CUDA Visual profiler Use on early verson of final program?

I have a very long program that I wish to speed up using CUDA/NVIDIA. I have profiled it using the Linux c profiler, but that seems to be somewhat limiting. Sooner or later I will go to the CUDA Viusal profiler. I was wondering if I could use the Cudaprof now to profile this c program that has no CUDA code in it.

I understand that I will get no CUDA outputs, but that is okay, since all I want to know for now is the location of the program’s bottlenecks.

In the future when subprogram(s) are re-written in CUDA, then I will get the CUDA output info. I just want to get used to using the CUDA Visual profiler for now.

I am guessing that I will have to compile using nvcc instead of gcc so that the cuda visual profiler will work on this code. Is this correct?


The CUDA visual profiler only instruments and profiles code which runs on the GPU. You can’t use it for host code, and it doesn’t profile host code in CUDA programs either.