[PATCH] D151340: AMDGPU: Refine undef handling for llvm.amdgcn.class intrinsic

[Jay Foad via Phabricator via llvm-commits]

foad added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp:471
 
+    // llvm.amdgcn.class(undef, _) -> undef
+    if (IC.getSimplifyQuery().isUndefValue(Src0))
----------------
arsenm wrote:
> foad wrote:
> > I don't think this is sound, e.g if the RHS is 0 (but not a ConstantInt) then the result should be 0, not undef.
> > 
> > Perhaps you could fold it to `RHS != 0`?
> icmp ne x, undef folds to undef, so that would be the same thing 
No we're talking about the case where LHS is undef.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D151340/new/

https://reviews.llvm.org/D151340