mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-04-19 22:39:03 +00:00
ec7c2e912e
* format * improving pipeline * fix typo * format * adding thread group * adding thread group * adding thread group * adding gemm pipeline * tweak * refactor * refactor * add missing type convert * refactor * refactor * refactor * clean * fix build * refactor * format * clean up * use remove_cvref_t * clean * clean up * clean up * clean up
155 lines
6.6 KiB
C++
155 lines
6.6 KiB
C++
#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 "device_gemm_xdl_cshuffle.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_f32_f32_f32_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
|
|
void add_device_gemm_xdl_f32_f32_f32_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
|
|
void add_device_gemm_xdl_f32_f32_f32_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
|
|
void add_device_gemm_xdl_f32_f32_f32_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
|
|
|
|
void add_device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
|
|
void add_device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
|
|
void add_device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
|
|
void add_device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
|
|
|
|
void add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
|
|
void add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
|
|
void add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
|
|
void add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
|
|
|
|
} // namespace device_gemm_instance
|
|
} // namespace device
|
|
} // namespace tensor_operation
|
|
} // namespace ck
|
|
|
|
int main()
|
|
{
|
|
using ADataType = float;
|
|
using BDataType = float;
|
|
using CDataType = float;
|
|
|
|
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_f32_f32_f32_km_kn_mn_instances(gemmPtrs);
|
|
ck::tensor_operation::device::device_gemm_instance::
|
|
add_device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instances(gemmPtrs);
|
|
ck::tensor_operation::device::device_gemm_instance::
|
|
add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instances(gemmPtrs);
|
|
|
|
for(auto& gemmPtr : gemmPtrs)
|
|
{
|
|
res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
|
|
ADataType,
|
|
BDataType,
|
|
CDataType,
|
|
ColumnMajor,
|
|
RowMajor,
|
|
RowMajor,
|
|
PassThrough,
|
|
PassThrough,
|
|
PassThrough>{}(gemmPtr);
|
|
}
|
|
|
|
gemmPtrs.clear();
|
|
ck::tensor_operation::device::device_gemm_instance::
|
|
add_device_gemm_xdl_f32_f32_f32_km_nk_mn_instances(gemmPtrs);
|
|
ck::tensor_operation::device::device_gemm_instance::
|
|
add_device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instances(gemmPtrs);
|
|
ck::tensor_operation::device::device_gemm_instance::
|
|
add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instances(gemmPtrs);
|
|
|
|
for(auto& gemmPtr : gemmPtrs)
|
|
{
|
|
res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
|
|
ADataType,
|
|
BDataType,
|
|
CDataType,
|
|
ColumnMajor,
|
|
ColumnMajor,
|
|
RowMajor,
|
|
PassThrough,
|
|
PassThrough,
|
|
PassThrough>{}(gemmPtr);
|
|
}
|
|
|
|
gemmPtrs.clear();
|
|
ck::tensor_operation::device::device_gemm_instance::
|
|
add_device_gemm_xdl_f32_f32_f32_mk_kn_mn_instances(gemmPtrs);
|
|
ck::tensor_operation::device::device_gemm_instance::
|
|
add_device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instances(gemmPtrs);
|
|
ck::tensor_operation::device::device_gemm_instance::
|
|
add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instances(gemmPtrs);
|
|
|
|
for(auto& gemmPtr : gemmPtrs)
|
|
{
|
|
res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
|
|
ADataType,
|
|
BDataType,
|
|
CDataType,
|
|
RowMajor,
|
|
RowMajor,
|
|
RowMajor,
|
|
PassThrough,
|
|
PassThrough,
|
|
PassThrough>{}(gemmPtr);
|
|
}
|
|
|
|
gemmPtrs.clear();
|
|
ck::tensor_operation::device::device_gemm_instance::
|
|
add_device_gemm_xdl_f32_f32_f32_mk_nk_mn_instances(gemmPtrs);
|
|
ck::tensor_operation::device::device_gemm_instance::
|
|
add_device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instances(gemmPtrs);
|
|
ck::tensor_operation::device::device_gemm_instance::
|
|
add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instances(gemmPtrs);
|
|
|
|
for(auto& gemmPtr : gemmPtrs)
|
|
{
|
|
res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
|
|
ADataType,
|
|
BDataType,
|
|
CDataType,
|
|
RowMajor,
|
|
ColumnMajor,
|
|
RowMajor,
|
|
PassThrough,
|
|
PassThrough,
|
|
PassThrough>{}(gemmPtr);
|
|
}
|
|
|
|
std::cout << "TestGemm ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
|
|
return res ? 0 : 1;
|
|
}
|