PPro patch for egcs

[H.J. Lu]

> > +
> 
> I think that you misunderstood my comment.  The problem is that the
> Intel floating point conditional move instructions are not capable
> of checking the overflow bit.  This prevents them from being used
> in situtations requiring signed integer comparisons.  I've submitted
> a patch to prevent them from being used in this situation.
> 

I think there may be a better way than disabling the signed integer
comparisons for floating point conditional move all together. We may
be able to translate CC into something fcmovxx can handle like what I
did for fcomi. But it may require changes in other places in the x86
backend. I will give it a try this weekend.

-- 
H.J. Lu (hjl@gnu.org)