diff --git a/lib/Transforms/Scalar/LoopUnrollPass.cpp b/lib/Transforms/Scalar/LoopUnrollPass.cpp
index f20f7dc5d6d..f0a661cb699 100644
--- a/lib/Transforms/Scalar/LoopUnrollPass.cpp
+++ b/lib/Transforms/Scalar/LoopUnrollPass.cpp
@@ -132,8 +132,10 @@ bool LoopUnroll::runOnLoop(Loop *L, LPPassManager &LPM) {
     unsigned LoopSize = ApproximateLoopSize(L, NumCalls);
     DEBUG(dbgs() << "  Loop Size = " << LoopSize << "\n");
     if (NumCalls != 0) {
-      DEBUG(dbgs() << "  Not unrolling loop with function calls.\n");
-      return false;
+      // Even for a loop that contains calls, it can still be profitable to
+      // unroll if the loop is really, REALLY small.
+      DEBUG(dbgs() <<"  Using lower threshold for loop with function calls.\n");
+      CurrentThreshold = OptSizeUnrollThreshold;
     }
     uint64_t Size = (uint64_t)LoopSize*Count;
     if (TripCount != 1 && Size > CurrentThreshold) {
diff --git a/test/Transforms/LoopUnroll/call.ll b/test/Transforms/LoopUnroll/call.ll
new file mode 100644
index 00000000000..5dcb3bc2af4
--- /dev/null
+++ b/test/Transforms/LoopUnroll/call.ll
@@ -0,0 +1,51 @@
+; RUN: opt < %s -S -loop-unroll | FileCheck %s
+
+@id = internal global i32 0
+@val = internal global [4 x i32] zeroinitializer, align 16
+
+; CHECK: @test
+define i32 @test(i32 %k) nounwind ssp {
+; CHECK-NOT: call i32 @test(i32 %t.06)
+; CHECK: call i32 @test(i32 0)
+; CHECK-NOT: call i32 @test(i32 %t.06)
+; CHECK: call i32 @test(i32 1)
+; CHECK-NOT: call i32 @test(i32 %t.06)
+; CHECK: call i32 @test(i32 2)
+; CHECK-NOT: call i32 @test(i32 %t.06)
+; CHECK: call i32 @test(i32 3)
+; CHECK-NOT: call i32 @test(i32 %t.06)
+
+bb.nph:
+  %0 = load i32* @id, align 4
+  %1 = add nsw i32 %0, 1
+  store i32 %1, i32* @id, align 4
+  %2 = sext i32 %k to i64
+  %3 = getelementptr inbounds [4 x i32]* @val, i64 0, i64 %2
+  store i32 %1, i32* %3, align 4
+  br label %bb
+
+bb:                                               ; preds = %bb2, %bb.nph
+  %indvar = phi i64 [ 0, %bb.nph ], [ %indvar.next, %bb2 ]
+  %scevgep = getelementptr [4 x i32]* @val, i64 0, i64 %indvar
+  %4 = load i32* %scevgep, align 4
+  %5 = icmp eq i32 %4, 0
+  br i1 %5, label %bb1, label %bb2
+
+bb1:                                              ; preds = %bb
+  %t.06 = trunc i64 %indvar to i32
+  %6 = tail call i32 @test(i32 %t.06) nounwind
+  br label %bb2
+
+bb2:                                              ; preds = %bb1, %bb
+  %indvar.next = add i64 %indvar, 1
+  %exitcond = icmp eq i64 %indvar.next, 4
+  br i1 %exitcond, label %bb4, label %bb
+
+bb4:                                              ; preds = %bb2
+  %.pre = load i32* @id, align 4
+  %7 = add nsw i32 %.pre, -1
+  store i32 %7, i32* @id, align 4
+  store i32 0, i32* %3, align 4
+  ret i32 undef
+; CHECK: }
+}