h3-blackwell-runtime/benchmarks/gb10-post-fc2-warmed-step-stats_cuda_gpu_kern_sum.csv

1	Time (%)	Total Time (ns)	Instances	Avg (ns)	Med (ns)	Min (ns)	Max (ns)	StdDev (ns)	Name
2	57.9	12095370432	50	241907408.6	241960816.0	238904576	245704832	1430443.8	void qk_int_sv_f8_attn_kernel<(unsigned int)128, (unsigned int)64, (unsigned int)32, (unsigned int)64, (unsigned int)128, (DataType)1, (QuantGranularity)2, (QuantGranularity)2, float, (bool)1, __nv_bfloat16, (ComputeUnit)1, (MaskMode)0, (bool)0, (bool)1, (bool)0, (bool)1>(signed char *, signed char *, signed char *, T11 *, float *, float *, float *, float *, float *, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, float)
3	19.9	4167855808	200	20839279.0	20371696.0	7702240	35071936	9954629.7	cutlass3x_sm120_bstensorop_s16864gemm_block_scaled_ue4m3xe2m1_ue4m3xe2m1_f32_bf16_bf16_128x128x256_1x1x1_0_tnn_align32_o_vs16_bias_bf16_relu
4	3.0	616933984	50	12338679.7	12314672.0	11865984	12928448	384444.2	<unnamed>::partial_absmax_swiglu_bf16_kernel(const unsigned short *, float *, long, long)
5	2.9	606392320	50	12127846.4	12134352.0	11797920	12481600	272664.4	void comfy::<unnamed>::rope_kernel<__nv_bfloat16, __nv_bfloat16, __nv_bfloat16, (bool)1, (bool)1, (bool)1, (bool)1, (bool)0>(const T1 *, const T1 *, const T2 *, const T3 *, const T3 *, T1 *, T1 *, long, long, long, int, int, long, long, long, long, long, long, long, long, long, long, long, long, long, long, long, long, long, long, long, long, long, long, long, long, long, long, long, float)
6	2.6	549298592	100	5492985.9	5476528.0	5449216	5824160	55130.6	_gate_add_kernel
7	2.5	528098848	50	10561977.0	10468928.0	10341376	11079968	201841.8	<unnamed>::quantize_nvfp4_swiglu_bf16_kernel(const unsigned short *, const float *, unsigned char *, unsigned char *, long, long, long, long, long)
8	1.8	371120608	100	3711206.1	3518080.0	3408320	4344608	319034.1	void <unnamed>::partial_absmax_modulated_bf16_kernel<float>(const unsigned short *, const T1 *, const T1 *, const int *, float *, long, long)
9	1.7	358825984	102	3517901.8	3507296.0	45056	3844544	361846.1	void at::native::<unnamed>::vectorized_layer_norm_kernel<c10::BFloat16, float, (bool)1>(int, T2, const T1 *, const T1 *, const T1 *, T2 *, T2 *, T1 *)
10	1.4	286555520	50	5731110.4	5711552.0	5660672	5912992	64887.3	void MeanScaleKernel<(unsigned int)64, (bool)0, __nv_bfloat16>(T3 *, signed char *, float *, float *, float, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int)
11	1.3	264825568	50	5296511.4	5240688.0	5065568	5700544	172507.0	void TransposePadPermuteKernel<(unsigned int)128, (unsigned int)64, (bool)1, __nv_bfloat16>(T4 *, T4 *, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int)
12	1.2	248751328	100	2487513.3	2438752.0	2417696	2845728	108528.2	void <unnamed>::quantize_nvfp4_modulated_bf16_kernel<float>(const unsigned short *, const T1 *, const T1 *, const int *, const float *, unsigned char *, unsigned char *, long, long, long, long, long, long)
13	0.9	192832480	50	3856649.6	3831584.0	3724512	4098912	116494.0	void QuantInt8Kernel<(unsigned int)128, (unsigned int)64, (unsigned int)1, (bool)0, (bool)1, __nv_bfloat16>(T6 *, T6 *, signed char *, float *, float, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int)
14	0.9	189654560	50	3793091.2	3766976.0	3707360	4001984	83318.4	void QuantInt8Kernel<(unsigned int)128, (unsigned int)32, (unsigned int)1, (bool)0, (bool)0, __nv_bfloat16>(T6 *, T6 *, signed char *, float *, float, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int)
15	0.7	138351232	50	2767024.6	2761616.0	2747456	2899360	25808.7	void comfy::<unnamed>::quantize_nvfp4_kernel<__nv_bfloat16, __nv_fp4x2_e2m1, __nv_fp8_e4m3, (bool)1, (bool)1>(const T1 *, const float *, T2 *, T3 *, unsigned long, unsigned long, unsigned long, unsigned long, float)
16	0.6	119167296	50	2383345.9	2358464.0	2338976	2551328	51731.7	void at::native::reduce_kernel<(int)128, (int)4, at::native::ReduceOp<c10::BFloat16, at::native::MeanOps<c10::BFloat16, float, float, c10::BFloat16>, unsigned int, c10::BFloat16, (int)4, (int)4>>(T3)
17	0.5	108606272	50	2172125.4	2172144.0	2156896	2190848	8057.6	<unnamed>::partial_absmax_bf16_vec_kernel(const uint4 *, const unsigned short *, float *, long, long)
18	0.1	14493152	200	72465.8	48272.0	40608	206816	42011.8	void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<unsigned char>, std::array<char *, (unsigned long)1>>(int, T2, T3)
19	0.0	6883840	2	3441920.0	3441920.0	61632	6822208	4780449.1	void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<float>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)
20	0.0	5532736	60	92212.3	2912.0	1280	5035136	649588.6	void at::native::vectorized_elementwise_kernel<(int)4, at::native::bfloat16_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)
21	0.0	5508544	2	2754272.0	2754272.0	51296	5457248	3822585.3	void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)
22	0.0	4915936	1	4915936.0	4915936.0	4915936	4915936	0.0	void cutlass::Kernel2<cutlass_80_simt_sgemm_256x128_8x4_tn_align1>(T1::Params)
23	0.0	4410976	1	4410976.0	4410976.0	4410976	4410976	0.0	void magma_sgemmEx_kernel<float, float, float, (bool)1, (bool)0, (int)6, (int)4, (int)6, (int)3, (int)4>(int, int, int, Tensor, int, Tensor, int, Tensor, int, Tensor, int, int, int, const T1 *, const T1 *, T1, T1, int, cublasLtEpilogue_t, int, const void *, long)
24	0.0	3766816	1	3766816.0	3766816.0	3766816	3766816	0.0	void at::native::<unnamed>::CatArrayBatchedCopy_alignedK_contig<at::native::<unnamed>::OpaqueType<(unsigned int)2>, unsigned int, (int)2, (int)128, (int)1, (int)8>(T1 *, at::native::<unnamed>::CatArrInputTensorMetadata<T1, T2, T4, T5>, at::native::<unnamed>::TensorSizeStride<T2, (unsigned int)4>, int, T2)
25	0.0	1542400	51	30243.1	26560.0	4640	93376	13266.2	std::enable_if<!T7, void>::type internal::gemvx::kernel<int, int, float, float, float, float, (bool)0, (bool)1, (bool)1, (bool)0, (int)6, (bool)0, cublasGemvParamsEx<int, cublasGemvTensorStridedBatched<const float>, cublasGemvTensorStridedBatched<const float>, cublasGemvTensorStridedBatched<float>, float>>(T13)
26	0.0	873696	1	873696.0	873696.0	873696	873696	0.0	void at::native::<unnamed>::CatArrayBatchedCopy_alignedK_contig<at::native::<unnamed>::OpaqueType<(unsigned int)4>, unsigned int, (int)3, (int)128, (int)1, (int)16>(T1 *, at::native::<unnamed>::CatArrInputTensorMetadata<T1, T2, T4, T5>, at::native::<unnamed>::TensorSizeStride<T2, (unsigned int)4>, int, T2)
27	0.0	730688	201	3635.3	2336.0	1920	100448	7972.3	void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)
28	0.0	501280	6	83546.7	81136.0	1600	172960	89454.6	void at::native::vectorized_elementwise_kernel<(int)4, at::native::CUDAFunctor_add<float>, std::array<char *, (unsigned long)3>>(int, T2, T3)
29	0.0	405472	200	2027.4	1920.0	1824	7584	687.4	void at::native::vectorized_elementwise_kernel<(int)4, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)3>>(int, T2, T3)
30	0.0	400288	150	2668.6	2176.0	1920	9696	1031.2	<unnamed>::final_scale_bf16_compat_kernel(const float *, float *, long, float)
31	0.0	358944	102	3519.1	4304.0	1600	11488	2037.3	void at::native::index_elementwise_kernel<(int)128, (int)4, void at::native::gpu_index_kernel<void at::native::index_kernel_impl<at::native::OpaqueType<(int)4>>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>)::[lambda(char *, const char *, long) (instance 1)]>(at::TensorIteratorBase &, c10::ArrayRef<long>, c10::ArrayRef<long>, const T1 &, bool)::[lambda(int) (instance 1)]>(long, T3)
32	0.0	224736	55	4086.1	1536.0	1504	71520	12382.0	void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)
33	0.0	218368	5	43673.6	27680.0	2240	88064	35553.0	void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)
34	0.0	217312	51	4261.0	3456.0	3040	22080	3140.9	void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 4)]::operator ()() const::[lambda(long) (instance 1)], std::array<char *, (unsigned long)2>, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)1>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)
35	0.0	126112	2	63056.0	63056.0	39712	86400	33013.4	void at::native::vectorized_elementwise_kernel<(int)4, at::native::neg_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)
36	0.0	114816	51	2251.3	1504.0	1472	14048	2514.1	void at::native::unrolled_elementwise_kernel<at::native::CUDAFunctor_add<float>, std::array<char *, (unsigned long)3>, (int)4, TrivialOffsetCalculator<(int)2, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithCast<(int)2>, at::native::memory::StoreWithCast<(int)1>>(int, T1, T2, T4, T5, T6, T7)
37	0.0	108896	52	2094.2	1952.0	1920	3904	385.3	void at::native::vectorized_elementwise_kernel<(int)4, at::native::<unnamed>::launch_clamp_scalar(at::TensorIteratorBase &, c10::Scalar, c10::Scalar, at::native::detail::ClampLimits)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 7)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)
38	0.0	104288	51	2044.9	1312.0	1280	3744	1021.8	void at::native::vectorized_elementwise_kernel<(int)4, at::native::floor_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)
39	0.0	99968	1	99968.0	99968.0	99968	99968	0.0	void cutlass::Kernel2<cutlass_80_simt_sgemm_128x32_8x5_tn_align1>(T1::Params)
40	0.0	97952	50	1959.0	1952.0	1920	2048	29.1	<unnamed>::final_scale_warp_kernel(const float *, float *, long, float)
41	0.0	89600	50	1792.0	896.0	832	8672	1561.9	void at::native::vectorized_elementwise_kernel<(int)4, at::native::FillFunctor<float>, std::array<char *, (unsigned long)1>>(int, T2, T3)
42	0.0	87776	2	43888.0	43888.0	3904	83872	56545.9	void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)
43	0.0	87744	60	1462.4	1440.0	1088	3104	258.8	void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)
44	0.0	83392	51	1635.1	1280.0	1216	8192	1333.7	void at::native::elementwise_kernel<(int)128, (int)2, void at::native::gpu_kernel_impl_nocast<at::native::<unnamed>::lerp_tensor_kernel(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float, float, float) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)
45	0.0	78720	51	1543.5	1344.0	1312	6144	751.5	void at::native::vectorized_elementwise_kernel<(int)2, at::native::<unnamed>::launch_clamp_scalar(at::TensorIteratorBase &, c10::Scalar, c10::Scalar, at::native::detail::ClampLimits)::[lambda() (instance 1)]::operator ()() const::[lambda() (instance 4)]::operator ()() const::[lambda(long) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)
46	0.0	73152	51	1434.4	1120.0	1088	7584	1226.3	void at::native::vectorized_elementwise_kernel<(int)2, at::native::CUDAFunctorOnSelf_add<long>, std::array<char *, (unsigned long)2>>(int, T2, T3)
47	0.0	52384	1	52384.0	52384.0	52384	52384	0.0	void at::native::<unnamed>::CatArrayBatchedCopy<at::native::<unnamed>::OpaqueType<(unsigned int)4>, unsigned int, (int)2, (int)64, (int)64>(T1 *, at::native::<unnamed>::CatArrInputTensorMetadata<T1, T2, T4, T5>, at::native::<unnamed>::TensorSizeStride<T2, (unsigned int)4>, int, T2)
48	0.0	48832	1	48832.0	48832.0	48832	48832	0.0	void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl_nocast<at::native::direct_copy_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 3)]::operator ()() const::[lambda() (instance 12)]::operator ()() const::[lambda(c10::BFloat16) (instance 1)]>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)
49	0.0	44096	1	44096.0	44096.0	44096	44096	0.0	void cutlass::Kernel2<cutlass_80_simt_sgemm_128x64_8x5_tt_align1>(T1::Params)
50	0.0	39456	1	39456.0	39456.0	39456	39456	0.0	void at::native::vectorized_elementwise_kernel<(int)4, at::native::sin_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)
51	0.0	31872	1	31872.0	31872.0	31872	31872	0.0	void at::native::vectorized_elementwise_kernel<(int)4, at::native::cos_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 2)]::operator ()() const::[lambda(float) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)
52	0.0	16448	7	2349.7	1376.0	1056	6688	2004.8	void at::native::vectorized_elementwise_kernel<(int)4, at::native::CUDAFunctorOnSelf_add<float>, std::array<char *, (unsigned long)2>>(int, T2, T3)
53	0.0	11424	5	2284.8	2080.0	1824	3392	628.1	void at::native::vectorized_elementwise_kernel<(int)4, at::native::BinaryFunctor<float, float, float, at::native::binary_internal::DivFunctor<float>>, std::array<char *, (unsigned long)3>>(int, T2, T3)
54	0.0	8960	2	4480.0	4480.0	3360	5600	1583.9	void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl_nocast<at::native::BinaryFunctor<c10::BFloat16, c10::BFloat16, c10::BFloat16, at::native::binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)
55	0.0	6112	2	3056.0	3056.0	1184	4928	2647.4	void at::native::unrolled_elementwise_kernel<at::native::CUDAFunctor_add<float>, std::array<char *, (unsigned long)3>, (int)4, TrivialOffsetCalculator<(int)2, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithoutCast, at::native::memory::StoreWithoutCast>(int, T1, T2, T4, T5, T6, T7)
56	0.0	5728	1	5728.0	5728.0	5728	5728	0.0	void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl<at::native::BinaryFunctor<c10::BFloat16, c10::BFloat16, c10::BFloat16, at::native::binary_internal::MulFunctor<float>>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)
57	0.0	4672	2	2336.0	2336.0	1824	2848	724.1	void at::native::unrolled_elementwise_kernel<at::native::AUnaryFunctor<float, float, bool, at::native::<unnamed>::CompareEqFunctor<float>>, std::array<char *, (unsigned long)2>, (int)4, TrivialOffsetCalculator<(int)1, unsigned int>, TrivialOffsetCalculator<(int)1, unsigned int>, at::native::memory::LoadWithoutCast, at::native::memory::StoreWithoutCast>(int, T1, T2, T4, T5, T6, T7)
58	0.0	4128	1	4128.0	4128.0	4128	4128	0.0	void at::native::elementwise_kernel<(int)128, (int)4, void at::native::gpu_kernel_impl_nocast<at::native::CUDAFunctor_add<c10::BFloat16>>(at::TensorIteratorBase &, const T1 &)::[lambda(int) (instance 1)]>(int, T3)
59	0.0	4000	1	4000.0	4000.0	4000	4000	0.0	void cublasLt::splitKreduce_kernel<(int)32, (int)16, int, float, float, float, float, (bool)0, float, float, float, (bool)1, (bool)1, (bool)0, (bool)0>(cublasLt::cublasSplitKParams<T6>, const T4 *, const T10 *, T9 *, T5 *, const T6 *, const T6 *, const T11 *, const T4 *, T11 *, void *, long, T6 *, int *, T6 *, T6 *, const T6 *, const T6 *, const T6 *, const T6 *, const T6 *)
60	0.0	3872	2	1936.0	1936.0	1024	2848	1289.8	void at::native::vectorized_elementwise_kernel<(int)4, at::native::CUDAFunctorOnOther_add<float>, std::array<char *, (unsigned long)2>>(int, T2, T3)
61	0.0	1984	1	1984.0	1984.0	1984	1984	0.0	void at::native::vectorized_elementwise_kernel<(int)4, at::native::BUnaryFunctor<float, float, float, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)
62	0.0	1664	1	1664.0	1664.0	1664	1664	0.0	void at::native::vectorized_elementwise_kernel<(int)4, at::native::neg_kernel_cuda(at::TensorIteratorBase &)::[lambda() (instance 2)]::operator ()() const::[lambda() (instance 9)]::operator ()() const::[lambda(c10::BFloat16) (instance 1)], std::array<char *, (unsigned long)2>>(int, T2, T3)
63	0.0	1312	1	1312.0	1312.0	1312	1312	0.0	void at::native::vectorized_elementwise_kernel<(int)4, at::native::AUnaryFunctor<c10::BFloat16, c10::BFloat16, c10::BFloat16, at::native::binary_internal::MulFunctor<float>>, std::array<char *, (unsigned long)2>>(int, T2, T3)