gcc/libm floating-point bug?

[David Schultz]

On Thu, May 22, 2003, Anti wrote:
> 
> > > I feel some will screem if we take away the ability to use
> > > -march=pentium4 in places they know for sure will work.  Unix is about
> > > mechanisms, not policy.
> > 
> > Well, we've got a compiler here with a broken mechanism.  Deciding whether or 
> > not to act on it sounds like a policy decision to me.  I just hope 5.1 
> > doesn't get shipped with such an easy way to break stuff.
> 
> 
> p4 should expand to "-march=pentium4 -mno-sse2" (could set it to pentium3, but
> then people would see pentium3 being used when they specified p4 and think it's
> a bug)...
> 
> any educated person with a legit reason to use -march=pentium4 without disabling
> sse2 can always add it to CFLAGS instead of setting it in CPUTYPE...
> 
> don't see why anyone wouldn't want to put an end to all the breakage and bug
> reports due to this when the fix is so simple and of no real harm...

When I last checked, -march=pentium4 was slower than
-march=pentium3, and -msse2 was pretty much a wash in terms of
performance, even though SSE2 shaved a few bytes off of FP code
size.  This implies that downgrading to -march=pentium3 is the
better option, unless our goal is to dupe people who don't know
well enough to read the documentation.