ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-21 21:39:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

use single threaded tensor generator (#161)

Browse Source 
[ROCm/composable_kernel commit: f015c77687]
This commit is contained in:
Anthony Chang
2022-03-31 11:28:30 +08:00
committed by GitHub
parent 297ef9795d
commit 8bb6c6e120
20 changed files with 26 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
profiler/include/profile_batched_gemm_impl.hpp
View File

@@ -103,7 +103,7 @@ bool profile_batched_gemm_impl(int do_verification,
std::cout << "b_g_k_n: " << b_g_k_n.mDesc << std::endl;
std::cout << "c_g_m_n: " << c_g_m_n_host_result.mDesc << std::endl;
std::size_t num_thread = std::thread::hardware_concurrency();
std::size_t num_thread = 1;
switch(init_method)
{
case 0: break;

2
profiler/include/profile_gemm_bias_2d_impl.hpp
View File

@@ -98,7 +98,7 @@ void profile_gemm_bias_2d_impl(int do_verification,
std::cout << "c0_m_n: " << c0_m_n.mDesc << std::endl;
std::cout << "c_m_n: " << c_m_n_host_result.mDesc << std::endl;
std::size_t num_thread = std::thread::hardware_concurrency();
std::size_t num_thread = 1;
switch(init_method)
{
case 0: break;

2
profiler/include/profile_gemm_bias_relu_impl.hpp
View File

@@ -83,7 +83,7 @@ void profile_gemm_bias_relu_impl(int do_verification,
std::cout << "c_m_n: " << c_m_n_host_result.mDesc << std::endl;
std::cout << "c0_n: " << c0_n.mDesc << std::endl;
std::size_t num_thread = std::thread::hardware_concurrency();
std::size_t num_thread = 1;
switch(init_method)
{
case 0: break;

6
profiler/include/profile_gemm_impl.hpp
View File

@@ -120,7 +120,7 @@ void profile_gemm_impl(int do_verification,
std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
std::cout << "c_m_n: " << c_m_n_device_result.mDesc << std::endl;
std::size_t num_thread = std::thread::hardware_concurrency();
std::size_t num_thread = 1;
switch(init_method)
{
case 0: break;
@@ -408,6 +408,10 @@ void profile_gemm_impl(int do_verification,
if(gemm_ptr->IsSupportedArgument(argument_ptr.get()))
{
// re-init C to zero before profiling next kernel
c_m_n_device_result.GenerateTensorValue(GeneratorTensor_0<CDataType>{}, num_thread);
c_device_buf.ToDevice(c_m_n_device_result.mData.data());
std::string gemm_name = gemm_ptr->GetTypeString();
float ave_time = invoker_ptr->Run(argument_ptr.get(), nrepeat);

2
profiler/include/profile_gemm_reduce_impl.hpp
View File

@@ -98,7 +98,7 @@ bool profile_gemm_reduce_impl(int do_verification,
std::cout << "d0_m: " << d0_m_host_result.mDesc << std::endl;
std::cout << "d1_m: " << d1_m_host_result.mDesc << std::endl;
std::size_t num_thread = std::thread::hardware_concurrency();
std::size_t num_thread = 1;
switch(init_method)
{
case 0: break;

2
profiler/include/profile_grouped_gemm_impl.hpp
View File

@@ -95,7 +95,7 @@ void profile_grouped_gemm_impl(int do_verification,
<< "]:" << b_k_n[i].mDesc << ", c_m_n_device_results[" << i
<< "]:" << c_m_n_device_results[i].mDesc << std::endl;
std::size_t num_thread = std::thread::hardware_concurrency();
std::size_t num_thread = 1;
switch(init_method)
{
case 0: break;

2
profiler/include/profile_reduce_impl.hpp
View File

@@ -242,7 +242,7 @@ void profile_reduce_impl_impl(bool do_verification,
size_t invariant_total_length = out.mDesc.GetElementSize();
size_t reduce_total_length = in.mDesc.GetElementSize() / invariant_total_length;
std::size_t num_thread = std::thread::hardware_concurrency();
std::size_t num_thread = 1;
if(do_verification)
{