{ "scope": "Steady full-name groups; unresolved does not imply zero utilization. Shape grouping incomplete.", "groups": [ { "source": "original", "kernel_name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "family_name_hint": "matrix_multiply", "count": 7618, "summed_gpu_ms": 917.505966, "mean_us": 120.43921843003413, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void cutlass::device_kernel, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple, cute::C<(int)160>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)1, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)1, (bool)0, (bool)0, cutlass::bfloat16_t, (int)1>, flash::CollectiveEpilogueFwd, cute::C<(int)128>, cute::C<(int)160>>, cute::tuple, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)1, (bool)1, (bool)0, (bool)0, (int)1>, flash::VarlenDynamicPersistentTileScheduler<(int)128, (int)160, (int)256, (int)128, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)1>>>>(T1::Params)", "family_name_hint": "attention", "count": 848, "summed_gpu_ms": 840.687064, "mean_us": 991.3762547169811, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": "raw/ncu_attention_fa3_pilot_v1" }, { "source": "original", "kernel_name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "family_name_hint": "communication", "count": 4971, "summed_gpu_ms": 454.471394, "mean_us": 91.4245411386039, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "N/A", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 13895, "summed_gpu_ms": 384.26423, "mean_us": 27.6548564231738, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize256x32x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 288, "summed_gpu_ms": 383.824099, "mean_us": 1332.7225659722224, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 672, "summed_gpu_ms": 332.943711, "mean_us": 495.45195089285716, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "family_name_hint": "matrix_multiply", "count": 848, "summed_gpu_ms": 228.023722, "mean_us": 268.89589858490564, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, const T1 *, T2 *)", "family_name_hint": "unknown", "count": 3168, "summed_gpu_ms": 101.231038, "mean_us": 31.95424179292929, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "fmha_cutlassF_f32_aligned_32x128_gmem_sm80(PyTorchMemEffAttention::AttentionKernel::Params)", "family_name_hint": "attention", "count": 48, "summed_gpu_ms": 93.787351, "mean_us": 1953.9031458333332, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": "raw/ncu_attention_fa3_pilot_v1" }, { "source": "original", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 1980, "summed_gpu_ms": 80.488442, "mean_us": 40.65072828282828, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "family_name_hint": "unknown", "count": 1696, "summed_gpu_ms": 65.722455, "mean_us": 38.7514475235849, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 1465, "summed_gpu_ms": 64.085437, "mean_us": 43.74432559726963, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 1474, "summed_gpu_ms": 63.967672, "mean_us": 43.39733514246947, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel>(at::TensorIterator &, c10::ArrayRef, c10::ArrayRef)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef, c10::ArrayRef, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "family_name_hint": "elementwise", "count": 1696, "summed_gpu_ms": 50.882687, "mean_us": 30.001584316037736, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void cudnn::engines_precompiled::nhwcToNchwKernel(cudnn::engines_precompiled::nhwc2nchw_params_t, const T1 *, T2 *)", "family_name_hint": "unknown", "count": 1632, "summed_gpu_ms": 47.09927, "mean_us": 28.85984681372549, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize64x64x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 480, "summed_gpu_ms": 45.873365, "mean_us": 95.56951041666667, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 1392, "summed_gpu_ms": 38.896366, "mean_us": 27.94279166666667, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor>, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 1507, "summed_gpu_ms": 37.886606, "mean_us": 25.140415394824153, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::CUDAFunctorOnSelf_add, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 1454, "summed_gpu_ms": 37.843735, "mean_us": 26.0273280605227, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 2450, "summed_gpu_ms": 33.987061, "mean_us": 13.872269795918369, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize64x192x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 48, "summed_gpu_ms": 32.151325, "mean_us": 669.8192708333333, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::reduce_kernel<(int)128, (int)4, at::native::ReduceOp, unsigned int, float, (int)4, (int)4>>(T3)", "family_name_hint": "unknown", "count": 1440, "summed_gpu_ms": 31.842517, "mean_us": 22.112859027777777, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "sm80_xmma_fprop_implicit_gemm_indexed_wo_smem_tf32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x16x64_stage1_warpsize4x1x1_g1_tensor16x8x8_alignc4_execute_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 48, "summed_gpu_ms": 26.17438, "mean_us": 545.2995833333333, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_on_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 96, "summed_gpu_ms": 25.021707, "mean_us": 260.64278125, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void cutlass::device_kernel, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple, cute::C<(int)176>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)0>, flash::CollectiveEpilogueFwd, cute::C<(int)128>, cute::C<(int)176>>, cute::tuple, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)0, (bool)0, (bool)0, (bool)0>, flash::StaticPersistentTileScheduler<(bool)0>>>>(T1::Params)", "family_name_hint": "attention", "count": 846, "summed_gpu_ms": 24.708999, "mean_us": 29.20685460992908, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": "raw/ncu_attention_fa3_pilot_v1" }, { "source": "original", "kernel_name": "void at::native::::CatArrayBatchedCopy_alignedK_contig::OpaqueType<(unsigned int)4>, unsigned int, (int)4, (int)128, (int)1, (int)16>(T1 *, at::native::::CatArrInputTensorMetadata, at::native::::TensorSizeStride, int, T2)", "family_name_hint": "unknown", "count": 149, "summed_gpu_ms": 22.147394, "mean_us": 148.64022818791946, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_9", "family_name_hint": "unknown", "count": 847, "summed_gpu_ms": 20.424998, "mean_us": 24.11451948051948, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_12", "family_name_hint": "unknown", "count": 845, "summed_gpu_ms": 19.397364, "mean_us": 22.955460355029587, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "triton_poi_fused_gelu_view_13", "family_name_hint": "unknown", "count": 844, "summed_gpu_ms": 18.894503, "mean_us": 22.386851895734598, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_1", "family_name_hint": "unknown", "count": 848, "summed_gpu_ms": 18.807847, "mean_us": 22.179064858490566, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::CUDAFunctor_add, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 654, "summed_gpu_ms": 18.530802, "mean_us": 28.33455963302752, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::::upsample_nearest2d_out_frame(const T1 *, T1 *, unsigned long, unsigned long, unsigned long, unsigned long, unsigned long, float, float)", "family_name_hint": "unknown", "count": 144, "summed_gpu_ms": 17.882843, "mean_us": 124.18640972222222, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "rotary_kernel", "family_name_hint": "unknown", "count": 1696, "summed_gpu_ms": 16.601388, "mean_us": 9.78855424528302, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "nvjet_sm90_tst_256x144_64x4_2x1_v_bz_coopA_bias_TNT", "family_name_hint": "matrix_multiply", "count": 88, "summed_gpu_ms": 16.302988, "mean_us": 185.26122727272727, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "triton_poi_fused_all_to_all_single_clone_split_with_sizes_stack_transpose_view_4", "family_name_hint": "unknown", "count": 848, "summed_gpu_ms": 15.709778, "mean_us": 18.525681603773585, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "triton_poi_fused__to_copy_add_mul_split_squeeze_view_14", "family_name_hint": "unknown", "count": 844, "summed_gpu_ms": 12.442972, "mean_us": 14.742857819905213, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "ncclDevKernel_AllGather_RING_LL(ncclDevKernelArgsStorage<(unsigned long)4096>)", "family_name_hint": "communication", "count": 114, "summed_gpu_ms": 12.404196, "mean_us": 108.80873684210526, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "N/A", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "triton_poi_fused_all_to_all_single_clone_transpose_unsqueeze_view_5", "family_name_hint": "unknown", "count": 1693, "summed_gpu_ms": 11.367168, "mean_us": 6.714216184288246, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "triton_poi_fused_clone_transpose_view_8", "family_name_hint": "unknown", "count": 1692, "summed_gpu_ms": 10.415643, "mean_us": 6.155817375886525, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "triton_poi_fused__to_copy_all_to_all_single_clone_div_mul_sqrt_transpose_view_11", "family_name_hint": "unknown", "count": 847, "summed_gpu_ms": 9.320439, "mean_us": 11.004060212514759, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_3", "family_name_hint": "unknown", "count": 848, "summed_gpu_ms": 9.007955, "mean_us": 10.622588443396225, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "triton_poi_fused_add_view_7", "family_name_hint": "unknown", "count": 848, "summed_gpu_ms": 8.998389, "mean_us": 10.61130778301887, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "triton_poi_fused_silu_view_6", "family_name_hint": "unknown", "count": 848, "summed_gpu_ms": 8.767877, "mean_us": 10.339477594339623, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 116, "summed_gpu_ms": 8.180667, "mean_us": 70.52299137931035, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_2", "family_name_hint": "unknown", "count": 848, "summed_gpu_ms": 8.150009, "mean_us": 9.61085966981132, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "ncclDevKernel_AllReduce_Sum_f32_RING_LL(ncclDevKernelArgsStorage<(unsigned long)4096>)", "family_name_hint": "communication", "count": 32, "summed_gpu_ms": 7.8781, "mean_us": 246.190625, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "N/A", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, const T1 *, T2 *)", "family_name_hint": "unknown", "count": 192, "summed_gpu_ms": 7.390843, "mean_us": 38.493973958333335, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "triton_red_fused__to_copy_pow_sum_view_10", "family_name_hint": "unknown", "count": 847, "summed_gpu_ms": 4.35853, "mean_us": 5.145844155844156, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::::launch_clamp_scalar(at::TensorIteratorBase &, c10::Scalar, c10::Scalar, at::native::detail::ClampLimits)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 1448, "summed_gpu_ms": 2.367554, "mean_us": 1.6350511049723757, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void cutlass::Kernel2(T1::Params)", "family_name_hint": "matrix_multiply", "count": 48, "summed_gpu_ms": 1.97786, "mean_us": 41.205416666666665, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void flash::prepare_varlen_num_blocks_kernel<(int)1, (bool)0>(int, int, int, const int *, const int *, const int *, const int *, const int *, const int *, int, int, int, int, int, cutlass::FastDivmod, cutlass::FastDivmod, int *, int *, int *, int *, int *, bool, bool, bool, int)", "family_name_hint": "attention", "count": 848, "summed_gpu_ms": 1.919233, "mean_us": 2.263246462264151, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": "raw/ncu_attention_fa3_pilot_v1" }, { "source": "original", "kernel_name": "nvjet_sm90_tst_64x8_64x16_2x1_v_bz_bias_TNT", "family_name_hint": "matrix_multiply", "count": 22, "summed_gpu_ms": 1.716068, "mean_us": 78.00309090909091, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::unrolled_elementwise_kernel, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)", "family_name_hint": "elementwise", "count": 53, "summed_gpu_ms": 1.465158, "mean_us": 27.64449056603774, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::::vectorized_layer_norm_kernel(int, T2, const T1 *, const T1 *, const T1 *, T2 *, T2 *, T1 *)", "family_name_hint": "unknown", "count": 18, "summed_gpu_ms": 1.346052, "mean_us": 74.78066666666668, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::unrolled_elementwise_kernel, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)", "family_name_hint": "elementwise", "count": 848, "summed_gpu_ms": 1.307354, "mean_us": 1.541691037735849, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::::CatArrayBatchedCopy_alignedK_contig::OpaqueType<(unsigned int)4>, unsigned int, (int)1, (int)128, (int)1, (int)16>(T1 *, at::native::::CatArrInputTensorMetadata, at::native::::TensorSizeStride, int, T2)", "family_name_hint": "unknown", "count": 848, "summed_gpu_ms": 1.247844, "mean_us": 1.4715141509433962, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void cutlass::Kernel2(T1::Params)", "family_name_hint": "matrix_multiply", "count": 64, "summed_gpu_ms": 1.008263, "mean_us": 15.754109375, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 848, "summed_gpu_ms": 0.921769, "mean_us": 1.086991745283019, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void cutlass::Kernel2(T1::Params)", "family_name_hint": "matrix_multiply", "count": 48, "summed_gpu_ms": 0.898114, "mean_us": 18.710708333333333, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "sm80_xmma_fprop_implicit_gemm_tf32f32_tf32f32_f32_nhwckrsc_nchw_tilesize128x64x32_stage5_warpsize2x2x1_g1_tensor16x8x8_execute_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 48, "summed_gpu_ms": 0.777634, "mean_us": 16.200708333333335, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 89, "summed_gpu_ms": 0.773315, "mean_us": 8.688932584269663, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::CUDAFunctor_add, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 22, "summed_gpu_ms": 0.719714, "mean_us": 32.71427272727273, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 6)]::operator ()() const::[lambda(c10::BFloat16) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 66, "summed_gpu_ms": 0.660513, "mean_us": 10.007772727272728, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "nvjet_sm90_tst_64x8_64x16_4x1_v_bz_bias_TNT", "family_name_hint": "matrix_multiply", "count": 44, "summed_gpu_ms": 0.531841, "mean_us": 12.087295454545455, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::::CatArrayBatchedCopy::OpaqueType<(unsigned int)4>, unsigned int, (int)4, (int)64, (int)64>(T1 *, at::native::::CatArrInputTensorMetadata, at::native::::TensorSizeStride, int, T2)", "family_name_hint": "unknown", "count": 49, "summed_gpu_ms": 0.443041, "mean_us": 9.04165306122449, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "sm80_xmma_gemm_f32f32_f32f32_f32_tn_n_tilesize128x32x8_stage3_warpsize2x2x1_ffma_aligna4_alignc4_execute_kernel__5x_cublas", "family_name_hint": "matrix_multiply", "count": 5, "summed_gpu_ms": 0.369889, "mean_us": 73.9778, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "nvjet_sm90_tst_64x48_64x15_1x4_h_bz_bias_TNT", "family_name_hint": "matrix_multiply", "count": 18, "summed_gpu_ms": 0.301216, "mean_us": 16.734222222222222, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void cask_plugin__5x_cudnn::xmma__5x_cudnn::init_device_workspace_kernel>(T1, bool)", "family_name_hint": "matrix_multiply", "count": 96, "summed_gpu_ms": 0.259776, "mean_us": 2.706, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::reduce_kernel<(int)512, (int)1, at::native::ReduceOp, unsigned int, float, (int)4, (int)4>>(T3)", "family_name_hint": "unknown", "count": 15, "summed_gpu_ms": 0.245344, "mean_us": 16.356266666666667, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::unrolled_elementwise_kernel, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.168992, "mean_us": 42.248, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::round_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.116032, "mean_us": 29.008, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AbsFunctor, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.106144, "mean_us": 26.536, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::::distribution_elementwise_grid_stride_kernel(const at::TensorBase &, double, double, T1)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, T3, T4)::[lambda(curandStatePhilox4_32_10 *) (instance 2)], void at::native::::distribution_nullary_kernel(const at::TensorBase &, double, double, T1)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, T3, T4)::[lambda(curandStatePhilox4_32_10 *) (instance 2)], void at::native::templates::cuda::normal_kernel(const at::TensorBase &, double, double, T1)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, T4, const T5 &, T6)::[lambda(int, float) (instance 1)]>(long, at::PhiloxCudaState, T3, T4)", "family_name_hint": "elementwise", "count": 18, "summed_gpu_ms": 0.097184, "mean_us": 5.399111111111111, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 18, "summed_gpu_ms": 0.094752, "mean_us": 5.264, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl::pow_tensor_tensor_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 6)]::operator ()() const::[lambda(double, double) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 22, "summed_gpu_ms": 0.077953, "mean_us": 3.543318181818182, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::AUnaryFunctor::CompareEqFunctor>, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.075104, "mean_us": 18.776, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::BinaryFunctor::CompareEqFunctor>, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.074976, "mean_us": 18.744, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::unrolled_elementwise_kernel, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)", "family_name_hint": "elementwise", "count": 22, "summed_gpu_ms": 0.060256, "mean_us": 2.738909090909091, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)2, at::native::sin_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 1)]::operator ()() const::[lambda(double) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 22, "summed_gpu_ms": 0.059041, "mean_us": 2.683681818181818, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)2, at::native::cos_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 1)]::operator ()() const::[lambda(double) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 22, "summed_gpu_ms": 0.057952, "mean_us": 2.634181818181818, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::BinaryFunctor>, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.052, "mean_us": 13.0, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::reduce_kernel<(int)512, (int)1, at::native::ReduceOp, unsigned int, bool, (int)4, (int)4>>(T3)", "family_name_hint": "unknown", "count": 4, "summed_gpu_ms": 0.048576, "mean_us": 12.144, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 18, "summed_gpu_ms": 0.042784, "mean_us": 2.376888888888889, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 22, "summed_gpu_ms": 0.037088, "mean_us": 1.6858181818181817, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 22, "summed_gpu_ms": 0.033568, "mean_us": 1.5258181818181817, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::::CatArrayBatchedCopy_vectorized::OpaqueType<(unsigned int)8>, unsigned int, (int)2, (int)128, (int)1, (int)16, (int)2>(char *, at::native::::CatArrInputTensorMetadata, at::native::::TensorSizeStride, int, T2)", "family_name_hint": "unknown", "count": 22, "summed_gpu_ms": 0.025472, "mean_us": 1.1578181818181819, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "sm80_xmma_gemm_f32f32_f32f32_f32_nn_n_tilesize32x32x8_stage3_warpsize1x2x1_ffma_aligna4_alignc4_execute_kernel__5x_cublas", "family_name_hint": "matrix_multiply", "count": 10, "summed_gpu_ms": 0.025024, "mean_us": 2.5024, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)2, at::native::neg_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 6)]::operator ()() const::[lambda(double) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 22, "summed_gpu_ms": 0.023008, "mean_us": 1.0458181818181818, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)2, at::native::BUnaryFunctor>, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 22, "summed_gpu_ms": 0.023008, "mean_us": 1.0458181818181818, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)2, at::native::FillFunctor, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 22, "summed_gpu_ms": 0.017856, "mean_us": 0.8116363636363636, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void ::elementwise_kernel_with_index(T1, T2, function_traits::result_type *)", "family_name_hint": "elementwise", "count": 22, "summed_gpu_ms": 0.017696, "mean_us": 0.8043636363636364, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel>(at::TensorIterator &, c10::ArrayRef, c10::ArrayRef)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef, c10::ArrayRef, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "family_name_hint": "elementwise", "count": 10, "summed_gpu_ms": 0.015008, "mean_us": 1.5008, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel>(at::TensorIteratorBase &, c10::ArrayRef, c10::ArrayRef)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef, c10::ArrayRef, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "family_name_hint": "elementwise", "count": 10, "summed_gpu_ms": 0.014944, "mean_us": 1.4944000000000002, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 15, "summed_gpu_ms": 0.01232, "mean_us": 0.8213333333333334, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::reduce_kernel<(int)512, (int)1, at::native::ReduceOp, unsigned int, bool, (int)4, (int)4>>(T3)", "family_name_hint": "unknown", "count": 5, "summed_gpu_ms": 0.011104, "mean_us": 2.2208, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::reduce_kernel<(int)512, (int)1, at::native::ReduceOp>, unsigned int, float, (int)4, (int)4>>(T3)", "family_name_hint": "unknown", "count": 5, "summed_gpu_ms": 0.010208, "mean_us": 2.0416, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::neg_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 10, "summed_gpu_ms": 0.009857, "mean_us": 0.9857, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void gemvNSP_kernel, cublasGemvTensorStridedBatched, cublasGemvTensorStridedBatched, float>>(T10)", "family_name_hint": "unknown", "count": 5, "summed_gpu_ms": 0.008768, "mean_us": 1.7535999999999998, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void ::elementwise_kernel_with_index(T1, T2, function_traits::result_type *)", "family_name_hint": "elementwise", "count": 10, "summed_gpu_ms": 0.007872, "mean_us": 0.7872, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "std::enable_if::type internal::gemvx::kernel, cublasGemvTensorStridedBatched, cublasGemvTensorStridedBatched, float>>(T13)", "family_name_hint": "unknown", "count": 5, "summed_gpu_ms": 0.00752, "mean_us": 1.504, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl_nocast::CompareEqFunctor>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 5, "summed_gpu_ms": 0.005376, "mean_us": 1.0752000000000002, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "original", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, void at::native::compare_scalar_kernel(at::TensorIteratorBase &, at::native::::OpType, T1)::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 5, "summed_gpu_ms": 0.004992, "mean_us": 0.9984, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_original_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "family_name_hint": "matrix_multiply", "count": 13126, "summed_gpu_ms": 1574.520988, "mean_us": 119.95436446746915, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void cutlass::device_kernel, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple, cute::C<(int)160>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)1, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)1, (bool)0, (bool)0, cutlass::bfloat16_t, (int)1>, flash::CollectiveEpilogueFwd, cute::C<(int)128>, cute::C<(int)160>>, cute::tuple, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)1, (bool)1, (bool)0, (bool)0, (int)1>, flash::VarlenDynamicPersistentTileScheduler<(int)128, (int)160, (int)256, (int)128, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)1>>>>(T1::Params)", "family_name_hint": "attention", "count": 1459, "summed_gpu_ms": 1428.09285, "mean_us": 978.8162097326937, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": "raw/ncu_attention_fa3_pilot_v1" }, { "source": "rebased", "kernel_name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "family_name_hint": "communication", "count": 7420, "summed_gpu_ms": 695.087986, "mean_us": 93.67762614555257, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "N/A", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "family_name_hint": "matrix_multiply", "count": 1461, "summed_gpu_ms": 388.515463, "mean_us": 265.92434154688567, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 13916, "summed_gpu_ms": 384.311752, "mean_us": 27.616538660534637, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize256x32x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 288, "summed_gpu_ms": 383.262111, "mean_us": 1330.77121875, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 672, "summed_gpu_ms": 337.404766, "mean_us": 502.0904255952381, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "family_name_hint": "unknown", "count": 2918, "summed_gpu_ms": 113.005016, "mean_us": 38.72687320082248, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, const T1 *, T2 *)", "family_name_hint": "unknown", "count": 3168, "summed_gpu_ms": 101.16054, "mean_us": 31.931988636363634, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "fmha_cutlassF_f32_aligned_32x128_gmem_sm80(PyTorchMemEffAttention::AttentionKernel::Params)", "family_name_hint": "attention", "count": 48, "summed_gpu_ms": 93.262298, "mean_us": 1942.9645416666667, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": "raw/ncu_attention_fa3_pilot_v1" }, { "source": "rebased", "kernel_name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel>(at::TensorIterator &, c10::ArrayRef, c10::ArrayRef)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef, c10::ArrayRef, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "family_name_hint": "elementwise", "count": 2921, "summed_gpu_ms": 86.738054, "mean_us": 29.694643615200274, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 2002, "summed_gpu_ms": 81.826604, "mean_us": 40.872429570429574, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 1490, "summed_gpu_ms": 65.249574, "mean_us": 43.79166040268457, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 1480, "summed_gpu_ms": 64.239183, "mean_us": 43.40485337837838, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void cudnn::engines_precompiled::nhwcToNchwKernel(cudnn::engines_precompiled::nhwc2nchw_params_t, const T1 *, T2 *)", "family_name_hint": "unknown", "count": 1632, "summed_gpu_ms": 47.092331, "mean_us": 28.855594975490195, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize64x64x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 480, "summed_gpu_ms": 45.808385, "mean_us": 95.43413541666668, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void cutlass::device_kernel, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple, cute::C<(int)176>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)0>, flash::CollectiveEpilogueFwd, cute::C<(int)128>, cute::C<(int)176>>, cute::tuple, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)0, (bool)0, (bool)0, (bool)0>, flash::StaticPersistentTileScheduler<(bool)0>>>>(T1::Params)", "family_name_hint": "attention", "count": 1458, "summed_gpu_ms": 42.067651, "mean_us": 28.852984224965706, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": "raw/ncu_attention_fa3_pilot_v1" }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 1392, "summed_gpu_ms": 38.899033, "mean_us": 27.94470761494253, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor>, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 1549, "summed_gpu_ms": 37.983605, "mean_us": 24.521371852808265, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::CUDAFunctorOnSelf_add, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 1460, "summed_gpu_ms": 37.832062, "mean_us": 25.912371232876712, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_9", "family_name_hint": "unknown", "count": 1458, "summed_gpu_ms": 36.281512, "mean_us": 24.884438957475997, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 2477, "summed_gpu_ms": 34.015324, "mean_us": 13.732468308437626, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_12", "family_name_hint": "unknown", "count": 1458, "summed_gpu_ms": 33.739949, "mean_us": 23.141254458161864, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_1", "family_name_hint": "unknown", "count": 1461, "summed_gpu_ms": 32.613571, "mean_us": 22.32277275838467, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "triton_poi_fused_gelu_view_13", "family_name_hint": "unknown", "count": 1458, "summed_gpu_ms": 32.262376, "mean_us": 22.12782990397805, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize64x192x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 48, "summed_gpu_ms": 32.146688, "mean_us": 669.7226666666667, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::reduce_kernel<(int)128, (int)4, at::native::ReduceOp, unsigned int, float, (int)4, (int)4>>(T3)", "family_name_hint": "unknown", "count": 1440, "summed_gpu_ms": 31.792113, "mean_us": 22.07785625, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "rotary_kernel", "family_name_hint": "unknown", "count": 2922, "summed_gpu_ms": 28.249024, "mean_us": 9.667701574264203, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "nvjet_sm90_tst_256x144_64x4_2x1_v_bz_coopA_bias_TNT", "family_name_hint": "matrix_multiply", "count": 148, "summed_gpu_ms": 27.029812, "mean_us": 182.63386486486485, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "triton_poi_fused_clone_transpose_view_8", "family_name_hint": "unknown", "count": 2917, "summed_gpu_ms": 26.988606, "mean_us": 9.252178950977031, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "triton_poi_fused_all_to_all_single_clone_split_with_sizes_stack_transpose_view_4", "family_name_hint": "unknown", "count": 1461, "summed_gpu_ms": 26.772754, "mean_us": 18.32495140314853, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "sm80_xmma_fprop_implicit_gemm_indexed_wo_smem_tf32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x16x64_stage1_warpsize4x1x1_g1_tensor16x8x8_alignc4_execute_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 48, "summed_gpu_ms": 26.103825, "mean_us": 543.8296875, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_on_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 96, "summed_gpu_ms": 25.14021, "mean_us": 261.8771875, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "triton_poi_fused__to_copy_add_mul_split_squeeze_view_14", "family_name_hint": "unknown", "count": 1458, "summed_gpu_ms": 22.54566, "mean_us": 15.463415637860082, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::::CatArrayBatchedCopy_alignedK_contig::OpaqueType<(unsigned int)4>, unsigned int, (int)4, (int)128, (int)1, (int)16>(T1 *, at::native::::CatArrInputTensorMetadata, at::native::::TensorSizeStride, int, T2)", "family_name_hint": "unknown", "count": 152, "summed_gpu_ms": 22.227653, "mean_us": 146.23455921052633, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "triton_poi_fused_all_to_all_single_clone_transpose_unsqueeze_view_5", "family_name_hint": "unknown", "count": 2917, "summed_gpu_ms": 19.443414, "mean_us": 6.665551594103531, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::CUDAFunctor_add, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 675, "summed_gpu_ms": 18.595479, "mean_us": 27.54885777777778, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::::upsample_nearest2d_out_frame(const T1 *, T1 *, unsigned long, unsigned long, unsigned long, unsigned long, unsigned long, float, float)", "family_name_hint": "unknown", "count": 144, "summed_gpu_ms": 17.900723, "mean_us": 124.31057638888889, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "triton_poi_fused_silu_view_6", "family_name_hint": "unknown", "count": 1459, "summed_gpu_ms": 16.748419, "mean_us": 11.479382453735436, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "triton_poi_fused_add_view_7", "family_name_hint": "unknown", "count": 1459, "summed_gpu_ms": 16.376758, "mean_us": 11.224645647703907, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "triton_poi_fused__to_copy_all_to_all_single_clone_div_mul_sqrt_transpose_view_11", "family_name_hint": "unknown", "count": 1458, "summed_gpu_ms": 15.84636, "mean_us": 10.868559670781893, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_3", "family_name_hint": "unknown", "count": 1461, "summed_gpu_ms": 15.504394, "mean_us": 10.612179329226558, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_2", "family_name_hint": "unknown", "count": 1461, "summed_gpu_ms": 13.972858, "mean_us": 9.56390006844627, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 198, "summed_gpu_ms": 13.577673, "mean_us": 68.57410606060607, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "ncclDevKernel_AllGather_RING_LL(ncclDevKernelArgsStorage<(unsigned long)4096>)", "family_name_hint": "communication", "count": 130, "summed_gpu_ms": 12.832044, "mean_us": 98.70803076923077, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "N/A", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "ncclDevKernel_AllReduce_Sum_f32_RING_LL(ncclDevKernelArgsStorage<(unsigned long)4096>)", "family_name_hint": "communication", "count": 32, "summed_gpu_ms": 10.231326, "mean_us": 319.7289375, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "N/A", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "triton_red_fused__to_copy_pow_sum_view_10", "family_name_hint": "unknown", "count": 1458, "summed_gpu_ms": 7.463236, "mean_us": 5.11881755829904, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, const T1 *, T2 *)", "family_name_hint": "unknown", "count": 192, "summed_gpu_ms": 7.383703, "mean_us": 38.45678645833333, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void flash::prepare_varlen_num_blocks_kernel<(int)1, (bool)0>(int, int, int, const int *, const int *, const int *, const int *, const int *, const int *, int, int, int, int, int, cutlass::FastDivmod, cutlass::FastDivmod, int *, int *, int *, int *, int *, bool, bool, bool, int)", "family_name_hint": "attention", "count": 1459, "summed_gpu_ms": 3.277581, "mean_us": 2.246457162440027, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": "raw/ncu_attention_fa3_pilot_v1" }, { "source": "rebased", "kernel_name": "nvjet_sm90_tst_64x8_64x16_2x1_v_bz_bias_TNT", "family_name_hint": "matrix_multiply", "count": 37, "summed_gpu_ms": 2.882888, "mean_us": 77.91589189189189, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::unrolled_elementwise_kernel, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)", "family_name_hint": "elementwise", "count": 97, "summed_gpu_ms": 2.733191, "mean_us": 28.177226804123713, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::::vectorized_layer_norm_kernel(int, T2, const T1 *, const T1 *, const T1 *, T2 *, T2 *, T1 *)", "family_name_hint": "unknown", "count": 34, "summed_gpu_ms": 2.543528, "mean_us": 74.80964705882353, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::::launch_clamp_scalar(at::TensorIteratorBase &, c10::Scalar, c10::Scalar, at::native::detail::ClampLimits)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 1448, "summed_gpu_ms": 2.359108, "mean_us": 1.6292182320441988, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::unrolled_elementwise_kernel, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)", "family_name_hint": "elementwise", "count": 1459, "summed_gpu_ms": 2.223804, "mean_us": 1.5241973954763537, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::::CatArrayBatchedCopy_alignedK_contig::OpaqueType<(unsigned int)4>, unsigned int, (int)1, (int)128, (int)1, (int)16>(T1 *, at::native::::CatArrInputTensorMetadata, at::native::::TensorSizeStride, int, T2)", "family_name_hint": "unknown", "count": 1459, "summed_gpu_ms": 2.143306, "mean_us": 1.4690239890335848, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void cutlass::Kernel2(T1::Params)", "family_name_hint": "matrix_multiply", "count": 48, "summed_gpu_ms": 1.966691, "mean_us": 40.97272916666667, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 1459, "summed_gpu_ms": 1.581648, "mean_us": 1.0840630568882796, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 153, "summed_gpu_ms": 1.412804, "mean_us": 9.234013071895424, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::CUDAFunctor_add, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 37, "summed_gpu_ms": 1.214982, "mean_us": 32.83735135135135, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 6)]::operator ()() const::[lambda(c10::BFloat16) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 111, "summed_gpu_ms": 1.096515, "mean_us": 9.878513513513512, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void cutlass::Kernel2(T1::Params)", "family_name_hint": "matrix_multiply", "count": 64, "summed_gpu_ms": 1.004931, "mean_us": 15.702046875, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void cutlass::Kernel2(T1::Params)", "family_name_hint": "matrix_multiply", "count": 48, "summed_gpu_ms": 0.8929, "mean_us": 18.602083333333333, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "nvjet_sm90_tst_64x8_64x16_4x1_v_bz_bias_TNT", "family_name_hint": "matrix_multiply", "count": 74, "summed_gpu_ms": 0.892227, "mean_us": 12.057121621621622, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "sm80_xmma_fprop_implicit_gemm_tf32f32_tf32f32_f32_nhwckrsc_nchw_tilesize128x64x32_stage5_warpsize2x2x1_g1_tensor16x8x8_execute_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 48, "summed_gpu_ms": 0.772453, "mean_us": 16.092770833333333, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::::CatArrayBatchedCopy::OpaqueType<(unsigned int)4>, unsigned int, (int)4, (int)64, (int)64>(T1 *, at::native::::CatArrInputTensorMetadata, at::native::::TensorSizeStride, int, T2)", "family_name_hint": "unknown", "count": 82, "summed_gpu_ms": 0.715779, "mean_us": 8.729012195121951, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "sm80_xmma_gemm_f32f32_f32f32_f32_tn_n_tilesize128x32x8_stage3_warpsize2x2x1_ffma_aligna4_alignc4_execute_kernel__5x_cublas", "family_name_hint": "matrix_multiply", "count": 8, "summed_gpu_ms": 0.593568, "mean_us": 74.196, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "nvjet_sm90_tst_64x48_64x15_1x4_h_bz_bias_TNT", "family_name_hint": "matrix_multiply", "count": 34, "summed_gpu_ms": 0.559555, "mean_us": 16.4575, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::reduce_kernel<(int)512, (int)1, at::native::ReduceOp, unsigned int, float, (int)4, (int)4>>(T3)", "family_name_hint": "unknown", "count": 24, "summed_gpu_ms": 0.392001, "mean_us": 16.333375, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void cask_plugin__5x_cudnn::xmma__5x_cudnn::init_device_workspace_kernel>(T1, bool)", "family_name_hint": "matrix_multiply", "count": 96, "summed_gpu_ms": 0.261824, "mean_us": 2.7273333333333336, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 34, "summed_gpu_ms": 0.178562, "mean_us": 5.251823529411765, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::unrolled_elementwise_kernel, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.170688, "mean_us": 42.672, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::::distribution_elementwise_grid_stride_kernel(const at::TensorBase &, double, double, T1)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, T3, T4)::[lambda(curandStatePhilox4_32_10 *) (instance 2)], void at::native::::distribution_nullary_kernel(const at::TensorBase &, double, double, T1)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, T3, T4)::[lambda(curandStatePhilox4_32_10 *) (instance 2)], void at::native::templates::cuda::normal_kernel(const at::TensorBase &, double, double, T1)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, T4, const T5 &, T6)::[lambda(int, float) (instance 1)]>(long, at::PhiloxCudaState, T3, T4)", "family_name_hint": "elementwise", "count": 30, "summed_gpu_ms": 0.159744, "mean_us": 5.3248, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl::pow_tensor_tensor_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 6)]::operator ()() const::[lambda(double, double) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 37, "summed_gpu_ms": 0.125664, "mean_us": 3.396324324324324, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::round_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.114913, "mean_us": 28.72825, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AbsFunctor, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.104577, "mean_us": 26.14425, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)2, at::native::sin_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 1)]::operator ()() const::[lambda(double) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 37, "summed_gpu_ms": 0.097824, "mean_us": 2.643891891891892, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::unrolled_elementwise_kernel, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)", "family_name_hint": "elementwise", "count": 37, "summed_gpu_ms": 0.097249, "mean_us": 2.6283513513513515, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)2, at::native::cos_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 1)]::operator ()() const::[lambda(double) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 37, "summed_gpu_ms": 0.095712, "mean_us": 2.5868108108108108, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 34, "summed_gpu_ms": 0.080608, "mean_us": 2.3708235294117648, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::AUnaryFunctor::CompareEqFunctor>, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.074593, "mean_us": 18.64825, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::BinaryFunctor::CompareEqFunctor>, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.073632, "mean_us": 18.408, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 37, "summed_gpu_ms": 0.059904, "mean_us": 1.619027027027027, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 37, "summed_gpu_ms": 0.056768, "mean_us": 1.5342702702702702, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::BinaryFunctor>, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.05216, "mean_us": 13.04, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::reduce_kernel<(int)512, (int)1, at::native::ReduceOp, unsigned int, bool, (int)4, (int)4>>(T3)", "family_name_hint": "unknown", "count": 4, "summed_gpu_ms": 0.04944, "mean_us": 12.36, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::::CatArrayBatchedCopy_vectorized::OpaqueType<(unsigned int)8>, unsigned int, (int)2, (int)128, (int)1, (int)16, (int)2>(char *, at::native::::CatArrInputTensorMetadata, at::native::::TensorSizeStride, int, T2)", "family_name_hint": "unknown", "count": 37, "summed_gpu_ms": 0.04192, "mean_us": 1.132972972972973, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "sm80_xmma_gemm_f32f32_f32f32_f32_nn_n_tilesize32x32x8_stage3_warpsize1x2x1_ffma_aligna4_alignc4_execute_kernel__5x_cublas", "family_name_hint": "matrix_multiply", "count": 16, "summed_gpu_ms": 0.039456, "mean_us": 2.466, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)2, at::native::neg_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 6)]::operator ()() const::[lambda(double) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 37, "summed_gpu_ms": 0.037024, "mean_us": 1.0006486486486486, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)2, at::native::BUnaryFunctor>, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 37, "summed_gpu_ms": 0.036768, "mean_us": 0.9937297297297297, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)2, at::native::FillFunctor, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 37, "summed_gpu_ms": 0.031008, "mean_us": 0.8380540540540541, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void ::elementwise_kernel_with_index(T1, T2, function_traits::result_type *)", "family_name_hint": "elementwise", "count": 37, "summed_gpu_ms": 0.026752, "mean_us": 0.7230270270270269, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel>(at::TensorIteratorBase &, c10::ArrayRef, c10::ArrayRef)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef, c10::ArrayRef, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "family_name_hint": "elementwise", "count": 16, "summed_gpu_ms": 0.023648, "mean_us": 1.478, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel>(at::TensorIterator &, c10::ArrayRef, c10::ArrayRef)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef, c10::ArrayRef, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "family_name_hint": "elementwise", "count": 16, "summed_gpu_ms": 0.0232, "mean_us": 1.45, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 24, "summed_gpu_ms": 0.019424, "mean_us": 0.8093333333333333, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::reduce_kernel<(int)512, (int)1, at::native::ReduceOp, unsigned int, bool, (int)4, (int)4>>(T3)", "family_name_hint": "unknown", "count": 8, "summed_gpu_ms": 0.017632, "mean_us": 2.204, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::reduce_kernel<(int)512, (int)1, at::native::ReduceOp>, unsigned int, float, (int)4, (int)4>>(T3)", "family_name_hint": "unknown", "count": 8, "summed_gpu_ms": 0.016, "mean_us": 2.0, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::neg_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 16, "summed_gpu_ms": 0.015744, "mean_us": 0.984, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void gemvNSP_kernel, cublasGemvTensorStridedBatched, cublasGemvTensorStridedBatched, float>>(T10)", "family_name_hint": "unknown", "count": 8, "summed_gpu_ms": 0.01392, "mean_us": 1.74, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void ::elementwise_kernel_with_index(T1, T2, function_traits::result_type *)", "family_name_hint": "elementwise", "count": 16, "summed_gpu_ms": 0.012416, "mean_us": 0.776, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "std::enable_if::type internal::gemvx::kernel, cublasGemvTensorStridedBatched, cublasGemvTensorStridedBatched, float>>(T13)", "family_name_hint": "unknown", "count": 8, "summed_gpu_ms": 0.012321, "mean_us": 1.540125, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl_nocast::CompareEqFunctor>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 8, "summed_gpu_ms": 0.008576, "mean_us": 1.072, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "rebased", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, void at::native::compare_scalar_kernel(at::TensorIteratorBase &, at::native::::OpType, T1)::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 8, "summed_gpu_ms": 0.008032, "mean_us": 1.004, "status": "unresolved", "main_bottleneck": null, "missing": [ "clock-anchor and complete DiT tail validation for assigned steady window", "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Validate recaptured trace coverage first, then join actual operator context and collect counters", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_rebased_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "nvjet_sm90_tst_128x200_64x5_2x1_v_bz_coopA_bias_TNT", "family_name_hint": "matrix_multiply", "count": 28800, "summed_gpu_ms": 3432.218463, "mean_us": 119.17425218749999, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void cutlass::device_kernel, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple, cute::C<(int)160>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)1, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)1, (bool)0, (bool)0, cutlass::bfloat16_t, (int)1>, flash::CollectiveEpilogueFwd, cute::C<(int)128>, cute::C<(int)160>>, cute::tuple, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)1, (bool)1, (bool)0, (bool)0, (int)1>, flash::VarlenDynamicPersistentTileScheduler<(int)128, (int)160, (int)256, (int)128, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)1>>>>(T1::Params)", "family_name_hint": "attention", "count": 3200, "summed_gpu_ms": 3149.761347, "mean_us": 984.3004209375, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": "raw/ncu_attention_fa3_pilot_v1" }, { "source": "worker_flush_v2", "kernel_name": "ncclDevKernel_SendRecv(ncclDevKernelArgsStorage<(unsigned long)4096>)", "family_name_hint": "communication", "count": 14384, "summed_gpu_ms": 1232.356857, "mean_us": 85.6755323275862, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "N/A", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "nvjet_sm90_tst_256x152_64x4_1x2_h_bz_coopA_bias_TNT", "family_name_hint": "matrix_multiply", "count": 3200, "summed_gpu_ms": 853.086899, "mean_us": 266.5896559375, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 13972, "summed_gpu_ms": 384.421003, "mean_us": 27.51367041225308, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize256x32x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 288, "summed_gpu_ms": 383.406434, "mean_us": 1331.2723402777779, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 672, "summed_gpu_ms": 337.359273, "mean_us": 502.0227276785714, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_gather_kernel<(int)16, long>(char *, char *, T2 *, int, long, long, long, long, bool)", "family_name_hint": "unknown", "count": 6400, "summed_gpu_ms": 247.727709, "mean_us": 38.70745453125, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel>(at::TensorIterator &, c10::ArrayRef, c10::ArrayRef)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef, c10::ArrayRef, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "family_name_hint": "elementwise", "count": 6400, "summed_gpu_ms": 191.355589, "mean_us": 29.899310781249998, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, const T1 *, T2 *)", "family_name_hint": "unknown", "count": 3168, "summed_gpu_ms": 101.056065, "mean_us": 31.899010416666666, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void cutlass::device_kernel, cute::C<(int)1>, cute::C<(int)1>>, cute::tuple, cute::C<(int)176>, cute::C<(int)128>>, (int)128, cutlass::bfloat16_t, float, cutlass::arch::Sm90, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)0, (bool)1, (bool)1, (bool)0, (bool)0, (bool)0>, flash::CollectiveEpilogueFwd, cute::C<(int)128>, cute::C<(int)176>>, cute::tuple, cute::C<(int)1>, cute::C<(int)1>>, cutlass::bfloat16_t, cutlass::arch::Sm90, (int)256, (bool)0, (bool)0, (bool)0, (bool)0>, flash::StaticPersistentTileScheduler<(bool)0>>>>(T1::Params)", "family_name_hint": "attention", "count": 3200, "summed_gpu_ms": 93.039508, "mean_us": 29.07484625, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": "raw/ncu_attention_fa3_pilot_v1" }, { "source": "worker_flush_v2", "kernel_name": "fmha_cutlassF_f32_aligned_32x128_gmem_sm80(PyTorchMemEffAttention::AttentionKernel::Params)", "family_name_hint": "attention", "count": 48, "summed_gpu_ms": 92.675485, "mean_us": 1930.7392708333332, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": "raw/ncu_attention_fa3_pilot_v1" }, { "source": "worker_flush_v2", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 2064, "summed_gpu_ms": 85.579279, "mean_us": 41.462828972868216, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_9", "family_name_hint": "unknown", "count": 3200, "summed_gpu_ms": 82.64547, "mean_us": 25.826709375, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_view_12", "family_name_hint": "unknown", "count": 3200, "summed_gpu_ms": 73.792145, "mean_us": 23.060045312499998, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "triton_red_fused__to_copy_add_mul_native_layer_norm_split_squeeze_1", "family_name_hint": "unknown", "count": 3200, "summed_gpu_ms": 71.370039, "mean_us": 22.3031371875, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "triton_poi_fused_gelu_view_13", "family_name_hint": "unknown", "count": 3200, "summed_gpu_ms": 71.156884, "mean_us": 22.23652625, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 1536, "summed_gpu_ms": 68.972207, "mean_us": 44.903780598958335, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 1520, "summed_gpu_ms": 64.480655, "mean_us": 42.42148355263158, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "rotary_kernel", "family_name_hint": "unknown", "count": 6400, "summed_gpu_ms": 62.212757, "mean_us": 9.72074328125, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "triton_poi_fused_all_to_all_single_clone_split_with_sizes_stack_transpose_view_4", "family_name_hint": "unknown", "count": 3200, "summed_gpu_ms": 58.729221, "mean_us": 18.3528815625, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "nvjet_sm90_tst_256x144_64x4_2x1_v_bz_coopA_bias_TNT", "family_name_hint": "matrix_multiply", "count": 320, "summed_gpu_ms": 58.543511, "mean_us": 182.948471875, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "triton_poi_fused__to_copy_add_mul_split_squeeze_view_14", "family_name_hint": "unknown", "count": 3200, "summed_gpu_ms": 49.037935, "mean_us": 15.3243546875, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void cudnn::engines_precompiled::nhwcToNchwKernel(cudnn::engines_precompiled::nhwc2nchw_params_t, const T1 *, T2 *)", "family_name_hint": "unknown", "count": 1632, "summed_gpu_ms": 47.074274, "mean_us": 28.844530637254902, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize64x64x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 480, "summed_gpu_ms": 45.933336, "mean_us": 95.69445, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "triton_poi_fused_all_to_all_single_clone_transpose_unsqueeze_view_5", "family_name_hint": "unknown", "count": 6400, "summed_gpu_ms": 42.839876, "mean_us": 6.693730625, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "triton_poi_fused_clone_transpose_view_8", "family_name_hint": "unknown", "count": 6400, "summed_gpu_ms": 39.675898, "mean_us": 6.1993590625, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 1392, "summed_gpu_ms": 38.873203, "mean_us": 27.92615158045977, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor>, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 1668, "summed_gpu_ms": 38.071534, "mean_us": 22.82466067146283, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::CUDAFunctorOnSelf_add, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 1476, "summed_gpu_ms": 37.802149, "mean_us": 25.6112120596206, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "triton_poi_fused_silu_view_6", "family_name_hint": "unknown", "count": 3200, "summed_gpu_ms": 36.534415, "mean_us": 11.4170046875, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "triton_poi_fused__to_copy_all_to_all_single_clone_div_mul_sqrt_transpose_view_11", "family_name_hint": "unknown", "count": 3200, "summed_gpu_ms": 35.005508, "mean_us": 10.939221250000001, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_3", "family_name_hint": "unknown", "count": 3200, "summed_gpu_ms": 34.654719, "mean_us": 10.8295996875, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 2552, "summed_gpu_ms": 34.057951, "mean_us": 13.345592084639499, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "triton_poi_fused_add_view_7", "family_name_hint": "unknown", "count": 3200, "summed_gpu_ms": 33.243808, "mean_us": 10.38869, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize64x192x32_warpgroupsize1x1x1_g1_execute_segment_k_off_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 48, "summed_gpu_ms": 32.145728, "mean_us": 669.7026666666666, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::reduce_kernel<(int)128, (int)4, at::native::ReduceOp, unsigned int, float, (int)4, (int)4>>(T3)", "family_name_hint": "unknown", "count": 1440, "summed_gpu_ms": 31.863354, "mean_us": 22.127329166666666, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "triton_red_fused__to_copy_div_mul_pow_split_with_sizes_sqrt_sum_view_2", "family_name_hint": "unknown", "count": 3200, "summed_gpu_ms": 30.674257, "mean_us": 9.5857053125, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl_nocast(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 432, "summed_gpu_ms": 29.435922, "mean_us": 68.13870833333333, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "sm80_xmma_fprop_implicit_gemm_indexed_wo_smem_tf32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x16x64_stage1_warpsize4x1x1_g1_tensor16x8x8_alignc4_execute_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 48, "summed_gpu_ms": 26.140915, "mean_us": 544.6023958333334, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "sm90_xmma_fprop_implicit_gemm_f32f32_tf32f32_f32_nhwckrsc_nhwc_tilesize128x128x32_warpgroupsize1x1x1_g1_execute_segment_k_on_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 96, "summed_gpu_ms": 24.845262, "mean_us": 258.8048125, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "ncclDevKernel_AllGather_RING_LL(ncclDevKernelArgsStorage<(unsigned long)4096>)", "family_name_hint": "communication", "count": 176, "summed_gpu_ms": 23.648364, "mean_us": 134.36570454545455, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "N/A", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::::CatArrayBatchedCopy_alignedK_contig::OpaqueType<(unsigned int)4>, unsigned int, (int)4, (int)128, (int)1, (int)16>(T1 *, at::native::::CatArrInputTensorMetadata, at::native::::TensorSizeStride, int, T2)", "family_name_hint": "unknown", "count": 160, "summed_gpu_ms": 22.550569, "mean_us": 140.94105625, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::CUDAFunctor_add, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 736, "summed_gpu_ms": 18.67982, "mean_us": 25.380190217391306, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::::upsample_nearest2d_out_frame(const T1 *, T1 *, unsigned long, unsigned long, unsigned long, unsigned long, unsigned long, float, float)", "family_name_hint": "unknown", "count": 144, "summed_gpu_ms": 17.873421, "mean_us": 124.12097916666667, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "triton_red_fused__to_copy_pow_sum_view_10", "family_name_hint": "unknown", "count": 3200, "summed_gpu_ms": 16.402317, "mean_us": 5.125724062500001, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "ncclDevKernel_AllReduce_Sum_f32_RING_LL(ncclDevKernelArgsStorage<(unsigned long)4096>)", "family_name_hint": "communication", "count": 32, "summed_gpu_ms": 11.141755, "mean_us": 348.17984375, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "N/A", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void cudnn::engines_precompiled::nchwToNhwcKernel(cudnn::engines_precompiled::nchw2nhwc_params_t, const T1 *, T2 *)", "family_name_hint": "unknown", "count": 192, "summed_gpu_ms": 7.361136, "mean_us": 38.33925, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void flash::prepare_varlen_num_blocks_kernel<(int)1, (bool)0>(int, int, int, const int *, const int *, const int *, const int *, const int *, const int *, int, int, int, int, int, cutlass::FastDivmod, cutlass::FastDivmod, int *, int *, int *, int *, int *, bool, bool, bool, int)", "family_name_hint": "attention", "count": 3200, "summed_gpu_ms": 7.162384, "mean_us": 2.238245, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": "raw/ncu_attention_fa3_pilot_v1" }, { "source": "worker_flush_v2", "kernel_name": "void at::native::unrolled_elementwise_kernel, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)", "family_name_hint": "elementwise", "count": 224, "summed_gpu_ms": 6.406732, "mean_us": 28.60148214285714, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "nvjet_sm90_tst_64x8_64x16_2x1_v_bz_bias_TNT", "family_name_hint": "matrix_multiply", "count": 80, "summed_gpu_ms": 6.190926, "mean_us": 77.386575, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::::vectorized_layer_norm_kernel(int, T2, const T1 *, const T1 *, const T1 *, T2 *, T2 *, T1 *)", "family_name_hint": "unknown", "count": 80, "summed_gpu_ms": 5.992427, "mean_us": 74.90533749999999, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::unrolled_elementwise_kernel, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)", "family_name_hint": "elementwise", "count": 3200, "summed_gpu_ms": 4.924317, "mean_us": 1.5388490625, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::::CatArrayBatchedCopy_alignedK_contig::OpaqueType<(unsigned int)4>, unsigned int, (int)1, (int)128, (int)1, (int)16>(T1 *, at::native::::CatArrInputTensorMetadata, at::native::::TensorSizeStride, int, T2)", "family_name_hint": "unknown", "count": 3200, "summed_gpu_ms": 4.618377, "mean_us": 1.4432428124999999, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 3200, "summed_gpu_ms": 3.460525, "mean_us": 1.0814140625, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 336, "summed_gpu_ms": 3.268553, "mean_us": 9.72783630952381, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::CUDAFunctor_add, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 80, "summed_gpu_ms": 2.617604, "mean_us": 32.72005, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::::silu_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 6)]::operator ()() const::[lambda(c10::BFloat16) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 240, "summed_gpu_ms": 2.375045, "mean_us": 9.896020833333335, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::::launch_clamp_scalar(at::TensorIteratorBase &, c10::Scalar, c10::Scalar, at::native::detail::ClampLimits)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 1448, "summed_gpu_ms": 2.338918, "mean_us": 1.6152748618784531, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void cutlass::Kernel2(T1::Params)", "family_name_hint": "matrix_multiply", "count": 48, "summed_gpu_ms": 1.949635, "mean_us": 40.61739583333333, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "nvjet_sm90_tst_64x8_64x16_4x1_v_bz_bias_TNT", "family_name_hint": "matrix_multiply", "count": 160, "summed_gpu_ms": 1.928162, "mean_us": 12.0510125, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::::CatArrayBatchedCopy::OpaqueType<(unsigned int)4>, unsigned int, (int)4, (int)64, (int)64>(T1 *, at::native::::CatArrInputTensorMetadata, at::native::::TensorSizeStride, int, T2)", "family_name_hint": "unknown", "count": 176, "summed_gpu_ms": 1.493572, "mean_us": 8.486204545454546, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "nvjet_sm90_tst_64x48_64x15_1x4_h_bz_bias_TNT", "family_name_hint": "matrix_multiply", "count": 80, "summed_gpu_ms": 1.332161, "mean_us": 16.6520125, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "sm80_xmma_gemm_f32f32_f32f32_f32_tn_n_tilesize128x32x8_stage3_warpsize2x2x1_ffma_aligna4_alignc4_execute_kernel__5x_cublas", "family_name_hint": "matrix_multiply", "count": 16, "summed_gpu_ms": 1.186051, "mean_us": 74.1281875, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void cutlass::Kernel2(T1::Params)", "family_name_hint": "matrix_multiply", "count": 64, "summed_gpu_ms": 1.000675, "mean_us": 15.635546875, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void cutlass::Kernel2(T1::Params)", "family_name_hint": "matrix_multiply", "count": 48, "summed_gpu_ms": 0.888258, "mean_us": 18.505375, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::reduce_kernel<(int)512, (int)1, at::native::ReduceOp, unsigned int, float, (int)4, (int)4>>(T3)", "family_name_hint": "unknown", "count": 48, "summed_gpu_ms": 0.784321, "mean_us": 16.340020833333334, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "sm80_xmma_fprop_implicit_gemm_tf32f32_tf32f32_f32_nhwckrsc_nchw_tilesize128x64x32_stage5_warpsize2x2x1_g1_tensor16x8x8_execute_kernel__5x_cudnn", "family_name_hint": "matrix_multiply", "count": 48, "summed_gpu_ms": 0.767361, "mean_us": 15.9866875, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 80, "summed_gpu_ms": 0.417696, "mean_us": 5.2212, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::::distribution_elementwise_grid_stride_kernel(const at::TensorBase &, double, double, T1)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, T3, T4)::[lambda(curandStatePhilox4_32_10 *) (instance 2)], void at::native::::distribution_nullary_kernel(const at::TensorBase &, double, double, T1)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, T3, T4)::[lambda(curandStatePhilox4_32_10 *) (instance 2)], void at::native::templates::cuda::normal_kernel(const at::TensorBase &, double, double, T1)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, T4, const T5 &, T6)::[lambda(int, float) (instance 1)]>(long, at::PhiloxCudaState, T3, T4)", "family_name_hint": "elementwise", "count": 64, "summed_gpu_ms": 0.344898, "mean_us": 5.38903125, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl::pow_tensor_tensor_kernel(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 6)]::operator ()() const::[lambda(double, double) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 80, "summed_gpu_ms": 0.277728, "mean_us": 3.4716, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void cask_plugin__5x_cudnn::xmma__5x_cudnn::init_device_workspace_kernel>(T1, bool)", "family_name_hint": "matrix_multiply", "count": 96, "summed_gpu_ms": 0.258016, "mean_us": 2.6876666666666664, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)2, at::native::cos_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 1)]::operator ()() const::[lambda(double) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 80, "summed_gpu_ms": 0.212864, "mean_us": 2.6608, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)2, at::native::sin_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 1)]::operator ()() const::[lambda(double) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 80, "summed_gpu_ms": 0.211169, "mean_us": 2.6396125, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::unrolled_elementwise_kernel, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)", "family_name_hint": "elementwise", "count": 80, "summed_gpu_ms": 0.208641, "mean_us": 2.6080124999999996, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 80, "summed_gpu_ms": 0.192576, "mean_us": 2.4072, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::unrolled_elementwise_kernel, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.169312, "mean_us": 42.328, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 80, "summed_gpu_ms": 0.132832, "mean_us": 1.6604, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 80, "summed_gpu_ms": 0.121376, "mean_us": 1.5172, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::round_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.115649, "mean_us": 28.91225, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::AbsFunctor, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.10528, "mean_us": 26.32, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::::CatArrayBatchedCopy_vectorized::OpaqueType<(unsigned int)8>, unsigned int, (int)2, (int)128, (int)1, (int)16, (int)2>(char *, at::native::::CatArrInputTensorMetadata, at::native::::TensorSizeStride, int, T2)", "family_name_hint": "unknown", "count": 80, "summed_gpu_ms": 0.09328, "mean_us": 1.166, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)2, at::native::neg_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 6)]::operator ()() const::[lambda(double) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 80, "summed_gpu_ms": 0.080193, "mean_us": 1.0024125, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)2, at::native::BUnaryFunctor>, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 80, "summed_gpu_ms": 0.080032, "mean_us": 1.0004, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "sm80_xmma_gemm_f32f32_f32f32_f32_nn_n_tilesize32x32x8_stage3_warpsize1x2x1_ffma_aligna4_alignc4_execute_kernel__5x_cublas", "family_name_hint": "matrix_multiply", "count": 32, "summed_gpu_ms": 0.079456, "mean_us": 2.483, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::AUnaryFunctor::CompareEqFunctor>, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.076032, "mean_us": 19.008, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::BinaryFunctor::CompareEqFunctor>, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.07392, "mean_us": 18.48, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)2, at::native::FillFunctor, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 80, "summed_gpu_ms": 0.065856, "mean_us": 0.8232, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void ::elementwise_kernel_with_index(T1, T2, function_traits::result_type *)", "family_name_hint": "elementwise", "count": 80, "summed_gpu_ms": 0.060384, "mean_us": 0.7547999999999999, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, at::native::BinaryFunctor>, std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 4, "summed_gpu_ms": 0.052416, "mean_us": 13.104, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel>(at::TensorIteratorBase &, c10::ArrayRef, c10::ArrayRef)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef, c10::ArrayRef, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "family_name_hint": "elementwise", "count": 32, "summed_gpu_ms": 0.049216, "mean_us": 1.538, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::reduce_kernel<(int)512, (int)1, at::native::ReduceOp, unsigned int, bool, (int)4, (int)4>>(T3)", "family_name_hint": "unknown", "count": 4, "summed_gpu_ms": 0.048865, "mean_us": 12.21625, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel>(at::TensorIterator &, c10::ArrayRef, c10::ArrayRef)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef, c10::ArrayRef, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)", "family_name_hint": "elementwise", "count": 32, "summed_gpu_ms": 0.04736, "mean_us": 1.48, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 48, "summed_gpu_ms": 0.038496, "mean_us": 0.802, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::reduce_kernel<(int)512, (int)1, at::native::ReduceOp, unsigned int, bool, (int)4, (int)4>>(T3)", "family_name_hint": "unknown", "count": 16, "summed_gpu_ms": 0.035232, "mean_us": 2.202, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::reduce_kernel<(int)512, (int)1, at::native::ReduceOp>, unsigned int, float, (int)4, (int)4>>(T3)", "family_name_hint": "unknown", "count": 16, "summed_gpu_ms": 0.03296, "mean_us": 2.06, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)4, at::native::neg_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 32, "summed_gpu_ms": 0.02928, "mean_us": 0.915, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void gemvNSP_kernel, cublasGemvTensorStridedBatched, cublasGemvTensorStridedBatched, float>>(T10)", "family_name_hint": "unknown", "count": 16, "summed_gpu_ms": 0.027456, "mean_us": 1.716, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "std::enable_if::type internal::gemvx::kernel, cublasGemvTensorStridedBatched, cublasGemvTensorStridedBatched, float>>(T13)", "family_name_hint": "unknown", "count": 16, "summed_gpu_ms": 0.024576, "mean_us": 1.536, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void ::elementwise_kernel_with_index(T1, T2, function_traits::result_type *)", "family_name_hint": "elementwise", "count": 32, "summed_gpu_ms": 0.024576, "mean_us": 0.768, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl_nocast::CompareEqFunctor>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)", "family_name_hint": "elementwise", "count": 16, "summed_gpu_ms": 0.017568, "mean_us": 1.098, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null }, { "source": "worker_flush_v2", "kernel_name": "void at::native::vectorized_elementwise_kernel<(int)8, void at::native::compare_scalar_kernel(at::TensorIteratorBase &, at::native::::OpType, T1)::[lambda(float) (instance 1)], std::array>(int, T2, T3)", "family_name_hint": "elementwise", "count": 16, "summed_gpu_ms": 0.015744, "mean_us": 0.984, "status": "unresolved", "main_bottleneck": null, "missing": [ "actual shape/dtype/layout and stage mapping for this group", "matching counter evidence in representative context", "counterfactual comparison" ], "mfu_applicability": "undetermined until operator/shape mapping", "alternative_explanations": [ "launch/dependency limits", "memory traffic/cache behavior", "compute or execution resource limits" ], "next_action": "Join actual operator context, profile representative group, then test one variable", "evidence": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernels_steady.csv", "related_but_not_group_matched_evidence": null } ] }