[llvm] r223862 - [ARM] Combine base-updating/post-incrementing vector load/stores.

Ahmed Bougacha ahmed.bougacha at gmail.com
Sat Dec 13 13:40:48 PST 2014


On Sat, Dec 13, 2014 at 1:37 PM, Renato Golin <renato.golin at linaro.org> wrote:
> Hi Ahmed,
>
> Just tested and it seems to work. Just let the bot go back to green to
> commit again.

Thanks for the help! I'll add tests and recommit later.

- Ahmed

> Thanks!
> Renato
>
> On 13 Dec 2014 20:39, "Ahmed Bougacha" <ahmed.bougacha at gmail.com> wrote:
>>
>> I think I found the problem, but I don't have an ARM device to try it
>> out right now.  Examining the assembly generates unbroken code though
>> (the problem seems to be with extloads/truncstores; from what I've
>> seen the problematic instructions in anagram are a 32bit load+vmovl
>> turning into a 128bit postinc load).
>>
>> I attached an updated patch; it would be great if you could try it out.
>>
>> If not I'll run the test on monday and reapply if it works.
>>
>> - Ahmed
>>
>>
>> On Sat, Dec 13, 2014 at 12:25 PM, Renato Golin <renato.golin at linaro.org>
>> wrote:
>> > On 13 December 2014 at 19:07, Ahmed Bougacha <ahmed.bougacha at gmail.com>
>> > wrote:
>> >> Interesting; I suspected this commit was the culprit, since it was the
>> >> only one related to ARM on the blamelist.  I checked locally, and for
>> >> some reason I got the same anagram binary before and after the commit
>> >> range.
>> >
>> > Yeah, so I just ran again on ToT failing and reverting your patch,
>> > succeeding. I've reverted on r224198.
>> >
>> > Let me know if you need help reducing it.
>> >
>> > cheers,
>> > --renato



More information about the llvm-commits mailing list