This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [autovect] [patch] Make store motion use alias oracle
- From: Zdenek Dvorak <rakdver at atrey dot karlin dot mff dot cuni dot cz>
- To: Revital1 Eres <ERES at il dot ibm dot com>
- Cc: Dorit Nuzman <DORIT at il dot ibm dot com>, Ayal Zaks <ZAKS at il dot ibm dot com>, Victor Kaplansky <VICTORK at il dot ibm dot com>, gcc-patches at gcc dot gnu dot org
- Date: Mon, 12 Mar 2007 16:43:40 +0100
- Subject: Re: [autovect] [patch] Make store motion use alias oracle
- References: <OF0BB46B6A.0C923C74-ONC225729C.00543FA2-C225729C.005448BF@il.ibm.com>
Hello,
> I would like to ask to insert Zdenek Dvorak's make
> store motion use alias oracle patch into autovect branch.
> http://gcc.gnu.org/ml/gcc-patches/2007-01/msg02605.html (with the fix :
> http://gcc.gnu.org/ml/gcc-patches/2007-01/msg02627.html)
>
> This patch should simplify stmts in loops and thus open opportunities
> for vectorization.
>
> This patch inserts additional constrain to the movement_possibility
> function which causes the ifcvt to not be applied when it contains memory
> references (and as a result to some testcases faulires). This problem
> is fixed in the attached patch which was bootstrap and tested on ppc
> together with the above lim patch. (I intend to re-submit this fix once
> the above patch will be inserted to the branch)
>
> (See attached file: patch_fix_ifcvt.txt)
>
> OK for autovect?
I think tree-if-conv.c should not use movement_possibility at all, and
rather use its own function to specify what constructions it handles.
As it does not move anything anywhere, using movement_possibility looks
a bit weird, and it is hard to specify what semantics you exactly
expect.
Zdenek