 build mode | Simple, implicit headers | CUTLASS, implicit headers | +| --- | ---: | ---: | +| Stock, no PCH | 37.23 (35.75–42.89) | 48.82 (42.95–54.31) | +| Stock + existing host PCH | 35.37 (35.03–36.43) | 43.72 (41.50–48.83) | +| Patched compiler + host PCH | 36.01 (34.74–38.47) | 45.76 (41.73–51.06) | +| Patched compiler + host and NVCC PCH | **24.17 (23.93–29.14)** | **34.03 (28.44–43.66)** | + +Comparing the last two rows isolates NVCC PCH on the same patched compiler: +**1.49× simple / 1.34× CUTLASS**. The CUTLASS cached builds took 39–44s in the +first repetition and 28–29s in the second; underlying storage/cache state was +not controlled. The four-sample median is not a guarantee for every fresh +container, and its range overlaps the stock host-cache range. + +The fresh production-image CUDA 13.3 baseline measured CUTLASS at **48.75s +without PCH** and **45.32s with the existing host PCH**. Moving from that current +host-cache setup to CUDA 13.4 plus the supplied PCH path measured **34.03s**: +**1.33×, or 24.9% less compile time**, including the toolkit change. The old +benchmark's `inline_simple` case deliberately omits implicit CUDA headers and +is not comparable to the new `simple` fixture; its 10.37s host-cache result is +retained only as a separate minimal-header diagnostic. + +A supplemental run tested sequentially reading the PCH files before compilation, +with the read time included in the total. It did not improve the medians: + +| Two builds per cell | Direct shared-volume PCH | Prefetch + compilation | +| --- | ---: | ---: | +| Simple | 25.24s | 25.70s | +| CUTLASS | 32.90s | 35.44s | + +Prefetch itself took 1.51–1.60s for simple and 2.61–7.73s for CUTLASS. Keep this +supplemental run separate from the main matrix; its faster CUTLASS samples do +not replace the main result. + +Warm diagnostic phase timings explain the limit. Combined `cudafe++`/`cicc` +time fell from a median 23.82s to 7.61s for simple and 29.25s to 10.51s for +CUTLASS. GCC preprocessing and compilation still occupied roughly 13–19s. +These are second compilations after timing and correctness checks, so they +illustrate the remaining work rather than measuring first-use storage overhead. + +One-time CUDA PCH creation took **44.74s for simple** and **51.80s for CUTLASS**. +Each profile produced two files, totaling **2.52 GB** and **3.00 GB** respectively +(2.35 and 2.79 GiB). The host warmup took 72.26s. These stage timings exclude +volume commits and flag-capture setup. Storage and profile count need budgeting +when expanding to multiple architectures and submission flag sets. + +All **492 correctness checks** passed across **62 distinct single-use T4 +containers**: 32 main builds, 12 supplemental builds, 16 CUDA 13.3 baseline +builds, and two reuse smoke builds. Maximum GEMM absolute error was **8.53e-6**. +All **18 NVCC-PCH builds** confirmed both frontend hits in their actual timed +logs and in independent diagnostics. Different module names, kernel sources, +and user-header values reused the same profile and produced the expected +distinct outputs. No kernel object or shared-library cache was used. + +[Complete measurements, compiler hashes, commands, warmup logs, and validation](benchmarks/nvcc-pch.json). + +Modal runs: + +- [Main 32-build matrix](https://modal.com/apps/gpu-mode/main/ap-sCbGWCCsGHByigNhMrPUYA) +- [Prefetch and phase diagnostics](https://modal.com/apps/gpu-mode/main/ap-P9owg8O1R3RlDGy0ozOSE1) +- [Production-image CUDA 13.3 baseline](https://modal.com/apps/gpu-mode/main/ap-42e4FtDMIK2OGSSHHcAg51) +- [Shared-volume warmup](https://modal.com/apps/gpu-mode/main/ap-ypnw7T8Xntoe9WGjwycFyh) +- [Cross-module reuse smoke](https://modal.com/apps/gpu-mode/main/ap-ya4TMM2KIT440BrHtTc77E) diff --git a/docs/nvcc-pch-cache.md b/docs/nvcc-pch-cache.md new file mode 100644 index 00000000..242c77a3 --- /dev/null +++ b/docs/nvcc-pch-cache.md @@ -0,0 +1,205 @@ +# NVCC framework header cache + +KernelBot can use NVIDIA's private PCH compiler drop to precompile the framework +headers that PyTorch inserts into `load_inline()` CUDA sources. Matching builds +reuse those headers from a shared, read-only Modal Volume. Users keep their +existing submissions and compiler flags. + +This is a separately selectable CUDA 13.4 toolkit on the existing runner base +image. Both the toolchain and the NVCC cache are disabled by default. The existing +[host PCH cache](modal-pch-cache.md) +and [CPU-to-GPU artifact transfer](python-cpu-compilation.md) work alongside it. +The [initial benchmark](nvcc-pch-benchmark.md) used broader, workload-specific +headers, including a CUTLASS header. Its speedups do not measure this narrower +framework-only cache. [Release validation results](benchmarks/nvcc-pch-release.json) +record measurements for the implemented CPU-to-GPU path. + +## Compatibility + +The forwarding header contains, in order, `torch/types.h`, `cuda.h`, and +`cuda_runtime.h`. These are the exact implicit CUDA includes in the pinned +PyTorch 2.12 image; `torch/types.h` includes ATen transitively. CUTLASS and user +headers are not precompiled. Adding arbitrary common headers before user code +could change macro and include-order behavior, so header-frequency data alone +does not make an include eligible. + +The wrapper requires this source prefix and an exact match for the compiler +flags and header-search environment. It preserves flag order, macros, and the +real module name in the compiler invocation. Cache lookup omits generated +source/output/dependency paths and the per-module name macro. +Unknown profiles, custom include paths, compiler overrides, and sources without +the prefix compile normally. `no_implicit_headers=True` does not gain injected +framework headers. A rejected cache retries the original compile once; genuine +submission errors still fail. + +The compiler is NVIDIA's `13.4.99` drop, paired with toolkit package `13.4.3-1` +and CUDA 13.4.92 compiler packages. The complete toolkit preserves cuTile, +cuBLAS, sanitizer, and other existing development tools when CUDA paths switch. +CUDA 13.3's `ptxas` cannot assemble the emitted PTX 9.4. The installer +verifies all four supplied binary hashes and explicitly selects `nvvm-next` for +the supported GPUs. Image-build smoke checks cover every deployed architecture +and the architecture-specific Blackwell profiles. + +## Observed profiles and candidate coverage + +The [profile inventory](benchmarks/nvcc-pch-profiles.json) covers the +2026-11-13 private-diff snapshot: **366,803 unique submissions** in 367,118 rows. +The public baseline is excluded; this is not the complete GPUMODE submission +history. Runner metadata identifies 124,736 NVIDIA submissions, leaves 242,065 +missing or unrecognized, and identifies two other runners. Those cohorts remain +separate throughout the analysis. + +A conservative static pass examines calls with an explicit `cuda_sources` +keyword. It resolves simple literals and unambiguous assignments in their lexical +scope, rejecting reassigned, mutated, escaped, or otherwise unresolved values. +Calls using `**kwargs` are never assumed to have default flags. An eligible +candidate also needs a verified PyTorch `load_inline` import, resolved CUDA +source, implicit headers enabled, no custom include paths, and flags from the +bounded inventory allowlist. Flags retain their original order. This finds +20,411 eligible candidate submissions among 70,461 static CUDA `load_inline` +submissions with known NVIDIA runners, and 25,996 among 112,520 with unknown +runners. + +The default catalog uses the six most frequent eligible ordered tuples in the +**known NVIDIA cohort**: + +| Ordered `extra_cuda_cflags` | Known NVIDIA submissions | Unknown-runner submissions | +| --- | ---: | ---: | +| `['-O3']` | 5,419 | 10,457 | +| `['-O3', '--use_fast_math']` | 3,358 | 6,695 | +| `[]` | 2,670 | 2,556 | +| `['-O3', '--use_fast_math', '-arch=sm_100a']` | 2,609 | 113 | +| `['-O3', '-arch=sm_100', '--use_fast_math']` | 771 | 99 | +| `['-O3', '-std=c++17', '-arch=sm_80', '--use_fast_math']` | 628 | 174 | + +The three architecture-neutral tuples are warmed for `sm75`, `sm80`, `sm89`, +`sm90a`, and `sm100`; the three explicit-architecture tuples add one profile each, +for **18 default profiles**. The warmer captures PyTorch's remaining generated +flags. Different ordering, include paths, or generated architecture flags still +require their own exact profile. + +Each table row counts a submission once per distinct tuple. The coverage below +uses a set union per submission, so a submission containing several selected +tuples is counted once: + +| Runner cohort | Static CUDA `load_inline` submissions | Any eligible call with a selected tuple | All observed calls eligible and selected | +| --- | ---: | ---: | ---: | +| Known NVIDIA | 70,461 | 15,421 (21.89%) | 15,176 | +| Unknown | 112,520 | 18,283 (16.25%) | 16,356 | +| Combined analyzed cohorts | 182,981 | 33,704 (18.42%) | 31,532 | + +These are **static flag-tuple candidate counts**, not executed NVCC command +counts or measured cache-hit rates. Matching the real architecture, framework +prefix, toolchain, and environment remains necessary. Dynamic Python and +positional CUDA arguments fall outside this inventory's coverage. In particular, +88,425 analyzed submissions contain a call with `no_implicit_headers=True`; those +calls keep their existing behavior and do not receive the master header. + +Resolved source includes also show why header popularity cannot define the +master header. The following frequencies count known-NVIDIA static CUDA +`load_inline` submissions once per explicit include across their resolved C++ +and CUDA source strings; they do not count PyTorch's generated implicit includes: + +| Explicit source include | Submissions | In the master header | +| --- | ---: | --- | +| `cuda_runtime.h` | 50,015 | Yes | +| `cuda_fp16.h` | 38,863 | No | +| `cuda.h` | 35,932 | Yes | +| `torch/extension.h` | 24,752 | No | +| `ATen/ATen.h` | 3,448 | Transitive through `torch/types.h` | +| `cutlass/cutlass.h` | 263 | No; CUTLASS excluded | +| `cute/tensor.hpp` | 232 | No; CuTe excluded | +| `torch/types.h` | 159 | Yes; normally inserted by PyTorch | + +Only the exact three-header implicit prefix described above is precompiled. +The published inventory contains aggregates, without submission IDs or private +source code. + +## Immutable cache and artifacts + +The cache layout is +`/kernelbot-nvcc-pch///`. +Each completed entry contains `gpumode.cuh`, both frontend PCH files, and a ready +manifest. The toolchain fingerprint covers the compiler drop, supporting CUDA +tools, installed CUDA package versions and headers, and the wrapper and warmer +sources. Profile keys add the ordered flags and relevant environment. + +Only the trusted CPU warmer can write the Volume. Submission workers mount it +read-only with Modal API access restricted. The warmer accepts fixed profile +names rather than submission-supplied code or flags, and publishes a manifest +only after both frontend files are complete. New generations coexist with old +ones so an image rollback retains its matching cache. + +CPU artifact packets include the NVCC toolchain identity. GPU replay checks that +identity and ordinary source/header dependencies. A PCH dependency uses its small +trusted manifest and file sizes instead of reading several GB to hash it on each +submission. Arbitrary PCH paths are not eligible for this shortcut. Changed user +headers and compiler environment variables still invalidate replay. + +## Prepare and validate a release + +Use KernelBot's Modal workspace. Keep the NVIDIA binaries outside this public +repository. The input directory must contain `nvcc`, `cudafe++`, +`nvvm/bin/cicc`, and `nvvm-next/bin/cicc`. + +```sh +export PYTHONPATH=src:src/runners +uv run python scripts/upload_nvcc_toolchain.py --source /private/path/to/binaries +uv run modal run scripts/validate_nvcc_pch.py --output /tmp/nvcc-pch-validation.json +``` + +The validator selects the new toolchain and isolated cache Volumes. Its default +matrix runs PCH on/off on T4, H100, and B200, plus T4 checks for unknown flags, +missing/corrupt caches, disabled implicit headers, changed user headers, and a +custom NVCC override. It exercises real CPU compilation, GPU artifact replay, +correctness, benchmarking, and ranking evaluation without publishing scores. +Use `--profiles` to select trusted catalog entries and `--repeats` for repeated +PCH-on/off pairs in fresh containers. Successful compilation alone is not a +cache hit: the validator requires both NVIDIA frontends to report consumption. +The [profile inventory](benchmarks/nvcc-pch-profiles.json) describes the sampled +submission population and static candidate coverage. Those counts are not +executed NVCC command counts or measured cache-hit rates. + +## Deploy and roll back + +First warm and deploy a separate canary app using a single consistent image: + +```sh +export KERNELBOT_MODAL_APP=discord-bot-runner-nvcc-pch-canary +export KERNELBOT_NVCC_TOOLCHAIN=1 +export KERNELBOT_NVCC_PCH=1 +uv run modal run src/runners/modal_runner.py::warm_pch +uv run modal run src/runners/modal_runner.py::warm_nvcc_pch +uv run modal deploy src/runners/modal_runner_archs.py +``` + +`ModalLauncher(app_name=...)` can target that app for non-scoring validation. +Do not split one scoring population randomly across compiler versions. Promote +the validated toolchain as a consistent runner release, recording the image and +toolchain fingerprint with the rollout. + +The main/dev deployment workflow reads these repository/environment variables: + +| Variable | Default | Purpose | +| --- | --- | --- | +| `KERNELBOT_NVCC_TOOLCHAIN` | `0` | Select the pinned CUDA 13.4 compiler image. | +| `KERNELBOT_NVCC_PCH` | `0` | Enable automatic NVCC PCH lookup. Requires the toolchain. | +| `KERNELBOT_NVCC_PCH_PROFILES` | empty | Comma-separated trusted profiles; empty uses the bounded default set. | +| `KERNELBOT_NVCC_TOOLCHAIN_VOLUME` | `kernelbot-nvcc-toolchains` | Private compiler input Volume. | +| `KERNELBOT_NVCC_PCH_VOLUME` | `kernelbot-nvcc-pch` | Device PCH Volume. | +| `KERNELBOT_PCH_VOLUME` | `kernelbot-torch-pch` | Host PCH Volume. | + +`warm_nvcc_pch --profiles ...` limits manual warmup to specific catalog entries. +Warm before replacing workers, since existing containers retain their Volume +snapshot. + +For a cache-only rollback, deploy the same toolchain with +`KERNELBOT_NVCC_PCH=0`. For a compiler rollback, deploy with both switches `0` to +restore the CUDA 13.3 image. Individual submissions can disable lookup with +`KERNELBOT_NVCC_PCH_DISABLE=1` or override `PYTORCH_NVCC`. + +Compilation emits `[kernelbot-nvcc-pch]` JSON events for `hit`, `miss`, `bypass`, +`disabled`, `fallback`, or `error`. A hit requires NVIDIA diagnostics confirming +both `cicc` and `cudafe` consumed the cache. Events include profile identity and +elapsed compiler time. `KERNELBOT_NVCC_PCH_EVENTS` optionally writes JSONL to a +container-local file. diff --git a/scripts/nvcc_pch_benchmark.py b/scripts/nvcc_pch_benchmark.py new file mode 100644 index 00000000..3450df44 --- /dev/null +++ b/scripts/nvcc_pch_benchmark.py @@ -0,0 +1,656 @@ +"""Isolated benchmark of NVIDIA's experimental NVCC PCH build. + +NVIDIA_PCH_BINARIES=/path/to/x86_64_Linux_release PYTHONPATH=src:src/runners:scripts \ + uv run modal run scripts/nvcc_pch_benchmark.py --probe-only + +No production app or production PCH volume is modified. +""" + +import json +import os +import statistics +from pathlib import Path + +import modal + +os.environ["KERNELBOT_PCH_VOLUME"] = os.environ.get( + "KERNELBOT_HOST_PCH_BENCHMARK_VOLUME", "kernelbot-nvcc-host-pch-benchmark-cuda134-20261115" +) +if os.environ["KERNELBOT_PCH_VOLUME"] == "kernelbot-torch-pch": + raise RuntimeError("Refusing to use the production host PCH volume for this benchmark") +from modal_pch_benchmark import CPP, CUDA, GEMM_CPP, GEMM_CUDA # noqa: E402 +from modal_runner import PCH_MOUNT, cuda_image_base, pch_volume # noqa: E402 + +app = modal.App("kernelbot-nvcc-pch-benchmark") +NVCC_MOUNT = "/kernelbot-nvcc-pch" +nvcc_volume = modal.Volume.from_name( + os.environ.get("KERNELBOT_NVCC_PCH_VOLUME", "kernelbot-nvcc-pch-benchmark-20261115"), + create_if_missing=True, +) +image = ( + cuda_image_base.apt_install( + "cuda-compiler-13-4", "cuda-cudart-dev-13-4", "cuda-driver-dev-13-4" + ) + .run_commands( + "ln -sfn /usr/local/cuda-13.4 /usr/local/cuda", + "nvcc --version", + "python /opt/kernelbot-pch/compiler.py --kernelbot-fingerprint", + ) + .add_local_python_source( + "libkernelbot", "modal_runner", "modal_runner_archs", "modal_pch_benchmark" + ) +) +if modal.is_local(): + binary_root = os.environ.get("NVIDIA_PCH_BINARIES") + if not binary_root: + raise RuntimeError( + "Set NVIDIA_PCH_BINARIES to the extracted x86_64_Linux_release directory" + ) + image = image.add_local_dir(binary_root, "/opt/nvidia-pch-binaries") + +CASES = ("simple", "cutlass") +DEFAULT_MODES = ("stock_no_pch", "stock_host_pch", "patched_host_pch", "patched_all_pch") +MODES = (*DEFAULT_MODES, "patched_all_pch_prefetch") +CUDA_FLAGS = [ + "-U__CUDA_NO_HALF_OPERATORS__", + "-U__CUDA_NO_HALF_CONVERSIONS__", + "-U__CUDA_NO_HALF2_OPERATORS__", + "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", + "--expt-relaxed-constexpr", + "-O3", + "-I/opt/cutlass/include", + "-I/opt/cutlass/tools/util/include", +] +OBSERVER = r"""#!/usr/bin/env python3 +import json, os, subprocess, sys, time +kind = "nvcc" if sys.argv[0].endswith("nvcc") else "cxx" +compiler = "/usr/local/cuda/bin/nvcc" if kind == "nvcc" else os.environ["PCH_TEST_COMPILER"] +args = sys.argv[1:] +if kind == "nvcc" and os.environ.get("PCH_CAPTURE") == "1" and "-c" in args: + with open(os.environ["PCH_TEST_COMMANDS"] + ".capture", "w") as f: + json.dump(args, f) + sys.exit(1) +if kind == "nvcc" and "-c" in args and os.environ.get("PCH_NVCC_ENTRY"): + entry = os.environ["PCH_NVCC_ENTRY"] + args = [*args, "-include", entry + "/gpumode.cuh", "--pch", "--pch-dir", entry + "/pch"] +started = time.perf_counter() +code = subprocess.call([compiler, *args]) +if "-c" in args: + with open(os.environ["PCH_TEST_COMMANDS"] + "." + kind, "a") as f: + f.write(json.dumps({ + "args": args, "seconds": time.perf_counter()-started, "returncode": code + }) + "\n") +sys.exit(code) +""" + +LOADER = r""" +import json, os, time +from pathlib import Path +import torch +from torch.utils.cpp_extension import load_inline +root = Path(os.environ["PCH_TEST_ROOT"]) +case = os.environ["PCH_TEST_CASE"] +variant = int(os.environ["PCH_TEST_VARIANT"]) +options = json.loads((root / "options.json").read_text()) +started = time.perf_counter() +prefetch_bytes = 0 +if os.environ.get("PCH_PREFETCH") == "1": + for pch in sorted((Path(os.environ["PCH_NVCC_ENTRY"]) / "pch").glob("*.pch")): + with pch.open("rb") as source: + while block := source.read(16 * 1024 * 1024): + prefetch_bytes += len(block) + assert prefetch_bytes > 0 and prefetch_bytes == int(os.environ["PCH_EXPECTED_BYTES"]), \ + "Prefetch did not read exactly the warmed PCH payload" +prefetch_seconds = time.perf_counter() - started if prefetch_bytes else 0 +compile_started = time.perf_counter() +module = load_inline(cpp_sources=(root / "cpp.txt").read_text(), + cuda_sources=(root / "cuda.txt").read_text(), functions=["add_cuda", "version"], + no_implicit_headers=False, verbose=True, build_directory=str(root), **options) +seconds = time.perf_counter() - started +load_inline_seconds = time.perf_counter() - compile_started +assert module.version() == variant + 1 +outputs, errors = [], [] +if case == "cutlass": + torch.manual_seed(17) + with torch.cuda.stream(torch.cuda.Stream()): + for m, n, k in ((128, 128, 64), (129, 136, 64), (256, 256, 128), (512, 512, 256)): + a = torch.randn(m, k, device="cuda", dtype=torch.float16) / 4 + b = torch.randn(n, k, device="cuda", dtype=torch.float16) / 4 + for zero in (False, True): + if zero: + a.zero_() + actual = module.add_cuda(a, b) + reference = (a.double() @ b.double().T) * (variant + 1) + torch.testing.assert_close(actual.double(), reference, rtol=2e-4, atol=2e-4) + errors.append(float((actual.double() - reference).abs().max())) + a = torch.ones(8, 8, device="cuda", dtype=torch.float16) + actual = module.add_cuda(a, a) + torch.testing.assert_close( + actual, torch.full_like(actual, 8 * (variant + 1)), rtol=0, atol=0 + ) + outputs.append(float(actual[0, 0])) +else: + for n in (1, 127, 128, 129, 1025): + a = torch.arange(n, device="cuda", dtype=torch.float16) % 8 + b = torch.ones_like(a) + actual = module.add_cuda(a, b) + torch.testing.assert_close(actual, a + b + (variant + 1), rtol=0, atol=0) + outputs.append(float(actual[0])) +(root / "result.json").write_text(json.dumps(dict( + seconds=seconds, load_inline_seconds=load_inline_seconds, + prefetch_seconds=prefetch_seconds, prefetch_bytes=prefetch_bytes, outputs=outputs, errors=errors +))) +""" + + +def prepare_build(root, case, variant, name, host_pch, entry=None, capture=False): + cpp = GEMM_CPP if case == "cutlass" else CPP + cuda = ( + GEMM_CUDA.replace("TILE_N", str(128 if variant == 0 else 64)) + if case == "cutlass" + else CUDA.replace("THREADS", str(256 if variant == 0 else 128)) + ) + header = f"#pragma once\nconstexpr int kVariant = {variant + 1};\n" + (root / "variant.h").write_text(header) + (root / "cpp.txt").write_text(cpp) + (root / "cuda.txt").write_text(cuda) + (root / "options.json").write_text( + json.dumps( + { + "name": name, + "extra_cuda_cflags": CUDA_FLAGS if case == "cutlass" else [], + } + ) + ) + for kind in ("cxx", "nvcc"): + p = root / f"observe-{kind}" + p.write_text(OBSERVER) + p.chmod(0o755) + env = dict(os.environ) + for key in ("MAX_JOBS", "KERNELBOT_PCH_DISABLE", "KERNELBOT_PCH_TRACE", "KERNELBOT_PCH_WRITE"): + env.pop(key, None) + env.update( + { + "TORCH_CUDA_ARCH_LIST": "7.5", + "CXX": str(root / "observe-cxx"), + "PYTORCH_NVCC": str(root / "observe-nvcc"), + "PCH_TEST_COMPILER": "/opt/kernelbot-pch/compiler.py" if host_pch else "/usr/bin/g++", + "PCH_TEST_ROOT": str(root), + "PCH_TEST_CASE": case, + "PCH_TEST_VARIANT": str(variant), + "PCH_TEST_COMMANDS": str(root / "commands"), + "PCH_CAPTURE": "1" if capture else "0", + "PCH_NVCC_ENTRY": str(entry) if entry else "", + } + ) + return env, cpp + cuda + header + + +def pch_identity(case, patched): + header = ( + "#pragma once\n#include \n#include \n" + "#include \n#include \n" + ) + header += ( + "#include \n" + if case == "cutlass" + else "#include \n" + ) + header += "#pragma hdrstop\n" + return { + "patch": patched, + "fingerprint": Path("/opt/kernelbot-pch/fingerprint").read_text().strip(), + "header": header, + "case": case, + "cuda_flags": CUDA_FLAGS if case == "cutlass" else [], + "arch": "7.5", + "protocol": 2, + } + + +def install_patch(): + """Overlay only the four supplied binaries inside this disposable container.""" + import hashlib + import re + import shutil + + rows = [] + for source_name, target_name in ( + ("nvcc", "bin/nvcc"), + ("cudafe++", "bin/cudafe++"), + ("nvvm/bin/cicc", "nvvm/bin/cicc"), + ("nvvm-next/bin/cicc", "nvvm-next/bin/cicc"), + ): + source = Path("/opt/nvidia-pch-binaries") / source_name + target = Path("/usr/local/cuda") / target_name + existed = target.exists() + target.parent.mkdir(parents=True, exist_ok=True) + shutil.copyfile(source, target) + target.chmod(0o755) + rows.append( + { + "file": str(target), + "replaced_existing": existed, + "sha256": hashlib.sha256(source.read_bytes()).hexdigest(), + } + ) + profile_path = Path("/usr/local/cuda/bin/nvcc.profile") + # The experiment uses SM75 and SM100, both on NVIDIA's newer backend. + # General production dispatch must also handle SM70 and earlier separately. + profile, count = re.subn( + r"^CICC_PATH\s*=.*$", + "CICC_PATH = $(TOP)/nvvm-next/bin", + profile_path.read_text(), + flags=re.MULTILINE, + ) + assert count == 1, "Expected exactly one CICC_PATH assignment in CUDA's profile" + profile_path.write_text(profile) + rows.append( + { + "file": str(profile_path), + "replaced_existing": True, + "sha256": hashlib.sha256(profile.encode()).hexdigest(), + } + ) + return rows + + +@app.function(image=image, cpu=4, memory=16384, timeout=300, single_use_containers=True) +def probe(): + import shutil + import subprocess + import tempfile + + stock_version = subprocess.check_output(["/usr/local/cuda/bin/nvcc", "--version"], text=True) + paths = subprocess.check_output( + ["find", "/usr/local/cuda/", "-name", "cicc", "-o", "-name", "nvcc.profile"], text=True + ) + patched = install_patch() + version = subprocess.check_output(["/usr/local/cuda/bin/nvcc", "--version"], text=True) + help_text = subprocess.check_output(["/usr/local/cuda/bin/nvcc", "--help"], text=True) + lines = help_text.splitlines() + indices = { + j + for i, line in enumerate(lines) + if "pch" in line.lower() + for j in range(max(0, i - 2), min(len(lines), i + 14)) + } + profiles = {str(p): p.read_text() for p in Path("/usr/local/cuda/bin").glob("*.profile")} + smoke = [] + with tempfile.TemporaryDirectory() as directory: + root = Path(directory) + source = root / "smoke.cu" + source.write_text("__global__ void kernelbot_nvcc_smoke() {}\n") + for arch in ("sm_75", "sm_100"): + output = root / f"{arch}.o" + command = [ + "/usr/local/cuda/bin/nvcc", + f"-arch={arch}", + "-c", + str(source), + "-o", + str(output), + ] + dry = subprocess.run([*command, "--dryrun"], capture_output=True, text=True, timeout=60) + compile_result = subprocess.run( + [*command, "-v"], capture_output=True, text=True, timeout=60 + ) + dispatched = "nvvm-next/bin" in dry.stderr and "$(_NVVM_BRANCH_)" not in dry.stderr + smoke.append( + { + "arch": arch, + "command": command, + "dryrun_returncode": dry.returncode, + "dryrun": dry.stdout + dry.stderr, + "next_backend_selected": dispatched, + "returncode": compile_result.returncode, + "log": compile_result.stdout + compile_result.stderr, + "object_bytes": output.stat().st_size if output.exists() else 0, + } + ) + return { + "stock_version": stock_version, + "patched_version": version, + "paths": paths, + "patch": patched, + "nvcc_profiles": profiles, + "pch_help": "\n".join(lines[i] for i in sorted(indices)), + "strace": shutil.which("strace"), + "smoke": smoke, + "compatible": all( + r["returncode"] == 0 and r["next_backend_selected"] and r["object_bytes"] for r in smoke + ), + } + + +@app.function( + image=image, + cpu=4, + memory=16384, + timeout=1800, + single_use_containers=True, + max_containers=1, + volumes={PCH_MOUNT: pch_volume, NVCC_MOUNT: nvcc_volume}, +) +def warm(cases: list[str]): + import hashlib + import subprocess + import tempfile + import time + + patched = install_patch() + pch_volume.reload() + nvcc_volume.reload() + print("Warming host CUDA-linked default profile", flush=True) + host_started = time.perf_counter() + host = subprocess.run( + ["python3", "/opt/kernelbot-pch/warm.py", "cuda-default"], + capture_output=True, + text=True, + timeout=300, + ) + if host.returncode: + raise RuntimeError(host.stdout + host.stderr) + host_seconds = time.perf_counter() - host_started + pch_volume.commit() + print(f"Host profile ready in {host_seconds:.2f}s", flush=True) + rows = [] + for case in cases: + identity = pch_identity(case, patched) + key = hashlib.sha256(json.dumps(identity, sort_keys=True).encode()).hexdigest() + entry = Path(NVCC_MOUNT) / key + entry.mkdir(parents=True, exist_ok=True) + (entry / "gpumode.cuh").write_text(identity["header"]) + (entry / "pch").mkdir(exist_ok=True) + with tempfile.TemporaryDirectory() as directory: + root = Path(directory) + print(f"Capturing exact NVCC flags for {case}", flush=True) + env, _ = prepare_build(root, case, 0, "kb_nvcc_pch_warmer", True, capture=True) + captured = subprocess.run( + ["python3", "-c", LOADER], env=env, capture_output=True, text=True, timeout=180 + ) + capture_path = root / "commands.capture" + if not capture_path.exists(): + raise RuntimeError( + "Failed to capture NVCC command: " + captured.stdout + captured.stderr + ) + args = json.loads(capture_path.read_text()) + flags = [] + i = 0 + while i < len(args): + if args[i] in {"-c", "-o", "--dependency-output", "-MF", "-MT", "-MQ"}: + i += 2 + continue + if args[i] not in {"--generate-dependencies-with-compile", "-MMD", "-MD"}: + flags.append(args[i]) + i += 1 + command = [ + "/usr/local/cuda/bin/nvcc", + *flags, + "--pch-dir", + str(entry / "pch"), + "--pch", + "--pch-verbose", + "-x", + "cu", + "-c", + str(entry / "gpumode.cuh"), + "-o", + str(root / "warm.o"), + ] + print(f"Creating {case} PCH in {entry}", flush=True) + tick = time.perf_counter() + result = subprocess.run( + command, cwd=root, env=env, capture_output=True, text=True, timeout=300 + ) + seconds = time.perf_counter() - tick + row = { + "case": case, + "key": key, + "command": command, + "seconds": seconds, + "stdout": result.stdout, + "stderr": result.stderr, + "returncode": result.returncode, + "files": [ + {"path": str(p.relative_to(entry)), "bytes": p.stat().st_size} + for p in sorted((entry / "pch").rglob("*")) + if p.is_file() + ], + } + rows.append(row) + print( + f"{case} warmup returned {result.returncode} after {seconds:.2f}s; " + f"{len(row['files'])} files", + flush=True, + ) + if result.returncode or not row["files"]: + raise RuntimeError(json.dumps(row)) + (entry / "metadata.json").write_text( + json.dumps({"identity": identity, "warmup": row}, indent=2) + ) + (Path(NVCC_MOUNT) / f"{case}.json").write_text(json.dumps({"key": key})) + nvcc_volume.commit() + return {"host_seconds": host_seconds, "host_log": host.stdout + host.stderr, "device": rows} + + +@app.function( + image=image, + gpu="T4", + cpu=4, + memory=16384, + timeout=900, + max_containers=4, + single_use_containers=True, + restrict_modal_access=True, + volumes={ + PCH_MOUNT: pch_volume.with_mount_options(read_only=True), + NVCC_MOUNT: nvcc_volume.with_mount_options(read_only=True), + }, +) +def trial(case: str, variant: int, mode: str, repeat: int): + import hashlib + import re + import subprocess + import tempfile + import time + + import torch + + started = time.perf_counter() + assert mode in MODES and case in CASES + patched = install_patch() if mode.startswith("patched") else None + nvcc_version = subprocess.check_output(["/usr/local/cuda/bin/nvcc", "--version"], text=True) + entry = None + if mode.startswith("patched_all_pch"): + key = json.loads((Path(NVCC_MOUNT) / f"{case}.json").read_text())["key"] + entry = Path(NVCC_MOUNT) / key + expected = pch_identity(case, patched) + metadata = json.loads((entry / "metadata.json").read_text()) + assert metadata["identity"] == expected, "Warm PCH identity does not match; rerun warmup" + assert (entry / "gpumode.cuh").read_text() == expected["header"] + assert key == hashlib.sha256(json.dumps(expected, sort_keys=True).encode()).hexdigest() + with tempfile.TemporaryDirectory() as directory: + root = Path(directory) + name = f"kb_nvcc_{case}_{variant}" + env, source = prepare_build(root, case, variant, name, mode != "stock_no_pch", entry) + env["PCH_PREFETCH"] = "1" if mode == "patched_all_pch_prefetch" else "0" + if entry: + env["PCH_EXPECTED_BYTES"] = str( + sum(f["bytes"] for f in metadata["warmup"]["files"] if f["path"].endswith(".pch")) + ) + run_started = time.perf_counter() + result = subprocess.run( + ["python3", "-c", LOADER], env=env, capture_output=True, text=True, timeout=300 + ) + process_seconds = time.perf_counter() - run_started + log = result.stdout + result.stderr + row = { + "case": case, + "variant": variant, + "mode": mode, + "repeat": repeat, + "extension_name": name, + "returncode": result.returncode, + "log": log, + "source_sha256": hashlib.sha256(source.encode()).hexdigest(), + "container_id": os.environ["MODAL_TASK_ID"], + "image_id": os.environ["MODAL_IMAGE_ID"], + "gpu": torch.cuda.get_device_name(), + "torch": str(torch.__version__), + "nvcc_version": nvcc_version, + "patch": patched, + "pch_entry": str(entry) if entry else None, + "process_seconds": process_seconds, + } + if entry: + timed_used = re.findall(r'using precompiled header file "([^"]+)"', log) + timed_created = re.findall(r'creating precompiled header file "([^"]+)"', log) + row["timed_device_pch_files_used"] = timed_used + row["timed_device_pch_consumed"] = ( + result.returncode == 0 + and not timed_created + and {Path(p).name.split("_", 1)[0] for p in timed_used} == {"cicc", "cudafe"} + and all(Path(p).is_relative_to(entry / "pch") for p in timed_used) + ) + for kind in ("cxx", "nvcc"): + path = root / f"commands.{kind}" + row[f"{kind}_commands"] = ( + [json.loads(s) for s in path.read_text().splitlines()] if path.exists() else [] + ) + if result.returncode: + return row + measurement = json.loads((root / "result.json").read_text()) + row.update( + { + "seconds": measurement["seconds"], + "outputs": measurement["outputs"], + "load_inline_seconds": measurement["load_inline_seconds"], + "prefetch_seconds": measurement["prefetch_seconds"], + "prefetch_bytes": measurement["prefetch_bytes"], + "max_absolute_error": max(measurement["errors"], default=0), + "correctness_checks": 10 if case == "cutlass" else 6, + } + ) + # GCC -H prints ! for a valid consumed PCH. Keep this compilation untimed. + cxx_args = list(row["cxx_commands"][0]["args"]) + cxx_args[cxx_args.index("-o") + 1] = str(root / "diagnostic.o") + diagnostic = subprocess.run( + [env["PCH_TEST_COMPILER"], *cxx_args, "-H"], + env=env, + cwd=root, + capture_output=True, + text=True, + timeout=180, + ) + row["host_pch_consumed"] = "! /kernelbot-pch/" in diagnostic.stderr + assert diagnostic.returncode == 0, diagnostic.stderr[-6000:] + assert row["host_pch_consumed"] == (mode != "stock_no_pch"), diagnostic.stderr[-6000:] + row["host_hit_keys"] = re.findall(r"\[kernelbot-pch\] hit (\w+)", log) + # Preserve verbose NVCC diagnostics independently of the measured build. + if mode.startswith("patched"): + args = list(row["nvcc_commands"][0]["args"]) + args[args.index("-o") + 1] = str(root / "diagnostic.cuda.o") + phase_file = root / "nvcc-diagnostic.csv" + extra = ["-v", "--time", str(phase_file), *(["--pch-verbose"] if entry else [])] + diagnostic = subprocess.run( + ["/usr/local/cuda/bin/nvcc", *args, *extra], + env=env, + cwd=root, + capture_output=True, + text=True, + timeout=180, + ) + row["nvcc_diagnostic"] = { + "returncode": diagnostic.returncode, + "stdout": diagnostic.stdout, + "stderr": diagnostic.stderr, + "phase_csv": phase_file.read_text() if phase_file.exists() else None, + } + used = re.findall(r'using precompiled header file "([^"]+)"', diagnostic.stderr) + created = re.findall(r'creating precompiled header file "([^"]+)"', diagnostic.stderr) + row["device_pch_files_used"] = used + row["device_pch_files_created"] = created + frontends = {Path(path).name.split("_", 1)[0] for path in used} + row["device_pch_consumed"] = bool(entry) and ( + frontends == {"cicc", "cudafe"} + and not created + and diagnostic.returncode == 0 + and all(Path(path).is_relative_to(entry / "pch") for path in used) + ) + row["remote_seconds"] = time.perf_counter() - started + return row + + +@app.local_entrypoint() +def main( + output: str = "/tmp/kernelbot-nvcc-pch.json", + probe_only: bool = False, + warm_only: bool = False, + skip_warm: bool = False, + cases: str = "simple,cutlass", + modes: str = ",".join(DEFAULT_MODES), + repeats: int = 2, +): + selected = cases.split(",") + selected_modes = modes.split(",") + assert set(selected) <= set(CASES) and set(selected_modes) <= set(MODES) + assert repeats >= 1 + result = {"probe": probe.remote(), "rows": []} + + def save(): + Path(output).write_text(json.dumps(result, indent=2) + "\n") + + save() + if probe_only: + print(json.dumps(result["probe"], indent=2)) + return + assert result["probe"]["compatible"], ( + f"Patched compiler compatibility checks failed; inspect {output}" + ) + if not skip_warm: + result["warmup"] = warm.remote(selected) + save() + print(json.dumps(result["warmup"], indent=2)) + if warm_only: + return + jobs = [ + (case, variant, mode, repeat) + for repeat in range(repeats) + for case in selected + for variant in (0, 1) + for mode in (selected_modes if (variant + repeat) % 2 == 0 else selected_modes[::-1]) + ] + for row in trial.starmap(jobs): + result["rows"].append(row) + save() + print( + json.dumps( + { + k: row.get(k) + for k in ("case", "variant", "mode", "repeat", "returncode", "seconds") + } + ) + ) + assert len({r["container_id"] for r in result["rows"]}) == len(result["rows"]) + for case in selected: + for mode in selected_modes: + group = [r for r in result["rows"] if r["case"] == case and r["mode"] == mode] + good = [r["seconds"] for r in group if r["returncode"] == 0] + print( + f"{case}/{mode}: median={statistics.median(good) if good else None}; " + f"passed={len(good)}/{len(group)}" + ) + assert all(row["returncode"] == 0 for row in result["rows"]), ( + f"Failed builds retained in {output}" + ) + assert all( + row.get("device_pch_consumed") + for row in result["rows"] + if row["mode"].startswith("patched_all_pch") + ), f"NVCC PCH consumption was not verified; inspect diagnostics in {output}" + assert all( + row.get("timed_device_pch_consumed") + for row in result["rows"] + if row["mode"].startswith("patched_all_pch") + ), f"NVCC PCH consumption was not verified in the timed build; inspect logs in {output}" diff --git a/scripts/upload_nvcc_toolchain.py b/scripts/upload_nvcc_toolchain.py new file mode 100644 index 00000000..57863608 --- /dev/null +++ b/scripts/upload_nvcc_toolchain.py @@ -0,0 +1,52 @@ +"""Upload the checksum-pinned NVIDIA drop to KernelBot's private Modal volume. + +PYTHONPATH=src/runners python scripts/upload_nvcc_toolchain.py --source /path/to/binaries +This does not deploy a runner or publish the binaries outside the Modal workspace. +""" + +import argparse +import hashlib +import json +import os +import tempfile +from pathlib import Path + +import modal +from nvcc_toolchain import BINARY_SHA256, RELEASE + + +def upload(source: Path, volume_name: str): + for name, expected in BINARY_SHA256.items(): + if hashlib.sha256((source / name).read_bytes()).hexdigest() != expected: + raise ValueError(f"Compiler checksum mismatch: {name}") + manifest = json.dumps({"release": RELEASE, "sha256": BINARY_SHA256}, sort_keys=True) + "\n" + volume = modal.Volume.from_name(volume_name, create_if_missing=True) + manifest_path = f"/{RELEASE}/manifest.json" + try: + existing = b"".join(volume.read_file(manifest_path)).decode() + except FileNotFoundError: + existing = None + if existing is not None: + if existing != manifest: + raise ValueError("Existing compiler release has a different manifest") + print(f"{RELEASE} is already present in {volume_name}") + return + with tempfile.TemporaryDirectory() as directory: + metadata = Path(directory) / "manifest.json" + metadata.write_text(manifest) + with volume.batch_upload(force=False) as batch: + for name in BINARY_SHA256: + batch.put_file(source / name, f"/{RELEASE}/{name}") + batch.put_file(metadata, manifest_path) + print(f"Uploaded verified {RELEASE} to private volume {volume_name}") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument("--source", type=Path, required=True) + parser.add_argument( + "--volume", + default=os.environ.get("KERNELBOT_NVCC_TOOLCHAIN_VOLUME", "kernelbot-nvcc-toolchains"), + ) + args = parser.parse_args() + upload(args.source, args.volume) diff --git a/scripts/validate_nvcc_pch.py b/scripts/validate_nvcc_pch.py new file mode 100644 index 00000000..2d8bbbcd --- /dev/null +++ b/scripts/validate_nvcc_pch.py @@ -0,0 +1,612 @@ +"""Validate the real CPU-to-GPU submission path without publishing leaderboard scores. + +KERNELBOT_NVCC_TOOLCHAIN=1 KERNELBOT_NVCC_PCH=1 \ + PYTHONPATH=src:src/runners uv run modal run scripts/validate_nvcc_pch.py + +Publish the verified private toolchain Volume with upload_nvcc_toolchain.py first. +The default matrix runs all edge cases on T4 and PCH-on/off on H100 and B200. +The two baseline cases execute the normal test, benchmark, and leaderboard +validation phases in an ephemeral app; no service or database is contacted. +Use --profiles with trusted nvcc_common.PROFILES names to select baseline +flag sets, and --repeats to repeat baseline pairs in fresh containers. +The optional CUTLASS pair uses the generic sm75-O3 framework profile: + --gpus T4 --cases cutlass,cutlass-disabled --repeats 2 --concurrency 1 +""" + +import ast +import asyncio +import dataclasses +import hashlib +import json +import os +import shutil +import statistics +import subprocess +import time +from contextlib import contextmanager +from pathlib import Path + +os.environ["KERNELBOT_MODAL_APP"] = os.environ.get( + "KERNELBOT_NVCC_VALIDATION_APP", "kernelbot-nvcc-pch-release-validation" +) +if os.environ["KERNELBOT_MODAL_APP"] == "discord-bot-runner": + raise RuntimeError("The validation app must be separate from the production runner") +os.environ.setdefault("KERNELBOT_NVCC_TOOLCHAIN", "1") +os.environ.setdefault("KERNELBOT_NVCC_PCH", "1") +os.environ.setdefault("KERNELBOT_PCH_VOLUME", "kernelbot-nvcc-release-host-pch") +os.environ.setdefault("KERNELBOT_NVCC_PCH_VOLUME", "kernelbot-nvcc-release-device-pch") + +from modal_runner import ( # noqa: E402 + NVCC_PCH_MOUNT, + app, + cuda_image, + submission_volumes, + warm_nvcc_pch, + warm_pch, +) + +from libkernelbot.consts import GPU_TO_SM, SubmissionMode, Timeout # noqa: E402 +from libkernelbot.python_precompile import ( # noqa: E402 + compile_python, + run_with_cpu_artifacts, +) +from libkernelbot.run_eval import run_config # noqa: E402 +from libkernelbot.task import build_task_config, make_task_definition # noqa: E402 + +# Serialized GPU functions refer to helpers from this module by name. Mount the +# same harness source on both stages so fresh containers can import those helpers. +cuda_image = cuda_image.add_local_file(Path(__file__).resolve(), "/root/validate_nvcc_pch.py") + +GPUS = ("T4", "L4", "A100", "H100", "B200") +DEFAULT_GPUS = ("T4", "H100", "B200") +GPU_RESOURCES = {"A100": "A100-80GB", "H100": "H100!"} +BASELINE_CASES = ("default", "disabled") +CUTLASS_CASES = ("cutlass", "cutlass-disabled") +CASES = ( + *BASELINE_CASES, + "no-implicit", + "unknown-flags", + "changed-header", + "rejected-cache", + "empty-cache", + "custom-nvcc", +) +ALL_CASES = (*CASES, *CUTLASS_CASES) +FINGERPRINT = Path("/opt/kernelbot-pch/nvcc-fingerprint") +GPU_ARCHES = { + "T4": {"75"}, + "L4": {"89"}, + "A100": {"80"}, + "H100": {"90", "90a"}, + "B200": {"100", "100a"}, +} + + +def profile_catalog() -> dict: + # Only the local orchestrator needs the profile catalog. The remote image + # keeps these stdlib modules in /opt/kernelbot-pch instead of a Python package. + from pch.nvcc_common import PROFILES + + return PROFILES + + +def make_submission(example: Path, case: str, extra_cuda_cflags=()) -> str: + source = (example / "submission_cuda_inline.py").read_text() + source = source.replace("name='add_cuda'", "name='nvcc_release_vectoradd'") + if extra_cuda_cflags: + source = source.replace( + " verbose=True,", + f" verbose=True,\n extra_cuda_cflags={list(extra_cuda_cflags)!r},", + ) + if case == "no-implicit": + source = source.replace( + 'add_cuda_source = """', + 'add_cuda_source = """\n#include \n#include ', + ).replace(" verbose=True,", " verbose=True,\n no_implicit_headers=True,") + elif case == "unknown-flags": + source = source.replace( + " verbose=True,", + " verbose=True,\n extra_cuda_cflags=['-DKERNELBOT_RELEASE_PROFILE_MISS=1'],", + ) + elif case == "changed-header": + # Identical source and options in both containers, but a genuinely changed + # user dependency. Replaying the CPU binary would produce wrong answers. + source = ( + "import torch\nfrom pathlib import Path\n" + "Path('/tmp/kernelbot-nvcc-release-user.h').write_text(\n" + " '#define OFFSET ' + ('0' if torch.cuda.is_available() else '1') + '\\n')\n" + + source.replace( + 'add_cuda_source = """', + 'add_cuda_source = """\n#include "/tmp/kernelbot-nvcc-release-user.h"', + ).replace("C[idx] = A[idx] + B[idx];", "C[idx] = A[idx] + B[idx] + OFFSET;") + ) + elif case == "custom-nvcc": + source = "import os\nos.environ['PYTORCH_NVCC'] = '/usr/local/cuda/bin/nvcc'\n" + source + return source + + +def cutlass_submission() -> str: + # Reuse the benchmark's actual SM75 tensor-core GEMM without importing its + # Modal app or its workload-specific PCH. Only these two string literals run. + benchmark = Path(__file__).with_name("modal_pch_benchmark.py") + constants = {} + for node in ast.parse(benchmark.read_text()).body: + if isinstance(node, ast.Assign): + for target in node.targets: + if isinstance(target, ast.Name) and target.id in {"GEMM_CPP", "GEMM_CUDA"}: + constants[target.id] = ( + ast.literal_eval(node.value) + .replace('#include "variant.h"', "constexpr int kVariant = 1;") + .replace("TILE_N", "128") + ) + return ( + "# GEMM fixture adapted from scripts/modal_pch_benchmark.py.\n" + "import torch\nfrom torch.utils.cpp_extension import load_inline\n" + "module = load_inline(name='nvcc_release_cutlass',\n" + f" cpp_sources={constants['GEMM_CPP']!r},\n" + f" cuda_sources={constants['GEMM_CUDA']!r},\n" + " functions=['add_cuda', 'version'], extra_cuda_cflags=['-O3'],\n" + " no_implicit_headers=False, verbose=True)\n" + "assert module.version() == 1\n" + "def custom_kernel(data):\n" + " a, b, nondefault = data\n" + " if not nondefault:\n" + " return module.add_cuda(a, b)\n" + " current = torch.cuda.current_stream(a.device)\n" + " stream = torch.cuda.Stream(device=a.device)\n" + " stream.wait_stream(current)\n" + " with torch.cuda.stream(stream):\n" + " result = module.add_cuda(a, b)\n" + " a.record_stream(stream)\n" + " b.record_stream(stream)\n" + " current.wait_stream(stream)\n" + " result.record_stream(current)\n" + " return result\n" + ) + + +CUTLASS_REFERENCE = """ +import torch + +def generate_input(m, n, k, seed, zero=0, ones=0, nondefault=1): + generator = torch.Generator(device='cuda').manual_seed(seed) + a = torch.randn(m, k, device='cuda', dtype=torch.float16, generator=generator) / 4 + b = torch.randn(n, k, device='cuda', dtype=torch.float16, generator=generator) / 4 + if zero: + a.zero_() + if ones: + a.fill_(1) + b.fill_(1) + return a, b, nondefault + +def check_implementation(data, output): + a, b, _ = data + expected = a.double() @ b.double().T + try: + torch.testing.assert_close(output.double(), expected, rtol=2e-4, atol=2e-4) + except AssertionError as error: + return False, str(error) + error = float((output.double() - expected).abs().max()) + return True, f'max_absolute_error={error}' +""" + + +def configure_cutlass(config): + config["sources"]["reference.py"] = CUTLASS_REFERENCE + config["sources"]["task.py"] = "TestSpec = dict\n" + config["tests"] = [ + {"m": m, "n": n, "k": k, "seed": 17, "zero": zero, "nondefault": 1} + for m, n, k in ((128, 128, 64), (129, 136, 64), (256, 256, 128), (512, 512, 256)) + for zero in (0, 1) + ] + [{"m": 8, "n": 8, "k": 8, "seed": 17, "ones": 1, "nondefault": 1}] + # Time the kernel on the usual stream; correctness tests above exercise a + # separate stream, random/zero inputs, and tile tails against float64 Torch. + config["benchmarks"] = [{"m": 256, "n": 256, "k": 128, "seed": 54352, "nondefault": 0}] + + +def make_config(gpu: str, case: str, profile: str | None = None) -> dict: + example = Path(__file__).resolve().parents[1] / "examples/vectoradd_py" + if case in CUTLASS_CASES: + profile = profile or "sm75-O3" + if gpu != "T4" or profile != "sm75-O3": + raise ValueError("The CUTLASS fixture requires T4 and the generic sm75-O3 profile") + arch, flags = profile_catalog()[profile or f"sm{GPU_TO_SM[gpu]}-default"] + if arch not in GPU_ARCHES[gpu]: + raise ValueError(f"Profile {profile} targets SM{arch}, which is incompatible with {gpu}") + task = make_task_definition(example).task + mode = ( + SubmissionMode.LEADERBOARD + if case in (*BASELINE_CASES, *CUTLASS_CASES) + else SubmissionMode.TEST + ) + config = build_task_config( + task=task, + submission_content=( + cutlass_submission() if case in CUTLASS_CASES else make_submission(example, case, flags) + ), + arch=arch, + mode=mode, + ) + config["benchmarks"] = [{"size": 512, "seed": 54352}] + if case in CUTLASS_CASES: + configure_cutlass(config) + return config + + +def resource_usage() -> dict: + import resource + + result = { + "self_peak_rss_kib": resource.getrusage(resource.RUSAGE_SELF).ru_maxrss, + "child_peak_rss_kib": resource.getrusage(resource.RUSAGE_CHILDREN).ru_maxrss, + } + for name in ("memory.peak", "memory.max"): + path = Path("/sys/fs/cgroup") / name + if path.exists(): + result[name] = path.read_text().strip() + return result + + +def read_events(path: Path) -> list[dict]: + return [json.loads(line) for line in path.read_text().splitlines()] if path.exists() else [] + + +def prepare_rejected_cache(destination: Path, arch: str) -> dict: + """Damage disposable sparse copies; never modify files on the shared Volume.""" + generation = FINGERPRINT.read_text().strip() + manifests = list((Path(NVCC_PCH_MOUNT) / generation).glob("*/manifest.json")) + for path in manifests: + manifest = json.loads(path.read_text()) + if manifest.get("profile") != f"sm{arch}-default": + continue + entry = destination / generation / path.parent.name + (entry / "pch").mkdir(parents=True) + shutil.copyfile(path, entry / "manifest.json") + shutil.copyfile(path.parent / "gpumode.cuh", entry / "gpumode.cuh") + total = 0 + for item in manifest["files"]: + target = entry / item["path"] + if not target.resolve().is_relative_to((entry / "pch").resolve()): + raise ValueError("Unexpected PCH path in trusted manifest") + target.parent.mkdir(exist_ok=True) + with target.open("wb") as handle: + # Preserve manifest size validation while making the PCH invalid. + # Sparse local files do not copy the multi-GB shared payload. + handle.truncate(item["bytes"]) + total += item["bytes"] + return {"profile": manifest["profile"], "sparse_bytes": total} + raise RuntimeError(f"Warm sm{arch}-default before running the rejection test") + + +@contextmanager +def validation_environment(case: str, arch: str): + previous = dict(os.environ) + event_path = Path("/tmp/kernelbot-nvcc-release-events.jsonl") + event_path.unlink(missing_ok=True) + metadata = {} + try: + os.environ.update( + KERNELBOT_NVCC_PCH="1", + KERNELBOT_NVCC_PCH_DISABLE="1" if case in {"disabled", "cutlass-disabled"} else "0", + KERNELBOT_NVCC_PCH_EVENTS=str(event_path), + ) + os.environ.pop("KERNELBOT_NVCC_PCH_ROOT", None) + if case in {"rejected-cache", "empty-cache"}: + destination = Path("/tmp") / f"kernelbot-nvcc-release-{case}" + destination.mkdir(exist_ok=False) + if case == "rejected-cache": + metadata = prepare_rejected_cache(destination, arch) + os.environ["KERNELBOT_NVCC_PCH_ROOT"] = str(destination) + yield event_path, metadata + finally: + os.environ.clear() + os.environ.update(previous) + + +def toolchain_info() -> dict: + return { + "generation": FINGERPRINT.read_text().strip(), + "nvcc_version": subprocess.check_output( + ["/usr/local/cuda/bin/nvcc", "--version"], text=True + ), + "container_id": os.environ["MODAL_TASK_ID"], + "image_id": os.environ["MODAL_IMAGE_ID"], + } + + +@app.function( + image=cuda_image, + cpu=4, + memory=8192, + timeout=Timeout.COMPILE + 30, + max_containers=3, + single_use_containers=True, + restrict_modal_access=True, + volumes=submission_volumes, +) +def validate_cpu(config: dict, case: str) -> dict: + started = time.perf_counter() + with validation_environment(case, config["arch"]) as (event_path, metadata): + packet = compile_python(config) + return { + "packet": packet, + "events": read_events(event_path), + "toolchain": toolchain_info(), + "memory": resource_usage(), + "case_metadata": metadata, + "wall_seconds": time.perf_counter() - started, + "requested_memory_mib": 8192, + "max_jobs": 2, + } + + +def validate_gpu(config: dict, case: str, packet: dict) -> dict: + started = time.perf_counter() + with validation_environment(case, config["arch"]) as (event_path, metadata): + result = run_with_cpu_artifacts({**config, "cpu_compile": packet}, run_config) + return { + "result": dataclasses.asdict(result), + "events": read_events(event_path), + "toolchain": toolchain_info(), + "memory": resource_usage(), + "case_metadata": metadata, + "wall_seconds": time.perf_counter() - started, + } + + +gpu_functions = { + gpu: app.function( + image=cuda_image, + gpu=GPU_RESOURCES.get(gpu, gpu), + cpu=4, + memory=16384, + timeout=600, + max_containers=1, + single_use_containers=True, + restrict_modal_access=True, + volumes=submission_volumes, + name=f"validate_nvcc_{gpu.lower()}", + serialized=True, + )(validate_gpu) + for gpu in GPUS +} + + +def verify_hits(events, profile): + for event in events: + if event["status"] == "hit": + assert event["frontends"] == ["cicc", "cudafe"], event + if profile: + assert event["profile"] == profile, event + + +def verify_result(case: str, cpu: dict, gpu: dict, profile: str | None = None) -> None: + packet = cpu["packet"] + assert packet["info"]["status"] == "compiled", packet["info"] + assert packet["info"]["artifacts"] > 0 and packet["artifacts"] + assert cpu["toolchain"]["generation"] == gpu["toolchain"]["generation"] + assert cpu["toolchain"]["image_id"] == gpu["toolchain"]["image_id"] + result = gpu["result"] + assert result["success"], result["error"] + assert result["runs"], "No evaluation phases ran" + for name, evaluation in result["runs"].items(): + run = evaluation["run"] + assert run and run["success"] and run["passed"], (name, evaluation) + assert run["result"].get("check") == "pass", (name, run["result"]) + info = result["cpu_compile"] + cpu_statuses = {event["status"] for event in cpu["events"]} + gpu_statuses = {event["status"] for event in gpu["events"]} + if case == "changed-header": + assert info["fallbacks"] > 0 and info["reused"] == 0, info + assert "dependency changed" in info["reason"].lower(), info + assert "hit" in gpu_statuses, gpu["events"] + else: + assert info["status"] == "reused" and info["reused"] >= 2, info + assert info["fallbacks"] == 0, info + assert not gpu["events"], gpu["events"] + expected_status = { + "default": "hit", + "disabled": "disabled", + "no-implicit": "bypass", + "unknown-flags": "miss", + "changed-header": "hit", + "rejected-cache": "fallback", + "empty-cache": "miss", + "cutlass": "hit", + "cutlass-disabled": "disabled", + } + if case == "custom-nvcc": + assert not cpu["events"], cpu["events"] + else: + assert cpu_statuses == {expected_status[case]}, cpu["events"] + verify_hits(cpu["events"] + gpu["events"], profile) + if case not in {"default", "changed-header", "cutlass"}: + assert "hit" not in cpu_statuses, cpu["events"] + if case == "rejected-cache": + assert any( + event["status"] == "fallback" and event["reason"] == "cache_rejected" + for event in cpu["events"] + ), cpu["events"] + if case in (*BASELINE_CASES, *CUTLASS_CASES): + assert set(result["runs"]) == {"test", "benchmark", "leaderboard"} + measurements = result["runs"]["benchmark"]["run"]["result"] + assert float(measurements["benchmark.0.mean"]) > 0, measurements + + +def compact_cpu(cpu: dict) -> dict: + packet = cpu["packet"] + binary = packet["artifacts"] + return cpu | { + "packet": {key: value for key, value in packet.items() if key != "artifacts"} + | {"artifact_bytes": len(binary), "artifact_sha256": hashlib.sha256(binary).hexdigest()} + } + + +def paired_jobs(gpu, profile, pair, cases, repeats): + return [ + (gpu, case, profile, repeat) + for repeat in range(1, repeats + 1) + for case in pair[:: 1 if repeat % 2 else -1] + if case in cases + ] + + +def matrix_jobs(gpus, cases, edge_gpu, profiles, repeats): + catalog = profile_catalog() + selected = profiles or [f"sm{GPU_TO_SM[gpu]}-default" for gpu in gpus] + if len(selected) > 12 or not set(selected) <= catalog.keys(): + raise ValueError("Choose at most 12 fixed profiles from " + ", ".join(catalog)) + jobs = [] + for profile in selected: + arch, _ = catalog[profile] + matching = [gpu for gpu in gpus if arch in GPU_ARCHES[gpu]] + if len(matching) != 1: + raise ValueError(f"Select a compatible GPU for {profile} (SM{arch})") + jobs.extend(paired_jobs(matching[0], profile, BASELINE_CASES, cases, repeats)) + if set(cases) & set(CUTLASS_CASES): + if "T4" not in gpus: + raise ValueError("Select T4 for the CUTLASS fixture") + jobs.extend(paired_jobs("T4", "sm75-O3", CUTLASS_CASES, cases, repeats)) + if edge_gpu in gpus: + jobs.extend( + (edge_gpu, case, f"sm{GPU_TO_SM[edge_gpu]}-default", 1) + for case in cases + if case not in (*BASELINE_CASES, *CUTLASS_CASES) + ) + if not jobs: + raise ValueError("No jobs selected; edge cases run only on edge_gpu") + return jobs + + +def comparisons(rows): + """Keep latency observations separate from the functional pass/fail gates.""" + grouped = {} + failed = set() + for row in rows: + if row["case"] in (*BASELINE_CASES, *CUTLASS_CASES): + workload = "cutlass" if row["case"] in CUTLASS_CASES else "vectoradd" + arm = "disabled" if row["case"].endswith("disabled") else "default" + key = (row["gpu"], row["profile"], workload) + if not row["passed"]: + failed.add(key) + else: + grouped.setdefault(key, {}).setdefault(arm, []).append( + row["cpu"]["packet"]["info"]["duration"] + ) + results = [] + for (gpu, profile, workload), arms in sorted(grouped.items()): + if ( + (gpu, profile, workload) in failed + or set(arms) != set(BASELINE_CASES) + or len(arms["default"]) != len(arms["disabled"]) + ): + continue + on, off = (statistics.median(arms[case]) for case in BASELINE_CASES) + results.append( + { + "gpu": gpu, + "profile": profile, + "workload": workload, + "cpu_compile_seconds": arms, + "median_pch_on_seconds": on, + "median_pch_off_seconds": off, + "speedup": off / on, + "reduction_percent": 100 * (1 - on / off), + "scope": "CPU compile_python including import, dependency validation and packing", + } + ) + return results + + +@app.local_entrypoint() +async def main( + gpus: str = ",".join(DEFAULT_GPUS), + cases: str = ",".join(CASES), + edge_gpu: str = "T4", + profiles: str = "", + repeats: int = 1, + skip_warm: bool = False, + concurrency: int = 2, + output: str = "/tmp/kernelbot-nvcc-pch-release-validation.json", +): + selected_gpus, selected_cases = gpus.split(","), cases.split(",") + if not set(selected_gpus) <= set(GPUS) or not set(selected_cases) <= set(ALL_CASES): + raise ValueError(f"Choose GPUs from {GPUS} and cases from {ALL_CASES}") + if edge_gpu not in GPUS or not 1 <= concurrency <= 3 or not 1 <= repeats <= 5: + raise ValueError("Choose a supported edge_gpu, concurrency 1–3, and repeats 1–5") + selected_profiles = list(dict.fromkeys(profiles.split(","))) if profiles else [] + jobs = matrix_jobs(selected_gpus, selected_cases, edge_gpu, selected_profiles, repeats) + result = { + "app": os.environ["KERNELBOT_MODAL_APP"], + "baseline_repeats": repeats, + "concurrency": concurrency, + "rows": [], + } + + def save(): + result["comparisons"] = comparisons(result["rows"]) + Path(output).write_text(json.dumps(result, default=str, indent=2) + "\n") + + if not skip_warm: + result["host_warmup"] = await warm_pch.remote.aio(profiles="cuda-default") + warm_profiles = ",".join(dict.fromkeys(job[2] for job in jobs)) + result["nvcc_warmup"] = await warm_nvcc_pch.remote.aio(profiles=warm_profiles) + save() + gate = asyncio.Semaphore(concurrency) + + async def run_one(gpu_name, case, profile, repeat): + async with gate: + arch, flags = profile_catalog()[profile] + row = { + "gpu": gpu_name, + "case": case, + "workload": "cutlass" if case in CUTLASS_CASES else "vectoradd", + "profile": profile, + "arch": arch, + "extra_cuda_cflags": flags, + "repeat": repeat, + "passed": False, + "state": "running", + } + started = time.perf_counter() + result["rows"].append(row) + save() + try: + config = make_config(gpu_name, case, profile) + row["sources_sha256"] = hashlib.sha256( + json.dumps(config["sources"], sort_keys=True).encode() + ).hexdigest() + cpu = await validate_cpu.remote.aio(config=config, case=case) + row["cpu"] = compact_cpu(cpu) + save() + gpu = await gpu_functions[gpu_name].remote.aio( + config=config, case=case, packet=cpu["packet"] + ) + row["gpu_run"] = gpu + verify_result(case, cpu, gpu, profile) + row["passed"] = True + except Exception as exc: + row["error"] = f"{type(exc).__name__}: {exc}" + row["wall_seconds"] = time.perf_counter() - started + row["state"] = "passed" if row["passed"] else "failed" + save() + print( + json.dumps( + { + key: row.get(key) + for key in ("gpu", "case", "profile", "repeat", "passed", "error") + } + ), + flush=True, + ) + + await asyncio.gather(*(run_one(*job) for job in jobs)) + identifiers = [ + stage["toolchain"]["container_id"] + for row in result["rows"] + for stage in (row.get("cpu"), row.get("gpu_run")) + if stage + ] + assert len(identifiers) == len(set(identifiers)), "Expected fresh CPU and GPU containers" + assert all(row["passed"] for row in result["rows"]), f"Validation failures saved in {output}" + print(f"Passed {len(result['rows'])} CPU-to-GPU cases; results: {output}", flush=True) diff --git a/src/libkernelbot/inline_artifacts.py b/src/libkernelbot/inline_artifacts.py index e7f2ece5..69acdbfb 100644 --- a/src/libkernelbot/inline_artifacts.py +++ b/src/libkernelbot/inline_artifacts.py @@ -6,15 +6,96 @@ import json import os import platform +import re import subprocess import sysconfig from pathlib import Path +NVCC_FINGERPRINT = Path("/opt/kernelbot-pch/nvcc-fingerprint") +NVCC_PCH_ROOT = Path("/kernelbot-nvcc-pch") +MAX_CACHE_METADATA_BYTES = 64 * 1024 + def _digest(value: bytes) -> str: return hashlib.sha256(value).hexdigest() +def nvcc_fingerprint() -> str | None: + """Bind artifacts to the immutable compiler/wrapper manifest, when installed.""" + try: + contents = NVCC_FINGERPRINT.read_bytes() + except FileNotFoundError: + return None + if not contents.strip(): + raise RuntimeError("NVCC toolchain fingerprint is empty") + return _digest(contents) + + +def _small_cache_file(path: Path) -> bytes: + if path.is_symlink() or path.stat().st_size > MAX_CACHE_METADATA_BYTES: + raise RuntimeError("Invalid NVCC PCH cache metadata") + return path.read_bytes() + + +def _pch_dependency_digest(path: Path) -> str: + """Use a trusted, immutable cache manifest instead of rehashing GB-sized PCHs.""" + path = path.resolve() + root = NVCC_PCH_ROOT.resolve() + if not path.is_relative_to(root): + raise RuntimeError("PCH dependency is outside the trusted NVCC cache") + parts = path.relative_to(root).parts + if ( + len(parts) != 4 + or not all(re.fullmatch(r"[0-9a-f]{64}", part) for part in parts[:2]) + or parts[2] != "pch" + ): + raise RuntimeError("Invalid NVCC PCH dependency path") + generation, key, _, _ = parts + if generation != NVCC_FINGERPRINT.read_text().strip(): + raise RuntimeError("NVCC PCH dependency belongs to a different toolchain") + entry = root / generation / key + manifest_bytes = _small_cache_file(entry / "manifest.json") + manifest = json.loads(manifest_bytes) + if ( + manifest.get("version") != 1 + or manifest.get("generation") != generation + or manifest.get("key") != key + or manifest.get("header_sha256") != _digest(_small_cache_file(entry / "gpumode.cuh")) + ): + raise RuntimeError("NVCC PCH dependency manifest does not match its cache entry") + files = manifest.get("files", []) + if len(files) != 2: + raise RuntimeError("NVCC PCH dependency manifest is incomplete") + frontends, names = set(), set() + for item in files: + name, size = item.get("path", ""), item.get("bytes") + match = re.fullmatch(r"pch/(cicc|cudafe)_[A-Za-z0-9_]+\.pch", name) + if not match or type(size) is not int or size <= 0: + raise RuntimeError("Invalid NVCC PCH dependency file metadata") + cache_file = entry / name + if ( + cache_file.resolve() != cache_file + or not cache_file.is_file() + or cache_file.stat().st_size != size + ): + raise RuntimeError("NVCC PCH dependency file changed") + names.add(name) + frontends.add(match[1]) + if frontends != {"cicc", "cudafe"} or path.relative_to(entry).as_posix() not in names: + raise RuntimeError("NVCC PCH dependency is not covered by the cache manifest") + return f"nvcc-pch-v1:{_digest(manifest_bytes)}:{path.stat().st_size}" + + +def _dependency_digest(path: Path) -> str: + if path.suffix == ".pch": + return _pch_dependency_digest(path) + digest = hashlib.sha256() + with path.open("rb") as source: + for block in iter(lambda: source.read(1024 * 1024), b""): + digest.update(block) + return digest.hexdigest() + + def _save_module(directory: Path, module, work: Path) -> None: directory.mkdir(exist_ok=True) binary = Path(module.__file__).read_bytes() @@ -64,7 +145,7 @@ def _dependencies(build: Path, work: Path) -> dict[str, str]: if path.parent == build.resolve() and path.name in {"main.cpp", "cuda.cu", "sycl.sycl"}: continue # Generated from the source strings already in the request key. name = str(path.relative_to(work)) if path.is_relative_to(work) else str(path) - dependencies[name] = _digest(path.read_bytes()) + dependencies[name] = _dependency_digest(path) return dependencies @@ -73,7 +154,7 @@ def _check_dependencies(dependencies: dict[str, str] | None, work: Path) -> None raise RuntimeError("Artifact has no compiler dependency information") for name, digest in dependencies.items(): path = work / name - if _digest(path.read_bytes()) != digest: + if _dependency_digest(path) != digest: raise RuntimeError(f"Build dependency changed: {path}") @@ -124,18 +205,30 @@ def dispatch(*args, **kwargs): current_runtime = { **runtime, "cuda_home": getattr(cpp_extension, "CUDA_HOME", None), + "nvcc_toolchain": nvcc_fingerprint(), "build_environment": { name: os.environ.get(name) for name in ( "TORCH_CUDA_ARCH_LIST", "CXX", "CC", + "PYTORCH_NVCC", "CUDA_HOME", + "CUDA_PATH", + "CUDA_ROOT", "CPATH", "CPLUS_INCLUDE_PATH", "C_INCLUDE_PATH", + "GCC_EXEC_PREFIX", + "COMPILER_PATH", + "PATH", + "NVCC_CCBIN", + "NVVMIR_LIBRARY_DIR", "NVCC_PREPEND_FLAGS", "NVCC_APPEND_FLAGS", + "KERNELBOT_NVCC_PCH", + "KERNELBOT_NVCC_PCH_DISABLE", + "KERNELBOT_NVCC_PCH_ROOT", "LIBRARY_PATH", "LD_LIBRARY_PATH", ) diff --git a/src/libkernelbot/python_precompile.py b/src/libkernelbot/python_precompile.py index 6d2ea1cd..558c20cf 100644 --- a/src/libkernelbot/python_precompile.py +++ b/src/libkernelbot/python_precompile.py @@ -14,7 +14,7 @@ from pathlib import Path from libkernelbot.consts import Timeout -from libkernelbot.inline_artifacts import prepare_environment +from libkernelbot.inline_artifacts import nvcc_fingerprint, prepare_environment from libkernelbot.run_eval import CPUCompileInfo MAX_ARTIFACT_BYTES = 64 * 1024 * 1024 @@ -67,11 +67,12 @@ def should_precompile(config: dict) -> bool: def build_identity(config: dict) -> dict: return { - "version": 2, + "version": 3, "sources": hashlib.sha256( json.dumps(config["sources"], sort_keys=True).encode() ).hexdigest(), "image": Path("/opt/kernelbot-pch/fingerprint").read_text().strip(), + "nvcc_toolchain": nvcc_fingerprint(), } diff --git a/src/runners/modal_runner.py b/src/runners/modal_runner.py index a457a44c..46af9156 100644 --- a/src/runners/modal_runner.py +++ b/src/runners/modal_runner.py @@ -6,6 +6,7 @@ from pathlib import Path from modal import App, Image, Volume +from nvcc_toolchain import BINARY_SHA256, RELEASE, install_nvcc_toolchain from libkernelbot.consts import Timeout from libkernelbot.python_precompile import compile_python, run_with_cpu_artifacts @@ -15,6 +16,10 @@ # IMPORTANT: This has to stay in separate file or modal breaks app = App(os.environ.get("KERNELBOT_MODAL_APP", "discord-bot-runner")) cuda_version = "13.3.0" +NVCC_TOOLCHAIN_ENABLED = os.environ.get("KERNELBOT_NVCC_TOOLCHAIN") == "1" +NVCC_PCH_ENABLED = os.environ.get("KERNELBOT_NVCC_PCH") == "1" +if NVCC_PCH_ENABLED and not NVCC_TOOLCHAIN_ENABLED: + raise ValueError("NVCC PCH requires KERNELBOT_NVCC_TOOLCHAIN=1") flavor = "devel" operating_sys = "ubuntu24.04" tag = f"{cuda_version}-{flavor}-{operating_sys}" @@ -98,14 +103,16 @@ f"rm /tmp/{mathdx_archive}" ), ) - .env({ - "CUTLASS_PATH": "/opt/cutlass", - "MATHDX_HOME": "/opt/mathdx", - "CPLUS_INCLUDE_PATH": ( - "/opt/mathdx/include:/opt/mathdx/external/cutlass/include:" - "/opt/cutlass/include:/opt/cutlass/tools/util/include" - ), - }) + .env( + { + "CUTLASS_PATH": "/opt/cutlass", + "MATHDX_HOME": "/opt/mathdx", + "CPLUS_INCLUDE_PATH": ( + "/opt/mathdx/include:/opt/mathdx/external/cutlass/include:" + "/opt/cutlass/include:/opt/cutlass/tools/util/include" + ), + } + ) .run_commands( "python -m pip check", 'python -c "import cuda.tile, nvmath"', @@ -143,10 +150,75 @@ "ncu --version", ) +# This switch selects a separately versioned runner image. Keeping it separate +# from PCH enablement allows a same-compiler comparison and independent rollback. +NVCC_PCH_MOUNT = "/kernelbot-nvcc-pch" +nvcc_pch_volume = Volume.from_name( + os.environ.get("KERNELBOT_NVCC_PCH_VOLUME", "kernelbot-nvcc-pch"), create_if_missing=True +) +submission_volumes = {PCH_MOUNT: pch_volume.with_mount_options(read_only=True)} +if NVCC_TOOLCHAIN_ENABLED: + binary_volume = Volume.from_name( + os.environ.get("KERNELBOT_NVCC_TOOLCHAIN_VOLUME", "kernelbot-nvcc-toolchains") + ) + cuda_image = cuda_image.apt_install( + # Preserve the full devel-image surface (tileiras, cuBLAS, sanitizer, + # profiler tools, and all library headers) when switching CUDA_HOME. + "cuda-toolkit-13-4=13.4.3-1", + "cuda-nvcc-13-4=13.4.92-1", + "cuda-cudart-dev-13-4=13.4.92-1", + "cuda-driver-dev-13-4=13.4.92-1", + "libnvvm-13-4=13.4.92-1", + "cuda-crt-13-4=13.4.92-1", + "libnvptxcompiler-13-4=13.4.92-1", + ).run_commands( + "ln -sfn /usr/local/cuda-13.4 /usr/local/cuda", + # The toolkit meta-package may install a newer Nsight. Keep the tested + # profiler on both CUDA's and the system's command lookup paths. + "ln -sfn /opt/nvidia/nsight-compute/2025.2.1/ncu /usr/local/cuda/bin/ncu", + "ln -sfn /opt/nvidia/nsight-compute/2025.2.1/ncu /usr/local/bin/ncu", + "ncu --version", + "tileiras --version", + "compute-sanitizer --version", + ( + "printf '#include \\n#include \\n" + "int main() { cublasHandle_t h; return cublasCreate(&h); }\\n' | " + "nvcc -std=c++17 -x cu - -lcublas -o /tmp/cublas-toolkit-smoke && " + "rm /tmp/cublas-toolkit-smoke" + ), + "python /opt/kernelbot-pch/compiler.py --kernelbot-fingerprint", + ) + for filename in ("nvcc.py", "nvcc_common.py", "nvcc_warm.py"): + cuda_image = cuda_image.add_local_file( + pch_files / filename, f"/opt/kernelbot-pch/{filename}", copy=True + ) + cuda_image = ( + cuda_image.run_commands("chmod +x /opt/kernelbot-pch/nvcc.py") + .run_function( + install_nvcc_toolchain, + volumes={"/nvcc-toolchain-input": binary_volume.with_mount_options(read_only=True)}, + kwargs={"release": RELEASE, "expected_hashes": BINARY_SHA256}, + cpu=4, + memory=8192, + timeout=600, + ) + .env( + { + "PYTORCH_NVCC": "/opt/kernelbot-pch/nvcc.py", + "KERNELBOT_NVCC_TOOLCHAIN": "1", + "KERNELBOT_NVCC_PCH": "1" if NVCC_PCH_ENABLED else "0", + "KERNELBOT_NVCC_PCH_ROOT": NVCC_PCH_MOUNT, + } + ) + ) + submission_volumes[NVCC_PCH_MOUNT] = nvcc_pch_volume.with_mount_options(read_only=True) + +cuda_image_base = cuda_image cuda_image = cuda_image.add_local_python_source( "libkernelbot", "modal_runner", "modal_runner_archs", + "nvcc_toolchain", ) MODAL_RUN_TIMEOUT_SECONDS = 60 * 60 @@ -207,7 +279,7 @@ def modal_run_config( # noqa: C901 timeout=Timeout.COMPILE + 30, single_use_containers=True, restrict_modal_access=True, - volumes={PCH_MOUNT: pch_volume.with_mount_options(read_only=True)}, + volumes=submission_volumes, ) def compile_python_submission(config: dict) -> dict: return compile_python(config) @@ -236,3 +308,26 @@ def warm_pch(profiles: str = ""): subprocess.run(["python3", "/opt/kernelbot-pch/warm.py", *selected], check=True) pch_volume.commit() return "Torch PCH profiles committed" + + +@app.function( + image=cuda_image, + cpu=4, + memory=16384, + timeout=1800, + max_containers=1, + volumes={NVCC_PCH_MOUNT: nvcc_pch_volume}, +) +def warm_nvcc_pch(profiles: str = ""): + """Publish fixed, trusted profiles before replacing submission workers.""" + if not NVCC_TOOLCHAIN_ENABLED: + return "NVCC toolchain disabled; no CUDA profiles warmed" + nvcc_pch_volume.reload() + subprocess.run( + ["python3", "/opt/kernelbot-pch/nvcc_warm.py", *profiles.split(",")] + if profiles + else ["python3", "/opt/kernelbot-pch/nvcc_warm.py"], + check=True, + ) + nvcc_pch_volume.commit() + return "NVCC PCH profiles committed" diff --git a/src/runners/modal_runner_archs.py b/src/runners/modal_runner_archs.py index 64f327f1..04fcfb90 100644 --- a/src/runners/modal_runner_archs.py +++ b/src/runners/modal_runner_archs.py @@ -1,6 +1,12 @@ # This file contains wrapper functions for running # Modal apps on specific devices. We will fix this later. -from modal_runner import MODAL_RUN_TIMEOUT_SECONDS, PCH_MOUNT, app, cuda_image, modal_run_config, pch_volume +from modal_runner import ( + MODAL_RUN_TIMEOUT_SECONDS, + app, + cuda_image, + modal_run_config, + submission_volumes, +) from libkernelbot.validation_runtime import run_validation_config @@ -14,7 +20,7 @@ name=f"run_cuda_script_{gpu_slug}", serialized=True, restrict_modal_access=True, - volumes={PCH_MOUNT: pch_volume.with_mount_options(read_only=True)}, + volumes=submission_volumes, timeout=MODAL_RUN_TIMEOUT_SECONDS, )(modal_run_config) pytorch_functions[gpu_slug] = app.function( @@ -23,7 +29,7 @@ name=f"run_pytorch_script_{gpu_slug}", serialized=True, restrict_modal_access=True, - volumes={PCH_MOUNT: pch_volume.with_mount_options(read_only=True)}, + volumes=submission_volumes, timeout=MODAL_RUN_TIMEOUT_SECONDS, )(modal_run_config) @@ -33,6 +39,6 @@ name="run_validation_script_b200", serialized=True, restrict_modal_access=True, - volumes={PCH_MOUNT: pch_volume.with_mount_options(read_only=True)}, + volumes=submission_volumes, timeout=MODAL_RUN_TIMEOUT_SECONDS, )(run_validation_config) diff --git a/src/runners/nvcc_toolchain.py b/src/runners/nvcc_toolchain.py new file mode 100644 index 00000000..d25e5407 --- /dev/null +++ b/src/runners/nvcc_toolchain.py @@ -0,0 +1,114 @@ +"""Pinned private NVIDIA compiler drop used by the opt-in Modal runner image.""" + +RELEASE = "nvidia-pch-13.4.99-20261017" +BINARY_SHA256 = { + "nvcc": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298", + "cudafe++": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442", + "nvvm/bin/cicc": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050", + "nvvm-next/bin/cicc": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6", +} + + +def install_nvcc_toolchain(release: str, expected_hashes: dict[str, str]): # noqa: C901 + """Image-build step: consume a private read-only volume, never submission input.""" + import hashlib + import json + import re + import shutil + import subprocess + import tempfile + from pathlib import Path + + source = Path("/nvcc-toolchain-input") / release + cuda = Path("/usr/local/cuda") + targets = { + "nvcc": "bin/nvcc", + "cudafe++": "bin/cudafe++", + "nvvm/bin/cicc": "nvvm/bin/cicc", + "nvvm-next/bin/cicc": "nvvm-next/bin/cicc", + } + if expected_hashes.keys() != targets.keys(): + raise ValueError("Expected checksums for exactly the four NVIDIA compiler binaries") + for name, expected in expected_hashes.items(): + data = (source / name).read_bytes() + if hashlib.sha256(data).hexdigest() != expected: + raise ValueError(f"NVIDIA compiler checksum mismatch: {name}") + for name, target_name in targets.items(): + target = cuda / target_name + target.parent.mkdir(parents=True, exist_ok=True) + shutil.copyfile(source / name, target) + target.chmod(0o755) + + # All supported Modal runner GPUs are SM75 or newer. The vendor profile + # otherwise keeps selecting the old backend despite the second binary. + profile_path = cuda / "bin/nvcc.profile" + profile, count = re.subn( + r"^CICC_PATH\s*=.*$", + "CICC_PATH = $(TOP)/nvvm-next/bin", + profile_path.read_text(), + flags=re.MULTILINE, + ) + if count != 1: + raise ValueError("Expected exactly one CICC_PATH in the CUDA compiler profile") + profile_path.write_text(profile) + + compiler = str(cuda / "bin/nvcc") + version = subprocess.check_output([compiler, "--version"], text=True) + if "V13.4.99" not in version: + raise ValueError(f"Unexpected NVIDIA compiler version: {version}") + smoke = [] + with tempfile.TemporaryDirectory() as directory: + root = Path(directory) + source_file = root / "smoke.cu" + source_file.write_text("__global__ void kernel(int* x) { x[0] = 7; }\n") + for arch in ("75", "80", "89", "90a", "100", "100a"): + command = [compiler, f"-arch=sm_{arch}", "-c", str(source_file), "-o"] + command.append(str(root / f"sm_{arch}.o")) + dryrun = subprocess.check_output( + [*command, "--dryrun"], stderr=subprocess.STDOUT, text=True + ) + if "nvvm-next/bin" not in dryrun: + raise ValueError(f"Wrong CUDA compiler backend for SM{arch}") + subprocess.run(command, check=True, timeout=60) + smoke.append(arch) + + package_output = subprocess.check_output( + [ + "dpkg-query", + "--show", + "--showformat=${Package}\t${Version}\t${Status}\n", + "*13-4*", + ], + text=True, + ) + cuda_packages = {} + for line in package_output.splitlines(): + name, package_version, status = line.split("\t", 2) + if status == "install ok installed": + cuda_packages[name] = package_version + if not cuda_packages: + raise ValueError("No installed CUDA 13.4 packages found for the toolchain fingerprint") + + artifact_root = Path("/opt/kernelbot-pch") + files = [ + *(cuda / name for name in targets.values()), + profile_path, + *(cuda / "bin" / name for name in ("ptxas", "nvlink", "fatbinary")), + cuda / "nvvm/libdevice/libdevice.10.bc", + *(artifact_root / name for name in ("nvcc.py", "nvcc_common.py", "nvcc_warm.py")), + ] + metadata = { + "version": 1, + "release": release, + "nvcc_version": version, + "smoke_architectures": smoke, + "cuda_packages": cuda_packages, + "headers_fingerprint": (artifact_root / "fingerprint").read_text().strip(), + "files": {str(p): hashlib.sha256(p.read_bytes()).hexdigest() for p in files}, + } + fingerprint = hashlib.sha256(json.dumps(metadata, sort_keys=True).encode()).hexdigest() + (artifact_root / "nvcc-toolchain.json").write_text( + json.dumps({**metadata, "fingerprint": fingerprint}, indent=2) + "\n" + ) + (artifact_root / "nvcc-fingerprint").write_text(fingerprint + "\n") + print(f"Installed {release}; toolchain fingerprint {fingerprint}", flush=True) diff --git a/src/runners/pch/nvcc.py b/src/runners/pch/nvcc.py new file mode 100644 index 00000000..1f11a234 --- /dev/null +++ b/src/runners/pch/nvcc.py @@ -0,0 +1,128 @@ +#!/usr/bin/env python3 +"""Read-only, opt-in NVCC PCH wrapper for compatible load_inline CUDA sources.""" + +import json +import os +import re +import subprocess +import sys +import time +from pathlib import Path + +import nvcc_common as common + +USED = re.compile(r'using precompiled header file "([^"]+)"') +CREATED = re.compile(r'creating precompiled header file "([^"]+)"') +REJECTION = re.compile( + r"invalid (?:PCH|precompiled header)|cannot open precompiled header|" + r"precompiled header[^\n]*(?:invalid|incomplete|not generated|have changed)|" + r"memory usage conflict with precompiled header|" + r"command line options do not match those used when precompiled header|" + r"initial sequence of preprocessing directives is not compatible|" + r"could not publish precompiled header|" + r"source preceding the #pragma hdrstop does not match|" + r"incompatible precompiled header \(PCH\)", + re.IGNORECASE, +) + + +def event(status, started, code, **details): + record = { + "status": status, + "seconds": time.perf_counter() - started, + "returncode": code, + **details, + } + print("[kernelbot-nvcc-pch] " + json.dumps(record, sort_keys=True), file=sys.stderr) + if destination := os.environ.get("KERNELBOT_NVCC_PCH_EVENTS"): + try: + with open(destination, "a") as output: + output.write(json.dumps(record, sort_keys=True) + "\n") + except OSError: + pass # Observability must not break a submission compilation. + + +def run_compiler(args, capture=False): + return subprocess.run([common.COMPILER, *args], capture_output=capture, text=capture) + + +def passthrough(args, status, reason, started, **details): + result = run_compiler(args) + if "-c" in args: + event(status, started, result.returncode, reason=reason, **details) + return result.returncode + + +def consumed(log, entry): + paths = USED.findall(log) + frontends = {Path(path).name.split("_", 1)[0] for path in paths} + valid = ( + frontends == {"cicc", "cudafe"} + and not CREATED.search(log) + and all(Path(path).is_relative_to(entry / "pch") for path in paths) + ) + return valid, sorted(frontends) + + +def main(args): + started = time.perf_counter() + if ( + os.environ.get("KERNELBOT_NVCC_PCH") != "1" + or os.environ.get("KERNELBOT_NVCC_PCH_DISABLE") == "1" + ): + return passthrough(args, "disabled", "disabled", started) + flags = common.compile_flags(args) + if flags is None: + return passthrough(args, "bypass", "unsupported_invocation_or_headers", started) + details = {} + try: + fingerprint = common.generation() + key = common.cache_key(flags, fingerprint) + details = {"generation": fingerprint, "key": key} + entry = common.cache_root() / fingerprint / key + manifest = common.read_manifest(entry, flags, fingerprint) + except (OSError, ValueError, KeyError, TypeError): + return passthrough(args, "miss", "profile_unavailable", started, **details) + details["profile"] = manifest["profile"] + cached_args = [ + *args, + "-include", + str(entry / "gpumode.cuh"), + "--pch", + "--pch-dir", + str(entry / "pch"), + "--pch-verbose", + ] + tick = time.perf_counter() + result = run_compiler(cached_args, capture=True) + cached_seconds = time.perf_counter() - tick + log = result.stdout + result.stderr + hit, frontends = consumed(log, entry) + if result.returncode and (CREATED.search(log) or REJECTION.search(log)): + # Cache-specific failure only. Genuine user compiler errors remain errors. + # Original argv removes both the forwarding header and every PCH option. + print("[kernelbot-nvcc-pch] rejected profile; compiling normally", file=sys.stderr) + return passthrough( + args, + "fallback", + "cache_rejected", + started, + cached_attempt_seconds=cached_seconds, + **details, + ) + sys.stdout.write(result.stdout) + sys.stderr.write(result.stderr) + status = "error" if result.returncode else ("hit" if hit else "fallback") + event( + status, + started, + result.returncode, + frontends=frontends, + reason="consumed" if hit else "not_consumed", + **details, + ) + return result.returncode + + +if __name__ == "__main__": + sys.exit(main(sys.argv[1:])) diff --git a/src/runners/pch/nvcc_common.py b/src/runners/pch/nvcc_common.py new file mode 100644 index 00000000..8cc1211b --- /dev/null +++ b/src/runners/pch/nvcc_common.py @@ -0,0 +1,189 @@ +"""Shared, stdlib-only identity rules for trusted NVCC header profiles.""" + +import hashlib +import json +import os +import re +from pathlib import Path + +VERSION = 1 +COMPILER = "/usr/local/cuda/bin/nvcc" +FINGERPRINT = Path("/opt/kernelbot-pch/nvcc-fingerprint") +# Match PyTorch 2.12 load_inline's CUDA prefix exactly. In particular, ATen is +# included transitively by torch/types.h rather than inserted before it here. +HEADERS = ("torch/types.h", "cuda.h", "cuda_runtime.h") +HEADER = "#pragma once\n" + "".join(f"#include <{h}>\n" for h in HEADERS) +HEADER += "#pragma hdrstop\n" +HEADER_SHA256 = hashlib.sha256(HEADER.encode()).hexdigest() +ARCHES = ("75", "80", "89", "90a", "100") +# Ordered tuples observed in the 2026-11-13 submission snapshot. Preserve every +# option and its position; the warmer captures PyTorch's remaining flags. +FLAG_PROFILES = { + "default": [], + "O3": ["-O3"], + "O3-fast-math": ["-O3", "--use_fast_math"], +} +PROFILES = { + f"sm{arch}-{name}": (arch, flags) for arch in ARCHES for name, flags in FLAG_PROFILES.items() +} +EXPLICIT_PROFILES = { + "sm100a-O3-fast-math-arch": ("100a", ["-O3", "--use_fast_math", "-arch=sm_100a"]), + "sm100-O3-arch-fast-math": ("100", ["-O3", "-arch=sm_100", "--use_fast_math"]), + "sm80-O3-cxx17-arch-fast-math": ( + "80", + ["-O3", "-std=c++17", "-arch=sm_80", "--use_fast_math"], + ), +} +PROFILES.update(EXPLICIT_PROFILES) +DEFAULT_PROFILES = tuple(PROFILES) +ENVIRONMENT_KEYS = ( + "CPATH", + "CPLUS_INCLUDE_PATH", + "C_INCLUDE_PATH", + "GCC_EXEC_PREFIX", + "COMPILER_PATH", + "NVCC_CCBIN", + "NVVMIR_LIBRARY_DIR", + "PATH", + "LD_LIBRARY_PATH", + "LIBRARY_PATH", + "CUDA_HOME", + "CUDA_PATH", + "CUDA_ROOT", +) +HEX = re.compile(r"[0-9a-f]{64}\Z") +PCH_NAME = re.compile(r"(?:cicc|cudafe)_[0-9a-f_]+\.pch\Z") +PREFIX = re.compile( + r"\A[ \t\r\n]*" + + "".join(r"#[ \t]*include[ \t]*<" + re.escape(h) + r">[ \t]*\r?\n[ \t\r\n]*" for h in HEADERS) +) + + +def cache_root(): + return Path(os.environ.get("KERNELBOT_NVCC_PCH_ROOT", "/kernelbot-nvcc-pch")) + + +def generation(): + value = FINGERPRINT.read_text().strip() + if not HEX.fullmatch(value): + raise ValueError("Invalid NVCC toolchain fingerprint") + return value + + +def environment(): + return {key: os.environ.get(key) for key in ENVIRONMENT_KEYS} + + +def compile_flags(args): + """Return only compatible single CUDA compile flags; never mutate the command.""" + if args.count("-c") != 1 or args.count("-o") != 1: + return None + if any(os.environ.get(k) for k in ("NVCC_PREPEND_FLAGS", "NVCC_APPEND_FLAGS")): + return None + if any( + a.startswith( + ( + "@", + "-ccbin", + "--compiler-bindir", + "--pch", + "--use-pch", + "--create-pch", + "-include", + "--pre-include", + ) + ) + for a in args + ): + return None + try: + source = Path(args[args.index("-c") + 1]) + if source.suffix != ".cu": + return None + # PyTorch inserts this prefix before user code. Do not move user macros or + # introduce Torch/CUTLASS headers into deliberately minimal CUDA sources. + with source.open() as stream: + if not PREFIX.match(stream.read(4096)): + return None + except (IndexError, OSError, UnicodeError): + return None + return strip_outputs(args) + + +def strip_outputs(args, drop_module=True): + flags = [] + paired = {"-c", "-o", "-MF", "-MT", "-MQ", "--dependency-output"} + standalone = {"-MD", "-MMD", "--generate-dependencies-with-compile"} + index = 0 + while index < len(args): + arg = args[index] + if arg in paired: + if index + 1 >= len(args): + return None + index += 2 + continue + if drop_module and re.fullmatch(r"-DTORCH_EXTENSION_NAME=[A-Za-z_]\w*", arg): + # Keep this macro in the real invocation. NVIDIA validates whether + # it was used in a PCH; the curated forwarding header does not use it. + index += 1 + continue + if arg not in standalone and not arg.startswith("--dependency-output="): + flags.append(arg) + index += 1 + return flags + + +def identity(flags, fingerprint): + return { + "version": VERSION, + "generation": fingerprint, + "header_sha256": HEADER_SHA256, + "flags": flags, + "environment": environment(), + } + + +def cache_key(flags, fingerprint): + value = json.dumps(identity(flags, fingerprint), sort_keys=True).encode() + return hashlib.sha256(value).hexdigest() + + +def read_manifest(entry, flags, fingerprint): + """Validate ready metadata and cheap file stats, without reading multi-GB blobs.""" + manifest = json.loads((entry / "manifest.json").read_text()) + return validate_manifest(manifest, entry, flags, fingerprint) + + +def validate_manifest(manifest, entry, flags, fingerprint): + if not isinstance(manifest, dict): + raise ValueError("Invalid PCH manifest") + expected = identity(flags, fingerprint) + if any(manifest.get(k) != value for k, value in expected.items()): + raise ValueError("PCH profile identity changed") + if manifest.get("key") != entry.name or entry.name != cache_key(flags, fingerprint): + raise ValueError("PCH profile key changed") + if manifest.get("profile") not in PROFILES: + raise ValueError("Unrecognized trusted PCH profile") + if (entry / "gpumode.cuh").read_text() != HEADER: + raise ValueError("PCH forwarding header changed") + validate_files(entry, manifest.get("files", [])) + return manifest + + +def validate_files(entry, files): + if len(files) != 2: + raise ValueError("PCH profile must contain both frontends") + frontends = set() + for item in files: + path = Path(item["path"]) + if path.parent != Path("pch") or not PCH_NAME.fullmatch(path.name): + raise ValueError("Invalid PCH manifest path") + target = entry / path + if not target.resolve().is_relative_to(entry.resolve()): + raise ValueError("PCH file escapes its profile") + size = item["bytes"] + if not isinstance(size, int) or size <= 0 or target.stat().st_size != size: + raise ValueError("PCH file is missing or truncated") + frontends.add(path.name.split("_", 1)[0]) + if frontends != {"cicc", "cudafe"}: + raise ValueError("PCH profile is missing a frontend") diff --git a/src/runners/pch/nvcc_warm.py b/src/runners/pch/nvcc_warm.py new file mode 100644 index 00000000..5c3de7ab --- /dev/null +++ b/src/runners/pch/nvcc_warm.py @@ -0,0 +1,137 @@ +"""Trusted CPU-only NVCC PCH warmer; accepts curated profile names, never user flags.""" + +import json +import os +import re +import shutil +import subprocess +import sys +import tempfile +import time +from pathlib import Path + +import nvcc_common as common + +CAPTURE = """#!/usr/bin/env python3 +import json, os, subprocess, sys +from pathlib import Path +compiler = "/usr/local/cuda/bin/nvcc" if sys.argv[0].endswith("nvcc") else "/usr/bin/g++" +if "-c" in sys.argv: + if compiler.endswith("nvcc"): + with open(os.environ["KERNELBOT_CAPTURE_NVCC"], "w") as output: + json.dump(sys.argv[1:], output) + sys.exit(1) + Path(sys.argv[sys.argv.index("-o") + 1]).touch() + sys.exit(0) +sys.exit(subprocess.call([compiler, *sys.argv[1:]])) +""" +BUILD = """ +import json, os +from torch.utils.cpp_extension import load_inline +load_inline(name="kernelbot_nvcc_warmer", cpp_sources="int warm() { return 1; }", + cuda_sources="__global__ void warm_cuda() {}", functions=["warm"], + extra_cuda_cflags=json.loads(os.environ["KERNELBOT_CAPTURE_FLAGS"]), + build_directory=os.environ["KERNELBOT_CAPTURE_BUILD"], verbose=True) +""" + + +def torch_arch(arch): + match = re.fullmatch(r"([1-9][0-9]*)([0-9])(a?)", arch) + if match is None: + raise ValueError("Invalid curated CUDA architecture") + major, minor, suffix = match.groups() + return f"{major}.{minor}{suffix}" + + +def capture(profile, work): + arch, extra = common.PROFILES[profile] + for kind in ("cxx", "nvcc"): + path = work / f"capture-{kind}" + path.write_text(CAPTURE) + path.chmod(0o755) + capture_path = work / "args.json" + environment = os.environ | { + "CXX": str(work / "capture-cxx"), + "PYTORCH_NVCC": str(work / "capture-nvcc"), + "TORCH_CUDA_ARCH_LIST": torch_arch(arch), + "MAX_JOBS": "2", + "KERNELBOT_CAPTURE_NVCC": str(capture_path), + "KERNELBOT_CAPTURE_FLAGS": json.dumps(extra), + "KERNELBOT_CAPTURE_BUILD": str(work), + } + result = subprocess.run( + ["python3", "-c", BUILD], env=environment, capture_output=True, text=True, timeout=180 + ) + if not capture_path.exists(): + raise RuntimeError("Could not capture PyTorch NVCC flags: " + result.stdout + result.stderr) + args = json.loads(capture_path.read_text()) + flags = common.compile_flags(args) + if flags is None: + raise RuntimeError("Trusted warm invocation did not pass runtime eligibility rules") + return flags, common.strip_outputs(args, drop_module=False) + + +def warm_profile(profile, fingerprint): + with tempfile.TemporaryDirectory(prefix="kernelbot-nvcc-warm-") as directory: + work = Path(directory) + flags, warm_flags = capture(profile, work) + key = common.cache_key(flags, fingerprint) + entry = common.cache_root() / fingerprint / key + if (entry / "manifest.json").exists(): + return common.read_manifest(entry, flags, fingerprint) + # Only incomplete, unpublished entries may be repaired by the single writer. + entry.mkdir(parents=True, exist_ok=True) + shutil.rmtree(entry / "pch", ignore_errors=True) + (entry / "pch").mkdir() + (entry / "gpumode.cuh").write_text(common.HEADER) + command = [ + common.COMPILER, + *warm_flags, + "--pch", + "--pch-verbose", + "--pch-dir", + str(entry / "pch"), + "-x", + "cu", + "-c", + str(entry / "gpumode.cuh"), + "-o", + str(work / "header.o"), + ] + print(f"Warming {profile}: {key}", flush=True) + started = time.perf_counter() + result = subprocess.run(command, capture_output=True, text=True, timeout=600) + if result.returncode: + raise RuntimeError(result.stdout + result.stderr) + files = [ + {"path": str(path.relative_to(entry)), "bytes": path.stat().st_size} + for path in sorted((entry / "pch").glob("*.pch")) + ] + manifest = common.identity(flags, fingerprint) | { + "key": key, + "profile": profile, + "files": files, + "creation_seconds": time.perf_counter() - started, + } + common.validate_manifest(manifest, entry, flags, fingerprint) + temporary = entry / "manifest.json.tmp" + temporary.write_text(json.dumps(manifest, sort_keys=True, indent=2) + "\n") + # Validate the exact ready metadata before making it visible to workers. + # The parent Modal warmer commits the completed volume snapshot afterward. + temporary.replace(entry / "manifest.json") + print(json.dumps(manifest), flush=True) + return manifest + + +def warm(profiles=()): + selected = list(profiles) or list(common.DEFAULT_PROFILES) + if not set(selected) <= common.PROFILES.keys(): + raise ValueError(f"Choose fixed profiles from {sorted(common.PROFILES)}") + if any(os.environ.get(k) for k in ("NVCC_PREPEND_FLAGS", "NVCC_APPEND_FLAGS")): + raise ValueError("Trusted warmup does not accept NVCC environment flags") + fingerprint = common.generation() + return [warm_profile(profile, fingerprint) for profile in selected] + + +if __name__ == "__main__": + warm(sys.argv[1:]) diff --git a/tests/test_inline_artifacts.py b/tests/test_inline_artifacts.py index 869c1640..99d590f3 100644 --- a/tests/test_inline_artifacts.py +++ b/tests/test_inline_artifacts.py @@ -30,6 +30,9 @@ def original(name, cpp_sources, cuda_sources=None, extra_cuda_cflags=None, extra monkeypatch.setenv("KERNELBOT_INLINE_WORKDIR", str(tmp_path)) monkeypatch.setenv("TORCH_CUDA_ARCH_LIST", "7.5") monkeypatch.setenv("KERNELBOT_INLINE_MODE", "capture") + fingerprint = tmp_path / "nvcc-fingerprint" + fingerprint.write_text("a" * 64 + "\n") + monkeypatch.setattr("libkernelbot.inline_artifacts.NVCC_FINGERPRINT", fingerprint) with patch("libkernelbot.inline_artifacts._dependencies", return_value={}): install() extension.load_inline("example", "source", cuda_sources="cuda") @@ -53,7 +56,18 @@ def test_loads_saved_module_once_without_compiling(extension, tmp_path): assert extension.builds.call_count == 1 -@pytest.mark.parametrize("change", ["source", "flags", "arch", "manifest", "binary", "linker"]) +@pytest.mark.parametrize( + "change", + [ + "source", + "flags", + "arch", + "manifest", + "binary", + "linker", + "nvcc_toolchain", + ], +) def test_incompatible_artifacts_fall_back_before_loading(extension, tmp_path, monkeypatch, change): arguments = {"name": "example", "cpp_sources": "source", "cuda_sources": "cuda"} manifest_path = next((tmp_path / "artifacts").glob("*/manifest.json")) @@ -69,11 +83,41 @@ def test_incompatible_artifacts_fall_back_before_loading(extension, tmp_path, mo manifest_path.write_text(json.dumps(manifest)) elif change == "binary": manifest_path.with_name("extension.so").write_bytes(b"corrupted") - else: + elif change == "linker": arguments["extra_ldflags"] = ["-lcustom"] + elif change == "nvcc_toolchain": + (tmp_path / "nvcc-fingerprint").write_text("b" * 64 + "\n") with patch("importlib.util.spec_from_file_location") as load: extension.load_inline(**arguments) load.assert_not_called() assert extension.builds.call_count == 2 event = json.loads((tmp_path / "artifacts/events.jsonl").read_text()) assert event["mode"] == "fallback" and event["reason"] + + +@pytest.mark.parametrize( + "name,value", + [ + ("PYTORCH_NVCC", "/different/compiler-wrapper"), + ("KERNELBOT_NVCC_PCH", "1"), + ("KERNELBOT_NVCC_PCH_DISABLE", "1"), + ("KERNELBOT_NVCC_PCH_ROOT", "/different/cache"), + ("CUDA_PATH", "/different/cuda"), + ("CUDA_ROOT", "/different/cuda"), + ("GCC_EXEC_PREFIX", "/different/gcc"), + ("COMPILER_PATH", "/different/compiler"), + ("PATH", "/different/bin"), + ("NVCC_CCBIN", "/different/cxx"), + ("NVVMIR_LIBRARY_DIR", "/different/nvvm"), + ], +) +def test_nvcc_environment_change_rebuilds_before_loading( + extension, tmp_path, monkeypatch, name, value +): + monkeypatch.setenv(name, value) + with patch("importlib.util.spec_from_file_location") as load: + extension.load_inline("example", "source", cuda_sources="cuda") + load.assert_not_called() + assert extension.builds.call_count == 2 + event = json.loads((tmp_path / "artifacts/events.jsonl").read_text()) + assert event["mode"] == "fallback" diff --git a/tests/test_nvcc_pch.py b/tests/test_nvcc_pch.py new file mode 100644 index 00000000..d37ddc98 --- /dev/null +++ b/tests/test_nvcc_pch.py @@ -0,0 +1,321 @@ +"""NVCC cache eligibility, immutable identity, and failure recovery without CUDA.""" + +import importlib.util +import json +import subprocess +import sys +from pathlib import Path + +import pytest + +SOURCE = Path(__file__).parents[1] / "src/runners/pch" + + +def load(name, path): + spec = importlib.util.spec_from_file_location(name, path) + module = importlib.util.module_from_spec(spec) + sys.modules[name] = module + spec.loader.exec_module(module) + return module + + +common = load("nvcc_common", SOURCE / "nvcc_common.py") +wrapper = load("kernelbot_nvcc_wrapper", SOURCE / "nvcc.py") +warmer = load("kernelbot_nvcc_warmer", SOURCE / "nvcc_warm.py") +GENERATION = "a" * 64 + + +@pytest.fixture(autouse=True) +def clean_flags(monkeypatch): + for key in ("NVCC_PREPEND_FLAGS", "NVCC_APPEND_FLAGS", "KERNELBOT_NVCC_PCH_DISABLE"): + monkeypatch.delenv(key, raising=False) + monkeypatch.setenv("KERNELBOT_NVCC_PCH", "1") + + +def command(tmp_path, name="first", extra=()): + source = tmp_path / f"{name}.cu" + prefix = "".join(f"#include <{h}>\n" for h in common.HEADERS) + source.write_text(prefix + '#include "user.h"\n__global__ void kernel() {}\n') + return [ + "-MD", + "-MF", + f"{name}.d", + f"-DTORCH_EXTENSION_NAME={name}", + "-DTORCH_API_INCLUDE_EXTENSION_H", + "-isystem", + "/torch/include", + "-gencode=arch=compute_75,code=sm_75", + "--compiler-options", + "-fPIC", + "-std=c++20", + *extra, + "-c", + str(source), + "-o", + f"{name}.o", + ] + + +def ready(tmp_path, monkeypatch, args): + root = tmp_path / "volume" + monkeypatch.setenv("KERNELBOT_NVCC_PCH_ROOT", str(root)) + fingerprint = tmp_path / "fingerprint" + fingerprint.write_text(GENERATION) + monkeypatch.setattr(common, "FINGERPRINT", fingerprint) + flags = common.compile_flags(args) + key = common.cache_key(flags, GENERATION) + entry = root / GENERATION / key + (entry / "pch").mkdir(parents=True) + (entry / "gpumode.cuh").write_text(common.HEADER) + files = [] + for name in ("cicc_ab_12.pch", "cudafe_ab_12.pch"): + path = entry / "pch" / name + path.write_bytes(b"PCH stub") + files.append({"path": "pch/" + name, "bytes": path.stat().st_size}) + manifest = common.identity(flags, GENERATION) | { + "key": key, + "profile": "sm75-default", + "files": files, + } + (entry / "manifest.json").write_text(json.dumps(manifest)) + return entry, manifest + + +def fake_compiler(monkeypatch, responses): + calls = [] + + def run(args, capture=False): + calls.append((args, capture)) + code, stderr = responses[len(calls) - 1] + return subprocess.CompletedProcess(args, code, "", stderr) + + monkeypatch.setattr(wrapper, "run_compiler", run) + return calls + + +def hit_log(entry): + return "\n".join( + f'using precompiled header file "{path}"' for path in sorted((entry / "pch").glob("*.pch")) + ) + + +def test_different_modules_and_user_sources_share_only_header_profile(tmp_path): + first = command(tmp_path) + second = command(tmp_path, "second") + source = Path(second[second.index("-c") + 1]) + source.write_text(source.read_text().replace("kernel() {}", "kernel() { int changed = 2; }")) + one, two = common.compile_flags(first), common.compile_flags(second) + assert common.cache_key(one, GENERATION) == common.cache_key(two, GENERATION) + assert "-DTORCH_EXTENSION_NAME=first" in first + + +def test_pytorch_212_cuda_prefix_is_eligible_without_extra_headers(tmp_path): + args = command(tmp_path) + Path(args[args.index("-c") + 1]).write_text( + "#include \n" + "#include \n" + "#include \n" + "__global__ void kernel() {}\n" + ) + assert common.compile_flags(args) is not None + assert common.HEADERS == ("torch/types.h", "cuda.h", "cuda_runtime.h") + + +@pytest.mark.parametrize( + "extra", + [ + ["-O3"], + ["-DUSER_MACRO=1"], + ["-I/custom/headers"], + ["--use_fast_math"], + ["-std=c++17"], + ["-gencode=arch=compute_100,code=sm_100"], + ], +) +def test_flags_must_match_exactly(tmp_path, extra): + baseline = common.compile_flags(command(tmp_path)) + changed = common.compile_flags(command(tmp_path, "changed", extra)) + assert common.cache_key(baseline, GENERATION) != common.cache_key(changed, GENERATION) + + +def test_toolchain_and_environment_invalidate_profile(tmp_path, monkeypatch): + flags = common.compile_flags(command(tmp_path)) + key = common.cache_key(flags, GENERATION) + assert key != common.cache_key(flags, "b" * 64) + monkeypatch.setenv("CPATH", "/different/headers") + assert key != common.cache_key(flags, GENERATION) + + +@pytest.mark.parametrize( + "prefix", + [ + "#include \n", + "#define USER_SETTING 1\n", + "#include #include \n", + ], +) +def test_no_implicit_headers_and_macro_reordering_bypass(tmp_path, prefix): + args = command(tmp_path) + Path(args[args.index("-c") + 1]).write_text(prefix + "__global__ void kernel() {}\n") + assert common.compile_flags(args) is None + + +@pytest.mark.parametrize( + "extra", [["@flags.rsp"], ["-include", "custom.h"], ["--compiler-bindir=/custom"], ["--pch"]] +) +def test_custom_compiler_and_hidden_flags_bypass(tmp_path, extra): + assert common.compile_flags(command(tmp_path, extra=extra)) is None + + +def test_environment_injected_options_bypass(tmp_path, monkeypatch): + monkeypatch.setenv("NVCC_APPEND_FLAGS", "-include /user/header.h") + assert common.compile_flags(command(tmp_path)) is None + + +def test_unknown_profile_never_creates_volume_files(tmp_path, monkeypatch): + args = command(tmp_path) + monkeypatch.setenv("KERNELBOT_NVCC_PCH_ROOT", str(tmp_path / "absent")) + monkeypatch.setattr(common, "generation", lambda: GENERATION) + calls = fake_compiler(monkeypatch, [(0, "")]) + assert wrapper.main(args) == 0 + assert calls == [(args, False)] + assert not (tmp_path / "absent").exists() + + +def test_hit_keeps_original_flags_and_reports_both_frontends(tmp_path, monkeypatch): + args = command(tmp_path) + entry, _ = ready(tmp_path, monkeypatch, args) + events = tmp_path / "events.jsonl" + monkeypatch.setenv("KERNELBOT_NVCC_PCH_EVENTS", str(events)) + calls = fake_compiler(monkeypatch, [(0, hit_log(entry))]) + assert wrapper.main(args) == 0 + assert calls[0][0][: len(args)] == args + assert calls[0][0][-1] == "--pch-verbose" + record = json.loads(events.read_text()) + assert record["status"] == "hit" and record["frontends"] == ["cicc", "cudafe"] + + +def test_rejected_pch_retries_once_without_injected_header(tmp_path, monkeypatch): + args = command(tmp_path) + ready(tmp_path, monkeypatch, args) + calls = fake_compiler(monkeypatch, [(1, "invalid precompiled header file"), (0, "")]) + assert wrapper.main(args) == 0 + assert len(calls) == 2 and calls[1] == (args, False) + + +def test_user_compile_error_is_not_retried(tmp_path, monkeypatch): + args = command(tmp_path) + entry, _ = ready(tmp_path, monkeypatch, args) + calls = fake_compiler( + monkeypatch, [(2, hit_log(entry) + '\nerror: identifier "typo" undefined')] + ) + assert wrapper.main(args) == 2 + assert len(calls) == 1 + + +def test_partial_pch_consumption_is_not_claimed_as_hit(tmp_path, monkeypatch): + args = command(tmp_path) + entry, _ = ready(tmp_path, monkeypatch, args) + fake_compiler(monkeypatch, [(0, hit_log(entry).splitlines()[0])]) + events = tmp_path / "events.jsonl" + monkeypatch.setenv("KERNELBOT_NVCC_PCH_EVENTS", str(events)) + assert wrapper.main(args) == 0 + assert json.loads(events.read_text())["status"] == "fallback" + + +@pytest.mark.parametrize("change", ["truncated", "header", "generation", "escape", "partial"]) +def test_invalid_ready_metadata_falls_back_before_compiler(tmp_path, monkeypatch, change): + args = command(tmp_path) + entry, manifest = ready(tmp_path, monkeypatch, args) + if change == "truncated": + (entry / manifest["files"][0]["path"]).write_bytes(b"broken") + elif change == "header": + (entry / "gpumode.cuh").write_text("#include \n") + elif change == "generation": + manifest["generation"] = "b" * 64 + elif change == "escape": + manifest["files"][0]["path"] = "../foreign.pch" + else: + manifest["files"].pop() + (entry / "manifest.json").write_text(json.dumps(manifest)) + calls = fake_compiler(monkeypatch, [(0, "")]) + assert wrapper.main(args) == 0 + assert calls == [(args, False)] + + +def test_kill_switch_does_not_access_cache(tmp_path, monkeypatch): + args = command(tmp_path) + monkeypatch.setenv("KERNELBOT_NVCC_PCH_DISABLE", "1") + monkeypatch.setattr(common, "generation", lambda: pytest.fail("Cache accessed while disabled")) + calls = fake_compiler(monkeypatch, [(0, "")]) + assert wrapper.main(args) == 0 + assert calls == [(args, False)] + + +def test_warmer_rejects_untrusted_selectors_before_any_compile(monkeypatch): + monkeypatch.setattr(warmer, "warm_profile", lambda *args: pytest.fail("Unexpected compilation")) + with pytest.raises(ValueError, match="fixed profiles"): + warmer.warm(["-include /user/header.h"]) + + +def test_profile_catalog_has_bounded_architecture_and_flag_sets(): + assert len(common.PROFILES) == len(common.DEFAULT_PROFILES) == 18 + assert set(common.DEFAULT_PROFILES) == set(common.PROFILES) + assert common.ARCHES == ("75", "80", "89", "90a", "100") + assert list(common.FLAG_PROFILES.values()) == [[], ["-O3"], ["-O3", "--use_fast_math"]] + assert common.PROFILES["sm90a-default"] == ("90a", []) + assert common.PROFILES["sm100-O3-fast-math"] == ("100", ["-O3", "--use_fast_math"]) + assert "cutlass" not in common.HEADER.lower() + + +def test_observed_explicit_architecture_profiles_preserve_exact_ordered_flags(): + assert common.EXPLICIT_PROFILES == { + "sm100a-O3-fast-math-arch": ("100a", ["-O3", "--use_fast_math", "-arch=sm_100a"]), + "sm100-O3-arch-fast-math": ("100", ["-O3", "-arch=sm_100", "--use_fast_math"]), + "sm80-O3-cxx17-arch-fast-math": ( + "80", + ["-O3", "-std=c++17", "-arch=sm_80", "--use_fast_math"], + ), + } + + +@pytest.mark.parametrize( + ("arch", "expected"), [("75", "7.5"), ("90a", "9.0a"), ("100", "10.0"), ("100a", "10.0a")] +) +def test_capture_uses_curated_architecture_including_blackwell_suffix(arch, expected): + assert warmer.torch_arch(arch) == expected + + +def test_ready_validation_never_reads_pch_payloads(tmp_path, monkeypatch): + args = command(tmp_path) + entry, _ = ready(tmp_path, monkeypatch, args) + original_open = Path.open + + def guarded_open(path, *args, **kwargs): + assert path.suffix != ".pch", "PCH payload read during metadata validation" + return original_open(path, *args, **kwargs) + + monkeypatch.setattr(Path, "open", guarded_open) + assert common.read_manifest(entry, common.compile_flags(args), GENERATION)["files"] + + +def test_warm_command_retains_module_macro_and_publishes_only_complete_entry(tmp_path, monkeypatch): + args = command(tmp_path) + flags = common.compile_flags(args) + warm_flags = common.strip_outputs(args, drop_module=False) + monkeypatch.setenv("KERNELBOT_NVCC_PCH_ROOT", str(tmp_path / "warm")) + monkeypatch.setattr(warmer, "capture", lambda *args: (flags, warm_flags)) + + def compile_header(argv, **kwargs): + assert "-DTORCH_EXTENSION_NAME=first" in argv + pch_dir = Path(argv[argv.index("--pch-dir") + 1]) + assert not (pch_dir.parent / "manifest.json").exists() + (pch_dir / "cicc_ab_12.pch").write_bytes(b"device") + (pch_dir / "cudafe_ab_12.pch").write_bytes(b"frontend") + return subprocess.CompletedProcess(argv, 0, "", "") + + monkeypatch.setattr(warmer.subprocess, "run", compile_header) + result = warmer.warm_profile("sm75-default", GENERATION) + entry = common.cache_root() / GENERATION / result["key"] + assert common.read_manifest(entry, flags, GENERATION) == result + assert not (entry / "manifest.json.tmp").exists() diff --git a/tests/test_nvcc_release_validation.py b/tests/test_nvcc_release_validation.py new file mode 100644 index 00000000..8367988f --- /dev/null +++ b/tests/test_nvcc_release_validation.py @@ -0,0 +1,212 @@ +"""Exercise release-matrix selection and pass gates without running Modal jobs.""" + +import ast +import copy +import os +import runpy +from pathlib import Path +from types import SimpleNamespace +from unittest.mock import patch + +import pytest + + +@pytest.fixture(scope="module") +def harness(): + root = Path(__file__).parents[1] + with pytest.MonkeyPatch.context() as monkeypatch, patch.dict(os.environ): + monkeypatch.syspath_prepend(str(root / "src/runners")) + namespace = runpy.run_path(str(root / "scripts/validate_nvcc_pch.py")) + namespace["profile_catalog"]() + return SimpleNamespace(**namespace) + + +def test_default_matrix_preserves_twelve_fresh_cpu_gpu_cases(harness): + jobs = harness.matrix_jobs(harness.DEFAULT_GPUS, harness.CASES, "T4", [], 1) + assert len(jobs) == 12 + assert {case for gpu, case, _, _ in jobs if gpu == "T4"} == set(harness.CASES) + assert {case for gpu, case, _, _ in jobs if gpu != "T4"} == set(harness.BASELINE_CASES) + + +def test_all_gpu_matrix_adds_l4_and_production_a100_resource(harness): + jobs = harness.matrix_jobs(harness.GPUS, harness.CASES, "T4", [], 1) + assert len(jobs) == 16 + assert harness.GPU_RESOURCES["A100"] == "A100-80GB" + assert harness.GPU_RESOURCES.get("L4", "L4") == "L4" + assert {profile for gpu, _, profile, _ in jobs if gpu == "A100"} == {"sm80-default"} + assert {profile for gpu, _, profile, _ in jobs if gpu == "L4"} == {"sm89-default"} + assert harness.make_config("A100", "default")["arch"] == "80" + assert harness.make_config("L4", "default")["arch"] == "89" + + +def test_observed_sm80_profile_runs_only_on_a100(harness): + profile = "sm80-O3-cxx17-arch-fast-math" + jobs = harness.matrix_jobs(harness.GPUS, harness.BASELINE_CASES, "T4", [profile], 1) + assert jobs == [ + ("A100", "default", profile, 1), + ("A100", "disabled", profile, 1), + ] + assert harness.make_config("A100", "default", profile)["arch"] == "80" + with pytest.raises(ValueError, match="incompatible"): + harness.make_config("L4", "default", profile) + + +def test_selected_profiles_repeat_paired_baselines_and_alternate_order(harness): + jobs = harness.matrix_jobs(["B200"], harness.BASELINE_CASES, "T4", ["sm100-O3"], 2) + assert jobs == [ + ("B200", "default", "sm100-O3", 1), + ("B200", "disabled", "sm100-O3", 1), + ("B200", "disabled", "sm100-O3", 2), + ("B200", "default", "sm100-O3", 2), + ] + + +def test_cutlass_pair_uses_generic_o3_profile_without_expanding_default_matrix(harness): + jobs = harness.matrix_jobs(["T4"], harness.CUTLASS_CASES, "T4", [], 2) + assert jobs == [ + ("T4", "cutlass", "sm75-O3", 1), + ("T4", "cutlass-disabled", "sm75-O3", 1), + ("T4", "cutlass-disabled", "sm75-O3", 2), + ("T4", "cutlass", "sm75-O3", 2), + ] + assert not set(harness.CASES) & set(harness.CUTLASS_CASES) + with pytest.raises(ValueError, match="Select T4"): + harness.matrix_jobs(["B200"], harness.CUTLASS_CASES, "T4", [], 1) + + +def test_cutlass_fixture_has_real_gemm_and_tail_zero_stream_checks(harness): + on = harness.make_config("T4", "cutlass") + off = harness.make_config("T4", "cutlass-disabled") + assert on == off + assert on["arch"] == "75" + source = on["sources"]["submission.py"] + assert "cutlass::gemm::device::Gemm" in source + assert "c10::cuda::getCurrentCUDAStream" in source + assert "torch.cuda.stream(stream)" in source + tree = ast.parse(source) + call = next( + node + for node in ast.walk(tree) + if isinstance(node, ast.Call) + and isinstance(node.func, ast.Name) + and node.func.id == "load_inline" + ) + kwargs = {keyword.arg: keyword.value for keyword in call.keywords} + assert ast.literal_eval(kwargs["extra_cuda_cflags"]) == ["-O3"] + assert ast.literal_eval(kwargs["no_implicit_headers"]) is False + assert len(on["tests"]) == 9 + assert any(test["m"] == 129 and test["n"] == 136 for test in on["tests"]) + assert sum(test.get("zero", 0) for test in on["tests"]) == 4 + assert all(test["nondefault"] == 1 for test in on["tests"]) + assert on["benchmarks"][0]["nondefault"] == 0 + assert "a.double() @ b.double().T" in on["sources"]["reference.py"] + for name, contents in on["sources"].items(): + ast.parse(contents, filename=name) + + +def test_explicit_arch_profile_sets_architecture_and_exact_extra_flags(harness, monkeypatch): + profile, flags = "sm100a-test", ["-O3", "--use_fast_math", "-arch=sm_100a"] + catalog = {profile: ("100a", flags)} + monkeypatch.setitem(harness.make_config.__globals__, "profile_catalog", lambda: catalog) + jobs = harness.matrix_jobs(["B200"], harness.BASELINE_CASES, "T4", [profile], 1) + assert {job[0] for job in jobs} == {"B200"} + on = harness.make_config("B200", "default", profile) + off = harness.make_config("B200", "disabled", profile) + assert on == off + assert on["arch"] == "100a" + tree = ast.parse(on["sources"]["submission.py"]) + calls = [ + node + for node in ast.walk(tree) + if isinstance(node, ast.Call) + and isinstance(node.func, ast.Name) + and node.func.id == "load_inline" + ] + kwargs = {keyword.arg: keyword.value for keyword in calls[0].keywords} + assert ast.literal_eval(kwargs["extra_cuda_cflags"]) == flags + + +@pytest.mark.parametrize("profile", ["sm100-default", "-O3; arbitrary-command"]) +def test_profile_selection_rejects_unknown_or_incompatible_architecture(harness, profile): + with pytest.raises(ValueError): + harness.matrix_jobs(["T4"], harness.BASELINE_CASES, "T4", [profile], 1) + + +def test_comparison_requires_paired_successful_observations(harness): + def row(case, seconds, passed=True): + return { + "gpu": "T4", + "profile": "sm75-default", + "case": case, + "passed": passed, + "cpu": {"packet": {"info": {"duration": seconds}}}, + } + + rows = [row("default", 10), row("default", 20), row("disabled", 25), row("disabled", 35)] + result = harness.comparisons(rows)[0] + assert result["median_pch_on_seconds"] == 15 + assert result["median_pch_off_seconds"] == 30 + assert result["speedup"] == 2 + assert result["reduction_percent"] == 50 + assert result["cpu_compile_seconds"]["default"] == [10, 20] + assert harness.comparisons(rows[:-1]) == [] + assert harness.comparisons(rows + [row("default", 100, False)]) == [] + + # A vectoradd profile run and the CUTLASS fixture must remain separate even + # when both use the same generic framework-header profile and GPU. + cutlass = [row("cutlass", 40), row("cutlass-disabled", 50)] + results = harness.comparisons(rows + cutlass) + assert {item["workload"]: item["speedup"] for item in results} == { + "vectoradd": 2, + "cutlass": 1.25, + } + + +@pytest.fixture +def successful_pair(): + toolchain = {"generation": "same-generation", "image_id": "same-image"} + event = {"status": "hit", "frontends": ["cicc", "cudafe"], "profile": "sm75-default"} + run = {"success": True, "passed": True, "result": {"check": "pass", "benchmark.0.mean": 1}} + return ( + { + "packet": {"info": {"status": "compiled", "artifacts": 1}, "artifacts": b"artifact"}, + "toolchain": toolchain, + "events": [event], + }, + { + "toolchain": toolchain, + "events": [], + "result": { + "success": True, + "error": "", + "runs": {name: {"run": run} for name in ("test", "benchmark", "leaderboard")}, + "cpu_compile": {"status": "reused", "reused": 6, "fallbacks": 0}, + }, + }, + ) + + +def test_success_requires_matching_profile_and_no_gpu_compilation(harness, successful_pair): + cpu, gpu = successful_pair + harness.verify_result("default", cpu, gpu, "sm75-default") + with pytest.raises(AssertionError): + harness.verify_result("default", cpu, gpu, "sm75-O3") + gpu["events"] = [{"status": "bypass"}] + with pytest.raises(AssertionError): + harness.verify_result("default", cpu, gpu, "sm75-default") + + +def test_changed_header_requires_rejection_and_gpu_pch_hit(harness, successful_pair): + cpu, gpu = copy.deepcopy(successful_pair) + gpu["events"] = cpu["events"] + info = { + "status": "fallback", + "reused": 0, + "fallbacks": 2, + "reason": "Build dependency changed: /tmp/user.h", + } + gpu["result"]["cpu_compile"] = info + harness.verify_result("changed-header", cpu, gpu, "sm75-default") + info["reused"] = 1 + with pytest.raises(AssertionError): + harness.verify_result("changed-header", cpu, gpu, "sm75-default") diff --git a/tests/test_nvcc_toolchain.py b/tests/test_nvcc_toolchain.py new file mode 100644 index 00000000..195c4531 --- /dev/null +++ b/tests/test_nvcc_toolchain.py @@ -0,0 +1,241 @@ +"""Private compiler installation and cache-generation release gates, without CUDA.""" + +import builtins +import hashlib +import importlib.util +import json +import subprocess +from pathlib import Path +from types import SimpleNamespace + +import pytest + +SPEC = importlib.util.spec_from_file_location( + "kernelbot_nvcc_toolchain", Path(__file__).parents[1] / "src/runners/nvcc_toolchain.py" +) +toolchain = importlib.util.module_from_spec(SPEC) +SPEC.loader.exec_module(toolchain) + + +@pytest.fixture +def installation(tmp_path, monkeypatch): # noqa: C901 + roots = { + "/nvcc-toolchain-input": tmp_path / "input", + "/usr/local/cuda": tmp_path / "cuda", + "/opt/kernelbot-pch": tmp_path / "pch", + } + source = roots["/nvcc-toolchain-input"] / toolchain.RELEASE + cuda, artifacts = roots["/usr/local/cuda"], roots["/opt/kernelbot-pch"] + targets = { + "nvcc": "bin/nvcc", + "cudafe++": "bin/cudafe++", + "nvvm/bin/cicc": "nvvm/bin/cicc", + "nvvm-next/bin/cicc": "nvvm-next/bin/cicc", + } + + def write(path, data): + path.parent.mkdir(parents=True, exist_ok=True) + path.write_bytes(data) + + hashes = {} + for name, target in targets.items(): + data = f"pinned private compiler: {name}".encode() + write(source / name, data) + hashes[name] = hashlib.sha256(data).hexdigest() + write(cuda / target, f"original toolkit: {name}".encode()) + write(cuda / "bin/nvcc.profile", b"CICC_PATH = $(TOP)/nvvm/bin\nOTHER = keep\n") + for name in ("bin/ptxas", "bin/nvlink", "bin/fatbinary", "nvvm/libdevice/libdevice.10.bc"): + write(cuda / name, name.encode()) + for name in ("fingerprint", "nvcc.py", "nvcc_common.py", "nvcc_warm.py"): + write(artifacts / name, name.encode()) + + # The image-build function imports pathlib internally. Redirect only that + # import; pytest and the standard library keep their ordinary Path objects. + original_import = builtins.__import__ + + def sandbox_import(name, globals=None, locals=None, fromlist=(), level=0): + if name == "pathlib" and globals and globals.get("__name__") == toolchain.__name__: + return SimpleNamespace(Path=lambda path: roots.get(str(path), Path(path))) + return original_import(name, globals, locals, fromlist, level) + + monkeypatch.setattr(builtins, "__import__", sandbox_import) + state = SimpleNamespace( + source=source, + cuda=cuda, + artifacts=artifacts, + targets=targets, + hashes=hashes, + calls=[], + version="Cuda compilation tools, release 13.4, V13.4.99\n", + backend="nvvm-next/bin/cicc", + packages=( + "cuda-toolkit-13-4\t13.4.3-1\tinstall ok installed\n" + "cuda-nvcc-13-4\t13.4.92-1\tinstall ok installed\n" + "libcublas-13-4\t13.4.0.1-1\tinstall ok installed\n" + "removed-13-4\t1.0-1\tdeinstall ok config-files\n" + ), + fail_arch=None, + ) + + def output(command, **kwargs): + state.calls.append(command) + if command[0] == "dpkg-query": + return state.packages + return state.version if "--version" in command else state.backend + + def run(command, **kwargs): + state.calls.append(command) + if state.fail_arch and f"-arch=sm_{state.fail_arch}" in command: + raise subprocess.CalledProcessError(1, command) + Path(command[command.index("-o") + 1]).write_bytes(b"compiled object") + return subprocess.CompletedProcess(command, 0) + + monkeypatch.setattr(subprocess, "check_output", output) + monkeypatch.setattr(subprocess, "run", run) + state.install = lambda: toolchain.install_nvcc_toolchain(toolchain.RELEASE, hashes) + return state + + +def test_installs_verified_compilers_and_publishes_reproducible_generation(installation): + state = installation + state.install() + for name, target in state.targets.items(): + installed = state.cuda / target + assert installed.read_bytes() == (state.source / name).read_bytes() + assert installed.stat().st_mode & 0o777 == 0o755 + profile = (state.cuda / "bin/nvcc.profile").read_text() + assert "CICC_PATH = $(TOP)/nvvm-next/bin" in profile + assert "OTHER = keep" in profile + + manifest = json.loads((state.artifacts / "nvcc-toolchain.json").read_text()) + generation = manifest.pop("fingerprint") + assert manifest["smoke_architectures"] == ["75", "80", "89", "90a", "100", "100a"] + compiled = [call[1] for call in state.calls if "-c" in call and "--dryrun" not in call] + assert compiled == [f"-arch=sm_{arch}" for arch in manifest["smoke_architectures"]] + assert manifest["headers_fingerprint"] == "fingerprint" + assert manifest["cuda_packages"] == { + "cuda-toolkit-13-4": "13.4.3-1", + "cuda-nvcc-13-4": "13.4.92-1", + "libcublas-13-4": "13.4.0.1-1", + } + assert generation == hashlib.sha256(json.dumps(manifest, sort_keys=True).encode()).hexdigest() + assert (state.artifacts / "nvcc-fingerprint").read_text().strip() == generation + state.install() + assert (state.artifacts / "nvcc-fingerprint").read_text().strip() == generation + + +@pytest.mark.parametrize("damaged", ["nvcc", "cudafe++", "nvvm/bin/cicc", "nvvm-next/bin/cicc"]) +def test_checksum_failure_changes_no_installed_binary(installation, damaged): + state = installation + original = {target: (state.cuda / target).read_bytes() for target in state.targets.values()} + (state.source / damaged).write_bytes(b"unverified compiler") + with pytest.raises(ValueError, match="checksum mismatch"): + state.install() + assert original == { + target: (state.cuda / target).read_bytes() for target in state.targets.values() + } + assert not state.calls + assert not (state.artifacts / "nvcc-fingerprint").exists() + + +@pytest.mark.parametrize( + "missing", ["nvcc", "cudafe++", "nvvm/bin/cicc", "nvvm-next/bin/cicc", None] +) +def test_checksum_manifest_must_cover_exactly_installed_binaries(installation, missing): + state = installation + original = {target: (state.cuda / target).read_bytes() for target in state.targets.values()} + if missing: + del state.hashes[missing] + else: + state.hashes["unrecognized-compiler"] = "a" * 64 + with pytest.raises(ValueError, match="exactly the four NVIDIA compiler binaries"): + state.install() + assert original == { + target: (state.cuda / target).read_bytes() for target in state.targets.values() + } + assert not state.calls + assert not (state.artifacts / "nvcc-fingerprint").exists() + + +@pytest.mark.parametrize("profile", ["OTHER = keep\n", "CICC_PATH = one\nCICC_PATH = two\n"]) +def test_ambiguous_backend_profile_blocks_release(installation, profile): + state = installation + (state.cuda / "bin/nvcc.profile").write_text(profile) + with pytest.raises(ValueError, match="exactly one CICC_PATH"): + state.install() + assert not state.calls + assert not (state.artifacts / "nvcc-fingerprint").exists() + + +@pytest.mark.parametrize( + ("field", "value", "message"), + [ + ( + "version", + "Cuda compilation tools, release 13.4, V13.4.92\n", + "Unexpected NVIDIA compiler", + ), + ("backend", "nvvm/bin/cicc", "Wrong CUDA compiler backend"), + ], +) +def test_version_and_backend_mismatch_block_release(installation, field, value, message): + state = installation + setattr(state, field, value) + with pytest.raises(ValueError, match=message): + state.install() + assert not (state.artifacts / "nvcc-toolchain.json").exists() + assert not (state.artifacts / "nvcc-fingerprint").exists() + + +def test_failed_architecture_smoke_blocks_release(installation): + state = installation + state.fail_arch = "100a" + with pytest.raises(subprocess.CalledProcessError): + state.install() + assert not (state.artifacts / "nvcc-toolchain.json").exists() + assert not (state.artifacts / "nvcc-fingerprint").exists() + + +def test_supporting_library_upgrade_invalidates_cache_generation(installation): + state = installation + state.install() + before = (state.artifacts / "nvcc-fingerprint").read_text() + state.packages = state.packages.replace( + "libcublas-13-4\t13.4.0.1-1", "libcublas-13-4\t13.4.0.2-1" + ) + state.install() + assert (state.artifacts / "nvcc-fingerprint").read_text() != before + + +@pytest.mark.parametrize("packages", ["", "removed-13-4\t1.0-1\tdeinstall ok config-files\n"]) +def test_missing_installed_package_inventory_blocks_release(installation, packages): + state = installation + state.packages = packages + with pytest.raises(ValueError, match="No installed CUDA 13.4 packages"): + state.install() + assert not (state.artifacts / "nvcc-fingerprint").exists() + + +@pytest.mark.parametrize( + ("root", "name"), + [ + ("artifacts", "fingerprint"), + ("artifacts", "nvcc.py"), + ("artifacts", "nvcc_common.py"), + ("artifacts", "nvcc_warm.py"), + ("cuda", "bin/ptxas"), + ("cuda", "bin/nvlink"), + ("cuda", "bin/fatbinary"), + ("cuda", "nvvm/libdevice/libdevice.10.bc"), + ], +) +def test_related_toolchain_and_wrapper_changes_invalidate_cache_generation( + installation, root, name +): + state = installation + state.install() + before = (state.artifacts / "nvcc-fingerprint").read_text() + path = getattr(state, root) / name + path.write_bytes(path.read_bytes() + b" changed") + state.install() + assert (state.artifacts / "nvcc-fingerprint").read_text() != before diff --git a/tests/test_python_precompile.py b/tests/test_python_precompile.py index 1908f468..2ec65609 100644 --- a/tests/test_python_precompile.py +++ b/tests/test_python_precompile.py @@ -1,4 +1,5 @@ import asyncio +import hashlib import json import os import subprocess @@ -9,9 +10,15 @@ import pytest from libkernelbot.consts import get_gpu_by_name -from libkernelbot.inline_artifacts import _check_dependencies, _dependencies +from libkernelbot.inline_artifacts import ( + _check_dependencies, + _dependencies, + _dependency_digest, + nvcc_fingerprint, +) from libkernelbot.launchers.modal import ModalLauncher from libkernelbot.python_precompile import ( + build_identity, compile_python, cuda_arch, pack_artifacts, @@ -133,7 +140,7 @@ def test_cpu_import_failures_return_fallback(config, monkeypatch, error): assert packet["artifacts"] == b"" -@pytest.mark.parametrize("changed", ["sources", "image", "version"]) +@pytest.mark.parametrize("changed", ["sources", "image", "version", "nvcc_toolchain"]) def test_gpu_rejects_stale_build_without_failing_submission(config, result, changed): packet = { "identity": {"image": "image"}, @@ -226,3 +233,123 @@ def test_missing_dependency_information_cannot_be_reused(tmp_path, monkeypatch): monkeypatch.setenv("KERNELBOT_INLINE_WORKDIR", str(tmp_path)) with pytest.raises(RuntimeError, match="no compiler dependency"): _check_dependencies(None, tmp_path) + + +def test_cpu_packet_identity_tracks_experimental_compiler(config, tmp_path, monkeypatch): + fingerprint = tmp_path / "nvcc-fingerprint" + monkeypatch.setattr("libkernelbot.inline_artifacts.NVCC_FINGERPRINT", fingerprint) + with patch("libkernelbot.python_precompile.Path.read_text", return_value="host headers"): + original = build_identity(config) + assert original["nvcc_toolchain"] is None + fingerprint.write_bytes(b"first compiler and wrapper digest\n") + first = build_identity(config) + fingerprint.write_bytes(b"second compiler and wrapper digest\n") + second = build_identity(config) + assert first["image"] == second["image"] == original["image"] + assert first["sources"] == second["sources"] == original["sources"] + assert len({first["nvcc_toolchain"], second["nvcc_toolchain"], None}) == 3 + + +def test_empty_nvcc_fingerprint_cannot_authorize_reuse(tmp_path, monkeypatch): + fingerprint = tmp_path / "nvcc-fingerprint" + fingerprint.write_bytes(b"\n") + monkeypatch.setattr("libkernelbot.inline_artifacts.NVCC_FINGERPRINT", fingerprint) + with pytest.raises(RuntimeError, match="fingerprint is empty"): + nvcc_fingerprint() + + +@pytest.fixture +def trusted_pch(tmp_path, monkeypatch): + generation, key = "a" * 64, "b" * 64 + root = tmp_path / "trusted-cache" + entry = root / generation / key + (entry / "pch").mkdir(parents=True) + header = entry / "gpumode.cuh" + header.write_text("#include \n#pragma hdrstop\n") + files = [] + for frontend in ("cicc", "cudafe"): + name = f"pch/{frontend}_123456.pch" + (entry / name).write_bytes(b"compiler cache contents") + files.append({"path": name, "bytes": (entry / name).stat().st_size}) + manifest = { + "version": 1, + "generation": generation, + "key": key, + "profile": "sm75-default", + "flags": ["-std=c++20"], + "environment": {}, + "header_sha256": hashlib.sha256(header.read_bytes()).hexdigest(), + "files": files, + } + (entry / "manifest.json").write_text(json.dumps(manifest)) + fingerprint = tmp_path / "nvcc-fingerprint" + fingerprint.write_text(generation + "\n") + monkeypatch.setattr("libkernelbot.inline_artifacts.NVCC_FINGERPRINT", fingerprint) + monkeypatch.setattr("libkernelbot.inline_artifacts.NVCC_PCH_ROOT", root) + return entry, manifest, fingerprint + + +def test_trusted_pch_dependencies_do_not_read_cache_bodies(trusted_pch, tmp_path, monkeypatch): + entry, _, _ = trusted_pch + original_open = Path.open + + def small_files_only(path, *args, **kwargs): + assert path.suffix != ".pch", "Artifact replay must not read multi-GB PCH bodies" + return original_open(path, *args, **kwargs) + + monkeypatch.setattr(Path, "open", small_files_only) + cpu, gpu = tmp_path / "cpu", tmp_path / "gpu" + build = cpu / "build" + build.mkdir(parents=True) + gpu.mkdir() + for work in (cpu, gpu): + (work / "user.h").write_text("#define VALUE 1\n") + pch = entry / "pch/cicc_123456.pch" + output = f"cuda.o: #deps 2\n {pch}\n {cpu / 'user.h'}\n" + with patch( + "libkernelbot.inline_artifacts.subprocess.run", return_value=SimpleNamespace(stdout=output) + ): + dependencies = _dependencies(build, cpu) + assert dependencies[str(pch)].startswith("nvcc-pch-v1:") + assert len(dependencies) == 2 + _check_dependencies(dependencies, gpu) + (gpu / "user.h").write_text("#define VALUE 2\n") + with pytest.raises(RuntimeError, match="Build dependency changed"): + _check_dependencies(dependencies, gpu) + + +@pytest.mark.parametrize("change", ["manifest", "header", "generation", "size", "missing"]) +def test_trusted_pch_dependency_changes_reject_replay(trusted_pch, tmp_path, change): + entry, manifest, fingerprint = trusted_pch + path = entry / "pch/cicc_123456.pch" + dependencies = {str(path): _dependency_digest(path)} + if change == "manifest": + manifest["flags"] = ["-std=c++20", "-O3"] + (entry / "manifest.json").write_text(json.dumps(manifest)) + elif change == "header": + (entry / "gpumode.cuh").write_text("#include \n") + elif change == "generation": + fingerprint.write_text("c" * 64 + "\n") + elif change == "size": + path.write_bytes(b"truncated") + else: + (entry / "manifest.json").unlink() + with pytest.raises((RuntimeError, FileNotFoundError)): + _check_dependencies(dependencies, tmp_path) + + +def test_untrusted_pch_is_not_skipped_or_read(tmp_path, monkeypatch): + path = tmp_path / "user.pch" + path.write_bytes(b"not trusted") + monkeypatch.setenv("KERNELBOT_NVCC_PCH_ROOT", str(tmp_path)) + with patch.object(Path, "open", side_effect=AssertionError("Must not read PCH contents")): + with pytest.raises(RuntimeError, match="outside the trusted"): + _dependency_digest(path) + + +def test_pch_file_must_be_named_by_the_complete_ready_manifest(trusted_pch): + entry, _, _ = trusted_pch + path = entry / "pch/unknown.pch" + path.write_bytes(b"unlisted cache") + with pytest.raises(RuntimeError, match="not covered"): + _dependency_digest(path) From a5f078686035620527a41ae0cdc59b0dd8c9816b Mon Sep 17 00:00:00 2001 From: Mark Saroufim Date: Sun, 15 Nov 2026 14:08:59 -0800 Subject: [PATCH 2/4] Enable NVCC PCH by default and keep benchmark evidence out of the patch --- .github/workflows/modal-deploy.yml | 6 +- docs/benchmarks/nvcc-pch-profiles.json | 1461 ----- docs/benchmarks/nvcc-pch.json | 6957 ------------------------ docs/modal-pch-cache.md | 35 +- docs/nvcc-pch-benchmark.md | 214 - docs/nvcc-pch-cache.md | 205 - scripts/nvcc_pch_benchmark.py | 656 --- scripts/validate_nvcc_pch.py | 612 --- src/runners/modal_runner.py | 4 +- src/runners/nvcc_toolchain.py | 2 +- tests/test_nvcc_pch.py | 395 +- tests/test_nvcc_release_validation.py | 212 - tests/test_nvcc_toolchain.py | 262 +- 13 files changed, 239 insertions(+), 10782 deletions(-) delete mode 100644 docs/benchmarks/nvcc-pch-profiles.json delete mode 100644 docs/benchmarks/nvcc-pch.json delete mode 100644 docs/nvcc-pch-benchmark.md delete mode 100644 docs/nvcc-pch-cache.md delete mode 100644 scripts/nvcc_pch_benchmark.py delete mode 100644 scripts/validate_nvcc_pch.py delete mode 100644 tests/test_nvcc_release_validation.py diff --git a/.github/workflows/modal-deploy.yml b/.github/workflows/modal-deploy.yml index 3851790c..821c4dd1 100644 --- a/.github/workflows/modal-deploy.yml +++ b/.github/workflows/modal-deploy.yml @@ -14,9 +14,9 @@ jobs: env: MODAL_TOKEN_ID: ${{ secrets.MODAL_TOKEN_ID }} MODAL_TOKEN_SECRET: ${{ secrets.MODAL_TOKEN_SECRET }} - # Enable only after validating this toolchain in the dev leaderboard. - KERNELBOT_NVCC_TOOLCHAIN: ${{ vars.KERNELBOT_NVCC_TOOLCHAIN || '0' }} - KERNELBOT_NVCC_PCH: ${{ vars.KERNELBOT_NVCC_PCH || '0' }} + # Set to 0 for rollback; ordinary submissions use the shared cache automatically. + KERNELBOT_NVCC_TOOLCHAIN: ${{ vars.KERNELBOT_NVCC_TOOLCHAIN || '1' }} + KERNELBOT_NVCC_PCH: ${{ vars.KERNELBOT_NVCC_PCH || '1' }} KERNELBOT_NVCC_PCH_PROFILES: ${{ vars.KERNELBOT_NVCC_PCH_PROFILES }} KERNELBOT_NVCC_TOOLCHAIN_VOLUME: ${{ vars.KERNELBOT_NVCC_TOOLCHAIN_VOLUME || 'kernelbot-nvcc-toolchains' }} KERNELBOT_NVCC_PCH_VOLUME: ${{ vars.KERNELBOT_NVCC_PCH_VOLUME || 'kernelbot-nvcc-pch' }} diff --git a/docs/benchmarks/nvcc-pch-profiles.json b/docs/benchmarks/nvcc-pch-profiles.json deleted file mode 100644 index fc7188dc..00000000 --- a/docs/benchmarks/nvcc-pch-profiles.json +++ /dev/null @@ -1,1461 +0,0 @@ -{ - "scope": "2026-11-13 private-diff snapshot only (expected 366803 unique submissions); public baseline excluded. Static candidate inventory of calls with explicit cuda_sources keyword; submission code was never executed. Counts do not represent actual NVCC argv, guaranteed cache hits, or all KernelBot submissions.", - "method": { - "flag_order": "preserved exactly", - "name_resolution": "single unconditional assignment/import in lexical scope before call; ambiguous, reassigned, mutated, aliased-to-mutated, and escaped variables unresolved", - "eligibility": "proven torch.utils.cpp_extension.load_inline origin; no kwargs unpacking; resolved nonempty CUDA source; implicit headers enabled; no custom include paths; flags entirely in listed initial allowlist", - "defaults": "missing or literal None extra_cuda_cflags means empty tuple; **kwargs never assumed default", - "runner": "known NVIDIA separated from missing or unrecognized runner; explicit AMD/CPU excluded from analysis", - "limitations": [ - "Conservative static analysis is not Python execution or a proof of runtime eligibility.", - "Only explicit cuda_sources keyword calls are inventoried; positional CUDA arguments are outside scope.", - "Dynamic sources, flags, monkeypatches, environment changes, and runtime branches can change executed behavior.", - "Top-N candidate coverage uses per-submission set union; row frequencies must not be summed as unique coverage.", - "Custom include paths, arbitrary backend/compiler flags, response files, and unknown flags are outside the initial warm-catalog allowlist. Fixed recognized NVIDIA architecture tokens and literal relocatable-device-code=false are included.", - "Source prefilter and analysis failures are reported rather than treated as default flags." - ], - "initial_allowlist": [ - "--expt-extended-lambda", - "--expt-relaxed-constexpr", - "--extended-lambda", - "--generate-code=arch=compute_100,code=sm_100", - "--generate-code=arch=compute_100a,code=sm_100a", - "--generate-code=arch=compute_103,code=sm_103", - "--generate-code=arch=compute_103a,code=sm_103a", - "--generate-code=arch=compute_120,code=sm_120", - "--generate-code=arch=compute_120a,code=sm_120a", - "--generate-code=arch=compute_75,code=sm_75", - "--generate-code=arch=compute_80,code=sm_80", - "--generate-code=arch=compute_86,code=sm_86", - "--generate-code=arch=compute_89,code=sm_89", - "--generate-code=arch=compute_90,code=sm_90", - "--generate-code=arch=compute_90a,code=sm_90a", - "--generate-line-info", - "--gpu-architecture=sm_100", - "--gpu-architecture=sm_100a", - "--gpu-architecture=sm_103", - "--gpu-architecture=sm_103a", - "--gpu-architecture=sm_120", - "--gpu-architecture=sm_120a", - "--gpu-architecture=sm_75", - "--gpu-architecture=sm_80", - "--gpu-architecture=sm_86", - "--gpu-architecture=sm_89", - "--gpu-architecture=sm_90", - "--gpu-architecture=sm_90a", - "--relocatable-device-code=false", - "--std=c++17", - "--std=c++20", - "--use_fast_math", - "-O0", - "-O1", - "-O2", - "-O3", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF_OPERATORS__", - "-arch=sm_100", - "-arch=sm_100a", - "-arch=sm_103", - "-arch=sm_103a", - "-arch=sm_120", - "-arch=sm_120a", - "-arch=sm_75", - "-arch=sm_80", - "-arch=sm_86", - "-arch=sm_89", - "-arch=sm_90", - "-arch=sm_90a", - "-gencode=arch=compute_100,code=sm_100", - "-gencode=arch=compute_100a,code=sm_100a", - "-gencode=arch=compute_103,code=sm_103", - "-gencode=arch=compute_103a,code=sm_103a", - "-gencode=arch=compute_120,code=sm_120", - "-gencode=arch=compute_120a,code=sm_120a", - "-gencode=arch=compute_75,code=sm_75", - "-gencode=arch=compute_80,code=sm_80", - "-gencode=arch=compute_86,code=sm_86", - "-gencode=arch=compute_89,code=sm_89", - "-gencode=arch=compute_90,code=sm_90", - "-gencode=arch=compute_90a,code=sm_90a", - "-lineinfo", - "-rdc=false", - "-std=c++17", - "-std=c++20" - ] - }, - "counts": { - "rows": 367118, - "unique_submissions": 366803, - "duplicate_submission_rows": 315 - }, - "cohorts": { - "unknown": { - "all_submissions": 242065, - "source_prefilter_candidates": 118885, - "static_cuda_load_inline_submissions": 112520, - "static_cuda_load_inline_calls": 133869, - "statically_resolved_flag_submissions": 70311, - "eligible_candidate_calls": 33082, - "eligible_candidate_submissions": 25996, - "parse_or_analysis_failure": 5322 - }, - "nvidia": { - "all_submissions": 124736, - "source_prefilter_candidates": 70854, - "static_cuda_load_inline_submissions": 70461, - "static_cuda_load_inline_calls": 74676, - "statically_resolved_flag_submissions": 64735, - "eligible_candidate_calls": 20717, - "eligible_candidate_submissions": 20411, - "parse_or_analysis_failure": 16 - }, - "other": { - "all_submissions": 2 - }, - "combined_analyzed": { - "all_submissions": 366801, - "source_prefilter_candidates": 189739, - "static_cuda_load_inline_submissions": 182981, - "static_cuda_load_inline_calls": 208545, - "statically_resolved_flag_submissions": 135046, - "eligible_candidate_calls": 53799, - "eligible_candidate_submissions": 46407, - "parse_or_analysis_failure": 5338 - } - }, - "coverage_warning": "Flag-tuple candidate coverage only. Exact framework prefix, generated architecture flags, include order, compiler generation, and environment must still match at runtime; these figures are not measured PCH hit rates. Known NVIDIA and unknown runner cohorts remain separate.", - "selected_default_profile_count": 18, - "selected_distinct_ordered_flag_tuples": 6, - "selection": "Six most frequent eligible ordered flag tuples in the known-NVIDIA runner cohort. Three neutral tuples are warmed across five runner architectures; the three explicit-architecture tuples each add one profile.", - "selected_default_profiles": [ - { - "name": "sm75-default", - "architecture": "75", - "extra_cuda_cflags": [] - }, - { - "name": "sm75-O3", - "architecture": "75", - "extra_cuda_cflags": [ - "-O3" - ] - }, - { - "name": "sm75-O3-fast-math", - "architecture": "75", - "extra_cuda_cflags": [ - "-O3", - "--use_fast_math" - ] - }, - { - "name": "sm80-default", - "architecture": "80", - "extra_cuda_cflags": [] - }, - { - "name": "sm80-O3", - "architecture": "80", - "extra_cuda_cflags": [ - "-O3" - ] - }, - { - "name": "sm80-O3-fast-math", - "architecture": "80", - "extra_cuda_cflags": [ - "-O3", - "--use_fast_math" - ] - }, - { - "name": "sm89-default", - "architecture": "89", - "extra_cuda_cflags": [] - }, - { - "name": "sm89-O3", - "architecture": "89", - "extra_cuda_cflags": [ - "-O3" - ] - }, - { - "name": "sm89-O3-fast-math", - "architecture": "89", - "extra_cuda_cflags": [ - "-O3", - "--use_fast_math" - ] - }, - { - "name": "sm90a-default", - "architecture": "90a", - "extra_cuda_cflags": [] - }, - { - "name": "sm90a-O3", - "architecture": "90a", - "extra_cuda_cflags": [ - "-O3" - ] - }, - { - "name": "sm90a-O3-fast-math", - "architecture": "90a", - "extra_cuda_cflags": [ - "-O3", - "--use_fast_math" - ] - }, - { - "name": "sm100-default", - "architecture": "100", - "extra_cuda_cflags": [] - }, - { - "name": "sm100-O3", - "architecture": "100", - "extra_cuda_cflags": [ - "-O3" - ] - }, - { - "name": "sm100-O3-fast-math", - "architecture": "100", - "extra_cuda_cflags": [ - "-O3", - "--use_fast_math" - ] - }, - { - "name": "sm100a-O3-fast-math-arch", - "architecture": "100a", - "extra_cuda_cflags": [ - "-O3", - "--use_fast_math", - "-arch=sm_100a" - ] - }, - { - "name": "sm100-O3-arch-fast-math", - "architecture": "100", - "extra_cuda_cflags": [ - "-O3", - "-arch=sm_100", - "--use_fast_math" - ] - }, - { - "name": "sm80-O3-cxx17-arch-fast-math", - "architecture": "80", - "extra_cuda_cflags": [ - "-O3", - "-std=c++17", - "-arch=sm_80", - "--use_fast_math" - ] - } - ], - "master_header_direct_includes": [ - "torch/types.h", - "cuda.h", - "cuda_runtime.h" - ], - "selected_catalog_flags": [ - { - "flags": [ - "-O3" - ], - "known_nvidia_unique_submissions": 5419, - "unknown_runner_unique_submissions": 10457, - "combined_unique_submissions": 15876 - }, - { - "flags": [ - "-O3", - "--use_fast_math" - ], - "known_nvidia_unique_submissions": 3358, - "unknown_runner_unique_submissions": 6695, - "combined_unique_submissions": 10053 - }, - { - "flags": [], - "known_nvidia_unique_submissions": 2670, - "unknown_runner_unique_submissions": 2556, - "combined_unique_submissions": 5226 - }, - { - "flags": [ - "-O3", - "--use_fast_math", - "-arch=sm_100a" - ], - "known_nvidia_unique_submissions": 2609, - "unknown_runner_unique_submissions": 113, - "combined_unique_submissions": 2722 - }, - { - "flags": [ - "-O3", - "-arch=sm_100", - "--use_fast_math" - ], - "known_nvidia_unique_submissions": 771, - "unknown_runner_unique_submissions": 99, - "combined_unique_submissions": 870 - }, - { - "flags": [ - "-O3", - "-std=c++17", - "-arch=sm_80", - "--use_fast_math" - ], - "known_nvidia_unique_submissions": 628, - "unknown_runner_unique_submissions": 174, - "combined_unique_submissions": 802 - } - ], - "selected_catalog_candidate_coverage": { - "unknown": { - "unique_submissions_with_any_candidate_call_covered": 18283, - "unique_submissions_with_all_observed_calls_eligible_and_covered": 16356, - "static_cuda_load_inline_submissions": 112520, - "eligible_candidate_submissions": 25996, - "fraction_of_static_cuda_load_inline_submissions": 0.1624866690366157 - }, - "nvidia": { - "unique_submissions_with_any_candidate_call_covered": 15421, - "unique_submissions_with_all_observed_calls_eligible_and_covered": 15176, - "static_cuda_load_inline_submissions": 70461, - "eligible_candidate_submissions": 20411, - "fraction_of_static_cuda_load_inline_submissions": 0.21885865940023558 - }, - "combined_analyzed": { - "unique_submissions_with_any_candidate_call_covered": 33704, - "unique_submissions_with_all_observed_calls_eligible_and_covered": 31532, - "static_cuda_load_inline_submissions": 182981, - "eligible_candidate_submissions": 46407, - "fraction_of_static_cuda_load_inline_submissions": 0.18419398735387826 - } - }, - "top_eligible_flags": { - "unknown": [ - { - "flags": [ - "-O3" - ], - "unique_submissions": 10457 - }, - { - "flags": [ - "-O3", - "--use_fast_math" - ], - "unique_submissions": 6695 - }, - { - "flags": [], - "unique_submissions": 2556 - }, - { - "flags": [ - "-O3", - "--use_fast_math", - "-gencode=arch=compute_100a,code=sm_100a", - "--expt-relaxed-constexpr" - ], - "unique_submissions": 770 - }, - { - "flags": [ - "-O3", - "--use_fast_math", - "-lineinfo", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-gencode=arch=compute_100a,code=sm_100a" - ], - "unique_submissions": 593 - }, - { - "flags": [ - "-O2" - ], - "unique_submissions": 491 - }, - { - "flags": [ - "-O2", - "--use_fast_math" - ], - "unique_submissions": 488 - }, - { - "flags": [ - "-O3", - "-gencode=arch=compute_100a,code=sm_100a", - "--use_fast_math", - "--expt-relaxed-constexpr", - "--relocatable-device-code=false" - ], - "unique_submissions": 392 - }, - { - "flags": [ - "-O3", - "--use_fast_math", - "-gencode=arch=compute_100a,code=sm_100a", - "-lineinfo", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__" - ], - "unique_submissions": 310 - }, - { - "flags": [ - "-O3", - "--use_fast_math", - "-gencode=arch=compute_100a,code=sm_100a", - "--expt-relaxed-constexpr", - "--relocatable-device-code=false" - ], - "unique_submissions": 298 - }, - { - "flags": [ - "-O3", - "-lineinfo" - ], - "unique_submissions": 248 - }, - { - "flags": [ - "-O3", - "-gencode=arch=compute_100a,code=sm_100a", - "--use_fast_math", - "--expt-relaxed-constexpr", - "--relocatable-device-code=false", - "-lineinfo" - ], - "unique_submissions": 247 - } - ], - "nvidia": [ - { - "flags": [ - "-O3" - ], - "unique_submissions": 5419 - }, - { - "flags": [ - "-O3", - "--use_fast_math" - ], - "unique_submissions": 3358 - }, - { - "flags": [], - "unique_submissions": 2670 - }, - { - "flags": [ - "-O3", - "--use_fast_math", - "-arch=sm_100a" - ], - "unique_submissions": 2609 - }, - { - "flags": [ - "-O3", - "-arch=sm_100", - "--use_fast_math" - ], - "unique_submissions": 771 - }, - { - "flags": [ - "-O3", - "-std=c++17", - "-arch=sm_80", - "--use_fast_math" - ], - "unique_submissions": 628 - }, - { - "flags": [ - "-O3", - "-gencode=arch=compute_100a,code=sm_100a", - "--use_fast_math", - "--expt-relaxed-constexpr", - "--relocatable-device-code=false", - "-lineinfo" - ], - "unique_submissions": 540 - }, - { - "flags": [ - "-O3", - "--use_fast_math", - "-arch=sm_100a", - "-std=c++17" - ], - "unique_submissions": 434 - }, - { - "flags": [ - "-O3", - "--use_fast_math", - "-lineinfo", - "-gencode=arch=compute_100,code=sm_100", - "-arch=sm_100" - ], - "unique_submissions": 362 - }, - { - "flags": [ - "-O3", - "--use_fast_math", - "-gencode=arch=compute_100a,code=sm_100a" - ], - "unique_submissions": 285 - }, - { - "flags": [ - "-O3", - "--use_fast_math", - "-gencode=arch=compute_100a,code=sm_100a", - "--expt-relaxed-constexpr", - "--relocatable-device-code=false" - ], - "unique_submissions": 282 - }, - { - "flags": [ - "-O3", - "--use_fast_math", - "-lineinfo", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-gencode=arch=compute_100a,code=sm_100a" - ], - "unique_submissions": 234 - } - ], - "combined_analyzed": [ - { - "flags": [ - "-O3" - ], - "unique_submissions": 15876 - }, - { - "flags": [ - "-O3", - "--use_fast_math" - ], - "unique_submissions": 10053 - }, - { - "flags": [], - "unique_submissions": 5226 - }, - { - "flags": [ - "-O3", - "--use_fast_math", - "-arch=sm_100a" - ], - "unique_submissions": 2722 - }, - { - "flags": [ - "-O3", - "-arch=sm_100", - "--use_fast_math" - ], - "unique_submissions": 870 - }, - { - "flags": [ - "-O3", - "--use_fast_math", - "-lineinfo", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-gencode=arch=compute_100a,code=sm_100a" - ], - "unique_submissions": 827 - }, - { - "flags": [ - "-O3", - "--use_fast_math", - "-gencode=arch=compute_100a,code=sm_100a", - "--expt-relaxed-constexpr" - ], - "unique_submissions": 826 - }, - { - "flags": [ - "-O3", - "-std=c++17", - "-arch=sm_80", - "--use_fast_math" - ], - "unique_submissions": 802 - }, - { - "flags": [ - "-O3", - "-gencode=arch=compute_100a,code=sm_100a", - "--use_fast_math", - "--expt-relaxed-constexpr", - "--relocatable-device-code=false", - "-lineinfo" - ], - "unique_submissions": 787 - }, - { - "flags": [ - "-O3", - "--use_fast_math", - "-gencode=arch=compute_100a,code=sm_100a", - "--expt-relaxed-constexpr", - "--relocatable-device-code=false" - ], - "unique_submissions": 580 - }, - { - "flags": [ - "-O2" - ], - "unique_submissions": 561 - }, - { - "flags": [ - "-O2", - "--use_fast_math" - ], - "unique_submissions": 522 - } - ] - }, - "ineligibility_reasons_unique_submissions_overlapping": { - "unknown": { - "custom_include_paths": 900, - "flags_outside_initial_allowlist": 24563, - "cuda_sources_unresolved": 21583, - "unverified_load_inline_origin": 10897, - "cuda_sources_empty": 576, - "flags_unresolved_or_ambiguous": 41236, - "no_implicit_headers_true": 52839, - "include_paths_unresolved": 3437, - "kwargs_unpacking": 1400 - }, - "nvidia": { - "custom_include_paths": 146, - "flags_outside_initial_allowlist": 7091, - "cuda_sources_unresolved": 13229, - "unverified_load_inline_origin": 6229, - "flags_unresolved_or_ambiguous": 5842, - "no_implicit_headers_true": 35586, - "include_paths_unresolved": 625, - "cuda_sources_empty": 88, - "kwargs_unpacking": 6 - }, - "combined_analyzed": { - "custom_include_paths": 1046, - "flags_outside_initial_allowlist": 31654, - "cuda_sources_unresolved": 34812, - "unverified_load_inline_origin": 17126, - "flags_unresolved_or_ambiguous": 47078, - "no_implicit_headers_true": 88425, - "include_paths_unresolved": 4062, - "cuda_sources_empty": 664, - "kwargs_unpacking": 1406 - } - }, - "source_header_count_scope": "Distinct known-NVIDIA static CUDA load_inline submissions with each explicit include in their resolved C++ or CUDA source strings; PyTorch-generated implicit includes are not counted.", - "known_nvidia_framework_headers": [ - { - "header": "cuda_runtime.h", - "unique_submissions": 50015, - "direct_master_include": true, - "cutlass_or_cute_excluded": false - }, - { - "header": "cuda_fp16.h", - "unique_submissions": 38863, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "cuda.h", - "unique_submissions": 35932, - "direct_master_include": true, - "cutlass_or_cute_excluded": false - }, - { - "header": "cudaTypedefs.h", - "unique_submissions": 29600, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "torch/library.h", - "unique_submissions": 29011, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "ATen/core/Tensor.h", - "unique_submissions": 28375, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "torch/extension.h", - "unique_submissions": 24752, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "ATen/cuda/CUDAContext.h", - "unique_submissions": 6201, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "cuda_fp4.h", - "unique_submissions": 3846, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "cuda_fp8.h", - "unique_submissions": 3768, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "cublasLt.h", - "unique_submissions": 3753, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "cublas_v2.h", - "unique_submissions": 3597, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "ATen/ATen.h", - "unique_submissions": 3448, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "cuda_bf16.h", - "unique_submissions": 2550, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "c10/cuda/CUDAGuard.h", - "unique_submissions": 1525, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "ATen/cuda/CUDABlas.h", - "unique_submissions": 1170, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "torch/csrc/autograd/python_variable.h", - "unique_submissions": 933, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "cuda/atomic", - "unique_submissions": 648, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "ATen/cuda/CUDAUtils.h", - "unique_submissions": 491, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "c10/cuda/CUDAException.h", - "unique_submissions": 418, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "cuda/ptx", - "unique_submissions": 320, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "c10/cuda/CUDACachingAllocator.h", - "unique_submissions": 318, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "cuda_pipeline.h", - "unique_submissions": 312, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "torch/torch.h", - "unique_submissions": 277, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "cutlass/cutlass.h", - "unique_submissions": 263, - "direct_master_include": false, - "cutlass_or_cute_excluded": true - }, - { - "header": "cutlass/gemm/kernel/gemm_universal.hpp", - "unique_submissions": 244, - "direct_master_include": false, - "cutlass_or_cute_excluded": true - }, - { - "header": "cutlass/gemm/collective/collective_builder.hpp", - "unique_submissions": 244, - "direct_master_include": false, - "cutlass_or_cute_excluded": true - }, - { - "header": "cutlass/epilogue/collective/collective_builder.hpp", - "unique_submissions": 244, - "direct_master_include": false, - "cutlass_or_cute_excluded": true - }, - { - "header": "cutlass/gemm/device/gemm_universal_adapter.h", - "unique_submissions": 244, - "direct_master_include": false, - "cutlass_or_cute_excluded": true - }, - { - "header": "cutlass/util/packed_stride.hpp", - "unique_submissions": 243, - "direct_master_include": false, - "cutlass_or_cute_excluded": true - }, - { - "header": "cute/tensor.hpp", - "unique_submissions": 232, - "direct_master_include": false, - "cutlass_or_cute_excluded": true - }, - { - "header": "cuda/pipeline", - "unique_submissions": 195, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "cuda/barrier", - "unique_submissions": 159, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "torch/types.h", - "unique_submissions": 159, - "direct_master_include": true, - "cutlass_or_cute_excluded": false - }, - { - "header": "cutlass/gemm/dispatch_policy.hpp", - "unique_submissions": 157, - "direct_master_include": false, - "cutlass_or_cute_excluded": true - }, - { - "header": "ATen/ops/cudnn_convolution.h", - "unique_submissions": 140, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "cutlass/detail/sm100_blockscaled_layout.hpp", - "unique_submissions": 137, - "direct_master_include": false, - "cutlass_or_cute_excluded": true - }, - { - "header": "c10/cuda/CUDAStream.h", - "unique_submissions": 74, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "ATen/cuda/CUDAContextLight.h", - "unique_submissions": 71, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - }, - { - "header": "cuda_runtime_api.h", - "unique_submissions": 64, - "direct_master_include": false, - "cutlass_or_cute_excluded": false - } - ], - "runner_eligible_flags": [ - { - "runner_class": "unknown", - "flags": [ - "-O3" - ], - "unique_submissions": 10457 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "--use_fast_math" - ], - "unique_submissions": 6695 - }, - { - "runner_class": "B200", - "flags": [ - "-O3" - ], - "unique_submissions": 2613 - }, - { - "runner_class": "B200", - "flags": [ - "-O3", - "--use_fast_math", - "-arch=sm_100a" - ], - "unique_submissions": 2608 - }, - { - "runner_class": "unknown", - "flags": [], - "unique_submissions": 2556 - }, - { - "runner_class": "A100", - "flags": [ - "-O3" - ], - "unique_submissions": 1862 - }, - { - "runner_class": "B200", - "flags": [], - "unique_submissions": 1054 - }, - { - "runner_class": "B200", - "flags": [ - "-O3", - "--use_fast_math" - ], - "unique_submissions": 922 - }, - { - "runner_class": "A100", - "flags": [ - "-O3", - "--use_fast_math" - ], - "unique_submissions": 794 - }, - { - "runner_class": "B200", - "flags": [ - "-O3", - "-arch=sm_100", - "--use_fast_math" - ], - "unique_submissions": 771 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "--use_fast_math", - "-gencode=arch=compute_100a,code=sm_100a", - "--expt-relaxed-constexpr" - ], - "unique_submissions": 770 - }, - { - "runner_class": "NVIDIA_OTHER", - "flags": [ - "-O3", - "--use_fast_math" - ], - "unique_submissions": 737 - }, - { - "runner_class": "A100", - "flags": [], - "unique_submissions": 688 - }, - { - "runner_class": "A100", - "flags": [ - "-O3", - "-std=c++17", - "-arch=sm_80", - "--use_fast_math" - ], - "unique_submissions": 628 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "--use_fast_math", - "-lineinfo", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-gencode=arch=compute_100a,code=sm_100a" - ], - "unique_submissions": 593 - }, - { - "runner_class": "H100", - "flags": [], - "unique_submissions": 563 - }, - { - "runner_class": "NVIDIA_OTHER", - "flags": [ - "-O3", - "-gencode=arch=compute_100a,code=sm_100a", - "--use_fast_math", - "--expt-relaxed-constexpr", - "--relocatable-device-code=false", - "-lineinfo" - ], - "unique_submissions": 540 - }, - { - "runner_class": "L4", - "flags": [ - "-O3" - ], - "unique_submissions": 507 - }, - { - "runner_class": "unknown", - "flags": [ - "-O2" - ], - "unique_submissions": 491 - }, - { - "runner_class": "unknown", - "flags": [ - "-O2", - "--use_fast_math" - ], - "unique_submissions": 488 - }, - { - "runner_class": "L4", - "flags": [ - "-O3", - "--use_fast_math" - ], - "unique_submissions": 469 - }, - { - "runner_class": "NVIDIA_OTHER", - "flags": [ - "-O3", - "--use_fast_math", - "-arch=sm_100a", - "-std=c++17" - ], - "unique_submissions": 434 - }, - { - "runner_class": "H100", - "flags": [ - "-O3", - "--use_fast_math" - ], - "unique_submissions": 431 - }, - { - "runner_class": "H100", - "flags": [ - "-O3" - ], - "unique_submissions": 414 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "-gencode=arch=compute_100a,code=sm_100a", - "--use_fast_math", - "--expt-relaxed-constexpr", - "--relocatable-device-code=false" - ], - "unique_submissions": 392 - }, - { - "runner_class": "NVIDIA_OTHER", - "flags": [ - "-O3", - "--use_fast_math", - "-lineinfo", - "-gencode=arch=compute_100,code=sm_100", - "-arch=sm_100" - ], - "unique_submissions": 362 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "--use_fast_math", - "-gencode=arch=compute_100a,code=sm_100a", - "-lineinfo", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__" - ], - "unique_submissions": 310 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "--use_fast_math", - "-gencode=arch=compute_100a,code=sm_100a", - "--expt-relaxed-constexpr", - "--relocatable-device-code=false" - ], - "unique_submissions": 298 - }, - { - "runner_class": "T4", - "flags": [], - "unique_submissions": 286 - }, - { - "runner_class": "B200", - "flags": [ - "-O3", - "--use_fast_math", - "-gencode=arch=compute_100a,code=sm_100a" - ], - "unique_submissions": 285 - }, - { - "runner_class": "B200", - "flags": [ - "-O3", - "--use_fast_math", - "-gencode=arch=compute_100a,code=sm_100a", - "--expt-relaxed-constexpr", - "--relocatable-device-code=false" - ], - "unique_submissions": 265 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "-lineinfo" - ], - "unique_submissions": 248 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "-gencode=arch=compute_100a,code=sm_100a", - "--use_fast_math", - "--expt-relaxed-constexpr", - "--relocatable-device-code=false", - "-lineinfo" - ], - "unique_submissions": 247 - }, - { - "runner_class": "NVIDIA_OTHER", - "flags": [ - "-O3", - "--use_fast_math", - "-lineinfo", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-gencode=arch=compute_100a,code=sm_100a" - ], - "unique_submissions": 234 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "--use_fast_math", - "-lineinfo" - ], - "unique_submissions": 227 - }, - { - "runner_class": "B200", - "flags": [ - "-O3", - "-std=c++17", - "-gencode=arch=compute_100a,code=sm_100a", - "--expt-relaxed-constexpr" - ], - "unique_submissions": 223 - }, - { - "runner_class": "unknown", - "flags": [ - "-O2", - "--use_fast_math", - "-gencode=arch=compute_100a,code=sm_100a" - ], - "unique_submissions": 217 - }, - { - "runner_class": "B200", - "flags": [ - "-O3", - "-lineinfo" - ], - "unique_submissions": 198 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "-std=c++17" - ], - "unique_submissions": 192 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "-gencode=arch=compute_100a,code=sm_100a" - ], - "unique_submissions": 186 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "-arch=sm_100a", - "--use_fast_math" - ], - "unique_submissions": 185 - }, - { - "runner_class": "B200", - "flags": [ - "-O3", - "-arch=sm_100" - ], - "unique_submissions": 179 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "-std=c++17", - "-arch=sm_80", - "--use_fast_math" - ], - "unique_submissions": 174 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "--use_fast_math", - "-std=c++17", - "-gencode=arch=compute_100a,code=sm_100a" - ], - "unique_submissions": 141 - }, - { - "runner_class": "NVIDIA_OTHER", - "flags": [ - "-O3", - "-std=c++17", - "--use_fast_math", - "-lineinfo", - "-arch=sm_100a" - ], - "unique_submissions": 141 - }, - { - "runner_class": "NVIDIA_OTHER", - "flags": [ - "-O3", - "--use_fast_math", - "-arch=sm_100a", - "-std=c++17", - "-lineinfo", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__" - ], - "unique_submissions": 138 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "--use_fast_math", - "-gencode=arch=compute_100a,code=sm_100a", - "--expt-relaxed-constexpr", - "--std=c++20" - ], - "unique_submissions": 133 - }, - { - "runner_class": "A100", - "flags": [ - "-O3", - "-arch=sm_80" - ], - "unique_submissions": 131 - }, - { - "runner_class": "A100", - "flags": [ - "-arch=sm_80" - ], - "unique_submissions": 113 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "--use_fast_math", - "-arch=sm_100a" - ], - "unique_submissions": 113 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "--std=c++17" - ], - "unique_submissions": 113 - }, - { - "runner_class": "unknown", - "flags": [ - "-arch=sm_80" - ], - "unique_submissions": 112 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "--use_fast_math", - "-std=c++17" - ], - "unique_submissions": 109 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "-std=c++17", - "-gencode=arch=compute_100a,code=sm_100a", - "--expt-relaxed-constexpr" - ], - "unique_submissions": 107 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "--use_fast_math", - "-arch=sm_100" - ], - "unique_submissions": 104 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "-arch=sm_100", - "--use_fast_math" - ], - "unique_submissions": 99 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "--use_fast_math", - "-gencode=arch=compute_100a,code=sm_100a" - ], - "unique_submissions": 95 - }, - { - "runner_class": "unknown", - "flags": [ - "-O3", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_100a,code=sm_100a" - ], - "unique_submissions": 93 - }, - { - "runner_class": "NVIDIA_OTHER", - "flags": [ - "-O3", - "--use_fast_math", - "-gencode=arch=compute_100a,code=sm_100a", - "-lineinfo", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__" - ], - "unique_submissions": 93 - }, - { - "runner_class": "unknown", - "flags": [ - "-arch=sm_89", - "-O3", - "--use_fast_math" - ], - "unique_submissions": 86 - } - ] -} diff --git a/docs/benchmarks/nvcc-pch.json b/docs/benchmarks/nvcc-pch.json deleted file mode 100644 index 2dc7192c..00000000 --- a/docs/benchmarks/nvcc-pch.json +++ /dev/null @@ -1,6957 +0,0 @@ -{ - "summary": { - "primary": [ - { - "case": "cutlass", - "mode": "patched_all_pch", - "n": 4, - "median_seconds": 34.026361068499995, - "min_seconds": 28.440589126, - "max_seconds": 43.662042164, - "seconds": [ - 43.662042164, - 39.165971053, - 28.886751083999997, - 28.440589126 - ] - }, - { - "case": "cutlass", - "mode": "patched_host_pch", - "n": 4, - "median_seconds": 45.7583980275, - "min_seconds": 41.732421651, - "max_seconds": 51.061138988, - "seconds": [ - 51.061138988, - 42.159665512000004, - 41.732421651, - 49.357130543 - ] - }, - { - "case": "cutlass", - "mode": "stock_host_pch", - "n": 4, - "median_seconds": 43.724847137, - "min_seconds": 41.497131416, - "max_seconds": 48.826632251999996, - "seconds": [ - 48.826632251999996, - 41.497131416, - 43.52756049, - 43.922133784 - ] - }, - { - "case": "cutlass", - "mode": "stock_no_pch", - "n": 4, - "median_seconds": 48.817288012, - "min_seconds": 42.949985459, - "max_seconds": 54.309780212999996, - "seconds": [ - 54.136435518, - 42.949985459, - 54.309780212999996, - 43.498140506 - ] - }, - { - "case": "simple", - "mode": "patched_all_pch", - "n": 4, - "median_seconds": 24.172627273, - "min_seconds": 23.934274335, - "max_seconds": 29.137589784, - "seconds": [ - 24.183967813, - 24.161286733, - 23.934274335, - 29.137589784 - ] - }, - { - "case": "simple", - "mode": "patched_host_pch", - "n": 4, - "median_seconds": 36.013780205, - "min_seconds": 34.736912296, - "max_seconds": 38.467215558, - "seconds": [ - 34.736912296, - 35.908356786, - 36.119203624, - 38.467215558 - ] - }, - { - "case": "simple", - "mode": "stock_host_pch", - "n": 4, - "median_seconds": 35.3712442095, - "min_seconds": 35.025937535000004, - "max_seconds": 36.431860271, - "seconds": [ - 36.431860271, - 35.256968721999996, - 35.485519697, - 35.025937535000004 - ] - }, - { - "case": "simple", - "mode": "stock_no_pch", - "n": 4, - "median_seconds": 37.2270242755, - "min_seconds": 35.746856564, - "max_seconds": 42.885632301, - "seconds": [ - 36.414967365, - 42.885632301, - 38.039081186, - 35.746856564 - ] - } - ], - "supplemental": [ - { - "case": "cutlass", - "mode": "patched_all_pch", - "n": 2, - "median_seconds": 32.899585346500004, - "min_seconds": 32.528592897, - "max_seconds": 33.270577796, - "seconds": [ - 32.528592897, - 33.270577796 - ] - }, - { - "case": "cutlass", - "mode": "patched_all_pch_prefetch", - "n": 2, - "median_seconds": 35.439373548499994, - "min_seconds": 30.342368112999996, - "max_seconds": 40.536378983999995, - "seconds": [ - 30.342368112999996, - 40.536378983999995 - ] - }, - { - "case": "cutlass", - "mode": "patched_host_pch", - "n": 2, - "median_seconds": 50.212558552000004, - "min_seconds": 49.549131939, - "max_seconds": 50.875985165, - "seconds": [ - 50.875985165, - 49.549131939 - ] - }, - { - "case": "simple", - "mode": "patched_all_pch", - "n": 2, - "median_seconds": 25.238124405, - "min_seconds": 22.962655791, - "max_seconds": 27.513593019, - "seconds": [ - 27.513593019, - 22.962655791 - ] - }, - { - "case": "simple", - "mode": "patched_all_pch_prefetch", - "n": 2, - "median_seconds": 25.7022750095, - "min_seconds": 25.672850840000002, - "max_seconds": 25.731699179, - "seconds": [ - 25.731699179, - 25.672850840000002 - ] - }, - { - "case": "simple", - "mode": "patched_host_pch", - "n": 2, - "median_seconds": 40.63829330600001, - "min_seconds": 37.205667172000005, - "max_seconds": 44.070919440000004, - "seconds": [ - 37.205667172000005, - 44.070919440000004 - ] - } - ], - "production_cuda133": [ - { - "case": "cutlass_gemm_implicit", - "enabled": false, - "n": 4, - "median_seconds": 48.7533205525, - "min_seconds": 46.312597678, - "max_seconds": 54.753927797, - "seconds": [ - 54.753927797, - 46.312597678, - 51.057156396, - 46.449484709 - ] - }, - { - "case": "cutlass_gemm_implicit", - "enabled": true, - "n": 4, - "median_seconds": 45.322798237, - "min_seconds": 43.304864159, - "max_seconds": 55.171705347, - "seconds": [ - 55.171705347, - 46.287869813, - 43.304864159, - 44.357726661 - ] - }, - { - "case": "inline_simple", - "enabled": false, - "n": 4, - "median_seconds": 20.7435791125, - "min_seconds": 19.752127795, - "max_seconds": 21.796966691, - "seconds": [ - 21.796966691, - 19.861022294999998, - 19.752127795, - 21.62613593 - ] - }, - { - "case": "inline_simple", - "enabled": true, - "n": 4, - "median_seconds": 10.365594108499998, - "min_seconds": 10.092071702999998, - "max_seconds": 11.21638811, - "seconds": [ - 10.092071702999998, - 10.122365914999998, - 11.21638811, - 10.608822302 - ] - } - ], - "validated_timed_pch_builds": 18, - "fresh_gpu_containers": 62, - "correctness_checks": 492, - "max_absolute_error": 8.529314072802663e-06 - }, - "validation": { - "timed_pch_proof": [ - { - "container_id": "ta-01M7GG9SJ3GHSRNQV0Z0ACJC3R", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ] - }, - { - "container_id": "ta-01M7GGBPJYFMHYHKCZZFP9G16R", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ] - }, - { - "container_id": "ta-01M7GGJYZRTAM0NXNGM9XQ7FTR", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" - ] - }, - { - "container_id": "ta-01M7GGKGST0M04EKKWTZSZAZMR", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" - ] - }, - { - "container_id": "ta-01M7GGP9SHQZVXA5GZ2ZE3HXKR", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ] - }, - { - "container_id": "ta-01M7GGSSKT9DDF905Z9S6J9SER", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ] - }, - { - "container_id": "ta-01M7GGT09WZH9JV14JHDV0YQVR", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" - ] - }, - { - "container_id": "ta-01M7GGY1T72T9C66SB78QVCPJR", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" - ] - }, - { - "container_id": "ta-01M7GGED50E6R503HVDSM8MGRR", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ] - }, - { - "container_id": "ta-01M7GGED5AZC6SWFBFCVMC7C5R", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ] - }, - { - "container_id": "ta-01M7GGED5AR5589PP71ECYZT5R", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ] - }, - { - "container_id": "ta-01M7GGGJTBKKMAZG8E43P3FYXR", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ] - }, - { - "container_id": "ta-01M7GGH8YZHK6M1B0FWN50S9BR", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" - ] - }, - { - "container_id": "ta-01M7GGJEA1CBKA78JQGA7X6YTR", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" - ] - }, - { - "container_id": "ta-01M7GGKN4QWSM62X7YPQY4B47R", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" - ] - }, - { - "container_id": "ta-01M7GGM20MGZZK7K4NM6A452NR", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" - ] - }, - { - "container_id": "ta-01M7GG5X2J0T5Z7DF200Z9AWXR", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ] - }, - { - "container_id": "ta-01M7GG5X2JWF1E5D985MF82VDR", - "timed_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ] - } - ] - }, - "primary": { - "probe": { - "stock_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patched_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "paths": "/usr/local/cuda/bin/nvcc.profile\n/usr/local/cuda/nvvm/bin/cicc\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "nvcc_profiles": { - "/usr/local/cuda/bin/nvcc.profile": "\nTOP = $(_HERE_)/..\n\nCICC_PATH = $(TOP)/nvvm-next/bin\nNVVMIR_LIBRARY_DIR = $(TOP)/nvvm/libdevice\n\nLD_LIBRARY_PATH += $(TOP)/lib:\nPATH += $(CICC_PATH):$(_HERE_):\n\nINCLUDES += \"-I$(TOP)/$(_TARGET_DIR_)/include\" $(_SPACE_)\nSYSTEM_INCLUDES += \"-isystem\" \"$(TOP)/$(_TARGET_DIR_)/include/cccl\" $(_SPACE_)\n\nLIBRARIES =+ $(_SPACE_) \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)/stubs\" \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)\"\n\nCUDAFE_FLAGS +=\nPTXAS_FLAGS +=\n" - }, - "pch_help": " Specify options directly to library manager.\n\n--pch --pch \n Enable precompiled header (PCH) processing. When specified, each compilation\n searches the PCH directory (see --pch-dir) for a reusable precompiled header,\n and creates a new one when none is found.\n\n--pch-dir --pch-dir \n Specify the directory in which precompiled header files are read from and\n written to. By default this is the current working directory. Ignored unless\n --pch, --create-pch, or --use-pch is enabled.\n\n--pch-verbose --pch-verbose \n Print (to stderr) messages related to precompiled header processing. Ignored\n unless --pch, --create-pch, or --use-pch is enabled.\n\n--create-pch --create-pch \n Manually create a precompiled header (PCH) at the first PCH terminal stop-point\n (i.e., #pragma hdrstop) and save it under the specified name. Intermediate\n stop points (#pragma hdrstop++) are ignored. The directory into which the\n PCH file is written is determined by --pch-dir (or is the current working\n directory if not specified). This option forces a PCH file to be created\n (any existing PCH file with the same name is overwritten). This option is\n mutually exclusive with '--use-pch' and '--pch'.\n\n--use-pch --use-pch \n If a precompiled header (PCH) with the specified name exists in the directory\n determined by '--pch-dir' (or the current working directory if not specified),\n try to use it in compilation.This option is mutually exclusive with '--create-pch'\n and '--pch'.\n\n--ptxas-options ,... (-Xptxas) \n Specify options directly to ptxas, the PTX optimizing assembler.\n\n--nvlink-options ,... (-Xnvlink) \n Specify options directly to nvlink.\n\n--nvprune-options ,... (-Xnvprune) \n Specify options directly to nvprune (used with -prune).\n\n\nMiscellaneous options for guiding the compiler driver.\n======================================================", - "strace": null, - "smoke": [ - { - "arch": "sm_75", - "command": [ - "/usr/local/cuda/bin/nvcc", - "-arch=sm_75", - "-c", - "/tmp/tmpw1rin_68/smoke.cu", - "-o", - "/tmp/tmpw1rin_68/sm_75.o" - ], - "dryrun_returncode": 0, - "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpw1rin_68/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw1rin_68/smoke.cu\" --orig_src_path_name \"/tmp/tmpw1rin_68/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpw1rin_68/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw1rin_68/smoke.cu\" --orig_src_path_name \"/tmp/tmpw1rin_68/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_0000001e_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_0000001e_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_0000001e_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpw1rin_68/sm_75.o\" \n", - "next_backend_selected": true, - "returncode": 0, - "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpw1rin_68/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw1rin_68/smoke.cu\" --orig_src_path_name \"/tmp/tmpw1rin_68/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpw1rin_68/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw1rin_68/smoke.cu\" --orig_src_path_name \"/tmp/tmpw1rin_68/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000021_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000021_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000021_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpw1rin_68/sm_75.o\" \n", - "object_bytes": 10200 - }, - { - "arch": "sm_100", - "command": [ - "/usr/local/cuda/bin/nvcc", - "-arch=sm_100", - "-c", - "/tmp/tmpw1rin_68/smoke.cu", - "-o", - "/tmp/tmpw1rin_68/sm_100.o" - ], - "dryrun_returncode": 0, - "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpw1rin_68/smoke.cu\" -o \"/tmp/tmpxft_00000036_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw1rin_68/smoke.cu\" --orig_src_path_name \"/tmp/tmpw1rin_68/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000036_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000036_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000036_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpw1rin_68/smoke.cu\" -o \"/tmp/tmpxft_00000036_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw1rin_68/smoke.cu\" --orig_src_path_name \"/tmp/tmpw1rin_68/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000036_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000036_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000036_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000036_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000036_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000036_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000036_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000036_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000036_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000036_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpw1rin_68/sm_100.o\" \n", - "next_backend_selected": true, - "returncode": 0, - "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpw1rin_68/smoke.cu\" -o \"/tmp/tmpxft_00000039_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw1rin_68/smoke.cu\" --orig_src_path_name \"/tmp/tmpw1rin_68/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000039_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000039_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000039_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpw1rin_68/smoke.cu\" -o \"/tmp/tmpxft_00000039_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw1rin_68/smoke.cu\" --orig_src_path_name \"/tmp/tmpw1rin_68/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000039_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000039_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000039_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000039_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000039_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000039_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000039_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000039_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000039_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000039_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpw1rin_68/sm_100.o\" \n", - "object_bytes": 12504 - } - ], - "compatible": true - }, - "rows": [ - { - "case": "simple", - "variant": 0, - "mode": "stock_no_pch", - "repeat": 0, - "extension_name": "kb_nvcc_simple_0", - "returncode": 0, - "log": "[1/3] /tmp/tmp_wiyksu7/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp_wiyksu7/main.cpp -o main.o \n[2/3] /tmp/tmp_wiyksu7/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmp_wiyksu7/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmp_wiyksu7/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", - "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", - "container_id": "ta-01M7GG9R0KZAS84KFWFH0WTG6R", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patch": null, - "pch_entry": null, - "process_seconds": 40.325069411, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp_wiyksu7/main.cpp", - "-o", - "main.o" - ], - "seconds": 19.397623978, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp_wiyksu7/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 35.671998917, - "returncode": 0 - } - ], - "seconds": 36.414967365, - "outputs": [ - 2.0, - 2.0, - 2.0, - 2.0, - 2.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": false, - "host_hit_keys": [], - "remote_seconds": 58.20478887 - }, - { - "case": "simple", - "variant": 0, - "mode": "stock_host_pch", - "repeat": 0, - "extension_name": "kb_nvcc_simple_0", - "returncode": 0, - "log": "[1/3] /tmp/tmpz0wlsxxa/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpz0wlsxxa/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpz0wlsxxa/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpz0wlsxxa/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpz0wlsxxa/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", - "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", - "container_id": "ta-01M7GG9SJ3NSEV91JE6RJNCHTR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patch": null, - "pch_entry": null, - "process_seconds": 40.810135701, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpz0wlsxxa/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.101947039, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpz0wlsxxa/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 35.389258092, - "returncode": 0 - } - ], - "seconds": 36.431860271, - "outputs": [ - 2.0, - 2.0, - 2.0, - 2.0, - 2.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "remote_seconds": 48.346788907000004 - }, - { - "case": "simple", - "variant": 0, - "mode": "patched_host_pch", - "repeat": 0, - "extension_name": "kb_nvcc_simple_0", - "returncode": 0, - "log": "[1/3] /tmp/tmpqhcvxumg/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpqhcvxumg/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpqhcvxumg/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpqhcvxumg/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpqhcvxumg/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", - "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", - "container_id": "ta-01M7GG9SJE0C3GB9N1TYQPQWNR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": null, - "process_seconds": 38.605369446, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpqhcvxumg/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.167082585, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpqhcvxumg/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 33.756942244, - "returncode": 0 - } - ], - "seconds": 34.736912296, - "outputs": [ - 2.0, - 2.0, - 2.0, - 2.0, - 2.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "remote_seconds": 47.10341623 - }, - { - "case": "simple", - "variant": 0, - "mode": "patched_all_pch", - "repeat": 0, - "extension_name": "kb_nvcc_simple_0", - "returncode": 0, - "log": "[1/3] /tmp/tmpxpe57igv/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpxpe57igv/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpxpe57igv/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpxpe57igv/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000031_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000031_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmpxpe57igv/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", - "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", - "container_id": "ta-01M7GG9SJ3GHSRNQV0Z0ACJC3R", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", - "process_seconds": 28.188046862, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpxpe57igv/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.079369766, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpxpe57igv/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" - ], - "seconds": 23.198122242, - "returncode": 0 - } - ], - "seconds": 24.183967813, - "outputs": [ - 2.0, - 2.0, - 2.0, - 2.0, - 2.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmpxpe57igv/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmpxpe57igv/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxpe57igv/cuda.cu\" --orig_src_path_name \"/tmp/tmpxpe57igv/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxpe57igv/cuda.cu\" --orig_src_path_name \"/tmp/tmpxpe57igv/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpxpe57igv/diagnostic.cuda.o\" \n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 58.327971263 - }, - { - "case": "simple", - "variant": 1, - "mode": "patched_all_pch", - "repeat": 0, - "extension_name": "kb_nvcc_simple_1", - "returncode": 0, - "log": "[1/3] /tmp/tmp9723ol1n/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp9723ol1n/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp9723ol1n/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmp9723ol1n/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000031_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000031_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmp9723ol1n/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", - "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", - "container_id": "ta-01M7GGBPJYFMHYHKCZZFP9G16R", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", - "process_seconds": 27.839054124999997, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp9723ol1n/main.cpp", - "-o", - "main.o" - ], - "seconds": 7.848992351999999, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp9723ol1n/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" - ], - "seconds": 23.174968698, - "returncode": 0 - } - ], - "seconds": 24.161286733, - "outputs": [ - 3.0, - 3.0, - 3.0, - 3.0, - 3.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp9723ol1n/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp9723ol1n/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp9723ol1n/cuda.cu\" --orig_src_path_name \"/tmp/tmp9723ol1n/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp9723ol1n/cuda.cu\" --orig_src_path_name \"/tmp/tmp9723ol1n/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmp9723ol1n/diagnostic.cuda.o\" \n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 56.157136401 - }, - { - "case": "simple", - "variant": 1, - "mode": "patched_host_pch", - "repeat": 0, - "extension_name": "kb_nvcc_simple_1", - "returncode": 0, - "log": "[1/3] /tmp/tmp4eloe17n/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp4eloe17n/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp4eloe17n/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmp4eloe17n/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmp4eloe17n/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", - "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", - "container_id": "ta-01M7GGGG38BJT0QQKXKPVXKHBR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": null, - "process_seconds": 39.721209833, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp4eloe17n/main.cpp", - "-o", - "main.o" - ], - "seconds": 7.6305212120000006, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp4eloe17n/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 34.912277029, - "returncode": 0 - } - ], - "seconds": 35.908356786, - "outputs": [ - 3.0, - 3.0, - 3.0, - 3.0, - 3.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "remote_seconds": 49.137765439000006 - }, - { - "case": "simple", - "variant": 1, - "mode": "stock_host_pch", - "repeat": 0, - "extension_name": "kb_nvcc_simple_1", - "returncode": 0, - "log": "[1/3] /tmp/tmpsrn85hw5/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpsrn85hw5/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpsrn85hw5/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpsrn85hw5/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpsrn85hw5/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", - "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", - "container_id": "ta-01M7GGGHRFMCBN61Z5SVD1QG4R", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patch": null, - "pch_entry": null, - "process_seconds": 38.687573193, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpsrn85hw5/main.cpp", - "-o", - "main.o" - ], - "seconds": 7.539243877, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpsrn85hw5/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 34.245309142, - "returncode": 0 - } - ], - "seconds": 35.256968721999996, - "outputs": [ - 3.0, - 3.0, - 3.0, - 3.0, - 3.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "remote_seconds": 45.988208858 - }, - { - "case": "simple", - "variant": 1, - "mode": "stock_no_pch", - "repeat": 0, - "extension_name": "kb_nvcc_simple_1", - "returncode": 0, - "log": "[1/3] /tmp/tmpjbfcorzj/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpjbfcorzj/main.cpp -o main.o \n[2/3] /tmp/tmpjbfcorzj/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpjbfcorzj/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpjbfcorzj/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", - "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", - "container_id": "ta-01M7GGGKF3C7V0H9KR3G5ESY1R", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patch": null, - "pch_entry": null, - "process_seconds": 47.066089143999996, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpjbfcorzj/main.cpp", - "-o", - "main.o" - ], - "seconds": 23.122218314, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpjbfcorzj/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 41.880367325, - "returncode": 0 - } - ], - "seconds": 42.885632301, - "outputs": [ - 3.0, - 3.0, - 3.0, - 3.0, - 3.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": false, - "host_hit_keys": [], - "remote_seconds": 67.5727541 - }, - { - "case": "cutlass", - "variant": 0, - "mode": "stock_no_pch", - "repeat": 0, - "extension_name": "kb_nvcc_cutlass_0", - "returncode": 0, - "log": "[1/3] /tmp/tmp11nt69um/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp11nt69um/main.cpp -o main.o \n[2/3] /tmp/tmp11nt69um/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmp11nt69um/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmp11nt69um/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", - "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", - "container_id": "ta-01M7GGGKV69BR167MB9GHB1FWR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patch": null, - "pch_entry": null, - "process_seconds": 60.260855773, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp11nt69um/main.cpp", - "-o", - "main.o" - ], - "seconds": 24.870952887999998, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmp11nt69um/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 53.1093533, - "returncode": 0 - } - ], - "seconds": 54.136435518, - "outputs": [ - 8.0 - ], - "max_absolute_error": 4.2646570364013314e-06, - "correctness_checks": 10, - "host_pch_consumed": false, - "host_hit_keys": [], - "remote_seconds": 82.230074779 - }, - { - "case": "cutlass", - "variant": 0, - "mode": "stock_host_pch", - "repeat": 0, - "extension_name": "kb_nvcc_cutlass_0", - "returncode": 0, - "log": "[1/3] /tmp/tmpqd3a7m0i/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpqd3a7m0i/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpqd3a7m0i/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpqd3a7m0i/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpqd3a7m0i/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", - "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", - "container_id": "ta-01M7GGJ4Y2SED4Z9ZS48M6HFPR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patch": null, - "pch_entry": null, - "process_seconds": 53.059010687, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpqd3a7m0i/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.62824865, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmpqd3a7m0i/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 47.631198614, - "returncode": 0 - } - ], - "seconds": 48.826632251999996, - "outputs": [ - 8.0 - ], - "max_absolute_error": 4.2646570364013314e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "remote_seconds": 61.13100136399999 - }, - { - "case": "cutlass", - "variant": 0, - "mode": "patched_host_pch", - "repeat": 0, - "extension_name": "kb_nvcc_cutlass_0", - "returncode": 0, - "log": "[1/3] /tmp/tmppx4mtayi/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmppx4mtayi/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmppx4mtayi/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmppx4mtayi/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmppx4mtayi/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", - "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", - "container_id": "ta-01M7GGJ6GEPN98RD1T5H1HNMTR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": null, - "process_seconds": 56.536430032999995, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmppx4mtayi/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.464728404999999, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmppx4mtayi/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 49.374024531, - "returncode": 0 - } - ], - "seconds": 51.061138988, - "outputs": [ - 8.0 - ], - "max_absolute_error": 4.2646570364013314e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "remote_seconds": 66.024979734 - }, - { - "case": "cutlass", - "variant": 0, - "mode": "patched_all_pch", - "repeat": 0, - "extension_name": "kb_nvcc_cutlass_0", - "returncode": 0, - "log": "[1/3] /tmp/tmpvj9dykih/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpvj9dykih/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpvj9dykih/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpvj9dykih/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n[3/3] /tmp/tmpvj9dykih/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", - "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", - "container_id": "ta-01M7GGJYZRTAM0NXNGM9XQ7FTR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", - "process_seconds": 48.695631634, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpvj9dykih/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.893383368000002, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmpvj9dykih/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch" - ], - "seconds": 42.391933361999996, - "returncode": 0 - } - ], - "seconds": 43.662042164, - "outputs": [ - 8.0 - ], - "max_absolute_error": 4.2646570364013314e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpvj9dykih/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpvj9dykih/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpvj9dykih/cuda.cu\" --orig_src_path_name \"/tmp/tmpvj9dykih/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpvj9dykih/cuda.cu\" --orig_src_path_name \"/tmp/tmpvj9dykih/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpvj9dykih/diagnostic.cuda.o\" \n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 88.982580958 - }, - { - "case": "cutlass", - "variant": 1, - "mode": "patched_all_pch", - "repeat": 0, - "extension_name": "kb_nvcc_cutlass_1", - "returncode": 0, - "log": "[1/3] /tmp/tmp17x8fwhm/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp17x8fwhm/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp17x8fwhm/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmp17x8fwhm/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000031_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n\"/tmp/tmpxft_00000031_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n[3/3] /tmp/tmp17x8fwhm/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", - "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", - "container_id": "ta-01M7GGKGST0M04EKKWTZSZAZMR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", - "process_seconds": 43.799417716, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp17x8fwhm/main.cpp", - "-o", - "main.o" - ], - "seconds": 9.442032742999999, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmp17x8fwhm/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch" - ], - "seconds": 37.932805739, - "returncode": 0 - } - ], - "seconds": 39.165971053, - "outputs": [ - 16.0 - ], - "max_absolute_error": 8.529314072802663e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmp17x8fwhm/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmp17x8fwhm/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp17x8fwhm/cuda.cu\" --orig_src_path_name \"/tmp/tmp17x8fwhm/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp17x8fwhm/cuda.cu\" --orig_src_path_name \"/tmp/tmp17x8fwhm/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmp17x8fwhm/diagnostic.cuda.o\" \n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 82.974530772 - }, - { - "case": "cutlass", - "variant": 1, - "mode": "patched_host_pch", - "repeat": 0, - "extension_name": "kb_nvcc_cutlass_1", - "returncode": 0, - "log": "[1/3] /tmp/tmpqxtqfy1o/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpqxtqfy1o/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpqxtqfy1o/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpqxtqfy1o/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpqxtqfy1o/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", - "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", - "container_id": "ta-01M7GGMAPNEZDRDHFYXWX097RR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": null, - "process_seconds": 46.039022967, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpqxtqfy1o/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.110150218, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmpqxtqfy1o/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 41.130776678, - "returncode": 0 - } - ], - "seconds": 42.159665512000004, - "outputs": [ - 16.0 - ], - "max_absolute_error": 8.529314072802663e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "remote_seconds": 54.831509720999996 - }, - { - "case": "cutlass", - "variant": 1, - "mode": "stock_host_pch", - "repeat": 0, - "extension_name": "kb_nvcc_cutlass_1", - "returncode": 0, - "log": "[1/3] /tmp/tmpi_7ydo9h/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpi_7ydo9h/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpi_7ydo9h/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpi_7ydo9h/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpi_7ydo9h/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", - "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", - "container_id": "ta-01M7GGMJPMNY2K94DA3P3SCGPR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patch": null, - "pch_entry": null, - "process_seconds": 45.051432037000005, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpi_7ydo9h/main.cpp", - "-o", - "main.o" - ], - "seconds": 7.547227215, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmpi_7ydo9h/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 40.564446911, - "returncode": 0 - } - ], - "seconds": 41.497131416, - "outputs": [ - 16.0 - ], - "max_absolute_error": 8.529314072802663e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "remote_seconds": 52.125646299 - }, - { - "case": "cutlass", - "variant": 1, - "mode": "stock_no_pch", - "repeat": 0, - "extension_name": "kb_nvcc_cutlass_1", - "returncode": 0, - "log": "[1/3] /tmp/tmpcu5fn3dj/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpcu5fn3dj/main.cpp -o main.o \n[2/3] /tmp/tmpcu5fn3dj/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpcu5fn3dj/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpcu5fn3dj/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", - "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", - "container_id": "ta-01M7GGP2866GKDMKQ7CB7P4DWR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patch": null, - "pch_entry": null, - "process_seconds": 46.651000024999995, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpcu5fn3dj/main.cpp", - "-o", - "main.o" - ], - "seconds": 18.774265045, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmpcu5fn3dj/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 42.147979744, - "returncode": 0 - } - ], - "seconds": 42.949985459, - "outputs": [ - 16.0 - ], - "max_absolute_error": 8.529314072802663e-06, - "correctness_checks": 10, - "host_pch_consumed": false, - "host_hit_keys": [], - "remote_seconds": 64.96884930200001 - }, - { - "case": "simple", - "variant": 0, - "mode": "patched_all_pch", - "repeat": 1, - "extension_name": "kb_nvcc_simple_0", - "returncode": 0, - "log": "[1/3] /tmp/tmpjwxy1kiy/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpjwxy1kiy/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpjwxy1kiy/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpjwxy1kiy/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmpjwxy1kiy/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", - "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", - "container_id": "ta-01M7GGP9SHQZVXA5GZ2ZE3HXKR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", - "process_seconds": 27.58442731, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpjwxy1kiy/main.cpp", - "-o", - "main.o" - ], - "seconds": 7.728120469, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpjwxy1kiy/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" - ], - "seconds": 23.012643256, - "returncode": 0 - } - ], - "seconds": 23.934274335, - "outputs": [ - 2.0, - 2.0, - 2.0, - 2.0, - 2.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmpjwxy1kiy/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmpjwxy1kiy/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpjwxy1kiy/cuda.cu\" --orig_src_path_name \"/tmp/tmpjwxy1kiy/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpjwxy1kiy/cuda.cu\" --orig_src_path_name \"/tmp/tmpjwxy1kiy/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpjwxy1kiy/diagnostic.cuda.o\" \n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 55.646309089000006 - }, - { - "case": "simple", - "variant": 0, - "mode": "patched_host_pch", - "repeat": 1, - "extension_name": "kb_nvcc_simple_0", - "returncode": 0, - "log": "[1/3] /tmp/tmponu3k1v4/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmponu3k1v4/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmponu3k1v4/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmponu3k1v4/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmponu3k1v4/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", - "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", - "container_id": "ta-01M7GGPAVJ1N65VC6G7VXNW35R", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": null, - "process_seconds": 39.487048599000005, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmponu3k1v4/main.cpp", - "-o", - "main.o" - ], - "seconds": 7.554983864, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmponu3k1v4/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 35.113119433, - "returncode": 0 - } - ], - "seconds": 36.119203624, - "outputs": [ - 2.0, - 2.0, - 2.0, - 2.0, - 2.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "remote_seconds": 47.540994509 - }, - { - "case": "simple", - "variant": 0, - "mode": "stock_host_pch", - "repeat": 1, - "extension_name": "kb_nvcc_simple_0", - "returncode": 0, - "log": "[1/3] /tmp/tmp40_sti7n/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp40_sti7n/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp40_sti7n/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmp40_sti7n/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmp40_sti7n/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", - "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", - "container_id": "ta-01M7GGPCR34H6B7QV57HMT3TFR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patch": null, - "pch_entry": null, - "process_seconds": 39.490719059, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp40_sti7n/main.cpp", - "-o", - "main.o" - ], - "seconds": 7.914581139, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp40_sti7n/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 34.503693313, - "returncode": 0 - } - ], - "seconds": 35.485519697, - "outputs": [ - 2.0, - 2.0, - 2.0, - 2.0, - 2.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "remote_seconds": 47.474009400999996 - }, - { - "case": "simple", - "variant": 0, - "mode": "stock_no_pch", - "repeat": 1, - "extension_name": "kb_nvcc_simple_0", - "returncode": 0, - "log": "[1/3] /tmp/tmplhkbmlgw/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmplhkbmlgw/main.cpp -o main.o \n[2/3] /tmp/tmplhkbmlgw/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmplhkbmlgw/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmplhkbmlgw/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", - "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", - "container_id": "ta-01M7GGR0BSXMZF6601Q2HH4GDR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patch": null, - "pch_entry": null, - "process_seconds": 41.56644511099999, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmplhkbmlgw/main.cpp", - "-o", - "main.o" - ], - "seconds": 19.695273711, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmplhkbmlgw/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 37.226831019, - "returncode": 0 - } - ], - "seconds": 38.039081186, - "outputs": [ - 2.0, - 2.0, - 2.0, - 2.0, - 2.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": false, - "host_hit_keys": [], - "remote_seconds": 59.653186801 - }, - { - "case": "simple", - "variant": 1, - "mode": "stock_no_pch", - "repeat": 1, - "extension_name": "kb_nvcc_simple_1", - "returncode": 0, - "log": "[1/3] /tmp/tmpjq7sj4w6/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpjq7sj4w6/main.cpp -o main.o \n[2/3] /tmp/tmpjq7sj4w6/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpjq7sj4w6/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpjq7sj4w6/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", - "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", - "container_id": "ta-01M7GGR1YGMM34BW5HTWRC9WQR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patch": null, - "pch_entry": null, - "process_seconds": 39.52563073, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpjq7sj4w6/main.cpp", - "-o", - "main.o" - ], - "seconds": 19.178143019, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpjq7sj4w6/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 35.017214447, - "returncode": 0 - } - ], - "seconds": 35.746856564, - "outputs": [ - 3.0, - 3.0, - 3.0, - 3.0, - 3.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": false, - "host_hit_keys": [], - "remote_seconds": 57.896333988 - }, - { - "case": "simple", - "variant": 1, - "mode": "stock_host_pch", - "repeat": 1, - "extension_name": "kb_nvcc_simple_1", - "returncode": 0, - "log": "[1/3] /tmp/tmptwzsi6e7/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmptwzsi6e7/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmptwzsi6e7/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmptwzsi6e7/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmptwzsi6e7/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", - "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", - "container_id": "ta-01M7GGR6WK131M0DZZXP133CPR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patch": null, - "pch_entry": null, - "process_seconds": 38.652557074, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmptwzsi6e7/main.cpp", - "-o", - "main.o" - ], - "seconds": 7.423724915000001, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmptwzsi6e7/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 34.067612735, - "returncode": 0 - } - ], - "seconds": 35.025937535000004, - "outputs": [ - 3.0, - 3.0, - 3.0, - 3.0, - 3.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "remote_seconds": 45.924620759 - }, - { - "case": "simple", - "variant": 1, - "mode": "patched_host_pch", - "repeat": 1, - "extension_name": "kb_nvcc_simple_1", - "returncode": 0, - "log": "[1/3] /tmp/tmpihl2xip2/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpihl2xip2/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpihl2xip2/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpihl2xip2/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpihl2xip2/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", - "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", - "container_id": "ta-01M7GGR8JJ7XAQGGNEN85CYRER", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": null, - "process_seconds": 42.065910112, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpihl2xip2/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.606916758, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpihl2xip2/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 37.376972288000005, - "returncode": 0 - } - ], - "seconds": 38.467215558, - "outputs": [ - 3.0, - 3.0, - 3.0, - 3.0, - 3.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "remote_seconds": 50.520850166 - }, - { - "case": "simple", - "variant": 1, - "mode": "patched_all_pch", - "repeat": 1, - "extension_name": "kb_nvcc_simple_1", - "returncode": 0, - "log": "[1/3] /tmp/tmp858f0ai0/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp858f0ai0/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp858f0ai0/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmp858f0ai0/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000031_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000031_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmp858f0ai0/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", - "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", - "container_id": "ta-01M7GGSSKT9DDF905Z9S6J9SER", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", - "process_seconds": 33.745011223999995, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp858f0ai0/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.940535134, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp858f0ai0/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" - ], - "seconds": 27.845098587000003, - "returncode": 0 - } - ], - "seconds": 29.137589784, - "outputs": [ - 3.0, - 3.0, - 3.0, - 3.0, - 3.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp858f0ai0/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp858f0ai0/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp858f0ai0/cuda.cu\" --orig_src_path_name \"/tmp/tmp858f0ai0/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp858f0ai0/cuda.cu\" --orig_src_path_name \"/tmp/tmp858f0ai0/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmp858f0ai0/diagnostic.cuda.o\" \n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 72.54307103100001 - }, - { - "case": "cutlass", - "variant": 0, - "mode": "patched_all_pch", - "repeat": 1, - "extension_name": "kb_nvcc_cutlass_0", - "returncode": 0, - "log": "[1/3] /tmp/tmp6b6mdyz9/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp6b6mdyz9/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp6b6mdyz9/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmp6b6mdyz9/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n[3/3] /tmp/tmp6b6mdyz9/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", - "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", - "container_id": "ta-01M7GGT09WZH9JV14JHDV0YQVR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", - "process_seconds": 32.546973025, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp6b6mdyz9/main.cpp", - "-o", - "main.o" - ], - "seconds": 7.5847706829999995, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmp6b6mdyz9/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch" - ], - "seconds": 27.954299207, - "returncode": 0 - } - ], - "seconds": 28.886751083999997, - "outputs": [ - 8.0 - ], - "max_absolute_error": 4.2646570364013314e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmp6b6mdyz9/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmp6b6mdyz9/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp6b6mdyz9/cuda.cu\" --orig_src_path_name \"/tmp/tmp6b6mdyz9/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp6b6mdyz9/cuda.cu\" --orig_src_path_name \"/tmp/tmp6b6mdyz9/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmp6b6mdyz9/diagnostic.cuda.o\" \n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 65.119184206 - }, - { - "case": "cutlass", - "variant": 0, - "mode": "patched_host_pch", - "repeat": 1, - "extension_name": "kb_nvcc_cutlass_0", - "returncode": 0, - "log": "[1/3] /tmp/tmpx6o9xcvw/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpx6o9xcvw/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpx6o9xcvw/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpx6o9xcvw/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpx6o9xcvw/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", - "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", - "container_id": "ta-01M7GGT1KA1EXM9Z1NHPNBF6FR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": null, - "process_seconds": 45.630522881, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpx6o9xcvw/main.cpp", - "-o", - "main.o" - ], - "seconds": 7.483017078999999, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmpx6o9xcvw/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 40.705319495, - "returncode": 0 - } - ], - "seconds": 41.732421651, - "outputs": [ - 8.0 - ], - "max_absolute_error": 4.2646570364013314e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "remote_seconds": 53.773416114 - }, - { - "case": "cutlass", - "variant": 0, - "mode": "stock_host_pch", - "repeat": 1, - "extension_name": "kb_nvcc_cutlass_0", - "returncode": 0, - "log": "[1/3] /tmp/tmpli06qb2k/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpli06qb2k/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpli06qb2k/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpli06qb2k/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpli06qb2k/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", - "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", - "container_id": "ta-01M7GGT2A6N9TPEAVPNNM3K8TR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patch": null, - "pch_entry": null, - "process_seconds": 47.514512463, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpli06qb2k/main.cpp", - "-o", - "main.o" - ], - "seconds": 7.684575478, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmpli06qb2k/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 42.481320051, - "returncode": 0 - } - ], - "seconds": 43.52756049, - "outputs": [ - 8.0 - ], - "max_absolute_error": 4.2646570364013314e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "remote_seconds": 54.892269899999995 - }, - { - "case": "cutlass", - "variant": 0, - "mode": "stock_no_pch", - "repeat": 1, - "extension_name": "kb_nvcc_cutlass_0", - "returncode": 0, - "log": "[1/3] /tmp/tmppeys7t2i/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmppeys7t2i/main.cpp -o main.o \n[2/3] /tmp/tmppeys7t2i/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmppeys7t2i/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmppeys7t2i/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", - "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", - "container_id": "ta-01M7GGVWMY5WT8TDPYNZFRJY0R", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patch": null, - "pch_entry": null, - "process_seconds": 59.069312495, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmppeys7t2i/main.cpp", - "-o", - "main.o" - ], - "seconds": 22.843029878, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmppeys7t2i/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 53.267142170999996, - "returncode": 0 - } - ], - "seconds": 54.309780212999996, - "outputs": [ - 8.0 - ], - "max_absolute_error": 4.2646570364013314e-06, - "correctness_checks": 10, - "host_pch_consumed": false, - "host_hit_keys": [], - "remote_seconds": 82.28517314700001 - }, - { - "case": "cutlass", - "variant": 1, - "mode": "stock_no_pch", - "repeat": 1, - "extension_name": "kb_nvcc_cutlass_1", - "returncode": 0, - "log": "[1/3] /tmp/tmphrakn43k/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmphrakn43k/main.cpp -o main.o \n[2/3] /tmp/tmphrakn43k/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmphrakn43k/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmphrakn43k/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", - "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", - "container_id": "ta-01M7GGVZYE7ZXED0JB12EWFP6R", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patch": null, - "pch_entry": null, - "process_seconds": 47.475423516999996, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmphrakn43k/main.cpp", - "-o", - "main.o" - ], - "seconds": 18.90647545, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmphrakn43k/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 42.640482535, - "returncode": 0 - } - ], - "seconds": 43.498140506, - "outputs": [ - 16.0 - ], - "max_absolute_error": 8.529314072802663e-06, - "correctness_checks": 10, - "host_pch_consumed": false, - "host_hit_keys": [], - "remote_seconds": 65.75153526199999 - }, - { - "case": "cutlass", - "variant": 1, - "mode": "stock_host_pch", - "repeat": 1, - "extension_name": "kb_nvcc_cutlass_1", - "returncode": 0, - "log": "[1/3] /tmp/tmp1hzlz2oz/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp1hzlz2oz/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp1hzlz2oz/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmp1hzlz2oz/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmp1hzlz2oz/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", - "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", - "container_id": "ta-01M7GGW6GV9V5H8EYB13ANYRBR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patch": null, - "pch_entry": null, - "process_seconds": 47.481454979000006, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp1hzlz2oz/main.cpp", - "-o", - "main.o" - ], - "seconds": 7.983763389000001, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmp1hzlz2oz/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 42.884354589999994, - "returncode": 0 - } - ], - "seconds": 43.922133784, - "outputs": [ - 16.0 - ], - "max_absolute_error": 8.529314072802663e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "remote_seconds": 54.927763373 - }, - { - "case": "cutlass", - "variant": 1, - "mode": "patched_host_pch", - "repeat": 1, - "extension_name": "kb_nvcc_cutlass_1", - "returncode": 0, - "log": "[1/3] /tmp/tmpimrh22t3/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpimrh22t3/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpimrh22t3/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpimrh22t3/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpimrh22t3/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", - "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", - "container_id": "ta-01M7GGW978S61BB6WXTVDCE8GR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": null, - "process_seconds": 54.218881587999995, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpimrh22t3/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.671524286999999, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmpimrh22t3/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 48.066379684000005, - "returncode": 0 - } - ], - "seconds": 49.357130543, - "outputs": [ - 16.0 - ], - "max_absolute_error": 8.529314072802663e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "remote_seconds": 63.463944837999996 - }, - { - "case": "cutlass", - "variant": 1, - "mode": "patched_all_pch", - "repeat": 1, - "extension_name": "kb_nvcc_cutlass_1", - "returncode": 0, - "log": "[1/3] /tmp/tmpxzjxiynb/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpxzjxiynb/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpxzjxiynb/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpxzjxiynb/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n[3/3] /tmp/tmpxzjxiynb/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", - "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", - "container_id": "ta-01M7GGY1T72T9C66SB78QVCPJR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", - "process_seconds": 32.155385849999995, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpxzjxiynb/main.cpp", - "-o", - "main.o" - ], - "seconds": 7.430922593, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmpxzjxiynb/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch" - ], - "seconds": 27.405767641000004, - "returncode": 0 - } - ], - "seconds": 28.440589126, - "outputs": [ - 16.0 - ], - "max_absolute_error": 8.529314072802663e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpxzjxiynb/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpxzjxiynb/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzjxiynb/cuda.cu\" --orig_src_path_name \"/tmp/tmpxzjxiynb/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzjxiynb/cuda.cu\" --orig_src_path_name \"/tmp/tmpxzjxiynb/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpxzjxiynb/diagnostic.cuda.o\" \n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 65.992117256 - } - ] - }, - "supplemental": { - "probe": { - "stock_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patched_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "paths": "/usr/local/cuda/bin/nvcc.profile\n/usr/local/cuda/nvvm/bin/cicc\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "nvcc_profiles": { - "/usr/local/cuda/bin/nvcc.profile": "\nTOP = $(_HERE_)/..\n\nCICC_PATH = $(TOP)/nvvm-next/bin\nNVVMIR_LIBRARY_DIR = $(TOP)/nvvm/libdevice\n\nLD_LIBRARY_PATH += $(TOP)/lib:\nPATH += $(CICC_PATH):$(_HERE_):\n\nINCLUDES += \"-I$(TOP)/$(_TARGET_DIR_)/include\" $(_SPACE_)\nSYSTEM_INCLUDES += \"-isystem\" \"$(TOP)/$(_TARGET_DIR_)/include/cccl\" $(_SPACE_)\n\nLIBRARIES =+ $(_SPACE_) \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)/stubs\" \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)\"\n\nCUDAFE_FLAGS +=\nPTXAS_FLAGS +=\n" - }, - "pch_help": " Specify options directly to library manager.\n\n--pch --pch \n Enable precompiled header (PCH) processing. When specified, each compilation\n searches the PCH directory (see --pch-dir) for a reusable precompiled header,\n and creates a new one when none is found.\n\n--pch-dir --pch-dir \n Specify the directory in which precompiled header files are read from and\n written to. By default this is the current working directory. Ignored unless\n --pch, --create-pch, or --use-pch is enabled.\n\n--pch-verbose --pch-verbose \n Print (to stderr) messages related to precompiled header processing. Ignored\n unless --pch, --create-pch, or --use-pch is enabled.\n\n--create-pch --create-pch \n Manually create a precompiled header (PCH) at the first PCH terminal stop-point\n (i.e., #pragma hdrstop) and save it under the specified name. Intermediate\n stop points (#pragma hdrstop++) are ignored. The directory into which the\n PCH file is written is determined by --pch-dir (or is the current working\n directory if not specified). This option forces a PCH file to be created\n (any existing PCH file with the same name is overwritten). This option is\n mutually exclusive with '--use-pch' and '--pch'.\n\n--use-pch --use-pch \n If a precompiled header (PCH) with the specified name exists in the directory\n determined by '--pch-dir' (or the current working directory if not specified),\n try to use it in compilation.This option is mutually exclusive with '--create-pch'\n and '--pch'.\n\n--ptxas-options ,... (-Xptxas) \n Specify options directly to ptxas, the PTX optimizing assembler.\n\n--nvlink-options ,... (-Xnvlink) \n Specify options directly to nvlink.\n\n--nvprune-options ,... (-Xnvprune) \n Specify options directly to nvprune (used with -prune).\n\n\nMiscellaneous options for guiding the compiler driver.\n======================================================", - "strace": null, - "smoke": [ - { - "arch": "sm_75", - "command": [ - "/usr/local/cuda/bin/nvcc", - "-arch=sm_75", - "-c", - "/tmp/tmpxzw56gh8/smoke.cu", - "-o", - "/tmp/tmpxzw56gh8/sm_75.o" - ], - "dryrun_returncode": 0, - "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpxzw56gh8/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzw56gh8/smoke.cu\" --orig_src_path_name \"/tmp/tmpxzw56gh8/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpxzw56gh8/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzw56gh8/smoke.cu\" --orig_src_path_name \"/tmp/tmpxzw56gh8/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_0000001e_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_0000001e_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_0000001e_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpxzw56gh8/sm_75.o\" \n", - "next_backend_selected": true, - "returncode": 0, - "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpxzw56gh8/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzw56gh8/smoke.cu\" --orig_src_path_name \"/tmp/tmpxzw56gh8/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpxzw56gh8/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzw56gh8/smoke.cu\" --orig_src_path_name \"/tmp/tmpxzw56gh8/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000021_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000021_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000021_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpxzw56gh8/sm_75.o\" \n", - "object_bytes": 10200 - }, - { - "arch": "sm_100", - "command": [ - "/usr/local/cuda/bin/nvcc", - "-arch=sm_100", - "-c", - "/tmp/tmpxzw56gh8/smoke.cu", - "-o", - "/tmp/tmpxzw56gh8/sm_100.o" - ], - "dryrun_returncode": 0, - "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpxzw56gh8/smoke.cu\" -o \"/tmp/tmpxft_00000036_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzw56gh8/smoke.cu\" --orig_src_path_name \"/tmp/tmpxzw56gh8/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000036_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000036_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000036_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpxzw56gh8/smoke.cu\" -o \"/tmp/tmpxft_00000036_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzw56gh8/smoke.cu\" --orig_src_path_name \"/tmp/tmpxzw56gh8/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000036_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000036_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000036_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000036_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000036_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000036_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000036_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000036_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000036_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000036_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpxzw56gh8/sm_100.o\" \n", - "next_backend_selected": true, - "returncode": 0, - "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpxzw56gh8/smoke.cu\" -o \"/tmp/tmpxft_00000039_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzw56gh8/smoke.cu\" --orig_src_path_name \"/tmp/tmpxzw56gh8/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000039_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000039_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000039_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpxzw56gh8/smoke.cu\" -o \"/tmp/tmpxft_00000039_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpxzw56gh8/smoke.cu\" --orig_src_path_name \"/tmp/tmpxzw56gh8/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000039_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000039_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000039_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000039_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000039_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000039_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000039_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000039_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000039_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000039_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpxzw56gh8/sm_100.o\" \n", - "object_bytes": 12504 - } - ], - "compatible": true - }, - "rows": [ - { - "case": "simple", - "variant": 0, - "mode": "patched_host_pch", - "repeat": 0, - "extension_name": "kb_nvcc_simple_0", - "returncode": 0, - "log": "[1/3] /tmp/tmpix5k2j49/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpix5k2j49/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpix5k2j49/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpix5k2j49/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpix5k2j49/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", - "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", - "container_id": "ta-01M7GGED50386E0QEMXCQGKJAR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": null, - "process_seconds": 41.332697233000005, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpix5k2j49/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.343771207, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpix5k2j49/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 36.214367632000005, - "returncode": 0 - } - ], - "seconds": 37.205667172000005, - "outputs": [ - 2.0, - 2.0, - 2.0, - 2.0, - 2.0 - ], - "load_inline_seconds": 37.205656627, - "prefetch_seconds": 0, - "prefetch_bytes": 0, - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpix5k2j49/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpix5k2j49/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpix5k2j49/cuda.cu\" --orig_src_path_name \"/tmp/tmpix5k2j49/cuda.cu\" --allow_managed --relaxed_constexpr --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ \"$CICC_PATH/cicc\" --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpix5k2j49/cuda.cu\" --orig_src_path_name \"/tmp/tmpix5k2j49/cuda.cu\" --allow_managed --relaxed_constexpr -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpix5k2j49/diagnostic.cuda.o\" \n", - "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmpix5k2j49/cuda.cu , gcc (preprocessing 1) , /tmp/tmpix5k2j49/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 2824.1750 , ms\n/tmp/tmpix5k2j49/cuda.cu , gcc (preprocessing 4) , /tmp/tmpix5k2j49/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 2755.5020 , ms\n/tmp/tmpix5k2j49/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 12239.0156 , ms\n/tmp/tmpix5k2j49/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 8393.6445 , ms\n/tmp/tmpix5k2j49/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 25.3120 , ms\n/tmp/tmpix5k2j49/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 4.2280 , ms\n/tmp/tmpix5k2j49/cuda.cu , gcc (compiling) , , /tmp/tmpix5k2j49/diagnostic.cuda.o , compute_75 , nvcc , 7372.6270 , ms\n , nvcc (driver) , , /tmp/tmpix5k2j49/diagnostic.cuda.o , , nvcc , 65.8086 , ms\n" - }, - "device_pch_files_used": [], - "device_pch_files_created": [], - "device_pch_consumed": false, - "remote_seconds": 83.449806407 - }, - { - "case": "simple", - "variant": 0, - "mode": "patched_all_pch", - "repeat": 0, - "extension_name": "kb_nvcc_simple_0", - "returncode": 0, - "log": "[1/3] /tmp/tmp279h84nf/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp279h84nf/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp279h84nf/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmp279h84nf/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmp279h84nf/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", - "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", - "container_id": "ta-01M7GGED50E6R503HVDSM8MGRR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", - "process_seconds": 31.842433575999998, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp279h84nf/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.092909477, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp279h84nf/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" - ], - "seconds": 26.555807986, - "returncode": 0 - } - ], - "seconds": 27.513593019, - "outputs": [ - 2.0, - 2.0, - 2.0, - 2.0, - 2.0 - ], - "load_inline_seconds": 27.513590667000003, - "prefetch_seconds": 0, - "prefetch_bytes": 0, - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp279h84nf/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp279h84nf/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp279h84nf/cuda.cu\" --orig_src_path_name \"/tmp/tmp279h84nf/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp279h84nf/cuda.cu\" --orig_src_path_name \"/tmp/tmp279h84nf/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmp279h84nf/diagnostic.cuda.o\" \n", - "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmp279h84nf/cuda.cu , gcc (preprocessing 1) , /tmp/tmp279h84nf/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 2803.0601 , ms\n/tmp/tmp279h84nf/cuda.cu , gcc (preprocessing 4) , /tmp/tmp279h84nf/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 2761.7849 , ms\n/tmp/tmp279h84nf/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 3816.4561 , ms\n/tmp/tmp279h84nf/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 4133.8110 , ms\n/tmp/tmp279h84nf/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 31.0950 , ms\n/tmp/tmp279h84nf/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 25.4210 , ms\n/tmp/tmp279h84nf/cuda.cu , gcc (compiling) , , /tmp/tmp279h84nf/diagnostic.cuda.o , compute_75 , nvcc , 7562.0918 , ms\n , nvcc (driver) , , /tmp/tmp279h84nf/diagnostic.cuda.o , , nvcc , 65.3496 , ms\n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 61.302708119 - }, - { - "case": "simple", - "variant": 0, - "mode": "patched_all_pch_prefetch", - "repeat": 0, - "extension_name": "kb_nvcc_simple_0", - "returncode": 0, - "log": "[1/3] /tmp/tmpcucf0lnk/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpcucf0lnk/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpcucf0lnk/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpcucf0lnk/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmpcucf0lnk/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", - "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", - "container_id": "ta-01M7GGED5AZC6SWFBFCVMC7C5R", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", - "process_seconds": 30.038420162, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpcucf0lnk/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.254390696, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpcucf0lnk/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" - ], - "seconds": 23.209440734000005, - "returncode": 0 - } - ], - "seconds": 25.731699179, - "outputs": [ - 2.0, - 2.0, - 2.0, - 2.0, - 2.0 - ], - "load_inline_seconds": 24.222713579, - "prefetch_seconds": 1.5089852529999996, - "prefetch_bytes": 2524303360, - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmpcucf0lnk/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmpcucf0lnk/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpcucf0lnk/cuda.cu\" --orig_src_path_name \"/tmp/tmpcucf0lnk/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpcucf0lnk/cuda.cu\" --orig_src_path_name \"/tmp/tmpcucf0lnk/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpcucf0lnk/diagnostic.cuda.o\" \n", - "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmpcucf0lnk/cuda.cu , gcc (preprocessing 1) , /tmp/tmpcucf0lnk/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 2736.3269 , ms\n/tmp/tmpcucf0lnk/cuda.cu , gcc (preprocessing 4) , /tmp/tmpcucf0lnk/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 2780.0220 , ms\n/tmp/tmpcucf0lnk/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 3874.4480 , ms\n/tmp/tmpcucf0lnk/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 4437.3799 , ms\n/tmp/tmpcucf0lnk/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 26.7620 , ms\n/tmp/tmpcucf0lnk/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 4.7330 , ms\n/tmp/tmpcucf0lnk/cuda.cu , gcc (compiling) , , /tmp/tmpcucf0lnk/diagnostic.cuda.o , compute_75 , nvcc , 7862.6460 , ms\n , nvcc (driver) , , /tmp/tmpcucf0lnk/diagnostic.cuda.o , , nvcc , 67.0508 , ms\n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 60.145700806 - }, - { - "case": "simple", - "variant": 1, - "mode": "patched_all_pch_prefetch", - "repeat": 0, - "extension_name": "kb_nvcc_simple_1", - "returncode": 0, - "log": "[1/3] /tmp/tmp__6x16w0/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp__6x16w0/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp__6x16w0/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmp__6x16w0/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000031_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000031_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmp__6x16w0/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", - "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", - "container_id": "ta-01M7GGED5AR5589PP71ECYZT5R", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", - "process_seconds": 30.059914977, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp__6x16w0/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.255181534999998, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp__6x16w0/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" - ], - "seconds": 23.040583837, - "returncode": 0 - } - ], - "seconds": 25.672850840000002, - "outputs": [ - 3.0, - 3.0, - 3.0, - 3.0, - 3.0 - ], - "load_inline_seconds": 24.072621637, - "prefetch_seconds": 1.6002290010000007, - "prefetch_bytes": 2524303360, - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp__6x16w0/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp__6x16w0/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp__6x16w0/cuda.cu\" --orig_src_path_name \"/tmp/tmp__6x16w0/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp__6x16w0/cuda.cu\" --orig_src_path_name \"/tmp/tmp__6x16w0/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmp__6x16w0/diagnostic.cuda.o\" \n", - "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmp__6x16w0/cuda.cu , gcc (preprocessing 1) , /tmp/tmp__6x16w0/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 2714.1531 , ms\n/tmp/tmp__6x16w0/cuda.cu , gcc (preprocessing 4) , /tmp/tmp__6x16w0/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 2679.6211 , ms\n/tmp/tmp__6x16w0/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 3953.6370 , ms\n/tmp/tmp__6x16w0/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 4338.0518 , ms\n/tmp/tmp__6x16w0/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 32.9190 , ms\n/tmp/tmp__6x16w0/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 4.1430 , ms\n/tmp/tmp__6x16w0/cuda.cu , gcc (compiling) , , /tmp/tmp__6x16w0/diagnostic.cuda.o , compute_75 , nvcc , 7706.5312 , ms\n , nvcc (driver) , , /tmp/tmp__6x16w0/diagnostic.cuda.o , , nvcc , 66.0625 , ms\n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 60.086438656999995 - }, - { - "case": "simple", - "variant": 1, - "mode": "patched_all_pch", - "repeat": 0, - "extension_name": "kb_nvcc_simple_1", - "returncode": 0, - "log": "[1/3] /tmp/tmppd82nejq/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmppd82nejq/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmppd82nejq/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmppd82nejq/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmppd82nejq/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", - "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", - "container_id": "ta-01M7GGGJTBKKMAZG8E43P3FYXR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", - "process_seconds": 26.685320548, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmppd82nejq/main.cpp", - "-o", - "main.o" - ], - "seconds": 7.648132214000001, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmppd82nejq/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" - ], - "seconds": 22.029563485, - "returncode": 0 - } - ], - "seconds": 22.962655791, - "outputs": [ - 3.0, - 3.0, - 3.0, - 3.0, - 3.0 - ], - "load_inline_seconds": 22.962653443, - "prefetch_seconds": 0, - "prefetch_bytes": 0, - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmppd82nejq/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmppd82nejq/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmppd82nejq/cuda.cu\" --orig_src_path_name \"/tmp/tmppd82nejq/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmppd82nejq/cuda.cu\" --orig_src_path_name \"/tmp/tmppd82nejq/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmppd82nejq/diagnostic.cuda.o\" \n", - "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmppd82nejq/cuda.cu , gcc (preprocessing 1) , /tmp/tmppd82nejq/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 2792.3230 , ms\n/tmp/tmppd82nejq/cuda.cu , gcc (preprocessing 4) , /tmp/tmppd82nejq/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 2763.0911 , ms\n/tmp/tmppd82nejq/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 3405.5300 , ms\n/tmp/tmppd82nejq/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 3858.2410 , ms\n/tmp/tmppd82nejq/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 23.1240 , ms\n/tmp/tmppd82nejq/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 4.2590 , ms\n/tmp/tmppd82nejq/cuda.cu , gcc (compiling) , , /tmp/tmppd82nejq/diagnostic.cuda.o , compute_75 , nvcc , 7438.9702 , ms\n , nvcc (driver) , , /tmp/tmppd82nejq/diagnostic.cuda.o , , nvcc , 65.9883 , ms\n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 55.045973556 - }, - { - "case": "simple", - "variant": 1, - "mode": "patched_host_pch", - "repeat": 0, - "extension_name": "kb_nvcc_simple_1", - "returncode": 0, - "log": "[1/3] /tmp/tmpdogjj3qw/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpdogjj3qw/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpdogjj3qw/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpdogjj3qw/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpdogjj3qw/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", - "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", - "container_id": "ta-01M7GGGJTBSQFTS3HEY1Y4K4GR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": null, - "process_seconds": 49.306690329, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpdogjj3qw/main.cpp", - "-o", - "main.o" - ], - "seconds": 12.670156114000001, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpdogjj3qw/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 42.628599909, - "returncode": 0 - } - ], - "seconds": 44.070919440000004, - "outputs": [ - 3.0, - 3.0, - 3.0, - 3.0, - 3.0 - ], - "load_inline_seconds": 44.070915819, - "prefetch_seconds": 0, - "prefetch_bytes": 0, - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpdogjj3qw/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpdogjj3qw/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpdogjj3qw/cuda.cu\" --orig_src_path_name \"/tmp/tmpdogjj3qw/cuda.cu\" --allow_managed --relaxed_constexpr --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ \"$CICC_PATH/cicc\" --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpdogjj3qw/cuda.cu\" --orig_src_path_name \"/tmp/tmpdogjj3qw/cuda.cu\" --allow_managed --relaxed_constexpr -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpdogjj3qw/diagnostic.cuda.o\" \n", - "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmpdogjj3qw/cuda.cu , gcc (preprocessing 1) , /tmp/tmpdogjj3qw/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 3497.3870 , ms\n/tmp/tmpdogjj3qw/cuda.cu , gcc (preprocessing 4) , /tmp/tmpdogjj3qw/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 3515.5879 , ms\n/tmp/tmpdogjj3qw/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 15524.1338 , ms\n/tmp/tmpdogjj3qw/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 11493.0625 , ms\n/tmp/tmpdogjj3qw/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 33.9710 , ms\n/tmp/tmpdogjj3qw/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 5.8130 , ms\n/tmp/tmpdogjj3qw/cuda.cu , gcc (compiling) , , /tmp/tmpdogjj3qw/diagnostic.cuda.o , compute_75 , nvcc , 9197.6523 , ms\n , nvcc (driver) , , /tmp/tmpdogjj3qw/diagnostic.cuda.o , , nvcc , 84.5664 , ms\n" - }, - "device_pch_files_used": [], - "device_pch_files_created": [], - "device_pch_consumed": false, - "remote_seconds": 101.697851061 - }, - { - "case": "cutlass", - "variant": 0, - "mode": "patched_host_pch", - "repeat": 0, - "extension_name": "kb_nvcc_cutlass_0", - "returncode": 0, - "log": "[1/3] /tmp/tmpd36naltj/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpd36naltj/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpd36naltj/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpd36naltj/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpd36naltj/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", - "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", - "container_id": "ta-01M7GGGZXAYNTYNS7JADW3KA9R", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": null, - "process_seconds": 55.738259471999996, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpd36naltj/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.778172018, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmpd36naltj/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 49.517795254999996, - "returncode": 0 - } - ], - "seconds": 50.875985165, - "outputs": [ - 8.0 - ], - "load_inline_seconds": 50.875981628, - "prefetch_seconds": 0, - "prefetch_bytes": 0, - "max_absolute_error": 4.2646570364013314e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpd36naltj/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpd36naltj/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpd36naltj/cuda.cu\" --orig_src_path_name \"/tmp/tmpd36naltj/cuda.cu\" --allow_managed --relaxed_constexpr --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ \"$CICC_PATH/cicc\" --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpd36naltj/cuda.cu\" --orig_src_path_name \"/tmp/tmpd36naltj/cuda.cu\" --allow_managed --relaxed_constexpr -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpd36naltj/diagnostic.cuda.o\" \n", - "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmpd36naltj/cuda.cu , gcc (preprocessing 1) , /tmp/tmpd36naltj/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 4387.8960 , ms\n/tmp/tmpd36naltj/cuda.cu , gcc (preprocessing 4) , /tmp/tmpd36naltj/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 4520.6260 , ms\n/tmp/tmpd36naltj/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 19127.1426 , ms\n/tmp/tmpd36naltj/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 11011.8525 , ms\n/tmp/tmpd36naltj/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 413.6620 , ms\n/tmp/tmpd36naltj/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 7.7100 , ms\n/tmp/tmpd36naltj/cuda.cu , gcc (compiling) , , /tmp/tmpd36naltj/diagnostic.cuda.o , compute_75 , nvcc , 9960.2764 , ms\n , nvcc (driver) , , /tmp/tmpd36naltj/diagnostic.cuda.o , , nvcc , 110.8203 , ms\n" - }, - "device_pch_files_used": [], - "device_pch_files_created": [], - "device_pch_consumed": false, - "remote_seconds": 114.717228026 - }, - { - "case": "cutlass", - "variant": 0, - "mode": "patched_all_pch", - "repeat": 0, - "extension_name": "kb_nvcc_cutlass_0", - "returncode": 0, - "log": "[1/3] /tmp/tmp8ber6ya5/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp8ber6ya5/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp8ber6ya5/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmp8ber6ya5/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n[3/3] /tmp/tmp8ber6ya5/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", - "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", - "container_id": "ta-01M7GGH8YZHK6M1B0FWN50S9BR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", - "process_seconds": 36.453414843000004, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp8ber6ya5/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.150167573, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmp8ber6ya5/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch" - ], - "seconds": 31.500741393, - "returncode": 0 - } - ], - "seconds": 32.528592897, - "outputs": [ - 8.0 - ], - "load_inline_seconds": 32.528590723, - "prefetch_seconds": 0, - "prefetch_bytes": 0, - "max_absolute_error": 4.2646570364013314e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmp8ber6ya5/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmp8ber6ya5/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp8ber6ya5/cuda.cu\" --orig_src_path_name \"/tmp/tmp8ber6ya5/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp8ber6ya5/cuda.cu\" --orig_src_path_name \"/tmp/tmp8ber6ya5/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmp8ber6ya5/diagnostic.cuda.o\" \n", - "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmp8ber6ya5/cuda.cu , gcc (preprocessing 1) , /tmp/tmp8ber6ya5/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 3192.6299 , ms\n/tmp/tmp8ber6ya5/cuda.cu , gcc (preprocessing 4) , /tmp/tmp8ber6ya5/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 3075.6389 , ms\n/tmp/tmp8ber6ya5/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 5036.9600 , ms\n/tmp/tmp8ber6ya5/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 5079.5298 , ms\n/tmp/tmp8ber6ya5/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 320.1950 , ms\n/tmp/tmp8ber6ya5/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 5.3060 , ms\n/tmp/tmp8ber6ya5/cuda.cu , gcc (compiling) , , /tmp/tmp8ber6ya5/diagnostic.cuda.o , compute_75 , nvcc , 9192.4521 , ms\n , nvcc (driver) , , /tmp/tmp8ber6ya5/diagnostic.cuda.o , , nvcc , 96.7441 , ms\n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 71.035927235 - }, - { - "case": "cutlass", - "variant": 0, - "mode": "patched_all_pch_prefetch", - "repeat": 0, - "extension_name": "kb_nvcc_cutlass_0", - "returncode": 0, - "log": "[1/3] /tmp/tmpz4s1y_ge/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpz4s1y_ge/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpz4s1y_ge/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpz4s1y_ge/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n[3/3] /tmp/tmpz4s1y_ge/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_0.so\n", - "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", - "container_id": "ta-01M7GGJEA1CBKA78JQGA7X6YTR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", - "process_seconds": 34.245046614, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpz4s1y_ge/main.cpp", - "-o", - "main.o" - ], - "seconds": 7.674453068000002, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmpz4s1y_ge/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch" - ], - "seconds": 26.635627733, - "returncode": 0 - } - ], - "seconds": 30.342368112999996, - "outputs": [ - 8.0 - ], - "load_inline_seconds": 27.733861292, - "prefetch_seconds": 2.6085065270000003, - "prefetch_bytes": 2999697408, - "max_absolute_error": 4.2646570364013314e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpz4s1y_ge/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpz4s1y_ge/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpz4s1y_ge/cuda.cu\" --orig_src_path_name \"/tmp/tmpz4s1y_ge/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpz4s1y_ge/cuda.cu\" --orig_src_path_name \"/tmp/tmpz4s1y_ge/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpz4s1y_ge/diagnostic.cuda.o\" \n", - "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmpz4s1y_ge/cuda.cu , gcc (preprocessing 1) , /tmp/tmpz4s1y_ge/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 3394.4260 , ms\n/tmp/tmpz4s1y_ge/cuda.cu , gcc (preprocessing 4) , /tmp/tmpz4s1y_ge/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 3355.1160 , ms\n/tmp/tmpz4s1y_ge/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 4308.4219 , ms\n/tmp/tmpz4s1y_ge/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 5538.6201 , ms\n/tmp/tmpz4s1y_ge/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 329.7930 , ms\n/tmp/tmpz4s1y_ge/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 5.9670 , ms\n/tmp/tmpz4s1y_ge/cuda.cu , gcc (compiling) , , /tmp/tmpz4s1y_ge/diagnostic.cuda.o , compute_75 , nvcc , 9146.8916 , ms\n , nvcc (driver) , , /tmp/tmpz4s1y_ge/diagnostic.cuda.o , , nvcc , 92.5684 , ms\n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 68.732929376 - }, - { - "case": "cutlass", - "variant": 1, - "mode": "patched_all_pch_prefetch", - "repeat": 0, - "extension_name": "kb_nvcc_cutlass_1", - "returncode": 0, - "log": "[1/3] /tmp/tmpw_st2x9m/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpw_st2x9m/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpw_st2x9m/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpw_st2x9m/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n[3/3] /tmp/tmpw_st2x9m/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", - "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", - "container_id": "ta-01M7GGKN4QWSM62X7YPQY4B47R", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", - "process_seconds": 44.962129579, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpw_st2x9m/main.cpp", - "-o", - "main.o" - ], - "seconds": 9.114189858, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmpw_st2x9m/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch" - ], - "seconds": 31.421957751, - "returncode": 0 - } - ], - "seconds": 40.536378983999995, - "outputs": [ - 16.0 - ], - "load_inline_seconds": 32.801416029, - "prefetch_seconds": 7.734961381, - "prefetch_bytes": 2999697408, - "max_absolute_error": 8.529314072802663e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpw_st2x9m/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpw_st2x9m/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw_st2x9m/cuda.cu\" --orig_src_path_name \"/tmp/tmpw_st2x9m/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpw_st2x9m/cuda.cu\" --orig_src_path_name \"/tmp/tmpw_st2x9m/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpw_st2x9m/diagnostic.cuda.o\" \n", - "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmpw_st2x9m/cuda.cu , gcc (preprocessing 1) , /tmp/tmpw_st2x9m/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 4109.4219 , ms\n/tmp/tmpw_st2x9m/cuda.cu , gcc (preprocessing 4) , /tmp/tmpw_st2x9m/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 4190.4790 , ms\n/tmp/tmpw_st2x9m/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 4666.5859 , ms\n/tmp/tmpw_st2x9m/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 6280.6641 , ms\n/tmp/tmpw_st2x9m/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 381.9780 , ms\n/tmp/tmpw_st2x9m/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 7.4310 , ms\n/tmp/tmpw_st2x9m/cuda.cu , gcc (compiling) , , /tmp/tmpw_st2x9m/diagnostic.cuda.o , compute_75 , nvcc , 9935.6123 , ms\n , nvcc (driver) , , /tmp/tmpw_st2x9m/diagnostic.cuda.o , , nvcc , 106.5527 , ms\n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 83.664074216 - }, - { - "case": "cutlass", - "variant": 1, - "mode": "patched_all_pch", - "repeat": 0, - "extension_name": "kb_nvcc_cutlass_1", - "returncode": 0, - "log": "[1/3] /tmp/tmpcylmnezm/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpcylmnezm/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpcylmnezm/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpcylmnezm/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000031_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n\"/tmp/tmpxft_00000031_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n[3/3] /tmp/tmpcylmnezm/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", - "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", - "container_id": "ta-01M7GGM20MGZZK7K4NM6A452NR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", - "process_seconds": 37.669464049, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpcylmnezm/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.499430686, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmpcylmnezm/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch" - ], - "seconds": 32.01834994, - "returncode": 0 - } - ], - "seconds": 33.270577796, - "outputs": [ - 16.0 - ], - "load_inline_seconds": 33.270573987, - "prefetch_seconds": 0, - "prefetch_bytes": 0, - "max_absolute_error": 8.529314072802663e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpcylmnezm/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh\" -m64 \"/tmp/tmpcylmnezm/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpcylmnezm/cuda.cu\" --orig_src_path_name \"/tmp/tmpcylmnezm/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpcylmnezm/cuda.cu\" --orig_src_path_name \"/tmp/tmpcylmnezm/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpcylmnezm/diagnostic.cuda.o\" \n", - "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmpcylmnezm/cuda.cu , gcc (preprocessing 1) , /tmp/tmpcylmnezm/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 4109.8291 , ms\n/tmp/tmpcylmnezm/cuda.cu , gcc (preprocessing 4) , /tmp/tmpcylmnezm/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 4238.2021 , ms\n/tmp/tmpcylmnezm/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 4641.4351 , ms\n/tmp/tmpcylmnezm/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 6262.5859 , ms\n/tmp/tmpcylmnezm/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 390.5320 , ms\n/tmp/tmpcylmnezm/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 7.3630 , ms\n/tmp/tmpcylmnezm/cuda.cu , gcc (compiling) , , /tmp/tmpcylmnezm/diagnostic.cuda.o , compute_75 , nvcc , 9785.2178 , ms\n , nvcc (driver) , , /tmp/tmpcylmnezm/diagnostic.cuda.o , , nvcc , 108.5156 , ms\n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 77.071578788 - }, - { - "case": "cutlass", - "variant": 1, - "mode": "patched_host_pch", - "repeat": 0, - "extension_name": "kb_nvcc_cutlass_1", - "returncode": 0, - "log": "[1/3] /tmp/tmpanf9brnl/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpanf9brnl/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpanf9brnl/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -U__CUDA_NO_HALF_OPERATORS__ -U__CUDA_NO_HALF_CONVERSIONS__ -U__CUDA_NO_HALF2_OPERATORS__ -U__CUDA_NO_BFLOAT16_CONVERSIONS__ --expt-relaxed-constexpr -O3 -I/opt/cutlass/include -I/opt/cutlass/tools/util/include -std=c++20 -c /tmp/tmpanf9brnl/cuda.cu -o cuda.cuda.o \n[3/3] /tmp/tmpanf9brnl/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_cutlass_1.so\n", - "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", - "container_id": "ta-01M7GGMR8RYJMZVXGZ6MZKK5ER", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": null, - "process_seconds": 53.932866476, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpanf9brnl/main.cpp", - "-o", - "main.o" - ], - "seconds": 8.41651323, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_cutlass_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "-c", - "/tmp/tmpanf9brnl/cuda.cu", - "-o", - "cuda.cuda.o" - ], - "seconds": 48.115408051, - "returncode": 0 - } - ], - "seconds": 49.549131939, - "outputs": [ - 16.0 - ], - "load_inline_seconds": 49.549129623, - "prefetch_seconds": 0, - "prefetch_bytes": 0, - "max_absolute_error": 8.529314072802663e-06, - "correctness_checks": 10, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpanf9brnl/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC -O3 -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_cutlass_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_HALF_OPERATORS__\" -U \"__CUDA_NO_HALF_CONVERSIONS__\" -U \"__CUDA_NO_HALF2_OPERATORS__\" -U \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpanf9brnl/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpanf9brnl/cuda.cu\" --orig_src_path_name \"/tmp/tmpanf9brnl/cuda.cu\" --allow_managed --relaxed_constexpr --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ \"$CICC_PATH/cicc\" --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpanf9brnl/cuda.cu\" --orig_src_path_name \"/tmp/tmpanf9brnl/cuda.cu\" --allow_managed --relaxed_constexpr -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -O3 -Wno-psabi -I\"/opt/cutlass/include\" -I\"/opt/cutlass/tools/util/include\" \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpanf9brnl/diagnostic.cuda.o\" \n", - "phase_csv": "source file name , phase name , phase input files , phase output file , arch , tool, metric , unit\n/tmp/tmpanf9brnl/cuda.cu , gcc (preprocessing 1) , /tmp/tmpanf9brnl/cuda.cu , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , compute_75 , nvcc , 4388.6689 , ms\n/tmp/tmpanf9brnl/cuda.cu , gcc (preprocessing 4) , /tmp/tmpanf9brnl/cuda.cu , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , , nvcc , 4251.5449 , ms\n/tmp/tmpanf9brnl/cuda.cu , cudafe++ , /tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii , /tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp , compute_75 , nvcc , 17644.5254 , ms\n/tmp/tmpanf9brnl/cuda.cu , cicc , /tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , compute_75 , nvcc , 10723.0273 , ms\n/tmp/tmpanf9brnl/cuda.cu , ptxas , /tmp/tmpxft_00000057_00000000-6_cuda.ptx , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , sm_75 , nvcc , 391.4860 , ms\n/tmp/tmpanf9brnl/cuda.cu , fatbinary , /tmp/tmpxft_00000057_00000000-8_cuda.cubin , /tmp/tmpxft_00000057_00000000-3_cuda.fatbin , , nvcc , 7.5860 , ms\n/tmp/tmpanf9brnl/cuda.cu , gcc (compiling) , , /tmp/tmpanf9brnl/diagnostic.cuda.o , compute_75 , nvcc , 9803.3232 , ms\n , nvcc (driver) , , /tmp/tmpanf9brnl/diagnostic.cuda.o , , nvcc , 102.2070 , ms\n" - }, - "device_pch_files_used": [], - "device_pch_files_created": [], - "device_pch_consumed": false, - "remote_seconds": 112.023578118 - } - ] - }, - "production_cuda133": [ - { - "case": "inline_simple", - "variant": 0, - "enabled": false, - "repeat": 0, - "extension_name": "pch_matrix_inline_simple_0_0", - "seconds": 21.796966691, - "timing_samples": [ - 21.796966691 - ], - "host_compile_seconds": 20.979295318000002, - "pch_consumed": false, - "hit_keys": [], - "cache_miss": false, - "correctness_checks": 6, - "outputs": [ - 2.0, - 2.0, - 2.0, - 2.0, - 2.0 - ], - "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", - "container_id": "ta-01M7GFSN13SKZYRAX01JEAZB2R", - "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_seconds": [ - 3.0216056989999984 - ], - "max_absolute_error": 0, - "remote_seconds": 45.022717201000006 - }, - { - "case": "inline_simple", - "variant": 0, - "enabled": true, - "repeat": 0, - "extension_name": "pch_matrix_inline_simple_0_1", - "seconds": 10.092071702999998, - "timing_samples": [ - 10.092071702999998 - ], - "host_compile_seconds": 8.931718365, - "pch_consumed": true, - "hit_keys": [ - "cd04740b9ab36ac9a2a8d533e50fc30250ab89f341186829da169aefb0865f38" - ], - "cache_miss": false, - "correctness_checks": 6, - "outputs": [ - 2.0, - 2.0, - 2.0, - 2.0, - 2.0 - ], - "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", - "container_id": "ta-01M7GFSN1CJ9VQMYM83YJZ6YCR", - "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_seconds": [ - 2.8047221360000005 - ], - "max_absolute_error": 0, - "remote_seconds": 22.803371427000002 - }, - { - "case": "inline_simple", - "variant": 1, - "enabled": true, - "repeat": 0, - "extension_name": "pch_matrix_inline_simple_1_1", - "seconds": 10.122365914999998, - "timing_samples": [ - 10.122365914999998 - ], - "host_compile_seconds": 8.959541608, - "pch_consumed": true, - "hit_keys": [ - "cd04740b9ab36ac9a2a8d533e50fc30250ab89f341186829da169aefb0865f38" - ], - "cache_miss": false, - "correctness_checks": 6, - "outputs": [ - 3.0, - 3.0, - 3.0, - 3.0, - 3.0 - ], - "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", - "container_id": "ta-01M7GFSN13X2GVZ36HQVSN6S2R", - "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_seconds": [ - 2.7680481429999997 - ], - "max_absolute_error": 0, - "remote_seconds": 22.595921241 - }, - { - "case": "inline_simple", - "variant": 1, - "enabled": false, - "repeat": 0, - "extension_name": "pch_matrix_inline_simple_1_0", - "seconds": 19.861022294999998, - "timing_samples": [ - 19.861022294999998 - ], - "host_compile_seconds": 19.03845912, - "pch_consumed": false, - "hit_keys": [], - "cache_miss": false, - "correctness_checks": 6, - "outputs": [ - 3.0, - 3.0, - 3.0, - 3.0, - 3.0 - ], - "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", - "container_id": "ta-01M7GFSPSNZV6D195BH3PEWKYR", - "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_seconds": [ - 2.0814953679999997 - ], - "max_absolute_error": 0, - "remote_seconds": 41.715831361 - }, - { - "case": "cutlass_gemm_implicit", - "variant": 0, - "enabled": false, - "repeat": 0, - "extension_name": "pch_matrix_cutlass_gemm_implicit_0_0", - "seconds": 54.753927797, - "timing_samples": [ - 54.753927797 - ], - "host_compile_seconds": 23.725823061, - "pch_consumed": false, - "hit_keys": [], - "cache_miss": false, - "correctness_checks": 10, - "outputs": [ - 8.0 - ], - "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", - "container_id": "ta-01M7GFTQD7VSWT2AGTQ9041GCR", - "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_seconds": [ - 53.95127715 - ], - "max_absolute_error": 4.2646570364013314e-06, - "remote_seconds": 77.437517225 - }, - { - "case": "cutlass_gemm_implicit", - "variant": 0, - "enabled": true, - "repeat": 0, - "extension_name": "pch_matrix_cutlass_gemm_implicit_0_1", - "seconds": 55.171705347, - "timing_samples": [ - 55.171705347 - ], - "host_compile_seconds": 7.705170879000001, - "pch_consumed": true, - "hit_keys": [ - "cd04740b9ab36ac9a2a8d533e50fc30250ab89f341186829da169aefb0865f38" - ], - "cache_miss": false, - "correctness_checks": 10, - "outputs": [ - 8.0 - ], - "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", - "container_id": "ta-01M7GFTQJPYCBMYG758BQ09FXR", - "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_seconds": [ - 54.050255391 - ], - "max_absolute_error": 4.2646570364013314e-06, - "remote_seconds": 66.164975964 - }, - { - "case": "cutlass_gemm_implicit", - "variant": 1, - "enabled": true, - "repeat": 0, - "extension_name": "pch_matrix_cutlass_gemm_implicit_1_1", - "seconds": 46.287869813, - "timing_samples": [ - 46.287869813 - ], - "host_compile_seconds": 7.971731351, - "pch_consumed": true, - "hit_keys": [ - "cd04740b9ab36ac9a2a8d533e50fc30250ab89f341186829da169aefb0865f38" - ], - "cache_miss": false, - "correctness_checks": 10, - "outputs": [ - 16.0 - ], - "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", - "container_id": "ta-01M7GFVD2473KRK7WSX8NNEP2R", - "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_seconds": [ - 45.135409772 - ], - "max_absolute_error": 8.529314072802663e-06, - "remote_seconds": 58.033208515 - }, - { - "case": "cutlass_gemm_implicit", - "variant": 1, - "enabled": false, - "repeat": 0, - "extension_name": "pch_matrix_cutlass_gemm_implicit_1_0", - "seconds": 46.312597678, - "timing_samples": [ - 46.312597678 - ], - "host_compile_seconds": 19.034647205000002, - "pch_consumed": false, - "hit_keys": [], - "cache_miss": false, - "correctness_checks": 10, - "outputs": [ - 16.0 - ], - "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", - "container_id": "ta-01M7GFVEKV0YSM0R3DR3S4RAER", - "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_seconds": [ - 45.396892644999994 - ], - "max_absolute_error": 8.529314072802663e-06, - "remote_seconds": 68.968569437 - }, - { - "case": "inline_simple", - "variant": 0, - "enabled": false, - "repeat": 1, - "extension_name": "pch_matrix_inline_simple_0_0", - "seconds": 19.752127795, - "timing_samples": [ - 19.752127795 - ], - "host_compile_seconds": 19.002610604, - "pch_consumed": false, - "hit_keys": [], - "cache_miss": false, - "correctness_checks": 6, - "outputs": [ - 2.0, - 2.0, - 2.0, - 2.0, - 2.0 - ], - "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", - "container_id": "ta-01M7GFWZZJ4FJB7NPRQJTQ8JSR", - "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_seconds": [ - 1.7609109730000005 - ], - "max_absolute_error": 0, - "remote_seconds": 42.332975843999996 - }, - { - "case": "inline_simple", - "variant": 0, - "enabled": true, - "repeat": 1, - "extension_name": "pch_matrix_inline_simple_0_1", - "seconds": 11.21638811, - "timing_samples": [ - 11.21638811 - ], - "host_compile_seconds": 10.272402451000001, - "pch_consumed": true, - "hit_keys": [ - "cd04740b9ab36ac9a2a8d533e50fc30250ab89f341186829da169aefb0865f38" - ], - "cache_miss": false, - "correctness_checks": 6, - "outputs": [ - 2.0, - 2.0, - 2.0, - 2.0, - 2.0 - ], - "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", - "container_id": "ta-01M7GG0KV2AHWV84NN3ECQX0QR", - "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_seconds": [ - 1.6058505829999996 - ], - "max_absolute_error": 0, - "remote_seconds": 22.183825709 - }, - { - "case": "inline_simple", - "variant": 1, - "enabled": true, - "repeat": 1, - "extension_name": "pch_matrix_inline_simple_1_1", - "seconds": 10.608822302, - "timing_samples": [ - 10.608822302 - ], - "host_compile_seconds": 9.567511885, - "pch_consumed": true, - "hit_keys": [ - "cd04740b9ab36ac9a2a8d533e50fc30250ab89f341186829da169aefb0865f38" - ], - "cache_miss": false, - "correctness_checks": 6, - "outputs": [ - 3.0, - 3.0, - 3.0, - 3.0, - 3.0 - ], - "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", - "container_id": "ta-01M7GG0KV2QXH009EK3QFRJ3KR", - "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_seconds": [ - 1.722032843 - ], - "max_absolute_error": 0, - "remote_seconds": 21.330559701 - }, - { - "case": "inline_simple", - "variant": 1, - "enabled": false, - "repeat": 1, - "extension_name": "pch_matrix_inline_simple_1_0", - "seconds": 21.62613593, - "timing_samples": [ - 21.62613593 - ], - "host_compile_seconds": 20.728185403, - "pch_consumed": false, - "hit_keys": [], - "cache_miss": false, - "correctness_checks": 6, - "outputs": [ - 3.0, - 3.0, - 3.0, - 3.0, - 3.0 - ], - "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", - "container_id": "ta-01M7GG0KP789FE1B50S50BZD8R", - "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_seconds": [ - 1.7904973860000002 - ], - "max_absolute_error": 0, - "remote_seconds": 45.89947318 - }, - { - "case": "cutlass_gemm_implicit", - "variant": 0, - "enabled": false, - "repeat": 1, - "extension_name": "pch_matrix_cutlass_gemm_implicit_0_0", - "seconds": 51.057156396, - "timing_samples": [ - 51.057156396 - ], - "host_compile_seconds": 20.795003843, - "pch_consumed": false, - "hit_keys": [], - "cache_miss": false, - "correctness_checks": 10, - "outputs": [ - 8.0 - ], - "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", - "container_id": "ta-01M7GG0KP7MXTNR2TTPF0PZM5R", - "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_seconds": [ - 50.191617069 - ], - "max_absolute_error": 4.2646570364013314e-06, - "remote_seconds": 76.48378077699999 - }, - { - "case": "cutlass_gemm_implicit", - "variant": 0, - "enabled": true, - "repeat": 1, - "extension_name": "pch_matrix_cutlass_gemm_implicit_0_1", - "seconds": 43.304864159, - "timing_samples": [ - 43.304864159 - ], - "host_compile_seconds": 7.688989933999999, - "pch_consumed": true, - "hit_keys": [ - "cd04740b9ab36ac9a2a8d533e50fc30250ab89f341186829da169aefb0865f38" - ], - "cache_miss": false, - "correctness_checks": 10, - "outputs": [ - 8.0 - ], - "source_sha256": "c0cc1e80088a516e7d3b0a21890de234e88d4cf0b2b278c6e8e74d7912d4a14d", - "container_id": "ta-01M7GG1H2G9JENDFKF2E6J86FR", - "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_seconds": [ - 42.357097376 - ], - "max_absolute_error": 4.2646570364013314e-06, - "remote_seconds": 55.385258896 - }, - { - "case": "cutlass_gemm_implicit", - "variant": 1, - "enabled": true, - "repeat": 1, - "extension_name": "pch_matrix_cutlass_gemm_implicit_1_1", - "seconds": 44.357726661, - "timing_samples": [ - 44.357726661 - ], - "host_compile_seconds": 7.677523335, - "pch_consumed": true, - "hit_keys": [ - "cd04740b9ab36ac9a2a8d533e50fc30250ab89f341186829da169aefb0865f38" - ], - "cache_miss": false, - "correctness_checks": 10, - "outputs": [ - 16.0 - ], - "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", - "container_id": "ta-01M7GG1H2G5ZZTD7T5RB3RJ71R", - "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_seconds": [ - 43.322019976 - ], - "max_absolute_error": 8.529314072802663e-06, - "remote_seconds": 55.589907571000005 - }, - { - "case": "cutlass_gemm_implicit", - "variant": 1, - "enabled": false, - "repeat": 1, - "extension_name": "pch_matrix_cutlass_gemm_implicit_1_0", - "seconds": 46.449484709, - "timing_samples": [ - 46.449484709 - ], - "host_compile_seconds": 19.092076731, - "pch_consumed": false, - "hit_keys": [], - "cache_miss": false, - "correctness_checks": 10, - "outputs": [ - 16.0 - ], - "source_sha256": "b926f64a92cfd59840083d417ec09e2db0dd93574c97203ad4ac0d08925ccaee", - "container_id": "ta-01M7GG27HD6MS78NKS8C24741R", - "image_id": "im-Mzzu4ioxJcWCYCVbBhpB2X", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_seconds": [ - 45.656016197 - ], - "max_absolute_error": 8.529314072802663e-06, - "remote_seconds": 69.458907044 - } - ], - "reuse_smoke": { - "probe": { - "stock_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patched_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "paths": "/usr/local/cuda/nvvm/bin/cicc\n/usr/local/cuda/bin/nvcc.profile\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "nvcc_profiles": { - "/usr/local/cuda/bin/nvcc.profile": "\nTOP = $(_HERE_)/..\n\nCICC_PATH = $(TOP)/nvvm-next/bin\nNVVMIR_LIBRARY_DIR = $(TOP)/nvvm/libdevice\n\nLD_LIBRARY_PATH += $(TOP)/lib:\nPATH += $(CICC_PATH):$(_HERE_):\n\nINCLUDES += \"-I$(TOP)/$(_TARGET_DIR_)/include\" $(_SPACE_)\nSYSTEM_INCLUDES += \"-isystem\" \"$(TOP)/$(_TARGET_DIR_)/include/cccl\" $(_SPACE_)\n\nLIBRARIES =+ $(_SPACE_) \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)/stubs\" \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)\"\n\nCUDAFE_FLAGS +=\nPTXAS_FLAGS +=\n" - }, - "pch_help": " Specify options directly to library manager.\n\n--pch --pch \n Enable precompiled header (PCH) processing. When specified, each compilation\n searches the PCH directory (see --pch-dir) for a reusable precompiled header,\n and creates a new one when none is found.\n\n--pch-dir --pch-dir \n Specify the directory in which precompiled header files are read from and\n written to. By default this is the current working directory. Ignored unless\n --pch, --create-pch, or --use-pch is enabled.\n\n--pch-verbose --pch-verbose \n Print (to stderr) messages related to precompiled header processing. Ignored\n unless --pch, --create-pch, or --use-pch is enabled.\n\n--create-pch --create-pch \n Manually create a precompiled header (PCH) at the first PCH terminal stop-point\n (i.e., #pragma hdrstop) and save it under the specified name. Intermediate\n stop points (#pragma hdrstop++) are ignored. The directory into which the\n PCH file is written is determined by --pch-dir (or is the current working\n directory if not specified). This option forces a PCH file to be created\n (any existing PCH file with the same name is overwritten). This option is\n mutually exclusive with '--use-pch' and '--pch'.\n\n--use-pch --use-pch \n If a precompiled header (PCH) with the specified name exists in the directory\n determined by '--pch-dir' (or the current working directory if not specified),\n try to use it in compilation.This option is mutually exclusive with '--create-pch'\n and '--pch'.\n\n--ptxas-options ,... (-Xptxas) \n Specify options directly to ptxas, the PTX optimizing assembler.\n\n--nvlink-options ,... (-Xnvlink) \n Specify options directly to nvlink.\n\n--nvprune-options ,... (-Xnvprune) \n Specify options directly to nvprune (used with -prune).\n\n\nMiscellaneous options for guiding the compiler driver.\n======================================================", - "strace": null, - "smoke": [ - { - "arch": "sm_75", - "command": [ - "/usr/local/cuda/bin/nvcc", - "-arch=sm_75", - "-c", - "/tmp/tmp0a2ctqns/smoke.cu", - "-o", - "/tmp/tmp0a2ctqns/sm_75.o" - ], - "dryrun_returncode": 0, - "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmp0a2ctqns/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp0a2ctqns/smoke.cu\" --orig_src_path_name \"/tmp/tmp0a2ctqns/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmp0a2ctqns/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp0a2ctqns/smoke.cu\" --orig_src_path_name \"/tmp/tmp0a2ctqns/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_0000001e_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_0000001e_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_0000001e_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmp0a2ctqns/sm_75.o\" \n", - "next_backend_selected": true, - "returncode": 0, - "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmp0a2ctqns/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp0a2ctqns/smoke.cu\" --orig_src_path_name \"/tmp/tmp0a2ctqns/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmp0a2ctqns/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp0a2ctqns/smoke.cu\" --orig_src_path_name \"/tmp/tmp0a2ctqns/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000021_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000021_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000021_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmp0a2ctqns/sm_75.o\" \n", - "object_bytes": 10200 - }, - { - "arch": "sm_100", - "command": [ - "/usr/local/cuda/bin/nvcc", - "-arch=sm_100", - "-c", - "/tmp/tmp0a2ctqns/smoke.cu", - "-o", - "/tmp/tmp0a2ctqns/sm_100.o" - ], - "dryrun_returncode": 0, - "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmp0a2ctqns/smoke.cu\" -o \"/tmp/tmpxft_00000036_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp0a2ctqns/smoke.cu\" --orig_src_path_name \"/tmp/tmp0a2ctqns/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000036_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000036_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000036_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmp0a2ctqns/smoke.cu\" -o \"/tmp/tmpxft_00000036_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp0a2ctqns/smoke.cu\" --orig_src_path_name \"/tmp/tmp0a2ctqns/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000036_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000036_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000036_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000036_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000036_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000036_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000036_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000036_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000036_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000036_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmp0a2ctqns/sm_100.o\" \n", - "next_backend_selected": true, - "returncode": 0, - "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmp0a2ctqns/smoke.cu\" -o \"/tmp/tmpxft_00000039_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp0a2ctqns/smoke.cu\" --orig_src_path_name \"/tmp/tmp0a2ctqns/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000039_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000039_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000039_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmp0a2ctqns/smoke.cu\" -o \"/tmp/tmpxft_00000039_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp0a2ctqns/smoke.cu\" --orig_src_path_name \"/tmp/tmp0a2ctqns/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000039_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000039_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000039_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000039_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000039_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000039_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000039_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000039_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000039_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000039_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmp0a2ctqns/sm_100.o\" \n", - "object_bytes": 12504 - } - ], - "compatible": true - }, - "rows": [ - { - "case": "simple", - "variant": 0, - "mode": "patched_all_pch", - "repeat": 0, - "extension_name": "kb_nvcc_simple_0", - "returncode": 0, - "log": "[1/3] /tmp/tmp8ze6n0e1/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmp8ze6n0e1/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmp8ze6n0e1/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmp8ze6n0e1/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmp8ze6n0e1/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_0.so\n", - "source_sha256": "e1725cde5c1d7ecb860a63e4b5da483ae9b2b8971558238aac98bc5224f81897", - "container_id": "ta-01M7GG5X2J0T5Z7DF200Z9AWXR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", - "process_seconds": 33.416754872, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp8ze6n0e1/main.cpp", - "-o", - "main.o" - ], - "seconds": 10.638259281000002, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_0", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmp8ze6n0e1/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" - ], - "seconds": 27.723024821, - "returncode": 0 - } - ], - "seconds": 28.746708839999997, - "outputs": [ - 2.0, - 2.0, - 2.0, - 2.0, - 2.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp8ze6n0e1/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_0\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmp8ze6n0e1/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp8ze6n0e1/cuda.cu\" --orig_src_path_name \"/tmp/tmp8ze6n0e1/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmp8ze6n0e1/cuda.cu\" --orig_src_path_name \"/tmp/tmp8ze6n0e1/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmp8ze6n0e1/diagnostic.cuda.o\" \n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 65.051380577 - }, - { - "case": "simple", - "variant": 1, - "mode": "patched_all_pch", - "repeat": 0, - "extension_name": "kb_nvcc_simple_1", - "returncode": 0, - "log": "[1/3] /tmp/tmpzsqcjep0/observe-cxx -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpzsqcjep0/main.cpp -o main.o \n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/3] /tmp/tmpzsqcjep0/observe-nvcc -MD -MF cuda.cuda.o.d -DTORCH_EXTENSION_NAME=kb_nvcc_simple_1 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 --compiler-options '-fPIC' -std=c++20 -c /tmp/tmpzsqcjep0/cuda.cu -o cuda.cuda.o \n\"/tmp/tmpxft_00000032_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n\"/tmp/tmpxft_00000032_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n[3/3] /tmp/tmpzsqcjep0/observe-cxx main.o cuda.cuda.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kb_nvcc_simple_1.so\n", - "source_sha256": "952bded8170418af7c3100b0089ad48753da78bb142adf39bde626a965ccebba", - "container_id": "ta-01M7GG5X2JWF1E5D985MF82VDR", - "image_id": "im-f00RGIxc0I83yIrG8DurE1", - "gpu": "Tesla T4", - "torch": "2.12.0+cu130", - "nvcc_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "pch_entry": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", - "process_seconds": 33.46370826, - "cxx_commands": [ - { - "args": [ - "-MMD", - "-MF", - "main.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpzsqcjep0/main.cpp", - "-o", - "main.o" - ], - "seconds": 10.227091994, - "returncode": 0 - } - ], - "nvcc_commands": [ - { - "args": [ - "-MD", - "-MF", - "cuda.cuda.o.d", - "-DTORCH_EXTENSION_NAME=kb_nvcc_simple_1", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "-c", - "/tmp/tmpzsqcjep0/cuda.cu", - "-o", - "cuda.cuda.o", - "-include", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", - "--pch", - "--pch-dir", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch" - ], - "seconds": 27.899412197000004, - "returncode": 0 - } - ], - "seconds": 28.910585837000003, - "outputs": [ - 3.0, - 3.0, - 3.0, - 3.0, - 3.0 - ], - "max_absolute_error": 0, - "correctness_checks": 6, - "host_pch_consumed": true, - "host_hit_keys": [ - "3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087" - ], - "nvcc_diagnostic": { - "returncode": 0, - "stdout": "", - "stderr": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmpzsqcjep0/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" \n#$ gcc -std=c++20 -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ -D__CUDACC_RELAXED_CONSTEXPR__ -fPIC \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D \"TORCH_EXTENSION_NAME=kb_nvcc_simple_1\" -D \"TORCH_API_INCLUDE_EXTENSION_H\" -D \"__CUDA_NO_HALF_OPERATORS__\" -D \"__CUDA_NO_HALF_CONVERSIONS__\" -D \"__CUDA_NO_BFLOAT16_CONVERSIONS__\" -D \"__CUDA_NO_HALF2_OPERATORS__\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -include \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh\" -m64 \"/tmp/tmpzsqcjep0/cuda.cu\" -o \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n#$ -- Filter Dependencies -- > cuda.cuda.o.d\n#$ cudafe++ --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpzsqcjep0/cuda.cu\" --orig_src_path_name \"/tmp/tmpzsqcjep0/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cudafe_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" --stub_file_name \"tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --simt-only \"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\" \n\"/tmp/tmpxft_00000057_00000000-5_cuda.cpp4.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch\"\n#$ \"$CICC_PATH/cicc\" --mmap_address=0xff45670000 --c++20 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpzsqcjep0/cuda.cu\" --orig_src_path_name \"/tmp/tmpzsqcjep0/cuda.cu\" --allow_managed --relaxed_constexpr --extended_pch --pch --extended_pch_name=cicc_ \"--pch_dir=/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch\" --pch_messages --pch_verbose -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000057_00000000-3_cuda.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000057_00000000-4_cuda.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.gpu\" \"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\"\n\"/tmp/tmpxft_00000057_00000000-7_cuda.cpp1.ii\": using precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.ptx\" -o \"/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000057_00000000-8_cuda.cubin\" --embedded-fatbin=\"/tmp/tmpxft_00000057_00000000-3_cuda.fatbin.c\" \n#$ rm /tmp/tmpxft_00000057_00000000-3_cuda.fatbin\n#$ gcc -std=c++20 -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -fPIC -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include\" -isystem \"/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include\" -isystem \"/usr/local/cuda/include\" -isystem \"/usr/local/include/python3.13\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000057_00000000-6_cuda.cudafe1.cpp\" -o \"/tmp/tmpzsqcjep0/diagnostic.cuda.o\" \n" - }, - "device_pch_files_used": [ - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch" - ], - "device_pch_files_created": [], - "device_pch_consumed": true, - "remote_seconds": 64.866187693 - } - ] - }, - "warmup": { - "probe": { - "stock_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Tue_Sep_01_08:45:21_PDT_2026\nCuda compilation tools, release 13.4, V13.4.92\nBuild cuda_13.4.r13.4/compiler.38855100_0\n", - "patched_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "paths": "/usr/local/cuda/nvvm/bin/cicc\n/usr/local/cuda/bin/nvcc.profile\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "754cfcd08d966d73d0e8dc7e61b548979a76ad4eb6ec9fcc76aa477b23177c98" - } - ], - "nvcc_profiles": { - "/usr/local/cuda/bin/nvcc.profile": "\nTOP = $(_HERE_)/..\n\nCICC_PATH = $(TOP)/nvvm-next/bin\nNVVMIR_LIBRARY_DIR = $(TOP)/nvvm/libdevice\n\nLD_LIBRARY_PATH += $(TOP)/lib:\nPATH += $(CICC_PATH):$(_HERE_):\n\nINCLUDES += \"-I$(TOP)/$(_TARGET_DIR_)/include\" $(_SPACE_)\nSYSTEM_INCLUDES += \"-isystem\" \"$(TOP)/$(_TARGET_DIR_)/include/cccl\" $(_SPACE_)\n\nLIBRARIES =+ $(_SPACE_) \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)/stubs\" \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)\"\n\nCUDAFE_FLAGS +=\nPTXAS_FLAGS +=\n" - }, - "pch_help": " Specify options directly to library manager.\n\n--pch --pch \n Enable precompiled header (PCH) processing. When specified, each compilation\n searches the PCH directory (see --pch-dir) for a reusable precompiled header,\n and creates a new one when none is found.\n\n--pch-dir --pch-dir \n Specify the directory in which precompiled header files are read from and\n written to. By default this is the current working directory. Ignored unless\n --pch, --create-pch, or --use-pch is enabled.\n\n--pch-verbose --pch-verbose \n Print (to stderr) messages related to precompiled header processing. Ignored\n unless --pch, --create-pch, or --use-pch is enabled.\n\n--create-pch --create-pch \n Manually create a precompiled header (PCH) at the first PCH terminal stop-point\n (i.e., #pragma hdrstop) and save it under the specified name. Intermediate\n stop points (#pragma hdrstop++) are ignored. The directory into which the\n PCH file is written is determined by --pch-dir (or is the current working\n directory if not specified). This option forces a PCH file to be created\n (any existing PCH file with the same name is overwritten). This option is\n mutually exclusive with '--use-pch' and '--pch'.\n\n--use-pch --use-pch \n If a precompiled header (PCH) with the specified name exists in the directory\n determined by '--pch-dir' (or the current working directory if not specified),\n try to use it in compilation.This option is mutually exclusive with '--create-pch'\n and '--pch'.\n\n--ptxas-options ,... (-Xptxas) \n Specify options directly to ptxas, the PTX optimizing assembler.\n\n--nvlink-options ,... (-Xnvlink) \n Specify options directly to nvlink.\n\n--nvprune-options ,... (-Xnvprune) \n Specify options directly to nvprune (used with -prune).\n\n\nMiscellaneous options for guiding the compiler driver.\n======================================================", - "strace": null, - "smoke": [ - { - "arch": "sm_75", - "command": [ - "/usr/local/cuda/bin/nvcc", - "-arch=sm_75", - "-c", - "/tmp/tmps_9b9md0/smoke.cu", - "-o", - "/tmp/tmps_9b9md0/sm_75.o" - ], - "dryrun_returncode": 0, - "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmps_9b9md0/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmps_9b9md0/smoke.cu\" --orig_src_path_name \"/tmp/tmps_9b9md0/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmps_9b9md0/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmps_9b9md0/smoke.cu\" --orig_src_path_name \"/tmp/tmps_9b9md0/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_0000001e_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_0000001e_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_0000001e_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmps_9b9md0/sm_75.o\" \n", - "next_backend_selected": true, - "returncode": 0, - "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmps_9b9md0/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmps_9b9md0/smoke.cu\" --orig_src_path_name \"/tmp/tmps_9b9md0/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmps_9b9md0/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmps_9b9md0/smoke.cu\" --orig_src_path_name \"/tmp/tmps_9b9md0/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000021_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000021_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000021_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmps_9b9md0/sm_75.o\" \n", - "object_bytes": 10200 - }, - { - "arch": "sm_100", - "command": [ - "/usr/local/cuda/bin/nvcc", - "-arch=sm_100", - "-c", - "/tmp/tmps_9b9md0/smoke.cu", - "-o", - "/tmp/tmps_9b9md0/sm_100.o" - ], - "dryrun_returncode": 0, - "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmps_9b9md0/smoke.cu\" -o \"/tmp/tmpxft_00000036_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmps_9b9md0/smoke.cu\" --orig_src_path_name \"/tmp/tmps_9b9md0/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000036_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000036_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000036_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmps_9b9md0/smoke.cu\" -o \"/tmp/tmpxft_00000036_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmps_9b9md0/smoke.cu\" --orig_src_path_name \"/tmp/tmps_9b9md0/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000036_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000036_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000036_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000036_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000036_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000036_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000036_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000036_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000036_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000036_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000036_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmps_9b9md0/sm_100.o\" \n", - "next_backend_selected": true, - "returncode": 0, - "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm-next/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm-next/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmps_9b9md0/smoke.cu\" -o \"/tmp/tmpxft_00000039_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmps_9b9md0/smoke.cu\" --orig_src_path_name \"/tmp/tmps_9b9md0/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000039_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000039_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000039_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmps_9b9md0/smoke.cu\" -o \"/tmp/tmpxft_00000039_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmps_9b9md0/smoke.cu\" --orig_src_path_name \"/tmp/tmps_9b9md0/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000039_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000039_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000039_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000039_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000039_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000039_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000039_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000039_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000039_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000039_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000039_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmps_9b9md0/sm_100.o\" \n", - "object_bytes": 12504 - } - ], - "compatible": true - }, - "rows": [], - "warmup": { - "host_seconds": 72.258458109, - "host_log": "[1/2] /opt/kernelbot-pch/compiler.py -MMD -MF main.o.d -DTORCH_EXTENSION_NAME=kernelbot_pch_warm_1_0 -DTORCH_API_INCLUDE_EXTENSION_H -isystem /usr/local/lib/python3.13/site-packages/torch/include -isystem /usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include -isystem /usr/local/cuda/include -isystem /usr/local/include/python3.13 -fPIC -std=c++20 -c /tmp/tmpnw106ee9/main.cpp -o main.o \n[kernelbot-pch] built 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[kernelbot-pch] hit 3ed56c55bed9ebe922aad94b1f690838afa7c84c59de200cc7d9f22fcac1c087\n[2/2] /opt/kernelbot-pch/compiler.py main.o -shared -L/usr/local/lib/python3.13/site-packages/torch/lib -lc10 -lc10_cuda -ltorch_cpu -ltorch_cuda -ltorch -ltorch_python -L/usr/local/cuda/lib64 -lcudart -o kernelbot_pch_warm_1_0.so\nW1115 20:57:29.119000 26 site-packages/torch/utils/cpp_extension.py:140] No CUDA runtime is found, using CUDA_HOME='/usr/local/cuda'\n", - "device": [ - { - "case": "simple", - "key": "8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767", - "command": [ - "/usr/local/cuda/bin/nvcc", - "-DTORCH_EXTENSION_NAME=kb_nvcc_pch_warmer", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - "--pch-dir", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch", - "--pch", - "--pch-verbose", - "-x", - "cu", - "-c", - "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh", - "-o", - "/tmp/tmp_hj5fj_n/warm.o" - ], - "seconds": 44.738656711999994, - "stdout": "", - "stderr": "/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh:1:9: warning: #pragma once in main file\n 1 | #pragma once\n | ^~~~\n\"/tmp/tmpxft_00000046_00000000-5_gpumode.cpp4.ii\": creating precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch.tmp.77\"\n/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/gpumode.cuh:1:9: warning: #pragma once in main file\n 1 | #pragma once\n | ^~~~\n\"/tmp/tmpxft_00000046_00000000-7_gpumode.cpp1.ii\": creating precompiled header file \"/kernelbot-nvcc-pch/8d64863a141aedc9fcf48b817eb0490f6df680c4822c9b96a878f514cf3af767/pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch.tmp.82\"\n", - "returncode": 0, - "files": [ - { - "path": "pch/cicc_77bf1b54_e1da27a0_2616be65_5685f066_515d41a1_a286da56_a7ed5f5b_54a3d61d.pch", - "bytes": 1229062144 - }, - { - "path": "pch/cudafe_f9e840ce_8534098d_5e074ec1_55dfe246_732a305f_33e53e13_233aad6c_e9a4b621.pch", - "bytes": 1295241216 - } - ] - }, - { - "case": "cutlass", - "key": "9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97", - "command": [ - "/usr/local/cuda/bin/nvcc", - "-DTORCH_EXTENSION_NAME=kb_nvcc_pch_warmer", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include", - "-isystem", - "/usr/local/lib/python3.13/site-packages/torch/include/torch/csrc/api/include", - "-isystem", - "/usr/local/cuda/include", - "-isystem", - "/usr/local/include/python3.13", - "-D__CUDA_NO_HALF_OPERATORS__", - "-D__CUDA_NO_HALF_CONVERSIONS__", - "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", - "-D__CUDA_NO_HALF2_OPERATORS__", - "--expt-relaxed-constexpr", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", - "-std=c++20", - "--pch-dir", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch", - "--pch", - "--pch-verbose", - "-x", - "cu", - "-c", - "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh", - "-o", - "/tmp/tmpw_lkzwtz/warm.o" - ], - "seconds": 51.79863217900001, - "stdout": "", - "stderr": "/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh:1:9: warning: #pragma once in main file\n 1 | #pragma once\n | ^~~~\n\"/tmp/tmpxft_00000070_00000000-5_gpumode.cpp4.ii\": creating precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch.tmp.119\"\n/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/gpumode.cuh:1:9: warning: #pragma once in main file\n 1 | #pragma once\n | ^~~~\n\"/tmp/tmpxft_00000070_00000000-7_gpumode.cpp1.ii\": creating precompiled header file \"/kernelbot-nvcc-pch/9464c5d67bdbe683c43a0bf77bbbe522c4bcd98667195f8a7ede846afb0ffa97/pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch.tmp.124\"\n", - "returncode": 0, - "files": [ - { - "path": "pch/cicc_12a1b0ff_c22a04ca_8eaeb977_fc578b5a_95970d99_c4945202_c712c395_138525a4.pch", - "bytes": 1443643392 - }, - { - "path": "pch/cudafe_a25988fd_0cee8499_6f0ac52c_e3cfc779_6610693d_53cac836_77e36ea0_67c27b43.pch", - "bytes": 1556054016 - } - ] - } - ] - } - }, - "initial_cuda133_incompatibility": { - "probe": { - "stock_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Fri_Apr_24_07:22:02_PM_PDT_2026\nCuda compilation tools, release 13.3, V13.3.33\nBuild cuda_13.3.r13.3/compiler.37862127_0\n", - "patched_version": "nvcc: NVIDIA (R) Cuda compiler driver\nCopyright (c) 2005-2026 NVIDIA Corporation\nBuilt on Sat_Oct_17_21:27:34_UTC_2026\nCuda compilation tools, release 13.4, V13.4.99\nBuild r615_00.r13.4/compiler.39158877_3\n", - "paths": "/usr/local/cuda/bin/nvcc.profile\n/usr/local/cuda/nvvm/bin/cicc\n", - "patch": [ - { - "file": "/usr/local/cuda/bin/nvcc", - "replaced_existing": true, - "sha256": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298" - }, - { - "file": "/usr/local/cuda/bin/cudafe++", - "replaced_existing": true, - "sha256": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442" - }, - { - "file": "/usr/local/cuda/nvvm/bin/cicc", - "replaced_existing": true, - "sha256": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050" - }, - { - "file": "/usr/local/cuda/nvvm-next/bin/cicc", - "replaced_existing": false, - "sha256": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - { - "file": "/usr/local/cuda/bin/nvcc.profile", - "replaced_existing": true, - "sha256": "2cb235ee19bab06e9a280fa888d489c4ac964b40279db1f050e8fda2e9fe9f60" - } - ], - "nvcc_profiles": { - "/usr/local/cuda/bin/nvcc.profile": "\nTOP = $(_HERE_)/..\n\nCICC_PATH = $(TOP)/$(_NVVM_BRANCH_)/bin\nNVVMIR_LIBRARY_DIR = $(TOP)/nvvm/libdevice\n\nLD_LIBRARY_PATH += $(TOP)/lib:\nPATH += $(CICC_PATH):$(_HERE_):\n\nINCLUDES += \"-I$(TOP)/$(_TARGET_DIR_)/include\" $(_SPACE_)\nSYSTEM_INCLUDES += \"-isystem\" \"$(TOP)/$(_TARGET_DIR_)/include/cccl\" $(_SPACE_)\n\nLIBRARIES =+ $(_SPACE_) \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)/stubs\" \"-L$(TOP)/$(_TARGET_DIR_)/lib$(_TARGET_SIZE_)\"\n\nCUDAFE_FLAGS +=\nPTXAS_FLAGS +=\n" - }, - "pch_help": " Specify options directly to library manager.\n\n--pch --pch \n Enable precompiled header (PCH) processing. When specified, each compilation\n searches the PCH directory (see --pch-dir) for a reusable precompiled header,\n and creates a new one when none is found.\n\n--pch-dir --pch-dir \n Specify the directory in which precompiled header files are read from and\n written to. By default this is the current working directory. Ignored unless\n --pch, --create-pch, or --use-pch is enabled.\n\n--pch-verbose --pch-verbose \n Print (to stderr) messages related to precompiled header processing. Ignored\n unless --pch, --create-pch, or --use-pch is enabled.\n\n--create-pch --create-pch \n Manually create a precompiled header (PCH) at the first PCH terminal stop-point\n (i.e., #pragma hdrstop) and save it under the specified name. Intermediate\n stop points (#pragma hdrstop++) are ignored. The directory into which the\n PCH file is written is determined by --pch-dir (or is the current working\n directory if not specified). This option forces a PCH file to be created\n (any existing PCH file with the same name is overwritten). This option is\n mutually exclusive with '--use-pch' and '--pch'.\n\n--use-pch --use-pch \n If a precompiled header (PCH) with the specified name exists in the directory\n determined by '--pch-dir' (or the current working directory if not specified),\n try to use it in compilation.This option is mutually exclusive with '--create-pch'\n and '--pch'.\n\n--ptxas-options ,... (-Xptxas) \n Specify options directly to ptxas, the PTX optimizing assembler.\n\n--nvlink-options ,... (-Xnvlink) \n Specify options directly to nvlink.\n\n--nvprune-options ,... (-Xnvprune) \n Specify options directly to nvprune (used with -prune).\n\n\nMiscellaneous options for guiding the compiler driver.\n======================================================", - "strace": null, - "smoke": [ - { - "arch": "sm_75", - "command": [ - "/usr/local/cuda/bin/nvcc", - "-arch=sm_75", - "-c", - "/tmp/tmpyfjiygy_/smoke.cu", - "-o", - "/tmp/tmpyfjiygy_/sm_75.o" - ], - "dryrun_returncode": 0, - "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpyfjiygy_/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpyfjiygy_/smoke.cu\" --orig_src_path_name \"/tmp/tmpyfjiygy_/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_0000001e_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpyfjiygy_/smoke.cu\" -o \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpyfjiygy_/smoke.cu\" --orig_src_path_name \"/tmp/tmpyfjiygy_/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_0000001e_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_0000001e_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_0000001e_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=75,file=/tmp/tmpxft_0000001e_00000000-8_smoke.sm_75.cubin\" \"--image3=kind=ptx,sm=75,file=/tmp/tmpxft_0000001e_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_0000001e_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_0000001e_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_0000001e_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpyfjiygy_/sm_75.o\" \n", - "next_backend_selected": false, - "returncode": 255, - "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpyfjiygy_/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpyfjiygy_/smoke.cu\" --orig_src_path_name \"/tmp/tmpyfjiygy_/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000021_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=750 -D__CUDA_ARCH_LIST__=750 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpyfjiygy_/smoke.cu\" -o \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpyfjiygy_/smoke.cu\" --orig_src_path_name \"/tmp/tmpyfjiygy_/smoke.cu\" --allow_managed -arch compute_75 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000021_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000021_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000021_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000021_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_75 -m64 \"/tmp/tmpxft_00000021_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000021_00000000-8_smoke.sm_75.cubin\" \nptxas /tmp/tmpxft_00000021_00000000-6_smoke.ptx, line 9; fatal : Unsupported .version 9.4; current version is '9.3'\nptxas fatal : Ptx assembly aborted due to errors\n# --error 0xff --\n", - "object_bytes": 0 - }, - { - "arch": "sm_100", - "command": [ - "/usr/local/cuda/bin/nvcc", - "-arch=sm_100", - "-c", - "/tmp/tmpyfjiygy_/smoke.cu", - "-o", - "/tmp/tmpyfjiygy_/sm_100.o" - ], - "dryrun_returncode": 0, - "dryrun": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpyfjiygy_/smoke.cu\" -o \"/tmp/tmpxft_00000030_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpyfjiygy_/smoke.cu\" --orig_src_path_name \"/tmp/tmpyfjiygy_/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000030_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000030_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000030_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000030_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpyfjiygy_/smoke.cu\" -o \"/tmp/tmpxft_00000030_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpyfjiygy_/smoke.cu\" --orig_src_path_name \"/tmp/tmpyfjiygy_/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000030_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000030_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000030_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000030_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000030_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000030_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000030_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000030_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000030_00000000-8_smoke.sm_100.cubin\" \n#$ fatbinary -64 --cicc-cmdline=\"-ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 \" \"--image3=kind=elf,sm=100,file=/tmp/tmpxft_00000030_00000000-8_smoke.sm_100.cubin\" \"--image3=kind=ptx,sm=100,file=/tmp/tmpxft_00000030_00000000-6_smoke.ptx\" --embedded-fatbin=\"/tmp/tmpxft_00000030_00000000-3_smoke.fatbin.c\" \n#$ rm /tmp/tmpxft_00000030_00000000-3_smoke.fatbin\n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -c -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -Wno-psabi \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -m64 \"/tmp/tmpxft_00000030_00000000-6_smoke.cudafe1.cpp\" -o \"/tmp/tmpyfjiygy_/sm_100.o\" \n", - "next_backend_selected": false, - "returncode": 255, - "log": "#$ _NVVM_BRANCH_=nvvm\n#$ _SPACE_= \n#$ _CUDART_=cudart\n#$ CUDA_ROOT=/usr/local/cuda\n#$ _HERE_=/usr/local/cuda/bin\n#$ _THERE_=/usr/local/cuda/bin\n#$ _TARGET_SIZE_=\n#$ _TARGET_DIR_=\n#$ _TARGET_DIR_=targets/x86_64-linux\n#$ TOP=/usr/local/cuda/bin/..\n#$ CICC_PATH=/usr/local/cuda/bin/../nvvm/bin\n#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda/bin/../nvvm/libdevice\n#$ LD_LIBRARY_PATH=/usr/local/cuda/bin/../lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64\n#$ PATH=/usr/local/cuda/bin/../nvvm/bin:/usr/local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin\n#$ INCLUDES=\"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \n#$ SYSTEM_INCLUDES=\"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" \n#$ LIBRARIES= \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib/stubs\" \"-L/usr/local/cuda/bin/../targets/x86_64-linux/lib\"\n#$ CUDAFE_FLAGS=\n#$ PTXAS_FLAGS=\n#$ gcc -D__CUDA_ARCH_LIST__=1000 -E -x c++ -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpyfjiygy_/smoke.cu\" -o \"/tmp/tmpxft_00000033_00000000-5_smoke.cpp4.ii\" \n#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpyfjiygy_/smoke.cu\" --orig_src_path_name \"/tmp/tmpyfjiygy_/smoke.cu\" --allow_managed --m64 --parse_templates --gen_c_file_name \"/tmp/tmpxft_00000033_00000000-6_smoke.cudafe1.cpp\" --stub_file_name \"tmpxft_00000033_00000000-6_smoke.cudafe1.stub.c\" --gen_module_id_file --module_id_file_name \"/tmp/tmpxft_00000033_00000000-4_smoke.module_id\" --simt-only \"/tmp/tmpxft_00000033_00000000-5_smoke.cpp4.ii\" \n#$ gcc -D__CUDA_ARCH__=1000 -D__CUDA_ARCH_LIST__=1000 -E -x c++ -DCUDA_DOUBLE_MATH_FUNCTIONS -D__CUDACC__ -D__NVCC__ \"-I/usr/local/cuda/bin/../targets/x86_64-linux/include\" \"-isystem\" \"/usr/local/cuda/bin/../targets/x86_64-linux/include/cccl\" -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=4 -D__CUDACC_VER_BUILD__=99 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=4 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include \"cuda_runtime.h\" -m64 \"/tmp/tmpyfjiygy_/smoke.cu\" -o \"/tmp/tmpxft_00000033_00000000-7_smoke.cpp1.ii\" \n#$ \"$CICC_PATH/cicc\" --c++17 --static-host-stub --device-hidden-visibility --gnu_version=130300 --display_error_number --orig_src_file_name \"/tmp/tmpyfjiygy_/smoke.cu\" --orig_src_path_name \"/tmp/tmpyfjiygy_/smoke.cu\" --allow_managed -arch compute_100 -m64 --no-version-ident -ftz=0 -prec_div=1 -prec_sqrt=1 -fmad=1 --include_file_name \"tmpxft_00000033_00000000-3_smoke.fatbin.c\" -tused --module_id_file_name \"/tmp/tmpxft_00000033_00000000-4_smoke.module_id\" --gen_c_file_name \"/tmp/tmpxft_00000033_00000000-6_smoke.cudafe1.c\" --stub_file_name \"/tmp/tmpxft_00000033_00000000-6_smoke.cudafe1.stub.c\" --gen_device_file_name \"/tmp/tmpxft_00000033_00000000-6_smoke.cudafe1.gpu\" \"/tmp/tmpxft_00000033_00000000-7_smoke.cpp1.ii\" --simt-only -o \"/tmp/tmpxft_00000033_00000000-6_smoke.ptx\"\n#$ ptxas -arch=sm_100 -m64 \"/tmp/tmpxft_00000033_00000000-6_smoke.ptx\" -o \"/tmp/tmpxft_00000033_00000000-8_smoke.sm_100.cubin\" \nptxas /tmp/tmpxft_00000033_00000000-6_smoke.ptx, line 9; fatal : Unsupported .version 9.4; current version is '9.3'\nptxas fatal : Ptx assembly aborted due to errors\n# --error 0xff --\n", - "object_bytes": 0 - } - ], - "compatible": false - }, - "rows": [] - }, - "binary_sha256": { - "pid-drop/cudart/bin/x86_64_Linux_release/nvcc": "254c308d113e2cdd0159ddfc02b54630b36be2013c0b928b6513e4bae723a298", - "pid-drop/cudart/bin/x86_64_Linux_release/cudafe++": "f64ef97248879dccf4cd92aef02d56a8ea915fbc20abd288e63dd9c199f24442", - "pid-drop/cudart/bin/x86_64_Linux_release/nvvm/bin/cicc": "9b890cf57ba86bc2d3c56714cf24e7c5fe52d0919c3b13512a3eea75ac2c3050", - "pid-drop/cudart/bin/x86_64_Linux_release/nvvm-next/bin/cicc": "5cebc486361825fe2625031922b8b8d481556cc4051006f1abd0c8f440e8a1c6" - }, - "modal_runs": { - "primary": "https://modal.com/apps/gpu-mode/main/ap-sCbGWCCsGHByigNhMrPUYA", - "supplemental": "https://modal.com/apps/gpu-mode/main/ap-P9owg8O1R3RlDGy0ozOSE1", - "warmup": "https://modal.com/apps/gpu-mode/main/ap-ypnw7T8Xntoe9WGjwycFyh", - "reuse_smoke": "https://modal.com/apps/gpu-mode/main/ap-ya4TMM2KIT440BrHtTc77E", - "production_cuda133": "https://modal.com/apps/gpu-mode/main/ap-42e4FtDMIK2OGSSHHcAg51" - } -} diff --git a/docs/modal-pch-cache.md b/docs/modal-pch-cache.md index 4fabd3b0..3c8e7995 100644 --- a/docs/modal-pch-cache.md +++ b/docs/modal-pch-cache.md @@ -5,10 +5,9 @@ The runner sets `CXX` to a GCC wrapper that loads a precompiled changes. Leave `use_pch=False`; PyTorch's separate native PCH option is unnecessary. The submission path assumes `load_inline()`. -An additional, opt-in [NVCC framework header cache](nvcc-pch-cache.md) can cache -the CUDA-side Torch/runtime includes with NVIDIA's private compiler drop. +NVCC also caches the implicit CUDA-side framework headers by default (see below). -This host cache only precompiles the Torch host header. CUDA/CUTLASS templates, user headers, kernel +The host cache only precompiles the Torch host header. CUDA/CUTLASS templates, user headers, kernel objects, and shared libraries still compile in each submission container. Different sources and module names share a PCH when their host flags match; changing a user header after the Torch include still changes the compiled code. @@ -31,6 +30,36 @@ speedup. Set `CXX=/usr/bin/g++` before building to bypass the wrapper entirely. ## Deployment +### CUDA-side framework cache + +The runner uses NVIDIA's checksum-pinned NVCC 13.4.99 drop with the complete CUDA +13.4 toolkit. `PYTORCH_NVCC` automatically reuses `torch/types.h`, `cuda.h`, and +`cuda_runtime.h` from `kernelbot-nvcc-pch`. CUTLASS and user headers stay outside +this cache. The trusted warmer builds 18 profiles from six observed flag tuples; +submission workers mount them read-only. Unknown flags, custom compiler inputs, +and sources without the implicit header prefix compile normally. Corrupt caches +fall back to normal compilation. CPU artifact replay checks compiler identity +and user-header dependencies. + +Upload the private compiler drop once, without committing binaries: + +```sh +PYTHONPATH=src/runners uv run python scripts/upload_nvcc_toolchain.py --source /private/path/to/binaries +``` + +CI warms both caches before deployment. `KERNELBOT_NVCC_PCH_PROFILES` selects a +comma-separated subset; empty warms all 18. Manual warmup uses +`modal run src/runners/modal_runner.py::warm_nvcc_pch --profiles ...`. +Set `KERNELBOT_NVCC_PCH=0` to disable device PCH. Set both +`KERNELBOT_NVCC_TOOLCHAIN=0` and `KERNELBOT_NVCC_PCH=0` to restore CUDA 13.3. +`KERNELBOT_NVCC_PCH_DISABLE=1` also disables lookup within a submission. +The corresponding `KERNELBOT_NVCC_TOOLCHAIN_VOLUME`, `KERNELBOT_NVCC_PCH_VOLUME`, +and `KERNELBOT_PCH_VOLUME` variables override the private input, device, and host +volumes. Retain old cache generations for rollback and replace workers after +warming so they see the new Volume snapshot. + +### Host cache + The main/dev workflow warms profiles before deploying GPU runners. Manually, run both commands in the same Modal profile/environment: diff --git a/docs/nvcc-pch-benchmark.md b/docs/nvcc-pch-benchmark.md deleted file mode 100644 index 8bb8e20d..00000000 --- a/docs/nvcc-pch-benchmark.md +++ /dev/null @@ -1,214 +0,0 @@ -# Experimental NVCC PCH benchmark - -`scripts/nvcc_pch_benchmark.py` tests NVIDIA's supplied PCH compiler against -ordinary `torch.utils.cpp_extension.load_inline()` builds. The existing -[Modal PCH cache](modal-pch-cache.md) precompiles the host-side Torch header with -GCC. This experiment adds NVCC header reuse from a separate shared Modal Volume. -It measures compilation latency; it does not optimize kernel execution. - -Measured on 2026-11-15: adding the supplied compiler and shared NVCC PCH to -an existing host-PCH setup reduced median build time **31.7% for the simple -implicit-header case (35.37s → 24.17s)** and **22.2% for CUTLASS -(43.72s → 34.03s)** on the same CUDA 13.4 toolkit. This is a useful incremental -improvement, approximately **1.46× and 1.29×**, with substantial CUTLASS timing -variation. Replacing the compiler without enabling its PCH did not improve -these medians. The experiment did not modify production. - -These are results for the original workload-specific forwarding headers. The -simple profile also precompiled `cuda_fp16.h`; the CUTLASS profile precompiled -`cutlass/gemm/device/gemm.h`. The [shipping runtime](nvcc-pch-cache.md) caches -only PyTorch's three implicit CUDA headers and leaves CUTLASS headers in each -submission. These POC speedups do not establish the final runtime's speedup. -Use the [release validation results](benchmarks/nvcc-pch-release.json) for that -separate implementation and execution path. - -## Toolkit compatibility - -The supplied compiler reports CUDA **13.4.99**. The initial attempt to overlay -its four binaries on KernelBot's CUDA 13.3 image failed: the compiler emitted -PTX 9.4, while the installed assembler supported PTX 9.3. Those failures are -compatibility results, not PCH performance measurements. - -The benchmark therefore installs the official CUDA 13.4 compiler/toolkit -dependencies, which resolved to compiler **13.4.92**, before applying the -experimental binaries. The stock and patched timing arms use this same 13.4 -environment. The official `cuda-nvcc-13-4` package supplies matching `ptxas`, -`nvlink`, and `fatbinary`; `libnvvm-13-4` supplies `cicc` and `libdevice`. -The JSON probe records the actual stock and experimental compiler versions. - -Both the official 13.3 and 13.4 `nvcc.profile` files select `nvvm/bin/cicc`. -Copying the supplied second backend to `nvvm-next/bin/cicc` does not select it. -For this bounded SM75/SM100 experiment, patched containers explicitly set: - -```text -CICC_PATH = $(TOP)/nvvm-next/bin -``` - -The probe checks the dispatched path and compiles a minimal kernel for both -`sm_75` and `sm_100`. SM100 is a **compile-only Blackwell smoke test**; it is -not a Blackwell runtime or performance measurement. The separate -[runtime installer](nvcc-pch-cache.md) checks all supported runner architectures; -this experiment makes no claim about older targets or their backend dispatch. - -## Comparison - -| Mode | Compiler | GCC host PCH | NVCC PCH | -| --- | --- | --- | --- | -| `stock_no_pch` | Official 13.4 | Off | Off | -| `stock_host_pch` | Official 13.4 | On | Off | -| `patched_host_pch` | Experimental 13.4.99 | On | Off | -| `patched_all_pch` | Experimental 13.4.99 | On | On | - -Compare `patched_all_pch` with `patched_host_pch` to isolate the added NVCC PCH -path. Compare it with `stock_host_pch` to estimate the combined effect of the -compiler replacement and NVCC PCH on the existing host-cache setup. - -Each timed build uses a fresh T4 container with four CPU cores, 16 GiB RAM, a -fresh build directory, and normal Ninja parallelism. Both workloads retain -PyTorch's default implicit CUDA headers: - -- `simple`: FP16 vector addition, with two launch/source variants. -- `cutlass`: an actual SM75 tensor-core GEMM, with two threadblock tile variants. - -Both variants change a user header and have distinct expected outputs. The -GEMM checks random and zero inputs, tile tails, and a non-default CUDA stream -against a float64 PyTorch reference. Each trial records `load_inline()` wall -time, individual C++/NVCC command times, correctness results, compiler versions, -binary hashes, and container/image identities. Warmup and diagnostic builds are -outside the reported build time. The command order alternates between variants -and repetitions to reduce ordering bias. - -Untimed diagnostic compilations verify actual cache consumption: GCC `-H` -must report a valid PCH, and NVCC `--pch-verbose` must report using both -`cudafe_` and `cicc_` PCH files from the expected read-only directory without -creating one. The actual timed-build logs independently confirm both hits. A successful build by itself is insufficient evidence of reuse. - -## Run - -Use an authenticated Modal environment. Extract the supplied package privately; -the directory below must contain `nvcc`, `cudafe++`, `nvvm/bin/cicc`, and -`nvvm-next/bin/cicc`. Do not commit the supplied binaries or credentials. - -```sh -export NVIDIA_PCH_BINARIES=/path/to/extracted/x86_64_Linux_release -export PYTHONPATH=src:src/runners:scripts - -# Optional: choose fresh, dedicated benchmark volumes. -export KERNELBOT_HOST_PCH_BENCHMARK_VOLUME=kernelbot-nvcc-host-pch-my-test -export KERNELBOT_NVCC_PCH_VOLUME=kernelbot-nvcc-pch-my-test - -# Check compiler versions, backend dispatch, and SM75/SM100 compilation. -uv run modal run scripts/nvcc_pch_benchmark.py \ - --probe-only --output /tmp/nvcc-pch-probe.json - -# Warm shared headers, then run all four modes and both source variants. -uv run modal run scripts/nvcc_pch_benchmark.py \ - --cases simple,cutlass --repeats 2 --output /tmp/nvcc-pch-results.json -``` - -`--warm-only` prepares and commits headers without timing trials. A subsequent -`--skip-warm` run reuses those entries and validates their identity. The harness -uses the temporary `kernelbot-nvcc-pch-benchmark` app and dedicated volumes; it -does not deploy over the production runner. Only the trusted warmer writes the -volumes. Trials mount both caches read-only and restrict Modal API access. - -## Integration requirements and limits - -In this benchmark, the warmer captures PyTorch's NVCC flags, then compiles a -forwarding `gpumode.cuh` containing the common Torch/CUDA headers and the workload's CUDA -or CUTLASS headers. The header ends with `#pragma hdrstop`. The warmer must specify -`-x cu -c gpumode.cuh -o /tmp/header.o` along with the matched flags and PCH -options; the bare `.cuh` command in the high-level setup note was rejected by -this NVCC build. Trials pass -`-include gpumode.cuh --pch --pch-dir ...`. Cache identity includes supplied -binary/profile hashes, the installed-header fingerprint, forwarding-header -contents, workload flags, and architecture. Exact commands are retained in the -results. The warmup and trials use different extension names, so successful -reuse also tests that per-module naming does not invalidate these profiles. - -The separate [runtime](nvcc-pch-cache.md) uses curated submission flag sets, -immutable compiler/header identities, and fallback on a miss. Its release -validator exercises the CPU precompile worker's 8 GiB memory budget and GPU -artifact replay; these original GPU trials used 16 GiB and did not exercise -that path. The runtime also narrows the forwarding header to preserve the -implicit include sequence. User source, kernel objects, and shared libraries -remain outside the shared PCH cache. - -This matrix covers two fixtures and one runtime GPU, not the submission -population. It excludes scheduling, image startup, PCH preparation, and the -existing [CPU-to-GPU artifact transfer](python-cpu-compilation.md) from the -`load_inline()` metric. Compiler subprocess times overlap under Ninja and must -not be summed to estimate wall time. The controlled CUDA 13.4 matrix isolates the cache/compiler changes. A -separate CUDA 13.3 baseline below measures the combined toolkit migration and -PCH effect for the same CUTLASS fixture. Fresh containers do not guarantee -cold underlying host or Modal storage caches. Two repetitions -per variant provide a small sample; retain individual measurements alongside -medians and avoid generalizing small differences. - -## Results - -Main matrix: four builds per cell, two source/tile variants × two repetitions. -Values are median `load_inline()` seconds, with the full min–max range. - -| CUDA 13.4 build mode | Simple, implicit headers | CUTLASS, implicit headers | -| --- | ---: | ---: | -| Stock, no PCH | 37.23 (35.75–42.89) | 48.82 (42.95–54.31) | -| Stock + existing host PCH | 35.37 (35.03–36.43) | 43.72 (41.50–48.83) | -| Patched compiler + host PCH | 36.01 (34.74–38.47) | 45.76 (41.73–51.06) | -| Patched compiler + host and NVCC PCH | **24.17 (23.93–29.14)** | **34.03 (28.44–43.66)** | - -Comparing the last two rows isolates NVCC PCH on the same patched compiler: -**1.49× simple / 1.34× CUTLASS**. The CUTLASS cached builds took 39–44s in the -first repetition and 28–29s in the second; underlying storage/cache state was -not controlled. The four-sample median is not a guarantee for every fresh -container, and its range overlaps the stock host-cache range. - -The fresh production-image CUDA 13.3 baseline measured CUTLASS at **48.75s -without PCH** and **45.32s with the existing host PCH**. Moving from that current -host-cache setup to CUDA 13.4 plus the supplied PCH path measured **34.03s**: -**1.33×, or 24.9% less compile time**, including the toolkit change. The old -benchmark's `inline_simple` case deliberately omits implicit CUDA headers and -is not comparable to the new `simple` fixture; its 10.37s host-cache result is -retained only as a separate minimal-header diagnostic. - -A supplemental run tested sequentially reading the PCH files before compilation, -with the read time included in the total. It did not improve the medians: - -| Two builds per cell | Direct shared-volume PCH | Prefetch + compilation | -| --- | ---: | ---: | -| Simple | 25.24s | 25.70s | -| CUTLASS | 32.90s | 35.44s | - -Prefetch itself took 1.51–1.60s for simple and 2.61–7.73s for CUTLASS. Keep this -supplemental run separate from the main matrix; its faster CUTLASS samples do -not replace the main result. - -Warm diagnostic phase timings explain the limit. Combined `cudafe++`/`cicc` -time fell from a median 23.82s to 7.61s for simple and 29.25s to 10.51s for -CUTLASS. GCC preprocessing and compilation still occupied roughly 13–19s. -These are second compilations after timing and correctness checks, so they -illustrate the remaining work rather than measuring first-use storage overhead. - -One-time CUDA PCH creation took **44.74s for simple** and **51.80s for CUTLASS**. -Each profile produced two files, totaling **2.52 GB** and **3.00 GB** respectively -(2.35 and 2.79 GiB). The host warmup took 72.26s. These stage timings exclude -volume commits and flag-capture setup. Storage and profile count need budgeting -when expanding to multiple architectures and submission flag sets. - -All **492 correctness checks** passed across **62 distinct single-use T4 -containers**: 32 main builds, 12 supplemental builds, 16 CUDA 13.3 baseline -builds, and two reuse smoke builds. Maximum GEMM absolute error was **8.53e-6**. -All **18 NVCC-PCH builds** confirmed both frontend hits in their actual timed -logs and in independent diagnostics. Different module names, kernel sources, -and user-header values reused the same profile and produced the expected -distinct outputs. No kernel object or shared-library cache was used. - -[Complete measurements, compiler hashes, commands, warmup logs, and validation](benchmarks/nvcc-pch.json). - -Modal runs: - -- [Main 32-build matrix](https://modal.com/apps/gpu-mode/main/ap-sCbGWCCsGHByigNhMrPUYA) -- [Prefetch and phase diagnostics](https://modal.com/apps/gpu-mode/main/ap-P9owg8O1R3RlDGy0ozOSE1) -- [Production-image CUDA 13.3 baseline](https://modal.com/apps/gpu-mode/main/ap-42e4FtDMIK2OGSSHHcAg51) -- [Shared-volume warmup](https://modal.com/apps/gpu-mode/main/ap-ypnw7T8Xntoe9WGjwycFyh) -- [Cross-module reuse smoke](https://modal.com/apps/gpu-mode/main/ap-ya4TMM2KIT440BrHtTc77E) diff --git a/docs/nvcc-pch-cache.md b/docs/nvcc-pch-cache.md deleted file mode 100644 index 242c77a3..00000000 --- a/docs/nvcc-pch-cache.md +++ /dev/null @@ -1,205 +0,0 @@ -# NVCC framework header cache - -KernelBot can use NVIDIA's private PCH compiler drop to precompile the framework -headers that PyTorch inserts into `load_inline()` CUDA sources. Matching builds -reuse those headers from a shared, read-only Modal Volume. Users keep their -existing submissions and compiler flags. - -This is a separately selectable CUDA 13.4 toolkit on the existing runner base -image. Both the toolchain and the NVCC cache are disabled by default. The existing -[host PCH cache](modal-pch-cache.md) -and [CPU-to-GPU artifact transfer](python-cpu-compilation.md) work alongside it. -The [initial benchmark](nvcc-pch-benchmark.md) used broader, workload-specific -headers, including a CUTLASS header. Its speedups do not measure this narrower -framework-only cache. [Release validation results](benchmarks/nvcc-pch-release.json) -record measurements for the implemented CPU-to-GPU path. - -## Compatibility - -The forwarding header contains, in order, `torch/types.h`, `cuda.h`, and -`cuda_runtime.h`. These are the exact implicit CUDA includes in the pinned -PyTorch 2.12 image; `torch/types.h` includes ATen transitively. CUTLASS and user -headers are not precompiled. Adding arbitrary common headers before user code -could change macro and include-order behavior, so header-frequency data alone -does not make an include eligible. - -The wrapper requires this source prefix and an exact match for the compiler -flags and header-search environment. It preserves flag order, macros, and the -real module name in the compiler invocation. Cache lookup omits generated -source/output/dependency paths and the per-module name macro. -Unknown profiles, custom include paths, compiler overrides, and sources without -the prefix compile normally. `no_implicit_headers=True` does not gain injected -framework headers. A rejected cache retries the original compile once; genuine -submission errors still fail. - -The compiler is NVIDIA's `13.4.99` drop, paired with toolkit package `13.4.3-1` -and CUDA 13.4.92 compiler packages. The complete toolkit preserves cuTile, -cuBLAS, sanitizer, and other existing development tools when CUDA paths switch. -CUDA 13.3's `ptxas` cannot assemble the emitted PTX 9.4. The installer -verifies all four supplied binary hashes and explicitly selects `nvvm-next` for -the supported GPUs. Image-build smoke checks cover every deployed architecture -and the architecture-specific Blackwell profiles. - -## Observed profiles and candidate coverage - -The [profile inventory](benchmarks/nvcc-pch-profiles.json) covers the -2026-11-13 private-diff snapshot: **366,803 unique submissions** in 367,118 rows. -The public baseline is excluded; this is not the complete GPUMODE submission -history. Runner metadata identifies 124,736 NVIDIA submissions, leaves 242,065 -missing or unrecognized, and identifies two other runners. Those cohorts remain -separate throughout the analysis. - -A conservative static pass examines calls with an explicit `cuda_sources` -keyword. It resolves simple literals and unambiguous assignments in their lexical -scope, rejecting reassigned, mutated, escaped, or otherwise unresolved values. -Calls using `**kwargs` are never assumed to have default flags. An eligible -candidate also needs a verified PyTorch `load_inline` import, resolved CUDA -source, implicit headers enabled, no custom include paths, and flags from the -bounded inventory allowlist. Flags retain their original order. This finds -20,411 eligible candidate submissions among 70,461 static CUDA `load_inline` -submissions with known NVIDIA runners, and 25,996 among 112,520 with unknown -runners. - -The default catalog uses the six most frequent eligible ordered tuples in the -**known NVIDIA cohort**: - -| Ordered `extra_cuda_cflags` | Known NVIDIA submissions | Unknown-runner submissions | -| --- | ---: | ---: | -| `['-O3']` | 5,419 | 10,457 | -| `['-O3', '--use_fast_math']` | 3,358 | 6,695 | -| `[]` | 2,670 | 2,556 | -| `['-O3', '--use_fast_math', '-arch=sm_100a']` | 2,609 | 113 | -| `['-O3', '-arch=sm_100', '--use_fast_math']` | 771 | 99 | -| `['-O3', '-std=c++17', '-arch=sm_80', '--use_fast_math']` | 628 | 174 | - -The three architecture-neutral tuples are warmed for `sm75`, `sm80`, `sm89`, -`sm90a`, and `sm100`; the three explicit-architecture tuples add one profile each, -for **18 default profiles**. The warmer captures PyTorch's remaining generated -flags. Different ordering, include paths, or generated architecture flags still -require their own exact profile. - -Each table row counts a submission once per distinct tuple. The coverage below -uses a set union per submission, so a submission containing several selected -tuples is counted once: - -| Runner cohort | Static CUDA `load_inline` submissions | Any eligible call with a selected tuple | All observed calls eligible and selected | -| --- | ---: | ---: | ---: | -| Known NVIDIA | 70,461 | 15,421 (21.89%) | 15,176 | -| Unknown | 112,520 | 18,283 (16.25%) | 16,356 | -| Combined analyzed cohorts | 182,981 | 33,704 (18.42%) | 31,532 | - -These are **static flag-tuple candidate counts**, not executed NVCC command -counts or measured cache-hit rates. Matching the real architecture, framework -prefix, toolchain, and environment remains necessary. Dynamic Python and -positional CUDA arguments fall outside this inventory's coverage. In particular, -88,425 analyzed submissions contain a call with `no_implicit_headers=True`; those -calls keep their existing behavior and do not receive the master header. - -Resolved source includes also show why header popularity cannot define the -master header. The following frequencies count known-NVIDIA static CUDA -`load_inline` submissions once per explicit include across their resolved C++ -and CUDA source strings; they do not count PyTorch's generated implicit includes: - -| Explicit source include | Submissions | In the master header | -| --- | ---: | --- | -| `cuda_runtime.h` | 50,015 | Yes | -| `cuda_fp16.h` | 38,863 | No | -| `cuda.h` | 35,932 | Yes | -| `torch/extension.h` | 24,752 | No | -| `ATen/ATen.h` | 3,448 | Transitive through `torch/types.h` | -| `cutlass/cutlass.h` | 263 | No; CUTLASS excluded | -| `cute/tensor.hpp` | 232 | No; CuTe excluded | -| `torch/types.h` | 159 | Yes; normally inserted by PyTorch | - -Only the exact three-header implicit prefix described above is precompiled. -The published inventory contains aggregates, without submission IDs or private -source code. - -## Immutable cache and artifacts - -The cache layout is -`/kernelbot-nvcc-pch///`. -Each completed entry contains `gpumode.cuh`, both frontend PCH files, and a ready -manifest. The toolchain fingerprint covers the compiler drop, supporting CUDA -tools, installed CUDA package versions and headers, and the wrapper and warmer -sources. Profile keys add the ordered flags and relevant environment. - -Only the trusted CPU warmer can write the Volume. Submission workers mount it -read-only with Modal API access restricted. The warmer accepts fixed profile -names rather than submission-supplied code or flags, and publishes a manifest -only after both frontend files are complete. New generations coexist with old -ones so an image rollback retains its matching cache. - -CPU artifact packets include the NVCC toolchain identity. GPU replay checks that -identity and ordinary source/header dependencies. A PCH dependency uses its small -trusted manifest and file sizes instead of reading several GB to hash it on each -submission. Arbitrary PCH paths are not eligible for this shortcut. Changed user -headers and compiler environment variables still invalidate replay. - -## Prepare and validate a release - -Use KernelBot's Modal workspace. Keep the NVIDIA binaries outside this public -repository. The input directory must contain `nvcc`, `cudafe++`, -`nvvm/bin/cicc`, and `nvvm-next/bin/cicc`. - -```sh -export PYTHONPATH=src:src/runners -uv run python scripts/upload_nvcc_toolchain.py --source /private/path/to/binaries -uv run modal run scripts/validate_nvcc_pch.py --output /tmp/nvcc-pch-validation.json -``` - -The validator selects the new toolchain and isolated cache Volumes. Its default -matrix runs PCH on/off on T4, H100, and B200, plus T4 checks for unknown flags, -missing/corrupt caches, disabled implicit headers, changed user headers, and a -custom NVCC override. It exercises real CPU compilation, GPU artifact replay, -correctness, benchmarking, and ranking evaluation without publishing scores. -Use `--profiles` to select trusted catalog entries and `--repeats` for repeated -PCH-on/off pairs in fresh containers. Successful compilation alone is not a -cache hit: the validator requires both NVIDIA frontends to report consumption. -The [profile inventory](benchmarks/nvcc-pch-profiles.json) describes the sampled -submission population and static candidate coverage. Those counts are not -executed NVCC command counts or measured cache-hit rates. - -## Deploy and roll back - -First warm and deploy a separate canary app using a single consistent image: - -```sh -export KERNELBOT_MODAL_APP=discord-bot-runner-nvcc-pch-canary -export KERNELBOT_NVCC_TOOLCHAIN=1 -export KERNELBOT_NVCC_PCH=1 -uv run modal run src/runners/modal_runner.py::warm_pch -uv run modal run src/runners/modal_runner.py::warm_nvcc_pch -uv run modal deploy src/runners/modal_runner_archs.py -``` - -`ModalLauncher(app_name=...)` can target that app for non-scoring validation. -Do not split one scoring population randomly across compiler versions. Promote -the validated toolchain as a consistent runner release, recording the image and -toolchain fingerprint with the rollout. - -The main/dev deployment workflow reads these repository/environment variables: - -| Variable | Default | Purpose | -| --- | --- | --- | -| `KERNELBOT_NVCC_TOOLCHAIN` | `0` | Select the pinned CUDA 13.4 compiler image. | -| `KERNELBOT_NVCC_PCH` | `0` | Enable automatic NVCC PCH lookup. Requires the toolchain. | -| `KERNELBOT_NVCC_PCH_PROFILES` | empty | Comma-separated trusted profiles; empty uses the bounded default set. | -| `KERNELBOT_NVCC_TOOLCHAIN_VOLUME` | `kernelbot-nvcc-toolchains` | Private compiler input Volume. | -| `KERNELBOT_NVCC_PCH_VOLUME` | `kernelbot-nvcc-pch` | Device PCH Volume. | -| `KERNELBOT_PCH_VOLUME` | `kernelbot-torch-pch` | Host PCH Volume. | - -`warm_nvcc_pch --profiles ...` limits manual warmup to specific catalog entries. -Warm before replacing workers, since existing containers retain their Volume -snapshot. - -For a cache-only rollback, deploy the same toolchain with -`KERNELBOT_NVCC_PCH=0`. For a compiler rollback, deploy with both switches `0` to -restore the CUDA 13.3 image. Individual submissions can disable lookup with -`KERNELBOT_NVCC_PCH_DISABLE=1` or override `PYTORCH_NVCC`. - -Compilation emits `[kernelbot-nvcc-pch]` JSON events for `hit`, `miss`, `bypass`, -`disabled`, `fallback`, or `error`. A hit requires NVIDIA diagnostics confirming -both `cicc` and `cudafe` consumed the cache. Events include profile identity and -elapsed compiler time. `KERNELBOT_NVCC_PCH_EVENTS` optionally writes JSONL to a -container-local file. diff --git a/scripts/nvcc_pch_benchmark.py b/scripts/nvcc_pch_benchmark.py deleted file mode 100644 index 3450df44..00000000 --- a/scripts/nvcc_pch_benchmark.py +++ /dev/null @@ -1,656 +0,0 @@ -"""Isolated benchmark of NVIDIA's experimental NVCC PCH build. - -NVIDIA_PCH_BINARIES=/path/to/x86_64_Linux_release PYTHONPATH=src:src/runners:scripts \ - uv run modal run scripts/nvcc_pch_benchmark.py --probe-only - -No production app or production PCH volume is modified. -""" - -import json -import os -import statistics -from pathlib import Path - -import modal - -os.environ["KERNELBOT_PCH_VOLUME"] = os.environ.get( - "KERNELBOT_HOST_PCH_BENCHMARK_VOLUME", "kernelbot-nvcc-host-pch-benchmark-cuda134-20261115" -) -if os.environ["KERNELBOT_PCH_VOLUME"] == "kernelbot-torch-pch": - raise RuntimeError("Refusing to use the production host PCH volume for this benchmark") -from modal_pch_benchmark import CPP, CUDA, GEMM_CPP, GEMM_CUDA # noqa: E402 -from modal_runner import PCH_MOUNT, cuda_image_base, pch_volume # noqa: E402 - -app = modal.App("kernelbot-nvcc-pch-benchmark") -NVCC_MOUNT = "/kernelbot-nvcc-pch" -nvcc_volume = modal.Volume.from_name( - os.environ.get("KERNELBOT_NVCC_PCH_VOLUME", "kernelbot-nvcc-pch-benchmark-20261115"), - create_if_missing=True, -) -image = ( - cuda_image_base.apt_install( - "cuda-compiler-13-4", "cuda-cudart-dev-13-4", "cuda-driver-dev-13-4" - ) - .run_commands( - "ln -sfn /usr/local/cuda-13.4 /usr/local/cuda", - "nvcc --version", - "python /opt/kernelbot-pch/compiler.py --kernelbot-fingerprint", - ) - .add_local_python_source( - "libkernelbot", "modal_runner", "modal_runner_archs", "modal_pch_benchmark" - ) -) -if modal.is_local(): - binary_root = os.environ.get("NVIDIA_PCH_BINARIES") - if not binary_root: - raise RuntimeError( - "Set NVIDIA_PCH_BINARIES to the extracted x86_64_Linux_release directory" - ) - image = image.add_local_dir(binary_root, "/opt/nvidia-pch-binaries") - -CASES = ("simple", "cutlass") -DEFAULT_MODES = ("stock_no_pch", "stock_host_pch", "patched_host_pch", "patched_all_pch") -MODES = (*DEFAULT_MODES, "patched_all_pch_prefetch") -CUDA_FLAGS = [ - "-U__CUDA_NO_HALF_OPERATORS__", - "-U__CUDA_NO_HALF_CONVERSIONS__", - "-U__CUDA_NO_HALF2_OPERATORS__", - "-U__CUDA_NO_BFLOAT16_CONVERSIONS__", - "--expt-relaxed-constexpr", - "-O3", - "-I/opt/cutlass/include", - "-I/opt/cutlass/tools/util/include", -] -OBSERVER = r"""#!/usr/bin/env python3 -import json, os, subprocess, sys, time -kind = "nvcc" if sys.argv[0].endswith("nvcc") else "cxx" -compiler = "/usr/local/cuda/bin/nvcc" if kind == "nvcc" else os.environ["PCH_TEST_COMPILER"] -args = sys.argv[1:] -if kind == "nvcc" and os.environ.get("PCH_CAPTURE") == "1" and "-c" in args: - with open(os.environ["PCH_TEST_COMMANDS"] + ".capture", "w") as f: - json.dump(args, f) - sys.exit(1) -if kind == "nvcc" and "-c" in args and os.environ.get("PCH_NVCC_ENTRY"): - entry = os.environ["PCH_NVCC_ENTRY"] - args = [*args, "-include", entry + "/gpumode.cuh", "--pch", "--pch-dir", entry + "/pch"] -started = time.perf_counter() -code = subprocess.call([compiler, *args]) -if "-c" in args: - with open(os.environ["PCH_TEST_COMMANDS"] + "." + kind, "a") as f: - f.write(json.dumps({ - "args": args, "seconds": time.perf_counter()-started, "returncode": code - }) + "\n") -sys.exit(code) -""" - -LOADER = r""" -import json, os, time -from pathlib import Path -import torch -from torch.utils.cpp_extension import load_inline -root = Path(os.environ["PCH_TEST_ROOT"]) -case = os.environ["PCH_TEST_CASE"] -variant = int(os.environ["PCH_TEST_VARIANT"]) -options = json.loads((root / "options.json").read_text()) -started = time.perf_counter() -prefetch_bytes = 0 -if os.environ.get("PCH_PREFETCH") == "1": - for pch in sorted((Path(os.environ["PCH_NVCC_ENTRY"]) / "pch").glob("*.pch")): - with pch.open("rb") as source: - while block := source.read(16 * 1024 * 1024): - prefetch_bytes += len(block) - assert prefetch_bytes > 0 and prefetch_bytes == int(os.environ["PCH_EXPECTED_BYTES"]), \ - "Prefetch did not read exactly the warmed PCH payload" -prefetch_seconds = time.perf_counter() - started if prefetch_bytes else 0 -compile_started = time.perf_counter() -module = load_inline(cpp_sources=(root / "cpp.txt").read_text(), - cuda_sources=(root / "cuda.txt").read_text(), functions=["add_cuda", "version"], - no_implicit_headers=False, verbose=True, build_directory=str(root), **options) -seconds = time.perf_counter() - started -load_inline_seconds = time.perf_counter() - compile_started -assert module.version() == variant + 1 -outputs, errors = [], [] -if case == "cutlass": - torch.manual_seed(17) - with torch.cuda.stream(torch.cuda.Stream()): - for m, n, k in ((128, 128, 64), (129, 136, 64), (256, 256, 128), (512, 512, 256)): - a = torch.randn(m, k, device="cuda", dtype=torch.float16) / 4 - b = torch.randn(n, k, device="cuda", dtype=torch.float16) / 4 - for zero in (False, True): - if zero: - a.zero_() - actual = module.add_cuda(a, b) - reference = (a.double() @ b.double().T) * (variant + 1) - torch.testing.assert_close(actual.double(), reference, rtol=2e-4, atol=2e-4) - errors.append(float((actual.double() - reference).abs().max())) - a = torch.ones(8, 8, device="cuda", dtype=torch.float16) - actual = module.add_cuda(a, a) - torch.testing.assert_close( - actual, torch.full_like(actual, 8 * (variant + 1)), rtol=0, atol=0 - ) - outputs.append(float(actual[0, 0])) -else: - for n in (1, 127, 128, 129, 1025): - a = torch.arange(n, device="cuda", dtype=torch.float16) % 8 - b = torch.ones_like(a) - actual = module.add_cuda(a, b) - torch.testing.assert_close(actual, a + b + (variant + 1), rtol=0, atol=0) - outputs.append(float(actual[0])) -(root / "result.json").write_text(json.dumps(dict( - seconds=seconds, load_inline_seconds=load_inline_seconds, - prefetch_seconds=prefetch_seconds, prefetch_bytes=prefetch_bytes, outputs=outputs, errors=errors -))) -""" - - -def prepare_build(root, case, variant, name, host_pch, entry=None, capture=False): - cpp = GEMM_CPP if case == "cutlass" else CPP - cuda = ( - GEMM_CUDA.replace("TILE_N", str(128 if variant == 0 else 64)) - if case == "cutlass" - else CUDA.replace("THREADS", str(256 if variant == 0 else 128)) - ) - header = f"#pragma once\nconstexpr int kVariant = {variant + 1};\n" - (root / "variant.h").write_text(header) - (root / "cpp.txt").write_text(cpp) - (root / "cuda.txt").write_text(cuda) - (root / "options.json").write_text( - json.dumps( - { - "name": name, - "extra_cuda_cflags": CUDA_FLAGS if case == "cutlass" else [], - } - ) - ) - for kind in ("cxx", "nvcc"): - p = root / f"observe-{kind}" - p.write_text(OBSERVER) - p.chmod(0o755) - env = dict(os.environ) - for key in ("MAX_JOBS", "KERNELBOT_PCH_DISABLE", "KERNELBOT_PCH_TRACE", "KERNELBOT_PCH_WRITE"): - env.pop(key, None) - env.update( - { - "TORCH_CUDA_ARCH_LIST": "7.5", - "CXX": str(root / "observe-cxx"), - "PYTORCH_NVCC": str(root / "observe-nvcc"), - "PCH_TEST_COMPILER": "/opt/kernelbot-pch/compiler.py" if host_pch else "/usr/bin/g++", - "PCH_TEST_ROOT": str(root), - "PCH_TEST_CASE": case, - "PCH_TEST_VARIANT": str(variant), - "PCH_TEST_COMMANDS": str(root / "commands"), - "PCH_CAPTURE": "1" if capture else "0", - "PCH_NVCC_ENTRY": str(entry) if entry else "", - } - ) - return env, cpp + cuda + header - - -def pch_identity(case, patched): - header = ( - "#pragma once\n#include \n#include \n" - "#include \n#include \n" - ) - header += ( - "#include \n" - if case == "cutlass" - else "#include \n" - ) - header += "#pragma hdrstop\n" - return { - "patch": patched, - "fingerprint": Path("/opt/kernelbot-pch/fingerprint").read_text().strip(), - "header": header, - "case": case, - "cuda_flags": CUDA_FLAGS if case == "cutlass" else [], - "arch": "7.5", - "protocol": 2, - } - - -def install_patch(): - """Overlay only the four supplied binaries inside this disposable container.""" - import hashlib - import re - import shutil - - rows = [] - for source_name, target_name in ( - ("nvcc", "bin/nvcc"), - ("cudafe++", "bin/cudafe++"), - ("nvvm/bin/cicc", "nvvm/bin/cicc"), - ("nvvm-next/bin/cicc", "nvvm-next/bin/cicc"), - ): - source = Path("/opt/nvidia-pch-binaries") / source_name - target = Path("/usr/local/cuda") / target_name - existed = target.exists() - target.parent.mkdir(parents=True, exist_ok=True) - shutil.copyfile(source, target) - target.chmod(0o755) - rows.append( - { - "file": str(target), - "replaced_existing": existed, - "sha256": hashlib.sha256(source.read_bytes()).hexdigest(), - } - ) - profile_path = Path("/usr/local/cuda/bin/nvcc.profile") - # The experiment uses SM75 and SM100, both on NVIDIA's newer backend. - # General production dispatch must also handle SM70 and earlier separately. - profile, count = re.subn( - r"^CICC_PATH\s*=.*$", - "CICC_PATH = $(TOP)/nvvm-next/bin", - profile_path.read_text(), - flags=re.MULTILINE, - ) - assert count == 1, "Expected exactly one CICC_PATH assignment in CUDA's profile" - profile_path.write_text(profile) - rows.append( - { - "file": str(profile_path), - "replaced_existing": True, - "sha256": hashlib.sha256(profile.encode()).hexdigest(), - } - ) - return rows - - -@app.function(image=image, cpu=4, memory=16384, timeout=300, single_use_containers=True) -def probe(): - import shutil - import subprocess - import tempfile - - stock_version = subprocess.check_output(["/usr/local/cuda/bin/nvcc", "--version"], text=True) - paths = subprocess.check_output( - ["find", "/usr/local/cuda/", "-name", "cicc", "-o", "-name", "nvcc.profile"], text=True - ) - patched = install_patch() - version = subprocess.check_output(["/usr/local/cuda/bin/nvcc", "--version"], text=True) - help_text = subprocess.check_output(["/usr/local/cuda/bin/nvcc", "--help"], text=True) - lines = help_text.splitlines() - indices = { - j - for i, line in enumerate(lines) - if "pch" in line.lower() - for j in range(max(0, i - 2), min(len(lines), i + 14)) - } - profiles = {str(p): p.read_text() for p in Path("/usr/local/cuda/bin").glob("*.profile")} - smoke = [] - with tempfile.TemporaryDirectory() as directory: - root = Path(directory) - source = root / "smoke.cu" - source.write_text("__global__ void kernelbot_nvcc_smoke() {}\n") - for arch in ("sm_75", "sm_100"): - output = root / f"{arch}.o" - command = [ - "/usr/local/cuda/bin/nvcc", - f"-arch={arch}", - "-c", - str(source), - "-o", - str(output), - ] - dry = subprocess.run([*command, "--dryrun"], capture_output=True, text=True, timeout=60) - compile_result = subprocess.run( - [*command, "-v"], capture_output=True, text=True, timeout=60 - ) - dispatched = "nvvm-next/bin" in dry.stderr and "$(_NVVM_BRANCH_)" not in dry.stderr - smoke.append( - { - "arch": arch, - "command": command, - "dryrun_returncode": dry.returncode, - "dryrun": dry.stdout + dry.stderr, - "next_backend_selected": dispatched, - "returncode": compile_result.returncode, - "log": compile_result.stdout + compile_result.stderr, - "object_bytes": output.stat().st_size if output.exists() else 0, - } - ) - return { - "stock_version": stock_version, - "patched_version": version, - "paths": paths, - "patch": patched, - "nvcc_profiles": profiles, - "pch_help": "\n".join(lines[i] for i in sorted(indices)), - "strace": shutil.which("strace"), - "smoke": smoke, - "compatible": all( - r["returncode"] == 0 and r["next_backend_selected"] and r["object_bytes"] for r in smoke - ), - } - - -@app.function( - image=image, - cpu=4, - memory=16384, - timeout=1800, - single_use_containers=True, - max_containers=1, - volumes={PCH_MOUNT: pch_volume, NVCC_MOUNT: nvcc_volume}, -) -def warm(cases: list[str]): - import hashlib - import subprocess - import tempfile - import time - - patched = install_patch() - pch_volume.reload() - nvcc_volume.reload() - print("Warming host CUDA-linked default profile", flush=True) - host_started = time.perf_counter() - host = subprocess.run( - ["python3", "/opt/kernelbot-pch/warm.py", "cuda-default"], - capture_output=True, - text=True, - timeout=300, - ) - if host.returncode: - raise RuntimeError(host.stdout + host.stderr) - host_seconds = time.perf_counter() - host_started - pch_volume.commit() - print(f"Host profile ready in {host_seconds:.2f}s", flush=True) - rows = [] - for case in cases: - identity = pch_identity(case, patched) - key = hashlib.sha256(json.dumps(identity, sort_keys=True).encode()).hexdigest() - entry = Path(NVCC_MOUNT) / key - entry.mkdir(parents=True, exist_ok=True) - (entry / "gpumode.cuh").write_text(identity["header"]) - (entry / "pch").mkdir(exist_ok=True) - with tempfile.TemporaryDirectory() as directory: - root = Path(directory) - print(f"Capturing exact NVCC flags for {case}", flush=True) - env, _ = prepare_build(root, case, 0, "kb_nvcc_pch_warmer", True, capture=True) - captured = subprocess.run( - ["python3", "-c", LOADER], env=env, capture_output=True, text=True, timeout=180 - ) - capture_path = root / "commands.capture" - if not capture_path.exists(): - raise RuntimeError( - "Failed to capture NVCC command: " + captured.stdout + captured.stderr - ) - args = json.loads(capture_path.read_text()) - flags = [] - i = 0 - while i < len(args): - if args[i] in {"-c", "-o", "--dependency-output", "-MF", "-MT", "-MQ"}: - i += 2 - continue - if args[i] not in {"--generate-dependencies-with-compile", "-MMD", "-MD"}: - flags.append(args[i]) - i += 1 - command = [ - "/usr/local/cuda/bin/nvcc", - *flags, - "--pch-dir", - str(entry / "pch"), - "--pch", - "--pch-verbose", - "-x", - "cu", - "-c", - str(entry / "gpumode.cuh"), - "-o", - str(root / "warm.o"), - ] - print(f"Creating {case} PCH in {entry}", flush=True) - tick = time.perf_counter() - result = subprocess.run( - command, cwd=root, env=env, capture_output=True, text=True, timeout=300 - ) - seconds = time.perf_counter() - tick - row = { - "case": case, - "key": key, - "command": command, - "seconds": seconds, - "stdout": result.stdout, - "stderr": result.stderr, - "returncode": result.returncode, - "files": [ - {"path": str(p.relative_to(entry)), "bytes": p.stat().st_size} - for p in sorted((entry / "pch").rglob("*")) - if p.is_file() - ], - } - rows.append(row) - print( - f"{case} warmup returned {result.returncode} after {seconds:.2f}s; " - f"{len(row['files'])} files", - flush=True, - ) - if result.returncode or not row["files"]: - raise RuntimeError(json.dumps(row)) - (entry / "metadata.json").write_text( - json.dumps({"identity": identity, "warmup": row}, indent=2) - ) - (Path(NVCC_MOUNT) / f"{case}.json").write_text(json.dumps({"key": key})) - nvcc_volume.commit() - return {"host_seconds": host_seconds, "host_log": host.stdout + host.stderr, "device": rows} - - -@app.function( - image=image, - gpu="T4", - cpu=4, - memory=16384, - timeout=900, - max_containers=4, - single_use_containers=True, - restrict_modal_access=True, - volumes={ - PCH_MOUNT: pch_volume.with_mount_options(read_only=True), - NVCC_MOUNT: nvcc_volume.with_mount_options(read_only=True), - }, -) -def trial(case: str, variant: int, mode: str, repeat: int): - import hashlib - import re - import subprocess - import tempfile - import time - - import torch - - started = time.perf_counter() - assert mode in MODES and case in CASES - patched = install_patch() if mode.startswith("patched") else None - nvcc_version = subprocess.check_output(["/usr/local/cuda/bin/nvcc", "--version"], text=True) - entry = None - if mode.startswith("patched_all_pch"): - key = json.loads((Path(NVCC_MOUNT) / f"{case}.json").read_text())["key"] - entry = Path(NVCC_MOUNT) / key - expected = pch_identity(case, patched) - metadata = json.loads((entry / "metadata.json").read_text()) - assert metadata["identity"] == expected, "Warm PCH identity does not match; rerun warmup" - assert (entry / "gpumode.cuh").read_text() == expected["header"] - assert key == hashlib.sha256(json.dumps(expected, sort_keys=True).encode()).hexdigest() - with tempfile.TemporaryDirectory() as directory: - root = Path(directory) - name = f"kb_nvcc_{case}_{variant}" - env, source = prepare_build(root, case, variant, name, mode != "stock_no_pch", entry) - env["PCH_PREFETCH"] = "1" if mode == "patched_all_pch_prefetch" else "0" - if entry: - env["PCH_EXPECTED_BYTES"] = str( - sum(f["bytes"] for f in metadata["warmup"]["files"] if f["path"].endswith(".pch")) - ) - run_started = time.perf_counter() - result = subprocess.run( - ["python3", "-c", LOADER], env=env, capture_output=True, text=True, timeout=300 - ) - process_seconds = time.perf_counter() - run_started - log = result.stdout + result.stderr - row = { - "case": case, - "variant": variant, - "mode": mode, - "repeat": repeat, - "extension_name": name, - "returncode": result.returncode, - "log": log, - "source_sha256": hashlib.sha256(source.encode()).hexdigest(), - "container_id": os.environ["MODAL_TASK_ID"], - "image_id": os.environ["MODAL_IMAGE_ID"], - "gpu": torch.cuda.get_device_name(), - "torch": str(torch.__version__), - "nvcc_version": nvcc_version, - "patch": patched, - "pch_entry": str(entry) if entry else None, - "process_seconds": process_seconds, - } - if entry: - timed_used = re.findall(r'using precompiled header file "([^"]+)"', log) - timed_created = re.findall(r'creating precompiled header file "([^"]+)"', log) - row["timed_device_pch_files_used"] = timed_used - row["timed_device_pch_consumed"] = ( - result.returncode == 0 - and not timed_created - and {Path(p).name.split("_", 1)[0] for p in timed_used} == {"cicc", "cudafe"} - and all(Path(p).is_relative_to(entry / "pch") for p in timed_used) - ) - for kind in ("cxx", "nvcc"): - path = root / f"commands.{kind}" - row[f"{kind}_commands"] = ( - [json.loads(s) for s in path.read_text().splitlines()] if path.exists() else [] - ) - if result.returncode: - return row - measurement = json.loads((root / "result.json").read_text()) - row.update( - { - "seconds": measurement["seconds"], - "outputs": measurement["outputs"], - "load_inline_seconds": measurement["load_inline_seconds"], - "prefetch_seconds": measurement["prefetch_seconds"], - "prefetch_bytes": measurement["prefetch_bytes"], - "max_absolute_error": max(measurement["errors"], default=0), - "correctness_checks": 10 if case == "cutlass" else 6, - } - ) - # GCC -H prints ! for a valid consumed PCH. Keep this compilation untimed. - cxx_args = list(row["cxx_commands"][0]["args"]) - cxx_args[cxx_args.index("-o") + 1] = str(root / "diagnostic.o") - diagnostic = subprocess.run( - [env["PCH_TEST_COMPILER"], *cxx_args, "-H"], - env=env, - cwd=root, - capture_output=True, - text=True, - timeout=180, - ) - row["host_pch_consumed"] = "! /kernelbot-pch/" in diagnostic.stderr - assert diagnostic.returncode == 0, diagnostic.stderr[-6000:] - assert row["host_pch_consumed"] == (mode != "stock_no_pch"), diagnostic.stderr[-6000:] - row["host_hit_keys"] = re.findall(r"\[kernelbot-pch\] hit (\w+)", log) - # Preserve verbose NVCC diagnostics independently of the measured build. - if mode.startswith("patched"): - args = list(row["nvcc_commands"][0]["args"]) - args[args.index("-o") + 1] = str(root / "diagnostic.cuda.o") - phase_file = root / "nvcc-diagnostic.csv" - extra = ["-v", "--time", str(phase_file), *(["--pch-verbose"] if entry else [])] - diagnostic = subprocess.run( - ["/usr/local/cuda/bin/nvcc", *args, *extra], - env=env, - cwd=root, - capture_output=True, - text=True, - timeout=180, - ) - row["nvcc_diagnostic"] = { - "returncode": diagnostic.returncode, - "stdout": diagnostic.stdout, - "stderr": diagnostic.stderr, - "phase_csv": phase_file.read_text() if phase_file.exists() else None, - } - used = re.findall(r'using precompiled header file "([^"]+)"', diagnostic.stderr) - created = re.findall(r'creating precompiled header file "([^"]+)"', diagnostic.stderr) - row["device_pch_files_used"] = used - row["device_pch_files_created"] = created - frontends = {Path(path).name.split("_", 1)[0] for path in used} - row["device_pch_consumed"] = bool(entry) and ( - frontends == {"cicc", "cudafe"} - and not created - and diagnostic.returncode == 0 - and all(Path(path).is_relative_to(entry / "pch") for path in used) - ) - row["remote_seconds"] = time.perf_counter() - started - return row - - -@app.local_entrypoint() -def main( - output: str = "/tmp/kernelbot-nvcc-pch.json", - probe_only: bool = False, - warm_only: bool = False, - skip_warm: bool = False, - cases: str = "simple,cutlass", - modes: str = ",".join(DEFAULT_MODES), - repeats: int = 2, -): - selected = cases.split(",") - selected_modes = modes.split(",") - assert set(selected) <= set(CASES) and set(selected_modes) <= set(MODES) - assert repeats >= 1 - result = {"probe": probe.remote(), "rows": []} - - def save(): - Path(output).write_text(json.dumps(result, indent=2) + "\n") - - save() - if probe_only: - print(json.dumps(result["probe"], indent=2)) - return - assert result["probe"]["compatible"], ( - f"Patched compiler compatibility checks failed; inspect {output}" - ) - if not skip_warm: - result["warmup"] = warm.remote(selected) - save() - print(json.dumps(result["warmup"], indent=2)) - if warm_only: - return - jobs = [ - (case, variant, mode, repeat) - for repeat in range(repeats) - for case in selected - for variant in (0, 1) - for mode in (selected_modes if (variant + repeat) % 2 == 0 else selected_modes[::-1]) - ] - for row in trial.starmap(jobs): - result["rows"].append(row) - save() - print( - json.dumps( - { - k: row.get(k) - for k in ("case", "variant", "mode", "repeat", "returncode", "seconds") - } - ) - ) - assert len({r["container_id"] for r in result["rows"]}) == len(result["rows"]) - for case in selected: - for mode in selected_modes: - group = [r for r in result["rows"] if r["case"] == case and r["mode"] == mode] - good = [r["seconds"] for r in group if r["returncode"] == 0] - print( - f"{case}/{mode}: median={statistics.median(good) if good else None}; " - f"passed={len(good)}/{len(group)}" - ) - assert all(row["returncode"] == 0 for row in result["rows"]), ( - f"Failed builds retained in {output}" - ) - assert all( - row.get("device_pch_consumed") - for row in result["rows"] - if row["mode"].startswith("patched_all_pch") - ), f"NVCC PCH consumption was not verified; inspect diagnostics in {output}" - assert all( - row.get("timed_device_pch_consumed") - for row in result["rows"] - if row["mode"].startswith("patched_all_pch") - ), f"NVCC PCH consumption was not verified in the timed build; inspect logs in {output}" diff --git a/scripts/validate_nvcc_pch.py b/scripts/validate_nvcc_pch.py deleted file mode 100644 index 2d8bbbcd..00000000 --- a/scripts/validate_nvcc_pch.py +++ /dev/null @@ -1,612 +0,0 @@ -"""Validate the real CPU-to-GPU submission path without publishing leaderboard scores. - -KERNELBOT_NVCC_TOOLCHAIN=1 KERNELBOT_NVCC_PCH=1 \ - PYTHONPATH=src:src/runners uv run modal run scripts/validate_nvcc_pch.py - -Publish the verified private toolchain Volume with upload_nvcc_toolchain.py first. -The default matrix runs all edge cases on T4 and PCH-on/off on H100 and B200. -The two baseline cases execute the normal test, benchmark, and leaderboard -validation phases in an ephemeral app; no service or database is contacted. -Use --profiles with trusted nvcc_common.PROFILES names to select baseline -flag sets, and --repeats to repeat baseline pairs in fresh containers. -The optional CUTLASS pair uses the generic sm75-O3 framework profile: - --gpus T4 --cases cutlass,cutlass-disabled --repeats 2 --concurrency 1 -""" - -import ast -import asyncio -import dataclasses -import hashlib -import json -import os -import shutil -import statistics -import subprocess -import time -from contextlib import contextmanager -from pathlib import Path - -os.environ["KERNELBOT_MODAL_APP"] = os.environ.get( - "KERNELBOT_NVCC_VALIDATION_APP", "kernelbot-nvcc-pch-release-validation" -) -if os.environ["KERNELBOT_MODAL_APP"] == "discord-bot-runner": - raise RuntimeError("The validation app must be separate from the production runner") -os.environ.setdefault("KERNELBOT_NVCC_TOOLCHAIN", "1") -os.environ.setdefault("KERNELBOT_NVCC_PCH", "1") -os.environ.setdefault("KERNELBOT_PCH_VOLUME", "kernelbot-nvcc-release-host-pch") -os.environ.setdefault("KERNELBOT_NVCC_PCH_VOLUME", "kernelbot-nvcc-release-device-pch") - -from modal_runner import ( # noqa: E402 - NVCC_PCH_MOUNT, - app, - cuda_image, - submission_volumes, - warm_nvcc_pch, - warm_pch, -) - -from libkernelbot.consts import GPU_TO_SM, SubmissionMode, Timeout # noqa: E402 -from libkernelbot.python_precompile import ( # noqa: E402 - compile_python, - run_with_cpu_artifacts, -) -from libkernelbot.run_eval import run_config # noqa: E402 -from libkernelbot.task import build_task_config, make_task_definition # noqa: E402 - -# Serialized GPU functions refer to helpers from this module by name. Mount the -# same harness source on both stages so fresh containers can import those helpers. -cuda_image = cuda_image.add_local_file(Path(__file__).resolve(), "/root/validate_nvcc_pch.py") - -GPUS = ("T4", "L4", "A100", "H100", "B200") -DEFAULT_GPUS = ("T4", "H100", "B200") -GPU_RESOURCES = {"A100": "A100-80GB", "H100": "H100!"} -BASELINE_CASES = ("default", "disabled") -CUTLASS_CASES = ("cutlass", "cutlass-disabled") -CASES = ( - *BASELINE_CASES, - "no-implicit", - "unknown-flags", - "changed-header", - "rejected-cache", - "empty-cache", - "custom-nvcc", -) -ALL_CASES = (*CASES, *CUTLASS_CASES) -FINGERPRINT = Path("/opt/kernelbot-pch/nvcc-fingerprint") -GPU_ARCHES = { - "T4": {"75"}, - "L4": {"89"}, - "A100": {"80"}, - "H100": {"90", "90a"}, - "B200": {"100", "100a"}, -} - - -def profile_catalog() -> dict: - # Only the local orchestrator needs the profile catalog. The remote image - # keeps these stdlib modules in /opt/kernelbot-pch instead of a Python package. - from pch.nvcc_common import PROFILES - - return PROFILES - - -def make_submission(example: Path, case: str, extra_cuda_cflags=()) -> str: - source = (example / "submission_cuda_inline.py").read_text() - source = source.replace("name='add_cuda'", "name='nvcc_release_vectoradd'") - if extra_cuda_cflags: - source = source.replace( - " verbose=True,", - f" verbose=True,\n extra_cuda_cflags={list(extra_cuda_cflags)!r},", - ) - if case == "no-implicit": - source = source.replace( - 'add_cuda_source = """', - 'add_cuda_source = """\n#include \n#include ', - ).replace(" verbose=True,", " verbose=True,\n no_implicit_headers=True,") - elif case == "unknown-flags": - source = source.replace( - " verbose=True,", - " verbose=True,\n extra_cuda_cflags=['-DKERNELBOT_RELEASE_PROFILE_MISS=1'],", - ) - elif case == "changed-header": - # Identical source and options in both containers, but a genuinely changed - # user dependency. Replaying the CPU binary would produce wrong answers. - source = ( - "import torch\nfrom pathlib import Path\n" - "Path('/tmp/kernelbot-nvcc-release-user.h').write_text(\n" - " '#define OFFSET ' + ('0' if torch.cuda.is_available() else '1') + '\\n')\n" - + source.replace( - 'add_cuda_source = """', - 'add_cuda_source = """\n#include "/tmp/kernelbot-nvcc-release-user.h"', - ).replace("C[idx] = A[idx] + B[idx];", "C[idx] = A[idx] + B[idx] + OFFSET;") - ) - elif case == "custom-nvcc": - source = "import os\nos.environ['PYTORCH_NVCC'] = '/usr/local/cuda/bin/nvcc'\n" + source - return source - - -def cutlass_submission() -> str: - # Reuse the benchmark's actual SM75 tensor-core GEMM without importing its - # Modal app or its workload-specific PCH. Only these two string literals run. - benchmark = Path(__file__).with_name("modal_pch_benchmark.py") - constants = {} - for node in ast.parse(benchmark.read_text()).body: - if isinstance(node, ast.Assign): - for target in node.targets: - if isinstance(target, ast.Name) and target.id in {"GEMM_CPP", "GEMM_CUDA"}: - constants[target.id] = ( - ast.literal_eval(node.value) - .replace('#include "variant.h"', "constexpr int kVariant = 1;") - .replace("TILE_N", "128") - ) - return ( - "# GEMM fixture adapted from scripts/modal_pch_benchmark.py.\n" - "import torch\nfrom torch.utils.cpp_extension import load_inline\n" - "module = load_inline(name='nvcc_release_cutlass',\n" - f" cpp_sources={constants['GEMM_CPP']!r},\n" - f" cuda_sources={constants['GEMM_CUDA']!r},\n" - " functions=['add_cuda', 'version'], extra_cuda_cflags=['-O3'],\n" - " no_implicit_headers=False, verbose=True)\n" - "assert module.version() == 1\n" - "def custom_kernel(data):\n" - " a, b, nondefault = data\n" - " if not nondefault:\n" - " return module.add_cuda(a, b)\n" - " current = torch.cuda.current_stream(a.device)\n" - " stream = torch.cuda.Stream(device=a.device)\n" - " stream.wait_stream(current)\n" - " with torch.cuda.stream(stream):\n" - " result = module.add_cuda(a, b)\n" - " a.record_stream(stream)\n" - " b.record_stream(stream)\n" - " current.wait_stream(stream)\n" - " result.record_stream(current)\n" - " return result\n" - ) - - -CUTLASS_REFERENCE = """ -import torch - -def generate_input(m, n, k, seed, zero=0, ones=0, nondefault=1): - generator = torch.Generator(device='cuda').manual_seed(seed) - a = torch.randn(m, k, device='cuda', dtype=torch.float16, generator=generator) / 4 - b = torch.randn(n, k, device='cuda', dtype=torch.float16, generator=generator) / 4 - if zero: - a.zero_() - if ones: - a.fill_(1) - b.fill_(1) - return a, b, nondefault - -def check_implementation(data, output): - a, b, _ = data - expected = a.double() @ b.double().T - try: - torch.testing.assert_close(output.double(), expected, rtol=2e-4, atol=2e-4) - except AssertionError as error: - return False, str(error) - error = float((output.double() - expected).abs().max()) - return True, f'max_absolute_error={error}' -""" - - -def configure_cutlass(config): - config["sources"]["reference.py"] = CUTLASS_REFERENCE - config["sources"]["task.py"] = "TestSpec = dict\n" - config["tests"] = [ - {"m": m, "n": n, "k": k, "seed": 17, "zero": zero, "nondefault": 1} - for m, n, k in ((128, 128, 64), (129, 136, 64), (256, 256, 128), (512, 512, 256)) - for zero in (0, 1) - ] + [{"m": 8, "n": 8, "k": 8, "seed": 17, "ones": 1, "nondefault": 1}] - # Time the kernel on the usual stream; correctness tests above exercise a - # separate stream, random/zero inputs, and tile tails against float64 Torch. - config["benchmarks"] = [{"m": 256, "n": 256, "k": 128, "seed": 54352, "nondefault": 0}] - - -def make_config(gpu: str, case: str, profile: str | None = None) -> dict: - example = Path(__file__).resolve().parents[1] / "examples/vectoradd_py" - if case in CUTLASS_CASES: - profile = profile or "sm75-O3" - if gpu != "T4" or profile != "sm75-O3": - raise ValueError("The CUTLASS fixture requires T4 and the generic sm75-O3 profile") - arch, flags = profile_catalog()[profile or f"sm{GPU_TO_SM[gpu]}-default"] - if arch not in GPU_ARCHES[gpu]: - raise ValueError(f"Profile {profile} targets SM{arch}, which is incompatible with {gpu}") - task = make_task_definition(example).task - mode = ( - SubmissionMode.LEADERBOARD - if case in (*BASELINE_CASES, *CUTLASS_CASES) - else SubmissionMode.TEST - ) - config = build_task_config( - task=task, - submission_content=( - cutlass_submission() if case in CUTLASS_CASES else make_submission(example, case, flags) - ), - arch=arch, - mode=mode, - ) - config["benchmarks"] = [{"size": 512, "seed": 54352}] - if case in CUTLASS_CASES: - configure_cutlass(config) - return config - - -def resource_usage() -> dict: - import resource - - result = { - "self_peak_rss_kib": resource.getrusage(resource.RUSAGE_SELF).ru_maxrss, - "child_peak_rss_kib": resource.getrusage(resource.RUSAGE_CHILDREN).ru_maxrss, - } - for name in ("memory.peak", "memory.max"): - path = Path("/sys/fs/cgroup") / name - if path.exists(): - result[name] = path.read_text().strip() - return result - - -def read_events(path: Path) -> list[dict]: - return [json.loads(line) for line in path.read_text().splitlines()] if path.exists() else [] - - -def prepare_rejected_cache(destination: Path, arch: str) -> dict: - """Damage disposable sparse copies; never modify files on the shared Volume.""" - generation = FINGERPRINT.read_text().strip() - manifests = list((Path(NVCC_PCH_MOUNT) / generation).glob("*/manifest.json")) - for path in manifests: - manifest = json.loads(path.read_text()) - if manifest.get("profile") != f"sm{arch}-default": - continue - entry = destination / generation / path.parent.name - (entry / "pch").mkdir(parents=True) - shutil.copyfile(path, entry / "manifest.json") - shutil.copyfile(path.parent / "gpumode.cuh", entry / "gpumode.cuh") - total = 0 - for item in manifest["files"]: - target = entry / item["path"] - if not target.resolve().is_relative_to((entry / "pch").resolve()): - raise ValueError("Unexpected PCH path in trusted manifest") - target.parent.mkdir(exist_ok=True) - with target.open("wb") as handle: - # Preserve manifest size validation while making the PCH invalid. - # Sparse local files do not copy the multi-GB shared payload. - handle.truncate(item["bytes"]) - total += item["bytes"] - return {"profile": manifest["profile"], "sparse_bytes": total} - raise RuntimeError(f"Warm sm{arch}-default before running the rejection test") - - -@contextmanager -def validation_environment(case: str, arch: str): - previous = dict(os.environ) - event_path = Path("/tmp/kernelbot-nvcc-release-events.jsonl") - event_path.unlink(missing_ok=True) - metadata = {} - try: - os.environ.update( - KERNELBOT_NVCC_PCH="1", - KERNELBOT_NVCC_PCH_DISABLE="1" if case in {"disabled", "cutlass-disabled"} else "0", - KERNELBOT_NVCC_PCH_EVENTS=str(event_path), - ) - os.environ.pop("KERNELBOT_NVCC_PCH_ROOT", None) - if case in {"rejected-cache", "empty-cache"}: - destination = Path("/tmp") / f"kernelbot-nvcc-release-{case}" - destination.mkdir(exist_ok=False) - if case == "rejected-cache": - metadata = prepare_rejected_cache(destination, arch) - os.environ["KERNELBOT_NVCC_PCH_ROOT"] = str(destination) - yield event_path, metadata - finally: - os.environ.clear() - os.environ.update(previous) - - -def toolchain_info() -> dict: - return { - "generation": FINGERPRINT.read_text().strip(), - "nvcc_version": subprocess.check_output( - ["/usr/local/cuda/bin/nvcc", "--version"], text=True - ), - "container_id": os.environ["MODAL_TASK_ID"], - "image_id": os.environ["MODAL_IMAGE_ID"], - } - - -@app.function( - image=cuda_image, - cpu=4, - memory=8192, - timeout=Timeout.COMPILE + 30, - max_containers=3, - single_use_containers=True, - restrict_modal_access=True, - volumes=submission_volumes, -) -def validate_cpu(config: dict, case: str) -> dict: - started = time.perf_counter() - with validation_environment(case, config["arch"]) as (event_path, metadata): - packet = compile_python(config) - return { - "packet": packet, - "events": read_events(event_path), - "toolchain": toolchain_info(), - "memory": resource_usage(), - "case_metadata": metadata, - "wall_seconds": time.perf_counter() - started, - "requested_memory_mib": 8192, - "max_jobs": 2, - } - - -def validate_gpu(config: dict, case: str, packet: dict) -> dict: - started = time.perf_counter() - with validation_environment(case, config["arch"]) as (event_path, metadata): - result = run_with_cpu_artifacts({**config, "cpu_compile": packet}, run_config) - return { - "result": dataclasses.asdict(result), - "events": read_events(event_path), - "toolchain": toolchain_info(), - "memory": resource_usage(), - "case_metadata": metadata, - "wall_seconds": time.perf_counter() - started, - } - - -gpu_functions = { - gpu: app.function( - image=cuda_image, - gpu=GPU_RESOURCES.get(gpu, gpu), - cpu=4, - memory=16384, - timeout=600, - max_containers=1, - single_use_containers=True, - restrict_modal_access=True, - volumes=submission_volumes, - name=f"validate_nvcc_{gpu.lower()}", - serialized=True, - )(validate_gpu) - for gpu in GPUS -} - - -def verify_hits(events, profile): - for event in events: - if event["status"] == "hit": - assert event["frontends"] == ["cicc", "cudafe"], event - if profile: - assert event["profile"] == profile, event - - -def verify_result(case: str, cpu: dict, gpu: dict, profile: str | None = None) -> None: - packet = cpu["packet"] - assert packet["info"]["status"] == "compiled", packet["info"] - assert packet["info"]["artifacts"] > 0 and packet["artifacts"] - assert cpu["toolchain"]["generation"] == gpu["toolchain"]["generation"] - assert cpu["toolchain"]["image_id"] == gpu["toolchain"]["image_id"] - result = gpu["result"] - assert result["success"], result["error"] - assert result["runs"], "No evaluation phases ran" - for name, evaluation in result["runs"].items(): - run = evaluation["run"] - assert run and run["success"] and run["passed"], (name, evaluation) - assert run["result"].get("check") == "pass", (name, run["result"]) - info = result["cpu_compile"] - cpu_statuses = {event["status"] for event in cpu["events"]} - gpu_statuses = {event["status"] for event in gpu["events"]} - if case == "changed-header": - assert info["fallbacks"] > 0 and info["reused"] == 0, info - assert "dependency changed" in info["reason"].lower(), info - assert "hit" in gpu_statuses, gpu["events"] - else: - assert info["status"] == "reused" and info["reused"] >= 2, info - assert info["fallbacks"] == 0, info - assert not gpu["events"], gpu["events"] - expected_status = { - "default": "hit", - "disabled": "disabled", - "no-implicit": "bypass", - "unknown-flags": "miss", - "changed-header": "hit", - "rejected-cache": "fallback", - "empty-cache": "miss", - "cutlass": "hit", - "cutlass-disabled": "disabled", - } - if case == "custom-nvcc": - assert not cpu["events"], cpu["events"] - else: - assert cpu_statuses == {expected_status[case]}, cpu["events"] - verify_hits(cpu["events"] + gpu["events"], profile) - if case not in {"default", "changed-header", "cutlass"}: - assert "hit" not in cpu_statuses, cpu["events"] - if case == "rejected-cache": - assert any( - event["status"] == "fallback" and event["reason"] == "cache_rejected" - for event in cpu["events"] - ), cpu["events"] - if case in (*BASELINE_CASES, *CUTLASS_CASES): - assert set(result["runs"]) == {"test", "benchmark", "leaderboard"} - measurements = result["runs"]["benchmark"]["run"]["result"] - assert float(measurements["benchmark.0.mean"]) > 0, measurements - - -def compact_cpu(cpu: dict) -> dict: - packet = cpu["packet"] - binary = packet["artifacts"] - return cpu | { - "packet": {key: value for key, value in packet.items() if key != "artifacts"} - | {"artifact_bytes": len(binary), "artifact_sha256": hashlib.sha256(binary).hexdigest()} - } - - -def paired_jobs(gpu, profile, pair, cases, repeats): - return [ - (gpu, case, profile, repeat) - for repeat in range(1, repeats + 1) - for case in pair[:: 1 if repeat % 2 else -1] - if case in cases - ] - - -def matrix_jobs(gpus, cases, edge_gpu, profiles, repeats): - catalog = profile_catalog() - selected = profiles or [f"sm{GPU_TO_SM[gpu]}-default" for gpu in gpus] - if len(selected) > 12 or not set(selected) <= catalog.keys(): - raise ValueError("Choose at most 12 fixed profiles from " + ", ".join(catalog)) - jobs = [] - for profile in selected: - arch, _ = catalog[profile] - matching = [gpu for gpu in gpus if arch in GPU_ARCHES[gpu]] - if len(matching) != 1: - raise ValueError(f"Select a compatible GPU for {profile} (SM{arch})") - jobs.extend(paired_jobs(matching[0], profile, BASELINE_CASES, cases, repeats)) - if set(cases) & set(CUTLASS_CASES): - if "T4" not in gpus: - raise ValueError("Select T4 for the CUTLASS fixture") - jobs.extend(paired_jobs("T4", "sm75-O3", CUTLASS_CASES, cases, repeats)) - if edge_gpu in gpus: - jobs.extend( - (edge_gpu, case, f"sm{GPU_TO_SM[edge_gpu]}-default", 1) - for case in cases - if case not in (*BASELINE_CASES, *CUTLASS_CASES) - ) - if not jobs: - raise ValueError("No jobs selected; edge cases run only on edge_gpu") - return jobs - - -def comparisons(rows): - """Keep latency observations separate from the functional pass/fail gates.""" - grouped = {} - failed = set() - for row in rows: - if row["case"] in (*BASELINE_CASES, *CUTLASS_CASES): - workload = "cutlass" if row["case"] in CUTLASS_CASES else "vectoradd" - arm = "disabled" if row["case"].endswith("disabled") else "default" - key = (row["gpu"], row["profile"], workload) - if not row["passed"]: - failed.add(key) - else: - grouped.setdefault(key, {}).setdefault(arm, []).append( - row["cpu"]["packet"]["info"]["duration"] - ) - results = [] - for (gpu, profile, workload), arms in sorted(grouped.items()): - if ( - (gpu, profile, workload) in failed - or set(arms) != set(BASELINE_CASES) - or len(arms["default"]) != len(arms["disabled"]) - ): - continue - on, off = (statistics.median(arms[case]) for case in BASELINE_CASES) - results.append( - { - "gpu": gpu, - "profile": profile, - "workload": workload, - "cpu_compile_seconds": arms, - "median_pch_on_seconds": on, - "median_pch_off_seconds": off, - "speedup": off / on, - "reduction_percent": 100 * (1 - on / off), - "scope": "CPU compile_python including import, dependency validation and packing", - } - ) - return results - - -@app.local_entrypoint() -async def main( - gpus: str = ",".join(DEFAULT_GPUS), - cases: str = ",".join(CASES), - edge_gpu: str = "T4", - profiles: str = "", - repeats: int = 1, - skip_warm: bool = False, - concurrency: int = 2, - output: str = "/tmp/kernelbot-nvcc-pch-release-validation.json", -): - selected_gpus, selected_cases = gpus.split(","), cases.split(",") - if not set(selected_gpus) <= set(GPUS) or not set(selected_cases) <= set(ALL_CASES): - raise ValueError(f"Choose GPUs from {GPUS} and cases from {ALL_CASES}") - if edge_gpu not in GPUS or not 1 <= concurrency <= 3 or not 1 <= repeats <= 5: - raise ValueError("Choose a supported edge_gpu, concurrency 1–3, and repeats 1–5") - selected_profiles = list(dict.fromkeys(profiles.split(","))) if profiles else [] - jobs = matrix_jobs(selected_gpus, selected_cases, edge_gpu, selected_profiles, repeats) - result = { - "app": os.environ["KERNELBOT_MODAL_APP"], - "baseline_repeats": repeats, - "concurrency": concurrency, - "rows": [], - } - - def save(): - result["comparisons"] = comparisons(result["rows"]) - Path(output).write_text(json.dumps(result, default=str, indent=2) + "\n") - - if not skip_warm: - result["host_warmup"] = await warm_pch.remote.aio(profiles="cuda-default") - warm_profiles = ",".join(dict.fromkeys(job[2] for job in jobs)) - result["nvcc_warmup"] = await warm_nvcc_pch.remote.aio(profiles=warm_profiles) - save() - gate = asyncio.Semaphore(concurrency) - - async def run_one(gpu_name, case, profile, repeat): - async with gate: - arch, flags = profile_catalog()[profile] - row = { - "gpu": gpu_name, - "case": case, - "workload": "cutlass" if case in CUTLASS_CASES else "vectoradd", - "profile": profile, - "arch": arch, - "extra_cuda_cflags": flags, - "repeat": repeat, - "passed": False, - "state": "running", - } - started = time.perf_counter() - result["rows"].append(row) - save() - try: - config = make_config(gpu_name, case, profile) - row["sources_sha256"] = hashlib.sha256( - json.dumps(config["sources"], sort_keys=True).encode() - ).hexdigest() - cpu = await validate_cpu.remote.aio(config=config, case=case) - row["cpu"] = compact_cpu(cpu) - save() - gpu = await gpu_functions[gpu_name].remote.aio( - config=config, case=case, packet=cpu["packet"] - ) - row["gpu_run"] = gpu - verify_result(case, cpu, gpu, profile) - row["passed"] = True - except Exception as exc: - row["error"] = f"{type(exc).__name__}: {exc}" - row["wall_seconds"] = time.perf_counter() - started - row["state"] = "passed" if row["passed"] else "failed" - save() - print( - json.dumps( - { - key: row.get(key) - for key in ("gpu", "case", "profile", "repeat", "passed", "error") - } - ), - flush=True, - ) - - await asyncio.gather(*(run_one(*job) for job in jobs)) - identifiers = [ - stage["toolchain"]["container_id"] - for row in result["rows"] - for stage in (row.get("cpu"), row.get("gpu_run")) - if stage - ] - assert len(identifiers) == len(set(identifiers)), "Expected fresh CPU and GPU containers" - assert all(row["passed"] for row in result["rows"]), f"Validation failures saved in {output}" - print(f"Passed {len(result['rows'])} CPU-to-GPU cases; results: {output}", flush=True) diff --git a/src/runners/modal_runner.py b/src/runners/modal_runner.py index 46af9156..b9812b26 100644 --- a/src/runners/modal_runner.py +++ b/src/runners/modal_runner.py @@ -16,8 +16,8 @@ # IMPORTANT: This has to stay in separate file or modal breaks app = App(os.environ.get("KERNELBOT_MODAL_APP", "discord-bot-runner")) cuda_version = "13.3.0" -NVCC_TOOLCHAIN_ENABLED = os.environ.get("KERNELBOT_NVCC_TOOLCHAIN") == "1" -NVCC_PCH_ENABLED = os.environ.get("KERNELBOT_NVCC_PCH") == "1" +NVCC_TOOLCHAIN_ENABLED = os.environ.get("KERNELBOT_NVCC_TOOLCHAIN", "1") == "1" +NVCC_PCH_ENABLED = os.environ.get("KERNELBOT_NVCC_PCH", "1" if NVCC_TOOLCHAIN_ENABLED else "0") == "1" if NVCC_PCH_ENABLED and not NVCC_TOOLCHAIN_ENABLED: raise ValueError("NVCC PCH requires KERNELBOT_NVCC_TOOLCHAIN=1") flavor = "devel" diff --git a/src/runners/nvcc_toolchain.py b/src/runners/nvcc_toolchain.py index d25e5407..54764add 100644 --- a/src/runners/nvcc_toolchain.py +++ b/src/runners/nvcc_toolchain.py @@ -1,4 +1,4 @@ -"""Pinned private NVIDIA compiler drop used by the opt-in Modal runner image.""" +"""Pinned private NVIDIA compiler drop used by the Modal runner image.""" RELEASE = "nvidia-pch-13.4.99-20261017" BINARY_SHA256 = { diff --git a/tests/test_nvcc_pch.py b/tests/test_nvcc_pch.py index d37ddc98..f45572ca 100644 --- a/tests/test_nvcc_pch.py +++ b/tests/test_nvcc_pch.py @@ -1,321 +1,184 @@ -"""NVCC cache eligibility, immutable identity, and failure recovery without CUDA.""" +"""NVCC eligibility, immutable cache consumption, and fallback without CUDA.""" import importlib.util import json import subprocess import sys from pathlib import Path +from types import SimpleNamespace import pytest -SOURCE = Path(__file__).parents[1] / "src/runners/pch" - -def load(name, path): - spec = importlib.util.spec_from_file_location(name, path) +def load(name, filename): + spec = importlib.util.spec_from_file_location( + name, Path(__file__).parents[1] / "src/runners/pch" / filename + ) module = importlib.util.module_from_spec(spec) sys.modules[name] = module spec.loader.exec_module(module) return module -common = load("nvcc_common", SOURCE / "nvcc_common.py") -wrapper = load("kernelbot_nvcc_wrapper", SOURCE / "nvcc.py") -warmer = load("kernelbot_nvcc_warmer", SOURCE / "nvcc_warm.py") +common = load("nvcc_common", "nvcc_common.py") +wrapper = load("kernelbot_nvcc_wrapper", "nvcc.py") +warmer = load("kernelbot_nvcc_warmer", "nvcc_warm.py") GENERATION = "a" * 64 -@pytest.fixture(autouse=True) -def clean_flags(monkeypatch): +@pytest.fixture +def cache(tmp_path, monkeypatch): for key in ("NVCC_PREPEND_FLAGS", "NVCC_APPEND_FLAGS", "KERNELBOT_NVCC_PCH_DISABLE"): monkeypatch.delenv(key, raising=False) monkeypatch.setenv("KERNELBOT_NVCC_PCH", "1") - - -def command(tmp_path, name="first", extra=()): - source = tmp_path / f"{name}.cu" - prefix = "".join(f"#include <{h}>\n" for h in common.HEADERS) - source.write_text(prefix + '#include "user.h"\n__global__ void kernel() {}\n') - return [ - "-MD", - "-MF", - f"{name}.d", - f"-DTORCH_EXTENSION_NAME={name}", - "-DTORCH_API_INCLUDE_EXTENSION_H", - "-isystem", - "/torch/include", - "-gencode=arch=compute_75,code=sm_75", - "--compiler-options", - "-fPIC", - "-std=c++20", - *extra, - "-c", - str(source), - "-o", - f"{name}.o", - ] - - -def ready(tmp_path, monkeypatch, args): - root = tmp_path / "volume" - monkeypatch.setenv("KERNELBOT_NVCC_PCH_ROOT", str(root)) + source = tmp_path / "first.cu" + source.write_text("#include \n#include \n#include \n") + args = ["-MD", "-MF", "first.d", "-DTORCH_EXTENSION_NAME=first", "-isystem", "/torch", + "-gencode=arch=compute_75,code=sm_75", "-std=c++20", "-c", str(source), "-o", "first.o"] + flags = common.compile_flags(args) + assert flags is not None # Actual Torch 2.12 prefix, independent of implementation constants. fingerprint = tmp_path / "fingerprint" fingerprint.write_text(GENERATION) monkeypatch.setattr(common, "FINGERPRINT", fingerprint) - flags = common.compile_flags(args) - key = common.cache_key(flags, GENERATION) - entry = root / GENERATION / key + monkeypatch.setenv("KERNELBOT_NVCC_PCH_ROOT", str(tmp_path / "volume")) + entry = common.cache_root() / GENERATION / common.cache_key(flags, GENERATION) (entry / "pch").mkdir(parents=True) (entry / "gpumode.cuh").write_text(common.HEADER) - files = [] - for name in ("cicc_ab_12.pch", "cudafe_ab_12.pch"): - path = entry / "pch" / name - path.write_bytes(b"PCH stub") - files.append({"path": "pch/" + name, "bytes": path.stat().st_size}) + files = [{"path": f"pch/{name}_ab.pch", "bytes": 4} for name in ("cicc", "cudafe")] + for item in files: + (entry / item["path"]).write_bytes(b"stub") manifest = common.identity(flags, GENERATION) | { - "key": key, - "profile": "sm75-default", - "files": files, - } + "key": entry.name, "profile": "sm75-default", "files": files} (entry / "manifest.json").write_text(json.dumps(manifest)) - return entry, manifest - - -def fake_compiler(monkeypatch, responses): - calls = [] + events = tmp_path / "events.jsonl" + monkeypatch.setenv("KERNELBOT_NVCC_PCH_EVENTS", str(events)) + log = "\n".join(f'using precompiled header file "{entry / item["path"]}"' for item in files) + calls, responses = [], [] def run(args, capture=False): calls.append((args, capture)) - code, stderr = responses[len(calls) - 1] + code, stderr = responses.pop(0) if responses else (0, log) return subprocess.CompletedProcess(args, code, "", stderr) monkeypatch.setattr(wrapper, "run_compiler", run) - return calls - - -def hit_log(entry): - return "\n".join( - f'using precompiled header file "{path}"' for path in sorted((entry / "pch").glob("*.pch")) - ) - - -def test_different_modules_and_user_sources_share_only_header_profile(tmp_path): - first = command(tmp_path) - second = command(tmp_path, "second") - source = Path(second[second.index("-c") + 1]) - source.write_text(source.read_text().replace("kernel() {}", "kernel() { int changed = 2; }")) - one, two = common.compile_flags(first), common.compile_flags(second) - assert common.cache_key(one, GENERATION) == common.cache_key(two, GENERATION) - assert "-DTORCH_EXTENSION_NAME=first" in first - - -def test_pytorch_212_cuda_prefix_is_eligible_without_extra_headers(tmp_path): - args = command(tmp_path) - Path(args[args.index("-c") + 1]).write_text( - "#include \n" - "#include \n" - "#include \n" - "__global__ void kernel() {}\n" - ) - assert common.compile_flags(args) is not None - assert common.HEADERS == ("torch/types.h", "cuda.h", "cuda_runtime.h") - - -@pytest.mark.parametrize( - "extra", - [ - ["-O3"], - ["-DUSER_MACRO=1"], - ["-I/custom/headers"], - ["--use_fast_math"], - ["-std=c++17"], - ["-gencode=arch=compute_100,code=sm_100"], - ], -) -def test_flags_must_match_exactly(tmp_path, extra): - baseline = common.compile_flags(command(tmp_path)) - changed = common.compile_flags(command(tmp_path, "changed", extra)) - assert common.cache_key(baseline, GENERATION) != common.cache_key(changed, GENERATION) - - -def test_toolchain_and_environment_invalidate_profile(tmp_path, monkeypatch): - flags = common.compile_flags(command(tmp_path)) - key = common.cache_key(flags, GENERATION) - assert key != common.cache_key(flags, "b" * 64) - monkeypatch.setenv("CPATH", "/different/headers") - assert key != common.cache_key(flags, GENERATION) - - -@pytest.mark.parametrize( - "prefix", - [ - "#include \n", - "#define USER_SETTING 1\n", - "#include #include \n", - ], -) -def test_no_implicit_headers_and_macro_reordering_bypass(tmp_path, prefix): - args = command(tmp_path) - Path(args[args.index("-c") + 1]).write_text(prefix + "__global__ void kernel() {}\n") - assert common.compile_flags(args) is None - - -@pytest.mark.parametrize( - "extra", [["@flags.rsp"], ["-include", "custom.h"], ["--compiler-bindir=/custom"], ["--pch"]] -) -def test_custom_compiler_and_hidden_flags_bypass(tmp_path, extra): - assert common.compile_flags(command(tmp_path, extra=extra)) is None - - -def test_environment_injected_options_bypass(tmp_path, monkeypatch): - monkeypatch.setenv("NVCC_APPEND_FLAGS", "-include /user/header.h") - assert common.compile_flags(command(tmp_path)) is None - - -def test_unknown_profile_never_creates_volume_files(tmp_path, monkeypatch): - args = command(tmp_path) - monkeypatch.setenv("KERNELBOT_NVCC_PCH_ROOT", str(tmp_path / "absent")) - monkeypatch.setattr(common, "generation", lambda: GENERATION) - calls = fake_compiler(monkeypatch, [(0, "")]) - assert wrapper.main(args) == 0 - assert calls == [(args, False)] - assert not (tmp_path / "absent").exists() - - -def test_hit_keeps_original_flags_and_reports_both_frontends(tmp_path, monkeypatch): - args = command(tmp_path) - entry, _ = ready(tmp_path, monkeypatch, args) - events = tmp_path / "events.jsonl" - monkeypatch.setenv("KERNELBOT_NVCC_PCH_EVENTS", str(events)) - calls = fake_compiler(monkeypatch, [(0, hit_log(entry))]) - assert wrapper.main(args) == 0 - assert calls[0][0][: len(args)] == args - assert calls[0][0][-1] == "--pch-verbose" - record = json.loads(events.read_text()) - assert record["status"] == "hit" and record["frontends"] == ["cicc", "cudafe"] - - -def test_rejected_pch_retries_once_without_injected_header(tmp_path, monkeypatch): - args = command(tmp_path) - ready(tmp_path, monkeypatch, args) - calls = fake_compiler(monkeypatch, [(1, "invalid precompiled header file"), (0, "")]) - assert wrapper.main(args) == 0 - assert len(calls) == 2 and calls[1] == (args, False) - - -def test_user_compile_error_is_not_retried(tmp_path, monkeypatch): - args = command(tmp_path) - entry, _ = ready(tmp_path, monkeypatch, args) - calls = fake_compiler( - monkeypatch, [(2, hit_log(entry) + '\nerror: identifier "typo" undefined')] - ) - assert wrapper.main(args) == 2 - assert len(calls) == 1 - - -def test_partial_pch_consumption_is_not_claimed_as_hit(tmp_path, monkeypatch): - args = command(tmp_path) - entry, _ = ready(tmp_path, monkeypatch, args) - fake_compiler(monkeypatch, [(0, hit_log(entry).splitlines()[0])]) - events = tmp_path / "events.jsonl" - monkeypatch.setenv("KERNELBOT_NVCC_PCH_EVENTS", str(events)) - assert wrapper.main(args) == 0 - assert json.loads(events.read_text())["status"] == "fallback" + return SimpleNamespace(**locals()) + + +def test_identity_reuses_modules_but_invalidates_compiler_inputs(cache, monkeypatch): + c = cache + other = c.args.copy() + other[other.index("-DTORCH_EXTENSION_NAME=first")] = "-DTORCH_EXTENSION_NAME=second" + c.source.write_text(c.source.read_text() + "__global__ void changed_kernel() {}\n") + assert common.compile_flags(other) == c.flags + key = common.cache_key(c.flags, GENERATION) + for flag in ("-O3", "-DUSER=1", "-I/custom", "--use_fast_math", "-std=c++17", "-arch=sm_100"): + assert common.cache_key(common.compile_flags([*other, flag]), GENERATION) != key + assert common.cache_key(c.flags, "b" * 64) != key + monkeypatch.setenv("CPATH", "/different") + assert common.cache_key(c.flags, GENERATION) != key + + +@pytest.mark.parametrize("extra", [["@flags.rsp"], ["-include", "custom.h"], + ["--compiler-bindir=/custom"], ["--pch"]]) +def test_hidden_options_bypass(cache, extra): + assert common.compile_flags([*cache.args, *extra]) is None + + +@pytest.mark.parametrize("prefix", ["#include \n", "#define USER_SETTING 1\n"]) +def test_header_order_is_preserved(cache, prefix): + cache.source.write_text(prefix + cache.source.read_text()) + assert common.compile_flags(cache.args) is None + + +@pytest.mark.parametrize("mode", ["disabled", "missing", "environment"]) +def test_passthrough_never_writes_cache(cache, monkeypatch, mode): + monkeypatch.setenv("KERNELBOT_NVCC_PCH_ROOT", str(cache.tmp_path / "absent")) + if mode == "disabled": + monkeypatch.setenv("KERNELBOT_NVCC_PCH_DISABLE", "1") + monkeypatch.setattr(common, "generation", lambda: pytest.fail("Disabled cache accessed")) + elif mode == "environment": + monkeypatch.setenv("NVCC_APPEND_FLAGS", "-include custom.h") + assert wrapper.main(cache.args) == 0 + assert cache.calls == [(cache.args, False)] + assert not common.cache_root().exists() + + +@pytest.mark.parametrize("kind,code,status", [ + ("hit", 0, "hit"), ("partial", 0, "fallback"), + ("rejected", 1, "fallback"), ("user-error", 2, "error")]) +def test_consumption_evidence_and_cache_only_retry(cache, kind, code, status): + log = {"partial": cache.log.splitlines()[0], "rejected": "invalid precompiled header file"} + log["user-error"] = cache.log + "\nerror: identifier typo undefined" + cache.responses.append((code, log.get(kind, cache.log))) + assert wrapper.main(cache.args) == (0 if kind == "rejected" else code) + assert cache.calls[0][0][:len(cache.args)] == cache.args + assert len(cache.calls) == (2 if kind == "rejected" else 1) + if kind == "rejected": + assert cache.calls[1] == (cache.args, False) + event = json.loads(cache.events.read_text()) + assert event["status"] == status + if status == "hit": + assert event["frontends"] == ["cicc", "cudafe"] @pytest.mark.parametrize("change", ["truncated", "header", "generation", "escape", "partial"]) -def test_invalid_ready_metadata_falls_back_before_compiler(tmp_path, monkeypatch, change): - args = command(tmp_path) - entry, manifest = ready(tmp_path, monkeypatch, args) +def test_bad_ready_metadata_falls_back_before_compiler(cache, change): + c = cache if change == "truncated": - (entry / manifest["files"][0]["path"]).write_bytes(b"broken") + (c.entry / c.files[0]["path"]).write_bytes(b"bad") elif change == "header": - (entry / "gpumode.cuh").write_text("#include \n") + (c.entry / "gpumode.cuh").write_text("#include \n") elif change == "generation": - manifest["generation"] = "b" * 64 + c.manifest["generation"] = "b" * 64 elif change == "escape": - manifest["files"][0]["path"] = "../foreign.pch" + c.files[0]["path"] = "../foreign.pch" else: - manifest["files"].pop() - (entry / "manifest.json").write_text(json.dumps(manifest)) - calls = fake_compiler(monkeypatch, [(0, "")]) - assert wrapper.main(args) == 0 - assert calls == [(args, False)] - - -def test_kill_switch_does_not_access_cache(tmp_path, monkeypatch): - args = command(tmp_path) - monkeypatch.setenv("KERNELBOT_NVCC_PCH_DISABLE", "1") - monkeypatch.setattr(common, "generation", lambda: pytest.fail("Cache accessed while disabled")) - calls = fake_compiler(monkeypatch, [(0, "")]) - assert wrapper.main(args) == 0 - assert calls == [(args, False)] - - -def test_warmer_rejects_untrusted_selectors_before_any_compile(monkeypatch): - monkeypatch.setattr(warmer, "warm_profile", lambda *args: pytest.fail("Unexpected compilation")) - with pytest.raises(ValueError, match="fixed profiles"): - warmer.warm(["-include /user/header.h"]) - + c.files.pop() + (c.entry / "manifest.json").write_text(json.dumps(c.manifest)) + assert wrapper.main(c.args) == 0 and c.calls == [(c.args, False)] -def test_profile_catalog_has_bounded_architecture_and_flag_sets(): - assert len(common.PROFILES) == len(common.DEFAULT_PROFILES) == 18 - assert set(common.DEFAULT_PROFILES) == set(common.PROFILES) - assert common.ARCHES == ("75", "80", "89", "90a", "100") - assert list(common.FLAG_PROFILES.values()) == [[], ["-O3"], ["-O3", "--use_fast_math"]] - assert common.PROFILES["sm90a-default"] == ("90a", []) - assert common.PROFILES["sm100-O3-fast-math"] == ("100", ["-O3", "--use_fast_math"]) - assert "cutlass" not in common.HEADER.lower() +def test_validation_does_not_read_gigabyte_payloads(cache, monkeypatch): + original = Path.open -def test_observed_explicit_architecture_profiles_preserve_exact_ordered_flags(): - assert common.EXPLICIT_PROFILES == { - "sm100a-O3-fast-math-arch": ("100a", ["-O3", "--use_fast_math", "-arch=sm_100a"]), - "sm100-O3-arch-fast-math": ("100", ["-O3", "-arch=sm_100", "--use_fast_math"]), - "sm80-O3-cxx17-arch-fast-math": ( - "80", - ["-O3", "-std=c++17", "-arch=sm_80", "--use_fast_math"], - ), - } + def guarded(path, *args, **kwargs): + assert path.suffix != ".pch" + return original(path, *args, **kwargs) + monkeypatch.setattr(Path, "open", guarded) + assert common.read_manifest(cache.entry, cache.flags, GENERATION)["files"] -@pytest.mark.parametrize( - ("arch", "expected"), [("75", "7.5"), ("90a", "9.0a"), ("100", "10.0"), ("100a", "10.0a")] -) -def test_capture_uses_curated_architecture_including_blackwell_suffix(arch, expected): - assert warmer.torch_arch(arch) == expected - -def test_ready_validation_never_reads_pch_payloads(tmp_path, monkeypatch): - args = command(tmp_path) - entry, _ = ready(tmp_path, monkeypatch, args) - original_open = Path.open - - def guarded_open(path, *args, **kwargs): - assert path.suffix != ".pch", "PCH payload read during metadata validation" - return original_open(path, *args, **kwargs) - - monkeypatch.setattr(Path, "open", guarded_open) - assert common.read_manifest(entry, common.compile_flags(args), GENERATION)["files"] - - -def test_warm_command_retains_module_macro_and_publishes_only_complete_entry(tmp_path, monkeypatch): - args = command(tmp_path) - flags = common.compile_flags(args) - warm_flags = common.strip_outputs(args, drop_module=False) - monkeypatch.setenv("KERNELBOT_NVCC_PCH_ROOT", str(tmp_path / "warm")) - monkeypatch.setattr(warmer, "capture", lambda *args: (flags, warm_flags)) +@pytest.mark.parametrize("complete", [True, False]) +def test_warmer_retains_module_macro_and_publishes_atomically(cache, monkeypatch, complete): + c = cache + monkeypatch.setenv("KERNELBOT_NVCC_PCH_ROOT", str(c.tmp_path / "warm")) + captured = c.flags, common.strip_outputs(c.args, False) + monkeypatch.setattr(warmer, "capture", lambda *a: captured) def compile_header(argv, **kwargs): assert "-DTORCH_EXTENSION_NAME=first" in argv - pch_dir = Path(argv[argv.index("--pch-dir") + 1]) - assert not (pch_dir.parent / "manifest.json").exists() - (pch_dir / "cicc_ab_12.pch").write_bytes(b"device") - (pch_dir / "cudafe_ab_12.pch").write_bytes(b"frontend") + entry = Path(argv[argv.index("--pch-dir") + 1]).parent + assert not (entry / "manifest.json").exists() + for item in c.files[:2 if complete else 1]: + (entry / item["path"]).write_bytes(b"stub") return subprocess.CompletedProcess(argv, 0, "", "") monkeypatch.setattr(warmer.subprocess, "run", compile_header) - result = warmer.warm_profile("sm75-default", GENERATION) - entry = common.cache_root() / GENERATION / result["key"] - assert common.read_manifest(entry, flags, GENERATION) == result - assert not (entry / "manifest.json.tmp").exists() + entry = common.cache_root() / GENERATION / c.entry.name + if not complete: + with pytest.raises(ValueError): + warmer.warm_profile("sm75-default", GENERATION) + assert not (entry / "manifest.json").exists() + else: + result = warmer.warm_profile("sm75-default", GENERATION) + assert common.read_manifest(entry, c.flags, GENERATION) == result + + +def test_warmer_accepts_only_curated_profiles(monkeypatch): + monkeypatch.setattr(warmer, "warm_profile", lambda *a: pytest.fail("Unexpected compilation")) + with pytest.raises(ValueError, match="fixed profiles"): + warmer.warm(["-include /user/header.h"]) + assert common.HEADERS == ("torch/types.h", "cuda.h", "cuda_runtime.h") + for arch, expected in (("75", "7.5"), ("90a", "9.0a"), ("100", "10.0"), ("100a", "10.0a")): + assert warmer.torch_arch(arch) == expected diff --git a/tests/test_nvcc_release_validation.py b/tests/test_nvcc_release_validation.py deleted file mode 100644 index 8367988f..00000000 --- a/tests/test_nvcc_release_validation.py +++ /dev/null @@ -1,212 +0,0 @@ -"""Exercise release-matrix selection and pass gates without running Modal jobs.""" - -import ast -import copy -import os -import runpy -from pathlib import Path -from types import SimpleNamespace -from unittest.mock import patch - -import pytest - - -@pytest.fixture(scope="module") -def harness(): - root = Path(__file__).parents[1] - with pytest.MonkeyPatch.context() as monkeypatch, patch.dict(os.environ): - monkeypatch.syspath_prepend(str(root / "src/runners")) - namespace = runpy.run_path(str(root / "scripts/validate_nvcc_pch.py")) - namespace["profile_catalog"]() - return SimpleNamespace(**namespace) - - -def test_default_matrix_preserves_twelve_fresh_cpu_gpu_cases(harness): - jobs = harness.matrix_jobs(harness.DEFAULT_GPUS, harness.CASES, "T4", [], 1) - assert len(jobs) == 12 - assert {case for gpu, case, _, _ in jobs if gpu == "T4"} == set(harness.CASES) - assert {case for gpu, case, _, _ in jobs if gpu != "T4"} == set(harness.BASELINE_CASES) - - -def test_all_gpu_matrix_adds_l4_and_production_a100_resource(harness): - jobs = harness.matrix_jobs(harness.GPUS, harness.CASES, "T4", [], 1) - assert len(jobs) == 16 - assert harness.GPU_RESOURCES["A100"] == "A100-80GB" - assert harness.GPU_RESOURCES.get("L4", "L4") == "L4" - assert {profile for gpu, _, profile, _ in jobs if gpu == "A100"} == {"sm80-default"} - assert {profile for gpu, _, profile, _ in jobs if gpu == "L4"} == {"sm89-default"} - assert harness.make_config("A100", "default")["arch"] == "80" - assert harness.make_config("L4", "default")["arch"] == "89" - - -def test_observed_sm80_profile_runs_only_on_a100(harness): - profile = "sm80-O3-cxx17-arch-fast-math" - jobs = harness.matrix_jobs(harness.GPUS, harness.BASELINE_CASES, "T4", [profile], 1) - assert jobs == [ - ("A100", "default", profile, 1), - ("A100", "disabled", profile, 1), - ] - assert harness.make_config("A100", "default", profile)["arch"] == "80" - with pytest.raises(ValueError, match="incompatible"): - harness.make_config("L4", "default", profile) - - -def test_selected_profiles_repeat_paired_baselines_and_alternate_order(harness): - jobs = harness.matrix_jobs(["B200"], harness.BASELINE_CASES, "T4", ["sm100-O3"], 2) - assert jobs == [ - ("B200", "default", "sm100-O3", 1), - ("B200", "disabled", "sm100-O3", 1), - ("B200", "disabled", "sm100-O3", 2), - ("B200", "default", "sm100-O3", 2), - ] - - -def test_cutlass_pair_uses_generic_o3_profile_without_expanding_default_matrix(harness): - jobs = harness.matrix_jobs(["T4"], harness.CUTLASS_CASES, "T4", [], 2) - assert jobs == [ - ("T4", "cutlass", "sm75-O3", 1), - ("T4", "cutlass-disabled", "sm75-O3", 1), - ("T4", "cutlass-disabled", "sm75-O3", 2), - ("T4", "cutlass", "sm75-O3", 2), - ] - assert not set(harness.CASES) & set(harness.CUTLASS_CASES) - with pytest.raises(ValueError, match="Select T4"): - harness.matrix_jobs(["B200"], harness.CUTLASS_CASES, "T4", [], 1) - - -def test_cutlass_fixture_has_real_gemm_and_tail_zero_stream_checks(harness): - on = harness.make_config("T4", "cutlass") - off = harness.make_config("T4", "cutlass-disabled") - assert on == off - assert on["arch"] == "75" - source = on["sources"]["submission.py"] - assert "cutlass::gemm::device::Gemm" in source - assert "c10::cuda::getCurrentCUDAStream" in source - assert "torch.cuda.stream(stream)" in source - tree = ast.parse(source) - call = next( - node - for node in ast.walk(tree) - if isinstance(node, ast.Call) - and isinstance(node.func, ast.Name) - and node.func.id == "load_inline" - ) - kwargs = {keyword.arg: keyword.value for keyword in call.keywords} - assert ast.literal_eval(kwargs["extra_cuda_cflags"]) == ["-O3"] - assert ast.literal_eval(kwargs["no_implicit_headers"]) is False - assert len(on["tests"]) == 9 - assert any(test["m"] == 129 and test["n"] == 136 for test in on["tests"]) - assert sum(test.get("zero", 0) for test in on["tests"]) == 4 - assert all(test["nondefault"] == 1 for test in on["tests"]) - assert on["benchmarks"][0]["nondefault"] == 0 - assert "a.double() @ b.double().T" in on["sources"]["reference.py"] - for name, contents in on["sources"].items(): - ast.parse(contents, filename=name) - - -def test_explicit_arch_profile_sets_architecture_and_exact_extra_flags(harness, monkeypatch): - profile, flags = "sm100a-test", ["-O3", "--use_fast_math", "-arch=sm_100a"] - catalog = {profile: ("100a", flags)} - monkeypatch.setitem(harness.make_config.__globals__, "profile_catalog", lambda: catalog) - jobs = harness.matrix_jobs(["B200"], harness.BASELINE_CASES, "T4", [profile], 1) - assert {job[0] for job in jobs} == {"B200"} - on = harness.make_config("B200", "default", profile) - off = harness.make_config("B200", "disabled", profile) - assert on == off - assert on["arch"] == "100a" - tree = ast.parse(on["sources"]["submission.py"]) - calls = [ - node - for node in ast.walk(tree) - if isinstance(node, ast.Call) - and isinstance(node.func, ast.Name) - and node.func.id == "load_inline" - ] - kwargs = {keyword.arg: keyword.value for keyword in calls[0].keywords} - assert ast.literal_eval(kwargs["extra_cuda_cflags"]) == flags - - -@pytest.mark.parametrize("profile", ["sm100-default", "-O3; arbitrary-command"]) -def test_profile_selection_rejects_unknown_or_incompatible_architecture(harness, profile): - with pytest.raises(ValueError): - harness.matrix_jobs(["T4"], harness.BASELINE_CASES, "T4", [profile], 1) - - -def test_comparison_requires_paired_successful_observations(harness): - def row(case, seconds, passed=True): - return { - "gpu": "T4", - "profile": "sm75-default", - "case": case, - "passed": passed, - "cpu": {"packet": {"info": {"duration": seconds}}}, - } - - rows = [row("default", 10), row("default", 20), row("disabled", 25), row("disabled", 35)] - result = harness.comparisons(rows)[0] - assert result["median_pch_on_seconds"] == 15 - assert result["median_pch_off_seconds"] == 30 - assert result["speedup"] == 2 - assert result["reduction_percent"] == 50 - assert result["cpu_compile_seconds"]["default"] == [10, 20] - assert harness.comparisons(rows[:-1]) == [] - assert harness.comparisons(rows + [row("default", 100, False)]) == [] - - # A vectoradd profile run and the CUTLASS fixture must remain separate even - # when both use the same generic framework-header profile and GPU. - cutlass = [row("cutlass", 40), row("cutlass-disabled", 50)] - results = harness.comparisons(rows + cutlass) - assert {item["workload"]: item["speedup"] for item in results} == { - "vectoradd": 2, - "cutlass": 1.25, - } - - -@pytest.fixture -def successful_pair(): - toolchain = {"generation": "same-generation", "image_id": "same-image"} - event = {"status": "hit", "frontends": ["cicc", "cudafe"], "profile": "sm75-default"} - run = {"success": True, "passed": True, "result": {"check": "pass", "benchmark.0.mean": 1}} - return ( - { - "packet": {"info": {"status": "compiled", "artifacts": 1}, "artifacts": b"artifact"}, - "toolchain": toolchain, - "events": [event], - }, - { - "toolchain": toolchain, - "events": [], - "result": { - "success": True, - "error": "", - "runs": {name: {"run": run} for name in ("test", "benchmark", "leaderboard")}, - "cpu_compile": {"status": "reused", "reused": 6, "fallbacks": 0}, - }, - }, - ) - - -def test_success_requires_matching_profile_and_no_gpu_compilation(harness, successful_pair): - cpu, gpu = successful_pair - harness.verify_result("default", cpu, gpu, "sm75-default") - with pytest.raises(AssertionError): - harness.verify_result("default", cpu, gpu, "sm75-O3") - gpu["events"] = [{"status": "bypass"}] - with pytest.raises(AssertionError): - harness.verify_result("default", cpu, gpu, "sm75-default") - - -def test_changed_header_requires_rejection_and_gpu_pch_hit(harness, successful_pair): - cpu, gpu = copy.deepcopy(successful_pair) - gpu["events"] = cpu["events"] - info = { - "status": "fallback", - "reused": 0, - "fallbacks": 2, - "reason": "Build dependency changed: /tmp/user.h", - } - gpu["result"]["cpu_compile"] = info - harness.verify_result("changed-header", cpu, gpu, "sm75-default") - info["reused"] = 1 - with pytest.raises(AssertionError): - harness.verify_result("changed-header", cpu, gpu, "sm75-default") diff --git a/tests/test_nvcc_toolchain.py b/tests/test_nvcc_toolchain.py index 195c4531..7f4aefc8 100644 --- a/tests/test_nvcc_toolchain.py +++ b/tests/test_nvcc_toolchain.py @@ -1,4 +1,4 @@ -"""Private compiler installation and cache-generation release gates, without CUDA.""" +"""Compiler installation must fail closed and fingerprint all supporting inputs.""" import builtins import hashlib @@ -11,46 +11,30 @@ import pytest SPEC = importlib.util.spec_from_file_location( - "kernelbot_nvcc_toolchain", Path(__file__).parents[1] / "src/runners/nvcc_toolchain.py" -) + "kernelbot_nvcc_toolchain", Path(__file__).parents[1] / "src/runners/nvcc_toolchain.py") toolchain = importlib.util.module_from_spec(SPEC) SPEC.loader.exec_module(toolchain) @pytest.fixture def installation(tmp_path, monkeypatch): # noqa: C901 - roots = { - "/nvcc-toolchain-input": tmp_path / "input", - "/usr/local/cuda": tmp_path / "cuda", - "/opt/kernelbot-pch": tmp_path / "pch", - } + roots = {name: tmp_path / str(i) for i, name in enumerate( + ("/nvcc-toolchain-input", "/usr/local/cuda", "/opt/kernelbot-pch"))} source = roots["/nvcc-toolchain-input"] / toolchain.RELEASE cuda, artifacts = roots["/usr/local/cuda"], roots["/opt/kernelbot-pch"] - targets = { - "nvcc": "bin/nvcc", - "cudafe++": "bin/cudafe++", - "nvvm/bin/cicc": "nvvm/bin/cicc", - "nvvm-next/bin/cicc": "nvvm-next/bin/cicc", - } - - def write(path, data): - path.parent.mkdir(parents=True, exist_ok=True) - path.write_bytes(data) - + targets = {name: name if "/" in name else "bin/" + name for name in toolchain.BINARY_SHA256} hashes = {} + files = {cuda / "bin/nvcc.profile": "CICC_PATH = $(TOP)/nvvm/bin\n"} for name, target in targets.items(): - data = f"pinned private compiler: {name}".encode() - write(source / name, data) - hashes[name] = hashlib.sha256(data).hexdigest() - write(cuda / target, f"original toolkit: {name}".encode()) - write(cuda / "bin/nvcc.profile", b"CICC_PATH = $(TOP)/nvvm/bin\nOTHER = keep\n") + files[source / name], files[cuda / target] = name, "original" + hashes[name] = hashlib.sha256(name.encode()).hexdigest() for name in ("bin/ptxas", "bin/nvlink", "bin/fatbinary", "nvvm/libdevice/libdevice.10.bc"): - write(cuda / name, name.encode()) + files[cuda / name] = name for name in ("fingerprint", "nvcc.py", "nvcc_common.py", "nvcc_warm.py"): - write(artifacts / name, name.encode()) - - # The image-build function imports pathlib internally. Redirect only that - # import; pytest and the standard library keep their ordinary Path objects. + files[artifacts / name] = name + for path, text in files.items(): + path.parent.mkdir(parents=True, exist_ok=True) + path.write_text(text) original_import = builtins.__import__ def sandbox_import(name, globals=None, locals=None, fromlist=(), level=0): @@ -58,24 +42,11 @@ def sandbox_import(name, globals=None, locals=None, fromlist=(), level=0): return SimpleNamespace(Path=lambda path: roots.get(str(path), Path(path))) return original_import(name, globals, locals, fromlist, level) - monkeypatch.setattr(builtins, "__import__", sandbox_import) - state = SimpleNamespace( - source=source, - cuda=cuda, - artifacts=artifacts, - targets=targets, - hashes=hashes, - calls=[], - version="Cuda compilation tools, release 13.4, V13.4.99\n", - backend="nvvm-next/bin/cicc", - packages=( - "cuda-toolkit-13-4\t13.4.3-1\tinstall ok installed\n" - "cuda-nvcc-13-4\t13.4.92-1\tinstall ok installed\n" - "libcublas-13-4\t13.4.0.1-1\tinstall ok installed\n" - "removed-13-4\t1.0-1\tdeinstall ok config-files\n" - ), - fail_arch=None, - ) + state = SimpleNamespace(source=source, cuda=cuda, artifacts=artifacts, targets=targets, + hashes=hashes, calls=[], version="V13.4.99", + backend="nvvm-next/bin/cicc", + packages="cuda-toolkit-13-4\t13.4.3-1\tinstall ok installed\n", + fail_arch=None) def output(command, **kwargs): state.calls.append(command) @@ -87,155 +58,66 @@ def run(command, **kwargs): state.calls.append(command) if state.fail_arch and f"-arch=sm_{state.fail_arch}" in command: raise subprocess.CalledProcessError(1, command) - Path(command[command.index("-o") + 1]).write_bytes(b"compiled object") return subprocess.CompletedProcess(command, 0) + monkeypatch.setattr(builtins, "__import__", sandbox_import) monkeypatch.setattr(subprocess, "check_output", output) monkeypatch.setattr(subprocess, "run", run) state.install = lambda: toolchain.install_nvcc_toolchain(toolchain.RELEASE, hashes) return state -def test_installs_verified_compilers_and_publishes_reproducible_generation(installation): - state = installation - state.install() - for name, target in state.targets.items(): - installed = state.cuda / target - assert installed.read_bytes() == (state.source / name).read_bytes() - assert installed.stat().st_mode & 0o777 == 0o755 - profile = (state.cuda / "bin/nvcc.profile").read_text() - assert "CICC_PATH = $(TOP)/nvvm-next/bin" in profile - assert "OTHER = keep" in profile - - manifest = json.loads((state.artifacts / "nvcc-toolchain.json").read_text()) - generation = manifest.pop("fingerprint") - assert manifest["smoke_architectures"] == ["75", "80", "89", "90a", "100", "100a"] - compiled = [call[1] for call in state.calls if "-c" in call and "--dryrun" not in call] - assert compiled == [f"-arch=sm_{arch}" for arch in manifest["smoke_architectures"]] - assert manifest["headers_fingerprint"] == "fingerprint" - assert manifest["cuda_packages"] == { - "cuda-toolkit-13-4": "13.4.3-1", - "cuda-nvcc-13-4": "13.4.92-1", - "libcublas-13-4": "13.4.0.1-1", - } - assert generation == hashlib.sha256(json.dumps(manifest, sort_keys=True).encode()).hexdigest() - assert (state.artifacts / "nvcc-fingerprint").read_text().strip() == generation - state.install() - assert (state.artifacts / "nvcc-fingerprint").read_text().strip() == generation - - -@pytest.mark.parametrize("damaged", ["nvcc", "cudafe++", "nvvm/bin/cicc", "nvvm-next/bin/cicc"]) -def test_checksum_failure_changes_no_installed_binary(installation, damaged): - state = installation - original = {target: (state.cuda / target).read_bytes() for target in state.targets.values()} - (state.source / damaged).write_bytes(b"unverified compiler") - with pytest.raises(ValueError, match="checksum mismatch"): - state.install() - assert original == { - target: (state.cuda / target).read_bytes() for target in state.targets.values() - } - assert not state.calls - assert not (state.artifacts / "nvcc-fingerprint").exists() - - -@pytest.mark.parametrize( - "missing", ["nvcc", "cudafe++", "nvvm/bin/cicc", "nvvm-next/bin/cicc", None] -) -def test_checksum_manifest_must_cover_exactly_installed_binaries(installation, missing): - state = installation - original = {target: (state.cuda / target).read_bytes() for target in state.targets.values()} - if missing: - del state.hashes[missing] +def test_install_checks_every_architecture_and_is_reproducible(installation): + s = installation + s.install() + for name, target in s.targets.items(): + assert (s.cuda / target).read_bytes() == (s.source / name).read_bytes() + assert (s.cuda / target).stat().st_mode & 0o777 == 0o755 + compiled = [c[1] for c in s.calls if "-c" in c and "--dryrun" not in c] + assert compiled == [f"-arch=sm_{a}" for a in ("75", "80", "89", "90a", "100", "100a")] + before = (s.artifacts / "nvcc-fingerprint").read_text() + manifest = json.loads((s.artifacts / "nvcc-toolchain.json").read_text()) + assert manifest["fingerprint"] == before.strip() + s.install() + assert (s.artifacts / "nvcc-fingerprint").read_text() == before + + +@pytest.mark.parametrize("failure", [*toolchain.BINARY_SHA256, "missing-hash", "extra-hash", + "profile", "version", "backend", "smoke", "packages"]) +def test_bad_installation_never_publishes_ready_fingerprint(installation, failure): + s = installation + if failure in s.hashes: + (s.source / failure).write_bytes(b"corrupt") + elif failure == "missing-hash": + del s.hashes["nvcc"] + elif failure == "extra-hash": + s.hashes["unexpected"] = "0" * 64 + elif failure == "profile": + (s.cuda / "bin/nvcc.profile").write_text("CICC_PATH=one\nCICC_PATH=two\n") + else: + field, value = {"version": ("version", "V13.3"), "backend": ("backend", "nvvm/bin/cicc"), + "smoke": ("fail_arch", "100a"), "packages": ("packages", "")}[failure] + setattr(s, field, value) + with pytest.raises((ValueError, subprocess.CalledProcessError)): + s.install() + assert not (s.artifacts / "nvcc-fingerprint").exists() + if failure in toolchain.BINARY_SHA256 or failure.endswith("hash"): + assert not s.calls + assert all((s.cuda / target).read_text() == "original" for target in s.targets.values()) + + +@pytest.mark.parametrize("root,name", [ + ("artifacts", name) for name in ("fingerprint", "nvcc.py", "nvcc_common.py", "nvcc_warm.py") +] + [("cuda", name) for name in ("bin/ptxas", "bin/nvlink", "bin/fatbinary", + "nvvm/libdevice/libdevice.10.bc")] + [(None, "packages")]) +def test_changed_supporting_input_invalidates_generation(installation, root, name): + s = installation + s.install() + before = (s.artifacts / "nvcc-fingerprint").read_text() + if root: + path = getattr(s, root) / name + path.write_bytes(path.read_bytes() + b" changed") else: - state.hashes["unrecognized-compiler"] = "a" * 64 - with pytest.raises(ValueError, match="exactly the four NVIDIA compiler binaries"): - state.install() - assert original == { - target: (state.cuda / target).read_bytes() for target in state.targets.values() - } - assert not state.calls - assert not (state.artifacts / "nvcc-fingerprint").exists() - - -@pytest.mark.parametrize("profile", ["OTHER = keep\n", "CICC_PATH = one\nCICC_PATH = two\n"]) -def test_ambiguous_backend_profile_blocks_release(installation, profile): - state = installation - (state.cuda / "bin/nvcc.profile").write_text(profile) - with pytest.raises(ValueError, match="exactly one CICC_PATH"): - state.install() - assert not state.calls - assert not (state.artifacts / "nvcc-fingerprint").exists() - - -@pytest.mark.parametrize( - ("field", "value", "message"), - [ - ( - "version", - "Cuda compilation tools, release 13.4, V13.4.92\n", - "Unexpected NVIDIA compiler", - ), - ("backend", "nvvm/bin/cicc", "Wrong CUDA compiler backend"), - ], -) -def test_version_and_backend_mismatch_block_release(installation, field, value, message): - state = installation - setattr(state, field, value) - with pytest.raises(ValueError, match=message): - state.install() - assert not (state.artifacts / "nvcc-toolchain.json").exists() - assert not (state.artifacts / "nvcc-fingerprint").exists() - - -def test_failed_architecture_smoke_blocks_release(installation): - state = installation - state.fail_arch = "100a" - with pytest.raises(subprocess.CalledProcessError): - state.install() - assert not (state.artifacts / "nvcc-toolchain.json").exists() - assert not (state.artifacts / "nvcc-fingerprint").exists() - - -def test_supporting_library_upgrade_invalidates_cache_generation(installation): - state = installation - state.install() - before = (state.artifacts / "nvcc-fingerprint").read_text() - state.packages = state.packages.replace( - "libcublas-13-4\t13.4.0.1-1", "libcublas-13-4\t13.4.0.2-1" - ) - state.install() - assert (state.artifacts / "nvcc-fingerprint").read_text() != before - - -@pytest.mark.parametrize("packages", ["", "removed-13-4\t1.0-1\tdeinstall ok config-files\n"]) -def test_missing_installed_package_inventory_blocks_release(installation, packages): - state = installation - state.packages = packages - with pytest.raises(ValueError, match="No installed CUDA 13.4 packages"): - state.install() - assert not (state.artifacts / "nvcc-fingerprint").exists() - - -@pytest.mark.parametrize( - ("root", "name"), - [ - ("artifacts", "fingerprint"), - ("artifacts", "nvcc.py"), - ("artifacts", "nvcc_common.py"), - ("artifacts", "nvcc_warm.py"), - ("cuda", "bin/ptxas"), - ("cuda", "bin/nvlink"), - ("cuda", "bin/fatbinary"), - ("cuda", "nvvm/libdevice/libdevice.10.bc"), - ], -) -def test_related_toolchain_and_wrapper_changes_invalidate_cache_generation( - installation, root, name -): - state = installation - state.install() - before = (state.artifacts / "nvcc-fingerprint").read_text() - path = getattr(state, root) / name - path.write_bytes(path.read_bytes() + b" changed") - state.install() - assert (state.artifacts / "nvcc-fingerprint").read_text() != before + s.packages = s.packages.replace("13.4.3-1", "13.4.3-2") + s.install() + assert (s.artifacts / "nvcc-fingerprint").read_text() != before From 4861e9154c421386f64400685c8c29a6024bfb27 Mon Sep 17 00:00:00 2001 From: Mark Saroufim Date: Sun, 15 Nov 2026 14:24:47 -0800 Subject: [PATCH 3/4] Support multiple device PCH variants for explicit CUDA architectures --- src/libkernelbot/inline_artifacts.py | 4 ++-- src/runners/pch/nvcc_common.py | 12 ++++++++---- tests/test_nvcc_pch.py | 14 ++++++++++++-- tests/test_python_precompile.py | 17 +++++++++++------ 4 files changed, 33 insertions(+), 14 deletions(-) diff --git a/src/libkernelbot/inline_artifacts.py b/src/libkernelbot/inline_artifacts.py index 69acdbfb..ce03a631 100644 --- a/src/libkernelbot/inline_artifacts.py +++ b/src/libkernelbot/inline_artifacts.py @@ -64,13 +64,13 @@ def _pch_dependency_digest(path: Path) -> str: ): raise RuntimeError("NVCC PCH dependency manifest does not match its cache entry") files = manifest.get("files", []) - if len(files) != 2: + if len(files) < 2: raise RuntimeError("NVCC PCH dependency manifest is incomplete") frontends, names = set(), set() for item in files: name, size = item.get("path", ""), item.get("bytes") match = re.fullmatch(r"pch/(cicc|cudafe)_[A-Za-z0-9_]+\.pch", name) - if not match or type(size) is not int or size <= 0: + if not match or name in names or type(size) is not int or size <= 0: raise RuntimeError("Invalid NVCC PCH dependency file metadata") cache_file = entry / name if ( diff --git a/src/runners/pch/nvcc_common.py b/src/runners/pch/nvcc_common.py index 8cc1211b..605e1ee0 100644 --- a/src/runners/pch/nvcc_common.py +++ b/src/runners/pch/nvcc_common.py @@ -171,18 +171,22 @@ def validate_manifest(manifest, entry, flags, fingerprint): def validate_files(entry, files): - if len(files) != 2: + if len(files) < 2: raise ValueError("PCH profile must contain both frontends") - frontends = set() + frontends, paths = set(), set() for item in files: path = Path(item["path"]) - if path.parent != Path("pch") or not PCH_NAME.fullmatch(path.name): + if path in paths or path.parent != Path("pch") or not PCH_NAME.fullmatch(path.name): raise ValueError("Invalid PCH manifest path") + paths.add(path) target = entry / path if not target.resolve().is_relative_to(entry.resolve()): raise ValueError("PCH file escapes its profile") size = item["bytes"] - if not isinstance(size, int) or size <= 0 or target.stat().st_size != size: + if ( + type(size) is not int or size <= 0 + or not target.is_file() or target.stat().st_size != size + ): raise ValueError("PCH file is missing or truncated") frontends.add(path.name.split("_", 1)[0]) if frontends != {"cicc", "cudafe"}: diff --git a/tests/test_nvcc_pch.py b/tests/test_nvcc_pch.py index f45572ca..c668e1fb 100644 --- a/tests/test_nvcc_pch.py +++ b/tests/test_nvcc_pch.py @@ -121,7 +121,8 @@ def test_consumption_evidence_and_cache_only_retry(cache, kind, code, status): assert event["frontends"] == ["cicc", "cudafe"] -@pytest.mark.parametrize("change", ["truncated", "header", "generation", "escape", "partial"]) +@pytest.mark.parametrize("change", [ + "truncated", "header", "generation", "escape", "partial", "duplicate", "duplicate-alias"]) def test_bad_ready_metadata_falls_back_before_compiler(cache, change): c = cache if change == "truncated": @@ -132,13 +133,22 @@ def test_bad_ready_metadata_falls_back_before_compiler(cache, change): c.manifest["generation"] = "b" * 64 elif change == "escape": c.files[0]["path"] = "../foreign.pch" + elif change.startswith("duplicate"): + path = c.files[0]["path"] + path = path.replace("pch/", "pch/./") if change.endswith("alias") else path + c.files.append({"path": path, "bytes": 4}) else: c.files.pop() (c.entry / "manifest.json").write_text(json.dumps(c.manifest)) assert wrapper.main(c.args) == 0 and c.calls == [(c.args, False)] -def test_validation_does_not_read_gigabyte_payloads(cache, monkeypatch): +@pytest.mark.parametrize("count", [2, 3]) +def test_validation_does_not_read_gigabyte_payloads(cache, monkeypatch, count): + if count == 3: + (cache.entry / "pch/cicc_cd.pch").write_bytes(b"stub") + cache.files.append({"path": "pch/cicc_cd.pch", "bytes": 4}) + (cache.entry / "manifest.json").write_text(json.dumps(cache.manifest)) original = Path.open def guarded(path, *args, **kwargs): diff --git a/tests/test_python_precompile.py b/tests/test_python_precompile.py index 2ec65609..f3884319 100644 --- a/tests/test_python_precompile.py +++ b/tests/test_python_precompile.py @@ -258,8 +258,8 @@ def test_empty_nvcc_fingerprint_cannot_authorize_reuse(tmp_path, monkeypatch): nvcc_fingerprint() -@pytest.fixture -def trusted_pch(tmp_path, monkeypatch): +@pytest.fixture(params=[2, 3]) +def trusted_pch(tmp_path, monkeypatch, request): generation, key = "a" * 64, "b" * 64 root = tmp_path / "trusted-cache" entry = root / generation / key @@ -267,8 +267,8 @@ def trusted_pch(tmp_path, monkeypatch): header = entry / "gpumode.cuh" header.write_text("#include \n#pragma hdrstop\n") files = [] - for frontend in ("cicc", "cudafe"): - name = f"pch/{frontend}_123456.pch" + for name in ("cicc_123456", "cudafe_123456", "cicc_654321")[:request.param]: + name = f"pch/{name}.pch" (entry / name).write_bytes(b"compiler cache contents") files.append({"path": name, "bytes": (entry / name).stat().st_size}) manifest = { @@ -318,7 +318,8 @@ def small_files_only(path, *args, **kwargs): _check_dependencies(dependencies, gpu) -@pytest.mark.parametrize("change", ["manifest", "header", "generation", "size", "missing"]) +@pytest.mark.parametrize("change", [ + "manifest", "header", "generation", "size", "missing", "duplicate"]) def test_trusted_pch_dependency_changes_reject_replay(trusted_pch, tmp_path, change): entry, manifest, fingerprint = trusted_pch path = entry / "pch/cicc_123456.pch" @@ -332,9 +333,13 @@ def test_trusted_pch_dependency_changes_reject_replay(trusted_pch, tmp_path, cha fingerprint.write_text("c" * 64 + "\n") elif change == "size": path.write_bytes(b"truncated") + elif change == "duplicate": + manifest["files"].append(manifest["files"][0].copy()) + (entry / "manifest.json").write_text(json.dumps(manifest)) else: (entry / "manifest.json").unlink() - with pytest.raises((RuntimeError, FileNotFoundError)): + message = "Invalid NVCC PCH dependency file metadata" if change == "duplicate" else None + with pytest.raises((RuntimeError, FileNotFoundError), match=message): _check_dependencies(dependencies, tmp_path) From 9ca0c6a7df99060e2c6acf982ce38ff1c97d5e5c Mon Sep 17 00:00:00 2001 From: Mark Saroufim Date: Sun, 15 Nov 2026 14:35:52 -0800 Subject: [PATCH 4/4] Exclude historical C++17 tuple from framework PCH warmup --- docs/modal-pch-cache.md | 4 ++-- src/runners/pch/nvcc_common.py | 5 +---- 2 files changed, 3 insertions(+), 6 deletions(-) diff --git a/docs/modal-pch-cache.md b/docs/modal-pch-cache.md index 3c8e7995..56c98fd5 100644 --- a/docs/modal-pch-cache.md +++ b/docs/modal-pch-cache.md @@ -35,7 +35,7 @@ speedup. Set `CXX=/usr/bin/g++` before building to bypass the wrapper entirely. The runner uses NVIDIA's checksum-pinned NVCC 13.4.99 drop with the complete CUDA 13.4 toolkit. `PYTORCH_NVCC` automatically reuses `torch/types.h`, `cuda.h`, and `cuda_runtime.h` from `kernelbot-nvcc-pch`. CUTLASS and user headers stay outside -this cache. The trusted warmer builds 18 profiles from six observed flag tuples; +this cache. The trusted warmer builds 17 profiles from five observed flag tuples; submission workers mount them read-only. Unknown flags, custom compiler inputs, and sources without the implicit header prefix compile normally. Corrupt caches fall back to normal compilation. CPU artifact replay checks compiler identity @@ -48,7 +48,7 @@ PYTHONPATH=src/runners uv run python scripts/upload_nvcc_toolchain.py --source / ``` CI warms both caches before deployment. `KERNELBOT_NVCC_PCH_PROFILES` selects a -comma-separated subset; empty warms all 18. Manual warmup uses +comma-separated subset; empty warms all 17. Manual warmup uses `modal run src/runners/modal_runner.py::warm_nvcc_pch --profiles ...`. Set `KERNELBOT_NVCC_PCH=0` to disable device PCH. Set both `KERNELBOT_NVCC_TOOLCHAIN=0` and `KERNELBOT_NVCC_PCH=0` to restore CUDA 13.3. diff --git a/src/runners/pch/nvcc_common.py b/src/runners/pch/nvcc_common.py index 605e1ee0..3948e66f 100644 --- a/src/runners/pch/nvcc_common.py +++ b/src/runners/pch/nvcc_common.py @@ -29,11 +29,8 @@ EXPLICIT_PROFILES = { "sm100a-O3-fast-math-arch": ("100a", ["-O3", "--use_fast_math", "-arch=sm_100a"]), "sm100-O3-arch-fast-math": ("100", ["-O3", "-arch=sm_100", "--use_fast_math"]), - "sm80-O3-cxx17-arch-fast-math": ( - "80", - ["-O3", "-std=c++17", "-arch=sm_80", "--use_fast_math"], - ), } +# Historical C++17 tuples cannot precompile the current Torch headers. PROFILES.update(EXPLICIT_PROFILES) DEFAULT_PROFILES = tuple(PROFILES) ENVIRONMENT_KEYS = (