Index: lib/Target/AMDGPU/SIISelLowering.h
===================================================================
--- lib/Target/AMDGPU/SIISelLowering.h
+++ lib/Target/AMDGPU/SIISelLowering.h
@@ -123,6 +123,7 @@
   SDValue lowerADDRSPACECAST(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const;
+  SDValue lowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerTRAP(SDValue Op, SelectionDAG &DAG) const;
   SDValue lowerDEBUGTRAP(SDValue Op, SelectionDAG &DAG) const;
Index: lib/Target/AMDGPU/SIISelLowering.cpp
===================================================================
--- lib/Target/AMDGPU/SIISelLowering.cpp
+++ lib/Target/AMDGPU/SIISelLowering.cpp
@@ -630,6 +630,9 @@
     setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v2i16, Custom);
     setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v2f16, Custom);
 
+    setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v4f16, Custom);
+    setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v4i16, Custom);
+
     setOperationAction(ISD::SHL, MVT::v4i16, Custom);
     setOperationAction(ISD::SRA, MVT::v4i16, Custom);
     setOperationAction(ISD::SRL, MVT::v4i16, Custom);
@@ -3882,6 +3885,8 @@
     return lowerINSERT_VECTOR_ELT(Op, DAG);
   case ISD::EXTRACT_VECTOR_ELT:
     return lowerEXTRACT_VECTOR_ELT(Op, DAG);
+  case ISD::VECTOR_SHUFFLE:
+    return lowerVECTOR_SHUFFLE(Op, DAG);
   case ISD::BUILD_VECTOR:
     return lowerBUILD_VECTOR(Op, DAG);
   case ISD::FP_ROUND:
@@ -4665,6 +4670,57 @@
   return DAG.getAnyExtOrTrunc(Elt, SL, ResultVT);
 }
 
+static bool elementPairIsContiguous(ArrayRef<int> Mask, int Elt) {
+  assert(Elt % 2 == 0);
+  return Mask[Elt + 1] == Mask[Elt] + 1 && (Mask[Elt] % 2 == 0);
+}
+
+SDValue SITargetLowering::lowerVECTOR_SHUFFLE(SDValue Op,
+                                              SelectionDAG &DAG) const {
+  SDLoc SL(Op);
+  EVT ResultVT = Op.getValueType();
+  ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(Op);
+
+  EVT PackVT = ResultVT.isInteger() ? MVT::v2i16 : MVT::v2f16;
+  EVT EltVT = PackVT.getVectorElementType();
+  int SrcNumElts = Op.getOperand(0).getValueType().getVectorNumElements();
+
+  // vector_shuffle <0,1,6,7> lhs, rhs
+  // -> concat_vectors (extract_subvector lhs, 0), (extract_subvector rhs, 2)
+  //
+  // vector_shuffle <6,7,2,3> lhs, rhs
+  // -> concat_vectors (extract_subvector rhs, 2), (extract_subvector lhs, 2)
+  //
+  // vector_shuffle <6,7,0,1> lhs, rhs
+  // -> concat_vectors (extract_subvector rhs, 2), (extract_subvector lhs, 0)
+
+  // Avoid scalarizing when both halves are reading from consecutive elements.
+  SmallVector<SDValue, 4> Pieces;
+  for (int I = 0, N = ResultVT.getVectorNumElements(); I != N; I += 2) {
+    if (elementPairIsContiguous(SVN->getMask(), I)) {
+      int VecIdx = SVN->getMaskElt(I) / SrcNumElts;
+      int EltIdx = SVN->getMaskElt(I) % SrcNumElts;
+      SDValue SubVec = DAG.getNode(ISD::EXTRACT_SUBVECTOR, SL,
+                                    PackVT, SVN->getOperand(VecIdx),
+                                    DAG.getConstant(EltIdx, SL, MVT::i32));
+      Pieces.push_back(SubVec);
+    } else {
+      int VecIdx = SVN->getMaskElt(I) / SrcNumElts;
+      int EltIdx0 = SVN->getMaskElt(I) % SrcNumElts;
+      int EltIdx1 = SVN->getMaskElt(I + 1) % SrcNumElts;
+
+      SDValue Vec = SVN->getOperand(VecIdx);
+      SDValue Elt0 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SL, EltVT,
+                                 Vec, DAG.getConstant(EltIdx0, SL, MVT::i32));
+      SDValue Elt1 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SL, EltVT,
+                                 Vec, DAG.getConstant(EltIdx1, SL, MVT::i32));
+      Pieces.push_back(DAG.getBuildVector(PackVT, SL, { Elt0, Elt1 }));
+    }
+  }
+
+  return DAG.getNode(ISD::CONCAT_VECTORS, SL, ResultVT, Pieces);
+}
+
 SDValue SITargetLowering::lowerBUILD_VECTOR(SDValue Op,
                                             SelectionDAG &DAG) const {
   SDLoc SL(Op);
Index: test/CodeGen/AMDGPU/vector_shuffle.packed.ll
===================================================================
--- test/CodeGen/AMDGPU/vector_shuffle.packed.ll
+++ test/CodeGen/AMDGPU/vector_shuffle.packed.ll
@@ -7,9 +7,7 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX9-NEXT:    v_mov_b32_e32 v0, v1
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -24,10 +22,8 @@
 ; GFX9-NEXT:    global_load_dwordx2 v[4:5], v[0:1], off
 ; GFX9-NEXT:    global_load_dwordx2 v[1:2], v[2:3], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v5
+; GFX9-NEXT:    v_mov_b32_e32 v0, v5
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v5
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v2
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -101,12 +97,9 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v[0:1], off offset:4
-; GFX9-NEXT:    global_load_dword v1, v[2:3], off offset:4
-; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 16, v0
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -118,15 +111,13 @@
 ; GFX9-LABEL: shuffle_v4f16_35u5:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dword v0, v[0:1], off offset:4
-; GFX9-NEXT:    global_load_dword v1, v[2:3], off
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
 ; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_sdwa v0, v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
-; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
+; GFX9-NEXT:    v_and_b32_sdwa v0, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
+; GFX9-NEXT:    v_lshlrev_b32_e32 v1, 16, v3
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -138,15 +129,15 @@
 ; GFX9-LABEL: shuffle_v4f16_357u:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dword v4, v[0:1], off offset:4
-; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[2:3], off
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    global_load_dword v2, v[2:3], off offset:4
+; GFX9-NEXT:    v_mov_b32_e32 v3, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_and_b32_sdwa v3, v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v2
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
+; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v3
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -160,9 +151,6 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
@@ -192,13 +180,6 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[1:2], v[2:3], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, v2, v0
-; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v1, v2, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v1
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -214,13 +195,7 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[1:2], v[2:3], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0xffff
-; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, v1, v0
-; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 16, v2
-; GFX9-NEXT:    v_and_b32_e32 v1, v1, v2
-; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v1
+; GFX9-NEXT:    v_mov_b32_e32 v1, v2
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -232,15 +207,9 @@
 ; GFX9-LABEL: shuffle_v4f16_2301:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
+; GFX9-NEXT:    global_load_dwordx2 v[1:2], v[0:1], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v1, v2, v1
-; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v2, v2, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v2
+; GFX9-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -254,10 +223,7 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
-; GFX9-NEXT:    v_mov_b32_e32 v1, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, v1
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -269,17 +235,11 @@
 ; GFX9-LABEL: shuffle_v4f16_2345:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
-; GFX9-NEXT:    global_load_dwordx2 v[2:3], v[2:3], off
+; GFX9-NEXT:    global_load_dwordx2 v[4:5], v[0:1], off
+; GFX9-NEXT:    global_load_dwordx2 v[1:2], v[2:3], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
+; GFX9-NEXT:    v_mov_b32_e32 v0, v5
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 16, v2
-; GFX9-NEXT:    v_and_b32_e32 v1, v0, v1
-; GFX9-NEXT:    v_and_b32_e32 v2, v0, v2
-; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v2
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -291,17 +251,10 @@
 ; GFX9-LABEL: shuffle_v4f16_2367:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
-; GFX9-NEXT:    global_load_dwordx2 v[2:3], v[2:3], off
-; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
+; GFX9-NEXT:    global_load_dwordx2 v[4:5], v[0:1], off
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[2:3], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 16, v1
-; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 16, v3
-; GFX9-NEXT:    v_and_b32_e32 v1, v0, v1
-; GFX9-NEXT:    v_and_b32_e32 v3, v0, v3
-; GFX9-NEXT:    v_lshl_or_b32 v0, v2, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v3
+; GFX9-NEXT:    v_mov_b32_e32 v0, v5
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -313,17 +266,10 @@
 ; GFX9-LABEL: shuffle_v4f16_4501:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    global_load_dwordx2 v[1:2], v[2:3], off
+; GFX9-NEXT:    global_load_dwordx2 v[4:5], v[0:1], off
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[2:3], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
-; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v1, v2, v1
-; GFX9-NEXT:    v_and_b32_e32 v2, v2, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v4, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v3, 16, v2
+; GFX9-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -337,15 +283,8 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
 ; GFX9-NEXT:    global_load_dwordx2 v[2:3], v[2:3], off
-; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v1, v0, v1
-; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 16, v2
-; GFX9-NEXT:    v_and_b32_e32 v2, v0, v2
-; GFX9-NEXT:    v_lshl_or_b32 v0, v4, 16, v2
-; GFX9-NEXT:    v_lshl_or_b32 v1, v3, 16, v1
+; GFX9-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -359,9 +298,6 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[2:3], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
@@ -387,17 +323,11 @@
 ; GFX9-LABEL: shuffle_v4f16_6701:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    global_load_dwordx2 v[1:2], v[2:3], off
+; GFX9-NEXT:    global_load_dwordx2 v[4:5], v[0:1], off
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[2:3], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0xffff
-; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
-; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 16, v2
-; GFX9-NEXT:    v_and_b32_e32 v2, v1, v2
-; GFX9-NEXT:    v_and_b32_e32 v1, v1, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v4, 16, v2
-; GFX9-NEXT:    v_lshl_or_b32 v1, v3, 16, v1
+; GFX9-NEXT:    v_mov_b32_e32 v0, v1
+; GFX9-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -411,15 +341,8 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
 ; GFX9-NEXT:    global_load_dwordx2 v[2:3], v[2:3], off
-; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v1, v0, v1
-; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 16, v3
-; GFX9-NEXT:    v_and_b32_e32 v3, v0, v3
-; GFX9-NEXT:    v_lshl_or_b32 v0, v4, 16, v3
-; GFX9-NEXT:    v_lshl_or_b32 v1, v2, 16, v1
+; GFX9-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -431,15 +354,9 @@
 ; GFX9-LABEL: shuffle_v4f16_6745:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[2:3], off
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
+; GFX9-NEXT:    global_load_dwordx2 v[1:2], v[2:3], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v1, v2, v1
-; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v2, v2, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v2
+; GFX9-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -453,10 +370,7 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[2:3], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
-; GFX9-NEXT:    v_mov_b32_e32 v1, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, v1
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -468,16 +382,13 @@
 ; GFX9-LABEL: shuffle_v4f16_2356:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
-; GFX9-NEXT:    global_load_dwordx2 v[2:3], v[2:3], off
-; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
-; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v1, v0, v1
+; GFX9-NEXT:    global_load_dwordx2 v[4:5], v[0:1], off
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[2:3], off
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_sdwa v2, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v4, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v3, 16, v2
+; GFX9-NEXT:    v_and_b32_sdwa v0, v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, v5
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -493,12 +404,9 @@
 ; GFX9-NEXT:    global_load_dwordx2 v[2:3], v[2:3], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
-; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v1, v0, v1
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_sdwa v2, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v2
-; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v1
+; GFX9-NEXT:    v_and_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -510,15 +418,15 @@
 ; GFX9-LABEL: shuffle_v4f16_3456:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dword v4, v[0:1], off offset:4
-; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[2:3], off
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
+; GFX9-NEXT:    global_load_dwordx2 v[2:3], v[2:3], off
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_sdwa v3, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_and_b32_sdwa v1, v4, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_sdwa v2, v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v3
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
+; GFX9-NEXT:    v_and_b32_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX9-NEXT:    v_lshl_or_b32 v1, v3, 16, v2
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -530,15 +438,15 @@
 ; GFX9-LABEL: shuffle_v4f16_5634:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dword v0, v[0:1], off offset:4
-; GFX9-NEXT:    global_load_dwordx2 v[1:2], v[2:3], off
-; GFX9-NEXT:    v_mov_b32_e32 v3, 0xffff
+; GFX9-NEXT:    global_load_dwordx2 v[4:5], v[0:1], off
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[2:3], off
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_sdwa v0, v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_and_b32_sdwa v3, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_sdwa v3, v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v2, 16, v3
+; GFX9-NEXT:    v_and_b32_sdwa v0, v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v3
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -550,16 +458,16 @@
 ; GFX9-LABEL: shuffle_v4f16_5734:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dword v0, v[0:1], off offset:4
-; GFX9-NEXT:    global_load_dwordx2 v[1:2], v[2:3], off
-; GFX9-NEXT:    v_mov_b32_e32 v3, 0xffff
+; GFX9-NEXT:    global_load_dwordx2 v[4:5], v[0:1], off
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[2:3], off
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_and_b32_sdwa v4, v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_and_b32_sdwa v3, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v2
-; GFX9-NEXT:    v_and_b32_sdwa v2, v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v2
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v4
+; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
+; GFX9-NEXT:    v_and_b32_sdwa v0, v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
+; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v3
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -571,14 +479,13 @@
 ; GFX9-LABEL: shuffle_v4i16_2356:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
-; GFX9-NEXT:    global_load_dwordx2 v[2:3], v[2:3], off
-; GFX9-NEXT:    v_mov_b32_e32 v4, 0xffff
-; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_bfi_b32 v0, v4, v1, v1
+; GFX9-NEXT:    global_load_dwordx2 v[4:5], v[0:1], off
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[2:3], off
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_sdwa v1, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v3, 16, v1
+; GFX9-NEXT:    v_and_b32_sdwa v0, v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, v5
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x i16>, <4 x i16> addrspace(1)* %arg0
   %val1 = load <4 x i16>, <4 x i16> addrspace(1)* %arg1
@@ -594,9 +501,7 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[1:2], v[2:3], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0xffff
-; GFX9-NEXT:    v_bfi_b32 v0, v1, v0, v0
-; GFX9-NEXT:    v_bfi_b32 v1, v1, v2, v2
+; GFX9-NEXT:    v_mov_b32_e32 v1, v2
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x i16>, <4 x i16> addrspace(1)* %arg0
   %val1 = load <4 x i16>, <4 x i16> addrspace(1)* %arg1
@@ -660,11 +565,9 @@
 ; GFX9-LABEL: shuffle_v4f16_6161:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dword v0, v[0:1], off
-; GFX9-NEXT:    global_load_dword v1, v[2:3], off offset:4
-; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
+; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[2:3], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v1
 ; GFX9-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v0
@@ -679,14 +582,12 @@
 ; GFX9-LABEL: shuffle_v4f16_2333:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
+; GFX9-NEXT:    global_load_dwordx2 v[1:2], v[0:1], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v0, v2, v1
-; GFX9-NEXT:    v_and_b32_e32 v1, v2, v3
-; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v3, 16, v1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v2
+; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v0
+; GFX9-NEXT:    v_lshl_or_b32 v1, v0, 16, v1
+; GFX9-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -698,14 +599,12 @@
 ; GFX9-LABEL: shuffle_v4f16_6667:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
+; GFX9-NEXT:    global_load_dwordx2 v[1:2], v[0:1], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v0, v2, v1
-; GFX9-NEXT:    v_and_b32_e32 v1, v2, v3
-; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v3, 16, v1
+; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v2
+; GFX9-NEXT:    v_and_b32_e32 v1, 0xffff, v0
+; GFX9-NEXT:    v_lshl_or_b32 v1, v0, 16, v1
+; GFX9-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <4 x half>, <4 x half> addrspace(1)* %arg0
   %val1 = load <4 x half>, <4 x half> addrspace(1)* %arg1
@@ -719,9 +618,6 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v[0:1], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v1, v0
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <8 x half>, <8 x half> addrspace(1)* %arg0
@@ -749,15 +645,7 @@
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v[0:1], off offset:8
 ; GFX9-NEXT:    global_load_dword v1, v[2:3], off
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, v2, v0
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v1, v2, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
-; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v1
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <8 x half>, <8 x half> addrspace(1)* %arg0
   %val1 = load <8 x half>, <8 x half> addrspace(1)* %arg1
@@ -769,17 +657,9 @@
 ; GFX9-LABEL: shuffle_v8f16_10_11_2_3:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dword v0, v[0:1], off offset:4
-; GFX9-NEXT:    global_load_dword v1, v[2:3], off offset:4
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
-; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
+; GFX9-NEXT:    global_load_dword v1, v[0:1], off offset:4
+; GFX9-NEXT:    global_load_dword v0, v[2:3], off offset:4
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
-; GFX9-NEXT:    v_and_b32_e32 v1, v2, v1
-; GFX9-NEXT:    v_and_b32_e32 v2, v2, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v4, 16, v1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v3, 16, v2
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <8 x half>, <8 x half> addrspace(1)* %arg0
   %val1 = load <8 x half>, <8 x half> addrspace(1)* %arg1
@@ -791,15 +671,12 @@
 ; GFX9-LABEL: shuffle_v8f16_13_14_2_3:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dword v4, v[0:1], off offset:4
-; GFX9-NEXT:    global_load_dwordx4 v[0:3], v[2:3], off
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    global_load_dword v1, v[0:1], off offset:4
+; GFX9-NEXT:    global_load_dwordx4 v[2:5], v[2:3], off
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v4
-; GFX9-NEXT:    v_and_b32_e32 v4, v0, v4
-; GFX9-NEXT:    v_and_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v4
-; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_and_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_lshl_or_b32 v0, v5, 16, v0
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <8 x half>, <8 x half> addrspace(1)* %arg0
   %val1 = load <8 x half>, <8 x half> addrspace(1)* %arg1
@@ -812,12 +689,8 @@
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0xffff
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, v2, v0
-; GFX9-NEXT:    v_and_b32_e32 v2, v2, v1
-; GFX9-NEXT:    v_lshl_or_b32 v0, v3, 16, v0
+; GFX9-NEXT:    v_and_b32_e32 v2, 0xffff, v1
 ; GFX9-NEXT:    v_lshl_or_b32 v1, v1, 16, v2
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <3 x half>, <3 x half> addrspace(1)* %arg0
@@ -846,17 +719,14 @@
 ; GFX9-LABEL: shuffle_v6f16_452367:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    global_load_dwordx3 v[4:6], v[0:1], off
-; GFX9-NEXT:    global_load_dword v2, v[2:3], off
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0xffff
+; GFX9-NEXT:    v_mov_b32_e32 v4, v3
+; GFX9-NEXT:    v_mov_b32_e32 v3, v2
+; GFX9-NEXT:    global_load_dwordx3 v[0:2], v[0:1], off
+; GFX9-NEXT:    global_load_dword v3, v[3:4], off
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v4, 16, v5
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 16, v6
-; GFX9-NEXT:    v_and_b32_e32 v3, v0, v6
-; GFX9-NEXT:    v_and_b32_e32 v5, v0, v5
-; GFX9-NEXT:    v_lshl_or_b32 v0, v1, 16, v3
-; GFX9-NEXT:    v_lshl_or_b32 v1, v4, 16, v5
+; GFX9-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v2, v3
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val0 = load <6 x half>, <6 x half> addrspace(1)* %arg0
   %val1 = load <6 x half>, <6 x half> addrspace(1)* %arg1
@@ -888,9 +758,6 @@
 ; GFX9-NEXT:    v_pk_fma_f16 v2, v1, v2, v7 op_sel_hi:[0,1,1]
 ; GFX9-NEXT:    v_pk_fma_f16 v0, v0, v3, v6 op_sel:[1,0,0]
 ; GFX9-NEXT:    v_pk_fma_f16 v1, v1, v3, v2 op_sel:[1,0,0]
-; GFX9-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX9-NEXT:    v_lshl_or_b32 v0, v2, 16, v0
 ; GFX9-NEXT:    global_store_dwordx2 v[4:5], v[0:1], off
 ; GFX9-NEXT:    s_endpgm
 entry: