diff --git a/example/65_gemm_multiply_multiply/gemm_add_add_wmma_fp16.cpp b/example/65_gemm_multiply_multiply/gemm_add_add_wmma_fp16.cpp index 68de2ab9b6..54abab2f60 100644 --- a/example/65_gemm_multiply_multiply/gemm_add_add_wmma_fp16.cpp +++ b/example/65_gemm_multiply_multiply/gemm_add_add_wmma_fp16.cpp @@ -8,7 +8,7 @@ #include "ck/ck.hpp" #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" -#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle_v3.hpp" +#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle_v3.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp" @@ -25,7 +25,6 @@ template using S = ck::Sequence; using F16 = ck::half_t; -using FP8 = ck::f8_t; using F32 = float; using Row = ck::tensor_layout::gemm::RowMajor; @@ -71,14 +70,13 @@ using CDEElementOp = AddAdd; static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::Default; -using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultiD_Xdl_CShuffle_V3 +using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultipleD_Wmma_CShuffleV3 // clang-format off -///######| ALayout| BLayout| DsLayout| ELayout| AData| BData| DsData| EData| AccData| CShuffle| A| B| CDE| GEMM| 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| Type| DataType| 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| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| -///######| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| -///######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | S| -///###### RCR - < Row, Col, DsLayout, ELayout, A0DataType, B0DataType, DsDataType, EDataType, AccDataType, CShuffleDataType, AElementOp, BElementOp, CDEElementOp, GemmSpec, 256, 256, 128, 128, 16, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, S<8, 8, 8>, ck::BlockGemmPipelineScheduler::Interwave, ck::BlockGemmPipelineVersion::v1, FP8>; + //#########################| ALayout| BLayout| DsLayout| ELayout| AData| BData| DsData| EData| AccData| CShuffle| A| B| CDE| GemmSpec| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CShuffleBlockTransfer| CDEShuffleBlockTransfer| BlkGemm| BlkGemm| + //#########################| | | | | Type| Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| | Size| Block| Block| Block| | | Wmma| Wmma| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraN| MRepeat| NRepeat| ClusterLengths| ScalarPerVectors| PipeSched| PipelineVer| + //#########################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | Lengths_AK0_M_AK1| ArrangeOrder| | | PerVector| PerVector_AK1| | Lengths_BK0_N_BK1| ArrangeOrder| | | PerVector| PerVector_BK1| | PerShuffle| PerShuffle| _MBlock_MPerBlock| | | | + //#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | _NBlock_NPerBlock| S| | | + < A0Layout, B0Layout, DsLayout, ELayout, A0DataType, B0DataType, DsDataType, EDataType, AccDataType, CShuffleDataType, AElementOp, BElementOp, CDEElementOp, GemmSpec, 128, 128, 64, 64, 8, 8, 16, 16, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 4>, S<8, 8, 8>, ck::BlockGemmPipelineScheduler::Interwave, ck::BlockGemmPipelineVersion::v1>; // clang-format on int main(int argc, char* argv[]) @@ -184,7 +182,6 @@ int main(int argc, char* argv[]) b0_device_buf.ToDevice(b0_k_n.mData.data()); d0_device_buf.ToDevice(d0_m_n.mData.data()); d1_device_buf.ToDevice(d1_m_n.mData.data()); - e_device_buf.ToDevice(e_m_n_device_result.mData.data()); auto a_element_op = AElementOp{}; auto b_element_op = BElementOp{}; @@ -220,11 +217,12 @@ int main(int argc, char* argv[]) "not support this GEMM problem"); } - float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel, 20, 50}); + float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel, 0, 20, 50}); - std::size_t flop = std::size_t(2) * M * N * K; - std::size_t num_btype = - sizeof(A0DataType) * M * K + sizeof(B0DataType) * K * N + sizeof(EDataType) * M * N; + std::size_t flop = std::size_t(2) * M * N * K; + std::size_t num_btype = sizeof(A0DataType) * M * K + sizeof(B0DataType) * K * N + + sizeof(D0DataType) * M * N + sizeof(D1DataType) * M * N + + sizeof(EDataType) * M * N; float tflops = static_cast(flop) / 1.E9 / ave_time; @@ -233,8 +231,6 @@ int main(int argc, char* argv[]) std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s" << std::endl; - e_device_buf.FromDevice(e_m_n_device_result.mData.data()); - if(do_verification) { Tensor c_m_n({M, N}); diff --git a/example/65_gemm_multiply_multiply/gemm_add_add_xdl_fp16.cpp b/example/65_gemm_multiply_multiply/gemm_add_add_xdl_fp16.cpp index 580f38a79f..086ea45d10 100644 --- a/example/65_gemm_multiply_multiply/gemm_add_add_xdl_fp16.cpp +++ b/example/65_gemm_multiply_multiply/gemm_add_add_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 @@ -184,7 +184,6 @@ int main(int argc, char* argv[]) b0_device_buf.ToDevice(b0_k_n.mData.data()); d0_device_buf.ToDevice(d0_m_n.mData.data()); d1_device_buf.ToDevice(d1_m_n.mData.data()); - e_device_buf.ToDevice(e_m_n_device_result.mData.data()); auto a_element_op = AElementOp{}; auto b_element_op = BElementOp{}; @@ -220,11 +219,12 @@ int main(int argc, char* argv[]) "not support this GEMM problem"); } - float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel, 20, 50}); + float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel, 0, 20, 50}); - std::size_t flop = std::size_t(2) * M * N * K; - std::size_t num_btype = - sizeof(A0DataType) * M * K + sizeof(B0DataType) * K * N + sizeof(EDataType) * M * N; + std::size_t flop = std::size_t(2) * M * N * K; + std::size_t num_btype = sizeof(A0DataType) * M * K + sizeof(B0DataType) * K * N + + sizeof(D0DataType) * M * N + sizeof(D1DataType) * M * N + + sizeof(EDataType) * M * N; float tflops = static_cast(flop) / 1.E9 / ave_time; @@ -233,8 +233,6 @@ int main(int argc, char* argv[]) std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s" << std::endl; - e_device_buf.FromDevice(e_m_n_device_result.mData.data()); - if(do_verification) { Tensor c_m_n({M, N});