Index: lib/Target/AMDGPU/SIISelLowering.cpp
===================================================================
--- lib/Target/AMDGPU/SIISelLowering.cpp
+++ lib/Target/AMDGPU/SIISelLowering.cpp
@@ -8998,6 +8998,42 @@
     Ops.push_back(ImpDef.getValue(1));
     return DAG.getMachineNode(Opcode, SDLoc(Node), Node->getVTList(), Ops);
   }
+  case AMDGPU::BUFFER_LOAD_SHORT_D16_HI_OFFSET:
+  case AMDGPU::BUFFER_LOAD_UBYTE_D16_HI_OFFSET:
+  case AMDGPU::BUFFER_LOAD_SBYTE_D16_HI_OFFSET:
+  case AMDGPU::BUFFER_LOAD_SHORT_D16_HI_OFFEN:
+  case AMDGPU::BUFFER_LOAD_UBYTE_D16_HI_OFFEN:
+  case AMDGPU::BUFFER_LOAD_SBYTE_D16_HI_OFFEN: {
+    // For these loads that write to the HI part of a register,
+    // we should chain them to the op that writes to the LO part
+    // of the register to maintain the order.
+    unsigned NumOps = Node->getNumOperands();
+    SDValue OldChain = Node->getOperand(NumOps-1);
+
+    if (OldChain.getValueType() != MVT::Other)
+      break;
+
+    // Look for the chain to replace to.
+    SDValue Lo = Node->getOperand(NumOps-2);
+    SDNode *LoNode = Lo.getNode();
+    if (LoNode->getNumValues() == 1 ||
+        LoNode->getValueType(LoNode->getNumValues() - 1) != MVT::Other)
+      break;
+
+    SDValue NewChain = Lo.getValue(LoNode->getNumValues() - 1);
+    if (NewChain == OldChain) // Already replaced.
+      break;
+
+    SmallVector<SDValue, 4> Ops;
+    for (unsigned I = 0; I < NumOps-1; ++I)
+      Ops.push_back(Node->getOperand(I));
+    // Repalce the Chain.
+    Ops.push_back(NewChain);
+    MachineSDNode *NewNode = DAG.getMachineNode(Opcode, SDLoc(Node),
+                                                Node->getVTList(), Ops);
+    DAG.setNodeMemRefs(NewNode, Node->memoperands());
+    return NewNode;
+  }
   default:
     break;
   }
Index: test/CodeGen/AMDGPU/chain-hi-to-lo.ll
===================================================================
--- /dev/null
+++ test/CodeGen/AMDGPU/chain-hi-to-lo.ll
@@ -0,0 +1,18 @@
+; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN %s
+
+; GCN-LABEL: {{^}}chain_hi_to_lo:
+; GCN: buffer_load_ushort [[DST:v[0-9]+]], off, [[RSRC:s\[[0-9]+:[0-9]+\]]], [[SOFF:s[0-9]+]] offset:2
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: buffer_load_short_d16_hi [[DST]], off, [[RSRC]], [[SOFF]]
+
+define amdgpu_kernel void @chain_hi_to_lo() {
+bb:
+  %loads = load <2 x half>, <2 x half> addrspace(5)* null, align 2
+  %shuffled = shufflevector <2 x half> %loads, <2 x half> undef, <2 x i32> <i32 1, i32 0>
+  br label %bb1
+
+bb1:
+  call void asm sideeffect "; use $0", "v"(<2 x half> %shuffled)
+  br label %bb1
+}
+