[PATCH] D98512: [LoopVectorize] Simplify scalar cost calculation in getInstructionCost

[Sander de Smalen via Phabricator via llvm-commits]

sdesmalen added inline comments.


================
Comment at: llvm/lib/Transforms/Vectorize/LoopVectorize.cpp:7242
 
+#ifndef NDEBUG
+bool LoopVectorizationCostModel::hasSingleCopyAfterVectorization(
----------------
I'd prefer to either remove the ifndef here so that it's always defined (and the compiler can optimize it away if it isn't called), or implementing the lambda suggestion from Florian.


================
Comment at: llvm/lib/Transforms/Vectorize/LoopVectorize.cpp:7557
     // is unknown. Assume that it is the same as 'mul'.
-    return VF.getKnownMinValue() * TTI.getArithmeticInstrCost(
-                                       Instruction::Mul, VectorTy, CostKind) +
+    return VF.getFixedValue() * TTI.getArithmeticInstrCost(Instruction::Mul,
+                                                           VectorTy, CostKind) +
----------------
This should probably return `TTI.getArithmeticInstrCost(..)` now, because the VF.getFixedValue() multiplier and possible scalarization overhead are now represented as part of `VectorTy` and the call to `TTI.getArithmeticInstrCost`. There be no reason why this default opcode should behave differently from the other opcodes now that you've done the work to verify the right type is passed to the `TTI.get`*Cost calls.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D98512/new/

https://reviews.llvm.org/D98512