[X86] Fold cmpeq/ne(and(X,Y),Y) --> cmpeq/ne(and(~X,Y),0)

Followup to D100177, handle an similar (demorgan inverse style) case from PR47797 as well

The AVX512 test cases could be further improved if we folded not(iX bitcast(vXi1)) -> (iX bitcast(not(vXi1)))

Alive2: https://alive2.llvm.org/ce/z/AnA_-W

5 files changed, 84 insertions, 65 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 321d4791029a..3e260027bb83 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -48163,9 +48163,9 @@ static SDValue combineSetCC(SDNode *N, SelectionDAG &DAG,
                            DAG.getNode(X86ISD::SETCC, DL, MVT::i8, X86CC, V));
     }
 
-    // cmpeq(or(X,Y),X) --> cmpeq(and(~X,Y),0)
-    // cmpne(or(X,Y),X) --> cmpne(and(~X,Y),0)
     if (OpVT.isScalarInteger()) {
+      // cmpeq(or(X,Y),X) --> cmpeq(and(~X,Y),0)
+      // cmpne(or(X,Y),X) --> cmpne(and(~X,Y),0)
       auto MatchOrCmpEq = [&](SDValue N0, SDValue N1) {
         if (N0.getOpcode() == ISD::OR && N0->hasOneUse()) {
           if (N0.getOperand(0) == N1)
@@ -48181,6 +48181,24 @@ static SDValue combineSetCC(SDNode *N, SelectionDAG &DAG,
         return DAG.getSetCC(DL, VT, AndN, DAG.getConstant(0, DL, OpVT), CC);
       if (SDValue AndN = MatchOrCmpEq(RHS, LHS))
         return DAG.getSetCC(DL, VT, AndN, DAG.getConstant(0, DL, OpVT), CC);
+
+      // cmpeq(and(X,Y),Y) --> cmpeq(and(~X,Y),0)
+      // cmpne(and(X,Y),Y) --> cmpne(and(~X,Y),0)
+      auto MatchAndCmpEq = [&](SDValue N0, SDValue N1) {
+        if (N0.getOpcode() == ISD::AND && N0->hasOneUse()) {
+          if (N0.getOperand(0) == N1)
+            return DAG.getNode(ISD::AND, DL, OpVT, N1,
+                               DAG.getNOT(DL, N0.getOperand(1), OpVT));
+          if (N0.getOperand(1) == N1)
+            return DAG.getNode(ISD::AND, DL, OpVT, N1,
+                               DAG.getNOT(DL, N0.getOperand(0), OpVT));
+        }
+        return SDValue();
+      };
+      if (SDValue AndN = MatchAndCmpEq(LHS, RHS))
+        return DAG.getSetCC(DL, VT, AndN, DAG.getConstant(0, DL, OpVT), CC);
+      if (SDValue AndN = MatchAndCmpEq(RHS, LHS))
+        return DAG.getSetCC(DL, VT, AndN, DAG.getConstant(0, DL, OpVT), CC);
     }
   }
 
diff --git a/llvm/test/CodeGen/X86/bmi.ll b/llvm/test/CodeGen/X86/bmi.ll
index 8aea6f99c144..7708c7255661 100644
--- a/llvm/test/CodeGen/X86/bmi.ll
+++ b/llvm/test/CodeGen/X86/bmi.ll
@@ -157,15 +157,15 @@ define i1 @and_cmp_const(i32 %x) {
 ; X86-LABEL: and_cmp_const:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    andl $43, %eax
-; X86-NEXT:    cmpl $43, %eax
+; X86-NEXT:    notl %eax
+; X86-NEXT:    testb $43, %al
 ; X86-NEXT:    sete %al
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: and_cmp_const:
 ; X64:       # %bb.0:
-; X64-NEXT:    andl $43, %edi
-; X64-NEXT:    cmpl $43, %edi
+; X64-NEXT:    notl %edi
+; X64-NEXT:    testb $43, %dil
 ; X64-NEXT:    sete %al
 ; X64-NEXT:    retq
   %and = and i32 %x, 43
diff --git a/llvm/test/CodeGen/X86/movmsk-cmp.ll b/llvm/test/CodeGen/X86/movmsk-cmp.ll
index 748624f36bf7..3d0b7cd50fcf 100644
--- a/llvm/test/CodeGen/X86/movmsk-cmp.ll
+++ b/llvm/test/CodeGen/X86/movmsk-cmp.ll
@@ -1019,8 +1019,8 @@ define i1 @allzeros_v2i64_not(<2 x i64> %a0) {
 ; KNL-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; KNL-NEXT:    vptestnmq %zmm0, %zmm0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    andb $3, %al
-; KNL-NEXT:    cmpb $3, %al
+; KNL-NEXT:    notb %al
+; KNL-NEXT:    testb $3, %al
 ; KNL-NEXT:    setne %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -1840,8 +1840,8 @@ define i1 @allones_v4i32_and1(<4 x i32> %arg) {
 ; KNL-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; KNL-NEXT:    vptestmd {{.*}}(%rip){1to16}, %zmm0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    andb $15, %al
-; KNL-NEXT:    cmpb $15, %al
+; KNL-NEXT:    notb %al
+; KNL-NEXT:    testb $15, %al
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -2156,8 +2156,8 @@ define i1 @allones_v2i64_and1(<2 x i64> %arg) {
 ; KNL-NEXT:    vmovdqa {{.*#+}} xmm1 = [1,1]
 ; KNL-NEXT:    vptestmq %zmm1, %zmm0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    andb $3, %al
-; KNL-NEXT:    cmpb $3, %al
+; KNL-NEXT:    notb %al
+; KNL-NEXT:    testb $3, %al
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -2254,8 +2254,8 @@ define i1 @allones_v4i64_and1(<4 x i64> %arg) {
 ; KNL-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; KNL-NEXT:    vptestmq {{.*}}(%rip){1to8}, %zmm0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    andb $15, %al
-; KNL-NEXT:    cmpb $15, %al
+; KNL-NEXT:    notb %al
+; KNL-NEXT:    testb $15, %al
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -3161,8 +3161,8 @@ define i1 @allones_v4i32_and4(<4 x i32> %arg) {
 ; KNL-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; KNL-NEXT:    vptestmd {{.*}}(%rip){1to16}, %zmm0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    andb $15, %al
-; KNL-NEXT:    cmpb $15, %al
+; KNL-NEXT:    notb %al
+; KNL-NEXT:    testb $15, %al
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -3477,8 +3477,8 @@ define i1 @allones_v2i64_and4(<2 x i64> %arg) {
 ; KNL-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,4]
 ; KNL-NEXT:    vptestmq %zmm1, %zmm0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    andb $3, %al
-; KNL-NEXT:    cmpb $3, %al
+; KNL-NEXT:    notb %al
+; KNL-NEXT:    testb $3, %al
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -3575,8 +3575,8 @@ define i1 @allones_v4i64_and4(<4 x i64> %arg) {
 ; KNL-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; KNL-NEXT:    vptestmq {{.*}}(%rip){1to8}, %zmm0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    andb $15, %al
-; KNL-NEXT:    cmpb $15, %al
+; KNL-NEXT:    notb %al
+; KNL-NEXT:    testb $15, %al
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -3999,8 +3999,8 @@ define i1 @movmsk_v8i16(<8 x i16> %x, <8 x i16> %y) {
 ; SSE-NEXT:    pcmpgtw %xmm1, %xmm0
 ; SSE-NEXT:    packsswb %xmm0, %xmm0
 ; SSE-NEXT:    pmovmskb %xmm0, %eax
-; SSE-NEXT:    andb $-109, %al
-; SSE-NEXT:    cmpb $-109, %al
+; SSE-NEXT:    notb %al
+; SSE-NEXT:    testb $-109, %al
 ; SSE-NEXT:    sete %al
 ; SSE-NEXT:    retq
 ;
@@ -4009,8 +4009,8 @@ define i1 @movmsk_v8i16(<8 x i16> %x, <8 x i16> %y) {
 ; AVX1OR2-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0
 ; AVX1OR2-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
 ; AVX1OR2-NEXT:    vpmovmskb %xmm0, %eax
-; AVX1OR2-NEXT:    andb $-109, %al
-; AVX1OR2-NEXT:    cmpb $-109, %al
+; AVX1OR2-NEXT:    notb %al
+; AVX1OR2-NEXT:    testb $-109, %al
 ; AVX1OR2-NEXT:    sete %al
 ; AVX1OR2-NEXT:    retq
 ;
@@ -4020,8 +4020,8 @@ define i1 @movmsk_v8i16(<8 x i16> %x, <8 x i16> %y) {
 ; KNL-NEXT:    vpmovsxwq %xmm0, %zmm0
 ; KNL-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    andb $-109, %al
-; KNL-NEXT:    cmpb $-109, %al
+; KNL-NEXT:    notb %al
+; KNL-NEXT:    testb $-109, %al
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -4030,8 +4030,8 @@ define i1 @movmsk_v8i16(<8 x i16> %x, <8 x i16> %y) {
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpcmpgtw %xmm1, %xmm0, %k0
 ; SKX-NEXT:    kmovd %k0, %eax
-; SKX-NEXT:    andb $-109, %al
-; SKX-NEXT:    cmpb $-109, %al
+; SKX-NEXT:    notb %al
+; SKX-NEXT:    testb $-109, %al
 ; SKX-NEXT:    sete %al
 ; SKX-NEXT:    retq
   %cmp = icmp sgt <8 x i16> %x, %y
@@ -4138,8 +4138,8 @@ define i1 @movmsk_and_v2i64(<2 x i64> %x, <2 x i64> %y) {
 ; KNL-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; KNL-NEXT:    vpcmpneqq %zmm1, %zmm0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    andb $3, %al
-; KNL-NEXT:    cmpb $3, %al
+; KNL-NEXT:    notb %al
+; KNL-NEXT:    testb $3, %al
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -4279,8 +4279,8 @@ define i1 @movmsk_and_v2f64(<2 x double> %x, <2 x double> %y) {
 ; KNL-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; KNL-NEXT:    vcmplepd %zmm0, %zmm1, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    andb $3, %al
-; KNL-NEXT:    cmpb $3, %al
+; KNL-NEXT:    notb %al
+; KNL-NEXT:    testb $3, %al
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
diff --git a/llvm/test/CodeGen/X86/pr27202.ll b/llvm/test/CodeGen/X86/pr27202.ll
index 2496a4bae258..3bd3be62fb4c 100644
--- a/llvm/test/CodeGen/X86/pr27202.ll
+++ b/llvm/test/CodeGen/X86/pr27202.ll
@@ -4,9 +4,8 @@
 define i1 @foo(i32 %i) optsize {
 ; CHECK-LABEL: foo:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    movl $305419896, %eax # imm = 0x12345678
-; CHECK-NEXT:    andl %eax, %edi
-; CHECK-NEXT:    cmpl %eax, %edi
+; CHECK-NEXT:    notl %edi
+; CHECK-NEXT:    testl $305419896, %edi # imm = 0x12345678
 ; CHECK-NEXT:    sete %al
 ; CHECK-NEXT:    retq
   %and = and i32 %i, 305419896
@@ -17,9 +16,8 @@ define i1 @foo(i32 %i) optsize {
 define i1 @foo_pgso(i32 %i) !prof !14 {
 ; CHECK-LABEL: foo_pgso:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    movl $305419896, %eax # imm = 0x12345678
-; CHECK-NEXT:    andl %eax, %edi
-; CHECK-NEXT:    cmpl %eax, %edi
+; CHECK-NEXT:    notl %edi
+; CHECK-NEXT:    testl $305419896, %edi # imm = 0x12345678
 ; CHECK-NEXT:    sete %al
 ; CHECK-NEXT:    retq
   %and = and i32 %i, 305419896
diff --git a/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll b/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
index e558660e4de8..4e0410f97346 100644
--- a/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
+++ b/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
@@ -33,8 +33,8 @@ define i1 @trunc_v2i64_v2i1(<2 x i64>) {
 ; AVX512F-NEXT:    vpsllq $63, %xmm0, %xmm0
 ; AVX512F-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; AVX512F-NEXT:    kmovw %k0, %eax
-; AVX512F-NEXT:    andb $3, %al
-; AVX512F-NEXT:    cmpb $3, %al
+; AVX512F-NEXT:    notb %al
+; AVX512F-NEXT:    testb $3, %al
 ; AVX512F-NEXT:    sete %al
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
@@ -44,8 +44,8 @@ define i1 @trunc_v2i64_v2i1(<2 x i64>) {
 ; AVX512BW-NEXT:    vpsllq $63, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    andb $3, %al
-; AVX512BW-NEXT:    cmpb $3, %al
+; AVX512BW-NEXT:    notb %al
+; AVX512BW-NEXT:    testb $3, %al
 ; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
@@ -55,7 +55,8 @@ define i1 @trunc_v2i64_v2i1(<2 x i64>) {
 ; AVX512VL-NEXT:    vpsllq $63, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vptestmq %xmm0, %xmm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    cmpb $3, %al
+; AVX512VL-NEXT:    notb %al
+; AVX512VL-NEXT:    testb $3, %al
 ; AVX512VL-NEXT:    sete %al
 ; AVX512VL-NEXT:    retq
   %a = trunc <2 x i64> %0 to <2 x i1>
@@ -85,8 +86,8 @@ define i1 @trunc_v4i32_v4i1(<4 x i32>) {
 ; AVX512F-NEXT:    vpslld $31, %xmm0, %xmm0
 ; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; AVX512F-NEXT:    kmovw %k0, %eax
-; AVX512F-NEXT:    andb $15, %al
-; AVX512F-NEXT:    cmpb $15, %al
+; AVX512F-NEXT:    notb %al
+; AVX512F-NEXT:    testb $15, %al
 ; AVX512F-NEXT:    sete %al
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
@@ -96,8 +97,8 @@ define i1 @trunc_v4i32_v4i1(<4 x i32>) {
 ; AVX512BW-NEXT:    vpslld $31, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    andb $15, %al
-; AVX512BW-NEXT:    cmpb $15, %al
+; AVX512BW-NEXT:    notb %al
+; AVX512BW-NEXT:    testb $15, %al
 ; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
@@ -107,7 +108,8 @@ define i1 @trunc_v4i32_v4i1(<4 x i32>) {
 ; AVX512VL-NEXT:    vpslld $31, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vptestmd %xmm0, %xmm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    cmpb $15, %al
+; AVX512VL-NEXT:    notb %al
+; AVX512VL-NEXT:    testb $15, %al
 ; AVX512VL-NEXT:    sete %al
 ; AVX512VL-NEXT:    retq
   %a = trunc <4 x i32> %0 to <4 x i1>
@@ -244,8 +246,8 @@ define i1 @trunc_v4i64_v4i1(<4 x i64>) {
 ; AVX512F-NEXT:    vpsllq $63, %ymm0, %ymm0
 ; AVX512F-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; AVX512F-NEXT:    kmovw %k0, %eax
-; AVX512F-NEXT:    andb $15, %al
-; AVX512F-NEXT:    cmpb $15, %al
+; AVX512F-NEXT:    notb %al
+; AVX512F-NEXT:    testb $15, %al
 ; AVX512F-NEXT:    sete %al
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
@@ -255,8 +257,8 @@ define i1 @trunc_v4i64_v4i1(<4 x i64>) {
 ; AVX512BW-NEXT:    vpsllq $63, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    andb $15, %al
-; AVX512BW-NEXT:    cmpb $15, %al
+; AVX512BW-NEXT:    notb %al
+; AVX512BW-NEXT:    testb $15, %al
 ; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
@@ -266,7 +268,8 @@ define i1 @trunc_v4i64_v4i1(<4 x i64>) {
 ; AVX512VL-NEXT:    vpsllq $63, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vptestmq %ymm0, %ymm0, %k0
 ; AVX512VL-NEXT:    kmovd %k0, %eax
-; AVX512VL-NEXT:    cmpb $15, %al
+; AVX512VL-NEXT:    notb %al
+; AVX512VL-NEXT:    testb $15, %al
 ; AVX512VL-NEXT:    sete %al
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
@@ -874,8 +877,8 @@ define i1 @icmp_v2i64_v2i1(<2 x i64>) {
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512F-NEXT:    vptestnmq %zmm0, %zmm0, %k0
 ; AVX512F-NEXT:    kmovw %k0, %eax
-; AVX512F-NEXT:    andb $3, %al
-; AVX512F-NEXT:    cmpb $3, %al
+; AVX512F-NEXT:    notb %al
+; AVX512F-NEXT:    testb $3, %al
 ; AVX512F-NEXT:    sete %al
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
@@ -885,8 +888,8 @@ define i1 @icmp_v2i64_v2i1(<2 x i64>) {
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vptestnmq %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    andb $3, %al
-; AVX512BW-NEXT:    cmpb $3, %al
+; AVX512BW-NEXT:    notb %al
+; AVX512BW-NEXT:    testb $3, %al
 ; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
@@ -930,8 +933,8 @@ define i1 @icmp_v4i32_v4i1(<4 x i32>) {
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512F-NEXT:    vptestnmd %zmm0, %zmm0, %k0
 ; AVX512F-NEXT:    kmovw %k0, %eax
-; AVX512F-NEXT:    andb $15, %al
-; AVX512F-NEXT:    cmpb $15, %al
+; AVX512F-NEXT:    notb %al
+; AVX512F-NEXT:    testb $15, %al
 ; AVX512F-NEXT:    sete %al
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
@@ -941,8 +944,8 @@ define i1 @icmp_v4i32_v4i1(<4 x i32>) {
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vptestnmd %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    andb $15, %al
-; AVX512BW-NEXT:    cmpb $15, %al
+; AVX512BW-NEXT:    notb %al
+; AVX512BW-NEXT:    testb $15, %al
 ; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
@@ -1109,8 +1112,8 @@ define i1 @icmp_v4i64_v4i1(<4 x i64>) {
 ; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512F-NEXT:    vptestnmq %zmm0, %zmm0, %k0
 ; AVX512F-NEXT:    kmovw %k0, %eax
-; AVX512F-NEXT:    andb $15, %al
-; AVX512F-NEXT:    cmpb $15, %al
+; AVX512F-NEXT:    notb %al
+; AVX512F-NEXT:    testb $15, %al
 ; AVX512F-NEXT:    sete %al
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
@@ -1120,8 +1123,8 @@ define i1 @icmp_v4i64_v4i1(<4 x i64>) {
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512BW-NEXT:    vptestnmq %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    kmovd %k0, %eax
-; AVX512BW-NEXT:    andb $15, %al
-; AVX512BW-NEXT:    cmpb $15, %al
+; AVX512BW-NEXT:    notb %al
+; AVX512BW-NEXT:    testb $15, %al
 ; AVX512BW-NEXT:    sete %al
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq