diff --git a/llvm/lib/Target/DirectX/DXILLegalizePass.cpp b/llvm/lib/Target/DirectX/DXILLegalizePass.cpp
index c9ff7137fdac1..c73648f21e8d7 100644
--- a/llvm/lib/Target/DirectX/DXILLegalizePass.cpp
+++ b/llvm/lib/Target/DirectX/DXILLegalizePass.cpp
@@ -563,7 +563,7 @@ legalizeGetHighLowi64Bytes(Instruction &I,
 }
 
 static void
-legalizeLoadStoreOnArrayAllocas(Instruction &I,
+legalizeScalarLoadStoreOnArrays(Instruction &I,
                                 SmallVectorImpl<Instruction *> &ToRemove,
                                 DenseMap<Value *, Value *> &) {
 
@@ -581,23 +581,31 @@ legalizeLoadStoreOnArrayAllocas(Instruction &I,
   } else
     return;
 
-  assert(LoadStoreTy->isSingleValueType() &&
-         "Expected load/store type to be a single-valued type");
+  // If the load/store is not of a single-value type (i.e., scalar or vector)
+  // then we do not modify it. It shouldn't be a vector either because the
+  // dxil-data-scalarization pass is expected to run before this, but it's not
+  // incorrect to apply this transformation to vector load/stores.
+  if (!LoadStoreTy->isSingleValueType())
+    return;
 
-  auto *AllocaPtrOp = dyn_cast<AllocaInst>(PtrOp);
-  if (!AllocaPtrOp)
+  Type *ArrayTy;
+  if (auto *GlobalVarPtrOp = dyn_cast<GlobalVariable>(PtrOp))
+    ArrayTy = GlobalVarPtrOp->getValueType();
+  else if (auto *AllocaPtrOp = dyn_cast<AllocaInst>(PtrOp))
+    ArrayTy = AllocaPtrOp->getAllocatedType();
+  else
     return;
 
-  Type *Ty = AllocaPtrOp->getAllocatedType();
-  if (!isa<ArrayType>(Ty))
+  if (!isa<ArrayType>(ArrayTy))
     return;
-  assert(!isa<ArrayType>(Ty->getArrayElementType()) &&
-         "Expected allocated type of AllocaInst to be a flat ArrayType");
 
-  IRBuilder<> Builder(&I);
-  Value *Zero = Builder.getInt32(0);
-  Value *GEP = Builder.CreateGEP(Ty, AllocaPtrOp, {Zero, Zero}, "",
-                                 GEPNoWrapFlags::all());
+  assert(ArrayTy->getArrayElementType() == LoadStoreTy &&
+         "Expected array element type to be the same as to the scalar load or "
+         "store type");
+
+  Value *Zero = ConstantInt::get(Type::getInt32Ty(I.getContext()), 0);
+  Value *GEP = GetElementPtrInst::Create(
+      ArrayTy, PtrOp, {Zero, Zero}, GEPNoWrapFlags::all(), "", I.getIterator());
   I.setOperand(PtrOpIndex, GEP);
 }
 
@@ -651,7 +659,7 @@ class DXILLegalizationPipeline {
     // downcastI64toI32InsertExtractElements needs to handle.
     LegalizationPipeline[Stage2].push_back(
         downcastI64toI32InsertExtractElements);
-    LegalizationPipeline[Stage2].push_back(legalizeLoadStoreOnArrayAllocas);
+    LegalizationPipeline[Stage2].push_back(legalizeScalarLoadStoreOnArrays);
   }
 };
 
diff --git a/llvm/test/CodeGen/DirectX/legalize-load-store-array-alloca.ll b/llvm/test/CodeGen/DirectX/legalize-load-store-array-alloca.ll
index b25b3de901d91..c6789ac7886d5 100644
--- a/llvm/test/CodeGen/DirectX/legalize-load-store-array-alloca.ll
+++ b/llvm/test/CodeGen/DirectX/legalize-load-store-array-alloca.ll
@@ -21,3 +21,21 @@ define void @store() {
   store i32 0, ptr %a, align 4
   ret void
 }
+
+@g = local_unnamed_addr addrspace(3) global [4 x i32] zeroinitializer, align 4
+define void @load_whole_global () {
+; CHECK-LABEL: define void @load_whole_global
+; CHECK-NEXT:    load [4 x i32], ptr addrspace(3) @g, align 4
+; CHECK-NEXT:    ret void
+  %l = load [4 x i32], ptr addrspace(3) @g, align 4
+  ret void
+}
+
+define void @load_global_index0 () {
+; CHECK-LABEL: define void @load_global_index0
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds nuw [4 x i32], ptr addrspace(3) @g, i32 0, i32 0
+; CHECK-NEXT:    load i32, ptr addrspace(3) [[GEP]], align 4
+; CHECK-NEXT:    ret void
+  %l = load i32, ptr addrspace(3) @g, align 4
+  ret void
+}
diff --git a/llvm/test/CodeGen/DirectX/llc-vector-load-scalarize.ll b/llvm/test/CodeGen/DirectX/llc-vector-load-scalarize.ll
index 27a892591a867..0c91c53227763 100644
--- a/llvm/test/CodeGen/DirectX/llc-vector-load-scalarize.ll
+++ b/llvm/test/CodeGen/DirectX/llc-vector-load-scalarize.ll
@@ -24,7 +24,8 @@
 define <4 x i32> @load_array_vec_test() #0 {
 ; CHECK-LABEL: define <4 x i32> @load_array_vec_test(
 ; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {
-; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr addrspace(3) @arrayofVecData.scalarized.1dim, align 4
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds nuw [8 x i32], ptr addrspace(3) @arrayofVecData.scalarized.1dim, i32 0, i32 0
+; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr addrspace(3) [[GEP]], align 4
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(3) getelementptr ([8 x i32], ptr addrspace(3) @arrayofVecData.scalarized.1dim, i32 0, i32 1), align 4
 ; CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr addrspace(3) getelementptr ([8 x i32], ptr addrspace(3) @arrayofVecData.scalarized.1dim, i32 0, i32 2), align 4
 ; CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr addrspace(3) getelementptr ([8 x i32], ptr addrspace(3) @arrayofVecData.scalarized.1dim, i32 0, i32 3), align 4
@@ -52,7 +53,8 @@ define <4 x i32> @load_array_vec_test() #0 {
 define <4 x i32> @load_vec_test() #0 {
 ; CHECK-LABEL: define <4 x i32> @load_vec_test(
 ; CHECK-SAME: ) #[[ATTR0]] {
-; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(3) @vecData.scalarized, align 4
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds nuw [4 x i32], ptr addrspace(3) @vecData.scalarized, i32 0, i32 0
+; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(3) [[GEP]], align 4
 ; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr addrspace(3) getelementptr ([4 x i32], ptr addrspace(3) @vecData.scalarized, i32 0, i32 1), align 4
 ; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(3) getelementptr ([4 x i32], ptr addrspace(3) @vecData.scalarized, i32 0, i32 2), align 4
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(3) getelementptr ([4 x i32], ptr addrspace(3) @vecData.scalarized, i32 0, i32 3), align 4
@@ -203,7 +205,8 @@ define <4 x i32> @load_static_array_of_vec_from_i8_gep_test(i32 %index) #0 {
 define <4 x i32> @multid_load_test() #0 {
 ; CHECK-LABEL: define <4 x i32> @multid_load_test(
 ; CHECK-SAME: ) #[[ATTR0]] {
-; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(3) @groupshared2dArrayofVectors.scalarized.1dim, align 4
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds nuw [36 x i32], ptr addrspace(3) @groupshared2dArrayofVectors.scalarized.1dim, i32 0, i32 0
+; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(3) [[GEP]], align 4
 ; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr addrspace(3) getelementptr ([36 x i32], ptr addrspace(3) @groupshared2dArrayofVectors.scalarized.1dim, i32 0, i32 1), align 4
 ; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(3) getelementptr ([36 x i32], ptr addrspace(3) @groupshared2dArrayofVectors.scalarized.1dim, i32 0, i32 2), align 4
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(3) getelementptr ([36 x i32], ptr addrspace(3) @groupshared2dArrayofVectors.scalarized.1dim, i32 0, i32 3), align 4
diff --git a/llvm/test/CodeGen/DirectX/scalar-store.ll b/llvm/test/CodeGen/DirectX/scalar-store.ll
index a124c665ad15e..4394235ffe4bd 100644
--- a/llvm/test/CodeGen/DirectX/scalar-store.ll
+++ b/llvm/test/CodeGen/DirectX/scalar-store.ll
@@ -14,7 +14,8 @@
 
 ; CHECK-LABEL: store_array_vec_test
 define void @store_array_vec_test () local_unnamed_addr #0 {
-; CHECK-NEXT:    store float 1.000000e+00, ptr addrspace(3) @arrayofVecData.scalarized.1dim, align 16
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds nuw [6 x float], ptr addrspace(3) @arrayofVecData.scalarized.1dim, i32 0, i32 0
+; CHECK-NEXT:    store float 1.000000e+00, ptr addrspace(3) [[GEP]], align 16
 ; CHECK-NEXT:    store float 2.000000e+00, ptr addrspace(3) getelementptr ([6 x float], ptr addrspace(3) @arrayofVecData.scalarized.1dim, i32 0, i32 1), align 4
 ; CHECK-NEXT:    store float 3.000000e+00, ptr addrspace(3) getelementptr ([6 x float], ptr addrspace(3) @arrayofVecData.scalarized.1dim, i32 0, i32 2), align 8
 ; CHECK-NEXT:    store float 2.000000e+00, ptr addrspace(3) getelementptr inbounds ([6 x float], ptr addrspace(3) @arrayofVecData.scalarized.1dim, i32 0, i32 3), align 16
@@ -30,7 +31,8 @@ define void @store_array_vec_test () local_unnamed_addr #0 {
 ; CHECK-LABEL: store_vec_test
 define void @store_vec_test(<4 x i32> %inputVec) #0 {
 ; CHECK-NEXT:    [[INPUTVEC_I01:%.*]] = extractelement <4 x i32> %inputVec, i32 0
-; CHECK-NEXT:    store i32 [[INPUTVEC_I01]], ptr addrspace(3) @vecData.scalarized, align 4
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds nuw [4 x i32], ptr addrspace(3) @vecData.scalarized, i32 0, i32 0
+; CHECK-NEXT:    store i32 [[INPUTVEC_I01]], ptr addrspace(3) [[GEP]], align 4
 ; CHECK-NEXT:    [[INPUTVEC_I12:%.*]] = extractelement <4 x i32> %inputVec, i32 1
 ; CHECK-NEXT:    store i32 [[INPUTVEC_I12]], ptr addrspace(3) getelementptr ([4 x i32], ptr addrspace(3) @vecData.scalarized, i32 0, i32 1), align 4
 ; CHECK-NEXT:    [[INPUTVEC_I23:%.*]] = extractelement <4 x i32> %inputVec, i32 2