diff --git a/i/llvm/include/llvm/CodeGen/TargetLowering.h b/llvm/include/llvm/CodeGen/TargetLowering.h
--- a/i/llvm/include/llvm/CodeGen/TargetLowering.h
+++ b/llvm/include/llvm/CodeGen/TargetLowering.h
@@ -605,6 +605,12 @@
     return false;
   }
 
+  /// Return the maximum number of "x & (x - 1)" operations that can be done
+  /// instead of deferring to a custom CTPOP.
+  virtual unsigned getCustomCtpopCost(EVT VT, ISD::CondCode Cond) const {
+    return 1;
+  }
+
   /// Return true if instruction generated for equality comparison is folded
   /// with instruction generated for signed comparison.
   virtual bool isEqualityCmpFoldedWithSignedCmp() const { return true; }
diff --git a/i/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
--- a/i/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -3396,21 +3396,34 @@
   EVT CTVT = CTPOP.getValueType();
   SDValue CTOp = CTPOP.getOperand(0);
 
+  // If this is a vector CTPOP, keep the CTPOP if it is legal.
+  // TODO: Should we check if CTPOP is legal(or custom) for scalars?
+  if (VT.isVector() && TLI.isOperationLegal(ISD::CTPOP, CTVT))
+    return SDValue();
+
+
   // (ctpop x) u< 2 -> (x & x-1) == 0
   // (ctpop x) u> 1 -> (x & x-1) != 0
-  if ((Cond == ISD::SETULT && C1 == 2) || (Cond == ISD::SETUGT && C1 == 1)) {
-    // If this is a vector CTPOP, keep the CTPOP if it is legal.
+  if (Cond == ISD::SETULT || Cond == ISD::SETUGT) {
     // This based on X86's custom lowering for vector CTPOP which produces more
     // instructions than the expansion here.
-    // TODO: Should we check if CTPOP is legal(or custom) for scalars?
-    if (VT.isVector() && TLI.isOperationLegal(ISD::CTPOP, CTVT))
+
+    unsigned CostLimit = TLI.getCustomCtpopCost(CTVT, Cond);
+    if (C1.ugt(CostLimit + (Cond == ISD::SETULT)))
       return SDValue();
+    if (C1 == 0 && (Cond == ISD::SETULT))
+      return SDValue(); // This is handled elsewhere.
+
+    unsigned Passes = C1.getLimitedValue() - (Cond == ISD::SETULT);
 
     SDValue NegOne = DAG.getAllOnesConstant(dl, CTVT);
-    SDValue Add = DAG.getNode(ISD::ADD, dl, CTVT, CTOp, NegOne);
-    SDValue And = DAG.getNode(ISD::AND, dl, CTVT, CTOp, Add);
+    SDValue Result = CTOp;
+    for (unsigned i = 0; i < Passes; i++) {
+      SDValue Add = DAG.getNode(ISD::ADD, dl, CTVT, Result, NegOne);
+      Result = DAG.getNode(ISD::AND, dl, CTVT, Result, Add);
+    }
     ISD::CondCode CC = Cond == ISD::SETULT ? ISD::SETEQ : ISD::SETNE;
-    return DAG.getSetCC(dl, VT, And, DAG.getConstant(0, dl, CTVT), CC);
+    return DAG.getSetCC(dl, VT, Result, DAG.getConstant(0, dl, CTVT), CC);
   }
 
   // If ctpop is not supported, expand a power-of-2 comparison based on it.
@@ -3418,11 +3431,8 @@
     // For scalars, keep CTPOP if it is legal or custom.
     if (!VT.isVector() && TLI.isOperationLegalOrCustom(ISD::CTPOP, CTVT))
       return SDValue();
-    // For vectors, keep CTPOP only if it is legal.
     // This is based on X86's custom lowering for CTPOP which produces more
     // instructions than the expansion here.
-    if (VT.isVector() && TLI.isOperationLegal(ISD::CTPOP, CTVT))
-      return SDValue();
 
     // (ctpop x) == 1 --> (x != 0) && ((x & x-1) == 0)
     // (ctpop x) != 1 --> (x == 0) || ((x & x-1) != 0)
diff --git a/i/llvm/test/CodeGen/X86/vector-popcnt-128-ult-ugt.ll b/llvm/test/CodeGen/X86/vector-popcnt-128-ult-ugt.ll
--- a/i/llvm/test/CodeGen/X86/vector-popcnt-128-ult-ugt.ll
+++ b/llvm/test/CodeGen/X86/vector-popcnt-128-ult-ugt.ll
@@ -38,98 +38,16 @@
 }
 
 define <16 x i8> @ugt_0_v16i8(<16 x i8> %0) {
-; SSE2-LABEL: ugt_0_v16i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $1, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    psubb %xmm1, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    pand %xmm1, %xmm2
-; SSE2-NEXT:    psrlw $2, %xmm0
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    paddb %xmm2, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $4, %xmm1
-; SSE2-NEXT:    paddb %xmm0, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm0
-; SSE2-NEXT:    pcmpeqb %xmm0, %xmm1
-; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE2-NEXT:    pxor %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    retq
-;
-; SSE3-LABEL: ugt_0_v16i8:
-; SSE3:       # %bb.0:
-; SSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSE3-NEXT:    psrlw $1, %xmm1
-; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE3-NEXT:    psubb %xmm1, %xmm0
-; SSE3-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE3-NEXT:    movdqa %xmm0, %xmm2
-; SSE3-NEXT:    pand %xmm1, %xmm2
-; SSE3-NEXT:    psrlw $2, %xmm0
-; SSE3-NEXT:    pand %xmm1, %xmm0
-; SSE3-NEXT:    paddb %xmm2, %xmm0
-; SSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSE3-NEXT:    psrlw $4, %xmm1
-; SSE3-NEXT:    paddb %xmm0, %xmm1
-; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE3-NEXT:    pxor %xmm0, %xmm0
-; SSE3-NEXT:    pcmpeqb %xmm0, %xmm1
-; SSE3-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE3-NEXT:    pxor %xmm0, %xmm1
-; SSE3-NEXT:    movdqa %xmm1, %xmm0
-; SSE3-NEXT:    retq
-;
-; SSSE3-LABEL: ugt_0_v16i8:
-; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT:    movdqa %xmm0, %xmm2
-; SSSE3-NEXT:    pand %xmm1, %xmm2
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSSE3-NEXT:    movdqa %xmm3, %xmm4
-; SSSE3-NEXT:    pshufb %xmm2, %xmm4
-; SSSE3-NEXT:    psrlw $4, %xmm0
-; SSSE3-NEXT:    pand %xmm1, %xmm0
-; SSSE3-NEXT:    pshufb %xmm0, %xmm3
-; SSSE3-NEXT:    paddb %xmm4, %xmm3
-; SSSE3-NEXT:    pxor %xmm1, %xmm1
-; SSSE3-NEXT:    pcmpeqb %xmm3, %xmm1
-; SSSE3-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSSE3-NEXT:    pxor %xmm1, %xmm0
-; SSSE3-NEXT:    retq
-;
-; SSE41-LABEL: ugt_0_v16i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    pand %xmm1, %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pshufb %xmm2, %xmm4
-; SSE41-NEXT:    psrlw $4, %xmm0
-; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    pshufb %xmm0, %xmm3
-; SSE41-NEXT:    paddb %xmm4, %xmm3
-; SSE41-NEXT:    pxor %xmm1, %xmm1
-; SSE41-NEXT:    pcmpeqb %xmm3, %xmm1
-; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    retq
+; SSE-LABEL: ugt_0_v16i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    pcmpeqb %xmm1, %xmm0
+; SSE-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: ugt_0_v16i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX1-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
@@ -138,14 +56,6 @@
 ;
 ; AVX2-LABEL: ugt_0_v16i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX2-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX2-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
@@ -154,9 +64,6 @@
 ;
 ; AVX512VPOPCNTDQ-LABEL: ugt_0_v16i8:
 ; AVX512VPOPCNTDQ:       # %bb.0:
-; AVX512VPOPCNTDQ-NEXT:    vpmovzxbd {{.*#+}} zmm0 = 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,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512VPOPCNTDQ-NEXT:    vpopcntd %zmm0, %zmm0
-; AVX512VPOPCNTDQ-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512VPOPCNTDQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512VPOPCNTDQ-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; AVX512VPOPCNTDQ-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
@@ -166,13 +73,9 @@
 ;
 ; AVX512VPOPCNTDQVL-LABEL: ugt_0_v16i8:
 ; AVX512VPOPCNTDQVL:       # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = 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,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512VPOPCNTDQVL-NEXT:    vpopcntd %zmm0, %zmm0
-; AVX512VPOPCNTDQVL-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512VPOPCNTDQVL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512VPOPCNTDQVL-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; AVX512VPOPCNTDQVL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
-; AVX512VPOPCNTDQVL-NEXT:    vzeroupper
 ; AVX512VPOPCNTDQVL-NEXT:    retq
 ;
 ; BITALG_NOVLX-LABEL: ugt_0_v16i8:
@@ -200,127 +103,17 @@
 }
 
 define <16 x i8> @ult_1_v16i8(<16 x i8> %0) {
-; SSE2-LABEL: ult_1_v16i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $1, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    psubb %xmm1, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    pand %xmm1, %xmm2
-; SSE2-NEXT:    psrlw $2, %xmm0
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    paddb %xmm2, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $4, %xmm1
-; SSE2-NEXT:    paddb %xmm0, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm0
-; SSE2-NEXT:    pcmpeqb %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    retq
-;
-; SSE3-LABEL: ult_1_v16i8:
-; SSE3:       # %bb.0:
-; SSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSE3-NEXT:    psrlw $1, %xmm1
-; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE3-NEXT:    psubb %xmm1, %xmm0
-; SSE3-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE3-NEXT:    movdqa %xmm0, %xmm2
-; SSE3-NEXT:    pand %xmm1, %xmm2
-; SSE3-NEXT:    psrlw $2, %xmm0
-; SSE3-NEXT:    pand %xmm1, %xmm0
-; SSE3-NEXT:    paddb %xmm2, %xmm0
-; SSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSE3-NEXT:    psrlw $4, %xmm1
-; SSE3-NEXT:    paddb %xmm0, %xmm1
-; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE3-NEXT:    pxor %xmm0, %xmm0
-; SSE3-NEXT:    pcmpeqb %xmm0, %xmm1
-; SSE3-NEXT:    movdqa %xmm1, %xmm0
-; SSE3-NEXT:    retq
-;
-; SSSE3-LABEL: ult_1_v16i8:
-; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT:    movdqa %xmm0, %xmm2
-; SSSE3-NEXT:    pand %xmm1, %xmm2
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSSE3-NEXT:    movdqa %xmm3, %xmm4
-; SSSE3-NEXT:    pshufb %xmm2, %xmm4
-; SSSE3-NEXT:    psrlw $4, %xmm0
-; SSSE3-NEXT:    pand %xmm1, %xmm0
-; SSSE3-NEXT:    pshufb %xmm0, %xmm3
-; SSSE3-NEXT:    paddb %xmm4, %xmm3
-; SSSE3-NEXT:    pxor %xmm0, %xmm0
-; SSSE3-NEXT:    pcmpeqb %xmm3, %xmm0
-; SSSE3-NEXT:    retq
-;
-; SSE41-LABEL: ult_1_v16i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    pand %xmm1, %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pshufb %xmm2, %xmm4
-; SSE41-NEXT:    psrlw $4, %xmm0
-; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    pshufb %xmm0, %xmm3
-; SSE41-NEXT:    paddb %xmm4, %xmm3
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqb %xmm3, %xmm0
-; SSE41-NEXT:    retq
-;
-; AVX1-LABEL: ult_1_v16i8:
-; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX1-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: ult_1_v16i8:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX2-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX2-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    retq
-;
-; AVX512VPOPCNTDQ-LABEL: ult_1_v16i8:
-; AVX512VPOPCNTDQ:       # %bb.0:
-; AVX512VPOPCNTDQ-NEXT:    vpmovzxbd {{.*#+}} zmm0 = 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,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512VPOPCNTDQ-NEXT:    vpopcntd %zmm0, %zmm0
-; AVX512VPOPCNTDQ-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512VPOPCNTDQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
-; AVX512VPOPCNTDQ-NEXT:    vzeroupper
-; AVX512VPOPCNTDQ-NEXT:    retq
+; SSE-LABEL: ult_1_v16i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    pcmpeqb %xmm1, %xmm0
+; SSE-NEXT:    retq
 ;
-; AVX512VPOPCNTDQVL-LABEL: ult_1_v16i8:
-; AVX512VPOPCNTDQVL:       # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = 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,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512VPOPCNTDQVL-NEXT:    vpopcntd %zmm0, %zmm0
-; AVX512VPOPCNTDQVL-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512VPOPCNTDQVL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
-; AVX512VPOPCNTDQVL-NEXT:    vzeroupper
-; AVX512VPOPCNTDQVL-NEXT:    retq
+; AVX-LABEL: ult_1_v16i8:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    retq
 ;
 ; BITALG_NOVLX-LABEL: ult_1_v16i8:
 ; BITALG_NOVLX:       # %bb.0:
@@ -2480,115 +2273,16 @@
 }
 
 define <8 x i16> @ugt_0_v8i16(<8 x i16> %0) {
-; SSE2-LABEL: ugt_0_v8i16:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $1, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    psubb %xmm1, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    pand %xmm1, %xmm2
-; SSE2-NEXT:    psrlw $2, %xmm0
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    paddb %xmm2, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $4, %xmm1
-; SSE2-NEXT:    paddb %xmm0, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psllw $8, %xmm0
-; SSE2-NEXT:    paddb %xmm1, %xmm0
-; SSE2-NEXT:    psrlw $8, %xmm0
-; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    pcmpeqw %xmm1, %xmm0
-; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE2-NEXT:    pxor %xmm1, %xmm0
-; SSE2-NEXT:    retq
-;
-; SSE3-LABEL: ugt_0_v8i16:
-; SSE3:       # %bb.0:
-; SSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSE3-NEXT:    psrlw $1, %xmm1
-; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE3-NEXT:    psubb %xmm1, %xmm0
-; SSE3-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE3-NEXT:    movdqa %xmm0, %xmm2
-; SSE3-NEXT:    pand %xmm1, %xmm2
-; SSE3-NEXT:    psrlw $2, %xmm0
-; SSE3-NEXT:    pand %xmm1, %xmm0
-; SSE3-NEXT:    paddb %xmm2, %xmm0
-; SSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSE3-NEXT:    psrlw $4, %xmm1
-; SSE3-NEXT:    paddb %xmm0, %xmm1
-; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE3-NEXT:    movdqa %xmm1, %xmm0
-; SSE3-NEXT:    psllw $8, %xmm0
-; SSE3-NEXT:    paddb %xmm1, %xmm0
-; SSE3-NEXT:    psrlw $8, %xmm0
-; SSE3-NEXT:    pxor %xmm1, %xmm1
-; SSE3-NEXT:    pcmpeqw %xmm1, %xmm0
-; SSE3-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE3-NEXT:    pxor %xmm1, %xmm0
-; SSE3-NEXT:    retq
-;
-; SSSE3-LABEL: ugt_0_v8i16:
-; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT:    movdqa %xmm0, %xmm2
-; SSSE3-NEXT:    pand %xmm1, %xmm2
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSSE3-NEXT:    movdqa %xmm3, %xmm4
-; SSSE3-NEXT:    pshufb %xmm2, %xmm4
-; SSSE3-NEXT:    psrlw $4, %xmm0
-; SSSE3-NEXT:    pand %xmm1, %xmm0
-; SSSE3-NEXT:    pshufb %xmm0, %xmm3
-; SSSE3-NEXT:    paddb %xmm4, %xmm3
-; SSSE3-NEXT:    movdqa %xmm3, %xmm0
-; SSSE3-NEXT:    psllw $8, %xmm0
-; SSSE3-NEXT:    paddb %xmm3, %xmm0
-; SSSE3-NEXT:    psrlw $8, %xmm0
-; SSSE3-NEXT:    pxor %xmm1, %xmm1
-; SSSE3-NEXT:    pcmpeqw %xmm0, %xmm1
-; SSSE3-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSSE3-NEXT:    pxor %xmm1, %xmm0
-; SSSE3-NEXT:    retq
-;
-; SSE41-LABEL: ugt_0_v8i16:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    pand %xmm1, %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pshufb %xmm2, %xmm4
-; SSE41-NEXT:    psrlw $4, %xmm0
-; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    pshufb %xmm0, %xmm3
-; SSE41-NEXT:    paddb %xmm4, %xmm3
-; SSE41-NEXT:    movdqa %xmm3, %xmm0
-; SSE41-NEXT:    psllw $8, %xmm0
-; SSE41-NEXT:    paddb %xmm3, %xmm0
-; SSE41-NEXT:    psrlw $8, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm1
-; SSE41-NEXT:    pcmpeqw %xmm0, %xmm1
-; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    retq
+; SSE-LABEL: ugt_0_v8i16:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    pcmpeqw %xmm1, %xmm0
+; SSE-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: ugt_0_v8i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX1-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpsllw $8, %xmm0, %xmm1
-; AVX1-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
@@ -2597,17 +2291,6 @@
 ;
 ; AVX2-LABEL: ugt_0_v8i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX2-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX2-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpsllw $8, %xmm0, %xmm1
-; AVX2-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
@@ -2616,9 +2299,6 @@
 ;
 ; AVX512VPOPCNTDQ-LABEL: ugt_0_v8i16:
 ; AVX512VPOPCNTDQ:       # %bb.0:
-; AVX512VPOPCNTDQ-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
-; AVX512VPOPCNTDQ-NEXT:    vpopcntd %zmm0, %zmm0
-; AVX512VPOPCNTDQ-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512VPOPCNTDQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512VPOPCNTDQ-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; AVX512VPOPCNTDQ-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
@@ -2628,13 +2308,9 @@
 ;
 ; AVX512VPOPCNTDQVL-LABEL: ugt_0_v8i16:
 ; AVX512VPOPCNTDQVL:       # %bb.0:
-; AVX512VPOPCNTDQVL-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
-; AVX512VPOPCNTDQVL-NEXT:    vpopcntd %ymm0, %ymm0
-; AVX512VPOPCNTDQVL-NEXT:    vpmovdw %ymm0, %xmm0
 ; AVX512VPOPCNTDQVL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512VPOPCNTDQVL-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
 ; AVX512VPOPCNTDQVL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
-; AVX512VPOPCNTDQVL-NEXT:    vzeroupper
 ; AVX512VPOPCNTDQVL-NEXT:    retq
 ;
 ; BITALG_NOVLX-LABEL: ugt_0_v8i16:
@@ -2662,147 +2338,17 @@
 }
 
 define <8 x i16> @ult_1_v8i16(<8 x i16> %0) {
-; SSE2-LABEL: ult_1_v8i16:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $1, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    psubb %xmm1, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    pand %xmm1, %xmm2
-; SSE2-NEXT:    psrlw $2, %xmm0
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    paddb %xmm2, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $4, %xmm1
-; SSE2-NEXT:    paddb %xmm0, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    psllw $8, %xmm0
-; SSE2-NEXT:    paddb %xmm1, %xmm0
-; SSE2-NEXT:    psrlw $8, %xmm0
-; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    pcmpeqw %xmm1, %xmm0
-; SSE2-NEXT:    retq
-;
-; SSE3-LABEL: ult_1_v8i16:
-; SSE3:       # %bb.0:
-; SSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSE3-NEXT:    psrlw $1, %xmm1
-; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE3-NEXT:    psubb %xmm1, %xmm0
-; SSE3-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE3-NEXT:    movdqa %xmm0, %xmm2
-; SSE3-NEXT:    pand %xmm1, %xmm2
-; SSE3-NEXT:    psrlw $2, %xmm0
-; SSE3-NEXT:    pand %xmm1, %xmm0
-; SSE3-NEXT:    paddb %xmm2, %xmm0
-; SSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSE3-NEXT:    psrlw $4, %xmm1
-; SSE3-NEXT:    paddb %xmm0, %xmm1
-; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE3-NEXT:    movdqa %xmm1, %xmm0
-; SSE3-NEXT:    psllw $8, %xmm0
-; SSE3-NEXT:    paddb %xmm1, %xmm0
-; SSE3-NEXT:    psrlw $8, %xmm0
-; SSE3-NEXT:    pxor %xmm1, %xmm1
-; SSE3-NEXT:    pcmpeqw %xmm1, %xmm0
-; SSE3-NEXT:    retq
-;
-; SSSE3-LABEL: ult_1_v8i16:
-; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT:    movdqa %xmm0, %xmm2
-; SSSE3-NEXT:    pand %xmm1, %xmm2
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSSE3-NEXT:    movdqa %xmm3, %xmm4
-; SSSE3-NEXT:    pshufb %xmm2, %xmm4
-; SSSE3-NEXT:    psrlw $4, %xmm0
-; SSSE3-NEXT:    pand %xmm1, %xmm0
-; SSSE3-NEXT:    pshufb %xmm0, %xmm3
-; SSSE3-NEXT:    paddb %xmm4, %xmm3
-; SSSE3-NEXT:    movdqa %xmm3, %xmm1
-; SSSE3-NEXT:    psllw $8, %xmm1
-; SSSE3-NEXT:    paddb %xmm3, %xmm1
-; SSSE3-NEXT:    psrlw $8, %xmm1
-; SSSE3-NEXT:    pxor %xmm0, %xmm0
-; SSSE3-NEXT:    pcmpeqw %xmm1, %xmm0
-; SSSE3-NEXT:    retq
-;
-; SSE41-LABEL: ult_1_v8i16:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    pand %xmm1, %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pshufb %xmm2, %xmm4
-; SSE41-NEXT:    psrlw $4, %xmm0
-; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    pshufb %xmm0, %xmm3
-; SSE41-NEXT:    paddb %xmm4, %xmm3
-; SSE41-NEXT:    movdqa %xmm3, %xmm1
-; SSE41-NEXT:    psllw $8, %xmm1
-; SSE41-NEXT:    paddb %xmm3, %xmm1
-; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqw %xmm1, %xmm0
-; SSE41-NEXT:    retq
-;
-; AVX1-LABEL: ult_1_v8i16:
-; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX1-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpsllw $8, %xmm0, %xmm1
-; AVX1-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: ult_1_v8i16:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX2-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX2-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpsllw $8, %xmm0, %xmm1
-; AVX2-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    retq
-;
-; AVX512VPOPCNTDQ-LABEL: ult_1_v8i16:
-; AVX512VPOPCNTDQ:       # %bb.0:
-; AVX512VPOPCNTDQ-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
-; AVX512VPOPCNTDQ-NEXT:    vpopcntd %zmm0, %zmm0
-; AVX512VPOPCNTDQ-NEXT:    vpmovdw %zmm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
-; AVX512VPOPCNTDQ-NEXT:    vzeroupper
-; AVX512VPOPCNTDQ-NEXT:    retq
+; SSE-LABEL: ult_1_v8i16:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    pcmpeqw %xmm1, %xmm0
+; SSE-NEXT:    retq
 ;
-; AVX512VPOPCNTDQVL-LABEL: ult_1_v8i16:
-; AVX512VPOPCNTDQVL:       # %bb.0:
-; AVX512VPOPCNTDQVL-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
-; AVX512VPOPCNTDQVL-NEXT:    vpopcntd %ymm0, %ymm0
-; AVX512VPOPCNTDQVL-NEXT:    vpmovdw %ymm0, %xmm0
-; AVX512VPOPCNTDQVL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
-; AVX512VPOPCNTDQVL-NEXT:    vzeroupper
-; AVX512VPOPCNTDQVL-NEXT:    retq
+; AVX-LABEL: ult_1_v8i16:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    retq
 ;
 ; BITALG_NOVLX-LABEL: ult_1_v8i16:
 ; BITALG_NOVLX:       # %bb.0:
@@ -7752,127 +7298,17 @@
 }
 
 define <4 x i32> @ugt_0_v4i32(<4 x i32> %0) {
-; SSE2-LABEL: ugt_0_v4i32:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $1, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    psubb %xmm1, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    pand %xmm1, %xmm2
-; SSE2-NEXT:    psrlw $2, %xmm0
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    paddb %xmm2, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $4, %xmm1
-; SSE2-NEXT:    paddb %xmm0, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm0
-; SSE2-NEXT:    movdqa %xmm1, %xmm2
-; SSE2-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; SSE2-NEXT:    psadbw %xmm0, %xmm2
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE2-NEXT:    psadbw %xmm0, %xmm1
-; SSE2-NEXT:    packuswb %xmm2, %xmm1
-; SSE2-NEXT:    pcmpeqd %xmm0, %xmm1
-; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE2-NEXT:    pxor %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    retq
-;
-; SSE3-LABEL: ugt_0_v4i32:
-; SSE3:       # %bb.0:
-; SSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSE3-NEXT:    psrlw $1, %xmm1
-; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE3-NEXT:    psubb %xmm1, %xmm0
-; SSE3-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE3-NEXT:    movdqa %xmm0, %xmm2
-; SSE3-NEXT:    pand %xmm1, %xmm2
-; SSE3-NEXT:    psrlw $2, %xmm0
-; SSE3-NEXT:    pand %xmm1, %xmm0
-; SSE3-NEXT:    paddb %xmm2, %xmm0
-; SSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSE3-NEXT:    psrlw $4, %xmm1
-; SSE3-NEXT:    paddb %xmm0, %xmm1
-; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE3-NEXT:    pxor %xmm0, %xmm0
-; SSE3-NEXT:    movdqa %xmm1, %xmm2
-; SSE3-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; SSE3-NEXT:    psadbw %xmm0, %xmm2
-; SSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE3-NEXT:    psadbw %xmm0, %xmm1
-; SSE3-NEXT:    packuswb %xmm2, %xmm1
-; SSE3-NEXT:    pcmpeqd %xmm0, %xmm1
-; SSE3-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE3-NEXT:    pxor %xmm0, %xmm1
-; SSE3-NEXT:    movdqa %xmm1, %xmm0
-; SSE3-NEXT:    retq
-;
-; SSSE3-LABEL: ugt_0_v4i32:
-; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT:    movdqa %xmm0, %xmm2
-; SSSE3-NEXT:    pand %xmm1, %xmm2
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSSE3-NEXT:    movdqa %xmm3, %xmm4
-; SSSE3-NEXT:    pshufb %xmm2, %xmm4
-; SSSE3-NEXT:    psrlw $4, %xmm0
-; SSSE3-NEXT:    pand %xmm1, %xmm0
-; SSSE3-NEXT:    pshufb %xmm0, %xmm3
-; SSSE3-NEXT:    paddb %xmm4, %xmm3
-; SSSE3-NEXT:    pxor %xmm0, %xmm0
-; SSSE3-NEXT:    movdqa %xmm3, %xmm1
-; SSSE3-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSSE3-NEXT:    psadbw %xmm0, %xmm1
-; SSSE3-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1]
-; SSSE3-NEXT:    psadbw %xmm0, %xmm3
-; SSSE3-NEXT:    packuswb %xmm1, %xmm3
-; SSSE3-NEXT:    pcmpeqd %xmm0, %xmm3
-; SSSE3-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSSE3-NEXT:    pxor %xmm3, %xmm0
-; SSSE3-NEXT:    retq
-;
-; SSE41-LABEL: ugt_0_v4i32:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    pand %xmm1, %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pshufb %xmm2, %xmm4
-; SSE41-NEXT:    psrlw $4, %xmm0
-; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    pshufb %xmm0, %xmm3
-; SSE41-NEXT:    paddb %xmm4, %xmm3
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm1 = xmm3[0],zero,xmm3[1],zero
-; SSE41-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm0[2],xmm3[3],xmm0[3]
-; SSE41-NEXT:    psadbw %xmm0, %xmm3
-; SSE41-NEXT:    psadbw %xmm0, %xmm1
-; SSE41-NEXT:    packuswb %xmm3, %xmm1
-; SSE41-NEXT:    pcmpeqd %xmm0, %xmm1
-; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    retq
+; SSE-LABEL: ugt_0_v4i32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    pcmpeqd %xmm1, %xmm0
+; SSE-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE-NEXT:    pxor %xmm1, %xmm0
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: ugt_0_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX1-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX1-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; AVX1-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
@@ -7880,20 +7316,7 @@
 ;
 ; AVX2-LABEL: ugt_0_v4i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX2-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX2-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX2-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; AVX2-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
@@ -7910,182 +7333,51 @@
 ; AVX512VPOPCNTDQ-NEXT:    vzeroupper
 ; AVX512VPOPCNTDQ-NEXT:    retq
 ;
-; AVX512VPOPCNTDQVL-LABEL: ugt_0_v4i32:
-; AVX512VPOPCNTDQVL:       # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT:    vpopcntd %xmm0, %xmm0
-; AVX512VPOPCNTDQVL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQVL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
-; AVX512VPOPCNTDQVL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
-; AVX512VPOPCNTDQVL-NEXT:    retq
-;
-; BITALG_NOVLX-LABEL: ugt_0_v4i32:
-; BITALG_NOVLX:       # %bb.0:
-; BITALG_NOVLX-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; BITALG_NOVLX-NEXT:    vpopcntb %zmm0, %zmm0
-; BITALG_NOVLX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; BITALG_NOVLX-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
-; BITALG_NOVLX-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; BITALG_NOVLX-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; BITALG_NOVLX-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
-; BITALG_NOVLX-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
-; BITALG_NOVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
-; BITALG_NOVLX-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; BITALG_NOVLX-NEXT:    vzeroupper
-; BITALG_NOVLX-NEXT:    retq
-;
-; BITALG-LABEL: ugt_0_v4i32:
-; BITALG:       # %bb.0:
-; BITALG-NEXT:    vpopcntb %xmm0, %xmm0
-; BITALG-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; BITALG-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
-; BITALG-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; BITALG-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; BITALG-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
-; BITALG-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
-; BITALG-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
-; BITALG-NEXT:    retq
-  %2 = tail call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %0)
-  %3 = icmp ugt <4 x i32> %2, <i32 0, i32 0, i32 0, i32 0>
-  %4 = sext <4 x i1> %3 to <4 x i32>
-  ret <4 x i32> %4
-}
-
-define <4 x i32> @ult_1_v4i32(<4 x i32> %0) {
-; SSE2-LABEL: ult_1_v4i32:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $1, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    psubb %xmm1, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    pand %xmm1, %xmm2
-; SSE2-NEXT:    psrlw $2, %xmm0
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    paddb %xmm2, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $4, %xmm1
-; SSE2-NEXT:    paddb %xmm0, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm0
-; SSE2-NEXT:    movdqa %xmm1, %xmm2
-; SSE2-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; SSE2-NEXT:    psadbw %xmm0, %xmm2
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE2-NEXT:    psadbw %xmm0, %xmm1
-; SSE2-NEXT:    packuswb %xmm2, %xmm1
-; SSE2-NEXT:    pcmpeqd %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    retq
-;
-; SSE3-LABEL: ult_1_v4i32:
-; SSE3:       # %bb.0:
-; SSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSE3-NEXT:    psrlw $1, %xmm1
-; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE3-NEXT:    psubb %xmm1, %xmm0
-; SSE3-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE3-NEXT:    movdqa %xmm0, %xmm2
-; SSE3-NEXT:    pand %xmm1, %xmm2
-; SSE3-NEXT:    psrlw $2, %xmm0
-; SSE3-NEXT:    pand %xmm1, %xmm0
-; SSE3-NEXT:    paddb %xmm2, %xmm0
-; SSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSE3-NEXT:    psrlw $4, %xmm1
-; SSE3-NEXT:    paddb %xmm0, %xmm1
-; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE3-NEXT:    pxor %xmm0, %xmm0
-; SSE3-NEXT:    movdqa %xmm1, %xmm2
-; SSE3-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; SSE3-NEXT:    psadbw %xmm0, %xmm2
-; SSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE3-NEXT:    psadbw %xmm0, %xmm1
-; SSE3-NEXT:    packuswb %xmm2, %xmm1
-; SSE3-NEXT:    pcmpeqd %xmm0, %xmm1
-; SSE3-NEXT:    movdqa %xmm1, %xmm0
-; SSE3-NEXT:    retq
-;
-; SSSE3-LABEL: ult_1_v4i32:
-; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT:    movdqa %xmm0, %xmm3
-; SSSE3-NEXT:    pand %xmm2, %xmm3
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSSE3-NEXT:    movdqa %xmm1, %xmm4
-; SSSE3-NEXT:    pshufb %xmm3, %xmm4
-; SSSE3-NEXT:    psrlw $4, %xmm0
-; SSSE3-NEXT:    pand %xmm2, %xmm0
-; SSSE3-NEXT:    pshufb %xmm0, %xmm1
-; SSSE3-NEXT:    paddb %xmm4, %xmm1
-; SSSE3-NEXT:    pxor %xmm0, %xmm0
-; SSSE3-NEXT:    movdqa %xmm1, %xmm2
-; SSSE3-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; SSSE3-NEXT:    psadbw %xmm0, %xmm2
-; SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSSE3-NEXT:    psadbw %xmm0, %xmm1
-; SSSE3-NEXT:    packuswb %xmm2, %xmm1
-; SSSE3-NEXT:    pcmpeqd %xmm0, %xmm1
-; SSSE3-NEXT:    movdqa %xmm1, %xmm0
-; SSSE3-NEXT:    retq
-;
-; SSE41-LABEL: ult_1_v4i32:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    pand %xmm1, %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pshufb %xmm2, %xmm4
-; SSE41-NEXT:    psrlw $4, %xmm0
-; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    pshufb %xmm0, %xmm3
-; SSE41-NEXT:    paddb %xmm4, %xmm3
-; SSE41-NEXT:    pxor %xmm1, %xmm1
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm3[0],zero,xmm3[1],zero
-; SSE41-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm1[2],xmm3[3],xmm1[3]
-; SSE41-NEXT:    psadbw %xmm1, %xmm3
-; SSE41-NEXT:    psadbw %xmm1, %xmm0
-; SSE41-NEXT:    packuswb %xmm3, %xmm0
-; SSE41-NEXT:    pcmpeqd %xmm1, %xmm0
-; SSE41-NEXT:    retq
+; AVX512VPOPCNTDQVL-LABEL: ugt_0_v4i32:
+; AVX512VPOPCNTDQVL:       # %bb.0:
+; AVX512VPOPCNTDQVL-NEXT:    vpopcntd %xmm0, %xmm0
+; AVX512VPOPCNTDQVL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512VPOPCNTDQVL-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; AVX512VPOPCNTDQVL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
+; AVX512VPOPCNTDQVL-NEXT:    retq
+;
+; BITALG_NOVLX-LABEL: ugt_0_v4i32:
+; BITALG_NOVLX:       # %bb.0:
+; BITALG_NOVLX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; BITALG_NOVLX-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; BITALG_NOVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
+; BITALG_NOVLX-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; BITALG_NOVLX-NEXT:    vzeroupper
+; BITALG_NOVLX-NEXT:    retq
+;
+; BITALG-LABEL: ugt_0_v4i32:
+; BITALG:       # %bb.0:
+; BITALG-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; BITALG-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
+; BITALG-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
+; BITALG-NEXT:    retq
+  %2 = tail call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %0)
+  %3 = icmp ugt <4 x i32> %2, <i32 0, i32 0, i32 0, i32 0>
+  %4 = sext <4 x i1> %3 to <4 x i32>
+  ret <4 x i32> %4
+}
+
+define <4 x i32> @ult_1_v4i32(<4 x i32> %0) {
+; SSE-LABEL: ult_1_v4i32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    pcmpeqd %xmm1, %xmm0
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: ult_1_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX1-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX1-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; AVX1-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: ult_1_v4i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX2-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX2-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX2-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; AVX2-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
@@ -8107,27 +7399,13 @@
 ;
 ; BITALG_NOVLX-LABEL: ult_1_v4i32:
 ; BITALG_NOVLX:       # %bb.0:
-; BITALG_NOVLX-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; BITALG_NOVLX-NEXT:    vpopcntb %zmm0, %zmm0
 ; BITALG_NOVLX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; BITALG_NOVLX-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
-; BITALG_NOVLX-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; BITALG_NOVLX-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; BITALG_NOVLX-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; BITALG_NOVLX-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
-; BITALG_NOVLX-NEXT:    vzeroupper
 ; BITALG_NOVLX-NEXT:    retq
 ;
 ; BITALG-LABEL: ult_1_v4i32:
 ; BITALG:       # %bb.0:
-; BITALG-NEXT:    vpopcntb %xmm0, %xmm0
 ; BITALG-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; BITALG-NEXT:    vpsadbw %xmm1, %xmm2, %xmm2
-; BITALG-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; BITALG-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
-; BITALG-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; BITALG-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; BITALG-NEXT:    retq
   %2 = tail call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %0)
@@ -20090,108 +19368,45 @@
 define <2 x i64> @ugt_0_v2i64(<2 x i64> %0) {
 ; SSE2-LABEL: ugt_0_v2i64:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $1, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    psubb %xmm1, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    pand %xmm1, %xmm2
-; SSE2-NEXT:    psrlw $2, %xmm0
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    paddb %xmm2, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $4, %xmm1
-; SSE2-NEXT:    paddb %xmm0, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm0
-; SSE2-NEXT:    psadbw %xmm0, %xmm1
+; SSE2-NEXT:    pxor %xmm1, %xmm1
 ; SSE2-NEXT:    pcmpeqd %xmm0, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,0,3,2]
-; SSE2-NEXT:    pand %xmm0, %xmm1
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,0,3,2]
+; SSE2-NEXT:    pand %xmm1, %xmm2
 ; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE2-NEXT:    pxor %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE3-LABEL: ugt_0_v2i64:
 ; SSE3:       # %bb.0:
-; SSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSE3-NEXT:    psrlw $1, %xmm1
-; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE3-NEXT:    psubb %xmm1, %xmm0
-; SSE3-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE3-NEXT:    movdqa %xmm0, %xmm2
-; SSE3-NEXT:    pand %xmm1, %xmm2
-; SSE3-NEXT:    psrlw $2, %xmm0
-; SSE3-NEXT:    pand %xmm1, %xmm0
-; SSE3-NEXT:    paddb %xmm2, %xmm0
-; SSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSE3-NEXT:    psrlw $4, %xmm1
-; SSE3-NEXT:    paddb %xmm0, %xmm1
-; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE3-NEXT:    pxor %xmm0, %xmm0
-; SSE3-NEXT:    psadbw %xmm0, %xmm1
+; SSE3-NEXT:    pxor %xmm1, %xmm1
 ; SSE3-NEXT:    pcmpeqd %xmm0, %xmm1
-; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,0,3,2]
-; SSE3-NEXT:    pand %xmm0, %xmm1
+; SSE3-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,0,3,2]
+; SSE3-NEXT:    pand %xmm1, %xmm2
 ; SSE3-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE3-NEXT:    pxor %xmm0, %xmm1
-; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    pxor %xmm2, %xmm0
 ; SSE3-NEXT:    retq
 ;
 ; SSSE3-LABEL: ugt_0_v2i64:
 ; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT:    movdqa %xmm0, %xmm2
+; SSSE3-NEXT:    pxor %xmm1, %xmm1
+; SSSE3-NEXT:    pcmpeqd %xmm0, %xmm1
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,0,3,2]
 ; SSSE3-NEXT:    pand %xmm1, %xmm2
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSSE3-NEXT:    movdqa %xmm3, %xmm4
-; SSSE3-NEXT:    pshufb %xmm2, %xmm4
-; SSSE3-NEXT:    psrlw $4, %xmm0
-; SSSE3-NEXT:    pand %xmm1, %xmm0
-; SSSE3-NEXT:    pshufb %xmm0, %xmm3
-; SSSE3-NEXT:    paddb %xmm4, %xmm3
-; SSSE3-NEXT:    pxor %xmm0, %xmm0
-; SSSE3-NEXT:    psadbw %xmm0, %xmm3
-; SSSE3-NEXT:    pcmpeqd %xmm0, %xmm3
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,0,3,2]
-; SSSE3-NEXT:    pand %xmm3, %xmm1
 ; SSSE3-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSSE3-NEXT:    pxor %xmm1, %xmm0
+; SSSE3-NEXT:    pxor %xmm2, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: ugt_0_v2i64:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    pand %xmm1, %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pshufb %xmm2, %xmm4
-; SSE41-NEXT:    psrlw $4, %xmm0
-; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    pshufb %xmm0, %xmm3
-; SSE41-NEXT:    paddb %xmm4, %xmm3
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    psadbw %xmm0, %xmm3
-; SSE41-NEXT:    pcmpeqq %xmm0, %xmm3
-; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm3, %xmm0
+; SSE41-NEXT:    pxor %xmm1, %xmm1
+; SSE41-NEXT:    pcmpeqq %xmm1, %xmm0
+; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: ugt_0_v2i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX1-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqq %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
@@ -20199,16 +19414,7 @@
 ;
 ; AVX2-LABEL: ugt_0_v2i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX2-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX2-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqq %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
@@ -20235,10 +19441,7 @@
 ;
 ; BITALG_NOVLX-LABEL: ugt_0_v2i64:
 ; BITALG_NOVLX:       # %bb.0:
-; BITALG_NOVLX-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; BITALG_NOVLX-NEXT:    vpopcntb %zmm0, %zmm0
 ; BITALG_NOVLX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
 ; BITALG_NOVLX-NEXT:    vpcmpeqq %xmm1, %xmm0, %xmm0
 ; BITALG_NOVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; BITALG_NOVLX-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
@@ -20247,9 +19450,7 @@
 ;
 ; BITALG-LABEL: ugt_0_v2i64:
 ; BITALG:       # %bb.0:
-; BITALG-NEXT:    vpopcntb %xmm0, %xmm0
 ; BITALG-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
 ; BITALG-NEXT:    vpcmpeqq %xmm1, %xmm0, %xmm0
 ; BITALG-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; BITALG-NEXT:    retq
@@ -20262,116 +19463,43 @@
 define <2 x i64> @ult_1_v2i64(<2 x i64> %0) {
 ; SSE2-LABEL: ult_1_v2i64:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $1, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    psubb %xmm1, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    pand %xmm1, %xmm2
-; SSE2-NEXT:    psrlw $2, %xmm0
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    paddb %xmm2, %xmm0
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrlw $4, %xmm1
-; SSE2-NEXT:    paddb %xmm0, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm0
-; SSE2-NEXT:    psadbw %xmm0, %xmm1
+; SSE2-NEXT:    pxor %xmm1, %xmm1
 ; SSE2-NEXT:    pcmpeqd %xmm0, %xmm1
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,0,3,2]
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    pand %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE3-LABEL: ult_1_v2i64:
 ; SSE3:       # %bb.0:
-; SSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSE3-NEXT:    psrlw $1, %xmm1
-; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE3-NEXT:    psubb %xmm1, %xmm0
-; SSE3-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
-; SSE3-NEXT:    movdqa %xmm0, %xmm2
-; SSE3-NEXT:    pand %xmm1, %xmm2
-; SSE3-NEXT:    psrlw $2, %xmm0
-; SSE3-NEXT:    pand %xmm1, %xmm0
-; SSE3-NEXT:    paddb %xmm2, %xmm0
-; SSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSE3-NEXT:    psrlw $4, %xmm1
-; SSE3-NEXT:    paddb %xmm0, %xmm1
-; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE3-NEXT:    pxor %xmm0, %xmm0
-; SSE3-NEXT:    psadbw %xmm0, %xmm1
+; SSE3-NEXT:    pxor %xmm1, %xmm1
 ; SSE3-NEXT:    pcmpeqd %xmm0, %xmm1
 ; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,0,3,2]
-; SSE3-NEXT:    pand %xmm0, %xmm1
-; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    pand %xmm1, %xmm0
 ; SSE3-NEXT:    retq
 ;
 ; SSSE3-LABEL: ult_1_v2i64:
 ; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT:    movdqa %xmm0, %xmm2
-; SSSE3-NEXT:    pand %xmm1, %xmm2
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSSE3-NEXT:    movdqa %xmm3, %xmm4
-; SSSE3-NEXT:    pshufb %xmm2, %xmm4
-; SSSE3-NEXT:    psrlw $4, %xmm0
+; SSSE3-NEXT:    pxor %xmm1, %xmm1
+; SSSE3-NEXT:    pcmpeqd %xmm0, %xmm1
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,0,3,2]
 ; SSSE3-NEXT:    pand %xmm1, %xmm0
-; SSSE3-NEXT:    pshufb %xmm0, %xmm3
-; SSSE3-NEXT:    paddb %xmm4, %xmm3
-; SSSE3-NEXT:    pxor %xmm0, %xmm0
-; SSSE3-NEXT:    psadbw %xmm0, %xmm3
-; SSSE3-NEXT:    pcmpeqd %xmm0, %xmm3
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,0,3,2]
-; SSSE3-NEXT:    pand %xmm3, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: ult_1_v2i64:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand %xmm2, %xmm3
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; SSE41-NEXT:    movdqa %xmm1, %xmm4
-; SSE41-NEXT:    pshufb %xmm3, %xmm4
-; SSE41-NEXT:    psrlw $4, %xmm0
-; SSE41-NEXT:    pand %xmm2, %xmm0
-; SSE41-NEXT:    pshufb %xmm0, %xmm1
-; SSE41-NEXT:    paddb %xmm4, %xmm1
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    psadbw %xmm0, %xmm1
-; SSE41-NEXT:    pcmpeqq %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm1, %xmm1
+; SSE41-NEXT:    pcmpeqq %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: ult_1_v2i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX1-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqq %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: ult_1_v2i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX2-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX2-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqq %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
@@ -20393,19 +19521,13 @@
 ;
 ; BITALG_NOVLX-LABEL: ult_1_v2i64:
 ; BITALG_NOVLX:       # %bb.0:
-; BITALG_NOVLX-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; BITALG_NOVLX-NEXT:    vpopcntb %zmm0, %zmm0
 ; BITALG_NOVLX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
 ; BITALG_NOVLX-NEXT:    vpcmpeqq %xmm1, %xmm0, %xmm0
-; BITALG_NOVLX-NEXT:    vzeroupper
 ; BITALG_NOVLX-NEXT:    retq
 ;
 ; BITALG-LABEL: ult_1_v2i64:
 ; BITALG:       # %bb.0:
-; BITALG-NEXT:    vpopcntb %xmm0, %xmm0
 ; BITALG-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT:    vpsadbw %xmm1, %xmm0, %xmm0
 ; BITALG-NEXT:    vpcmpeqq %xmm1, %xmm0, %xmm0
 ; BITALG-NEXT:    retq
   %2 = tail call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %0)
diff --git a/i/llvm/test/CodeGen/X86/vector-popcnt-256-ult-ugt.ll b/llvm/test/CodeGen/X86/vector-popcnt-256-ult-ugt.ll
--- a/i/llvm/test/CodeGen/X86/vector-popcnt-256-ult-ugt.ll
+++ b/llvm/test/CodeGen/X86/vector-popcnt-256-ult-ugt.ll
@@ -21,40 +21,18 @@
 define <32 x i8> @ugt_0_v32i8(<32 x i8> %0) {
 ; AVX1-LABEL: ugt_0_v32i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX1-NEXT:    vpand %xmm1, %xmm4, %xmm4
-; AVX1-NEXT:    vpshufb %xmm4, %xmm3, %xmm4
-; AVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm4
-; AVX1-NEXT:    vpshufb %xmm4, %xmm3, %xmm4
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX1-NEXT:    vpaddb %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpcmpeqb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpxor %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpxor %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX1-NEXT:    vpcmpeqb %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: ugt_0_v32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
@@ -63,14 +41,6 @@
 ;
 ; AVX512VPOPCNTDQ-LABEL: ugt_0_v32i8:
 ; AVX512VPOPCNTDQ:       # %bb.0:
-; AVX512VPOPCNTDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VPOPCNTDQ-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VPOPCNTDQ-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512VPOPCNTDQ-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VPOPCNTDQ-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX512VPOPCNTDQ-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
 ; AVX512VPOPCNTDQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512VPOPCNTDQ-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
 ; AVX512VPOPCNTDQ-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
@@ -79,14 +49,6 @@
 ;
 ; AVX512VPOPCNTDQVL-LABEL: ugt_0_v32i8:
 ; AVX512VPOPCNTDQVL:       # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VPOPCNTDQVL-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VPOPCNTDQVL-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512VPOPCNTDQVL-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VPOPCNTDQVL-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512VPOPCNTDQVL-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX512VPOPCNTDQVL-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX512VPOPCNTDQVL-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
 ; AVX512VPOPCNTDQVL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512VPOPCNTDQVL-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
 ; AVX512VPOPCNTDQVL-NEXT:    vpternlogq $15, %ymm0, %ymm0, %ymm0
@@ -118,65 +80,27 @@
 define <32 x i8> @ult_1_v32i8(<32 x i8> %0) {
 ; AVX1-LABEL: ult_1_v32i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX1-NEXT:    vpand %xmm1, %xmm4, %xmm4
-; AVX1-NEXT:    vpshufb %xmm4, %xmm3, %xmm4
-; AVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm4
-; AVX1-NEXT:    vpshufb %xmm4, %xmm3, %xmm4
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX1-NEXT:    vpaddb %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm2, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpcmpeqb %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpcmpeqb %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: ult_1_v32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512VPOPCNTDQ-LABEL: ult_1_v32i8:
 ; AVX512VPOPCNTDQ:       # %bb.0:
-; AVX512VPOPCNTDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VPOPCNTDQ-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VPOPCNTDQ-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512VPOPCNTDQ-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VPOPCNTDQ-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX512VPOPCNTDQ-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
 ; AVX512VPOPCNTDQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512VPOPCNTDQ-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
 ; AVX512VPOPCNTDQ-NEXT:    retq
 ;
 ; AVX512VPOPCNTDQVL-LABEL: ult_1_v32i8:
 ; AVX512VPOPCNTDQVL:       # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VPOPCNTDQVL-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VPOPCNTDQVL-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512VPOPCNTDQVL-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VPOPCNTDQVL-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512VPOPCNTDQVL-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX512VPOPCNTDQVL-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX512VPOPCNTDQVL-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
 ; AVX512VPOPCNTDQVL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512VPOPCNTDQVL-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
 ; AVX512VPOPCNTDQVL-NEXT:    retq
@@ -1569,49 +1493,18 @@
 define <16 x i16> @ugt_0_v16i16(<16 x i16> %0) {
 ; AVX1-LABEL: ugt_0_v16i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX1-NEXT:    vpand %xmm1, %xmm4, %xmm4
-; AVX1-NEXT:    vpshufb %xmm4, %xmm3, %xmm4
-; AVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
-; AVX1-NEXT:    vpsllw $8, %xmm2, %xmm4
-; AVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
-; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm4
-; AVX1-NEXT:    vpshufb %xmm4, %xmm3, %xmm4
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX1-NEXT:    vpaddb %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpsllw $8, %xmm0, %xmm1
-; AVX1-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpcmpeqw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpxor %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpxor %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX1-NEXT:    vpcmpeqw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: ugt_0_v16i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
-; AVX2-NEXT:    vpsllw $8, %ymm0, %ymm1
-; AVX2-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
-; AVX2-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
@@ -1620,9 +1513,6 @@
 ;
 ; AVX512VPOPCNTDQ-LABEL: ugt_0_v16i16:
 ; AVX512VPOPCNTDQ:       # %bb.0:
-; AVX512VPOPCNTDQ-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512VPOPCNTDQ-NEXT:    vpopcntd %zmm0, %zmm0
-; AVX512VPOPCNTDQ-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512VPOPCNTDQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512VPOPCNTDQ-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; AVX512VPOPCNTDQ-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
@@ -1631,9 +1521,6 @@
 ;
 ; AVX512VPOPCNTDQVL-LABEL: ugt_0_v16i16:
 ; AVX512VPOPCNTDQVL:       # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512VPOPCNTDQVL-NEXT:    vpopcntd %zmm0, %zmm0
-; AVX512VPOPCNTDQVL-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512VPOPCNTDQVL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512VPOPCNTDQVL-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; AVX512VPOPCNTDQVL-NEXT:    vpternlogq $15, %ymm0, %ymm0, %ymm0
@@ -1665,64 +1552,27 @@
 define <16 x i16> @ult_1_v16i16(<16 x i16> %0) {
 ; AVX1-LABEL: ult_1_v16i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX1-NEXT:    vpand %xmm1, %xmm4, %xmm4
-; AVX1-NEXT:    vpshufb %xmm4, %xmm3, %xmm4
-; AVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
-; AVX1-NEXT:    vpsllw $8, %xmm2, %xmm4
-; AVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
-; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm4
-; AVX1-NEXT:    vpshufb %xmm4, %xmm3, %xmm4
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX1-NEXT:    vpaddb %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpsllw $8, %xmm0, %xmm1
-; AVX1-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm2, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpcmpeqw %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpcmpeqw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: ult_1_v16i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
-; AVX2-NEXT:    vpsllw $8, %ymm0, %ymm1
-; AVX2-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
-; AVX2-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512VPOPCNTDQ-LABEL: ult_1_v16i16:
 ; AVX512VPOPCNTDQ:       # %bb.0:
-; AVX512VPOPCNTDQ-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512VPOPCNTDQ-NEXT:    vpopcntd %zmm0, %zmm0
-; AVX512VPOPCNTDQ-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512VPOPCNTDQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512VPOPCNTDQ-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; AVX512VPOPCNTDQ-NEXT:    retq
 ;
 ; AVX512VPOPCNTDQVL-LABEL: ult_1_v16i16:
 ; AVX512VPOPCNTDQVL:       # %bb.0:
-; AVX512VPOPCNTDQVL-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512VPOPCNTDQVL-NEXT:    vpopcntd %zmm0, %zmm0
-; AVX512VPOPCNTDQVL-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512VPOPCNTDQVL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512VPOPCNTDQVL-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
 ; AVX512VPOPCNTDQVL-NEXT:    retq
@@ -4380,56 +4230,19 @@
 define <8 x i32> @ugt_0_v8i32(<8 x i32> %0) {
 ; AVX1-LABEL: ugt_0_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX1-NEXT:    vpand %xmm1, %xmm4, %xmm4
-; AVX1-NEXT:    vpshufb %xmm4, %xmm3, %xmm4
-; AVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
-; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm5 = xmm2[2],xmm4[2],xmm2[3],xmm4[3]
-; AVX1-NEXT:    vpsadbw %xmm4, %xmm5, %xmm5
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
-; AVX1-NEXT:    vpsadbw %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vpackuswb %xmm5, %xmm2, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm5
-; AVX1-NEXT:    vpshufb %xmm5, %xmm3, %xmm5
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX1-NEXT:    vpaddb %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm1 = xmm0[2],xmm4[2],xmm0[3],xmm4[3]
-; AVX1-NEXT:    vpsadbw %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; AVX1-NEXT:    vpsadbw %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vpxor %xmm1, %xmm2, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpcmpeqd %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX1-NEXT:    vpxor %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpcmpeqd %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: ugt_0_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
-; AVX2-NEXT:    vpsadbw %ymm1, %ymm2, %ymm2
-; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
-; AVX2-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpcmpeqd %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
 ; AVX2-NEXT:    vpxor %ymm1, %ymm0, %ymm0
@@ -4455,14 +4268,7 @@
 ;
 ; BITALG_NOVLX-LABEL: ugt_0_v8i32:
 ; BITALG_NOVLX:       # %bb.0:
-; BITALG_NOVLX-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; BITALG_NOVLX-NEXT:    vpopcntb %zmm0, %zmm0
 ; BITALG_NOVLX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
-; BITALG_NOVLX-NEXT:    vpsadbw %ymm1, %ymm2, %ymm2
-; BITALG_NOVLX-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
-; BITALG_NOVLX-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
-; BITALG_NOVLX-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
 ; BITALG_NOVLX-NEXT:    vpcmpeqd %ymm1, %ymm0, %ymm0
 ; BITALG_NOVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; BITALG_NOVLX-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
@@ -4470,13 +4276,7 @@
 ;
 ; BITALG-LABEL: ugt_0_v8i32:
 ; BITALG:       # %bb.0:
-; BITALG-NEXT:    vpopcntb %ymm0, %ymm0
 ; BITALG-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
-; BITALG-NEXT:    vpsadbw %ymm1, %ymm2, %ymm2
-; BITALG-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
-; BITALG-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
-; BITALG-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
 ; BITALG-NEXT:    vpcmpeqd %ymm1, %ymm0, %ymm0
 ; BITALG-NEXT:    vpternlogq $15, %ymm0, %ymm0, %ymm0
 ; BITALG-NEXT:    retq
@@ -4489,53 +4289,16 @@
 define <8 x i32> @ult_1_v8i32(<8 x i32> %0) {
 ; AVX1-LABEL: ult_1_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX1-NEXT:    vpand %xmm1, %xmm4, %xmm4
-; AVX1-NEXT:    vpshufb %xmm4, %xmm3, %xmm4
-; AVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
-; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm5 = xmm2[2],xmm4[2],xmm2[3],xmm4[3]
-; AVX1-NEXT:    vpsadbw %xmm4, %xmm5, %xmm5
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
-; AVX1-NEXT:    vpsadbw %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vpackuswb %xmm5, %xmm2, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm5
-; AVX1-NEXT:    vpshufb %xmm5, %xmm3, %xmm5
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX1-NEXT:    vpaddb %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm1 = xmm0[2],xmm4[2],xmm0[3],xmm4[3]
-; AVX1-NEXT:    vpsadbw %xmm4, %xmm1, %xmm1
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; AVX1-NEXT:    vpsadbw %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm2, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpcmpeqd %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpcmpeqd %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: ult_1_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
-; AVX2-NEXT:    vpsadbw %ymm1, %ymm2, %ymm2
-; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
-; AVX2-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpcmpeqd %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
@@ -4556,26 +4319,13 @@
 ;
 ; BITALG_NOVLX-LABEL: ult_1_v8i32:
 ; BITALG_NOVLX:       # %bb.0:
-; BITALG_NOVLX-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; BITALG_NOVLX-NEXT:    vpopcntb %zmm0, %zmm0
 ; BITALG_NOVLX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
-; BITALG_NOVLX-NEXT:    vpsadbw %ymm1, %ymm2, %ymm2
-; BITALG_NOVLX-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
-; BITALG_NOVLX-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
-; BITALG_NOVLX-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
 ; BITALG_NOVLX-NEXT:    vpcmpeqd %ymm1, %ymm0, %ymm0
 ; BITALG_NOVLX-NEXT:    retq
 ;
 ; BITALG-LABEL: ult_1_v8i32:
 ; BITALG:       # %bb.0:
-; BITALG-NEXT:    vpopcntb %ymm0, %ymm0
 ; BITALG-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
-; BITALG-NEXT:    vpsadbw %ymm1, %ymm2, %ymm2
-; BITALG-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
-; BITALG-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
-; BITALG-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
 ; BITALG-NEXT:    vpcmpeqd %ymm1, %ymm0, %ymm0
 ; BITALG-NEXT:    retq
   %2 = tail call <8 x i32> @llvm.ctpop.v8i32(<8 x i32> %0)
@@ -11178,44 +10928,19 @@
 define <4 x i64> @ugt_0_v4i64(<4 x i64> %0) {
 ; AVX1-LABEL: ugt_0_v4i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX1-NEXT:    vpand %xmm1, %xmm4, %xmm4
-; AVX1-NEXT:    vpshufb %xmm4, %xmm3, %xmm4
-; AVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
-; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpsadbw %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm5
-; AVX1-NEXT:    vpshufb %xmm5, %xmm3, %xmm5
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX1-NEXT:    vpaddb %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vpsadbw %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqq %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqq %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vpxor %xmm1, %xmm2, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpcmpeqq %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
+; AVX1-NEXT:    vpxor %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpcmpeqq %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: ugt_0_v4i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpcmpeqq %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
 ; AVX2-NEXT:    vpxor %ymm1, %ymm0, %ymm0
@@ -11241,10 +10966,7 @@
 ;
 ; BITALG_NOVLX-LABEL: ugt_0_v4i64:
 ; BITALG_NOVLX:       # %bb.0:
-; BITALG_NOVLX-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; BITALG_NOVLX-NEXT:    vpopcntb %zmm0, %zmm0
 ; BITALG_NOVLX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
 ; BITALG_NOVLX-NEXT:    vpcmpeqq %ymm1, %ymm0, %ymm0
 ; BITALG_NOVLX-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; BITALG_NOVLX-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
@@ -11252,9 +10974,7 @@
 ;
 ; BITALG-LABEL: ugt_0_v4i64:
 ; BITALG:       # %bb.0:
-; BITALG-NEXT:    vpopcntb %ymm0, %ymm0
 ; BITALG-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
 ; BITALG-NEXT:    vpcmpeqq %ymm1, %ymm0, %ymm0
 ; BITALG-NEXT:    vpternlogq $15, %ymm0, %ymm0, %ymm0
 ; BITALG-NEXT:    retq
@@ -11267,41 +10987,16 @@
 define <4 x i64> @ult_1_v4i64(<4 x i64> %0) {
 ; AVX1-LABEL: ult_1_v4i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX1-NEXT:    vpand %xmm1, %xmm4, %xmm4
-; AVX1-NEXT:    vpshufb %xmm4, %xmm3, %xmm4
-; AVX1-NEXT:    vpaddb %xmm2, %xmm4, %xmm2
-; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpsadbw %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm5
-; AVX1-NEXT:    vpshufb %xmm5, %xmm3, %xmm5
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX1-NEXT:    vpaddb %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vpsadbw %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqq %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqq %xmm4, %xmm2, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpcmpeqq %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpcmpeqq %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: ult_1_v4i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpcmpeqq %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
@@ -11322,18 +11017,13 @@
 ;
 ; BITALG_NOVLX-LABEL: ult_1_v4i64:
 ; BITALG_NOVLX:       # %bb.0:
-; BITALG_NOVLX-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; BITALG_NOVLX-NEXT:    vpopcntb %zmm0, %zmm0
 ; BITALG_NOVLX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG_NOVLX-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
 ; BITALG_NOVLX-NEXT:    vpcmpeqq %ymm1, %ymm0, %ymm0
 ; BITALG_NOVLX-NEXT:    retq
 ;
 ; BITALG-LABEL: ult_1_v4i64:
 ; BITALG:       # %bb.0:
-; BITALG-NEXT:    vpopcntb %ymm0, %ymm0
 ; BITALG-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
 ; BITALG-NEXT:    vpcmpeqq %ymm1, %ymm0, %ymm0
 ; BITALG-NEXT:    retq
   %2 = tail call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %0)
diff --git a/i/llvm/test/CodeGen/X86/vector-popcnt-512-ult-ugt.ll b/llvm/test/CodeGen/X86/vector-popcnt-512-ult-ugt.ll
--- a/i/llvm/test/CodeGen/X86/vector-popcnt-512-ult-ugt.ll
+++ b/llvm/test/CodeGen/X86/vector-popcnt-512-ult-ugt.ll
@@ -20,76 +20,32 @@
 define <64 x i8> @ugt_0_v64i8(<64 x i8> %0) {
 ; AVX512F-LABEL: ugt_0_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512F-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512F-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512F-NEXT:    vpand %ymm1, %ymm4, %ymm4
-; AVX512F-NEXT:    vpshufb %ymm4, %ymm3, %ymm4
-; AVX512F-NEXT:    vpaddb %ymm2, %ymm4, %ymm2
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512F-NEXT:    vpand %ymm1, %ymm0, %ymm4
-; AVX512F-NEXT:    vpshufb %ymm4, %ymm3, %ymm4
-; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX512F-NEXT:    vpaddb %ymm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512F-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpcmpeqb %ymm1, %ymm2, %ymm1
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
+; AVX512F-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512F-NEXT:    vpcmpeqb %ymm2, %ymm1, %ymm1
+; AVX512F-NEXT:    vpcmpeqb %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: ugt_0_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm2
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
-; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
-; AVX512BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vptestmb %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VPOPCNTDQ-NOBW-LABEL: ugt_0_v64i8:
 ; AVX512VPOPCNTDQ-NOBW:       # %bb.0:
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpand %ymm1, %ymm4, %ymm4
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpshufb %ymm4, %ymm3, %ymm4
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpaddb %ymm2, %ymm4, %ymm2
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpand %ymm1, %ymm0, %ymm4
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpshufb %ymm4, %ymm3, %ymm4
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpaddb %ymm4, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpcmpeqb %ymm1, %ymm2, %ymm1
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512VPOPCNTDQ-NOBW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
+; AVX512VPOPCNTDQ-NOBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VPOPCNTDQ-NOBW-NEXT:    vpcmpeqb %ymm2, %ymm1, %ymm1
+; AVX512VPOPCNTDQ-NOBW-NEXT:    vpcmpeqb %ymm2, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NOBW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512VPOPCNTDQ-NOBW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512VPOPCNTDQ-NOBW-NEXT:    retq
 ;
 ; AVX512VPOPCNTDQ-BW-LABEL: ugt_0_v64i8:
 ; AVX512VPOPCNTDQ-BW:       # %bb.0:
-; AVX512VPOPCNTDQ-BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VPOPCNTDQ-BW-NEXT:    vpandq %zmm1, %zmm0, %zmm2
-; AVX512VPOPCNTDQ-BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512VPOPCNTDQ-BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
-; AVX512VPOPCNTDQ-BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
-; AVX512VPOPCNTDQ-BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
-; AVX512VPOPCNTDQ-BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
-; AVX512VPOPCNTDQ-BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
 ; AVX512VPOPCNTDQ-BW-NEXT:    vptestmb %zmm0, %zmm0, %k0
 ; AVX512VPOPCNTDQ-BW-NEXT:    vpmovm2b %k0, %zmm0
 ; AVX512VPOPCNTDQ-BW-NEXT:    retq
@@ -109,74 +65,30 @@
 define <64 x i8> @ult_1_v64i8(<64 x i8> %0) {
 ; AVX512F-LABEL: ult_1_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512F-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512F-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512F-NEXT:    vpand %ymm1, %ymm4, %ymm4
-; AVX512F-NEXT:    vpshufb %ymm4, %ymm3, %ymm4
-; AVX512F-NEXT:    vpaddb %ymm2, %ymm4, %ymm2
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512F-NEXT:    vpand %ymm1, %ymm0, %ymm4
-; AVX512F-NEXT:    vpshufb %ymm4, %ymm3, %ymm4
-; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX512F-NEXT:    vpaddb %ymm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512F-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpcmpeqb %ymm1, %ymm2, %ymm1
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
+; AVX512F-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512F-NEXT:    vpcmpeqb %ymm2, %ymm1, %ymm1
+; AVX512F-NEXT:    vpcmpeqb %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: ult_1_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm2
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
-; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
-; AVX512BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vptestnmb %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VPOPCNTDQ-NOBW-LABEL: ult_1_v64i8:
 ; AVX512VPOPCNTDQ-NOBW:       # %bb.0:
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpand %ymm1, %ymm4, %ymm4
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpshufb %ymm4, %ymm3, %ymm4
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpaddb %ymm2, %ymm4, %ymm2
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpand %ymm1, %ymm0, %ymm4
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpshufb %ymm4, %ymm3, %ymm4
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpaddb %ymm4, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpcmpeqb %ymm1, %ymm2, %ymm1
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512VPOPCNTDQ-NOBW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
+; AVX512VPOPCNTDQ-NOBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VPOPCNTDQ-NOBW-NEXT:    vpcmpeqb %ymm2, %ymm1, %ymm1
+; AVX512VPOPCNTDQ-NOBW-NEXT:    vpcmpeqb %ymm2, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NOBW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512VPOPCNTDQ-NOBW-NEXT:    retq
 ;
 ; AVX512VPOPCNTDQ-BW-LABEL: ult_1_v64i8:
 ; AVX512VPOPCNTDQ-BW:       # %bb.0:
-; AVX512VPOPCNTDQ-BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VPOPCNTDQ-BW-NEXT:    vpandq %zmm1, %zmm0, %zmm2
-; AVX512VPOPCNTDQ-BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512VPOPCNTDQ-BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
-; AVX512VPOPCNTDQ-BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
-; AVX512VPOPCNTDQ-BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
-; AVX512VPOPCNTDQ-BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
-; AVX512VPOPCNTDQ-BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
 ; AVX512VPOPCNTDQ-BW-NEXT:    vptestnmb %zmm0, %zmm0, %k0
 ; AVX512VPOPCNTDQ-BW-NEXT:    vpmovm2b %k0, %zmm0
 ; AVX512VPOPCNTDQ-BW-NEXT:    retq
@@ -1597,80 +1509,32 @@
 define <32 x i16> @ugt_0_v32i16(<32 x i16> %0) {
 ; AVX512F-LABEL: ugt_0_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512F-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512F-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512F-NEXT:    vpand %ymm1, %ymm4, %ymm4
-; AVX512F-NEXT:    vpshufb %ymm4, %ymm3, %ymm4
-; AVX512F-NEXT:    vpaddb %ymm2, %ymm4, %ymm2
-; AVX512F-NEXT:    vpsllw $8, %ymm2, %ymm4
-; AVX512F-NEXT:    vpaddb %ymm2, %ymm4, %ymm2
-; AVX512F-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512F-NEXT:    vpand %ymm1, %ymm0, %ymm4
-; AVX512F-NEXT:    vpshufb %ymm4, %ymm3, %ymm4
-; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX512F-NEXT:    vpaddb %ymm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsllw $8, %ymm0, %ymm1
-; AVX512F-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
-; AVX512F-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; AVX512F-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512F-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpcmpeqw %ymm1, %ymm2, %ymm1
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
+; AVX512F-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512F-NEXT:    vpcmpeqw %ymm2, %ymm1, %ymm1
+; AVX512F-NEXT:    vpcmpeqw %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: ugt_0_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm2
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
-; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
-; AVX512BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpsllw $8, %zmm0, %zmm1
-; AVX512BW-NEXT:    vpaddb %zmm0, %zmm1, %zmm0
-; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vptestmw %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    vpmovm2w %k0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VPOPCNTDQ-NOBW-LABEL: ugt_0_v32i16:
 ; AVX512VPOPCNTDQ-NOBW:       # %bb.0:
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpopcntd %zmm1, %zmm1
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpmovdw %zmm1, %ymm1
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpopcntd %zmm0, %zmm0
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512VPOPCNTDQ-NOBW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512VPOPCNTDQ-NOBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpcmpeqw %ymm2, %ymm0, %ymm0
 ; AVX512VPOPCNTDQ-NOBW-NEXT:    vpcmpeqw %ymm2, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512VPOPCNTDQ-NOBW-NEXT:    vpcmpeqw %ymm2, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NOBW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512VPOPCNTDQ-NOBW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512VPOPCNTDQ-NOBW-NEXT:    retq
 ;
 ; AVX512VPOPCNTDQ-BW-LABEL: ugt_0_v32i16:
 ; AVX512VPOPCNTDQ-BW:       # %bb.0:
-; AVX512VPOPCNTDQ-BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VPOPCNTDQ-BW-NEXT:    vpandq %zmm1, %zmm0, %zmm2
-; AVX512VPOPCNTDQ-BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512VPOPCNTDQ-BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
-; AVX512VPOPCNTDQ-BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
-; AVX512VPOPCNTDQ-BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
-; AVX512VPOPCNTDQ-BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
-; AVX512VPOPCNTDQ-BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
-; AVX512VPOPCNTDQ-BW-NEXT:    vpsllw $8, %zmm0, %zmm1
-; AVX512VPOPCNTDQ-BW-NEXT:    vpaddb %zmm0, %zmm1, %zmm0
-; AVX512VPOPCNTDQ-BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
 ; AVX512VPOPCNTDQ-BW-NEXT:    vptestmw %zmm0, %zmm0, %k0
 ; AVX512VPOPCNTDQ-BW-NEXT:    vpmovm2w %k0, %zmm0
 ; AVX512VPOPCNTDQ-BW-NEXT:    retq
@@ -1690,78 +1554,30 @@
 define <32 x i16> @ult_1_v32i16(<32 x i16> %0) {
 ; AVX512F-LABEL: ult_1_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512F-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512F-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512F-NEXT:    vpand %ymm1, %ymm4, %ymm4
-; AVX512F-NEXT:    vpshufb %ymm4, %ymm3, %ymm4
-; AVX512F-NEXT:    vpaddb %ymm2, %ymm4, %ymm2
-; AVX512F-NEXT:    vpsllw $8, %ymm2, %ymm4
-; AVX512F-NEXT:    vpaddb %ymm2, %ymm4, %ymm2
-; AVX512F-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512F-NEXT:    vpand %ymm1, %ymm0, %ymm4
-; AVX512F-NEXT:    vpshufb %ymm4, %ymm3, %ymm4
-; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX512F-NEXT:    vpaddb %ymm4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsllw $8, %ymm0, %ymm1
-; AVX512F-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
-; AVX512F-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; AVX512F-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512F-NEXT:    vpcmpeqw %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpcmpeqw %ymm1, %ymm2, %ymm1
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
+; AVX512F-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512F-NEXT:    vpcmpeqw %ymm2, %ymm1, %ymm1
+; AVX512F-NEXT:    vpcmpeqw %ymm2, %ymm0, %ymm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: ult_1_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm2
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
-; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
-; AVX512BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpsllw $8, %zmm0, %zmm1
-; AVX512BW-NEXT:    vpaddb %zmm0, %zmm1, %zmm0
-; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vptestnmw %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    vpmovm2w %k0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VPOPCNTDQ-NOBW-LABEL: ult_1_v32i16:
 ; AVX512VPOPCNTDQ-NOBW:       # %bb.0:
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpopcntd %zmm1, %zmm1
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpmovdw %zmm1, %ymm1
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpopcntd %zmm0, %zmm0
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512VPOPCNTDQ-NOBW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512VPOPCNTDQ-NOBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vpcmpeqw %ymm2, %ymm0, %ymm0
 ; AVX512VPOPCNTDQ-NOBW-NEXT:    vpcmpeqw %ymm2, %ymm1, %ymm1
-; AVX512VPOPCNTDQ-NOBW-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512VPOPCNTDQ-NOBW-NEXT:    vpcmpeqw %ymm2, %ymm0, %ymm0
+; AVX512VPOPCNTDQ-NOBW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512VPOPCNTDQ-NOBW-NEXT:    retq
 ;
 ; AVX512VPOPCNTDQ-BW-LABEL: ult_1_v32i16:
 ; AVX512VPOPCNTDQ-BW:       # %bb.0:
-; AVX512VPOPCNTDQ-BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VPOPCNTDQ-BW-NEXT:    vpandq %zmm1, %zmm0, %zmm2
-; AVX512VPOPCNTDQ-BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512VPOPCNTDQ-BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
-; AVX512VPOPCNTDQ-BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
-; AVX512VPOPCNTDQ-BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
-; AVX512VPOPCNTDQ-BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
-; AVX512VPOPCNTDQ-BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
-; AVX512VPOPCNTDQ-BW-NEXT:    vpsllw $8, %zmm0, %zmm1
-; AVX512VPOPCNTDQ-BW-NEXT:    vpaddb %zmm0, %zmm1, %zmm0
-; AVX512VPOPCNTDQ-BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
 ; AVX512VPOPCNTDQ-BW-NEXT:    vptestnmw %zmm0, %zmm0, %k0
 ; AVX512VPOPCNTDQ-BW-NEXT:    vpmovm2w %k0, %zmm0
 ; AVX512VPOPCNTDQ-BW-NEXT:    retq
@@ -4638,53 +4454,12 @@
 define <16 x i32> @ugt_0_v16i32(<16 x i32> %0) {
 ; AVX512F-LABEL: ugt_0_v16i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512F-NEXT:    vpand %ymm2, %ymm1, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512F-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
-; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpand %ymm2, %ymm1, %ymm1
-; AVX512F-NEXT:    vpshufb %ymm1, %ymm4, %ymm1
-; AVX512F-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpunpckhdq {{.*#+}} ymm5 = ymm1[2],ymm3[2],ymm1[3],ymm3[3],ymm1[6],ymm3[6],ymm1[7],ymm3[7]
-; AVX512F-NEXT:    vpsadbw %ymm3, %ymm5, %ymm5
-; AVX512F-NEXT:    vpunpckldq {{.*#+}} ymm1 = ymm1[0],ymm3[0],ymm1[1],ymm3[1],ymm1[4],ymm3[4],ymm1[5],ymm3[5]
-; AVX512F-NEXT:    vpsadbw %ymm3, %ymm1, %ymm1
-; AVX512F-NEXT:    vpackuswb %ymm5, %ymm1, %ymm1
-; AVX512F-NEXT:    vpand %ymm2, %ymm0, %ymm5
-; AVX512F-NEXT:    vpshufb %ymm5, %ymm4, %ymm5
-; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpand %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
-; AVX512F-NEXT:    vpaddb %ymm5, %ymm0, %ymm0
-; AVX512F-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm3[2],ymm0[3],ymm3[3],ymm0[6],ymm3[6],ymm0[7],ymm3[7]
-; AVX512F-NEXT:    vpsadbw %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm3[0],ymm0[1],ymm3[1],ymm0[4],ymm3[4],ymm0[5],ymm3[5]
-; AVX512F-NEXT:    vpsadbw %ymm3, %ymm0, %ymm0
-; AVX512F-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k1
 ; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: ugt_0_v16i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm2
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
-; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
-; AVX512BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpunpckhdq {{.*#+}} zmm2 = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
-; AVX512BW-NEXT:    vpsadbw %zmm1, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpunpckldq {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
-; AVX512BW-NEXT:    vpsadbw %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpackuswb %zmm2, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k1
 ; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; AVX512BW-NEXT:    retq
@@ -4698,13 +4473,6 @@
 ;
 ; BITALG-LABEL: ugt_0_v16i32:
 ; BITALG:       # %bb.0:
-; BITALG-NEXT:    vpopcntb %zmm0, %zmm0
-; BITALG-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT:    vpunpckhdq {{.*#+}} zmm2 = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
-; BITALG-NEXT:    vpsadbw %zmm1, %zmm2, %zmm2
-; BITALG-NEXT:    vpunpckldq {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
-; BITALG-NEXT:    vpsadbw %zmm1, %zmm0, %zmm0
-; BITALG-NEXT:    vpackuswb %zmm2, %zmm0, %zmm0
 ; BITALG-NEXT:    vptestmd %zmm0, %zmm0, %k1
 ; BITALG-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; BITALG-NEXT:    retq
@@ -4717,53 +4485,12 @@
 define <16 x i32> @ult_1_v16i32(<16 x i32> %0) {
 ; AVX512F-LABEL: ult_1_v16i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512F-NEXT:    vpand %ymm2, %ymm1, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512F-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
-; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpand %ymm2, %ymm1, %ymm1
-; AVX512F-NEXT:    vpshufb %ymm1, %ymm4, %ymm1
-; AVX512F-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpunpckhdq {{.*#+}} ymm5 = ymm1[2],ymm3[2],ymm1[3],ymm3[3],ymm1[6],ymm3[6],ymm1[7],ymm3[7]
-; AVX512F-NEXT:    vpsadbw %ymm3, %ymm5, %ymm5
-; AVX512F-NEXT:    vpunpckldq {{.*#+}} ymm1 = ymm1[0],ymm3[0],ymm1[1],ymm3[1],ymm1[4],ymm3[4],ymm1[5],ymm3[5]
-; AVX512F-NEXT:    vpsadbw %ymm3, %ymm1, %ymm1
-; AVX512F-NEXT:    vpackuswb %ymm5, %ymm1, %ymm1
-; AVX512F-NEXT:    vpand %ymm2, %ymm0, %ymm5
-; AVX512F-NEXT:    vpshufb %ymm5, %ymm4, %ymm5
-; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpand %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
-; AVX512F-NEXT:    vpaddb %ymm5, %ymm0, %ymm0
-; AVX512F-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm3[2],ymm0[3],ymm3[3],ymm0[6],ymm3[6],ymm0[7],ymm3[7]
-; AVX512F-NEXT:    vpsadbw %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm3[0],ymm0[1],ymm3[1],ymm0[4],ymm3[4],ymm0[5],ymm3[5]
-; AVX512F-NEXT:    vpsadbw %ymm3, %ymm0, %ymm0
-; AVX512F-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    vptestnmd %zmm0, %zmm0, %k1
 ; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: ult_1_v16i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm2
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
-; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
-; AVX512BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpunpckhdq {{.*#+}} zmm2 = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
-; AVX512BW-NEXT:    vpsadbw %zmm1, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpunpckldq {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
-; AVX512BW-NEXT:    vpsadbw %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpackuswb %zmm2, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vptestnmd %zmm0, %zmm0, %k1
 ; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; AVX512BW-NEXT:    retq
@@ -4777,13 +4504,6 @@
 ;
 ; BITALG-LABEL: ult_1_v16i32:
 ; BITALG:       # %bb.0:
-; BITALG-NEXT:    vpopcntb %zmm0, %zmm0
-; BITALG-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT:    vpunpckhdq {{.*#+}} zmm2 = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
-; BITALG-NEXT:    vpsadbw %zmm1, %zmm2, %zmm2
-; BITALG-NEXT:    vpunpckldq {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
-; BITALG-NEXT:    vpsadbw %zmm1, %zmm0, %zmm0
-; BITALG-NEXT:    vpackuswb %zmm2, %zmm0, %zmm0
 ; BITALG-NEXT:    vptestnmd %zmm0, %zmm0, %k1
 ; BITALG-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; BITALG-NEXT:    retq
@@ -9779,41 +9499,12 @@
 define <8 x i64> @ugt_0_v8i64(<8 x i64> %0) {
 ; AVX512F-LABEL: ugt_0_v8i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512F-NEXT:    vpand %ymm2, %ymm1, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512F-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
-; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpand %ymm2, %ymm1, %ymm1
-; AVX512F-NEXT:    vpshufb %ymm1, %ymm4, %ymm1
-; AVX512F-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpsadbw %ymm3, %ymm1, %ymm1
-; AVX512F-NEXT:    vpand %ymm2, %ymm0, %ymm5
-; AVX512F-NEXT:    vpshufb %ymm5, %ymm4, %ymm5
-; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpand %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
-; AVX512F-NEXT:    vpaddb %ymm5, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsadbw %ymm3, %ymm0, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    vptestmq %zmm0, %zmm0, %k1
 ; AVX512F-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: ugt_0_v8i64:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm2
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
-; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
-; AVX512BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpsadbw %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1
 ; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; AVX512BW-NEXT:    retq
@@ -9827,9 +9518,6 @@
 ;
 ; BITALG-LABEL: ugt_0_v8i64:
 ; BITALG:       # %bb.0:
-; BITALG-NEXT:    vpopcntb %zmm0, %zmm0
-; BITALG-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT:    vpsadbw %zmm1, %zmm0, %zmm0
 ; BITALG-NEXT:    vptestmq %zmm0, %zmm0, %k1
 ; BITALG-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; BITALG-NEXT:    retq
@@ -9842,41 +9530,12 @@
 define <8 x i64> @ult_1_v8i64(<8 x i64> %0) {
 ; AVX512F-LABEL: ult_1_v8i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512F-NEXT:    vpand %ymm2, %ymm1, %ymm3
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512F-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
-; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpand %ymm2, %ymm1, %ymm1
-; AVX512F-NEXT:    vpshufb %ymm1, %ymm4, %ymm1
-; AVX512F-NEXT:    vpaddb %ymm3, %ymm1, %ymm1
-; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpsadbw %ymm3, %ymm1, %ymm1
-; AVX512F-NEXT:    vpand %ymm2, %ymm0, %ymm5
-; AVX512F-NEXT:    vpshufb %ymm5, %ymm4, %ymm5
-; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512F-NEXT:    vpand %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
-; AVX512F-NEXT:    vpaddb %ymm5, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsadbw %ymm3, %ymm0, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512F-NEXT:    vptestnmq %zmm0, %zmm0, %k1
 ; AVX512F-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: ult_1_v8i64:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm2
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
-; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
-; AVX512BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpsadbw %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vptestnmq %zmm0, %zmm0, %k1
 ; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; AVX512BW-NEXT:    retq
@@ -9890,9 +9549,6 @@
 ;
 ; BITALG-LABEL: ult_1_v8i64:
 ; BITALG:       # %bb.0:
-; BITALG-NEXT:    vpopcntb %zmm0, %zmm0
-; BITALG-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; BITALG-NEXT:    vpsadbw %zmm1, %zmm0, %zmm0
 ; BITALG-NEXT:    vptestnmq %zmm0, %zmm0, %k1
 ; BITALG-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; BITALG-NEXT:    retq