This is the mail archive of the
mailing list for the GCC project.
Re: [RFC] Fix PR19401: always completely peel loops
- 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: Giovanni Bajo <giovannibajo at libero dot it>, <gcc-patches at gcc dot gnu dot org>
- Date: Thu, 13 Jan 2005 13:24:38 +0100 (CET)
- Subject: Re: [RFC] Fix PR19401: always completely peel loops
On Thu, 13 Jan 2005, Zdenek Dvorak wrote:
> I am not persuaded that enabling complete loop unrolling unconditionally
> as done by this patch is a good idea.
> I would prefer:
> 1) Unrolling loops completely whenever this does not cause a code
It's hard to prove - especially for the std::pow issue, which is
the first case in the PR19401 audit trail.
> 2) Making all possible loops to be completely unrolled at -O3.
Yes, that makes sense - we already limit code size growth anyway.
Of course this would be done also with the patch to enable
-funroll-loops at -O3.
> 3) Possibly having a separate flag controlling complete loop unrolling.
I'd very much like this at least, if 1) does not work for the testcases.
As I said - previously I used -fpeel-loops for this, but that doesn't
work for the tree unroller -- so simply changing the cunroll guard
to contain flag_unroll_loops || flag_peel_loops would work also.
Richard Guenther <richard dot guenther at uni-tuebingen dot de>