Skip to content

GitLab

Commit 77e63b34 authored by Bartlomiej Wroblewski's avatar Bartlomiej Wroblewski
Browse files

Review: Introduce AK0 and BK0 instead of a single K0

parent 03cd2692
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 191 additions and 162 deletions
example/01_gemm/gemm_dpp_fp16.cpp
View file @ 77e63b34
......@@ -24,11 +24,11 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
// clang-format off
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDpp
// ######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MDpp| NDpp| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
// ######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise|Spacialization| Size| Block| Block| Block| | Dpp| Dpp| PerWave| PerWave| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
// ######| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
// ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmDefault, 128, 64, 64, 8, 8, 32, 8, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>;
// ######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MDpp| NDpp| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
// ######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise|Spacialization| Size| Block| Block| Block| | | Dpp| Dpp| PerWave| PerWave| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
// ######| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
// ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmDefault, 128, 64, 64, 64, 8, 2, 32, 8, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, 5, 1>;
// // clang-format on
using ReferenceGemmInstance = ck::tensor_operation::host::
......
include/ck/tensor_operation/gpu/block/blockwise_gemm_dpp.hpp
View file @ 77e63b34
......@@ -29,7 +29,7 @@ template <index_t BlockSize,
index_t MRepeat,
index_t NRepeat,
index_t KPack>
struct BlockwiseGemmDpp_k0mk1_k0nk1_m0n0m1n1m2n2
struct BlockwiseGemmDpp_ak0mak1_bk0nbk1_m0n0m1n1m2n2
{
static constexpr auto I0 = Number<0>{};
static constexpr auto I1 = Number<1>{};
......@@ -127,7 +127,7 @@ struct BlockwiseGemmDpp_k0mk1_k0nk1_m0n0m1n1m2n2
return make_tuple(c_thread_m, c_thread_n);
}
__host__ __device__ BlockwiseGemmDpp_k0mk1_k0nk1_m0n0m1n1m2n2()
__host__ __device__ BlockwiseGemmDpp_ak0mak1_bk0nbk1_m0n0m1n1m2n2()
{
static_assert(AK0MK1BlockDesc::IsKnownAtCompileTime() &&
BK0NK1BlockDesc::IsKnownAtCompileTime(),
......
include/ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp
View file @ 77e63b34
......@@ -33,8 +33,9 @@ template <typename ADataType,
ck::index_t BlockSize,
ck::index_t MPerBlock,
ck::index_t NPerBlock,
ck::index_t K0PerBlock,
ck::index_t K1,
ck::index_t KPerBlock,
ck::index_t AK1,
ck::index_t BK1,
ck::index_t MPerDpp,
ck::index_t NPerDpp,
ck::index_t MDppPerWave,
......@@ -67,7 +68,7 @@ struct DeviceGemmDpp : public DeviceGemm<ALayout,
BElementwiseOperation,
CElementwiseOperation>
{
using GridwiseGemm = GridwiseGemm_k0mk1_k0nk1_mn_dpp<
using GridwiseGemm = GridwiseGemm_ak0mak1_bk0nbk1_mn_dpp<
BlockSize,
ADataType,
AccDataType,
......@@ -82,10 +83,11 @@ struct DeviceGemmDpp : public DeviceGemm<ALayout,
GemmSpec,
MPerBlock,
NPerBlock,
K0PerBlock,
KPerBlock,
MPerDpp,
NPerDpp,
K1,
AK1,
BK1,
MDppPerWave,
NDppPerWave,
ABlockTransferThreadClusterLengths_K0_M_K1,
......@@ -242,8 +244,9 @@ struct DeviceGemmDpp : public DeviceGemm<ALayout,
<< BlockSize << ", "
<< MPerBlock << ", "
<< NPerBlock << ", "
<< K0PerBlock << ", "
<< K1 << ", "
<< KPerBlock << ", "
<< AK1 << ", "
<< BK1 << ", "
<< MPerDpp << ", "
<< NPerDpp << ", "
<< MDppPerWave << ", "
......
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp
View file @ 77e63b34
......@@ -30,17 +30,17 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
// clang-format off
using device_gemm_dpp_f16_f16_f16_km_kn_mn_instances = std::tuple<
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MDpp| NDpp| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | Dpp| Dpp| PerWave| PerWave| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 8, 8, 16, 16, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 8, 8, 32, 8, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 64, 8, 8, 32, 8, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 32, 32, 4, 8, 32, 8, 1, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 1, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 1, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 8, 8, 32, 8, 2, 4, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 32, 32, 4, 8, 32, 8, 1, 4, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 16, 16, 2, 8, 16, 16, 1, 1, S<2, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 1, true, S<2, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 1, true, 5, 1>
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MDpp| NDpp| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | | Dpp| Dpp| PerWave| PerWave| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 64, 4, 4, 16, 16, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 64, 4, 4, 32, 8, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 64, 64, 4, 4, 32, 8, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 32, 32, 32, 4, 4, 32, 8, 1, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 64, 4, 4, 32, 8, 2, 4, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, true, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 32, 32, 32, 4, 4, 32, 8, 1, 4, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, true, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 16, 16, 16, 4, 4, 16, 16, 1, 1, S<2, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, true, S<2, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, true, 5, 1>
>;
// clang-format on
......
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp
View file @ 77e63b34
......@@ -30,17 +30,17 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
// clang-format off
using device_gemm_dpp_f16_f16_f16_km_nk_mn_instances = std::tuple<
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MDpp| NDpp| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | Dpp| Dpp| PerWave| PerWave| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 8, 8, 16, 16, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 8, 8, 32, 8, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 64, 8, 8, 32, 8, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 32, 32, 4, 8, 32, 8, 1, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 1, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 8, 8, 32, 8, 2, 4, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 32, 32, 4, 8, 32, 8, 1, 4, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, S<4, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 16, 16, 2, 8, 16, 16, 1, 1, S<2, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 1, true, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MDpp| NDpp| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | | Dpp| Dpp| PerWave| PerWave| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 64, 4, 8, 16, 16, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 64, 4, 8, 32, 8, 2, 4, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 64, 64, 4, 8, 32, 8, 2, 2, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 32, 32, 32, 4, 8, 32, 8, 1, 1, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 64, 4, 8, 32, 8, 2, 4, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 32, 32, 32, 4, 8, 32, 8, 1, 4, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, true, S<4, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Col, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 16, 16, 16, 4, 8, 16, 16, 1, 1, S<2, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, true, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>
>;
// clang-format on
......
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp
View file @ 77e63b34
......@@ -29,17 +29,17 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
// clang-format off
using device_gemm_dpp_f16_f16_f16_mk_kn_mn_instances = std::tuple<
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MDpp| NDpp| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | Dpp| Dpp| PerWave| PerWave| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 8, 8, 16, 16, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 8, 8, 32, 8, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 64, 8, 8, 32, 8, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 32, 32, 4, 8, 32, 8, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 1, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 8, 8, 32, 8, 2, 4, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 32, 32, 4, 8, 32, 8, 1, 4, S<4, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 16, 16, 2, 8, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<2, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 1, true, 5, 1>
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MDpp| NDpp| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | | Dpp| Dpp| PerWave| PerWave| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 64, 8, 4, 16, 16, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 64, 8, 4, 32, 8, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 64, 64, 8, 4, 32, 8, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 32, 32, 32, 8, 4, 32, 8, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 64, 8, 4, 32, 8, 2, 4, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 32, 32, 32, 8, 4, 32, 8, 1, 4, S<4, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 8, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 4, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 16, 16, 16, 8, 4, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<2, 16, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 4, true, 5, 1>
>;
// clang-format on
......
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp
View file @ 77e63b34
......@@ -30,17 +30,17 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
// clang-format off
using device_gemm_dpp_f16_f16_f16_mk_nk_mn_instances = std::tuple<
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| MPer| NPer| MDpp| NDpp| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | Dpp| Dpp| PerWave| PerWave| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 8, 8, 16, 16, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 8, 8, 32, 8, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 64, 8, 8, 32, 8, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 32, 32, 4, 8, 32, 8, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 8, 8, 32, 8, 2, 4, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 32, 32, 4, 8, 32, 8, 1, 4, S<4, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 16, 16, 2, 8, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>
// ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MDpp| NDpp| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
// ########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Specialization| Size| Block| Block| Block| | | Dpp| Dpp| PerWave| PerWave| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
// ########| | | | | | | | Operation| Operation| Operation| | | | | | | | | | | | Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
// ########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 64, 8, 8, 16, 16, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 64, 8, 8, 32, 8, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 64, 64, 8, 8, 32, 8, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 32, 32, 32, 8, 8, 32, 8, 1, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 64, 8, 8, 32, 8, 2, 4, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 32, 32, 32, 8, 8, 32, 8, 1, 4, S<4, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<4, 8, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>,
DeviceGemmDpp< F16, F16, F16, F32, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 32, 16, 16, 16, 8, 8, 16, 16, 1, 1, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, S<2, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, true, 5, 1>
>;
// clang-format on
......
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