Is it possible to collect performance metrics in CUDA per region inside the kernel?

Question

I have the following fragment of code for which I want to count the number of executed instructions.

#include 
#include 

__global__ v