From ed51c0638b595dc1a09b8da27315ead7062edabf Mon Sep 17 00:00:00 2001 From: "Po-Yen, Chen" Date: Fri, 19 Aug 2022 14:31:46 -0400 Subject: [PATCH] Re-format template argument in example code --- example/01_gemm/gemm_dl_fp16.cpp | 15 +++---- example/01_gemm/gemm_dl_fp32.cpp | 14 +++---- example/01_gemm/gemm_dl_int8.cpp | 12 +++--- example/01_gemm/gemm_xdl_bf16.cpp | 67 ++++++------------------------- example/01_gemm/gemm_xdl_fp16.cpp | 17 ++++---- example/01_gemm/gemm_xdl_fp64.cpp | 14 +++---- example/01_gemm/gemm_xdl_int8.cpp | 57 ++++---------------------- 7 files changed, 51 insertions(+), 145 deletions(-) diff --git a/example/01_gemm/gemm_dl_fp16.cpp b/example/01_gemm/gemm_dl_fp16.cpp index 89f37b02c1..03be1880f3 100644 --- a/example/01_gemm/gemm_dl_fp16.cpp +++ b/example/01_gemm/gemm_dl_fp16.cpp @@ -5,9 +5,6 @@ #include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp" -using F16 = ck::half_t; -using F32 = float; - using ADataType = ck::half_t; using BDataType = ck::half_t; using CDataType = ck::half_t; @@ -24,12 +21,12 @@ using CElementOp = PassThrough; static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; // clang-format off -using DeviceGemmInstance = ck::tensor_operation::device:: - // ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer| - // ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector| - // ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | | - // ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | - DeviceGemmDl< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 16, 2, 4, 4, 1, S<8, 2>, S<8, 2>, S<2, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<2, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>; +using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl +// ######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer| +// ######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector| +// ######| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | | +// ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | + < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmDefault, 256, 128, 128, 16, 2, 4, 4, 1, S<8, 2>, S<8, 2>, S<2, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<2, 1, 4, 2>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>; // clang-format on using ReferenceGemmInstance = ck::tensor_operation::host:: diff --git a/example/01_gemm/gemm_dl_fp32.cpp b/example/01_gemm/gemm_dl_fp32.cpp index 5e5821ebac..b217011401 100644 --- a/example/01_gemm/gemm_dl_fp32.cpp +++ b/example/01_gemm/gemm_dl_fp32.cpp @@ -5,8 +5,6 @@ #include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp" -using F32 = float; - using ADataType = float; using BDataType = float; using CDataType = float; @@ -23,12 +21,12 @@ using CElementOp = PassThrough; static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; // clang-format off -using DeviceGemmInstance = ck::tensor_operation::device:: - // ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer| - // ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector| - // ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | | - // ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | - DeviceGemmDl< F32, F32, F32, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 16, 1, 4, 4, 1, S<8, 2>, S<8, 2>, S<2, 1, 4, 1>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<2, 1, 4, 1>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 1, 2, 3, 4, 5>, 5, 4>; +using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl +// ######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer| +// ######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector| +// ######| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | | +// ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | + < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmDefault, 256, 128, 128, 16, 1, 4, 4, 1, S<8, 2>, S<8, 2>, S<2, 1, 4, 1>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<2, 1, 4, 1>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 1, 2, 3, 4, 5>, 5, 4>; // clang-format on using ReferenceGemmInstance = ck::tensor_operation::host:: diff --git a/example/01_gemm/gemm_dl_int8.cpp b/example/01_gemm/gemm_dl_int8.cpp index df53110d17..a867cf3b67 100644 --- a/example/01_gemm/gemm_dl_int8.cpp +++ b/example/01_gemm/gemm_dl_int8.cpp @@ -21,12 +21,12 @@ using CElementOp = PassThrough; static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; // clang-format off -using DeviceGemmInstance = ck::tensor_operation::device:: - // #########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer| - // #########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector| - // #########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | | - // #########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | - DeviceGemmDl< int8_t, int8_t, int8_t, int32_t, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 16, 4, 4, 4, 1, S<8, 2>, S<8, 2>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 4>; +using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl +// ######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| M11N11Thread| M11N11Thread| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer| +// ######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector| +// ######| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | | +// ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | + < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmDefault, 256, 128, 128, 16, 4, 4, 4, 1, S<8, 2>, S<8, 2>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<2, 1, 4, 4>, S<8, 1, 32, 1>, S<0, 3, 1, 2>, S<0, 3, 1, 2>, S<1, 1, 4, 1>, S<0, 3, 1, 2>, S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>, 5, 4>; // clang-format on using ReferenceGemmInstance = ck::tensor_operation::host:: diff --git a/example/01_gemm/gemm_xdl_bf16.cpp b/example/01_gemm/gemm_xdl_bf16.cpp index f344c77719..9ac5012087 100644 --- a/example/01_gemm/gemm_xdl_bf16.cpp +++ b/example/01_gemm/gemm_xdl_bf16.cpp @@ -5,13 +5,11 @@ #include "ck/tensor_operation/gpu/device/device_gemm_xdl_cshuffle.hpp" -using BF16 = ck::bhalf_t; -using F32 = float; - -using ADataType = BF16; -using BDataType = BF16; -using CDataType = BF16; -using AccDataType = F32; +using ADataType = ck::bhalf_t; +using BDataType = ck::bhalf_t; +using CDataType = ck::bhalf_t; +using AccDataType = float; +using CShuffleDataType = ck::bhalf_t; using ALayout = Row; using BLayout = Col; @@ -25,56 +23,15 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa // clang-format off using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle - , // typename ABlockTransferThreadClusterLengths_AK0_M_AK1 - S<1, 0, 2>, // typename ABlockTransferThreadClusterArrangeOrder - S<1, 0, 2>, // typename ABlockTransferSrcAccessOrder - 2, // index_t ABlockTransferSrcVectorDim - 8, // index_t ABlockTransferSrcScalarPerVector - 8, // index_t ABlockTransferDstScalarPerVector_AK1 - 1, // index_t ABlockLdsExtraM - S<4, 64, 1>, // typename BBlockTransferThreadClusterLengths_BK0_N_BK1 - S<1, 0, 2>, // typename BBlockTransferThreadClusterArrangeOrder - S<1, 0, 2>, // typename BBlockTransferSrcAccessOrder - 2, // index_t BBlockTransferSrcVectorDim - 8, // index_t BBlockTransferSrcScalarPerVector - 8, // index_t BBlockTransferDstScalarPerVector_BK1 - 1, // index_t BBlockLdsExtraN - 1, // index_t CShuffleMXdlPerWavePerShuffle - 1, // index_t CShuffleNXdlPerWavePerShuffle - S<1, 32, 1, 8>, // typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock - 8>; // index_t CShuffleBlockTransferScalarPerVector_NPerBlock +//######| ALayout| BLayout| CLayout| AData| BData| CData| AccData| CShuffle| A| B| C| GEMM| NumGemmK| 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| +//######| | | | 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, 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>; // clang-format on -using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm; +using ReferenceGemmInstance = ck::tensor_operation::host:: + ReferenceGemm; #include "run_gemm_example.inc" diff --git a/example/01_gemm/gemm_xdl_fp16.cpp b/example/01_gemm/gemm_xdl_fp16.cpp index 9e8e2dc97f..68f02ce0ba 100644 --- a/example/01_gemm/gemm_xdl_fp16.cpp +++ b/example/01_gemm/gemm_xdl_fp16.cpp @@ -6,14 +6,11 @@ #include "ck/tensor_operation/gpu/device/device_gemm_xdl.hpp" #include "ck/tensor_operation/gpu/device/device_gemm_xdl_cshuffle.hpp" -using F16 = ck::half_t; -using F32 = float; - -using ADataType = F16; -using BDataType = F16; -using AccDataType = F32; -using CShuffleDataType = F32; -using CDataType = F16; +using ADataType = ck::half_t; +using BDataType = ck::half_t; +using AccDataType = float; +using CShuffleDataType = float; +using CDataType = ck::half_t; using ALayout = Row; using BLayout = Col; @@ -25,8 +22,8 @@ using CElementOp = PassThrough; static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; +// clang-format off using DeviceGemmInstance0 = ck::tensor_operation::device::DeviceGemmXdl - // clang-format off //######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| //######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise|Spacialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar| //######| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| @@ -34,8 +31,8 @@ using DeviceGemmInstance0 = ck::tensor_operation::device::DeviceGemmXdl < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 7, 1>; // clang-format on +// clang-format off using DeviceGemmInstance1 = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle - // clang-format off //######| ALayout| BLayout| CLayout| AData| BData| CData| AccData| CShuffle| A| B| C| GEMM| NumGemmK| 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| //######| | | | 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| diff --git a/example/01_gemm/gemm_xdl_fp64.cpp b/example/01_gemm/gemm_xdl_fp64.cpp index bf37108dcb..6ef851c4f9 100644 --- a/example/01_gemm/gemm_xdl_fp64.cpp +++ b/example/01_gemm/gemm_xdl_fp64.cpp @@ -5,8 +5,6 @@ #include "ck/tensor_operation/gpu/device/device_gemm_xdl.hpp" -using F64 = double; - using ADataType = double; using BDataType = double; using CDataType = double; @@ -24,14 +22,14 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa // clang-format off using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl -//##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| -//##########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise|Spacialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar| -//##########| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| -//##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | +//######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| +//######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise|Spacialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar| +//######| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector| +//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | #if 0 - < F64, F64, F64, F64, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 32, 32, 4, 1, 16, 16, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 7, 1>; + < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmDefault, 64, 32, 32, 4, 1, 16, 16, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 1, true, 7, 1>; #else - < F64, F64, F64, F64, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 4, 2, 16, 16, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, 7, 1>; + < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmDefault, 256, 128, 128, 4, 2, 16, 16, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, 7, 1>; #endif // clang-format on diff --git a/example/01_gemm/gemm_xdl_int8.cpp b/example/01_gemm/gemm_xdl_int8.cpp index dde4f6a32a..cbe8169006 100644 --- a/example/01_gemm/gemm_xdl_int8.cpp +++ b/example/01_gemm/gemm_xdl_int8.cpp @@ -22,57 +22,16 @@ using CElementOp = PassThrough; static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; // clang-format off -using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle< - ALayout, // typename ALayout - BLayout, // typename BLayout - CLayout, // typename CLayout - ADataType, // typename ADataType - BDataType, // typename BDataType - CDataType, // typename CDataType - AccDataType, // typename GemmAccDataType - CShuffleDataType, // typename CShuffleDataType - PassThrough, // typename AElementwiseOperation - PassThrough, // typename BElementwiseOperation - PassThrough, // typename CElementwiseOperation - GemmDefault, // GemmSpecialization GemmSpec - 1, // index_t NumGemmKPrefetchStage - 256, // index_t BlockSize - 256, // index_t MPerBlock - 128, // index_t NPerBlock - 64, // index_t KPerBlock - 16, // index_t AK1 - 16, // index_t BK1 - 32, // index_t MPerXDL - 32, // index_t NPerXDL - 4, // index_t MXdlPerWave - 2, // index_t NXdlPerWave - S<4, 64, 1>, // typename ABlockTransferThreadClusterLengths_AK0_M_AK1 - S<1, 0, 2>, // typename ABlockTransferThreadClusterArrangeOrder - S<1, 0, 2>, // typename ABlockTransferSrcAccessOrder - 2, // index_t ABlockTransferSrcVectorDim - 16, // index_t ABlockTransferSrcScalarPerVector - 16, // index_t ABlockTransferDstScalarPerVector_AK1 - 1, // index_t ABlockLdsExtraM - S<4, 64, 1>, // typename BBlockTransferThreadClusterLengths_BK0_N_BK1 - S<1, 0, 2>, // typename BBlockTransferThreadClusterArrangeOrder - S<1, 0, 2>, // typename BBlockTransferSrcAccessOrder - 2, // index_t BBlockTransferSrcVectorDim - 8, // index_t BBlockTransferSrcScalarPerVector - 8, // index_t BBlockTransferDstScalarPerVector_BK1 - 1, // index_t BBlockLdsExtraN - 1, // index_t CShuffleMXdlPerWavePerShuffle - 1, // index_t CShuffleNXdlPerWavePerShuffle - S<1, 64, 1, 4>, // typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock - 16>; // index_t CShuffleBlockTransferScalarPerVector_NPerBlock +using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle +//######| ALayout| BLayout| CLayout| AData| BData| CData| AccData| CShuffle| A| B| C| GEMM| NumGemmK| 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| +//######| | | | 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, 64, 16, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 64, 1, 4>, 16>; // clang-format on -using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm; +using ReferenceGemmInstance = ck::tensor_operation::host:: + ReferenceGemm; #include "run_gemm_example.inc"