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

[dewar@gnat.com]

<<An n% increase in speed makes n hours on a 4-day run. But then you're
right that it is probably not worth the effort to implement optimizations
that get you 1 or 2 per cent. If it's 10%, it starts getting interesting.
>>

warning: 10% is a VERY heavy burden, very few individual optimizations
can make this much effect, so I would actually accept a much smaller
threshhold. For example, if the a/b/c transformation made a 2% imrpovement
to established programs, quake or SPEC :-) then it is on the radar as
significant.