diff --git a/lib/CodeGen/SelectionDAG/FastISel.cpp b/lib/CodeGen/SelectionDAG/FastISel.cpp
index 2b624adac94..13f0cb3900f 100644
--- a/lib/CodeGen/SelectionDAG/FastISel.cpp
+++ b/lib/CodeGen/SelectionDAG/FastISel.cpp
@@ -1675,6 +1675,7 @@ unsigned FastISel::fastEmit_ri_(MVT VT, unsigned Opcode, unsigned Op0,
   if (ResultReg)
     return ResultReg;
   unsigned MaterialReg = fastEmit_i(ImmType, ImmType, ISD::Constant, Imm);
+  bool IsImmKill = true;
   if (!MaterialReg) {
     // This is a bit ugly/slow, but failing here means falling out of
     // fast-isel, which would be very slow.
@@ -1683,9 +1684,12 @@ unsigned FastISel::fastEmit_ri_(MVT VT, unsigned Opcode, unsigned Op0,
     MaterialReg = getRegForValue(ConstantInt::get(ITy, Imm));
     if (!MaterialReg)
       return 0;
+    // If this constant was already materialized, then we don't want to kill it.
+    // In this case we will have a use.
+    if (!MRI.use_empty(MaterialReg))
+      IsImmKill = false;
   }
-  return fastEmit_rr(VT, VT, Opcode, Op0, Op0IsKill, MaterialReg,
-                     /*IsKill=*/true);
+  return fastEmit_rr(VT, VT, Opcode, Op0, Op0IsKill, MaterialReg, IsImmKill);
 }
 
 unsigned FastISel::createResultReg(const TargetRegisterClass *RC) {
diff --git a/test/CodeGen/ARM/fast-isel-shift-materialize.ll b/test/CodeGen/ARM/fast-isel-shift-materialize.ll
new file mode 100644
index 00000000000..ab36a979d71
--- /dev/null
+++ b/test/CodeGen/ARM/fast-isel-shift-materialize.ll
@@ -0,0 +1,25 @@
+; RUN: llc %s -o - -verify-machineinstrs -fast-isel=true | FileCheck %s
+
+target datalayout = "e-m:o-p:32:32-f64:32:64-v64:32:64-v128:32:128-a:0:32-n32-S32"
+target triple = "thumbv7-apple-ios8.0.0"
+
+; When materializing the '2' for the shifts below, the second shift kills the vreg
+; we materialize in to.  However, the first shift was also killing that vreg.
+; This fails the machine verifier because the second use then thinks its reading an
+; undefined register.
+
+; CHECK: lsr.w
+; CHECK: lsr.w
+
+declare void @foo(i32, i32)
+
+; Function Attrs: minsize
+define i32 @test(i32 %tmp3, i32 %tmp9) #0 {
+bb:
+  %tmp4 = lshr i32 %tmp3, 2
+  %tmp10 = lshr i32 %tmp9, 2
+  call void @foo(i32 %tmp10, i32 %tmp4)
+  unreachable
+}
+
+attributes #0 = { minsize }