Merge pull request #250 from oleksandr-pavlyk/measure-cold-with-blocking-kernel-to-start-cpu-timer-in-kernel-timer-start

oleksandr-pavlyk · web-flow · commit 3de9dc95da3f · 2025-07-22T15:32:09.000-05:00
Include host work of benched fn in CPU time when using blocking kernel
diff --git a/nvbench/detail/measure_cold.cuh b/nvbench/detail/measure_cold.cuh
@@ -153,18 +153,16 @@ struct measure_cold_base::kernel_launch_timer
       m_measure.gpu_frequency_start();
     }
     m_measure.m_cuda_timer.start(m_measure.m_launch.get_stream());
-    if (m_disable_blocking_kernel)
-    {
-      m_measure.m_cpu_timer.start();
-    }
+    // start CPU timer irrespective of use of blocking kernel
+    // Ref: https://github.com/NVIDIA/nvbench/issues/249
+    m_measure.m_cpu_timer.start();
   }
 
   __forceinline__ void stop()
   {
     m_measure.m_cuda_timer.stop(m_measure.m_launch.get_stream());
     if (!m_disable_blocking_kernel)
     {
-      m_measure.m_cpu_timer.start();
       m_measure.unblock_stream();
     }
     if (m_measure.m_check_throttling)

Original file line number	Diff line number	Diff line change
`@@ -153,18 +153,16 @@ struct measure_cold_base::kernel_launch_timer`
`153`	`153`	`m_measure.gpu_frequency_start();`
`154`	`154`	`}`
`155`	`155`	`m_measure.m_cuda_timer.start(m_measure.m_launch.get_stream());`
`156`		`- if (m_disable_blocking_kernel)`
`157`		`- {`
`158`		`- m_measure.m_cpu_timer.start();`
`159`		`- }`
	`156`	`+ // start CPU timer irrespective of use of blocking kernel`
	`157`	`+ // Ref: https://github.com/NVIDIA/nvbench/issues/249`
	`158`	`+ m_measure.m_cpu_timer.start();`
`160`	`159`	`}`
`161`	`160`
`162`	`161`	`__forceinline__ void stop()`
`163`	`162`	`{`
`164`	`163`	`m_measure.m_cuda_timer.stop(m_measure.m_launch.get_stream());`
`165`	`164`	`if (!m_disable_blocking_kernel)`
`166`	`165`	`{`
`167`		`- m_measure.m_cpu_timer.start();`
`168`	`166`	`m_measure.unblock_stream();`
`169`	`167`	`}`
`170`	`168`	`if (m_measure.m_check_throttling)`