[llvm] [LV][NFC] Remove unnecessary multiply in expandVPWidenIntOrFpInduction (PR #170159)

[Florian Hahn via llvm-commits]

================
@@ -3454,9 +3454,11 @@ expandVPWidenIntOrFpInduction(VPWidenIntOrFpInductionRecipe *WidenIVR,
     Init = Builder.createWidenCast(Instruction::UIToFP, Init, StepTy);
 
   VPValue *SplatStart = Builder.createNaryOp(VPInstruction::Broadcast, Start);
-  VPValue *SplatStep = Builder.createNaryOp(VPInstruction::Broadcast, Step);
 
-  Init = Builder.createNaryOp(MulOp, {Init, SplatStep}, Flags);
+  if (!match(Step, m_One())) {
+    VPValue *SplatStep = Builder.createNaryOp(VPInstruction::Broadcast, Step);
+    Init = Builder.createNaryOp(MulOp, {Init, SplatStep}, Flags);
+  }
----------------
fhahn wrote:

We have this fold here: https://github.com/llvm/llvm-project/blob/main/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp#L1251

`m_One` should probably look through `m_Broadcast`, or `m_Broadcast(Constant)` should get folded to just `Constant`, so the existing folds take care of this?

https://github.com/llvm/llvm-project/pull/170159