From 2d72cefc7cf1e492384347c19cca5d094e10f8b9 Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@intel.com>
Date: Mon, 8 Jan 2018 06:53:52 +0000
Subject: [PATCH] [X86] Add patterns to allow 512-bit BWI compare instructions
 to be used for 128/256-bit compares when VLX is not available.

llvm-svn: 321984
---
 lib/Target/X86/X86ISelLowering.cpp         |   7 +-
 lib/Target/X86/X86InstrAVX512.td           |  26 +++++
 test/CodeGen/X86/x86-interleaved-access.ll | 112 ++++++++++-----------
 3 files changed, 81 insertions(+), 64 deletions(-)

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 3c19960860d..3f10c978047 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -17840,12 +17840,7 @@ static SDValue LowerVSETCC(SDValue Op, const X86Subtarget &Subtarget,
     // In AVX-512 architecture setcc returns mask with i1 elements,
     // But there is no compare instruction for i8 and i16 elements in KNL.
     // In this case use SSE compare
-    bool UseAVX512Inst =
-      (OpVT.is512BitVector() ||
-       OpVT.getScalarSizeInBits() >= 32 ||
-       (Subtarget.hasBWI() && Subtarget.hasVLX()));
-
-    if (UseAVX512Inst)
+    if (OpVT.getScalarSizeInBits() >= 32 || Subtarget.hasBWI())
       return LowerIntVSETCC_AVX512(Op, DAG);
 
     return DAG.getNode(ISD::TRUNCATE, dl, VT,
diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index caf5091dac6..b7061bedcf1 100644
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -3035,6 +3035,32 @@ let Predicates = [HasAVX512, NoVLX] in {
   defm : axv512_icmp_packed_cc_no_vlx_lowering<X86cmpmu, "VPCMPUQ", v2i64x_info, v8i64_info>;
 }
 
+let Predicates = [HasBWI, NoVLX] in {
+  defm : axv512_icmp_packed_no_vlx_lowering<X86pcmpgtm, "VPCMPGTB", v32i8x_info, v64i8_info>;
+  defm : axv512_icmp_packed_no_vlx_lowering<X86pcmpeqm, "VPCMPEQB", v32i8x_info, v64i8_info>;
+
+  defm : axv512_icmp_packed_no_vlx_lowering<X86pcmpgtm, "VPCMPGTB", v16i8x_info, v64i8_info>;
+  defm : axv512_icmp_packed_no_vlx_lowering<X86pcmpeqm, "VPCMPEQB", v16i8x_info, v64i8_info>;
+
+  defm : axv512_icmp_packed_no_vlx_lowering<X86pcmpgtm, "VPCMPGTW", v16i16x_info, v32i16_info>;
+  defm : axv512_icmp_packed_no_vlx_lowering<X86pcmpeqm, "VPCMPEQW", v16i16x_info, v32i16_info>;
+
+  defm : axv512_icmp_packed_no_vlx_lowering<X86pcmpgtm, "VPCMPGTW", v8i16x_info, v32i16_info>;
+  defm : axv512_icmp_packed_no_vlx_lowering<X86pcmpeqm, "VPCMPEQW", v8i16x_info, v32i16_info>;
+
+  defm : axv512_icmp_packed_cc_no_vlx_lowering<X86cmpm, "VPCMPB", v32i8x_info, v64i8_info>;
+  defm : axv512_icmp_packed_cc_no_vlx_lowering<X86cmpmu, "VPCMPUB", v32i8x_info, v64i8_info>;
+
+  defm : axv512_icmp_packed_cc_no_vlx_lowering<X86cmpm, "VPCMPB", v16i8x_info, v64i8_info>;
+  defm : axv512_icmp_packed_cc_no_vlx_lowering<X86cmpmu, "VPCMPUB", v16i8x_info, v64i8_info>;
+
+  defm : axv512_icmp_packed_cc_no_vlx_lowering<X86cmpm, "VPCMPW", v16i16x_info, v32i16_info>;
+  defm : axv512_icmp_packed_cc_no_vlx_lowering<X86cmpmu, "VPCMPUW", v16i16x_info, v32i16_info>;
+
+  defm : axv512_icmp_packed_cc_no_vlx_lowering<X86cmpm, "VPCMPW", v8i16x_info, v32i16_info>;
+  defm : axv512_icmp_packed_cc_no_vlx_lowering<X86cmpmu, "VPCMPUW", v8i16x_info, v32i16_info>;
+}
+
 // Mask setting all 0s or 1s
 multiclass avx512_mask_setop<RegisterClass KRC, ValueType VT, PatFrag Val> {
   let Predicates = [HasAVX512] in
diff --git a/test/CodeGen/X86/x86-interleaved-access.ll b/test/CodeGen/X86/x86-interleaved-access.ll
index e02258a788a..01480f64ac6 100644
--- a/test/CodeGen/X86/x86-interleaved-access.ll
+++ b/test/CodeGen/X86/x86-interleaved-access.ll
@@ -615,7 +615,7 @@ define <16 x i1> @interleaved_load_vf16_i8_stride4(<64 x i8>* %ptr) {
 ; AVX512-NEXT:    vpshufb %xmm5, %xmm4, %xmm6
 ; AVX512-NEXT:    vpshufb %xmm5, %xmm0, %xmm5
 ; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1]
-; AVX512-NEXT:    vpblendd {{.*#+}} xmm3 = xmm5[0,1],xmm3[2,3]
+; AVX512-NEXT:    vpblendd {{.*#+}} xmm8 = xmm5[0,1],xmm3[2,3]
 ; AVX512-NEXT:    vmovdqa {{.*#+}} xmm5 = <u,u,u,u,1,5,9,13,u,u,u,u,u,u,u,u>
 ; AVX512-NEXT:    vpshufb %xmm5, %xmm2, %xmm6
 ; AVX512-NEXT:    vpshufb %xmm5, %xmm1, %xmm5
@@ -625,16 +625,15 @@ define <16 x i1> @interleaved_load_vf16_i8_stride4(<64 x i8>* %ptr) {
 ; AVX512-NEXT:    vpshufb %xmm6, %xmm0, %xmm6
 ; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm6 = xmm6[0],xmm7[0],xmm6[1],xmm7[1]
 ; AVX512-NEXT:    vpblendd {{.*#+}} xmm5 = xmm6[0,1],xmm5[2,3]
-; AVX512-NEXT:    vpcmpeqb %xmm5, %xmm3, %xmm3
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm5 = <u,u,u,u,2,6,10,14,u,u,u,u,u,u,u,u>
-; AVX512-NEXT:    vpshufb %xmm5, %xmm2, %xmm6
-; AVX512-NEXT:    vpshufb %xmm5, %xmm1, %xmm5
-; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1]
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm6 = <2,6,10,14,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512-NEXT:    vpshufb %xmm6, %xmm4, %xmm7
-; AVX512-NEXT:    vpshufb %xmm6, %xmm0, %xmm6
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm6 = <u,u,u,u,2,6,10,14,u,u,u,u,u,u,u,u>
+; AVX512-NEXT:    vpshufb %xmm6, %xmm2, %xmm7
+; AVX512-NEXT:    vpshufb %xmm6, %xmm1, %xmm6
 ; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm6 = xmm6[0],xmm7[0],xmm6[1],xmm7[1]
-; AVX512-NEXT:    vpblendd {{.*#+}} xmm5 = xmm6[0,1],xmm5[2,3]
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm7 = <2,6,10,14,u,u,u,u,u,u,u,u,u,u,u,u>
+; AVX512-NEXT:    vpshufb %xmm7, %xmm4, %xmm3
+; AVX512-NEXT:    vpshufb %xmm7, %xmm0, %xmm7
+; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm3 = xmm7[0],xmm3[0],xmm7[1],xmm3[1]
+; AVX512-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1],xmm6[2,3]
 ; AVX512-NEXT:    vmovdqa {{.*#+}} xmm6 = <u,u,u,u,3,7,11,15,u,u,u,u,u,u,u,u>
 ; AVX512-NEXT:    vpshufb %xmm6, %xmm2, %xmm2
 ; AVX512-NEXT:    vpshufb %xmm6, %xmm1, %xmm1
@@ -644,9 +643,8 @@ define <16 x i1> @interleaved_load_vf16_i8_stride4(<64 x i8>* %ptr) {
 ; AVX512-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
 ; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1]
 ; AVX512-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
-; AVX512-NEXT:    vpcmpeqb %xmm0, %xmm5, %xmm0
-; AVX512-NEXT:    vpmovb2m %zmm3, %k0
-; AVX512-NEXT:    vpmovb2m %zmm0, %k1
+; AVX512-NEXT:    vpcmpeqb %zmm5, %zmm8, %k0
+; AVX512-NEXT:    vpcmpeqb %zmm0, %zmm3, %k1
 ; AVX512-NEXT:    kxnorw %k1, %k0, %k0
 ; AVX512-NEXT:    vpmovm2b %k0, %zmm0
 ; AVX512-NEXT:    # kill: def %xmm0 killed %xmm0 killed %zmm0
@@ -869,27 +867,27 @@ define <32 x i1> @interleaved_load_vf32_i8_stride4(<128 x i8>* %ptr) {
 ; AVX512-NEXT:    vmovdqa64 (%rdi), %zmm0
 ; AVX512-NEXT:    vmovdqa64 64(%rdi), %zmm7
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512-NEXT:    vextracti128 $1, %ymm1, %xmm9
+; AVX512-NEXT:    vextracti128 $1, %ymm1, %xmm10
 ; AVX512-NEXT:    vmovdqa {{.*#+}} xmm6 = <u,u,u,u,0,4,8,12,u,u,u,u,u,u,u,u>
-; AVX512-NEXT:    vpshufb %xmm6, %xmm9, %xmm3
+; AVX512-NEXT:    vpshufb %xmm6, %xmm10, %xmm3
 ; AVX512-NEXT:    vpshufb %xmm6, %xmm1, %xmm4
 ; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
-; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm10
+; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm11
 ; AVX512-NEXT:    vmovdqa {{.*#+}} xmm2 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512-NEXT:    vpshufb %xmm2, %xmm10, %xmm5
+; AVX512-NEXT:    vpshufb %xmm2, %xmm11, %xmm5
 ; AVX512-NEXT:    vpshufb %xmm2, %xmm0, %xmm3
 ; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm3 = xmm3[0],xmm5[0],xmm3[1],xmm5[1]
 ; AVX512-NEXT:    vpblendd {{.*#+}} xmm8 = xmm3[0,1],xmm4[2,3]
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm7, %ymm5
-; AVX512-NEXT:    vextracti128 $1, %ymm5, %xmm11
-; AVX512-NEXT:    vpshufb %xmm6, %xmm11, %xmm3
-; AVX512-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,0,1]
 ; AVX512-NEXT:    vextracti128 $1, %ymm5, %xmm12
-; AVX512-NEXT:    vpshufb %xmm6, %xmm12, %xmm6
+; AVX512-NEXT:    vpshufb %xmm6, %xmm12, %xmm3
+; AVX512-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,0,1]
+; AVX512-NEXT:    vextracti128 $1, %ymm5, %xmm13
+; AVX512-NEXT:    vpshufb %xmm6, %xmm13, %xmm6
 ; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm3 = xmm6[0],xmm3[0],xmm6[1],xmm3[1]
 ; AVX512-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm3
-; AVX512-NEXT:    vextracti128 $1, %ymm7, %xmm13
-; AVX512-NEXT:    vpshufb %xmm2, %xmm13, %xmm4
+; AVX512-NEXT:    vextracti128 $1, %ymm7, %xmm14
+; AVX512-NEXT:    vpshufb %xmm2, %xmm14, %xmm4
 ; AVX512-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,0,1]
 ; AVX512-NEXT:    vextracti128 $1, %ymm7, %xmm7
 ; AVX512-NEXT:    vpshufb %xmm2, %xmm7, %xmm2
@@ -897,67 +895,65 @@ define <32 x i1> @interleaved_load_vf32_i8_stride4(<128 x i8>* %ptr) {
 ; AVX512-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
 ; AVX512-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm3[6,7]
 ; AVX512-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3],ymm2[4,5,6,7]
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm3 = <u,u,u,u,1,5,9,13,u,u,u,u,u,u,u,u>
-; AVX512-NEXT:    vpshufb %xmm3, %xmm9, %xmm4
-; AVX512-NEXT:    vpshufb %xmm3, %xmm1, %xmm2
-; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1]
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm4 = <1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512-NEXT:    vpshufb %xmm4, %xmm10, %xmm5
-; AVX512-NEXT:    vpshufb %xmm4, %xmm0, %xmm6
-; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm5 = xmm6[0],xmm5[0],xmm6[1],xmm5[1]
-; AVX512-NEXT:    vpblendd {{.*#+}} xmm2 = xmm5[0,1],xmm2[2,3]
-; AVX512-NEXT:    vpshufb %xmm3, %xmm11, %xmm5
-; AVX512-NEXT:    vpshufb %xmm3, %xmm12, %xmm3
-; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm3 = xmm3[0],xmm5[0],xmm3[1],xmm5[1]
-; AVX512-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm3
-; AVX512-NEXT:    vpshufb %xmm4, %xmm13, %xmm5
-; AVX512-NEXT:    vpshufb %xmm4, %xmm7, %xmm4
-; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
-; AVX512-NEXT:    vinserti128 $1, %xmm4, %ymm0, %ymm4
-; AVX512-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6,7]
-; AVX512-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
-; AVX512-NEXT:    vpcmpeqb %ymm2, %ymm8, %ymm8
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,u,u,u,2,6,10,14,u,u,u,u,u,u,u,u>
-; AVX512-NEXT:    vpshufb %xmm2, %xmm9, %xmm3
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,u,u,u,1,5,9,13,u,u,u,u,u,u,u,u>
+; AVX512-NEXT:    vpshufb %xmm2, %xmm10, %xmm3
 ; AVX512-NEXT:    vpshufb %xmm2, %xmm1, %xmm4
 ; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm4 = <2,6,10,14,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512-NEXT:    vpshufb %xmm4, %xmm10, %xmm5
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm4 = <1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u>
+; AVX512-NEXT:    vpshufb %xmm4, %xmm11, %xmm5
 ; AVX512-NEXT:    vpshufb %xmm4, %xmm0, %xmm6
 ; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm5 = xmm6[0],xmm5[0],xmm6[1],xmm5[1]
 ; AVX512-NEXT:    vpblendd {{.*#+}} xmm3 = xmm5[0,1],xmm3[2,3]
-; AVX512-NEXT:    vpshufb %xmm2, %xmm11, %xmm5
-; AVX512-NEXT:    vpshufb %xmm2, %xmm12, %xmm2
+; AVX512-NEXT:    vpshufb %xmm2, %xmm12, %xmm5
+; AVX512-NEXT:    vpshufb %xmm2, %xmm13, %xmm2
 ; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm2[0],xmm5[0],xmm2[1],xmm5[1]
 ; AVX512-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX512-NEXT:    vpshufb %xmm4, %xmm13, %xmm5
+; AVX512-NEXT:    vpshufb %xmm4, %xmm14, %xmm5
+; AVX512-NEXT:    vpshufb %xmm4, %xmm7, %xmm4
+; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
+; AVX512-NEXT:    vinserti128 $1, %xmm4, %ymm0, %ymm4
+; AVX512-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5],ymm2[6,7]
+; AVX512-NEXT:    vpblendd {{.*#+}} ymm9 = ymm3[0,1,2,3],ymm2[4,5,6,7]
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,u,u,u,2,6,10,14,u,u,u,u,u,u,u,u>
+; AVX512-NEXT:    vpshufb %xmm2, %xmm10, %xmm3
+; AVX512-NEXT:    vpshufb %xmm2, %xmm1, %xmm4
+; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
+; AVX512-NEXT:    vmovdqa {{.*#+}} xmm4 = <2,6,10,14,u,u,u,u,u,u,u,u,u,u,u,u>
+; AVX512-NEXT:    vpshufb %xmm4, %xmm11, %xmm5
+; AVX512-NEXT:    vpshufb %xmm4, %xmm0, %xmm6
+; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm5 = xmm6[0],xmm5[0],xmm6[1],xmm5[1]
+; AVX512-NEXT:    vpblendd {{.*#+}} xmm3 = xmm5[0,1],xmm3[2,3]
+; AVX512-NEXT:    vpshufb %xmm2, %xmm12, %xmm5
+; AVX512-NEXT:    vpshufb %xmm2, %xmm13, %xmm2
+; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm2[0],xmm5[0],xmm2[1],xmm5[1]
+; AVX512-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
+; AVX512-NEXT:    vpshufb %xmm4, %xmm14, %xmm5
 ; AVX512-NEXT:    vpshufb %xmm4, %xmm7, %xmm4
 ; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
 ; AVX512-NEXT:    vinserti128 $1, %xmm4, %ymm0, %ymm4
 ; AVX512-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5],ymm2[6,7]
 ; AVX512-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
 ; AVX512-NEXT:    vmovdqa {{.*#+}} xmm3 = <u,u,u,u,3,7,11,15,u,u,u,u,u,u,u,u>
-; AVX512-NEXT:    vpshufb %xmm3, %xmm9, %xmm4
+; AVX512-NEXT:    vpshufb %xmm3, %xmm10, %xmm4
 ; AVX512-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1]
 ; AVX512-NEXT:    vmovdqa {{.*#+}} xmm4 = <3,7,11,15,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512-NEXT:    vpshufb %xmm4, %xmm10, %xmm5
+; AVX512-NEXT:    vpshufb %xmm4, %xmm11, %xmm5
 ; AVX512-NEXT:    vpshufb %xmm4, %xmm0, %xmm0
 ; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm5[0],xmm0[1],xmm5[1]
 ; AVX512-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
-; AVX512-NEXT:    vpshufb %xmm3, %xmm11, %xmm1
-; AVX512-NEXT:    vpshufb %xmm3, %xmm12, %xmm3
+; AVX512-NEXT:    vpshufb %xmm3, %xmm12, %xmm1
+; AVX512-NEXT:    vpshufb %xmm3, %xmm13, %xmm3
 ; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
 ; AVX512-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX512-NEXT:    vpshufb %xmm4, %xmm13, %xmm3
+; AVX512-NEXT:    vpshufb %xmm4, %xmm14, %xmm3
 ; AVX512-NEXT:    vpshufb %xmm4, %xmm7, %xmm4
 ; AVX512-NEXT:    vpunpckldq {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
 ; AVX512-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm3
 ; AVX512-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5],ymm1[6,7]
 ; AVX512-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
-; AVX512-NEXT:    vpcmpeqb %ymm0, %ymm2, %ymm0
-; AVX512-NEXT:    vpmovb2m %zmm8, %k0
-; AVX512-NEXT:    vpmovb2m %zmm0, %k1
+; AVX512-NEXT:    vpcmpeqb %zmm9, %zmm8, %k0
+; AVX512-NEXT:    vpcmpeqb %zmm0, %zmm2, %k1
 ; AVX512-NEXT:    kxnord %k1, %k0, %k0
 ; AVX512-NEXT:    vpmovm2b %k0, %zmm0
 ; AVX512-NEXT:    # kill: def %ymm0 killed %ymm0 killed %zmm0