OpenCL profiling function in kernel

Question

As far as I know, the kernel can be profiled by opencl profiling API. So I just get the kernel-level performance. But if the kernel call other function, how i get profiling of this callee.

static B {

}

kernel A {

B();

}

Answer 1

Profile the kernel with the function call and with the function call commented out. Subtract the execution time. However be careful that the compiler doesn't empty the kernel (disable compiler optimization) in case no result is written to global memory. Also be aware of time measurement error and uncertainty, if the function call takes way less time than the rest of the kernel execution; repeat and averagt the time measurements to get more accurate results.