ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-16 08:44:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

print hipOccupancyDefined Grid size for best in ckProfiler

Browse Source
This commit is contained in:
ozturkosu
2025-06-22 19:29:34 -04:00
parent b0e4ef4f96
commit 256aec203e
4 changed files with 37 additions and 61 deletions
Download Patch File Download Diff File Expand all files Collapse all files

53
profiler/include/profiler/profile_gemm_universal_streamk_impl.hpp Executable file → Normal file
View File

@@ -251,14 +251,16 @@ bool profile_gemm_universal_streamk_impl(int do_verification,
float gb_per_sec = num_btype / 1.E6 / ave_time;
// // const auto actual_launch_grid_dims = argument_ptr->GetLaunchGridDims();
// const auto* typed_argument_ptr = dynamic_cast<const GridwiseGemm_xdl_cshuffle_streamk_v3<...>::Argument*>(argument_ptr)
// const auto* typed_argument_ptr = dynamic_cast<const
// GridwiseGemm_xdl_cshuffle_streamk_v3<...>::Argument*>(argument_ptr)
// Get actual launch grid dims from argument
dim3 actual_launch_grid_dims = argument_ptr->GetLaunchGridDims();
std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops
<< " TFlops, " << gb_per_sec << " GB/s, " << op_name << ", Grid_size ";
<< " TFlops, " << gb_per_sec << " GB/s, " << op_name
<< ", Grid_size ";
if(argument_ptr->HasLaunchGridDims() && actual_launch_grid_dims.x > 0)
{
std::cout << actual_launch_grid_dims.x;
@@ -267,31 +269,9 @@ bool profile_gemm_universal_streamk_impl(int do_verification,
{
std::cout << grid_size_curr;
}
std::cout << ", streamk selection strategy " << streamk_sel_curr << std::endl;
// if (typed_argument_ptr)
// {
// const auto actual_launch_grid_dims = typed_argument_ptr->GetLaunchGridDims();
// std::cout << "Actual Grid Dimensions: " << actual_launch_grid_dims.x << "x"
// << actual_launch_grid_dims.y << "x" << actual_launch_grid_dims.z << std::endl;
// }
// else
// {
// std::cerr << "Error: Failed to cast argument_ptr to the correct type." << std::endl;
// }
// std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops
// << " TFlops, " << gb_per_sec << " GB/s, " << op_name << ", Grid_size "
// << actual_launch_grid_dims.x // Use the x-dimension of the actual launch grid
// << ", streamk selection strategy "
// << streamk_sel_curr << std::endl;
// std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops
// << " TFlops, " << gb_per_sec << " GB/s, " << op_name << ", Grid_size "
// << grid_size_curr << ", streamk selection strategy"
// << streamk_sel_curr << std::endl;
#if defined CK_ENABLE_FP8
// set softer tolerances for fp8
if constexpr(is_same_v<ADataType, f8_t> || is_same_v<BDataType, f8_t> ||
@@ -313,11 +293,22 @@ bool profile_gemm_universal_streamk_impl(int do_verification,
if(tflops > best_tflops)
{
best_op_name = op_name;
best_tflops = tflops;
best_ave_time = ave_time;
best_gb_per_sec = gb_per_sec;
best_grid_size = grid_size_curr;
best_op_name = op_name;
best_tflops = tflops;
best_ave_time = ave_time;
best_gb_per_sec = gb_per_sec;
best_grid_size = grid_size_curr;
if(argument_ptr->HasLaunchGridDims() && actual_launch_grid_dims.x > 0)
{
best_grid_size = actual_launch_grid_dims.x;
}
else
{
best_grid_size = grid_size_curr;
}
best_streamk_sel = streamk_sel_curr;
}
}