[PATCH] D143420: AMDGPU: Replace certain llvm.amdgcn.class uses with llvm.is.fpclass

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp:438
+    if (isa<UndefValue>(Src0))
+      return IC.replaceInstUsesWith(II, UndefValue::get(II.getType()));
 
----------------
rampitec wrote:
> foad wrote:
> > I'm not convinced this is sound. E.g. if Src1 is dynamically zero (resp. all-ones) then the result should be false (resp. true) even if Src0 is undef. But that's a pre-existing problem.
> In fact in the wake of Poison replacing Undef I am not sure that isa<UndefValue> is a test to go anymore.
I thought I fixed this before (maybe that’s what I implemented for the generic intrinsic). I can fix this separately 


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D143420/new/

https://reviews.llvm.org/D143420