mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-18 03:49:41 +00:00
Code refactor (#175)
* format
* improving pipeline
* fix typo
* format
* adding thread group
* adding thread group
* adding thread group
* adding gemm pipeline
* tweak
* refactor
* refactor
* add missing type convert
* refactor
* refactor
* refactor
* clean
* fix build
* refactor
* format
* clean up
* use remove_cvref_t
* clean
* clean up
* clean up
* clean up
[ROCm/composable_kernel commit: ec7c2e912e]
This commit is contained in:
Chao Liu
@@ -11,8 +11,7 @@
|
||||
#include "host_tensor.hpp"
|
||||
#include "host_tensor_generator.hpp"
|
||||
#include "device_tensor.hpp"
|
||||
#include "device_gemm_xdl.hpp"
|
||||
#include "device_gemm_xdl_c_shuffle.hpp"
|
||||
#include "device_gemm_xdl_cshuffle.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "reference_gemm.hpp"
|
||||
#include "gemm_specialization.hpp"
|
||||
@@ -37,47 +36,51 @@ using ALayout = ck::tensor_layout::gemm::RowMajor;
|
||||
using BLayout = ck::tensor_layout::gemm::ColumnMajor;
|
||||
using CLayout = ck::tensor_layout::gemm::RowMajor;
|
||||
|
||||
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
|
||||
|
||||
// clang-format off
|
||||
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl_C_Shuffle<
|
||||
ADataType, // ADataType
|
||||
BDataType, // BDataType
|
||||
CDataType, // CDataType
|
||||
AccDataType, // AccDataType
|
||||
CDataType, // CShuffleDataType
|
||||
ALayout, // ALayout
|
||||
BLayout, // BLayout
|
||||
CLayout, // CLayout
|
||||
PassThrough, // AElementwiseOperation
|
||||
PassThrough, // BElementwiseOperation
|
||||
PassThrough, // CElementwiseOperation
|
||||
256, // BlockSize
|
||||
256, // MPerBlock
|
||||
128, // NPerBlock
|
||||
32, // KPerBlock
|
||||
8, // AK1
|
||||
8, // BK1
|
||||
32, // MPerXDL
|
||||
32, // NPerXDL
|
||||
4, // MXdlPerWave
|
||||
2, // NXdlPerWave
|
||||
S<4, 64, 1>, // ABlockTransferThreadClusterLengths_K0_M_K1
|
||||
S<1, 0, 2>, // ABlockTransferThreadClusterArrangeOrder
|
||||
S<1, 0, 2>, // ABlockTransferSrcAccessOrder
|
||||
2, // ABlockTransferSrcVectorDim
|
||||
8, // ABlockTransferSrcScalarPerVector
|
||||
8, // ABlockTransferDstScalarPerVector_K1
|
||||
true, // ABlockLdsAddExtraM
|
||||
S<4, 64, 1>, // BBlockTransferThreadClusterLengths_K0_N_K1
|
||||
S<1, 0, 2>, // BBlockTransferThreadClusterArrangeOrder
|
||||
S<1, 0, 2>, // BBlockTransferSrcAccessOrder
|
||||
2, // BBlockTransferSrcVectorDim
|
||||
8, // BBlockTransferSrcScalarPerVector
|
||||
8, // BBlockTransferDstScalarPerVector_K1
|
||||
true, // BBlockLdsAddExtraN
|
||||
1, // CShuffleMXdlPerWavePerShuffle
|
||||
1, // CShuffleNXdlPerWavePerShuffle
|
||||
S<1, 1, 32, 1, 1, 8>, // CBlockTransferClusterLengths_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl
|
||||
8>; // CBlockTransferScalarPerVector_NWaveNPerXdl
|
||||
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
|
||||
CDataType, // 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
|
||||
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
|
||||
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
|
||||
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
|
||||
// clang-format on
|
||||
|
||||
using ReferenceGemmInstance = ck::tensor_operation::host::
|
||||
|
||||
@@ -4,7 +4,6 @@
|
||||
#include <cstdlib>
|
||||
#include <stdlib.h>
|
||||
#include <half.hpp>
|
||||
|
||||
#include "check_err.hpp"
|
||||
#include "config.hpp"
|
||||
#include "device.hpp"
|
||||
@@ -12,7 +11,6 @@
|
||||
#include "host_tensor_generator.hpp"
|
||||
#include "device_tensor.hpp"
|
||||
#include "device_gemm_xdl.hpp"
|
||||
#include "device_gemm_xdl_c_shuffle.hpp"
|
||||
#include "device_gemm_xdl_cshuffle.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "reference_gemm.hpp"
|
||||
@@ -46,11 +44,11 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
|
||||
|
||||
// clang-format off
|
||||
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
|
||||
//######| ALayout| BLayout| CLayout|AData| BData| CData| GemmAcc| 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| DataType| DataType| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| ExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MPerBlock| 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_NPerBlock| _NPerBlock|
|
||||
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
< Row, Col, Row, F16, F16, F16, F32, F32, 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>;
|
||||
//######| 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|
|
||||
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
|
||||
< Row, Col, Row, F16, F16, F16, F32, F32, 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::
|
||||
|
||||
@@ -11,8 +11,7 @@
|
||||
#include "host_tensor.hpp"
|
||||
#include "host_tensor_generator.hpp"
|
||||
#include "device_tensor.hpp"
|
||||
#include "device_gemm_xdl.hpp"
|
||||
#include "device_gemm_xdl_c_shuffle.hpp"
|
||||
#include "device_gemm_xdl_cshuffle.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "reference_gemm.hpp"
|
||||
#include "gemm_specialization.hpp"
|
||||
@@ -20,64 +19,63 @@
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using F32 = float;
|
||||
|
||||
using Row = ck::tensor_layout::gemm::RowMajor;
|
||||
using Col = ck::tensor_layout::gemm::ColumnMajor;
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
using ADataType = int8_t;
|
||||
using BDataType = int8_t;
|
||||
using CDataType = int32_t;
|
||||
using CDataType = int8_t;
|
||||
using AccDataType = int32_t;
|
||||
using CShuffleDataType = int32_t;
|
||||
using CShuffleDataType = int8_t;
|
||||
|
||||
using ALayout = ck::tensor_layout::gemm::RowMajor;
|
||||
using BLayout = ck::tensor_layout::gemm::ColumnMajor;
|
||||
using CLayout = ck::tensor_layout::gemm::RowMajor;
|
||||
|
||||
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
|
||||
|
||||
// clang-format off
|
||||
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl_C_Shuffle<
|
||||
ADataType, // ADataType
|
||||
BDataType, // BDataType
|
||||
CDataType, // CDataType
|
||||
AccDataType, // AccDataType
|
||||
CShuffleDataType, // CShuffleDataType
|
||||
ALayout, // ALayout
|
||||
BLayout, // BLayout
|
||||
CLayout, // CLayout
|
||||
PassThrough, // AElementwiseOperation
|
||||
PassThrough, // BElementwiseOperation
|
||||
PassThrough, // CElementwiseOperation
|
||||
256, // BlockSize
|
||||
256, // MPerBlock
|
||||
128, // NPerBlock
|
||||
64, // KPerBlock
|
||||
16, // AK1
|
||||
16, // BK1
|
||||
32, // MPerXDL
|
||||
32, // NPerXDL
|
||||
4, // MXdlPerWave
|
||||
2, // NXdlPerWave
|
||||
S<4, 64, 1>, // ABlockTransferThreadClusterLengths_K0_M_K1
|
||||
S<1, 0, 2>, // ABlockTransferThreadClusterArrangeOrder
|
||||
S<1, 0, 2>, // ABlockTransferSrcAccessOrder
|
||||
2, // ABlockTransferSrcVectorDim
|
||||
16, // ABlockTransferSrcScalarPerVector
|
||||
16, // ABlockTransferDstScalarPerVector_K1
|
||||
true, // ABlockLdsAddExtraM
|
||||
S<4, 64, 1>, // BBlockTransferThreadClusterLengths_K0_N_K1
|
||||
S<1, 0, 2>, // BBlockTransferThreadClusterArrangeOrder
|
||||
S<1, 0, 2>, // BBlockTransferSrcAccessOrder
|
||||
2, // BBlockTransferSrcVectorDim
|
||||
16, // BBlockTransferSrcScalarPerVector
|
||||
16, // BBlockTransferDstScalarPerVector_K1
|
||||
true, // BBlockLdsAddExtraN
|
||||
1, // CShuffleMXdlPerWavePerShuffle
|
||||
1, // CShuffleNXdlPerWavePerShuffle
|
||||
S<1, 1, 32, 1, 1, 8>, // CBlockTransferClusterLengths_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl
|
||||
4>; // CBlockTransferScalarPerVector_NWaveNPerXdl
|
||||
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
|
||||
// clang-format on
|
||||
|
||||
using ReferenceGemmInstance = ck::tensor_operation::host::
|
||||
|
||||
@@ -13,74 +13,91 @@
|
||||
#include "host_tensor_generator.hpp"
|
||||
#include "host_gemm.hpp"
|
||||
#include "device_tensor.hpp"
|
||||
#include "device_gemm_xdl.hpp"
|
||||
#include "device_gemm_xdl_c_shuffle.hpp"
|
||||
#include "device_gemm_xdl_cshuffle.hpp"
|
||||
#include "element_wise_operation.hpp"
|
||||
#include "reference_gemm.hpp"
|
||||
#include "gemm_specialization.hpp"
|
||||
|
||||
struct RequantReluRequant
|
||||
{
|
||||
// FIXME: We just need one scale for Relu / Leaky Relu / PRelu
|
||||
RequantReluRequant(float scaleGemm, float scaleRelu)
|
||||
: scaleGemm_(scaleGemm), scaleRelu_(scaleRelu)
|
||||
{
|
||||
}
|
||||
|
||||
__host__ __device__ constexpr void operator()(float& y, const float& x) const
|
||||
{
|
||||
float gemm_requant = scaleGemm_ * x;
|
||||
float relu = gemm_requant > 0 ? gemm_requant : 0;
|
||||
float relu_requant = scaleRelu_ * relu;
|
||||
y = relu_requant > 127 ? 127 : relu_requant < -128 ? -128 : relu_requant;
|
||||
}
|
||||
|
||||
float scaleGemm_;
|
||||
float scaleRelu_;
|
||||
};
|
||||
|
||||
template <ck::index_t... Is>
|
||||
using S = ck::Sequence<Is...>;
|
||||
|
||||
using F32 = float;
|
||||
|
||||
using Row = ck::tensor_layout::gemm::RowMajor;
|
||||
using Col = ck::tensor_layout::gemm::ColumnMajor;
|
||||
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
using RequantReluRequant = ck::tensor_operation::element_wise::RequantReluRequant;
|
||||
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
|
||||
|
||||
using ADataType = int8_t;
|
||||
using BDataType = int8_t;
|
||||
using CDataType = int8_t;
|
||||
using AccDataType = int32_t;
|
||||
using CShuffleDataType = int32_t;
|
||||
using CShuffleDataType = float;
|
||||
|
||||
using ALayout = ck::tensor_layout::gemm::RowMajor;
|
||||
using BLayout = ck::tensor_layout::gemm::ColumnMajor;
|
||||
using CLayout = ck::tensor_layout::gemm::RowMajor;
|
||||
|
||||
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
|
||||
|
||||
// clang-format off
|
||||
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl_C_Shuffle<
|
||||
ADataType, // ADataType
|
||||
BDataType, // BDataType
|
||||
CDataType, // CDataType
|
||||
AccDataType, // AccDataType
|
||||
CShuffleDataType, // CShuffleDataType
|
||||
ALayout, // ALayout
|
||||
BLayout, // BLayout
|
||||
CLayout, // CLayout
|
||||
PassThrough, // AElementwiseOperation
|
||||
PassThrough, // BElementwiseOperation
|
||||
RequantReluRequant, // CElementwiseOperation
|
||||
256, // BlockSize
|
||||
256, // MPerBlock
|
||||
128, // NPerBlock
|
||||
64, // KPerBlock
|
||||
16, // AK1
|
||||
16, // BK1
|
||||
32, // MPerXDL
|
||||
32, // NPerXDL
|
||||
4, // MXdlPerWave
|
||||
2, // NXdlPerWave
|
||||
S<4, 64, 1>, // ABlockTransferThreadClusterLengths_K0_M_K1
|
||||
S<1, 0, 2>, // ABlockTransferThreadClusterArrangeOrder
|
||||
S<1, 0, 2>, // ABlockTransferSrcAccessOrder
|
||||
2, // ABlockTransferSrcVectorDim
|
||||
16, // ABlockTransferSrcScalarPerVector
|
||||
16, // ABlockTransferDstScalarPerVector_K1
|
||||
true, // ABlockLdsAddExtraM
|
||||
S<4, 64, 1>, // BBlockTransferThreadClusterLengths_K0_N_K1
|
||||
S<1, 0, 2>, // BBlockTransferThreadClusterArrangeOrder
|
||||
S<1, 0, 2>, // BBlockTransferSrcAccessOrder
|
||||
2, // BBlockTransferSrcVectorDim
|
||||
16, // BBlockTransferSrcScalarPerVector
|
||||
16, // BBlockTransferDstScalarPerVector_K1
|
||||
true, // BBlockLdsAddExtraN
|
||||
1, // CShuffleMXdlPerWavePerShuffle
|
||||
1, // CShuffleNXdlPerWavePerShuffle
|
||||
S<1, 1, 64, 1, 1, 4>, // CBlockTransferClusterLengths_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl
|
||||
16>; // CBlockTransferScalarPerVector_NWaveNPerXdl
|
||||
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,
|
||||
RequantReluRequant, // 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, // bool 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, // bool 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>
|
||||
// clang-format on
|
||||
|
||||
using ReferenceGemmInstance = ck::tensor_operation::host::
|
||||
|
||||
Reference in New Issue
Block a user