ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-11 01:27:34 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'origin/feature/multiple-d-gemms' into 64-implement-device_gemm_multiply_multiply_instance-for-rdna4

Browse Source
This commit is contained in:
Zoltan Lakatos
2025-06-30 11:06:09 +00:00
parent 8b694c3441 4f1910117b
commit eaa0452b80
11 changed files with 572 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
test/gemm_add/CMakeLists.txt
View File

@@ -39,3 +39,8 @@ add_gtest_executable(test_gemm_multiply_multiply_wmma test_gemm_multiply_multipl
if(result EQUAL 0)
target_link_libraries(test_gemm_multiply_multiply_wmma PRIVATE utility device_gemm_multiply_multiply_instance)
endif()
add_gtest_executable(test_gemm_bilinear_wmma test_gemm_bilinear_wmma.cpp)
if(result EQUAL 0)
target_link_libraries(test_gemm_bilinear_wmma PRIVATE utility device_gemm_bilinear_instance)
endif()

72
test/gemm_add/test_gemm_bilinear_wmma.cpp Normal file
View File

@@ -0,0 +1,72 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include "gtest/gtest.h"
#include "ck/ck.hpp"
#include "profiler/profile_gemm_bilinear_impl.hpp"
#include "test_gemm_common.hpp"
template <typename Tuple>
class TestGemmBilinear : public ::testing::Test
{
private:
using ADataType = std::tuple_element_t<0, Tuple>;
using BDataType = std::tuple_element_t<1, Tuple>;
using AccDataType = std::tuple_element_t<2, Tuple>;
using D0DataType = std::tuple_element_t<3, Tuple>;
using EDataType = std::tuple_element_t<4, Tuple>;
using ALayout = std::tuple_element_t<5, Tuple>;
using BLayout = std::tuple_element_t<6, Tuple>;
using D0Layout = std::tuple_element_t<7, Tuple>;
using ELayout = std::tuple_element_t<8, Tuple>;
constexpr static auto ProfileGemmBilinearImpl =
ck::profiler::profile_gemm_bilinear_impl<ADataType,
BDataType,
AccDataType,
D0DataType,
EDataType,
ALayout,
BLayout,
D0Layout,
ELayout>;
public:
void Run()
{
std::vector<std::vector<ck::index_t>> lengths = {
{16, 32, 64}, {512, 2048, 4096}, {2048, 1024, 16}};
bool all_success = true;
for(auto length : lengths)
{
int M = length[0];
int N = length[1];
int K = length[2];
int StrideA = ck::is_same_v<ALayout, Row> ? K : M;
int StrideB = ck::is_same_v<BLayout, Row> ? N : K;
int StrideD0 = ck::is_same_v<D0Layout, Row> ? N : M;
int StrideE = ck::is_same_v<ELayout, Row> ? N : M;
all_success =
all_success &
ProfileGemmBilinearImpl(
1, 1, false, true, M, N, K, StrideA, StrideB, StrideD0, StrideE, 1.F, 1.F);
}
EXPECT_TRUE(all_success);
}
};
using KernelTypes = ::testing::Types<std::tuple<F16, F16, F32, F16, F16, Row, Row, Row, Row>,
std::tuple<F16, F16, F32, F16, F16, Row, Col, Row, Row>,
std::tuple<F16, F16, F32, F16, F16, Col, Row, Row, Row>,
std::tuple<F16, F16, F32, F16, F16, Col, Col, Row, Row>,
std::tuple<I8, I8, I32, I8, I8, Row, Row, Row, Row>,
std::tuple<I8, I8, I32, I8, I8, Row, Col, Row, Row>,
std::tuple<I8, I8, I32, I8, I8, Col, Row, Row, Row>,
std::tuple<I8, I8, I32, I8, I8, Col, Col, Row, Row>>;
TYPED_TEST_SUITE(TestGemmBilinear, KernelTypes);
TYPED_TEST(TestGemmBilinear, Test) { this->Run(); }

1
test/gemm_add/test_gemm_common.hpp
View File

@@ -8,6 +8,7 @@ using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
using I8 = int8_t;
using I32 = int32_t;
using BF16 = ck::bhalf_t;
using F16 = ck::half_t;
using F32 = float;