[X86][SSE] Recognise vXi1 boolean anyof/allof reduction patterns

[RKSimon]

Summary

Currently combineHorizontalPredicateResult only handles anyof/allof reduction patterns of legal types, which can be tricky to match as type legalization of bools can introduce bitcasts/truncs/extensions.

This patch extends to combineHorizontalPredicateResult to recognise vXi1 bool reductions as well and uses the existing combineBitcastvxi1 helper to create the MOVMSK necessary to then compare the signmask result.

This ensures the accuracy of the reduction costs added in D60403 which assume the MOVMSK generation.

[spatel]

LGTM - see inline for a minor potential follow-up.

If we can assume vXi1 IR, then I can probably abandon my recent movmsk efforts:
D59669
D59912

test/CodeGen/X86/vector-compare-all_of.ll

1295–1296

I think the other path prefers to use packss and a 128-bit movmsk for AVX1 here which could be a slight win since it avoids ymm?

[RKSimon]

In D60610#1464251, @spatel wrote:

LGTM - see inline for a minor potential follow-up.

If we can assume vXi1 IR, then I can probably abandon my recent movmsk efforts:
D59669
D59912

Let's keep those open for now - both still improve code beyond the vector reductions which this patch handles.

test/CodeGen/X86/vector-compare-all_of.ll

1295–1296

Yes a single packss to a movmskps is better - I think we can take some of the code from D59912 to tweak this.

lib/Target/X86/X86ISelLowering.cpp

static SDValue combineHorizontalPredicateResult(SDNode *Extract,
                                                 const X86Subtarget &Subtarget) {
   // Bail without SSE2 or with AVX512VL (which uses predicate registers).
   if (!Subtarget.hasSSE2() || Subtarget.hasVLX())
     return SDValue();
 
   EVT ExtractVT = Extract->getValueType(0);
   unsigned BitWidth = ExtractVT.getSizeInBits();
   if (ExtractVT != MVT::i64 && ExtractVT != MVT::i32 && ExtractVT != MVT::i16 &&
-      ExtractVT != MVT::i8)
+      ExtractVT != MVT::i8 && ExtractVT != MVT::i1)
     return SDValue();
 
   // Check for OR(any_of) and AND(all_of) horizontal reduction patterns.
   ISD::NodeType BinOp;
   SDValue Match = DAG.matchBinOpReduction(Extract, BinOp, {ISD::OR, ISD::AND});
   if (!Match)
     return SDValue();
 
   // EXTRACT_VECTOR_ELT can require implicit extension of the vector element
   // which we can't support here for now.
   if (Match.getScalarValueSizeInBits() != BitWidth)
     return SDValue();
 
+  SDValue Movmsk;
+  SDLoc DL(Extract);
+  unsigned NumElts = Match.getValueType().getVectorNumElements();
+
+  if (ExtractVT == MVT::i1) {
+    // Special case for (pre-legalization) vXi1 reductions.
+    // Use combineBitcastvxi1 to create the MOVMSK.
+    if (NumElts > 32)
+      return SDValue();
+    if (NumElts == 32 && !Subtarget.hasInt256()) {
+      SDValue Lo, Hi;
+      std::tie(Lo, Hi) = DAG.SplitVector(Match, DL);
+      Match = DAG.getNode(BinOp, DL, Lo.getValueType(), Lo, Hi);
+      NumElts = 16;
+    }
+    EVT MovmskVT = EVT::getIntegerVT(*DAG.getContext(), NumElts);
+    Movmsk = combineBitcastvxi1(DAG, MovmskVT, Match, DL, Subtarget);
+    if (!Movmsk)
+      return SDValue();
+    Movmsk = DAG.getZExtOrTrunc(Movmsk, DL, MVT::i32);
+  } else {
     unsigned MatchSizeInBits = Match.getValueSizeInBits();
-  if (!(MatchSizeInBits == 128 || (MatchSizeInBits == 256 && Subtarget.hasAVX())))
+    if (!(MatchSizeInBits == 128 ||
+          (MatchSizeInBits == 256 && Subtarget.hasAVX())))
       return SDValue();
 
-  // Make sure this isn't a vector of 1 element. The perf win from using MOVMSK
-  // diminishes with less elements in the reduction, but it is generally better
-  // to get the comparison over to the GPRs as soon as possible to reduce the
-  // number of vector ops.
+    // Make sure this isn't a vector of 1 element. The perf win from using
+    // MOVMSK diminishes with less elements in the reduction, but it is
+    // generally better to get the comparison over to the GPRs as soon as
+    // possible to reduce the number of vector ops.
     if (Match.getValueType().getVectorNumElements() < 2)
       return SDValue();
 
     // Check that we are extracting a reduction of all sign bits.
     if (DAG.ComputeNumSignBits(Match) != BitWidth)
       return SDValue();
 
-  SDLoc DL(Extract);
     if (MatchSizeInBits == 256 && BitWidth < 32 && !Subtarget.hasInt256()) {
       SDValue Lo, Hi;
       std::tie(Lo, Hi) = DAG.SplitVector(Match, DL);
       Match = DAG.getNode(BinOp, DL, Lo.getValueType(), Lo, Hi);
       MatchSizeInBits = Match.getValueSizeInBits();
     }
 
     // For 32/64 bit comparisons use MOVMSKPS/MOVMSKPD, else PMOVMSKB.
     MVT MaskSrcVT;
     if (64 == BitWidth || 32 == BitWidth)
       MaskSrcVT = MVT::getVectorVT(MVT::getFloatingPointVT(BitWidth),
                                    MatchSizeInBits / BitWidth);
     else
       MaskSrcVT = MVT::getVectorVT(MVT::i8, MatchSizeInBits / 8);
 
+    SDValue BitcastLogicOp = DAG.getBitcast(MaskSrcVT, Match);
+    Movmsk = getPMOVMSKB(DL, BitcastLogicOp, DAG, Subtarget);
+    NumElts = MaskSrcVT.getVectorNumElements();
+  }
+  assert(NumElts <= 32 && "Not expecting more than 32 elements");
+
   SDValue CmpC;
   ISD::CondCode CondCode;
   if (BinOp == ISD::OR) {
     // any_of -> MOVMSK != 0
     CmpC = DAG.getConstant(0, DL, MVT::i32);
     CondCode = ISD::CondCode::SETNE;
   } else {
     // all_of -> MOVMSK == ((1 << NumElts) - 1)
-    uint64_t NumElts = MaskSrcVT.getVectorNumElements();
-    assert(NumElts <= 32 && "Not expecting more than 32 elements");
     CmpC = DAG.getConstant((1ULL << NumElts) - 1, DL, MVT::i32);
     CondCode = ISD::CondCode::SETEQ;
   }
 
   // The setcc produces an i8 of 0/1, so extend that to the result width and
   // negate to get the final 0/-1 mask value.
   const TargetLowering &TLI = DAG.getTargetLoweringInfo();
-  SDValue BitcastLogicOp = DAG.getBitcast(MaskSrcVT, Match);
-  SDValue Movmsk = getPMOVMSKB(DL, BitcastLogicOp, DAG, Subtarget);
-  EVT SetccVT = TLI.getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(),
-                                       MVT::i32);
+  EVT SetccVT =
+      TLI.getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), MVT::i32);
   SDValue Setcc = DAG.getSetCC(DL, SetccVT, Movmsk, CmpC, CondCode);
   SDValue Zext = DAG.getZExtOrTrunc(Setcc, DL, ExtractVT);
   SDValue Zero = DAG.getConstant(0, DL, ExtractVT);
   return DAG.getNode(ISD::SUB, DL, ExtractVT, Zero, Zext);
 }
 
 static SDValue combineBasicSADPattern(SDNode *Extract, SelectionDAG &DAG,
                                       const X86Subtarget &Subtarget) {

test/CodeGen/X86/vector-compare-all_of.ll

; AVX512-NEXT:    retq
   %11 = extractelement <32 x i8> %10, i32 0
   ret i8 %11
 }
 
 define i1 @bool_reduction_v2f64(<2 x double> %x, <2 x double> %y) {
 ; SSE-LABEL: bool_reduction_v2f64:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    cmpltpd %xmm0, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
-; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    movmskpd %xmm1, %eax
+; SSE-NEXT:    cmpb $3, %al
+; SSE-NEXT:    sete %al
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: bool_reduction_v2f64:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vcmpltpd %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; AVX-NEXT:    vandpd %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX-NEXT:    # kill: def $al killed $al killed $eax
+; AVX-NEXT:    vmovmskpd %xmm0, %eax
+; AVX-NEXT:    cmpb $3, %al
+; AVX-NEXT:    sete %al
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v2f64:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vcmpltpd %xmm0, %xmm1, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z}
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpsllq $63, %xmm0, %xmm0
 ; AVX512-NEXT:    vptestmq %xmm0, %xmm0, %k0 {%k1}
 ; AVX512-NEXT:    kmovd %k0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %a = fcmp ogt <2 x double> %x, %y
   %b = shufflevector <2 x i1> %a, <2 x i1> undef, <2 x i32> <i32 1, i32 undef>
   %c = and <2 x i1> %a, %b
   %d = extractelement <2 x i1> %c, i32 0
   ret i1 %d
 }
 
 define i1 @bool_reduction_v4f32(<4 x float> %x, <4 x float> %y) {
 ; SSE-LABEL: bool_reduction_v4f32:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    cmpeqps %xmm1, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
-; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    movmskps %xmm0, %eax
+; SSE-NEXT:    cmpb $15, %al
+; SSE-NEXT:    sete %al
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: bool_reduction_v4f32:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vcmpeqps %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; AVX-NEXT:    vandpd %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpermilps {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vandpd %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX-NEXT:    # kill: def $al killed $al killed $eax
+; AVX-NEXT:    vmovmskps %xmm0, %eax
+; AVX-NEXT:    cmpb $15, %al
+; AVX-NEXT:    sete %al
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v4f32:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
; AVX512-NEXT:    retq
   ret i1 %d
 }
 
 define i1 @bool_reduction_v4f64(<4 x double> %x, <4 x double> %y) {
 ; SSE-LABEL: bool_reduction_v4f64:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    cmplepd %xmm1, %xmm3
 ; SSE-NEXT:    cmplepd %xmm0, %xmm2
-; SSE-NEXT:    movapd %xmm2, %xmm0
-; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm3[0,2]
-; SSE-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,2],xmm2[0,2]
-; SSE-NEXT:    andps %xmm0, %xmm3
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,1,2,3]
-; SSE-NEXT:    pand %xmm3, %xmm0
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
-; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    packssdw %xmm3, %xmm2
+; SSE-NEXT:    movmskps %xmm2, %eax
+; SSE-NEXT:    cmpb $15, %al
+; SSE-NEXT:    sete %al
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: bool_reduction_v4f64:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vcmplepd %ymm0, %ymm1, %ymm0
-; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX-NEXT:    vpackssdw %xmm1, %xmm0, %xmm2
-; AVX-NEXT:    vpackssdw %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpand %xmm2, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX-NEXT:    # kill: def $al killed $al killed $eax
+; AVX-NEXT:    vmovmskpd %ymm0, %eax
+; AVX-NEXT:    cmpb $15, %al
+; AVX-NEXT:    sete %al
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v4f64:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vcmplepd %ymm0, %ymm1, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}

 
 define i1 @bool_reduction_v2i64(<2 x i64> %x, <2 x i64> %y) {
 ; SSE-LABEL: bool_reduction_v2i64:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
 ; SSE-NEXT:    pxor %xmm2, %xmm1
 ; SSE-NEXT:    pxor %xmm2, %xmm0
 ; SSE-NEXT:    pcmpgtq %xmm1, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    pextrb $0, %xmm1, %eax
-; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    movmskpd %xmm0, %eax
+; SSE-NEXT:    cmpb $3, %al
+; SSE-NEXT:    sete %al
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: bool_reduction_v2i64:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
 ; AVX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
 ; AVX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
 ; AVX-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX-NEXT:    # kill: def $al killed $al killed $eax
+; AVX-NEXT:    vmovmskpd %xmm0, %eax
+; AVX-NEXT:    cmpb $3, %al
+; AVX-NEXT:    sete %al
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v2i64:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpnleuq %xmm1, %xmm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z}
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]

 }
 
 define i1 @bool_reduction_v4i32(<4 x i32> %x, <4 x i32> %y) {
 ; SSE-LABEL: bool_reduction_v4i32:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pcmpeqd %xmm1, %xmm0
 ; SSE-NEXT:    pcmpeqd %xmm1, %xmm1
 ; SSE-NEXT:    pxor %xmm0, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; SSE-NEXT:    pandn %xmm1, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    pextrb $0, %xmm1, %eax
-; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    movmskps %xmm1, %eax
+; SSE-NEXT:    cmpb $15, %al
+; SSE-NEXT:    sete %al
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: bool_reduction_v4i32:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm1
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX-NEXT:    vpandn %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX-NEXT:    # kill: def $al killed $al killed $eax
+; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vmovmskps %xmm0, %eax
+; AVX-NEXT:    cmpb $15, %al
+; AVX-NEXT:    sete %al
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v4i32:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpneqd %xmm1, %xmm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
; AVX512-NEXT:    retq
   %d = extractelement <4 x i1> %c, i32 0
   ret i1 %d
 }
 
 define i1 @bool_reduction_v8i16(<8 x i16> %x, <8 x i16> %y) {
 ; SSE-LABEL: bool_reduction_v8i16:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pcmpgtw %xmm0, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    psrld $16, %xmm0
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
-; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    packsswb %xmm0, %xmm1
+; SSE-NEXT:    pmovmskb %xmm1, %eax
+; SSE-NEXT:    cmpb $-1, %al
+; SSE-NEXT:    sete %al
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: bool_reduction_v8i16:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX-NEXT:    # kill: def $al killed $al killed $eax
+; AVX-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
+; AVX-NEXT:    vpmovmskb %xmm0, %eax
+; AVX-NEXT:    cmpb $-1, %al
+; AVX-NEXT:    sete %al
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v8i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpgtw %xmm0, %xmm1, %k1
 ; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
 ; AVX512-NEXT:    vextracti128 $1, %ymm1, %xmm1
; AVX512-NEXT:    retq
   ret i1 %e
 }
 
 define i1 @bool_reduction_v16i8(<16 x i8> %x, <16 x i8> %y) {
 ; SSE-LABEL: bool_reduction_v16i8:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pcmpgtb %xmm1, %xmm0
 ; SSE-NEXT:    pmovmskb %xmm0, %eax
-; SSE-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; SSE-NEXT:    cmpw $-1, %ax
 ; SSE-NEXT:    sete %al
-; SSE-NEXT:    negb %al
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: bool_reduction_v16i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpmovmskb %xmm0, %eax
-; AVX-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; AVX-NEXT:    cmpw $-1, %ax
 ; AVX-NEXT:    sete %al
-; AVX-NEXT:    negb %al
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v16i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0
 ; AVX512-NEXT:    kshiftrw $8, %k0, %k1
 ; AVX512-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0 {%k1}
 ; AVX512-NEXT:    kshiftrw $4, %k0, %k1
; AVX512-NEXT:    retq
   ret i1 %f
 }
 
 define i1 @bool_reduction_v4i64(<4 x i64> %x, <4 x i64> %y) {
 ; SSE-LABEL: bool_reduction_v4i64:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pcmpgtq %xmm1, %xmm3
 ; SSE-NEXT:    pcmpgtq %xmm0, %xmm2
-; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2],xmm3[0,2]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
-; SSE-NEXT:    pand %xmm2, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    pextrb $0, %xmm1, %eax
-; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    packssdw %xmm3, %xmm2
+; SSE-NEXT:    movmskps %xmm2, %eax
+; SSE-NEXT:    cmpb $15, %al
+; SSE-NEXT:    sete %al
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: bool_reduction_v4i64:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm1
-; AVX1-NEXT:    vpackssdw %xmm0, %xmm2, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX1-NEXT:    # kill: def $al killed $al killed $eax
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT:    vmovmskpd %ymm0, %eax
+; AVX1-NEXT:    cmpb $15, %al
+; AVX1-NEXT:    sete %al
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: bool_reduction_v4i64:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpcmpgtq %ymm0, %ymm1, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm2
-; AVX2-NEXT:    vpackssdw %xmm0, %xmm1, %xmm0
-; AVX2-NEXT:    vpand %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX2-NEXT:    # kill: def $al killed $al killed $eax
+; AVX2-NEXT:    vmovmskpd %ymm0, %eax
+; AVX2-NEXT:    cmpb $15, %al
+; AVX2-NEXT:    sete %al
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v4i64:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpgtq %ymm0, %ymm1, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}

 define i1 @bool_reduction_v8i32(<8 x i32> %x, <8 x i32> %y) {
 ; SSE-LABEL: bool_reduction_v8i32:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pminud %xmm1, %xmm3
 ; SSE-NEXT:    pcmpeqd %xmm1, %xmm3
 ; SSE-NEXT:    pminud %xmm0, %xmm2
 ; SSE-NEXT:    pcmpeqd %xmm0, %xmm2
 ; SSE-NEXT:    packssdw %xmm3, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
-; SSE-NEXT:    pand %xmm2, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    psrld $16, %xmm0
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
-; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    packsswb %xmm0, %xmm2
+; SSE-NEXT:    pmovmskb %xmm2, %eax
+; SSE-NEXT:    cmpb $-1, %al
+; SSE-NEXT:    sete %al
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: bool_reduction_v8i32:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpminud %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpeqd %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpminud %xmm1, %xmm0, %xmm1
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX1-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX1-NEXT:    # kill: def $al killed $al killed $eax
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT:    vmovmskps %ymm0, %eax
+; AVX1-NEXT:    cmpb $-1, %al
+; AVX1-NEXT:    sete %al
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: bool_reduction_v8i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpminud %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vpcmpeqd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX2-NEXT:    vpand %xmm0, %xmm1, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX2-NEXT:    # kill: def $al killed $al killed $eax
+; AVX2-NEXT:    vmovmskps %ymm0, %eax
+; AVX2-NEXT:    cmpb $-1, %al
+; AVX2-NEXT:    sete %al
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v8i32:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpleud %ymm1, %ymm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}

 
 define i1 @bool_reduction_v16i16(<16 x i16> %x, <16 x i16> %y) {
 ; SSE-LABEL: bool_reduction_v16i16:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pcmpeqw %xmm3, %xmm1
 ; SSE-NEXT:    pcmpeqw %xmm2, %xmm0
 ; SSE-NEXT:    packsswb %xmm1, %xmm0
 ; SSE-NEXT:    pmovmskb %xmm0, %eax
-; SSE-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; SSE-NEXT:    cmpw $-1, %ax
 ; SSE-NEXT:    sete %al
-; SSE-NEXT:    negb %al
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: bool_reduction_v16i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpeqw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpacksswb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %eax
-; AVX1-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; AVX1-NEXT:    cmpw $-1, %ax
 ; AVX1-NEXT:    sete %al
-; AVX1-NEXT:    negb %al
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: bool_reduction_v16i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpmovmskb %xmm0, %eax
-; AVX2-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; AVX2-NEXT:    cmpw $-1, %ax
 ; AVX2-NEXT:    sete %al
-; AVX2-NEXT:    negb %al
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v16i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpeqw %ymm1, %ymm0, %k0
 ; AVX512-NEXT:    kshiftrw $8, %k0, %k1
 ; AVX512-NEXT:    vpcmpeqw %ymm1, %ymm0, %k0 {%k1}
; AVX512-NEXT:    retq
   %e = and <16 x i1> %s4, %d
   %f = extractelement <16 x i1> %e, i32 0
   ret i1 %f
 }
 
 define i1 @bool_reduction_v32i8(<32 x i8> %x, <32 x i8> %y) {
 ; SSE-LABEL: bool_reduction_v32i8:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    pcmpeqb %xmm2, %xmm0
 ; SSE-NEXT:    pcmpeqb %xmm3, %xmm1
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    pmovmskb %xmm1, %eax
-; SSE-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; SSE-NEXT:    pcmpeqb %xmm2, %xmm0
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    pmovmskb %xmm0, %eax
+; SSE-NEXT:    cmpw $-1, %ax
 ; SSE-NEXT:    sete %al
-; SSE-NEXT:    negb %al
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: bool_reduction_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpeqb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %eax
-; AVX1-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; AVX1-NEXT:    cmpw $-1, %ax
 ; AVX1-NEXT:    sete %al
-; AVX1-NEXT:    negb %al
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: bool_reduction_v32i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmovmskb %ymm0, %eax
 ; AVX2-NEXT:    cmpl $-1, %eax
 ; AVX2-NEXT:    sete %al
-; AVX2-NEXT:    negb %al
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpeqb %ymm1, %ymm0, %k0
 ; AVX512-NEXT:    kshiftrd $16, %k0, %k1
 ; AVX512-NEXT:    vpcmpeqb %ymm1, %ymm0, %k0 {%k1}

test/CodeGen/X86/vector-compare-any_of.ll

; AVX512-NEXT:    retq
   %11 = extractelement <32 x i8> %10, i32 0
   ret i8 %11
 }
 
 define i1 @bool_reduction_v2f64(<2 x double> %x, <2 x double> %y) {
 ; SSE-LABEL: bool_reduction_v2f64:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    cmpltpd %xmm0, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
-; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    movmskpd %xmm1, %eax
+; SSE-NEXT:    testb %al, %al
+; SSE-NEXT:    setne %al
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: bool_reduction_v2f64:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vcmpltpd %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; AVX-NEXT:    vorpd %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX-NEXT:    # kill: def $al killed $al killed $eax
+; AVX-NEXT:    vmovmskpd %xmm0, %eax
+; AVX-NEXT:    testb %al, %al
+; AVX-NEXT:    setne %al
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v2f64:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vcmpltpd %xmm0, %xmm1, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z}
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
; AVX512-NEXT:    retq
   %d = extractelement <2 x i1> %c, i32 0
   ret i1 %d
 }
 
 define i1 @bool_reduction_v4f32(<4 x float> %x, <4 x float> %y) {
 ; SSE-LABEL: bool_reduction_v4f32:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    cmpeqps %xmm1, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
-; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    movmskps %xmm0, %eax
+; SSE-NEXT:    testb %al, %al
+; SSE-NEXT:    setne %al
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: bool_reduction_v4f32:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vcmpeqps %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; AVX-NEXT:    vorpd %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpermilps {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vorpd %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX-NEXT:    # kill: def $al killed $al killed $eax
+; AVX-NEXT:    vmovmskps %xmm0, %eax
+; AVX-NEXT:    testb %al, %al
+; AVX-NEXT:    setne %al
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v4f32:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
; AVX512-NEXT:    retq
   ret i1 %d
 }
 
 define i1 @bool_reduction_v4f64(<4 x double> %x, <4 x double> %y) {
 ; SSE-LABEL: bool_reduction_v4f64:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    cmplepd %xmm1, %xmm3
 ; SSE-NEXT:    cmplepd %xmm0, %xmm2
-; SSE-NEXT:    movapd %xmm2, %xmm0
-; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm3[0,2]
-; SSE-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,2],xmm2[0,2]
-; SSE-NEXT:    orps %xmm0, %xmm3
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,1,2,3]
-; SSE-NEXT:    por %xmm3, %xmm0
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
-; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    packssdw %xmm3, %xmm2
+; SSE-NEXT:    movmskps %xmm2, %eax
+; SSE-NEXT:    testb %al, %al
+; SSE-NEXT:    setne %al
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: bool_reduction_v4f64:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vcmplepd %ymm0, %ymm1, %ymm0
-; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX-NEXT:    vpackssdw %xmm1, %xmm0, %xmm2
-; AVX-NEXT:    vpackssdw %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpor %xmm2, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX-NEXT:    # kill: def $al killed $al killed $eax
+; AVX-NEXT:    vmovmskpd %ymm0, %eax
+; AVX-NEXT:    testb %al, %al
+; AVX-NEXT:    setne %al
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v4f64:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vcmplepd %ymm0, %ymm1, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}

 
 define i1 @bool_reduction_v2i64(<2 x i64> %x, <2 x i64> %y) {
 ; SSE-LABEL: bool_reduction_v2i64:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
 ; SSE-NEXT:    pxor %xmm2, %xmm1
 ; SSE-NEXT:    pxor %xmm2, %xmm0
 ; SSE-NEXT:    pcmpgtq %xmm1, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    pextrb $0, %xmm1, %eax
-; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    movmskpd %xmm0, %eax
+; SSE-NEXT:    testb %al, %al
+; SSE-NEXT:    setne %al
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: bool_reduction_v2i64:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
 ; AVX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
 ; AVX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
 ; AVX-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX-NEXT:    # kill: def $al killed $al killed $eax
+; AVX-NEXT:    vmovmskpd %xmm0, %eax
+; AVX-NEXT:    testb %al, %al
+; AVX-NEXT:    setne %al
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v2i64:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpnleuq %xmm1, %xmm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z}
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]

 }
 
 define i1 @bool_reduction_v4i32(<4 x i32> %x, <4 x i32> %y) {
 ; SSE-LABEL: bool_reduction_v4i32:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pcmpeqd %xmm1, %xmm0
 ; SSE-NEXT:    pcmpeqd %xmm1, %xmm1
 ; SSE-NEXT:    pxor %xmm0, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    pextrb $0, %xmm1, %eax
-; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    movmskps %xmm1, %eax
+; SSE-NEXT:    testb %al, %al
+; SSE-NEXT:    setne %al
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: bool_reduction_v4i32:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX-NEXT:    # kill: def $al killed $al killed $eax
+; AVX-NEXT:    vmovmskps %xmm0, %eax
+; AVX-NEXT:    testb %al, %al
+; AVX-NEXT:    setne %al
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v4i32:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpneqd %xmm1, %xmm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
; AVX512-NEXT:    retq
   %d = extractelement <4 x i1> %c, i32 0
   ret i1 %d
 }
 
 define i1 @bool_reduction_v8i16(<8 x i16> %x, <8 x i16> %y) {
 ; SSE-LABEL: bool_reduction_v8i16:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pcmpgtw %xmm0, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    psrld $16, %xmm0
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
-; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    packsswb %xmm0, %xmm1
+; SSE-NEXT:    pmovmskb %xmm1, %eax
+; SSE-NEXT:    testb %al, %al
+; SSE-NEXT:    setne %al
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: bool_reduction_v8i16:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX-NEXT:    # kill: def $al killed $al killed $eax
+; AVX-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
+; AVX-NEXT:    vpmovmskb %xmm0, %eax
+; AVX-NEXT:    testb %al, %al
+; AVX-NEXT:    setne %al
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v8i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpgtw %xmm0, %xmm1, %k1
 ; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
 ; AVX512-NEXT:    vextracti128 $1, %ymm1, %xmm1
; AVX512-NEXT:    retq
   ret i1 %e
 }
 
 define i1 @bool_reduction_v16i8(<16 x i8> %x, <16 x i8> %y) {
 ; SSE-LABEL: bool_reduction_v16i8:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pcmpgtb %xmm1, %xmm0
 ; SSE-NEXT:    pmovmskb %xmm0, %eax
-; SSE-NEXT:    negl %eax
-; SSE-NEXT:    sbbb %al, %al
+; SSE-NEXT:    testw %ax, %ax
+; SSE-NEXT:    setne %al
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: bool_reduction_v16i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpmovmskb %xmm0, %eax
-; AVX-NEXT:    negl %eax
-; AVX-NEXT:    sbbb %al, %al
+; AVX-NEXT:    testw %ax, %ax
+; AVX-NEXT:    setne %al
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v16i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0
 ; AVX512-NEXT:    kshiftrw $8, %k0, %k1
 ; AVX512-NEXT:    korw %k0, %k1, %k0
 ; AVX512-NEXT:    kshiftrw $4, %k0, %k1
; AVX512-NEXT:    retq
   ret i1 %f
 }
 
 define i1 @bool_reduction_v4i64(<4 x i64> %x, <4 x i64> %y) {
 ; SSE-LABEL: bool_reduction_v4i64:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pcmpgtq %xmm1, %xmm3
 ; SSE-NEXT:    pcmpgtq %xmm0, %xmm2
-; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2],xmm3[0,2]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
-; SSE-NEXT:    por %xmm2, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    pextrb $0, %xmm1, %eax
-; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    packssdw %xmm3, %xmm2
+; SSE-NEXT:    movmskps %xmm2, %eax
+; SSE-NEXT:    testb %al, %al
+; SSE-NEXT:    setne %al
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: bool_reduction_v4i64:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm1
-; AVX1-NEXT:    vpackssdw %xmm0, %xmm2, %xmm0
-; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX1-NEXT:    # kill: def $al killed $al killed $eax
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT:    vmovmskpd %ymm0, %eax
+; AVX1-NEXT:    testb %al, %al
+; AVX1-NEXT:    setne %al
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: bool_reduction_v4i64:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpcmpgtq %ymm0, %ymm1, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm2
-; AVX2-NEXT:    vpackssdw %xmm0, %xmm1, %xmm0
-; AVX2-NEXT:    vpor %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX2-NEXT:    # kill: def $al killed $al killed $eax
+; AVX2-NEXT:    vmovmskpd %ymm0, %eax
+; AVX2-NEXT:    testb %al, %al
+; AVX2-NEXT:    setne %al
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v4i64:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpgtq %ymm0, %ymm1, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}

 define i1 @bool_reduction_v8i32(<8 x i32> %x, <8 x i32> %y) {
 ; SSE-LABEL: bool_reduction_v8i32:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pminud %xmm1, %xmm3
 ; SSE-NEXT:    pcmpeqd %xmm1, %xmm3
 ; SSE-NEXT:    pminud %xmm0, %xmm2
 ; SSE-NEXT:    pcmpeqd %xmm0, %xmm2
 ; SSE-NEXT:    packssdw %xmm3, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
-; SSE-NEXT:    por %xmm2, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    psrld $16, %xmm0
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
-; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    packsswb %xmm0, %xmm2
+; SSE-NEXT:    pmovmskb %xmm2, %eax
+; SSE-NEXT:    testb %al, %al
+; SSE-NEXT:    setne %al
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: bool_reduction_v8i32:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpminud %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpeqd %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpminud %xmm1, %xmm0, %xmm1
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX1-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX1-NEXT:    # kill: def $al killed $al killed $eax
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT:    vmovmskps %ymm0, %eax
+; AVX1-NEXT:    testb %al, %al
+; AVX1-NEXT:    setne %al
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: bool_reduction_v8i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpminud %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vpcmpeqd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX2-NEXT:    # kill: def $al killed $al killed $eax
+; AVX2-NEXT:    vmovmskps %ymm0, %eax
+; AVX2-NEXT:    testb %al, %al
+; AVX2-NEXT:    setne %al
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v8i32:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpleud %ymm1, %ymm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}

 
 define i1 @bool_reduction_v16i16(<16 x i16> %x, <16 x i16> %y) {
 ; SSE-LABEL: bool_reduction_v16i16:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pcmpeqw %xmm3, %xmm1
 ; SSE-NEXT:    pcmpeqw %xmm2, %xmm0
 ; SSE-NEXT:    packsswb %xmm1, %xmm0
 ; SSE-NEXT:    pmovmskb %xmm0, %eax
-; SSE-NEXT:    negl %eax
-; SSE-NEXT:    sbbb %al, %al
+; SSE-NEXT:    testw %ax, %ax
+; SSE-NEXT:    setne %al
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: bool_reduction_v16i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpeqw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpacksswb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %eax
-; AVX1-NEXT:    negl %eax
-; AVX1-NEXT:    sbbb %al, %al
+; AVX1-NEXT:    testw %ax, %ax
+; AVX1-NEXT:    setne %al
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: bool_reduction_v16i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpmovmskb %xmm0, %eax
-; AVX2-NEXT:    negl %eax
-; AVX2-NEXT:    sbbb %al, %al
+; AVX2-NEXT:    testw %ax, %ax
+; AVX2-NEXT:    setne %al
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v16i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpeqw %ymm1, %ymm0, %k0
 ; AVX512-NEXT:    kshiftrw $8, %k0, %k1
 ; AVX512-NEXT:    korw %k0, %k1, %k0
; AVX512-NEXT:    retq
   %e = or <16 x i1> %s4, %d
   %f = extractelement <16 x i1> %e, i32 0
   ret i1 %f
 }
 
 define i1 @bool_reduction_v32i8(<32 x i8> %x, <32 x i8> %y) {
 ; SSE-LABEL: bool_reduction_v32i8:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    pcmpeqb %xmm2, %xmm0
 ; SSE-NEXT:    pcmpeqb %xmm3, %xmm1
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    pmovmskb %xmm1, %eax
-; SSE-NEXT:    negl %eax
-; SSE-NEXT:    sbbb %al, %al
+; SSE-NEXT:    pcmpeqb %xmm2, %xmm0
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    pmovmskb %xmm0, %eax
+; SSE-NEXT:    testw %ax, %ax
+; SSE-NEXT:    setne %al
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: bool_reduction_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpeqb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %eax
-; AVX1-NEXT:    negl %eax
-; AVX1-NEXT:    sbbb %al, %al
+; AVX1-NEXT:    testw %ax, %ax
+; AVX1-NEXT:    setne %al
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: bool_reduction_v32i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmovmskb %ymm0, %eax
-; AVX2-NEXT:    negl %eax
-; AVX2-NEXT:    sbbb %al, %al
+; AVX2-NEXT:    testl %eax, %eax
+; AVX2-NEXT:    setne %al
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: bool_reduction_v32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpeqb %ymm1, %ymm0, %k0
 ; AVX512-NEXT:    kshiftrd $16, %k0, %k1
 ; AVX512-NEXT:    kord %k0, %k1, %k0