ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-12 17:26:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

Ensure MX GEMM Instances can be Cross-Compiled for Multiple Architectures (#2171)

Browse Source 
* Re-enable MX GEMM instances

* Fix compilation error when building MX GEMM for multiple architectures
This commit is contained in:
Andriy Roshchenko
2025-05-08 13:26:03 -06:00
committed by GitHub
parent c757046d49
commit cb27e7c77f
7 changed files with 11 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
library/src/tensor_operation_instance/gpu/CMakeLists.txt
View File

@@ -115,9 +115,12 @@ function(add_instance_library INSTANCE_NAME)
list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack+ gfx908:xnack- gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx942 gfx1030 gfx950)
elseif(source MATCHES "mha")
list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx908 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1200 gfx1201 gfx10.3-generic gfx11-generic gfx12-generic)
elseif(source MATCHES "_mx")
endif()
if(source MATCHES "_mx")
list(REMOVE_ITEM INST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack- gfx908:xnack+ gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx942 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152 gfx1200 gfx1201 gfx10.3-generic gfx11-generic gfx12-generic)
endif()
#only build the fp8 gemm instances for gfx90a if the build argument is set, otherwise only build for gfx942/gfx950
if(NOT CK_USE_FP8_ON_UNSUPPORTED_ARCH)
if(source MATCHES "gemm_xdl_universal" AND source MATCHES "f8")

4
library/src/tensor_operation_instance/gpu/gemm_mx/device_gemm_mx_xdl_bf8_f8_f16/device_gemm_mx_xdl_bf8_f8_f16_mk_kn_mn.hpp
View File

@@ -40,18 +40,16 @@ static constexpr auto ScaleBlockSize = 32;
template <BlockGemmPipelineScheduler BlkGemmPipeSched, GemmSpecialization GemmSpec>
using device_gemm_mx_xdl_bf8_f8_f16_mk_kn_mn_instances = std::tuple<
// clang-format off
// clang-format off
//#########################| ALayout| BLayout| CLayout|AData|AScale|BData|BScale| CData| AccData| Cshuffle| A| B| C| GEMM| Scale Block| Block| MPer| NPer| KPer| AK1| BK1|MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| Block-wiseGemm| Block-wiseGemm|
//#########################| | | | Type| Data| Type| Data| Type| Type| Type| Elementwise| Elementwise| Elementwise|Specialization| Size| 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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| Pipeline| Pipeline|
//#########################| | | | | Type| | Type| | | | Operation| Operation| Operation| | | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision|
//#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
#if defined(__gfx950__)
DeviceGemmMX_Xdl_CShuffleV3< Row, Row, Row, BF8, E8M0, F8, E8M0, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 128, 64, 16, 128, 16, 4, 16, 16, 2, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<32, 4, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
DeviceGemmMX_Xdl_CShuffleV3< Row, Row, Row, BF8, E8M0, F8, E8M0, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 256, 256, 256, 256, 16, 4, 32, 32, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, false, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, false, 1, 1, S<1, 32, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
DeviceGemmMX_Xdl_CShuffleV3< Row, Row, Row, BF8, E8M0, F8, E8M0, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 256, 64, 64, 256, 16, 4, 32, 32, 1, 1, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<32, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
DeviceGemmMX_Xdl_CShuffleV3< Row, Row, Row, BF8, E8M0, F8, E8M0, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 256, 128, 128, 128, 16, 4, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<32, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, 0, 1, 1, S<1, 32, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
DeviceGemmMX_Xdl_CShuffleV3< Row, Row, Row, BF8, E8M0, F8, E8M0, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 128, 16, 32, 512, 16, 8, 16, 16, 1, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<64, 2, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 8, 0, 1, 1, S<1, 16, 1, 8>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>
#endif
// clang-format on
>;

4
library/src/tensor_operation_instance/gpu/gemm_mx/device_gemm_mx_xdl_f8_f8_bf16/device_gemm_mx_xdl_f8_f8_bf16_km_nk_mn.hpp
View File

@@ -39,12 +39,11 @@ static constexpr auto ScaleBlockSize = 32;
template <BlockGemmPipelineScheduler BlkGemmPipeSched, GemmSpecialization GemmSpec>
using device_gemm_mx_xdl_f8_f8_bf16_km_nk_mn_instances = std::tuple<
// clang-format off
// clang-format off
//#########################| ALayout| BLayout| CLayout|AData|AScale|BData|BScale| CData| AccData| Cshuffle| A| B| C| GEMM| Scale Block| Block| MPer| NPer| KPer| AK1| BK1|MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| Block-wiseGemm| Block-wiseGemm|
//#########################| | | | Type| Data| Type| Data| Type| Type| Type| Elementwise| Elementwise| Elementwise|Specialization| Size| 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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| Pipeline| Pipeline|
//#########################| | | | | Type| | Type| | | | Operation| Operation| Operation| | | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision|
//#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
#if defined(__gfx950__)
DeviceGemmMX_Xdl_CShuffleV3< Col, Col, Row, F8, E8M0, F8, E8M0, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 256, 128, 128, 128, 4, 16, 32, 32, 2, 2, S<32, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 32, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
DeviceGemmMX_Xdl_CShuffleV3< Col, Col, Row, F8, E8M0, F8, E8M0, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 256, 16, 256, 128, 4, 16, 16, 16, 1, 4, S<32, 4, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 16, 1, 16>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
DeviceGemmMX_Xdl_CShuffleV3< Col, Col, Row, F8, E8M0, F8, E8M0, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 256, 128, 128, 64, 4, 16, 32, 32, 2, 2, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 8, 4, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 32, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
@@ -52,7 +51,6 @@ using device_gemm_mx_xdl_f8_f8_bf16_km_nk_mn_instances = std::tuple<
DeviceGemmMX_Xdl_CShuffleV3< Col, Col, Row, F8, E8M0, F8, E8M0, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 256, 256, 256, 128, 8, 16, 16, 16, 8, 8, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 2, S<1, 32, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
DeviceGemmMX_Xdl_CShuffleV3< Col, Col, Row, F8, E8M0, F8, E8M0, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 256, 256, 256, 64, 4, 16, 32, 32, 4, 4, S<16, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 32, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
DeviceGemmMX_Xdl_CShuffleV3< Col, Col, Row, F8, E8M0, F8, E8M0, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 128, 128, 128, 128, 4, 16, 16, 16, 4, 8, S<32, 4, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 16, 4, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>
#endif
// clang-format on
>;

7
library/src/tensor_operation_instance/gpu/gemm_mx/device_gemm_mx_xdl_f8_f8_bf16/device_gemm_mx_xdl_f8_f8_bf16_mk_nk_mn.hpp
View File

@@ -39,21 +39,16 @@ static constexpr auto ScaleBlockSize = 32;
template <BlockGemmPipelineScheduler BlkGemmPipeSched, GemmSpecialization GemmSpec>
using device_gemm_mx_xdl_f8_f8_bf16_mk_nk_mn_instances = std::tuple<
// clang-format off
// clang-format off
//#########################| ALayout| BLayout| CLayout|AData|AScale|BData|BScale| CData| AccData| Cshuffle| A| B| C| GEMM| Scale Block| Block| MPer| NPer| KPer| AK1| BK1|MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| Block-wiseGemm| Block-wiseGemm|
//#########################| | | | Type| Data| Type| Data| Type| Type| Type| Elementwise| Elementwise| Elementwise|Specialization| Size| 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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| Pipeline| Pipeline|
//#########################| | | | | Type| | Type| | | | Operation| Operation| Operation| | | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision|
//#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
#if defined(__gfx950__)
DeviceGemmMX_Xdl_CShuffleV3< Row, Col, Row, F8, E8M0, F8, E8M0, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 128, 128, 16, 128, 16, 16, 16, 16, 4, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
DeviceGemmMX_Xdl_CShuffleV3< Row, Col, Row, F8, E8M0, F8, E8M0, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 256, 128, 128, 256, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, false, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, false, 1, 1, S<1, 32, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
DeviceGemmMX_Xdl_CShuffleV3< Row, Col, Row, F8, E8M0, F8, E8M0, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 256, 256, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 32, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
DeviceGemmMX_Xdl_CShuffleV3< Row, Col, Row, F8, E8M0, F8, E8M0, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 256, 128, 128, 128, 16, 16, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 32, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
DeviceGemmMX_Xdl_CShuffleV3< Row, Col, Row, F8, E8M0, F8, E8M0, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 64, 16, 16, 512, 16, 16, 16, 16, 1, 1, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>
//Require verification
//DeviceGemmMX_Xdl_CShuffleV3< Row, Col, Row, F8, E8M0, F8, E8M0, BF16, F32, BF16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 256, 256, 256, 128, 16, 16, 16, 16, 8, 8, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 2, S<1, 32, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>
#endif
// clang-format on
>;

7
library/src/tensor_operation_instance/gpu/gemm_mx/device_gemm_mx_xdl_f8_f8_f16/device_gemm_mx_xdl_f8_f8_f16_mk_nk_mn.hpp
View File

@@ -39,21 +39,16 @@ static constexpr auto ScaleBlockSize = 32;
template <BlockGemmPipelineScheduler BlkGemmPipeSched, GemmSpecialization GemmSpec>
using device_gemm_mx_xdl_f8_f8_f16_mk_nk_mn_instances = std::tuple<
// clang-format off
// clang-format off
//#########################| ALayout| BLayout| CLayout|AData|AScale|BData|BScale| CData| AccData| Cshuffle| A| B| C| GEMM| Scale Block| Block| MPer| NPer| KPer| AK1| BK1|MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| Block-wiseGemm| Block-wiseGemm|
//#########################| | | | Type| Data| Type| Data| Type| Type| Type| Elementwise| Elementwise| Elementwise|Specialization| Size| 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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector| Pipeline| Pipeline|
//#########################| | | | | Type| | Type| | | | Operation| Operation| Operation| | | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl| Scheduler| Verision|
//#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
#if defined(__gfx950__)
DeviceGemmMX_Xdl_CShuffleV3< Row, Col, Row, F8, E8M0, F8, E8M0, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 128, 128, 16, 128, 16, 16, 16, 16, 4, 1, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 16, 1, 8>, 2, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
DeviceGemmMX_Xdl_CShuffleV3< Row, Col, Row, F8, E8M0, F8, E8M0, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 256, 128, 128, 256, 16, 16, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, false, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, false, 1, 1, S<1, 32, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
DeviceGemmMX_Xdl_CShuffleV3< Row, Col, Row, F8, E8M0, F8, E8M0, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 256, 256, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 32, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
DeviceGemmMX_Xdl_CShuffleV3< Row, Col, Row, F8, E8M0, F8, E8M0, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 256, 128, 128, 128, 16, 16, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 32, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
DeviceGemmMX_Xdl_CShuffleV3< Row, Col, Row, F8, E8M0, F8, E8M0, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 64, 16, 16, 512, 16, 16, 16, 16, 1, 1, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<8, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 1, S<1, 16, 1, 4>, 4, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>
//Require verification
//DeviceGemmMX_Xdl_CShuffleV3< Row, Col, Row, F8, E8M0, F8, E8M0, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmSpec, ScaleBlockSize, 256, 256, 256, 128, 16, 16, 16, 16, 8, 8, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, 1, 2, S<1, 32, 1, 8>, 8, BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
#endif
// clang-format on
>;