diff --git a/lib/CodeGen/RegisterCoalescer.cpp b/lib/CodeGen/RegisterCoalescer.cpp
index 3493645f7f2..b7d18ba1a14 100644
--- a/lib/CodeGen/RegisterCoalescer.cpp
+++ b/lib/CodeGen/RegisterCoalescer.cpp
@@ -553,10 +553,12 @@ bool RegisterCoalescer::AdjustCopiesBackFrom(const CoalescerPair &CP,
   if (UIdx != -1) {
     ValLREndInst->getOperand(UIdx).setIsKill(false);
   }
-
-  // If the copy instruction was killing the destination register before the
-  // merge, find the last use and trim the live range. That will also add the
-  // isKill marker.
+  
+  // Rewrite the copy. If the copy instruction was killing the destination
+  // register before the merge, find the last use and trim the live range. That
+  // will also add the isKill marker.
+  CopyMI->substituteRegister(IntA.reg, IntB.reg, CP.getSubIdx(),
+                             *TRI);
   if (ALR->end == CopyIdx)
     LIS->shrinkToUses(&IntA);
 
diff --git a/test/CodeGen/ARM/2012-01-26-CoalescerBug.ll b/test/CodeGen/ARM/2012-01-26-CoalescerBug.ll
new file mode 100644
index 00000000000..ec5b2e9de7c
--- /dev/null
+++ b/test/CodeGen/ARM/2012-01-26-CoalescerBug.ll
@@ -0,0 +1,21 @@
+; RUN: llc < %s -verify-coalescing
+; PR11861
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:64:128-a0:0:64-n32-S64"
+target triple = "armv7-none-linux-eabi"
+
+define arm_aapcs_vfpcc void @foo() nounwind uwtable align 2 {
+  br label %1
+
+; <label>:1                                       ; preds = %1, %0
+  %2 = phi <4 x float> [ undef, %0 ], [ %11, %1 ]
+  %3 = bitcast <4 x float> %2 to <2 x i64>
+  %4 = shufflevector <2 x i64> %3, <2 x i64> undef, <1 x i32> zeroinitializer
+  %5 = xor <2 x i32> zeroinitializer, <i32 -1, i32 -1>
+  %6 = bitcast <2 x i32> zeroinitializer to <2 x float>
+  %7 = shufflevector <2 x float> zeroinitializer, <2 x float> %6, <2 x i32> <i32 0, i32 2>
+  %8 = shufflevector <2 x i64> %3, <2 x i64> undef, <1 x i32> <i32 1>
+  %9 = bitcast <2 x float> %7 to <1 x i64>
+  %10 = shufflevector <1 x i64> %9, <1 x i64> %8, <2 x i32> <i32 0, i32 1>
+  %11 = bitcast <2 x i64> %10 to <4 x float>
+  br label %1
+}