ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-10 17:17:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

log print added to splitk benchmarks

Browse Source
This commit is contained in:
Zoltan Lakatos
2025-06-05 13:12:07 +00:00
parent 4e070857ad
commit 8658ca6cf3
3 changed files with 24 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
profiler/include/profiler/profile_gemm_add_fastgelu_impl.hpp
View File

@@ -33,7 +33,7 @@ template <typename ADataType,
typename ELayout>
bool profile_gemm_add_fastgelu_impl(int do_verification,
int init_method,
bool /*do_log*/,
bool do_log,
bool time_kernel,
int M,
int N,
@@ -213,6 +213,17 @@ bool profile_gemm_add_fastgelu_impl(int do_verification,
{
e_device_buf.FromDevice(e_m_n_device_result.mData.data());
if(do_log)
{
LogRangeAsType<float>(
std::cout << "e_m_n_device_result: ", e_m_n_device_result.mData, ",")
<< std::endl;
LogRangeAsType<float>(
std::cout << "e_m_n_host_result: ", e_m_n_host_result.mData, ",")
<< std::endl;
}
pass = pass && ck::utils::check_err(e_m_n_device_result, e_m_n_host_result);
}
}

13
profiler/include/profiler/profile_gemm_add_multiply_impl.hpp
View File

@@ -35,7 +35,7 @@ template <typename ADataType,
typename ELayout>
bool profile_gemm_add_multiply_impl(int do_verification,
int init_method,
bool /*do_log*/,
bool do_log,
bool time_kernel,
int M,
int N,
@@ -223,6 +223,17 @@ bool profile_gemm_add_multiply_impl(int do_verification,
{
e_device_buf.FromDevice(e_m_n_device_result.mData.data());
if(do_log)
{
LogRangeAsType<float>(
std::cout << "e_m_n_device_result: ", e_m_n_device_result.mData, ",")
<< std::endl;
LogRangeAsType<float>(
std::cout << "e_m_n_host_result: ", e_m_n_host_result.mData, ",")
<< std::endl;
}
pass = pass && ck::utils::check_err(e_m_n_device_result, e_m_n_host_result);
}
}

2
test/gemm_add/test_gemm_add_multiply_wmma.cpp
View File

@@ -11,8 +11,6 @@ using Col = ck::tensor_layout::gemm::ColumnMajor;
using F16 = ck::half_t;
using F32 = float;
// TODO: inerit TestGemmAddMultiply from TestGemmD0Common after changes are rebased on top of multipleD feature branch.
// After that clean test...
template <typename Tuple>
class TestGemmAddMultiply : public ::testing::Test
{