diff --git a/llvm/lib/Transforms/Scalar/Scalarizer.cpp b/llvm/lib/Transforms/Scalar/Scalarizer.cpp
--- a/llvm/lib/Transforms/Scalar/Scalarizer.cpp
+++ b/llvm/lib/Transforms/Scalar/Scalarizer.cpp
@@ -111,6 +111,7 @@
   ValueVector *CachePtr;
   ValueVector Tmp;
   unsigned Size;
+  bool BroadcastedAddress;
 };
 
 // FCmpSpliiter(FCI)(Builder, X, Y, Name) uses Builder to create an FCmp
@@ -280,7 +281,8 @@
 
 Scatterer::Scatterer(BasicBlock *bb, BasicBlock::iterator bbi, Value *v,
                      Type *PtrElemTy, ValueVector *cachePtr)
-    : BB(bb), BBI(bbi), V(v), PtrElemTy(PtrElemTy), CachePtr(cachePtr) {
+    : BB(bb), BBI(bbi), V(v), PtrElemTy(PtrElemTy), CachePtr(cachePtr),
+      BroadcastedAddress(false) {
   Type *Ty = V->getType();
   if (Ty->isPointerTy()) {
     assert(cast<PointerType>(Ty)->isOpaqueOrPointeeTypeMatches(PtrElemTy) &&
@@ -292,16 +294,31 @@
     Tmp.resize(Size, nullptr);
   else if (CachePtr->empty())
     CachePtr->resize(Size, nullptr);
-  else
+  else if (PtrElemTy && CachePtr->size() == 1) {
+    BroadcastedAddress = true;
+  } else
     assert(Size == CachePtr->size() && "Inconsistent vector sizes");
 }
 
 // Return component I, creating a new Value for it if necessary.
 Value *Scatterer::operator[](unsigned I) {
   ValueVector &CV = (CachePtr ? *CachePtr : Tmp);
+
+  // A broadcasted value is a value that already exists in a 1-element vector
+  // and is propagated across several elements for loading /storing.
+  if (BroadcastedAddress) {
+    assert(CV[0] && "value to broadcast");
+    IRBuilder<> Builder(BB, BBI);
+    Type *VectorElemTy = cast<VectorType>(PtrElemTy)->getElementType();
+    Type *NewPtrTy = PointerType::get(
+        VectorElemTy, CV[0]->getType()->getPointerAddressSpace());
+    return Builder.CreateBitCast(CV[0], NewPtrTy, CV[0]->getName() + ".bcast");
+  }
+
   // Try to reuse a previous value.
   if (CV[I])
     return CV[I];
+
   IRBuilder<> Builder(BB, BBI);
   if (PtrElemTy) {
     Type *VectorElemTy = cast<VectorType>(PtrElemTy)->getElementType();
@@ -432,6 +449,7 @@
       PotentiallyDeadInstrs.emplace_back(Old);
     }
   }
+  assert((SV.empty() || SV.size() == CV.size()) && "consistent update");
   SV = CV;
   Gathered.push_back(GatherList::value_type(Op, &SV));
 }
diff --git a/llvm/test/Transforms/Scalarizer/global-bug.ll b/llvm/test/Transforms/Scalarizer/global-bug.ll
--- a/llvm/test/Transforms/Scalarizer/global-bug.ll
+++ b/llvm/test/Transforms/Scalarizer/global-bug.ll
@@ -1,3 +1,4 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt < %s -passes='function(scalarizer)' -S | FileCheck %s
 
 @a = dso_local global i16 0, align 1
diff --git a/llvm/test/Transforms/Scalarizer/vector-of-pointer-to-vector.ll b/llvm/test/Transforms/Scalarizer/vector-of-pointer-to-vector.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/Transforms/Scalarizer/vector-of-pointer-to-vector.ll
@@ -0,0 +1,54 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt %s -passes='function(scalarizer,dce)' -scalarize-load-store -S | FileCheck %s
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+
+define <1 x i32> @f1(<1 x <1 x i32>*> %src, i32 %index) {
+; CHECK-LABEL: @f1(
+; CHECK-NEXT:    [[INDEX_IS_0:%.*]] = icmp eq i32 [[INDEX:%.*]], 0
+; CHECK-NEXT:    [[SRC_I0:%.*]] = extractelement <1 x <1 x i32>*> [[SRC:%.*]], i32 0
+; CHECK-NEXT:    [[DOTUPTO0:%.*]] = select i1 [[INDEX_IS_0]], <1 x i32>* [[SRC_I0]], <1 x i32>* undef
+; CHECK-NEXT:    [[DOTUPTO0_BCAST:%.*]] = bitcast <1 x i32>* [[DOTUPTO0]] to i32*
+; CHECK-NEXT:    [[DOTI0:%.*]] = load i32, i32* [[DOTUPTO0_BCAST]], align 4
+; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <1 x i32> poison, i32 [[DOTI0]], i32 0
+; CHECK-NEXT:    ret <1 x i32> [[TMP1]]
+;
+  %1 = extractelement <1 x <1 x i32>*> %src, i32 %index
+  %2 = load <1 x i32>, <1 x i32>* %1, align 4
+  ret <1 x i32> %2
+}
+
+define <2 x i32> @f2(<1 x <2 x i32>*> %src, i32 %index) {
+; CHECK-LABEL: @f2(
+; CHECK-NEXT:    [[INDEX_IS_0:%.*]] = icmp eq i32 [[INDEX:%.*]], 0
+; CHECK-NEXT:    [[SRC_I0:%.*]] = extractelement <1 x <2 x i32>*> [[SRC:%.*]], i32 0
+; CHECK-NEXT:    [[DOTUPTO0:%.*]] = select i1 [[INDEX_IS_0]], <2 x i32>* [[SRC_I0]], <2 x i32>* undef
+; CHECK-NEXT:    [[DOTUPTO0_BCAST:%.*]] = bitcast <2 x i32>* [[DOTUPTO0]] to i32*
+; CHECK-NEXT:    [[DOTI0:%.*]] = load i32, i32* [[DOTUPTO0_BCAST]], align 4
+; CHECK-NEXT:    [[DOTUPTO0_BCAST1:%.*]] = bitcast <2 x i32>* [[DOTUPTO0]] to i32*
+; CHECK-NEXT:    [[DOTI1:%.*]] = load i32, i32* [[DOTUPTO0_BCAST1]], align 4
+; CHECK-NEXT:    [[DOTUPTO02:%.*]] = insertelement <2 x i32> poison, i32 [[DOTI0]], i32 0
+; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x i32> [[DOTUPTO02]], i32 [[DOTI1]], i32 1
+; CHECK-NEXT:    ret <2 x i32> [[TMP1]]
+;
+  %1 = extractelement <1 x <2 x i32>*> %src, i32 %index
+  %2 = load <2 x i32>, <2 x i32>* %1, align 4
+  ret <2 x i32> %2
+}
+
+define void @f3(<1 x <2 x i32>*> %src, i32 %index, <2 x i32> %val) {
+; CHECK-LABEL: @f3(
+; CHECK-NEXT:    [[VAL_I0:%.*]] = extractelement <2 x i32> [[VAL:%.*]], i32 0
+; CHECK-NEXT:    [[VAL_I1:%.*]] = extractelement <2 x i32> [[VAL]], i32 1
+; CHECK-NEXT:    [[INDEX_IS_0:%.*]] = icmp eq i32 [[INDEX:%.*]], 0
+; CHECK-NEXT:    [[SRC_I0:%.*]] = extractelement <1 x <2 x i32>*> [[SRC:%.*]], i32 0
+; CHECK-NEXT:    [[DOTUPTO0:%.*]] = select i1 [[INDEX_IS_0]], <2 x i32>* [[SRC_I0]], <2 x i32>* undef
+; CHECK-NEXT:    [[DOTUPTO0_BCAST:%.*]] = bitcast <2 x i32>* [[DOTUPTO0]] to i32*
+; CHECK-NEXT:    store i32 [[VAL_I0]], i32* [[DOTUPTO0_BCAST]], align 4
+; CHECK-NEXT:    [[DOTUPTO0_BCAST1:%.*]] = bitcast <2 x i32>* [[DOTUPTO0]] to i32*
+; CHECK-NEXT:    store i32 [[VAL_I1]], i32* [[DOTUPTO0_BCAST1]], align 4
+; CHECK-NEXT:    ret void
+;
+  %1 = extractelement <1 x <2 x i32>*> %src, i32 %index
+  store <2 x i32> %val, <2 x i32>* %1, align 4
+  ret void
+}