How to profile kernel launch overhead?

Developer Tools Nsight Systems Profiling Linux Targets
1

Does cudaLaunchKernel in nsight system reflect the kernel launch overhead?

I’m trying to call cusparseSpMV kernel in the for loop, the loop size can 10^6.
So it will launch kernel this many times. I want to profile the kernel launch overhead.
How can I do this?

2

Please take a look at https://developer.nvidia.com/blog/understanding-the-visualization-of-overhead-and-latency-in-nsight-systems/ where I tried to go into depth on latency.