diff --git a/lib/Target/R600/SIFixSGPRCopies.cpp b/lib/Target/R600/SIFixSGPRCopies.cpp
index f6b8b783d02..5f714535abe 100644
--- a/lib/Target/R600/SIFixSGPRCopies.cpp
+++ b/lib/Target/R600/SIFixSGPRCopies.cpp
@@ -259,14 +259,17 @@ bool SIFixSGPRCopies::runOnMachineFunction(MachineFunction &MF) {
         break;
       }
       case AMDGPU::INSERT_SUBREG: {
-        const TargetRegisterClass *DstRC, *SrcRC;
+        const TargetRegisterClass *DstRC, *Src0RC, *Src1RC;
         DstRC = MRI.getRegClass(MI.getOperand(0).getReg());
-        SrcRC = MRI.getRegClass(MI.getOperand(1).getReg());
-        if (!TRI->isSGPRClass(DstRC) || !TRI->hasVGPRs(SrcRC))
-          break;
-        DEBUG(dbgs() << " Fixing INSERT_SUBREG:\n");
-        DEBUG(MI.print(dbgs()));
-        TII->moveToVALU(MI);
+        Src0RC = MRI.getRegClass(MI.getOperand(1).getReg());
+        Src1RC = MRI.getRegClass(MI.getOperand(2).getReg());
+        if (TRI->isSGPRClass(DstRC) &&
+            (TRI->hasVGPRs(Src0RC) || TRI->hasVGPRs(Src1RC))) {
+          DEBUG(dbgs() << " Fixing INSERT_SUBREG:\n");
+          DEBUG(MI.print(dbgs()));
+          TII->moveToVALU(MI);
+        }
+        break;
       }
       }
     }
diff --git a/lib/Target/R600/SIInstrInfo.cpp b/lib/Target/R600/SIInstrInfo.cpp
index c6ccbea5873..4a9e3467c5a 100644
--- a/lib/Target/R600/SIInstrInfo.cpp
+++ b/lib/Target/R600/SIInstrInfo.cpp
@@ -691,6 +691,7 @@ bool SIInstrInfo::canReadVGPR(const MachineInstr &MI, unsigned OpNo) const {
   case AMDGPU::COPY:
   case AMDGPU::REG_SEQUENCE:
   case AMDGPU::PHI:
+  case AMDGPU::INSERT_SUBREG:
     return RI.hasVGPRs(getOpRegClass(MI, 0));
   default:
     return RI.hasVGPRs(getOpRegClass(MI, OpNo));
@@ -924,6 +925,23 @@ void SIInstrInfo::legalizeOperands(MachineInstr *MI) const {
     }
   }
 
+  // Legalize INSERT_SUBREG
+  // src0 must have the same register class as dst
+  if (MI->getOpcode() == AMDGPU::INSERT_SUBREG) {
+    unsigned Dst = MI->getOperand(0).getReg();
+    unsigned Src0 = MI->getOperand(1).getReg();
+    const TargetRegisterClass *DstRC = MRI.getRegClass(Dst);
+    const TargetRegisterClass *Src0RC = MRI.getRegClass(Src0);
+    if (DstRC != Src0RC) {
+      MachineBasicBlock &MBB = *MI->getParent();
+      unsigned NewSrc0 = MRI.createVirtualRegister(DstRC);
+      BuildMI(MBB, MI, MI->getDebugLoc(), get(AMDGPU::COPY), NewSrc0)
+              .addReg(Src0);
+      MI->getOperand(1).setReg(NewSrc0);
+    }
+    return;
+  }
+
   // Legalize MUBUF* instructions
   // FIXME: If we start using the non-addr64 instructions for compute, we
   // may need to legalize them here.