Index: lib/Target/AMDGPU/SIISelLowering.cpp
===================================================================
--- lib/Target/AMDGPU/SIISelLowering.cpp
+++ lib/Target/AMDGPU/SIISelLowering.cpp
@@ -3464,10 +3464,6 @@
   return false;
 }
 
-static bool isDwordAligned(unsigned Alignment) {
-  return Alignment % 4 == 0;
-}
-
 //===----------------------------------------------------------------------===//
 // Custom DAG Lowering Operations
 //===----------------------------------------------------------------------===//
@@ -5385,21 +5381,23 @@
          AMDGPUASI.PRIVATE_ADDRESS : AMDGPUASI.GLOBAL_ADDRESS;
 
   unsigned NumElements = MemVT.getVectorNumElements();
+
   if (AS == AMDGPUASI.CONSTANT_ADDRESS ||
       AS == AMDGPUASI.CONSTANT_ADDRESS_32BIT) {
-    if (!Op->isDivergent())
+    if (!Op->isDivergent() && Alignment >= 4)
       return SDValue();
     // Non-uniform loads will be selected to MUBUF instructions, so they
     // have the same legalization requirements as global and private
     // loads.
     //
   }
+
   if (AS == AMDGPUASI.CONSTANT_ADDRESS ||
       AS == AMDGPUASI.CONSTANT_ADDRESS_32BIT ||
       AS == AMDGPUASI.GLOBAL_ADDRESS) {
     if (Subtarget->getScalarizeGlobalBehavior() && !Op->isDivergent() &&
         !Load->isVolatile() && isMemOpHasNoClobberedMemOperand(Load) &&
-        isDwordAligned(Alignment))
+        Alignment >= 4)
       return SDValue();
     // Non-uniform loads will be selected to MUBUF instructions, so they
     // have the same legalization requirements as global and private
Index: test/CodeGen/AMDGPU/load-constant-i16.ll
===================================================================
--- test/CodeGen/AMDGPU/load-constant-i16.ll
+++ test/CodeGen/AMDGPU/load-constant-i16.ll
@@ -72,6 +72,18 @@
   ret void
 }
 
+; FUNC-LABEL: {{^}}constant_load_v16i16_align2:
+; GCN-HSA: flat_load_dwordx4
+; GCN-HSA: flat_load_dwordx4
+; GCN-HSA: flat_store_dwordx4
+; GCN-HSA: flat_store_dwordx4
+define amdgpu_kernel void @constant_load_v16i16_align2(<16 x i16> addrspace(4)* %ptr0) #0 {
+entry:
+  %ld =  load <16 x i16>, <16 x i16> addrspace(4)* %ptr0, align 2
+  store <16 x i16> %ld, <16 x i16> addrspace(1)* undef, align 32
+  ret void
+}
+
 ; FUNC-LABEL: {{^}}constant_zextload_i16_to_i32:
 ; GCN-NOHSA: buffer_load_ushort
 ; GCN-NOHSA: buffer_store_dword
Index: test/CodeGen/AMDGPU/load-global-i16.ll
===================================================================
--- test/CodeGen/AMDGPU/load-global-i16.ll
+++ test/CodeGen/AMDGPU/load-global-i16.ll
@@ -83,6 +83,18 @@
   ret void
 }
 
+; GCN-LABEL: {{^}}global_load_v16i16_align2:
+; GCN-HSA: flat_load_dwordx4
+; GCN-HSA: flat_load_dwordx4
+; GCN-HSA: flat_store_dwordx4
+; GCN-HSA: flat_store_dwordx4
+define amdgpu_kernel void @global_load_v16i16_align2(<16 x i16> addrspace(1)* %in, <16 x i16> addrspace(1)* %out) #0 {
+entry:
+  %ld =  load <16 x i16>, <16 x i16> addrspace(1)* %in, align 2
+  store <16 x i16> %ld, <16 x i16> addrspace(1)* %out, align 32
+  ret void
+}
+
 ; FUNC-LABEL: {{^}}global_zextload_i16_to_i32:
 ; GCN-NOHSA: buffer_load_ushort
 ; GCN-NOHSA: buffer_store_dword