diff --git a/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp b/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp
index 52506fa..4a38324 100644
--- a/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp
+++ b/lib/Target/X86/MCTargetDesc/X86MCCodeEmitter.cpp
@@ -572,8 +572,14 @@
   // Classify VEX_B, VEX_4V, VEX_R, VEX_X
   unsigned NumOps = Desc.getNumOperands();
   unsigned CurOp = 0;
-  if (NumOps > 1 && Desc.getOperandConstraint(1, MCOI::TIED_TO) != -1)
+  if (NumOps > 1 && Desc.getOperandConstraint(1, MCOI::TIED_TO) == 0)
     ++CurOp;
+  else if (NumOps > 3 && Desc.getOperandConstraint(2, MCOI::TIED_TO) == 0) {
+    assert(Desc.getOperandConstraint(NumOps - 1, MCOI::TIED_TO) == 1);
+    // Special case for GATHER with 2 TIED_TO operands
+    // Skip the first 2 operands: dst, mask_wb
+    CurOp += 2;
+  }
 
   switch (TSFlags & X86II::FormMask) {
   case X86II::MRMInitReg: llvm_unreachable("FIXME: Remove this!");
@@ -971,11 +977,14 @@
   // FIXME: This should be handled during MCInst lowering.
   unsigned NumOps = Desc.getNumOperands();
   unsigned CurOp = 0;
-  if (NumOps > 1 && Desc.getOperandConstraint(1, MCOI::TIED_TO) != -1)
+  if (NumOps > 1 && Desc.getOperandConstraint(1, MCOI::TIED_TO) == 0)
     ++CurOp;
-  else if (NumOps > 2 && Desc.getOperandConstraint(NumOps-1, MCOI::TIED_TO)== 0)
-    // Skip the last source operand that is tied_to the dest reg. e.g. LXADD32
-    --NumOps;
+  else if (NumOps > 3 && Desc.getOperandConstraint(2, MCOI::TIED_TO) == 0) {
+    assert(Desc.getOperandConstraint(NumOps - 1, MCOI::TIED_TO) == 1);
+    // Special case for GATHER with 2 TIED_TO operands
+    // Skip the first 2 operands: dst, mask_wb
+    CurOp += 2;
+  }
 
   // Keep track of the current byte being emitted.
   unsigned CurByte = 0;
diff --git a/lib/Target/X86/X86CodeEmitter.cpp b/lib/Target/X86/X86CodeEmitter.cpp
index 3079dfa..977cc50 100644
--- a/lib/Target/X86/X86CodeEmitter.cpp
+++ b/lib/Target/X86/X86CodeEmitter.cpp
@@ -935,8 +935,15 @@
   // Classify VEX_B, VEX_4V, VEX_R, VEX_X
   unsigned NumOps = Desc->getNumOperands();
   unsigned CurOp = 0;
-  if (NumOps > 1 && Desc->getOperandConstraint(1, MCOI::TIED_TO) != -1)
+  if (NumOps > 1 && Desc->getOperandConstraint(1, MCOI::TIED_TO) == 0)
     ++CurOp;
+  else if (NumOps > 3 && Desc->getOperandConstraint(2, MCOI::TIED_TO) == 0) {
+    assert(Desc->getOperandConstraint(NumOps - 1, MCOI::TIED_TO) == 1);
+    // Special case for GATHER with 2 TIED_TO operands
+    // Skip the first 2 operands: dst, mask_wb
+    CurOp += 2;
+  }
+
   switch (TSFlags & X86II::FormMask) {
     case X86II::MRMInitReg:
       // Duplicate register.
@@ -1117,11 +1124,14 @@
   // If this is a two-address instruction, skip one of the register operands.
   unsigned NumOps = Desc->getNumOperands();
   unsigned CurOp = 0;
-  if (NumOps > 1 && Desc->getOperandConstraint(1, MCOI::TIED_TO) != -1)
+  if (NumOps > 1 && Desc->getOperandConstraint(1, MCOI::TIED_TO) == 0)
     ++CurOp;
-  else if (NumOps > 2 && Desc->getOperandConstraint(NumOps-1,MCOI::TIED_TO)== 0)
-    // Skip the last source operand that is tied_to the dest reg. e.g. LXADD32
-    --NumOps;
+  else if (NumOps > 3 && Desc->getOperandConstraint(2, MCOI::TIED_TO) == 0) {
+    assert(Desc->getOperandConstraint(NumOps - 1, MCOI::TIED_TO) == 1);
+    // Special case for GATHER with 2 TIED_TO operands
+    // Skip the first 2 operands: dst, mask_wb
+    CurOp += 2;
+  }
 
   uint64_t TSFlags = Desc->TSFlags;
 
diff --git a/lib/Target/X86/X86ISelDAGToDAG.cpp b/lib/Target/X86/X86ISelDAGToDAG.cpp
index 8170130..5186482 100644
--- a/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -1966,14 +1966,22 @@
   if (!Scale)
     return 0;
 
+  SDVTList VTs = CurDAG->getVTList(VSrc.getValueType(), VSrc.getValueType(),
+                                   MVT::Other);
+
   // Memory Operands: Base, Scale, Index, Disp, Segment
   SDValue Disp = CurDAG->getTargetConstant(0, MVT::i32);
   SDValue Segment = CurDAG->getRegister(0, MVT::i32);
   const SDValue Ops[] = { VSrc, Base, getI8Imm(Scale->getSExtValue()), VIdx,
                           Disp, Segment, VMask, Chain};
   SDNode *ResNode = CurDAG->getMachineNode(Opc, Node->getDebugLoc(),
-                                           VSrc.getValueType(), MVT::Other,
-                                           Ops, array_lengthof(Ops));
+                                           VTs, Ops, array_lengthof(Ops));
+  // Node has 2 outputs: VDst and MVT::Other.
+  // ResNode has 3 outputs: VDst, VMask_wb, and MVT::Other.
+  // We replace VDst of Node with VDst of ResNode, and Other of Node with Other
+  // of ResNode.
+  ReplaceUses(SDValue(Node, 0), SDValue(ResNode, 0));
+  ReplaceUses(SDValue(Node, 1), SDValue(ResNode, 2));
   return ResNode;
 }
 
@@ -2034,7 +2042,8 @@
       }
       SDNode *RetVal = SelectGather(Node, Opc);
       if (RetVal)
-        return RetVal;
+        // We already called ReplaceUses inside SelectGather.
+        return NULL;
       break;
     }
     }
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 5e96eec..0ad9241 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -8038,19 +8038,19 @@
 // VGATHER - GATHER Operations
 multiclass avx2_gather<bits<8> opc, string OpcodeStr,
                        RegisterClass RC256, X86MemOperand memop256> {
-  def rm  : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
+  def rm  : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
             (ins VR128:$src1, v128mem:$src2, VR128:$mask),
             !strconcat(OpcodeStr,
               "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
             []>, VEX_4VOp3;
-  def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst),
+  def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
             (ins RC256:$src1, memop256:$src2, RC256:$mask),
             !strconcat(OpcodeStr,
               "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
             []>, VEX_4VOp3, VEX_L;
 }
 
-let Constraints = "$src1 = $dst" in {
+let Constraints = "$src1 = $dst, $mask = $mask_wb" in {
   defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, v128mem>, VEX_W;
   defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, v256mem>, VEX_W;
   defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, v256mem>;
diff --git a/test/CodeGen/X86/avx2-intrinsics-x86.ll b/test/CodeGen/X86/avx2-intrinsics-x86.ll
index 459dbb2..a6141b0 100644
--- a/test/CodeGen/X86/avx2-intrinsics-x86.ll
+++ b/test/CodeGen/X86/avx2-intrinsics-x86.ll
@@ -1136,3 +1136,22 @@
 }
 declare <4 x i32> @llvm.x86.avx2.gather.q.d.256(<4 x i32>, i8*,
                       <4 x i64>, <4 x i32>, i8) nounwind readonly
+
+; PR13298
+define <8 x float>  @test_gather_mask(<8 x float> %a0, float* %a,
+                                      <8 x i32> %idx, <8 x float> %mask,
+                                      float* nocapture %out) {
+; CHECK: test_gather_mask
+; CHECK: vmovdqa %ymm2, [[DEST:%.*]]
+; CHECK: vgatherdps [[DEST]]
+;; gather with mask
+  %a_i8 = bitcast float* %a to i8*
+  %res = call <8 x float> @llvm.x86.avx2.gather.d.ps.256(<8 x float> %a0,
+                           i8* %a_i8, <8 x i32> %idx, <8 x float> %mask, i8 4) ;
+
+;; for debugging, we'll just dump out the mask
+  %out_ptr = bitcast float * %out to <8 x float> *
+  store <8 x float> %mask, <8 x float> * %out_ptr, align 4
+
+  ret <8 x float> %res
+}
diff --git a/utils/TableGen/X86RecognizableInstr.cpp b/utils/TableGen/X86RecognizableInstr.cpp
index 0ab21c8..1cc67c2 100644
--- a/utils/TableGen/X86RecognizableInstr.cpp
+++ b/utils/TableGen/X86RecognizableInstr.cpp
@@ -277,8 +277,8 @@
 }
   
 void RecognizableInstr::processInstr(DisassemblerTables &tables,
-	const CodeGenInstruction &insn,
-                                   InstrUID uid)
+                                     const CodeGenInstruction &insn,
+                                     InstrUID uid)
 {
   // Ignore "asm parser only" instructions.
   if (insn.TheDef->getValueAsBit("isAsmParserOnly"))
@@ -508,13 +508,13 @@
   return false;
 }
   
-void RecognizableInstr::handleOperand(
-  bool optional,
-  unsigned &operandIndex,
-  unsigned &physicalOperandIndex,
-  unsigned &numPhysicalOperands,
-  unsigned *operandMapping,
-  OperandEncoding (*encodingFromString)(const std::string&, bool hasOpSizePrefix)) {
+void RecognizableInstr::handleOperand(bool optional, unsigned &operandIndex,
+                                      unsigned &physicalOperandIndex,
+                                      unsigned &numPhysicalOperands,
+                                      const unsigned *operandMapping,
+                                      OperandEncoding (*encodingFromString)
+                                        (const std::string&,
+                                         bool hasOpSizePrefix)) {
   if (optional) {
     if (physicalOperandIndex >= numPhysicalOperands)
       return;
@@ -563,7 +563,6 @@
     
   const std::vector<CGIOperandList::OperandInfo> &OperandList = *Operands;
   
-  unsigned operandIndex;
   unsigned numOperands = OperandList.size();
   unsigned numPhysicalOperands = 0;
   
@@ -575,12 +574,13 @@
   
   assert(numOperands <= X86_MAX_OPERANDS && "X86_MAX_OPERANDS is not large enough");
   
-  for (operandIndex = 0; operandIndex < numOperands; ++operandIndex) {
+  for (unsigned operandIndex = 0; operandIndex < numOperands; ++operandIndex) {
     if (OperandList[operandIndex].Constraints.size()) {
       const CGIOperandList::ConstraintInfo &Constraint =
         OperandList[operandIndex].Constraints[0];
       if (Constraint.isTied()) {
-        operandMapping[operandIndex] = Constraint.getTiedOperand();
+        operandMapping[operandIndex] = operandIndex;
+        operandMapping[Constraint.getTiedOperand()] = operandIndex;
       } else {
         ++numPhysicalOperands;
         operandMapping[operandIndex] = operandIndex;
@@ -621,7 +621,7 @@
                 class##EncodingFromString);
   
   // operandIndex should always be < numOperands
-  operandIndex = 0;
+  unsigned operandIndex = 0;
   // physicalOperandIndex should always be < numPhysicalOperands
   unsigned physicalOperandIndex = 0;
     
diff --git a/utils/TableGen/X86RecognizableInstr.h b/utils/TableGen/X86RecognizableInstr.h
index 6c0a234..542e510 100644
--- a/utils/TableGen/X86RecognizableInstr.h
+++ b/utils/TableGen/X86RecognizableInstr.h
@@ -204,7 +204,7 @@
                      unsigned &operandIndex,
                      unsigned &physicalOperandIndex,
                      unsigned &numPhysicalOperands,
-                     unsigned *operandMapping,
+                     const unsigned *operandMapping,
                      OperandEncoding (*encodingFromString)
                        (const std::string&,
                         bool hasOpSizePrefix));