ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-20 06:49:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add compiler flags for ROCm versions 6.2+ (#1429)

Browse Source 
* add compiler flags to fix compiler issues

* fix typo.

* disable test_smfmac_op on all devices except gfx942

* specify full path to compiler in CI
This commit is contained in:
Illia Silin
2024-08-01 08:27:52 -07:00
committed by GitHub
parent 6648fd3b04
commit d311c95396
3 changed files with 51 additions and 48 deletions
Download Patch File Download Diff File Expand all files Collapse all files

68
test/smfmac_op/smfmac_op_xdl.cpp
View File

@@ -13,6 +13,7 @@
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "test/smfmac_op/smfmac_op_util.hpp"
#include "ck/host_utility/device_prop.hpp"
using BF16 = ck::bhalf_t;
using F16 = ck::half_t;
@@ -38,40 +39,43 @@ class TestSmfmac : public ::testing::Test
void Run()
{
bool pass = true;
constexpr auto matmul_default = ck::smfmac_op_util::matmul<Src1Type,
Src1VecSize,
Src2Type,
Src2VecSize,
GPUAccType,
AccVecSize,
DstType,
M,
N,
K>;
bool pass = true;
if(ck::get_device_name() == "gfx942")
{
constexpr auto matmul_default = ck::smfmac_op_util::matmul<Src1Type,
Src1VecSize,
Src2Type,
Src2VecSize,
GPUAccType,
AccVecSize,
DstType,
M,
N,
K>;
constexpr auto smfmac_kernel_container = std::make_tuple(matmul_default);
ck::static_for<0, std::tuple_size_v<decltype(smfmac_kernel_container)>, 1>{}([&](auto i) {
pass &= ck::smfmac_op_util::TestSmfmac<
std::tuple_element_t<i.value, decltype(smfmac_kernel_container)>,
Src1Type,
Src2Type,
DstType,
GPUAccType,
CPUAccType,
decltype(Row{}),
decltype(Row{}),
decltype(Row{}),
PassThrough,
PassThrough,
PassThrough,
AccVecSize,
M,
N,
K>{}(std::get<ck::Number<i>{}>(smfmac_kernel_container));
});
constexpr auto smfmac_kernel_container = std::make_tuple(matmul_default);
ck::static_for<0, std::tuple_size_v<decltype(smfmac_kernel_container)>, 1>{}(
[&](auto i) {
pass &= ck::smfmac_op_util::TestSmfmac<
std::tuple_element_t<i.value, decltype(smfmac_kernel_container)>,
Src1Type,
Src2Type,
DstType,
GPUAccType,
CPUAccType,
decltype(Row{}),
decltype(Row{}),
decltype(Row{}),
PassThrough,
PassThrough,
PassThrough,
AccVecSize,
M,
N,
K>{}(std::get<ck::Number<i>{}>(smfmac_kernel_container));
});
}
EXPECT_TRUE(pass);
}
};