[llvm] 4b33b23 - Reapply "[LV] Vectorize (some) early and multiple exit loops"" w/fix for builder

[Philip Reames via llvm-commits]

On 4/9/21 2:44 PM, Florian Hahn wrote:
> Hi,
>
>> On 8 Apr 2021, at 19:17, Philip Reames <listmail at philipreames.com 
>> <mailto:listmail at philipreames.com>> wrote:
>>
>> This isn't a regression in this change, we're exposing an existing 
>> issue at higher frequency.  I reduced a test which doesn't involve 
>> this change, and filed a bug for it. 
>> https://bugs.llvm.org/show_bug.cgi?id=49900
>>
>>
> Oh that’s interesting, thanks for taking a look. I think there might 
> be an existing bug for a similar/identical issue. Let me try to find 
> it and link it in Guggilla.
>>
>> Florian, I'm willing to take a shot at fixing this, but need a bit of 
>> help.  Would you have a couple minutes to debate ideas on how the 
>> vectorizer should be structured to avoid querying SCEV over partially 
>> constructed IR?
>>
> Unfortunately I won’t have time in the next few days, but I’ll try to 
> take a look early next week. Perhaps it would be worth syncing up mid 
> next week?

Sure.  I don't see this as urgent, so simply suggest a time late next 
week which works for you.

One near term question.  I have a slightly ugly workaround (described on 
the bug), for this particular manifestation of the broader issue.  Do 
you think it's worth working around?  (e.g. Did the frequency change by 
enough to be concerning?)

>
> Cheers,
> Florian
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210409/6f9ce008/attachment.html>