I've never tried -ffast-math, but it might improve the speed of some parts of the code.
One should be careful with NaNs, as they do occur in certain operations, as they can be in the MS. Those files should be compiled with -ffast-math -fno-finite-math-only.
Diff:
The gridder is now compiled with -ffast-math. MSGridder could not be compiled with -ffast-math because it needs to check for/handle infs/nans. Some of clean algorithms could still be tried.
The clean algorithms have been modified to not use NaNs as magic value and are now also compiled with -ffast-math.