This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [patch] for PR19401
- From: Richard Guenther <rguenth at tat dot physik dot uni-tuebingen dot de>
- To: Zdenek Dvorak <rakdver at atrey dot karlin dot mff dot cuni dot cz>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Fri, 14 Jan 2005 10:38:32 +0100 (CET)
- Subject: Re: [patch] for PR19401
On Fri, 14 Jan 2005, Zdenek Dvorak wrote:
> Hello,
>
> following the discussions on the gcc mailing list, there is another
> proposal for the fix for this PR:
>
> 1) Always unroll loop completely if it does not cause code growth
> (of course this is heuristics only, so it may be possible to
> construct examples where the code size is increased, but it should
> not be common).
> 2) Enables complete unrolling of all loops at -O3 and with -fpeel-loops.
Ok, this certainly solves my problems in a way I do anyway with gcc 3.4.
Is it possible to schedule another sra pass after loop only if we
completely unrolled some loop in a function? This would maybe help
PR18754 and should be not too intrusive if it can be conditionalized
on a previously completely unrolled loop. But of course this works
only if we go through all passes per-function (or BB) -- do we?
Can you comment on
http://gcc.gnu.org/ml/gcc-patches/2005-01/msg00656.html
which is originally your patch, still suitable for 3.4?
Thanks,
Richard.
--
Richard Guenther <richard dot guenther at uni-tuebingen dot de>
WWW: http://www.tat.physik.uni-tuebingen.de/~rguenth/