[PATCH] D130487: [PowerPC] Fix vector_shuffle combines when inputs are scalar_to_vector of differing types.

[Kamau Bridgeman via Phabricator via llvm-commits]

kamaub added inline comments.


================
Comment at: llvm/lib/Target/PowerPC/PPCISelLowering.cpp:15088
+                          LHS.getValueType().getScalarSizeInBits();
+      int LHSScalarSize = SToVLHS.getValueType().getScalarSizeInBits();
+      LHSLastElt = LHSScalarSize / (ShuffleEltWidth + 1);
----------------
nemanjai wrote:
> I think we should have an early exit here if the valid lane width is zero:
> ```
> if (LHSValidLaneWidth == 0)
>   return false;
> ```
> Since it is not really reasonable to do this transform if we are pulling in more bits than the original `scalar_to_vector` actually defined. Similarly with the RHS below.
move `LHSScalarSize` up and use it in this block


================
Comment at: llvm/lib/Target/PowerPC/PPCISelLowering.cpp:15100
+                          RHS.getValueType().getScalarSizeInBits();
+      int RHSScalarSize = SToVRHS.getValueType().getScalarSizeInBits();
+      RHSLastElt = RHSScalarSize / (ShuffleEltWidth + 1) + RHSFirstElt;
----------------
Same here, please raise `RHSScalarSize` and reuse it.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D130487/new/

https://reviews.llvm.org/D130487