Re: gcc/libm floating-point bug?

From: David O'Brien <obrien_at_FreeBSD.ORG>
Date: Wed, 21 May 2003 18:10:14 -0700
On Tue, May 20, 2003 at 03:12:27PM -0400, Jon Lido wrote:
> On Tuesday 20 May 2003 02:00 pm, David Schultz wrote:
> > On Tue, May 20, 2003, Jon Lido wrote:
> > > Well, I do have a P4, and had built everything with -march=pentium4. 
> > > However, rebuilding the kernel and modules with -march=pentium3 produces
> > > the same results.
> >
> > This isn't a kernel problem, so you need to rebuild libm and libc
> > without -march=pentium4.  You really don't want to be using the
> > Pentium 4 optimizations in gcc 3.2 anyway; the generated code is
> > generally slower.  gcc 3.3 has fixes for a number of the bugs, but
> > I don't know about the performance problems.
> 
> Yes, this was the problem.  I rebuilt world with -march=pentium3 and that did 
> the trick.

Honest question of you -- I'll assume you're subscribed to
freebsd-current_at_.  How have you missed all the warnings from myself and
others not to trust the -march=pentium4 optimizations?  I honestly want
to know so we can figure out a better way of getting the word out.


> I'm not sure how CPUTYPE gets handled, but perhaps p4 should expand to 
> -march=pentium3, if possible.

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.
Received on Wed May 21 2003 - 16:10:30 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:08 UTC