{
  "schema_version": 1,
  "method": "one warmed backward pass; CUDA device event durations from torch.profiler",
  "rows": [
    {
      "shape": {
        "batch": 4,
        "length": 2048,
        "width": 2048
      },
      "architecture": "rglru",
      "backend": "fused_grouped_prefix32_hierarchical_chunk16",
      "cuda_events": 35,
      "summed_cuda_duration_us": 1537.0,
      "categories": {
        "copy": {
          "events": 9,
          "duration_us": 486.0
        },
        "framework_elementwise": {
          "events": 6,
          "duration_us": 286.0
        },
        "framework_reduce": {
          "events": 4,
          "duration_us": 188.0
        },
        "gemm": {
          "events": 4,
          "duration_us": 181.0
        },
        "memset": {
          "events": 7,
          "duration_us": 1.0
        },
        "reverse_prefix": {
          "events": 3,
          "duration_us": 46.0
        },
        "reverse_replay": {
          "events": 1,
          "duration_us": 282.0
        },
        "reverse_summary": {
          "events": 1,
          "duration_us": 67.0
        }
      },
      "top_cuda_events": [
        {
          "name": "_reverse_chunk_replay",
          "events": 1,
          "duration_us": 282.0
        },
        {
          "name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)",
          "events": 3,
          "duration_us": 216.0
        },
        {
          "name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)",
          "events": 4,
          "duration_us": 149.0
        },
        {
          "name": "void at::native::reduce_kernel<128, 4, at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4> >(at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4>)",
          "events": 3,
          "duration_us": 138.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}>(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}>(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})",
          "events": 2,
          "duration_us": 121.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> >, at::detail::Array<char*, 3> >(int, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> >, at::detail::Array<char*, 3>)",
          "events": 1,
          "duration_us": 108.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add<c10::BFloat16>, at::detail::Array<char*, 3> >(int, at::native::CUDAFunctor_add<c10::BFloat16>, at::detail::Array<char*, 3>)",
          "events": 2,
          "duration_us": 102.0
        },
        {
          "name": "sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize256x128x64_warpgroupsize2x1x1_execute_segment_k_off_kernel__51_cublas",
          "events": 2,
          "duration_us": 93.0
        },
        {
          "name": "sm90_xmma_gemm_bf16bf16_bf16f32_f32_nt_n_tilesize64x64x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__51_cublas",
          "events": 2,
          "duration_us": 88.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > const&)::{lambda(int)#1})",
          "events": 1,
          "duration_us": 74.0
        },
        {
          "name": "_reverse_chunk_summary",
          "events": 1,
          "duration_us": 67.0
        },
        {
          "name": "void at::native::reduce_kernel<512, 1, at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4> >(at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4>)",
          "events": 1,
          "duration_us": 50.0
        },
        {
          "name": "_affine_group_local",
          "events": 1,
          "duration_us": 38.0
        },
        {
          "name": "_affine_group_correct",
          "events": 1,
          "duration_us": 6.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor<float>, at::detail::Array<char*, 1> >(int, at::native::FillFunctor<float>, at::detail::Array<char*, 1>)",
          "events": 2,
          "duration_us": 2.0
        },
        {
          "name": "_affine_group_outer",
          "events": 1,
          "duration_us": 2.0
        },
        {
          "name": "Memset (Device)",
          "events": 7,
          "duration_us": 1.0
        }
      ]
    },
    {
      "shape": {
        "batch": 4,
        "length": 2048,
        "width": 2048
      },
      "architecture": "samu",
      "backend": "fused_output_fused_write_shared_sfu_chunk32",
      "cuda_events": 49,
      "summed_cuda_duration_us": 1023.0,
      "categories": {
        "copy": {
          "events": 12,
          "duration_us": 368.0
        },
        "formal_coordinate": {
          "events": 2,
          "duration_us": 3.0
        },
        "framework_elementwise": {
          "events": 24,
          "duration_us": 255.0
        },
        "framework_reduce": {
          "events": 3,
          "duration_us": 60.0
        },
        "gemm": {
          "events": 3,
          "duration_us": 63.0
        },
        "memset": {
          "events": 1,
          "duration_us": 0.0
        },
        "reverse_prefix": {
          "events": 1,
          "duration_us": 32.0
        },
        "reverse_replay": {
          "events": 1,
          "duration_us": 191.0
        },
        "reverse_summary": {
          "events": 1,
          "duration_us": 44.0
        },
        "shared_control_reduce": {
          "events": 1,
          "duration_us": 7.0
        }
      },
      "top_cuda_events": [
        {
          "name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)",
          "events": 5,
          "duration_us": 215.0
        },
        {
          "name": "_samu_reverse_chunk_replay",
          "events": 1,
          "duration_us": 191.0
        },
        {
          "name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)",
          "events": 3,
          "duration_us": 149.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> >, at::detail::Array<char*, 3> >(int, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> >, at::detail::Array<char*, 3>)",
          "events": 2,
          "duration_us": 110.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > const&)::{lambda(int)#1})",
          "events": 2,
          "duration_us": 76.0
        },
        {
          "name": "void at::native::reduce_kernel<512, 1, at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4> >(at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4>)",
          "events": 2,
          "duration_us": 53.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add<c10::BFloat16>, at::detail::Array<char*, 3> >(int, at::native::CUDAFunctor_add<c10::BFloat16>, at::detail::Array<char*, 3>)",
          "events": 1,
          "duration_us": 49.0
        },
        {
          "name": "_samu_reverse_chunk_summary",
          "events": 1,
          "duration_us": 44.0
        },
        {
          "name": "_samu_reverse_chunk_prefix",
          "events": 1,
          "duration_us": 32.0
        },
        {
          "name": "void cutlass::Kernel<cutlass_80_tensorop_bf16_s16816gemm_bf16_256x64_32x4_nt_align2>(cutlass_80_tensorop_bf16_s16816gemm_bf16_256x64_32x4_nt_align2::Params)",
          "events": 1,
          "duration_us": 32.0
        },
        {
          "name": "void cutlass::Kernel<cutlass_80_tensorop_bf16_s16816gemm_bf16_256x128_32x3_nn_align2>(cutlass_80_tensorop_bf16_s16816gemm_bf16_256x128_32x3_nn_align2::Params)",
          "events": 1,
          "duration_us": 29.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor<float>, at::detail::Array<char*, 1> >(int, at::native::FillFunctor<float>, at::detail::Array<char*, 1>)",
          "events": 8,
          "duration_us": 8.0
        },
        {
          "name": "_reduce_shared_control_gradients",
          "events": 1,
          "duration_us": 7.0
        },
        {
          "name": "void at::native::reduce_kernel<256, 2, at::native::ReduceOp<c10::BFloat16, at::native::func_wrapper_t<c10::BFloat16, at::native::sum_functor<c10::BFloat16, float, c10::BFloat16>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, c10::BFloat16, 4> >(at::native::ReduceOp<c10::BFloat16, at::native::func_wrapper_t<c10::BFloat16, at::native::sum_functor<c10::BFloat16, float, c10::BFloat16>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, c10::BFloat16, 4>)",
          "events": 1,
          "duration_us": 7.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float> > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float> > const&)::{lambda(int)#1})",
          "events": 4,
          "duration_us": 5.0
        },
        {
          "name": "_formal_coordinate_backward",
          "events": 1,
          "duration_us": 2.0
        },
        {
          "name": "void splitKreduce_kernel<32, 16, int, __nv_bfloat16, __nv_bfloat16, float, __nv_bfloat16, true, false, false>(cublasSplitKParams<float>, __nv_bfloat16 const*, __nv_bfloat16 const*, __nv_bfloat16*, float const*, float const*, __nv_bfloat16 const*, __nv_bfloat16 const*, __nv_bfloat16*, void*, long, float*, int*)",
          "events": 1,
          "duration_us": 2.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}>(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}>(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1} const&)::{lambda(int)#1})",
          "events": 2,
          "duration_us": 2.0
        },
        {
          "name": "Memcpy DtoD (Device -> Device)",
          "events": 2,
          "duration_us": 2.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add<float>, at::detail::Array<char*, 3> >(int, at::native::CUDAFunctor_add<float>, at::detail::Array<char*, 3>)",
          "events": 2,
          "duration_us": 2.0
        },
        {
          "name": "_formal_coordinate_reduce",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::neg_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2> >(int, at::native::neg_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>)",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::compare_scalar_kernel<float>(at::TensorIteratorBase&, at::native::(anonymous namespace)::OpType, float)::{lambda(float)#1}, at::detail::Array<char*, 2> >(int, at::native::compare_scalar_kernel<float>(at::TensorIteratorBase&, at::native::(anonymous namespace)::OpType, float)::{lambda(float)#1}, at::detail::Array<char*, 2>)",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::(anonymous namespace)::where_kernel_impl(at::TensorIterator&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(bool, float, float)#1}>(at::TensorIteratorBase&, at::native::(anonymous namespace)::where_kernel_impl(at::TensorIterator&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(bool, float, float)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::(anonymous namespace)::where_kernel_impl(at::TensorIterator&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(bool, float, float)#1}>(at::TensorIteratorBase&, at::native::(anonymous namespace)::where_kernel_impl(at::TensorIterator&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(bool, float, float)#1} const&)::{lambda(int)#1})",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::AUnaryFunctor<float, float, bool, at::native::(anonymous namespace)::CompareEqFunctor<float> >, at::detail::Array<char*, 2> >(int, at::native::AUnaryFunctor<float, float, bool, at::native::(anonymous namespace)::CompareEqFunctor<float> >, at::detail::Array<char*, 2>)",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::(anonymous namespace)::masked_fill_kernel(at::TensorIterator&, c10::Scalar const&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(float, bool)#1}>(at::TensorIteratorBase&, at::native::(anonymous namespace)::masked_fill_kernel(at::TensorIterator&, c10::Scalar const&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(float, bool)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::(anonymous namespace)::masked_fill_kernel(at::TensorIterator&, c10::Scalar const&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(float, bool)#1}>(at::TensorIteratorBase&, at::native::(anonymous namespace)::masked_fill_kernel(at::TensorIterator&, c10::Scalar const&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(float, bool)#1} const&)::{lambda(int)#1})",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "Memset (Device)",
          "events": 1,
          "duration_us": 0.0
        }
      ]
    },
    {
      "shape": {
        "batch": 1,
        "length": 8192,
        "width": 2560
      },
      "architecture": "rglru",
      "backend": "fused_grouped_prefix32_hierarchical_chunk16",
      "cuda_events": 35,
      "summed_cuda_duration_us": 1929.0,
      "categories": {
        "copy": {
          "events": 9,
          "duration_us": 597.0
        },
        "framework_elementwise": {
          "events": 6,
          "duration_us": 360.0
        },
        "framework_reduce": {
          "events": 4,
          "duration_us": 215.0
        },
        "gemm": {
          "events": 4,
          "duration_us": 236.0
        },
        "memset": {
          "events": 7,
          "duration_us": 3.0
        },
        "reverse_prefix": {
          "events": 3,
          "duration_us": 70.0
        },
        "reverse_replay": {
          "events": 1,
          "duration_us": 368.0
        },
        "reverse_summary": {
          "events": 1,
          "duration_us": 80.0
        }
      },
      "top_cuda_events": [
        {
          "name": "_reverse_chunk_replay",
          "events": 1,
          "duration_us": 368.0
        },
        {
          "name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)",
          "events": 3,
          "duration_us": 264.0
        },
        {
          "name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)",
          "events": 4,
          "duration_us": 181.0
        },
        {
          "name": "void at::native::reduce_kernel<128, 4, at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4> >(at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4>)",
          "events": 3,
          "duration_us": 155.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}>(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}>(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})",
          "events": 2,
          "duration_us": 152.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> >, at::detail::Array<char*, 3> >(int, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> >, at::detail::Array<char*, 3>)",
          "events": 1,
          "duration_us": 135.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add<c10::BFloat16>, at::detail::Array<char*, 3> >(int, at::native::CUDAFunctor_add<c10::BFloat16>, at::detail::Array<char*, 3>)",
          "events": 2,
          "duration_us": 130.0
        },
        {
          "name": "sm90_xmma_gemm_bf16bf16_bf16f32_f32_nt_n_tilesize64x64x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__51_cublas",
          "events": 2,
          "duration_us": 127.0
        },
        {
          "name": "sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize256x128x64_warpgroupsize2x1x1_execute_segment_k_off_kernel__51_cublas",
          "events": 2,
          "duration_us": 109.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > const&)::{lambda(int)#1})",
          "events": 1,
          "duration_us": 93.0
        },
        {
          "name": "_reverse_chunk_summary",
          "events": 1,
          "duration_us": 80.0
        },
        {
          "name": "_affine_group_local",
          "events": 1,
          "duration_us": 61.0
        },
        {
          "name": "void at::native::reduce_kernel<512, 1, at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4> >(at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4>)",
          "events": 1,
          "duration_us": 60.0
        },
        {
          "name": "_affine_group_outer",
          "events": 1,
          "duration_us": 5.0
        },
        {
          "name": "_affine_group_correct",
          "events": 1,
          "duration_us": 4.0
        },
        {
          "name": "Memset (Device)",
          "events": 7,
          "duration_us": 3.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor<float>, at::detail::Array<char*, 1> >(int, at::native::FillFunctor<float>, at::detail::Array<char*, 1>)",
          "events": 2,
          "duration_us": 2.0
        }
      ]
    },
    {
      "shape": {
        "batch": 1,
        "length": 8192,
        "width": 2560
      },
      "architecture": "samu",
      "backend": "fused_output_fused_write_shared_sfu_grouped_prefix64_hierarchical_chunk32",
      "cuda_events": 51,
      "summed_cuda_duration_us": 1212.0,
      "categories": {
        "copy": {
          "events": 12,
          "duration_us": 454.0
        },
        "formal_coordinate": {
          "events": 2,
          "duration_us": 3.0
        },
        "framework_elementwise": {
          "events": 24,
          "duration_us": 312.0
        },
        "framework_reduce": {
          "events": 3,
          "duration_us": 70.0
        },
        "gemm": {
          "events": 3,
          "duration_us": 82.0
        },
        "memset": {
          "events": 1,
          "duration_us": 0.0
        },
        "reverse_prefix": {
          "events": 3,
          "duration_us": 26.0
        },
        "reverse_replay": {
          "events": 1,
          "duration_us": 194.0
        },
        "reverse_summary": {
          "events": 1,
          "duration_us": 65.0
        },
        "shared_control_reduce": {
          "events": 1,
          "duration_us": 6.0
        }
      },
      "top_cuda_events": [
        {
          "name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)",
          "events": 5,
          "duration_us": 267.0
        },
        {
          "name": "_samu_reverse_chunk_replay",
          "events": 1,
          "duration_us": 194.0
        },
        {
          "name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)",
          "events": 3,
          "duration_us": 183.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> >, at::detail::Array<char*, 3> >(int, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> >, at::detail::Array<char*, 3>)",
          "events": 2,
          "duration_us": 136.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > const&)::{lambda(int)#1})",
          "events": 2,
          "duration_us": 95.0
        },
        {
          "name": "_samu_reverse_chunk_summary",
          "events": 1,
          "duration_us": 65.0
        },
        {
          "name": "void at::native::reduce_kernel<512, 1, at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4> >(at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4>)",
          "events": 2,
          "duration_us": 64.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add<c10::BFloat16>, at::detail::Array<char*, 3> >(int, at::native::CUDAFunctor_add<c10::BFloat16>, at::detail::Array<char*, 3>)",
          "events": 1,
          "duration_us": 61.0
        },
        {
          "name": "void cutlass::Kernel<cutlass_80_tensorop_bf16_s16816gemm_bf16_256x64_64x3_nt_align2>(cutlass_80_tensorop_bf16_s16816gemm_bf16_256x64_64x3_nt_align2::Params)",
          "events": 1,
          "duration_us": 42.0
        },
        {
          "name": "void cutlass::Kernel<cutlass_80_tensorop_bf16_s16816gemm_bf16_256x128_32x3_nn_align2>(cutlass_80_tensorop_bf16_s16816gemm_bf16_256x128_32x3_nn_align2::Params)",
          "events": 1,
          "duration_us": 37.0
        },
        {
          "name": "_complex_affine_group_local",
          "events": 1,
          "duration_us": 23.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor<float>, at::detail::Array<char*, 1> >(int, at::native::FillFunctor<float>, at::detail::Array<char*, 1>)",
          "events": 8,
          "duration_us": 8.0
        },
        {
          "name": "_reduce_shared_control_gradients",
          "events": 1,
          "duration_us": 6.0
        },
        {
          "name": "void at::native::reduce_kernel<256, 2, at::native::ReduceOp<c10::BFloat16, at::native::func_wrapper_t<c10::BFloat16, at::native::sum_functor<c10::BFloat16, float, c10::BFloat16>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, c10::BFloat16, 4> >(at::native::ReduceOp<c10::BFloat16, at::native::func_wrapper_t<c10::BFloat16, at::native::sum_functor<c10::BFloat16, float, c10::BFloat16>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, c10::BFloat16, 4>)",
          "events": 1,
          "duration_us": 6.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float> > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float> > const&)::{lambda(int)#1})",
          "events": 4,
          "duration_us": 5.0
        },
        {
          "name": "void splitKreduce_kernel<32, 16, int, __nv_bfloat16, __nv_bfloat16, float, __nv_bfloat16, true, false, false>(cublasSplitKParams<float>, __nv_bfloat16 const*, __nv_bfloat16 const*, __nv_bfloat16*, float const*, float const*, __nv_bfloat16 const*, __nv_bfloat16 const*, __nv_bfloat16*, void*, long, float*, int*)",
          "events": 1,
          "duration_us": 3.0
        },
        {
          "name": "_complex_affine_group_correct",
          "events": 1,
          "duration_us": 2.0
        },
        {
          "name": "_formal_coordinate_backward",
          "events": 1,
          "duration_us": 2.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}>(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}>(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1} const&)::{lambda(int)#1})",
          "events": 2,
          "duration_us": 2.0
        },
        {
          "name": "Memcpy DtoD (Device -> Device)",
          "events": 2,
          "duration_us": 2.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add<float>, at::detail::Array<char*, 3> >(int, at::native::CUDAFunctor_add<float>, at::detail::Array<char*, 3>)",
          "events": 2,
          "duration_us": 2.0
        },
        {
          "name": "_complex_affine_group_outer",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "_formal_coordinate_reduce",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::neg_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2> >(int, at::native::neg_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>)",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::compare_scalar_kernel<float>(at::TensorIteratorBase&, at::native::(anonymous namespace)::OpType, float)::{lambda(float)#1}, at::detail::Array<char*, 2> >(int, at::native::compare_scalar_kernel<float>(at::TensorIteratorBase&, at::native::(anonymous namespace)::OpType, float)::{lambda(float)#1}, at::detail::Array<char*, 2>)",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::(anonymous namespace)::where_kernel_impl(at::TensorIterator&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(bool, float, float)#1}>(at::TensorIteratorBase&, at::native::(anonymous namespace)::where_kernel_impl(at::TensorIterator&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(bool, float, float)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::(anonymous namespace)::where_kernel_impl(at::TensorIterator&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(bool, float, float)#1}>(at::TensorIteratorBase&, at::native::(anonymous namespace)::where_kernel_impl(at::TensorIterator&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(bool, float, float)#1} const&)::{lambda(int)#1})",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::AUnaryFunctor<float, float, bool, at::native::(anonymous namespace)::CompareEqFunctor<float> >, at::detail::Array<char*, 2> >(int, at::native::AUnaryFunctor<float, float, bool, at::native::(anonymous namespace)::CompareEqFunctor<float> >, at::detail::Array<char*, 2>)",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::(anonymous namespace)::masked_fill_kernel(at::TensorIterator&, c10::Scalar const&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(float, bool)#1}>(at::TensorIteratorBase&, at::native::(anonymous namespace)::masked_fill_kernel(at::TensorIterator&, c10::Scalar const&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(float, bool)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::(anonymous namespace)::masked_fill_kernel(at::TensorIterator&, c10::Scalar const&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(float, bool)#1}>(at::TensorIteratorBase&, at::native::(anonymous namespace)::masked_fill_kernel(at::TensorIterator&, c10::Scalar const&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(float, bool)#1} const&)::{lambda(int)#1})",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "Memset (Device)",
          "events": 1,
          "duration_us": 0.0
        }
      ]
    },
    {
      "shape": {
        "batch": 1,
        "length": 32768,
        "width": 1024
      },
      "architecture": "rglru",
      "backend": "fused_grouped_prefix32_hierarchical_chunk16",
      "cuda_events": 36,
      "summed_cuda_duration_us": 3029.0,
      "categories": {
        "copy": {
          "events": 9,
          "duration_us": 943.0
        },
        "framework_elementwise": {
          "events": 6,
          "duration_us": 577.0
        },
        "framework_reduce": {
          "events": 4,
          "duration_us": 310.0
        },
        "gemm": {
          "events": 4,
          "duration_us": 400.0
        },
        "memset": {
          "events": 8,
          "duration_us": 1.0
        },
        "reverse_prefix": {
          "events": 3,
          "duration_us": 102.0
        },
        "reverse_replay": {
          "events": 1,
          "duration_us": 569.0
        },
        "reverse_summary": {
          "events": 1,
          "duration_us": 127.0
        }
      },
      "top_cuda_events": [
        {
          "name": "_reverse_chunk_replay",
          "events": 1,
          "duration_us": 569.0
        },
        {
          "name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)",
          "events": 3,
          "duration_us": 413.0
        },
        {
          "name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)",
          "events": 4,
          "duration_us": 286.0
        },
        {
          "name": "sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize256x128x64_warpgroupsize2x1x1_execute_segment_k_off_kernel__51_cublas",
          "events": 2,
          "duration_us": 250.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}>(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}>(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})",
          "events": 2,
          "duration_us": 244.0
        },
        {
          "name": "void at::native::reduce_kernel<128, 4, at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4> >(at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4>)",
          "events": 3,
          "duration_us": 222.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> >, at::detail::Array<char*, 3> >(int, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> >, at::detail::Array<char*, 3>)",
          "events": 1,
          "duration_us": 213.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add<c10::BFloat16>, at::detail::Array<char*, 3> >(int, at::native::CUDAFunctor_add<c10::BFloat16>, at::detail::Array<char*, 3>)",
          "events": 2,
          "duration_us": 207.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > const&)::{lambda(int)#1})",
          "events": 1,
          "duration_us": 155.0
        },
        {
          "name": "sm90_xmma_gemm_bf16bf16_bf16f32_f32_nt_n_tilesize64x64x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__51_cublas",
          "events": 2,
          "duration_us": 150.0
        },
        {
          "name": "_reverse_chunk_summary",
          "events": 1,
          "duration_us": 127.0
        },
        {
          "name": "void at::native::reduce_kernel<512, 1, at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4> >(at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4>)",
          "events": 1,
          "duration_us": 88.0
        },
        {
          "name": "_affine_group_local",
          "events": 1,
          "duration_us": 79.0
        },
        {
          "name": "_affine_group_outer",
          "events": 1,
          "duration_us": 17.0
        },
        {
          "name": "_affine_group_correct",
          "events": 1,
          "duration_us": 6.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor<float>, at::detail::Array<char*, 1> >(int, at::native::FillFunctor<float>, at::detail::Array<char*, 1>)",
          "events": 2,
          "duration_us": 2.0
        },
        {
          "name": "Memset (Device)",
          "events": 8,
          "duration_us": 1.0
        }
      ]
    },
    {
      "shape": {
        "batch": 1,
        "length": 32768,
        "width": 1024
      },
      "architecture": "samu",
      "backend": "fused_output_fused_write_shared_sfu_grouped_prefix64_hierarchical_chunk32",
      "cuda_events": 51,
      "summed_cuda_duration_us": 1995.0,
      "categories": {
        "copy": {
          "events": 12,
          "duration_us": 708.0
        },
        "formal_coordinate": {
          "events": 2,
          "duration_us": 3.0
        },
        "framework_elementwise": {
          "events": 24,
          "duration_us": 491.0
        },
        "framework_reduce": {
          "events": 3,
          "duration_us": 110.0
        },
        "gemm": {
          "events": 3,
          "duration_us": 116.0
        },
        "memset": {
          "events": 1,
          "duration_us": 0.0
        },
        "reverse_prefix": {
          "events": 3,
          "duration_us": 129.0
        },
        "reverse_replay": {
          "events": 1,
          "duration_us": 331.0
        },
        "reverse_summary": {
          "events": 1,
          "duration_us": 85.0
        },
        "shared_control_reduce": {
          "events": 1,
          "duration_us": 22.0
        }
      },
      "top_cuda_events": [
        {
          "name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)",
          "events": 5,
          "duration_us": 421.0
        },
        {
          "name": "_samu_reverse_chunk_replay",
          "events": 1,
          "duration_us": 331.0
        },
        {
          "name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)",
          "events": 3,
          "duration_us": 283.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> >, at::detail::Array<char*, 3> >(int, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> >, at::detail::Array<char*, 3>)",
          "events": 2,
          "duration_us": 214.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > const&)::{lambda(int)#1})",
          "events": 2,
          "duration_us": 157.0
        },
        {
          "name": "_complex_affine_group_local",
          "events": 1,
          "duration_us": 117.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add<c10::BFloat16>, at::detail::Array<char*, 3> >(int, at::native::CUDAFunctor_add<c10::BFloat16>, at::detail::Array<char*, 3>)",
          "events": 1,
          "duration_us": 100.0
        },
        {
          "name": "void at::native::reduce_kernel<512, 1, at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4> >(at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4>)",
          "events": 2,
          "duration_us": 92.0
        },
        {
          "name": "_samu_reverse_chunk_summary",
          "events": 1,
          "duration_us": 85.0
        },
        {
          "name": "void cutlass::Kernel<cutlass_80_tensorop_bf16_s16816gemm_bf16_128x256_32x3_nn_align2>(cutlass_80_tensorop_bf16_s16816gemm_bf16_128x256_32x3_nn_align2::Params)",
          "events": 1,
          "duration_us": 56.0
        },
        {
          "name": "void cutlass::Kernel<cutlass_80_tensorop_bf16_s16816gemm_bf16_256x64_32x4_nt_align2>(cutlass_80_tensorop_bf16_s16816gemm_bf16_256x64_32x4_nt_align2::Params)",
          "events": 1,
          "duration_us": 56.0
        },
        {
          "name": "_reduce_shared_control_gradients",
          "events": 1,
          "duration_us": 22.0
        },
        {
          "name": "void at::native::reduce_kernel<256, 2, at::native::ReduceOp<c10::BFloat16, at::native::func_wrapper_t<c10::BFloat16, at::native::sum_functor<c10::BFloat16, float, c10::BFloat16>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, c10::BFloat16, 4> >(at::native::ReduceOp<c10::BFloat16, at::native::func_wrapper_t<c10::BFloat16, at::native::sum_functor<c10::BFloat16, float, c10::BFloat16>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, c10::BFloat16, 4>)",
          "events": 1,
          "duration_us": 18.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor<float>, at::detail::Array<char*, 1> >(int, at::native::FillFunctor<float>, at::detail::Array<char*, 1>)",
          "events": 8,
          "duration_us": 8.0
        },
        {
          "name": "_complex_affine_group_outer",
          "events": 1,
          "duration_us": 8.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float> > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float> > const&)::{lambda(int)#1})",
          "events": 4,
          "duration_us": 5.0
        },
        {
          "name": "_complex_affine_group_correct",
          "events": 1,
          "duration_us": 4.0
        },
        {
          "name": "void splitKreduce_kernel<32, 16, int, __nv_bfloat16, __nv_bfloat16, float, __nv_bfloat16, true, false, false>(cublasSplitKParams<float>, __nv_bfloat16 const*, __nv_bfloat16 const*, __nv_bfloat16*, float const*, float const*, __nv_bfloat16 const*, __nv_bfloat16 const*, __nv_bfloat16*, void*, long, float*, int*)",
          "events": 1,
          "duration_us": 4.0
        },
        {
          "name": "_formal_coordinate_backward",
          "events": 1,
          "duration_us": 2.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}>(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}>(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1} const&)::{lambda(int)#1})",
          "events": 2,
          "duration_us": 2.0
        },
        {
          "name": "Memcpy DtoD (Device -> Device)",
          "events": 2,
          "duration_us": 2.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add<float>, at::detail::Array<char*, 3> >(int, at::native::CUDAFunctor_add<float>, at::detail::Array<char*, 3>)",
          "events": 2,
          "duration_us": 2.0
        },
        {
          "name": "_formal_coordinate_reduce",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::neg_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2> >(int, at::native::neg_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>)",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::compare_scalar_kernel<float>(at::TensorIteratorBase&, at::native::(anonymous namespace)::OpType, float)::{lambda(float)#1}, at::detail::Array<char*, 2> >(int, at::native::compare_scalar_kernel<float>(at::TensorIteratorBase&, at::native::(anonymous namespace)::OpType, float)::{lambda(float)#1}, at::detail::Array<char*, 2>)",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::(anonymous namespace)::where_kernel_impl(at::TensorIterator&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(bool, float, float)#1}>(at::TensorIteratorBase&, at::native::(anonymous namespace)::where_kernel_impl(at::TensorIterator&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(bool, float, float)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::(anonymous namespace)::where_kernel_impl(at::TensorIterator&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(bool, float, float)#1}>(at::TensorIteratorBase&, at::native::(anonymous namespace)::where_kernel_impl(at::TensorIterator&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(bool, float, float)#1} const&)::{lambda(int)#1})",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::AUnaryFunctor<float, float, bool, at::native::(anonymous namespace)::CompareEqFunctor<float> >, at::detail::Array<char*, 2> >(int, at::native::AUnaryFunctor<float, float, bool, at::native::(anonymous namespace)::CompareEqFunctor<float> >, at::detail::Array<char*, 2>)",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::(anonymous namespace)::masked_fill_kernel(at::TensorIterator&, c10::Scalar const&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(float, bool)#1}>(at::TensorIteratorBase&, at::native::(anonymous namespace)::masked_fill_kernel(at::TensorIterator&, c10::Scalar const&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(float, bool)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::(anonymous namespace)::masked_fill_kernel(at::TensorIterator&, c10::Scalar const&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(float, bool)#1}>(at::TensorIteratorBase&, at::native::(anonymous namespace)::masked_fill_kernel(at::TensorIterator&, c10::Scalar const&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(float, bool)#1} const&)::{lambda(int)#1})",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "Memset (Device)",
          "events": 1,
          "duration_us": 0.0
        }
      ]
    },
    {
      "shape": {
        "batch": 1,
        "length": 32768,
        "width": 2048
      },
      "architecture": "rglru",
      "backend": "fused_grouped_prefix32_hierarchical_chunk16",
      "cuda_events": 36,
      "summed_cuda_duration_us": 5664.0,
      "categories": {
        "copy": {
          "events": 9,
          "duration_us": 1835.0
        },
        "framework_elementwise": {
          "events": 6,
          "duration_us": 1156.0
        },
        "framework_reduce": {
          "events": 4,
          "duration_us": 507.0
        },
        "gemm": {
          "events": 4,
          "duration_us": 658.0
        },
        "memset": {
          "events": 8,
          "duration_us": 3.0
        },
        "reverse_prefix": {
          "events": 3,
          "duration_us": 144.0
        },
        "reverse_replay": {
          "events": 1,
          "duration_us": 1116.0
        },
        "reverse_summary": {
          "events": 1,
          "duration_us": 245.0
        }
      },
      "top_cuda_events": [
        {
          "name": "_reverse_chunk_replay",
          "events": 1,
          "duration_us": 1116.0
        },
        {
          "name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)",
          "events": 3,
          "duration_us": 806.0
        },
        {
          "name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)",
          "events": 4,
          "duration_us": 557.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}>(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}>(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1} const&)::{lambda(int)#1})",
          "events": 2,
          "duration_us": 472.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> >, at::detail::Array<char*, 3> >(int, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> >, at::detail::Array<char*, 3>)",
          "events": 1,
          "duration_us": 427.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add<c10::BFloat16>, at::detail::Array<char*, 3> >(int, at::native::CUDAFunctor_add<c10::BFloat16>, at::detail::Array<char*, 3>)",
          "events": 2,
          "duration_us": 416.0
        },
        {
          "name": "void at::native::reduce_kernel<128, 4, at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4> >(at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4>)",
          "events": 3,
          "duration_us": 341.0
        },
        {
          "name": "sm90_xmma_gemm_bf16bf16_bf16f32_f32_nt_n_tilesize64x64x64_warpgroupsize1x1x1_execute_segment_k_off_kernel__51_cublas",
          "events": 2,
          "duration_us": 336.0
        },
        {
          "name": "sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize256x128x64_warpgroupsize2x1x1_execute_segment_k_off_kernel__51_cublas",
          "events": 2,
          "duration_us": 322.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > const&)::{lambda(int)#1})",
          "events": 1,
          "duration_us": 311.0
        },
        {
          "name": "_reverse_chunk_summary",
          "events": 1,
          "duration_us": 245.0
        },
        {
          "name": "void at::native::reduce_kernel<512, 1, at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4> >(at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4>)",
          "events": 1,
          "duration_us": 166.0
        },
        {
          "name": "_affine_group_local",
          "events": 1,
          "duration_us": 102.0
        },
        {
          "name": "_affine_group_correct",
          "events": 1,
          "duration_us": 24.0
        },
        {
          "name": "_affine_group_outer",
          "events": 1,
          "duration_us": 18.0
        },
        {
          "name": "Memset (Device)",
          "events": 8,
          "duration_us": 3.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor<float>, at::detail::Array<char*, 1> >(int, at::native::FillFunctor<float>, at::detail::Array<char*, 1>)",
          "events": 2,
          "duration_us": 2.0
        }
      ]
    },
    {
      "shape": {
        "batch": 1,
        "length": 32768,
        "width": 2048
      },
      "architecture": "samu",
      "backend": "fused_output_fused_write_shared_sfu_grouped_prefix64_hierarchical_chunk32",
      "cuda_events": 51,
      "summed_cuda_duration_us": 3777.0,
      "categories": {
        "copy": {
          "events": 12,
          "duration_us": 1375.0
        },
        "formal_coordinate": {
          "events": 2,
          "duration_us": 3.0
        },
        "framework_elementwise": {
          "events": 24,
          "duration_us": 966.0
        },
        "framework_reduce": {
          "events": 3,
          "duration_us": 190.0
        },
        "gemm": {
          "events": 3,
          "duration_us": 202.0
        },
        "memset": {
          "events": 1,
          "duration_us": 1.0
        },
        "reverse_prefix": {
          "events": 3,
          "duration_us": 250.0
        },
        "reverse_replay": {
          "events": 1,
          "duration_us": 602.0
        },
        "reverse_summary": {
          "events": 1,
          "duration_us": 166.0
        },
        "shared_control_reduce": {
          "events": 1,
          "duration_us": 22.0
        }
      },
      "top_cuda_events": [
        {
          "name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)",
          "events": 5,
          "duration_us": 827.0
        },
        {
          "name": "_samu_reverse_chunk_replay",
          "events": 1,
          "duration_us": 602.0
        },
        {
          "name": "void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1> >(int, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#12}::operator()() const::{lambda(c10::BFloat16)#1}, at::detail::Array<char*, 2>, TrivialOffsetCalculator<1, unsigned int>, TrivialOffsetCalculator<1, unsigned int>, at::native::memory::LoadWithCast<1>, at::native::memory::StoreWithCast<1>)",
          "events": 3,
          "duration_us": 544.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> >, at::detail::Array<char*, 3> >(int, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> >, at::detail::Array<char*, 3>)",
          "events": 2,
          "duration_us": 427.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float> > const&)::{lambda(int)#1})",
          "events": 2,
          "duration_us": 313.0
        },
        {
          "name": "_complex_affine_group_local",
          "events": 1,
          "duration_us": 235.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add<c10::BFloat16>, at::detail::Array<char*, 3> >(int, at::native::CUDAFunctor_add<c10::BFloat16>, at::detail::Array<char*, 3>)",
          "events": 1,
          "duration_us": 205.0
        },
        {
          "name": "void at::native::reduce_kernel<512, 1, at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4> >(at::native::ReduceOp<float, at::native::func_wrapper_t<float, at::native::sum_functor<float, float, float>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, float, 4>)",
          "events": 2,
          "duration_us": 171.0
        },
        {
          "name": "_samu_reverse_chunk_summary",
          "events": 1,
          "duration_us": 166.0
        },
        {
          "name": "void cutlass::Kernel<cutlass_80_tensorop_bf16_s16816gemm_bf16_256x128_32x3_nn_align2>(cutlass_80_tensorop_bf16_s16816gemm_bf16_256x128_32x3_nn_align2::Params)",
          "events": 1,
          "duration_us": 108.0
        },
        {
          "name": "void cutlass::Kernel<cutlass_80_tensorop_bf16_s16816gemm_bf16_256x64_32x4_nt_align2>(cutlass_80_tensorop_bf16_s16816gemm_bf16_256x64_32x4_nt_align2::Params)",
          "events": 1,
          "duration_us": 91.0
        },
        {
          "name": "_reduce_shared_control_gradients",
          "events": 1,
          "duration_us": 22.0
        },
        {
          "name": "void at::native::reduce_kernel<256, 2, at::native::ReduceOp<c10::BFloat16, at::native::func_wrapper_t<c10::BFloat16, at::native::sum_functor<c10::BFloat16, float, c10::BFloat16>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, c10::BFloat16, 4> >(at::native::ReduceOp<c10::BFloat16, at::native::func_wrapper_t<c10::BFloat16, at::native::sum_functor<c10::BFloat16, float, c10::BFloat16>::operator()(at::TensorIterator&)::{lambda(float, float)#1}>, unsigned int, c10::BFloat16, 4>)",
          "events": 1,
          "duration_us": 19.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor<float>, at::detail::Array<char*, 1> >(int, at::native::FillFunctor<float>, at::detail::Array<char*, 1>)",
          "events": 8,
          "duration_us": 8.0
        },
        {
          "name": "_complex_affine_group_outer",
          "events": 1,
          "duration_us": 8.0
        },
        {
          "name": "_complex_affine_group_correct",
          "events": 1,
          "duration_us": 7.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float> > const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float> > >(at::TensorIteratorBase&, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float> > const&)::{lambda(int)#1})",
          "events": 4,
          "duration_us": 5.0
        },
        {
          "name": "void splitKreduce_kernel<32, 16, int, __nv_bfloat16, __nv_bfloat16, float, __nv_bfloat16, true, false, false>(cublasSplitKParams<float>, __nv_bfloat16 const*, __nv_bfloat16 const*, __nv_bfloat16*, float const*, float const*, __nv_bfloat16 const*, __nv_bfloat16 const*, __nv_bfloat16*, void*, long, float*, int*)",
          "events": 1,
          "duration_us": 3.0
        },
        {
          "name": "_formal_coordinate_backward",
          "events": 1,
          "duration_us": 2.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}>(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}>(at::TensorIteratorBase&, at::native::direct_copy_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1} const&)::{lambda(int)#1})",
          "events": 2,
          "duration_us": 2.0
        },
        {
          "name": "Memcpy DtoD (Device -> Device)",
          "events": 2,
          "duration_us": 2.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctor_add<float>, at::detail::Array<char*, 3> >(int, at::native::CUDAFunctor_add<float>, at::detail::Array<char*, 3>)",
          "events": 2,
          "duration_us": 2.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::(anonymous namespace)::where_kernel_impl(at::TensorIterator&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(bool, float, float)#1}>(at::TensorIteratorBase&, at::native::(anonymous namespace)::where_kernel_impl(at::TensorIterator&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(bool, float, float)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::(anonymous namespace)::where_kernel_impl(at::TensorIterator&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(bool, float, float)#1}>(at::TensorIteratorBase&, at::native::(anonymous namespace)::where_kernel_impl(at::TensorIterator&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(bool, float, float)#1} const&)::{lambda(int)#1})",
          "events": 1,
          "duration_us": 2.0
        },
        {
          "name": "Memset (Device)",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "_formal_coordinate_reduce",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::neg_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2> >(int, at::native::neg_kernel_cuda(at::TensorIteratorBase&)::{lambda()#2}::operator()() const::{lambda()#7}::operator()() const::{lambda(float)#1}, at::detail::Array<char*, 2>)",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::compare_scalar_kernel<float>(at::TensorIteratorBase&, at::native::(anonymous namespace)::OpType, float)::{lambda(float)#1}, at::detail::Array<char*, 2> >(int, at::native::compare_scalar_kernel<float>(at::TensorIteratorBase&, at::native::(anonymous namespace)::OpType, float)::{lambda(float)#1}, at::detail::Array<char*, 2>)",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::vectorized_elementwise_kernel<4, at::native::AUnaryFunctor<float, float, bool, at::native::(anonymous namespace)::CompareEqFunctor<float> >, at::detail::Array<char*, 2> >(int, at::native::AUnaryFunctor<float, float, bool, at::native::(anonymous namespace)::CompareEqFunctor<float> >, at::detail::Array<char*, 2>)",
          "events": 1,
          "duration_us": 1.0
        },
        {
          "name": "void at::native::elementwise_kernel<128, 2, at::native::gpu_kernel_impl<at::native::(anonymous namespace)::masked_fill_kernel(at::TensorIterator&, c10::Scalar const&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(float, bool)#1}>(at::TensorIteratorBase&, at::native::(anonymous namespace)::masked_fill_kernel(at::TensorIterator&, c10::Scalar const&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(float, bool)#1} const&)::{lambda(int)#1}>(int, at::native::gpu_kernel_impl<at::native::(anonymous namespace)::masked_fill_kernel(at::TensorIterator&, c10::Scalar const&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(float, bool)#1}>(at::TensorIteratorBase&, at::native::(anonymous namespace)::masked_fill_kernel(at::TensorIterator&, c10::Scalar const&)::{lambda()#1}::operator()() const::{lambda()#7}::operator()() const::{lambda(float, bool)#1} const&)::{lambda(int)#1})",
          "events": 1,
          "duration_us": 1.0
        }
      ]
    }
  ]
}