This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: DFA scheduler for i386.md
- From: Jan Hubicka <jh at suse dot cz>
- To: law at redhat dot com
- Cc: Vladimir Makarov <vmakarov at redhat dot com>, Jan Hubicka <jh at suse dot cz>,gcc at gcc dot gnu dot org
- Date: Sat, 20 Apr 2002 00:13:34 +0200
- Subject: Re: DFA scheduler for i386.md
- References: <3CC04575.C9769DE4@redhat.com> <6929.1019237274@porcupine.cygnus.com>
> In message <3CC04575.C9769DE4@redhat.com>, Vladimir Makarov writes:
> > I meant i686. Sure, even out-of-order execution/speculative
> > processors has some bottlenecks which can not resolved by insn
> > scheduler. But as for i686, I ran Speck for P4 with and without
> > scheduling several months ago. There is a very small difference (as I
> > remember only 1-2%).
> That's rather odd since I poked at P3 a little I saw 2-4% from scheduling
> on FP benchmarks on i686. It's also the case that 1-2% is nothing to sneeze
> at if we get it consistently across a variety of codes.
THis seems to be consistent with my benchamrks. P3 benefits a lot from
scheduling for decoders as it is limited there and it has long latencies
and limitations on unretired instructions, so in FP it benefits there as well.
In Athlon the benefits are lower. For P4 I don't know, as we don't scheudle
for that chip yet.
Honza
>
> It may be the case that we can't do a lot for P4, I don't really know.
>
>
> jeff
>