[llvm] r347917 - [DAGCombiner] narrow truncated binops

[Friedman, Eli via llvm-commits]

On 12/5/2018 5:47 PM, David Jones wrote:
>
> On Wed, Dec 5, 2018 at 5:40 PM Sanjay Patel <spatel at rotateright.com 
> <mailto:spatel at rotateright.com>> wrote:
>
>     Hi David,
>
>     Thanks for reporting the problem. I don’t have any guesses as to
>     how this could cause OOM. Cc’ing some people that might.
>
>
> +a few more, thanks! :-)
>
>     If the problem is only showing up on PPC, a quick hack would be to
>     add a TLI hook and disable the transform for PPC. I’d prefer that
>     to a full revert (especially since I’ve already enhanced this code
>     to handle some vectors, and another enhancement is in
>     progress)...unless we have a test case that shows it’s not a
>     PPC-specific bug?
>
>
> Agreed on avoiding a revert... this looks like a pretty hard-fought 
> change, based on https://bugs.llvm.org/show_bug.cgi?id=32023.
>
> I do believe this is PPC-specific (although, as is often the case, 
> your change may simply be tickling a bug somewhere else).

If it's PPC-specific, my first guess here would be there's some 
PowerPC-specific target combine that's trying to combine the opposite 
way, or something like that.  Assuming that's the case, it probably runs 
out of memory pretty quickly, so it should be fast to bisect if you 
restrict the memory usage with ulimit.

-Eli

-- 
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20181205/cb6cbda0/attachment.html>