What is acceptable for -ffast-math? A numerical viewpoint

[dewar@gnat.com]

<<Having optimization change program results is a trickier case.


But *all* the "optimizations" you suggest *do* change program results!