[PATCH] D114541: [CodeGen] Add scalable vector support for lowering of llvm.get.active.lane.mask

[Sjoerd Meijer via Phabricator via llvm-commits]

SjoerdMeijer accepted this revision.
SjoerdMeijer added a comment.
This revision is now accepted and ready to land.

> although the code quality leaves a lot to be desired.

We never want or actually should fallback to this code, it's just a safety net, and thus I don't think we need to worry at all about code quality. Also, this is just a straightforward expansion how get.active.lane.mask is semantically defined in the LLVM langref; thus I am not even sure there's a more optimal sequence (which again I think is irrelevant). At the moment, this intrinsic is emitted by the LV when MVE is targeted, and when it is emitted, we should always lower it to the MVE target specific intrinsic variant. I guess you're going to do the same for SVE. That's the reason why should never reach this expansion here. But this lowering was added here as a safety net in case we for example one day end up with a user facing version of get.active.lane.mask which a target can't or don't know how to lower.

Long story short, LGTM


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D114541/new/

https://reviews.llvm.org/D114541