[PATCH] D83891: AMDGPU/GlobalISel: Try to promote to use packed saturating add/sub

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

In D83891#2199198 <https://reviews.llvm.org/D83891#2199198>, @foad wrote:

> Looks OK technically, but...
>
>> This produces worse results right now for i8 vectors
>
> So is there any benefit to committing this now?

So I don't have another random patch hanging around to handle this one obscure case that I'm sure will get forgotten. This isn't a unique problem, all of the code that legalizes to packed operations is garbage until we have some vector and bit op combines


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D83891/new/

https://reviews.llvm.org/D83891