Skip to content

GitLab

"git@developer.sourcefind.cn:orangecat/ollama.git" did not exist on "5b2cf16397d302f97a079ec925adb0e81bcc2bd5"
Unverified Commit 3e6c2610 authored by ltqin's avatar ltqin Committed by GitHub
Browse files

Add FP64 XDL GEMM built-in function (#199) 



* add intrin_mfma_f64_16x16x4f64

* add example

* gemm reference add double data type

* chang init data

* fix M N PerXdlops

* fix ifdef

* add comparsion config

* add conv fwd example

* format log out

* change rc matrix egister layout

* reorganize example

* reorganize example 2

* format,because merge develop

* fix call impl adding acc data type

* lost ;

* add compiler warning

* change example tunning parameters

* add test for fp64

* add instance

* add test/gemm/gemm_fp64.cpp

* fix get name issue

* remove some tunning parameter

* fix conflict

* format

* use integer value for GEMM test

* add acc data type

* remove typeid because fp16

* fix streamconfig etc bug from merging develop

* format

* remove test_gemm_xdl_fp64

* add AccDataType

* AccDataType problem
Co-authored-by: default avatarqinletao <letaoqin@amd.com>
Co-authored-by: default avatarChao Liu <chao.liu2@amd.com>
parent 97c4d486
Hide whitespace changes
Inline Side-by-side
Showing with 361 additions and 24 deletions
library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp 0 → 100644
View file @ 3e6c2610
#include <stdlib.h>
#include "config.hpp"
#include "device_gemm_xdl.hpp"
#include "element_wise_operation.hpp"
#include "device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_gemm_instance {
using F64 = double;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
using device_gemm_xdl_f64_f64_f64_mk_kn_mn_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| CThreadTransfer| CThreadTransfer|
//##########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise|Spacialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##########| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmXdl< F64, F64, F64, F64, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 4, 2, 16, 16, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, 7, 1>,
DeviceGemmXdl< F64, F64, F64, F64, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 128, 64, 4, 2, 16, 16, 4, 4, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, 7, 1>,
DeviceGemmXdl< F64, F64, F64, F64, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 128, 4, 2, 16, 16, 4, 4, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 32, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 4, 2, true, 7, 1>,
DeviceGemmXdl< F64, F64, F64, F64, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 64, 4, 2, 16, 16, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 1, 2, true, 7, 1>,
DeviceGemmXdl< F64, F64, F64, F64, Row, Row, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 64, 128, 4, 2, 16, 16, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 64, 1>, S<0, 2, 1>, S<0, 2, 1>, 1, 2, 2, true, 7, 1>
// clang-format on
>;
void add_device_gemm_xdl_f64_f64_f64_mk_kn_mn_instances(
std::vector<DeviceGemmPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances, device_gemm_xdl_f64_f64_f64_mk_kn_mn_instances{});
}
} // namespace device_gemm_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp 0 → 100644
View file @ 3e6c2610
#include <stdlib.h>
#include "config.hpp"
#include "device_gemm_xdl.hpp"
#include "element_wise_operation.hpp"
#include "device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_gemm_instance {
using F64 = double;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
using device_gemm_xdl_f64_f64_f64_mk_nk_mn_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| CThreadTransfer| CThreadTransfer|
//##########| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise|Spacialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| SrcDstVectorDim| DstScalar|
//##########| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | | PerVector|
//##########| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceGemmXdl< F64, F64, F64, F64, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 128, 4, 2, 16, 16, 4, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, 7, 1>,
DeviceGemmXdl< F64, F64, F64, F64, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 128, 64, 4, 2, 16, 16, 4, 4, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, 7, 1>,
DeviceGemmXdl< F64, F64, F64, F64, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 64, 128, 4, 2, 16, 16, 4, 4, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, 7, 1>,
DeviceGemmXdl< F64, F64, F64, F64, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 64, 4, 2, 16, 16, 4, 4, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, 7, 1>,
DeviceGemmXdl< F64, F64, F64, F64, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 128, 64, 4, 2, 16, 16, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, 7, 1>,
DeviceGemmXdl< F64, F64, F64, F64, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 256, 64, 128, 4, 2, 16, 16, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, 7, 1>,
DeviceGemmXdl< F64, F64, F64, F64, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 128, 32, 4, 2, 16, 16, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, 7, 1>,
DeviceGemmXdl< F64, F64, F64, F64, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 128, 32, 128, 4, 2, 16, 16, 2, 4, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, 7, 1>,
DeviceGemmXdl< F64, F64, F64, F64, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 64, 32, 4, 2, 16, 16, 4, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, 7, 1>,
DeviceGemmXdl< F64, F64, F64, F64, Row, Col, Row, PassThrough, PassThrough, PassThrough, GemmDefault, 64, 32, 64, 4, 2, 16, 16, 2, 4, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 2, 2, true, 7, 1>
// clang-format on
>;
void add_device_gemm_xdl_f64_f64_f64_mk_nk_mn_instances(
std::vector<DeviceGemmPtr<PassThrough, PassThrough, PassThrough>>& instances)
{
add_device_operation_instances(instances, device_gemm_xdl_f64_f64_f64_mk_nk_mn_instances{});
}
} // namespace device_gemm_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
profiler/include/profile_gemm_impl.hpp
View file @ 3e6c2610
...@@ -98,6 +98,7 @@ namespace profiler { ...@@ -98,6 +98,7 @@ namespace profiler {
template <typename ADataType, template <typename ADataType,
typename BDataType, typename BDataType,
typename CDataType, typename CDataType,
typename AccDataType,
typename ALayout, typename ALayout,
typename BLayout, typename BLayout,
typename CLayout> typename CLayout>
...@@ -511,8 +512,14 @@ void profile_gemm_impl(int do_verification, ...@@ -511,8 +512,14 @@ void profile_gemm_impl(int do_verification,
bf16_to_f32_(b_k_n, b_f32_k_n); bf16_to_f32_(b_k_n, b_f32_k_n);
bf16_to_f32_(c_m_n_device_result, c_m_n_device_f32_result); bf16_to_f32_(c_m_n_device_result, c_m_n_device_f32_result);
using ReferenceGemmInstance = ck::tensor_operation::host:: using ReferenceGemmInstance =
ReferenceGemm<float, float, float, AElementOp, BElementOp, CElementOp>; ck::tensor_operation::host::ReferenceGemm<float,
float,
float,
float,
AElementOp,
BElementOp,
CElementOp>;
auto ref_gemm = ReferenceGemmInstance{}; auto ref_gemm = ReferenceGemmInstance{};
auto ref_invoker = ref_gemm.MakeInvoker(); auto ref_invoker = ref_gemm.MakeInvoker();
...@@ -544,6 +551,7 @@ void profile_gemm_impl(int do_verification, ...@@ -544,6 +551,7 @@ void profile_gemm_impl(int do_verification,
ck::tensor_operation::host::ReferenceGemm<ADataType, ck::tensor_operation::host::ReferenceGemm<ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
AElementOp, AElementOp,
BElementOp, BElementOp,
CElementOp>; CElementOp>;
......
profiler/include/profile_gemm_reduce_impl.hpp
View file @ 3e6c2610
...@@ -144,8 +144,13 @@ bool profile_gemm_reduce_impl(int do_verification, ...@@ -144,8 +144,13 @@ bool profile_gemm_reduce_impl(int do_verification,
if(do_verification) if(do_verification)
{ {
using ReferenceGemmInstance = ck::tensor_operation::host:: using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
ReferenceGemm<ADataType, BDataType, CDataType, AElementOp, BElementOp, CElementOp>; BDataType,
CDataType,
DDataType,
AElementOp,
BElementOp,
CElementOp>;
auto ref_gemm = ReferenceGemmInstance{}; auto ref_gemm = ReferenceGemmInstance{};
auto ref_invoker = ref_gemm.MakeInvoker(); auto ref_invoker = ref_gemm.MakeInvoker();
......
profiler/include/profile_grouped_gemm_impl.hpp
View file @ 3e6c2610
...@@ -43,6 +43,7 @@ namespace profiler { ...@@ -43,6 +43,7 @@ namespace profiler {
template <typename ADataType, template <typename ADataType,
typename BDataType, typename BDataType,
typename CDataType, typename CDataType,
typename AccDataType,
typename ALayout, typename ALayout,
typename BLayout, typename BLayout,
typename CLayout> typename CLayout>
...@@ -271,6 +272,7 @@ void profile_grouped_gemm_impl(int do_verification, ...@@ -271,6 +272,7 @@ void profile_grouped_gemm_impl(int do_verification,
ck::tensor_operation::host::ReferenceGemm<ADataType, ck::tensor_operation::host::ReferenceGemm<ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
AElementOp, AElementOp,
BElementOp, BElementOp,
CElementOp>; CElementOp>;
......
profiler/src/profile_gemm.cpp
View file @ 3e6c2610
...@@ -68,6 +68,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -68,6 +68,7 @@ int profile_gemm(int argc, char* argv[])
ck::profiler::profile_gemm_impl<ck::half_t, ck::profiler::profile_gemm_impl<ck::half_t,
ck::half_t, ck::half_t,
ck::half_t, ck::half_t,
float,
ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::RowMajor,
ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::RowMajor,
ck::tensor_layout::gemm::RowMajor>( ck::tensor_layout::gemm::RowMajor>(
...@@ -88,6 +89,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -88,6 +89,7 @@ int profile_gemm(int argc, char* argv[])
ck::profiler::profile_gemm_impl<ck::half_t, ck::profiler::profile_gemm_impl<ck::half_t,
ck::half_t, ck::half_t,
ck::half_t, ck::half_t,
float,
ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::RowMajor,
ck::tensor_layout::gemm::ColumnMajor, ck::tensor_layout::gemm::ColumnMajor,
ck::tensor_layout::gemm::RowMajor>( ck::tensor_layout::gemm::RowMajor>(
...@@ -108,6 +110,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -108,6 +110,7 @@ int profile_gemm(int argc, char* argv[])
ck::profiler::profile_gemm_impl<ck::half_t, ck::profiler::profile_gemm_impl<ck::half_t,
ck::half_t, ck::half_t,
ck::half_t, ck::half_t,
float,
ck::tensor_layout::gemm::ColumnMajor, ck::tensor_layout::gemm::ColumnMajor,
ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::RowMajor,
ck::tensor_layout::gemm::RowMajor>( ck::tensor_layout::gemm::RowMajor>(
...@@ -128,6 +131,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -128,6 +131,7 @@ int profile_gemm(int argc, char* argv[])
ck::profiler::profile_gemm_impl<ck::half_t, ck::profiler::profile_gemm_impl<ck::half_t,
ck::half_t, ck::half_t,
ck::half_t, ck::half_t,
float,
ck::tensor_layout::gemm::ColumnMajor, ck::tensor_layout::gemm::ColumnMajor,
ck::tensor_layout::gemm::ColumnMajor, ck::tensor_layout::gemm::ColumnMajor,
ck::tensor_layout::gemm::RowMajor>( ck::tensor_layout::gemm::RowMajor>(
...@@ -146,6 +150,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -146,6 +150,7 @@ int profile_gemm(int argc, char* argv[])
else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_KN_MN) else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_KN_MN)
{ {
ck::profiler::profile_gemm_impl<float, ck::profiler::profile_gemm_impl<float,
float,
float, float,
float, float,
ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::RowMajor,
...@@ -166,6 +171,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -166,6 +171,7 @@ int profile_gemm(int argc, char* argv[])
else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_NK_MN) else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_NK_MN)
{ {
ck::profiler::profile_gemm_impl<float, ck::profiler::profile_gemm_impl<float,
float,
float, float,
float, float,
ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::RowMajor,
...@@ -186,6 +192,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -186,6 +192,7 @@ int profile_gemm(int argc, char* argv[])
else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::KM_KN_MN) else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::KM_KN_MN)
{ {
ck::profiler::profile_gemm_impl<float, ck::profiler::profile_gemm_impl<float,
float,
float, float,
float, float,
ck::tensor_layout::gemm::ColumnMajor, ck::tensor_layout::gemm::ColumnMajor,
...@@ -206,6 +213,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -206,6 +213,7 @@ int profile_gemm(int argc, char* argv[])
else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::KM_NK_MN) else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::KM_NK_MN)
{ {
ck::profiler::profile_gemm_impl<float, ck::profiler::profile_gemm_impl<float,
float,
float, float,
float, float,
ck::tensor_layout::gemm::ColumnMajor, ck::tensor_layout::gemm::ColumnMajor,
...@@ -228,6 +236,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -228,6 +236,7 @@ int profile_gemm(int argc, char* argv[])
ck::profiler::profile_gemm_impl<int8_t, ck::profiler::profile_gemm_impl<int8_t,
int8_t, int8_t,
int8_t, int8_t,
int32_t,
ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::RowMajor,
ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::RowMajor,
ck::tensor_layout::gemm::RowMajor>( ck::tensor_layout::gemm::RowMajor>(
...@@ -248,6 +257,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -248,6 +257,7 @@ int profile_gemm(int argc, char* argv[])
ck::profiler::profile_gemm_impl<int8_t, ck::profiler::profile_gemm_impl<int8_t,
int8_t, int8_t,
int8_t, int8_t,
int32_t,
ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::RowMajor,
ck::tensor_layout::gemm::ColumnMajor, ck::tensor_layout::gemm::ColumnMajor,
ck::tensor_layout::gemm::RowMajor>( ck::tensor_layout::gemm::RowMajor>(
...@@ -268,6 +278,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -268,6 +278,7 @@ int profile_gemm(int argc, char* argv[])
ck::profiler::profile_gemm_impl<int8_t, ck::profiler::profile_gemm_impl<int8_t,
int8_t, int8_t,
int8_t, int8_t,
int32_t,
ck::tensor_layout::gemm::ColumnMajor, ck::tensor_layout::gemm::ColumnMajor,
ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::RowMajor,
ck::tensor_layout::gemm::RowMajor>( ck::tensor_layout::gemm::RowMajor>(
...@@ -288,6 +299,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -288,6 +299,7 @@ int profile_gemm(int argc, char* argv[])
ck::profiler::profile_gemm_impl<int8_t, ck::profiler::profile_gemm_impl<int8_t,
int8_t, int8_t,
int8_t, int8_t,
int32_t,
ck::tensor_layout::gemm::ColumnMajor, ck::tensor_layout::gemm::ColumnMajor,
ck::tensor_layout::gemm::ColumnMajor, ck::tensor_layout::gemm::ColumnMajor,
ck::tensor_layout::gemm::RowMajor>( ck::tensor_layout::gemm::RowMajor>(
...@@ -308,6 +320,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -308,6 +320,7 @@ int profile_gemm(int argc, char* argv[])
ck::profiler::profile_gemm_impl<ck::bhalf_t, ck::profiler::profile_gemm_impl<ck::bhalf_t,
ck::bhalf_t, ck::bhalf_t,
ck::bhalf_t, ck::bhalf_t,
float,
ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::RowMajor,
ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::RowMajor,
ck::tensor_layout::gemm::RowMajor>( ck::tensor_layout::gemm::RowMajor>(
...@@ -328,6 +341,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -328,6 +341,7 @@ int profile_gemm(int argc, char* argv[])
ck::profiler::profile_gemm_impl<ck::bhalf_t, ck::profiler::profile_gemm_impl<ck::bhalf_t,
ck::bhalf_t, ck::bhalf_t,
ck::bhalf_t, ck::bhalf_t,
float,
ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::RowMajor,
ck::tensor_layout::gemm::ColumnMajor, ck::tensor_layout::gemm::ColumnMajor,
ck::tensor_layout::gemm::RowMajor>( ck::tensor_layout::gemm::RowMajor>(
...@@ -348,6 +362,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -348,6 +362,7 @@ int profile_gemm(int argc, char* argv[])
ck::profiler::profile_gemm_impl<ck::bhalf_t, ck::profiler::profile_gemm_impl<ck::bhalf_t,
ck::bhalf_t, ck::bhalf_t,
ck::bhalf_t, ck::bhalf_t,
float,
ck::tensor_layout::gemm::ColumnMajor, ck::tensor_layout::gemm::ColumnMajor,
ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::RowMajor,
ck::tensor_layout::gemm::RowMajor>( ck::tensor_layout::gemm::RowMajor>(
...@@ -368,6 +383,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -368,6 +383,7 @@ int profile_gemm(int argc, char* argv[])
ck::profiler::profile_gemm_impl<ck::bhalf_t, ck::profiler::profile_gemm_impl<ck::bhalf_t,
ck::bhalf_t, ck::bhalf_t,
ck::bhalf_t, ck::bhalf_t,
float,
ck::tensor_layout::gemm::ColumnMajor, ck::tensor_layout::gemm::ColumnMajor,
ck::tensor_layout::gemm::ColumnMajor, ck::tensor_layout::gemm::ColumnMajor,
ck::tensor_layout::gemm::RowMajor>( ck::tensor_layout::gemm::RowMajor>(
......
profiler/src/profile_grouped_gemm.cpp
View file @ 3e6c2610
...@@ -79,6 +79,7 @@ int profile_grouped_gemm(int argc, char* argv[]) ...@@ -79,6 +79,7 @@ int profile_grouped_gemm(int argc, char* argv[])
if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN) if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
{ {
ck::profiler::profile_grouped_gemm_impl<ck::half_t, ck::profiler::profile_grouped_gemm_impl<ck::half_t,
ck::half_t,
ck::half_t, ck::half_t,
ck::half_t, ck::half_t,
ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::RowMajor,
...@@ -97,6 +98,7 @@ int profile_grouped_gemm(int argc, char* argv[]) ...@@ -97,6 +98,7 @@ int profile_grouped_gemm(int argc, char* argv[])
else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_NK_MN) else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_NK_MN)
{ {
ck::profiler::profile_grouped_gemm_impl<ck::half_t, ck::profiler::profile_grouped_gemm_impl<ck::half_t,
ck::half_t,
ck::half_t, ck::half_t,
ck::half_t, ck::half_t,
ck::tensor_layout::gemm::RowMajor, ck::tensor_layout::gemm::RowMajor,
...@@ -115,6 +117,7 @@ int profile_grouped_gemm(int argc, char* argv[]) ...@@ -115,6 +117,7 @@ int profile_grouped_gemm(int argc, char* argv[])
else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_KN_MN) else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_KN_MN)
{ {
ck::profiler::profile_grouped_gemm_impl<ck::half_t, ck::profiler::profile_grouped_gemm_impl<ck::half_t,
ck::half_t,
ck::half_t, ck::half_t,
ck::half_t, ck::half_t,
ck::tensor_layout::gemm::ColumnMajor, ck::tensor_layout::gemm::ColumnMajor,
...@@ -133,6 +136,7 @@ int profile_grouped_gemm(int argc, char* argv[]) ...@@ -133,6 +136,7 @@ int profile_grouped_gemm(int argc, char* argv[])
else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_NK_MN) else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_NK_MN)
{ {
ck::profiler::profile_grouped_gemm_impl<ck::half_t, ck::profiler::profile_grouped_gemm_impl<ck::half_t,
ck::half_t,
ck::half_t, ck::half_t,
ck::half_t, ck::half_t,
ck::tensor_layout::gemm::ColumnMajor, ck::tensor_layout::gemm::ColumnMajor,
......
test/gemm/gemm_dl_fp16.cpp
View file @ 3e6c2610
...@@ -43,9 +43,10 @@ void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(std::vector<DeviceGemmNoO ...@@ -43,9 +43,10 @@ void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(std::vector<DeviceGemmNoO
int main() int main()
{ {
using ADataType = ck::half_t; using ADataType = ck::half_t;
using BDataType = ck::half_t; using BDataType = ck::half_t;
using CDataType = ck::half_t; using CDataType = ck::half_t;
using AccDataType = float;
using RowMajor = ck::tensor_layout::gemm::RowMajor; using RowMajor = ck::tensor_layout::gemm::RowMajor;
using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor; using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;
...@@ -63,6 +64,7 @@ int main() ...@@ -63,6 +64,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
RowMajor, RowMajor,
...@@ -81,6 +83,7 @@ int main() ...@@ -81,6 +83,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
ColumnMajor, ColumnMajor,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
...@@ -99,6 +102,7 @@ int main() ...@@ -99,6 +102,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
RowMajor, RowMajor,
RowMajor, RowMajor,
RowMajor, RowMajor,
...@@ -117,6 +121,7 @@ int main() ...@@ -117,6 +121,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
RowMajor, RowMajor,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
......
test/gemm/gemm_dl_fp32.cpp
View file @ 3e6c2610
...@@ -43,9 +43,10 @@ void add_device_gemm_dl_f32_f32_f32_mk_kn_mn_instances(std::vector<DeviceGemmNoO ...@@ -43,9 +43,10 @@ void add_device_gemm_dl_f32_f32_f32_mk_kn_mn_instances(std::vector<DeviceGemmNoO
int main() int main()
{ {
using ADataType = float; using ADataType = float;
using BDataType = float; using BDataType = float;
using CDataType = float; using CDataType = float;
using AccDataType = float;
using RowMajor = ck::tensor_layout::gemm::RowMajor; using RowMajor = ck::tensor_layout::gemm::RowMajor;
using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor; using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;
...@@ -61,6 +62,7 @@ int main() ...@@ -61,6 +62,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
RowMajor, RowMajor,
...@@ -79,6 +81,7 @@ int main() ...@@ -79,6 +81,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
ColumnMajor, ColumnMajor,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
...@@ -97,6 +100,7 @@ int main() ...@@ -97,6 +100,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
RowMajor, RowMajor,
RowMajor, RowMajor,
RowMajor, RowMajor,
...@@ -115,6 +119,7 @@ int main() ...@@ -115,6 +119,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
RowMajor, RowMajor,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
......
test/gemm/gemm_dl_int8.cpp
View file @ 3e6c2610
...@@ -43,9 +43,10 @@ void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPt ...@@ -43,9 +43,10 @@ void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPt
int main() int main()
{ {
using ADataType = int8_t; using ADataType = int8_t;
using BDataType = int8_t; using BDataType = int8_t;
using CDataType = int8_t; using CDataType = int8_t;
using AccDataType = int;
using RowMajor = ck::tensor_layout::gemm::RowMajor; using RowMajor = ck::tensor_layout::gemm::RowMajor;
using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor; using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;
...@@ -61,6 +62,7 @@ int main() ...@@ -61,6 +62,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
RowMajor, RowMajor,
...@@ -79,6 +81,7 @@ int main() ...@@ -79,6 +81,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
ColumnMajor, ColumnMajor,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
...@@ -97,6 +100,7 @@ int main() ...@@ -97,6 +100,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
RowMajor, RowMajor,
RowMajor, RowMajor,
RowMajor, RowMajor,
...@@ -115,6 +119,7 @@ int main() ...@@ -115,6 +119,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
RowMajor, RowMajor,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
......
test/gemm/gemm_util.hpp
View file @ 3e6c2610
...@@ -111,6 +111,7 @@ template <typename DeviceGemmPtr_, ...@@ -111,6 +111,7 @@ template <typename DeviceGemmPtr_,
typename ADataType, typename ADataType,
typename BDataType, typename BDataType,
typename CDataType, typename CDataType,
typename AccDataType,
typename ALayout, typename ALayout,
typename BLayout, typename BLayout,
typename CLayout, typename CLayout,
...@@ -186,6 +187,7 @@ struct TestGemm ...@@ -186,6 +187,7 @@ struct TestGemm
ck::tensor_operation::host::ReferenceGemm<ADataType, ck::tensor_operation::host::ReferenceGemm<ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
AElementwiseOperation, AElementwiseOperation,
BElementwiseOperation, BElementwiseOperation,
CElementwiseOperation>; CElementwiseOperation>;
...@@ -215,6 +217,11 @@ struct TestGemm ...@@ -215,6 +217,11 @@ struct TestGemm
res = ck::utils::check_err(c_device.mData, c_host.mData); res = ck::utils::check_err(c_device.mData, c_host.mData);
std::cout << (res ? "SUCCESS" : "FAILURE") << std::endl; std::cout << (res ? "SUCCESS" : "FAILURE") << std::endl;
} }
else if(std::is_same<CDataType, double>::value)
{
res = ck::utils::check_err(c_device.mData, c_host.mData);
std::cout << (res ? "SUCCESS" : "FAILURE") << std::endl;
}
return res; return res;
} }
...@@ -311,6 +318,7 @@ struct TestGemmBF16 ...@@ -311,6 +318,7 @@ struct TestGemmBF16
// use fp32 host kernel to verify bf16 device kernel // use fp32 host kernel to verify bf16 device kernel
using ReferenceGemmInstance = using ReferenceGemmInstance =
ck::tensor_operation::host::ReferenceGemm<float, ck::tensor_operation::host::ReferenceGemm<float,
float,
float, float,
float, float,
AElementwiseOperation, AElementwiseOperation,
......
test/gemm/gemm_xdl_fp16.cpp
View file @ 3e6c2610
...@@ -52,9 +52,10 @@ void add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances( ...@@ -52,9 +52,10 @@ void add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(
int main() int main()
{ {
using ADataType = ck::half_t; using ADataType = ck::half_t;
using BDataType = ck::half_t; using BDataType = ck::half_t;
using CDataType = ck::half_t; using CDataType = ck::half_t;
using AccDataType = float;
using RowMajor = ck::tensor_layout::gemm::RowMajor; using RowMajor = ck::tensor_layout::gemm::RowMajor;
using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor; using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;
...@@ -74,6 +75,7 @@ int main() ...@@ -74,6 +75,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
RowMajor, RowMajor,
...@@ -96,6 +98,7 @@ int main() ...@@ -96,6 +98,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
ColumnMajor, ColumnMajor,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
...@@ -118,6 +121,7 @@ int main() ...@@ -118,6 +121,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
RowMajor, RowMajor,
RowMajor, RowMajor,
RowMajor, RowMajor,
...@@ -142,6 +146,7 @@ int main() ...@@ -142,6 +146,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
RowMajor, RowMajor,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
......
test/gemm/gemm_xdl_fp32.cpp
View file @ 3e6c2610
...@@ -53,9 +53,10 @@ void add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instances(std::vector<De ...@@ -53,9 +53,10 @@ void add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instances(std::vector<De
int main() int main()
{ {
using ADataType = float; using ADataType = float;
using BDataType = float; using BDataType = float;
using CDataType = float; using CDataType = float;
using AccDataType = float;
using RowMajor = ck::tensor_layout::gemm::RowMajor; using RowMajor = ck::tensor_layout::gemm::RowMajor;
using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor; using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;
...@@ -75,6 +76,7 @@ int main() ...@@ -75,6 +76,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
RowMajor, RowMajor,
...@@ -97,6 +99,7 @@ int main() ...@@ -97,6 +99,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
ColumnMajor, ColumnMajor,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
...@@ -119,6 +122,7 @@ int main() ...@@ -119,6 +122,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
RowMajor, RowMajor,
RowMajor, RowMajor,
RowMajor, RowMajor,
...@@ -141,6 +145,7 @@ int main() ...@@ -141,6 +145,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
RowMajor, RowMajor,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
......
test/gemm/gemm_xdl_fp64.cpp 0 → 100644
View file @ 3e6c2610
#include <algorithm>
#include <cstdlib>
#include <half.hpp>
#include <iostream>
#include <numeric>
#include <tuple>
#include <vector>
#include "gemm_util.hpp"
#include "config.hpp"
#include "print.hpp"
#include "device.hpp"
#include "host_tensor.hpp"
#include "host_tensor_generator.hpp"
#include "host_gemm.hpp"
#include "device_tensor.hpp"
#include "device_gemm_xdl.hpp"
#include "element_wise_operation.hpp"
#include "reference_gemm.hpp"
#include "gemm_specialization.hpp"
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using DeviceGemmNoOpPtr =
ck::tensor_operation::device::DeviceGemmPtr<ck::tensor_operation::element_wise::PassThrough,
ck::tensor_operation::element_wise::PassThrough,
ck::tensor_operation::element_wise::PassThrough>;
namespace ck {
namespace tensor_operation {
namespace device {
namespace device_gemm_instance {
void add_device_gemm_xdl_f64_f64_f64_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
void add_device_gemm_xdl_f64_f64_f64_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
void add_device_gemm_xdl_f64_f64_f64_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
void add_device_gemm_xdl_f64_f64_f64_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
} // namespace device_gemm_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
inline std::string get_device_name()
{
hipDeviceProp_t props{};
int device;
auto status = hipGetDevice(&device);
if(status != hipSuccess)
{
return std::string();
}
status = hipGetDeviceProperties(&props, device);
if(status != hipSuccess)
{
return std::string();
}
const std::string name(props.gcnArchName);
return name;
}
int main()
{
if(get_device_name().find("gfx90a") == std::string::npos)
{
std::cout << "TestGemm ..... SUCCESS" << std::endl;
return 0;
}
using ADataType = double;
using BDataType = double;
using CDataType = double;
using AccDataType = double;
using RowMajor = ck::tensor_layout::gemm::RowMajor;
using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;
bool res = true;
std::vector<DeviceGemmNoOpPtr> gemmPtrs;
ck::tensor_operation::device::device_gemm_instance::
add_device_gemm_xdl_f64_f64_f64_km_kn_mn_instances(gemmPtrs);
for(auto& gemmPtr : gemmPtrs)
{
res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
ADataType,
BDataType,
CDataType,
AccDataType,
ColumnMajor,
RowMajor,
RowMajor,
PassThrough,
PassThrough,
PassThrough>{}(gemmPtr);
}
gemmPtrs.clear();
ck::tensor_operation::device::device_gemm_instance::
add_device_gemm_xdl_f64_f64_f64_km_nk_mn_instances(gemmPtrs);
for(auto& gemmPtr : gemmPtrs)
{
res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
ADataType,
BDataType,
CDataType,
AccDataType,
ColumnMajor,
ColumnMajor,
RowMajor,
PassThrough,
PassThrough,
PassThrough>{}(gemmPtr);
}
gemmPtrs.clear();
ck::tensor_operation::device::device_gemm_instance::
add_device_gemm_xdl_f64_f64_f64_mk_kn_mn_instances(gemmPtrs);
for(auto& gemmPtr : gemmPtrs)
{
res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
ADataType,
BDataType,
CDataType,
AccDataType,
RowMajor,
RowMajor,
RowMajor,
PassThrough,
PassThrough,
PassThrough>{}(gemmPtr);
}
gemmPtrs.clear();
ck::tensor_operation::device::device_gemm_instance::
add_device_gemm_xdl_f64_f64_f64_mk_nk_mn_instances(gemmPtrs);
for(auto& gemmPtr : gemmPtrs)
{
res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
ADataType,
BDataType,
CDataType,
AccDataType,
RowMajor,
ColumnMajor,
RowMajor,
PassThrough,
PassThrough,
PassThrough>{}(gemmPtr);
}
std::cout << "TestGemm ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
return res ? 0 : 1;
}
test/gemm/gemm_xdl_int8.cpp
View file @ 3e6c2610
...@@ -42,9 +42,10 @@ void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(std::vector<Devic ...@@ -42,9 +42,10 @@ void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(std::vector<Devic
int main() int main()
{ {
using ADataType = int8_t; using ADataType = int8_t;
using BDataType = int8_t; using BDataType = int8_t;
using CDataType = int8_t; using CDataType = int8_t;
using AccDataType = int32_t;
using RowMajor = ck::tensor_layout::gemm::RowMajor; using RowMajor = ck::tensor_layout::gemm::RowMajor;
using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor; using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;
...@@ -61,6 +62,7 @@ int main() ...@@ -61,6 +62,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
RowMajor, RowMajor,
...@@ -79,6 +81,7 @@ int main() ...@@ -79,6 +81,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
ColumnMajor, ColumnMajor,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
...@@ -97,6 +100,7 @@ int main() ...@@ -97,6 +100,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
RowMajor, RowMajor,
RowMajor, RowMajor,
RowMajor, RowMajor,
...@@ -115,6 +119,7 @@ int main() ...@@ -115,6 +119,7 @@ int main()
ADataType, ADataType,
BDataType, BDataType,
CDataType, CDataType,
AccDataType,
RowMajor, RowMajor,
ColumnMajor, ColumnMajor,
RowMajor, RowMajor,
......
test/grouped_gemm/grouped_gemm_fp16.cpp
View file @ 3e6c2610
...@@ -151,8 +151,13 @@ bool TestGroupedGemm(DeviceGroupedGemmPtr_& groupedGemmPtr) ...@@ -151,8 +151,13 @@ bool TestGroupedGemm(DeviceGroupedGemmPtr_& groupedGemmPtr)
{ {
c_tensors_device[i]->FromDevice(c_device_tensors[i].mData.data()); c_tensors_device[i]->FromDevice(c_device_tensors[i].mData.data());
using ReferenceGemmInstance = ck::tensor_operation::host:: using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
ReferenceGemm<ADataType, BDataType, CDataType, PassThrough, PassThrough, PassThrough>; BDataType,
CDataType,
AccDataType,
PassThrough,
PassThrough,
PassThrough>;
auto ref_gemm = ReferenceGemmInstance{}; auto ref_gemm = ReferenceGemmInstance{};
auto ref_invoker = ref_gemm.MakeInvoker(); auto ref_invoker = ref_gemm.MakeInvoker();
......
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