Index: lib/Target/AMDGPU/SIISelLowering.cpp
===================================================================
--- lib/Target/AMDGPU/SIISelLowering.cpp
+++ lib/Target/AMDGPU/SIISelLowering.cpp
@@ -429,6 +429,8 @@
   PointerType *PtrTy = PointerType::get(Ty, AMDGPUAS::CONSTANT_ADDRESS);
   SDValue BasePtr = DAG.getCopyFromReg(Chain, SL,
                                        MRI.getLiveInVirtReg(InputPtrReg), PtrVT);
+  Chain = BasePtr.getValue(1);
+
   SDValue Ptr = DAG.getNode(ISD::ADD, SL, PtrVT, BasePtr,
                             DAG.getConstant(Offset, SL, PtrVT));
   SDValue PtrOffset = DAG.getUNDEF(getPointerTy(AMDGPUAS::CONSTANT_ADDRESS));
@@ -451,7 +453,12 @@
                                true, // isNonTemporal
                                true, // isInvariant
                                Align); // Alignment
-    return DAG.getNode(ISD::FP16_TO_FP, SL, VT, Load);
+    SDValue Ops[] = {
+      DAG.getNode(ISD::FP16_TO_FP, SL, VT, Load),
+      Load.getValue(1)
+    };
+
+    return DAG.getMergeValues(Ops, SL);
   }
 
   ISD::LoadExtType ExtTy = Signed ? ISD::SEXTLOAD : ISD::ZEXTLOAD;
@@ -569,6 +576,8 @@
 
   AnalyzeFormalArguments(CCInfo, Splits);
 
+  SmallVector<SDValue, 16> Chains;
+
   for (unsigned i = 0, e = Ins.size(), ArgIdx = 0; i != e; ++i) {
 
     const ISD::InputArg &Arg = Ins[i];
@@ -586,8 +595,9 @@
       const unsigned Offset = 36 + VA.getLocMemOffset();
       // The first 36 bytes of the input buffer contains information about
       // thread group and global sizes.
-      SDValue Arg = LowerParameter(DAG, VT, MemVT,  DL, DAG.getRoot(),
+      SDValue Arg = LowerParameter(DAG, VT, MemVT,  DL, Chain,
                                    Offset, Ins[i].Flags.isSExt());
+      Chains.push_back(Arg.getValue(1));
 
       const PointerType *ParamTy =
         dyn_cast<PointerType>(FType->getParamType(Ins[i].getOrigArgIndex()));
@@ -613,7 +623,9 @@
       Reg = TRI->getMatchingSuperReg(Reg, AMDGPU::sub0,
                                      &AMDGPU::SReg_64RegClass);
       Reg = MF.addLiveIn(Reg, &AMDGPU::SReg_64RegClass);
-      InVals.push_back(DAG.getCopyFromReg(Chain, DL, Reg, VT));
+      SDValue Copy = DAG.getCopyFromReg(Chain, DL, Reg, VT);
+      InVals.push_back(Copy);
+      Chains.push_back(Copy.getValue(1));
       continue;
     }
 
@@ -621,6 +633,7 @@
 
     Reg = MF.addLiveIn(Reg, RC);
     SDValue Val = DAG.getCopyFromReg(Chain, DL, Reg, VT);
+    Chains.push_back(Val.getValue(1));
 
     if (Arg.VT.isVector()) {
 
@@ -633,7 +646,10 @@
       for (unsigned j = 1; j != NumElements; ++j) {
         Reg = ArgLocs[ArgIdx++].getLocReg();
         Reg = MF.addLiveIn(Reg, RC);
-        Regs.push_back(DAG.getCopyFromReg(Chain, DL, Reg, VT));
+
+        SDValue Copy = DAG.getCopyFromReg(Chain, DL, Reg, VT);
+        Regs.push_back(Copy);
+        Chains.push_back(Copy.getValue(1));
       }
 
       // Fill up the missing vector elements
@@ -652,7 +668,11 @@
         AMDGPU::SGPR_32RegClass.begin(), AMDGPU::SGPR_32RegClass.getNumRegs()));
     Info->ScratchOffsetReg = AMDGPU::SGPR_32RegClass.getRegister(ScratchIdx);
   }
-  return Chain;
+
+  if (Chains.empty())
+    return Chain;
+
+  return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Chains);
 }
 
 MachineBasicBlock * SITargetLowering::EmitInstrWithCustomInserter(