This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [Patch, i386] Turn on -fschedule-insns and -fsched-pressure by default for x86-64
- From: Jakub Jelinek <jakub at redhat dot com>
- To: Dominique Dhumieres <dominiq at lps dot ens dot fr>
- Cc: gcc-patches at gcc dot gnu dot org, howarth at bromo dot med dot uc dot edu, sebpop at gmail dot com, hjl dot tools at gmail dot com, Changpeng dot Fang at amd dot com
- Date: Tue, 7 Dec 2010 20:09:16 +0100
- Subject: Re: [Patch, i386] Turn on -fschedule-insns and -fsched-pressure by default for x86-64
- References: <20101207190445.DCD773BE19@mailhost.lps.ens.fr>
- Reply-to: Jakub Jelinek <jakub at redhat dot com>
On Tue, Dec 07, 2010 at 08:04:45PM +0100, Dominique Dhumieres wrote:
> the vectorization still occurs with -fschedule-insns and -fsched-pressure. One possible
> explanation is that with -fschedule-insns and -fsched-pressure the vectorization is done
> on the implicit inner loops of length 3, for a 2/3 speedup, while without them the
> vectorization is done on the inner explicit loop of length 9, for a 5/9 speedup.
> With this ver naive evaluation I get a 6/5 (~20%) slowdown with -fschedule-insns and
That doesn't make any sense, because scheduling is a RTL pass, while vectorization is
done much earlier on GIMPLE.
Jakub