jFP religious wars

[Robert Dewar]

R. D. Flowers wrote:

> Really sort of an inverse of -fast-math -- an insistence on total 
> floating point correctness at the expense of any optimization that would 
> water the correctness down any. Without it, roundings etc. thought to be 
> minor would be allowed to happen in the middle end. I guess.

Rounding is not some pedantic issue, proper rounding means that
you can reason much more clearly about floating-point semantics.