using tic/toc to estimate speed up in C++ and GPU

I am preparing a pilot program which will eventually get transcribed into C++ and some of the code on to a GPU. I want to try to estimate how much speed up I will gain when I do this. However, using tic/toc, some of my matlab code is 'mexed', some is vectorised and some uses slow for loops. There will be varying degrees of speed up for various pieces of code. Is there an easy way to estimate speedup? I note there used to be a matlab function called 'flops' which might have served the purpose. Another way would be to deliberately run everything 'unmexed' and unvectorised and then compare the speedup for simple for loops and then estimate for the more complex cases...

Thanks in advance for any help Chris Walker