Index: lib/CodeGen/CGBuiltin.cpp
===================================================================
--- lib/CodeGen/CGBuiltin.cpp
+++ lib/CodeGen/CGBuiltin.cpp
@@ -7485,11 +7485,34 @@
 
 static Value *EmitX86SExtMask(CodeGenFunction &CGF, Value *Op, 
                               llvm::Type *DstTy) {
-  unsigned NumberOfElements = DstTy->getVectorNumElements();
+  unsigned NumberOfElements = DstTy->getVectorNumElements(); 
   Value *Mask = getMaskVecValue(CGF, Op, NumberOfElements);
   return CGF.Builder.CreateSExt(Mask, DstTy, "vpmovm2");
 }
 
+static Value *EmitX86Mask(CodeGenFunction &CGF, ArrayRef<Value *> Ops) {
+  llvm::Type *Typ1 = Ops[0]->getType();
+  Value *Cmp = CGF.Builder.CreateICmp(CmpInst::Predicate::ICMP_SLT, Ops[0],
+                                      llvm::ConstantVector::getNullValue(Typ1));
+  Value *BitCast = CGF.Builder.CreateBitCast(
+      Cmp, llvm::Type::getIntNTy(CGF.Builder.getContext(),
+                                 Typ1->getVectorNumElements()));
+  return (Typ1->getVectorNumElements() < 32)
+             ? CGF.Builder.CreateZExt(
+                   BitCast, llvm::Type::getInt32Ty(CGF.getLLVMContext()))
+             : BitCast;
+}
+
+static Value *EmitX86MaskFloat(CodeGenFunction &CGF, ArrayRef<Value *> Ops) {
+  Value *FloatOp = Ops[0];
+  unsigned DstTypEle = FloatOp->getType()->getVectorNumElements();
+  llvm::Type *IntTy = llvm::Type::getIntNTy(
+      CGF.Builder.getContext(), FloatOp->getType()->getScalarSizeInBits());
+  llvm::Type *DstTyp = llvm::VectorType::get(IntTy, DstTypEle);
+  Value *BitCastFloatToInt = CGF.Builder.CreateBitCast(FloatOp, DstTyp);
+  return EmitX86Mask(CGF, BitCastFloatToInt);
+}
+
 Value *CodeGenFunction::EmitX86CpuIs(const CallExpr *E) {
   const Expr *CPUExpr = E->getArg(0)->IgnoreParenCasts();
   StringRef CPUStr = cast<clang::StringLiteral>(CPUExpr)->getString();
@@ -7860,6 +7883,14 @@
   case X86::BI__builtin_ia32_cvtmask2q512:
     return EmitX86SExtMask(*this, Ops[0], ConvertType(E->getType()));
 
+  case X86::BI__builtin_ia32_pmovmskb128:
+  case X86::BI__builtin_ia32_pmovmskb256:
+    return EmitX86Mask(*this, Ops);
+  case X86::BI__builtin_ia32_movmskps256:
+  case X86::BI__builtin_ia32_movmskpd256:
+  case X86::BI__builtin_ia32_movmskps:
+  case X86::BI__builtin_ia32_movmskpd:
+    return EmitX86MaskFloat(*this, Ops);
   case X86::BI__builtin_ia32_movdqa32store128_mask:
   case X86::BI__builtin_ia32_movdqa64store128_mask:
   case X86::BI__builtin_ia32_storeaps128_mask: