Floating Point Problem ffast-math Complex divide Pentium3

[FX Coudert]

> Option -ffpe-trap does not work on my Pentium3. Should it work on a
> Pentium3? The short program still prints out incorrect results.

It's not related to you chip, I guess -ffpe-trap and -ffast-math have 
incorrect interactions. That should be expected, since the first asks 
for a precise handling of IEEE exceptions, while the second says 
something like "don't bother about IEEE, do what you want but do it quick!".

> The program I extracted the code from runs without an error being raised,
> but has completely different output. My 64 bit athlon64 outputs all Nans.
> My Pentium3 outputs numbers that are close to being correct. Strange ??

Not especially. Agressive optimization is only possible for some types 
of numeric codes, with good stability. On others, it might (and does) 
wreak havoc.

FX