From: Craig Topper <craig.topper@intel.com>
Date: Mon, 27 Nov 2017 22:56:10 +0000 (+0000)
Subject: [X86] Teach getSetCCResultType to handle more than just SimpleVTs when looking at... 
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=9ee906e259cfb2883512c8b1ee8db75970e86227;p=llvm

[X86] Teach getSetCCResultType to handle more than just SimpleVTs when looking at larger than 512-bit vectors.

Which VTs are considered simple is determined by the superset of the legal types of all targets in LLVM. If we're looking at VTs that are going to be split down to 512-bits we should allow any VT not just simple ones since the simple list changes over time as new targets are added.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@319110 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 4580466f090..3dd265ff0c2 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -1716,25 +1716,22 @@ EVT X86TargetLowering::getSetCCResultType(const DataLayout &DL,
   if (!VT.isVector())
     return MVT::i8;
 
+  if (VT.getSizeInBits() >= 512) {
+    EVT EltVT = VT.getVectorElementType();
+    const unsigned NumElts = VT.getVectorNumElements();
+    if (Subtarget.hasAVX512())
+      if (EltVT == MVT::i32 || EltVT == MVT::i64 ||
+          EltVT == MVT::f32 || EltVT == MVT::f64)
+        return EVT::getVectorVT(Context, MVT::i1, NumElts);
+    if (Subtarget.hasBWI())
+      if (EltVT == MVT::i8 || EltVT == MVT::i16)
+        return EVT::getVectorVT(Context, MVT::i1, NumElts);
+  }
+
   if (VT.isSimple()) {
     MVT VVT = VT.getSimpleVT();
     const unsigned NumElts = VVT.getVectorNumElements();
     MVT EltVT = VVT.getVectorElementType();
-    if (VVT.getSizeInBits() >= 512) {
-      if (Subtarget.hasAVX512())
-        if (EltVT == MVT::i32 || EltVT == MVT::i64 ||
-            EltVT == MVT::f32 || EltVT == MVT::f64)
-          switch(NumElts) {
-          case  8: return MVT::v8i1;
-          case 16: return MVT::v16i1;
-        }
-      if (Subtarget.hasBWI())
-        if (EltVT == MVT::i8 || EltVT == MVT::i16)
-          switch(NumElts) {
-          case 32: return MVT::v32i1;
-          case 64: return MVT::v64i1;
-        }
-    }
 
     if (Subtarget.hasBWI() && Subtarget.hasVLX())
       return MVT::getVectorVT(MVT::i1, NumElts);
diff --git a/test/CodeGen/X86/vector-compare-results.ll b/test/CodeGen/X86/vector-compare-results.ll
index f8541c6b01c..b0d4127258c 100644
--- a/test/CodeGen/X86/vector-compare-results.ll
+++ b/test/CodeGen/X86/vector-compare-results.ll
@@ -2068,121 +2068,35 @@ define <16 x i1> @test_cmp_v16f64(<16 x double> %a0, <16 x double> %a1) nounwind
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
-; AVX512-LABEL: test_cmp_v16f64:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vextractf32x4 $3, %zmm2, %xmm4
-; AVX512-NEXT:    vextractf32x4 $3, %zmm0, %xmm5
-; AVX512-NEXT:    xorl %eax, %eax
-; AVX512-NEXT:    vucomisd %xmm4, %xmm5
-; AVX512-NEXT:    movq $-1, %rcx
-; AVX512-NEXT:    movl $0, %edx
-; AVX512-NEXT:    cmovaq %rcx, %rdx
-; AVX512-NEXT:    vmovq %rdx, %xmm6
-; AVX512-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm4[1,0]
-; AVX512-NEXT:    vpermilpd {{.*#+}} xmm5 = xmm5[1,0]
-; AVX512-NEXT:    vucomisd %xmm4, %xmm5
-; AVX512-NEXT:    movl $0, %edx
-; AVX512-NEXT:    cmovaq %rcx, %rdx
-; AVX512-NEXT:    vmovq %rdx, %xmm4
-; AVX512-NEXT:    vpunpcklqdq {{.*#+}} xmm4 = xmm6[0],xmm4[0]
-; AVX512-NEXT:    vextractf32x4 $2, %zmm2, %xmm5
-; AVX512-NEXT:    vextractf32x4 $2, %zmm0, %xmm6
-; AVX512-NEXT:    vucomisd %xmm5, %xmm6
-; AVX512-NEXT:    movl $0, %edx
-; AVX512-NEXT:    cmovaq %rcx, %rdx
-; AVX512-NEXT:    vmovq %rdx, %xmm7
-; AVX512-NEXT:    vpermilpd {{.*#+}} xmm5 = xmm5[1,0]
-; AVX512-NEXT:    vpermilpd {{.*#+}} xmm6 = xmm6[1,0]
-; AVX512-NEXT:    vucomisd %xmm5, %xmm6
-; AVX512-NEXT:    movl $0, %edx
-; AVX512-NEXT:    cmovaq %rcx, %rdx
-; AVX512-NEXT:    vmovq %rdx, %xmm5
-; AVX512-NEXT:    vpunpcklqdq {{.*#+}} xmm5 = xmm7[0],xmm5[0]
-; AVX512-NEXT:    vinserti128 $1, %xmm4, %ymm5, %ymm4
-; AVX512-NEXT:    vextractf128 $1, %ymm2, %xmm5
-; AVX512-NEXT:    vextractf128 $1, %ymm0, %xmm6
-; AVX512-NEXT:    vucomisd %xmm5, %xmm6
-; AVX512-NEXT:    movl $0, %edx
-; AVX512-NEXT:    cmovaq %rcx, %rdx
-; AVX512-NEXT:    vmovq %rdx, %xmm7
-; AVX512-NEXT:    vpermilpd {{.*#+}} xmm5 = xmm5[1,0]
-; AVX512-NEXT:    vpermilpd {{.*#+}} xmm6 = xmm6[1,0]
-; AVX512-NEXT:    vucomisd %xmm5, %xmm6
-; AVX512-NEXT:    movl $0, %edx
-; AVX512-NEXT:    cmovaq %rcx, %rdx
-; AVX512-NEXT:    vmovq %rdx, %xmm5
-; AVX512-NEXT:    vpunpcklqdq {{.*#+}} xmm5 = xmm7[0],xmm5[0]
-; AVX512-NEXT:    vucomisd %xmm2, %xmm0
-; AVX512-NEXT:    movl $0, %edx
-; AVX512-NEXT:    cmovaq %rcx, %rdx
-; AVX512-NEXT:    vmovq %rdx, %xmm6
-; AVX512-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512-NEXT:    vucomisd %xmm2, %xmm0
-; AVX512-NEXT:    movl $0, %edx
-; AVX512-NEXT:    cmovaq %rcx, %rdx
-; AVX512-NEXT:    vmovq %rdx, %xmm0
-; AVX512-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm6[0],xmm0[0]
-; AVX512-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm0
-; AVX512-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
-; AVX512-NEXT:    vpmovqd %zmm0, %ymm0
-; AVX512-NEXT:    vextractf32x4 $3, %zmm3, %xmm2
-; AVX512-NEXT:    vextractf32x4 $3, %zmm1, %xmm4
-; AVX512-NEXT:    vucomisd %xmm2, %xmm4
-; AVX512-NEXT:    movl $0, %edx
-; AVX512-NEXT:    cmovaq %rcx, %rdx
-; AVX512-NEXT:    vmovq %rdx, %xmm5
-; AVX512-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm4[1,0]
-; AVX512-NEXT:    vucomisd %xmm2, %xmm4
-; AVX512-NEXT:    movl $0, %edx
-; AVX512-NEXT:    cmovaq %rcx, %rdx
-; AVX512-NEXT:    vmovq %rdx, %xmm2
-; AVX512-NEXT:    vpunpcklqdq {{.*#+}} xmm2 = xmm5[0],xmm2[0]
-; AVX512-NEXT:    vextractf32x4 $2, %zmm3, %xmm4
-; AVX512-NEXT:    vextractf32x4 $2, %zmm1, %xmm5
-; AVX512-NEXT:    vucomisd %xmm4, %xmm5
-; AVX512-NEXT:    movl $0, %edx
-; AVX512-NEXT:    cmovaq %rcx, %rdx
-; AVX512-NEXT:    vmovq %rdx, %xmm6
-; AVX512-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm4[1,0]
-; AVX512-NEXT:    vpermilpd {{.*#+}} xmm5 = xmm5[1,0]
-; AVX512-NEXT:    vucomisd %xmm4, %xmm5
-; AVX512-NEXT:    movl $0, %edx
-; AVX512-NEXT:    cmovaq %rcx, %rdx
-; AVX512-NEXT:    vmovq %rdx, %xmm4
-; AVX512-NEXT:    vpunpcklqdq {{.*#+}} xmm4 = xmm6[0],xmm4[0]
-; AVX512-NEXT:    vinserti128 $1, %xmm2, %ymm4, %ymm2
-; AVX512-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX512-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX512-NEXT:    vucomisd %xmm4, %xmm5
-; AVX512-NEXT:    movl $0, %edx
-; AVX512-NEXT:    cmovaq %rcx, %rdx
-; AVX512-NEXT:    vmovq %rdx, %xmm6
-; AVX512-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm4[1,0]
-; AVX512-NEXT:    vpermilpd {{.*#+}} xmm5 = xmm5[1,0]
-; AVX512-NEXT:    vucomisd %xmm4, %xmm5
-; AVX512-NEXT:    movl $0, %edx
-; AVX512-NEXT:    cmovaq %rcx, %rdx
-; AVX512-NEXT:    vmovq %rdx, %xmm4
-; AVX512-NEXT:    vpunpcklqdq {{.*#+}} xmm4 = xmm6[0],xmm4[0]
-; AVX512-NEXT:    vucomisd %xmm3, %xmm1
-; AVX512-NEXT:    movl $0, %edx
-; AVX512-NEXT:    cmovaq %rcx, %rdx
-; AVX512-NEXT:    vmovq %rdx, %xmm5
-; AVX512-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm3[1,0]
-; AVX512-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm1[1,0]
-; AVX512-NEXT:    vucomisd %xmm3, %xmm1
-; AVX512-NEXT:    cmovaq %rcx, %rax
-; AVX512-NEXT:    vmovq %rax, %xmm1
-; AVX512-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm5[0],xmm1[0]
-; AVX512-NEXT:    vinserti128 $1, %xmm4, %ymm1, %ymm1
-; AVX512-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
-; AVX512-NEXT:    vpmovqd %zmm1, %ymm1
-; AVX512-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
-; AVX512-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512-NEXT:    vzeroupper
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: test_cmp_v16f64:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vcmpltpd %zmm0, %zmm2, %k0
+; AVX512F-NEXT:    vcmpltpd %zmm1, %zmm3, %k1
+; AVX512F-NEXT:    kunpckbw %k0, %k1, %k1
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512DQ-LABEL: test_cmp_v16f64:
+; AVX512DQ:       # BB#0:
+; AVX512DQ-NEXT:    vcmpltpd %zmm0, %zmm2, %k0
+; AVX512DQ-NEXT:    vcmpltpd %zmm1, %zmm3, %k1
+; AVX512DQ-NEXT:    kunpckbw %k0, %k1, %k0
+; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
+; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512DQ-NEXT:    vzeroupper
+; AVX512DQ-NEXT:    retq
+;
+; AVX512BW-LABEL: test_cmp_v16f64:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vcmpltpd %zmm0, %zmm2, %k0
+; AVX512BW-NEXT:    vcmpltpd %zmm1, %zmm3, %k1
+; AVX512BW-NEXT:    kunpckbw %k0, %k1, %k0
+; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0
+; AVX512BW-NEXT:    # kill: %XMM0<def> %XMM0<kill> %ZMM0<kill>
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
   %1 = fcmp ogt <16 x double> %a0, %a1
   ret <16 x i1> %1
 }
@@ -2869,205 +2783,11 @@ define <32 x i1> @test_cmp_v32f32(<32 x float> %a0, <32 x float> %a1) nounwind {
 ;
 ; AVX512BW-LABEL: test_cmp_v32f32:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vextractf32x4 $3, %zmm2, %xmm4
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm5 = xmm4[1,1,3,3]
-; AVX512BW-NEXT:    vextractf32x4 $3, %zmm0, %xmm6
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm7 = xmm6[1,1,3,3]
-; AVX512BW-NEXT:    xorl %eax, %eax
-; AVX512BW-NEXT:    vucomiss %xmm5, %xmm7
-; AVX512BW-NEXT:    movl $-1, %ecx
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vucomiss %xmm4, %xmm6
-; AVX512BW-NEXT:    movl $0, %esi
-; AVX512BW-NEXT:    cmoval %ecx, %esi
-; AVX512BW-NEXT:    vmovd %esi, %xmm5
-; AVX512BW-NEXT:    vpinsrd $1, %edx, %xmm5, %xmm8
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm7 = xmm4[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm5 = xmm6[1,0]
-; AVX512BW-NEXT:    vucomiss %xmm7, %xmm5
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $2, %edx, %xmm8, %xmm5
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm4 = xmm4[3,1,2,3]
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm6 = xmm6[3,1,2,3]
-; AVX512BW-NEXT:    vucomiss %xmm4, %xmm6
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $3, %edx, %xmm5, %xmm8
-; AVX512BW-NEXT:    vextractf32x4 $2, %zmm2, %xmm5
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm6 = xmm5[1,1,3,3]
-; AVX512BW-NEXT:    vextractf32x4 $2, %zmm0, %xmm7
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm4 = xmm7[1,1,3,3]
-; AVX512BW-NEXT:    vucomiss %xmm6, %xmm4
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vucomiss %xmm5, %xmm7
-; AVX512BW-NEXT:    movl $0, %esi
-; AVX512BW-NEXT:    cmoval %ecx, %esi
-; AVX512BW-NEXT:    vmovd %esi, %xmm4
-; AVX512BW-NEXT:    vpinsrd $1, %edx, %xmm4, %xmm9
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm6 = xmm5[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm7[1,0]
-; AVX512BW-NEXT:    vucomiss %xmm6, %xmm4
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $2, %edx, %xmm9, %xmm4
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm5 = xmm5[3,1,2,3]
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm6 = xmm7[3,1,2,3]
-; AVX512BW-NEXT:    vucomiss %xmm5, %xmm6
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $3, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vinserti128 $1, %xmm8, %ymm4, %ymm8
-; AVX512BW-NEXT:    vextractf128 $1, %ymm2, %xmm5
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm6 = xmm5[1,1,3,3]
-; AVX512BW-NEXT:    vextractf128 $1, %ymm0, %xmm7
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm4 = xmm7[1,1,3,3]
-; AVX512BW-NEXT:    vucomiss %xmm6, %xmm4
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vucomiss %xmm5, %xmm7
-; AVX512BW-NEXT:    movl $0, %esi
-; AVX512BW-NEXT:    cmoval %ecx, %esi
-; AVX512BW-NEXT:    vmovd %esi, %xmm4
-; AVX512BW-NEXT:    vpinsrd $1, %edx, %xmm4, %xmm9
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm6 = xmm5[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm7[1,0]
-; AVX512BW-NEXT:    vucomiss %xmm6, %xmm4
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $2, %edx, %xmm9, %xmm4
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm5 = xmm5[3,1,2,3]
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm6 = xmm7[3,1,2,3]
-; AVX512BW-NEXT:    vucomiss %xmm5, %xmm6
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $3, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm5 = xmm2[1,1,3,3]
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm6 = xmm0[1,1,3,3]
-; AVX512BW-NEXT:    vucomiss %xmm5, %xmm6
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vucomiss %xmm2, %xmm0
-; AVX512BW-NEXT:    movl $0, %esi
-; AVX512BW-NEXT:    cmoval %ecx, %esi
-; AVX512BW-NEXT:    vmovd %esi, %xmm5
-; AVX512BW-NEXT:    vpinsrd $1, %edx, %xmm5, %xmm5
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm6 = xmm2[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm7 = xmm0[1,0]
-; AVX512BW-NEXT:    vucomiss %xmm6, %xmm7
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $2, %edx, %xmm5, %xmm5
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512BW-NEXT:    vucomiss %xmm2, %xmm0
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $3, %edx, %xmm5, %xmm0
-; AVX512BW-NEXT:    vinserti128 $1, %xmm4, %ymm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm8, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpmovdw %zmm0, %ymm8
-; AVX512BW-NEXT:    vextractf32x4 $3, %zmm3, %xmm2
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm4 = xmm2[1,1,3,3]
-; AVX512BW-NEXT:    vextractf32x4 $3, %zmm1, %xmm5
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm6 = xmm5[1,1,3,3]
-; AVX512BW-NEXT:    vucomiss %xmm4, %xmm6
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vucomiss %xmm2, %xmm5
-; AVX512BW-NEXT:    movl $0, %esi
-; AVX512BW-NEXT:    cmoval %ecx, %esi
-; AVX512BW-NEXT:    vmovd %esi, %xmm4
-; AVX512BW-NEXT:    vpinsrd $1, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm6 = xmm2[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm7 = xmm5[1,0]
-; AVX512BW-NEXT:    vucomiss %xmm6, %xmm7
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $2, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm5 = xmm5[3,1,2,3]
-; AVX512BW-NEXT:    vucomiss %xmm2, %xmm5
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $3, %edx, %xmm4, %xmm2
-; AVX512BW-NEXT:    vextractf32x4 $2, %zmm3, %xmm4
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm5 = xmm4[1,1,3,3]
-; AVX512BW-NEXT:    vextractf32x4 $2, %zmm1, %xmm6
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm7 = xmm6[1,1,3,3]
-; AVX512BW-NEXT:    vucomiss %xmm5, %xmm7
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vucomiss %xmm4, %xmm6
-; AVX512BW-NEXT:    movl $0, %esi
-; AVX512BW-NEXT:    cmoval %ecx, %esi
-; AVX512BW-NEXT:    vmovd %esi, %xmm5
-; AVX512BW-NEXT:    vpinsrd $1, %edx, %xmm5, %xmm5
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm7 = xmm4[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm6[1,0]
-; AVX512BW-NEXT:    vucomiss %xmm7, %xmm0
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $2, %edx, %xmm5, %xmm0
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm4 = xmm4[3,1,2,3]
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm5 = xmm6[3,1,2,3]
-; AVX512BW-NEXT:    vucomiss %xmm4, %xmm5
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $3, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX512BW-NEXT:    vextractf128 $1, %ymm3, %xmm0
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm4 = xmm0[1,1,3,3]
-; AVX512BW-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm6 = xmm5[1,1,3,3]
-; AVX512BW-NEXT:    vucomiss %xmm4, %xmm6
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vucomiss %xmm0, %xmm5
-; AVX512BW-NEXT:    movl $0, %esi
-; AVX512BW-NEXT:    cmoval %ecx, %esi
-; AVX512BW-NEXT:    vmovd %esi, %xmm4
-; AVX512BW-NEXT:    vpinsrd $1, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm6 = xmm0[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm7 = xmm5[1,0]
-; AVX512BW-NEXT:    vucomiss %xmm6, %xmm7
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $2, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm5 = xmm5[3,1,2,3]
-; AVX512BW-NEXT:    vucomiss %xmm0, %xmm5
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $3, %edx, %xmm4, %xmm0
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm4 = xmm3[1,1,3,3]
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm5 = xmm1[1,1,3,3]
-; AVX512BW-NEXT:    vucomiss %xmm4, %xmm5
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vucomiss %xmm3, %xmm1
-; AVX512BW-NEXT:    movl $0, %esi
-; AVX512BW-NEXT:    cmoval %ecx, %esi
-; AVX512BW-NEXT:    vmovd %esi, %xmm4
-; AVX512BW-NEXT:    vpinsrd $1, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm5 = xmm3[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm6 = xmm1[1,0]
-; AVX512BW-NEXT:    vucomiss %xmm5, %xmm6
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmoval %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $2, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm3 = xmm3[3,1,2,3]
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX512BW-NEXT:    vucomiss %xmm3, %xmm1
-; AVX512BW-NEXT:    cmoval %ecx, %eax
-; AVX512BW-NEXT:    vpinsrd $3, %eax, %xmm4, %xmm1
-; AVX512BW-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpmovdw %zmm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm8, %zmm0
-; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512BW-NEXT:    vcmpltps %zmm0, %zmm2, %k0
+; AVX512BW-NEXT:    vcmpltps %zmm1, %zmm3, %k1
+; AVX512BW-NEXT:    kunpckwd %k0, %k1, %k0
+; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0
+; AVX512BW-NEXT:    # kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
 ; AVX512BW-NEXT:    retq
   %1 = fcmp ogt <32 x float> %a0, %a1
   ret <32 x i1> %1
@@ -3976,221 +3696,11 @@ define <32 x i1> @test_cmp_v32i32(<32 x i32> %a0, <32 x i32> %a1) nounwind {
 ;
 ; AVX512BW-LABEL: test_cmp_v32i32:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm2, %xmm4
-; AVX512BW-NEXT:    vpextrd $1, %xmm4, %ecx
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm0, %xmm5
-; AVX512BW-NEXT:    vpextrd $1, %xmm5, %edx
-; AVX512BW-NEXT:    xorl %eax, %eax
-; AVX512BW-NEXT:    cmpl %ecx, %edx
-; AVX512BW-NEXT:    movl $-1, %ecx
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vmovd %xmm4, %esi
-; AVX512BW-NEXT:    vmovd %xmm5, %edi
-; AVX512BW-NEXT:    cmpl %esi, %edi
-; AVX512BW-NEXT:    movl $0, %esi
-; AVX512BW-NEXT:    cmovgl %ecx, %esi
-; AVX512BW-NEXT:    vmovd %esi, %xmm6
-; AVX512BW-NEXT:    vpinsrd $1, %edx, %xmm6, %xmm6
-; AVX512BW-NEXT:    vpextrd $2, %xmm4, %edx
-; AVX512BW-NEXT:    vpextrd $2, %xmm5, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $2, %edx, %xmm6, %xmm6
-; AVX512BW-NEXT:    vpextrd $3, %xmm4, %edx
-; AVX512BW-NEXT:    vpextrd $3, %xmm5, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $3, %edx, %xmm6, %xmm4
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm2, %xmm5
-; AVX512BW-NEXT:    vpextrd $1, %xmm5, %edx
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm0, %xmm6
-; AVX512BW-NEXT:    vpextrd $1, %xmm6, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vmovd %xmm5, %esi
-; AVX512BW-NEXT:    vmovd %xmm6, %edi
-; AVX512BW-NEXT:    cmpl %esi, %edi
-; AVX512BW-NEXT:    movl $0, %esi
-; AVX512BW-NEXT:    cmovgl %ecx, %esi
-; AVX512BW-NEXT:    vmovd %esi, %xmm7
-; AVX512BW-NEXT:    vpinsrd $1, %edx, %xmm7, %xmm7
-; AVX512BW-NEXT:    vpextrd $2, %xmm5, %edx
-; AVX512BW-NEXT:    vpextrd $2, %xmm6, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $2, %edx, %xmm7, %xmm7
-; AVX512BW-NEXT:    vpextrd $3, %xmm5, %edx
-; AVX512BW-NEXT:    vpextrd $3, %xmm6, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $3, %edx, %xmm7, %xmm5
-; AVX512BW-NEXT:    vinserti128 $1, %xmm4, %ymm5, %ymm4
-; AVX512BW-NEXT:    vextracti128 $1, %ymm2, %xmm5
-; AVX512BW-NEXT:    vpextrd $1, %xmm5, %edx
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm6
-; AVX512BW-NEXT:    vpextrd $1, %xmm6, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vmovd %xmm5, %esi
-; AVX512BW-NEXT:    vmovd %xmm6, %edi
-; AVX512BW-NEXT:    cmpl %esi, %edi
-; AVX512BW-NEXT:    movl $0, %esi
-; AVX512BW-NEXT:    cmovgl %ecx, %esi
-; AVX512BW-NEXT:    vmovd %esi, %xmm7
-; AVX512BW-NEXT:    vpinsrd $1, %edx, %xmm7, %xmm7
-; AVX512BW-NEXT:    vpextrd $2, %xmm5, %edx
-; AVX512BW-NEXT:    vpextrd $2, %xmm6, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $2, %edx, %xmm7, %xmm7
-; AVX512BW-NEXT:    vpextrd $3, %xmm5, %edx
-; AVX512BW-NEXT:    vpextrd $3, %xmm6, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $3, %edx, %xmm7, %xmm5
-; AVX512BW-NEXT:    vpextrd $1, %xmm2, %edx
-; AVX512BW-NEXT:    vpextrd $1, %xmm0, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vmovd %xmm2, %esi
-; AVX512BW-NEXT:    vmovd %xmm0, %edi
-; AVX512BW-NEXT:    cmpl %esi, %edi
-; AVX512BW-NEXT:    movl $0, %esi
-; AVX512BW-NEXT:    cmovgl %ecx, %esi
-; AVX512BW-NEXT:    vmovd %esi, %xmm6
-; AVX512BW-NEXT:    vpinsrd $1, %edx, %xmm6, %xmm6
-; AVX512BW-NEXT:    vpextrd $2, %xmm2, %edx
-; AVX512BW-NEXT:    vpextrd $2, %xmm0, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $2, %edx, %xmm6, %xmm6
-; AVX512BW-NEXT:    vpextrd $3, %xmm2, %edx
-; AVX512BW-NEXT:    vpextrd $3, %xmm0, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $3, %edx, %xmm6, %xmm0
-; AVX512BW-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpmovdw %zmm0, %ymm0
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm3, %xmm2
-; AVX512BW-NEXT:    vpextrd $1, %xmm2, %edx
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm1, %xmm4
-; AVX512BW-NEXT:    vpextrd $1, %xmm4, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vmovd %xmm2, %esi
-; AVX512BW-NEXT:    vmovd %xmm4, %edi
-; AVX512BW-NEXT:    cmpl %esi, %edi
-; AVX512BW-NEXT:    movl $0, %esi
-; AVX512BW-NEXT:    cmovgl %ecx, %esi
-; AVX512BW-NEXT:    vmovd %esi, %xmm5
-; AVX512BW-NEXT:    vpinsrd $1, %edx, %xmm5, %xmm5
-; AVX512BW-NEXT:    vpextrd $2, %xmm2, %edx
-; AVX512BW-NEXT:    vpextrd $2, %xmm4, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $2, %edx, %xmm5, %xmm5
-; AVX512BW-NEXT:    vpextrd $3, %xmm2, %edx
-; AVX512BW-NEXT:    vpextrd $3, %xmm4, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $3, %edx, %xmm5, %xmm2
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm3, %xmm4
-; AVX512BW-NEXT:    vpextrd $1, %xmm4, %edx
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm1, %xmm5
-; AVX512BW-NEXT:    vpextrd $1, %xmm5, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vmovd %xmm4, %esi
-; AVX512BW-NEXT:    vmovd %xmm5, %edi
-; AVX512BW-NEXT:    cmpl %esi, %edi
-; AVX512BW-NEXT:    movl $0, %esi
-; AVX512BW-NEXT:    cmovgl %ecx, %esi
-; AVX512BW-NEXT:    vmovd %esi, %xmm6
-; AVX512BW-NEXT:    vpinsrd $1, %edx, %xmm6, %xmm6
-; AVX512BW-NEXT:    vpextrd $2, %xmm4, %edx
-; AVX512BW-NEXT:    vpextrd $2, %xmm5, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $2, %edx, %xmm6, %xmm6
-; AVX512BW-NEXT:    vpextrd $3, %xmm4, %edx
-; AVX512BW-NEXT:    vpextrd $3, %xmm5, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $3, %edx, %xmm6, %xmm4
-; AVX512BW-NEXT:    vinserti128 $1, %xmm2, %ymm4, %ymm2
-; AVX512BW-NEXT:    vextracti128 $1, %ymm3, %xmm4
-; AVX512BW-NEXT:    vpextrd $1, %xmm4, %edx
-; AVX512BW-NEXT:    vextracti128 $1, %ymm1, %xmm5
-; AVX512BW-NEXT:    vpextrd $1, %xmm5, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vmovd %xmm4, %esi
-; AVX512BW-NEXT:    vmovd %xmm5, %edi
-; AVX512BW-NEXT:    cmpl %esi, %edi
-; AVX512BW-NEXT:    movl $0, %esi
-; AVX512BW-NEXT:    cmovgl %ecx, %esi
-; AVX512BW-NEXT:    vmovd %esi, %xmm6
-; AVX512BW-NEXT:    vpinsrd $1, %edx, %xmm6, %xmm6
-; AVX512BW-NEXT:    vpextrd $2, %xmm4, %edx
-; AVX512BW-NEXT:    vpextrd $2, %xmm5, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $2, %edx, %xmm6, %xmm6
-; AVX512BW-NEXT:    vpextrd $3, %xmm4, %edx
-; AVX512BW-NEXT:    vpextrd $3, %xmm5, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $3, %edx, %xmm6, %xmm4
-; AVX512BW-NEXT:    vpextrd $1, %xmm3, %edx
-; AVX512BW-NEXT:    vpextrd $1, %xmm1, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vmovd %xmm3, %esi
-; AVX512BW-NEXT:    vmovd %xmm1, %edi
-; AVX512BW-NEXT:    cmpl %esi, %edi
-; AVX512BW-NEXT:    movl $0, %esi
-; AVX512BW-NEXT:    cmovgl %ecx, %esi
-; AVX512BW-NEXT:    vmovd %esi, %xmm5
-; AVX512BW-NEXT:    vpinsrd $1, %edx, %xmm5, %xmm5
-; AVX512BW-NEXT:    vpextrd $2, %xmm3, %edx
-; AVX512BW-NEXT:    vpextrd $2, %xmm1, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgl %ecx, %edx
-; AVX512BW-NEXT:    vpinsrd $2, %edx, %xmm5, %xmm5
-; AVX512BW-NEXT:    vpextrd $3, %xmm3, %edx
-; AVX512BW-NEXT:    vpextrd $3, %xmm1, %esi
-; AVX512BW-NEXT:    cmpl %edx, %esi
-; AVX512BW-NEXT:    cmovgl %ecx, %eax
-; AVX512BW-NEXT:    vpinsrd $3, %eax, %xmm5, %xmm1
-; AVX512BW-NEXT:    vinserti128 $1, %xmm4, %ymm1, %ymm1
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpmovdw %zmm1, %ymm1
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512BW-NEXT:    vpcmpgtd %zmm2, %zmm0, %k0
+; AVX512BW-NEXT:    vpcmpgtd %zmm3, %zmm1, %k1
+; AVX512BW-NEXT:    kunpckwd %k0, %k1, %k0
+; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0
+; AVX512BW-NEXT:    # kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
 ; AVX512BW-NEXT:    retq
   %1 = icmp sgt <32 x i32> %a0, %a1
   ret <32 x i1> %1
@@ -8228,229 +7738,15 @@ define <32 x i1> @test_cmp_v32f64(<32 x double> %a0, <32 x double> %a1) nounwind
 ;
 ; AVX512BW-LABEL: test_cmp_v32f64:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vextractf32x4 $3, %zmm4, %xmm8
-; AVX512BW-NEXT:    vextractf32x4 $3, %zmm0, %xmm9
-; AVX512BW-NEXT:    xorl %eax, %eax
-; AVX512BW-NEXT:    vucomisd %xmm8, %xmm9
-; AVX512BW-NEXT:    movq $-1, %rcx
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm10
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm8 = xmm8[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm9 = xmm9[1,0]
-; AVX512BW-NEXT:    vucomisd %xmm8, %xmm9
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm8
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm8 = xmm10[0],xmm8[0]
-; AVX512BW-NEXT:    vextractf32x4 $2, %zmm4, %xmm9
-; AVX512BW-NEXT:    vextractf32x4 $2, %zmm0, %xmm10
-; AVX512BW-NEXT:    vucomisd %xmm9, %xmm10
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm11
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm9 = xmm9[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm10 = xmm10[1,0]
-; AVX512BW-NEXT:    vucomisd %xmm9, %xmm10
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm9
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm9 = xmm11[0],xmm9[0]
-; AVX512BW-NEXT:    vinserti128 $1, %xmm8, %ymm9, %ymm8
-; AVX512BW-NEXT:    vextractf128 $1, %ymm4, %xmm9
-; AVX512BW-NEXT:    vextractf128 $1, %ymm0, %xmm10
-; AVX512BW-NEXT:    vucomisd %xmm9, %xmm10
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm11
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm9 = xmm9[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm10 = xmm10[1,0]
-; AVX512BW-NEXT:    vucomisd %xmm9, %xmm10
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm9
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm9 = xmm11[0],xmm9[0]
-; AVX512BW-NEXT:    vucomisd %xmm4, %xmm0
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm10
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm4[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512BW-NEXT:    vucomisd %xmm4, %xmm0
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm0
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm10[0],xmm0[0]
-; AVX512BW-NEXT:    vinserti128 $1, %xmm9, %ymm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm8, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm8
-; AVX512BW-NEXT:    vextractf32x4 $3, %zmm5, %xmm4
-; AVX512BW-NEXT:    vextractf32x4 $3, %zmm1, %xmm0
-; AVX512BW-NEXT:    vucomisd %xmm4, %xmm0
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm9
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm4[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512BW-NEXT:    vucomisd %xmm4, %xmm0
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm0
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm9 = xmm9[0],xmm0[0]
-; AVX512BW-NEXT:    vextractf32x4 $2, %zmm5, %xmm4
-; AVX512BW-NEXT:    vextractf32x4 $2, %zmm1, %xmm0
-; AVX512BW-NEXT:    vucomisd %xmm4, %xmm0
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm10
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm4[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512BW-NEXT:    vucomisd %xmm4, %xmm0
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm0
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm10[0],xmm0[0]
-; AVX512BW-NEXT:    vinserti128 $1, %xmm9, %ymm0, %ymm9
-; AVX512BW-NEXT:    vextractf128 $1, %ymm5, %xmm4
-; AVX512BW-NEXT:    vextractf128 $1, %ymm1, %xmm0
-; AVX512BW-NEXT:    vucomisd %xmm4, %xmm0
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm10
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm4[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512BW-NEXT:    vucomisd %xmm4, %xmm0
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm0
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm10[0],xmm0[0]
-; AVX512BW-NEXT:    vucomisd %xmm5, %xmm1
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm4
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm5 = xmm5[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm1[1,0]
-; AVX512BW-NEXT:    vucomisd %xmm5, %xmm1
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm1
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm4[0],xmm1[0]
-; AVX512BW-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm9, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm8, %zmm0
-; AVX512BW-NEXT:    vpmovdw %zmm0, %ymm8
-; AVX512BW-NEXT:    vextractf32x4 $3, %zmm6, %xmm1
-; AVX512BW-NEXT:    vextractf32x4 $3, %zmm2, %xmm4
-; AVX512BW-NEXT:    vucomisd %xmm1, %xmm4
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm5
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm1[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm4[1,0]
-; AVX512BW-NEXT:    vucomisd %xmm1, %xmm4
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm1
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm5[0],xmm1[0]
-; AVX512BW-NEXT:    vextractf32x4 $2, %zmm6, %xmm4
-; AVX512BW-NEXT:    vextractf32x4 $2, %zmm2, %xmm5
-; AVX512BW-NEXT:    vucomisd %xmm4, %xmm5
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm0
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm4[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm5 = xmm5[1,0]
-; AVX512BW-NEXT:    vucomisd %xmm4, %xmm5
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm4
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm4[0]
-; AVX512BW-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
-; AVX512BW-NEXT:    vextractf128 $1, %ymm6, %xmm1
-; AVX512BW-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX512BW-NEXT:    vucomisd %xmm1, %xmm4
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm5
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm1[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm4[1,0]
-; AVX512BW-NEXT:    vucomisd %xmm1, %xmm4
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm1
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm5[0],xmm1[0]
-; AVX512BW-NEXT:    vucomisd %xmm6, %xmm2
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm4
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm5 = xmm6[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512BW-NEXT:    vucomisd %xmm5, %xmm2
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm2
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm2 = xmm4[0],xmm2[0]
-; AVX512BW-NEXT:    vinserti128 $1, %xmm1, %ymm2, %ymm1
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
-; AVX512BW-NEXT:    vextractf32x4 $3, %zmm7, %xmm1
-; AVX512BW-NEXT:    vextractf32x4 $3, %zmm3, %xmm2
-; AVX512BW-NEXT:    vucomisd %xmm1, %xmm2
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm4
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm1[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512BW-NEXT:    vucomisd %xmm1, %xmm2
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm1
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm4[0],xmm1[0]
-; AVX512BW-NEXT:    vextractf32x4 $2, %zmm7, %xmm2
-; AVX512BW-NEXT:    vextractf32x4 $2, %zmm3, %xmm4
-; AVX512BW-NEXT:    vucomisd %xmm2, %xmm4
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm5
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm4[1,0]
-; AVX512BW-NEXT:    vucomisd %xmm2, %xmm4
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm2
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm2 = xmm5[0],xmm2[0]
-; AVX512BW-NEXT:    vinserti128 $1, %xmm1, %ymm2, %ymm1
-; AVX512BW-NEXT:    vextractf128 $1, %ymm7, %xmm2
-; AVX512BW-NEXT:    vextractf128 $1, %ymm3, %xmm4
-; AVX512BW-NEXT:    vucomisd %xmm2, %xmm4
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm5
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm4 = xmm4[1,0]
-; AVX512BW-NEXT:    vucomisd %xmm2, %xmm4
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm2
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm2 = xmm5[0],xmm2[0]
-; AVX512BW-NEXT:    vucomisd %xmm7, %xmm3
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovaq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm4
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm5 = xmm7[1,0]
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm3[1,0]
-; AVX512BW-NEXT:    vucomisd %xmm5, %xmm3
-; AVX512BW-NEXT:    cmovaq %rcx, %rax
-; AVX512BW-NEXT:    vmovq %rax, %xmm3
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm3 = xmm4[0],xmm3[0]
-; AVX512BW-NEXT:    vinserti128 $1, %xmm2, %ymm3, %ymm2
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm1, %zmm2, %zmm1
-; AVX512BW-NEXT:    vpmovqd %zmm1, %ymm1
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpmovdw %zmm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm8, %zmm0
-; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512BW-NEXT:    vcmpltpd %zmm0, %zmm4, %k0
+; AVX512BW-NEXT:    vcmpltpd %zmm1, %zmm5, %k1
+; AVX512BW-NEXT:    kunpckbw %k0, %k1, %k0
+; AVX512BW-NEXT:    vcmpltpd %zmm2, %zmm6, %k1
+; AVX512BW-NEXT:    vcmpltpd %zmm3, %zmm7, %k2
+; AVX512BW-NEXT:    kunpckbw %k1, %k2, %k1
+; AVX512BW-NEXT:    kunpckwd %k0, %k1, %k0
+; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0
+; AVX512BW-NEXT:    # kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
 ; AVX512BW-NEXT:    retq
   %1 = fcmp ogt <32 x double> %a0, %a1
   ret <32 x i1> %1
@@ -9565,261 +8861,15 @@ define <32 x i1> @test_cmp_v32i64(<32 x i64> %a0, <32 x i64> %a1) nounwind {
 ;
 ; AVX512BW-LABEL: test_cmp_v32i64:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm4, %xmm8
-; AVX512BW-NEXT:    vpextrq $1, %xmm8, %rcx
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm0, %xmm9
-; AVX512BW-NEXT:    vpextrq $1, %xmm9, %rdx
-; AVX512BW-NEXT:    xorl %eax, %eax
-; AVX512BW-NEXT:    cmpq %rcx, %rdx
-; AVX512BW-NEXT:    movq $-1, %rcx
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm10
-; AVX512BW-NEXT:    vmovq %xmm8, %rdx
-; AVX512BW-NEXT:    vmovq %xmm9, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm8
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm8 = xmm8[0],xmm10[0]
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm4, %xmm9
-; AVX512BW-NEXT:    vpextrq $1, %xmm9, %rdx
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm0, %xmm10
-; AVX512BW-NEXT:    vpextrq $1, %xmm10, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm11
-; AVX512BW-NEXT:    vmovq %xmm9, %rdx
-; AVX512BW-NEXT:    vmovq %xmm10, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm9
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm9 = xmm9[0],xmm11[0]
-; AVX512BW-NEXT:    vinserti128 $1, %xmm8, %ymm9, %ymm8
-; AVX512BW-NEXT:    vextracti128 $1, %ymm4, %xmm9
-; AVX512BW-NEXT:    vpextrq $1, %xmm9, %rdx
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm10
-; AVX512BW-NEXT:    vpextrq $1, %xmm10, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm11
-; AVX512BW-NEXT:    vmovq %xmm9, %rdx
-; AVX512BW-NEXT:    vmovq %xmm10, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm9
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm9 = xmm9[0],xmm11[0]
-; AVX512BW-NEXT:    vpextrq $1, %xmm4, %rdx
-; AVX512BW-NEXT:    vpextrq $1, %xmm0, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm10
-; AVX512BW-NEXT:    vmovq %xmm4, %rdx
-; AVX512BW-NEXT:    vmovq %xmm0, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm0
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm10[0]
-; AVX512BW-NEXT:    vinserti128 $1, %xmm9, %ymm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm8, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm8
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm5, %xmm4
-; AVX512BW-NEXT:    vpextrq $1, %xmm4, %rdx
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm1, %xmm0
-; AVX512BW-NEXT:    vpextrq $1, %xmm0, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm9
-; AVX512BW-NEXT:    vmovq %xmm4, %rdx
-; AVX512BW-NEXT:    vmovq %xmm0, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm0
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm9 = xmm0[0],xmm9[0]
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm5, %xmm4
-; AVX512BW-NEXT:    vpextrq $1, %xmm4, %rdx
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm1, %xmm0
-; AVX512BW-NEXT:    vpextrq $1, %xmm0, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm10
-; AVX512BW-NEXT:    vmovq %xmm4, %rdx
-; AVX512BW-NEXT:    vmovq %xmm0, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm0
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm10[0]
-; AVX512BW-NEXT:    vinserti128 $1, %xmm9, %ymm0, %ymm9
-; AVX512BW-NEXT:    vextracti128 $1, %ymm5, %xmm0
-; AVX512BW-NEXT:    vpextrq $1, %xmm0, %rdx
-; AVX512BW-NEXT:    vextracti128 $1, %ymm1, %xmm4
-; AVX512BW-NEXT:    vpextrq $1, %xmm4, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm10
-; AVX512BW-NEXT:    vmovq %xmm0, %rdx
-; AVX512BW-NEXT:    vmovq %xmm4, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm0
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm10[0]
-; AVX512BW-NEXT:    vpextrq $1, %xmm5, %rdx
-; AVX512BW-NEXT:    vpextrq $1, %xmm1, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm4
-; AVX512BW-NEXT:    vmovq %xmm5, %rdx
-; AVX512BW-NEXT:    vmovq %xmm1, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm1
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm4[0]
-; AVX512BW-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm9, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm8, %zmm0
-; AVX512BW-NEXT:    vpmovdw %zmm0, %ymm8
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm6, %xmm1
-; AVX512BW-NEXT:    vpextrq $1, %xmm1, %rdx
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm2, %xmm4
-; AVX512BW-NEXT:    vpextrq $1, %xmm4, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm5
-; AVX512BW-NEXT:    vmovq %xmm1, %rdx
-; AVX512BW-NEXT:    vmovq %xmm4, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm1
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm5[0]
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm6, %xmm4
-; AVX512BW-NEXT:    vpextrq $1, %xmm4, %rdx
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm2, %xmm5
-; AVX512BW-NEXT:    vpextrq $1, %xmm5, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm0
-; AVX512BW-NEXT:    vmovq %xmm4, %rdx
-; AVX512BW-NEXT:    vmovq %xmm5, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm4
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm4[0],xmm0[0]
-; AVX512BW-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX512BW-NEXT:    vextracti128 $1, %ymm6, %xmm0
-; AVX512BW-NEXT:    vpextrq $1, %xmm0, %rdx
-; AVX512BW-NEXT:    vextracti128 $1, %ymm2, %xmm4
-; AVX512BW-NEXT:    vpextrq $1, %xmm4, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm5
-; AVX512BW-NEXT:    vmovq %xmm0, %rdx
-; AVX512BW-NEXT:    vmovq %xmm4, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm0
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm5[0]
-; AVX512BW-NEXT:    vpextrq $1, %xmm6, %rdx
-; AVX512BW-NEXT:    vpextrq $1, %xmm2, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm4
-; AVX512BW-NEXT:    vmovq %xmm6, %rdx
-; AVX512BW-NEXT:    vmovq %xmm2, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm2
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm4[0]
-; AVX512BW-NEXT:    vinserti128 $1, %xmm0, %ymm2, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm1
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm7, %xmm0
-; AVX512BW-NEXT:    vpextrq $1, %xmm0, %rdx
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm3, %xmm2
-; AVX512BW-NEXT:    vpextrq $1, %xmm2, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm4
-; AVX512BW-NEXT:    vmovq %xmm0, %rdx
-; AVX512BW-NEXT:    vmovq %xmm2, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm0
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm4[0]
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm7, %xmm2
-; AVX512BW-NEXT:    vpextrq $1, %xmm2, %rdx
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm3, %xmm4
-; AVX512BW-NEXT:    vpextrq $1, %xmm4, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm5
-; AVX512BW-NEXT:    vmovq %xmm2, %rdx
-; AVX512BW-NEXT:    vmovq %xmm4, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm2
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm5[0]
-; AVX512BW-NEXT:    vinserti128 $1, %xmm0, %ymm2, %ymm2
-; AVX512BW-NEXT:    vextracti128 $1, %ymm7, %xmm0
-; AVX512BW-NEXT:    vpextrq $1, %xmm0, %rdx
-; AVX512BW-NEXT:    vextracti128 $1, %ymm3, %xmm4
-; AVX512BW-NEXT:    vpextrq $1, %xmm4, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm5
-; AVX512BW-NEXT:    vmovq %xmm0, %rdx
-; AVX512BW-NEXT:    vmovq %xmm4, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm0
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm5[0]
-; AVX512BW-NEXT:    vpextrq $1, %xmm7, %rdx
-; AVX512BW-NEXT:    vpextrq $1, %xmm3, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    movl $0, %edx
-; AVX512BW-NEXT:    cmovgq %rcx, %rdx
-; AVX512BW-NEXT:    vmovq %rdx, %xmm4
-; AVX512BW-NEXT:    vmovq %xmm7, %rdx
-; AVX512BW-NEXT:    vmovq %xmm3, %rsi
-; AVX512BW-NEXT:    cmpq %rdx, %rsi
-; AVX512BW-NEXT:    cmovgq %rcx, %rax
-; AVX512BW-NEXT:    vmovq %rax, %xmm3
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm4[0]
-; AVX512BW-NEXT:    vinserti128 $1, %xmm0, %ymm3, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512BW-NEXT:    vpmovdw %zmm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm8, %zmm0
-; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512BW-NEXT:    vpcmpgtq %zmm4, %zmm0, %k0
+; AVX512BW-NEXT:    vpcmpgtq %zmm5, %zmm1, %k1
+; AVX512BW-NEXT:    kunpckbw %k0, %k1, %k0
+; AVX512BW-NEXT:    vpcmpgtq %zmm6, %zmm2, %k1
+; AVX512BW-NEXT:    vpcmpgtq %zmm7, %zmm3, %k2
+; AVX512BW-NEXT:    kunpckbw %k1, %k2, %k1
+; AVX512BW-NEXT:    kunpckwd %k0, %k1, %k0
+; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0
+; AVX512BW-NEXT:    # kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
 ; AVX512BW-NEXT:    retq
   %1 = icmp sgt <32 x i64> %a0, %a1
   ret <32 x i1> %1