[X86] Use min/max for vector ult/ugt compares if avoids a sign flip.

[craig.topper]

Summary

Currently we only use min/max to help with ule/uge compares because it removes an invert of the result that would otherwise be needed. But we can also use it for ult/ugt compares if it will prevent the need for a sign bit flip needed to use pcmpgt at the cost of requiring an invert after the compare.

I also refactored the code so that the max/min code is self contained and does its own return instead of setting up a flag to manipulate the rest of the function's behavior.

Most of the test cases look ok with this. I did notice that we added instructions when one of the operands being sign flipped is a constant vector that we were able to constant fold the flip into.

I also noticed that sometimes the SSE min/max clobbers a register that is needed after the compare. This resulted in an extra move being inserted before the min/max to preserve the register. We could try to detect this and switch from min to max and change the compare operands to use the operand that gets reused in the compare.

[RKSimon]

I've been hitting some of the same issues while finishing the TRUNC(CLAMP()) truncation -> PACK support. Would it be a good idea to enable custom lowering of all (U,S)(MAX,MIN) on SSE2+ targets? It cleans out a lot of CMPGT/CMPGE mismatches and makes it easier to match a lot of this stuff using the SMIN etc. opcodes directly.

[craig.topper]

Maybe it's too early in the morning, but I'm not sure I followed. A lot of these cases aren't min/max how does custom handling help?

[RKSimon]

In D42935#999124, @craig.topper wrote:

Maybe it's too early in the morning, but I'm not sure I followed. A lot of these cases aren't min/max how does custom handling help?

No, it might be me - I'm starting to see phantom minmax patterns everywhere....

But that test_ult_byte regression is annoying.

lib/Target/X86/X86ISelLowering.cpp

18008

Use IsOperationLegal instead? I think that'd add VLX v2i64/v4i64 support?

test/CodeGen/X86/vec_setcc-2.ll

150

Interesting that this didn't commute and fold the load? You'd need a MOVDQArr instead I guess.

[craig.topper]

lib/Target/X86/X86ISelLowering.cpp

18008

hasAVX512() && VET == MVT::i64 should cover VLX as well since AVX512 is a subset of VLX.

But I agree is OperationLegal would be cleaner.

[craig.topper]

Use isOperationLegal

[RKSimon]

LGTM, with a couple of minor thoughts.

Having looked through, I think the PSUBUS regressions can be blamed on PR31293 and shouldn't block this patch.

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

16301

Is this necessary? Is there anyway that ComputeKnownBits/NumSignBits + SimplifyDemandedBits could be used to remove it?

[craig.topper]

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

16301

I believe it's getting tripped up because the vpternlog is a (v16i8 (bitcast (xor (bitcast v2i64))) after legalize. I don't think computeNumSignBits can cross both of those bitcasts very well.

Probably doesn't help that LegalVectorOps visits nodes bottom up so the compare is legalized first, it creates the xor, the xor gets legalized with the bitcatsts. Then the truncate gets legalized, but its too late.

Meanwhile LegalizeDAG visits nodes top down so would probably be able to handle this case since the truncate would legalize first.

lib/Target/X86/X86ISelLowering.cpp

if (BC0.getOpcode() == ISD::AND) {
         if (llvm::all_of(EltBits, [](APInt &V) { return V.isPowerOf2(); })) {
           Cond = ISD::SETEQ;
           Op1 = DAG.getBitcast(VT, BC0.getOperand(1));
         }
       }
     }
   }
 
-  // We are handling one of the integer comparisons here. Since SSE only has
-  // GT and EQ comparisons for integer, swapping operands and multiple
-  // operations may be required for some comparisons.
-  unsigned Opc = (Cond == ISD::SETEQ || Cond == ISD::SETNE) ? X86ISD::PCMPEQ
-                                                            : X86ISD::PCMPGT;
-  bool Swap = Cond == ISD::SETLT || Cond == ISD::SETULT ||
-              Cond == ISD::SETGE || Cond == ISD::SETUGE;
-  bool Invert = Cond == ISD::SETNE ||
-                (Cond != ISD::SETEQ && ISD::isTrueWhenEqual(Cond));
-
   // If both operands are known non-negative, then an unsigned compare is the
   // same as a signed compare and there's no need to flip signbits.
   // TODO: We could check for more general simplifications here since we're
   // computing known bits.
   bool FlipSigns = ISD::isUnsignedIntSetCC(Cond) &&
                    !(DAG.SignBitIsZero(Op0) && DAG.SignBitIsZero(Op1));
 
-  // Special case: Use min/max operations for SETULE/SETUGE
+  // Special case: Use min/max operations for unsigned compares. We only want
+  // to do this for unsigned compares if we need to flip signs or if it allows
+  // use to avoid an invert.
   MVT VET = VT.getVectorElementType();
-  bool HasMinMax =
-      (Subtarget.hasAVX512() && VET == MVT::i64) ||
-      (Subtarget.hasSSE41() && (VET == MVT::i16 || VET == MVT::i32)) ||
-      (Subtarget.hasSSE2() && (VET == MVT::i8));
-  bool MinMax = false;
-  if (HasMinMax) {
+  bool UseMinMax = ISD::isUnsignedIntSetCC(Cond) &&
+                   (FlipSigns || ISD::isTrueWhenEqual(Cond)) &&
+                   ((Subtarget.hasAVX512() && VET == MVT::i64) ||
+                    (Subtarget.hasSSE41() && (VET == MVT::i16 ||
+                                              VET == MVT::i32)) ||
+                    (Subtarget.hasSSE2() && (VET == MVT::i8)));
+  if (UseMinMax) {
+    bool Invert = false;
+    unsigned Opc;
     switch (Cond) {
-    default: break;
-    case ISD::SETULE: Opc = ISD::UMIN; MinMax = true; break;
-    case ISD::SETUGE: Opc = ISD::UMAX; MinMax = true; break;
+    default: llvm_unreachable("Unexpected condition code");
+    case ISD::SETUGT: Invert = true; LLVM_FALLTHROUGH;
+    case ISD::SETULE: Opc = ISD::UMIN; break;
+    case ISD::SETULT: Invert = true; LLVM_FALLTHROUGH;
+    case ISD::SETUGE: Opc = ISD::UMAX; break;
     }
 
-    if (MinMax)
-      Swap = Invert = FlipSigns = false;
+    SDValue Result = DAG.getNode(Opc, dl, VT, Op0, Op1);
+    Result = DAG.getNode(X86ISD::PCMPEQ, dl, VT, Op0, Result);
+
+    // If the logical-not of the result is required, perform that now.
+    if (Invert)
+      Result = DAG.getNOT(dl, Result, VT);
+
+    return Result;
   }
 
+  // We are handling one of the integer comparisons here. Since SSE only has
+  // GT and EQ comparisons for integer, swapping operands and multiple
+  // operations may be required for some comparisons.
+  unsigned Opc = (Cond == ISD::SETEQ || Cond == ISD::SETNE) ? X86ISD::PCMPEQ
+                                                            : X86ISD::PCMPGT;
+  bool Swap = Cond == ISD::SETLT || Cond == ISD::SETULT ||
+              Cond == ISD::SETGE || Cond == ISD::SETUGE;
+  bool Invert = Cond == ISD::SETNE ||
+                (Cond != ISD::SETEQ && ISD::isTrueWhenEqual(Cond));
+
   bool HasSubus = Subtarget.hasSSE2() && (VET == MVT::i8 || VET == MVT::i16);
   bool Subus = false;
-  if (!MinMax && HasSubus) {
+  if (HasSubus) {
     // As another special case, use PSUBUS[BW] when it's profitable. E.g. for
     // Op0 u<= Op1:
     //   t = psubus Op0, Op1
     //   pcmpeq t, <0..0>
     switch (Cond) {
     default: break;
     case ISD::SETULT: {
       // If the comparison is against a constant we can turn this into a
#endif
   }
 
   SDValue Result = DAG.getNode(Opc, dl, VT, Op0, Op1);
 
   // If the logical-not of the result is required, perform that now.
   if (Invert)
     Result = DAG.getNOT(dl, Result, VT);
 
-  if (MinMax)
-    Result = DAG.getNode(X86ISD::PCMPEQ, dl, VT, Op0, Result);
-
   if (Subus)
     Result = DAG.getNode(X86ISD::PCMPEQ, dl, VT, Result,
                          getZeroVector(VT, Subtarget, DAG, dl));
 
   return Result;
 }
 
 SDValue X86TargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) const {

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

; SKX-NEXT:    retq
   %t2 = extractelement <4 x i1> %t1, i32 3
   %res = zext i1 %t2 to i8
   ret i8 %res
 }
 
 define zeroext i8 @test_extractelement_v32i1(<32 x i8> %a, <32 x i8> %b) {
 ; KNL-LABEL: test_extractelement_v32i1:
 ; KNL:       ## %bb.0:
-; KNL-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; KNL-NEXT:    vpxor %ymm2, %ymm1, %ymm1
-; KNL-NEXT:    vpxor %ymm2, %ymm0, %ymm0
-; KNL-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
+; KNL-NEXT:    vpminub %ymm1, %ymm0, %ymm1
+; KNL-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
+; KNL-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
+; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; KNL-NEXT:    kshiftrw $2, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    andl $1, %eax
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test_extractelement_v32i1:
 ; SKX:       ## %bb.0:
 ; SKX-NEXT:    vpcmpnleub %ymm1, %ymm0, %k0
 ; SKX-NEXT:    kshiftrd $2, %k0, %k0
 ; SKX-NEXT:    kmovd %k0, %eax
 ; SKX-NEXT:    andl $1, %eax
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
   %t1 = icmp ugt <32 x i8> %a, %b
   %t2 = extractelement <32 x i1> %t1, i32 2
   %res = zext i1 %t2 to i8
   ret i8 %res
 }
 
 define zeroext i8 @test_extractelement_v64i1(<64 x i8> %a, <64 x i8> %b) {
 ; KNL-LABEL: test_extractelement_v64i1:
 ; KNL:       ## %bb.0:
-; KNL-NEXT:    vmovdqa {{.*#+}} ymm0 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; KNL-NEXT:    vpxor %ymm0, %ymm3, %ymm2
-; KNL-NEXT:    vpxor %ymm0, %ymm1, %ymm0
-; KNL-NEXT:    vpcmpgtb %ymm2, %ymm0, %ymm0
+; KNL-NEXT:    vpminub %ymm3, %ymm1, %ymm0
+; KNL-NEXT:    vpcmpeqb %ymm0, %ymm1, %ymm0
+; KNL-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
+; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    andb $1, %al
 ; KNL-NEXT:    movb $4, %cl
 ; KNL-NEXT:    subb %al, %cl
 ; KNL-NEXT:    movzbl %cl, %eax
 ; KNL-NEXT:    vzeroupper
; SKX-NEXT:    retq
   %t2 = extractelement <64 x i1> %t1, i32 63
   %res = select i1 %t2, i8 3, i8 4
   ret i8 %res
 }
 
 define zeroext i8 @extractelement_v64i1_alt(<64 x i8> %a, <64 x i8> %b) {
 ; KNL-LABEL: extractelement_v64i1_alt:
 ; KNL:       ## %bb.0:
-; KNL-NEXT:    vmovdqa {{.*#+}} ymm0 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; KNL-NEXT:    vpxor %ymm0, %ymm3, %ymm2
-; KNL-NEXT:    vpxor %ymm0, %ymm1, %ymm0
-; KNL-NEXT:    vpcmpgtb %ymm2, %ymm0, %ymm0
+; KNL-NEXT:    vpminub %ymm3, %ymm1, %ymm0
+; KNL-NEXT:    vpcmpeqb %ymm0, %ymm1, %ymm0
+; KNL-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
+; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    andb $1, %al
 ; KNL-NEXT:    movb $4, %cl
 ; KNL-NEXT:    subb %al, %cl
 ; KNL-NEXT:    movzbl %cl, %eax
 ; KNL-NEXT:    vzeroupper

 ; KNL-NEXT:    pushq %rbp
 ; KNL-NEXT:    .cfi_def_cfa_offset 16
 ; KNL-NEXT:    .cfi_offset %rbp, -16
 ; KNL-NEXT:    movq %rsp, %rbp
 ; KNL-NEXT:    .cfi_def_cfa_register %rbp
 ; KNL-NEXT:    andq $-32, %rsp
 ; KNL-NEXT:    subq $64, %rsp
 ; KNL-NEXT:    ## kill: def $edi killed $edi def $rdi
-; KNL-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; KNL-NEXT:    vpxor %ymm2, %ymm1, %ymm1
-; KNL-NEXT:    vpxor %ymm2, %ymm0, %ymm0
-; KNL-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
+; KNL-NEXT:    vpminub %ymm1, %ymm0, %ymm1
+; KNL-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
+; KNL-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; KNL-NEXT:    vmovdqa %ymm0, (%rsp)
 ; KNL-NEXT:    andl $31, %edi
 ; KNL-NEXT:    movzbl (%rsp,%rdi), %eax
 ; KNL-NEXT:    andl $1, %eax
 ; KNL-NEXT:    movq %rbp, %rsp
 ; KNL-NEXT:    popq %rbp
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq

 ; KNL-NEXT:    pushq %rbp
 ; KNL-NEXT:    .cfi_def_cfa_offset 16
 ; KNL-NEXT:    .cfi_offset %rbp, -16
 ; KNL-NEXT:    movq %rsp, %rbp
 ; KNL-NEXT:    .cfi_def_cfa_register %rbp
 ; KNL-NEXT:    andq $-32, %rsp
 ; KNL-NEXT:    subq $64, %rsp
 ; KNL-NEXT:    ## kill: def $esi killed $esi def $rsi
-; KNL-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; KNL-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; KNL-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
+; KNL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; KNL-NEXT:    vpminub %ymm1, %ymm0, %ymm1
+; KNL-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
+; KNL-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; KNL-NEXT:    andl $31, %esi
 ; KNL-NEXT:    testb %dil, %dil
 ; KNL-NEXT:    vmovdqa %ymm0, (%rsp)
 ; KNL-NEXT:    setne (%rsp,%rsi)
 ; KNL-NEXT:    vpmovsxbd (%rsp), %zmm0
 ; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; KNL-NEXT:    kmovw %k0, %ecx

 ; KNL-NEXT:    pushq %rbp
 ; KNL-NEXT:    .cfi_def_cfa_offset 16
 ; KNL-NEXT:    .cfi_offset %rbp, -16
 ; KNL-NEXT:    movq %rsp, %rbp
 ; KNL-NEXT:    .cfi_def_cfa_register %rbp
 ; KNL-NEXT:    andq $-64, %rsp
 ; KNL-NEXT:    subq $128, %rsp
 ; KNL-NEXT:    ## kill: def $esi killed $esi def $rsi
-; KNL-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; KNL-NEXT:    vpxor %ymm2, %ymm0, %ymm0
-; KNL-NEXT:    vpcmpgtb %ymm2, %ymm0, %ymm0
-; KNL-NEXT:    vpxor %ymm2, %ymm1, %ymm1
-; KNL-NEXT:    vpcmpgtb %ymm2, %ymm1, %ymm1
+; KNL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; KNL-NEXT:    vpminub %ymm2, %ymm0, %ymm3
+; KNL-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm0
+; KNL-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
+; KNL-NEXT:    vpminub %ymm2, %ymm1, %ymm2
+; KNL-NEXT:    vpcmpeqb %ymm2, %ymm1, %ymm1
+; KNL-NEXT:    vpternlogq $15, %zmm1, %zmm1, %zmm1
 ; KNL-NEXT:    andl $63, %esi
 ; KNL-NEXT:    testb %dil, %dil
 ; KNL-NEXT:    vmovdqa %ymm1, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    vmovdqa %ymm0, (%rsp)
 ; KNL-NEXT:    setne (%rsp,%rsi)
 ; KNL-NEXT:    vmovdqa (%rsp), %ymm0
 ; KNL-NEXT:    vmovdqa {{[0-9]+}}(%rsp), %ymm1
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm2

 ; KNL-NEXT:    vpinsrb $9, 168(%rbp), %xmm3, %xmm3
 ; KNL-NEXT:    vpinsrb $10, 176(%rbp), %xmm3, %xmm3
 ; KNL-NEXT:    vpinsrb $11, 184(%rbp), %xmm3, %xmm3
 ; KNL-NEXT:    vpinsrb $12, 192(%rbp), %xmm3, %xmm3
 ; KNL-NEXT:    vpinsrb $13, 200(%rbp), %xmm3, %xmm3
 ; KNL-NEXT:    vpinsrb $14, 208(%rbp), %xmm3, %xmm3
 ; KNL-NEXT:    vpinsrb $15, 216(%rbp), %xmm3, %xmm3
 ; KNL-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
-; KNL-NEXT:    vmovdqa {{.*#+}} ymm3 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; KNL-NEXT:    vpxor %ymm3, %ymm2, %ymm2
-; KNL-NEXT:    vpcmpgtb %ymm3, %ymm2, %ymm2
-; KNL-NEXT:    vpxor %ymm3, %ymm1, %ymm1
-; KNL-NEXT:    vpcmpgtb %ymm3, %ymm1, %ymm1
-; KNL-NEXT:    vpxor %ymm3, %ymm0, %ymm0
-; KNL-NEXT:    vpcmpgtb %ymm3, %ymm0, %ymm0
 ; KNL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; KNL-NEXT:    vpminub %ymm3, %ymm2, %ymm4
+; KNL-NEXT:    vpcmpeqb %ymm4, %ymm2, %ymm2
+; KNL-NEXT:    vpternlogq $15, %zmm2, %zmm2, %zmm2
+; KNL-NEXT:    vpminub %ymm3, %ymm1, %ymm4
+; KNL-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm1
+; KNL-NEXT:    vpternlogq $15, %zmm1, %zmm1, %zmm1
+; KNL-NEXT:    vpminub %ymm3, %ymm0, %ymm4
+; KNL-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm0
+; KNL-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; KNL-NEXT:    movl 744(%rbp), %eax
 ; KNL-NEXT:    andl $127, %eax
 ; KNL-NEXT:    cmpb $0, 736(%rbp)
 ; KNL-NEXT:    vmovdqa %ymm3, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    vmovdqa %ymm0, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    vmovdqa %ymm1, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    vmovdqa %ymm2, (%rsp)
 ; KNL-NEXT:    setne (%rsp,%rax)

 ; KNL-NEXT:    pushq %rbp
 ; KNL-NEXT:    .cfi_def_cfa_offset 16
 ; KNL-NEXT:    .cfi_offset %rbp, -16
 ; KNL-NEXT:    movq %rsp, %rbp
 ; KNL-NEXT:    .cfi_def_cfa_register %rbp
 ; KNL-NEXT:    andq $-128, %rsp
 ; KNL-NEXT:    subq $256, %rsp ## imm = 0x100
 ; KNL-NEXT:    ## kill: def $esi killed $esi def $rsi
-; KNL-NEXT:    vmovdqa {{.*#+}} ymm4 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; KNL-NEXT:    vpxor %ymm4, %ymm0, %ymm0
-; KNL-NEXT:    vpcmpgtb %ymm4, %ymm0, %ymm0
-; KNL-NEXT:    vpxor %ymm4, %ymm1, %ymm1
-; KNL-NEXT:    vpcmpgtb %ymm4, %ymm1, %ymm1
-; KNL-NEXT:    vpxor %ymm4, %ymm2, %ymm2
-; KNL-NEXT:    vpcmpgtb %ymm4, %ymm2, %ymm2
-; KNL-NEXT:    vpxor %ymm4, %ymm3, %ymm3
-; KNL-NEXT:    vpcmpgtb %ymm4, %ymm3, %ymm3
+; KNL-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; KNL-NEXT:    vpminub %ymm4, %ymm0, %ymm5
+; KNL-NEXT:    vpcmpeqb %ymm5, %ymm0, %ymm0
+; KNL-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
+; KNL-NEXT:    vpminub %ymm4, %ymm1, %ymm5
+; KNL-NEXT:    vpcmpeqb %ymm5, %ymm1, %ymm1
+; KNL-NEXT:    vpternlogq $15, %zmm1, %zmm1, %zmm1
+; KNL-NEXT:    vpminub %ymm4, %ymm2, %ymm5
+; KNL-NEXT:    vpcmpeqb %ymm5, %ymm2, %ymm2
+; KNL-NEXT:    vpternlogq $15, %zmm2, %zmm2, %zmm2
+; KNL-NEXT:    vpminub %ymm4, %ymm3, %ymm4
+; KNL-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm3
+; KNL-NEXT:    vpternlogq $15, %zmm3, %zmm3, %zmm3
 ; KNL-NEXT:    andl $127, %esi
 ; KNL-NEXT:    testb %dil, %dil
 ; KNL-NEXT:    vmovdqa %ymm3, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    vmovdqa %ymm2, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    vmovdqa %ymm1, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    vmovdqa %ymm0, (%rsp)
 ; KNL-NEXT:    setne (%rsp,%rsi)
 ; KNL-NEXT:    vmovdqa (%rsp), %ymm2

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

 ; VLX-LABEL: test_vpcmpultb_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltub %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp ult <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpultb_v16i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultb_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltub (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor (%rdi), %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxub (%rdi), %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <16 x i8>
   %2 = icmp ult <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultb_v16i1_v32i1_mask(i16 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultb_v16i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltub %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltub (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor (%rsi), %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxub (%rsi), %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX-LABEL: test_vpcmpultb_v16i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltub %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>
   %2 = icmp ult <16 x i8> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpultb_v16i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultb_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltub (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor (%rdi), %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxub (%rdi), %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltub %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %1 = bitcast <2 x i64> %__b to <16 x i8>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltub (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor (%rsi), %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxub (%rsi), %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <16 x i8>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltub %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmaxub %ymm1, %ymm0, %ymm1
+; NoVLX-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
+; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    shll $16, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltub (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultb_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor (%rdi), %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmaxub (%rdi), %ymm0, %ymm1
+; NoVLX-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
+; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    shll $16, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <32 x i8>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltub %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v32i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmaxub %ymm1, %ymm0, %ymm1
+; NoVLX-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
+; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    shrl $16, %edi
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    andl %edi, %ecx
 ; NoVLX-NEXT:    shll $16, %ecx
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltub (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultb_v32i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor (%rsi), %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmaxub (%rsi), %ymm0, %ymm1
+; NoVLX-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm1
+; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
 ; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    shrl $16, %edi
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm0
 ; NoVLX-NEXT:    vpmovsxbd %xmm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    andl %edi, %ecx
 ; NoVLX-NEXT:    shll $16, %ecx
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v8i1_v16i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
+; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp ult <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <16 x i1> %3 to i16
   ret i16 %4
 }
 
 define zeroext i16 @test_vpcmpultw_v8i1_v16i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultw_v8i1_v16i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v8i1_v16i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor (%rdi), %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxuw (%rdi), %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
+; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v8i1_v16i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
+; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v8i1_v16i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor (%rsi), %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxuw (%rsi), %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
+; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>

 ; VLX-LABEL: test_vpcmpultw_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
+; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp ult <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpultw_v8i1_v32i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultw_v8i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor (%rdi), %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxuw (%rdi), %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
+; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b
   %1 = bitcast <2 x i64> %load to <8 x i16>
   %2 = icmp ult <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_masked_vpcmpultw_v8i1_v32i1_mask(i8 zeroext %__u, <2 x i64> %__a, <2 x i64> %__b) local_unnamed_addr {
 ; VLX-LABEL: test_masked_vpcmpultw_v8i1_v32i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v8i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
+; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v8i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor (%rsi), %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxuw (%rsi), %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
+; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX-LABEL: test_vpcmpultw_v8i1_v64i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw %xmm1, %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
+; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %1 = bitcast <2 x i64> %__b to <8 x i16>
   %2 = icmp ult <8 x i16> %0, %1
   %3 = shufflevector <8 x i1> %2, <8 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpultw_v8i1_v64i1_mask_mem(<2 x i64> %__a, <2 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultw_v8i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw (%rdi), %xmm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor (%rdi), %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxuw (%rdi), %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
+; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>
   %load = load <2 x i64>, <2 x i64>* %__b

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw %xmm1, %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v8i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
+; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw (%rsi), %xmm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v8i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; NoVLX-NEXT:    vpxor (%rsi), %xmm1, %xmm1
-; NoVLX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
+; NoVLX-NEXT:    vpmaxuw (%rsi), %xmm0, %xmm1
+; NoVLX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwq %xmm0, %zmm0
+; NoVLX-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; NoVLX-NEXT:    kmovw %edi, %k1
 ; NoVLX-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <2 x i64> %__a to <8 x i16>

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm2 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm1
+; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp ult <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <32 x i1> %3 to i32
   ret i32 %4
 }
 
 define zeroext i32 @test_vpcmpultw_v16i1_v32i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultw_v16i1_v32i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm1 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor (%rdi), %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmaxuw (%rdi), %ymm0, %ymm1
+; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b
   %1 = bitcast <4 x i64> %load to <16 x i16>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v16i1_v32i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm2 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm1
+; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovd %k0, %eax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v16i1_v32i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm1 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor (%rsi), %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmaxuw (%rsi), %ymm0, %ymm1
+; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b

 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw %ymm1, %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm2 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm1
+; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>
   %2 = icmp ult <16 x i16> %0, %1
   %3 = shufflevector <16 x i1> %2, <16 x i1> zeroinitializer, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
   %4 = bitcast <64 x i1> %3 to i64
   ret i64 %4
 }
 
 define zeroext i64 @test_vpcmpultw_v16i1_v64i1_mask_mem(<4 x i64> %__a, <4 x i64>* %__b) local_unnamed_addr {
 ; VLX-LABEL: test_vpcmpultw_v16i1_v64i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuw (%rdi), %ymm0, %k0
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultw_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm1 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor (%rdi), %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmaxuw (%rdi), %ymm0, %ymm1
+; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw %ymm1, %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v16i1_v64i1_mask:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm2 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm1
+; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %1 = bitcast <4 x i64> %__b to <16 x i16>

 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuw (%rsi), %ymm0, %k0 {%k1}
 ; VLX-NEXT:    kmovq %k0, %rax
 ; VLX-NEXT:    vzeroupper
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultw_v16i1_v64i1_mask_mem:
 ; NoVLX:       # %bb.0: # %entry
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm1 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor (%rsi), %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmaxuw (%rsi), %ymm0, %ymm1
+; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <4 x i64> %__a to <16 x i16>
   %load = load <4 x i64>, <4 x i64>* %__b

 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
 ; NoVLX-NEXT:    vinserti128 $1, %xmm5, %ymm2, %ymm2
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm4 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %ymm4, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor %ymm4, %ymm2, %ymm2
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm2, %ymm0
+; NoVLX-NEXT:    vpmaxuw %ymm2, %ymm0, %ymm2
+; NoVLX-NEXT:    vpcmpeqw %ymm2, %ymm0, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    vpxor %ymm4, %ymm1, %ymm0
-; NoVLX-NEXT:    vpxor %ymm4, %ymm3, %ymm1
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmaxuw %ymm3, %ymm1, %ymm0
+; NoVLX-NEXT:    vpcmpeqw %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    shll $16, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>

 ; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm0
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vinserti128 $1, %xmm1, %ymm2, %ymm1
+; NoVLX-NEXT:    shrq $48, %rax
 ; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
 ; NoVLX-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm2 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor (%rdi), %ymm2, %ymm3
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm3, %ymm0
+; NoVLX-NEXT:    vpmaxuw (%rdi), %ymm0, %ymm2
+; NoVLX-NEXT:    vpcmpeqw %ymm2, %ymm0, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
-; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm0
-; NoVLX-NEXT:    vpxor 32(%rdi), %ymm2, %ymm1
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmaxuw 32(%rdi), %ymm1, %ymm0
+; NoVLX-NEXT:    vpcmpeqw %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    shll $16, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq
 entry:
   %0 = bitcast <8 x i64> %__a to <32 x i16>

 ; NoVLX:       # %bb.0: # %entry
 ; NoVLX-NEXT:    vextracti128 $1, %ymm1, %xmm9
 ; NoVLX-NEXT:    vextracti32x4 $3, %zmm0, %xmm3
 ; NoVLX-NEXT:    vmovq %xmm3, %rax
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    movq %rax, %rdx
 ; NoVLX-NEXT:    vmovd %eax, %xmm4
 ; NoVLX-NEXT:    shrl $16, %eax
-; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm5
+; NoVLX-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
 ; NoVLX-NEXT:    vmovq %xmm9, %rax
-; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm4
+; NoVLX-NEXT:    vextracti32x4 $2, %zmm1, %xmm5
 ; NoVLX-NEXT:    vextracti32x4 $3, %zmm1, %xmm7
 ; NoVLX-NEXT:    vextracti128 $1, %ymm0, %xmm6
 ; NoVLX-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm4, %xmm4
 ; NoVLX-NEXT:    vpextrq $1, %xmm3, %rdx
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm3
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm3
 ; NoVLX-NEXT:    movl %edx, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %edx, %xmm3, %xmm3
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    movq %rdx, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm3, %xmm3
 ; NoVLX-NEXT:    vmovq %xmm2, %rcx
 ; NoVLX-NEXT:    shrq $48, %rdx
 ; NoVLX-NEXT:    vpinsrw $7, %edx, %xmm3, %xmm10
 ; NoVLX-NEXT:    movl %ecx, %edx
 ; NoVLX-NEXT:    shrl $16, %edx
-; NoVLX-NEXT:    vmovd %ecx, %xmm5
-; NoVLX-NEXT:    vpinsrw $1, %edx, %xmm5, %xmm5
+; NoVLX-NEXT:    vmovd %ecx, %xmm4
+; NoVLX-NEXT:    vpinsrw $1, %edx, %xmm4, %xmm4
 ; NoVLX-NEXT:    movq %rcx, %rdx
 ; NoVLX-NEXT:    shrq $32, %rdx
-; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm5, %xmm5
+; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm4, %xmm4
 ; NoVLX-NEXT:    vpextrq $1, %xmm2, %rdx
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm5, %xmm2
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm4, %xmm2
 ; NoVLX-NEXT:    movl %edx, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %edx, %xmm2, %xmm2
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rdx, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    vmovq %xmm6, %rcx
 ; NoVLX-NEXT:    shrq $48, %rdx
-; NoVLX-NEXT:    vpinsrw $7, %edx, %xmm2, %xmm5
+; NoVLX-NEXT:    vpinsrw $7, %edx, %xmm2, %xmm4
 ; NoVLX-NEXT:    movl %ecx, %edx
 ; NoVLX-NEXT:    shrl $16, %edx
 ; NoVLX-NEXT:    vmovd %ecx, %xmm2
 ; NoVLX-NEXT:    vpinsrw $1, %edx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rdx
 ; NoVLX-NEXT:    shrq $32, %rdx
 ; NoVLX-NEXT:    vpinsrw $2, %edx, %xmm2, %xmm2
 ; NoVLX-NEXT:    vpextrq $1, %xmm6, %rdx

 ; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movl %edx, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
 ; NoVLX-NEXT:    vpinsrw $4, %edx, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rdx, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
 ; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm0, %xmm0
-; NoVLX-NEXT:    vmovq %xmm4, %rcx
+; NoVLX-NEXT:    vmovq %xmm5, %rcx
 ; NoVLX-NEXT:    shrq $48, %rdx
 ; NoVLX-NEXT:    vpinsrw $7, %edx, %xmm0, %xmm0
 ; NoVLX-NEXT:    movl %ecx, %edx
 ; NoVLX-NEXT:    shrl $16, %edx
 ; NoVLX-NEXT:    vmovd %ecx, %xmm2
 ; NoVLX-NEXT:    vpinsrw $1, %edx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rcx, %rdx
 ; NoVLX-NEXT:    shrq $32, %rdx

 ; NoVLX-NEXT:    shrl $16, %eax
 ; NoVLX-NEXT:    vpinsrw $4, %edx, %xmm1, %xmm1
 ; NoVLX-NEXT:    vpinsrw $5, %eax, %xmm1, %xmm1
 ; NoVLX-NEXT:    movq %rdx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $6, %eax, %xmm1, %xmm1
 ; NoVLX-NEXT:    shrq $48, %rdx
 ; NoVLX-NEXT:    vpinsrw $7, %edx, %xmm1, %xmm1
-; NoVLX-NEXT:    vpextrq $1, %xmm4, %rax
+; NoVLX-NEXT:    vpextrq $1, %xmm5, %rax
 ; NoVLX-NEXT:    vinserti128 $1, %xmm6, %ymm8, %ymm3
 ; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm2 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %ymm2, %ymm3, %ymm3
-; NoVLX-NEXT:    vpxor %ymm2, %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtw %ymm3, %ymm1, %ymm1
-; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
-; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
+; NoVLX-NEXT:    vpmaxuw %ymm1, %ymm3, %ymm1
+; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm3, %ymm1
 ; NoVLX-NEXT:    shrq $48, %rcx
-; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm7, %xmm1
+; NoVLX-NEXT:    vpinsrw $3, %ecx, %xmm7, %xmm2
 ; NoVLX-NEXT:    movl %eax, %ecx
 ; NoVLX-NEXT:    shrl $16, %ecx
-; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $4, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vpinsrw $5, %ecx, %xmm2, %xmm2
 ; NoVLX-NEXT:    movq %rax, %rcx
 ; NoVLX-NEXT:    shrq $32, %rcx
-; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm1, %xmm1
+; NoVLX-NEXT:    vpinsrw $6, %ecx, %xmm2, %xmm2
+; NoVLX-NEXT:    vpternlogq $15, %zmm1, %zmm1, %zmm1
+; NoVLX-NEXT:    vpmovsxwd %ymm1, %zmm1
+; NoVLX-NEXT:    vpslld $31, %zmm1, %zmm1
+; NoVLX-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    andl %edi, %ecx
 ; NoVLX-NEXT:    shrl $16, %edi
 ; NoVLX-NEXT:    shrq $48, %rax
-; NoVLX-NEXT:    vinserti128 $1, %xmm10, %ymm5, %ymm3
-; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm1
-; NoVLX-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
-; NoVLX-NEXT:    vpxor %ymm2, %ymm3, %ymm1
-; NoVLX-NEXT:    vpxor %ymm2, %ymm0, %ymm0
-; NoVLX-NEXT:    vpcmpgtw %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm10, %ymm4, %ymm1
+; NoVLX-NEXT:    vpinsrw $7, %eax, %xmm2, %xmm2
+; NoVLX-NEXT:    vinserti128 $1, %xmm0, %ymm2, %ymm0
+; NoVLX-NEXT:    vpmaxuw %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpcmpeqw %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %edx
 ; NoVLX-NEXT:    andl %edi, %edx
 ; NoVLX-NEXT:    shll $16, %edx
 ; NoVLX-NEXT:    movzwl %cx, %eax
 ; NoVLX-NEXT:    orl %edx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq

 ; NoVLX-NEXT:    vpinsrw $4, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpinsrw $5, %eax, %xmm0, %xmm0
 ; NoVLX-NEXT:    movq %rcx, %rax
 ; NoVLX-NEXT:    shrq $32, %rax
 ; NoVLX-NEXT:    vpinsrw $6, %eax, %xmm0, %xmm0
 ; NoVLX-NEXT:    shrq $48, %rcx
 ; NoVLX-NEXT:    vpinsrw $7, %ecx, %xmm0, %xmm0
 ; NoVLX-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
-; NoVLX-NEXT:    vmovdqa {{.*#+}} ymm1 = [32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768,32768]
-; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor (%rsi), %ymm1, %ymm4
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm4, %ymm0
+; NoVLX-NEXT:    vpmaxuw (%rsi), %ymm0, %ymm1
+; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    shrq $48, %rdx
-; NoVLX-NEXT:    vpinsrw $7, %edx, %xmm3, %xmm3
+; NoVLX-NEXT:    vpinsrw $7, %edx, %xmm3, %xmm1
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %eax
 ; NoVLX-NEXT:    andl %edi, %eax
 ; NoVLX-NEXT:    shrl $16, %edi
-; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm3, %ymm0
-; NoVLX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
-; NoVLX-NEXT:    vpxor 32(%rsi), %ymm1, %ymm1
-; NoVLX-NEXT:    vpcmpgtw %ymm0, %ymm1, %ymm0
+; NoVLX-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm0
+; NoVLX-NEXT:    vpmaxuw 32(%rsi), %ymm0, %ymm1
+; NoVLX-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
+; NoVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; NoVLX-NEXT:    vpmovsxwd %ymm0, %zmm0
+; NoVLX-NEXT:    vpslld $31, %zmm0, %zmm0
 ; NoVLX-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; NoVLX-NEXT:    kmovw %k0, %ecx
 ; NoVLX-NEXT:    andl %edi, %ecx
 ; NoVLX-NEXT:    shll $16, %ecx
 ; NoVLX-NEXT:    movzwl %ax, %eax
 ; NoVLX-NEXT:    orl %ecx, %eax
 ; NoVLX-NEXT:    vzeroupper
 ; NoVLX-NEXT:    retq

test/CodeGen/X86/psubus.ll

 ; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [32767,32767,32767,32767,32767,32767,32767,32767]
 ; SSE-NEXT:    psubusw %xmm2, %xmm0
 ; SSE-NEXT:    psubusw %xmm2, %xmm1
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test8:
 ; AVX1:       # %bb.0: # %vector.ph
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX1-NEXT:    vpxor %xmm2, %xmm1, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [65534,65534,65534,65534,65534,65534,65534,65534]
-; AVX1-NEXT:    vpcmpgtw %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpxor %xmm2, %xmm0, %xmm2
-; AVX1-NEXT:    vpcmpgtw %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [32766,32766,32766,32766,32766,32766,32766,32766]
+; AVX1-NEXT:    vpminuw %xmm2, %xmm1, %xmm3
+; AVX1-NEXT:    vpcmpeqw %xmm3, %xmm1, %xmm3
+; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpxor %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpminuw %xmm2, %xmm0, %xmm2
+; AVX1-NEXT:    vpcmpeqw %xmm2, %xmm0, %xmm2
+; AVX1-NEXT:    vpxor %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [32769,32769,32769,32769,32769,32769,32769,32769]
 ; AVX1-NEXT:    vpaddw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpaddw %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vandps %ymm0, %ymm2, %ymm0
 ; AVX1-NEXT:    retq
 ;

 ; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
 ; SSE-NEXT:    psubusb %xmm2, %xmm0
 ; SSE-NEXT:    psubusb %xmm2, %xmm1
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test11:
 ; AVX1:       # %bb.0: # %vector.ph
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX1-NEXT:    vpxor %xmm2, %xmm1, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254]
-; AVX1-NEXT:    vpcmpgtb %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpxor %xmm2, %xmm0, %xmm2
-; AVX1-NEXT:    vpcmpgtb %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [126,126,126,126,126,126,126,126,126,126,126,126,126,126,126,126]
+; AVX1-NEXT:    vpminub %xmm2, %xmm1, %xmm3
+; AVX1-NEXT:    vpcmpeqb %xmm3, %xmm1, %xmm3
+; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpxor %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm2
+; AVX1-NEXT:    vpcmpeqb %xmm2, %xmm0, %xmm2
+; AVX1-NEXT:    vpxor %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [129,129,129,129,129,129,129,129,129,129,129,129,129,129,129,129]
 ; AVX1-NEXT:    vpaddb %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpaddb %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vandps %ymm0, %ymm2, %ymm0
 ; AVX1-NEXT:    retq
 ;

 ; SSSE3-NEXT:    movdqa %xmm2, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: test13:
 ; SSE41:       # %bb.0: # %vector.ph
 ; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[2,3,0,1]
 ; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero
 ; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm4 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [2147483648,2147483648,2147483648,2147483648]
-; SSE41-NEXT:    movdqa %xmm4, %xmm6
-; SSE41-NEXT:    psubd %xmm1, %xmm4
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    movdqa %xmm4, %xmm0
+; SSE41-NEXT:    pmaxud %xmm1, %xmm0
+; SSE41-NEXT:    pcmpeqd %xmm4, %xmm0
+; SSE41-NEXT:    pcmpeqd %xmm5, %xmm5
 ; SSE41-NEXT:    pxor %xmm5, %xmm0
-; SSE41-NEXT:    por %xmm5, %xmm6
-; SSE41-NEXT:    pcmpgtd %xmm6, %xmm0
-; SSE41-NEXT:    movdqa %xmm2, %xmm1
-; SSE41-NEXT:    pxor %xmm5, %xmm1
-; SSE41-NEXT:    por %xmm3, %xmm5
-; SSE41-NEXT:    pcmpgtd %xmm5, %xmm1
-; SSE41-NEXT:    packssdw %xmm1, %xmm0
+; SSE41-NEXT:    movdqa {{.*#+}} xmm6 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; SSE41-NEXT:    pshufb %xmm6, %xmm0
+; SSE41-NEXT:    movdqa %xmm3, %xmm7
+; SSE41-NEXT:    pmaxud %xmm2, %xmm7
+; SSE41-NEXT:    pcmpeqd %xmm3, %xmm7
+; SSE41-NEXT:    pxor %xmm5, %xmm7
+; SSE41-NEXT:    pshufb %xmm6, %xmm7
+; SSE41-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm7[0]
 ; SSE41-NEXT:    psubd %xmm2, %xmm3
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; SSE41-NEXT:    pshufb %xmm1, %xmm4
-; SSE41-NEXT:    pshufb %xmm1, %xmm3
+; SSE41-NEXT:    psubd %xmm1, %xmm4
+; SSE41-NEXT:    pshufb %xmm6, %xmm4
+; SSE41-NEXT:    pshufb %xmm6, %xmm3
 ; SSE41-NEXT:    punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm3[0]
 ; SSE41-NEXT:    pandn %xmm4, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test13:
 ; AVX1:       # %bb.0: # %vector.ph
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
 ; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [2147483648,2147483648,2147483648,2147483648]
-; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm4
-; AVX1-NEXT:    vpxor %xmm3, %xmm1, %xmm5
-; AVX1-NEXT:    vpcmpgtd %xmm4, %xmm5, %xmm4
+; AVX1-NEXT:    vpmaxud %xmm1, %xmm0, %xmm3
+; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm0, %xmm3
+; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpxor %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX1-NEXT:    vpxor %xmm3, %xmm5, %xmm6
-; AVX1-NEXT:    vpor %xmm3, %xmm2, %xmm3
-; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm6, %xmm3
-; AVX1-NEXT:    vpackssdw %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpmaxud %xmm5, %xmm2, %xmm6
+; AVX1-NEXT:    vpcmpeqd %xmm6, %xmm2, %xmm6
+; AVX1-NEXT:    vpxor %xmm4, %xmm6, %xmm4
+; AVX1-NEXT:    vpackssdw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsubd %xmm5, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX1-NEXT:    vpandn %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test13:
 ; AVX2:       # %bb.0: # %vector.ph
 ; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm2 = [2147483648,2147483648,2147483648,2147483648,2147483648,2147483648,2147483648,2147483648]
-; AVX2-NEXT:    vpxor %ymm2, %ymm1, %ymm3
-; AVX2-NEXT:    vpor %ymm2, %ymm0, %ymm2
-; AVX2-NEXT:    vpcmpgtd %ymm2, %ymm3, %ymm2
+; AVX2-NEXT:    vpmaxud %ymm1, %ymm0, %ymm2
+; AVX2-NEXT:    vpcmpeqd %ymm2, %ymm0, %ymm2
+; AVX2-NEXT:    vpcmpeqd %ymm3, %ymm3, %ymm3
+; AVX2-NEXT:    vpxor %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
 ; AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubd %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-NEXT:    vpandn %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq

 ; SSSE3-NEXT:    pand %xmm5, %xmm1
 ; SSSE3-NEXT:    packuswb %xmm2, %xmm1
 ; SSSE3-NEXT:    packuswb %xmm3, %xmm1
 ; SSSE3-NEXT:    andnpd %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: test14:
 ; SSE41:       # %bb.0: # %vector.ph
-; SSE41-NEXT:    movdqa %xmm0, %xmm5
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[1,1,2,3]
+; SSE41-NEXT:    pmovzxbd {{.*#+}} xmm11 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero,xmm5[2],zero,zero,zero,xmm5[3],zero,zero,zero
 ; SSE41-NEXT:    pmovzxbd {{.*#+}} xmm8 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
-; SSE41-NEXT:    pmovzxbd {{.*#+}} xmm0 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero,xmm5[2],zero,zero,zero,xmm5[3],zero,zero,zero
-; SSE41-NEXT:    pshufd {{.*#+}} xmm6 = xmm5[2,3,0,1]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm6 = xmm0[2,3,0,1]
 ; SSE41-NEXT:    pmovzxbd {{.*#+}} xmm9 = xmm6[0],zero,zero,zero,xmm6[1],zero,zero,zero,xmm6[2],zero,zero,zero,xmm6[3],zero,zero,zero
-; SSE41-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[3,1,2,3]
-; SSE41-NEXT:    pmovzxbd {{.*#+}} xmm6 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero,xmm5[2],zero,zero,zero,xmm5[3],zero,zero,zero
-; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [2147483648,2147483648,2147483648,2147483648]
-; SSE41-NEXT:    movdqa %xmm4, %xmm7
-; SSE41-NEXT:    pxor %xmm5, %xmm7
-; SSE41-NEXT:    psubd %xmm6, %xmm4
-; SSE41-NEXT:    por %xmm5, %xmm6
-; SSE41-NEXT:    pcmpgtd %xmm7, %xmm6
-; SSE41-NEXT:    movdqa {{.*#+}} xmm10 = <u,u,u,u,0,4,8,12,u,u,u,u,u,u,u,u>
-; SSE41-NEXT:    pshufb %xmm10, %xmm6
-; SSE41-NEXT:    movdqa %xmm3, %xmm7
-; SSE41-NEXT:    pxor %xmm5, %xmm7
-; SSE41-NEXT:    psubd %xmm9, %xmm3
-; SSE41-NEXT:    por %xmm5, %xmm9
-; SSE41-NEXT:    pcmpgtd %xmm7, %xmm9
-; SSE41-NEXT:    pshufb %xmm10, %xmm9
-; SSE41-NEXT:    punpckldq {{.*#+}} xmm9 = xmm9[0],xmm6[0],xmm9[1],xmm6[1]
-; SSE41-NEXT:    movdqa %xmm1, %xmm6
-; SSE41-NEXT:    pxor %xmm5, %xmm6
-; SSE41-NEXT:    psubd %xmm0, %xmm1
-; SSE41-NEXT:    por %xmm5, %xmm0
-; SSE41-NEXT:    pcmpgtd %xmm6, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm6 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
-; SSE41-NEXT:    pshufb %xmm6, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; SSE41-NEXT:    pmovzxbd {{.*#+}} xmm10 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
+; SSE41-NEXT:    movdqa %xmm4, %xmm0
+; SSE41-NEXT:    pmaxud %xmm10, %xmm0
+; SSE41-NEXT:    pcmpeqd %xmm4, %xmm0
+; SSE41-NEXT:    pcmpeqd %xmm6, %xmm6
+; SSE41-NEXT:    pxor %xmm6, %xmm0
+; SSE41-NEXT:    movdqa {{.*#+}} xmm7 = <u,u,u,u,0,4,8,12,u,u,u,u,u,u,u,u>
+; SSE41-NEXT:    pshufb %xmm7, %xmm0
+; SSE41-NEXT:    movdqa %xmm3, %xmm5
+; SSE41-NEXT:    pmaxud %xmm9, %xmm5
+; SSE41-NEXT:    pcmpeqd %xmm3, %xmm5
+; SSE41-NEXT:    pxor %xmm6, %xmm5
+; SSE41-NEXT:    pshufb %xmm7, %xmm5
+; SSE41-NEXT:    punpckldq {{.*#+}} xmm5 = xmm5[0],xmm0[0],xmm5[1],xmm0[1]
+; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    pmaxud %xmm8, %xmm0
+; SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm6, %xmm0
+; SSE41-NEXT:    movdqa {{.*#+}} xmm12 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
+; SSE41-NEXT:    pshufb %xmm12, %xmm0
 ; SSE41-NEXT:    movdqa %xmm2, %xmm7
-; SSE41-NEXT:    pxor %xmm5, %xmm7
-; SSE41-NEXT:    por %xmm8, %xmm5
-; SSE41-NEXT:    pcmpgtd %xmm7, %xmm5
-; SSE41-NEXT:    pshufb %xmm6, %xmm5
-; SSE41-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm5[0],xmm0[1],xmm5[1]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm9[4,5,6,7]
-; SSE41-NEXT:    psubd %xmm8, %xmm2
+; SSE41-NEXT:    pmaxud %xmm11, %xmm7
+; SSE41-NEXT:    pcmpeqd %xmm2, %xmm7
+; SSE41-NEXT:    pxor %xmm6, %xmm7
+; SSE41-NEXT:    pshufb %xmm12, %xmm7
+; SSE41-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm7[0],xmm0[1],xmm7[1]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm5[4,5,6,7]
+; SSE41-NEXT:    psubd %xmm11, %xmm2
+; SSE41-NEXT:    psubd %xmm8, %xmm1
+; SSE41-NEXT:    psubd %xmm9, %xmm3
+; SSE41-NEXT:    psubd %xmm10, %xmm4
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
 ; SSE41-NEXT:    pand %xmm5, %xmm4
 ; SSE41-NEXT:    pand %xmm5, %xmm3
 ; SSE41-NEXT:    packuswb %xmm4, %xmm3
 ; SSE41-NEXT:    pand %xmm5, %xmm1
 ; SSE41-NEXT:    pand %xmm5, %xmm2
 ; SSE41-NEXT:    packuswb %xmm2, %xmm1
 ; SSE41-NEXT:    packuswb %xmm3, %xmm1
 ; SSE41-NEXT:    pandn %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test14:
 ; AVX1:       # %bb.0: # %vector.ph
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,2,3]
 ; AVX1-NEXT:    vpmovzxbd {{.*#+}} xmm8 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero
 ; AVX1-NEXT:    vpmovzxbd {{.*#+}} xmm9 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm5 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vpmovzxbd {{.*#+}} xmm10 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero,xmm5[2],zero,zero,zero,xmm5[3],zero,zero,zero
+; AVX1-NEXT:    vpmovzxbd {{.*#+}} xmm11 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero,xmm5[2],zero,zero,zero,xmm5[3],zero,zero,zero
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
 ; AVX1-NEXT:    vpmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [2147483648,2147483648,2147483648,2147483648]
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm7
-; AVX1-NEXT:    vpxor %xmm6, %xmm7, %xmm3
-; AVX1-NEXT:    vpor %xmm6, %xmm0, %xmm4
-; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpxor %xmm6, %xmm2, %xmm4
-; AVX1-NEXT:    vpor %xmm6, %xmm10, %xmm5
-; AVX1-NEXT:    vpcmpgtd %xmm4, %xmm5, %xmm4
-; AVX1-NEXT:    vpackssdw %xmm3, %xmm4, %xmm11
-; AVX1-NEXT:    vpxor %xmm6, %xmm1, %xmm4
-; AVX1-NEXT:    vpor %xmm6, %xmm9, %xmm5
-; AVX1-NEXT:    vpcmpgtd %xmm4, %xmm5, %xmm4
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX1-NEXT:    vpxor %xmm6, %xmm5, %xmm3
-; AVX1-NEXT:    vpor %xmm6, %xmm8, %xmm6
-; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm6, %xmm3
-; AVX1-NEXT:    vpackssdw %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpacksswb %xmm11, %xmm3, %xmm3
-; AVX1-NEXT:    vpsubd %xmm8, %xmm5, %xmm4
+; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm6
+; AVX1-NEXT:    vpmaxud %xmm0, %xmm6, %xmm7
+; AVX1-NEXT:    vpcmpeqd %xmm7, %xmm6, %xmm7
+; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX1-NEXT:    vpxor %xmm3, %xmm7, %xmm7
+; AVX1-NEXT:    vpmaxud %xmm11, %xmm2, %xmm4
+; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm2, %xmm4
+; AVX1-NEXT:    vpxor %xmm3, %xmm4, %xmm4
+; AVX1-NEXT:    vpackssdw %xmm7, %xmm4, %xmm10
+; AVX1-NEXT:    vpmaxud %xmm9, %xmm1, %xmm7
+; AVX1-NEXT:    vpcmpeqd %xmm7, %xmm1, %xmm7
+; AVX1-NEXT:    vpxor %xmm3, %xmm7, %xmm7
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
+; AVX1-NEXT:    vpmaxud %xmm8, %xmm4, %xmm5
+; AVX1-NEXT:    vpcmpeqd %xmm5, %xmm4, %xmm5
+; AVX1-NEXT:    vpxor %xmm3, %xmm5, %xmm3
+; AVX1-NEXT:    vpackssdw %xmm3, %xmm7, %xmm3
+; AVX1-NEXT:    vpacksswb %xmm10, %xmm3, %xmm3
+; AVX1-NEXT:    vpsubd %xmm8, %xmm4, %xmm4
 ; AVX1-NEXT:    vpsubd %xmm9, %xmm1, %xmm1
-; AVX1-NEXT:    vpsubd %xmm10, %xmm2, %xmm2
-; AVX1-NEXT:    vpsubd %xmm0, %xmm7, %xmm0
+; AVX1-NEXT:    vpsubd %xmm11, %xmm2, %xmm2
+; AVX1-NEXT:    vpsubd %xmm0, %xmm6, %xmm0
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
 ; AVX1-NEXT:    vpand %xmm5, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackuswb %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpand %xmm5, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm2
 ; AVX1-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackuswb %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    vpandn %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test14:
 ; AVX2:       # %bb.0: # %vector.ph
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[2,3,0,1]
 ; AVX2-NEXT:    vpmovzxbd {{.*#+}} ymm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero,xmm3[2],zero,zero,zero,xmm3[3],zero,zero,zero,xmm3[4],zero,zero,zero,xmm3[5],zero,zero,zero,xmm3[6],zero,zero,zero,xmm3[7],zero,zero,zero
 ; AVX2-NEXT:    vpmovzxbd {{.*#+}} ymm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm4 = [2147483648,2147483648,2147483648,2147483648,2147483648,2147483648,2147483648,2147483648]
-; AVX2-NEXT:    vpxor %ymm4, %ymm1, %ymm5
-; AVX2-NEXT:    vpor %ymm4, %ymm0, %ymm6
-; AVX2-NEXT:    vpcmpgtd %ymm5, %ymm6, %ymm5
-; AVX2-NEXT:    vextracti128 $1, %ymm5, %xmm6
-; AVX2-NEXT:    vpackssdw %xmm6, %xmm5, %xmm5
-; AVX2-NEXT:    vpxor %ymm4, %ymm2, %ymm6
-; AVX2-NEXT:    vpor %ymm4, %ymm3, %ymm4
-; AVX2-NEXT:    vpcmpgtd %ymm6, %ymm4, %ymm4
+; AVX2-NEXT:    vpmaxud %ymm0, %ymm1, %ymm4
+; AVX2-NEXT:    vpcmpeqd %ymm4, %ymm1, %ymm4
+; AVX2-NEXT:    vpcmpeqd %ymm5, %ymm5, %ymm5
+; AVX2-NEXT:    vpxor %ymm5, %ymm4, %ymm4
 ; AVX2-NEXT:    vextracti128 $1, %ymm4, %xmm6
 ; AVX2-NEXT:    vpackssdw %xmm6, %xmm4, %xmm4
-; AVX2-NEXT:    vpacksswb %xmm4, %xmm5, %xmm4
+; AVX2-NEXT:    vpmaxud %ymm3, %ymm2, %ymm6
+; AVX2-NEXT:    vpcmpeqd %ymm6, %ymm2, %ymm6
+; AVX2-NEXT:    vpxor %ymm5, %ymm6, %ymm5
+; AVX2-NEXT:    vextracti128 $1, %ymm5, %xmm6
+; AVX2-NEXT:    vpackssdw %xmm6, %xmm5, %xmm5
+; AVX2-NEXT:    vpacksswb %xmm5, %xmm4, %xmm4
 ; AVX2-NEXT:    vpsubd %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vpsubd %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
 ; AVX2-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
 ; AVX2-NEXT:    vpshufb %ymm1, %ymm2, %ymm1

 ; SSSE3-NEXT:    movdqa %xmm3, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: test15:
 ; SSE41:       # %bb.0: # %vector.ph
 ; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[2,3,0,1]
 ; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero
 ; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [2147483648,2147483648,2147483648,2147483648]
-; SSE41-NEXT:    movdqa %xmm0, %xmm5
-; SSE41-NEXT:    psubd %xmm1, %xmm0
-; SSE41-NEXT:    pxor %xmm4, %xmm1
-; SSE41-NEXT:    por %xmm4, %xmm5
-; SSE41-NEXT:    pcmpgtd %xmm1, %xmm5
-; SSE41-NEXT:    movdqa %xmm2, %xmm1
-; SSE41-NEXT:    pxor %xmm4, %xmm1
-; SSE41-NEXT:    por %xmm3, %xmm4
-; SSE41-NEXT:    pcmpgtd %xmm1, %xmm4
-; SSE41-NEXT:    packssdw %xmm4, %xmm5
+; SSE41-NEXT:    movdqa %xmm0, %xmm4
+; SSE41-NEXT:    pminud %xmm1, %xmm4
+; SSE41-NEXT:    pcmpeqd %xmm0, %xmm4
+; SSE41-NEXT:    pcmpeqd %xmm5, %xmm5
+; SSE41-NEXT:    pxor %xmm5, %xmm4
+; SSE41-NEXT:    movdqa {{.*#+}} xmm6 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; SSE41-NEXT:    pshufb %xmm6, %xmm4
+; SSE41-NEXT:    movdqa %xmm3, %xmm7
+; SSE41-NEXT:    pminud %xmm2, %xmm7
+; SSE41-NEXT:    pcmpeqd %xmm3, %xmm7
+; SSE41-NEXT:    pxor %xmm5, %xmm7
+; SSE41-NEXT:    pshufb %xmm6, %xmm7
+; SSE41-NEXT:    punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm7[0]
 ; SSE41-NEXT:    psubd %xmm2, %xmm3
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; SSE41-NEXT:    pshufb %xmm1, %xmm0
-; SSE41-NEXT:    pshufb %xmm1, %xmm3
+; SSE41-NEXT:    psubd %xmm1, %xmm0
+; SSE41-NEXT:    pshufb %xmm6, %xmm0
+; SSE41-NEXT:    pshufb %xmm6, %xmm3
 ; SSE41-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm3[0]
-; SSE41-NEXT:    pand %xmm5, %xmm0
+; SSE41-NEXT:    pand %xmm4, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test15:
 ; AVX1:       # %bb.0: # %vector.ph
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
 ; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [2147483648,2147483648,2147483648,2147483648]
-; AVX1-NEXT:    vpxor %xmm3, %xmm1, %xmm4
-; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm5
-; AVX1-NEXT:    vpcmpgtd %xmm4, %xmm5, %xmm4
+; AVX1-NEXT:    vpminud %xmm1, %xmm0, %xmm3
+; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm0, %xmm3
+; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpxor %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX1-NEXT:    vpxor %xmm3, %xmm5, %xmm6
-; AVX1-NEXT:    vpor %xmm3, %xmm2, %xmm3
-; AVX1-NEXT:    vpcmpgtd %xmm6, %xmm3, %xmm3
-; AVX1-NEXT:    vpackssdw %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpminud %xmm5, %xmm2, %xmm6
+; AVX1-NEXT:    vpcmpeqd %xmm6, %xmm2, %xmm6
+; AVX1-NEXT:    vpxor %xmm4, %xmm6, %xmm4
+; AVX1-NEXT:    vpackssdw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsubd %xmm5, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX1-NEXT:    vpand %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test15:
 ; AVX2:       # %bb.0: # %vector.ph
 ; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm2 = [2147483648,2147483648,2147483648,2147483648,2147483648,2147483648,2147483648,2147483648]
-; AVX2-NEXT:    vpxor %ymm2, %ymm1, %ymm3
-; AVX2-NEXT:    vpor %ymm2, %ymm0, %ymm2
-; AVX2-NEXT:    vpcmpgtd %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpminud %ymm1, %ymm0, %ymm2
+; AVX2-NEXT:    vpcmpeqd %ymm2, %ymm0, %ymm2
+; AVX2-NEXT:    vpcmpeqd %ymm3, %ymm3, %ymm3
+; AVX2-NEXT:    vpxor %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
 ; AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubd %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-NEXT:    vpand %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq

 ; SSSE3-NEXT:    movdqa %xmm3, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: test16:
 ; SSE41:       # %bb.0: # %vector.ph
 ; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[2,3,0,1]
 ; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero
 ; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [2147483648,2147483648,2147483648,2147483648]
-; SSE41-NEXT:    movdqa %xmm0, %xmm5
-; SSE41-NEXT:    psubd %xmm1, %xmm0
-; SSE41-NEXT:    pxor %xmm4, %xmm1
-; SSE41-NEXT:    por %xmm4, %xmm5
-; SSE41-NEXT:    pcmpgtd %xmm1, %xmm5
-; SSE41-NEXT:    movdqa %xmm2, %xmm1
-; SSE41-NEXT:    pxor %xmm4, %xmm1
-; SSE41-NEXT:    por %xmm3, %xmm4
-; SSE41-NEXT:    pcmpgtd %xmm1, %xmm4
-; SSE41-NEXT:    packssdw %xmm4, %xmm5
+; SSE41-NEXT:    movdqa %xmm1, %xmm4
+; SSE41-NEXT:    pmaxud %xmm0, %xmm4
+; SSE41-NEXT:    pcmpeqd %xmm1, %xmm4
+; SSE41-NEXT:    pcmpeqd %xmm5, %xmm5
+; SSE41-NEXT:    pxor %xmm5, %xmm4
+; SSE41-NEXT:    movdqa {{.*#+}} xmm6 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; SSE41-NEXT:    pshufb %xmm6, %xmm4
+; SSE41-NEXT:    movdqa %xmm2, %xmm7
+; SSE41-NEXT:    pmaxud %xmm3, %xmm7
+; SSE41-NEXT:    pcmpeqd %xmm2, %xmm7
+; SSE41-NEXT:    pxor %xmm5, %xmm7
+; SSE41-NEXT:    pshufb %xmm6, %xmm7
+; SSE41-NEXT:    punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm7[0]
 ; SSE41-NEXT:    psubd %xmm2, %xmm3
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; SSE41-NEXT:    pshufb %xmm1, %xmm0
-; SSE41-NEXT:    pshufb %xmm1, %xmm3
+; SSE41-NEXT:    psubd %xmm1, %xmm0
+; SSE41-NEXT:    pshufb %xmm6, %xmm0
+; SSE41-NEXT:    pshufb %xmm6, %xmm3
 ; SSE41-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm3[0]
-; SSE41-NEXT:    pand %xmm5, %xmm0
+; SSE41-NEXT:    pand %xmm4, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test16:
 ; AVX1:       # %bb.0: # %vector.ph
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
 ; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [2147483648,2147483648,2147483648,2147483648]
-; AVX1-NEXT:    vpxor %xmm3, %xmm1, %xmm4
-; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm5
-; AVX1-NEXT:    vpcmpgtd %xmm4, %xmm5, %xmm4
+; AVX1-NEXT:    vpmaxud %xmm0, %xmm1, %xmm3
+; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm1, %xmm3
+; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpxor %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX1-NEXT:    vpxor %xmm3, %xmm5, %xmm6
-; AVX1-NEXT:    vpor %xmm3, %xmm2, %xmm3
-; AVX1-NEXT:    vpcmpgtd %xmm6, %xmm3, %xmm3
-; AVX1-NEXT:    vpackssdw %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpmaxud %xmm2, %xmm5, %xmm6
+; AVX1-NEXT:    vpcmpeqd %xmm6, %xmm5, %xmm6
+; AVX1-NEXT:    vpxor %xmm4, %xmm6, %xmm4
+; AVX1-NEXT:    vpackssdw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsubd %xmm5, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX1-NEXT:    vpand %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test16:
 ; AVX2:       # %bb.0: # %vector.ph
 ; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm2 = [2147483648,2147483648,2147483648,2147483648,2147483648,2147483648,2147483648,2147483648]
-; AVX2-NEXT:    vpxor %ymm2, %ymm1, %ymm3
-; AVX2-NEXT:    vpor %ymm2, %ymm0, %ymm2
-; AVX2-NEXT:    vpcmpgtd %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpmaxud %ymm0, %ymm1, %ymm2
+; AVX2-NEXT:    vpcmpeqd %ymm2, %ymm1, %ymm2
+; AVX2-NEXT:    vpcmpeqd %ymm3, %ymm3, %ymm3
+; AVX2-NEXT:    vpxor %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
 ; AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubd %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-NEXT:    vpand %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq

test/CodeGen/X86/vec_cmp_uint-128.ll

 ; SSE42-LABEL: gt_v2i64:
 ; SSE42:       # %bb.0:
 ; SSE42-NEXT:    movdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
 ; SSE42-NEXT:    pxor %xmm2, %xmm1
 ; SSE42-NEXT:    pxor %xmm2, %xmm0
 ; SSE42-NEXT:    pcmpgtq %xmm1, %xmm0
 ; SSE42-NEXT:    retq
 ;
-; AVX-LABEL: gt_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:    retq
+; AVX1-LABEL: gt_v2i64:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
+; AVX1-NEXT:    vpxor %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpxor %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: gt_v2i64:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
+; AVX2-NEXT:    vpxor %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpxor %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    retq
 ;
 ; XOP-LABEL: gt_v2i64:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vpcomgtuq %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
+;
+; AVX512-LABEL: gt_v2i64:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512-NEXT:    vpminuq %zmm1, %zmm0, %zmm1
+; AVX512-NEXT:    vpcmpeqq %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
+; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    retq
   %1 = icmp ugt <2 x i64> %a, %b
   %2 = sext <2 x i1> %1 to <2 x i64>
   ret <2 x i64> %2
 }
 
 define <4 x i32> @gt_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
-; SSE-LABEL: gt_v4i32:
-; SSE:       # %bb.0:
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
-; SSE-NEXT:    pxor %xmm2, %xmm1
-; SSE-NEXT:    pxor %xmm2, %xmm0
-; SSE-NEXT:    pcmpgtd %xmm1, %xmm0
-; SSE-NEXT:    retq
+; SSE2-LABEL: gt_v4i32:
+; SSE2:       # %bb.0:
+; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
+; SSE2-NEXT:    pxor %xmm2, %xmm1
+; SSE2-NEXT:    pxor %xmm2, %xmm0
+; SSE2-NEXT:    pcmpgtd %xmm1, %xmm0
+; SSE2-NEXT:    retq
+;
+; SSE41-LABEL: gt_v4i32:
+; SSE41:       # %bb.0:
+; SSE41-NEXT:    pminud %xmm0, %xmm1
+; SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
+; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    retq
+;
+; SSE42-LABEL: gt_v4i32:
+; SSE42:       # %bb.0:
+; SSE42-NEXT:    pminud %xmm0, %xmm1
+; SSE42-NEXT:    pcmpeqd %xmm1, %xmm0
+; SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE42-NEXT:    pxor %xmm1, %xmm0
+; SSE42-NEXT:    retq
 ;
 ; AVX1-LABEL: gt_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
-; AVX1-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpminud %xmm1, %xmm0, %xmm1
+; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: gt_v4i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
-; AVX2-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpminud %xmm1, %xmm0, %xmm1
+; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOP-LABEL: gt_v4i32:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vpcomgtud %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; AVX512-LABEL: gt_v4i32:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
-; AVX512-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; AVX512-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; AVX512-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpminud %xmm1, %xmm0, %xmm1
+; AVX512-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
+; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = icmp ugt <4 x i32> %a, %b
   %2 = sext <4 x i1> %1 to <4 x i32>
   ret <4 x i32> %2
 }
 
 define <8 x i16> @gt_v8i16(<8 x i16> %a, <8 x i16> %b) nounwind {
-; SSE-LABEL: gt_v8i16:
-; SSE:       # %bb.0:
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE-NEXT:    pxor %xmm2, %xmm1
-; SSE-NEXT:    pxor %xmm2, %xmm0
-; SSE-NEXT:    pcmpgtw %xmm1, %xmm0
-; SSE-NEXT:    retq
+; SSE2-LABEL: gt_v8i16:
+; SSE2:       # %bb.0:
+; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; SSE2-NEXT:    pxor %xmm2, %xmm1
+; SSE2-NEXT:    pxor %xmm2, %xmm0
+; SSE2-NEXT:    pcmpgtw %xmm1, %xmm0
+; SSE2-NEXT:    retq
 ;
-; AVX-LABEL: gt_v8i16:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; AVX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; AVX-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    retq
+; SSE41-LABEL: gt_v8i16:
+; SSE41:       # %bb.0:
+; SSE41-NEXT:    pminuw %xmm0, %xmm1
+; SSE41-NEXT:    pcmpeqw %xmm1, %xmm0
+; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    retq
+;
+; SSE42-LABEL: gt_v8i16:
+; SSE42:       # %bb.0:
+; SSE42-NEXT:    pminuw %xmm0, %xmm1
+; SSE42-NEXT:    pcmpeqw %xmm1, %xmm0
+; SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE42-NEXT:    pxor %xmm1, %xmm0
+; SSE42-NEXT:    retq
+;
+; AVX1-LABEL: gt_v8i16:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpminuw %xmm1, %xmm0, %xmm1
+; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: gt_v8i16:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpminuw %xmm1, %xmm0, %xmm1
+; AVX2-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    retq
 ;
 ; XOP-LABEL: gt_v8i16:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vpcomgtuw %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
+;
+; AVX512-LABEL: gt_v8i16:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpminuw %xmm1, %xmm0, %xmm1
+; AVX512-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
+; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    retq
   %1 = icmp ugt <8 x i16> %a, %b
   %2 = sext <8 x i1> %1 to <8 x i16>
   ret <8 x i16> %2
 }
 
 define <16 x i8> @gt_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
 ; SSE-LABEL: gt_v16i8:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; SSE-NEXT:    pxor %xmm2, %xmm1
-; SSE-NEXT:    pxor %xmm2, %xmm0
-; SSE-NEXT:    pcmpgtb %xmm1, %xmm0
+; SSE-NEXT:    pminub %xmm0, %xmm1
+; SSE-NEXT:    pcmpeqb %xmm1, %xmm0
+; SSE-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE-NEXT:    pxor %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: gt_v16i8:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; AVX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; AVX-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    retq
+; AVX1-LABEL: gt_v16i8:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm1
+; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: gt_v16i8:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm1
+; AVX2-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    retq
 ;
 ; XOP-LABEL: gt_v16i8:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vpcomgtub %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
+;
+; AVX512-LABEL: gt_v16i8:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm1
+; AVX512-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
+; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    retq
   %1 = icmp ugt <16 x i8> %a, %b
   %2 = sext <16 x i1> %1 to <16 x i8>
   ret <16 x i8> %2
 }
 
 ;
 ; Less Than Or Equal
 ;

 ; SSE42:       # %bb.0:
 ; SSE42-NEXT:    movdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
 ; SSE42-NEXT:    pxor %xmm2, %xmm0
 ; SSE42-NEXT:    pxor %xmm1, %xmm2
 ; SSE42-NEXT:    pcmpgtq %xmm0, %xmm2
 ; SSE42-NEXT:    movdqa %xmm2, %xmm0
 ; SSE42-NEXT:    retq
 ;
-; AVX-LABEL: lt_v2i64:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
-; AVX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; AVX-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    retq
+; AVX1-LABEL: lt_v2i64:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
+; AVX1-NEXT:    vpxor %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: lt_v2i64:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
+; AVX2-NEXT:    vpxor %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    retq
 ;
 ; XOP-LABEL: lt_v2i64:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vpcomltuq %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
+;
+; AVX512-LABEL: lt_v2i64:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512-NEXT:    vpmaxuq %zmm1, %zmm0, %zmm1
+; AVX512-NEXT:    vpcmpeqq %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
+; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    retq
   %1 = icmp ult <2 x i64> %a, %b
   %2 = sext <2 x i1> %1 to <2 x i64>
   ret <2 x i64> %2
 }
 
 define <4 x i32> @lt_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
-; SSE-LABEL: lt_v4i32:
-; SSE:       # %bb.0:
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
-; SSE-NEXT:    pxor %xmm2, %xmm0
-; SSE-NEXT:    pxor %xmm1, %xmm2
-; SSE-NEXT:    pcmpgtd %xmm0, %xmm2
-; SSE-NEXT:    movdqa %xmm2, %xmm0
-; SSE-NEXT:    retq
+; SSE2-LABEL: lt_v4i32:
+; SSE2:       # %bb.0:
+; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
+; SSE2-NEXT:    pxor %xmm2, %xmm0
+; SSE2-NEXT:    pxor %xmm1, %xmm2
+; SSE2-NEXT:    pcmpgtd %xmm0, %xmm2
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    retq
+;
+; SSE41-LABEL: lt_v4i32:
+; SSE41:       # %bb.0:
+; SSE41-NEXT:    pmaxud %xmm0, %xmm1
+; SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
+; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    retq
+;
+; SSE42-LABEL: lt_v4i32:
+; SSE42:       # %bb.0:
+; SSE42-NEXT:    pmaxud %xmm0, %xmm1
+; SSE42-NEXT:    pcmpeqd %xmm1, %xmm0
+; SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE42-NEXT:    pxor %xmm1, %xmm0
+; SSE42-NEXT:    retq
 ;
 ; AVX1-LABEL: lt_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
-; AVX1-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpgtd %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpmaxud %xmm1, %xmm0, %xmm1
+; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: lt_v4i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
-; AVX2-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpcmpgtd %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vpmaxud %xmm1, %xmm0, %xmm1
+; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOP-LABEL: lt_v4i32:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vpcomltud %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; AVX512-LABEL: lt_v4i32:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
-; AVX512-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; AVX512-NEXT:    vpcmpgtd %xmm0, %xmm1, %xmm0
+; AVX512-NEXT:    vpmaxud %xmm1, %xmm0, %xmm1
+; AVX512-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
+; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = icmp ult <4 x i32> %a, %b
   %2 = sext <4 x i1> %1 to <4 x i32>
   ret <4 x i32> %2
 }
 
 define <8 x i16> @lt_v8i16(<8 x i16> %a, <8 x i16> %b) nounwind {
-; SSE-LABEL: lt_v8i16:
-; SSE:       # %bb.0:
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE-NEXT:    pxor %xmm2, %xmm0
-; SSE-NEXT:    pxor %xmm1, %xmm2
-; SSE-NEXT:    pcmpgtw %xmm0, %xmm2
-; SSE-NEXT:    movdqa %xmm2, %xmm0
-; SSE-NEXT:    retq
+; SSE2-LABEL: lt_v8i16:
+; SSE2:       # %bb.0:
+; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; SSE2-NEXT:    pxor %xmm2, %xmm0
+; SSE2-NEXT:    pxor %xmm1, %xmm2
+; SSE2-NEXT:    pcmpgtw %xmm0, %xmm2
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    retq
 ;
-; AVX-LABEL: lt_v8i16:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; AVX-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    retq
+; SSE41-LABEL: lt_v8i16:
+; SSE41:       # %bb.0:
+; SSE41-NEXT:    pmaxuw %xmm0, %xmm1
+; SSE41-NEXT:    pcmpeqw %xmm1, %xmm0
+; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    retq
+;
+; SSE42-LABEL: lt_v8i16:
+; SSE42:       # %bb.0:
+; SSE42-NEXT:    pmaxuw %xmm0, %xmm1
+; SSE42-NEXT:    pcmpeqw %xmm1, %xmm0
+; SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE42-NEXT:    pxor %xmm1, %xmm0
+; SSE42-NEXT:    retq
+;
+; AVX1-LABEL: lt_v8i16:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
+; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: lt_v8i16:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
+; AVX2-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    retq
 ;
 ; XOP-LABEL: lt_v8i16:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vpcomltuw %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
+;
+; AVX512-LABEL: lt_v8i16:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm1
+; AVX512-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
+; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    retq
   %1 = icmp ult <8 x i16> %a, %b
   %2 = sext <8 x i1> %1 to <8 x i16>
   ret <8 x i16> %2
 }
 
 define <16 x i8> @lt_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
 ; SSE-LABEL: lt_v16i8:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; SSE-NEXT:    pxor %xmm2, %xmm0
-; SSE-NEXT:    pxor %xmm1, %xmm2
-; SSE-NEXT:    pcmpgtb %xmm0, %xmm2
-; SSE-NEXT:    movdqa %xmm2, %xmm0
+; SSE-NEXT:    pmaxub %xmm0, %xmm1
+; SSE-NEXT:    pcmpeqb %xmm1, %xmm0
+; SSE-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE-NEXT:    pxor %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX-LABEL: lt_v16i8:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm2, %xmm1, %xmm1
-; AVX-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    retq
+; AVX1-LABEL: lt_v16i8:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
+; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: lt_v16i8:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
+; AVX2-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
+; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    retq
 ;
 ; XOP-LABEL: lt_v16i8:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vpcomltub %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
+;
+; AVX512-LABEL: lt_v16i8:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpmaxub %xmm1, %xmm0, %xmm1
+; AVX512-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
+; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    retq
   %1 = icmp ult <16 x i8> %a, %b
   %2 = sext <16 x i1> %1 to <16 x i8>
   ret <16 x i8> %2
 }

test/CodeGen/X86/vec_minmax_match.ll

; CHECK-NEXT:    retq
   %r = select <4 x i1> %cmp1, <4 x i32><i32 255, i32 255, i32 255, i32 255>, <4 x i32> %max
   ret <4 x i32> %r
 }
 
 define <4 x i32> @wrong_pred_for_smin_with_not(<4 x i32> %x) {
 ; CHECK-LABEL: wrong_pred_for_smin_with_not:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
-; CHECK-NEXT:    vpxor %xmm1, %xmm0, %xmm1
-; CHECK-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
-; CHECK-NEXT:    vpcmpgtd {{.*}}(%rip), %xmm0, %xmm0
-; CHECK-NEXT:    vmovaps {{.*#+}} xmm2 = [4294967291,4294967291,4294967291,4294967291]
-; CHECK-NEXT:    vblendvps %xmm0, %xmm1, %xmm2, %xmm0
+; CHECK-NEXT:    vpxor %xmm1, %xmm0, %xmm2
+; CHECK-NEXT:    vpminud {{.*}}(%rip), %xmm0, %xmm3
+; CHECK-NEXT:    vpcmpeqd %xmm3, %xmm0, %xmm0
+; CHECK-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vmovaps {{.*#+}} xmm1 = [4294967291,4294967291,4294967291,4294967291]
+; CHECK-NEXT:    vblendvps %xmm0, %xmm2, %xmm1, %xmm0
 ; CHECK-NEXT:    retq
   %not_x = xor <4 x i32> %x, <i32 -1, i32 -1, i32 -1, i32 -1>
   %cmp = icmp ugt <4 x i32> %x, <i32 4, i32 4, i32 4, i32 4>
   %sel = select <4 x i1> %cmp, <4 x i32> %not_x, <4 x i32> <i32 -5, i32 -5, i32 -5, i32 -5>
   ret <4 x i32> %sel
 }
 
 define <4 x i32> @wrong_pred_for_smin_with_subnsw(<4 x i32> %x, <4 x i32> %y) {

test/CodeGen/X86/vec_setcc-2.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -o - -mcpu=generic -mtriple=x86_64-apple-darwin -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE2
 ; RUN: llc < %s -o - -mcpu=generic -mtriple=x86_64-apple-darwin -mattr=+sse4.2 | FileCheck %s --check-prefixes=CHECK,SSE41
 
 ; For a setult against a constant, turn it into a setule and lower via psubusw.
 
 define void @loop_no_const_reload(<2 x i64>*  %in, <2 x i64>* %out, i32 %n) {
-; CHECK-LABEL: loop_no_const_reload:
-; CHECK:       ## %bb.0: ## %entry
-; CHECK-NEXT:    testl %edx, %edx
-; CHECK-NEXT:    je LBB0_3
-; CHECK-NEXT:  ## %bb.1: ## %for.body.preheader
-; CHECK-NEXT:    xorl %eax, %eax
-; CHECK-NEXT:    movdqa {{.*#+}} xmm0 = [25,25,25,25,25,25,25,25]
-; CHECK-NEXT:    pxor %xmm1, %xmm1
-; CHECK-NEXT:    .p2align 4, 0x90
-; CHECK-NEXT:  LBB0_2: ## %for.body
-; CHECK-NEXT:    ## =>This Inner Loop Header: Depth=1
-; CHECK-NEXT:    movdqa (%rdi,%rax), %xmm2
-; CHECK-NEXT:    psubusw %xmm0, %xmm2
-; CHECK-NEXT:    pcmpeqw %xmm1, %xmm2
-; CHECK-NEXT:    movdqa %xmm2, (%rsi,%rax)
-; CHECK-NEXT:    addq $16, %rax
-; CHECK-NEXT:    decl %edx
-; CHECK-NEXT:    jne LBB0_2
-; CHECK-NEXT:  LBB0_3: ## %for.end
-; CHECK-NEXT:    retq
+; SSE2-LABEL: loop_no_const_reload:
+; SSE2:       ## %bb.0: ## %entry
+; SSE2-NEXT:    testl %edx, %edx
+; SSE2-NEXT:    je LBB0_3
+; SSE2-NEXT:  ## %bb.1: ## %for.body.preheader
+; SSE2-NEXT:    xorl %eax, %eax
+; SSE2-NEXT:    movdqa {{.*#+}} xmm0 = [25,25,25,25,25,25,25,25]
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    .p2align 4, 0x90
+; SSE2-NEXT:  LBB0_2: ## %for.body
+; SSE2-NEXT:    ## =>This Inner Loop Header: Depth=1
+; SSE2-NEXT:    movdqa (%rdi,%rax), %xmm2
+; SSE2-NEXT:    psubusw %xmm0, %xmm2
+; SSE2-NEXT:    pcmpeqw %xmm1, %xmm2
+; SSE2-NEXT:    movdqa %xmm2, (%rsi,%rax)
+; SSE2-NEXT:    addq $16, %rax
+; SSE2-NEXT:    decl %edx
+; SSE2-NEXT:    jne LBB0_2
+; SSE2-NEXT:  LBB0_3: ## %for.end
+; SSE2-NEXT:    retq
+;
+; SSE41-LABEL: loop_no_const_reload:
+; SSE41:       ## %bb.0: ## %entry
+; SSE41-NEXT:    testl %edx, %edx
+; SSE41-NEXT:    je LBB0_3
+; SSE41-NEXT:  ## %bb.1: ## %for.body.preheader
+; SSE41-NEXT:    xorl %eax, %eax
+; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [26,26,26,26,26,26,26,26]
+; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE41-NEXT:    .p2align 4, 0x90
+; SSE41-NEXT:  LBB0_2: ## %for.body
+; SSE41-NEXT:    ## =>This Inner Loop Header: Depth=1
+; SSE41-NEXT:    movdqa (%rdi,%rax), %xmm2
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    pmaxuw %xmm0, %xmm3
+; SSE41-NEXT:    pcmpeqw %xmm2, %xmm3
+; SSE41-NEXT:    pxor %xmm1, %xmm3
+; SSE41-NEXT:    movdqa %xmm3, (%rsi,%rax)
+; SSE41-NEXT:    addq $16, %rax
+; SSE41-NEXT:    decl %edx
+; SSE41-NEXT:    jne LBB0_2
+; SSE41-NEXT:  LBB0_3: ## %for.end
+; SSE41-NEXT:    retq
 entry:
   %cmp9 = icmp eq i32 %n, 0
   br i1 %cmp9, label %for.end, label %for.body
 
 for.body:                                         ; preds = %for.body, %entry
   %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
   %arrayidx1 = getelementptr inbounds <2 x i64>, <2 x i64>* %in, i64 %indvars.iv
   %arrayidx1.val = load <2 x i64>, <2 x i64>* %arrayidx1, align 16

 
 for.end:                                          ; preds = %for.body, %entry
   ret void
 }
 
 ; Be careful if decrementing the constant would undeflow.
 
 define void @loop_const_folding_underflow(<2 x i64>*  %in, <2 x i64>* %out, i32 %n) {
-; CHECK-LABEL: loop_const_folding_underflow:
-; CHECK:       ## %bb.0: ## %entry
-; CHECK-NEXT:    testl %edx, %edx
-; CHECK-NEXT:    je LBB1_3
-; CHECK-NEXT:  ## %bb.1: ## %for.body.preheader
-; CHECK-NEXT:    xorl %eax, %eax
-; CHECK-NEXT:    movdqa {{.*#+}} xmm0 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32794,32794,32794,32794,32794,32794,32794]
-; CHECK-NEXT:    .p2align 4, 0x90
-; CHECK-NEXT:  LBB1_2: ## %for.body
-; CHECK-NEXT:    ## =>This Inner Loop Header: Depth=1
-; CHECK-NEXT:    movdqa (%rdi,%rax), %xmm2
-; CHECK-NEXT:    pxor %xmm0, %xmm2
-; CHECK-NEXT:    movdqa %xmm1, %xmm3
-; CHECK-NEXT:    pcmpgtw %xmm2, %xmm3
-; CHECK-NEXT:    movdqa %xmm3, (%rsi,%rax)
-; CHECK-NEXT:    addq $16, %rax
-; CHECK-NEXT:    decl %edx
-; CHECK-NEXT:    jne LBB1_2
-; CHECK-NEXT:  LBB1_3: ## %for.end
-; CHECK-NEXT:    retq
+; SSE2-LABEL: loop_const_folding_underflow:
+; SSE2:       ## %bb.0: ## %entry
+; SSE2-NEXT:    testl %edx, %edx
+; SSE2-NEXT:    je LBB1_3
+; SSE2-NEXT:  ## %bb.1: ## %for.body.preheader
+; SSE2-NEXT:    xorl %eax, %eax
+; SSE2-NEXT:    movdqa {{.*#+}} xmm0 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32794,32794,32794,32794,32794,32794,32794]
+; SSE2-NEXT:    .p2align 4, 0x90
+; SSE2-NEXT:  LBB1_2: ## %for.body
+; SSE2-NEXT:    ## =>This Inner Loop Header: Depth=1
+; SSE2-NEXT:    movdqa (%rdi,%rax), %xmm2
+; SSE2-NEXT:    pxor %xmm0, %xmm2
+; SSE2-NEXT:    movdqa %xmm1, %xmm3
+; SSE2-NEXT:    pcmpgtw %xmm2, %xmm3
+; SSE2-NEXT:    movdqa %xmm3, (%rsi,%rax)
+; SSE2-NEXT:    addq $16, %rax
+; SSE2-NEXT:    decl %edx
+; SSE2-NEXT:    jne LBB1_2
+; SSE2-NEXT:  LBB1_3: ## %for.end
+; SSE2-NEXT:    retq
+;
+; SSE41-LABEL: loop_const_folding_underflow:
+; SSE41:       ## %bb.0: ## %entry
+; SSE41-NEXT:    testl %edx, %edx
+; SSE41-NEXT:    je LBB1_3
+; SSE41-NEXT:  ## %bb.1: ## %for.body.preheader
+; SSE41-NEXT:    xorl %eax, %eax
+; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [0,26,26,26,26,26,26,26]
+; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE41-NEXT:    .p2align 4, 0x90
+; SSE41-NEXT:  LBB1_2: ## %for.body
+; SSE41-NEXT:    ## =>This Inner Loop Header: Depth=1
+; SSE41-NEXT:    movdqa (%rdi,%rax), %xmm2
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    pmaxuw %xmm0, %xmm3
+; SSE41-NEXT:    pcmpeqw %xmm2, %xmm3
+; SSE41-NEXT:    pxor %xmm1, %xmm3
+; SSE41-NEXT:    movdqa %xmm3, (%rsi,%rax)
+; SSE41-NEXT:    addq $16, %rax
+; SSE41-NEXT:    decl %edx
+; SSE41-NEXT:    jne LBB1_2
+; SSE41-NEXT:  LBB1_3: ## %for.end
+; SSE41-NEXT:    retq
 entry:
   %cmp9 = icmp eq i32 %n, 0
   br i1 %cmp9, label %for.end, label %for.body
 
 for.body:                                         ; preds = %for.body, %entry
   %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
   %arrayidx1 = getelementptr inbounds <2 x i64>, <2 x i64>* %in, i64 %indvars.iv
   %arrayidx1.val = load <2 x i64>, <2 x i64>* %arrayidx1, align 16
for.end:                                          ; preds = %for.body, %entry
   ret void
 }
 
 ; Test for PSUBUSB
 
 define <16 x i8> @test_ult_byte(<16 x i8> %a) {
 ; CHECK-LABEL: test_ult_byte:
 ; CHECK:       ## %bb.0: ## %entry
-; CHECK-NEXT:    psubusb {{.*}}(%rip), %xmm0
-; CHECK-NEXT:    pxor %xmm1, %xmm1
+; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [11,11,11,11,11,11,11,11,11,11,11,11,11,11,11,11]
+; CHECK-NEXT:    pmaxub %xmm0, %xmm1
 ; CHECK-NEXT:    pcmpeqb %xmm1, %xmm0
+; CHECK-NEXT:    pcmpeqd %xmm1, %xmm1
+; CHECK-NEXT:    pxor %xmm1, %xmm0
 ; CHECK-NEXT:    retq
 entry:
   %icmp = icmp ult <16 x i8> %a, <i8 11, i8 11, i8 11, i8 11, i8 11, i8 11, i8 11, i8 11, i8 11, i8 11, i8 11, i8 11, i8 11, i8 11, i8 11, i8 11>
   %sext = sext <16 x i1> %icmp to <16 x i8>
   ret <16 x i8> %sext
 }
 
 ; Only do this when we can turn the comparison into a setule.  I.e. not for
 ; register operands.
 
 define <8 x i16> @test_ult_register(<8 x i16> %a, <8 x i16> %b) {
-; CHECK-LABEL: test_ult_register:
-; CHECK:       ## %bb.0: ## %entry
-; CHECK-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; CHECK-NEXT:    pxor %xmm2, %xmm0
-; CHECK-NEXT:    pxor %xmm1, %xmm2
-; CHECK-NEXT:    pcmpgtw %xmm0, %xmm2
-; CHECK-NEXT:    movdqa %xmm2, %xmm0
-; CHECK-NEXT:    retq
+; SSE2-LABEL: test_ult_register:
+; SSE2:       ## %bb.0: ## %entry
+; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
+; SSE2-NEXT:    pxor %xmm2, %xmm0
+; SSE2-NEXT:    pxor %xmm1, %xmm2
+; SSE2-NEXT:    pcmpgtw %xmm0, %xmm2
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    retq
+;
+; SSE41-LABEL: test_ult_register:
+; SSE41:       ## %bb.0: ## %entry
+; SSE41-NEXT:    pmaxuw %xmm0, %xmm1
+; SSE41-NEXT:    pcmpeqw %xmm1, %xmm0
+; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    retq
 entry:
   %icmp = icmp ult <8 x i16> %a, %b
   %sext = sext <8 x i1> %icmp to <8 x i16>
   ret <8 x i16> %sext
 }