Index: lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- lib/Target/X86/X86ISelLowering.cpp
+++ lib/Target/X86/X86ISelLowering.cpp
@@ -28934,12 +28934,92 @@
                      EltNo);
 }
 
+// Try to match patterns such as
+// (i16 bitcast (v16i1 setcc v16i8 v1, v2, gt))
+// ->
+// (movmsk (v16i8 setcc v16i8 v1, v2, gt))
+// before the setcc result is scalarized on subtargets that don't have legal
+// vxi1 types.
+static SDValue combineBitcastOfSetCC(SelectionDAG &DAG, SDLoc DL, EVT VT,
+                                     SDValue SetCC, const X86Subtarget &Subtarget) {
+  if (!VT.isScalarInteger() || SetCC.getValueType().getScalarType() != MVT::i1)
+    return SDValue();
+  // With AVX512 vxi1 types are legal and we prefer using k-regs.
+  if (Subtarget.hasAVX512())
+    return SDValue();
+  SDValue N0 = SetCC->getOperand(0); // First operand the setcc compares
+  EVT InVT = N0.getValueType();      // Type of operands the setcc compares
+
+  // 128-bit PMOVMSK requires SSE2, 256-bit PMOVMSK requires AVX2.
+  // There are cases (see below) where we need to shuffle the result of the
+  // vector compare and it may be more profitable to bail-out if only SSE2.
+  // SSSE3 introduces PSHUFB which makes these shuffles cheaper.
+  switch (InVT.getSimpleVT().SimpleTy) {
+  default:
+    return SDValue();
+  case MVT::v16i8:
+  case MVT::v8i16:
+  case MVT::v4i32:
+  case MVT::v4f32:
+    if (!Subtarget.hasSSE2())
+      return SDValue();
+    break;
+  // For SSE2/v2x64 don't apply this combine because it is faster to extract
+  // the two elements and do a bitwise manipulation than shuffles+pmovmsk.
+  case MVT::v2i64:
+  case MVT::v2f64:
+    if (!Subtarget.hasSSSE3())
+      return SDValue();
+    break;
+  case MVT::v4i64:
+  case MVT::v8i32:
+  case MVT::v16i16:
+  case MVT::v32i8:
+  case MVT::v8f32:
+  case MVT::v4f64:
+    if (!Subtarget.hasAVX2())
+      return SDValue();
+    break;
+  };
+
+  SDValue V = DAG.getSetCC(
+      SDLoc(SetCC), InVT.changeVectorElementTypeToInteger(), N0,
+      SetCC->getOperand(1), cast<CondCodeSDNode>(SetCC->getOperand(2))->get());
+
+  // MOVMSK's operand must be a vector of i8's. So for types with larger
+  // elements, shuffle the upper element bytes to a consecutive sequence at the
+  // start of the vector.
+  // For example, t0 := (v4i32 setcc (v4i32 v1, v2, gt)) needs to be shuffled
+  // as:
+  // (v16i8 shuffle <0,4,8,12,u,u,...,u> (v16i8 bitcast t0), undef)
+  if (InVT != MVT::v16i8 && InVT != MVT::v32i8) {
+    unsigned Stride = InVT.getScalarSizeInBits() / 8;
+    MVT WideMaskVT = InVT.getSizeInBits() == 128 ? MVT::v16i8 : MVT::v32i8;
+    SmallVector<int, 32> Mask(WideMaskVT.getVectorNumElements(), -1);
+    for (unsigned i = 0, e = InVT.getVectorNumElements(); i != e; ++i)
+      Mask[i] = Stride * i;
+    V = DAG.getBitcast(WideMaskVT, V);
+    V = DAG.getVectorShuffle(WideMaskVT, DL, V, DAG.getUNDEF(WideMaskVT), Mask);
+  }
+  V = DAG.getNode(X86ISD::MOVMSK, DL, MVT::i32, V);
+  return DAG.getZExtOrTrunc(V, DL, VT);
+}
+
 static SDValue combineBitcast(SDNode *N, SelectionDAG &DAG,
-                              const X86Subtarget &Subtarget) {
+                              const X86Subtarget &Subtarget, bool LegalTypes) {
   SDValue N0 = N->getOperand(0);
   EVT VT = N->getValueType(0);
   EVT SrcVT = N0.getValueType();
 
+  // Try to match patterns such as
+  // (i16 bitcast (v16i1 setcc v16i8 v1, v2, gt))
+  // ->
+  // (movmsk (v16i8 setcc v16i8 v1, v2, gt))
+  // before the setcc result is scalarized on subtargets that don't have legal
+  // vxi1 types.
+  if (!LegalTypes && N0.getOpcode() == ISD::SETCC)
+    if (SDValue V = combineBitcastOfSetCC(DAG, SDLoc(N), VT, N0, Subtarget))
+      return V;
   // Since MMX types are special and don't usually play with other vector types,
   // it's better to handle them early to be sure we emit efficient code by
   // avoiding store-load conversions.
@@ -35044,7 +35124,7 @@
   case ISD::VSELECT:
   case ISD::SELECT:
   case X86ISD::SHRUNKBLEND: return combineSelect(N, DAG, DCI, Subtarget);
-  case ISD::BITCAST:        return combineBitcast(N, DAG, Subtarget);
+  case ISD::BITCAST:        return combineBitcast(N, DAG, Subtarget, !DCI.isBeforeLegalize());
   case X86ISD::CMOV:        return combineCMov(N, DAG, DCI, Subtarget);
   case ISD::ADD:            return combineAdd(N, DAG, Subtarget);
   case ISD::SUB:            return combineSub(N, DAG, Subtarget);
Index: test/CodeGen/X86/bitcast-setcc-128.ll
===================================================================
--- test/CodeGen/X86/bitcast-setcc-128.ll
+++ test/CodeGen/X86/bitcast-setcc-128.ll
@@ -8,91 +8,26 @@
 ; SSE2-LABEL: v8i16:
 ; SSE2:       ## BB#0:
 ; SSE2-NEXT:    pcmpgtw %xmm1, %xmm0
-; SSE2-NEXT:    pextrw $7, %xmm0, %eax
-; SSE2-NEXT:    andl $1, %eax
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    pextrw $6, %xmm0, %eax
-; SSE2-NEXT:    andl $1, %eax
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    pextrw $5, %xmm0, %eax
-; SSE2-NEXT:    andl $1, %eax
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    pextrw $4, %xmm0, %eax
-; SSE2-NEXT:    andl $1, %eax
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    pextrw $3, %xmm0, %eax
-; SSE2-NEXT:    andl $1, %eax
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    pextrw $2, %xmm0, %eax
-; SSE2-NEXT:    andl $1, %eax
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    pextrw $1, %xmm0, %eax
-; SSE2-NEXT:    andl $1, %eax
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    andl $1, %eax
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    packuswb %xmm0, %xmm0
+; SSE2-NEXT:    pmovmskb %xmm0, %eax
+; SSE2-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: v8i16:
 ; SSSE3:       ## BB#0:
 ; SSSE3-NEXT:    pcmpgtw %xmm1, %xmm0
-; SSSE3-NEXT:    pextrw $7, %xmm0, %eax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    pextrw $6, %xmm0, %eax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    pextrw $5, %xmm0, %eax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    pextrw $4, %xmm0, %eax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    pextrw $3, %xmm0, %eax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    pextrw $2, %xmm0, %eax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    pextrw $1, %xmm0, %eax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movd %xmm0, %eax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
+; SSSE3-NEXT:    pmovmskb %xmm0, %eax
+; SSSE3-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: v8i16:
 ; AVX1:       ## BB#0:
 ; AVX1-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrw $7, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrw $6, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrw $5, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrw $4, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrw $3, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrw $2, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrw $1, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vmovd %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX1-NEXT:    retq
 ;
 ; AVX512-LABEL: v8i16:
@@ -110,61 +45,27 @@
 ; SSE2-LABEL: v4i32:
 ; SSE2:       ## BB#0:
 ; SSE2-NEXT:    pcmpgtd %xmm1, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    andl $1, %eax
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    andl $1, %eax
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    andl $1, %eax
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    andl $1, %eax
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    packuswb %xmm0, %xmm0
+; SSE2-NEXT:    packuswb %xmm0, %xmm0
+; SSE2-NEXT:    pmovmskb %xmm0, %eax
+; SSE2-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: v4i32:
 ; SSSE3:       ## BB#0:
 ; SSSE3-NEXT:    pcmpgtd %xmm1, %xmm0
-; SSSE3-NEXT:    movd %xmm0, %eax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; SSSE3-NEXT:    movd %xmm1, %eax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSSE3-NEXT:    movd %xmm1, %eax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; SSSE3-NEXT:    movd %xmm0, %eax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; SSSE3-NEXT:    pmovmskb %xmm0, %eax
+; SSSE3-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: v4i32:
 ; AVX1:       ## BB#0:
 ; AVX1-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrd $3, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrd $2, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrd $1, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vmovd %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX1-NEXT:    retq
 ;
 ; AVX512-LABEL: v4i32:
@@ -183,63 +84,27 @@
 ; SSE2-LABEL: v4f32:
 ; SSE2:       ## BB#0:
 ; SSE2-NEXT:    cmpltps %xmm0, %xmm1
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    andl $1, %eax
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movaps %xmm1, %xmm0
-; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    andl $1, %eax
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    andl $1, %eax
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,1,2,3]
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    andl $1, %eax
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; SSE2-NEXT:    andps {{.*}}(%rip), %xmm1
+; SSE2-NEXT:    packuswb %xmm1, %xmm1
+; SSE2-NEXT:    packuswb %xmm1, %xmm1
+; SSE2-NEXT:    pmovmskb %xmm1, %eax
+; SSE2-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: v4f32:
 ; SSSE3:       ## BB#0:
 ; SSSE3-NEXT:    cmpltps %xmm0, %xmm1
-; SSSE3-NEXT:    movd %xmm1, %eax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movaps %xmm1, %xmm0
-; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; SSSE3-NEXT:    movd %xmm0, %eax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSSE3-NEXT:    movd %xmm0, %eax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,1,2,3]
-; SSSE3-NEXT:    movd %xmm1, %eax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; SSSE3-NEXT:    pmovmskb %xmm1, %eax
+; SSSE3-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: v4f32:
 ; AVX1:       ## BB#0:
 ; AVX1-NEXT:    vcmpltps %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vextractps $3, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vextractps $2, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vextractps $1, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vextractps $0, %xmm0, %eax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX1-NEXT:    retq
 ;
 ; AVX512-LABEL: v4f32:
@@ -258,165 +123,22 @@
 ; SSE2-LABEL: v16i8:
 ; SSE2:       ## BB#0:
 ; SSE2-NEXT:    pcmpgtb %xmm1, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-NEXT:    andb $1, %al
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-NEXT:    andb $1, %al
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-NEXT:    andb $1, %al
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-NEXT:    andb $1, %al
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-NEXT:    andb $1, %al
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-NEXT:    andb $1, %al
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-NEXT:    andb $1, %al
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-NEXT:    andb $1, %al
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-NEXT:    andb $1, %al
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-NEXT:    andb $1, %al
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-NEXT:    andb $1, %al
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-NEXT:    andb $1, %al
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-NEXT:    andb $1, %al
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-NEXT:    andb $1, %al
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSE2-NEXT:    movb -{{[0-9]+}}(%rsp), %cl
-; SSE2-NEXT:    andb $1, %cl
-; SSE2-NEXT:    movb %cl, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    andb $1, %al
-; SSE2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movzwl -{{[0-9]+}}(%rsp), %eax
+; SSE2-NEXT:    pmovmskb %xmm0, %eax
+; SSE2-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: v16i8:
 ; SSSE3:       ## BB#0:
 ; SSSE3-NEXT:    pcmpgtb %xmm1, %xmm0
-; SSSE3-NEXT:    movdqa %xmm0, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSSE3-NEXT:    andb $1, %al
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSSE3-NEXT:    andb $1, %al
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSSE3-NEXT:    andb $1, %al
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSSE3-NEXT:    andb $1, %al
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSSE3-NEXT:    andb $1, %al
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSSE3-NEXT:    andb $1, %al
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSSE3-NEXT:    andb $1, %al
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSSE3-NEXT:    andb $1, %al
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSSE3-NEXT:    andb $1, %al
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSSE3-NEXT:    andb $1, %al
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSSE3-NEXT:    andb $1, %al
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSSE3-NEXT:    andb $1, %al
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSSE3-NEXT:    andb $1, %al
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSSE3-NEXT:    andb $1, %al
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %cl
-; SSSE3-NEXT:    andb $1, %cl
-; SSSE3-NEXT:    movb %cl, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    andb $1, %al
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movzwl -{{[0-9]+}}(%rsp), %eax
+; SSSE3-NEXT:    pmovmskb %xmm0, %eax
+; SSSE3-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>
 ; SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: v16i8:
 ; AVX1:       ## BB#0:
 ; AVX1-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $15, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrb $14, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrb $13, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrb $12, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrb $11, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrb $10, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrb $9, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrb $8, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrb $7, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrb $6, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrb $5, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrb $4, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrb $3, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrb $2, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrb $1, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX1-NEXT:    andb $1, %al
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    movzwl -{{[0-9]+}}(%rsp), %eax
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>
 ; AVX1-NEXT:    retq
 ;
 ; AVX512-LABEL: v16i8:
@@ -467,26 +189,17 @@
 ; SSSE3-NEXT:    pand %xmm3, %xmm0
 ; SSSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
 ; SSSE3-NEXT:    por %xmm0, %xmm1
-; SSSE3-NEXT:    movq %xmm1, %rax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSSE3-NEXT:    movq %xmm0, %rax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u]
+; SSSE3-NEXT:    pmovmskb %xmm1, %eax
+; SSSE3-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: v2i64:
 ; AVX1:       ## BB#0:
 ; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vmovq %xmm0, %rax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX1-NEXT:    retq
 ;
 ; AVX512-LABEL: v2i64:
@@ -518,26 +231,17 @@
 ; SSSE3-LABEL: v2f64:
 ; SSSE3:       ## BB#0:
 ; SSSE3-NEXT:    cmpltpd %xmm0, %xmm1
-; SSSE3-NEXT:    movq %xmm1, %rax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSSE3-NEXT:    movq %xmm0, %rax
-; SSSE3-NEXT:    andl $1, %eax
-; SSSE3-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u]
+; SSSE3-NEXT:    pmovmskb %xmm1, %eax
+; SSSE3-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; SSSE3-NEXT:    retq
 ;
 ; AVX1-LABEL: v2f64:
 ; AVX1:       ## BB#0:
 ; AVX1-NEXT:    vcmpltpd %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpextrq $1, %xmm0, %rax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    vmovq %xmm0, %rax
-; AVX1-NEXT:    andl $1, %eax
-; AVX1-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX1-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX1-NEXT:    retq
 ;
 ; AVX512-LABEL: v2f64:
Index: test/CodeGen/X86/bitcast-setcc-256.ll
===================================================================
--- test/CodeGen/X86/bitcast-setcc-256.ll
+++ test/CodeGen/X86/bitcast-setcc-256.ll
@@ -7,56 +7,12 @@
 ; AVX2:       ## BB#0:
 ; AVX2-NEXT:    vpcmpgtw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $15, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrb $14, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrb $13, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrb $12, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrb $11, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrb $10, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrb $9, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrb $8, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrb $7, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrb $6, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrb $5, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrb $4, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrb $3, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrb $2, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrb $1, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    movzwl -{{[0-9]+}}(%rsp), %eax
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
+; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX2-NEXT:    vpmovmskb %ymm0, %eax
+; AVX2-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
@@ -77,32 +33,12 @@
 ; AVX2:       ## BB#0:
 ; AVX2-NEXT:    vpcmpgtd %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrw $7, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrw $6, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrw $5, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrw $4, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrw $3, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrw $2, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrw $1, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vmovd %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
+; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-NEXT:    vpmovmskb %ymm0, %eax
+; AVX2-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
@@ -123,32 +59,12 @@
 ; AVX2:       ## BB#0:
 ; AVX2-NEXT:    vcmpltps %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrw $7, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrw $6, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrw $5, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrw $4, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrw $3, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrw $2, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrw $1, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vmovd %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
+; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-NEXT:    vpmovmskb %ymm0, %eax
+; AVX2-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
@@ -167,117 +83,8 @@
 define i32 @v32i8(<32 x i8> %a, <32 x i8> %b) {
 ; AVX2-LABEL: v32i8:
 ; AVX2:       ## BB#0:
-; AVX2-NEXT:    pushq %rbp
-; AVX2-NEXT:  Lcfi0:
-; AVX2-NEXT:    .cfi_def_cfa_offset 16
-; AVX2-NEXT:  Lcfi1:
-; AVX2-NEXT:    .cfi_offset %rbp, -16
-; AVX2-NEXT:    movq %rsp, %rbp
-; AVX2-NEXT:  Lcfi2:
-; AVX2-NEXT:    .cfi_def_cfa_register %rbp
-; AVX2-NEXT:    andq $-32, %rsp
-; AVX2-NEXT:    subq $32, %rsp
 ; AVX2-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpextrb $15, %xmm1, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $14, %xmm1, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $13, %xmm1, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $12, %xmm1, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $11, %xmm1, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $10, %xmm1, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $9, %xmm1, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $8, %xmm1, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $7, %xmm1, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $6, %xmm1, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $5, %xmm1, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $4, %xmm1, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $3, %xmm1, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $2, %xmm1, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $1, %xmm1, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $0, %xmm1, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $15, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $14, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $13, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $12, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $11, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $10, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $9, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $8, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $7, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $6, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $5, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $4, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $3, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $2, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $1, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX2-NEXT:    andb $1, %al
-; AVX2-NEXT:    movb %al, (%rsp)
-; AVX2-NEXT:    movl (%rsp), %eax
-; AVX2-NEXT:    movq %rbp, %rsp
-; AVX2-NEXT:    popq %rbp
+; AVX2-NEXT:    vpmovmskb %ymm0, %eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
@@ -297,20 +104,12 @@
 ; AVX2:       ## BB#0:
 ; AVX2-NEXT:    vpcmpgtq %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrd $3, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrd $2, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrd $1, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vmovd %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
+; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-NEXT:    vpmovmskb %ymm0, %eax
+; AVX2-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
@@ -332,20 +131,12 @@
 ; AVX2:       ## BB#0:
 ; AVX2-NEXT:    vcmpltpd %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrd $3, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrd $2, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vpextrd $1, %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    vmovd %xmm0, %eax
-; AVX2-NEXT:    andl $1, %eax
-; AVX2-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
-; AVX2-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,8,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
+; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-NEXT:    vpmovmskb %ymm0, %eax
+; AVX2-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;