why visual profiler skip some GPU kernel function?

when profiling, some kernel is skipped (not shown in the method column), but I’m pretty sure such kernel get executed in release mode. Because if set timer before and kernel is launched , and after the kernel, the timer tells how much time elapsed for this kernel function.

any idea what might be the reason for visual profiler skipping the kernel?

maybe the kernel quit with an error (unspecified launch failure. 5 sec. limit or writing past the end of an array) You can copy the code from CUDA_SAFE_CALL into your kernel to check for errors, in release mode these macros do nothing, so you would not find out about these errors in release mode.

Great. I did not know how CUDA_SAFE_CALL works, this helps!