Index: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
@@ -10515,26 +10515,6 @@
   assert(Mask[0] < 2 && "We sort V1 to be the first input.");
   assert(Mask[1] >= 2 && "We sort V2 to be the second input.");
 
-  // If we have a blend of two same-type PACKUS operations and the blend aligns
-  // with the low and high halves, we can just merge the PACKUS operations.
-  // This is particularly important as it lets us merge shuffles that this
-  // routine itself creates.
-  auto GetPackNode = [](SDValue V) {
-    V = peekThroughBitcasts(V);
-    return V.getOpcode() == X86ISD::PACKUS ? V : SDValue();
-  };
-  if (SDValue V1Pack = GetPackNode(V1))
-    if (SDValue V2Pack = GetPackNode(V2)) {
-      EVT PackVT = V1Pack.getValueType();
-      if (PackVT == V2Pack.getValueType())
-        return DAG.getBitcast(MVT::v2i64,
-                              DAG.getNode(X86ISD::PACKUS, DL, PackVT,
-                                          Mask[0] == 0 ? V1Pack.getOperand(0)
-                                                       : V1Pack.getOperand(1),
-                                          Mask[1] == 2 ? V2Pack.getOperand(0)
-                                                       : V2Pack.getOperand(1)));
-    }
-
   // Try to use shift instructions.
   if (SDValue Shift = lowerVectorShuffleAsShift(DL, MVT::v2i64, V1, V2, Mask,
                                                 Zeroable, Subtarget, DAG))
@@ -28803,8 +28783,37 @@
   SDLoc DL(N);
   MVT VT = N.getSimpleValueType();
   SmallVector<int, 4> Mask;
-
   unsigned Opcode = N.getOpcode();
+
+  // Combine binary shuffle of 2 similar 'Horizontal' instructions into a
+  // single instruction.
+  if (VT.getScalarSizeInBits() == 64 &&
+      (Opcode == X86ISD::MOVSD || Opcode == X86ISD::UNPCKH ||
+       Opcode == X86ISD::UNPCKL)) {
+    auto BC0 = peekThroughBitcasts(N.getOperand(0));
+    auto BC1 = peekThroughBitcasts(N.getOperand(1));
+    EVT VT0 = BC0.getValueType();
+    EVT VT1 = BC1.getValueType();
+    unsigned Opcode0 = BC0.getOpcode();
+    unsigned Opcode1 = BC1.getOpcode();
+    if (Opcode0 == Opcode1 && VT0 == VT1 &&
+        (Opcode0 == X86ISD::FHADD || Opcode0 == X86ISD::HADD ||
+         Opcode0 == X86ISD::FHSUB || Opcode0 == X86ISD::HSUB ||
+         Opcode0 == X86ISD::PACKSS || Opcode0 == X86ISD::PACKUS)) {
+      SDValue Lo, Hi;
+      if (Opcode == X86ISD::MOVSD) {
+        Lo = BC1.getOperand(0);
+        Hi = BC0.getOperand(1);
+      } else {
+        Lo = BC0.getOperand(Opcode == X86ISD::UNPCKH ? 1 : 0);
+        Hi = BC1.getOperand(Opcode == X86ISD::UNPCKH ? 1 : 0);
+      }
+      SDValue Horiz = DAG.getNode(Opcode0, DL, VT0, Lo, Hi);
+      DCI.AddToWorklist(Horiz.getNode());
+      return DAG.getBitcast(VT, Horiz);
+    }
+  }
+
   switch (Opcode) {
   case X86ISD::PSHUFD:
   case X86ISD::PSHUFLW:
@@ -28813,17 +28822,6 @@
     assert(Mask.size() == 4);
     break;
   case X86ISD::UNPCKL: {
-    auto Op0 = N.getOperand(0);
-    auto Op1 = N.getOperand(1);
-    unsigned Opcode0 = Op0.getOpcode();
-    unsigned Opcode1 = Op1.getOpcode();
-
-    // Combine X86ISD::UNPCKL with 2 X86ISD::FHADD inputs into a single
-    // X86ISD::FHADD. This is generated by UINT_TO_FP v2f64 scalarization.
-    // TODO: Add other horizontal operations as required.
-    if (VT == MVT::v2f64 && Opcode0 == Opcode1 && Opcode0 == X86ISD::FHADD)
-      return DAG.getNode(Opcode0, DL, VT, Op0.getOperand(0), Op1.getOperand(0));
-
     // Combine X86ISD::UNPCKL and ISD::VECTOR_SHUFFLE into X86ISD::UNPCKH, in
     // which X86ISD::UNPCKL has a ISD::UNDEF operand, and ISD::VECTOR_SHUFFLE
     // moves upper half elements into the lower half part. For example:
@@ -28841,7 +28839,9 @@
     if (!VT.is128BitVector())
       return SDValue();
 
-    if (Op0.isUndef() && Opcode1 == ISD::VECTOR_SHUFFLE) {
+    auto Op0 = N.getOperand(0);
+    auto Op1 = N.getOperand(1);
+    if (Op0.isUndef() && Op1.getOpcode() == ISD::VECTOR_SHUFFLE) {
       ArrayRef<int> Mask = cast<ShuffleVectorSDNode>(Op1.getNode())->getMask();
 
       unsigned NumElts = VT.getVectorNumElements();
Index: llvm/trunk/test/CodeGen/X86/horizontal-shuffle.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/horizontal-shuffle.ll
+++ llvm/trunk/test/CodeGen/X86/horizontal-shuffle.ll
@@ -9,16 +9,12 @@
 define <4 x float> @test_unpackl_fhadd_128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, <4 x float> %a3) {
 ; X32-LABEL: test_unpackl_fhadd_128:
 ; X32:       ## BB#0:
-; X32-NEXT:    vhaddps %xmm1, %xmm0, %xmm0
-; X32-NEXT:    vhaddps %xmm3, %xmm2, %xmm1
-; X32-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; X32-NEXT:    vhaddps %xmm2, %xmm0, %xmm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackl_fhadd_128:
 ; X64:       ## BB#0:
-; X64-NEXT:    vhaddps %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vhaddps %xmm3, %xmm2, %xmm1
-; X64-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; X64-NEXT:    vhaddps %xmm2, %xmm0, %xmm0
 ; X64-NEXT:    retq
   %1 = call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %a0, <4 x float> %a1)
   %2 = call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %a2, <4 x float> %a3)
@@ -29,16 +25,12 @@
 define <2 x double> @test_unpackh_fhadd_128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, <2 x double> %a3) {
 ; X32-LABEL: test_unpackh_fhadd_128:
 ; X32:       ## BB#0:
-; X32-NEXT:    vhaddpd %xmm1, %xmm0, %xmm0
-; X32-NEXT:    vhaddpd %xmm3, %xmm2, %xmm1
-; X32-NEXT:    vunpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; X32-NEXT:    vhaddpd %xmm3, %xmm1, %xmm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackh_fhadd_128:
 ; X64:       ## BB#0:
-; X64-NEXT:    vhaddpd %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vhaddpd %xmm3, %xmm2, %xmm1
-; X64-NEXT:    vunpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; X64-NEXT:    vhaddpd %xmm3, %xmm1, %xmm0
 ; X64-NEXT:    retq
   %1 = call <2 x double> @llvm.x86.sse3.hadd.pd(<2 x double> %a0, <2 x double> %a1)
   %2 = call <2 x double> @llvm.x86.sse3.hadd.pd(<2 x double> %a2, <2 x double> %a3)
@@ -49,16 +41,12 @@
 define <2 x double> @test_unpackl_fhsub_128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, <2 x double> %a3) {
 ; X32-LABEL: test_unpackl_fhsub_128:
 ; X32:       ## BB#0:
-; X32-NEXT:    vhsubpd %xmm1, %xmm0, %xmm0
-; X32-NEXT:    vhsubpd %xmm3, %xmm2, %xmm1
-; X32-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; X32-NEXT:    vhsubpd %xmm2, %xmm0, %xmm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackl_fhsub_128:
 ; X64:       ## BB#0:
-; X64-NEXT:    vhsubpd %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vhsubpd %xmm3, %xmm2, %xmm1
-; X64-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; X64-NEXT:    vhsubpd %xmm2, %xmm0, %xmm0
 ; X64-NEXT:    retq
   %1 = call <2 x double> @llvm.x86.sse3.hsub.pd(<2 x double> %a0, <2 x double> %a1)
   %2 = call <2 x double> @llvm.x86.sse3.hsub.pd(<2 x double> %a2, <2 x double> %a3)
@@ -69,16 +57,12 @@
 define <4 x float> @test_unpackh_fhsub_128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, <4 x float> %a3) {
 ; X32-LABEL: test_unpackh_fhsub_128:
 ; X32:       ## BB#0:
-; X32-NEXT:    vhsubps %xmm1, %xmm0, %xmm0
-; X32-NEXT:    vhsubps %xmm3, %xmm2, %xmm1
-; X32-NEXT:    vunpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; X32-NEXT:    vhsubps %xmm3, %xmm1, %xmm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackh_fhsub_128:
 ; X64:       ## BB#0:
-; X64-NEXT:    vhsubps %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vhsubps %xmm3, %xmm2, %xmm1
-; X64-NEXT:    vunpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; X64-NEXT:    vhsubps %xmm3, %xmm1, %xmm0
 ; X64-NEXT:    retq
   %1 = call <4 x float> @llvm.x86.sse3.hsub.ps(<4 x float> %a0, <4 x float> %a1)
   %2 = call <4 x float> @llvm.x86.sse3.hsub.ps(<4 x float> %a2, <4 x float> %a3)
@@ -89,16 +73,12 @@
 define <8 x i16> @test_unpackl_hadd_128(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> %a2, <8 x i16> %a3) {
 ; X32-LABEL: test_unpackl_hadd_128:
 ; X32:       ## BB#0:
-; X32-NEXT:    vphaddw %xmm1, %xmm0, %xmm0
-; X32-NEXT:    vphaddw %xmm3, %xmm2, %xmm1
-; X32-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; X32-NEXT:    vphaddw %xmm2, %xmm0, %xmm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackl_hadd_128:
 ; X64:       ## BB#0:
-; X64-NEXT:    vphaddw %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vphaddw %xmm3, %xmm2, %xmm1
-; X64-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; X64-NEXT:    vphaddw %xmm2, %xmm0, %xmm0
 ; X64-NEXT:    retq
   %1 = call <8 x i16> @llvm.x86.ssse3.phadd.w.128(<8 x i16> %a0, <8 x i16> %a1)
   %2 = call <8 x i16> @llvm.x86.ssse3.phadd.w.128(<8 x i16> %a2, <8 x i16> %a3)
@@ -109,16 +89,12 @@
 define <4 x i32> @test_unpackh_hadd_128(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2, <4 x i32> %a3) {
 ; X32-LABEL: test_unpackh_hadd_128:
 ; X32:       ## BB#0:
-; X32-NEXT:    vphaddd %xmm1, %xmm0, %xmm0
-; X32-NEXT:    vphaddd %xmm3, %xmm2, %xmm1
-; X32-NEXT:    vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; X32-NEXT:    vphaddd %xmm3, %xmm1, %xmm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackh_hadd_128:
 ; X64:       ## BB#0:
-; X64-NEXT:    vphaddd %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vphaddd %xmm3, %xmm2, %xmm1
-; X64-NEXT:    vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; X64-NEXT:    vphaddd %xmm3, %xmm1, %xmm0
 ; X64-NEXT:    retq
   %1 = call <4 x i32> @llvm.x86.ssse3.phadd.d.128(<4 x i32> %a0, <4 x i32> %a1)
   %2 = call <4 x i32> @llvm.x86.ssse3.phadd.d.128(<4 x i32> %a2, <4 x i32> %a3)
@@ -129,16 +105,12 @@
 define <4 x i32> @test_unpackl_hsub_128(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2, <4 x i32> %a3) {
 ; X32-LABEL: test_unpackl_hsub_128:
 ; X32:       ## BB#0:
-; X32-NEXT:    vphsubd %xmm1, %xmm0, %xmm0
-; X32-NEXT:    vphsubd %xmm3, %xmm2, %xmm1
-; X32-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; X32-NEXT:    vphsubd %xmm2, %xmm0, %xmm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackl_hsub_128:
 ; X64:       ## BB#0:
-; X64-NEXT:    vphsubd %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vphsubd %xmm3, %xmm2, %xmm1
-; X64-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; X64-NEXT:    vphsubd %xmm2, %xmm0, %xmm0
 ; X64-NEXT:    retq
   %1 = call <4 x i32> @llvm.x86.ssse3.phsub.d.128(<4 x i32> %a0, <4 x i32> %a1)
   %2 = call <4 x i32> @llvm.x86.ssse3.phsub.d.128(<4 x i32> %a2, <4 x i32> %a3)
@@ -149,16 +121,12 @@
 define <8 x i16> @test_unpackh_hsub_128(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> %a2, <8 x i16> %a3) {
 ; X32-LABEL: test_unpackh_hsub_128:
 ; X32:       ## BB#0:
-; X32-NEXT:    vphsubw %xmm1, %xmm0, %xmm0
-; X32-NEXT:    vphsubw %xmm3, %xmm2, %xmm1
-; X32-NEXT:    vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; X32-NEXT:    vphsubw %xmm3, %xmm1, %xmm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackh_hsub_128:
 ; X64:       ## BB#0:
-; X64-NEXT:    vphsubw %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vphsubw %xmm3, %xmm2, %xmm1
-; X64-NEXT:    vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; X64-NEXT:    vphsubw %xmm3, %xmm1, %xmm0
 ; X64-NEXT:    retq
   %1 = call <8 x i16> @llvm.x86.ssse3.phsub.w.128(<8 x i16> %a0, <8 x i16> %a1)
   %2 = call <8 x i16> @llvm.x86.ssse3.phsub.w.128(<8 x i16> %a2, <8 x i16> %a3)
@@ -169,16 +137,12 @@
 define <16 x i8> @test_unpackl_packss_128(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> %a2, <8 x i16> %a3) {
 ; X32-LABEL: test_unpackl_packss_128:
 ; X32:       ## BB#0:
-; X32-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; X32-NEXT:    vpacksswb %xmm3, %xmm2, %xmm1
-; X32-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; X32-NEXT:    vpacksswb %xmm2, %xmm0, %xmm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackl_packss_128:
 ; X64:       ## BB#0:
-; X64-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vpacksswb %xmm3, %xmm2, %xmm1
-; X64-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; X64-NEXT:    vpacksswb %xmm2, %xmm0, %xmm0
 ; X64-NEXT:    retq
   %1 = call <16 x i8> @llvm.x86.sse2.packsswb.128(<8 x i16> %a0, <8 x i16> %a1)
   %2 = call <16 x i8> @llvm.x86.sse2.packsswb.128(<8 x i16> %a2, <8 x i16> %a3)
@@ -189,16 +153,12 @@
 define <8 x i16> @test_unpackh_packss_128(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2, <4 x i32> %a3) {
 ; X32-LABEL: test_unpackh_packss_128:
 ; X32:       ## BB#0:
-; X32-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
-; X32-NEXT:    vpackssdw %xmm3, %xmm2, %xmm1
-; X32-NEXT:    vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; X32-NEXT:    vpackssdw %xmm3, %xmm1, %xmm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackh_packss_128:
 ; X64:       ## BB#0:
-; X64-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vpackssdw %xmm3, %xmm2, %xmm1
-; X64-NEXT:    vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; X64-NEXT:    vpackssdw %xmm3, %xmm1, %xmm0
 ; X64-NEXT:    retq
   %1 = call <8 x i16> @llvm.x86.sse2.packssdw.128(<4 x i32> %a0, <4 x i32> %a1)
   %2 = call <8 x i16> @llvm.x86.sse2.packssdw.128(<4 x i32> %a2, <4 x i32> %a3)
@@ -245,16 +205,12 @@
 define <8 x float> @test_unpackl_fhadd_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, <8 x float> %a3) {
 ; X32-LABEL: test_unpackl_fhadd_256:
 ; X32:       ## BB#0:
-; X32-NEXT:    vhaddps %ymm1, %ymm0, %ymm0
-; X32-NEXT:    vhaddps %ymm3, %ymm2, %ymm1
-; X32-NEXT:    vunpcklpd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
+; X32-NEXT:    vhaddps %ymm2, %ymm0, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackl_fhadd_256:
 ; X64:       ## BB#0:
-; X64-NEXT:    vhaddps %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vhaddps %ymm3, %ymm2, %ymm1
-; X64-NEXT:    vunpcklpd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
+; X64-NEXT:    vhaddps %ymm2, %ymm0, %ymm0
 ; X64-NEXT:    retq
   %1 = call <8 x float> @llvm.x86.avx.hadd.ps.256(<8 x float> %a0, <8 x float> %a1)
   %2 = call <8 x float> @llvm.x86.avx.hadd.ps.256(<8 x float> %a2, <8 x float> %a3)
@@ -265,16 +221,12 @@
 define <4 x double> @test_unpackh_fhadd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, <4 x double> %a3) {
 ; X32-LABEL: test_unpackh_fhadd_256:
 ; X32:       ## BB#0:
-; X32-NEXT:    vhaddpd %ymm1, %ymm0, %ymm0
-; X32-NEXT:    vhaddpd %ymm3, %ymm2, %ymm1
-; X32-NEXT:    vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; X32-NEXT:    vhaddpd %ymm3, %ymm1, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackh_fhadd_256:
 ; X64:       ## BB#0:
-; X64-NEXT:    vhaddpd %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vhaddpd %ymm3, %ymm2, %ymm1
-; X64-NEXT:    vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; X64-NEXT:    vhaddpd %ymm3, %ymm1, %ymm0
 ; X64-NEXT:    retq
   %1 = call <4 x double> @llvm.x86.avx.hadd.pd.256(<4 x double> %a0, <4 x double> %a1)
   %2 = call <4 x double> @llvm.x86.avx.hadd.pd.256(<4 x double> %a2, <4 x double> %a3)
@@ -285,16 +237,12 @@
 define <4 x double> @test_unpackl_fhsub_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, <4 x double> %a3) {
 ; X32-LABEL: test_unpackl_fhsub_256:
 ; X32:       ## BB#0:
-; X32-NEXT:    vhsubpd %ymm1, %ymm0, %ymm0
-; X32-NEXT:    vhsubpd %ymm3, %ymm2, %ymm1
-; X32-NEXT:    vunpcklpd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
+; X32-NEXT:    vhsubpd %ymm2, %ymm0, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackl_fhsub_256:
 ; X64:       ## BB#0:
-; X64-NEXT:    vhsubpd %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vhsubpd %ymm3, %ymm2, %ymm1
-; X64-NEXT:    vunpcklpd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
+; X64-NEXT:    vhsubpd %ymm2, %ymm0, %ymm0
 ; X64-NEXT:    retq
   %1 = call <4 x double> @llvm.x86.avx.hsub.pd.256(<4 x double> %a0, <4 x double> %a1)
   %2 = call <4 x double> @llvm.x86.avx.hsub.pd.256(<4 x double> %a2, <4 x double> %a3)
@@ -305,16 +253,12 @@
 define <8 x float> @test_unpackh_fhsub_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, <8 x float> %a3) {
 ; X32-LABEL: test_unpackh_fhsub_256:
 ; X32:       ## BB#0:
-; X32-NEXT:    vhsubps %ymm1, %ymm0, %ymm0
-; X32-NEXT:    vhsubps %ymm3, %ymm2, %ymm1
-; X32-NEXT:    vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; X32-NEXT:    vhsubps %ymm3, %ymm1, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackh_fhsub_256:
 ; X64:       ## BB#0:
-; X64-NEXT:    vhsubps %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vhsubps %ymm3, %ymm2, %ymm1
-; X64-NEXT:    vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; X64-NEXT:    vhsubps %ymm3, %ymm1, %ymm0
 ; X64-NEXT:    retq
   %1 = call <8 x float> @llvm.x86.avx.hsub.ps.256(<8 x float> %a0, <8 x float> %a1)
   %2 = call <8 x float> @llvm.x86.avx.hsub.ps.256(<8 x float> %a2, <8 x float> %a3)
@@ -325,16 +269,12 @@
 define <16 x i16> @test_unpackl_hadd_256(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> %a2, <16 x i16> %a3) {
 ; X32-LABEL: test_unpackl_hadd_256:
 ; X32:       ## BB#0:
-; X32-NEXT:    vphaddw %ymm1, %ymm0, %ymm0
-; X32-NEXT:    vphaddw %ymm3, %ymm2, %ymm1
-; X32-NEXT:    vpunpcklqdq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
+; X32-NEXT:    vphaddw %ymm2, %ymm0, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackl_hadd_256:
 ; X64:       ## BB#0:
-; X64-NEXT:    vphaddw %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vphaddw %ymm3, %ymm2, %ymm1
-; X64-NEXT:    vpunpcklqdq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
+; X64-NEXT:    vphaddw %ymm2, %ymm0, %ymm0
 ; X64-NEXT:    retq
   %1 = call <16 x i16> @llvm.x86.avx2.phadd.w(<16 x i16> %a0, <16 x i16> %a1)
   %2 = call <16 x i16> @llvm.x86.avx2.phadd.w(<16 x i16> %a2, <16 x i16> %a3)
@@ -345,16 +285,12 @@
 define <8 x i32> @test_unpackh_hadd_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2, <8 x i32> %a3) {
 ; X32-LABEL: test_unpackh_hadd_256:
 ; X32:       ## BB#0:
-; X32-NEXT:    vphaddd %ymm1, %ymm0, %ymm0
-; X32-NEXT:    vphaddd %ymm3, %ymm2, %ymm1
-; X32-NEXT:    vpunpckhqdq {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; X32-NEXT:    vphaddd %ymm3, %ymm1, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackh_hadd_256:
 ; X64:       ## BB#0:
-; X64-NEXT:    vphaddd %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vphaddd %ymm3, %ymm2, %ymm1
-; X64-NEXT:    vpunpckhqdq {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; X64-NEXT:    vphaddd %ymm3, %ymm1, %ymm0
 ; X64-NEXT:    retq
   %1 = call <8 x i32> @llvm.x86.avx2.phadd.d(<8 x i32> %a0, <8 x i32> %a1)
   %2 = call <8 x i32> @llvm.x86.avx2.phadd.d(<8 x i32> %a2, <8 x i32> %a3)
@@ -365,16 +301,12 @@
 define <8 x i32> @test_unpackl_hsub_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2, <8 x i32> %a3) {
 ; X32-LABEL: test_unpackl_hsub_256:
 ; X32:       ## BB#0:
-; X32-NEXT:    vphsubd %ymm1, %ymm0, %ymm0
-; X32-NEXT:    vphsubd %ymm3, %ymm2, %ymm1
-; X32-NEXT:    vpunpcklqdq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
+; X32-NEXT:    vphsubd %ymm2, %ymm0, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackl_hsub_256:
 ; X64:       ## BB#0:
-; X64-NEXT:    vphsubd %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vphsubd %ymm3, %ymm2, %ymm1
-; X64-NEXT:    vpunpcklqdq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
+; X64-NEXT:    vphsubd %ymm2, %ymm0, %ymm0
 ; X64-NEXT:    retq
   %1 = call <8 x i32> @llvm.x86.avx2.phsub.d(<8 x i32> %a0, <8 x i32> %a1)
   %2 = call <8 x i32> @llvm.x86.avx2.phsub.d(<8 x i32> %a2, <8 x i32> %a3)
@@ -385,16 +317,12 @@
 define <16 x i16> @test_unpackh_hsub_256(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> %a2, <16 x i16> %a3) {
 ; X32-LABEL: test_unpackh_hsub_256:
 ; X32:       ## BB#0:
-; X32-NEXT:    vphsubw %ymm1, %ymm0, %ymm0
-; X32-NEXT:    vphsubw %ymm3, %ymm2, %ymm1
-; X32-NEXT:    vpunpckhqdq {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; X32-NEXT:    vphsubw %ymm3, %ymm1, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackh_hsub_256:
 ; X64:       ## BB#0:
-; X64-NEXT:    vphsubw %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vphsubw %ymm3, %ymm2, %ymm1
-; X64-NEXT:    vpunpckhqdq {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; X64-NEXT:    vphsubw %ymm3, %ymm1, %ymm0
 ; X64-NEXT:    retq
   %1 = call <16 x i16> @llvm.x86.avx2.phsub.w(<16 x i16> %a0, <16 x i16> %a1)
   %2 = call <16 x i16> @llvm.x86.avx2.phsub.w(<16 x i16> %a2, <16 x i16> %a3)
@@ -405,16 +333,12 @@
 define <32 x i8> @test_unpackl_packss_256(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> %a2, <16 x i16> %a3) {
 ; X32-LABEL: test_unpackl_packss_256:
 ; X32:       ## BB#0:
-; X32-NEXT:    vpacksswb %ymm1, %ymm0, %ymm0
-; X32-NEXT:    vpacksswb %ymm3, %ymm2, %ymm1
-; X32-NEXT:    vpunpcklqdq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
+; X32-NEXT:    vpacksswb %ymm2, %ymm0, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackl_packss_256:
 ; X64:       ## BB#0:
-; X64-NEXT:    vpacksswb %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vpacksswb %ymm3, %ymm2, %ymm1
-; X64-NEXT:    vpunpcklqdq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
+; X64-NEXT:    vpacksswb %ymm2, %ymm0, %ymm0
 ; X64-NEXT:    retq
   %1 = call <32 x i8> @llvm.x86.avx2.packsswb(<16 x i16> %a0, <16 x i16> %a1)
   %2 = call <32 x i8> @llvm.x86.avx2.packsswb(<16 x i16> %a2, <16 x i16> %a3)
@@ -425,16 +349,12 @@
 define <16 x i16> @test_unpackh_packss_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2, <8 x i32> %a3) {
 ; X32-LABEL: test_unpackh_packss_256:
 ; X32:       ## BB#0:
-; X32-NEXT:    vpackssdw %ymm1, %ymm0, %ymm0
-; X32-NEXT:    vpackssdw %ymm3, %ymm2, %ymm1
-; X32-NEXT:    vpunpckhqdq {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; X32-NEXT:    vpackssdw %ymm3, %ymm1, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackh_packss_256:
 ; X64:       ## BB#0:
-; X64-NEXT:    vpackssdw %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vpackssdw %ymm3, %ymm2, %ymm1
-; X64-NEXT:    vpunpckhqdq {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; X64-NEXT:    vpackssdw %ymm3, %ymm1, %ymm0
 ; X64-NEXT:    retq
   %1 = call <16 x i16> @llvm.x86.avx2.packssdw(<8 x i32> %a0, <8 x i32> %a1)
   %2 = call <16 x i16> @llvm.x86.avx2.packssdw(<8 x i32> %a2, <8 x i32> %a3)
@@ -445,16 +365,12 @@
 define <16 x i16> @test_unpackl_packus_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2, <8 x i32> %a3) {
 ; X32-LABEL: test_unpackl_packus_256:
 ; X32:       ## BB#0:
-; X32-NEXT:    vpackusdw %ymm1, %ymm0, %ymm0
-; X32-NEXT:    vpackusdw %ymm3, %ymm2, %ymm1
-; X32-NEXT:    vpunpcklqdq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
+; X32-NEXT:    vpackusdw %ymm2, %ymm0, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackl_packus_256:
 ; X64:       ## BB#0:
-; X64-NEXT:    vpackusdw %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vpackusdw %ymm3, %ymm2, %ymm1
-; X64-NEXT:    vpunpcklqdq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
+; X64-NEXT:    vpackusdw %ymm2, %ymm0, %ymm0
 ; X64-NEXT:    retq
   %1 = call <16 x i16> @llvm.x86.avx2.packusdw(<8 x i32> %a0, <8 x i32> %a1)
   %2 = call <16 x i16> @llvm.x86.avx2.packusdw(<8 x i32> %a2, <8 x i32> %a3)
@@ -465,16 +381,12 @@
 define <32 x i8> @test_unpackh_packus_256(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> %a2, <16 x i16> %a3) {
 ; X32-LABEL: test_unpackh_packus_256:
 ; X32:       ## BB#0:
-; X32-NEXT:    vpacksswb %ymm1, %ymm0, %ymm0
-; X32-NEXT:    vpacksswb %ymm3, %ymm2, %ymm1
-; X32-NEXT:    vpunpckhqdq {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; X32-NEXT:    vpacksswb %ymm3, %ymm1, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_unpackh_packus_256:
 ; X64:       ## BB#0:
-; X64-NEXT:    vpacksswb %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vpacksswb %ymm3, %ymm2, %ymm1
-; X64-NEXT:    vpunpckhqdq {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; X64-NEXT:    vpacksswb %ymm3, %ymm1, %ymm0
 ; X64-NEXT:    retq
   %1 = call <32 x i8> @llvm.x86.avx2.packsswb(<16 x i16> %a0, <16 x i16> %a1)
   %2 = call <32 x i8> @llvm.x86.avx2.packsswb(<16 x i16> %a2, <16 x i16> %a3)
Index: llvm/trunk/test/CodeGen/X86/vector-trunc.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/vector-trunc.ll
+++ llvm/trunk/test/CodeGen/X86/vector-trunc.ll
@@ -394,12 +394,10 @@
 ;
 ; AVX1-LABEL: trunc8i32_8i16_ashr:
 ; AVX1:       # BB#0: # %entry
-; AVX1-NEXT:    vpsrad $16, %xmm0, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vpsrad $16, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrad $16, %xmm0, %xmm0
-; AVX1-NEXT:    vpackssdw %xmm0, %xmm0, %xmm0
-; AVX1-NEXT:    vpackssdw %xmm0, %xmm1, %xmm1
-; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX1-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;