When I got the profile result with Nsight System, I found that ‘cudaEventRecord’ .
I didn’t call ‘cudaEventRecord’. Why is this kernel called?
Are these calls automatically called by the profiler?
It looks like you are making library calls, such as to cudnn.
CUDA libraries may use any CUDA API calls, including cudaEventRecord
Is this included in debugging?
Can I exclude “cudaEventRecord”?
I don’t know what that means. There are not separate versions of these libraries (e.g. cudnn) for debug vs. release builds.
If a cudaEventRecord is called from a library call that your code makes, I don’t know of any way to “exclude” it.