[X86][AVX512] Improve vector bool reductions

As predicate masks are legal on AVX512 targets, we avoid MOVMSK in these cases, but we can just bitcast the bool vector to the integer equivalent directly - avoiding expansion of the reduction to a shuffle pattern.

llvm-svn: 359386

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index d7e88da..420397f 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -34390,8 +34390,8 @@ static SDValue combineHorizontalMinMaxResult(SDNode *Extract, SelectionDAG &DAG,
 static SDValue combineHorizontalPredicateResult(SDNode *Extract,
                                                 SelectionDAG &DAG,
                                                 const X86Subtarget &Subtarget) {
-  // Bail without SSE2 or with AVX512VL (which uses predicate registers).
-  if (!Subtarget.hasSSE2() || Subtarget.hasVLX())
+  // Bail without SSE2.
+  if (!Subtarget.hasSSE2())
     return SDValue();
 
   EVT ExtractVT = Extract->getValueType(0);
@@ -34413,25 +34413,36 @@ static SDValue combineHorizontalPredicateResult(SDNode *Extract,
 
   SDValue Movmsk;
   SDLoc DL(Extract);
-  unsigned NumElts = Match.getValueType().getVectorNumElements();
+  EVT MatchVT = Match.getValueType();
+  unsigned NumElts = MatchVT.getVectorNumElements();
 
   if (ExtractVT == MVT::i1) {
     // Special case for (pre-legalization) vXi1 reductions.
-    // Use combineBitcastvxi1 to create the MOVMSK.
     if (NumElts > 32)
       return SDValue();
-    if (NumElts == 32 && !Subtarget.hasInt256()) {
-      SDValue Lo, Hi;
-      std::tie(Lo, Hi) = DAG.SplitVector(Match, DL);
-      Match = DAG.getNode(BinOp, DL, Lo.getValueType(), Lo, Hi);
-      NumElts = 16;
+    if (DAG.getTargetLoweringInfo().isTypeLegal(MatchVT)) {
+      // If this is a legal AVX512 predicate type then we can just bitcast.
+      EVT MovmskVT = EVT::getIntegerVT(*DAG.getContext(), NumElts);
+      Movmsk = DAG.getBitcast(MovmskVT, Match);
+    } else {
+      // Use combineBitcastvxi1 to create the MOVMSK.
+      if (NumElts == 32 && !Subtarget.hasInt256()) {
+        SDValue Lo, Hi;
+        std::tie(Lo, Hi) = DAG.SplitVector(Match, DL);
+        Match = DAG.getNode(BinOp, DL, Lo.getValueType(), Lo, Hi);
+        NumElts = 16;
+      }
+      EVT MovmskVT = EVT::getIntegerVT(*DAG.getContext(), NumElts);
+      Movmsk = combineBitcastvxi1(DAG, MovmskVT, Match, DL, Subtarget);
     }
-    EVT MovmskVT = EVT::getIntegerVT(*DAG.getContext(), NumElts);
-    Movmsk = combineBitcastvxi1(DAG, MovmskVT, Match, DL, Subtarget);
     if (!Movmsk)
       return SDValue();
     Movmsk = DAG.getZExtOrTrunc(Movmsk, DL, MVT::i32);
   } else {
+    // Bail with AVX512VL (which uses predicate registers).
+    if (Subtarget.hasVLX())
+      return SDValue();
+
     unsigned MatchSizeInBits = Match.getValueSizeInBits();
     if (!(MatchSizeInBits == 128 ||
           (MatchSizeInBits == 256 && Subtarget.hasAVX())))

diff --git a/llvm/test/CodeGen/X86/vector-compare-all_of.ll b/llvm/test/CodeGen/X86/vector-compare-all_of.ll
index b05d093..abd9495 100644 (file)
--- a/llvm/test/CodeGen/X86/vector-compare-all_of.ll
+++ b/llvm/test/CodeGen/X86/vector-compare-all_of.ll
@@ -935,14 +935,11 @@ define i1 @bool_reduction_v2f64(<2 x double> %x, <2 x double> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v2f64:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vcmpltpd %xmm0, %xmm1, %k1
-; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512-NEXT:    vpsllq $63, %xmm0, %xmm0
-; AVX512-NEXT:    vptestmq %xmm0, %xmm0, %k0 {%k1}
+; AVX512-NEXT:    vcmpltpd %xmm0, %xmm1, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    andb $3, %al
+; AVX512-NEXT:    cmpb $3, %al
+; AVX512-NEXT:    sete %al
 ; AVX512-NEXT:    retq
   %a = fcmp ogt <2 x double> %x, %y
   %b = shufflevector <2 x i1> %a, <2 x i1> undef, <2 x i32> <i32 1, i32 undef>
@@ -970,18 +967,11 @@ define i1 @bool_reduction_v4f32(<4 x float> %x, <4 x float> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v4f32:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
-; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512-NEXT:    vptestmd %xmm1, %xmm1, %k1 {%k1}
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512-NEXT:    vptestmd %xmm0, %xmm0, %k0 {%k1}
+; AVX512-NEXT:    vcmpeqps %xmm1, %xmm0, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    andb $15, %al
+; AVX512-NEXT:    cmpb $15, %al
+; AVX512-NEXT:    sete %al
 ; AVX512-NEXT:    retq
   %a = fcmp oeq <4 x float> %x, %y
   %s1 = shufflevector <4 x i1> %a, <4 x i1> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
@@ -1014,18 +1004,11 @@ define i1 @bool_reduction_v4f64(<4 x double> %x, <4 x double> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v4f64:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vcmplepd %ymm0, %ymm1, %k1
-; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512-NEXT:    vptestmd %xmm1, %xmm1, %k1 {%k1}
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512-NEXT:    vptestmd %xmm0, %xmm0, %k0 {%k1}
+; AVX512-NEXT:    vcmplepd %ymm0, %ymm1, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    andb $15, %al
+; AVX512-NEXT:    cmpb $15, %al
+; AVX512-NEXT:    sete %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = fcmp oge <4 x double> %x, %y
@@ -1060,22 +1043,10 @@ define i1 @bool_reduction_v8f32(<8 x float> %x, <8 x float> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v8f32:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vcmpneqps %ymm1, %ymm0, %k1
-; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512-NEXT:    vptestmd %ymm0, %ymm0, %k0 {%k1}
+; AVX512-NEXT:    vcmpneqps %ymm1, %ymm0, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    cmpb $-1, %al
+; AVX512-NEXT:    sete %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = fcmp une <8 x float> %x, %y
@@ -1114,14 +1085,11 @@ define i1 @bool_reduction_v2i64(<2 x i64> %x, <2 x i64> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v2i64:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpcmpnleuq %xmm1, %xmm0, %k1
-; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512-NEXT:    vpsllq $63, %xmm0, %xmm0
-; AVX512-NEXT:    vptestmq %xmm0, %xmm0, %k0 {%k1}
+; AVX512-NEXT:    vpcmpnleuq %xmm1, %xmm0, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    andb $3, %al
+; AVX512-NEXT:    cmpb $3, %al
+; AVX512-NEXT:    sete %al
 ; AVX512-NEXT:    retq
   %a = icmp ugt <2 x i64> %x, %y
   %b = shufflevector <2 x i1> %a, <2 x i1> undef, <2 x i32> <i32 1, i32 undef>
@@ -1153,18 +1121,11 @@ define i1 @bool_reduction_v4i32(<4 x i32> %x, <4 x i32> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v4i32:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpcmpneqd %xmm1, %xmm0, %k1
-; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512-NEXT:    vptestmd %xmm1, %xmm1, %k1 {%k1}
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512-NEXT:    vptestmd %xmm0, %xmm0, %k0 {%k1}
+; AVX512-NEXT:    vpcmpneqd %xmm1, %xmm0, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    andb $15, %al
+; AVX512-NEXT:    cmpb $15, %al
+; AVX512-NEXT:    sete %al
 ; AVX512-NEXT:    retq
   %a = icmp ne <4 x i32> %x, %y
   %s1 = shufflevector <4 x i1> %a, <4 x i1> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
@@ -1196,23 +1157,10 @@ define i1 @bool_reduction_v8i16(<8 x i16> %x, <8 x i16> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v8i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpcmpgtw %xmm0, %xmm1, %k1
-; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512-NEXT:    vptestmd %ymm0, %ymm0, %k0 {%k1}
+; AVX512-NEXT:    vpcmpgtw %xmm0, %xmm1, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    cmpb $-1, %al
+; AVX512-NEXT:    sete %al
 ; AVX512-NEXT:    retq
   %a = icmp slt <8 x i16> %x, %y
   %s1 = shufflevector <8 x i1> %a, <8 x i1> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
@@ -1245,16 +1193,8 @@ define i1 @bool_reduction_v16i8(<16 x i8> %x, <16 x i8> %y) {
 ; AVX512-LABEL: bool_reduction_v16i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0
-; AVX512-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512-NEXT:    kandw %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512-NEXT:    kandw %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512-NEXT:    kandw %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512-NEXT:    kandw %k0, %k1, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    kortestw %k0, %k0
+; AVX512-NEXT:    setb %al
 ; AVX512-NEXT:    retq
   %a = icmp sgt <16 x i8> %x, %y
   %s1 = shufflevector <16 x i1> %a, <16 x i1> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
@@ -1304,18 +1244,11 @@ define i1 @bool_reduction_v4i64(<4 x i64> %x, <4 x i64> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v4i64:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpcmpgtq %ymm0, %ymm1, %k1
-; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512-NEXT:    vptestmd %xmm1, %xmm1, %k1 {%k1}
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512-NEXT:    vptestmd %xmm0, %xmm0, %k0 {%k1}
+; AVX512-NEXT:    vpcmpgtq %ymm0, %ymm1, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    andb $15, %al
+; AVX512-NEXT:    cmpb $15, %al
+; AVX512-NEXT:    sete %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = icmp slt <4 x i64> %x, %y
@@ -1368,22 +1301,10 @@ define i1 @bool_reduction_v8i32(<8 x i32> %x, <8 x i32> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v8i32:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpcmpleud %ymm1, %ymm0, %k1
-; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512-NEXT:    vptestmd %ymm0, %ymm0, %k0 {%k1}
+; AVX512-NEXT:    vpcmpleud %ymm1, %ymm0, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    cmpb $-1, %al
+; AVX512-NEXT:    sete %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = icmp ule <8 x i32> %x, %y
@@ -1435,16 +1356,8 @@ define i1 @bool_reduction_v16i16(<16 x i16> %x, <16 x i16> %y) {
 ; AVX512-LABEL: bool_reduction_v16i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpeqw %ymm1, %ymm0, %k0
-; AVX512-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512-NEXT:    kandw %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512-NEXT:    kandw %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512-NEXT:    kandw %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512-NEXT:    kandw %k0, %k1, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    kortestw %k0, %k0
+; AVX512-NEXT:    setb %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = icmp eq <16 x i16> %x, %y
@@ -1496,18 +1409,8 @@ define i1 @bool_reduction_v32i8(<32 x i8> %x, <32 x i8> %y) {
 ; AVX512-LABEL: bool_reduction_v32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpeqb %ymm1, %ymm0, %k0
-; AVX512-NEXT:    kshiftrd $16, %k0, %k1
-; AVX512-NEXT:    kandd %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrd $8, %k0, %k1
-; AVX512-NEXT:    kandd %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrd $4, %k0, %k1
-; AVX512-NEXT:    kandd %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrd $2, %k0, %k1
-; AVX512-NEXT:    kandd %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrd $1, %k0, %k1
-; AVX512-NEXT:    kandd %k0, %k1, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    kortestd %k0, %k0
+; AVX512-NEXT:    setb %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = icmp eq <32 x i8> %x, %y

diff --git a/llvm/test/CodeGen/X86/vector-compare-any_of.ll b/llvm/test/CodeGen/X86/vector-compare-any_of.ll
index b526109..051e3c0 100644 (file)
--- a/llvm/test/CodeGen/X86/vector-compare-any_of.ll
+++ b/llvm/test/CodeGen/X86/vector-compare-any_of.ll
@@ -855,15 +855,10 @@ define i1 @bool_reduction_v2f64(<2 x double> %x, <2 x double> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v2f64:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vcmpltpd %xmm0, %xmm1, %k1
-; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512-NEXT:    vpsllq $63, %xmm0, %xmm0
-; AVX512-NEXT:    vptestmq %xmm0, %xmm0, %k0
-; AVX512-NEXT:    korw %k0, %k1, %k0
+; AVX512-NEXT:    vcmpltpd %xmm0, %xmm1, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    testb $3, %al
+; AVX512-NEXT:    setne %al
 ; AVX512-NEXT:    retq
   %a = fcmp ogt <2 x double> %x, %y
   %b = shufflevector <2 x i1> %a, <2 x i1> undef, <2 x i32> <i32 1, i32 undef>
@@ -891,20 +886,10 @@ define i1 @bool_reduction_v4f32(<4 x float> %x, <4 x float> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v4f32:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
-; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512-NEXT:    vptestmd %xmm1, %xmm1, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k1
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512-NEXT:    vptestmd %xmm0, %xmm0, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k0
+; AVX512-NEXT:    vcmpeqps %xmm1, %xmm0, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    testb $15, %al
+; AVX512-NEXT:    setne %al
 ; AVX512-NEXT:    retq
   %a = fcmp oeq <4 x float> %x, %y
   %s1 = shufflevector <4 x i1> %a, <4 x i1> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
@@ -937,20 +922,10 @@ define i1 @bool_reduction_v4f64(<4 x double> %x, <4 x double> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v4f64:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vcmplepd %ymm0, %ymm1, %k1
-; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512-NEXT:    vptestmd %xmm1, %xmm1, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k1
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512-NEXT:    vptestmd %xmm0, %xmm0, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k0
+; AVX512-NEXT:    vcmplepd %ymm0, %ymm1, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    testb $15, %al
+; AVX512-NEXT:    setne %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = fcmp oge <4 x double> %x, %y
@@ -985,25 +960,10 @@ define i1 @bool_reduction_v8f32(<8 x float> %x, <8 x float> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v8f32:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vcmpneqps %ymm1, %ymm0, %k1
-; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512-NEXT:    vptestmd %ymm1, %ymm1, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k1
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512-NEXT:    vptestmd %ymm1, %ymm1, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k1
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512-NEXT:    vptestmd %ymm0, %ymm0, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k0
+; AVX512-NEXT:    vcmpneqps %ymm1, %ymm0, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    testb %al, %al
+; AVX512-NEXT:    setne %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = fcmp une <8 x float> %x, %y
@@ -1042,15 +1002,10 @@ define i1 @bool_reduction_v2i64(<2 x i64> %x, <2 x i64> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v2i64:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpcmpnleuq %xmm1, %xmm0, %k1
-; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512-NEXT:    vpsllq $63, %xmm0, %xmm0
-; AVX512-NEXT:    vptestmq %xmm0, %xmm0, %k0
-; AVX512-NEXT:    korw %k0, %k1, %k0
+; AVX512-NEXT:    vpcmpnleuq %xmm1, %xmm0, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    testb $3, %al
+; AVX512-NEXT:    setne %al
 ; AVX512-NEXT:    retq
   %a = icmp ugt <2 x i64> %x, %y
   %b = shufflevector <2 x i1> %a, <2 x i1> undef, <2 x i32> <i32 1, i32 undef>
@@ -1082,20 +1037,10 @@ define i1 @bool_reduction_v4i32(<4 x i32> %x, <4 x i32> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v4i32:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpcmpneqd %xmm1, %xmm0, %k1
-; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512-NEXT:    vptestmd %xmm1, %xmm1, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k1
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512-NEXT:    vptestmd %xmm0, %xmm0, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k0
+; AVX512-NEXT:    vpcmpneqd %xmm1, %xmm0, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    testb $15, %al
+; AVX512-NEXT:    setne %al
 ; AVX512-NEXT:    retq
   %a = icmp ne <4 x i32> %x, %y
   %s1 = shufflevector <4 x i1> %a, <4 x i1> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
@@ -1127,26 +1072,10 @@ define i1 @bool_reduction_v8i16(<8 x i16> %x, <8 x i16> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v8i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpcmpgtw %xmm0, %xmm1, %k1
-; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512-NEXT:    vptestmd %ymm1, %ymm1, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k1
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512-NEXT:    vptestmd %ymm1, %ymm1, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k1
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512-NEXT:    vptestmd %ymm0, %ymm0, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k0
+; AVX512-NEXT:    vpcmpgtw %xmm0, %xmm1, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    testb %al, %al
+; AVX512-NEXT:    setne %al
 ; AVX512-NEXT:    retq
   %a = icmp slt <8 x i16> %x, %y
   %s1 = shufflevector <8 x i1> %a, <8 x i1> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
@@ -1179,16 +1108,8 @@ define i1 @bool_reduction_v16i8(<16 x i8> %x, <16 x i8> %y) {
 ; AVX512-LABEL: bool_reduction_v16i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0
-; AVX512-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512-NEXT:    korw %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512-NEXT:    korw %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512-NEXT:    korw %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512-NEXT:    korw %k0, %k1, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    kortestw %k0, %k0
+; AVX512-NEXT:    setne %al
 ; AVX512-NEXT:    retq
   %a = icmp sgt <16 x i8> %x, %y
   %s1 = shufflevector <16 x i1> %a, <16 x i1> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
@@ -1238,20 +1159,10 @@ define i1 @bool_reduction_v4i64(<4 x i64> %x, <4 x i64> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v4i64:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpcmpgtq %ymm0, %ymm1, %k1
-; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512-NEXT:    vptestmd %xmm1, %xmm1, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k1
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512-NEXT:    vptestmd %xmm0, %xmm0, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k0
+; AVX512-NEXT:    vpcmpgtq %ymm0, %ymm1, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    testb $15, %al
+; AVX512-NEXT:    setne %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = icmp slt <4 x i64> %x, %y
@@ -1304,25 +1215,10 @@ define i1 @bool_reduction_v8i32(<8 x i32> %x, <8 x i32> %y) {
 ;
 ; AVX512-LABEL: bool_reduction_v8i32:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpcmpleud %ymm1, %ymm0, %k1
-; AVX512-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512-NEXT:    vptestmd %ymm1, %ymm1, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k1
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512-NEXT:    vptestmd %ymm1, %ymm1, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k1
-; AVX512-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512-NEXT:    vptestmd %ymm0, %ymm0, %k0
-; AVX512-NEXT:    korw %k1, %k0, %k0
+; AVX512-NEXT:    vpcmpleud %ymm1, %ymm0, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    testb %al, %al
+; AVX512-NEXT:    setne %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = icmp ule <8 x i32> %x, %y
@@ -1374,16 +1270,8 @@ define i1 @bool_reduction_v16i16(<16 x i16> %x, <16 x i16> %y) {
 ; AVX512-LABEL: bool_reduction_v16i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpeqw %ymm1, %ymm0, %k0
-; AVX512-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512-NEXT:    korw %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512-NEXT:    korw %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512-NEXT:    korw %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512-NEXT:    korw %k0, %k1, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    kortestw %k0, %k0
+; AVX512-NEXT:    setne %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = icmp eq <16 x i16> %x, %y
@@ -1435,18 +1323,8 @@ define i1 @bool_reduction_v32i8(<32 x i8> %x, <32 x i8> %y) {
 ; AVX512-LABEL: bool_reduction_v32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpcmpeqb %ymm1, %ymm0, %k0
-; AVX512-NEXT:    kshiftrd $16, %k0, %k1
-; AVX512-NEXT:    kord %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrd $8, %k0, %k1
-; AVX512-NEXT:    kord %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrd $4, %k0, %k1
-; AVX512-NEXT:    kord %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrd $2, %k0, %k1
-; AVX512-NEXT:    kord %k0, %k1, %k0
-; AVX512-NEXT:    kshiftrd $1, %k0, %k1
-; AVX512-NEXT:    kord %k0, %k1, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    kortestd %k0, %k0
+; AVX512-NEXT:    setne %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = icmp eq <32 x i8> %x, %y

diff --git a/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll b/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
index fba7fe4..5f8eb82 100644 (file)
--- a/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
+++ b/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
@@ -30,27 +30,22 @@ define i1 @trunc_v2i64_v2i1(<2 x i64>) {
 ; AVX512BW-LABEL: trunc_v2i64_v2i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpsllq $63, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpsllq $63, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
+; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    andb $3, %al
+; AVX512BW-NEXT:    cmpb $3, %al
+; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: trunc_v2i64_v2i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllq $63, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestmq %xmm0, %xmm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512VL-NEXT:    vpsllq $63, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestmq %xmm0, %xmm0, %k0 {%k1}
+; AVX512VL-NEXT:    vptestmq %xmm0, %xmm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    andb $3, %al
+; AVX512VL-NEXT:    cmpb $3, %al
+; AVX512VL-NEXT:    sete %al
 ; AVX512VL-NEXT:    retq
   %a = trunc <2 x i64> %0 to <2 x i1>
   %b = call i1 @llvm.experimental.vector.reduce.and.v2i1(<2 x i1> %a)
@@ -77,35 +72,22 @@ define i1 @trunc_v4i32_v4i1(<4 x i32>) {
 ; AVX512BW-LABEL: trunc_v4i32_v4i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    andb $15, %al
+; AVX512BW-NEXT:    cmpb $15, %al
+; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: trunc_v4i32_v4i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestmd %xmm0, %xmm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512VL-NEXT:    vptestmd %xmm1, %xmm1, %k1 {%k1}
-; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestmd %xmm0, %xmm0, %k0 {%k1}
+; AVX512VL-NEXT:    vptestmd %xmm0, %xmm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    andb $15, %al
+; AVX512VL-NEXT:    cmpb $15, %al
+; AVX512VL-NEXT:    sete %al
 ; AVX512VL-NEXT:    retq
   %a = trunc <4 x i32> %0 to <4 x i1>
   %b = call i1 @llvm.experimental.vector.reduce.and.v4i1(<4 x i1> %a)
@@ -134,44 +116,20 @@ define i1 @trunc_v8i16_v8i1(<8 x i8>) {
 ; AVX512BW-LABEL: trunc_v8i16_v8i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpsllw $15, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpmovw2m %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1 {%k1}
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1 {%k1}
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
+; AVX512BW-NEXT:    vpmovw2m %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    cmpb $-1, %al
+; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: trunc_v8i16_v8i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllw $15, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpmovw2m %xmm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k0 {%k1}
+; AVX512VL-NEXT:    vpmovw2m %xmm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    cmpb $-1, %al
+; AVX512VL-NEXT:    sete %al
 ; AVX512VL-NEXT:    retq
   %a = trunc <8 x i8> %0 to <8 x i1>
   %b = call i1 @llvm.experimental.vector.reduce.and.v8i1(<8 x i1> %a)
@@ -195,39 +153,13 @@ define i1 @trunc_v16i8_v16i1(<16 x i8>) {
 ; AVX-NEXT:    sete %al
 ; AVX-NEXT:    retq
 ;
-; AVX512BW-LABEL: trunc_v16i8_v16i1:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpsllw $7, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpmovb2m %zmm0, %k0
-; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512BW-NEXT:    vpcmpgtb %zmm0, %zmm1, %k0 {%k1}
-; AVX512BW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512BW-NEXT:    kandw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512BW-NEXT:    kandw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512BW-NEXT:    kandw %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512BW-NEXT:    vzeroupper
-; AVX512BW-NEXT:    retq
-;
-; AVX512VL-LABEL: trunc_v16i8_v16i1:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpsllw $7, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpmovb2m %xmm0, %k0
-; AVX512VL-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512VL-NEXT:    kandw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512VL-NEXT:    kandw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512VL-NEXT:    kandw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512VL-NEXT:    kandw %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512VL-NEXT:    retq
+; AVX512-LABEL: trunc_v16i8_v16i1:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpsllw $7, %xmm0, %xmm0
+; AVX512-NEXT:    vpmovmskb %xmm0, %eax
+; AVX512-NEXT:    cmpw $-1, %ax
+; AVX512-NEXT:    sete %al
+; AVX512-NEXT:    retq
   %a = trunc <16 x i8> %0 to <16 x i1>
   %b = call i1 @llvm.experimental.vector.reduce.and.v16i1(<16 x i1> %a)
   ret i1 %b
@@ -257,35 +189,22 @@ define i1 @trunc_v4i64_v4i1(<4 x i64>) {
 ; AVX512BW-LABEL: trunc_v4i64_v4i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpsllq $63, %ymm0, %ymm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    andb $15, %al
+; AVX512BW-NEXT:    cmpb $15, %al
+; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: trunc_v4i64_v4i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllq $63, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmq %ymm0, %ymm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512VL-NEXT:    vptestmd %xmm1, %xmm1, %k1 {%k1}
-; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestmd %xmm0, %xmm0, %k0 {%k1}
+; AVX512VL-NEXT:    vptestmq %ymm0, %ymm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    andb $15, %al
+; AVX512VL-NEXT:    cmpb $15, %al
+; AVX512VL-NEXT:    sete %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = trunc <4 x i64> %0 to <4 x i1>
@@ -351,43 +270,20 @@ define i1 @trunc_v8i32_v8i1(<8 x i32>) {
 ; AVX512BW-LABEL: trunc_v8i32_v8i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1 {%k1}
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1 {%k1}
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
+; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    cmpb $-1, %al
+; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: trunc_v8i32_v8i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k0 {%k1}
+; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    cmpb $-1, %al
+; AVX512VL-NEXT:    sete %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = trunc <8 x i32> %0 to <8 x i1>
@@ -448,17 +344,8 @@ define i1 @trunc_v16i16_v16i1(<16 x i16>) {
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpsllw $15, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpmovw2m %zmm0, %k0
-; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512BW-NEXT:    vpcmpgtw %zmm0, %zmm1, %k0 {%k1}
-; AVX512BW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512BW-NEXT:    kandw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512BW-NEXT:    kandw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512BW-NEXT:    kandw %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    kortestw %k0, %k0
+; AVX512BW-NEXT:    setb %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
@@ -466,16 +353,8 @@ define i1 @trunc_v16i16_v16i1(<16 x i16>) {
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllw $15, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpmovw2m %ymm0, %k0
-; AVX512VL-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512VL-NEXT:    kandw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512VL-NEXT:    kandw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512VL-NEXT:    kandw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512VL-NEXT:    kandw %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    kortestw %k0, %k0
+; AVX512VL-NEXT:    setb %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = trunc <16 x i16> %0 to <16 x i1>
@@ -513,44 +392,14 @@ define i1 @trunc_v32i8_v32i1(<32 x i8>) {
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
-; AVX512BW-LABEL: trunc_v32i8_v32i1:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpsllw $7, %ymm0, %ymm0
-; AVX512BW-NEXT:    vpmovb2m %zmm0, %k0
-; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT:    kshiftrd $16, %k0, %k1
-; AVX512BW-NEXT:    vpcmpgtb %zmm0, %zmm1, %k0 {%k1}
-; AVX512BW-NEXT:    kshiftrd $8, %k0, %k1
-; AVX512BW-NEXT:    kandd %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrd $4, %k0, %k1
-; AVX512BW-NEXT:    kandd %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrd $2, %k0, %k1
-; AVX512BW-NEXT:    kandd %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrd $1, %k0, %k1
-; AVX512BW-NEXT:    kandd %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512BW-NEXT:    vzeroupper
-; AVX512BW-NEXT:    retq
-;
-; AVX512VL-LABEL: trunc_v32i8_v32i1:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpsllw $7, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpmovb2m %ymm0, %k0
-; AVX512VL-NEXT:    kshiftrd $16, %k0, %k1
-; AVX512VL-NEXT:    kandd %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrd $8, %k0, %k1
-; AVX512VL-NEXT:    kandd %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrd $4, %k0, %k1
-; AVX512VL-NEXT:    kandd %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrd $2, %k0, %k1
-; AVX512VL-NEXT:    kandd %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrd $1, %k0, %k1
-; AVX512VL-NEXT:    kandd %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512VL-NEXT:    vzeroupper
-; AVX512VL-NEXT:    retq
+; AVX512-LABEL: trunc_v32i8_v32i1:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpsllw $7, %ymm0, %ymm0
+; AVX512-NEXT:    vpmovmskb %ymm0, %eax
+; AVX512-NEXT:    cmpl $-1, %eax
+; AVX512-NEXT:    sete %al
+; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    retq
   %a = trunc <32 x i8> %0 to <32 x i1>
   %b = call i1 @llvm.experimental.vector.reduce.and.v32i1(<32 x i1> %a)
   ret i1 %b
@@ -629,48 +478,15 @@ define i1 @trunc_v8i64_v8i1(<8 x i64>) {
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
-; AVX512BW-LABEL: trunc_v8i64_v8i1:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1 {%k1}
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1 {%k1}
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512BW-NEXT:    vzeroupper
-; AVX512BW-NEXT:    retq
-;
-; AVX512VL-LABEL: trunc_v8i64_v8i1:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512VL-NEXT:    vptestmq %zmm0, %zmm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k0 {%k1}
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512VL-NEXT:    vzeroupper
-; AVX512VL-NEXT:    retq
+; AVX512-LABEL: trunc_v8i64_v8i1:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpsllq $63, %zmm0, %zmm0
+; AVX512-NEXT:    vptestmq %zmm0, %zmm0, %k0
+; AVX512-NEXT:    kmovd %k0, %eax
+; AVX512-NEXT:    cmpb $-1, %al
+; AVX512-NEXT:    sete %al
+; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    retq
   %a = trunc <8 x i64> %0 to <8 x i1>
   %b = call i1 @llvm.experimental.vector.reduce.and.v8i1(<8 x i1> %a)
   ret i1 %b
@@ -748,16 +564,8 @@ define i1 @trunc_v16i32_v16i1(<16 x i32>) {
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpslld $31, %zmm0, %zmm0
 ; AVX512-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512-NEXT:    kandw %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512-NEXT:    kandw %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512-NEXT:    kandw %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512-NEXT:    kandw %k1, %k0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    kortestw %k0, %k0
+; AVX512-NEXT:    setb %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = trunc <16 x i32> %0 to <16 x i1>
@@ -832,18 +640,8 @@ define i1 @trunc_v32i16_v32i1(<32 x i16>) {
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsllw $15, %zmm0, %zmm0
 ; AVX512-NEXT:    vpmovw2m %zmm0, %k0
-; AVX512-NEXT:    kshiftrd $16, %k0, %k1
-; AVX512-NEXT:    kandd %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrd $8, %k0, %k1
-; AVX512-NEXT:    kandd %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrd $4, %k0, %k1
-; AVX512-NEXT:    kandd %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrd $2, %k0, %k1
-; AVX512-NEXT:    kandd %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrd $1, %k0, %k1
-; AVX512-NEXT:    kandd %k1, %k0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    kortestd %k0, %k0
+; AVX512-NEXT:    setb %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = trunc <32 x i16> %0 to <32 x i1>
@@ -988,26 +786,21 @@ define i1 @icmp_v2i64_v2i1(<2 x i64>) {
 ; AVX512BW-LABEL: icmp_v2i64_v2i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vptestnmq %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpsllq $63, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
+; AVX512BW-NEXT:    vptestnmq %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    andb $3, %al
+; AVX512BW-NEXT:    cmpb $3, %al
+; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v2i64_v2i1:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vptestnmq %xmm0, %xmm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512VL-NEXT:    vpsllq $63, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestmq %xmm0, %xmm0, %k0 {%k1}
+; AVX512VL-NEXT:    vptestnmq %xmm0, %xmm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    andb $3, %al
+; AVX512VL-NEXT:    cmpb $3, %al
+; AVX512VL-NEXT:    sete %al
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <2 x i64> %0, zeroinitializer
   %b = call i1 @llvm.experimental.vector.reduce.and.v2i1(<2 x i1> %a)
@@ -1036,34 +829,21 @@ define i1 @icmp_v4i32_v4i1(<4 x i32>) {
 ; AVX512BW-LABEL: icmp_v4i32_v4i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vptestnmd %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; AVX512BW-NEXT:    vptestnmd %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    andb $15, %al
+; AVX512BW-NEXT:    cmpb $15, %al
+; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v4i32_v4i1:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vptestnmd %xmm0, %xmm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512VL-NEXT:    vptestmd %xmm1, %xmm1, %k1 {%k1}
-; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestmd %xmm0, %xmm0, %k0 {%k1}
+; AVX512VL-NEXT:    vptestnmd %xmm0, %xmm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    andb $15, %al
+; AVX512VL-NEXT:    cmpb $15, %al
+; AVX512VL-NEXT:    sete %al
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <4 x i32> %0, zeroinitializer
   %b = call i1 @llvm.experimental.vector.reduce.and.v4i1(<4 x i1> %a)
@@ -1097,43 +877,19 @@ define i1 @icmp_v8i16_v8i1(<8 x i8>) {
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm1 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
-; AVX512BW-NEXT:    vptestnmw %zmm1, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1 {%k1}
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1 {%k1}
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
+; AVX512BW-NEXT:    vptestnmw %zmm1, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    cmpb $-1, %al
+; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v8i16_v8i1:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vptestnmw {{.*}}(%rip), %xmm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k0 {%k1}
+; AVX512VL-NEXT:    vptestnmw {{.*}}(%rip), %xmm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    cmpb $-1, %al
+; AVX512VL-NEXT:    sete %al
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <8 x i8> %0, zeroinitializer
   %b = call i1 @llvm.experimental.vector.reduce.and.v8i1(<8 x i1> %a)
@@ -1163,33 +919,16 @@ define i1 @icmp_v16i8_v16i1(<16 x i8>) {
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vptestnmb %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512BW-NEXT:    vpcmpeqb %zmm1, %zmm0, %k0 {%k1}
-; AVX512BW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512BW-NEXT:    kandw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512BW-NEXT:    kandw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512BW-NEXT:    kandw %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    kortestw %k0, %k0
+; AVX512BW-NEXT:    setb %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v16i8_v16i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vptestnmb %xmm0, %xmm0, %k0
-; AVX512VL-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512VL-NEXT:    kandw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512VL-NEXT:    kandw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512VL-NEXT:    kandw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512VL-NEXT:    kandw %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    kortestw %k0, %k0
+; AVX512VL-NEXT:    setb %al
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <16 x i8> %0, zeroinitializer
   %b = call i1 @llvm.experimental.vector.reduce.and.v16i1(<16 x i1> %a)
@@ -1249,34 +988,21 @@ define i1 @icmp_v4i64_v4i1(<4 x i64>) {
 ; AVX512BW-LABEL: icmp_v4i64_v4i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT:    vptestnmq %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k1 {%k1}
-; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k0 {%k1}
+; AVX512BW-NEXT:    vptestnmq %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    andb $15, %al
+; AVX512BW-NEXT:    cmpb $15, %al
+; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v4i64_v4i1:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vptestnmq %ymm0, %ymm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512VL-NEXT:    vptestmd %xmm1, %xmm1, %k1 {%k1}
-; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestmd %xmm0, %xmm0, %k0 {%k1}
+; AVX512VL-NEXT:    vptestnmq %ymm0, %ymm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    andb $15, %al
+; AVX512VL-NEXT:    cmpb $15, %al
+; AVX512VL-NEXT:    sete %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <4 x i64> %0, zeroinitializer
@@ -1323,42 +1049,19 @@ define i1 @icmp_v8i32_v8i1(<8 x i32>) {
 ; AVX512BW-LABEL: icmp_v8i32_v8i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT:    vptestnmd %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1 {%k1}
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1 {%k1}
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
+; AVX512BW-NEXT:    vptestnmd %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    cmpb $-1, %al
+; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v8i32_v8i1:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vptestnmd %ymm0, %ymm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k0 {%k1}
+; AVX512VL-NEXT:    vptestnmd %ymm0, %ymm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    cmpb $-1, %al
+; AVX512VL-NEXT:    sete %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <8 x i32> %0, zeroinitializer
@@ -1407,33 +1110,16 @@ define i1 @icmp_v16i16_v16i1(<16 x i16>) {
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512BW-NEXT:    vptestnmw %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512BW-NEXT:    vpcmpeqw %zmm1, %zmm0, %k0 {%k1}
-; AVX512BW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512BW-NEXT:    kandw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512BW-NEXT:    kandw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512BW-NEXT:    kandw %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    kortestw %k0, %k0
+; AVX512BW-NEXT:    setb %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v16i16_v16i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vptestnmw %ymm0, %ymm0, %k0
-; AVX512VL-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512VL-NEXT:    kandw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512VL-NEXT:    kandw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512VL-NEXT:    kandw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512VL-NEXT:    kandw %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    kortestw %k0, %k0
+; AVX512VL-NEXT:    setb %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <16 x i16> %0, zeroinitializer
@@ -1479,37 +1165,16 @@ define i1 @icmp_v32i8_v32i1(<32 x i8>) {
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512BW-NEXT:    vptestnmb %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT:    kshiftrd $16, %k0, %k1
-; AVX512BW-NEXT:    vpcmpeqb %zmm1, %zmm0, %k0 {%k1}
-; AVX512BW-NEXT:    kshiftrd $8, %k0, %k1
-; AVX512BW-NEXT:    kandd %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrd $4, %k0, %k1
-; AVX512BW-NEXT:    kandd %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrd $2, %k0, %k1
-; AVX512BW-NEXT:    kandd %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrd $1, %k0, %k1
-; AVX512BW-NEXT:    kandd %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    kortestd %k0, %k0
+; AVX512BW-NEXT:    setb %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v32i8_v32i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vptestnmb %ymm0, %ymm0, %k0
-; AVX512VL-NEXT:    kshiftrd $16, %k0, %k1
-; AVX512VL-NEXT:    kandd %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrd $8, %k0, %k1
-; AVX512VL-NEXT:    kandd %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrd $4, %k0, %k1
-; AVX512VL-NEXT:    kandd %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrd $2, %k0, %k1
-; AVX512VL-NEXT:    kandd %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrd $1, %k0, %k1
-; AVX512VL-NEXT:    kandd %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    kortestd %k0, %k0
+; AVX512VL-NEXT:    setb %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <32 x i8> %0, zeroinitializer
@@ -1589,46 +1254,14 @@ define i1 @icmp_v8i64_v8i1(<8 x i64>) {
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
-; AVX512BW-LABEL: icmp_v8i64_v8i1:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vptestnmq %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1 {%k1}
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1 {%k1}
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0 {%k1}
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512BW-NEXT:    vzeroupper
-; AVX512BW-NEXT:    retq
-;
-; AVX512VL-LABEL: icmp_v8i64_v8i1:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vptestnmq %zmm0, %zmm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k1 {%k1}
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k0 {%k1}
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512VL-NEXT:    vzeroupper
-; AVX512VL-NEXT:    retq
+; AVX512-LABEL: icmp_v8i64_v8i1:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vptestnmq %zmm0, %zmm0, %k0
+; AVX512-NEXT:    kmovd %k0, %eax
+; AVX512-NEXT:    cmpb $-1, %al
+; AVX512-NEXT:    sete %al
+; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    retq
   %a = icmp eq <8 x i64> %0, zeroinitializer
   %b = call i1 @llvm.experimental.vector.reduce.and.v8i1(<8 x i1> %a)
   ret i1 %b
@@ -1686,16 +1319,8 @@ define i1 @icmp_v16i32_v16i1(<16 x i32>) {
 ; AVX512-LABEL: icmp_v16i32_v16i1:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vptestnmd %zmm0, %zmm0, %k0
-; AVX512-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512-NEXT:    kandw %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512-NEXT:    kandw %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512-NEXT:    kandw %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512-NEXT:    kandw %k1, %k0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    kortestw %k0, %k0
+; AVX512-NEXT:    setb %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = icmp eq <16 x i32> %0, zeroinitializer
@@ -1754,18 +1379,8 @@ define i1 @icmp_v32i16_v32i1(<32 x i16>) {
 ; AVX512-LABEL: icmp_v32i16_v32i1:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vptestnmw %zmm0, %zmm0, %k0
-; AVX512-NEXT:    kshiftrd $16, %k0, %k1
-; AVX512-NEXT:    kandd %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrd $8, %k0, %k1
-; AVX512-NEXT:    kandd %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrd $4, %k0, %k1
-; AVX512-NEXT:    kandd %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrd $2, %k0, %k1
-; AVX512-NEXT:    kandd %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrd $1, %k0, %k1
-; AVX512-NEXT:    kandd %k1, %k0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    kortestd %k0, %k0
+; AVX512-NEXT:    setb %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = icmp eq <32 x i16> %0, zeroinitializer

diff --git a/llvm/test/CodeGen/X86/vector-reduce-or-bool.ll b/llvm/test/CodeGen/X86/vector-reduce-or-bool.ll
index 859fbf0..f501a8b 100644 (file)
--- a/llvm/test/CodeGen/X86/vector-reduce-or-bool.ll
+++ b/llvm/test/CodeGen/X86/vector-reduce-or-bool.ll
@@ -30,29 +30,20 @@ define i1 @trunc_v2i64_v2i1(<2 x i64>) {
 ; AVX512BW-LABEL: trunc_v2i64_v2i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpsllq $63, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpsllq $63, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    testb $3, %al
+; AVX512BW-NEXT:    setne %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: trunc_v2i64_v2i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllq $63, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestmq %xmm0, %xmm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512VL-NEXT:    vpsllq $63, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vptestmq %xmm0, %xmm0, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    testb $3, %al
+; AVX512VL-NEXT:    setne %al
 ; AVX512VL-NEXT:    retq
   %a = trunc <2 x i64> %0 to <2 x i1>
   %b = call i1 @llvm.experimental.vector.reduce.or.v2i1(<2 x i1> %a)
@@ -79,39 +70,20 @@ define i1 @trunc_v4i32_v4i1(<4 x i32>) {
 ; AVX512BW-LABEL: trunc_v4i32_v4i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k1
-; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    testb $15, %al
+; AVX512BW-NEXT:    setne %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: trunc_v4i32_v4i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestmd %xmm0, %xmm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
-; AVX512VL-NEXT:    vmovdqa32 %xmm1, %xmm2 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpor %xmm2, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestmd %xmm0, %xmm0, %k1
-; AVX512VL-NEXT:    vmovdqa32 %xmm1, %xmm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vptestmd %xmm0, %xmm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    testb $15, %al
+; AVX512VL-NEXT:    setne %al
 ; AVX512VL-NEXT:    retq
   %a = trunc <4 x i32> %0 to <4 x i1>
   %b = call i1 @llvm.experimental.vector.reduce.or.v4i1(<4 x i1> %a)
@@ -140,50 +112,20 @@ define i1 @trunc_v8i16_v8i1(<8 x i8>) {
 ; AVX512BW-LABEL: trunc_v8i16_v8i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpsllw $15, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpmovw2m %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k0
+; AVX512BW-NEXT:    vpmovw2m %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    testb %al, %al
+; AVX512BW-NEXT:    setne %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: trunc_v8i16_v8i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllw $15, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpmovw2m %xmm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k0
+; AVX512VL-NEXT:    vpmovw2m %xmm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    testb %al, %al
+; AVX512VL-NEXT:    setne %al
 ; AVX512VL-NEXT:    retq
   %a = trunc <8 x i8> %0 to <8 x i1>
   %b = call i1 @llvm.experimental.vector.reduce.or.v8i1(<8 x i1> %a)
@@ -207,38 +149,13 @@ define i1 @trunc_v16i8_v16i1(<16 x i8>) {
 ; AVX-NEXT:    setne %al
 ; AVX-NEXT:    retq
 ;
-; AVX512BW-LABEL: trunc_v16i8_v16i1:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpsllw $7, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpmovb2m %zmm0, %k0
-; AVX512BW-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512BW-NEXT:    vzeroupper
-; AVX512BW-NEXT:    retq
-;
-; AVX512VL-LABEL: trunc_v16i8_v16i1:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpsllw $7, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpmovb2m %xmm0, %k0
-; AVX512VL-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512VL-NEXT:    korw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512VL-NEXT:    korw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512VL-NEXT:    korw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512VL-NEXT:    korw %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512VL-NEXT:    retq
+; AVX512-LABEL: trunc_v16i8_v16i1:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpsllw $7, %xmm0, %xmm0
+; AVX512-NEXT:    vpmovmskb %xmm0, %eax
+; AVX512-NEXT:    testw %ax, %ax
+; AVX512-NEXT:    setne %al
+; AVX512-NEXT:    retq
   %a = trunc <16 x i8> %0 to <16 x i1>
   %b = call i1 @llvm.experimental.vector.reduce.or.v16i1(<16 x i1> %a)
   ret i1 %b
@@ -268,39 +185,20 @@ define i1 @trunc_v4i64_v4i1(<4 x i64>) {
 ; AVX512BW-LABEL: trunc_v4i64_v4i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpsllq $63, %ymm0, %ymm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k1
-; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k0
+; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    testb $15, %al
+; AVX512BW-NEXT:    setne %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: trunc_v4i64_v4i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllq $63, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmq %ymm0, %ymm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512VL-NEXT:    vptestmd %xmm1, %xmm1, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestmd %xmm0, %xmm0, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k0
+; AVX512VL-NEXT:    vptestmq %ymm0, %ymm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    testb $15, %al
+; AVX512VL-NEXT:    setne %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = trunc <4 x i64> %0 to <4 x i1>
@@ -366,49 +264,20 @@ define i1 @trunc_v8i32_v8i1(<8 x i32>) {
 ; AVX512BW-LABEL: trunc_v8i32_v8i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k0
+; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    testb %al, %al
+; AVX512BW-NEXT:    setne %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: trunc_v8i32_v8i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
-; AVX512VL-NEXT:    vmovdqa32 %ymm1, %ymm2 {%k1} {z}
-; AVX512VL-NEXT:    vextracti128 $1, %ymm2, %xmm2
-; AVX512VL-NEXT:    vpslld $31, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpor %ymm2, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm1, %ymm2 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpor %ymm2, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm1, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    testb %al, %al
+; AVX512VL-NEXT:    setne %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = trunc <8 x i32> %0 to <8 x i1>
@@ -469,16 +338,8 @@ define i1 @trunc_v16i16_v16i1(<16 x i16>) {
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpsllw $15, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpmovw2m %zmm0, %k0
-; AVX512BW-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    kortestw %k0, %k0
+; AVX512BW-NEXT:    setne %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
@@ -486,16 +347,8 @@ define i1 @trunc_v16i16_v16i1(<16 x i16>) {
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllw $15, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpmovw2m %ymm0, %k0
-; AVX512VL-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512VL-NEXT:    korw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512VL-NEXT:    korw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512VL-NEXT:    korw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512VL-NEXT:    korw %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    kortestw %k0, %k0
+; AVX512VL-NEXT:    setne %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = trunc <16 x i16> %0 to <16 x i1>
@@ -533,43 +386,14 @@ define i1 @trunc_v32i8_v32i1(<32 x i8>) {
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
-; AVX512BW-LABEL: trunc_v32i8_v32i1:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpsllw $7, %ymm0, %ymm0
-; AVX512BW-NEXT:    vpmovb2m %zmm0, %k0
-; AVX512BW-NEXT:    kshiftrd $16, %k0, %k1
-; AVX512BW-NEXT:    kord %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrd $8, %k0, %k1
-; AVX512BW-NEXT:    kord %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrd $4, %k0, %k1
-; AVX512BW-NEXT:    kord %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrd $2, %k0, %k1
-; AVX512BW-NEXT:    kord %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrd $1, %k0, %k1
-; AVX512BW-NEXT:    kord %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512BW-NEXT:    vzeroupper
-; AVX512BW-NEXT:    retq
-;
-; AVX512VL-LABEL: trunc_v32i8_v32i1:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpsllw $7, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpmovb2m %ymm0, %k0
-; AVX512VL-NEXT:    kshiftrd $16, %k0, %k1
-; AVX512VL-NEXT:    kord %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrd $8, %k0, %k1
-; AVX512VL-NEXT:    kord %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrd $4, %k0, %k1
-; AVX512VL-NEXT:    kord %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrd $2, %k0, %k1
-; AVX512VL-NEXT:    kord %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrd $1, %k0, %k1
-; AVX512VL-NEXT:    kord %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512VL-NEXT:    vzeroupper
-; AVX512VL-NEXT:    retq
+; AVX512-LABEL: trunc_v32i8_v32i1:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpsllw $7, %ymm0, %ymm0
+; AVX512-NEXT:    vpmovmskb %ymm0, %eax
+; AVX512-NEXT:    testl %eax, %eax
+; AVX512-NEXT:    setne %al
+; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    retq
   %a = trunc <32 x i8> %0 to <32 x i1>
   %b = call i1 @llvm.experimental.vector.reduce.or.v32i1(<32 x i1> %a)
   ret i1 %b
@@ -648,54 +472,15 @@ define i1 @trunc_v8i64_v8i1(<8 x i64>) {
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
-; AVX512BW-LABEL: trunc_v8i64_v8i1:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512BW-NEXT:    vzeroupper
-; AVX512BW-NEXT:    retq
-;
-; AVX512VL-LABEL: trunc_v8i64_v8i1:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512VL-NEXT:    vptestmq %zmm0, %zmm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512VL-NEXT:    vzeroupper
-; AVX512VL-NEXT:    retq
+; AVX512-LABEL: trunc_v8i64_v8i1:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpsllq $63, %zmm0, %zmm0
+; AVX512-NEXT:    vptestmq %zmm0, %zmm0, %k0
+; AVX512-NEXT:    kmovd %k0, %eax
+; AVX512-NEXT:    testb %al, %al
+; AVX512-NEXT:    setne %al
+; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    retq
   %a = trunc <8 x i64> %0 to <8 x i1>
   %b = call i1 @llvm.experimental.vector.reduce.or.v8i1(<8 x i1> %a)
   ret i1 %b
@@ -773,16 +558,8 @@ define i1 @trunc_v16i32_v16i1(<16 x i32>) {
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpslld $31, %zmm0, %zmm0
 ; AVX512-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512-NEXT:    korw %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512-NEXT:    korw %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512-NEXT:    korw %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512-NEXT:    korw %k1, %k0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    kortestw %k0, %k0
+; AVX512-NEXT:    setne %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = trunc <16 x i32> %0 to <16 x i1>
@@ -857,18 +634,8 @@ define i1 @trunc_v32i16_v32i1(<32 x i16>) {
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsllw $15, %zmm0, %zmm0
 ; AVX512-NEXT:    vpmovw2m %zmm0, %k0
-; AVX512-NEXT:    kshiftrd $16, %k0, %k1
-; AVX512-NEXT:    kord %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrd $8, %k0, %k1
-; AVX512-NEXT:    kord %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrd $4, %k0, %k1
-; AVX512-NEXT:    kord %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrd $2, %k0, %k1
-; AVX512-NEXT:    kord %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrd $1, %k0, %k1
-; AVX512-NEXT:    kord %k1, %k0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    kortestd %k0, %k0
+; AVX512-NEXT:    setne %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = trunc <32 x i16> %0 to <32 x i1>
@@ -1013,28 +780,19 @@ define i1 @icmp_v2i64_v2i1(<2 x i64>) {
 ; AVX512BW-LABEL: icmp_v2i64_v2i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vptestnmq %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpsllq $63, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k0
+; AVX512BW-NEXT:    vptestnmq %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    testb $3, %al
+; AVX512BW-NEXT:    setne %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v2i64_v2i1:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vptestnmq %xmm0, %xmm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512VL-NEXT:    vpsllq $63, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestmq %xmm0, %xmm0, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k0
+; AVX512VL-NEXT:    vptestnmq %xmm0, %xmm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    testb $3, %al
+; AVX512VL-NEXT:    setne %al
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <2 x i64> %0, zeroinitializer
   %b = call i1 @llvm.experimental.vector.reduce.or.v2i1(<2 x i1> %a)
@@ -1063,38 +821,19 @@ define i1 @icmp_v4i32_v4i1(<4 x i32>) {
 ; AVX512BW-LABEL: icmp_v4i32_v4i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vptestnmd %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k1
-; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k0
+; AVX512BW-NEXT:    vptestnmd %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    testb $15, %al
+; AVX512BW-NEXT:    setne %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v4i32_v4i1:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vptestnmd %xmm0, %xmm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512VL-NEXT:    vptestmd %xmm1, %xmm1, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestmd %xmm0, %xmm0, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k0
+; AVX512VL-NEXT:    vptestnmd %xmm0, %xmm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    testb $15, %al
+; AVX512VL-NEXT:    setne %al
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <4 x i32> %0, zeroinitializer
   %b = call i1 @llvm.experimental.vector.reduce.or.v4i1(<4 x i1> %a)
@@ -1128,49 +867,19 @@ define i1 @icmp_v8i16_v8i1(<8 x i8>) {
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm1 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
-; AVX512BW-NEXT:    vptestnmw %zmm1, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k0
+; AVX512BW-NEXT:    vptestnmw %zmm1, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    testb %al, %al
+; AVX512BW-NEXT:    setne %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v8i16_v8i1:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vptestnmw {{.*}}(%rip), %xmm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k0
+; AVX512VL-NEXT:    vptestnmw {{.*}}(%rip), %xmm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    testb %al, %al
+; AVX512VL-NEXT:    setne %al
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <8 x i8> %0, zeroinitializer
   %b = call i1 @llvm.experimental.vector.reduce.or.v8i1(<8 x i1> %a)
@@ -1200,32 +909,16 @@ define i1 @icmp_v16i8_v16i1(<16 x i8>) {
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vptestnmb %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    kortestw %k0, %k0
+; AVX512BW-NEXT:    setne %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v16i8_v16i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vptestnmb %xmm0, %xmm0, %k0
-; AVX512VL-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512VL-NEXT:    korw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512VL-NEXT:    korw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512VL-NEXT:    korw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512VL-NEXT:    korw %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    kortestw %k0, %k0
+; AVX512VL-NEXT:    setne %al
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <16 x i8> %0, zeroinitializer
   %b = call i1 @llvm.experimental.vector.reduce.or.v16i1(<16 x i1> %a)
@@ -1285,38 +978,19 @@ define i1 @icmp_v4i64_v4i1(<4 x i64>) {
 ; AVX512BW-LABEL: icmp_v4i64_v4i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT:    vptestnmq %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k1
-; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k0
+; AVX512BW-NEXT:    vptestnmq %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    testb $15, %al
+; AVX512BW-NEXT:    setne %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v4i64_v4i1:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vptestnmq %ymm0, %ymm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512VL-NEXT:    vptestmd %xmm1, %xmm1, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512VL-NEXT:    vptestmd %xmm0, %xmm0, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k0
+; AVX512VL-NEXT:    vptestnmq %ymm0, %ymm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    testb $15, %al
+; AVX512VL-NEXT:    setne %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <4 x i64> %0, zeroinitializer
@@ -1363,48 +1037,19 @@ define i1 @icmp_v8i32_v8i1(<8 x i32>) {
 ; AVX512BW-LABEL: icmp_v8i32_v8i1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT:    vptestnmd %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k0
+; AVX512BW-NEXT:    vptestnmd %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    testb %al, %al
+; AVX512BW-NEXT:    setne %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v8i32_v8i1:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vptestnmd %ymm0, %ymm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k0
+; AVX512VL-NEXT:    vptestnmd %ymm0, %ymm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    testb %al, %al
+; AVX512VL-NEXT:    setne %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <8 x i32> %0, zeroinitializer
@@ -1453,32 +1098,16 @@ define i1 @icmp_v16i16_v16i1(<16 x i16>) {
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512BW-NEXT:    vptestnmw %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    kortestw %k0, %k0
+; AVX512BW-NEXT:    setne %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v16i16_v16i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vptestnmw %ymm0, %ymm0, %k0
-; AVX512VL-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512VL-NEXT:    korw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512VL-NEXT:    korw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512VL-NEXT:    korw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512VL-NEXT:    korw %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    kortestw %k0, %k0
+; AVX512VL-NEXT:    setne %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <16 x i16> %0, zeroinitializer
@@ -1525,36 +1154,16 @@ define i1 @icmp_v32i8_v32i1(<32 x i8>) {
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512BW-NEXT:    vptestnmb %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    kshiftrd $16, %k0, %k1
-; AVX512BW-NEXT:    kord %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrd $8, %k0, %k1
-; AVX512BW-NEXT:    kord %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrd $4, %k0, %k1
-; AVX512BW-NEXT:    kord %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrd $2, %k0, %k1
-; AVX512BW-NEXT:    kord %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrd $1, %k0, %k1
-; AVX512BW-NEXT:    kord %k1, %k0, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    kortestd %k0, %k0
+; AVX512BW-NEXT:    setne %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: icmp_v32i8_v32i1:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vptestnmb %ymm0, %ymm0, %k0
-; AVX512VL-NEXT:    kshiftrd $16, %k0, %k1
-; AVX512VL-NEXT:    kord %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrd $8, %k0, %k1
-; AVX512VL-NEXT:    kord %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrd $4, %k0, %k1
-; AVX512VL-NEXT:    kord %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrd $2, %k0, %k1
-; AVX512VL-NEXT:    kord %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrd $1, %k0, %k1
-; AVX512VL-NEXT:    kord %k1, %k0, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    kortestd %k0, %k0
+; AVX512VL-NEXT:    setne %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %a = icmp eq <32 x i8> %0, zeroinitializer
@@ -1634,52 +1243,14 @@ define i1 @icmp_v8i64_v8i1(<8 x i64>) {
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
-; AVX512BW-LABEL: icmp_v8i64_v8i1:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vptestnmq %zmm0, %zmm0, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpsllq $63, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
-; AVX512BW-NEXT:    korw %k0, %k1, %k0
-; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512BW-NEXT:    vzeroupper
-; AVX512BW-NEXT:    retq
-;
-; AVX512VL-LABEL: icmp_v8i64_v8i1:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vptestnmq %zmm0, %zmm0, %k1
-; AVX512VL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512VL-NEXT:    vpslld $31, %ymm0, %ymm0
-; AVX512VL-NEXT:    vptestmd %ymm0, %ymm0, %k0
-; AVX512VL-NEXT:    korw %k0, %k1, %k0
-; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512VL-NEXT:    vzeroupper
-; AVX512VL-NEXT:    retq
+; AVX512-LABEL: icmp_v8i64_v8i1:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vptestnmq %zmm0, %zmm0, %k0
+; AVX512-NEXT:    kmovd %k0, %eax
+; AVX512-NEXT:    testb %al, %al
+; AVX512-NEXT:    setne %al
+; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    retq
   %a = icmp eq <8 x i64> %0, zeroinitializer
   %b = call i1 @llvm.experimental.vector.reduce.or.v8i1(<8 x i1> %a)
   ret i1 %b
@@ -1737,16 +1308,8 @@ define i1 @icmp_v16i32_v16i1(<16 x i32>) {
 ; AVX512-LABEL: icmp_v16i32_v16i1:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vptestnmd %zmm0, %zmm0, %k0
-; AVX512-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512-NEXT:    korw %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512-NEXT:    korw %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512-NEXT:    korw %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrw $1, %k0, %k1
-; AVX512-NEXT:    korw %k1, %k0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    kortestw %k0, %k0
+; AVX512-NEXT:    setne %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = icmp eq <16 x i32> %0, zeroinitializer
@@ -1805,18 +1368,8 @@ define i1 @icmp_v32i16_v32i1(<32 x i16>) {
 ; AVX512-LABEL: icmp_v32i16_v32i1:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vptestnmw %zmm0, %zmm0, %k0
-; AVX512-NEXT:    kshiftrd $16, %k0, %k1
-; AVX512-NEXT:    kord %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrd $8, %k0, %k1
-; AVX512-NEXT:    kord %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrd $4, %k0, %k1
-; AVX512-NEXT:    kord %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrd $2, %k0, %k1
-; AVX512-NEXT:    kord %k1, %k0, %k0
-; AVX512-NEXT:    kshiftrd $1, %k0, %k1
-; AVX512-NEXT:    kord %k1, %k0, %k0
-; AVX512-NEXT:    kmovd %k0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    kortestd %k0, %k0
+; AVX512-NEXT:    setne %al
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %a = icmp eq <32 x i16> %0, zeroinitializer