ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-19 22:39:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

[CK_Tile] Updating gpu timer when doing flush cache (#2593)

Browse Source 
* Missed updating function names in example

* updating timer

* code cleanup

* addressing review comments

* updating tile_engine code

* addressing review comments
This commit is contained in:
Khushbu Agarwal
2025-07-31 16:43:33 -07:00
committed by GitHub
parent 546ef78d1d
commit 88d72178d6
11 changed files with 54 additions and 139 deletions
Download Patch File Download Diff File Expand all files Collapse all files

77
include/ck_tile/host/timer.hpp
View File

@@ -48,100 +48,31 @@ struct gpu_timer
hipEvent_t start_evt, stop_evt;
};
struct gpu_timer_new
{
CK_TILE_HOST gpu_timer_new(const hipStream_t& s)
{
for(auto& e : start_event)
{
HIP_CHECK_ERROR(hipEventCreate(&e));
}
for(auto& e : stop_event)
{
HIP_CHECK_ERROR(hipEventCreate(&e));
}
HIP_CHECK_ERROR(hipEventCreate(&event0));
HIP_CHECK_ERROR(hipEventRecord(event0, s));
}
CK_TILE_HOST ~gpu_timer_new() noexcept(false)
{
for(auto& e : start_event)
{
HIP_CHECK_ERROR(hipEventDestroy(e));
}
for(auto& e : stop_event)
{
HIP_CHECK_ERROR(hipEventDestroy(e));
}
HIP_CHECK_ERROR(hipEventDestroy(event0));
}
CK_TILE_HOST void start(const hipStream_t& s, int idx = 0)
{
HIP_CHECK_ERROR(hipEventRecord(start_event[idx % 2], s));
}
CK_TILE_HOST void stop(const hipStream_t& s, int idx = 0)
{
HIP_CHECK_ERROR(hipEventRecord(stop_event[idx % 2], s));
}
// return in ms
CK_TILE_HOST float duration(int idx = 0) const
{
float ms;
HIP_CHECK_ERROR(hipEventSynchronize(stop_event[idx % 2]));
HIP_CHECK_ERROR(hipEventElapsedTime(&ms, start_event[idx % 2], stop_event[idx % 2]));
return ms;
}
CK_TILE_HOST float is_exceed(int idx = 0) const
{
float ms;
HIP_CHECK_ERROR(hipEventElapsedTime(&ms, event0, stop_event[idx % 2]));
return ms;
}
private:
std::array<hipEvent_t, 2> start_event;
std::array<hipEvent_t, 2> stop_event;
hipEvent_t event0;
};
struct cpu_timer
{
// torch.utils.benchmark.Timer(), there is a sync inside each timer callback
CK_TILE_HOST void start(const hipStream_t& s, [[maybe_unused]] int idx = 0)
CK_TILE_HOST void start(const hipStream_t& s)
{
HIP_CHECK_ERROR(hipStreamSynchronize(s));
start_tick = std::chrono::high_resolution_clock::now();
time_event0 = std::chrono::high_resolution_clock::now();
start_tick = std::chrono::high_resolution_clock::now();
}
// torch.utils.benchmark.Timer(), there is a sync inside each timer callback
CK_TILE_HOST void stop(const hipStream_t& s, [[maybe_unused]] int idx = 0)
CK_TILE_HOST void stop(const hipStream_t& s)
{
HIP_CHECK_ERROR(hipStreamSynchronize(s));
stop_tick = std::chrono::high_resolution_clock::now();
}
// return in ms
CK_TILE_HOST float duration([[maybe_unused]] int idx = 0) const
CK_TILE_HOST float duration() const
{
double sec =
std::chrono::duration_cast<std::chrono::duration<double>>(stop_tick - start_tick)
.count();
return static_cast<float>(sec * 1e3);
}
// return in ms
CK_TILE_HOST float is_exceed([[maybe_unused]] int idx = 0) const
{
double sec =
std::chrono::duration_cast<std::chrono::duration<double>>(stop_tick - time_event0)
.count();
return static_cast<float>(sec * 1e3);
}
private:
std::chrono::time_point<std::chrono::high_resolution_clock> start_tick;
std::chrono::time_point<std::chrono::high_resolution_clock> time_event0;
std::chrono::time_point<std::chrono::high_resolution_clock> stop_tick;
};