Skip to content

GitLab

Unverified Commit 4173b984 authored by Rostyslav Geyyer's avatar Rostyslav Geyyer Committed by GitHub
Browse files

Merge branch 'develop' into lwpck-756

parents 6de7d10d 85e2e1e2
Hide whitespace changes
Inline Side-by-side
Showing with 872 additions and 16 deletions
library/src/tensor_operation_instance/gpu/gemm_bilinear/CMakeLists.txt
View file @ 4173b984
......@@ -4,5 +4,9 @@ add_instance_library(device_gemm_bilinear_instance
device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instance.cpp
device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instance.cpp
device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instance.cpp
device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instance.cpp
)
endif()
library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instance.cpp 0 → 100644
View file @ 4173b984
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using I8 = std::int8_t;
using I32 = std::int32_t;
using I8_Tuple = ck::Tuple<std::int8_t>;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
using Row_Tuple = ck::Tuple<Row>;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using Bilinear = ck::tensor_operation::element_wise::Bilinear;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
// e[m, n] = bilinear(a[m, k] * b[k, n], d[m, n])
using device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instances = std::tuple<
// clang-format off
//################################| A| B| Ds| E| AData| BData| DsData| EData| AccData| CShuffle| A| B| CDE| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | WMMA| WMMA| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//################################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmMultipleD_Wmma_CShuffle< Col, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmDefault, 256, 128, 128, 4, 16, 16, 16, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 2, S<1, 32, 1, 8>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmDefault, 128, 64, 64, 4, 16, 16, 16, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 2, S<1, 32, 1, 4>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmDefault, 64, 32, 32, 4, 16, 16, 16, 1, 2, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 2, S<1, 32, 1, 2>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmDefault, 32, 16, 16, 4, 16, 16, 16, 1, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 1, S<1, 16, 1, 2>, 8>,
// M/N/K padding
//################################| A| B| Ds| E| AData| BData| DsData| EData| AccData| CShuffle| A| B| CDE| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | WMMA| WMMA| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//################################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmMultipleD_Wmma_CShuffle< Col, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 256, 128, 128, 4, 16, 16, 16, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 2, S<1, 32, 1, 8>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 128, 64, 64, 4, 16, 16, 16, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 2, S<1, 32, 1, 4>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 64, 32, 32, 4, 16, 16, 16, 1, 2, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 2, S<1, 32, 1, 2>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 32, 16, 16, 4, 16, 16, 16, 1, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 1, S<1, 16, 1, 2>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 256, 128, 128, 8, 8, 16, 16, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 2, S<1, 32, 1, 8>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 128, 64, 64, 8, 8, 16, 16, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 2, S<1, 32, 1, 4>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 64, 32, 32, 8, 8, 16, 16, 1, 2, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 2, S<1, 32, 1, 2>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 32, 16, 16, 8, 8, 16, 16, 1, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 16, 1, 2>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 256, 128, 128, 8, 4, 16, 16, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 1, 1, 2, S<1, 32, 1, 8>, 4>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 128, 64, 64, 8, 4, 16, 16, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 1, 1, 2, S<1, 32, 1, 4>, 4>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 64, 32, 32, 8, 4, 16, 16, 1, 2, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 1, 1, 2, S<1, 32, 1, 2>, 4>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 32, 16, 16, 8, 4, 16, 16, 1, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 1, 1, 1, S<1, 16, 1, 2>, 4>
// clang-format on
>;
void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instances(
std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
Row,
Row_Tuple,
Row,
I8,
I8,
I8_Tuple,
I8,
PassThrough,
PassThrough,
Bilinear>>>& instances)
{
add_device_operation_instances(
instances, device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instance.cpp 0 → 100644
View file @ 4173b984
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using I8 = std::int8_t;
using I32 = std::int32_t;
using I8_Tuple = ck::Tuple<std::int8_t>;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
using Row_Tuple = ck::Tuple<Row>;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using Bilinear = ck::tensor_operation::element_wise::Bilinear;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
// e[m, n] = bilinear(a[m, k] * b[k, n], d[m, n])
using device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances = std::tuple<
// clang-format off
//################################| A| B| Ds| E| AData| BData| DsData| EData| AccData| CShuffle| A| B| CDE| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | WMMA| WMMA| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//################################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmMultipleD_Wmma_CShuffle< Col, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmDefault, 256, 128, 128, 4, 16, 16, 16, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 2, S<1, 32, 1, 8>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmDefault, 128, 64, 64, 4, 16, 16, 16, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 2, S<1, 32, 1, 4>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmDefault, 64, 32, 32, 4, 16, 16, 16, 1, 2, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 2, S<1, 32, 1, 2>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmDefault, 32, 16, 16, 4, 16, 16, 16, 1, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 16, 1, 2>, 8>,
// M/N/K padding
//################################| A| B| Ds| E| AData| BData| DsData| EData| AccData| CShuffle| A| B| CDE| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | WMMA| WMMA| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//################################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmMultipleD_Wmma_CShuffle< Col, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 256, 128, 128, 4, 16, 16, 16, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 2, S<1, 32, 1, 8>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 128, 64, 64, 4, 16, 16, 16, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 2, S<1, 32, 1, 4>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 64, 32, 32, 4, 16, 16, 16, 1, 2, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 2, S<1, 32, 1, 2>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 32, 16, 16, 4, 16, 16, 16, 1, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 16, 1, 2>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 256, 128, 128, 8, 8, 16, 16, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 2, S<1, 32, 1, 8>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 128, 64, 64, 8, 8, 16, 16, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 2, S<1, 32, 1, 4>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 64, 32, 32, 8, 8, 16, 16, 1, 2, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 2, S<1, 32, 1, 2>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 32, 16, 16, 8, 8, 16, 16, 1, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 2>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 256, 128, 128, 8, 4, 16, 16, 4, 2, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, 1, 2, S<1, 32, 1, 8>, 4>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 128, 64, 64, 8, 4, 16, 16, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, 1, 2, S<1, 32, 1, 4>, 4>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 64, 32, 32, 8, 4, 16, 16, 1, 2, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, 1, 2, S<1, 32, 1, 2>, 4>,
DeviceGemmMultipleD_Wmma_CShuffle< Col, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 32, 16, 16, 8, 4, 16, 16, 1, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, 1, 1, S<1, 16, 1, 2>, 4>
// clang-format on
>;
void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances(
std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
Col,
Row_Tuple,
Row,
I8,
I8,
I8_Tuple,
I8,
PassThrough,
PassThrough,
Bilinear>>>& instances)
{
add_device_operation_instances(
instances, device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instance.cpp 0 → 100644
View file @ 4173b984
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using I8 = std::int8_t;
using I32 = std::int32_t;
using I8_Tuple = ck::Tuple<std::int8_t>;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
using Row_Tuple = ck::Tuple<Row>;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using Bilinear = ck::tensor_operation::element_wise::Bilinear;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
// e[m, n] = bilinear(a[m, k] * b[k, n], d[m, n])
using device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instances = std::tuple<
// clang-format off
//################################| A| B| Ds| E| AData| BData| DsData| EData| AccData| CShuffle| A| B| CDE| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | WMMA| WMMA| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//################################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmMultipleD_Wmma_CShuffle< Row, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmDefault, 256, 128, 128, 4, 16, 16, 16, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 2, S<1, 32, 1, 8>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmDefault, 128, 64, 64, 4, 16, 16, 16, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 2, S<1, 32, 1, 4>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmDefault, 64, 32, 32, 4, 16, 16, 16, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 2, S<1, 32, 1, 2>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmDefault, 32, 16, 16, 4, 16, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 1, S<1, 16, 1, 2>, 8>,
// M/N/K padding
//################################| A| B| Ds| E| AData| BData| DsData| EData| AccData| CShuffle| A| B| CDE| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | WMMA| WMMA| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//################################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmMultipleD_Wmma_CShuffle< Row, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 256, 128, 128, 4, 16, 16, 16, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 2, S<1, 32, 1, 8>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 128, 64, 64, 4, 16, 16, 16, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 2, S<1, 32, 1, 4>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 64, 32, 32, 4, 16, 16, 16, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 2, S<1, 32, 1, 2>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 32, 16, 16, 4, 16, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 16, 1, 1, 1, S<1, 16, 1, 2>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 256, 128, 128, 8, 8, 16, 16, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 2, S<1, 32, 1, 8>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 128, 64, 64, 8, 8, 16, 16, 2, 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, 2, 8, 1, 1, 2, S<1, 32, 1, 4>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 64, 32, 32, 8, 8, 16, 16, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 2, S<1, 32, 1, 2>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 32, 16, 16, 8, 8, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 8, 1, 1, 1, S<1, 16, 1, 2>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 256, 128, 128, 8, 4, 16, 16, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 1, 1, 2, S<1, 32, 1, 8>, 4>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 128, 64, 64, 8, 4, 16, 16, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 1, 1, 2, S<1, 32, 1, 4>, 4>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 64, 32, 32, 8, 4, 16, 16, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 1, 1, 2, S<1, 32, 1, 2>, 4>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Row, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 32, 16, 16, 8, 4, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, 1, 1, 1, S<1, 16, 1, 2>, 4>
// clang-format on
>;
void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instances(
std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
Row,
Row_Tuple,
Row,
I8,
I8,
I8_Tuple,
I8,
PassThrough,
PassThrough,
Bilinear>>>& instances)
{
add_device_operation_instances(
instances, device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instance.cpp 0 → 100644
View file @ 4173b984
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using I8 = std::int8_t;
using I32 = std::int32_t;
using I8_Tuple = ck::Tuple<std::int8_t>;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
using Row_Tuple = ck::Tuple<Row>;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using Bilinear = ck::tensor_operation::element_wise::Bilinear;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
// e[m, n] = bilinear(a[m, k] * b[n, k], d[m, n])
using device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instances = std::tuple<
// clang-format off
// no padding
// N % 16 == 0 && K % 16 == 0
//################################| A| B| Ds| E| AData| BData| DsData| EData| AccData| CShuffle| A| B| CDE| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | WMMA| WMMA| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//################################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmDefault, 256, 128, 128, 4, 16, 16, 16, 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, 16, 16, 1, 1, 2, S<1, 32, 1, 8>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmDefault, 128, 64, 64, 4, 16, 16, 16, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 2, S<1, 32, 1, 4>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmDefault, 64, 32, 32, 4, 16, 16, 16, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 2, S<1, 32, 1, 2>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmDefault, 32, 16, 16, 4, 16, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 16, 1, 2>, 8>,
// M/N/K padding
// N % 16 == 0 && K % 16 == 0
//################################| A| B| Ds| E| AData| BData| DsData| EData| AccData| CShuffle| A| B| CDE| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | WMMA| WMMA| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//################################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 256, 128, 128, 4, 16, 16, 16, 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, 16, 16, 1, 1, 2, S<1, 32, 1, 8>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 128, 64, 64, 4, 16, 16, 16, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 2, S<1, 32, 1, 4>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 64, 32, 32, 4, 16, 16, 16, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 2, S<1, 32, 1, 2>, 16>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 32, 16, 16, 4, 16, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, 1, 1, S<1, 16, 1, 2>, 8>,
// M/N/K padding
// N % 8 == 0 && K % 8 == 0
//################################| A| B| Ds| E| AData| BData| DsData| EData| AccData| CShuffle| A| B| CDE| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | WMMA| WMMA| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//################################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 256, 128, 128, 8, 8, 16, 16, 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, 2, S<1, 32, 1, 8>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 128, 64, 64, 8, 8, 16, 16, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 2, S<1, 32, 1, 4>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 64, 32, 32, 8, 8, 16, 16, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 2, S<1, 32, 1, 2>, 8>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 32, 16, 16, 8, 8, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 2>, 8>,
// M/N/K padding
// N % 8 == 0 && K % 8 == 0
//################################| A| B| Ds| E| AData| BData| DsData| EData| AccData| CShuffle| A| B| CDE| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | WMMA| WMMA| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//################################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 256, 128, 128, 8, 4, 16, 16, 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, 2, S<1, 32, 1, 8>, 4>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 128, 64, 64, 8, 4, 16, 16, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, 1, 2, S<1, 32, 1, 4>, 4>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 64, 32, 32, 8, 4, 16, 16, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, 1, 2, S<1, 32, 1, 2>, 4>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 32, 16, 16, 8, 4, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 4, 4, 1, 1, 1, S<1, 16, 1, 2>, 4>,
// M/N/K padding
// N % 1 == 0 && K % 8 == 0
//################################| A| B| Ds| E| AData| BData| DsData| EData| AccData| CShuffle| A| B| CDE| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MRepeat| NRepeat| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//################################| Layout| Layout| Layout| Layout| Type| Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | WMMA| WMMA| | | ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//################################| | | | | | | | | | | Operation| Operation| Operation| | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 256, 128, 128, 8, 8, 16, 16, 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, 2, S<1, 32, 1, 8>, 1>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 128, 64, 64, 8, 8, 16, 16, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 2, S<1, 32, 1, 4>, 1>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 64, 32, 32, 8, 8, 16, 16, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 2, S<1, 32, 1, 2>, 1>,
DeviceGemmMultipleD_Wmma_CShuffle< Row, Col, Row_Tuple, Row, I8, I8, I8_Tuple, I8, I32, I32, PassThrough, PassThrough, Bilinear, GemmMNKPadding, 32, 16, 16, 8, 8, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 2>, 1>
// clang-format on
>;
void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instances(
std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
Col,
Row_Tuple,
Row,
I8,
I8,
I8_Tuple,
I8,
PassThrough,
PassThrough,
Bilinear>>>& instances)
{
add_device_operation_instances(
instances, device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instances{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp
View file @ 4173b984
......@@ -33,6 +33,18 @@ using MultiplyAdd = ck::tensor_operation::element_wise::MultiplyAdd;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
using device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_generic_instances =
std::tuple<
// clang-format off
// M/N/K padding
//##############################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| 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|
//##############################| 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, Row_Tuple, Row, F16, F8, F32, F32, F32_Tuple, F16, PassThrough, PassThrough, MultiplyAdd, GemmMNKPadding, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 1>
// clang-format on
>;
using device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instances =
std::tuple<
// clang-format off
......@@ -73,6 +85,10 @@ void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_m
PassThrough,
MultiplyAdd>>>& instances)
{
add_device_operation_instances(
instances,
device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_generic_instances{});
add_device_operation_instances(
instances,
device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instances{});
......
library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp
View file @ 4173b984
......@@ -33,6 +33,19 @@ using MultiplyAdd = ck::tensor_operation::element_wise::MultiplyAdd;
static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
using device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_generic_instances =
std::tuple<
// clang-format off
// M/N/K padding
// N % 8 == 0 && K % 1 == 0
//##############################| A| B| Ds| E| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| 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|
//##############################| 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, Col, Row_Tuple, Row, F16, F8, F32, F32, F32_Tuple, F16, PassThrough, PassThrough, MultiplyAdd, GemmMNKPadding, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 2, 1, 32>, 1>
// clang-format on
>;
using device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instances =
std::tuple<
// clang-format off
......@@ -55,7 +68,6 @@ using device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_i
DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F8, F32, F32, F32_Tuple, F16, PassThrough, PassThrough, MultiplyAdd, GemmMNKPadding, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 2, 1, 64>, 1>,
DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F8, F32, F32, F32_Tuple, F16, PassThrough, PassThrough, MultiplyAdd, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 2, 1, 32>, 1>,
DeviceGemmMultipleD_Xdl_CShuffle< Row, Col, Row_Tuple, Row, F16, F8, F32, F32, F32_Tuple, F16, PassThrough, PassThrough, MultiplyAdd, GemmMNKPadding, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 2, 1, 32>, 1>
// clang-format on
>;
......@@ -72,6 +84,10 @@ void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_m
PassThrough,
MultiplyAdd>>>& instances)
{
add_device_operation_instances(
instances,
device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_generic_instances{});
add_device_operation_instances(
instances,
device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instances{});
......
library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp
View file @ 4173b984
......@@ -30,6 +30,16 @@ using PassThrough = ck::tensor_operation::element_wise::PassThrough;
// static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
using device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_generic_instances = std::tuple<
// 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| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| Compute|
//#########################| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise|Specialization| 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| 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| |
//#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmMNPadding, 128, 32, 64, 4, 8, 32, 32, 1, 1, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 1, 8, true, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 2, 1, 8, true, 1, 1, S<1, 16, 1, 8>, 2, F16>
// clang-format on
>;
// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
using device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instances = std::tuple<
// clang-format off
......@@ -61,6 +71,8 @@ void add_device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instances(
DeviceGemmSplitK<Row, Row, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
instances)
{
add_device_operation_instances(instances,
device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_generic_instances{});
add_device_operation_instances(instances,
device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instances{});
}
......
library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp
View file @ 4173b984
......@@ -27,7 +27,17 @@ using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
static constexpr auto MNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
using device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_generic_instances = std::tuple<
// 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| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| Compute|
//#########################| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise|Specialization| 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| 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| |
//#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, MNKPadding, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 1, 8, true, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 1, 8, true, 1, 1, S<1, 16, 1, 4>, 2, F16>
// clang-format on
>;
// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
using device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instances = std::tuple<
......@@ -36,19 +46,19 @@ using device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instances = std::tuple<
//#########################| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise|Specialization| 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| 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| |
//#########################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 8>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 8>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 16, 1, 8>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 8>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 4>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 16, 1, 8>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 16, 1, 4>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 8>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 8>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 4>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 16, 1, 8>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 16, 1, 4>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 16, 1, 4>, 8, F16>
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, MNKPadding, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 8>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, MNKPadding, 256, 128, 256, 4, 8, 32, 32, 2, 4, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 8>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, MNKPadding, 128, 128, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 16, 1, 8>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, MNKPadding, 256, 128, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 8>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, MNKPadding, 128, 128, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 4>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, MNKPadding, 128, 64, 128, 4, 8, 32, 32, 2, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 16, 1, 8>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, MNKPadding, 64, 64, 64, 4, 8, 32, 32, 2, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 16, 1, 4>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, MNKPadding, 256, 128, 64, 4, 8, 32, 32, 2, 1, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 8>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, MNKPadding, 256, 64, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 8>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, MNKPadding, 128, 128, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 4>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, MNKPadding, 128, 32, 128, 4, 8, 32, 32, 1, 2, S<1, 4, 32, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 32, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 16, 1, 8>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, MNKPadding, 64, 64, 32, 4, 8, 32, 32, 2, 1, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 16, 1, 4>, 8, F16>,
DeviceGemmXdlSplitKCShuffle< F16, F8, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, MNKPadding, 64, 32, 64, 4, 8, 32, 32, 1, 2, S<1, 4, 16, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 16, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 16, 1, 4>, 8, F16>
// clang-format on
>;
......@@ -57,6 +67,8 @@ void add_device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instances(
DeviceGemmSplitK<Row, Col, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
instances)
{
add_device_operation_instances(instances,
device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_generic_instances{});
add_device_operation_instances(instances,
device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instances{});
}
......
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
View file @ 4173b984
add_instance_library(device_grouped_conv2d_fwd_instance
#xdl
# GNHWC, GKYXC, GNHWK
device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
......@@ -8,6 +9,13 @@ add_instance_library(device_grouped_conv2d_fwd_instance
device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
#dl
# GNHWC, GKYXC, GNHWK
device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
# WMMA
device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp
# NHWGC, GKYXC, NHWGK
device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
)
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp 0 → 100644
View file @ 4173b984
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_grouped_conv2d_fwd_dl_f16_instances<NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
Empty_Tuple,
PassThrough,
ConvFwdDefault>{});
add_device_operation_instances(instances,
device_grouped_conv2d_fwd_dl_f16_instances<NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
Empty_Tuple,
PassThrough,
ConvFwd1x1P0>{});
add_device_operation_instances(instances,
device_grouped_conv2d_fwd_dl_f16_instances<NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
Empty_Tuple,
PassThrough,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp 0 → 100644
View file @ 4173b984
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
F32,
F32,
Empty_Tuple,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_grouped_conv2d_fwd_dl_f32_instances<NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
Empty_Tuple,
PassThrough,
ConvFwdDefault>{});
add_device_operation_instances(instances,
device_grouped_conv2d_fwd_dl_f32_instances<NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
Empty_Tuple,
PassThrough,
ConvFwd1x1P0>{});
add_device_operation_instances(instances,
device_grouped_conv2d_fwd_dl_f32_instances<NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
Empty_Tuple,
PassThrough,
ConvFwd1x1S1P0>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp 0 → 100644
View file @ 4173b984
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv2d_fwd_wmma_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[g, n, hi ,wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
GNHWC,
GKYXC,
Empty_Tuple,
GNHWK,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_grouped_conv2d_fwd_wmma_f16_instances<GNHWC,
GKYXC,
Empty_Tuple,
GNHWK,
Empty_Tuple,
PassThrough,
ConvFwdDefault>{});
add_device_operation_instances(instances,
device_grouped_conv2d_fwd_wmma_f16_instances<GNHWC,
GKYXC,
Empty_Tuple,
GNHWK,
Empty_Tuple,
PassThrough,
ConvFwd1x1P0>{});
add_device_operation_instances(instances,
device_grouped_conv2d_fwd_wmma_f16_instances<GNHWC,
GKYXC,
Empty_Tuple,
GNHWK,
Empty_Tuple,
PassThrough,
ConvFwd1x1S1P0>{});
add_device_operation_instances(instances,
device_grouped_conv2d_fwd_wmma_f16_instances<GNHWC,
GKYXC,
Empty_Tuple,
GNHWK,
Empty_Tuple,
PassThrough,
ConvFwdOddC>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp 0 → 100644
View file @ 4173b984
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv2d_fwd_wmma_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[g, n, hi ,wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
GNHWC,
GKYXC,
Empty_Tuple,
GNHWK,
int8_t,
int8_t,
Empty_Tuple,
int8_t,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_grouped_conv2d_fwd_wmma_i8_instances<GNHWC,
GKYXC,
Empty_Tuple,
GNHWK,
Empty_Tuple,
PassThrough,
ConvFwdDefault>{});
add_device_operation_instances(instances,
device_grouped_conv2d_fwd_wmma_i8_instances<GNHWC,
GKYXC,
Empty_Tuple,
GNHWK,
Empty_Tuple,
PassThrough,
ConvFwd1x1P0>{});
add_device_operation_instances(instances,
device_grouped_conv2d_fwd_wmma_i8_instances<GNHWC,
GKYXC,
Empty_Tuple,
GNHWK,
Empty_Tuple,
PassThrough,
ConvFwd1x1S1P0>{});
add_device_operation_instances(instances,
device_grouped_conv2d_fwd_wmma_i8_instances<GNHWC,
GKYXC,
Empty_Tuple,
GNHWK,
Empty_Tuple,
PassThrough,
ConvFwdOddC>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/image_to_column/CMakeLists.txt 0 → 100644
View file @ 4173b984
add_instance_library(device_image_to_column_instance
device_image_to_column_nhwc_1d_instance.cpp
device_image_to_column_nhwc_2d_instance.cpp
device_image_to_column_nhwc_3d_instance.cpp
)
library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_1d_instance.cpp 0 → 100644
View file @ 4173b984
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_image_to_column_nhwc_1d_bf16_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, BF16, BF16>>>& instances)
{
add_device_operation_instances(instances, device_image_to_column_bf16_instances<1, GNWC>{});
}
void add_device_image_to_column_nhwc_1d_f16_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, F16, F16>>>& instances)
{
add_device_operation_instances(instances, device_image_to_column_f16_instances<1, GNWC>{});
}
void add_device_image_to_column_nhwc_1d_f32_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, F32, F32>>>& instances)
{
add_device_operation_instances(instances, device_image_to_column_f32_instances<1, GNWC>{});
}
void add_device_image_to_column_nhwc_1d_i8_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, int8_t, int8_t>>>& instances)
{
add_device_operation_instances(instances, device_image_to_column_i8_instances<1, GNWC>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_2d_instance.cpp 0 → 100644
View file @ 4173b984
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_image_to_column_nhwc_2d_bf16_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, BF16, BF16>>>& instances)
{
add_device_operation_instances(instances, device_image_to_column_bf16_instances<2, GNHWC>{});
}
void add_device_image_to_column_nhwc_2d_f16_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, F16, F16>>>& instances)
{
add_device_operation_instances(instances, device_image_to_column_f16_instances<2, GNHWC>{});
}
void add_device_image_to_column_nhwc_2d_f32_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, F32, F32>>>& instances)
{
add_device_operation_instances(instances, device_image_to_column_f32_instances<2, GNHWC>{});
}
void add_device_image_to_column_nhwc_2d_i8_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, int8_t, int8_t>>>& instances)
{
add_device_operation_instances(instances, device_image_to_column_i8_instances<2, GNHWC>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_3d_instance.cpp 0 → 100644
View file @ 4173b984
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_image_to_column_nhwc_3d_bf16_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, BF16, BF16>>>& instances)
{
add_device_operation_instances(instances, device_image_to_column_bf16_instances<3, GNDHWC>{});
}
void add_device_image_to_column_nhwc_3d_f16_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, F16, F16>>>& instances)
{
add_device_operation_instances(instances, device_image_to_column_f16_instances<3, GNDHWC>{});
}
void add_device_image_to_column_nhwc_3d_f32_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, F32, F32>>>& instances)
{
add_device_operation_instances(instances, device_image_to_column_f32_instances<3, GNDHWC>{});
}
void add_device_image_to_column_nhwc_3d_i8_instances(
std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, int8_t, int8_t>>>& instances)
{
add_device_operation_instances(instances, device_image_to_column_i8_instances<3, GNDHWC>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
profiler/README.md
View file @ 4173b984
......@@ -184,3 +184,41 @@ tflops: 95.337
GB/s: 69.2301
```
Note: This kernel use atomic add, this will cause output buffer to be accumulated multiple times, causing verification failure. To work around it, do not use CK's own timer and do verification at the same time.
## Profile image to column kernels
```bash
# arg1: tensor operation (" OP_NAME ": " OP_DESC ")
# arg2: data type (0: Input fp32, Weight fp32, Output fp32
# 1: Input fp16, Weight fp16, Output fp16
# 2: Input bf16, Weight bf16, Output bf16
# 3: Input int8, Weight int8, Output int8)
# arg3: tensor layout (0: Input[N, Hi, Wi, C], Output[N * Ho * Wo, Y * X * C])
# arg4: verification (0: no, 1: yes)
# arg5: initialization (0: no init, 1: integer value, 2: decimal value)
# arg6: print tensor value (0: no; 1: yes)
# arg7: time kernel (0: no, 1: yes)
# Following arguments (depending on number of spatial dims):
# Number of spatial dimensions (1=Conv1d, 2=Conv2d, 3=Conv3d)
# G, N, K, C,
# <filter spatial dimensions>, (ie Y, X for 2D)
# <input image spatial dimensions>, (ie Hi, Wi for 2D)
# <strides>, (ie Sy, Sx for 2D)
# <dilations>, (ie Dy, Dx for 2D)
# <left padding>, (ie LeftPy, LeftPx for 2D)
# <right padding>, (ie RightPy, RightPx for 2D)
################ op datatype layout verify init log time Ndims G N K C Y X Hi Wi Sy Sx Dy Dx LeftPy LeftPx RightPy RightPx
./bin/ckProfiler image_to_column 0 0 1 1 0 1 2 1 256 1 512 3 3 28 28 1 1 1 1 0 0 0 0
```
Result (MI210, FP32, NHWC)
```
input: dim 5, lengths {1, 256, 512, 28, 28}, strides {102760448, 401408, 1, 14336, 512}
output: dim 2, lengths {173056, 4608}, strides {4608, 1}
....
Best configuration parameters:
name: DeviceImageToColumn<128, 32, 64, 4>
avg_time: 3.12326
GB/s: 2042.59
```
profiler/include/profiler/profile_grouped_conv_fwd_impl.hpp
View file @ 4173b984
......@@ -215,7 +215,7 @@ bool profile_grouped_conv_fwd_impl(int do_verification,
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
DeviceOp>::GetInstances();
std::cout << "xdl found " << op_ptrs.size() << " instances" << std::endl;
std::cout << "ckProfiler found " << op_ptrs.size() << " instances" << std::endl;
for(auto& op_ptr : op_ptrs)
{
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment