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

Grouped Conv Bwd Data index calculation optimizations

Browse Source
This commit is contained in:
Bartlomiej Kocot
2025-09-23 21:50:42 +00:00
parent d876e87fe4
commit 0bdc7670eb
6 changed files with 404 additions and 79 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
profiler/include/profiler/profile_grouped_conv_bwd_data_impl.hpp
View File

@@ -137,8 +137,8 @@ bool profile_grouped_conv_bwd_data_impl(int do_verification,
auto invoker_ptr = op_ptr->MakeInvokerPointer();
float avg_time =
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
float avg_time = invoker_ptr->Run(argument_ptr.get(),
StreamConfig{nullptr, time_kernel, 0, 500, 100});
std::size_t flop = conv_param.GetFlops();
std::size_t num_btype = conv_param.GetByte<InDataType, WeiDataType, OutDataType>();
@@ -185,11 +185,17 @@ bool profile_grouped_conv_bwd_data_impl(int do_verification,
// Use higher threshold
rtol = std::max(rtol, rtol_split_k);
atol = std::max(atol, atol_split_k);
pass &= ck::utils::check_err(
in_device, in_host, "Error: Incorrect results!", rtol, atol);
std::cout << "Relative error threshold: " << rtol
<< " Absolute error threshold: " << atol << std::endl;
if(split_k_for_run > 1)
{
pass &= ck::utils::check_err(
in_device, in_host, "Error: Incorrect results!", rtol, atol);
std::cout << "Relative error threshold: " << rtol
<< " Absolute error threshold: " << atol << std::endl;
}
else
{
pass &= ck::utils::check_err(in_device, in_host, "Error: Incorrect results!");
}
if(do_log)
{