This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
AW: gcc 3.0 produces worser code on Ultrasparc
- To: "Alexandre Oliva" <aoliva at redhat dot com>, "Jakub Jelinek" <jakub at redhat dot com>
- Subject: AW: gcc 3.0 produces worser code on Ultrasparc
- From: "Heiko Wengler" <Heiko dot Wengler at do dot isst dot fhg dot de>
- Date: 3 Jul 2001 09:01:18 +0200
- Cc: "Bernd Schmidt" <bernds at redhat dot com>, "gcc at gcc dot gnu dot org" <gcc at gcc dot gnu dot org>
Hello!
> -----Ursprüngliche Nachricht-----
> Von: Alexandre Oliva [mailto:aoliva@redhat.com]
> Gesendet: Monday, July 02, 2001 11:50 PM
> An: Jakub Jelinek
> Cc: Bernd Schmidt; Heiko Wengler; gcc@gcc.gnu.org
> Betreff: Re: gcc 3.0 produces worser code on Ultrasparc
>
>
> On Jul 2, 2001, Jakub Jelinek <jakub@redhat.com> wrote:
>
> > On Mon, Jul 02, 2001 at 07:34:26PM +0100, Bernd Schmidt wrote:
> >> On 2 Jul 2001, Alexandre Oliva wrote:
> >> >
> >> > Just like GCC 2.95.3, GCC 3.0 doesn't fully support
> -mcpu=ultrasparc,
> >> > so all bets are off.
> >>
> >> What are the problems?
>
> > I'm not aware of any (what is not fully supported is -m64, not
> > -mcpu=ultrasparc).
>
> A number of people have been reporting ICEs with -mcpu=ultrasparc in
> GCC 3.0. I thought this would be exercising the same still-incomplete
> code that -m64 would, and so, it was expected that it wouldn't work.
Code optimized with -mv8 or -mcpu=ultrasparc is only 50% as fast as code
produced with
a standard -O2 without the ultrasparc flag.
Heiko