[PATCH] D126475: [AMDGPU] gfx11 vop3 and inherited vop instructions

[Dmitry Preobrazhensky via Phabricator via llvm-commits]

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

LGTM, though I’d have preferred to add tests for opcodes which have zero test coverage.



================
Comment at: llvm/lib/Target/AMDGPU/VOPInstructions.td:969
+}
+// In VOP1, we can have clamp and omod even if !HasModifiers
 class getVOP3Pat<VOPProfile P, SDPatternOperator node> {
----------------
Joe_Nash wrote:
> dp wrote:
> > This sounds rather cryptic to me. Why VOP1 only?
> In vop2, HasModifiers = HasModifiers || HasClamp || HasOmod
> In vop1, HasModifiers can be false if HasClamp or HasOmod is true. It is not ideal they are defined this way, but since this class aims to be a common interface for both, it must support both those cases. A later patch could refactor this to be more clear.
I see, thanks!


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D126475/new/

https://reviews.llvm.org/D126475