[llvm] [VPlan] Fix miscompile when widening induction with loop-variant step (PR #206010)

[Florian Hahn via llvm-commits]

================
@@ -686,7 +720,10 @@ createWidenInductionRecipe(PHINode *Phi, VPPhi *PhiR, VPIRValue *Start,
   // Update wide induction increments to use the same step as the corresponding
   // wide induction. This enables detecting induction increments directly in
   // VPlan and removes redundant splats.
-  if (match(BackedgeVal, m_Add(m_Specific(PhiR), m_VPValue())))
+  bool HasVariantStepIncrement =
+      hasVariantStepIncrement(BackedgeVal, PhiR, IndDesc, PSE, OrigLoop);
----------------
fhahn wrote:

Can check the SCEV itself for invariance, something like
```suggestion
  VPValue *IncStep;
  if (match(BackedgeVal, m_c_Add(m_Specific(PhiR), m_VPValue(IncStep)))) {
    const SCEV *IncStepSCEV = vputils::getSCEVExprForVPValue(IncStep, PSE);
    HasVariantStepIncrement = isa<SCEVCouldNotCompute>(IncStepSCEV) ||
                           !SE.isLoopInvariant(IncStepSCEV, &OrigLoop);
  }
```

https://github.com/llvm/llvm-project/pull/206010