ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-06-06 15:54:31 +00:00
Code Issues Packages Projects Releases Wiki Activity

added tests for all splitk fastgelus

Browse Source
This commit is contained in:
Zoltan Lakatos
2025-06-12 11:43:26 +00:00
parent 264e1b238a
commit b4d3e4112c
7 changed files with 204 additions and 37 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_wmma_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
View File

@@ -24,10 +24,10 @@ static constexpr auto Intrawave = BlockGemmPipelineScheduler::Intrawave;
static constexpr auto V1 = BlockGemmPipelineVersion::v1;
static constexpr auto V3 = BlockGemmPipelineVersion::v3;
// e = elementwise((a * b), d0)
// elementwise(c, d0) = fastgelu(c + d0)
// e = elementwise((a * b), d0, d1)
// elementwise(c, d0, d1) = fastgelu(c + d0 + d1)
// output: e[m, n]
// input: a[m, k], b[n, k], d0[m, n]
// input: a[m, k], b[n, k], d0[m, n], d1[m, n]
template <GemmSpecialization GemmSpec>
using device_gemm_add_add_fastgelu_wmma_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instances =

10
test/gemm_add/CMakeLists.txt
View File

@@ -20,7 +20,17 @@ if(result EQUAL 0)
target_link_libraries(test_gemm_add_fastgelu_xdl PRIVATE utility device_gemm_add_instance device_gemm_add_fastgelu_instance)
endif()
add_gtest_executable(test_gemm_fastgelu_wmma test_gemm_fastgelu_wmma.cpp)
if(result EQUAL 0)
target_link_libraries(test_gemm_fastgelu_wmma PRIVATE utility device_gemm_fastgelu_instance)
endif()
add_gtest_executable(test_gemm_add_fastgelu_wmma test_gemm_add_fastgelu_wmma.cpp)
if(result EQUAL 0)
target_link_libraries(test_gemm_add_fastgelu_wmma PRIVATE utility device_gemm_add_fastgelu_instance)
endif()
add_gtest_executable(test_gemm_add_add_fastgelu_wmma test_gemm_add_add_fastgelu_wmma.cpp)
if(result EQUAL 0)
target_link_libraries(test_gemm_add_add_fastgelu_wmma PRIVATE utility device_gemm_add_add_fastgelu_instance)
endif()

38
test/gemm_add/test_gemm_add_add_fastgelu_wmma.cpp Normal file
View File

@@ -0,0 +1,38 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
#include "gtest/gtest.h"
#include "ck/ck.hpp"
#include "profiler/profile_gemm_add_add_fastgelu_impl.hpp"
#include "test_gemm_common.hpp"
template <typename Tuple>
class TestGemmAddAddFastgelu : public TestGemmD0D1Common<Tuple>
{
using ProfileCall = typename TestGemmD0D1Common<Tuple>::ProfileCall;
public:
ProfileCall GetImpl() override
{
return ck::profiler::profile_gemm_add_add_fastgelu_impl<
typename TestGemmD0D1Common<Tuple>::ADataType,
typename TestGemmD0D1Common<Tuple>::BDataType,
typename TestGemmD0D1Common<Tuple>::AccDataType,
typename TestGemmD0D1Common<Tuple>::D0DataType,
typename TestGemmD0D1Common<Tuple>::D1DataType,
typename TestGemmD0D1Common<Tuple>::EDataType,
typename TestGemmD0D1Common<Tuple>::ALayout,
typename TestGemmD0D1Common<Tuple>::BLayout,
typename TestGemmD0D1Common<Tuple>::D0Layout,
typename TestGemmD0D1Common<Tuple>::D1Layout,
typename TestGemmD0D1Common<Tuple>::ELayout>;
}
};
using KernelTypes = ::testing::Types<std::tuple<F16, F16, F32, F16, F16, F16, Row, Row, Row, Row, Row>,
std::tuple<F16, F16, F32, F16, F16, F16, Row, Col, Row, Row, Row>,
std::tuple<F16, F16, F32, F16, F16, F16, Col, Row, Row, Row, Row>,
std::tuple<F16, F16, F32, F16, F16, F16, Col, Col, Row, Row, Row>>;
TYPED_TEST_SUITE(TestGemmAddAddFastgelu, KernelTypes);
TYPED_TEST(TestGemmAddAddFastgelu, Test_BF16FP16) { this->Run(); }

36
test/gemm_add/test_gemm_add_fastgelu_wmma.cpp
View File

@@ -9,29 +9,21 @@
template <typename Tuple>
class TestGemmAddFastgelu : public TestGemmD0Common<Tuple>
{
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>;
using ProfileCall = typename TestGemmD0Common<Tuple>::ProfileCall;
constexpr static auto ProfileGemmAddFastgeluImpl =
ck::profiler::profile_gemm_add_fastgelu_impl<ADataType,
BDataType,
AccDataType,
D0DataType,
EDataType,
ALayout,
BLayout,
D0Layout,
ELayout>;
decltype(ProfileGemmAddFastgeluImpl) GetImpl() override { return ProfileGemmAddFastgeluImpl; }
ProfileCall GetImpl() override
{
return ck::profiler::profile_gemm_add_fastgelu_impl<
typename TestGemmD0Common<Tuple>::ADataType,
typename TestGemmD0Common<Tuple>::BDataType,
typename TestGemmD0Common<Tuple>::AccDataType,
typename TestGemmD0Common<Tuple>::D0DataType,
typename TestGemmD0Common<Tuple>::EDataType,
typename TestGemmD0Common<Tuple>::ALayout,
typename TestGemmD0Common<Tuple>::BLayout,
typename TestGemmD0Common<Tuple>::D0Layout,
typename TestGemmD0Common<Tuple>::ELayout>;
}
};
using KernelTypes = ::testing::Types<std::tuple<F16, F16, F32, F16, F16, Row, Row, Row, Row>,

16
test/gemm_add/test_gemm_add_fastgelu_xdl.cpp
View File

@@ -9,6 +9,22 @@
template <typename Tuple>
class TestGemmAddFastgelu : public TestGemmD0Common<Tuple>
{
using ProfileCall = typename TestGemmD0Common<Tuple>::ProfileCall;
ProfileCall GetImpl() override
{
return ck::profiler::profile_gemm_add_fastgelu_impl<
typename TestGemmD0Common<Tuple>::ADataType,
typename TestGemmD0Common<Tuple>::BDataType,
typename TestGemmD0Common<Tuple>::AccDataType,
typename TestGemmD0Common<Tuple>::D0DataType,
typename TestGemmD0Common<Tuple>::EDataType,
typename TestGemmD0Common<Tuple>::ALayout,
typename TestGemmD0Common<Tuple>::BLayout,
typename TestGemmD0Common<Tuple>::D0Layout,
typename TestGemmD0Common<Tuple>::ELayout>;
}
private:
using ADataType = std::tuple_element_t<0, Tuple>;
using BDataType = std::tuple_element_t<1, Tuple>;

103
test/gemm_add/test_gemm_common.hpp
View File

@@ -3,7 +3,6 @@
#include "gtest/gtest.h"
#include "ck/ck.hpp"
#include "profiler/profile_gemm_add_impl.hpp"
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
@@ -13,6 +12,47 @@ using BF16 = ck::bhalf_t;
using F16 = ck::half_t;
using F32 = float;
template <typename Tuple>
class TestGemmCommon : public ::testing::Test
{
protected:
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 EDataType = std::tuple_element_t<3, Tuple>;
using ALayout = std::tuple_element_t<4, Tuple>;
using BLayout = std::tuple_element_t<5, Tuple>;
using ELayout = std::tuple_element_t<6, Tuple>;
using ProfileCall = bool(*const)(int, int, bool, bool, int, int, int, int, int, int);
virtual ProfileCall GetImpl() = 0;
void Run()
{
std::vector<std::vector<ck::index_t>> lengths = {
{16, 32, 64}, {2048, 4096, 8192}, {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 StrideE = ck::is_same_v<ELayout, Row> ? N : M;
all_success =
all_success &
GetImpl()(1, 1, false, false, M, N, K, StrideA, StrideB, StrideE);
}
EXPECT_TRUE(all_success);
}
};
template <typename Tuple>
class TestGemmD0Common : public ::testing::Test
{
@@ -27,17 +67,9 @@ class TestGemmD0Common : public ::testing::Test
using D0Layout = std::tuple_element_t<7, Tuple>;
using ELayout = std::tuple_element_t<8, Tuple>;
constexpr static auto ProfileGemmAddImpl = ck::profiler::profile_gemm_add_impl<ADataType,
BDataType,
AccDataType,
D0DataType,
EDataType,
ALayout,
BLayout,
D0Layout,
ELayout>;
using ProfileCall = bool(*const)(int, int, bool, bool, int, int, int, int, int, int, int);
virtual decltype(ProfileGemmAddImpl) GetImpl() = 0;
virtual ProfileCall GetImpl() = 0;
void Run()
{
@@ -58,7 +90,54 @@ class TestGemmD0Common : public ::testing::Test
all_success =
all_success &
GetImpl()(true, 1, false, false, M, N, K, StrideA, StrideB, StrideD0, StrideE);
GetImpl()(1, 1, false, false, M, N, K, StrideA, StrideB, StrideD0, StrideE);
}
EXPECT_TRUE(all_success);
}
};
template <typename Tuple>
class TestGemmD0D1Common : public ::testing::Test
{
protected:
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 D1DataType = std::tuple_element_t<4, Tuple>;
using EDataType = std::tuple_element_t<5, Tuple>;
using ALayout = std::tuple_element_t<6, Tuple>;
using BLayout = std::tuple_element_t<7, Tuple>;
using D0Layout = std::tuple_element_t<8, Tuple>;
using D1Layout = std::tuple_element_t<9, Tuple>;
using ELayout = std::tuple_element_t<10, Tuple>;
using ProfileCall = bool(*const)(int, int, bool, bool, int, int, int, int, int, int, int, int);
virtual ProfileCall GetImpl() = 0;
void Run()
{
std::vector<std::vector<ck::index_t>> lengths = {
{16, 32, 64}, {2048, 4096, 8192}, {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 StrideD1 = ck::is_same_v<D1Layout, Row> ? N : M;
int StrideE = ck::is_same_v<ELayout, Row> ? N : M;
all_success =
all_success &
GetImpl()(1, 1, false, false, M, N, K, StrideA, StrideB, StrideD0, StrideD1, StrideE);
}
EXPECT_TRUE(all_success);

32
test/gemm_add/test_gemm_fastgelu_wmma.cpp Normal file
View File

@@ -0,0 +1,32 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
#include "gtest/gtest.h"
#include "ck/ck.hpp"
#include "profiler/profile_gemm_fastgelu_impl.hpp"
#include "test_gemm_common.hpp"
template <typename Tuple>
class TestGemmFastgelu : public TestGemmCommon<Tuple>
{
using ProfileCall = typename TestGemmCommon<Tuple>::ProfileCall;
ProfileCall GetImpl() override
{
return ck::profiler::profile_gemm_fastgelu_impl<typename TestGemmCommon<Tuple>::ADataType,
typename TestGemmCommon<Tuple>::BDataType,
typename TestGemmCommon<Tuple>::AccDataType,
typename TestGemmCommon<Tuple>::EDataType,
typename TestGemmCommon<Tuple>::ALayout,
typename TestGemmCommon<Tuple>::BLayout,
typename TestGemmCommon<Tuple>::ELayout>;
}
};
using KernelTypes = ::testing::Types<std::tuple<F16, F16, F32, F16, Row, Row, Row>,
std::tuple<F16, F16, F32, F16, Row, Col, Row>,
std::tuple<F16, F16, F32, F16, Col, Row, Row>,
std::tuple<F16, F16, F32, F16, Col, Col, Row>>;
TYPED_TEST_SUITE(TestGemmFastgelu, KernelTypes);
TYPED_TEST(TestGemmFastgelu, Test_BF16FP16) { this->Run(); }