[PATCH] D61525: AMDGPU] gfx1010: prefer V_MUL_LO_U32 over V_MUL_LO_I32

[Nicolai Hähnle via Phabricator via llvm-commits]

nhaehnle added a comment.

In D61525#1490187 <https://reviews.llvm.org/D61525#1490187>, @rampitec wrote:

> In D61525#1490186 <https://reviews.llvm.org/D61525#1490186>, @arsenm wrote:
>
> > What does the signedness of the multiply even mean?
>
>
> It does not. There is problem with the instruction itself.


I believe the goal is to generally use the u32 suffix for arithmetic instructions where the signedness doesn't matter, but the ISA has generally been inconsistent about that and the i32 vs. u32 confusion here is simply because they simultaneously (1) changed some encodings around and (2) changed the canonical mnemonic of the instruction. So the _i32 and _u32 variants are really the same instruction semantically.

(The sign does matter for mul_hi)


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D61525/new/

https://reviews.llvm.org/D61525