[X86] Use CONCAT_VECTORS instead of INSERT_SUBVECTOR for padding v4i1/v2i1 vector...

The CONCAT_VECTORS will be lowered to INSERT_SUBVECTOR later. In the modified cases this seems to be enough to trick a later DAG combine into running in a different order than allows the ANDs to be removed.

I'll admit this is a bit of a hack that happens to work, but using CONCAT_VECTORS is more consistent with other legalization code anyway.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@321611 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index b8eea1c..378f226 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -30436,9 +30436,10 @@ static SDValue combineBitcast(SDNode *N, SelectionDAG &DAG,
     if ((SrcVT == MVT::v4i1 || SrcVT == MVT::v2i1) && VT.isScalarInteger() &&
         Subtarget.hasVLX()) {
       SDLoc dl(N);
-      N0 = DAG.getNode(ISD::INSERT_SUBVECTOR, dl, MVT::v8i1,
-                       DAG.getUNDEF(MVT::v8i1), N0,
-                       DAG.getIntPtrConstant(0, dl));
+      unsigned NumConcats = 8 / SrcVT.getVectorNumElements();
+      SmallVector<SDValue, 4> Ops(NumConcats, DAG.getUNDEF(SrcVT));
+      Ops[0] = N0;
+      N0 = DAG.getNode(ISD::CONCAT_VECTORS, dl, MVT::v8i1, Ops);
       N0 = DAG.getBitcast(MVT::i8, N0);
       return DAG.getNode(ISD::TRUNCATE, dl, VT, N0);
     }

diff --git a/test/CodeGen/X86/avx512vl-vec-masked-cmp.ll b/test/CodeGen/X86/avx512vl-vec-masked-cmp.ll
index c2bd80b..758bd47 100644 (file)
--- a/test/CodeGen/X86/avx512vl-vec-masked-cmp.ll
+++ b/test/CodeGen/X86/avx512vl-vec-masked-cmp.ll
@@ -5333,9 +5333,7 @@ define zeroext i4 @test_vpcmpeqq_v2i1_v4i1_mask(<2 x i64> %__a, <2 x i64> %__b)
 ; VLX-LABEL: test_vpcmpeqq_v2i1_v4i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq %xmm1, %xmm0, %k0
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v2i1_v4i1_mask:
@@ -5362,9 +5360,7 @@ define zeroext i4 @test_vpcmpeqq_v2i1_v4i1_mask_mem(<2 x i64> %__a, <2 x i64>* %
 ; VLX-LABEL: test_vpcmpeqq_v2i1_v4i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq (%rdi), %xmm0, %k0
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v2i1_v4i1_mask_mem:
@@ -5393,9 +5389,7 @@ define zeroext i4 @test_masked_vpcmpeqq_v2i1_v4i1_mask(i8 zeroext %__u, <2 x i64
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq %xmm1, %xmm0, %k0 {%k1}
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v2i1_v4i1_mask:
@@ -5433,9 +5427,7 @@ define zeroext i4 @test_masked_vpcmpeqq_v2i1_v4i1_mask_mem(i8 zeroext %__u, <2 x
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq (%rsi), %xmm0, %k0 {%k1}
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v2i1_v4i1_mask_mem:
@@ -5474,9 +5466,7 @@ define zeroext i4 @test_vpcmpeqq_v2i1_v4i1_mask_mem_b(<2 x i64> %__a, i64* %__b)
 ; VLX-LABEL: test_vpcmpeqq_v2i1_v4i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpeqq (%rdi){1to2}, %xmm0, %k0
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpeqq_v2i1_v4i1_mask_mem_b:
@@ -5507,9 +5497,7 @@ define zeroext i4 @test_masked_vpcmpeqq_v2i1_v4i1_mask_mem_b(i8 zeroext %__u, <2
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpeqq (%rsi){1to2}, %xmm0, %k0 {%k1}
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpeqq_v2i1_v4i1_mask_mem_b:
@@ -14449,9 +14437,7 @@ define zeroext i4 @test_vpcmpsgtq_v2i1_v4i1_mask(<2 x i64> %__a, <2 x i64> %__b)
 ; VLX-LABEL: test_vpcmpsgtq_v2i1_v4i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v2i1_v4i1_mask:
@@ -14478,9 +14464,7 @@ define zeroext i4 @test_vpcmpsgtq_v2i1_v4i1_mask_mem(<2 x i64> %__a, <2 x i64>*
 ; VLX-LABEL: test_vpcmpsgtq_v2i1_v4i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq (%rdi), %xmm0, %k0
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v2i1_v4i1_mask_mem:
@@ -14509,9 +14493,7 @@ define zeroext i4 @test_masked_vpcmpsgtq_v2i1_v4i1_mask(i8 zeroext %__u, <2 x i6
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0 {%k1}
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v2i1_v4i1_mask:
@@ -14549,9 +14531,7 @@ define zeroext i4 @test_masked_vpcmpsgtq_v2i1_v4i1_mask_mem(i8 zeroext %__u, <2
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq (%rsi), %xmm0, %k0 {%k1}
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v2i1_v4i1_mask_mem:
@@ -14590,9 +14570,7 @@ define zeroext i4 @test_vpcmpsgtq_v2i1_v4i1_mask_mem_b(<2 x i64> %__a, i64* %__b
 ; VLX-LABEL: test_vpcmpsgtq_v2i1_v4i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpgtq (%rdi){1to2}, %xmm0, %k0
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgtq_v2i1_v4i1_mask_mem_b:
@@ -14623,9 +14601,7 @@ define zeroext i4 @test_masked_vpcmpsgtq_v2i1_v4i1_mask_mem_b(i8 zeroext %__u, <
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpgtq (%rsi){1to2}, %xmm0, %k0 {%k1}
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgtq_v2i1_v4i1_mask_mem_b:
@@ -23693,9 +23669,7 @@ define zeroext i4 @test_vpcmpsgeq_v2i1_v4i1_mask(<2 x i64> %__a, <2 x i64> %__b)
 ; VLX-LABEL: test_vpcmpsgeq_v2i1_v4i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpleq %xmm0, %xmm1, %k0
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v2i1_v4i1_mask:
@@ -23724,9 +23698,7 @@ define zeroext i4 @test_vpcmpsgeq_v2i1_v4i1_mask_mem(<2 x i64> %__a, <2 x i64>*
 ; VLX-LABEL: test_vpcmpsgeq_v2i1_v4i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltq (%rdi), %xmm0, %k0
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v2i1_v4i1_mask_mem:
@@ -23758,9 +23730,7 @@ define zeroext i4 @test_masked_vpcmpsgeq_v2i1_v4i1_mask(i8 zeroext %__u, <2 x i6
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpleq %xmm0, %xmm1, %k0 {%k1}
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v2i1_v4i1_mask:
@@ -23798,9 +23768,7 @@ define zeroext i4 @test_masked_vpcmpsgeq_v2i1_v4i1_mask_mem(i8 zeroext %__u, <2
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltq (%rsi), %xmm0, %k0 {%k1}
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v2i1_v4i1_mask_mem:
@@ -23840,9 +23808,7 @@ define zeroext i4 @test_vpcmpsgeq_v2i1_v4i1_mask_mem_b(<2 x i64> %__a, i64* %__b
 ; VLX-LABEL: test_vpcmpsgeq_v2i1_v4i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpnltq (%rdi){1to2}, %xmm0, %k0
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpsgeq_v2i1_v4i1_mask_mem_b:
@@ -23875,9 +23841,7 @@ define zeroext i4 @test_masked_vpcmpsgeq_v2i1_v4i1_mask_mem_b(i8 zeroext %__u, <
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpnltq (%rsi){1to2}, %xmm0, %k0 {%k1}
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpsgeq_v2i1_v4i1_mask_mem_b:
@@ -33093,9 +33057,7 @@ define zeroext i4 @test_vpcmpultq_v2i1_v4i1_mask(<2 x i64> %__a, <2 x i64> %__b)
 ; VLX-LABEL: test_vpcmpultq_v2i1_v4i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq %xmm1, %xmm0, %k0
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v2i1_v4i1_mask:
@@ -33125,9 +33087,7 @@ define zeroext i4 @test_vpcmpultq_v2i1_v4i1_mask_mem(<2 x i64> %__a, <2 x i64>*
 ; VLX-LABEL: test_vpcmpultq_v2i1_v4i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq (%rdi), %xmm0, %k0
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v2i1_v4i1_mask_mem:
@@ -33159,9 +33119,7 @@ define zeroext i4 @test_masked_vpcmpultq_v2i1_v4i1_mask(i8 zeroext %__u, <2 x i6
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq %xmm1, %xmm0, %k0 {%k1}
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v2i1_v4i1_mask:
@@ -33202,9 +33160,7 @@ define zeroext i4 @test_masked_vpcmpultq_v2i1_v4i1_mask_mem(i8 zeroext %__u, <2
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq (%rsi), %xmm0, %k0 {%k1}
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v2i1_v4i1_mask_mem:
@@ -33246,9 +33202,7 @@ define zeroext i4 @test_vpcmpultq_v2i1_v4i1_mask_mem_b(<2 x i64> %__a, i64* %__b
 ; VLX-LABEL: test_vpcmpultq_v2i1_v4i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vpcmpltuq (%rdi){1to2}, %xmm0, %k0
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vpcmpultq_v2i1_v4i1_mask_mem_b:
@@ -33282,9 +33236,7 @@ define zeroext i4 @test_masked_vpcmpultq_v2i1_v4i1_mask_mem_b(i8 zeroext %__u, <
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vpcmpltuq (%rsi){1to2}, %xmm0, %k0 {%k1}
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vpcmpultq_v2i1_v4i1_mask_mem_b:
@@ -39979,9 +39931,7 @@ define zeroext i4 @test_vcmpoeqpd_v2i1_v4i1_mask(<2 x i64> %__a, <2 x i64> %__b)
 ; VLX-LABEL: test_vcmpoeqpd_v2i1_v4i1_mask:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd %xmm1, %xmm0, %k0
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v2i1_v4i1_mask:
@@ -40008,9 +39958,7 @@ define zeroext i4 @test_vcmpoeqpd_v2i1_v4i1_mask_mem(<2 x i64> %__a, <2 x i64>*
 ; VLX-LABEL: test_vcmpoeqpd_v2i1_v4i1_mask_mem:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd (%rdi), %xmm0, %k0
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v2i1_v4i1_mask_mem:
@@ -40038,9 +39986,7 @@ define zeroext i4 @test_vcmpoeqpd_v2i1_v4i1_mask_mem_b(<2 x i64> %__a, double* %
 ; VLX-LABEL: test_vcmpoeqpd_v2i1_v4i1_mask_mem_b:
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    vcmpeqpd (%rdi){1to2}, %xmm0, %k0
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_vcmpoeqpd_v2i1_v4i1_mask_mem_b:
@@ -40071,9 +40017,7 @@ define zeroext i4 @test_masked_vcmpoeqpd_v2i1_v4i1_mask(i2 zeroext %__u, <2 x i6
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd %xmm1, %xmm0, %k0 {%k1}
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v2i1_v4i1_mask:
@@ -40106,9 +40050,7 @@ define zeroext i4 @test_masked_vcmpoeqpd_v2i1_v4i1_mask_mem(i2 zeroext %__u, <2
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd (%rsi), %xmm0, %k0 {%k1}
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v2i1_v4i1_mask_mem:
@@ -40142,9 +40084,7 @@ define zeroext i4 @test_masked_vcmpoeqpd_v2i1_v4i1_mask_mem_b(i2 zeroext %__u, <
 ; VLX:       # %bb.0: # %entry
 ; VLX-NEXT:    kmovd %edi, %k1
 ; VLX-NEXT:    vcmpeqpd (%rsi){1to2}, %xmm0, %k0 {%k1}
-; VLX-NEXT:    kmovd %k0, %eax
-; VLX-NEXT:    andb $3, %al
-; VLX-NEXT:    movzbl %al, %eax
+; VLX-NEXT:    kmovb %k0, %eax
 ; VLX-NEXT:    retq
 ;
 ; NoVLX-LABEL: test_masked_vcmpoeqpd_v2i1_v4i1_mask_mem_b: