[PATCH] D84533: AMDGPU/GlobalISel: Legalize s33 G_ADD, G_SUB and G_MUL

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

I’ve been avoiding doing this since it’s really a workaround for narrowScalar not understanding what to do. Would it make more sense for narrow scalar to widen to the next multiple of 32 itself?

In D84533#2172336 <https://reviews.llvm.org/D84533#2172336>, @Petar.Avramovic wrote:

> > Would it make more sense for narrow scalar to widen to the next multiple of 32 itself?
>
> Probably yes, at least for add, sub and mul. It could calculate required scalar size and call widenScalar instead on reporting UnableToLegalize.
>  I am not so sure if this would be correct in general, maybe some opcodes prefer next-power-of-2 instead of multiple of NarrowSize?


This isn't a general property of widenScalar itself, but the add/sub/mul handling knows it can produce the requested size by widening, those implementations can be responsible for inserting an extend to the intermediate size it requires


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D84533/new/

https://reviews.llvm.org/D84533