You need to make sure the compiler doesn't optimise your code away thus invalidating your benchmark.
Write to a FP indicator after the benchmarking (i.e. USE the result somewhere) so that the compiler can't simply delete your loop because the result is not being used.
The compiler being this smart makes benchmarking sometimes really difficult and many have stumbled on problems with correct-looking code which simply does not benchmark anything useful.
HERE.
Shane.
