Index: lib/CodeGen/TargetInfo.cpp
===================================================================
--- lib/CodeGen/TargetInfo.cpp
+++ lib/CodeGen/TargetInfo.cpp
@@ -4379,12 +4379,16 @@
   // are returned indirectly.
   uint64_t Size = getContext().getTypeSize(RetTy);
   if (Size <= 32) {
-    // Return in the smallest viable integer type.
-    if (Size <= 8)
-      return ABIArgInfo::getDirect(llvm::Type::getInt8Ty(getVMContext()));
-    if (Size <= 16)
-      return ABIArgInfo::getDirect(llvm::Type::getInt16Ty(getVMContext()));
-    return ABIArgInfo::getDirect(llvm::Type::getInt32Ty(getVMContext()));
+    if (getDataLayout().isLittleEndian()) {
+      // Return in the smallest viable integer type.
+      if (Size <= 8)
+        return ABIArgInfo::getDirect(llvm::Type::getInt8Ty(getVMContext()));
+      if (Size <= 16)
+        return ABIArgInfo::getDirect(llvm::Type::getInt16Ty(getVMContext()));
+      return ABIArgInfo::getDirect(llvm::Type::getInt32Ty(getVMContext()));
+    } else
+      // Return in 32 bit integer integer type (as if loaded by LDR, AAPCS 5.4)
+      return ABIArgInfo::getDirect(llvm::Type::getInt32Ty(getVMContext()));
   }
 
   markAllocatedGPRs(1, 1);
Index: test/CodeGen/arm-be-result-return.c
===================================================================
--- test/CodeGen/arm-be-result-return.c
+++ test/CodeGen/arm-be-result-return.c
@@ -0,0 +1,37 @@
+// RUN: %clang_cc1 -triple armebv7-arm-none-eabi -emit-llvm -w -o - %s | FileCheck %s
+
+// this tests for AAPCS section 5.4:
+// A Composite Type not larger than 4 bytes is returned in r0.
+// The format is as if the result had been stored in memory at a
+// word-aligned address and then loaded into r0 with an LDR instruction
+
+extern union Us { short s; } us;
+union Us callee_us() { return us; }
+// CHECK-LABEL: callee_us()
+// CHECK: zext i16
+// CHECK: shl 
+// CHECK: ret i32
+
+void caller_us() {
+  us = callee_us();
+// CHECK-LABEL: caller_us()
+// CHECK: call i32
+// CHECK: lshr i32
+// CHECK: trunc i32
+}
+
+extern struct Ss { short s; } ss;
+struct Ss callee_ss() { return ss; }
+// CHECK-LABEL: callee_ss()
+// CHECK: zext i16
+// CHECK: shl 
+// CHECK: ret i32
+
+void caller_ss() {
+  ss = callee_ss();
+// CHECK-LABEL: caller_ss()
+// CHECK: call i32
+// CHECK: lshr i32
+// CHECK: trunc i32
+}
+