[X86][SSE] Use MOVMSK for all_of/any_of reduction patterns

[RKSimon]

Summary

This is a first attempt at using the MOVMSK instructions to replace all_of/any_of reduction patterns (i.e. an and/or + shuffle chain).

So far this only matches patterns where we are reducing an all/none bits source vector (i.e. a comparison result) but we should be able to expand on this in conjunction with improvements to 'bool vector' handling both in the x86 backend as well as the vectorizers etc.

[RKSimon]

Avoid this on avx512vl targets

[filcab]

I see that some simple tests grow by a bunch of instructions (also simple, but from looking at instruction tables, it seems to be marginally slower after this transform). These tests are like micro-benchmarks and don't really reflect real-world code, but should we only do this transform if we have a minimum number of shuffles?

Thanks,
Filipe

lib/Target/X86/X86ISelLowering.cpp

28639

Maybe that second part should be in an assert?
The way I see it, we haven't really done much yet, so either:

• The bitwidth changed (how? matchBinOpReduction bails if it encounters an extract_subvector (shuffles with different sizes are a thing in IR though, can't remember if also on SDAG)) but we can still recover if we use the new bitwidth we got (if it still matched, there shouldn't be a big problem, no?)
• The bitwidth changed and we can't recover (does this happen?)

28657

This raised a red flag when reading. Maybe join the two? Will it be that hard to read?
Otherwise, something like

MVT ElemVT = ...;
MaskVT = ...;

[RKSimon]

Updated based on Filipe's feedback.

[RKSimon]

I've limited the combine to only reduce vectors with 4 or more elements.

lib/Target/X86/X86ISelLowering.cpp

28639

So there are valid situations where Match.getScalarValueSizeInBits() != BitWidth as EXTRACT_VECTOR_ELT can perform implicit extension of the extracted vector element. I've added a comment to explain. If it proves useful we can add support for this case but don't see a need yet.

[RKSimon]

ping?

[andreadb]

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

739–746

I was expecting to see the usual cmpl+movw+cmov pattern here.
Is it because you would end up with a conditional move of i8 quantities?

[andreadb]

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

583–585

Rather than movw+cmovnew+cwtl, can we select instead movl+cmovnel?
The advantage is that we remove the partial register writes on AX, and we can get rid of the sign extend at the end (cwtl).

[RKSimon]

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

739–746

Hmm, it might be possible to lower this as:

pmovmskb %xmm0, %edi
xor %eax, %eax
cmpl $65535, %edi
sete %al
negl %eax
retq

[andreadb]

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

739–746

I was about to suggest cmp+set+sra, but neg is even better :-).

[RKSimon]

To avoid the partial register issue, I've updated the result select code to always perform it as i32/i64 followed by a truncation if necessary.

[andreadb]

In D28810#663514, @RKSimon wrote:

To avoid the partial register issue, I've updated the result select code to always perform it as i32/i64 followed by a truncation if necessary.

Nice! That also fixes the issue with partial register writes for OR reductions.

The patch looks good to me now.
Thanks Simon!

lib/Target/X86/X86ISelLowering.cpp

static SDValue createPSADBW(SelectionDAG &DAG, const SDValue &Zext0,
   Ops[0] = Zext1.getOperand(0);
   SDValue SadOp1 = DAG.getNode(ISD::CONCAT_VECTORS, DL, ExtendedVT, Ops);
 
   // Actually build the SAD
   MVT SadVT = MVT::getVectorVT(MVT::i64, RegSize / 64);
   return DAG.getNode(X86ISD::PSADBW, DL, SadVT, SadOp0, SadOp1);
 }
 
+// Attempt to replace an all_of/any_of style horizontal reduction with a MOVMSK.
+static SDValue combineHorizontalPredicateResult(SDNode *Extract,
+                                                SelectionDAG &DAG,
+                                                const X86Subtarget &Subtarget) {
+  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)
+    return SDValue();
+
+  // Check for OR(any_of) and AND(all_of) horizontal reduction patterns.
+  for (ISD::NodeType Op : {ISD::OR, ISD::AND}) {
+    SDValue Match = matchBinOpReduction(Extract, Op);
+    if (!Match)
+      continue;
+
+    // EXTRACT_VECTOR_ELT can require implicit extension of the vector element
+    // which we can't support here for now.
+    if (Match.getScalarValueSizeInBits() != BitWidth)
+      continue;
+
+    // We require AVX2 for PMOVMSKB for v16i16/v32i8;
+    unsigned MatchSizeInBits = Match.getValueSizeInBits();
+    if (!(MatchSizeInBits == 128 ||
+          (MatchSizeInBits == 256 &&
+           ((Subtarget.hasAVX() && BitWidth >= 32) || Subtarget.hasAVX2()))))
+      return SDValue();
+
+    // Don't bother performing this for 2-element vectors.
+    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();
+
+    // For 32/64 bit comparisons use MOVMSKPS/MOVMSKPD, else PMOVMSKB.
+    MVT MaskVT;
+    if (64 == BitWidth || 32 == BitWidth)
+      MaskVT = MVT::getVectorVT(MVT::getFloatingPointVT(BitWidth),
+                                MatchSizeInBits / BitWidth);
+    else
+      MaskVT = MVT::getVectorVT(MVT::i8, MatchSizeInBits / 8);
+
+    APInt CompareBits;
+    ISD::CondCode CondCode;
+    if (Op == ISD::OR) {
+      // any_of -> MOVMSK != 0
+      CompareBits = APInt::getNullValue(32);
+      CondCode = ISD::CondCode::SETNE;
+    } else {
+      // all_of -> MOVMSK == ((1 << NumElts) - 1)
+      CompareBits = APInt::getLowBitsSet(32, MaskVT.getVectorNumElements());
+      CondCode = ISD::CondCode::SETEQ;
+    }
+
+    SDLoc DL(Extract);
+    APInt ResOnes = APInt::getAllOnesValue(BitWidth);
+    APInt ResZero = APInt::getNullValue(BitWidth);
+
+    SDValue Res = DAG.getBitcast(MaskVT, Match);
+    Res = DAG.getNode(X86ISD::MOVMSK, DL, MVT::i32, Res);
+    Res = DAG.getSelectCC(DL, Res, DAG.getConstant(CompareBits, DL, MVT::i32),
+                          DAG.getConstant(ResOnes, DL, ExtractVT),
+                          DAG.getConstant(ResZero, DL, ExtractVT), CondCode);
+    return Res;
+  }
+
+  return SDValue();
+}
+
 static SDValue combineBasicSADPattern(SDNode *Extract, SelectionDAG &DAG,
                                       const X86Subtarget &Subtarget) {
   // PSADBW is only supported on SSE2 and up.
   if (!Subtarget.hasSSE2())
     return SDValue();
 
   // Verify the type we're extracting from is appropriate
   // TODO: There's nothing special about i32, any integer type above i16 should
static SDValue combineExtractVectorElt(SDNode *N, SelectionDAG &DAG,
   }
 
   // Check whether this extract is the root of a sum of absolute differences
   // pattern. This has to be done here because we really want it to happen
   // pre-legalization,
   if (SDValue SAD = combineBasicSADPattern(N, DAG, Subtarget))
     return SAD;
 
+  if (SDValue Cmp = combineHorizontalPredicateResult(N, DAG, Subtarget))
+    return Cmp;
+
   // Only operate on vectors of 4 elements, where the alternative shuffling
   // gets to be more expensive.
   if (InputVector.getValueType() != MVT::v4i32)
     return SDValue();
 
   // Check whether every use of InputVector is an EXTRACT_VECTOR_ELT with a
   // single use which is a sign-extend or zero-extend, and all elements are
   // used.

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

 ; SSE-NEXT:    cmpltpd %xmm1, %xmm3
 ; SSE-NEXT:    cmpltpd %xmm0, %xmm2
 ; SSE-NEXT:    andpd %xmm3, %xmm2
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
 ; SSE-NEXT:    pand %xmm2, %xmm0
 ; SSE-NEXT:    movd %xmm0, %rax
 ; SSE-NEXT:    retq
 ;
-; AVX1-LABEL: test_v4f64:
-; AVX1:       # BB#0:
-; AVX1-NEXT:    vcmpltpd %ymm0, %ymm1, %ymm0
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vandpd %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vandpd %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vmovq %xmm0, %rax
-; AVX1-NEXT:    vzeroupper
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: test_v4f64:
-; AVX2:       # BB#0:
-; AVX2-NEXT:    vcmpltpd %ymm0, %ymm1, %ymm0
-; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vandpd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vmovq %xmm0, %rax
-; AVX2-NEXT:    vzeroupper
-; AVX2-NEXT:    retq
+; AVX-LABEL: test_v4f64:
+; AVX:       # BB#0:
+; AVX-NEXT:    vcmpltpd %ymm0, %ymm1, %ymm0
+; AVX-NEXT:    vmovmskpd %ymm0, %eax
+; AVX-NEXT:    xorl %ecx, %ecx
+; AVX-NEXT:    cmpl $15, %eax
+; AVX-NEXT:    movq $-1, %rax
+; AVX-NEXT:    cmovneq %rcx, %rax
+; AVX-NEXT:    vzeroupper
+; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4f64:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vcmpltpd %ymm0, %ymm1, %k1
 ; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqa64 %ymm0, %ymm0 {%k1} {z}
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpand %ymm1, %ymm0, %ymm0

 }
 
 define i64 @test_v4f64_legal(<4 x double> %a0, <4 x double> %a1) {
 ; SSE-LABEL: test_v4f64_legal:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    cmpltpd %xmm1, %xmm3
 ; SSE-NEXT:    cmpltpd %xmm0, %xmm2
 ; SSE-NEXT:    packsswb %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:    movd %xmm1, %eax
+; SSE-NEXT:    movmskps %xmm2, %eax
+; SSE-NEXT:    xorl %ecx, %ecx
+; SSE-NEXT:    cmpl $15, %eax
+; SSE-NEXT:    movl $-1, %eax
+; SSE-NEXT:    cmovnel %ecx, %eax
 ; SSE-NEXT:    cltq
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4f64_legal:
 ; AVX:       # BB#0:
 ; AVX-NEXT:    vcmpltpd %ymm0, %ymm1, %ymm0
 ; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vmovd %xmm0, %eax
+; AVX-NEXT:    vmovmskps %xmm0, %eax
+; AVX-NEXT:    xorl %ecx, %ecx
+; AVX-NEXT:    cmpl $15, %eax
+; AVX-NEXT:    movl $-1, %eax
+; AVX-NEXT:    cmovnel %ecx, %eax
 ; AVX-NEXT:    cltq
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4f64_legal:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vcmpltpd %ymm0, %ymm1, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    retq
   %6 = sext i32 %5 to i64
   ret i64 %6
 }
 
 define i32 @test_v4f32(<4 x float> %a0, <4 x float> %a1) {
 ; SSE-LABEL: test_v4f32:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    cmpltps %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:    movd %xmm1, %eax
+; SSE-NEXT:    movmskps %xmm1, %eax
+; SSE-NEXT:    xorl %ecx, %ecx
+; SSE-NEXT:    cmpl $15, %eax
+; SSE-NEXT:    movl $-1, %eax
+; SSE-NEXT:    cmovnel %ecx, %eax
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4f32:
 ; AVX:       # BB#0:
 ; AVX-NEXT:    vcmpltps %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; AVX-NEXT:    vandpd %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vmovd %xmm0, %eax
+; AVX-NEXT:    vmovmskps %xmm0, %eax
+; AVX-NEXT:    xorl %ecx, %ecx
+; AVX-NEXT:    cmpl $15, %eax
+; AVX-NEXT:    movl $-1, %eax
+; AVX-NEXT:    cmovnel %ecx, %eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4f32:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vcmpltps %xmm0, %xmm1, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]

 }
 
 define i32 @test_v8f32(<8 x float> %a0, <8 x float> %a1) {
 ; SSE-LABEL: test_v8f32:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    cmpltps %xmm1, %xmm3
 ; SSE-NEXT:    cmpltps %xmm0, %xmm2
 ; SSE-NEXT:    andps %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:    movd %xmm1, %eax
+; SSE-NEXT:    movmskps %xmm2, %eax
+; SSE-NEXT:    xorl %ecx, %ecx
+; SSE-NEXT:    cmpl $15, %eax
+; SSE-NEXT:    movl $-1, %eax
+; SSE-NEXT:    cmovnel %ecx, %eax
 ; SSE-NEXT:    retq
 ;
-; AVX1-LABEL: test_v8f32:
-; AVX1:       # BB#0:
-; AVX1-NEXT:    vcmpltps %ymm0, %ymm1, %ymm0
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vandps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vandps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vandps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vmovd %xmm0, %eax
-; AVX1-NEXT:    vzeroupper
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: test_v8f32:
-; AVX2:       # BB#0:
-; AVX2-NEXT:    vcmpltps %ymm0, %ymm1, %ymm0
-; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vandps %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vmovd %xmm0, %eax
-; AVX2-NEXT:    vzeroupper
-; AVX2-NEXT:    retq
+; AVX-LABEL: test_v8f32:
+; AVX:       # BB#0:
+; AVX-NEXT:    vcmpltps %ymm0, %ymm1, %ymm0
+; AVX-NEXT:    vmovmskps %ymm0, %eax
+; AVX-NEXT:    xorl %ecx, %ecx
+; AVX-NEXT:    cmpl $255, %eax
+; AVX-NEXT:    movl $-1, %eax
+; AVX-NEXT:    cmovnel %ecx, %eax
+; AVX-NEXT:    vzeroupper
+; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8f32:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vcmpltps %ymm0, %ymm1, %k1
 ; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpand %ymm1, %ymm0, %ymm0

 }
 
 define i32 @test_v8f32_legal(<8 x float> %a0, <8 x float> %a1) {
 ; SSE-LABEL: test_v8f32_legal:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    cmpltps %xmm1, %xmm3
 ; SSE-NEXT:    cmpltps %xmm0, %xmm2
 ; SSE-NEXT:    packsswb %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:    movd %xmm0, %eax
+; SSE-NEXT:    pmovmskb %xmm2, %eax
+; SSE-NEXT:    xorl %ecx, %ecx
+; SSE-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; SSE-NEXT:    movw $-1, %ax
+; SSE-NEXT:    cmovnew %cx, %ax
 ; SSE-NEXT:    cwtl
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8f32_legal:
 ; AVX:       # BB#0:
 ; AVX-NEXT:    vcmpltps %ymm0, %ymm1, %ymm0
 ; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vmovd %xmm0, %eax
+; AVX-NEXT:    vpmovmskb %xmm0, %eax
+; AVX-NEXT:    xorl %ecx, %ecx
+; AVX-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; AVX-NEXT:    movw $-1, %ax
+; AVX-NEXT:    cmovnew %cx, %ax
 ; AVX-NEXT:    cwtl
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8f32_legal:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vcmpltps %ymm0, %ymm1, %k0
 ; AVX512-NEXT:    vpmovm2w %k0, %xmm0

 ;
 ; AVX1-LABEL: test_v4i64:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX1-NEXT:    vandps %ymm2, %ymm0, %ymm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vandps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vmovq %xmm0, %rax
+; AVX1-NEXT:    vmovmskpd %ymm0, %eax
+; AVX1-NEXT:    xorl %ecx, %ecx
+; AVX1-NEXT:    cmpl $15, %eax
+; AVX1-NEXT:    movq $-1, %rax
+; AVX1-NEXT:    cmovneq %rcx, %rax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v4i64:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtq %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vmovq %xmm0, %rax
+; AVX2-NEXT:    vmovmskpd %ymm0, %eax
+; AVX2-NEXT:    xorl %ecx, %ecx
+; AVX2-NEXT:    cmpl $15, %eax
+; AVX2-NEXT:    movq $-1, %rax
+; AVX2-NEXT:    cmovneq %rcx, %rax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4i64:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtq %ymm1, %ymm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqa64 %ymm0, %ymm0 {%k1} {z}

 }
 
 define i64 @test_v4i64_legal(<4 x i64> %a0, <4 x i64> %a1) {
 ; SSE-LABEL: test_v4i64_legal:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtq %xmm3, %xmm1
 ; SSE-NEXT:    pcmpgtq %xmm2, %xmm0
 ; SSE-NEXT:    packsswb %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:    movd %xmm0, %eax
+; SSE-NEXT:    movmskps %xmm0, %eax
+; SSE-NEXT:    xorl %ecx, %ecx
+; SSE-NEXT:    cmpl $15, %eax
+; SSE-NEXT:    movl $-1, %eax
+; SSE-NEXT:    cmovnel %ecx, %eax
 ; SSE-NEXT:    cltq
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v4i64_legal:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpacksswb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    vmovmskps %xmm0, %eax
+; AVX1-NEXT:    xorl %ecx, %ecx
+; AVX1-NEXT:    cmpl $15, %eax
+; AVX1-NEXT:    movl $-1, %eax
+; AVX1-NEXT:    cmovnel %ecx, %eax
 ; AVX1-NEXT:    cltq
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v4i64_legal:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtq %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    vmovmskps %xmm0, %eax
+; AVX2-NEXT:    xorl %ecx, %ecx
+; AVX2-NEXT:    cmpl $15, %eax
+; AVX2-NEXT:    movl $-1, %eax
+; AVX2-NEXT:    cmovnel %ecx, %eax
 ; AVX2-NEXT:    cltq
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4i64_legal:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtq %ymm1, %ymm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    retq
   %6 = sext i32 %5 to i64
   ret i64 %6
 }
 
 define i32 @test_v4i32(<4 x i32> %a0, <4 x i32> %a1) {
 ; SSE-LABEL: test_v4i32:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtd %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:    movd %xmm0, %eax
+; SSE-NEXT:    movmskps %xmm0, %eax
+; SSE-NEXT:    xorl %ecx, %ecx
+; SSE-NEXT:    cmpl $15, %eax
+; SSE-NEXT:    movl $-1, %eax
+; SSE-NEXT:    cmovnel %ecx, %eax
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4i32:
 ; AVX:       # BB#0:
 ; AVX-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vmovd %xmm0, %eax
+; AVX-NEXT:    vmovmskps %xmm0, %eax
+; AVX-NEXT:    xorl %ecx, %ecx
+; AVX-NEXT:    cmpl $15, %eax
+; AVX-NEXT:    movl $-1, %eax
+; AVX-NEXT:    cmovnel %ecx, %eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4i32:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtd %xmm1, %xmm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]

 }
 
 define i32 @test_v8i32(<8 x i32> %a0, <8 x i32> %a1) {
 ; SSE-LABEL: test_v8i32:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtd %xmm3, %xmm1
 ; SSE-NEXT:    pcmpgtd %xmm2, %xmm0
 ; SSE-NEXT:    pand %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:    movd %xmm0, %eax
+; SSE-NEXT:    movmskps %xmm0, %eax
+; SSE-NEXT:    xorl %ecx, %ecx
+; SSE-NEXT:    cmpl $15, %eax
+; SSE-NEXT:    movl $-1, %eax
+; SSE-NEXT:    cmovnel %ecx, %eax
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v8i32:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtd %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX1-NEXT:    vandps %ymm2, %ymm0, %ymm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vandps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vandps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    vmovmskps %ymm0, %eax
+; AVX1-NEXT:    xorl %ecx, %ecx
+; AVX1-NEXT:    cmpl $255, %eax
+; AVX1-NEXT:    movl $-1, %eax
+; AVX1-NEXT:    cmovnel %ecx, %eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v8i32:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    vmovmskps %ymm0, %eax
+; AVX2-NEXT:    xorl %ecx, %ecx
+; AVX2-NEXT:    cmpl $255, %eax
+; AVX2-NEXT:    movl $-1, %eax
+; AVX2-NEXT:    cmovnel %ecx, %eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i32:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtd %ymm1, %ymm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}

 }
 
 define i32 @test_v8i32_legal(<8 x i32> %a0, <8 x i32> %a1) {
 ; SSE-LABEL: test_v8i32_legal:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtd %xmm3, %xmm1
 ; SSE-NEXT:    pcmpgtd %xmm2, %xmm0
 ; SSE-NEXT:    packsswb %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:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    psrld $16, %xmm1
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    pmovmskb %xmm0, %eax
+; SSE-NEXT:    xorl %ecx, %ecx
+; SSE-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; SSE-NEXT:    movw $-1, %ax
+; SSE-NEXT:    cmovnew %cx, %ax
 ; SSE-NEXT:    cwtl
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v8i32_legal:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtd %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpacksswb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    xorl %ecx, %ecx
+; AVX1-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; AVX1-NEXT:    movw $-1, %ax
+; AVX1-NEXT:    cmovnew %cx, %ax
 ; AVX1-NEXT:    cwtl
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v8i32_legal:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtd %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    vpmovmskb %xmm0, %eax
+; AVX2-NEXT:    xorl %ecx, %ecx
+; AVX2-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; AVX2-NEXT:    movw $-1, %ax
+; AVX2-NEXT:    cmovnew %cx, %ax
 ; AVX2-NEXT:    cwtl
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i32_legal:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtd %ymm1, %ymm0, %k0
 ; AVX512-NEXT:    vpmovm2w %k0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
; AVX512-NEXT:    retq
   %8 = sext i16 %7 to i32
   ret i32 %8
 }
 
 define i16 @test_v8i16(<8 x i16> %a0, <8 x i16> %a1) {
 ; SSE-LABEL: test_v8i16:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtw %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:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    psrld $16, %xmm1
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    movd %xmm1, %eax
-; SSE-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
+; SSE-NEXT:    pmovmskb %xmm0, %eax
+; SSE-NEXT:    xorl %ecx, %ecx
+; SSE-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; SSE-NEXT:    movw $-1, %ax
+; SSE-NEXT:    cmovnew %cx, %ax
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i16:
 ; AVX:       # BB#0:
 ; AVX-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vmovd %xmm0, %eax
-; AVX-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
+; AVX-NEXT:    vpmovmskb %xmm0, %eax
+; AVX-NEXT:    xorl %ecx, %ecx
+; AVX-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; AVX-NEXT:    movw $-1, %ax
+; AVX-NEXT:    cmovnew %cx, %ax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i16:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtw %xmm1, %xmm0, %k0
 ; AVX512-NEXT:    vpmovm2w %k0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0

 }
 
 define i16 @test_v16i16(<16 x i16> %a0, <16 x i16> %a1) {
 ; SSE-LABEL: test_v16i16:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtw %xmm3, %xmm1
 ; SSE-NEXT:    pcmpgtw %xmm2, %xmm0
 ; SSE-NEXT:    pand %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:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    psrld $16, %xmm1
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    movd %xmm1, %eax
-; SSE-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
+; SSE-NEXT:    pmovmskb %xmm0, %eax
+; SSE-NEXT:    xorl %ecx, %ecx
+; SSE-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; SSE-NEXT:    movw $-1, %ax
+; SSE-NEXT:    cmovnew %cx, %ax
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v16i16:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; AVX1-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vandps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX1-NEXT:    vandps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX1-NEXT:    vandps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v16i16:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtw %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vmovd %xmm0, %eax
-; AVX2-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
+; AVX2-NEXT:    vpmovmskb %ymm0, %eax
+; AVX2-NEXT:    xorl %ecx, %ecx
+; AVX2-NEXT:    cmpl $-1, %eax
+; AVX2-NEXT:    movw $-1, %ax
+; AVX2-NEXT:    cmovnew %cx, %ax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i16:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtw %ymm1, %ymm0, %k0
 ; AVX512-NEXT:    vpmovm2w %k0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1

 }
 
 define i16 @test_v16i16_legal(<16 x i16> %a0, <16 x i16> %a1) {
 ; SSE-LABEL: test_v16i16_legal:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtw %xmm3, %xmm1
 ; SSE-NEXT:    pcmpgtw %xmm2, %xmm0
 ; SSE-NEXT:    packsswb %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:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    psrld $16, %xmm1
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    psrlw $8, %xmm0
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
+; SSE-NEXT:    pmovmskb %xmm0, %eax
+; SSE-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; SSE-NEXT:    movb $-1, %al
+; SSE-NEXT:    je .LBB14_2
+; SSE-NEXT:  # BB#1:
+; SSE-NEXT:    xorl %eax, %eax
+; SSE-NEXT:  .LBB14_2:
 ; SSE-NEXT:    movsbl %al, %eax
 ; SSE-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v16i16_legal:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpacksswb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    vpmovmskb %xmm0, %ecx
+; AVX1-NEXT:    movb $-1, %al
+; AVX1-NEXT:    cmpl $65535, %ecx # imm = 0xFFFF
+; AVX1-NEXT:    je .LBB14_2
+; AVX1-NEXT:  # BB#1:
+; AVX1-NEXT:    xorl %eax, %eax
+; AVX1-NEXT:  .LBB14_2:
 ; AVX1-NEXT:    movsbl %al, %eax
 ; AVX1-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v16i16_legal:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vpmovmskb %xmm0, %ecx
+; AVX2-NEXT:    movb $-1, %al
+; AVX2-NEXT:    cmpl $65535, %ecx # imm = 0xFFFF
+; AVX2-NEXT:    je .LBB14_2
+; AVX2-NEXT:  # BB#1:
+; AVX2-NEXT:    xorl %eax, %eax
+; AVX2-NEXT:  .LBB14_2:
 ; AVX2-NEXT:    movsbl %al, %eax
 ; AVX2-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i16_legal:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtw %ymm1, %ymm0, %k0
; AVX512-NEXT:    retq
   %10 = sext i8 %9 to i16
   ret i16 %10
 }
 
 define i8 @test_v16i8(<16 x i8> %a0, <16 x i8> %a1) {
 ; SSE-LABEL: test_v16i8:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtb %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:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    psrld $16, %xmm1
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    psrlw $8, %xmm0
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
+; SSE-NEXT:    pmovmskb %xmm0, %eax
+; SSE-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; SSE-NEXT:    movb $-1, %al
+; SSE-NEXT:    je .LBB15_2
+; SSE-NEXT:  # BB#1:
+; SSE-NEXT:    xorl %eax, %eax
+; SSE-NEXT:  .LBB15_2:
 ; SSE-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v16i8:
 ; AVX:       # BB#0:
 ; AVX-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    vpmovmskb %xmm0, %ecx
+; AVX-NEXT:    movb $-1, %al
+; AVX-NEXT:    cmpl $65535, %ecx # imm = 0xFFFF
+; AVX-NEXT:    je .LBB15_2
+; AVX-NEXT:  # BB#1:
+; AVX-NEXT:    xorl %eax, %eax
+; AVX-NEXT:  .LBB15_2:
 ; AVX-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i8:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0
 ; AVX512-NEXT:    vpmovm2b %k0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]

 }
 
 define i8 @test_v32i8(<32 x i8> %a0, <32 x i8> %a1) {
 ; SSE-LABEL: test_v32i8:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtb %xmm3, %xmm1
 ; SSE-NEXT:    pcmpgtb %xmm2, %xmm0
 ; SSE-NEXT:    pand %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:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    psrld $16, %xmm1
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    psrlw $8, %xmm0
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
+; SSE-NEXT:    pmovmskb %xmm0, %eax
+; SSE-NEXT:    cmpl $65535, %eax # imm = 0xFFFF
+; SSE-NEXT:    movb $-1, %al
+; SSE-NEXT:    je .LBB16_2
+; SSE-NEXT:  # BB#1:
+; SSE-NEXT:    xorl %eax, %eax
+; SSE-NEXT:  .LBB16_2:
 ; SSE-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtb %xmm2, %xmm3, %xmm2

 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX1-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    vpmovmskb %ymm0, %ecx
+; AVX2-NEXT:    movb $-1, %al
+; AVX2-NEXT:    cmpl $-1, %ecx
+; AVX2-NEXT:    je .LBB16_2
+; AVX2-NEXT:  # BB#1:
+; AVX2-NEXT:    xorl %eax, %eax
+; AVX2-NEXT:  .LBB16_2:
 ; AVX2-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i8:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtb %ymm1, %ymm0, %k0
 ; AVX512-NEXT:    vpmovm2b %k0, %ymm0

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

 ; SSE-NEXT:    cmpltpd %xmm1, %xmm3
 ; SSE-NEXT:    cmpltpd %xmm0, %xmm2
 ; SSE-NEXT:    orpd %xmm3, %xmm2
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
 ; SSE-NEXT:    por %xmm2, %xmm0
 ; SSE-NEXT:    movd %xmm0, %rax
 ; SSE-NEXT:    retq
 ;
-; AVX1-LABEL: test_v4f64:
-; AVX1:       # BB#0:
-; AVX1-NEXT:    vcmpltpd %ymm0, %ymm1, %ymm0
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vorpd %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vorpd %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vmovq %xmm0, %rax
-; AVX1-NEXT:    vzeroupper
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: test_v4f64:
-; AVX2:       # BB#0:
-; AVX2-NEXT:    vcmpltpd %ymm0, %ymm1, %ymm0
-; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vorpd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vmovq %xmm0, %rax
-; AVX2-NEXT:    vzeroupper
-; AVX2-NEXT:    retq
+; AVX-LABEL: test_v4f64:
+; AVX:       # BB#0:
+; AVX-NEXT:    vcmpltpd %ymm0, %ymm1, %ymm0
+; AVX-NEXT:    vmovmskpd %ymm0, %eax
+; AVX-NEXT:    negl %eax
+; AVX-NEXT:    sbbq %rax, %rax
+; AVX-NEXT:    vzeroupper
+; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4f64:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vcmpltpd %ymm0, %ymm1, %k1
 ; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqa64 %ymm0, %ymm0 {%k1} {z}
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpor %ymm1, %ymm0, %ymm0

 }
 
 define i64 @test_v4f64_legal(<4 x double> %a0, <4 x double> %a1) {
 ; SSE-LABEL: test_v4f64_legal:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    cmpltpd %xmm1, %xmm3
 ; SSE-NEXT:    cmpltpd %xmm0, %xmm2
 ; SSE-NEXT:    packsswb %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:    movd %xmm1, %eax
+; SSE-NEXT:    movmskps %xmm2, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbl %eax, %eax
 ; SSE-NEXT:    cltq
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4f64_legal:
 ; AVX:       # BB#0:
 ; AVX-NEXT:    vcmpltpd %ymm0, %ymm1, %ymm0
 ; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vmovd %xmm0, %eax
+; AVX-NEXT:    vmovmskps %xmm0, %eax
+; AVX-NEXT:    negl %eax
+; AVX-NEXT:    sbbl %eax, %eax
 ; AVX-NEXT:    cltq
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4f64_legal:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vcmpltpd %ymm0, %ymm1, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    retq
   %6 = sext i32 %5 to i64
   ret i64 %6
 }
 
 define i32 @test_v4f32(<4 x float> %a0, <4 x float> %a1) {
 ; SSE-LABEL: test_v4f32:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    cmpltps %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:    movd %xmm1, %eax
+; SSE-NEXT:    movmskps %xmm1, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbl %eax, %eax
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4f32:
 ; AVX:       # BB#0:
 ; AVX-NEXT:    vcmpltps %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; AVX-NEXT:    vorpd %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vmovd %xmm0, %eax
+; AVX-NEXT:    vmovmskps %xmm0, %eax
+; AVX-NEXT:    negl %eax
+; AVX-NEXT:    sbbl %eax, %eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4f32:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vcmpltps %xmm0, %xmm1, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]

 }
 
 define i32 @test_v8f32(<8 x float> %a0, <8 x float> %a1) {
 ; SSE-LABEL: test_v8f32:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    cmpltps %xmm1, %xmm3
 ; SSE-NEXT:    cmpltps %xmm0, %xmm2
 ; SSE-NEXT:    orps %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:    movd %xmm1, %eax
+; SSE-NEXT:    movmskps %xmm2, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbl %eax, %eax
 ; SSE-NEXT:    retq
 ;
-; AVX1-LABEL: test_v8f32:
-; AVX1:       # BB#0:
-; AVX1-NEXT:    vcmpltps %ymm0, %ymm1, %ymm0
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vmovd %xmm0, %eax
-; AVX1-NEXT:    vzeroupper
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: test_v8f32:
-; AVX2:       # BB#0:
-; AVX2-NEXT:    vcmpltps %ymm0, %ymm1, %ymm0
-; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vorps %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vmovd %xmm0, %eax
-; AVX2-NEXT:    vzeroupper
-; AVX2-NEXT:    retq
+; AVX-LABEL: test_v8f32:
+; AVX:       # BB#0:
+; AVX-NEXT:    vcmpltps %ymm0, %ymm1, %ymm0
+; AVX-NEXT:    vmovmskps %ymm0, %eax
+; AVX-NEXT:    negl %eax
+; AVX-NEXT:    sbbl %eax, %eax
+; AVX-NEXT:    vzeroupper
+; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8f32:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vcmpltps %ymm0, %ymm1, %k1
 ; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpor %ymm1, %ymm0, %ymm0

 }
 
 define i32 @test_v8f32_legal(<8 x float> %a0, <8 x float> %a1) {
 ; SSE-LABEL: test_v8f32_legal:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    cmpltps %xmm1, %xmm3
 ; SSE-NEXT:    cmpltps %xmm0, %xmm2
 ; SSE-NEXT:    packsswb %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:    movd %xmm0, %eax
+; SSE-NEXT:    pmovmskb %xmm2, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbw %ax, %ax
 ; SSE-NEXT:    cwtl
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8f32_legal:
 ; AVX:       # BB#0:
 ; AVX-NEXT:    vcmpltps %ymm0, %ymm1, %ymm0
 ; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vmovd %xmm0, %eax
+; AVX-NEXT:    vpmovmskb %xmm0, %eax
+; AVX-NEXT:    negl %eax
+; AVX-NEXT:    sbbw %ax, %ax
 ; AVX-NEXT:    cwtl
 ; AVX-NEXT:    vzeroupper
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8f32_legal:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vcmpltps %ymm0, %ymm1, %k0
 ; AVX512-NEXT:    vpmovm2w %k0, %xmm0

 ;
 ; AVX1-LABEL: test_v4i64:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm2, %ymm0, %ymm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vmovq %xmm0, %rax
+; AVX1-NEXT:    vmovmskpd %ymm0, %eax
+; AVX1-NEXT:    negl %eax
+; AVX1-NEXT:    sbbq %rax, %rax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v4i64:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtq %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vmovq %xmm0, %rax
+; AVX2-NEXT:    vmovmskpd %ymm0, %eax
+; AVX2-NEXT:    negl %eax
+; AVX2-NEXT:    sbbq %rax, %rax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4i64:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtq %ymm1, %ymm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqa64 %ymm0, %ymm0 {%k1} {z}

 }
 
 define i64 @test_v4i64_legal(<4 x i64> %a0, <4 x i64> %a1) {
 ; SSE-LABEL: test_v4i64_legal:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtq %xmm3, %xmm1
 ; SSE-NEXT:    pcmpgtq %xmm2, %xmm0
 ; SSE-NEXT:    packsswb %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:    movd %xmm0, %eax
+; SSE-NEXT:    movmskps %xmm0, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbl %eax, %eax
 ; SSE-NEXT:    cltq
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v4i64_legal:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpacksswb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    vmovmskps %xmm0, %eax
+; AVX1-NEXT:    negl %eax
+; AVX1-NEXT:    sbbl %eax, %eax
 ; AVX1-NEXT:    cltq
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v4i64_legal:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtq %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    vmovmskps %xmm0, %eax
+; AVX2-NEXT:    negl %eax
+; AVX2-NEXT:    sbbl %eax, %eax
 ; AVX2-NEXT:    cltq
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4i64_legal:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtq %ymm1, %ymm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX512-NEXT:    retq
   %6 = sext i32 %5 to i64
   ret i64 %6
 }
 
 define i32 @test_v4i32(<4 x i32> %a0, <4 x i32> %a1) {
 ; SSE-LABEL: test_v4i32:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtd %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:    movd %xmm0, %eax
+; SSE-NEXT:    movmskps %xmm0, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbl %eax, %eax
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4i32:
 ; AVX:       # BB#0:
 ; AVX-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vmovd %xmm0, %eax
+; AVX-NEXT:    vmovmskps %xmm0, %eax
+; AVX-NEXT:    negl %eax
+; AVX-NEXT:    sbbl %eax, %eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4i32:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtd %xmm1, %xmm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]

 }
 
 define i32 @test_v8i32(<8 x i32> %a0, <8 x i32> %a1) {
 ; SSE-LABEL: test_v8i32:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtd %xmm3, %xmm1
 ; SSE-NEXT:    pcmpgtd %xmm2, %xmm0
 ; SSE-NEXT:    por %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:    movd %xmm0, %eax
+; SSE-NEXT:    movmskps %xmm0, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbl %eax, %eax
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v8i32:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtd %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm2, %ymm0, %ymm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
-; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    vmovmskps %ymm0, %eax
+; AVX1-NEXT:    negl %eax
+; AVX1-NEXT:    sbbl %eax, %eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v8i32:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    vmovmskps %ymm0, %eax
+; AVX2-NEXT:    negl %eax
+; AVX2-NEXT:    sbbl %eax, %eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i32:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtd %ymm1, %ymm0, %k1
 ; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}

 }
 
 define i32 @test_v8i32_legal(<8 x i32> %a0, <8 x i32> %a1) {
 ; SSE-LABEL: test_v8i32_legal:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtd %xmm3, %xmm1
 ; SSE-NEXT:    pcmpgtd %xmm2, %xmm0
 ; SSE-NEXT:    packsswb %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:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    psrld $16, %xmm1
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movd %xmm1, %eax
+; SSE-NEXT:    pmovmskb %xmm0, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbw %ax, %ax
 ; SSE-NEXT:    cwtl
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v8i32_legal:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtd %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpacksswb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    negl %eax
+; AVX1-NEXT:    sbbw %ax, %ax
 ; AVX1-NEXT:    cwtl
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v8i32_legal:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtd %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    vpmovmskb %xmm0, %eax
+; AVX2-NEXT:    negl %eax
+; AVX2-NEXT:    sbbw %ax, %ax
 ; AVX2-NEXT:    cwtl
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i32_legal:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtd %ymm1, %ymm0, %k0
 ; AVX512-NEXT:    vpmovm2w %k0, %xmm0
; AVX512-NEXT:    retq
   %8 = sext i16 %7 to i32
   ret i32 %8
 }
 
 define i16 @test_v8i16(<8 x i16> %a0, <8 x i16> %a1) {
 ; SSE-LABEL: test_v8i16:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtw %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:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    psrld $16, %xmm1
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movd %xmm1, %eax
-; SSE-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
+; SSE-NEXT:    pmovmskb %xmm0, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbw %ax, %ax
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i16:
 ; AVX:       # BB#0:
 ; AVX-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vmovd %xmm0, %eax
-; AVX-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
+; AVX-NEXT:    vpmovmskb %xmm0, %eax
+; AVX-NEXT:    negl %eax
+; AVX-NEXT:    sbbw %ax, %ax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i16:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtw %xmm1, %xmm0, %k0
 ; AVX512-NEXT:    vpmovm2w %k0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0

 }
 
 define i16 @test_v16i16(<16 x i16> %a0, <16 x i16> %a1) {
 ; SSE-LABEL: test_v16i16:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtw %xmm3, %xmm1
 ; SSE-NEXT:    pcmpgtw %xmm2, %xmm0
 ; SSE-NEXT:    por %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:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    psrld $16, %xmm1
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movd %xmm1, %eax
-; SSE-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
+; SSE-NEXT:    pmovmskb %xmm0, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbw %ax, %ax
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v16i16:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; AVX1-NEXT:    vorps %ymm2, %ymm0, %ymm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v16i16:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtw %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vmovd %xmm0, %eax
-; AVX2-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
+; AVX2-NEXT:    vpmovmskb %ymm0, %eax
+; AVX2-NEXT:    negl %eax
+; AVX2-NEXT:    sbbw %ax, %ax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i16:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtw %ymm1, %ymm0, %k0
 ; AVX512-NEXT:    vpmovm2w %k0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1

 }
 
 define i16 @test_v16i16_legal(<16 x i16> %a0, <16 x i16> %a1) {
 ; SSE-LABEL: test_v16i16_legal:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtw %xmm3, %xmm1
 ; SSE-NEXT:    pcmpgtw %xmm2, %xmm0
 ; SSE-NEXT:    packsswb %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:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    psrld $16, %xmm1
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    psrlw $8, %xmm0
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
-; SSE-NEXT:    movsbl %al, %eax
-; SSE-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
+; SSE-NEXT:    pmovmskb %xmm0, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbw %ax, %ax
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v16i16_legal:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpacksswb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
-; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX1-NEXT:    movsbl %al, %eax
-; AVX1-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    negl %eax
+; AVX1-NEXT:    sbbw %ax, %ax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v16i16_legal:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
-; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX2-NEXT:    movsbl %al, %eax
-; AVX2-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
+; AVX2-NEXT:    vpmovmskb %xmm0, %eax
+; AVX2-NEXT:    negl %eax
+; AVX2-NEXT:    sbbw %ax, %ax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i16_legal:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtw %ymm1, %ymm0, %k0
 ; AVX512-NEXT:    vpmovm2b %k0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
; AVX512-NEXT:    retq
   %10 = sext i8 %9 to i16
   ret i16 %10
 }
 
 define i8 @test_v16i8(<16 x i8> %a0, <16 x i8> %a1) {
 ; SSE-LABEL: test_v16i8:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtb %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:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    psrld $16, %xmm1
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    psrlw $8, %xmm0
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
-; SSE-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
+; SSE-NEXT:    pmovmskb %xmm0, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbb %al, %al
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v16i8:
 ; AVX:       # BB#0:
 ; AVX-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
+; AVX-NEXT:    vpmovmskb %xmm0, %eax
+; AVX-NEXT:    negl %eax
+; AVX-NEXT:    sbbb %al, %al
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i8:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0
 ; AVX512-NEXT:    vpmovm2b %k0, %xmm0
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm0

 }
 
 define i8 @test_v32i8(<32 x i8> %a0, <32 x i8> %a1) {
 ; SSE-LABEL: test_v32i8:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pcmpgtb %xmm3, %xmm1
 ; SSE-NEXT:    pcmpgtb %xmm2, %xmm0
 ; SSE-NEXT:    por %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:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    psrld $16, %xmm1
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    psrlw $8, %xmm0
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    pextrb $0, %xmm0, %eax
-; SSE-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
+; SSE-NEXT:    pmovmskb %xmm0, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbb %al, %al
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0

 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX1-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
-; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX2-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
+; AVX2-NEXT:    vpmovmskb %ymm0, %eax
+; AVX2-NEXT:    negl %eax
+; AVX2-NEXT:    sbbb %al, %al
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i8:
 ; AVX512:       # BB#0:
 ; AVX512-NEXT:    vpcmpgtb %ymm1, %ymm0, %k0
 ; AVX512-NEXT:    vpmovm2b %k0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1