diff --git a/lib/CodeGen/LowerSubregs.cpp b/lib/CodeGen/LowerSubregs.cpp
index 9c23a5ac155..e9e60a00f0e 100644
--- a/lib/CodeGen/LowerSubregs.cpp
+++ b/lib/CodeGen/LowerSubregs.cpp
@@ -254,7 +254,13 @@ bool LowerSubregsInstructionPass::LowerInsert(MachineInstr *MI) {
     // Insert sub-register copy
     const TargetRegisterClass *TRC0= TRI.getPhysicalRegisterRegClass(DstSubReg);
     const TargetRegisterClass *TRC1= TRI.getPhysicalRegisterRegClass(InsReg);
-    TII.copyRegToReg(*MBB, MI, DstSubReg, InsReg, TRC0, TRC1);
+    if (MI->getOperand(2).isUndef())
+      // If the source register being inserted is undef, then this becomes an
+      // implicit_def.
+      BuildMI(*MBB, MI, MI->getDebugLoc(),
+              TII.get(TargetInstrInfo::IMPLICIT_DEF), DstSubReg);
+    else
+      TII.copyRegToReg(*MBB, MI, DstSubReg, InsReg, TRC0, TRC1);
     MachineBasicBlock::iterator CopyMI = MI;
     --CopyMI;
 
@@ -270,7 +276,7 @@ bool LowerSubregsInstructionPass::LowerInsert(MachineInstr *MI) {
     }
 
     // Make sure the inserted register gets killed
-    if (MI->getOperand(2).isKill())
+    if (MI->getOperand(2).isKill() && !MI->getOperand(2).isUndef())
       TransferKillFlag(MI, InsReg, TRI);
   }
 
diff --git a/test/CodeGen/Thumb2/2009-08-04-SubregLoweringBug.ll b/test/CodeGen/Thumb2/2009-08-04-SubregLoweringBug.ll
new file mode 100644
index 00000000000..f2a2729a5b7
--- /dev/null
+++ b/test/CodeGen/Thumb2/2009-08-04-SubregLoweringBug.ll
@@ -0,0 +1,34 @@
+; RUN: llvm-as < %s | llc -mtriple=thumbv7-apple-darwin9 -mattr=+neon,+neonfp 
+; RUN: llvm-as < %s | llc -mtriple=thumbv7-apple-darwin9 -mattr=+neon,+neonfp | grep fcpys | count 1
+; rdar://7117307
+
+	%struct.Hosp = type { i32, i32, i32, %struct.List, %struct.List, %struct.List, %struct.List }
+	%struct.List = type { %struct.List*, %struct.Patient*, %struct.List* }
+	%struct.Patient = type { i32, i32, i32, %struct.Village* }
+	%struct.Results = type { float, float, float }
+	%struct.Village = type { [4 x %struct.Village*], %struct.Village*, %struct.List, %struct.Hosp, i32, i32 }
+
+define arm_apcscc void @get_results(%struct.Results* noalias nocapture sret %agg.result, %struct.Village* %village) nounwind {
+entry:
+	br i1 undef, label %bb, label %bb6.preheader
+
+bb6.preheader:		; preds = %entry
+	call void @llvm.memcpy.i32(i8* undef, i8* undef, i32 12, i32 4)
+	br i1 undef, label %bb15, label %bb13
+
+bb:		; preds = %entry
+	ret void
+
+bb13:		; preds = %bb13, %bb6.preheader
+	%0 = fadd float undef, undef		; <float> [#uses=1]
+	%1 = fadd float undef, 1.000000e+00		; <float> [#uses=1]
+	br i1 undef, label %bb15, label %bb13
+
+bb15:		; preds = %bb13, %bb6.preheader
+	%r1.0.0.lcssa = phi float [ 0.000000e+00, %bb6.preheader ], [ %1, %bb13 ]		; <float> [#uses=1]
+	%r1.1.0.lcssa = phi float [ undef, %bb6.preheader ], [ %0, %bb13 ]		; <float> [#uses=0]
+	store float %r1.0.0.lcssa, float* undef, align 4
+	ret void
+}
+
+declare void @llvm.memcpy.i32(i8* nocapture, i8* nocapture, i32, i32) nounwind