[PATCH] D82474: [VectorCombine] try to form vector compare and binop to eliminate scalar ops

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added a comment.

Yes, old x86 targets really struggle with some v2f64 ops - cmppd included.



================
Comment at: llvm/lib/Transforms/Vectorize/VectorCombine.cpp:561
+                                                    : Instruction::ICmp;
+  auto *VecTy = cast<VectorType>(X->getType());
+  int OldCost = TTI.getVectorInstrCost(Ext0->getOpcode(), VecTy, Index0);
----------------
cast<FixedVectorType> (maybe a dyn_cast<>) ?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D82474/new/

https://reviews.llvm.org/D82474