[Fix] Pass stride in element size

Author: Mel-Chen

llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp

@@ -3259,9 +3259,12 @@ void VPWidenStridedLoadRecipe::execute(VPTransformState &State) {
 
   auto *PtrTy = Addr->getType();
   auto *StrideTy = Stride->getType();
+  const DataLayout &DL = Ingredient.getDataLayout();
+  Value *StrideInBytes = Builder.CreateMul(
+      Stride, ConstantInt::get(StrideTy, DL.getTypeAllocSize(ScalarDataTy)));
   CallInst *NewLI = Builder.CreateIntrinsic(
       Intrinsic::experimental_vp_strided_load, {DataTy, PtrTy, StrideTy},
-      {Addr, Stride, Mask, RunTimeVF}, nullptr, "wide.strided.load");
+      {Addr, StrideInBytes, Mask, RunTimeVF}, nullptr, "wide.strided.load");
   NewLI->addParamAttr(
       0, Attribute::getWithAlignment(NewLI->getContext(), Alignment));
   applyMetadata(*NewLI);

llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp

@@ -2743,8 +2743,8 @@ void VPlanTransforms::convertToStridedAccesses(VPlan &Plan, VPCostContext &Ctx,
       auto *LI = cast<LoadInst>(&Ingredient);
       const DataLayout &DL = LI->getDataLayout();
       auto *StrideTy = DL.getIndexType(LI->getPointerOperand()->getType());
-      VPValue *StrideVPV = Plan.getOrAddLiveIn(
-          ConstantInt::get(StrideTy, Stride * DL.getTypeAllocSize(ElementTy)));
+      VPValue *StrideVPV =
+          Plan.getOrAddLiveIn(ConstantInt::get(StrideTy, Stride));
       auto *StridedLoad = new VPWidenStridedLoadRecipe(
           *LI, NewPtr, StrideVPV, &Plan.getVF(), LoadR->getMask(), *LoadR,
           LoadR->getDebugLoc());

llvm/test/Transforms/LoopVectorize/RISCV/riscv-vector-reverse.ll

@@ -104,7 +104,7 @@ define void @vector_reverse_i64(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; CHECK-NEXT:      CLONE ir<[[ZEXT_IDX:%.+]]> = zext ir<[[IDX]]>
 ; CHECK-NEXT:      CLONE ir<[[LD_IDX:%.+]]> = getelementptr inbounds ir<%B>, ir<[[ZEXT_IDX]]>
 ; CHECK-NEXT:      vp<[[LD_PTR:%.+]]> = vector-pointer ir<[[LD_IDX]]>
-; CHECK-NEXT:      WIDEN ir<[[LD:%.+]]> = load vp<[[LD_PTR]]>, stride = ir<-4>, runtimeVF = vp<[[VF]]>
+; CHECK-NEXT:      WIDEN ir<[[LD:%.+]]> = load vp<[[LD_PTR]]>, stride = ir<-1>, runtimeVF = vp<[[VF]]>
 ; CHECK-NEXT:      WIDEN ir<[[ADD:%.+]]> = add ir<[[LD]]>, ir<1>
 ; CHECK-NEXT:      CLONE ir<[[ST_IDX:%.+]]> = getelementptr inbounds ir<%A>, ir<[[ZEXT_IDX]]>
 ; CHECK-NEXT:      vp<[[ST_PTR:%.+]]> = vector-end-pointer inbounds ir<[[ST_IDX]]>, vp<[[VF]]>
@@ -238,7 +238,7 @@ define void @vector_reverse_i64(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; CHECK-NEXT:    CLONE ir<[[ZEXT_IDX:%.+]]> = zext ir<[[IDX]]>
 ; CHECK-NEXT:    CLONE ir<[[LD_IDX:%.+]]> = getelementptr inbounds ir<%B>, ir<[[ZEXT_IDX]]>
 ; CHECK-NEXT:    vp<[[LD_PTR:%.+]]> = vector-pointer ir<[[LD_IDX]]>
-; CHECK-NEXT:    WIDEN ir<[[LD:%.+]]> = load vp<[[LD_PTR]]>, stride = ir<-4>, runtimeVF = ir<[[VF]]>
+; CHECK-NEXT:    WIDEN ir<[[LD:%.+]]> = load vp<[[LD_PTR]]>, stride = ir<-1>, runtimeVF = ir<[[VF]]>
 ; CHECK-NEXT:    WIDEN ir<[[ADD:%.+]]> = add ir<[[LD]]>, ir<1>
 ; CHECK-NEXT:    CLONE ir<[[ST_IDX:%.+]]> = getelementptr inbounds ir<%A>, ir<[[ZEXT_IDX]]>
 ; CHECK-NEXT:    vp<[[ST_PTR:%.+]]> = vector-end-pointer inbounds ir<[[ST_IDX]]>, ir<[[VF]]>
@@ -511,7 +511,7 @@ define void @vector_reverse_f32(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; CHECK-NEXT:      CLONE ir<[[ZEXT_IDX:%.+]]> = zext ir<[[IDX]]>
 ; CHECK-NEXT:      CLONE ir<[[LD_IDX:%.+]]> = getelementptr inbounds ir<%B>, ir<[[ZEXT_IDX]]>
 ; CHECK-NEXT:      vp<[[LD_PTR:%.+]]> = vector-pointer ir<[[LD_IDX]]>
-; CHECK-NEXT:      WIDEN ir<[[LD:%.+]]> = load vp<[[LD_PTR]]>, stride = ir<-4>, runtimeVF = vp<[[VF]]>
+; CHECK-NEXT:      WIDEN ir<[[LD:%.+]]> = load vp<[[LD_PTR]]>, stride = ir<-1>, runtimeVF = vp<[[VF]]>
 ; CHECK-NEXT:      WIDEN ir<[[ADD:%.+]]> = fadd ir<[[LD]]>, ir<1.000000e+00>
 ; CHECK-NEXT:      CLONE ir<[[ST_IDX:%.+]]> = getelementptr inbounds ir<%A>, ir<[[ZEXT_IDX]]>
 ; CHECK-NEXT:      vp<[[ST_PTR:%.+]]> = vector-end-pointer inbounds ir<[[ST_IDX]]>, vp<[[VF]]>
@@ -645,7 +645,7 @@ define void @vector_reverse_f32(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; CHECK-NEXT:    CLONE ir<[[ZEXT_IDX:%.+]]> = zext ir<[[IDX]]>
 ; CHECK-NEXT:    CLONE ir<[[LD_IDX:%.+]]> = getelementptr inbounds ir<%B>, ir<[[ZEXT_IDX]]>
 ; CHECK-NEXT:    vp<[[LD_PTR:%.+]]> = vector-pointer ir<[[LD_IDX]]>
-; CHECK-NEXT:    WIDEN ir<[[LD:%.+]]> = load vp<[[LD_PTR]]>, stride = ir<-4>, runtimeVF = ir<[[VF]]>
+; CHECK-NEXT:    WIDEN ir<[[LD:%.+]]> = load vp<[[LD_PTR]]>, stride = ir<-1>, runtimeVF = ir<[[VF]]>
 ; CHECK-NEXT:    WIDEN ir<[[ADD:%.+]]> = fadd ir<[[LD]]>, ir<1.000000e+00>
 ; CHECK-NEXT:    CLONE ir<[[ST_IDX:%.+]]> = getelementptr inbounds ir<%A>, ir<[[ZEXT_IDX]]>
 ; CHECK-NEXT:    vp<[[ST_PTR:%.+]]> = vector-end-pointer inbounds ir<[[ST_IDX]]>, ir<[[VF]]>