NVIDIA/cutlass
1
0
Fork 0
mirror of https://github.com/NVIDIA/cutlass.git synced 2026-05-21 21:38:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

3.6.0 update (#2005)

Browse Source 
* 3.6.0 update

* doc and swap stuff

---------

Co-authored-by: yuzhai <yuzhai@nvidia.com>
Co-authored-by: Haicheng Wu <haichengw@nvidia.com>
This commit is contained in:
Yujia Zhai
2024-12-24 22:34:40 -08:00
committed by GitHub
parent e1cd8c7866
commit 3d261a5974
258 changed files with 10863 additions and 3883 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
tools/profiler/src/gpu_timer.cpp
View File

@@ -33,9 +33,11 @@
*/
#include <stdexcept>
#include <cstring>
#include "cutlass/profiler/gpu_timer.h"
namespace cutlass {
namespace profiler {
@@ -52,32 +54,39 @@ GpuTimer::GpuTimer() {
}
}
GpuTimer::GpuTimer(GpuTimer&& gpu_timer) noexcept {
memcpy(events, gpu_timer.events, sizeof(events));
memset(gpu_timer.events, 0, sizeof(gpu_timer.events));
}
GpuTimer::~GpuTimer() {
for (auto & event : events) {
cudaEventDestroy(event);
for (const auto & event : events) {
if (event != nullptr) {
cudaEventDestroy(event);
}
}
}
/// Records a start event in the stream
void GpuTimer::start(cudaStream_t stream) {
cudaError_t result = cudaEventRecord(events[0], stream);
/// Records a start event in the stream, the flag is for cudaEventRecordWithFlags
void GpuTimer::start(cudaStream_t stream, const unsigned int flag) {
cudaError_t result = cudaEventRecordWithFlags(events[0], stream, flag);
if (result != cudaSuccess) {
throw std::runtime_error("Failed to record start event.");
}
}
/// Records a stop event in the stream
void GpuTimer::stop(cudaStream_t stream) {
cudaError_t result = cudaEventRecord(events[1], stream);
/// Records a stop event in the stream, the flag is for cudaEventRecordWithFlags
void GpuTimer::stop(cudaStream_t stream, const unsigned int flag) {
cudaError_t result = cudaEventRecordWithFlags(events[1], stream, flag);
if (result != cudaSuccess) {
throw std::runtime_error("Failed to record stop event.");
}
}
/// Records a stop event in the stream and synchronizes on the stream
void GpuTimer::stop_and_wait(cudaStream_t stream) {
/// Records a stop event in the stream and synchronizes on the stream, the flag is for cudaEventRecordWithFlags
void GpuTimer::stop_and_wait(cudaStream_t stream, const unsigned int flag) {
stop(stream);
stop(stream, flag);
cudaError_t result;
if (stream) {