[PATCH] D41096: [X86] Initial support for prefer-vector-width function attribute

Hal Finkel via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Wed Dec 13 09:54:53 PST 2017


hfinkel added a comment.

In https://reviews.llvm.org/D41096#953559, @octoploid wrote:

> In https://reviews.llvm.org/D41096#953086, @hfinkel wrote:
>
> > In https://reviews.llvm.org/D41096#953039, @craig.topper wrote:
> >
> > > Yes there are two variants of sklake-avx512, but there doesn't seem to be a good way of autodetecting this for march=native.
> >
> >
> > Your optimization manual suggests a relative timing test, so I'm guessing there's not (I wouldn't want to use that for -march=native because it wouldn't be deterministic). As a result, I think we'll just need to make a choice based on some combination of which is likely to be most common among our users and which is likely best on future hardware. Users will need to explicitly specify the architecture to get the other one.
>
>
> There is avx512_2ndFMA (PIROM offset 70h bit 0), see: https://www.intel.com/content/dam/www/public/us/en/documents/datasheets/xeon-scalable-datasheet-vol-1.pdf


Is that accessible in user space somehow?


https://reviews.llvm.org/D41096





More information about the llvm-commits mailing list