diff --git a/lib/Transforms/Scalar/ScalarReplAggregates.cpp b/lib/Transforms/Scalar/ScalarReplAggregates.cpp
index 8178c270759..5f8ab517df9 100644
--- a/lib/Transforms/Scalar/ScalarReplAggregates.cpp
+++ b/lib/Transforms/Scalar/ScalarReplAggregates.cpp
@@ -2481,19 +2481,22 @@ static bool isOnlyCopiedFromConstantGlobal(Value *V, MemTransferInst *&TheCopy,
     }
 
     if (CallSite CS = U) {
-      // If this is a readonly/readnone call site, then we know it is just a
-      // load and we can ignore it.
-      if (CS.onlyReadsMemory())
-        continue;
-
       // If this is the function being called then we treat it like a load and
       // ignore it.
       if (CS.isCallee(UI))
         continue;
 
+      // If this is a readonly/readnone call site, then we know it is just a
+      // load (but one that potentially returns the value itself), so we can
+      // ignore it if we know that the value isn't captured.
+      unsigned ArgNo = CS.getArgumentNo(UI);
+      if (CS.onlyReadsMemory() &&
+          (CS.getInstruction()->use_empty() ||
+           CS.paramHasAttr(ArgNo+1, Attribute::NoCapture)))
+        continue;
+
       // If this is being passed as a byval argument, the caller is making a
       // copy, so it is only a read of the alloca.
-      unsigned ArgNo = CS.getArgumentNo(UI);
       if (CS.paramHasAttr(ArgNo+1, Attribute::ByVal))
         continue;
     }
diff --git a/test/Transforms/ScalarRepl/2011-05-06-CapturedAlloca.ll b/test/Transforms/ScalarRepl/2011-05-06-CapturedAlloca.ll
new file mode 100644
index 00000000000..816cb60e82a
--- /dev/null
+++ b/test/Transforms/ScalarRepl/2011-05-06-CapturedAlloca.ll
@@ -0,0 +1,26 @@
+; RUN: opt < %s -instcombine -S | FileCheck %s
+; PR9820
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
+target triple = "x86_64-unknown-linux-gnu"
+
+@func_1.l_10 = internal unnamed_addr constant [4 x i32] [i32 1, i32 0, i32 0, i32 0], align 16
+
+define i32* @noop(i32* %p_29) nounwind readnone {
+entry:
+  ret i32* %p_29
+}
+
+define i32 @main() nounwind {
+entry:
+  %l_10 = alloca [4 x i32], align 16
+  %tmp = bitcast [4 x i32]* %l_10 to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %tmp, i8* bitcast ([4 x i32]* @func_1.l_10 to i8*), i64 16, i32 16, i1 false)
+; CHECK: call void @llvm.memcpy
+  %arrayidx = getelementptr inbounds [4 x i32]* %l_10, i64 0, i64 0
+  %call = call i32* @noop(i32* %arrayidx)
+  store i32 0, i32* %call
+  ret i32 0
+}
+
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) nounwind