[X86] Move promotion of vector and/or/xor from legalization to DAG combine

[craig.topper]

Summary

I've noticed that the bitcasts we introduce for these make computeKnownBits and computeNumSignBits not work well in LegalizeVectorOps. LegalizeVectorOps legalizes bottom up while LegalizeDAG legalizes top down. The bottom up strategy for LegalizeVectorOps means operands are legalized before their uses. So we promote and/or/xor before we legalize the operands that use them making computeKnownBits/computeNumSignBits in places like LowerTruncate suboptimal. I looked at changing LegalizeVectorOps to be top down as well, but that was more disruptive and caused some regressions. I also looked at just moving promotion of binops to LegalizeDAG, but that had a few issues one around matching AND,ANDN,OR into VSELECT because I had to create ANDN as vXi64, but the other nodes hadn't legalized yet, I didn't look too hard at fixing that.

This patch seems to produce better results overall than my other attempts. We now form broadcasts of constants better in some cases. For at least some of them the AND was being introduced in LegalizeDAG, promoted to vXi64, and the BUILD_VECTOR was also legalized there. I think we got bad ordering of that. Now the promotion is out of the legalizer so we handle this better.

In the longer term I think we really should evaluate whether we should be doing this promotion at all. It's really there to reduce isel pattern count, but I'm wondering if we'd be better served just eating the pattern cost or doing C++ based isel for vector and/or/xor in X86ISelDAGToDAG. The masked and/or/xor will definitely be difficult in patterns if a bitcast gets between the vselect and the and/or/xor node. That becomes a lot of permutations to cover.

[craig.topper]

test/CodeGen/X86/vector-trunc-usat.ll

734 ↗

(On Diff #169081)

Looks like we're now using a pcmpeq+xor+shr instead of a andn with a constant pool throughout this test. I wonder if this already happens with vXi64 types and now it happens with other vector types.

[RKSimon]

Entirely getting rid of the BITCASTs around logic ops (including X86ISD::ANDNP) would be wonderful - we miss a bit of reassociation later on because of this as well.

[craig.topper]

Rebase

[RKSimon]

The only issue I can see is the PANDN -> XOR+PSRL change from allbits/comparison results - is this something we want to fix before/after this patch?

[craig.topper]

Add a fix to prevent xor+psrl from being selected over andn.

[RKSimon]

LGTM with one minor query - cheers

test/CodeGen/X86/vector-trunc-packus.ll

2083 ↗

(On Diff #169580)

Any idea what this is?

llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

for (auto VT : { MVT::v2f64, MVT::v2i64 }) {
         continue;
 
       setOperationAction(ISD::INSERT_VECTOR_ELT,  VT, Custom);
       setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
     }
 
     // Promote v16i8, v8i16, v4i32 load, select, and, or, xor to v2i64.
     for (auto VT : { MVT::v16i8, MVT::v8i16, MVT::v4i32 }) {
-      setOperationPromotedToType(ISD::AND,    VT, MVT::v2i64);
-      setOperationPromotedToType(ISD::OR,     VT, MVT::v2i64);
-      setOperationPromotedToType(ISD::XOR,    VT, MVT::v2i64);
       setOperationPromotedToType(ISD::LOAD,   VT, MVT::v2i64);
     }
 
     // Custom lower v2i64 and v2f64 selects.
     setOperationAction(ISD::SELECT,             MVT::v2f64, Custom);
     setOperationAction(ISD::SELECT,             MVT::v2i64, Custom);
     setOperationAction(ISD::SELECT,             MVT::v4i32, Custom);
     setOperationAction(ISD::SELECT,             MVT::v8i16, Custom);
for (MVT VT : { MVT::v32i8, MVT::v16i16, MVT::v8i32, MVT::v4i64,
       setOperationAction(ISD::CONCAT_VECTORS,     VT, Custom);
     }
 
     if (HasInt256)
       setOperationAction(ISD::VSELECT,         MVT::v32i8, Legal);
 
     // Promote v32i8, v16i16, v8i32 select, and, or, xor to v4i64.
     for (auto VT : { MVT::v32i8, MVT::v16i16, MVT::v8i32 }) {
-      setOperationPromotedToType(ISD::AND,    VT, MVT::v4i64);
-      setOperationPromotedToType(ISD::OR,     VT, MVT::v4i64);
-      setOperationPromotedToType(ISD::XOR,    VT, MVT::v4i64);
       setOperationPromotedToType(ISD::LOAD,   VT, MVT::v4i64);
     }
 
     if (HasInt256) {
       // Custom legalize 2x32 to get a little better code.
       setOperationAction(ISD::MGATHER, MVT::v2f32, Custom);
       setOperationAction(ISD::MGATHER, MVT::v2i32, Custom);
 
for (auto VT : { MVT::v16i32, MVT::v8i64 }) {
       setOperationAction(ISD::SETCC,            VT, Custom);
 
       // The condition codes aren't legal in SSE/AVX and under AVX512 we use
       // setcc all the way to isel and prefer SETGT in some isel patterns.
       setCondCodeAction(ISD::SETLT, VT, Custom);
       setCondCodeAction(ISD::SETLE, VT, Custom);
     }
 
-    // Need to promote to 64-bit even though we have 32-bit masked instructions
-    // because the IR optimizers rearrange bitcasts around logic ops leaving
-    // too many variations to handle if we don't promote them.
-    setOperationPromotedToType(ISD::AND, MVT::v16i32, MVT::v8i64);
-    setOperationPromotedToType(ISD::OR,  MVT::v16i32, MVT::v8i64);
-    setOperationPromotedToType(ISD::XOR, MVT::v16i32, MVT::v8i64);
-
     if (Subtarget.hasDQI()) {
       setOperationAction(ISD::SINT_TO_FP, MVT::v8i64, Legal);
       setOperationAction(ISD::UINT_TO_FP, MVT::v8i64, Legal);
       setOperationAction(ISD::FP_TO_SINT, MVT::v8i64, Legal);
       setOperationAction(ISD::FP_TO_UINT, MVT::v8i64, Legal);
 
       setOperationAction(ISD::MUL,        MVT::v8i64, Legal);
     }
for (auto VT : { MVT::v64i8, MVT::v32i16 }) {
       setOperationAction(ISD::CTPOP,        VT, Custom);
       setOperationAction(ISD::CTLZ,         VT, Custom);
       setOperationAction(ISD::SMAX,         VT, Legal);
       setOperationAction(ISD::UMAX,         VT, Legal);
       setOperationAction(ISD::SMIN,         VT, Legal);
       setOperationAction(ISD::UMIN,         VT, Legal);
       setOperationAction(ISD::SETCC,        VT, Custom);
 
-      setOperationPromotedToType(ISD::AND,  VT, MVT::v8i64);
-      setOperationPromotedToType(ISD::OR,   VT, MVT::v8i64);
-      setOperationPromotedToType(ISD::XOR,  VT, MVT::v8i64);
-
       // The condition codes aren't legal in SSE/AVX and under AVX512 we use
       // setcc all the way to isel and prefer SETGT in some isel patterns.
       setCondCodeAction(ISD::SETLT, VT, Custom);
       setCondCodeAction(ISD::SETLE, VT, Custom);
     }
 
     for (auto ExtType : {ISD::ZEXTLOAD, ISD::SEXTLOAD}) {
       setLoadExtAction(ExtType, MVT::v32i16, MVT::v32i8, Legal);
static SDValue combineAndMaskToShift(SDNode *N, SelectionDAG &DAG,
   if (VT0 != VT1 || !VT0.isSimple() || !VT0.isInteger())
     return SDValue();
 
   APInt SplatVal;
   if (!ISD::isConstantSplatVector(Op1.getNode(), SplatVal) ||
       !SplatVal.isMask())
     return SDValue();
 
+  // Don't prevent creation of ANDN.
+  if (isBitwiseNot(Op0))
+    return SDValue();
+
   if (!SupportedVectorShiftWithImm(VT0.getSimpleVT(), Subtarget, ISD::SRL))
     return SDValue();
 
   unsigned EltBitWidth = VT0.getScalarSizeInBits();
   if (EltBitWidth != DAG.ComputeNumSignBits(Op0))
     return SDValue();
 
   SDLoc DL(N);
static SDValue combineParity(SDNode *N, SelectionDAG &DAG,
   SDValue Flags = DAG.getNode(X86ISD::XOR, DL, VTs, Lo, Hi).getValue(1);
 
   // Copy the inverse of the parity flag into a register with setcc.
   SDValue Setnp = getSETCC(X86::COND_NP, Flags, DL, DAG);
   // Zero extend to original type.
   return DAG.getNode(ISD::ZERO_EXTEND, DL, N->getValueType(0), Setnp);
 }
 
+// This promotes vectors and/or/xor to a vXi64 type. We used to do this during
+// op legalization, but DAG combine yields better results.
+// TODO: This is largely just to reduce the number of isel patterns. Maybe we
+// can just add all the patterns or do C++ based selection in X86ISelDAGToDAG?
+static SDValue promoteVecLogicOp(SDNode *N, SelectionDAG &DAG) {
+  MVT VT = N->getSimpleValueType(0);
+
+  if (!VT.is128BitVector() && !VT.is256BitVector() && !VT.is512BitVector())
+    return SDValue();
+
+  // Already correct type.
+  if (VT.getVectorElementType() == MVT::i64)
+    return SDValue();
+
+  MVT NewVT = MVT::getVectorVT(MVT::i64, VT.getSizeInBits() / 64);
+  SDValue Op0 = DAG.getBitcast(NewVT, N->getOperand(0));
+  SDValue Op1 = DAG.getBitcast(NewVT, N->getOperand(1));
+  return DAG.getBitcast(VT, DAG.getNode(N->getOpcode(), SDLoc(N), NewVT,
+                                        Op0, Op1));
+}
+
 static SDValue combineAnd(SDNode *N, SelectionDAG &DAG,
                           TargetLowering::DAGCombinerInfo &DCI,
                           const X86Subtarget &Subtarget) {
   EVT VT = N->getValueType(0);
 
   // If this is SSE1 only convert to FAND to avoid scalarization.
   if (Subtarget.hasSSE1() && !Subtarget.hasSSE2() && VT == MVT::v4i32) {
     return DAG.getBitcast(
static SDValue combineAnd(SDNode *N, SelectionDAG &DAG,
 
   // This must be done before legalization has expanded the ctpop.
   if (SDValue V = combineParity(N, DAG, Subtarget))
     return V;
 
   if (DCI.isBeforeLegalizeOps())
     return SDValue();
 
+  if (SDValue V = promoteVecLogicOp(N, DAG))
+    return V;
+
   if (SDValue R = combineCompareEqual(N, DAG, DCI, Subtarget))
     return R;
 
   if (SDValue FPLogic = convertIntLogicToFPLogic(N, DAG, Subtarget))
     return FPLogic;
 
   if (SDValue R = combineANDXORWithAllOnesIntoANDNP(N, DAG))
     return R;
return DAG.getBitcast(MVT::v4i32,
                           DAG.getNode(X86ISD::FOR, SDLoc(N), MVT::v4f32,
                                       DAG.getBitcast(MVT::v4f32, N0),
                                       DAG.getBitcast(MVT::v4f32, N1)));
   }
 
   if (DCI.isBeforeLegalizeOps())
     return SDValue();
 
+  if (SDValue V = promoteVecLogicOp(N, DAG))
+    return V;
+
   if (SDValue R = combineCompareEqual(N, DAG, DCI, Subtarget))
     return R;
 
   if (SDValue FPLogic = convertIntLogicToFPLogic(N, DAG, Subtarget))
     return FPLogic;
 
   if (SDValue R = combineLogicBlendIntoPBLENDV(N, DAG, Subtarget))
     return R;
static SDValue combineXor(SDNode *N, SelectionDAG &DAG,
   }
 
   if (SDValue Cmp = foldVectorXorShiftIntoCmp(N, DAG, Subtarget))
     return Cmp;
 
   if (DCI.isBeforeLegalizeOps())
     return SDValue();
 
+  if (SDValue V = promoteVecLogicOp(N, DAG))
+    return V;
+
   if (SDValue SetCC = foldXor1SetCC(N, DAG))
     return SetCC;
 
   if (SDValue RV = foldXorTruncShiftIntoCmp(N, DAG))
     return RV;
 
   if (SDValue FPLogic = convertIntLogicToFPLogic(N, DAG, Subtarget))
     return FPLogic;

llvm/trunk/test/CodeGen/X86/avx-logic.ll

 
 define <8 x i32> @and_disguised_i8_elts(<8 x i32> %x, <8 x i32> %y, <8 x i32> %z) {
 ; AVX1-LABEL: and_disguised_i8_elts:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpaddd %xmm1, %xmm0, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [1095216660735,1095216660735]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [255,255,255,255]
 ; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
 ; AVX1-NEXT:    vpaddd %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;

 
 define <8 x i32> @andn_disguised_i8_elts(<8 x i32> %x, <8 x i32> %y, <8 x i32> %z) {
 ; AVX1-LABEL: andn_disguised_i8_elts:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
 ; AVX1-NEXT:    vpaddd %xmm3, %xmm4, %xmm3
 ; AVX1-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [1095216660735,1095216660735]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [255,255,255,255]
 ; AVX1-NEXT:    vpandn %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpandn %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
 ; AVX1-NEXT:    vpaddd %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpaddd %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;

 
 define <8 x i32> @or_disguised_i8_elts(<8 x i32> %x, <8 x i32> %y, <8 x i32> %z) {
 ; AVX1-LABEL: or_disguised_i8_elts:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpaddd %xmm1, %xmm0, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [1095216660735,1095216660735]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [255,255,255,255]
 ; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
 ; AVX1-NEXT:    vpaddd %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;

 
 define <8 x i32> @xor_disguised_i8_elts(<8 x i32> %x, <8 x i32> %y, <8 x i32> %z) {
 ; AVX1-LABEL: xor_disguised_i8_elts:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpaddd %xmm1, %xmm0, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [1095216660735,1095216660735]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [255,255,255,255]
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
 ; AVX1-NEXT:    vpaddd %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;

 
 define <8 x i32> @or_disguised_i16_elts(<8 x i32> %x, <8 x i32> %y, <8 x i32> %z) {
 ; AVX1-LABEL: or_disguised_i16_elts:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpaddd %xmm1, %xmm0, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [281470681808895,281470681808895]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [65535,65535,65535,65535]
 ; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
 ; AVX1-NEXT:    vpaddd %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;

 
 define <8 x i32> @xor_disguised_i16_elts(<8 x i32> %x, <8 x i32> %y, <8 x i32> %z) {
 ; AVX1-LABEL: xor_disguised_i16_elts:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpaddd %xmm1, %xmm0, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [281470681808895,281470681808895]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [65535,65535,65535,65535]
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
 ; AVX1-NEXT:    vpaddd %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;

llvm/trunk/test/CodeGen/X86/avx512-ext.ll

; AVX512DQNOBW-NEXT:    retq
   %mask = icmp eq <32 x i16> %x, %y
   %1 = zext <32 x i1> %mask to <32 x i8>
   ret <32 x i8> %1
 }
 
 define <4 x i32> @zext_4xi1_to_4x32(<4 x i8> %x, <4 x i8> %y) #0 {
 ; ALL-LABEL: zext_4xi1_to_4x32:
 ; ALL:       # %bb.0:
-; ALL-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; ALL-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [255,255,255,255]
 ; ALL-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; ALL-NEXT:    vpand %xmm2, %xmm0, %xmm0
 ; ALL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; ALL-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; ALL-NEXT:    retq
   %mask = icmp eq <4 x i8> %x, %y
   %1 = zext <4 x i1> %mask to <4 x i32>
   ret <4 x i32> %1
 }
 
 define <2 x i64> @zext_2xi1_to_2xi64(<2 x i8> %x, <2 x i8> %y) #0 {
 ; ALL-LABEL: zext_2xi1_to_2xi64:
 ; ALL:       # %bb.0:
-; ALL-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
+; ALL-NEXT:    vpbroadcastq {{.*#+}} xmm2 = [255,255]
 ; ALL-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; ALL-NEXT:    vpand %xmm2, %xmm0, %xmm0
 ; ALL-NEXT:    vpcmpeqq %xmm1, %xmm0, %xmm0
 ; ALL-NEXT:    vpsrlq $63, %xmm0, %xmm0
 ; ALL-NEXT:    retq
   %mask = icmp eq <2 x i8> %x, %y
   %1 = zext <2 x i1> %mask to <2 x i64>
   ret <2 x i64> %1
 }

llvm/trunk/test/CodeGen/X86/avx512-insert-extract.ll

 
 define zeroext i8 @test_extractelement_v32i1(<32 x i8> %a, <32 x i8> %b) {
 ; KNL-LABEL: test_extractelement_v32i1:
 ; KNL:       ## %bb.0:
 ; KNL-NEXT:    vpminub %ymm1, %ymm0, %ymm1
 ; KNL-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
 ; KNL-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; KNL-NEXT:    kshiftrw $2, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    andl $1, %eax
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test_extractelement_v32i1:

 define zeroext i8 @test_extractelement_v64i1(<64 x i8> %a, <64 x i8> %b) {
 ; KNL-LABEL: test_extractelement_v64i1:
 ; KNL:       ## %bb.0:
 ; KNL-NEXT:    vpminub %ymm3, %ymm1, %ymm0
 ; KNL-NEXT:    vpcmpeqb %ymm0, %ymm1, %ymm0
 ; KNL-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    andb $1, %al
 ; KNL-NEXT:    movb $4, %cl
 ; KNL-NEXT:    subb %al, %cl
 ; KNL-NEXT:    movzbl %cl, %eax
 ; KNL-NEXT:    vzeroupper

 define zeroext i8 @extractelement_v64i1_alt(<64 x i8> %a, <64 x i8> %b) {
 ; KNL-LABEL: extractelement_v64i1_alt:
 ; KNL:       ## %bb.0:
 ; KNL-NEXT:    vpminub %ymm3, %ymm1, %ymm0
 ; KNL-NEXT:    vpcmpeqb %ymm0, %ymm1, %ymm0
 ; KNL-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    andb $1, %al
 ; KNL-NEXT:    movb $4, %cl
 ; KNL-NEXT:    subb %al, %cl
 ; KNL-NEXT:    movzbl %cl, %eax
 ; KNL-NEXT:    vzeroupper

llvm/trunk/test/CodeGen/X86/avx512-schedule.ll

; SKX-NEXT:    retq # sched: [7:1.00]
   %mask = icmp eq <32 x i16> %x, %y
   %1 = zext <32 x i1> %mask to <32 x i8>
   ret <32 x i8> %1
 }
 
 define <4 x i32> @zext_4xi1_to_4x32(<4 x i8> %x, <4 x i8> %y) #0 {
 ; GENERIC-LABEL: zext_4xi1_to_4x32:
 ; GENERIC:       # %bb.0:
-; GENERIC-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0] sched: [6:0.50]
+; GENERIC-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [255,255,255,255] sched: [7:0.50]
 ; GENERIC-NEXT:    vpand %xmm2, %xmm1, %xmm1 # sched: [1:0.33]
 ; GENERIC-NEXT:    vpand %xmm2, %xmm0, %xmm0 # sched: [1:0.33]
 ; GENERIC-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
 ; GENERIC-NEXT:    vpsrld $31, %xmm0, %xmm0 # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: zext_4xi1_to_4x32:
 ; SKX:       # %bb.0:
-; SKX-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0] sched: [6:0.50]
+; SKX-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [255,255,255,255] sched: [6:0.50]
 ; SKX-NEXT:    vpand %xmm2, %xmm1, %xmm1 # sched: [1:0.33]
 ; SKX-NEXT:    vpand %xmm2, %xmm0, %xmm0 # sched: [1:0.33]
 ; SKX-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
 ; SKX-NEXT:    vpsrld $31, %xmm0, %xmm0 # sched: [1:0.50]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %mask = icmp eq <4 x i8> %x, %y
   %1 = zext <4 x i1> %mask to <4 x i32>
   ret <4 x i32> %1
 }
 
 define <2 x i64> @zext_2xi1_to_2xi64(<2 x i8> %x, <2 x i8> %y) #0 {
 ; GENERIC-LABEL: zext_2xi1_to_2xi64:
 ; GENERIC:       # %bb.0:
-; GENERIC-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0] sched: [6:0.50]
+; GENERIC-NEXT:    vpbroadcastq {{.*#+}} xmm2 = [255,255] sched: [7:0.50]
 ; GENERIC-NEXT:    vpand %xmm2, %xmm1, %xmm1 # sched: [1:0.33]
 ; GENERIC-NEXT:    vpand %xmm2, %xmm0, %xmm0 # sched: [1:0.33]
 ; GENERIC-NEXT:    vpcmpeqq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
 ; GENERIC-NEXT:    vpsrlq $63, %xmm0, %xmm0 # sched: [1:1.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: zext_2xi1_to_2xi64:
 ; SKX:       # %bb.0:
-; SKX-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0] sched: [6:0.50]
+; SKX-NEXT:    vpbroadcastq {{.*#+}} xmm2 = [255,255] sched: [6:0.50]
 ; SKX-NEXT:    vpand %xmm2, %xmm1, %xmm1 # sched: [1:0.33]
 ; SKX-NEXT:    vpand %xmm2, %xmm0, %xmm0 # sched: [1:0.33]
 ; SKX-NEXT:    vpcmpeqq %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
 ; SKX-NEXT:    vpsrlq $63, %xmm0, %xmm0 # sched: [1:0.50]
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %mask = icmp eq <2 x i8> %x, %y
   %1 = zext <2 x i1> %mask to <2 x i64>
   ret <2 x i64> %1

llvm/trunk/test/CodeGen/X86/avx512vl-vec-masked-cmp.ll

 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeb_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp sge <16 x i8> %0, %1

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeb_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>

 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeb_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeb_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeb_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeb_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeb_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeb_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeb_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    shll $16, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeb_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    shll $16, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>

 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeb_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    shrl $16, %edi
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    andl %edi, %ecx
 ; NoVLX-NEXT:    shll $16, %ecx
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeb_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    shrl $16, %edi
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    andl %edi, %ecx
 ; NoVLX-NEXT:    shll $16, %ecx
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq

 ; VLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v8i1_v16i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v8i1_v16i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v8i1_v16i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v8i1_v16i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>

 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp sge <8 x i16> %0, %1

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>

 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rdi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rsi), %xmm1
 ; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>

 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp sge <16 x i16> %0, %1

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>

 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b

 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgew_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b

 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b

 ; NoVLX-LABEL: test_vpcmpsgew_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
 ; NoVLX-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
 ; NoVLX-NEXT:    vpcmpgtw %ymm2, %ymm3, %ymm2
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    vpternlogq $15, %zmm2, %zmm2, %zmm2
 ; NoVLX-NEXT:    vpmovsxwd %ymm2, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    shll $16, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>

 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; NoVLX-NEXT:    vmovdqa (%rdi), %ymm2
 ; NoVLX-NEXT:    vmovdqa 32(%rdi), %ymm3
 ; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm3, %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm2, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    vpternlogq $15, %zmm1, %zmm1, %zmm1
 ; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    shll $16, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>

 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm2
 ; NoVLX-NEXT:    vpternlogq $15, %zmm2, %zmm2, %zmm2
 ; NoVLX-NEXT:    vpmovsxwd %ymm2, %zmm2
-; NoVLX-NEXT:    vpslld $31, %zmm2, %zmm2
 ; NoVLX-NEXT:    vptestmd %zmm2, %zmm2, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    shrl $16, %edi
 ; NoVLX-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
 ; NoVLX-NEXT:    vextracti64x4 $1, %zmm1, %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    andl %edi, %ecx
 ; NoVLX-NEXT:    shll $16, %ecx
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgew_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vmovdqa (%rsi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm1
 ; NoVLX-NEXT:    vpternlogq $15, %zmm1, %zmm1, %zmm1
 ; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    shrl $16, %edi
 ; NoVLX-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
 ; NoVLX-NEXT:    vmovdqa 32(%rsi), %ymm1
 ; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    andl %edi, %ecx
 ; NoVLX-NEXT:    shll $16, %ecx
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp ult <16 x i8> %0, %1

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxub (%rdi), %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxub (%rsi), %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxub (%rdi), %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxub (%rsi), %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxub %ymm1, %ymm0, %ymm1
 ; NoVLX-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    shll $16, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxub (%rdi), %ymm0, %ymm1
 ; NoVLX-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    shll $16, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxub %ymm1, %ymm0, %ymm1
 ; NoVLX-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    shrl $16, %edi
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    andl %edi, %ecx
 ; NoVLX-NEXT:    shll $16, %ecx
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxub (%rsi), %ymm0, %ymm1
 ; NoVLX-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    shrl $16, %edi
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    andl %edi, %ecx
 ; NoVLX-NEXT:    shll $16, %ecx
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v8i1_v16i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v8i1_v16i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw (%rdi), %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v8i1_v16i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v8i1_v16i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw (%rsi), %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp ult <8 x i16> %0, %1

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw (%rdi), %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw (%rsi), %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw (%rdi), %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw (%rsi), %xmm0, %xmm1
 ; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
-; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm1
 ; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp ult <16 x i16> %0, %1

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw (%rdi), %ymm0, %ymm1
 ; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm1
 ; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw (%rsi), %ymm0, %ymm1
 ; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm1
 ; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw (%rdi), %ymm0, %ymm1
 ; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm1
 ; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw (%rsi), %ymm0, %ymm1
 ; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b

 ; NoVLX-LABEL: test_vpcmpultw_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
 ; NoVLX-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
 ; NoVLX-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm1
 ; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    vpmaxuw %ymm3, %ymm2, %ymm0
 ; NoVLX-NEXT:    vpcmpeqw %ymm0, %ymm2, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    shll $16, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>

 ;
 ; NoVLX-LABEL: test_vpcmpultw_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; NoVLX-NEXT:    vpmaxuw (%rdi), %ymm0, %ymm2
 ; NoVLX-NEXT:    vpcmpeqw %ymm2, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    vpmaxuw 32(%rdi), %ymm1, %ymm0
 ; NoVLX-NEXT:    vpcmpeqw %ymm0, %ymm1, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    shll $16, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm2
 ; NoVLX-NEXT:    vpcmpeqw %ymm2, %ymm0, %ymm2
 ; NoVLX-NEXT:    vpternlogq $15, %zmm2, %zmm2, %zmm2
 ; NoVLX-NEXT:    vpmovsxwd %ymm2, %zmm2
-; NoVLX-NEXT:    vpslld $31, %zmm2, %zmm2
 ; NoVLX-NEXT:    vptestmd %zmm2, %zmm2, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    shrl $16, %edi
 ; NoVLX-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
 ; NoVLX-NEXT:    vextracti64x4 $1, %zmm1, %ymm1
 ; NoVLX-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm1
 ; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    andl %edi, %ecx
 ; NoVLX-NEXT:    shll $16, %ecx
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq

 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vpmaxuw (%rsi), %ymm0, %ymm1
 ; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm1
 ; NoVLX-NEXT:    vpternlogq $15, %zmm1, %zmm1, %zmm1
 ; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    shrl $16, %edi
 ; NoVLX-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
 ; NoVLX-NEXT:    vpmaxuw 32(%rsi), %ymm0, %ymm1
 ; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
-; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    andl %edi, %ecx
 ; NoVLX-NEXT:    shll $16, %ecx
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq

llvm/trunk/test/CodeGen/X86/cast-vsel.ll

 ; AVX1-NEXT:    jne .LBB5_1
 ; AVX1-NEXT:  # %bb.2: # %for.end
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: example25:
 ; AVX2:       # %bb.0: # %vector.ph
 ; AVX2-NEXT:    movq $-4096, %rax # imm = 0xF000
-; AVX2-NEXT:    vbroadcastss {{.*#+}} ymm0 = [1,1,1,1,1,1,1,1]
 ; AVX2-NEXT:    .p2align 4, 0x90
 ; AVX2-NEXT:  .LBB5_1: # %vector.body
 ; AVX2-NEXT:    # =>This Inner Loop Header: Depth=1
-; AVX2-NEXT:    vmovups da+4096(%rax), %ymm1
-; AVX2-NEXT:    vcmpltps db+4096(%rax), %ymm1, %ymm1
-; AVX2-NEXT:    vmovups dc+4096(%rax), %ymm2
-; AVX2-NEXT:    vcmpltps dd+4096(%rax), %ymm2, %ymm2
-; AVX2-NEXT:    vandps %ymm0, %ymm2, %ymm2
-; AVX2-NEXT:    vandps %ymm2, %ymm1, %ymm1
-; AVX2-NEXT:    vmovups %ymm1, dj+4096(%rax)
+; AVX2-NEXT:    vmovups da+4096(%rax), %ymm0
+; AVX2-NEXT:    vcmpltps db+4096(%rax), %ymm0, %ymm0
+; AVX2-NEXT:    vmovups dc+4096(%rax), %ymm1
+; AVX2-NEXT:    vcmpltps dd+4096(%rax), %ymm1, %ymm1
+; AVX2-NEXT:    vandps %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpsrld $31, %ymm0, %ymm0
+; AVX2-NEXT:    vmovdqu %ymm0, dj+4096(%rax)
 ; AVX2-NEXT:    addq $32, %rax
 ; AVX2-NEXT:    jne .LBB5_1
 ; AVX2-NEXT:  # %bb.2: # %for.end
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 vector.ph:
   br label %vector.body
 

llvm/trunk/test/CodeGen/X86/combine-sdiv.ll

 ; AVX1-NEXT:    vpsraw $15, %xmm0, %xmm4
 ; AVX1-NEXT:    vpmulhuw %xmm3, %xmm4, %xmm3
 ; AVX1-NEXT:    vpaddw %xmm3, %xmm0, %xmm3
 ; AVX1-NEXT:    vpmulhw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0],xmm2[1,2,3,4,5,6,7]
 ; AVX1-NEXT:    vpsraw $1, %xmm3, %xmm3
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3,4,5,6],xmm3[7]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
-; AVX1-NEXT:    vmovaps {{.*#+}} ymm2 = [0,65535,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,65535]
+; AVX1-NEXT:    vbroadcastf128 {{.*#+}} ymm2 = [0,65535,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,65535]
+; AVX1-NEXT:    # ymm2 = mem[0,1,0,1]
 ; AVX1-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; AVX1-NEXT:    vandnps %ymm0, %ymm2, %ymm0
 ; AVX1-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: combine_vec_sdiv_by_pow2b_v16i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpsraw $15, %ymm0, %ymm1

 ; XOP-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
 ; XOP-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,65534,65535,65532,65533,65531,65530,65535]
 ; XOP-NEXT:    vpshaw %xmm2, %xmm1, %xmm1
 ; XOP-NEXT:    vpsraw $15, %xmm0, %xmm4
 ; XOP-NEXT:    vpshlw %xmm3, %xmm4, %xmm3
 ; XOP-NEXT:    vpaddw %xmm3, %xmm0, %xmm3
 ; XOP-NEXT:    vpshaw %xmm2, %xmm3, %xmm2
 ; XOP-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
-; XOP-NEXT:    vpcmov {{.*}}(%rip), %ymm0, %ymm1, %ymm0
+; XOP-NEXT:    vbroadcastf128 {{.*#+}} ymm2 = [0,65535,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,65535]
+; XOP-NEXT:    # ymm2 = mem[0,1,0,1]
+; XOP-NEXT:    vpcmov %ymm2, %ymm0, %ymm1, %ymm0
 ; XOP-NEXT:    retq
   %1 = sdiv <16 x i16> %x, <i16 1, i16 4, i16 2, i16 16, i16 8, i16 32, i16 64, i16 2, i16 1, i16 4, i16 2, i16 16, i16 8, i16 32, i16 64, i16 2>
   ret <16 x i16> %1
 }
 
 define <32 x i16> @combine_vec_sdiv_by_pow2b_v32i16(<32 x i16> %x) {
 ; SSE2-LABEL: combine_vec_sdiv_by_pow2b_v32i16:
 ; SSE2:       # %bb.0:

 ; AVX1-NEXT:    vpsraw $15, %xmm0, %xmm5
 ; AVX1-NEXT:    vpmulhuw %xmm4, %xmm5, %xmm5
 ; AVX1-NEXT:    vpaddw %xmm5, %xmm0, %xmm5
 ; AVX1-NEXT:    vpmulhw %xmm3, %xmm5, %xmm6
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm6 = xmm5[0],xmm6[1,2,3,4,5,6,7]
 ; AVX1-NEXT:    vpsraw $1, %xmm5, %xmm5
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm6[0,1],xmm5[2],xmm6[3,4,5,6],xmm5[7]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm5, %ymm2
-; AVX1-NEXT:    vmovaps {{.*#+}} ymm5 = [0,65535,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,65535]
+; AVX1-NEXT:    vbroadcastf128 {{.*#+}} ymm5 = [0,65535,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,65535]
+; AVX1-NEXT:    # ymm5 = mem[0,1,0,1]
 ; AVX1-NEXT:    vandps %ymm5, %ymm2, %ymm2
 ; AVX1-NEXT:    vandnps %ymm0, %ymm5, %ymm0
 ; AVX1-NEXT:    vorps %ymm0, %ymm2, %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vpsraw $15, %xmm2, %xmm6
 ; AVX1-NEXT:    vpmulhuw %xmm4, %xmm6, %xmm6
 ; AVX1-NEXT:    vpaddw %xmm6, %xmm2, %xmm2
 ; AVX1-NEXT:    vpmulhw %xmm3, %xmm2, %xmm6

 ; XOP-NEXT:    vpaddw %xmm3, %xmm2, %xmm2
 ; XOP-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,65534,65535,65532,65533,65531,65530,65535]
 ; XOP-NEXT:    vpshaw %xmm3, %xmm2, %xmm2
 ; XOP-NEXT:    vpsraw $15, %xmm0, %xmm5
 ; XOP-NEXT:    vpshlw %xmm4, %xmm5, %xmm5
 ; XOP-NEXT:    vpaddw %xmm5, %xmm0, %xmm5
 ; XOP-NEXT:    vpshaw %xmm3, %xmm5, %xmm5
 ; XOP-NEXT:    vinsertf128 $1, %xmm2, %ymm5, %ymm2
-; XOP-NEXT:    vmovdqa {{.*#+}} ymm5 = [0,65535,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,65535]
+; XOP-NEXT:    vbroadcastf128 {{.*#+}} ymm5 = [0,65535,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,65535]
+; XOP-NEXT:    # ymm5 = mem[0,1,0,1]
 ; XOP-NEXT:    vpcmov %ymm5, %ymm0, %ymm2, %ymm0
 ; XOP-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; XOP-NEXT:    vpsraw $15, %xmm2, %xmm6
 ; XOP-NEXT:    vpshlw %xmm4, %xmm6, %xmm6
 ; XOP-NEXT:    vpaddw %xmm6, %xmm2, %xmm2
 ; XOP-NEXT:    vpshaw %xmm3, %xmm2, %xmm2
 ; XOP-NEXT:    vpsraw $15, %xmm1, %xmm6
 ; XOP-NEXT:    vpshlw %xmm4, %xmm6, %xmm4

llvm/trunk/test/CodeGen/X86/combine-srl.ll

; AVX-NEXT:    retq
   %3 = lshr <4 x i32> %2, <i32 5, i32 5, i32 5, i32 5>
   ret <4 x i32> %3
 }
 
 define <4 x i32> @combine_vec_lshr_lzcnt_bit1(<4 x i32> %x) {
 ; SSE-LABEL: combine_vec_lshr_lzcnt_bit1:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pand {{.*}}(%rip), %xmm0
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE-NEXT:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    pand %xmm2, %xmm1
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSE-NEXT:    movdqa %xmm3, %xmm4
-; SSE-NEXT:    pshufb %xmm1, %xmm4
 ; SSE-NEXT:    movdqa %xmm0, %xmm1
 ; SSE-NEXT:    psrlw $4, %xmm1
-; SSE-NEXT:    pand %xmm2, %xmm1
 ; SSE-NEXT:    pxor %xmm2, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; SSE-NEXT:    movdqa %xmm3, %xmm4
 ; SSE-NEXT:    pshufb %xmm1, %xmm3
 ; SSE-NEXT:    pcmpeqb %xmm2, %xmm1
-; SSE-NEXT:    pand %xmm4, %xmm1
-; SSE-NEXT:    paddb %xmm3, %xmm1
-; SSE-NEXT:    movdqa %xmm0, %xmm3
-; SSE-NEXT:    pcmpeqb %xmm2, %xmm3
-; SSE-NEXT:    psrlw $8, %xmm3
-; SSE-NEXT:    pand %xmm1, %xmm3
+; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSE-NEXT:    pand %xmm0, %xmm5
+; SSE-NEXT:    pshufb %xmm5, %xmm4
+; SSE-NEXT:    pand %xmm1, %xmm4
+; SSE-NEXT:    paddb %xmm4, %xmm3
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    pcmpeqb %xmm2, %xmm1
 ; SSE-NEXT:    psrlw $8, %xmm1
-; SSE-NEXT:    paddw %xmm3, %xmm1
+; SSE-NEXT:    pand %xmm3, %xmm1
+; SSE-NEXT:    psrlw $8, %xmm3
+; SSE-NEXT:    paddw %xmm1, %xmm3
 ; SSE-NEXT:    pcmpeqw %xmm2, %xmm0
 ; SSE-NEXT:    psrld $16, %xmm0
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    psrld $16, %xmm1
-; SSE-NEXT:    paddd %xmm0, %xmm1
-; SSE-NEXT:    psrld $5, %xmm1
-; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    pand %xmm3, %xmm0
+; SSE-NEXT:    psrld $16, %xmm3
+; SSE-NEXT:    paddd %xmm3, %xmm0
+; SSE-NEXT:    psrld $5, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: combine_vec_lshr_lzcnt_bit1:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX-NEXT:    vpand %xmm1, %xmm4, %xmm1
-; AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX-NEXT:    vpcmpeqb %xmm4, %xmm1, %xmm5
-; AVX-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
-; AVX-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
-; AVX-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm2
-; AVX-NEXT:    vpsrlw $8, %xmm2, %xmm2
-; AVX-NEXT:    vpand %xmm2, %xmm1, %xmm2
+; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm1
+; AVX-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX-NEXT:    vpcmpeqb %xmm2, %xmm1, %xmm3
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm4
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm5 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX-NEXT:    vpshufb %xmm4, %xmm5, %xmm4
+; AVX-NEXT:    vpand %xmm3, %xmm4, %xmm3
+; AVX-NEXT:    vpshufb %xmm1, %xmm5, %xmm1
+; AVX-NEXT:    vpaddb %xmm1, %xmm3, %xmm1
+; AVX-NEXT:    vpcmpeqb %xmm2, %xmm0, %xmm3
+; AVX-NEXT:    vpsrlw $8, %xmm3, %xmm3
+; AVX-NEXT:    vpand %xmm3, %xmm1, %xmm3
 ; AVX-NEXT:    vpsrlw $8, %xmm1, %xmm1
-; AVX-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
-; AVX-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm0
+; AVX-NEXT:    vpaddw %xmm3, %xmm1, %xmm1
+; AVX-NEXT:    vpcmpeqw %xmm2, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm0
 ; AVX-NEXT:    vpand %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    vpsrld $5, %xmm0, %xmm0
 ; AVX-NEXT:    retq
   %1 = and <4 x i32> %x, <i32 4, i32 32, i32 64, i32 128>
   %2 = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %1, i1 0)

llvm/trunk/test/CodeGen/X86/gather-addresses.ll

 ; LIN-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; LIN-SSE2-NEXT:    movd %xmm1, %esi
 ; LIN-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
 ; LIN-SSE2-NEXT:    movd %xmm0, %edi
 ; LIN-SSE2-NEXT:    andl %ecx, %eax
 ; LIN-SSE2-NEXT:    andl %ecx, %edx
 ; LIN-SSE2-NEXT:    andl %ecx, %esi
 ; LIN-SSE2-NEXT:    andl %ecx, %edi
-; LIN-SSE2-NEXT:    movd %eax, %xmm0
-; LIN-SSE2-NEXT:    movd %edx, %xmm1
+; LIN-SSE2-NEXT:    movq %rax, %xmm0
+; LIN-SSE2-NEXT:    movq %rdx, %xmm1
 ; LIN-SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; LIN-SSE2-NEXT:    movd %edi, %xmm2
-; LIN-SSE2-NEXT:    movd %esi, %xmm1
+; LIN-SSE2-NEXT:    movq %rdi, %xmm2
+; LIN-SSE2-NEXT:    movq %rsi, %xmm1
 ; LIN-SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
 ; LIN-SSE2-NEXT:    retq
 ;
 ; LIN-SSE4-LABEL: old:
 ; LIN-SSE4:       # %bb.0:
 ; LIN-SSE4-NEXT:    movdqa (%rsi), %xmm0
 ; LIN-SSE4-NEXT:    pand (%rdx), %xmm0
 ; LIN-SSE4-NEXT:    movd %xmm0, %eax
 ; LIN-SSE4-NEXT:    pextrd $1, %xmm0, %edx
 ; LIN-SSE4-NEXT:    pextrd $2, %xmm0, %esi
 ; LIN-SSE4-NEXT:    pextrd $3, %xmm0, %edi
 ; LIN-SSE4-NEXT:    andl %ecx, %eax
 ; LIN-SSE4-NEXT:    andl %ecx, %edx
 ; LIN-SSE4-NEXT:    andl %ecx, %esi
 ; LIN-SSE4-NEXT:    andl %ecx, %edi
-; LIN-SSE4-NEXT:    movd %edx, %xmm1
-; LIN-SSE4-NEXT:    movd %eax, %xmm0
+; LIN-SSE4-NEXT:    movq %rdx, %xmm1
+; LIN-SSE4-NEXT:    movq %rax, %xmm0
 ; LIN-SSE4-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; LIN-SSE4-NEXT:    movd %edi, %xmm2
-; LIN-SSE4-NEXT:    movd %esi, %xmm1
+; LIN-SSE4-NEXT:    movq %rdi, %xmm2
+; LIN-SSE4-NEXT:    movq %rsi, %xmm1
 ; LIN-SSE4-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
 ; LIN-SSE4-NEXT:    retq
 ;
 ; WIN-SSE2-LABEL: old:
 ; WIN-SSE2:       # %bb.0:
 ; WIN-SSE2-NEXT:    movdqa (%rdx), %xmm0
 ; WIN-SSE2-NEXT:    pand (%r8), %xmm0
 ; WIN-SSE2-NEXT:    movd %xmm0, %eax
 ; WIN-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; WIN-SSE2-NEXT:    movd %xmm1, %ecx
 ; WIN-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; WIN-SSE2-NEXT:    movd %xmm1, %r8d
 ; WIN-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
 ; WIN-SSE2-NEXT:    movd %xmm0, %edx
 ; WIN-SSE2-NEXT:    andl %r9d, %eax
 ; WIN-SSE2-NEXT:    andl %r9d, %ecx
 ; WIN-SSE2-NEXT:    andl %r9d, %r8d
 ; WIN-SSE2-NEXT:    andl %r9d, %edx
-; WIN-SSE2-NEXT:    movd %eax, %xmm0
-; WIN-SSE2-NEXT:    movd %ecx, %xmm1
+; WIN-SSE2-NEXT:    movq %rax, %xmm0
+; WIN-SSE2-NEXT:    movq %rcx, %xmm1
 ; WIN-SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; WIN-SSE2-NEXT:    movd %edx, %xmm2
-; WIN-SSE2-NEXT:    movd %r8d, %xmm1
+; WIN-SSE2-NEXT:    movq %rdx, %xmm2
+; WIN-SSE2-NEXT:    movq %r8, %xmm1
 ; WIN-SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
 ; WIN-SSE2-NEXT:    retq
 ;
 ; WIN-SSE4-LABEL: old:
 ; WIN-SSE4:       # %bb.0:
 ; WIN-SSE4-NEXT:    movdqa (%rdx), %xmm0
 ; WIN-SSE4-NEXT:    pand (%r8), %xmm0
 ; WIN-SSE4-NEXT:    movd %xmm0, %eax
 ; WIN-SSE4-NEXT:    pextrd $1, %xmm0, %ecx
 ; WIN-SSE4-NEXT:    pextrd $2, %xmm0, %r8d
 ; WIN-SSE4-NEXT:    pextrd $3, %xmm0, %edx
 ; WIN-SSE4-NEXT:    andl %r9d, %eax
 ; WIN-SSE4-NEXT:    andl %r9d, %ecx
 ; WIN-SSE4-NEXT:    andl %r9d, %r8d
 ; WIN-SSE4-NEXT:    andl %r9d, %edx
-; WIN-SSE4-NEXT:    movd %ecx, %xmm1
-; WIN-SSE4-NEXT:    movd %eax, %xmm0
+; WIN-SSE4-NEXT:    movq %rcx, %xmm1
+; WIN-SSE4-NEXT:    movq %rax, %xmm0
 ; WIN-SSE4-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; WIN-SSE4-NEXT:    movd %edx, %xmm2
-; WIN-SSE4-NEXT:    movd %r8d, %xmm1
+; WIN-SSE4-NEXT:    movq %rdx, %xmm2
+; WIN-SSE4-NEXT:    movq %r8, %xmm1
 ; WIN-SSE4-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
 ; WIN-SSE4-NEXT:    retq
 ;
 ; LIN32-LABEL: old:
 ; LIN32:       # %bb.0:
 ; LIN32-NEXT:    pushl %edi
 ; LIN32-NEXT:    pushl %esi
 ; LIN32-NEXT:    movl {{[0-9]+}}(%esp), %eax

llvm/trunk/test/CodeGen/X86/horizontal-reduce-umax.ll

 define i16 @test_reduce_v8i16(<8 x i16> %a0) {
 ; X86-SSE2-LABEL: test_reduce_v8i16:
 ; X86-SSE2:       ## %bb.0:
 ; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X86-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
-; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
-; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    psrld $16, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v8i16:

 ;
 ; X64-SSE2-LABEL: test_reduce_v8i16:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
-; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
-; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v8i16:

 
 define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; X86-SSE2-LABEL: test_reduce_v16i16:
 ; X86-SSE2:       ## %bb.0:
 ; X86-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
-; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
-; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
-; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
-; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    psrld $16, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v16i16:

 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v16i16:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
-; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
-; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
-; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
-; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v16i16:
; X64-AVX512-NEXT:    retq
   %13 = extractelement <16 x i32> %12, i32 0
   ret i32 %13
 }
 
 define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; X86-SSE2-LABEL: test_reduce_v32i16:
 ; X86-SSE2:       ## %bb.0:
 ; X86-SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm2
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
-; X86-SSE2-NEXT:    pmaxsw %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm3
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X86-SSE2-NEXT:    pmaxsw %xmm3, %xmm1
-; X86-SSE2-NEXT:    movdqa %xmm4, %xmm2
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm2
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm0, %xmm2
-; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm2
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm2
-; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm2
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
 ; X86-SSE2-NEXT:    pmaxsw %xmm2, %xmm0
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
-; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
+; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
-; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
+; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
-; X86-SSE2-NEXT:    psrld $16, %xmm0
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
-; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
-; X86-SSE2-NEXT:    movd %xmm0, %eax
+; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
+; X86-SSE2-NEXT:    psrld $16, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
+; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
+; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v32i16:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pmaxuw %xmm3, %xmm1
 ; X86-SSE42-NEXT:    pmaxuw %xmm2, %xmm0
 ; X86-SSE42-NEXT:    pmaxuw %xmm1, %xmm0

 ; X86-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v32i16:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm2
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
-; X64-SSE2-NEXT:    pmaxsw %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm3
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm3, %xmm1
-; X64-SSE2-NEXT:    movdqa %xmm4, %xmm2
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm2
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm0, %xmm2
-; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm2
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm2
-; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm2
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
 ; X64-SSE2-NEXT:    pmaxsw %xmm2, %xmm0
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
-; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
+; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
-; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
+; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
-; X64-SSE2-NEXT:    psrld $16, %xmm0
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
-; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
-; X64-SSE2-NEXT:    movd %xmm0, %eax
+; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
+; X64-SSE2-NEXT:    psrld $16, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
+; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
+; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v32i16:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pmaxuw %xmm3, %xmm1
 ; X64-SSE42-NEXT:    pmaxuw %xmm2, %xmm0
 ; X64-SSE42-NEXT:    pmaxuw %xmm1, %xmm0

llvm/trunk/test/CodeGen/X86/horizontal-reduce-umin.ll

 define i16 @test_reduce_v8i16(<8 x i16> %a0) {
 ; X86-SSE2-LABEL: test_reduce_v8i16:
 ; X86-SSE2:       ## %bb.0:
 ; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X86-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
-; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
-; X86-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X86-SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    psrld $16, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v8i16:

 ;
 ; X64-SSE2-LABEL: test_reduce_v8i16:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
-; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
-; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v8i16:

 
 define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; X86-SSE2-LABEL: test_reduce_v16i16:
 ; X86-SSE2:       ## %bb.0:
 ; X86-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pminsw %xmm1, %xmm0
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
-; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
-; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
-; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
-; X86-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X86-SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    psrld $16, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v16i16:

 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v16i16:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
-; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
-; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
-; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
+; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
-; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
+; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v16i16:
; X64-AVX512-NEXT:    retq
   %13 = extractelement <16 x i32> %12, i32 0
   ret i32 %13
 }
 
 define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; X86-SSE2-LABEL: test_reduce_v32i16:
 ; X86-SSE2:       ## %bb.0:
 ; X86-SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm2
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
-; X86-SSE2-NEXT:    pminsw %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm3
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X86-SSE2-NEXT:    pminsw %xmm3, %xmm1
-; X86-SSE2-NEXT:    movdqa %xmm4, %xmm2
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm2
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm0, %xmm2
-; X86-SSE2-NEXT:    pminsw %xmm1, %xmm2
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm2
-; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm2
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
 ; X86-SSE2-NEXT:    pminsw %xmm2, %xmm0
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
-; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
+; X86-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
-; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
+; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
-; X86-SSE2-NEXT:    psrld $16, %xmm0
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
-; X86-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
-; X86-SSE2-NEXT:    movd %xmm0, %eax
+; X86-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
+; X86-SSE2-NEXT:    psrld $16, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
+; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
+; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v32i16:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pminuw %xmm3, %xmm1
 ; X86-SSE42-NEXT:    pminuw %xmm2, %xmm0
 ; X86-SSE42-NEXT:    pminuw %xmm1, %xmm0

 ; X86-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v32i16:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm2
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
-; X64-SSE2-NEXT:    pminsw %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm3
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X64-SSE2-NEXT:    pminsw %xmm3, %xmm1
-; X64-SSE2-NEXT:    movdqa %xmm4, %xmm2
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm2
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm0, %xmm2
-; X64-SSE2-NEXT:    pminsw %xmm1, %xmm2
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm2
-; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm2
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
 ; X64-SSE2-NEXT:    pminsw %xmm2, %xmm0
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
-; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
+; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
-; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
+; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
-; X64-SSE2-NEXT:    psrld $16, %xmm0
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
-; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
-; X64-SSE2-NEXT:    movd %xmm0, %eax
+; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
+; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
+; X64-SSE2-NEXT:    psrld $16, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
+; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
+; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
+; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v32i16:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pminuw %xmm3, %xmm1
 ; X64-SSE42-NEXT:    pminuw %xmm2, %xmm0
 ; X64-SSE42-NEXT:    pminuw %xmm1, %xmm0

llvm/trunk/test/CodeGen/X86/known-bits.ll

 ; X32-NEXT:    movzbl (%eax), %ecx
 ; X32-NEXT:    imull $101, %ecx, %eax
 ; X32-NEXT:    shrl $14, %eax
 ; X32-NEXT:    imull $177, %ecx, %ecx
 ; X32-NEXT:    shrl $14, %ecx
 ; X32-NEXT:    movzbl %al, %eax
 ; X32-NEXT:    vpxor %xmm0, %xmm0, %xmm0
 ; X32-NEXT:    vpinsrd $1, %eax, %xmm0, %xmm1
-; X32-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; X32-NEXT:    vbroadcastss {{.*#+}} xmm2 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
 ; X32-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; X32-NEXT:    movzbl %cl, %eax
 ; X32-NEXT:    vpinsrd $1, %eax, %xmm0, %xmm0
 ; X32-NEXT:    vpand %xmm2, %xmm0, %xmm0
 ; X32-NEXT:    vpextrd $1, %xmm1, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    vpextrd $1, %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    xorl %ecx, %ecx
 ; X32-NEXT:    vmovd %xmm1, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill

 ; X64-NEXT:    movzbl (%rdi), %eax
 ; X64-NEXT:    imull $101, %eax, %ecx
 ; X64-NEXT:    shrl $14, %ecx
 ; X64-NEXT:    imull $177, %eax, %eax
 ; X64-NEXT:    shrl $14, %eax
 ; X64-NEXT:    movzbl %cl, %ecx
 ; X64-NEXT:    vpxor %xmm0, %xmm0, %xmm0
 ; X64-NEXT:    vpinsrd $1, %ecx, %xmm0, %xmm1
-; X64-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; X64-NEXT:    vbroadcastss {{.*#+}} xmm2 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
 ; X64-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; X64-NEXT:    movzbl %al, %eax
 ; X64-NEXT:    vpinsrd $1, %eax, %xmm0, %xmm0
 ; X64-NEXT:    vpand %xmm2, %xmm0, %xmm0
 ; X64-NEXT:    vpextrd $1, %xmm1, %r8d
 ; X64-NEXT:    vpextrd $1, %xmm0, %r9d
 ; X64-NEXT:    xorl %esi, %esi
 ; X64-NEXT:    vmovd %xmm1, %r10d

llvm/trunk/test/CodeGen/X86/nontemporal-loads.ll

 ; AVX512VL-NEXT:    retq
   %1 = load <64 x i8>, <64 x i8>* %src, align 1, !nontemporal !1
   ret <64 x i8> %1
 }
 
 define <16 x i32> @test_masked_v16i32(i8 * %addr, <16 x i32> %old, <16 x i32> %mask1) {
 ; SSE2-LABEL: test_masked_v16i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm10
-; SSE2-NEXT:    pxor %xmm12, %xmm12
-; SSE2-NEXT:    pcmpeqd %xmm12, %xmm7
-; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE2-NEXT:    movdqa %xmm7, %xmm8
-; SSE2-NEXT:    pxor %xmm0, %xmm8
-; SSE2-NEXT:    pcmpeqd %xmm12, %xmm6
-; SSE2-NEXT:    movdqa %xmm6, %xmm9
-; SSE2-NEXT:    pxor %xmm0, %xmm9
-; SSE2-NEXT:    pcmpeqd %xmm12, %xmm5
-; SSE2-NEXT:    movdqa %xmm5, %xmm11
-; SSE2-NEXT:    pxor %xmm0, %xmm11
-; SSE2-NEXT:    pcmpeqd %xmm12, %xmm4
-; SSE2-NEXT:    pxor %xmm4, %xmm0
+; SSE2-NEXT:    pxor %xmm8, %xmm8
+; SSE2-NEXT:    pcmpeqd %xmm8, %xmm7
+; SSE2-NEXT:    pcmpeqd %xmm8, %xmm6
+; SSE2-NEXT:    pcmpeqd %xmm8, %xmm5
+; SSE2-NEXT:    pcmpeqd %xmm8, %xmm4
+; SSE2-NEXT:    pand %xmm4, %xmm0
 ; SSE2-NEXT:    pandn (%rdi), %xmm4
-; SSE2-NEXT:    pandn %xmm10, %xmm0
 ; SSE2-NEXT:    por %xmm4, %xmm0
+; SSE2-NEXT:    pand %xmm5, %xmm1
 ; SSE2-NEXT:    pandn 16(%rdi), %xmm5
-; SSE2-NEXT:    pandn %xmm1, %xmm11
-; SSE2-NEXT:    por %xmm5, %xmm11
+; SSE2-NEXT:    por %xmm5, %xmm1
+; SSE2-NEXT:    pand %xmm6, %xmm2
 ; SSE2-NEXT:    pandn 32(%rdi), %xmm6
-; SSE2-NEXT:    pandn %xmm2, %xmm9
-; SSE2-NEXT:    por %xmm6, %xmm9
+; SSE2-NEXT:    por %xmm6, %xmm2
+; SSE2-NEXT:    pand %xmm7, %xmm3
 ; SSE2-NEXT:    pandn 48(%rdi), %xmm7
-; SSE2-NEXT:    pandn %xmm3, %xmm8
-; SSE2-NEXT:    por %xmm7, %xmm8
-; SSE2-NEXT:    movdqa %xmm11, %xmm1
-; SSE2-NEXT:    movdqa %xmm9, %xmm2
-; SSE2-NEXT:    movdqa %xmm8, %xmm3
+; SSE2-NEXT:    por %xmm7, %xmm3
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_masked_v16i32:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm8
 ; SSE41-NEXT:    pxor %xmm0, %xmm0
 ; SSE41-NEXT:    pcmpeqd %xmm0, %xmm7
 ; SSE41-NEXT:    pcmpeqd %xmm9, %xmm9

llvm/trunk/test/CodeGen/X86/paddus.ll

; AVX-NEXT:    retq
   %3 = select <8 x i1> %2, <8 x i16> <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>, <8 x i16> %1
   ret <8 x i16> %3
 }
 
 define <8 x i16> @test23(<8 x i16> %x) {
 ; SSE2-LABEL: test23:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pxor %xmm0, %xmm2
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    pcmpgtw %xmm2, %xmm1
-; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm2, %xmm1
+; SSE2-NEXT:    pcmpgtw %xmm0, %xmm1
+; SSE2-NEXT:    por %xmm2, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: test23:
 ; SSSE3:       # %bb.0:
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSSE3-NEXT:    pxor %xmm2, %xmm0
 ; SSSE3-NEXT:    pxor %xmm0, %xmm2
-; SSSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSSE3-NEXT:    pcmpgtw %xmm2, %xmm1
-; SSSE3-NEXT:    por %xmm0, %xmm1
+; SSSE3-NEXT:    movdqa %xmm2, %xmm1
+; SSSE3-NEXT:    pcmpgtw %xmm0, %xmm1
+; SSSE3-NEXT:    por %xmm2, %xmm1
 ; SSSE3-NEXT:    movdqa %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: test23:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; SSE41-NEXT:    pxor %xmm0, %xmm1
 ; SSE41-NEXT:    pmaxuw %xmm1, %xmm0
; AVX512-NEXT:    retq
   %2 = icmp ugt <16 x i16> %x, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
   %3 = select <16 x i1> %2, <16 x i16> <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>, <16 x i16> %1
   ret <16 x i16> %3
 }
 
 define <16 x i16> @test29(<16 x i16> %x) {
 ; SSE2-LABEL: test29:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE2-NEXT:    pxor %xmm4, %xmm0
-; SSE2-NEXT:    pxor %xmm4, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
-; SSE2-NEXT:    pxor %xmm4, %xmm3
-; SSE2-NEXT:    movdqa %xmm1, %xmm2
-; SSE2-NEXT:    pcmpgtw %xmm3, %xmm2
-; SSE2-NEXT:    pxor %xmm0, %xmm4
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    pcmpgtw %xmm4, %xmm3
-; SSE2-NEXT:    por %xmm0, %xmm3
-; SSE2-NEXT:    por %xmm1, %xmm2
-; SSE2-NEXT:    movdqa %xmm3, %xmm0
+; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; SSE2-NEXT:    movdqa %xmm0, %xmm4
+; SSE2-NEXT:    pxor %xmm3, %xmm4
+; SSE2-NEXT:    pxor %xmm1, %xmm3
+; SSE2-NEXT:    movdqa %xmm3, %xmm2
+; SSE2-NEXT:    pcmpgtw %xmm1, %xmm2
+; SSE2-NEXT:    movdqa %xmm4, %xmm1
+; SSE2-NEXT:    pcmpgtw %xmm0, %xmm1
+; SSE2-NEXT:    por %xmm4, %xmm1
+; SSE2-NEXT:    por %xmm3, %xmm2
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm2, %xmm1
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: test29:
 ; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm4 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSSE3-NEXT:    pxor %xmm4, %xmm0
-; SSSE3-NEXT:    pxor %xmm4, %xmm1
-; SSSE3-NEXT:    movdqa %xmm1, %xmm3
-; SSSE3-NEXT:    pxor %xmm4, %xmm3
-; SSSE3-NEXT:    movdqa %xmm1, %xmm2
-; SSSE3-NEXT:    pcmpgtw %xmm3, %xmm2
-; SSSE3-NEXT:    pxor %xmm0, %xmm4
-; SSSE3-NEXT:    movdqa %xmm0, %xmm3
-; SSSE3-NEXT:    pcmpgtw %xmm4, %xmm3
-; SSSE3-NEXT:    por %xmm0, %xmm3
-; SSSE3-NEXT:    por %xmm1, %xmm2
-; SSSE3-NEXT:    movdqa %xmm3, %xmm0
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm3 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; SSSE3-NEXT:    movdqa %xmm0, %xmm4
+; SSSE3-NEXT:    pxor %xmm3, %xmm4
+; SSSE3-NEXT:    pxor %xmm1, %xmm3
+; SSSE3-NEXT:    movdqa %xmm3, %xmm2
+; SSSE3-NEXT:    pcmpgtw %xmm1, %xmm2
+; SSSE3-NEXT:    movdqa %xmm4, %xmm1
+; SSSE3-NEXT:    pcmpgtw %xmm0, %xmm1
+; SSSE3-NEXT:    por %xmm4, %xmm1
+; SSSE3-NEXT:    por %xmm3, %xmm2
+; SSSE3-NEXT:    movdqa %xmm1, %xmm0
 ; SSSE3-NEXT:    movdqa %xmm2, %xmm1
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: test29:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
 ; SSE41-NEXT:    pxor %xmm2, %xmm3
; AVX512-NEXT:    retq
   %2 = icmp ugt <32 x i16> %x, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
   %3 = select <32 x i1> %2, <32 x i16> <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>, <32 x i16> %1
   ret <32 x i16> %3
 }
 
 define <32 x i16> @test35(<32 x i16> %x) {
 ; SSE2-LABEL: test35:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE2-NEXT:    pxor %xmm4, %xmm0
-; SSE2-NEXT:    pxor %xmm4, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm2
-; SSE2-NEXT:    pxor %xmm4, %xmm3
-; SSE2-NEXT:    movdqa %xmm3, %xmm5
-; SSE2-NEXT:    pxor %xmm4, %xmm5
-; SSE2-NEXT:    movdqa %xmm3, %xmm8
-; SSE2-NEXT:    pcmpgtw %xmm5, %xmm8
-; SSE2-NEXT:    movdqa %xmm2, %xmm6
-; SSE2-NEXT:    pxor %xmm4, %xmm6
-; SSE2-NEXT:    movdqa %xmm2, %xmm5
-; SSE2-NEXT:    pcmpgtw %xmm6, %xmm5
+; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; SSE2-NEXT:    movdqa %xmm0, %xmm6
+; SSE2-NEXT:    pxor %xmm5, %xmm6
 ; SSE2-NEXT:    movdqa %xmm1, %xmm7
-; SSE2-NEXT:    pxor %xmm4, %xmm7
-; SSE2-NEXT:    movdqa %xmm1, %xmm6
-; SSE2-NEXT:    pcmpgtw %xmm7, %xmm6
-; SSE2-NEXT:    pxor %xmm0, %xmm4
-; SSE2-NEXT:    movdqa %xmm0, %xmm7
-; SSE2-NEXT:    pcmpgtw %xmm4, %xmm7
-; SSE2-NEXT:    por %xmm0, %xmm7
-; SSE2-NEXT:    por %xmm1, %xmm6
-; SSE2-NEXT:    por %xmm2, %xmm5
-; SSE2-NEXT:    por %xmm3, %xmm8
-; SSE2-NEXT:    movdqa %xmm7, %xmm0
-; SSE2-NEXT:    movdqa %xmm6, %xmm1
-; SSE2-NEXT:    movdqa %xmm5, %xmm2
+; SSE2-NEXT:    pxor %xmm5, %xmm7
+; SSE2-NEXT:    movdqa %xmm2, %xmm8
+; SSE2-NEXT:    pxor %xmm5, %xmm8
+; SSE2-NEXT:    pxor %xmm3, %xmm5
+; SSE2-NEXT:    movdqa %xmm5, %xmm4
+; SSE2-NEXT:    pcmpgtw %xmm3, %xmm4
 ; SSE2-NEXT:    movdqa %xmm8, %xmm3
+; SSE2-NEXT:    pcmpgtw %xmm2, %xmm3
+; SSE2-NEXT:    movdqa %xmm7, %xmm2
+; SSE2-NEXT:    pcmpgtw %xmm1, %xmm2
+; SSE2-NEXT:    movdqa %xmm6, %xmm1
+; SSE2-NEXT:    pcmpgtw %xmm0, %xmm1
+; SSE2-NEXT:    por %xmm6, %xmm1
+; SSE2-NEXT:    por %xmm7, %xmm2
+; SSE2-NEXT:    por %xmm8, %xmm3
+; SSE2-NEXT:    por %xmm5, %xmm4
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm2, %xmm1
+; SSE2-NEXT:    movdqa %xmm3, %xmm2
+; SSE2-NEXT:    movdqa %xmm4, %xmm3
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: test35:
 ; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm4 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSSE3-NEXT:    pxor %xmm4, %xmm0
-; SSSE3-NEXT:    pxor %xmm4, %xmm1
-; SSSE3-NEXT:    pxor %xmm4, %xmm2
-; SSSE3-NEXT:    pxor %xmm4, %xmm3
-; SSSE3-NEXT:    movdqa %xmm3, %xmm5
-; SSSE3-NEXT:    pxor %xmm4, %xmm5
-; SSSE3-NEXT:    movdqa %xmm3, %xmm8
-; SSSE3-NEXT:    pcmpgtw %xmm5, %xmm8
-; SSSE3-NEXT:    movdqa %xmm2, %xmm6
-; SSSE3-NEXT:    pxor %xmm4, %xmm6
-; SSSE3-NEXT:    movdqa %xmm2, %xmm5
-; SSSE3-NEXT:    pcmpgtw %xmm6, %xmm5
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm5 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; SSSE3-NEXT:    movdqa %xmm0, %xmm6
+; SSSE3-NEXT:    pxor %xmm5, %xmm6
 ; SSSE3-NEXT:    movdqa %xmm1, %xmm7
-; SSSE3-NEXT:    pxor %xmm4, %xmm7
-; SSSE3-NEXT:    movdqa %xmm1, %xmm6
-; SSSE3-NEXT:    pcmpgtw %xmm7, %xmm6
-; SSSE3-NEXT:    pxor %xmm0, %xmm4
-; SSSE3-NEXT:    movdqa %xmm0, %xmm7
-; SSSE3-NEXT:    pcmpgtw %xmm4, %xmm7
-; SSSE3-NEXT:    por %xmm0, %xmm7
-; SSSE3-NEXT:    por %xmm1, %xmm6
-; SSSE3-NEXT:    por %xmm2, %xmm5
-; SSSE3-NEXT:    por %xmm3, %xmm8
-; SSSE3-NEXT:    movdqa %xmm7, %xmm0
-; SSSE3-NEXT:    movdqa %xmm6, %xmm1
-; SSSE3-NEXT:    movdqa %xmm5, %xmm2
+; SSSE3-NEXT:    pxor %xmm5, %xmm7
+; SSSE3-NEXT:    movdqa %xmm2, %xmm8
+; SSSE3-NEXT:    pxor %xmm5, %xmm8
+; SSSE3-NEXT:    pxor %xmm3, %xmm5
+; SSSE3-NEXT:    movdqa %xmm5, %xmm4
+; SSSE3-NEXT:    pcmpgtw %xmm3, %xmm4
 ; SSSE3-NEXT:    movdqa %xmm8, %xmm3
+; SSSE3-NEXT:    pcmpgtw %xmm2, %xmm3
+; SSSE3-NEXT:    movdqa %xmm7, %xmm2
+; SSSE3-NEXT:    pcmpgtw %xmm1, %xmm2
+; SSSE3-NEXT:    movdqa %xmm6, %xmm1
+; SSSE3-NEXT:    pcmpgtw %xmm0, %xmm1
+; SSSE3-NEXT:    por %xmm6, %xmm1
+; SSSE3-NEXT:    por %xmm7, %xmm2
+; SSSE3-NEXT:    por %xmm8, %xmm3
+; SSSE3-NEXT:    por %xmm5, %xmm4
+; SSSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSSE3-NEXT:    movdqa %xmm2, %xmm1
+; SSSE3-NEXT:    movdqa %xmm3, %xmm2
+; SSSE3-NEXT:    movdqa %xmm4, %xmm3
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: test35:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; SSE41-NEXT:    movdqa %xmm0, %xmm5
 ; SSE41-NEXT:    pxor %xmm4, %xmm5
 ; SSE41-NEXT:    movdqa %xmm1, %xmm6

llvm/trunk/test/CodeGen/X86/psubus.ll

 ; AVX1-NEXT:    vpcmpeqd %xmm5, %xmm4, %xmm5
 ; AVX1-NEXT:    vpxor %xmm3, %xmm5, %xmm3
 ; AVX1-NEXT:    vpackssdw %xmm3, %xmm7, %xmm3
 ; AVX1-NEXT:    vpacksswb %xmm10, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsubd %xmm8, %xmm4, %xmm4
 ; AVX1-NEXT:    vpsubd %xmm9, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsubd %xmm11, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsubd %xmm0, %xmm6, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm5 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
 ; AVX1-NEXT:    vpand %xmm5, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpand %xmm5, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackuswb %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    vpandn %xmm0, %xmm3, %xmm0

llvm/trunk/test/CodeGen/X86/sat-add.ll

; ANY-NEXT:    retq
   %c = icmp ugt <16 x i8> %x, %noty
   %r = select <16 x i1> %c, <16 x i8> <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>, <16 x i8> %a
   ret <16 x i8> %r
 }
 
 define <8 x i16> @unsigned_sat_variable_v8i16_using_min(<8 x i16> %x, <8 x i16> %y) {
 ; SSE2-LABEL: unsigned_sat_variable_v8i16_using_min:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pcmpeqd %xmm2, %xmm2
-; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE2-NEXT:    pxor %xmm3, %xmm0
-; SSE2-NEXT:    pxor %xmm3, %xmm2
-; SSE2-NEXT:    pxor %xmm1, %xmm2
-; SSE2-NEXT:    pminsw %xmm2, %xmm0
-; SSE2-NEXT:    pxor %xmm3, %xmm0
+; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; SSE2-NEXT:    pxor %xmm2, %xmm0
+; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [32767,32767,32767,32767,32767,32767,32767,32767]
+; SSE2-NEXT:    pxor %xmm1, %xmm3
+; SSE2-NEXT:    pminsw %xmm3, %xmm0
+; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    paddw %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: unsigned_sat_variable_v8i16_using_min:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pcmpeqd %xmm2, %xmm2
 ; SSE41-NEXT:    pxor %xmm1, %xmm2
 ; SSE41-NEXT:    pminuw %xmm2, %xmm0
; ANY-NEXT:    retq
   %c = icmp ugt <8 x i16> %x, %a
   %r = select <8 x i1> %c, <8 x i16> <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>, <8 x i16> %a
   ret <8 x i16> %r
 }
 
 define <8 x i16> @unsigned_sat_variable_v8i16_using_cmp_notval(<8 x i16> %x, <8 x i16> %y) {
 ; SSE2-LABEL: unsigned_sat_variable_v8i16_using_cmp_notval:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pcmpeqd %xmm2, %xmm2
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    paddw %xmm1, %xmm3
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE2-NEXT:    pxor %xmm4, %xmm0
-; SSE2-NEXT:    pxor %xmm4, %xmm2
-; SSE2-NEXT:    pxor %xmm1, %xmm2
-; SSE2-NEXT:    pcmpgtw %xmm2, %xmm0
-; SSE2-NEXT:    por %xmm3, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    paddw %xmm1, %xmm2
+; SSE2-NEXT:    pxor {{.*}}(%rip), %xmm1
+; SSE2-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pcmpgtw %xmm1, %xmm0
+; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: unsigned_sat_variable_v8i16_using_cmp_notval:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pcmpeqd %xmm2, %xmm2
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
 ; SSE41-NEXT:    paddw %xmm1, %xmm3
 ; SSE41-NEXT:    pxor %xmm2, %xmm1
 ; SSE41-NEXT:    pminuw %xmm0, %xmm1
 ; SSE41-NEXT:    pcmpeqw %xmm1, %xmm0
 ; SSE41-NEXT:    pxor %xmm2, %xmm0
 ; SSE41-NEXT:    por %xmm3, %xmm0
 ; SSE41-NEXT:    retq
   %noty = xor <8 x i16> %y, <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>
   %a = add <8 x i16> %x, %y
   %c = icmp ugt <8 x i16> %x, %noty
   %r = select <8 x i1> %c, <8 x i16> <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>, <8 x i16> %a
   ret <8 x i16> %r
 }
 
 define <4 x i32> @unsigned_sat_variable_v4i32_using_min(<4 x i32> %x, <4 x i32> %y) {
 ; SSE2-LABEL: unsigned_sat_variable_v4i32_using_min:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pcmpeqd %xmm2, %xmm2
+; SSE2-NEXT:    pxor %xmm1, %xmm2
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [2147483648,2147483648,2147483648,2147483648]
-; SSE2-NEXT:    movdqa %xmm0, %xmm4
-; SSE2-NEXT:    pxor %xmm3, %xmm4
-; SSE2-NEXT:    pxor %xmm2, %xmm3
-; SSE2-NEXT:    pxor %xmm1, %xmm3
-; SSE2-NEXT:    pcmpgtd %xmm4, %xmm3
-; SSE2-NEXT:    pand %xmm3, %xmm0
-; SSE2-NEXT:    pxor %xmm2, %xmm3
-; SSE2-NEXT:    movdqa %xmm1, %xmm2
-; SSE2-NEXT:    pandn %xmm3, %xmm2
-; SSE2-NEXT:    por %xmm2, %xmm0
+; SSE2-NEXT:    pxor %xmm0, %xmm3
+; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [2147483647,2147483647,2147483647,2147483647]
+; SSE2-NEXT:    pxor %xmm1, %xmm4
+; SSE2-NEXT:    pcmpgtd %xmm3, %xmm4
+; SSE2-NEXT:    pand %xmm4, %xmm0
+; SSE2-NEXT:    pandn %xmm2, %xmm4
+; SSE2-NEXT:    por %xmm4, %xmm0
 ; SSE2-NEXT:    paddd %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: unsigned_sat_variable_v4i32_using_min:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pcmpeqd %xmm2, %xmm2
 ; SSE41-NEXT:    pxor %xmm1, %xmm2
 ; SSE41-NEXT:    pminud %xmm2, %xmm0
; SSE41-NEXT:    retq
   %c = icmp ugt <4 x i32> %x, %a
   %r = select <4 x i1> %c, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>, <4 x i32> %a
   ret <4 x i32> %r
 }
 
 define <4 x i32> @unsigned_sat_variable_v4i32_using_cmp_notval(<4 x i32> %x, <4 x i32> %y) {
 ; SSE2-LABEL: unsigned_sat_variable_v4i32_using_cmp_notval:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pcmpeqd %xmm2, %xmm2
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    paddd %xmm1, %xmm3
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [2147483648,2147483648,2147483648,2147483648]
-; SSE2-NEXT:    pxor %xmm4, %xmm0
-; SSE2-NEXT:    pxor %xmm4, %xmm2
-; SSE2-NEXT:    pxor %xmm1, %xmm2
-; SSE2-NEXT:    pcmpgtd %xmm2, %xmm0
-; SSE2-NEXT:    por %xmm3, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    paddd %xmm1, %xmm2
+; SSE2-NEXT:    pxor {{.*}}(%rip), %xmm1
+; SSE2-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pcmpgtd %xmm1, %xmm0
+; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: unsigned_sat_variable_v4i32_using_cmp_notval:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pcmpeqd %xmm2, %xmm2
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
 ; SSE41-NEXT:    paddd %xmm1, %xmm3
 ; SSE41-NEXT:    pxor %xmm2, %xmm1

llvm/trunk/test/CodeGen/X86/setcc-lowering.ll

entry:
   %3 = sext <8 x i1> %2 to <8 x i16>
   ret <8 x i16> %3
 }
 
 define void @pr26232(i64 %a, <16 x i1> %b) {
 ; AVX-LABEL: pr26232:
 ; AVX:       # %bb.0: # %allocas
 ; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; AVX-NEXT:    .p2align 4, 0x90
 ; AVX-NEXT:  .LBB1_1: # %for_loop599
 ; AVX-NEXT:    # =>This Inner Loop Header: Depth=1
 ; AVX-NEXT:    xorl %eax, %eax
 ; AVX-NEXT:    cmpq $65536, %rdi # imm = 0x10000
 ; AVX-NEXT:    setl %al
-; AVX-NEXT:    vmovd %eax, %xmm3
-; AVX-NEXT:    vpshufb %xmm1, %xmm3, %xmm3
-; AVX-NEXT:    vpand %xmm0, %xmm3, %xmm3
-; AVX-NEXT:    vpsllw $7, %xmm3, %xmm3
-; AVX-NEXT:    vpand %xmm2, %xmm3, %xmm3
-; AVX-NEXT:    vpmovmskb %xmm3, %eax
+; AVX-NEXT:    vmovd %eax, %xmm2
+; AVX-NEXT:    vpshufb %xmm1, %xmm2, %xmm2
+; AVX-NEXT:    vpand %xmm0, %xmm2, %xmm2
+; AVX-NEXT:    vpsllw $7, %xmm2, %xmm2
+; AVX-NEXT:    vpmovmskb %xmm2, %eax
 ; AVX-NEXT:    testw %ax, %ax
 ; AVX-NEXT:    jne .LBB1_1
 ; AVX-NEXT:  # %bb.2: # %for_exit600
 ; AVX-NEXT:    retq
 ;
 ; KNL-32-LABEL: pr26232:
 ; KNL-32:       # %bb.0: # %allocas
 ; KNL-32-NEXT:    pushl %esi

llvm/trunk/test/CodeGen/X86/sse2-intrinsics-canonical.ll

 ; SSE-NEXT:    pand %xmm2, %xmm3 ## encoding: [0x66,0x0f,0xdb,0xda]
 ; SSE-NEXT:    pand %xmm2, %xmm0 ## encoding: [0x66,0x0f,0xdb,0xc2]
 ; SSE-NEXT:    pmaxsw %xmm3, %xmm0 ## encoding: [0x66,0x0f,0xee,0xc3]
 ; SSE-NEXT:    psubw %xmm1, %xmm0 ## encoding: [0x66,0x0f,0xf9,0xc1]
 ; SSE-NEXT:    retl ## encoding: [0xc3]
 ;
 ; AVX2-LABEL: test_x86_sse2_psubus_b_64:
 ; AVX2:       ## %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
-; AVX2-NEXT:    ## encoding: [0xc5,0xf9,0x6f,0x15,A,A,A,A]
-; AVX2-NEXT:    ## fixup A - offset: 4, value: LCPI6_0, kind: FK_Data_4
+; AVX2-NEXT:    vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
+; AVX2-NEXT:    ## encoding: [0xc4,0xe2,0x79,0x79,0x15,A,A,A,A]
+; AVX2-NEXT:    ## fixup A - offset: 5, value: LCPI6_0, kind: FK_Data_4
 ; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm3 ## encoding: [0xc5,0xf1,0xdb,0xda]
 ; AVX2-NEXT:    vpand %xmm2, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0xdb,0xc2]
 ; AVX2-NEXT:    vpmaxuw %xmm3, %xmm0, %xmm0 ## encoding: [0xc4,0xe2,0x79,0x3e,0xc3]
 ; AVX2-NEXT:    vpsubw %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0xf9,0xc1]
 ; AVX2-NEXT:    retl ## encoding: [0xc3]
 ;
 ; SKX-LABEL: test_x86_sse2_psubus_b_64:
 ; SKX:       ## %bb.0:
-; SKX-NEXT:    vmovdqa LCPI6_0, %xmm2 ## EVEX TO VEX Compression xmm2 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
-; SKX-NEXT:    ## encoding: [0xc5,0xf9,0x6f,0x15,A,A,A,A]
-; SKX-NEXT:    ## fixup A - offset: 4, value: LCPI6_0, kind: FK_Data_4
+; SKX-NEXT:    vpbroadcastw LCPI6_0, %xmm2 ## EVEX TO VEX Compression xmm2 = [255,255,255,255,255,255,255,255]
+; SKX-NEXT:    ## encoding: [0xc4,0xe2,0x79,0x79,0x15,A,A,A,A]
+; SKX-NEXT:    ## fixup A - offset: 5, value: LCPI6_0, kind: FK_Data_4
 ; SKX-NEXT:    vpand %xmm2, %xmm1, %xmm3 ## EVEX TO VEX Compression encoding: [0xc5,0xf1,0xdb,0xda]
 ; SKX-NEXT:    vpand %xmm2, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xdb,0xc2]
 ; SKX-NEXT:    vpmaxuw %xmm3, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x3e,0xc3]
 ; SKX-NEXT:    vpsubw %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xf9,0xc1]
 ; SKX-NEXT:    retl ## encoding: [0xc3]
   %cmp = icmp ugt <8 x i8> %a0, %a1
   %sel = select <8 x i1> %cmp, <8 x i8> %a0, <8 x i8> %a1
   %sub = sub <8 x i8> %sel, %a1

llvm/trunk/test/CodeGen/X86/unfold-masked-merge-vector-variablemask-const.ll

 ; CHECK-SSE1-NEXT:    xorps %xmm1, %xmm2
 ; CHECK-SSE1-NEXT:    andnps %xmm2, %xmm0
 ; CHECK-SSE1-NEXT:    xorps %xmm1, %xmm0
 ; CHECK-SSE1-NEXT:    movaps %xmm0, (%rdi)
 ; CHECK-SSE1-NEXT:    retq
 ;
 ; CHECK-SSE2-LABEL: in_constant_varx_mone_invmask:
 ; CHECK-SSE2:       # %bb.0:
-; CHECK-SSE2-NEXT:    movdqa (%rdi), %xmm0
+; CHECK-SSE2-NEXT:    movdqa (%rdx), %xmm0
 ; CHECK-SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
-; CHECK-SSE2-NEXT:    movdqa (%rdx), %xmm2
+; CHECK-SSE2-NEXT:    movdqa (%rdi), %xmm2
 ; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm2
 ; CHECK-SSE2-NEXT:    pandn %xmm2, %xmm0
 ; CHECK-SSE2-NEXT:    pxor %xmm1, %xmm0
 ; CHECK-SSE2-NEXT:    retq
 ;
 ; CHECK-XOP-LABEL: in_constant_varx_mone_invmask:
 ; CHECK-XOP:       # %bb.0:
-; CHECK-XOP-NEXT:    vmovdqa (%rdi), %xmm0
+; CHECK-XOP-NEXT:    vmovdqa (%rdx), %xmm0
 ; CHECK-XOP-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
-; CHECK-XOP-NEXT:    vpxor (%rdx), %xmm1, %xmm2
+; CHECK-XOP-NEXT:    vpxor (%rdi), %xmm1, %xmm2
 ; CHECK-XOP-NEXT:    vpandn %xmm2, %xmm0, %xmm0
 ; CHECK-XOP-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; CHECK-XOP-NEXT:    retq
   %x = load <4 x i32>, <4 x i32> *%px, align 16
   %y = load <4 x i32>, <4 x i32> *%py, align 16
   %mask = load <4 x i32>, <4 x i32> *%pmask, align 16
   %notmask = xor <4 x i32> %mask, <i32 -1, i32 -1, i32 -1, i32 -1>
   %n0 = xor <4 x i32> %x, <i32 -1, i32 -1, i32 -1, i32 -1> ; %x

llvm/trunk/test/CodeGen/X86/v8i1-masks.ll

 ; X32-AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-AVX2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-AVX2-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X32-AVX2-NEXT:    vmovups (%edx), %ymm0
 ; X32-AVX2-NEXT:    vmovups (%ecx), %ymm1
 ; X32-AVX2-NEXT:    vcmpltps %ymm0, %ymm1, %ymm1
 ; X32-AVX2-NEXT:    vmovups (%eax), %ymm2
 ; X32-AVX2-NEXT:    vcmpltps %ymm0, %ymm2, %ymm0
-; X32-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1,1,1,1,1,1,1,1]
-; X32-AVX2-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; X32-AVX2-NEXT:    vandps %ymm1, %ymm0, %ymm0
-; X32-AVX2-NEXT:    vmovaps %ymm0, (%eax)
+; X32-AVX2-NEXT:    vpsrld $31, %ymm0, %ymm0
+; X32-AVX2-NEXT:    vmovdqa %ymm0, (%eax)
 ; X32-AVX2-NEXT:    vzeroupper
 ; X32-AVX2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: and_masks:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vmovups (%rdi), %ymm0
 ; X64-AVX2-NEXT:    vmovups (%rsi), %ymm1
 ; X64-AVX2-NEXT:    vcmpltps %ymm0, %ymm1, %ymm1
 ; X64-AVX2-NEXT:    vmovups (%rdx), %ymm2
 ; X64-AVX2-NEXT:    vcmpltps %ymm0, %ymm2, %ymm0
-; X64-AVX2-NEXT:    vbroadcastss {{.*#+}} ymm2 = [1,1,1,1,1,1,1,1]
-; X64-AVX2-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; X64-AVX2-NEXT:    vandps %ymm1, %ymm0, %ymm0
-; X64-AVX2-NEXT:    vmovaps %ymm0, (%rax)
+; X64-AVX2-NEXT:    vpsrld $31, %ymm0, %ymm0
+; X64-AVX2-NEXT:    vmovdqa %ymm0, (%rax)
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
   %v0 = load <8 x float>, <8 x float>* %a, align 16
   %v1 = load <8 x float>, <8 x float>* %b, align 16
   %m0 = fcmp olt <8 x float> %v1, %v0
   %v2 = load <8 x float>, <8 x float>* %c, align 16
   %m1 = fcmp olt <8 x float> %v2, %v0
   %mand = and <8 x i1> %m1, %m0

llvm/trunk/test/CodeGen/X86/vector-blend.ll

 ; SSE41-NEXT:    movaps {{.*#+}} xmm0 = [255,255,0,255,0,0,0,255,255,255,0,255,0,0,0,255]
 ; SSE41-NEXT:    pblendvb %xmm0, %xmm2, %xmm4
 ; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm1
 ; SSE41-NEXT:    movdqa %xmm4, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: constant_pblendvb_avx2:
 ; AVX1:       # %bb.0: # %entry
-; AVX1-NEXT:    vmovaps {{.*#+}} ymm2 = [255,255,0,255,0,0,0,255,255,255,0,255,0,0,0,255,255,255,0,255,0,0,0,255,255,255,0,255,0,0,0,255]
+; AVX1-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [-5.4861292804117373E+303,-5.4861292804117373E+303,-5.4861292804117373E+303,-5.4861292804117373E+303]
 ; AVX1-NEXT:    vandnps %ymm0, %ymm2, %ymm0
 ; AVX1-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; AVX1-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: constant_pblendvb_avx2:
 ; AVX2:       # %bb.0: # %entry
 ; AVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,0,255,0,0,0,255,255,255,0,255,0,0,0,255,255,255,0,255,0,0,0,255,255,255,0,255,0,0,0,255]

llvm/trunk/test/CodeGen/X86/vector-reduce-umax.ll

 define i16 @test_v8i16(<8 x i16> %a0) {
 ; SSE2-LABEL: test_v8i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE2-NEXT:    pxor %xmm2, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
-; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
+; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v8i16:

 
 define i16 @test_v16i16(<16 x i16> %a0) {
 ; SSE2-LABEL: test_v16i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
-; SSE2-NEXT:    pxor %xmm2, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE2-NEXT:    pxor %xmm2, %xmm0
-; SSE2-NEXT:    pxor %xmm2, %xmm1
-; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
+; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
-; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
+; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i16:
; AVX512VL-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.umax.i16.v16i16(<16 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v32i16(<32 x i16> %a0) {
 ; SSE2-LABEL: test_v32i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE2-NEXT:    pxor %xmm4, %xmm2
-; SSE2-NEXT:    pxor %xmm4, %xmm0
-; SSE2-NEXT:    pmaxsw %xmm2, %xmm0
 ; SSE2-NEXT:    pxor %xmm4, %xmm3
 ; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm3, %xmm1
-; SSE2-NEXT:    movdqa %xmm4, %xmm2
-; SSE2-NEXT:    pxor %xmm4, %xmm2
-; SSE2-NEXT:    pxor %xmm2, %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm2
-; SSE2-NEXT:    pmaxsw %xmm1, %xmm2
-; SSE2-NEXT:    pxor %xmm4, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
 ; SSE2-NEXT:    pxor %xmm4, %xmm2
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm2, %xmm0
-; SSE2-NEXT:    pxor %xmm4, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE2-NEXT:    pxor %xmm4, %xmm0
+; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm4, %xmm1
-; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; SSE2-NEXT:    pxor %xmm4, %xmm1
+; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrld $16, %xmm0
-; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
-; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
+; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    pxor %xmm4, %xmm1
+; SSE2-NEXT:    psrld $16, %xmm1
+; SSE2-NEXT:    pxor %xmm4, %xmm1
+; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; SSE2-NEXT:    pxor %xmm4, %xmm1
+; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxuw %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxuw %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxuw %xmm1, %xmm0
; AVX512VL-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.umax.i16.v32i16(<32 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v64i16(<64 x i16> %a0) {
 ; SSE2-LABEL: test_v64i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm8 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE2-NEXT:    pxor %xmm8, %xmm5
-; SSE2-NEXT:    pxor %xmm8, %xmm1
-; SSE2-NEXT:    pmaxsw %xmm5, %xmm1
-; SSE2-NEXT:    pxor %xmm8, %xmm7
-; SSE2-NEXT:    pxor %xmm8, %xmm3
-; SSE2-NEXT:    pmaxsw %xmm7, %xmm3
-; SSE2-NEXT:    pxor %xmm8, %xmm4
-; SSE2-NEXT:    pxor %xmm8, %xmm0
-; SSE2-NEXT:    pmaxsw %xmm4, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm6
 ; SSE2-NEXT:    pxor %xmm8, %xmm2
 ; SSE2-NEXT:    pmaxsw %xmm6, %xmm2
-; SSE2-NEXT:    movdqa %xmm8, %xmm4
 ; SSE2-NEXT:    pxor %xmm8, %xmm4
-; SSE2-NEXT:    pxor %xmm4, %xmm2
-; SSE2-NEXT:    pxor %xmm4, %xmm0
+; SSE2-NEXT:    pxor %xmm8, %xmm0
+; SSE2-NEXT:    pmaxsw %xmm4, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm2, %xmm0
-; SSE2-NEXT:    pxor %xmm4, %xmm3
-; SSE2-NEXT:    pxor %xmm4, %xmm1
+; SSE2-NEXT:    pxor %xmm8, %xmm7
+; SSE2-NEXT:    pxor %xmm8, %xmm3
+; SSE2-NEXT:    pmaxsw %xmm7, %xmm3
+; SSE2-NEXT:    pxor %xmm8, %xmm5
+; SSE2-NEXT:    pxor %xmm8, %xmm1
+; SSE2-NEXT:    pmaxsw %xmm5, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm3, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm0
-; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
+; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm8, %xmm1
-; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm8, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm8, %xmm1
+; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
-; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
+; SSE2-NEXT:    psrld $16, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrld $16, %xmm1
+; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
-; SSE2-NEXT:    pxor %xmm8, %xmm1
-; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm8, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    movd %xmm0, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v64i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxuw %xmm7, %xmm3
 ; SSE41-NEXT:    pmaxuw %xmm5, %xmm1
 ; SSE41-NEXT:    pmaxuw %xmm3, %xmm1

llvm/trunk/test/CodeGen/X86/vector-reduce-umin.ll

 define i16 @test_v8i16(<8 x i16> %a0) {
 ; SSE2-LABEL: test_v8i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE2-NEXT:    pxor %xmm2, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
-; SSE2-NEXT:    pminsw %xmm1, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
+; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v8i16:

 
 define i16 @test_v16i16(<16 x i16> %a0) {
 ; SSE2-LABEL: test_v16i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
-; SSE2-NEXT:    pxor %xmm2, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE2-NEXT:    pxor %xmm2, %xmm0
-; SSE2-NEXT:    pxor %xmm2, %xmm1
-; SSE2-NEXT:    pminsw %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
+; SSE2-NEXT:    pminsw %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
-; SSE2-NEXT:    pminsw %xmm1, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
+; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i16:
; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.umin.i16.v16i16(<16 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v32i16(<32 x i16> %a0) {
 ; SSE2-LABEL: test_v32i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE2-NEXT:    pxor %xmm4, %xmm2
-; SSE2-NEXT:    pxor %xmm4, %xmm0
-; SSE2-NEXT:    pminsw %xmm2, %xmm0
 ; SSE2-NEXT:    pxor %xmm4, %xmm3
 ; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    pminsw %xmm3, %xmm1
-; SSE2-NEXT:    movdqa %xmm4, %xmm2
-; SSE2-NEXT:    pxor %xmm4, %xmm2
-; SSE2-NEXT:    pxor %xmm2, %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm2
-; SSE2-NEXT:    pminsw %xmm1, %xmm2
-; SSE2-NEXT:    pxor %xmm4, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
 ; SSE2-NEXT:    pxor %xmm4, %xmm2
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
 ; SSE2-NEXT:    pminsw %xmm2, %xmm0
-; SSE2-NEXT:    pxor %xmm4, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE2-NEXT:    pxor %xmm4, %xmm0
+; SSE2-NEXT:    pminsw %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm4, %xmm1
-; SSE2-NEXT:    pminsw %xmm0, %xmm1
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; SSE2-NEXT:    pxor %xmm4, %xmm1
+; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psrld $16, %xmm0
-; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
-; SSE2-NEXT:    pminsw %xmm1, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
+; SSE2-NEXT:    pminsw %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    pxor %xmm4, %xmm1
+; SSE2-NEXT:    psrld $16, %xmm1
+; SSE2-NEXT:    pxor %xmm4, %xmm1
+; SSE2-NEXT:    pminsw %xmm0, %xmm1
+; SSE2-NEXT:    pxor %xmm4, %xmm1
+; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminuw %xmm3, %xmm1
 ; SSE41-NEXT:    pminuw %xmm2, %xmm0
 ; SSE41-NEXT:    pminuw %xmm1, %xmm0
; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.umin.i16.v32i16(<32 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v64i16(<64 x i16> %a0) {
 ; SSE2-LABEL: test_v64i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm8 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE2-NEXT:    pxor %xmm8, %xmm5
-; SSE2-NEXT:    pxor %xmm8, %xmm1
-; SSE2-NEXT:    pminsw %xmm5, %xmm1
-; SSE2-NEXT:    pxor %xmm8, %xmm7
-; SSE2-NEXT:    pxor %xmm8, %xmm3
-; SSE2-NEXT:    pminsw %xmm7, %xmm3
-; SSE2-NEXT:    pxor %xmm8, %xmm4
-; SSE2-NEXT:    pxor %xmm8, %xmm0
-; SSE2-NEXT:    pminsw %xmm4, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm6
 ; SSE2-NEXT:    pxor %xmm8, %xmm2
 ; SSE2-NEXT:    pminsw %xmm6, %xmm2
-; SSE2-NEXT:    movdqa %xmm8, %xmm4
 ; SSE2-NEXT:    pxor %xmm8, %xmm4
-; SSE2-NEXT:    pxor %xmm4, %xmm2
-; SSE2-NEXT:    pxor %xmm4, %xmm0
+; SSE2-NEXT:    pxor %xmm8, %xmm0
+; SSE2-NEXT:    pminsw %xmm4, %xmm0
 ; SSE2-NEXT:    pminsw %xmm2, %xmm0
-; SSE2-NEXT:    pxor %xmm4, %xmm3
-; SSE2-NEXT:    pxor %xmm4, %xmm1
+; SSE2-NEXT:    pxor %xmm8, %xmm7
+; SSE2-NEXT:    pxor %xmm8, %xmm3
+; SSE2-NEXT:    pminsw %xmm7, %xmm3
+; SSE2-NEXT:    pxor %xmm8, %xmm5
+; SSE2-NEXT:    pxor %xmm8, %xmm1
+; SSE2-NEXT:    pminsw %xmm5, %xmm1
 ; SSE2-NEXT:    pminsw %xmm3, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm0
-; SSE2-NEXT:    pminsw %xmm1, %xmm0
+; SSE2-NEXT:    pminsw %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
+; SSE2-NEXT:    pminsw %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm8, %xmm1
-; SSE2-NEXT:    pminsw %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm8, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm8, %xmm1
+; SSE2-NEXT:    pminsw %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
-; SSE2-NEXT:    pminsw %xmm1, %xmm0
+; SSE2-NEXT:    psrld $16, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrld $16, %xmm1
+; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
-; SSE2-NEXT:    pxor %xmm8, %xmm1
-; SSE2-NEXT:    pminsw %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm8, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    movd %xmm0, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v64i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminuw %xmm7, %xmm3
 ; SSE41-NEXT:    pminuw %xmm5, %xmm1
 ; SSE41-NEXT:    pminuw %xmm3, %xmm1

llvm/trunk/test/CodeGen/X86/vector-shift-lshr-128.ll

 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    psllw $5, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm2
 ; SSE2-NEXT:    pxor %xmm3, %xmm3
 ; SSE2-NEXT:    pcmpgtb %xmm1, %xmm3
 ; SSE2-NEXT:    movdqa %xmm3, %xmm4
 ; SSE2-NEXT:    pandn %xmm0, %xmm4
 ; SSE2-NEXT:    psrlw $4, %xmm0
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE2-NEXT:    pand %xmm3, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE2-NEXT:    por %xmm4, %xmm0
 ; SSE2-NEXT:    paddb %xmm1, %xmm1
 ; SSE2-NEXT:    pxor %xmm3, %xmm3
 ; SSE2-NEXT:    pcmpgtb %xmm1, %xmm3
 ; SSE2-NEXT:    movdqa %xmm3, %xmm4
 ; SSE2-NEXT:    pandn %xmm0, %xmm4
 ; SSE2-NEXT:    psrlw $2, %xmm0
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE2-NEXT:    pand %xmm3, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE2-NEXT:    por %xmm4, %xmm0
 ; SSE2-NEXT:    paddb %xmm1, %xmm1
 ; SSE2-NEXT:    pcmpgtb %xmm1, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm1
 ; SSE2-NEXT:    pandn %xmm0, %xmm1
 ; SSE2-NEXT:    psrlw $1, %xmm0
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE2-NEXT:    pand %xmm2, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: var_shift_v16i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm2
 ; SSE41-NEXT:    psllw $5, %xmm1
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3

 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    psllw $5, %xmm1
 ; X32-SSE-NEXT:    pxor %xmm2, %xmm2
 ; X32-SSE-NEXT:    pxor %xmm3, %xmm3
 ; X32-SSE-NEXT:    pcmpgtb %xmm1, %xmm3
 ; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
 ; X32-SSE-NEXT:    pandn %xmm0, %xmm4
 ; X32-SSE-NEXT:    psrlw $4, %xmm0
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
 ; X32-SSE-NEXT:    pand %xmm3, %xmm0
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
 ; X32-SSE-NEXT:    por %xmm4, %xmm0
 ; X32-SSE-NEXT:    paddb %xmm1, %xmm1
 ; X32-SSE-NEXT:    pxor %xmm3, %xmm3
 ; X32-SSE-NEXT:    pcmpgtb %xmm1, %xmm3
 ; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
 ; X32-SSE-NEXT:    pandn %xmm0, %xmm4
 ; X32-SSE-NEXT:    psrlw $2, %xmm0
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
 ; X32-SSE-NEXT:    pand %xmm3, %xmm0
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
 ; X32-SSE-NEXT:    por %xmm4, %xmm0
 ; X32-SSE-NEXT:    paddb %xmm1, %xmm1
 ; X32-SSE-NEXT:    pcmpgtb %xmm1, %xmm2
 ; X32-SSE-NEXT:    movdqa %xmm2, %xmm1
 ; X32-SSE-NEXT:    pandn %xmm0, %xmm1
 ; X32-SSE-NEXT:    psrlw $1, %xmm0
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
 ; X32-SSE-NEXT:    pand %xmm2, %xmm0
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
 ; X32-SSE-NEXT:    por %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %shift = lshr <16 x i8> %a, %b
   ret <16 x i8> %shift
 }
 
 ;
 ; Uniform Variable Shifts

llvm/trunk/test/CodeGen/X86/vector-shift-shl-128.ll

 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    psllw $5, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm2
 ; SSE2-NEXT:    pxor %xmm3, %xmm3
 ; SSE2-NEXT:    pcmpgtb %xmm1, %xmm3
 ; SSE2-NEXT:    movdqa %xmm3, %xmm4
 ; SSE2-NEXT:    pandn %xmm0, %xmm4
 ; SSE2-NEXT:    psllw $4, %xmm0
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE2-NEXT:    pand %xmm3, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE2-NEXT:    por %xmm4, %xmm0
 ; SSE2-NEXT:    paddb %xmm1, %xmm1
 ; SSE2-NEXT:    pxor %xmm3, %xmm3
 ; SSE2-NEXT:    pcmpgtb %xmm1, %xmm3
 ; SSE2-NEXT:    movdqa %xmm3, %xmm4
 ; SSE2-NEXT:    pandn %xmm0, %xmm4
 ; SSE2-NEXT:    psllw $2, %xmm0
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE2-NEXT:    pand %xmm3, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE2-NEXT:    por %xmm4, %xmm0
 ; SSE2-NEXT:    paddb %xmm1, %xmm1
 ; SSE2-NEXT:    pcmpgtb %xmm1, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm1
 ; SSE2-NEXT:    pandn %xmm0, %xmm1
 ; SSE2-NEXT:    paddb %xmm0, %xmm0
 ; SSE2-NEXT:    pand %xmm2, %xmm0
 ; SSE2-NEXT:    por %xmm1, %xmm0

 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    psllw $5, %xmm1
 ; X32-SSE-NEXT:    pxor %xmm2, %xmm2
 ; X32-SSE-NEXT:    pxor %xmm3, %xmm3
 ; X32-SSE-NEXT:    pcmpgtb %xmm1, %xmm3
 ; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
 ; X32-SSE-NEXT:    pandn %xmm0, %xmm4
 ; X32-SSE-NEXT:    psllw $4, %xmm0
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
 ; X32-SSE-NEXT:    pand %xmm3, %xmm0
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
 ; X32-SSE-NEXT:    por %xmm4, %xmm0
 ; X32-SSE-NEXT:    paddb %xmm1, %xmm1
 ; X32-SSE-NEXT:    pxor %xmm3, %xmm3
 ; X32-SSE-NEXT:    pcmpgtb %xmm1, %xmm3
 ; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
 ; X32-SSE-NEXT:    pandn %xmm0, %xmm4
 ; X32-SSE-NEXT:    psllw $2, %xmm0
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
 ; X32-SSE-NEXT:    pand %xmm3, %xmm0
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
 ; X32-SSE-NEXT:    por %xmm4, %xmm0
 ; X32-SSE-NEXT:    paddb %xmm1, %xmm1
 ; X32-SSE-NEXT:    pcmpgtb %xmm1, %xmm2
 ; X32-SSE-NEXT:    movdqa %xmm2, %xmm1
 ; X32-SSE-NEXT:    pandn %xmm0, %xmm1
 ; X32-SSE-NEXT:    paddb %xmm0, %xmm0
 ; X32-SSE-NEXT:    pand %xmm2, %xmm0
 ; X32-SSE-NEXT:    por %xmm1, %xmm0

llvm/trunk/test/CodeGen/X86/vector-shuffle-256-v16.ll

 ; AVX2OR512VL-NEXT:    retq
   %shuffle = shufflevector <16 x i16> %a, <16 x i16> %b, <16 x i32> <i32 7, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 7, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   ret <16 x i16> %shuffle
 }
 
 define <16 x i16> @shuffle_v16i16_00_17_02_19_04_21_06_23_08_25_10_27_12_29_14_31(<16 x i16> %a, <16 x i16> %b) {
 ; AVX1-LABEL: shuffle_v16i16_00_17_02_19_04_21_06_23_08_25_10_27_12_29_14_31:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovaps {{.*#+}} ymm2 = [65535,0,65535,0,65535,0,65535,0,65535,0,65535,0,65535,0,65535,0]
+; AVX1-NEXT:    vbroadcastss {{.*#+}} ymm2 = [9.18340949E-41,9.18340949E-41,9.18340949E-41,9.18340949E-41,9.18340949E-41,9.18340949E-41,9.18340949E-41,9.18340949E-41]
 ; AVX1-NEXT:    vandnps %ymm1, %ymm2, %ymm1
 ; AVX1-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2OR512VL-LABEL: shuffle_v16i16_00_17_02_19_04_21_06_23_08_25_10_27_12_29_14_31:
 ; AVX2OR512VL:       # %bb.0:
 ; AVX2OR512VL-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7],ymm0[8],ymm1[9],ymm0[10],ymm1[11],ymm0[12],ymm1[13],ymm0[14],ymm1[15]
 ; AVX2OR512VL-NEXT:    retq
   %shuffle = shufflevector <16 x i16> %a, <16 x i16> %b, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
   ret <16 x i16> %shuffle
 }
 
 define <16 x i16> @shuffle_v16i16_16_01_18_03_20_05_22_07_24_09_26_11_28_13_30_15(<16 x i16> %a, <16 x i16> %b) {
 ; AVX1-LABEL: shuffle_v16i16_16_01_18_03_20_05_22_07_24_09_26_11_28_13_30_15:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovaps {{.*#+}} ymm2 = [65535,0,65535,0,65535,0,65535,0,65535,0,65535,0,65535,0,65535,0]
+; AVX1-NEXT:    vbroadcastss {{.*#+}} ymm2 = [9.18340949E-41,9.18340949E-41,9.18340949E-41,9.18340949E-41,9.18340949E-41,9.18340949E-41,9.18340949E-41,9.18340949E-41]
 ; AVX1-NEXT:    vandnps %ymm0, %ymm2, %ymm0
 ; AVX1-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; AVX1-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2OR512VL-LABEL: shuffle_v16i16_16_01_18_03_20_05_22_07_24_09_26_11_28_13_30_15:
 ; AVX2OR512VL:       # %bb.0:
 ; AVX2OR512VL-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7],ymm1[8],ymm0[9],ymm1[10],ymm0[11],ymm1[12],ymm0[13],ymm1[14],ymm0[15]

llvm/trunk/test/CodeGen/X86/vector-trunc-math.ll

 ; AVX1-NEXT:    vpaddq %xmm6, %xmm2, %xmm5
 ; AVX1-NEXT:    vextractf128 $1, %ymm6, %xmm6
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm2
 ; AVX1-NEXT:    vpaddq %xmm6, %xmm2, %xmm2
 ; AVX1-NEXT:    vpaddq %xmm7, %xmm3, %xmm6
 ; AVX1-NEXT:    vextractf128 $1, %ymm7, %xmm7
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm3
 ; AVX1-NEXT:    vpaddq %xmm7, %xmm3, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm7 = [1.2598673968951787E-321,1.2598673968951787E-321]
+; AVX1-NEXT:    # xmm7 = mem[0,0]
 ; AVX1-NEXT:    vpand %xmm7, %xmm3, %xmm3
 ; AVX1-NEXT:    vpand %xmm7, %xmm6, %xmm6
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm6, %xmm3
 ; AVX1-NEXT:    vpand %xmm7, %xmm2, %xmm2
 ; AVX1-NEXT:    vpand %xmm7, %xmm5, %xmm5
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm5, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpand %xmm7, %xmm1, %xmm1

 ; AVX1-NEXT:    vpaddd %xmm2, %xmm0, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    vpaddd %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpaddd %xmm3, %xmm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
 ; AVX1-NEXT:    vpaddd %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
 ; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm3, %xmm4, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper

 ; SSE-NEXT:    packuswb %xmm2, %xmm0
 ; SSE-NEXT:    packuswb %xmm4, %xmm0
 ; SSE-NEXT:    paddb {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_add_const_v16i64_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
-; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm5 = [1.2598673968951787E-321,1.2598673968951787E-321]
+; AVX1-NEXT:    # xmm5 = mem[0,0]
+; AVX1-NEXT:    vandpd %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
 ; AVX1-NEXT:    vpackusdw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vandpd %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vandpd %xmm5, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm1, %xmm1
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vandpd %xmm5, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm0, %xmm0
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vandpd %xmm5, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpaddb {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: trunc_add_const_v16i64_v16i8:

 ; SSE-NEXT:    packuswb %xmm1, %xmm0
 ; SSE-NEXT:    packuswb %xmm2, %xmm0
 ; SSE-NEXT:    paddb {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_add_const_v16i32_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
+; AVX1-NEXT:    vandps %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vandps %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vandps %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vandps %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpaddb {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: trunc_add_const_v16i32_v16i8:
 ; AVX2:       # %bb.0:

 ; AVX1-NEXT:    vpsubq %xmm6, %xmm2, %xmm5
 ; AVX1-NEXT:    vextractf128 $1, %ymm6, %xmm6
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm2
 ; AVX1-NEXT:    vpsubq %xmm6, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsubq %xmm7, %xmm3, %xmm6
 ; AVX1-NEXT:    vextractf128 $1, %ymm7, %xmm7
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm3
 ; AVX1-NEXT:    vpsubq %xmm7, %xmm3, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm7 = [1.2598673968951787E-321,1.2598673968951787E-321]
+; AVX1-NEXT:    # xmm7 = mem[0,0]
 ; AVX1-NEXT:    vpand %xmm7, %xmm3, %xmm3
 ; AVX1-NEXT:    vpand %xmm7, %xmm6, %xmm6
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm6, %xmm3
 ; AVX1-NEXT:    vpand %xmm7, %xmm2, %xmm2
 ; AVX1-NEXT:    vpand %xmm7, %xmm5, %xmm5
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm5, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpand %xmm7, %xmm1, %xmm1

 ; AVX1-NEXT:    vpsubd %xmm2, %xmm0, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    vpsubd %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsubd %xmm3, %xmm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
 ; AVX1-NEXT:    vpsubd %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
 ; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm3, %xmm4, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper

 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
 ; AVX1-NEXT:    vpsubq {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vpsubq {{.*}}(%rip), %xmm2, %xmm6
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm2
 ; AVX1-NEXT:    vpsubq {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vpsubq {{.*}}(%rip), %xmm3, %xmm7
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm3
 ; AVX1-NEXT:    vpsubq {{.*}}(%rip), %xmm3, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm4 = [1.2598673968951787E-321,1.2598673968951787E-321]
+; AVX1-NEXT:    # xmm4 = mem[0,0]
 ; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpand %xmm4, %xmm7, %xmm7
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm7, %xmm3
 ; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpand %xmm4, %xmm6, %xmm6
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm6, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1

 ; AVX1-LABEL: trunc_sub_const_v16i32_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpsubd {{.*}}(%rip), %xmm0, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    vpsubd {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpsubd {{.*}}(%rip), %xmm1, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
 ; AVX1-NEXT:    vpsubd {{.*}}(%rip), %xmm1, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm4 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
 ; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpackusdw %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpand %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper

 ; AVX1-NEXT:    vpsrlq $32, %xmm3, %xmm6
 ; AVX1-NEXT:    vpmuludq %xmm4, %xmm6, %xmm6
 ; AVX1-NEXT:    vpsrlq $32, %xmm4, %xmm7
 ; AVX1-NEXT:    vpmuludq %xmm7, %xmm3, %xmm7
 ; AVX1-NEXT:    vpaddq %xmm6, %xmm7, %xmm6
 ; AVX1-NEXT:    vpsllq $32, %xmm6, %xmm6
 ; AVX1-NEXT:    vpmuludq %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpaddq %xmm6, %xmm3, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm4 = [1.2598673968951787E-321,1.2598673968951787E-321]
+; AVX1-NEXT:    # xmm4 = mem[0,0]
 ; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpand %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm4, %xmm5, %xmm3
 ; AVX1-NEXT:    vpackusdw %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1

 ; AVX1-NEXT:    vpmulld %xmm2, %xmm0, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    vpmulld %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmulld %xmm3, %xmm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
 ; AVX1-NEXT:    vpmulld %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
 ; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm3, %xmm4, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper

 ; AVX1-NEXT:    vpaddq %xmm0, %xmm7, %xmm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm3
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [14,15]
 ; AVX1-NEXT:    vpmuludq %xmm6, %xmm3, %xmm7
 ; AVX1-NEXT:    vpsrlq $32, %xmm3, %xmm3
 ; AVX1-NEXT:    vpmuludq %xmm6, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsllq $32, %xmm3, %xmm3
 ; AVX1-NEXT:    vpaddq %xmm3, %xmm7, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm6 = [1.2598673968951787E-321,1.2598673968951787E-321]
+; AVX1-NEXT:    # xmm6 = mem[0,0]
 ; AVX1-NEXT:    vpand %xmm6, %xmm3, %xmm3
 ; AVX1-NEXT:    vpand %xmm6, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm6, %xmm2, %xmm2
 ; AVX1-NEXT:    vpand %xmm6, %xmm4, %xmm3
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpand %xmm6, %xmm1, %xmm1

 ; AVX1-LABEL: trunc_mul_const_v16i32_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm0, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
 ; AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm1, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm4 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
 ; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpackusdw %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpand %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper

 ; SSE-NEXT:    pand %xmm8, %xmm0
 ; SSE-NEXT:    packuswb %xmm1, %xmm0
 ; SSE-NEXT:    packuswb %xmm2, %xmm0
 ; SSE-NEXT:    packuswb %xmm4, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_and_v16i64_v16i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vandps %ymm4, %ymm0, %ymm0
-; AVX1-NEXT:    vandps %ymm5, %ymm1, %ymm1
-; AVX1-NEXT:    vandps %ymm6, %ymm2, %ymm2
-; AVX1-NEXT:    vandps %ymm7, %ymm3, %ymm3
+; AVX1-NEXT:    vandpd %ymm4, %ymm0, %ymm0
+; AVX1-NEXT:    vandpd %ymm5, %ymm1, %ymm1
+; AVX1-NEXT:    vandpd %ymm6, %ymm2, %ymm2
+; AVX1-NEXT:    vandpd %ymm7, %ymm3, %ymm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX1-NEXT:    vmovaps {{.*#+}} xmm5 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
-; AVX1-NEXT:    vandps %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vandps %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm5 = [1.2598673968951787E-321,1.2598673968951787E-321]
+; AVX1-NEXT:    # xmm5 = mem[0,0]
+; AVX1-NEXT:    vandpd %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
 ; AVX1-NEXT:    vpackusdw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vandps %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vandps %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vandpd %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vandpd %xmm5, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT:    vandps %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vandps %xmm5, %xmm1, %xmm1
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vandpd %xmm5, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT:    vandps %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vandps %xmm5, %xmm0, %xmm0
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vandpd %xmm5, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: trunc_and_v16i64_v16i8:
 ; AVX2-SLOW:       # %bb.0:

 ; SSE-NEXT:    packuswb %xmm6, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_and_v16i32_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; AVX1-NEXT:    vandps %ymm3, %ymm1, %ymm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vmovaps {{.*#+}} xmm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
 ; AVX1-NEXT:    vandps %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vandps %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vandps %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vandps %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0

 ; SSE-NEXT:    packuswb %xmm2, %xmm0
 ; SSE-NEXT:    packuswb %xmm4, %xmm0
 ; SSE-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_and_const_v16i64_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
-; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm5 = [1.2598673968951787E-321,1.2598673968951787E-321]
+; AVX1-NEXT:    # xmm5 = mem[0,0]
+; AVX1-NEXT:    vandpd %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
 ; AVX1-NEXT:    vpackusdw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vandpd %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vandpd %xmm5, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm1, %xmm1
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vandpd %xmm5, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm0, %xmm0
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vandpd %xmm5, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: trunc_and_const_v16i64_v16i8:

 ; SSE-NEXT:    packuswb %xmm1, %xmm0
 ; SSE-NEXT:    packuswb %xmm2, %xmm0
 ; SSE-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_and_const_v16i32_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
+; AVX1-NEXT:    vandps %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vandps %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vandps %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vandps %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: trunc_and_const_v16i32_v16i8:
 ; AVX2:       # %bb.0:

 ; SSE-NEXT:    pand %xmm8, %xmm0
 ; SSE-NEXT:    packuswb %xmm1, %xmm0
 ; SSE-NEXT:    packuswb %xmm2, %xmm0
 ; SSE-NEXT:    packuswb %xmm4, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_xor_v16i64_v16i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vxorps %ymm4, %ymm0, %ymm0
-; AVX1-NEXT:    vxorps %ymm5, %ymm1, %ymm1
-; AVX1-NEXT:    vxorps %ymm6, %ymm2, %ymm2
-; AVX1-NEXT:    vxorps %ymm7, %ymm3, %ymm3
+; AVX1-NEXT:    vxorpd %ymm4, %ymm0, %ymm0
+; AVX1-NEXT:    vxorpd %ymm5, %ymm1, %ymm1
+; AVX1-NEXT:    vxorpd %ymm6, %ymm2, %ymm2
+; AVX1-NEXT:    vxorpd %ymm7, %ymm3, %ymm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX1-NEXT:    vmovaps {{.*#+}} xmm5 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
-; AVX1-NEXT:    vandps %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vandps %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm5 = [1.2598673968951787E-321,1.2598673968951787E-321]
+; AVX1-NEXT:    # xmm5 = mem[0,0]
+; AVX1-NEXT:    vandpd %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
 ; AVX1-NEXT:    vpackusdw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vandps %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vandps %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vandpd %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vandpd %xmm5, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT:    vandps %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vandps %xmm5, %xmm1, %xmm1
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vandpd %xmm5, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT:    vandps %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vandps %xmm5, %xmm0, %xmm0
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vandpd %xmm5, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: trunc_xor_v16i64_v16i8:
 ; AVX2-SLOW:       # %bb.0:

 ; SSE-NEXT:    packuswb %xmm2, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_xor_v16i32_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vxorps %ymm2, %ymm0, %ymm0
 ; AVX1-NEXT:    vxorps %ymm3, %ymm1, %ymm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vmovaps {{.*#+}} xmm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
 ; AVX1-NEXT:    vandps %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vandps %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vandps %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vandps %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0

 ; SSE-NEXT:    packuswb %xmm2, %xmm0
 ; SSE-NEXT:    packuswb %xmm4, %xmm0
 ; SSE-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_xor_const_v16i64_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
-; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm5 = [1.2598673968951787E-321,1.2598673968951787E-321]
+; AVX1-NEXT:    # xmm5 = mem[0,0]
+; AVX1-NEXT:    vandpd %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
 ; AVX1-NEXT:    vpackusdw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vandpd %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vandpd %xmm5, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm1, %xmm1
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vandpd %xmm5, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm0, %xmm0
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vandpd %xmm5, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: trunc_xor_const_v16i64_v16i8:

 ; SSE-NEXT:    packuswb %xmm1, %xmm0
 ; SSE-NEXT:    packuswb %xmm2, %xmm0
 ; SSE-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_xor_const_v16i32_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
+; AVX1-NEXT:    vandps %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vandps %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vandps %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vandps %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: trunc_xor_const_v16i32_v16i8:
 ; AVX2:       # %bb.0:

 ; SSE-NEXT:    pand %xmm8, %xmm0
 ; SSE-NEXT:    packuswb %xmm1, %xmm0
 ; SSE-NEXT:    packuswb %xmm2, %xmm0
 ; SSE-NEXT:    packuswb %xmm4, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_or_v16i64_v16i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vorps %ymm4, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm5, %ymm1, %ymm1
-; AVX1-NEXT:    vorps %ymm6, %ymm2, %ymm2
-; AVX1-NEXT:    vorps %ymm7, %ymm3, %ymm3
+; AVX1-NEXT:    vorpd %ymm4, %ymm0, %ymm0
+; AVX1-NEXT:    vorpd %ymm5, %ymm1, %ymm1
+; AVX1-NEXT:    vorpd %ymm6, %ymm2, %ymm2
+; AVX1-NEXT:    vorpd %ymm7, %ymm3, %ymm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX1-NEXT:    vmovaps {{.*#+}} xmm5 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
-; AVX1-NEXT:    vandps %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vandps %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm5 = [1.2598673968951787E-321,1.2598673968951787E-321]
+; AVX1-NEXT:    # xmm5 = mem[0,0]
+; AVX1-NEXT:    vandpd %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
 ; AVX1-NEXT:    vpackusdw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vandps %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vandps %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vandpd %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vandpd %xmm5, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT:    vandps %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vandps %xmm5, %xmm1, %xmm1
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vandpd %xmm5, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT:    vandps %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vandps %xmm5, %xmm0, %xmm0
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vandpd %xmm5, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: trunc_or_v16i64_v16i8:
 ; AVX2-SLOW:       # %bb.0:

 ; SSE-NEXT:    packuswb %xmm2, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_or_v16i32_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vorps %ymm2, %ymm0, %ymm0
 ; AVX1-NEXT:    vorps %ymm3, %ymm1, %ymm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vmovaps {{.*#+}} xmm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
 ; AVX1-NEXT:    vandps %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vandps %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vandps %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vandps %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0

 ; SSE-NEXT:    packuswb %xmm2, %xmm0
 ; SSE-NEXT:    packuswb %xmm4, %xmm0
 ; SSE-NEXT:    por {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_or_const_v16i64_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
-; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm5 = [1.2598673968951787E-321,1.2598673968951787E-321]
+; AVX1-NEXT:    # xmm5 = mem[0,0]
+; AVX1-NEXT:    vandpd %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
 ; AVX1-NEXT:    vpackusdw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vandpd %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vandpd %xmm5, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm1, %xmm1
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vandpd %xmm5, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpand %xmm5, %xmm0, %xmm0
+; AVX1-NEXT:    vandpd %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vandpd %xmm5, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: trunc_or_const_v16i64_v16i8:

 ; SSE-NEXT:    packuswb %xmm1, %xmm0
 ; SSE-NEXT:    packuswb %xmm2, %xmm0
 ; SSE-NEXT:    por {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_or_const_v16i32_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
+; AVX1-NEXT:    vandps %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vandps %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vandps %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vandps %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: trunc_or_const_v16i32_v16i8:
 ; AVX2:       # %bb.0:

llvm/trunk/test/CodeGen/X86/vector-trunc-packus.ll

 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,255]
 ; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm4, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm4, %xmm5
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm5, %ymm3
 ; AVX1-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm4, %xmm5
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm5, %ymm3
+; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm4, %xmm4
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
 ; AVX1-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
 ; AVX1-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm0, %xmm8
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm5, %xmm6
-; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm1, %xmm7
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vpand %xmm1, %xmm7, %xmm1
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
+; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm4, %xmm5
+; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm1, %xmm6
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
+; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm7, %xmm2
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm3 = [1.2598673968951787E-321,1.2598673968951787E-321]
+; AVX1-NEXT:    # xmm3 = mem[0,0]
+; AVX1-NEXT:    vpand %xmm3, %xmm7, %xmm7
+; AVX1-NEXT:    vpand %xmm7, %xmm2, %xmm2
+; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpand %xmm1, %xmm6, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpand %xmm4, %xmm5, %xmm2
-; AVX1-NEXT:    vpand %xmm2, %xmm6, %xmm2
-; AVX1-NEXT:    vpand %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vpand %xmm3, %xmm4, %xmm2
+; AVX1-NEXT:    vpand %xmm2, %xmm5, %xmm2
+; AVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm0, %xmm8, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm0, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovq %xmm0, (%rdi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;

llvm/trunk/test/CodeGen/X86/vector-trunc-ssat.ll

 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm5, %ymm3
 ; AVX1-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm4, %xmm1, %xmm4
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
 ; AVX1-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vmovapd {{.*#+}} xmm3 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm3 = [1.2598673968951787E-321,1.2598673968951787E-321]
+; AVX1-NEXT:    # xmm3 = mem[0,0]
 ; AVX1-NEXT:    vandpd %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vandpd %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vandpd %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vandpd %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0

llvm/trunk/test/CodeGen/X86/vector-trunc-usat.ll

 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
 ; AVX1-NEXT:    vpxor %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm5, %xmm3
 ; AVX1-NEXT:    vpxor %xmm4, %xmm1, %xmm4
 ; AVX1-NEXT:    vpcmpgtq %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
 ; AVX1-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vmovapd {{.*#+}} xmm3 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm3 = [1.2598673968951787E-321,1.2598673968951787E-321]
+; AVX1-NEXT:    # xmm3 = mem[0,0]
 ; AVX1-NEXT:    vandpd %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vandpd %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vandpd %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vandpd %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0

llvm/trunk/test/CodeGen/X86/vector-trunc.ll

 ; SSE41-NEXT:    packusdw %xmm2, %xmm0
 ; SSE41-NEXT:    packuswb %xmm0, %xmm0
 ; SSE41-NEXT:    movq %xmm0, (%rax)
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc8i64_8i8:
 ; AVX1:       # %bb.0: # %entry
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm3 = [1.2598673968951787E-321,1.2598673968951787E-321]
+; AVX1-NEXT:    # xmm3 = mem[0,0]
+; AVX1-NEXT:    vandpd %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vandpd %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vandpd %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vandpd %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm0, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovq %xmm0, (%rax)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: trunc8i64_8i8:

 ; SSE41-NEXT:    packusdw %xmm1, %xmm0
 ; SSE41-NEXT:    packuswb %xmm2, %xmm0
 ; SSE41-NEXT:    movdqu %xmm0, (%rax)
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc16i32_16i8:
 ; AVX1:       # %bb.0: # %entry
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm3 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
+; AVX1-NEXT:    vandps %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vandps %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vandps %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vandps %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackusdw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovdqu %xmm0, (%rax)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: trunc16i32_16i8:
 ; AVX2:       # %bb.0: # %entry

llvm/trunk/test/CodeGen/X86/vshift-6.ll

 ; X32-NEXT:    pand {{\.LCPI.*}}, %xmm0
 ; X32-NEXT:    por %xmm2, %xmm0
 ; X32-NEXT:    paddb %xmm1, %xmm1
 ; X32-NEXT:    pxor %xmm2, %xmm2
 ; X32-NEXT:    pcmpgtb %xmm1, %xmm2
 ; X32-NEXT:    movdqa %xmm2, %xmm4
 ; X32-NEXT:    pandn %xmm0, %xmm4
 ; X32-NEXT:    psllw $2, %xmm0
-; X32-NEXT:    pand {{\.LCPI.*}}, %xmm0
 ; X32-NEXT:    pand %xmm2, %xmm0
+; X32-NEXT:    pand {{\.LCPI.*}}, %xmm0
 ; X32-NEXT:    por %xmm4, %xmm0
 ; X32-NEXT:    paddb %xmm1, %xmm1
 ; X32-NEXT:    pcmpgtb %xmm1, %xmm3
 ; X32-NEXT:    movdqa %xmm3, %xmm1
 ; X32-NEXT:    pandn %xmm0, %xmm1
 ; X32-NEXT:    paddb %xmm0, %xmm0
 ; X32-NEXT:    pand %xmm3, %xmm0
 ; X32-NEXT:    por %xmm1, %xmm0

 ; X64-NEXT:    pand {{.*}}(%rip), %xmm0
 ; X64-NEXT:    por %xmm2, %xmm0
 ; X64-NEXT:    paddb %xmm1, %xmm1
 ; X64-NEXT:    pxor %xmm2, %xmm2
 ; X64-NEXT:    pcmpgtb %xmm1, %xmm2
 ; X64-NEXT:    movdqa %xmm2, %xmm4
 ; X64-NEXT:    pandn %xmm0, %xmm4
 ; X64-NEXT:    psllw $2, %xmm0
-; X64-NEXT:    pand {{.*}}(%rip), %xmm0
 ; X64-NEXT:    pand %xmm2, %xmm0
+; X64-NEXT:    pand {{.*}}(%rip), %xmm0
 ; X64-NEXT:    por %xmm4, %xmm0
 ; X64-NEXT:    paddb %xmm1, %xmm1
 ; X64-NEXT:    pcmpgtb %xmm1, %xmm3
 ; X64-NEXT:    movdqa %xmm3, %xmm1
 ; X64-NEXT:    pandn %xmm0, %xmm1
 ; X64-NEXT:    paddb %xmm0, %xmm0
 ; X64-NEXT:    pand %xmm3, %xmm0
 ; X64-NEXT:    por %xmm1, %xmm0
 ; X64-NEXT:    retq
 entry:
   store i8 %5, i8* %0
   %L5 = load i8, i8* %0
   %I8 = insertelement <16 x i8> <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>, i8 %L5, i32 7
   %B51 = shl <16 x i8> <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>, %I8
   ret <16 x i8> %B51
 }

llvm/trunk/test/CodeGen/X86/x86-interleaved-access.ll

 ; AVX1-NEXT:    vpalignr {{.*#+}} xmm3 = xmm3[11,12,13,14,15],xmm4[0,1,2,3,4,5,6,7,8,9,10]
 ; AVX1-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm8
 ; AVX1-NEXT:    vpalignr {{.*#+}} xmm4 = xmm4[11,12,13,14,15],xmm5[0,1,2,3,4,5,6,7,8,9,10]
 ; AVX1-NEXT:    vpalignr {{.*#+}} xmm1 = xmm1[11,12,13,14,15],xmm2[0,1,2,3,4,5,6,7,8,9,10]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm2
 ; AVX1-NEXT:    vpalignr {{.*#+}} xmm9 = xmm7[11,12,13,14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10]
 ; AVX1-NEXT:    vpalignr {{.*#+}} xmm4 = xmm6[11,12,13,14,15],xmm4[0,1,2,3,4,5,6,7,8,9,10]
-; AVX1-NEXT:    vmovaps {{.*#+}} ymm5 = [255,255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,255,255,255,255,255,255,255,255,255,255,255,0,0,0,0,0]
+; AVX1-NEXT:    vbroadcastf128 {{.*#+}} ymm5 = [255,255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,255,255,255,255,255,255,255,255,255,255,255,0,0,0,0,0]
+; AVX1-NEXT:    # ymm5 = mem[0,1,0,1]
 ; AVX1-NEXT:    vandnps %ymm2, %ymm5, %ymm2
 ; AVX1-NEXT:    vandps %ymm5, %ymm8, %ymm5
 ; AVX1-NEXT:    vorps %ymm2, %ymm5, %ymm2
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [128,128,128,128,128,128,11,12,13,14,15,128,128,128,128,128]
 ; AVX1-NEXT:    vpshufb %xmm5, %xmm3, %xmm3
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [5,6,7,8,9,10,128,128,128,128,128,0,1,2,3,4]
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm6, %xmm6
 ; AVX1-NEXT:    vpor %xmm3, %xmm6, %xmm3

 ; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm6, %ymm7
 ; AVX1-NEXT:    vpalignr {{.*#+}} xmm3 = xmm3[11,12,13,14,15],xmm12[0,1,2,3,4,5,6,7,8,9,10]
 ; AVX1-NEXT:    vpalignr {{.*#+}} xmm2 = xmm2[11,12,13,14,15],xmm11[0,1,2,3,4,5,6,7,8,9,10]
 ; AVX1-NEXT:    vpalignr {{.*#+}} xmm0 = xmm14[11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7,8,9,10]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm14
 ; AVX1-NEXT:    vpalignr {{.*#+}} xmm4 = xmm12[11,12,13,14,15],xmm4[0,1,2,3,4,5,6,7,8,9,10]
 ; AVX1-NEXT:    vpalignr {{.*#+}} xmm11 = xmm11[11,12,13,14,15],xmm13[0,1,2,3,4,5,6,7,8,9,10]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm11, %ymm12
-; AVX1-NEXT:    vmovaps {{.*#+}} ymm13 = [255,255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,255,255,255,255,255,255,255,255,255,255,255,0,0,0,0,0]
+; AVX1-NEXT:    vbroadcastf128 {{.*#+}} ymm13 = [255,255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,255,255,255,255,255,255,255,255,255,255,255,0,0,0,0,0]
+; AVX1-NEXT:    # ymm13 = mem[0,1,0,1]
 ; AVX1-NEXT:    vandnps %ymm12, %ymm13, %ymm12
 ; AVX1-NEXT:    vandps %ymm13, %ymm14, %ymm14
 ; AVX1-NEXT:    vorps %ymm12, %ymm14, %ymm12
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm14
 ; AVX1-NEXT:    vpalignr {{.*#+}} xmm4 = xmm15[11,12,13,14,15],xmm4[0,1,2,3,4,5,6,7,8,9,10]
 ; AVX1-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX1-NEXT:    vandnps %ymm14, %ymm13, %ymm14
 ; AVX1-NEXT:    vandps %ymm13, %ymm7, %ymm7