diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 640a98c622f80..6e6ac735b1f37 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -8474,6 +8474,8 @@ VPlanPtr LoopVectorizationPlanner::tryToBuildVPlanWithVPRecipes(
                                   CM.foldTailByMasking());
 
   VPlanTransforms::createLoopRegions(*Plan);
+  VPlanTransforms::removeInvariantStoresOfReduction(
+      *Plan, Legal->getReductionVars(), *LVer);
 
   // Don't use getDecisionAndClampRange here, because we don't know the UF
   // so this function is better to be conservative, rather than to split
@@ -8546,8 +8548,6 @@ VPlanPtr LoopVectorizationPlanner::tryToBuildVPlanWithVPRecipes(
   ReversePostOrderTraversal<VPBlockShallowTraversalWrapper<VPBlockBase *>> RPOT(
       HeaderVPBB);
 
-  auto *MiddleVPBB = Plan->getMiddleBlock();
-  VPBasicBlock::iterator MBIP = MiddleVPBB->getFirstNonPhi();
   // Mapping from VPValues in the initial plan to their widened VPValues. Needed
   // temporarily to update created block masks.
   DenseMap<VPValue *, VPValue *> Old2New;
@@ -8576,23 +8576,6 @@ VPlanPtr LoopVectorizationPlanner::tryToBuildVPlanWithVPRecipes(
       Instruction *Instr = cast<Instruction>(UnderlyingValue);
       Builder.setInsertPoint(SingleDef);
 
-      // The stores with invariant address inside the loop will be deleted, and
-      // in the exit block, a uniform store recipe will be created for the final
-      // invariant store of the reduction.
-      StoreInst *SI;
-      if ((SI = dyn_cast<StoreInst>(Instr)) &&
-          Legal->isInvariantAddressOfReduction(SI->getPointerOperand())) {
-        // Only create recipe for the final invariant store of the reduction.
-        if (Legal->isInvariantStoreOfReduction(SI)) {
-          auto *Recipe =
-              new VPReplicateRecipe(SI, R.operands(), true /* IsUniform */,
-                                    nullptr /*Mask*/, VPIRMetadata(*SI, LVer));
-          Recipe->insertBefore(*MiddleVPBB, MBIP);
-        }
-        R.eraseFromParent();
-        continue;
-      }
-
       VPRecipeBase *Recipe =
           RecipeBuilder.tryToCreateWidenRecipe(SingleDef, Range);
       if (!Recipe)
diff --git a/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp b/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp
index cef91c15dd873..849056659edb5 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp
@@ -634,6 +634,48 @@ void VPlanTransforms::createLoopRegions(VPlan &Plan) {
   TopRegion->getEntryBasicBlock()->setName("vector.body");
 }
 
+void VPlanTransforms::removeInvariantStoresOfReduction(
+    VPlan &Plan, const MapVector<PHINode *, RecurrenceDescriptor> &Rdxs,
+    LoopVersioning &LVer) {
+  auto *MiddleVPBB = Plan.getMiddleBlock();
+  VPBasicBlock::iterator MBIP = MiddleVPBB->getFirstNonPhi();
+
+  auto IsInvariantStore = [Rdxs](StoreInst *SI) {
+    return any_of(Rdxs, [SI](const auto &RdxDesc) {
+      return RdxDesc.second.IntermediateStore == SI;
+    });
+  };
+
+  auto IsInvariantAddr = [Rdxs](Value *V) {
+    return any_of(Rdxs, [V](const auto &RdxDesc) {
+      auto *SI = RdxDesc.second.IntermediateStore;
+      return SI && SI->getPointerOperand() == V;
+    });
+  };
+
+  for (VPBasicBlock *VPBB : VPBlockUtils::blocksOnly<VPBasicBlock>(
+           vp_depth_first_shallow(Plan.getVectorLoopRegion()->getEntry()))) {
+    for (VPRecipeBase &R : make_early_inc_range(*VPBB)) {
+      auto *Def = cast<VPSingleDefRecipe>(&R);
+
+      // The stores with invariant address inside the loop will be deleted, and
+      // in the exit block, a uniform store recipe will be created for the final
+      // invariant store of the reduction.
+      StoreInst *SI = dyn_cast_if_present<StoreInst>(Def->getUnderlyingValue());
+      if (!SI || !IsInvariantAddr(SI->getPointerOperand()))
+        continue;
+      if (IsInvariantStore(SI)) {
+        // Only create recipe for the final invariant store of the reduction.
+        auto *Recipe =
+            new VPReplicateRecipe(SI, Def->operands(), /*IsSingleScalar=*/true,
+                                  /*Mask=*/nullptr, VPIRMetadata(*SI, &LVer));
+        Recipe->insertBefore(*MiddleVPBB, MBIP);
+      }
+      Def->eraseFromParent();
+    }
+  }
+}
+
 // Likelyhood of bypassing the vectorized loop due to a runtime check block,
 // including memory overlap checks block and wrapping/unit-stride checks block.
 static constexpr uint32_t CheckBypassWeights[] = {1, 127};
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
index 69452a7e37572..a567ae127cf63 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
@@ -121,6 +121,12 @@ struct VPlanTransforms {
   /// flat CFG into a hierarchical CFG.
   LLVM_ABI_FOR_TEST static void createLoopRegions(VPlan &Plan);
 
+  /// Remove invariant stores of reductions, given \p Rdxs, in \p Plan. \p LVer
+  /// is used to create metadata from an existing store.
+  static void removeInvariantStoresOfReduction(
+      VPlan &Plan, const MapVector<PHINode *, RecurrenceDescriptor> &Rdxs,
+      LoopVersioning &LVer);
+
   /// Wrap runtime check block \p CheckBlock in a VPIRBB and \p Cond in a
   /// VPValue and connect the block to \p Plan, using the VPValue as branch
   /// condition.