NVIDIA/nvbench
1
0
Fork 0
mirror of https://github.com/NVIDIA/nvbench.git synced 2026-04-20 06:48:53 +00:00
Code Issues Packages Projects Releases Wiki Activity

Run once disables blocking kernel (#252)

Browse Source 
* Measure cold must not use block_kernel for single runs

Per https://github.com/NVIDIA/nvbench/issues/242, we should not
use blocking kernel when --run-once, or --profile is used to avoid
possible deadlocks when providing with external tools, also to avoid
deadlocking when Python programs load the program on the first execution.

* Measure hot should not use blocking kernel during warmup

This change follows suite of measure_cold, where it is prompted
by deadlock, see https://github.com/NVIDIA/nvbench/pull/241

* Remove setting of CUDA_MODULE_LOADING=EAGER

This is no longer necessary as warm-up runs in regular runs,
or the single run in (run-once/profile) no longer use blocking kernel.
This commit is contained in:
Oleksandr Pavlyk
2025-07-28 14:14:54 -05:00
committed by GitHub
parent 3de9dc95da
commit 2ab5e2d1be
3 changed files with 4 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
nvbench/detail/measure_cold.cuh
View File

@@ -220,7 +220,10 @@ private:
void run_trials()
{
kernel_launch_timer timer(*this);
// do not use blocking kernel if benchmark is only run once, e.g., when profiling
// ref: https://github.com/NVIDIA/nvbench/issue/242
const bool disable_blocking_kernel = m_run_once || m_disable_blocking_kernel;
kernel_launch_timer timer(*this, disable_blocking_kernel);
do
{
this->launch_kernel(timer);

9
nvbench/detail/measure_hot.cuh
View File

@@ -106,19 +106,10 @@ private:
// measurement.
void run_warmup()
{
if (!m_disable_blocking_kernel)
{
this->block_stream();
}
m_cuda_timer.start(m_launch.get_stream());
this->launch_kernel();
m_cuda_timer.stop(m_launch.get_stream());
if (!m_disable_blocking_kernel)
{
this->unblock_stream();
}
this->sync_stream();
this->check_skip_time(m_cuda_timer.get_duration());

12
nvbench/main.cuh
View File

@@ -172,20 +172,8 @@
namespace nvbench::detail
{
inline void set_env(const char *name, const char *value)
{
#ifdef _MSC_VER
_putenv_s(name, value);
#else
setenv(name, value, 1);
#endif
}
inline void main_initialize(int, char **)
{
// See NVIDIA/NVBench#136 for CUDA_MODULE_LOADING
set_env("CUDA_MODULE_LOADING", "EAGER");
// Initialize CUDA driver API if needed:
#ifdef NVBENCH_HAS_CUPTI
NVBENCH_DRIVER_API_CALL(cuInit(0));