[AVX512] Fix extractelement i1 lowering.

[igorb]

Summary

[AVX512] Fix extractelement i1 lowering.
The previous implementation (not custom) doesn't enforce zeroing off upper bits. The assumption is that i1 PRODUCER ( truncate and extractelement) must zero all upper bits, so i1 CONSUMER instructions ( test, zext, save, etc) can be done without additional zeroing.
Make extractelement i1 lowering custom for all vector i1.

lib/Target/X86/X86ISelLowering.cpp

if (!Subtarget.useSoftFloat() && Subtarget.hasBWI()) {
     setOperationAction(ISD::CONCAT_VECTORS,     MVT::v32i16, Custom);
     setOperationAction(ISD::CONCAT_VECTORS,     MVT::v64i8, Custom);
     setOperationAction(ISD::INSERT_SUBVECTOR,   MVT::v32i1, Custom);
     setOperationAction(ISD::INSERT_SUBVECTOR,   MVT::v64i1, Custom);
     setOperationAction(ISD::INSERT_SUBVECTOR,   MVT::v32i16, Custom);
     setOperationAction(ISD::INSERT_SUBVECTOR,   MVT::v64i8, Custom);
     setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v32i16, Custom);
     setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v64i8, Custom);
+    setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v32i1,  Custom);
+    setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v64i1, Custom);
     setOperationAction(ISD::SCALAR_TO_VECTOR,   MVT::v32i16, Custom);
     setOperationAction(ISD::SCALAR_TO_VECTOR,   MVT::v64i8, Custom);
     setOperationAction(ISD::SELECT,             MVT::v32i1, Custom);
     setOperationAction(ISD::SELECT,             MVT::v64i1, Custom);
     setOperationAction(ISD::SIGN_EXTEND,        MVT::v32i8, Custom);
     setOperationAction(ISD::ZERO_EXTEND,        MVT::v32i8, Custom);
     setOperationAction(ISD::SIGN_EXTEND,        MVT::v32i16, Custom);
     setOperationAction(ISD::ZERO_EXTEND,        MVT::v32i16, Custom);
for (auto ExtType : {ISD::ZEXTLOAD, ISD::SEXTLOAD, ISD::EXTLOAD}) {
       }
     }
   }
 
   if (!Subtarget.useSoftFloat() && Subtarget.hasVLX()) {
     addRegisterClass(MVT::v4i1,   &X86::VK4RegClass);
     addRegisterClass(MVT::v2i1,   &X86::VK2RegClass);
 
-    setOperationAction(ISD::ADD,                MVT::v2i1, Expand);
-    setOperationAction(ISD::ADD,                MVT::v4i1, Expand);
-    setOperationAction(ISD::SUB,                MVT::v2i1, Expand);
-    setOperationAction(ISD::SUB,                MVT::v4i1, Expand);
-    setOperationAction(ISD::MUL,                MVT::v2i1, Expand);
-    setOperationAction(ISD::MUL,                MVT::v4i1, Expand);
-
-    setOperationAction(ISD::TRUNCATE,           MVT::v2i1, Custom);
-    setOperationAction(ISD::TRUNCATE,           MVT::v4i1, Custom);
-    setOperationAction(ISD::SETCC,              MVT::v4i1, Custom);
-    setOperationAction(ISD::SETCC,              MVT::v2i1, Custom);
-    setOperationAction(ISD::CONCAT_VECTORS,     MVT::v4i1, Custom);
+    for (auto VT : { MVT::v2i1, MVT::v4i1 }) {
+      setOperationAction(ISD::ADD,                VT, Expand);
+      setOperationAction(ISD::SUB,                VT, Expand);
+      setOperationAction(ISD::MUL,                VT, Expand);
+      setOperationAction(ISD::VSELECT,            VT, Expand);
+
+      setOperationAction(ISD::TRUNCATE,           VT, Custom);
+      setOperationAction(ISD::SETCC,              VT, Custom);
+      setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
+      setOperationAction(ISD::SELECT,             VT, Custom);
+      setOperationAction(ISD::BUILD_VECTOR,       VT, Custom);
+      setOperationAction(ISD::VECTOR_SHUFFLE,     VT, Custom);
+    }
+
     setOperationAction(ISD::CONCAT_VECTORS,     MVT::v8i1, Custom);
+    setOperationAction(ISD::CONCAT_VECTORS,     MVT::v4i1, Custom);
     setOperationAction(ISD::INSERT_SUBVECTOR,   MVT::v8i1, Custom);
     setOperationAction(ISD::INSERT_SUBVECTOR,   MVT::v4i1, Custom);
-    setOperationAction(ISD::SELECT,             MVT::v4i1, Custom);
-    setOperationAction(ISD::SELECT,             MVT::v2i1, Custom);
-    setOperationAction(ISD::BUILD_VECTOR,       MVT::v4i1, Custom);
-    setOperationAction(ISD::BUILD_VECTOR,       MVT::v2i1, Custom);
-    setOperationAction(ISD::VECTOR_SHUFFLE,     MVT::v2i1, Custom);
-    setOperationAction(ISD::VECTOR_SHUFFLE,     MVT::v4i1, Custom);
-    setOperationAction(ISD::VSELECT,            MVT::v2i1, Expand);
-    setOperationAction(ISD::VSELECT,            MVT::v4i1, Expand);
 
     for (auto VT : { MVT::v4i32, MVT::v8i32 }) {
       setOperationAction(ISD::AND, VT, Legal);
       setOperationAction(ISD::OR,  VT, Legal);
       setOperationAction(ISD::XOR, VT, Legal);
     }
 
     for (auto VT : { MVT::v2i64, MVT::v4i64 }) {
if (!isa<ConstantSDNode>(Idx)) {
     MVT ExtVT = (VecVT == MVT::v8i1 ?  MVT::v8i64 : MVT::v16i32);
     SDValue Ext = DAG.getNode(ISD::ZERO_EXTEND, dl, ExtVT, Vec);
     SDValue Elt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl,
                               ExtVT.getVectorElementType(), Ext, Idx);
     return DAG.getNode(ISD::TRUNCATE, dl, EltVT, Elt);
   }
 
   unsigned IdxVal = cast<ConstantSDNode>(Idx)->getZExtValue();
-  if (!Subtarget.hasDQI() && (VecVT.getVectorNumElements() <= 8)) {
+  if ((!Subtarget.hasDQI() && (VecVT.getVectorNumElements() == 8)) ||
+      (VecVT.getVectorNumElements() < 8)) {
     // Use kshiftlw/rw instruction.
     VecVT = MVT::v16i1;
     Vec = DAG.getNode(ISD::INSERT_SUBVECTOR, dl, VecVT,
                       DAG.getUNDEF(VecVT),
                       Vec,
                       DAG.getIntPtrConstant(0, dl));
   }
   unsigned MaxSift = VecVT.getVectorNumElements() - 1;
+  if (MaxSift - IdxVal)
     Vec = DAG.getNode(X86ISD::VSHLI, dl, VecVT, Vec,
                       DAG.getConstant(MaxSift - IdxVal, dl, MVT::i8));
   Vec = DAG.getNode(X86ISD::VSRLI, dl, VecVT, Vec,
                     DAG.getConstant(MaxSift, dl, MVT::i8));
   return DAG.getNode(X86ISD::VEXTRACT, dl, MVT::i1, Vec,
                        DAG.getIntPtrConstant(0, dl));
 }
 
 SDValue
 X86TargetLowering::LowerEXTRACT_VECTOR_ELT(SDValue Op,

lib/Target/X86/X86InstrAVX512.td

def : Pat<(i8 (bitconvert (v8i1 VK8:$src))),
               (KMOVWrk (COPY_TO_REGCLASS VK8:$src, VK16)),
               sub_8bit)>;
   def : Pat<(i32 (zext (i8 (bitconvert (v8i1 VK8:$src))))),
             (KMOVWrk (COPY_TO_REGCLASS VK8:$src, VK16))>;
   def : Pat<(i32 (anyext (i8 (bitconvert (v8i1 VK8:$src))))),
             (KMOVWrk (COPY_TO_REGCLASS VK8:$src, VK16))>;
 }
 
-let Predicates = [HasAVX512] in {
-  def : Pat<(i1 (X86Vextract VK16:$src, (iPTR 0))),
-            (COPY_TO_REGCLASS VK16:$src, VK1)>;
-  def : Pat<(i1 (X86Vextract VK8:$src, (iPTR 0))),
-            (COPY_TO_REGCLASS VK8:$src, VK1)>;
-}
-let Predicates = [HasBWI] in {
-  def : Pat<(i1 (X86Vextract VK32:$src, (iPTR 0))),
-            (COPY_TO_REGCLASS VK32:$src, VK1)>;
-  def : Pat<(i1 (X86Vextract VK64:$src, (iPTR 0))),
-            (COPY_TO_REGCLASS VK64:$src, VK1)>;
-}
+def : Pat<(i1 (X86Vextract VK64:$src, (iPTR 0))), (COPY_TO_REGCLASS VK64:$src, VK1)>;
+def : Pat<(i1 (X86Vextract VK32:$src, (iPTR 0))), (COPY_TO_REGCLASS VK32:$src, VK1)>;
+def : Pat<(i1 (X86Vextract VK16:$src, (iPTR 0))), (COPY_TO_REGCLASS VK16:$src, VK1)>;
+def : Pat<(i1 (X86Vextract VK8:$src,  (iPTR 0))), (COPY_TO_REGCLASS VK8:$src,  VK1)>;
+def : Pat<(i1 (X86Vextract VK4:$src,  (iPTR 0))), (COPY_TO_REGCLASS VK4:$src,  VK1)>;
+def : Pat<(i1 (X86Vextract VK2:$src,  (iPTR 0))), (COPY_TO_REGCLASS VK2:$src,  VK1)>;
 
 // Mask unary operation
 // - KNOT
 multiclass avx512_mask_unop<bits<8> opc, string OpcodeStr,
                             RegisterClass KRC, SDPatternOperator OpNode,
                             Predicate prd> {
   let Predicates = [prd] in
     def rr : I<opc, MRMSrcReg, (outs KRC:$dst), (ins KRC:$src),

test/CodeGen/X86/avx512-ext.ll

 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %ebx
 ; KNL-NEXT:    kshiftlw $2, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %ebp
 ; KNL-NEXT:    kshiftlw $1, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %r14d
-; KNL-NEXT:    vptestmd %zmm5, %zmm5, %k2
-; KNL-NEXT:    kshiftlw $0, %k0, %k0
+; KNL-NEXT:    vptestmd %zmm5, %zmm5, %k1
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vmovd %r15d, %xmm4
 ; KNL-NEXT:    kmovw %k0, %r15d
-; KNL-NEXT:    kshiftlw $14, %k2, %k0
+; KNL-NEXT:    kshiftlw $14, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vpinsrb $1, %ecx, %xmm4, %xmm4
 ; KNL-NEXT:    kmovw %k0, %ecx
-; KNL-NEXT:    kshiftlw $15, %k2, %k0
+; KNL-NEXT:    kshiftlw $15, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vpinsrb $2, %r12d, %xmm4, %xmm4
 ; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    kshiftlw $13, %k2, %k0
+; KNL-NEXT:    kshiftlw $13, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vpinsrb $3, %edx, %xmm4, %xmm4
 ; KNL-NEXT:    kmovw %k0, %r12d
-; KNL-NEXT:    kshiftlw $12, %k2, %k0
+; KNL-NEXT:    kshiftlw $12, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vpinsrb $4, %r13d, %xmm4, %xmm4
 ; KNL-NEXT:    kmovw %k0, %edx
-; KNL-NEXT:    kshiftlw $11, %k2, %k0
+; KNL-NEXT:    kshiftlw $11, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vpinsrb $5, -{{[0-9]+}}(%rsp), %xmm4, %xmm4 ## 4-byte Folded Reload
 ; KNL-NEXT:    kmovw %k0, %r13d
-; KNL-NEXT:    kshiftlw $10, %k2, %k0
+; KNL-NEXT:    kshiftlw $10, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vpinsrb $6, %esi, %xmm4, %xmm4
 ; KNL-NEXT:    kmovw %k0, %esi
 ; KNL-NEXT:    movl %esi, -{{[0-9]+}}(%rsp) ## 4-byte Spill
-; KNL-NEXT:    kshiftlw $9, %k2, %k0
+; KNL-NEXT:    kshiftlw $9, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vpinsrb $7, %edi, %xmm4, %xmm4
 ; KNL-NEXT:    kmovw %k0, %esi
-; KNL-NEXT:    kshiftlw $8, %k2, %k0
+; KNL-NEXT:    kshiftlw $8, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vpinsrb $8, %r8d, %xmm4, %xmm4
 ; KNL-NEXT:    kmovw %k0, %edi
-; KNL-NEXT:    kshiftlw $7, %k2, %k0
+; KNL-NEXT:    kshiftlw $7, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vpinsrb $9, %r9d, %xmm4, %xmm4
 ; KNL-NEXT:    kmovw %k0, %r8d
-; KNL-NEXT:    kshiftlw $6, %k2, %k0
+; KNL-NEXT:    kshiftlw $6, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vpinsrb $10, %r10d, %xmm4, %xmm4
 ; KNL-NEXT:    kmovw %k0, %r9d
-; KNL-NEXT:    kshiftlw $5, %k2, %k0
+; KNL-NEXT:    kshiftlw $5, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vpinsrb $11, %r11d, %xmm4, %xmm4
 ; KNL-NEXT:    kmovw %k0, %r10d
-; KNL-NEXT:    kshiftlw $4, %k2, %k0
+; KNL-NEXT:    kshiftlw $4, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vpinsrb $12, %ebx, %xmm4, %xmm4
 ; KNL-NEXT:    kmovw %k0, %ebx
-; KNL-NEXT:    kshiftlw $3, %k2, %k0
+; KNL-NEXT:    kshiftlw $3, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vpinsrb $13, %ebp, %xmm4, %xmm4
 ; KNL-NEXT:    kmovw %k0, %ebp
-; KNL-NEXT:    kshiftlw $2, %k2, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $14, %r14d, %xmm4, %xmm4
-; KNL-NEXT:    kmovw %k0, %r11d
-; KNL-NEXT:    kshiftlw $1, %k2, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $15, %r15d, %xmm4, %xmm4
-; KNL-NEXT:    kmovw %k0, %r14d
-; KNL-NEXT:    vptestmd %zmm6, %zmm6, %k1
-; KNL-NEXT:    kshiftlw $0, %k2, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vmovd %eax, %xmm5
-; KNL-NEXT:    kmovw %k0, %r15d
-; KNL-NEXT:    kshiftlw $14, %k1, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $1, %ecx, %xmm5, %xmm5
-; KNL-NEXT:    kmovw %k0, %ecx
-; KNL-NEXT:    kshiftlw $15, %k1, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $2, %r12d, %xmm5, %xmm5
-; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    kshiftlw $13, %k1, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $3, %edx, %xmm5, %xmm5
-; KNL-NEXT:    kmovw %k0, %r12d
-; KNL-NEXT:    kshiftlw $12, %k1, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $4, %r13d, %xmm5, %xmm5
-; KNL-NEXT:    kmovw %k0, %edx
-; KNL-NEXT:    kshiftlw $11, %k1, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $5, -{{[0-9]+}}(%rsp), %xmm5, %xmm5 ## 4-byte Folded Reload
-; KNL-NEXT:    kmovw %k0, %r13d
-; KNL-NEXT:    kshiftlw $10, %k1, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $6, %esi, %xmm5, %xmm5
-; KNL-NEXT:    kmovw %k0, %esi
-; KNL-NEXT:    movl %esi, -{{[0-9]+}}(%rsp) ## 4-byte Spill
-; KNL-NEXT:    kshiftlw $9, %k1, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $7, %edi, %xmm5, %xmm5
-; KNL-NEXT:    kmovw %k0, %esi
-; KNL-NEXT:    kshiftlw $8, %k1, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $8, %r8d, %xmm5, %xmm5
-; KNL-NEXT:    kmovw %k0, %edi
-; KNL-NEXT:    kshiftlw $7, %k1, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $9, %r9d, %xmm5, %xmm5
-; KNL-NEXT:    kmovw %k0, %r8d
-; KNL-NEXT:    kshiftlw $6, %k1, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $10, %r10d, %xmm5, %xmm5
-; KNL-NEXT:    kmovw %k0, %r9d
-; KNL-NEXT:    kshiftlw $5, %k1, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $11, %ebx, %xmm5, %xmm5
-; KNL-NEXT:    kmovw %k0, %ebx
-; KNL-NEXT:    kshiftlw $4, %k1, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $12, %ebp, %xmm5, %xmm5
-; KNL-NEXT:    kmovw %k0, %ebp
-; KNL-NEXT:    kshiftlw $3, %k1, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $13, %r11d, %xmm5, %xmm5
-; KNL-NEXT:    kmovw %k0, %r10d
 ; KNL-NEXT:    kshiftlw $2, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $14, %r14d, %xmm5, %xmm5
+; KNL-NEXT:    vpinsrb $14, %r14d, %xmm4, %xmm4
 ; KNL-NEXT:    kmovw %k0, %r11d
 ; KNL-NEXT:    kshiftlw $1, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
-; KNL-NEXT:    vpinsrb $15, %r15d, %xmm5, %xmm5
+; KNL-NEXT:    vpinsrb $15, %r15d, %xmm4, %xmm4
 ; KNL-NEXT:    kmovw %k0, %r14d
-; KNL-NEXT:    vptestmd %zmm7, %zmm7, %k0
-; KNL-NEXT:    kshiftlw $0, %k1, %k1
+; KNL-NEXT:    vptestmd %zmm6, %zmm6, %k0
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
-; KNL-NEXT:    vmovd %eax, %xmm6
+; KNL-NEXT:    vmovd %eax, %xmm5
 ; KNL-NEXT:    kmovw %k1, %r15d
 ; KNL-NEXT:    kshiftlw $14, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
-; KNL-NEXT:    vpinsrb $1, %ecx, %xmm6, %xmm6
+; KNL-NEXT:    vpinsrb $1, %ecx, %xmm5, %xmm5
 ; KNL-NEXT:    kmovw %k1, %ecx
 ; KNL-NEXT:    kshiftlw $15, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
-; KNL-NEXT:    vpinsrb $2, %r12d, %xmm6, %xmm6
-; KNL-NEXT:    kmovw %k1, %r12d
+; KNL-NEXT:    vpinsrb $2, %r12d, %xmm5, %xmm5
+; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    kshiftlw $13, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
-; KNL-NEXT:    vpinsrb $3, %edx, %xmm6, %xmm6
-; KNL-NEXT:    kmovw %k1, %edx
+; KNL-NEXT:    vpinsrb $3, %edx, %xmm5, %xmm5
+; KNL-NEXT:    kmovw %k1, %r12d
 ; KNL-NEXT:    kshiftlw $12, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
-; KNL-NEXT:    vpinsrb $4, %r13d, %xmm6, %xmm6
-; KNL-NEXT:    kmovw %k1, %r13d
+; KNL-NEXT:    vpinsrb $4, %r13d, %xmm5, %xmm5
+; KNL-NEXT:    kmovw %k1, %edx
 ; KNL-NEXT:    kshiftlw $11, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
-; KNL-NEXT:    vpinsrb $5, -{{[0-9]+}}(%rsp), %xmm6, %xmm6 ## 4-byte Folded Reload
-; KNL-NEXT:    kmovw %k1, %eax
+; KNL-NEXT:    vpinsrb $5, -{{[0-9]+}}(%rsp), %xmm5, %xmm5 ## 4-byte Folded Reload
+; KNL-NEXT:    kmovw %k1, %r13d
 ; KNL-NEXT:    kshiftlw $10, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
-; KNL-NEXT:    vpinsrb $6, %esi, %xmm6, %xmm6
+; KNL-NEXT:    vpinsrb $6, %esi, %xmm5, %xmm5
 ; KNL-NEXT:    kmovw %k1, %esi
+; KNL-NEXT:    movl %esi, -{{[0-9]+}}(%rsp) ## 4-byte Spill
 ; KNL-NEXT:    kshiftlw $9, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
-; KNL-NEXT:    vpinsrb $7, %edi, %xmm6, %xmm6
-; KNL-NEXT:    kmovw %k1, %edi
+; KNL-NEXT:    vpinsrb $7, %edi, %xmm5, %xmm5
+; KNL-NEXT:    kmovw %k1, %esi
 ; KNL-NEXT:    kshiftlw $8, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
-; KNL-NEXT:    vpinsrb $8, %r8d, %xmm6, %xmm6
-; KNL-NEXT:    kmovw %k1, %r8d
+; KNL-NEXT:    vpinsrb $8, %r8d, %xmm5, %xmm5
+; KNL-NEXT:    kmovw %k1, %edi
 ; KNL-NEXT:    kshiftlw $7, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
-; KNL-NEXT:    vpinsrb $9, %r9d, %xmm6, %xmm6
-; KNL-NEXT:    kmovw %k1, %r9d
+; KNL-NEXT:    vpinsrb $9, %r9d, %xmm5, %xmm5
+; KNL-NEXT:    kmovw %k1, %r8d
 ; KNL-NEXT:    kshiftlw $6, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
-; KNL-NEXT:    vpinsrb $10, %ebx, %xmm6, %xmm6
-; KNL-NEXT:    kmovw %k1, %ebx
+; KNL-NEXT:    vpinsrb $10, %r10d, %xmm5, %xmm5
+; KNL-NEXT:    kmovw %k1, %r9d
 ; KNL-NEXT:    kshiftlw $5, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
-; KNL-NEXT:    vpinsrb $11, %ebp, %xmm6, %xmm6
-; KNL-NEXT:    kmovw %k1, %ebp
+; KNL-NEXT:    vpinsrb $11, %ebx, %xmm5, %xmm5
+; KNL-NEXT:    kmovw %k1, %ebx
 ; KNL-NEXT:    kshiftlw $4, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
-; KNL-NEXT:    vpinsrb $12, %r10d, %xmm6, %xmm6
-; KNL-NEXT:    kmovw %k1, %r10d
+; KNL-NEXT:    vpinsrb $12, %ebp, %xmm5, %xmm5
+; KNL-NEXT:    kmovw %k1, %ebp
 ; KNL-NEXT:    kshiftlw $3, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
-; KNL-NEXT:    vpinsrb $13, %r11d, %xmm6, %xmm6
-; KNL-NEXT:    kmovw %k1, %r11d
+; KNL-NEXT:    vpinsrb $13, %r11d, %xmm5, %xmm5
+; KNL-NEXT:    kmovw %k1, %r10d
 ; KNL-NEXT:    kshiftlw $2, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
-; KNL-NEXT:    vpinsrb $14, %r14d, %xmm6, %xmm6
-; KNL-NEXT:    kmovw %k1, %r14d
+; KNL-NEXT:    vpinsrb $14, %r14d, %xmm5, %xmm5
+; KNL-NEXT:    kmovw %k1, %r11d
 ; KNL-NEXT:    kshiftlw $1, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
-; KNL-NEXT:    vpinsrb $15, %r15d, %xmm6, %xmm6
-; KNL-NEXT:    kmovw %k1, %r15d
-; KNL-NEXT:    kshiftlw $0, %k0, %k0
+; KNL-NEXT:    vpinsrb $15, %r15d, %xmm5, %xmm5
+; KNL-NEXT:    kmovw %k1, %r14d
+; KNL-NEXT:    vptestmd %zmm7, %zmm7, %k1
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    vmovd %eax, %xmm6
+; KNL-NEXT:    kmovw %k0, %r15d
+; KNL-NEXT:    kshiftlw $14, %k1, %k0
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    vpinsrb $1, %ecx, %xmm6, %xmm6
+; KNL-NEXT:    kmovw %k0, %ecx
+; KNL-NEXT:    kshiftlw $15, %k1, %k0
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    vpinsrb $2, %r12d, %xmm6, %xmm6
+; KNL-NEXT:    kmovw %k0, %r12d
+; KNL-NEXT:    kshiftlw $13, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    vpinsrb $3, %edx, %xmm6, %xmm6
+; KNL-NEXT:    kmovw %k0, %edx
+; KNL-NEXT:    kshiftlw $12, %k1, %k0
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    vpinsrb $4, %r13d, %xmm6, %xmm6
+; KNL-NEXT:    kmovw %k0, %r13d
+; KNL-NEXT:    kshiftlw $11, %k1, %k0
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    vpinsrb $5, -{{[0-9]+}}(%rsp), %xmm6, %xmm6 ## 4-byte Folded Reload
+; KNL-NEXT:    kmovw %k0, %eax
+; KNL-NEXT:    kshiftlw $10, %k1, %k0
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    vpinsrb $6, %esi, %xmm6, %xmm6
+; KNL-NEXT:    kmovw %k0, %esi
+; KNL-NEXT:    kshiftlw $9, %k1, %k0
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    vpinsrb $7, %edi, %xmm6, %xmm6
+; KNL-NEXT:    kmovw %k0, %edi
+; KNL-NEXT:    kshiftlw $8, %k1, %k0
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    vpinsrb $8, %r8d, %xmm6, %xmm6
+; KNL-NEXT:    kmovw %k0, %r8d
+; KNL-NEXT:    kshiftlw $7, %k1, %k0
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    vpinsrb $9, %r9d, %xmm6, %xmm6
+; KNL-NEXT:    kmovw %k0, %r9d
+; KNL-NEXT:    kshiftlw $6, %k1, %k0
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    vpinsrb $10, %ebx, %xmm6, %xmm6
+; KNL-NEXT:    kmovw %k0, %ebx
+; KNL-NEXT:    kshiftlw $5, %k1, %k0
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    vpinsrb $11, %ebp, %xmm6, %xmm6
+; KNL-NEXT:    kmovw %k0, %ebp
+; KNL-NEXT:    kshiftlw $4, %k1, %k0
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    vpinsrb $12, %r10d, %xmm6, %xmm6
+; KNL-NEXT:    kmovw %k0, %r10d
+; KNL-NEXT:    kshiftlw $3, %k1, %k0
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    vpinsrb $13, %r11d, %xmm6, %xmm6
+; KNL-NEXT:    kmovw %k0, %r11d
+; KNL-NEXT:    kshiftlw $2, %k1, %k0
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    vpinsrb $14, %r14d, %xmm6, %xmm6
+; KNL-NEXT:    kmovw %k0, %r14d
+; KNL-NEXT:    kshiftlw $1, %k1, %k0
+; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    vpinsrb $15, %r15d, %xmm6, %xmm6
+; KNL-NEXT:    kmovw %k0, %r15d
+; KNL-NEXT:    kshiftrw $15, %k1, %k0
 ; KNL-NEXT:    vmovd %r12d, %xmm7
 ; KNL-NEXT:    kmovw %k0, %r12d
 ; KNL-NEXT:    vpinsrb $1, %ecx, %xmm7, %xmm7
 ; KNL-NEXT:    vpinsrb $2, %edx, %xmm7, %xmm7
 ; KNL-NEXT:    vpinsrb $3, %r13d, %xmm7, %xmm7
 ; KNL-NEXT:    vpinsrb $4, %eax, %xmm7, %xmm7
 ; KNL-NEXT:    vpinsrb $5, %esi, %xmm7, %xmm7
 ; KNL-NEXT:    vpinsrb $6, %edi, %xmm7, %xmm7

test/CodeGen/X86/avx512-insert-extract.ll

 ; SKX:       ## BB#0:
 ; SKX-NEXT:    vextracti32x4 $1, %ymm0, %xmm1
 ; SKX-NEXT:    vpinsrb $4, %edi, %xmm1, %xmm1
 ; SKX-NEXT:    vinserti32x4 $1, %xmm1, %ymm0, %ymm0
 ; SKX-NEXT:    retq
   %r = insertelement <32 x i8> %x, i8 %y, i32 20
   ret <32 x i8> %r
 }
+
+define zeroext i8 @test_extractelement_v2i1(<2 x i64> %a, <2 x i64> %b) {
+; KNL-LABEL: test_extractelement_v2i1:
+; KNL:       ## BB#0:
+; KNL-NEXT:    vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
+; KNL-NEXT:    vpxor %xmm2, %xmm1, %xmm1
+; KNL-NEXT:    vpxor %xmm2, %xmm0, %xmm0
+; KNL-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm0
+; KNL-NEXT:    vmovq %xmm0, %rax
+; KNL-NEXT:    testb $1, %al
+; KNL-NEXT:    sete %al
+; KNL-NEXT:    addb $3, %al
+; KNL-NEXT:    movzbl %al, %eax
+; KNL-NEXT:    retq
+;
+; SKX-LABEL: test_extractelement_v2i1:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpcmpnleuq %xmm1, %xmm0, %k0
+; SKX-NEXT:    kshiftlw $15, %k0, %k0
+; SKX-NEXT:    kshiftrw $15, %k0, %k0
+; SKX-NEXT:    kmovw %k0, %eax
+; SKX-NEXT:    testb %al, %al
+; SKX-NEXT:    sete %al
+; SKX-NEXT:    addb $3, %al
+; SKX-NEXT:    movzbl %al, %eax
+; SKX-NEXT:    retq
+  %t1 = icmp ugt <2 x i64> %a, %b
+  %t2 = extractelement <2 x i1> %t1, i32 0
+  %res = select i1 %t2, i8 3, i8 4
+  ret i8 %res
+}
+
+define zeroext i8 @test_extractelement_v4i1(<4 x i32> %a, <4 x i32> %b) {
+; KNL-LABEL: test_extractelement_v4i1:
+; KNL:       ## BB#0:
+; KNL-NEXT:    vpbroadcastd {{.*}}(%rip), %xmm2
+; KNL-NEXT:    vpxor %xmm2, %xmm1, %xmm1
+; KNL-NEXT:    vpxor %xmm2, %xmm0, %xmm0
+; KNL-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
+; KNL-NEXT:    vpextrd $3, %xmm0, %eax
+; KNL-NEXT:    andl $1, %eax
+; KNL-NEXT:    retq
+;
+; SKX-LABEL: test_extractelement_v4i1:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpcmpnleud %xmm1, %xmm0, %k0
+; SKX-NEXT:    kshiftlw $12, %k0, %k0
+; SKX-NEXT:    kshiftrw $15, %k0, %k0
+; SKX-NEXT:    kmovw %k0, %eax
+; SKX-NEXT:    retq
+  %t1 = icmp ugt <4 x i32> %a, %b
+  %t2 = extractelement <4 x i1> %t1, i32 3
+  %res = zext i1 %t2 to i8
+  ret i8 %res
+}
+
+define zeroext i8 @test_extractelement_v32i1(<32 x i8> %a, <32 x i8> %b) {
+; KNL-LABEL: test_extractelement_v32i1:
+; KNL:       ## BB#0:
+; KNL-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
+; KNL-NEXT:    vpxor %ymm2, %ymm1, %ymm1
+; KNL-NEXT:    vpxor %ymm2, %ymm0, %ymm0
+; KNL-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
+; KNL-NEXT:    vpextrb $2, %xmm0, %eax
+; KNL-NEXT:    andl $1, %eax
+; KNL-NEXT:    retq
+;
+; SKX-LABEL: test_extractelement_v32i1:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpcmpnleub %ymm1, %ymm0, %k0
+; SKX-NEXT:    kshiftld $29, %k0, %k0
+; SKX-NEXT:    kshiftrd $31, %k0, %k0
+; SKX-NEXT:    kmovw %k0, %eax
+; SKX-NEXT:    retq
+  %t1 = icmp ugt <32 x i8> %a, %b
+  %t2 = extractelement <32 x i1> %t1, i32 2
+  %res = zext i1 %t2 to i8
+  ret i8 %res
+}
+
+define zeroext i8 @test_extractelement_v64i1(<64 x i8> %a, <64 x i8> %b) {
+; KNL-LABEL: test_extractelement_v64i1:
+; KNL:       ## BB#0:
+; KNL-NEXT:    vmovdqa {{.*#+}} ymm0 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
+; KNL-NEXT:    vpxor %ymm0, %ymm3, %ymm2
+; KNL-NEXT:    vpxor %ymm0, %ymm1, %ymm0
+; KNL-NEXT:    vpcmpgtb %ymm2, %ymm0, %ymm0
+; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
+; KNL-NEXT:    vpextrb $15, %xmm0, %eax
+; KNL-NEXT:    testb $1, %al
+; KNL-NEXT:    sete %al
+; KNL-NEXT:    addb $3, %al
+; KNL-NEXT:    movzbl %al, %eax
+; KNL-NEXT:    retq
+;
+; SKX-LABEL: test_extractelement_v64i1:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vpcmpnleub %zmm1, %zmm0, %k0
+; SKX-NEXT:    kshiftrq $63, %k0, %k0
+; SKX-NEXT:    kmovw %k0, %eax
+; SKX-NEXT:    testb %al, %al
+; SKX-NEXT:    sete %al
+; SKX-NEXT:    addb $3, %al
+; SKX-NEXT:    movzbl %al, %eax
+; SKX-NEXT:    retq
+  %t1 = icmp ugt <64 x i8> %a, %b
+  %t2 = extractelement <64 x i1> %t1, i32 63
+  %res = select i1 %t2, i8 3, i8 4
+  ret i8 %res
+}

test/CodeGen/X86/avx512-mask-op.ll

 ; KNL-NEXT:    kshiftlw $2, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    vpinsrb $13, %eax, %xmm3, %xmm3
 ; KNL-NEXT:    kshiftlw $1, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    vpinsrb $14, %eax, %xmm3, %xmm3
-; KNL-NEXT:    kshiftlw $0, %k1, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    kshiftrw $15, %k1, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    vpinsrb $15, %eax, %xmm3, %xmm3
 ; KNL-NEXT:    vcmpltps %zmm0, %zmm2, %k2
 ; KNL-NEXT:    kshiftlw $14, %k2, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    kshiftlw $15, %k2, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0

 ; KNL-NEXT:    kshiftlw $2, %k2, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    vpinsrb $13, %eax, %xmm2, %xmm2
 ; KNL-NEXT:    kshiftlw $1, %k2, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    vpinsrb $14, %eax, %xmm2, %xmm2
-; KNL-NEXT:    kshiftlw $0, %k2, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    kshiftrw $15, %k2, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    vpinsrb $15, %eax, %xmm2, %xmm2
 ; KNL-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
 ; KNL-NEXT:    vpsllw $7, %ymm2, %ymm2
 ; KNL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
 ; KNL-NEXT:    vpxor %ymm3, %ymm3, %ymm3
 ; KNL-NEXT:    vpcmpgtb %ymm2, %ymm3, %ymm2
 ; KNL-NEXT:    vmovups 4(%rdi), %zmm3 {%k2} {z}

 ; KNL-NEXT:    kshiftlw $2, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    vpinsrb $13, %eax, %xmm4, %xmm4
 ; KNL-NEXT:    kshiftlw $1, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    vpinsrb $14, %eax, %xmm4, %xmm4
-; KNL-NEXT:    kshiftlw $0, %k0, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    vpinsrb $15, %eax, %xmm4, %xmm4
 ; KNL-NEXT:    vcmpltps %zmm3, %zmm0, %k0
 ; KNL-NEXT:    kshiftlw $14, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    kshiftlw $15, %k0, %k1

 ; KNL-NEXT:    kshiftlw $2, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    vpinsrb $13, %eax, %xmm3, %xmm3
 ; KNL-NEXT:    kshiftlw $1, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    vpinsrb $14, %eax, %xmm3, %xmm3
-; KNL-NEXT:    kshiftlw $0, %k0, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    vpinsrb $15, %eax, %xmm3, %xmm3
 ; KNL-NEXT:    vinserti128 $1, %xmm4, %ymm3, %ymm3
 ; KNL-NEXT:    vpor %ymm3, %ymm2, %ymm2
 ; KNL-NEXT:    vextracti128 $1, %ymm2, %xmm3
 ; KNL-NEXT:    vpmovsxbd %xmm3, %zmm3
 ; KNL-NEXT:    vpslld $31, %zmm3, %zmm3

 ; KNL-NEXT:    kshiftlw $2, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %esi
 ; KNL-NEXT:    kshiftlw $1, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    vmovd %r9d, %xmm3
 ; KNL-NEXT:    kmovw %k1, %r9d
 ; KNL-NEXT:    vptestmd %zmm2, %zmm2, %k2
-; KNL-NEXT:    kshiftlw $0, %k0, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vpinsrb $1, %r8d, %xmm3, %xmm2
 ; KNL-NEXT:    vpinsrb $2, %r10d, %xmm2, %xmm2
 ; KNL-NEXT:    vpinsrb $3, %r11d, %xmm2, %xmm2
 ; KNL-NEXT:    vpinsrb $4, %r14d, %xmm2, %xmm2
 ; KNL-NEXT:    vpinsrb $5, %r15d, %xmm2, %xmm2
 ; KNL-NEXT:    vpinsrb $6, %r12d, %xmm2, %xmm2
 ; KNL-NEXT:    vpinsrb $7, %r13d, %xmm2, %xmm2

 ; KNL-NEXT:    kshiftlw $2, %k2, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %ecx
 ; KNL-NEXT:    kshiftlw $1, %k2, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vmovd %r10d, %xmm2
 ; KNL-NEXT:    kmovw %k0, %r10d
 ; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k1
-; KNL-NEXT:    kshiftlw $0, %k2, %k0
-; KNL-NEXT:    kshiftrw $15, %k0, %k0
+; KNL-NEXT:    kshiftrw $15, %k2, %k0
 ; KNL-NEXT:    vpinsrb $1, %r8d, %xmm2, %xmm1
 ; KNL-NEXT:    vpinsrb $2, %r9d, %xmm1, %xmm1
 ; KNL-NEXT:    vpinsrb $3, %r11d, %xmm1, %xmm1
 ; KNL-NEXT:    vpinsrb $4, %r14d, %xmm1, %xmm1
 ; KNL-NEXT:    vpinsrb $5, %r15d, %xmm1, %xmm1
 ; KNL-NEXT:    vpinsrb $6, %r12d, %xmm1, %xmm1
 ; KNL-NEXT:    vpinsrb $7, %r13d, %xmm1, %xmm1
 ; KNL-NEXT:    vpinsrb $8, %edx, %xmm1, %xmm1

 ; KNL-NEXT:    kshiftlw $2, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %ecx
 ; KNL-NEXT:    kshiftlw $1, %k1, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vmovd %r10d, %xmm1
 ; KNL-NEXT:    kmovw %k0, %r10d
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kshiftlw $0, %k1, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    vpinsrb $1, %r8d, %xmm1, %xmm0
 ; KNL-NEXT:    vpinsrb $2, %r9d, %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $3, %r11d, %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $4, %r14d, %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $5, %r15d, %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $6, %r12d, %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $7, %r13d, %xmm0, %xmm0

 ; KNL-NEXT:    vpinsrb $5, %r15d, %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $6, %r12d, %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $7, %r13d, %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $8, %edx, %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $9, %esi, %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $10, %ebp, %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $11, %ebx, %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $12, %eax, %xmm0, %xmm0
-; KNL-NEXT:    kshiftlw $0, %k0, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    vpinsrb $13, %ecx, %xmm0, %xmm0
 ; KNL-NEXT:    vpinsrb $14, %r9d, %xmm0, %xmm0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0

test/CodeGen/X86/avx512-vec-cmp.ll

 ; KNL-NEXT:    kshiftlw $2, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
 ; KNL-NEXT:    kshiftlw $1, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm1
-; KNL-NEXT:    kshiftlw $0, %k0, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    vpinsrb $15, %eax, %xmm1, %xmm1
 ; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
 ; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
 ; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    vpcmpeqd %zmm2, %zmm0, %k0

 ; KNL-NEXT:    kshiftlw $2, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    vpinsrb $13, %eax, %xmm0, %xmm0
 ; KNL-NEXT:    kshiftlw $1, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    vpinsrb $14, %eax, %xmm0, %xmm0
-; KNL-NEXT:    kshiftlw $0, %k0, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; KNL-NEXT:    kmovw %k0, (%rsp)
 ; KNL-NEXT:    movl (%rsp), %eax

 ; KNL-NEXT:    kshiftlw $2, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
 ; KNL-NEXT:    kshiftlw $1, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm1
-; KNL-NEXT:    kshiftlw $0, %k0, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    vpinsrb $15, %eax, %xmm1, %xmm1
 ; KNL-NEXT:    vpmovsxbd %xmm1, %zmm1
 ; KNL-NEXT:    vpslld $31, %zmm1, %zmm1
 ; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
 ; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm0

 ; KNL-NEXT:    kshiftlw $2, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    vpinsrb $13, %eax, %xmm0, %xmm0
 ; KNL-NEXT:    kshiftlw $1, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    vpinsrb $14, %eax, %xmm0, %xmm0
-; KNL-NEXT:    kshiftlw $0, %k0, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; KNL-NEXT:    kmovw %k0, (%rsp)
 ; KNL-NEXT:    vpcmpeqw %ymm7, %ymm3, %ymm0

 ; KNL-NEXT:    kshiftlw $2, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    vpinsrb $13, %eax, %xmm0, %xmm0
 ; KNL-NEXT:    kshiftlw $1, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    vpinsrb $14, %eax, %xmm0, %xmm0
-; KNL-NEXT:    kshiftlw $0, %k0, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    vpcmpeqw %ymm6, %ymm2, %ymm0

 ; KNL-NEXT:    kshiftlw $2, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    vpinsrb $13, %eax, %xmm0, %xmm0
 ; KNL-NEXT:    kshiftlw $1, %k0, %k1
 ; KNL-NEXT:    kshiftrw $15, %k1, %k1
 ; KNL-NEXT:    kmovw %k1, %eax
 ; KNL-NEXT:    vpinsrb $14, %eax, %xmm0, %xmm0
-; KNL-NEXT:    kshiftlw $0, %k0, %k0
 ; KNL-NEXT:    kshiftrw $15, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
 ; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
 ; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; KNL-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp)
 ; KNL-NEXT:    movl (%rsp), %ecx

test/CodeGen/X86/masked_gather_scatter.ll

 ; KNL_32-NEXT:    popl %esi
 ; KNL_32-NEXT:    popl %ebx
 ; KNL_32-NEXT:    retl
 ;
 ; SKX-LABEL: test30:
 ; SKX:       # BB#0:
 ; SKX-NEXT:    vpslld $31, %xmm2, %xmm2
 ; SKX-NEXT:    vptestmd %xmm2, %xmm2, %k1
-; SKX-NEXT:    kmovb %k1, -{{[0-9]+}}(%rsp)
+; SKX-NEXT:    kshiftlw $15, %k1, %k0
+; SKX-NEXT:    kshiftrw $15, %k0, %k0
 ; SKX-NEXT:    vpmovsxdq %xmm1, %ymm1
 ; SKX-NEXT:    vpsllq $2, %ymm1, %ymm1
 ; SKX-NEXT:    vpaddq %ymm1, %ymm0, %ymm1
-; SKX-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; SKX-NEXT:    kmovw %k0, %eax
 ; SKX-NEXT:    # implicit-def: %XMM0
 ; SKX-NEXT:    testb %al, %al
 ; SKX-NEXT:    je .LBB29_2
 ; SKX-NEXT:  # BB#1: # %cond.load
 ; SKX-NEXT:    vmovq %xmm1, %rax
 ; SKX-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; SKX-NEXT:  .LBB29_2: # %else
-; SKX-NEXT:    kmovb %k1, -{{[0-9]+}}(%rsp)
-; SKX-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; SKX-NEXT:    kshiftlw $14, %k1, %k0
+; SKX-NEXT:    kshiftrw $15, %k0, %k0
+; SKX-NEXT:    kmovw %k0, %eax
 ; SKX-NEXT:    testb %al, %al
 ; SKX-NEXT:    je .LBB29_4
 ; SKX-NEXT:  # BB#3: # %cond.load1
 ; SKX-NEXT:    vpextrq $1, %xmm1, %rax
 ; SKX-NEXT:    vpinsrd $1, (%rax), %xmm0, %xmm0
 ; SKX-NEXT:  .LBB29_4: # %else2
-; SKX-NEXT:    kmovb %k1, -{{[0-9]+}}(%rsp)
-; SKX-NEXT:    movb -{{[0-9]+}}(%rsp), %al
+; SKX-NEXT:    kshiftlw $13, %k1, %k0
+; SKX-NEXT:    kshiftrw $15, %k0, %k0
+; SKX-NEXT:    kmovw %k0, %eax
 ; SKX-NEXT:    testb %al, %al
 ; SKX-NEXT:    je .LBB29_6
 ; SKX-NEXT:  # BB#5: # %cond.load4
 ; SKX-NEXT:    vextracti64x2 $1, %ymm1, %xmm1
 ; SKX-NEXT:    vmovq %xmm1, %rax
 ; SKX-NEXT:    vpinsrd $2, (%rax), %xmm0, %xmm0
 ; SKX-NEXT:  .LBB29_6: # %else5
 ; SKX-NEXT:    vpblendmd %xmm0, %xmm3, %xmm0 {%k1}
 ; SKX-NEXT:    retq
 ;
 ; SKX_32-LABEL: test30:
 ; SKX_32:       # BB#0:
 ; SKX_32-NEXT:    subl $12, %esp
 ; SKX_32-NEXT:  .Ltmp0:
 ; SKX_32-NEXT:    .cfi_def_cfa_offset 16
 ; SKX_32-NEXT:    vpslld $31, %xmm2, %xmm2
 ; SKX_32-NEXT:    vptestmd %xmm2, %xmm2, %k1
-; SKX_32-NEXT:    kmovb %k1, {{[0-9]+}}(%esp)
+; SKX_32-NEXT:    kshiftlw $15, %k1, %k0
+; SKX_32-NEXT:    kshiftrw $15, %k0, %k0
 ; SKX_32-NEXT:    vpslld $2, %xmm1, %xmm1
 ; SKX_32-NEXT:    vpaddd %xmm1, %xmm0, %xmm1
-; SKX_32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; SKX_32-NEXT:    kmovw %k0, %eax
 ; SKX_32-NEXT:    # implicit-def: %XMM0
 ; SKX_32-NEXT:    testb %al, %al
 ; SKX_32-NEXT:    je .LBB29_2
 ; SKX_32-NEXT:  # BB#1: # %cond.load
 ; SKX_32-NEXT:    vmovd %xmm1, %eax
 ; SKX_32-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; SKX_32-NEXT:  .LBB29_2: # %else
-; SKX_32-NEXT:    kmovb %k1, {{[0-9]+}}(%esp)
-; SKX_32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; SKX_32-NEXT:    kshiftlw $14, %k1, %k0
+; SKX_32-NEXT:    kshiftrw $15, %k0, %k0
+; SKX_32-NEXT:    kmovw %k0, %eax
 ; SKX_32-NEXT:    testb %al, %al
 ; SKX_32-NEXT:    je .LBB29_4
 ; SKX_32-NEXT:  # BB#3: # %cond.load1
 ; SKX_32-NEXT:    vpextrd $1, %xmm1, %eax
 ; SKX_32-NEXT:    vpinsrd $1, (%eax), %xmm0, %xmm0
 ; SKX_32-NEXT:  .LBB29_4: # %else2
 ; SKX_32-NEXT:    vmovdqa32 {{[0-9]+}}(%esp), %xmm2
-; SKX_32-NEXT:    kmovb %k1, (%esp)
-; SKX_32-NEXT:    movb (%esp), %al
+; SKX_32-NEXT:    kshiftlw $13, %k1, %k0
+; SKX_32-NEXT:    kshiftrw $15, %k0, %k0
+; SKX_32-NEXT:    kmovw %k0, %eax
 ; SKX_32-NEXT:    testb %al, %al
 ; SKX_32-NEXT:    je .LBB29_6
 ; SKX_32-NEXT:  # BB#5: # %cond.load4
 ; SKX_32-NEXT:    vpextrd $2, %xmm1, %eax
 ; SKX_32-NEXT:    vpinsrd $2, (%eax), %xmm0, %xmm0
 ; SKX_32-NEXT:  .LBB29_6: # %else5
 ; SKX_32-NEXT:    vpblendmd %xmm0, %xmm2, %xmm0 {%k1}
 ; SKX_32-NEXT:    addl $12, %esp

test/CodeGen/X86/masked_memop.ll

 ; AVX512F-NEXT:    kshiftlw $1, %k1, %k0
 ; AVX512F-NEXT:    kshiftrw $15, %k0, %k0
 ; AVX512F-NEXT:    kmovw %k0, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB50_30
 ; AVX512F-NEXT:  ## BB#29: ## %cond.load40
 ; AVX512F-NEXT:    vpinsrb $14, 14(%rdi), %xmm0, %xmm0
 ; AVX512F-NEXT:  LBB50_30: ## %else41
-; AVX512F-NEXT:    kshiftlw $0, %k1, %k0
-; AVX512F-NEXT:    kshiftrw $15, %k0, %k0
+; AVX512F-NEXT:    kshiftrw $15, %k1, %k0
 ; AVX512F-NEXT:    kmovw %k0, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB50_32
 ; AVX512F-NEXT:  ## BB#31: ## %cond.load43
 ; AVX512F-NEXT:    vpinsrb $15, 15(%rdi), %xmm0, %xmm0
 ; AVX512F-NEXT:  LBB50_32: ## %else44
 ; AVX512F-NEXT:    vpternlogd $255, %zmm1, %zmm1, %zmm1
 ; AVX512F-NEXT:    vmovdqa32 %zmm1, %zmm1 {%k1} {z}

 ; AVX512F-NEXT:    kmovw %k1, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB52_30
 ; AVX512F-NEXT:  ## BB#29: ## %cond.load40
 ; AVX512F-NEXT:    vpinsrb $14, 14(%rdi), %xmm0, %xmm6
 ; AVX512F-NEXT:    vpblendd {{.*#+}} ymm0 = ymm6[0,1,2,3],ymm0[4,5,6,7]
 ; AVX512F-NEXT:  LBB52_30: ## %else41
 ; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k1
-; AVX512F-NEXT:    kshiftlw $0, %k0, %k0
 ; AVX512F-NEXT:    kshiftrw $15, %k0, %k0
 ; AVX512F-NEXT:    kmovw %k0, -{{[0-9]+}}(%rsp) ## 2-byte Spill
 ; AVX512F-NEXT:    kmovw %k0, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB52_32
 ; AVX512F-NEXT:  ## BB#31: ## %cond.load43
 ; AVX512F-NEXT:    vpinsrb $15, 15(%rdi), %xmm0, %xmm1
 ; AVX512F-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]

 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB52_62
 ; AVX512F-NEXT:  ## BB#61: ## %cond.load88
 ; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm2
 ; AVX512F-NEXT:    vpinsrb $14, 30(%rdi), %xmm2, %xmm2
 ; AVX512F-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
 ; AVX512F-NEXT:  LBB52_62: ## %else89
 ; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512F-NEXT:    kshiftlw $0, %k1, %k1
 ; AVX512F-NEXT:    kshiftrw $15, %k1, %k1
 ; AVX512F-NEXT:    kmovw %k1, -{{[0-9]+}}(%rsp) ## 2-byte Spill
 ; AVX512F-NEXT:    kmovw %k1, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB52_64
 ; AVX512F-NEXT:  ## BB#63: ## %cond.load91
 ; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512F-NEXT:    vpinsrb $15, 31(%rdi), %xmm1, %xmm1

 ; AVX512F-NEXT:    kmovw %k1, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB52_94
 ; AVX512F-NEXT:  ## BB#93: ## %cond.load136
 ; AVX512F-NEXT:    vpinsrb $14, 46(%rdi), %xmm1, %xmm3
 ; AVX512F-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
 ; AVX512F-NEXT:  LBB52_94: ## %else137
 ; AVX512F-NEXT:    vptestmd %zmm2, %zmm2, %k1
-; AVX512F-NEXT:    kshiftlw $0, %k0, %k0
 ; AVX512F-NEXT:    kshiftrw $15, %k0, %k0
 ; AVX512F-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp) ## 2-byte Spill
 ; AVX512F-NEXT:    kmovw %k0, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB52_96
 ; AVX512F-NEXT:  ## BB#95: ## %cond.load139
 ; AVX512F-NEXT:    vpinsrb $15, 47(%rdi), %xmm1, %xmm2
 ; AVX512F-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]

 ; AVX512F-NEXT:    kmovw %k7, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB52_126
 ; AVX512F-NEXT:  ## BB#125: ## %cond.load184
 ; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm2
 ; AVX512F-NEXT:    vpinsrb $14, 62(%rdi), %xmm2, %xmm2
 ; AVX512F-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
 ; AVX512F-NEXT:  LBB52_126: ## %else185
-; AVX512F-NEXT:    kshiftlw $0, %k1, %k1
 ; AVX512F-NEXT:    kshiftrw $15, %k1, %k1
 ; AVX512F-NEXT:    kmovw %k1, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB52_128
 ; AVX512F-NEXT:  ## BB#127: ## %cond.load187
 ; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm2
 ; AVX512F-NEXT:    vpinsrb $15, 63(%rdi), %xmm2, %xmm2
 ; AVX512F-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1

 ; AVX512F-NEXT:    kmovw %k0, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB54_30
 ; AVX512F-NEXT:  ## BB#29: ## %cond.load40
 ; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512F-NEXT:    vpinsrw $6, 28(%rdi), %xmm1, %xmm1
 ; AVX512F-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
 ; AVX512F-NEXT:  LBB54_30: ## %else41
-; AVX512F-NEXT:    kshiftlw $0, %k1, %k0
-; AVX512F-NEXT:    kshiftrw $15, %k0, %k0
+; AVX512F-NEXT:    kshiftrw $15, %k1, %k0
 ; AVX512F-NEXT:    kmovw %k0, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB54_32
 ; AVX512F-NEXT:  ## BB#31: ## %cond.load43
 ; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512F-NEXT:    vpinsrw $7, 30(%rdi), %xmm1, %xmm1
 ; AVX512F-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
 ; AVX512F-NEXT:  LBB54_32: ## %else44

 ; AVX512F-NEXT:    kshiftlw $1, %k0, %k1
 ; AVX512F-NEXT:    kshiftrw $15, %k1, %k1
 ; AVX512F-NEXT:    kmovw %k1, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB56_30
 ; AVX512F-NEXT:  ## BB#29: ## %cond.store27
 ; AVX512F-NEXT:    vpextrb $14, %xmm1, 14(%rdi)
 ; AVX512F-NEXT:  LBB56_30: ## %else28
-; AVX512F-NEXT:    kshiftlw $0, %k0, %k0
 ; AVX512F-NEXT:    kshiftrw $15, %k0, %k0
 ; AVX512F-NEXT:    kmovw %k0, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB56_32
 ; AVX512F-NEXT:  ## BB#31: ## %cond.store29
 ; AVX512F-NEXT:    vpextrb $15, %xmm1, 15(%rdi)
 ; AVX512F-NEXT:  LBB56_32: ## %else30
 ; AVX512F-NEXT:    retq

 ; AVX512F-NEXT:    kshiftrw $15, %k1, %k1
 ; AVX512F-NEXT:    kmovw %k1, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB58_30
 ; AVX512F-NEXT:  ## BB#29: ## %cond.store27
 ; AVX512F-NEXT:    vpextrb $14, %xmm4, 14(%rdi)
 ; AVX512F-NEXT:  LBB58_30: ## %else28
 ; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k1
-; AVX512F-NEXT:    kshiftlw $0, %k0, %k0
 ; AVX512F-NEXT:    kshiftrw $15, %k0, %k0
 ; AVX512F-NEXT:    kmovw %k0, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB58_32
 ; AVX512F-NEXT:  ## BB#31: ## %cond.store29
 ; AVX512F-NEXT:    vpextrb $15, %xmm4, 15(%rdi)
 ; AVX512F-NEXT:  LBB58_32: ## %else30
 ; AVX512F-NEXT:    kshiftlw $15, %k1, %k0

 ; AVX512F-NEXT:    kmovw %k0, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB58_62
 ; AVX512F-NEXT:  ## BB#61: ## %cond.store59
 ; AVX512F-NEXT:    vextracti128 $1, %ymm4, %xmm1
 ; AVX512F-NEXT:    vpextrb $14, %xmm1, 30(%rdi)
 ; AVX512F-NEXT:  LBB58_62: ## %else60
 ; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512F-NEXT:    kshiftlw $0, %k1, %k1
 ; AVX512F-NEXT:    kshiftrw $15, %k1, %k1
 ; AVX512F-NEXT:    kmovw %k1, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB58_64
 ; AVX512F-NEXT:  ## BB#63: ## %cond.store61
 ; AVX512F-NEXT:    vextracti128 $1, %ymm4, %xmm0
 ; AVX512F-NEXT:    vpextrb $15, %xmm0, 31(%rdi)
 ; AVX512F-NEXT:  LBB58_64: ## %else62

 ; AVX512F-NEXT:    kshiftrw $15, %k1, %k1
 ; AVX512F-NEXT:    kmovw %k1, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB58_94
 ; AVX512F-NEXT:  ## BB#93: ## %cond.store91
 ; AVX512F-NEXT:    vpextrb $14, %xmm5, 46(%rdi)
 ; AVX512F-NEXT:  LBB58_94: ## %else92
 ; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k1
-; AVX512F-NEXT:    kshiftlw $0, %k0, %k0
 ; AVX512F-NEXT:    kshiftrw $15, %k0, %k0
 ; AVX512F-NEXT:    kmovw %k0, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB58_96
 ; AVX512F-NEXT:  ## BB#95: ## %cond.store93
 ; AVX512F-NEXT:    vpextrb $15, %xmm5, 47(%rdi)
 ; AVX512F-NEXT:  LBB58_96: ## %else94
 ; AVX512F-NEXT:    kshiftlw $15, %k1, %k0

 ; AVX512F-NEXT:    kshiftrw $15, %k0, %k0
 ; AVX512F-NEXT:    kmovw %k0, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB58_126
 ; AVX512F-NEXT:  ## BB#125: ## %cond.store123
 ; AVX512F-NEXT:    vextracti128 $1, %ymm5, %xmm0
 ; AVX512F-NEXT:    vpextrb $14, %xmm0, 62(%rdi)
 ; AVX512F-NEXT:  LBB58_126: ## %else124
-; AVX512F-NEXT:    kshiftlw $0, %k1, %k0
-; AVX512F-NEXT:    kshiftrw $15, %k0, %k0
+; AVX512F-NEXT:    kshiftrw $15, %k1, %k0
 ; AVX512F-NEXT:    kmovw %k0, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB58_128
 ; AVX512F-NEXT:  ## BB#127: ## %cond.store125
 ; AVX512F-NEXT:    vextracti128 $1, %ymm5, %xmm0
 ; AVX512F-NEXT:    vpextrb $15, %xmm0, 63(%rdi)
 ; AVX512F-NEXT:  LBB58_128: ## %else126
 ; AVX512F-NEXT:    retq

 ; AVX512F-NEXT:    kshiftrw $15, %k1, %k1
 ; AVX512F-NEXT:    kmovw %k1, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB60_30
 ; AVX512F-NEXT:  ## BB#29: ## %cond.store27
 ; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm0
 ; AVX512F-NEXT:    vpextrw $6, %xmm0, 28(%rdi)
 ; AVX512F-NEXT:  LBB60_30: ## %else28
-; AVX512F-NEXT:    kshiftlw $0, %k0, %k0
 ; AVX512F-NEXT:    kshiftrw $15, %k0, %k0
 ; AVX512F-NEXT:    kmovw %k0, %eax
 ; AVX512F-NEXT:    testb %al, %al
 ; AVX512F-NEXT:    je LBB60_32
 ; AVX512F-NEXT:  ## BB#31: ## %cond.store29
 ; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm0
 ; AVX512F-NEXT:    vpextrw $7, %xmm0, 30(%rdi)
 ; AVX512F-NEXT:  LBB60_32: ## %else30

test/CodeGen/X86/vector-compare-results.ll

 ; AVX512-NEXT:    kshiftlw $2, %k0, %k1
 ; AVX512-NEXT:    kshiftrw $15, %k1, %k1
 ; AVX512-NEXT:    kmovw %k1, %eax
 ; AVX512-NEXT:    vpinsrb $13, %eax, %xmm3, %xmm3
 ; AVX512-NEXT:    kshiftlw $1, %k0, %k1
 ; AVX512-NEXT:    kshiftrw $15, %k1, %k1
 ; AVX512-NEXT:    kmovw %k1, %eax
 ; AVX512-NEXT:    vpinsrb $14, %eax, %xmm3, %xmm3
-; AVX512-NEXT:    kshiftlw $0, %k0, %k0
 ; AVX512-NEXT:    kshiftrw $15, %k0, %k0
 ; AVX512-NEXT:    kmovw %k0, %eax
 ; AVX512-NEXT:    vpinsrb $15, %eax, %xmm3, %xmm3
 ; AVX512-NEXT:    vpcmpgtw %ymm6, %ymm2, %ymm2
 ; AVX512-NEXT:    vpmovsxwd %ymm2, %zmm2
 ; AVX512-NEXT:    vpslld $31, %zmm2, %zmm2
 ; AVX512-NEXT:    vptestmd %zmm2, %zmm2, %k0
 ; AVX512-NEXT:    kshiftlw $14, %k0, %k1

 ; AVX512-NEXT:    kshiftlw $2, %k0, %k1
 ; AVX512-NEXT:    kshiftrw $15, %k1, %k1
 ; AVX512-NEXT:    kmovw %k1, %eax
 ; AVX512-NEXT:    vpinsrb $13, %eax, %xmm2, %xmm2
 ; AVX512-NEXT:    kshiftlw $1, %k0, %k1
 ; AVX512-NEXT:    kshiftrw $15, %k1, %k1
 ; AVX512-NEXT:    kmovw %k1, %eax
 ; AVX512-NEXT:    vpinsrb $14, %eax, %xmm2, %xmm2
-; AVX512-NEXT:    kshiftlw $0, %k0, %k0
 ; AVX512-NEXT:    kshiftrw $15, %k0, %k0
 ; AVX512-NEXT:    kmovw %k0, %eax
 ; AVX512-NEXT:    vpinsrb $15, %eax, %xmm2, %xmm2
 ; AVX512-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
 ; AVX512-NEXT:    vpsllw $7, %ymm2, %ymm3
 ; AVX512-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
 ; AVX512-NEXT:    vpand %ymm2, %ymm3, %ymm3
 ; AVX512-NEXT:    vpxor %ymm6, %ymm6, %ymm6

 ; AVX512-NEXT:    kshiftlw $2, %k0, %k1
 ; AVX512-NEXT:    kshiftrw $15, %k1, %k1
 ; AVX512-NEXT:    kmovw %k1, %eax
 ; AVX512-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
 ; AVX512-NEXT:    kshiftlw $1, %k0, %k1
 ; AVX512-NEXT:    kshiftrw $15, %k1, %k1
 ; AVX512-NEXT:    kmovw %k1, %eax
 ; AVX512-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm1
-; AVX512-NEXT:    kshiftlw $0, %k0, %k0
 ; AVX512-NEXT:    kshiftrw $15, %k0, %k0
 ; AVX512-NEXT:    kmovw %k0, %eax
 ; AVX512-NEXT:    vpinsrb $15, %eax, %xmm1, %xmm1
 ; AVX512-NEXT:    vpcmpgtw %ymm4, %ymm0, %ymm0
 ; AVX512-NEXT:    vpmovsxwd %ymm0, %zmm0
 ; AVX512-NEXT:    vpslld $31, %zmm0, %zmm0
 ; AVX512-NEXT:    vptestmd %zmm0, %zmm0, %k0
 ; AVX512-NEXT:    kshiftlw $14, %k0, %k1

 ; AVX512-NEXT:    kshiftlw $2, %k0, %k1
 ; AVX512-NEXT:    kshiftrw $15, %k1, %k1
 ; AVX512-NEXT:    kmovw %k1, %eax
 ; AVX512-NEXT:    vpinsrb $13, %eax, %xmm0, %xmm0
 ; AVX512-NEXT:    kshiftlw $1, %k0, %k1
 ; AVX512-NEXT:    kshiftrw $15, %k1, %k1
 ; AVX512-NEXT:    kmovw %k1, %eax
 ; AVX512-NEXT:    vpinsrb $14, %eax, %xmm0, %xmm0
-; AVX512-NEXT:    kshiftlw $0, %k0, %k0
 ; AVX512-NEXT:    kshiftrw $15, %k0, %k0
 ; AVX512-NEXT:    kmovw %k0, %eax
 ; AVX512-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
 ; AVX512-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vpsllw $7, %ymm0, %ymm0
 ; AVX512-NEXT:    vpand %ymm2, %ymm0, %ymm0
 ; AVX512-NEXT:    vpcmpgtb %ymm0, %ymm6, %ymm1
 ; AVX512-NEXT:    vpsllw $7, %xmm1, %xmm0