# t/rank0.1790655533959855643.pt.trace.json.gz
kernels 520591 busy_ms 8203.9 window_ms 8302.2
eager kernel busy_ms 3205.3
graph replays by node count: n_nodes count mean_busy_us mean_span_us
    2702   144    29882.4    30036.1
     165   154      686.1      697.8
     132   146     2338.2     2345.1
     101   150     1539.0     1543.9
     137     8     2215.4     2221.9
top kernels: total_ms count mean_us name
    1883.90   20736    90.85 void exl3_gemm_kernel<4, false, 2, 16, 32, 256, 4, 3>(__half const*, unsigned short const*, void*, int, int, i
    1785.28    3312   539.03 void exl3_i8_gemm_kernel<4, 2, 128, 128, 128, 1, 4, 2, 8, false, false, true, false, true, true, true, 128, 2,
    1376.85   34816    39.55 void exl3_gemm_kernel<4, false, 2, 32, 32, 128, 4, 3>(__half const*, unsigned short const*, void*, int, int, i
     465.12   14400    32.30 fused_sigmoid_gating_delta_rule_update_kernel
     315.20    7879    40.01 void exl3_gemm_kernel<6, false, 1, 32, 32, 128, 4, 3>(__half const*, unsigned short const*, void*, int, int, i
     232.64     276   842.90 void exl3_gemm_kernel<6, false, 2, 32, 32, 128, 4, 3>(__half const*, unsigned short const*, void*, int, int, i
     200.64    2432    82.50 (anonymous namespace)::ni8_prefill_kernel((anonymous namespace)::Ni8Params)
     110.95     576   192.61 void arbi_serve_exl3_a_prep::a_prep_kernel<5, 32, true, true, arbi_serve_exl3_a_prep::SrcSiluMul<__half>, true
     103.67   41459     2.50 void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{la
      87.39    5603    15.60 void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_
      77.73   11995     6.48 void at::native::reduce_kernel<512, 1, at::native::ReduceOp<float, at::native::MeanOps<float, float, float, fl
      69.83     834    83.73 kernel_cutlass_kernel_split_d_tkvkernelscudaprefillturbo_attn_cute_bypassTurboAttnCuteBypass_object_at__tensor
      68.79     720    95.54 void arbi_serve_exl3_a_prep::a_prep_kernel<2, 32, true, true, arbi_serve_exl3_a_prep::SrcAddRMSNorm<__nv_bfloa
      68.35    2048    33.38 void exl3_gemm_kernel<4, false, 2, 16, 32, 128, 4, 3>(__half const*, unsigned short const*, void*, int, int, i
      60.69     432   140.48 chunk_gated_delta_rule_fwd_kernel_h_emit
      48.77     432   112.89 chunk_fwd_kernel_o
      47.79     566    84.43 Memcpy DtoH (Device -> Pinned)
      42.71    1615    26.45 _topc_mask_kernel
      40.97   21964     1.87 void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor<float, float, float, at::native::b
      39.93   24907     1.60 void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast<at::native::BinaryFunctor<float
      39.81     537    74.14 void at::native::(anonymous namespace)::cunn_SoftMaxForward<4, float, float, float, at::native::(anonymous nam
      37.36     864    43.24 void arbi_serve_exl3_a_prep::a_prep_kernel<2, 32, true, true, arbi_serve_exl3_a_prep::SrcHalf, true, 128>(arbi
      32.99     432    76.37 _kern_tiled
      32.39     432    74.98 recompute_w_u_fwd_kernel
      32.38     868    37.31 void exl3_gemm_kernel<6, false, 1, 16, 32, 128, 4, 3>(__half const*, unsigned short const*, void*, int, int, i
      31.58      37   853.44 void exl3_gemm_kernel<6, false, 2, 16, 32, 128, 4, 3>(__half const*, unsigned short const*, void*, int, int, i
      27.66    1615    17.13 _topc_partials_kernel
      26.78   23040     1.16 void at::native::vectorized_elementwise_kernel<4, at::native::float16_copy_kernel_cuda(at::TensorIteratorBase&
      22.94       8  2866.97 ampere_fp16_s1688gemm_fp16_128x64_sliced1x2_ldg8_f2f_nn
      22.87     432    52.93 chunk_gated_delta_rule_fwd_kkt_solve_kernel
      22.05    2448     9.01 fused_compress_store_kernel(__nv_bfloat16 const*, __nv_bfloat16 const*, unsigned char*, long const*, float con
      21.55    2352     9.16 (anonymous namespace)::tq_prefill_split_combine_kernel(float const*, float const*, __nv_bfloat16*, float*, int
      20.28    6528     3.11 void cutlass::Kernel2<cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_32x32_64x1_tn_align8>(cutlass_80_wmma_ten
      19.95    2400     8.31 (anonymous namespace)::ni8_vmax_kernel(unsigned char const*, int const*, int const*, int const*, int const*, i
      19.85    7200     2.76 _conv_window_commit_kernel
      19.37   11520     1.68 void at::native::vectorized_elementwise_kernel<4, at::native::sigmoid_kernel_cuda(at::TensorIteratorBase&)::{l
      19.00    6912     2.75 _causal_conv1d_verify_kernel
      18.75     288    65.12 triton_per_fused__to_copy_add_mean_mul_pow_rsqrt_silu_view_18
      18.03   17540     1.03 memcpy32_post
      17.61    4512     3.90 void cutlass::Kernel2<cutlass_80_wmma_tensorop_bf16_s161616gemm_bf16_32x32_128x2_nn_align8>(cutlass_80_wmma_te
      17.30    8512     2.03 void at::native::index_elementwise_kernel<128, 4, at::native::index_copy_kernel_impl<at::native::OpaqueType<2>
      16.09     150   107.25 _residual_gumbel_partial_kernel
      15.70   12599     1.25 void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add<c10::BFloat16>, std::array<char*
      15.33    7008     2.19 triton_red_fused__to_copy_add_exl3_gemm_group2_mean_mul_pow_rsqrt_1
      15.05     160    94.09 void exl3_i8_gemm_kernel_mb<8, 23, 48, 128, 128, 1, 8, 2, 8, false, false, true, false, false, false, true, 12
      14.66    1078    13.60 _qwrite_draw_partials_kernel
      13.83   11995     1.15 void at::native::vectorized_elementwise_kernel<4, at::native::(anonymous namespace)::pow_tensor_scalar_kernel_
      13.58   13824     0.98 triton_poi_fused__to_copy_exl3_add_rmsnorm_gemm2_8
      13.02    3029     4.30 void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_
      12.16    6960     1.75 triton_poi_fused__to_copy_add_exl3_linear_gemm_mean_mul_pow_rsqrt_silu_view_7
ring-ish kernels:
     200.64    2432    82.50 (anonymous namespace)::ni8_prefill_kernel((anonymous namespace)::Ni8Params)
      69.83     834    83.73 kernel_cutlass_kernel_split_d_tkvkernelscudaprefillturbo_attn_cute_bypassTurboAttnCuteBypass_object_at__tensor
      22.05    2448     9.01 fused_compress_store_kernel(__nv_bfloat16 const*, __nv_bfloat16 const*, unsigned char*, long const*, float con
      21.55    2352     9.16 (anonymous namespace)::tq_prefill_split_combine_kernel(float const*, float const*, __nv_bfloat16*, float*, int
      11.95     192    62.26 triton_poi_fused__to_copy_cat_gdn_prefill_conv_flat_index_select_mul_slice_split_with_sizes_squeeze_view_12
       8.42    6011     1.40 void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor<float>, std::array<char*, 1ul> >(int
       7.71      96    80.35 triton_poi_fused__to_copy_cat_gdn_prefill_conv_flat_index_select_mul_slice_split_with_sizes_squeeze_view_16
       7.66     160    47.88 ampere_bf16_s1688gemm_bf16_128x128_ldg8_f2f_stages_32x1_nn
       7.02    6912     1.02 triton_poi_fused_add_arange_gdn_replay_x_conv_save_mul_view_3
       6.85      50   137.03 ampere_fp16_s1688gemm_fp16_128x128_ldg8_f2f_stages_32x1_nn
       6.84     144    47.53 triton_poi_fused__to_copy_gdn_prefill_conv_flat_index_select_mul_squeeze_view_2
       5.95    6912     0.86 triton_poi_fused__to_copy_arange_gdn_verify_replay_save_split_split_with_sizes_view_4
       5.60    2259     2.48 void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor<c10::BFloat16>, std::array<char*, 1u
       5.54    6912     0.80 triton_poi_fused__to_copy_arange_gdn_verify_replay_save_split_split_with_sizes_view_5
       3.94    2400     1.64 void at::native::_scatter_gather_elementwise_kernel<128, 8, at::native::_cuda_scatter_gather_internal_kernel<f
       3.40      48    70.76 triton_poi_fused__to_copy_gdn_prefill_conv_flat_index_select_mul_squeeze_view_1
       2.50    3108     0.80 void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor<int>, std::array<char*, 1ul> >(int, 
       2.37     432     5.48 chunk_local_cumsum_scalar_kernel
       1.78    1500     1.18 void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_nocast<at::native::FillFunctor<c10::BF
       1.52    1762     0.86 void at::native::vectorized_elementwise_kernel<2, at::native::FillFunctor<long>, std::array<char*, 1ul> >(int,
       1.26     730     1.72 triton_per_fused__to_copy_add_arange_cat_copy_cos_dflash_varlen_attend_mean_mul_neg_pow_rsqrt_sin_slice_unsque
       1.13    1033     1.09 void at::native::vectorized_elementwise_kernel<2, at::native::(anonymous namespace)::where_kernel_impl(at::Ten
       1.11     730     1.52 triton_per_fused__to_copy_add_arange_as_strided_cat_cos_dflash_place_block_write_mean_mul_neg_pow_rsqrt_sin_sl
       0.74     354     2.10 void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast<at::native::FillFunctor<float> 
       0.74     288     2.56 triton_poi_fused__to_copy_add_arange_cat_index_copy_mul_slice_split_with_sizes_view_8
       0.67     192     3.49 triton_poi_fused__to_copy_add_arange_cat_index_copy_mul_slice_split_with_sizes_view_4
       0.51     308     1.67 at::native::mbtopk::computeDigitCumSum(short*, unsigned int*, unsigned int)
       0.47      96     4.90 void at::native::index_elementwise_kernel<128, 4, at::native::index_fill_kernel_impl<at::native::OpaqueType<4>
       0.45     288     1.56 triton_poi_fused__to_copy_arange_copy_select_silu_slice_split_split_with_sizes_view_15
       0.41     288     1.42 triton_poi_fused__to_copy_cat_gdn_prefill_conv_flat_index_select_mul_slice_split_with_sizes_squeeze_view_2
       0.39     467     0.85 void (anonymous namespace)::elementwise_kernel_with_index<int, at::native::arange_cuda_out(c10::Scalar const&,
       0.39     192     2.05 triton_poi_fused__to_copy_arange_copy_select_silu_slice_split_split_with_sizes_view_17
       0.38     288     1.33 triton_poi_fused__to_copy_arange_copy_select_silu_slice_split_split_with_sizes_view_0
       0.33      20    16.36 ampere_fp16_s16816gemm_fp16_128x64_ldg8_f2f_stages_32x6_nn
       0.30     192     1.57 triton_poi_fused__to_copy_arange_copy_select_silu_slice_split_split_with_sizes_view_16
       0.28     288     0.97 triton_poi_fused__to_copy_arange_copy_select_silu_slice_split_split_with_sizes_view_1
       0.27     292     0.93 triton_poi_fused__to_copy_add_arange_cat_copy_cos_dflash_varlen_attend_mean_mul_pow_rsqrt_sin_unsqueeze_view_9
       0.27     154     1.75 triton_poi_fused_add_clamp_copy__index_select_remainder_searchsorted_sub_view_1
       0.23      32     7.25 ampere_bf16_s16816gemm_bf16_128x64_ldg8_f2f_stages_32x6_nn
       0.23     150     1.54 void at::native::_scatter_gather_elementwise_kernel<128, 8, at::native::_cuda_scatter_gather_internal_kernel<t
       0.23      96     2.37 triton_poi_fused__to_copy_add_arange_cat_index_copy_mul_slice_split_with_sizes_view_3
       0.21     150     1.42 void at::native::vectorized_elementwise_kernel<4, at::native::(anonymous namespace)::where_kernel_impl(at::Ten
       0.21      96     2.17 triton_poi_fused__to_copy_gdn_prefill_conv_flat_index_select_mul_squeeze_view_3
       0.20     146     1.35 triton_poi_fused__unsafe_view_add_clone_copy__expand_remainder_view_2
       0.20     149     1.32 void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl_nocast<at::native::FillFunctor<int> >(
       0.17     154     1.13 void at::native::vectorized_elementwise_kernel<2, at::native::(anonymous namespace)::masked_fill_kernel(at::Te
       0.16     146     1.13 triton_poi_fused_add_clamp_copy_copy__remainder_sub_view_8
       0.16     146     1.11 triton_per_fused_add_clamp_cumsum_0
       0.15      96     1.54 void at::native::index_elementwise_kernel<128, 4, at::native::index_fill_kernel_impl<at::native::OpaqueType<2>
       0.15     146     1.00 triton_poi_fused__to_copy_add_arange_cat_copy_copy__cos_dflash_varlen_attend_mean_mul_pow_rsqrt_sin_unsqueeze_
       0.14     146     0.93 triton_poi_fused_add_clamp_copy__remainder_sub_view_19
       0.13      96     1.35 triton_poi_fused__to_copy_arange_copy_select_silu_slice_split_split_with_sizes_view_13
       0.13      96     1.31 triton_poi_fused__to_copy_arange_copy_select_silu_slice_split_split_with_sizes_view_14
       0.11     154     0.73 void at::native::mbtopk::fill<unsigned int, unsigned int>(unsigned int*, unsigned int, unsigned int)
       0.06      40     1.58 triton_per_fused__to_copy_add_arange_cat_clamp_copy_cos_dflash_varlen_attend_mean_mul_neg_pow_remainder_rsqrt_
       0.06      40     1.46 triton_per_fused__to_copy_add_arange_as_strided_cat_cos_dflash_place_block_write_expand_mean_mul_neg_pow_remai
       0.04      40     0.88 triton_poi_fused__to_copy_add_as_strided_dflash_place_block_write_expand_remainder_view_6
       0.03      32     1.02 triton_poi_fused__to_copy_add_arange_cat_clamp_copy_cos_dflash_varlen_attend_mean_mul_pow_remainder_rsqrt_sin_
       0.02      16     1.10 triton_poi_fused__to_copy_add_arange_as_strided_cat_cos_dflash_place_block_write_expand_mean_mul_pow_remainder
       0.02      16     1.02 triton_poi_fused__to_copy_add_arange_cat_clamp_copy_copy__cos_dflash_varlen_attend_mean_mul_pow_remainder_rsqr
       0.01       8     1.30 triton_poi_fused_add_clamp_cumsum_0
       0.01      12     0.87 void at::native::unrolled_elementwise_kernel<at::native::FillFunctor<long>, std::array<char*, 1ul>, 4, Trivial
       0.01       8     1.11 triton_poi_fused__to_copy_add_arange_as_strided_cat_clamp_copy__cos_dflash_place_block_write_expand_mean_mul_p
       0.00       3     1.05 void at::native::vectorized_elementwise_kernel<2, at::native::FillFunctor<int>, std::array<char*, 1ul> >(int, 
       0.00       3     0.92 void at::native::unrolled_elementwise_kernel<at::native::FillFunctor<int>, std::array<char*, 1ul>, 4, TrivialO
