This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Fix PR40648 -- Fix misaligned store vectorizer patch
- From: Richard Guenther <richard dot guenther at gmail dot com>
- To: Revital1 Eres <ERES at il dot ibm dot com>
- Cc: gcc-patches at gcc dot gnu dot org, ubizjak at gmail dot com
- Date: Tue, 27 Oct 2009 11:32:21 +0100
- Subject: Re: [PATCH] Fix PR40648 -- Fix misaligned store vectorizer patch
- References: <OFB84C78B2.44F012B0-ONC225765C.0031B677-C225765C.0036F3A7@il.ibm.com>
On Tue, Oct 27, 2009 at 11:00 AM, Revital1 Eres <ERES@il.ibm.com> wrote:
>
> Hello,
>
> This patch changes the decision of the vectorizer when to peel for
> alignment.
> The misaligned store vectorizer patch
> (http://gcc.gnu.org/ml/gcc-patches/2009-06/msg00492.html) set the
> decision of when to peel such that peeling was only considered if there
> is an unsupported misaligned access. ?This change caused degradation on
> targets that support misaligned accesses as it seems peeling is preferred
> when the access is misaligned. The attached patch
> tries to fix that by retrieving part of the the original behavior - peeling
> is considered if there is misaligned accesses. (misalign load or store as
> opposed to only stores in the former behavior). The decision of when to
> peel/version etc. should be integrated in the cost model using target
> dependent queries and I intend to address this next.
>
> Bootstrapped with vectorization enabled on powerpc64 and regtested
> on x86_64.
>
> OK for mainline?
Ok.
Thanks,
Richard.
> Thanks,
> Revital
>
> (See attached file: patch_fix_misalign_2710.txt)(See attached file:
> changelog.txt)