From: Craig Topper <craig.topper@intel.com>
Date: Tue, 5 Dec 2017 06:37:21 +0000 (+0000)
Subject: [X86] Use vector widening to support sign extend from i1 when the dest type is not... 
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=07b0498e27dc735187cd0121aaad03f8f3ca0f64;p=llvm

[X86] Use vector widening to support sign extend from i1 when the dest type is not 512-bits and vlx is not enabled.

Previously we used a wider element type and truncated. But its more efficient to keep the element type and drop unused elements.

If BWI isn't supported and we have a i16 or i8 type, we'll extend it to be i32 and still use a truncate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@319740 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 8e440743814..6512b6a7921 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -16147,7 +16147,7 @@ static  SDValue LowerZERO_EXTEND_AVX512(SDValue Op,
 
   // Widen to 512-bits if VLX is not supported.
   MVT WideVT = ExtVT;
-  if (!VT.is512BitVector() && !Subtarget.hasVLX()) {
+  if (!ExtVT.is512BitVector() && !Subtarget.hasVLX()) {
     NumElts *= 512 / ExtVT.getSizeInBits();
     InVT = MVT::getVectorVT(MVT::i1, NumElts);
     In = DAG.getNode(ISD::INSERT_SUBVECTOR, DL, InVT, DAG.getUNDEF(InVT),
@@ -18305,33 +18305,44 @@ static SDValue LowerSIGN_EXTEND_AVX512(SDValue Op,
   if (InVTElt != MVT::i1)
     return SDValue();
 
-  // SKX processor
-  if (((Subtarget.hasBWI() && VTElt.getSizeInBits() <= 16)) ||
-      ((Subtarget.hasDQI() && VTElt.getSizeInBits() >= 32)))
-    return DAG.getNode(X86ISD::VSEXT, dl, VT, In);
-
+  // Extend VT if the scalar type is v8/v16 and BWI is not supported.
   MVT ExtVT = VT;
-  if (!VT.is512BitVector() && !Subtarget.hasVLX()) {
-    ExtVT = MVT::getVectorVT(MVT::getIntegerVT(512/NumElts), NumElts);
-  } else if (VTElt == MVT::i16 || VTElt == MVT::i8) {
-    // If we don't have BWI support we need to extend 8/16-bit to 32-bit.
-    // Otherwise we end up with vselects we can't handle.
+  if (!Subtarget.hasBWI() && VTElt.getSizeInBits() <= 16)
     ExtVT = MVT::getVectorVT(MVT::i32, NumElts);
+
+  // Widen to 512-bits if VLX is not supported.
+  MVT WideVT = ExtVT;
+  if (!ExtVT.is512BitVector() && !Subtarget.hasVLX()) {
+    NumElts *= 512 / ExtVT.getSizeInBits();
+    InVT = MVT::getVectorVT(MVT::i1, NumElts);
+    In = DAG.getNode(ISD::INSERT_SUBVECTOR, dl, InVT, DAG.getUNDEF(InVT),
+                     In, DAG.getIntPtrConstant(0, dl));
+    WideVT = MVT::getVectorVT(ExtVT.getVectorElementType(), NumElts);
   }
 
   SDValue V;
-  if (Subtarget.hasDQI()) {
-    V = getExtendInVec(X86ISD::VSEXT, dl, ExtVT, In, DAG);
-    assert(!VT.is512BitVector() && "Unexpected vector type");
+  MVT WideEltVT = WideVT.getVectorElementType();
+  if ((Subtarget.hasDQI() && WideEltVT.getSizeInBits() >= 32) ||
+      (Subtarget.hasBWI() && WideEltVT.getSizeInBits() <= 16)) {
+    V = getExtendInVec(X86ISD::VSEXT, dl, WideVT, In, DAG);
   } else {
-    SDValue NegOne = getOnesVector(ExtVT, DAG, dl);
-    SDValue Zero = getZeroVector(ExtVT, Subtarget, DAG, dl);
-    V = DAG.getSelect(dl, ExtVT, In, NegOne, Zero);
-    if (ExtVT == VT)
-      return V;
+    SDValue NegOne = getOnesVector(WideVT, DAG, dl);
+    SDValue Zero = getZeroVector(WideVT, Subtarget, DAG, dl);
+    V = DAG.getSelect(dl, WideVT, In, NegOne, Zero);
   }
 
-  return DAG.getNode(X86ISD::VTRUNC, dl, VT, V);
+  // Truncate if we had to extend i16/i8 above.
+  if (VT != ExtVT) {
+    WideVT = MVT::getVectorVT(VTElt, NumElts);
+    V = DAG.getNode(X86ISD::VTRUNC, dl, WideVT, V);
+  }
+
+  // Extract back to 128/256-bit if we widened.
+  if (WideVT != VT)
+    V = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, V,
+                    DAG.getIntPtrConstant(0, dl));
+
+  return V;
 }
 
 // Lowering for SIGN_EXTEND_VECTOR_INREG and ZERO_EXTEND_VECTOR_INREG.
diff --git a/test/CodeGen/X86/avg-mask.ll b/test/CodeGen/X86/avg-mask.ll
index 4eacbdd4ccb..578d7aa7528 100644
--- a/test/CodeGen/X86/avg-mask.ll
+++ b/test/CodeGen/X86/avg-mask.ll
@@ -252,8 +252,8 @@ define <8 x i16> @avg_v8i16_mask(<8 x i16> %a, <8 x i16> %b, <8 x i16> %src, i8
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpavgw %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    kmovw %edi, %k1
-; AVX512F-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; AVX512F-NEXT:    vpmovqw %zmm1, %xmm1
+; AVX512F-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
+; AVX512F-NEXT:    vpmovdw %zmm1, %ymm1
 ; AVX512F-NEXT:    vpblendvb %xmm1, %xmm0, %xmm2, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
@@ -280,8 +280,8 @@ define <8 x i16> @avg_v8i16_maskz(<8 x i16> %a, <8 x i16> %b, i8 %mask) nounwind
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpavgw %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    kmovw %edi, %k1
-; AVX512F-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; AVX512F-NEXT:    vpmovqw %zmm1, %xmm1
+; AVX512F-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
+; AVX512F-NEXT:    vpmovdw %zmm1, %ymm1
 ; AVX512F-NEXT:    vpand %xmm0, %xmm1, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
diff --git a/test/CodeGen/X86/avx512-calling-conv.ll b/test/CodeGen/X86/avx512-calling-conv.ll
index e62367d2560..32b243d2eba 100644
--- a/test/CodeGen/X86/avx512-calling-conv.ll
+++ b/test/CodeGen/X86/avx512-calling-conv.ll
@@ -63,8 +63,9 @@ define <8 x i1> @test3(<8 x i1>%a, <8 x i1>%b) {
 ; KNL-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmq %zmm0, %zmm0, %k1
 ; KNL-NEXT:    vptestmq %zmm1, %zmm1, %k1 {%k1}
-; KNL-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; KNL-NEXT:    vpmovqw %zmm0, %xmm0
+; KNL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; KNL-NEXT:    vpmovdw %zmm0, %ymm0
+; KNL-NEXT:    ## kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test3:
@@ -85,8 +86,9 @@ define <8 x i1> @test3(<8 x i1>%a, <8 x i1>%b) {
 ; KNL_X32-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; KNL_X32-NEXT:    vptestmq %zmm0, %zmm0, %k1
 ; KNL_X32-NEXT:    vptestmq %zmm1, %zmm1, %k1 {%k1}
-; KNL_X32-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; KNL_X32-NEXT:    vpmovqw %zmm0, %xmm0
+; KNL_X32-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; KNL_X32-NEXT:    vpmovdw %zmm0, %ymm0
+; KNL_X32-NEXT:    ## kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
 ; KNL_X32-NEXT:    retl
   %c = and <8 x i1>%a, %b
   ret <8 x i1> %c
@@ -271,8 +273,9 @@ define <8 x i1> @test7a(<8 x i32>%a, <8 x i32>%b) {
 ; KNL-NEXT:    movb $85, %al
 ; KNL-NEXT:    kmovw %eax, %k1
 ; KNL-NEXT:    vptestmq %zmm0, %zmm0, %k1 {%k1}
-; KNL-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; KNL-NEXT:    vpmovqw %zmm0, %xmm0
+; KNL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; KNL-NEXT:    vpmovdw %zmm0, %ymm0
+; KNL-NEXT:    ## kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
 ; KNL-NEXT:    popq %rax
 ; KNL-NEXT:    retq
 ;
@@ -306,8 +309,9 @@ define <8 x i1> @test7a(<8 x i32>%a, <8 x i32>%b) {
 ; KNL_X32-NEXT:    movb $85, %al
 ; KNL_X32-NEXT:    kmovw %eax, %k1
 ; KNL_X32-NEXT:    vptestmq %zmm0, %zmm0, %k1 {%k1}
-; KNL_X32-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; KNL_X32-NEXT:    vpmovqw %zmm0, %xmm0
+; KNL_X32-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; KNL_X32-NEXT:    vpmovdw %zmm0, %ymm0
+; KNL_X32-NEXT:    ## kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
 ; KNL_X32-NEXT:    addl $12, %esp
 ; KNL_X32-NEXT:    retl
   %cmpRes = icmp sgt <8 x i32>%a, %b
diff --git a/test/CodeGen/X86/avx512-cvt.ll b/test/CodeGen/X86/avx512-cvt.ll
index 199d9143d9f..54342a10b95 100644
--- a/test/CodeGen/X86/avx512-cvt.ll
+++ b/test/CodeGen/X86/avx512-cvt.ll
@@ -1468,11 +1468,9 @@ define <16 x double> @sbto16f64(<16 x double> %a) {
 ; NOVLDQ-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
 ; NOVLDQ-NEXT:    vcmpltpd %zmm1, %zmm2, %k1
 ; NOVLDQ-NEXT:    vcmpltpd %zmm0, %zmm2, %k2
-; NOVLDQ-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
-; NOVLDQ-NEXT:    vpmovqd %zmm0, %ymm0
+; NOVLDQ-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
 ; NOVLDQ-NEXT:    vcvtdq2pd %ymm0, %zmm0
-; NOVLDQ-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NOVLDQ-NEXT:    vpmovqd %zmm1, %ymm1
+; NOVLDQ-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NOVLDQ-NEXT:    vcvtdq2pd %ymm1, %zmm1
 ; NOVLDQ-NEXT:    retq
 ;
@@ -1519,8 +1517,7 @@ define <8 x double> @sbto8f64(<8 x double> %a) {
 ; NOVLDQ:       # %bb.0:
 ; NOVLDQ-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
 ; NOVLDQ-NEXT:    vcmpltpd %zmm0, %zmm1, %k1
-; NOVLDQ-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NOVLDQ-NEXT:    vpmovqd %zmm0, %ymm0
+; NOVLDQ-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; NOVLDQ-NEXT:    vcvtdq2pd %ymm0, %zmm0
 ; NOVLDQ-NEXT:    retq
 ;
@@ -1559,8 +1556,7 @@ define <8 x float> @sbto8f32(<8 x float> %a) {
 ; NOVLDQ-NEXT:    # kill: %ymm0<def> %ymm0<kill> %zmm0<def>
 ; NOVLDQ-NEXT:    vxorps %xmm1, %xmm1, %xmm1
 ; NOVLDQ-NEXT:    vcmpltps %zmm0, %zmm1, %k1
-; NOVLDQ-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; NOVLDQ-NEXT:    vpmovqd %zmm0, %ymm0
+; NOVLDQ-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; NOVLDQ-NEXT:    vcvtdq2ps %ymm0, %ymm0
 ; NOVLDQ-NEXT:    retq
 ;
diff --git a/test/CodeGen/X86/avx512-mask-op.ll b/test/CodeGen/X86/avx512-mask-op.ll
index ab634d7d8d5..6feeb74e67c 100644
--- a/test/CodeGen/X86/avx512-mask-op.ll
+++ b/test/CodeGen/X86/avx512-mask-op.ll
@@ -1175,8 +1175,9 @@ define <8 x i1> @test18(i8 %a, i16 %y) {
 ; KNL-NEXT:    kshiftrw $1, %k1, %k1
 ; KNL-NEXT:    kshiftlw $7, %k0, %k0
 ; KNL-NEXT:    korw %k0, %k1, %k1
-; KNL-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; KNL-NEXT:    vpmovqw %zmm0, %xmm0
+; KNL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; KNL-NEXT:    vpmovdw %zmm0, %ymm0
+; KNL-NEXT:    ## kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
@@ -1241,8 +1242,9 @@ define <8 x i1> @test18(i8 %a, i16 %y) {
 ; AVX512DQ-NEXT:    kshiftrb $1, %k1, %k1
 ; AVX512DQ-NEXT:    kshiftlb $7, %k0, %k0
 ; AVX512DQ-NEXT:    korb %k0, %k1, %k0
-; AVX512DQ-NEXT:    vpmovm2q %k0, %zmm0
-; AVX512DQ-NEXT:    vpmovqw %zmm0, %xmm0
+; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
+; AVX512DQ-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512DQ-NEXT:    ## kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
   %b = bitcast i8 %a to <8 x i1>
@@ -2572,9 +2574,8 @@ define <4 x i16> @load_4i1(<4 x i1>* %a) {
 ; KNL:       ## %bb.0:
 ; KNL-NEXT:    movzbl (%rdi), %eax
 ; KNL-NEXT:    kmovw %eax, %k1
-; KNL-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; KNL-NEXT:    vpmovqd %zmm0, %ymm0
-; KNL-NEXT:    ## kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
+; KNL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; KNL-NEXT:    ## kill: %xmm0<def> %xmm0<kill> %zmm0<kill>
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
@@ -2588,9 +2589,8 @@ define <4 x i16> @load_4i1(<4 x i1>* %a) {
 ; AVX512BW:       ## %bb.0:
 ; AVX512BW-NEXT:    movzbl (%rdi), %eax
 ; AVX512BW-NEXT:    kmovd %eax, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
-; AVX512BW-NEXT:    ## kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
+; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512BW-NEXT:    ## kill: %xmm0<def> %xmm0<kill> %zmm0<kill>
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
diff --git a/test/CodeGen/X86/avx512-vec-cmp.ll b/test/CodeGen/X86/avx512-vec-cmp.ll
index 887ec006956..8bec292283b 100644
--- a/test/CodeGen/X86/avx512-vec-cmp.ll
+++ b/test/CodeGen/X86/avx512-vec-cmp.ll
@@ -874,8 +874,8 @@ define <8 x i32>@test28(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1, <8 x i64> %y1
 ; KNL-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
 ; KNL-NEXT:    vpcmpgtq %zmm3, %zmm2, %k1
 ; KNL-NEXT:    kxnorw %k1, %k0, %k1
-; KNL-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; KNL-NEXT:    vpmovqd %zmm0, %ymm0
+; KNL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; KNL-NEXT:    ## kill: %ymm0<def> %ymm0<kill> %zmm0<kill>
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test28:
diff --git a/test/CodeGen/X86/avx512vl-vec-masked-cmp.ll b/test/CodeGen/X86/avx512vl-vec-masked-cmp.ll
index 53e8a5c0644..7969a9ff1df 100644
--- a/test/CodeGen/X86/avx512vl-vec-masked-cmp.ll
+++ b/test/CodeGen/X86/avx512vl-vec-masked-cmp.ll
@@ -50208,8 +50208,7 @@ define zeroext i8 @test_masked_vcmpoeqps_v4i1_v8i1_mask(i4 zeroext %__u, <2 x i6
 ; NoVLX-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm2, %ymm2
+; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
 ; NoVLX-NEXT:    vcmpeqps %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vandps %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpextrb $4, %xmm0, %eax
@@ -50274,8 +50273,7 @@ define zeroext i8 @test_masked_vcmpoeqps_v4i1_v8i1_mask_mem(i4 zeroext %__u, <2
 ; NoVLX-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm1, %ymm1
+; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NoVLX-NEXT:    vcmpeqps (%rsi), %xmm0, %xmm0
 ; NoVLX-NEXT:    vandps %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpextrb $4, %xmm0, %eax
@@ -50341,8 +50339,7 @@ define zeroext i8 @test_masked_vcmpoeqps_v4i1_v8i1_mask_mem_b(i4 zeroext %__u, <
 ; NoVLX-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm1, %ymm1
+; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NoVLX-NEXT:    vbroadcastss (%rsi), %xmm2
 ; NoVLX-NEXT:    vcmpeqps %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vandps %xmm1, %xmm0, %xmm0
@@ -50581,8 +50578,7 @@ define zeroext i16 @test_masked_vcmpoeqps_v4i1_v16i1_mask(i4 zeroext %__u, <2 x
 ; NoVLX-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm2, %ymm2
+; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
 ; NoVLX-NEXT:    vcmpeqps %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vandps %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpextrb $4, %xmm0, %eax
@@ -50646,8 +50642,7 @@ define zeroext i16 @test_masked_vcmpoeqps_v4i1_v16i1_mask_mem(i4 zeroext %__u, <
 ; NoVLX-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm1, %ymm1
+; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NoVLX-NEXT:    vcmpeqps (%rsi), %xmm0, %xmm0
 ; NoVLX-NEXT:    vandps %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpextrb $4, %xmm0, %eax
@@ -50712,8 +50707,7 @@ define zeroext i16 @test_masked_vcmpoeqps_v4i1_v16i1_mask_mem_b(i4 zeroext %__u,
 ; NoVLX-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm1, %ymm1
+; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NoVLX-NEXT:    vbroadcastss (%rsi), %xmm2
 ; NoVLX-NEXT:    vcmpeqps %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vandps %xmm1, %xmm0, %xmm0
@@ -50915,8 +50909,7 @@ define zeroext i32 @test_masked_vcmpoeqps_v4i1_v32i1_mask(i4 zeroext %__u, <2 x
 ; NoVLX-NEXT:    movb %dil, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm2, %ymm2
+; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
 ; NoVLX-NEXT:    vcmpeqps %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vandps %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,4,8,12],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
@@ -50966,8 +50959,7 @@ define zeroext i32 @test_masked_vcmpoeqps_v4i1_v32i1_mask_mem(i4 zeroext %__u, <
 ; NoVLX-NEXT:    movb %dil, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm1, %ymm1
+; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NoVLX-NEXT:    vcmpeqps (%rsi), %xmm0, %xmm0
 ; NoVLX-NEXT:    vandps %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,4,8,12],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
@@ -51018,8 +51010,7 @@ define zeroext i32 @test_masked_vcmpoeqps_v4i1_v32i1_mask_mem_b(i4 zeroext %__u,
 ; NoVLX-NEXT:    movb %dil, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm1, %ymm1
+; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NoVLX-NEXT:    vbroadcastss (%rsi), %xmm2
 ; NoVLX-NEXT:    vcmpeqps %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vandps %xmm1, %xmm0, %xmm0
@@ -51219,8 +51210,7 @@ define zeroext i64 @test_masked_vcmpoeqps_v4i1_v64i1_mask(i4 zeroext %__u, <2 x
 ; NoVLX-NEXT:    movb %dil, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm2, %ymm2
+; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
 ; NoVLX-NEXT:    vcmpeqps %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    vandps %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    kxorw %k0, %k0, %k0
@@ -51276,8 +51266,7 @@ define zeroext i64 @test_masked_vcmpoeqps_v4i1_v64i1_mask_mem(i4 zeroext %__u, <
 ; NoVLX-NEXT:    movb %dil, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm1, %ymm1
+; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NoVLX-NEXT:    vcmpeqps (%rsi), %xmm0, %xmm0
 ; NoVLX-NEXT:    vandps %xmm1, %xmm0, %xmm0
 ; NoVLX-NEXT:    kxorw %k0, %k0, %k0
@@ -51334,8 +51323,7 @@ define zeroext i64 @test_masked_vcmpoeqps_v4i1_v64i1_mask_mem_b(i4 zeroext %__u,
 ; NoVLX-NEXT:    movb %dil, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm1, %ymm1
+; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NoVLX-NEXT:    vbroadcastss (%rsi), %xmm2
 ; NoVLX-NEXT:    vcmpeqps %xmm2, %xmm0, %xmm0
 ; NoVLX-NEXT:    vandps %xmm1, %xmm0, %xmm0
@@ -55620,8 +55608,7 @@ define zeroext i8 @test_masked_vcmpoeqpd_v4i1_v8i1_mask(i4 zeroext %__u, <4 x i6
 ; NoVLX-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm2, %ymm2
+; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
 ; NoVLX-NEXT:    vcmpeqpd %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovqd %zmm0, %ymm0
 ; NoVLX-NEXT:    vpand %xmm2, %xmm0, %xmm0
@@ -55688,8 +55675,7 @@ define zeroext i8 @test_masked_vcmpoeqpd_v4i1_v8i1_mask_mem(i4 zeroext %__u, <4
 ; NoVLX-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm1, %ymm1
+; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NoVLX-NEXT:    vcmpeqpd (%rsi), %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovqd %zmm0, %ymm0
 ; NoVLX-NEXT:    vpand %xmm1, %xmm0, %xmm0
@@ -55757,8 +55743,7 @@ define zeroext i8 @test_masked_vcmpoeqpd_v4i1_v8i1_mask_mem_b(i4 zeroext %__u, <
 ; NoVLX-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm1, %ymm1
+; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NoVLX-NEXT:    vbroadcastsd (%rsi), %ymm2
 ; NoVLX-NEXT:    vcmpeqpd %ymm2, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovqd %zmm0, %ymm0
@@ -56005,8 +55990,7 @@ define zeroext i16 @test_masked_vcmpoeqpd_v4i1_v16i1_mask(i4 zeroext %__u, <4 x
 ; NoVLX-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm2, %ymm2
+; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
 ; NoVLX-NEXT:    vcmpeqpd %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovqd %zmm0, %ymm0
 ; NoVLX-NEXT:    vpand %xmm2, %xmm0, %xmm0
@@ -56072,8 +56056,7 @@ define zeroext i16 @test_masked_vcmpoeqpd_v4i1_v16i1_mask_mem(i4 zeroext %__u, <
 ; NoVLX-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm1, %ymm1
+; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NoVLX-NEXT:    vcmpeqpd (%rsi), %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovqd %zmm0, %ymm0
 ; NoVLX-NEXT:    vpand %xmm1, %xmm0, %xmm0
@@ -56140,8 +56123,7 @@ define zeroext i16 @test_masked_vcmpoeqpd_v4i1_v16i1_mask_mem_b(i4 zeroext %__u,
 ; NoVLX-NEXT:    movb %dil, -{{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm1, %ymm1
+; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NoVLX-NEXT:    vbroadcastsd (%rsi), %ymm2
 ; NoVLX-NEXT:    vcmpeqpd %ymm2, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovqd %zmm0, %ymm0
@@ -56351,8 +56333,7 @@ define zeroext i32 @test_masked_vcmpoeqpd_v4i1_v32i1_mask(i4 zeroext %__u, <4 x
 ; NoVLX-NEXT:    movb %dil, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm2, %ymm2
+; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
 ; NoVLX-NEXT:    vcmpeqpd %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovqd %zmm0, %ymm0
 ; NoVLX-NEXT:    vpand %xmm2, %xmm0, %xmm0
@@ -56404,8 +56385,7 @@ define zeroext i32 @test_masked_vcmpoeqpd_v4i1_v32i1_mask_mem(i4 zeroext %__u, <
 ; NoVLX-NEXT:    movb %dil, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm1, %ymm1
+; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NoVLX-NEXT:    vcmpeqpd (%rsi), %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovqd %zmm0, %ymm0
 ; NoVLX-NEXT:    vpand %xmm1, %xmm0, %xmm0
@@ -56458,8 +56438,7 @@ define zeroext i32 @test_masked_vcmpoeqpd_v4i1_v32i1_mask_mem_b(i4 zeroext %__u,
 ; NoVLX-NEXT:    movb %dil, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm1, %ymm1
+; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NoVLX-NEXT:    vbroadcastsd (%rsi), %ymm2
 ; NoVLX-NEXT:    vcmpeqpd %ymm2, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovqd %zmm0, %ymm0
@@ -56667,8 +56646,7 @@ define zeroext i64 @test_masked_vcmpoeqpd_v4i1_v64i1_mask(i4 zeroext %__u, <4 x
 ; NoVLX-NEXT:    movb %dil, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm2, %ymm2
+; NoVLX-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
 ; NoVLX-NEXT:    vcmpeqpd %ymm1, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovqd %zmm0, %ymm0
 ; NoVLX-NEXT:    vpand %xmm2, %xmm0, %xmm0
@@ -56726,8 +56704,7 @@ define zeroext i64 @test_masked_vcmpoeqpd_v4i1_v64i1_mask_mem(i4 zeroext %__u, <
 ; NoVLX-NEXT:    movb %dil, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm1, %ymm1
+; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NoVLX-NEXT:    vcmpeqpd (%rsi), %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovqd %zmm0, %ymm0
 ; NoVLX-NEXT:    vpand %xmm1, %xmm0, %xmm0
@@ -56786,8 +56763,7 @@ define zeroext i64 @test_masked_vcmpoeqpd_v4i1_v64i1_mask_mem_b(i4 zeroext %__u,
 ; NoVLX-NEXT:    movb %dil, {{[0-9]+}}(%rsp)
 ; NoVLX-NEXT:    movzbl {{[0-9]+}}(%rsp), %eax
 ; NoVLX-NEXT:    kmovw %eax, %k1
-; NoVLX-NEXT:    vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; NoVLX-NEXT:    vpmovqd %zmm1, %ymm1
+; NoVLX-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
 ; NoVLX-NEXT:    vbroadcastsd (%rsi), %ymm2
 ; NoVLX-NEXT:    vcmpeqpd %ymm2, %ymm0, %ymm0
 ; NoVLX-NEXT:    vpmovqd %zmm0, %ymm0
diff --git a/test/CodeGen/X86/setcc-lowering.ll b/test/CodeGen/X86/setcc-lowering.ll
index 67e497aee0b..c422862ac14 100644
--- a/test/CodeGen/X86/setcc-lowering.ll
+++ b/test/CodeGen/X86/setcc-lowering.ll
@@ -29,8 +29,9 @@ define <8 x i16> @pr25080(<8 x i32> %a) {
 ; KNL-32-NEXT:    movb $15, %al
 ; KNL-32-NEXT:    kmovw %eax, %k1
 ; KNL-32-NEXT:    korw %k1, %k0, %k1
-; KNL-32-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; KNL-32-NEXT:    vpmovqw %zmm0, %xmm0
+; KNL-32-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; KNL-32-NEXT:    vpmovdw %zmm0, %ymm0
+; KNL-32-NEXT:    # kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
 ; KNL-32-NEXT:    retl
 entry:
   %0 = trunc <8 x i32> %a to <8 x i23>
diff --git a/test/CodeGen/X86/sse-fsignum.ll b/test/CodeGen/X86/sse-fsignum.ll
index 6712c0dccc5..0b6c205fd26 100644
--- a/test/CodeGen/X86/sse-fsignum.ll
+++ b/test/CodeGen/X86/sse-fsignum.ll
@@ -93,12 +93,10 @@ define void @signum32b(<8 x float>*) {
 ; AVX512F-NEXT:    vmovaps (%rdi), %ymm0
 ; AVX512F-NEXT:    vxorps %xmm1, %xmm1, %xmm1
 ; AVX512F-NEXT:    vcmpltps %zmm1, %zmm0, %k1
-; AVX512F-NEXT:    vpternlogq $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; AVX512F-NEXT:    vpmovqd %zmm2, %ymm2
+; AVX512F-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
 ; AVX512F-NEXT:    vcvtdq2ps %ymm2, %ymm2
 ; AVX512F-NEXT:    vcmpltps %zmm0, %zmm1, %k1
-; AVX512F-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512F-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
 ; AVX512F-NEXT:    vcvtdq2ps %ymm0, %ymm0
 ; AVX512F-NEXT:    vsubps %ymm0, %ymm2, %ymm0
 ; AVX512F-NEXT:    vmovaps %ymm0, (%rdi)
diff --git a/test/CodeGen/X86/vector-compare-results.ll b/test/CodeGen/X86/vector-compare-results.ll
index 5ceb4b1cb88..6ac0c7b3d33 100644
--- a/test/CodeGen/X86/vector-compare-results.ll
+++ b/test/CodeGen/X86/vector-compare-results.ll
@@ -615,16 +615,18 @@ define <8 x i1> @test_cmp_v8f64(<8 x double> %a0, <8 x double> %a1) nounwind {
 ; AVX512F-LABEL: test_cmp_v8f64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vcmpltpd %zmm0, %zmm1, %k1
-; AVX512F-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512F-NEXT:    vpmovqw %zmm0, %xmm0
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    # kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_cmp_v8f64:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vcmpltpd %zmm0, %zmm1, %k0
-; AVX512DQ-NEXT:    vpmovm2q %k0, %zmm0
-; AVX512DQ-NEXT:    vpmovqw %zmm0, %xmm0
+; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
+; AVX512DQ-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512DQ-NEXT:    # kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
 ;
@@ -792,16 +794,18 @@ define <8 x i1> @test_cmp_v8i64(<8 x i64> %a0, <8 x i64> %a1) nounwind {
 ; AVX512F-LABEL: test_cmp_v8i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpcmpgtq %zmm1, %zmm0, %k1
-; AVX512F-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512F-NEXT:    vpmovqw %zmm0, %xmm0
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    # kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_cmp_v8i64:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
-; AVX512DQ-NEXT:    vpmovm2q %k0, %zmm0
-; AVX512DQ-NEXT:    vpmovqw %zmm0, %xmm0
+; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
+; AVX512DQ-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512DQ-NEXT:    # kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
 ;
diff --git a/test/CodeGen/X86/vector-sext.ll b/test/CodeGen/X86/vector-sext.ll
index 88fc588d27e..24444ecdc19 100644
--- a/test/CodeGen/X86/vector-sext.ll
+++ b/test/CodeGen/X86/vector-sext.ll
@@ -1435,9 +1435,8 @@ define <4 x i32> @load_sext_4i1_to_4i32(<4 x i1> *%ptr) {
 ; AVX512F:       # %bb.0: # %entry
 ; AVX512F-NEXT:    movzbl (%rdi), %eax
 ; AVX512F-NEXT:    kmovw %eax, %k1
-; AVX512F-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512F-NEXT:    vpmovqd %zmm0, %ymm0
-; AVX512F-NEXT:    # kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT:    # kill: %xmm0<def> %xmm0<kill> %zmm0<kill>
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
@@ -1445,9 +1444,8 @@ define <4 x i32> @load_sext_4i1_to_4i32(<4 x i1> *%ptr) {
 ; AVX512BW:       # %bb.0: # %entry
 ; AVX512BW-NEXT:    movzbl (%rdi), %eax
 ; AVX512BW-NEXT:    kmovd %eax, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
-; AVX512BW-NEXT:    # kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
+; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512BW-NEXT:    # kill: %xmm0<def> %xmm0<kill> %zmm0<kill>
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
@@ -1999,8 +1997,9 @@ define <8 x i16> @load_sext_8i1_to_8i16(<8 x i1> *%ptr) {
 ; AVX512F:       # %bb.0: # %entry
 ; AVX512F-NEXT:    movzbl (%rdi), %eax
 ; AVX512F-NEXT:    kmovw %eax, %k1
-; AVX512F-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512F-NEXT:    vpmovqw %zmm0, %xmm0
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    # kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
@@ -2402,16 +2401,16 @@ define <8 x i32> @load_sext_8i1_to_8i32(<8 x i1> *%ptr) {
 ; AVX512F:       # %bb.0: # %entry
 ; AVX512F-NEXT:    movzbl (%rdi), %eax
 ; AVX512F-NEXT:    kmovw %eax, %k1
-; AVX512F-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512F-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT:    # kill: %ymm0<def> %ymm0<kill> %zmm0<kill>
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: load_sext_8i1_to_8i32:
 ; AVX512BW:       # %bb.0: # %entry
 ; AVX512BW-NEXT:    movzbl (%rdi), %eax
 ; AVX512BW-NEXT:    kmovd %eax, %k1
-; AVX512BW-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512BW-NEXT:    # kill: %ymm0<def> %ymm0<kill> %zmm0<kill>
 ; AVX512BW-NEXT:    retq
 ;
 ; X32-SSE41-LABEL: load_sext_8i1_to_8i32:
diff --git a/test/CodeGen/X86/vector-shuffle-v1.ll b/test/CodeGen/X86/vector-shuffle-v1.ll
index eba442e3138..b3219a33058 100644
--- a/test/CodeGen/X86/vector-shuffle-v1.ll
+++ b/test/CodeGen/X86/vector-shuffle-v1.ll
@@ -114,8 +114,9 @@ define <8 x i1> @shuf8i1_3_6_1_0_3_7_7_0(<8 x i64> %a, <8 x i64> %b, <8 x i64> %
 ; AVX512F-NEXT:    vpermq %zmm0, %zmm1, %zmm0
 ; AVX512F-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; AVX512F-NEXT:    vptestmq %zmm0, %zmm0, %k1
-; AVX512F-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512F-NEXT:    vpmovqw %zmm0, %xmm0
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    # kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
@@ -240,8 +241,9 @@ define <8 x i1> @shuf8i1_u_2_u_u_2_u_2_u(i8 %a) {
 ; AVX512F-NEXT:    vpbroadcastq %xmm0, %zmm0
 ; AVX512F-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; AVX512F-NEXT:    vptestmq %zmm0, %zmm0, %k1
-; AVX512F-NEXT:    vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512F-NEXT:    vpmovqw %zmm0, %xmm0
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    # kill: %xmm0<def> %xmm0<kill> %ymm0<kill>
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;