From ac4ecdacc5abe00aced110aa51184d672ffec6cf Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Micha=C5=82=20Kulikowski?= Date: Mon, 29 Sep 2025 18:05:04 +0200 Subject: [PATCH] [CK][Examples] Extending support for rdna3/4 in following examples: (#2884) * [CK][Examples] Extending support for rdna3/4 in following examples: -example_gemm_xdl_splitk_reduce_multi_d_fp16 -example_gemm_xdl_splitk_reduce_multi_d_bf16 -example_gemm_xdl_splitk_reduce_bf16A_i8B -example_gemm_xdl_splitk_reduce_bfp16 -example_splitk_gemm_bias_e_permute_xdl_fp32 -example_gemm_add_multiply_xdl_fp16 -example_complex_contraction_bilinear_xdl_fp32 -example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 -example_batched_gemm_bias_e_permute_xdl_fp16 -example_gemm_xdl_fp16 -example_gemm_xdl_fp16_av2 -example_gemm_xdl_wavelet_fp16 -example_gemm_add_add_fastgelu_xdl_bf16 -example_gemm_add_add_fastgelu_xdl_fp16 -example_gemm_add_add_fastgelu_xdl_fp32 -example_grouped_gemm_xdl_fp32 -example_grouped_gemm_xdl_fp16 -example_grouped_gemm_xdl_bf16 -example_cgemm_xdl_bf16 -example_cgemm_xdl_fp16 Signed-off-by: Michal Kulikowski * [CK][Examples] Extending support for rdna3/4 in following examples: -example_gemm_xdl_splitk_reduce_multi_d_fp16 -example_gemm_xdl_splitk_reduce_multi_d_bf16 -example_gemm_xdl_splitk_reduce_bf16A_i8B -example_gemm_xdl_splitk_reduce_bfp16 -example_splitk_gemm_bias_e_permute_xdl_fp32 -example_gemm_add_multiply_xdl_fp16 -example_complex_contraction_bilinear_xdl_fp32 -example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 -example_batched_gemm_bias_e_permute_xdl_fp16 -example_gemm_xdl_fp16 -example_gemm_xdl_fp16_av2 -example_gemm_xdl_wavelet_fp16 -example_gemm_add_add_fastgelu_xdl_bf16 -example_gemm_add_add_fastgelu_xdl_fp16 -example_gemm_add_add_fastgelu_xdl_fp32 -example_grouped_gemm_xdl_fp32 -example_grouped_gemm_xdl_fp16 -example_grouped_gemm_xdl_bf16 -example_cgemm_xdl_bf16 -example_cgemm_xdl_fp16 Signed-off-by: Michal Kulikowski --------- Signed-off-by: Michal Kulikowski [ROCm/composable_kernel commit: 2b684f0a7d2317b1b1f001716acb62f566cc71ee] --- example/01_gemm/gemm_xdl_fp16.cpp | 4 ++-- example/01_gemm/gemm_xdl_fp16_v2.cpp | 8 ++++---- example/01_gemm/gemm_xdl_wavelet_fp16.cpp | 4 ++-- .../gemm_add_add_fastgelu_xdl_bf16.cpp | 4 ++-- .../gemm_add_add_fastgelu_xdl_fp16.cpp | 4 ++-- .../gemm_add_add_fastgelu_xdl_fp32.cpp | 4 ++-- example/15_grouped_gemm/grouped_gemm_xdl_bf16.cpp | 4 ++-- example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp | 4 ++-- example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp | 4 ++-- example/22_cgemm/cgemm_xdl_bf16.cpp | 12 ++++++------ example/22_cgemm/cgemm_xdl_fp16.cpp | 12 ++++++------ .../grouped_gemm_bias_e_permute_xdl_fp16.cpp | 2 +- .../batched_gemm_bias_e_permute_xdl_fp16.cpp | 4 ++-- ...riangle_scale_softmax_gemm_permute_xdl_fp16.cpp | 14 +++++++------- .../35_splitK_gemm/gemm_xdl_splitk_reduce_bf16.cpp | 8 ++++---- .../gemm_xdl_splitk_reduce_bf16A_i8B.cpp | 8 ++++---- .../gemm_xdl_splitk_reduce_multi_d_bf16.cpp | 8 ++++---- .../gemm_xdl_splitk_reduce_multi_d_fp16.cpp | 8 ++++---- .../splitk_gemm_bias_e_permute_xdl_fp32.cpp | 4 ++-- .../gemm_add_multiply_xdl_fp16.cpp | 4 ++-- .../run_gemm_add_multiply_example.inc | 11 ++++++++++- .../common_instances.hpp | 4 ++-- 22 files changed, 74 insertions(+), 65 deletions(-) diff --git a/example/01_gemm/gemm_xdl_fp16.cpp b/example/01_gemm/gemm_xdl_fp16.cpp index 414683ffdf..66a0d98238 100644 --- a/example/01_gemm/gemm_xdl_fp16.cpp +++ b/example/01_gemm/gemm_xdl_fp16.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #include "common.hpp" @@ -37,7 +37,7 @@ using DeviceGemmInstance1 = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffl // ######| | | | Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| // ######| | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| // ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | - < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementOp, BElementOp, CElementOp, GemmDefault, 1, 256, 256, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 2, S<1, 16, 1, 16>, 8, ck::LoopScheduler::Interwave, ck::PipelineVersion::v1>; + < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementOp, BElementOp, CElementOp, GemmDefault, 1, 256, 256, 128, 32, 8, 2, 16, 16, 8, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 2, S<1, 16, 1, 16>, 4, ck::LoopScheduler::Interwave, ck::PipelineVersion::v1>; // clang-format on using DeviceGemmInstance = DeviceGemmInstance1; diff --git a/example/01_gemm/gemm_xdl_fp16_v2.cpp b/example/01_gemm/gemm_xdl_fp16_v2.cpp index ecd3b7be5d..59c059d014 100644 --- a/example/01_gemm/gemm_xdl_fp16_v2.cpp +++ b/example/01_gemm/gemm_xdl_fp16_v2.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #include "common.hpp" @@ -33,13 +33,13 @@ using DeviceGemmInstance = 2, 256, 256, 256, 32, 8, 4, - 32, 32, - 4, 4, + 16, 16, + 8, 8, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, - 1, 1, S<1, 32, 1, 8>, 8, + 1, 1, S<1, 32, 1, 8>, 4, ck::LoopScheduler::Default, ck::PipelineVersion::v1>; // clang-format on diff --git a/example/01_gemm/gemm_xdl_wavelet_fp16.cpp b/example/01_gemm/gemm_xdl_wavelet_fp16.cpp index d8672f6a0c..76a30657f0 100644 --- a/example/01_gemm/gemm_xdl_wavelet_fp16.cpp +++ b/example/01_gemm/gemm_xdl_wavelet_fp16.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #include "common.hpp" @@ -29,7 +29,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_WaveletM // ######| | | | Type| Type| Type| DataType| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| ThreadGroupSize| ThreadGroupSize| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| // ######| | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| // ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | - < ALayout, BLayout, CLayout, ADataType, BDataType, AccDataType, F16, CDataType, AElementOp, BElementOp, CElementOp, GemmDefault, 1, 256, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1,8>, 8>; + < ALayout, BLayout, CLayout, ADataType, BDataType, AccDataType, F16, CDataType, AElementOp, BElementOp, CElementOp, GemmDefault, 1, 256, 256, 256, 128, 32, 8, 8, 16, 16, 8, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1,8>, 4>; // clang-format on using DeviceGemmInstance = DeviceGemmInstance; diff --git a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_bf16.cpp b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_bf16.cpp index e630f67837..4e98bf3034 100644 --- a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_bf16.cpp +++ b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_bf16.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #include "common.hpp" @@ -32,7 +32,7 @@ using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultipleD_Xdl_C //######| | | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| //######| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| //######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | - < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>; + < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 16, 16, 8, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 4>; // clang-format on using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>; + < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 16, 16, 8, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 4>; // clang-format on using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, 1, 1, S<1, 32, 1, 8>, 4>; + < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 16, 16, 8, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, 1, 1, S<1, 32, 1, 8>, 2>; // clang-format on using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm #include @@ -54,7 +54,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemm_Xdl //######| | | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| //######| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| //######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | - < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>; + < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 16, 16, 8, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 4>; // clang-format on #include "run_grouped_gemm_example.inc" diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp index 90a12bc1dd..85ea8c2f2c 100644 --- a/example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp +++ b/example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #include #include @@ -54,7 +54,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemm_Xdl //######| | | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| //######| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| //######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | - < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>; + < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 256, 256, 128, 32, 8, 8, 16, 16, 8, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 4>; // clang-format on #include "run_grouped_gemm_example.inc" diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp index 28b0fcd0ce..fb047ae364 100644 --- a/example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp +++ b/example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #include #include @@ -54,7 +54,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemm_Xdl //######| | | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| //######| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| //######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | - < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 256, 256, 128, 16, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, 1, 1, S<1, 32, 1, 8>, 4>; + < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 256, 256, 128, 16, 4, 4, 16, 16, 8, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, 1, 1, S<1, 32, 1, 8>, 2>; // clang-format on #include "run_grouped_gemm_example.inc" diff --git a/example/22_cgemm/cgemm_xdl_bf16.cpp b/example/22_cgemm/cgemm_xdl_bf16.cpp index fa4482a984..716d36b487 100644 --- a/example/22_cgemm/cgemm_xdl_bf16.cpp +++ b/example/22_cgemm/cgemm_xdl_bf16.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #include @@ -48,10 +48,10 @@ using DeviceCGemmInstance = ck::tensor_operation::device::DeviceCGemm_4Gemm_Xdl_ 32, // index_t KPerBlock 8, // index_t AK1 8, // index_t BK1 - 32, // index_t MPerXDL - 32, // index_t NPerXDL - 4, // index_t MXdlPerWave - 2, // index_t NXdlPerWave + 16, // index_t MPerXDL + 16, // index_t NPerXDL + 8, // index_t MXdlPerWave + 4, // index_t NXdlPerWave S<4, 64, 1>, // typename ABlockTransferThreadClusterLengths_AK0_M_AK1 S<1, 0, 2>, // typename ABlockTransferThreadClusterArrangeOrder S<1, 0, 2>, // typename ABlockTransferSrcAccessOrder @@ -69,7 +69,7 @@ using DeviceCGemmInstance = ck::tensor_operation::device::DeviceCGemm_4Gemm_Xdl_ 1, // index_t CShuffleMXdlPerWavePerShuffle 1, // index_t CShuffleNXdlPerWavePerShuffle S<1, 32, 1, 8>, // typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock - 8>; // index_t CShuffleBlockTransferScalarPerVector_NPerBlock + 4>; // index_t CShuffleBlockTransferScalarPerVector_NPerBlock // clang-format on int main(int argc, char* argv[]) diff --git a/example/22_cgemm/cgemm_xdl_fp16.cpp b/example/22_cgemm/cgemm_xdl_fp16.cpp index 89a581e865..2996d87b28 100644 --- a/example/22_cgemm/cgemm_xdl_fp16.cpp +++ b/example/22_cgemm/cgemm_xdl_fp16.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #include @@ -47,10 +47,10 @@ using DeviceCGemmInstance = ck::tensor_operation::device::DeviceCGemm_4Gemm_Xdl_ 32, // index_t KPerBlock 8, // index_t AK1 8, // index_t BK1 - 32, // index_t MPerXDL - 32, // index_t NPerXDL - 4, // index_t MXdlPerWave - 2, // index_t NXdlPerWave + 16, // index_t MPerXDL + 16, // index_t NPerXDL + 8, // index_t MXdlPerWave + 4, // index_t NXdlPerWave S<4, 64, 1>, // typename ABlockTransferThreadClusterLengths_AK0_M_AK1 S<1, 0, 2>, // typename ABlockTransferThreadClusterArrangeOrder S<1, 0, 2>, // typename ABlockTransferSrcAccessOrder @@ -68,7 +68,7 @@ using DeviceCGemmInstance = ck::tensor_operation::device::DeviceCGemm_4Gemm_Xdl_ 1, // index_t CShuffleMXdlPerWavePerShuffle 1, // index_t CShuffleNXdlPerWavePerShuffle S<1, 32, 1, 8>, // typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock - 8>; // index_t CShuffleBlockTransferScalarPerVector_NPerBlock + 4>; // index_t CShuffleBlockTransferScalarPerVector_NPerBlock // clang-format on int main(int argc, char* argv[]) diff --git a/example/28_grouped_gemm_bias_e_permute/grouped_gemm_bias_e_permute_xdl_fp16.cpp b/example/28_grouped_gemm_bias_e_permute/grouped_gemm_bias_e_permute_xdl_fp16.cpp index 0bbbcb83aa..70c4a01185 100644 --- a/example/28_grouped_gemm_bias_e_permute/grouped_gemm_bias_e_permute_xdl_fp16.cpp +++ b/example/28_grouped_gemm_bias_e_permute/grouped_gemm_bias_e_permute_xdl_fp16.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #include #include diff --git a/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_xdl_fp16.cpp b/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_xdl_fp16.cpp index 427b397988..2b9afc342e 100644 --- a/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_xdl_fp16.cpp +++ b/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_xdl_fp16.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #include #include @@ -57,7 +57,7 @@ using DeviceOpInstanceKKNN = ck::tensor_operation::device:: //############################################| | | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Spacialization| Spacialization| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| //############################################| | | | | | | | | | | Operation| Operation| Operation| | | | | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| //############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | - DeviceBatchedContractionMultipleD_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, F16, F16, F32, F16, DsDataType, F16, AElementOp, BElementOp, CDEElementOp, GemmSpec, ABSpec, ABSpec, DESpec, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8>; + DeviceBatchedContractionMultipleD_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, F16, F16, F32, F16, DsDataType, F16, AElementOp, BElementOp, CDEElementOp, GemmSpec, ABSpec, ABSpec, DESpec, 1, 256, 256, 128, 32, 8, 8, 16, 16, 8, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 4>; // clang-format on using DeviceOpInstance = DeviceOpInstanceKKNN; diff --git a/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp b/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp index 5794924294..7738a6b6d4 100644 --- a/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp +++ b/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. /* Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o @@ -100,11 +100,11 @@ using DeviceGemmInstance = 8, // AK1 8, // BK1 2, // B1K1 - 32, // MPerXDL - 32, // NPerXDL - 1, // MXdlPerWave - 4, // NXdlPerWave - 2, // Gemm1NXdlPerWave + 16, // MPerXDL + 16, // NPerXDL + 2, // MXdlPerWave + 8, // NXdlPerWave + 4, // Gemm1NXdlPerWave S<4, 64, 1>, // ABlockTransfer S<1, 0, 2>, S<1, 0, 2>, @@ -129,7 +129,7 @@ using DeviceGemmInstance = 1, // CShuffleMXdlPerWavePerShuffle 2, // CShuffleNXdlPerWavePerShuffle S<1, 32, 1, 8>, // CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock - 8, // CShuffleBlockTransferScalarPerVector_NPerBlock + 4, // CShuffleBlockTransferScalarPerVector_NPerBlock MaskingSpec>; // MaskingSpecialization // Ref Gemm0: fp16 in, fp32 out diff --git a/example/35_splitK_gemm/gemm_xdl_splitk_reduce_bf16.cpp b/example/35_splitK_gemm/gemm_xdl_splitk_reduce_bf16.cpp index 7ceb1d09ef..1843198933 100644 --- a/example/35_splitK_gemm/gemm_xdl_splitk_reduce_bf16.cpp +++ b/example/35_splitK_gemm/gemm_xdl_splitk_reduce_bf16.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #include "common.hpp" @@ -35,13 +35,13 @@ using DeviceGemmV2Instance = 256, 128, 128, 64, 8, 4, - 32, 32, - 2, 2, + 16, 16, + 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, - 1, 1, S<1, 32, 1, 8>, 8, + 1, 1, S<1, 32, 1, 8>, 4, ck::BlockGemmPipelineScheduler::Intrawave,ck::BlockGemmPipelineVersion::v3>; // clang-format on diff --git a/example/35_splitK_gemm/gemm_xdl_splitk_reduce_bf16A_i8B.cpp b/example/35_splitK_gemm/gemm_xdl_splitk_reduce_bf16A_i8B.cpp index b5aeff65d6..1e4398b9f6 100644 --- a/example/35_splitK_gemm/gemm_xdl_splitk_reduce_bf16A_i8B.cpp +++ b/example/35_splitK_gemm/gemm_xdl_splitk_reduce_bf16A_i8B.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #include "common.hpp" @@ -35,13 +35,13 @@ using DeviceGemmV2Instance = 256, 128, 128, 64, 8, 4, - 32, 32, - 2, 2, + 16, 16, + 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, - 1, 1, S<1, 32, 1, 8>, 8, + 1, 1, S<1, 32, 1, 8>, 4, ck::BlockGemmPipelineScheduler::Intrawave,ck::BlockGemmPipelineVersion::v3, ReduceDataType>; // clang-format on diff --git a/example/35_splitK_gemm/gemm_xdl_splitk_reduce_multi_d_bf16.cpp b/example/35_splitK_gemm/gemm_xdl_splitk_reduce_multi_d_bf16.cpp index cb84f2a416..d5acde139a 100644 --- a/example/35_splitK_gemm/gemm_xdl_splitk_reduce_multi_d_bf16.cpp +++ b/example/35_splitK_gemm/gemm_xdl_splitk_reduce_multi_d_bf16.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #include "common.hpp" @@ -35,13 +35,13 @@ using DeviceGemmV2Instance = 256, 128, 128, 64, 8, 4, - 32, 32, - 2, 2, + 16, 16, + 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, - 1, 1, S<1, 32, 1, 8>, 8, + 1, 1, S<1, 32, 1, 8>, 4, ck::BlockGemmPipelineScheduler::Intrawave,ck::BlockGemmPipelineVersion::v3, ReduceDataType>; // clang-format on diff --git a/example/35_splitK_gemm/gemm_xdl_splitk_reduce_multi_d_fp16.cpp b/example/35_splitK_gemm/gemm_xdl_splitk_reduce_multi_d_fp16.cpp index 2ab8f77dc4..bb3c23f060 100644 --- a/example/35_splitK_gemm/gemm_xdl_splitk_reduce_multi_d_fp16.cpp +++ b/example/35_splitK_gemm/gemm_xdl_splitk_reduce_multi_d_fp16.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #include "common.hpp" @@ -35,13 +35,13 @@ using DeviceGemmV2Instance = 256, 128, 128, 64, 8, 4, - 32, 32, - 2, 2, + 16, 16, + 4, 4, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, - 1, 1, S<1, 32, 1, 8>, 8, + 1, 1, S<1, 32, 1, 8>, 4, ck::BlockGemmPipelineScheduler::Intrawave,ck::BlockGemmPipelineVersion::v2, ReduceDataType>; // clang-format on diff --git a/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp b/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp index 46843a4de8..32a7e4a76e 100644 --- a/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp +++ b/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #include #include @@ -56,7 +56,7 @@ using DeviceOpInstanceKKNN = ck::tensor_operation::device:: //############################################| | | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Spacialization| Spacialization| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| //############################################| | | | | | | | | | | Operation| Operation| Operation| | | | | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| //############################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | - DeviceSplitKContractionMultipleD_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmSpec, ABSpec, ABSpec, DESpec, 1, 256, 256, 128, 32, 4, 4, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 4, 4, 1, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 4, 4, 1, 1, 1, S<1, 32, 1, 4>, 4>; + DeviceSplitKContractionMultipleD_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmSpec, ABSpec, ABSpec, DESpec, 1, 256, 256, 128, 32, 4, 4, 16, 16, 8, 4, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 4, 4, 1, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 4, 4, 1, 1, 1, S<1, 32, 1, 4>, 2>; // clang-format on using DeviceOpInstance = DeviceOpInstanceKKNN; diff --git a/example/46_gemm_add_multiply/gemm_add_multiply_xdl_fp16.cpp b/example/46_gemm_add_multiply/gemm_add_multiply_xdl_fp16.cpp index 56417b101d..4d73f0c35f 100644 --- a/example/46_gemm_add_multiply/gemm_add_multiply_xdl_fp16.cpp +++ b/example/46_gemm_add_multiply/gemm_add_multiply_xdl_fp16.cpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #include "common.hpp" #include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp" @@ -31,7 +31,7 @@ using DeviceOpInstance = ck::tensor_operation::device:: //##############################| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| //##############################| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| //##############################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | - DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, DsLayout, Row, F16, F16, F32, F16, DsDataType, F16, PassThrough, PassThrough, CDEElementOp, GemmDefault, 1, 128, 128, 128, 32, 8, 2, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 8>; + DeviceGemmMultipleD_Xdl_CShuffle< Row, Row, DsLayout, Row, F16, F16, F32, F16, DsDataType, F16, PassThrough, PassThrough, CDEElementOp, GemmDefault, 1, 128, 128, 128, 32, 8, 2, 16, 16, 8, 4, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, 0, 1, 1, S<1, 16, 1, 8>, 4>; // clang-format on using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm, ck::half_t> && + std::is_same_v, ck::half_t>) + { + return ck::utils::check_err( + e_m_n_device_result, e_m_n_host_result, "Error: Incorrect results!", 5e-3, 1e-3); + } + else + { + return ck::utils::check_err(e_m_n_device_result, e_m_n_host_result); + } } return true; diff --git a/example/66_complex_contraction_bilinear/common_instances.hpp b/example/66_complex_contraction_bilinear/common_instances.hpp index 480ca5a0af..ed1c1dc303 100644 --- a/example/66_complex_contraction_bilinear/common_instances.hpp +++ b/example/66_complex_contraction_bilinear/common_instances.hpp @@ -1,5 +1,5 @@ // SPDX-License-Identifier: MIT -// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. +// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved. #pragma once @@ -37,7 +37,7 @@ using DeviceOpInstanceKK_Generic = ck::tensor_operation::device:: //#####################################| | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| Data| //#####################################| | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| Type| //#####################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | - DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmSpec, 1, 256, 256, 128, 16, 4, 4, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, 1, 1, S<1, 16, 1, 16>, 4, ComputeDataType>; + DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmSpec, 1, 256, 256, 128, 16, 4, 4, 16, 16, 8, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, 1, 1, S<1, 16, 1, 16>, 2, ComputeDataType>; // clang-format on template