This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [patch][graphite] Fix code generated by Graphite for vectorization
On Fri, Mar 19, 2010 at 17:27, Richard Guenther
<richard.guenther@gmail.com> wrote:
> On Fri, Mar 19, 2010 at 8:24 PM, Sebastian Pop <sebpop@gmail.com> wrote:
>> On Wed, Mar 17, 2010 at 22:56, Sebastian Pop <sebpop@gmail.com> wrote:
>>> On Wed, Mar 17, 2010 at 22:45, Sebastian Pop <sebpop@gmail.com> wrote:
>>>> After
>>>> ? ? ? ?* Merge from mainline (157009:157519).
>>>>
>>>> I get 33 loops vectorized with -O3 -ffast-math -fgraphite*
>>>> on the Graphite branch as well. ?So it is a change that
>>>> happened only on trunk (VTA changes to graphite maybe?).
>>>
>>> Nop, that's not the VTA change to graphite.
>>> I will delta debug it to figure out which patch does this.
>>>
>>
>> I finally got to this patch, before which we get 40 loops vectorized
>> and after which we only get 26 loops vectorized for gas_dyn.f90
>
> On top of -fgraphite-identity or with plain -O3?
This was with the two patches 0002 and 0003 and with -O3 -ffast-math
-fgraphite-identity
>
>> Author: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
>> Date: ? Fri Mar 5 20:37:44 2010 +0000
>>
>> ? ?Fix PR42326: Handle more carefully convert expressions in
>> chrec_fold_plus and chrec_fold_mult.
>>
>> ? ?2010-03-05 ?Sebastian Pop ?<sebastian.pop@amd.com>
>>
>> ? ? ? ?PR middle-end/42326
>> ? ? ? ?* tree-chrec.c (chrec_fold_plus_1): Do not handle convert expressions
>> ? ? ? ?that contain scevs.
>> ? ? ? ?(chrec_fold_multiply): Same.
>>
>> ? ? ? ?* gfortran.dg/graphite/pr42326.f90: New.
>> ? ? ? ?* gfortran.dg/graphite/pr42326-1.f90: New.
>>
>> ? ?git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@157244
>> 138bc75d-0d04-0410-961f-82ee72b054a4
>>
>> I will try to see if we could improve scev instead of just giving up
>> in these cases.
>>
>> Sebastian
>>
>