On Fri, Nov 18, 2016 at 5:57 PM, Bin.Cheng <amker.ch...@gmail.com> wrote:
> On Fri, Nov 18, 2016 at 4:52 PM, Michael Matz <m...@suse.de> wrote:
>> Hi,
>>
>> On Thu, 17 Nov 2016, Bin.Cheng wrote:
>>
>>> B) Depending on ilp, I think below test strings fail for long time with
>>> haswell:
>>> ! { dg-final { scan-tree-dump-times "Executing predictive commoning
>>> without unrolling" 1 "pcom" { target lp64 } } }
>>> ! { dg-final { scan-tree-dump-times "Executing predictive commoning
>>> without unrolling" 2 "pcom" { target ia32 } } }
>>> Because vectorizer choose vf==4 in this case, and there is no
>>> predictive commoning opportunities at all.
>>> Also the newly added test string fails in this case too because the
>>> prolog peeled iterates more than 1 times.
>>
>> Btw, this probably means that on haswell (or other archs with vf==4) mgrid
>> is slower than necessary. On mgrid you really really want predictive
>> commoning to happen. Vectorization isn't that interesting there.
> Interesting, I will check if there is difference between 2/4 vf. we
> do have cases that smaller vf is better and should be chosen, though
> for different reasons.
At some time in the past we had predictive commoning done before
vectorization (GCC 4.3 at least).
Patch is ok meanwhile.
Richard.
> Thanks,
> bin
>>
>>
>> Ciao,
>> Michael.
