[llvm] [VPlan] Use pointer index type for stride in convertToStridedAccesses (PR #199647)

[Luke Lau via llvm-commits]

================
@@ -6912,12 +6895,16 @@ void VPlanTransforms::convertToStridedAccesses(VPlan &Plan,
       VPBuilder Builder(LoadR);
       // Create the base pointer of strided access.
       VPValue *StartVPV = vputils::getOrCreateVPValueForSCEVExpr(Plan, Start);
+      const DataLayout &DL = PSE.getSE()->getDataLayout();
+      Type *IdxTy = DL.getIndexType(TypeInfo.inferScalarType(Ptr));
       VPValue *StrideInBytes =
-          Plan.getConstantInt(VectorLoop->getCanonicalIVType(),
-                              Step->getSExtValue(), /*IsSigned=*/true);
+          Plan.getConstantInt(IdxTy, Step->getSExtValue(), /*IsSigned=*/true);
       auto *AddRecPtr = cast<SCEVAddRecExpr>(PtrSCEV);
+      VPValue *CanIV = Builder.createScalarZExtOrTrunc(
+          VectorLoop->getCanonicalIV(), IdxTy, VectorLoop->getCanonicalIVType(),
+          DebugLoc::getUnknown());
       auto *Offset = Builder.createOverflowingOp(
-          Instruction::Mul, {VectorLoop->getCanonicalIV(), StrideInBytes},
+          Instruction::Mul, {CanIV, StrideInBytes},
----------------
lukel97 wrote:

IIUC every offset should correspond to an address from the original vp.gather. So if the canIV is very large then wouldn't the gather addresses have overflowed anyway? 

https://github.com/llvm/llvm-project/pull/199647