diff --git a/Android.mk b/Android.mk
index 13d765f57..c1a4956f9 100755
--- a/Android.mk
+++ b/Android.mk
@@ -20,7 +20,7 @@ LOCAL_SRC_FILES := \
 	$(wildcard $(LOCAL_PATH)/src/audio/*.c) \
 	$(wildcard $(LOCAL_PATH)/src/audio/android/*.c) \
 	$(wildcard $(LOCAL_PATH)/src/audio/dummy/*.c) \
-	$(LOCAL_PATH)/src/atomic/SDL_atomic.c \
+	$(LOCAL_PATH)/src/atomic/SDL_atomic.c.arm \
 	$(LOCAL_PATH)/src/atomic/SDL_spinlock.c.arm \
 	$(wildcard $(LOCAL_PATH)/src/core/android/*.c) \
 	$(wildcard $(LOCAL_PATH)/src/cpuinfo/*.c) \
diff --git a/include/SDL_atomic.h b/include/SDL_atomic.h
index b079896f9..6f68ebb27 100644
--- a/include/SDL_atomic.h
+++ b/include/SDL_atomic.h
@@ -149,6 +149,9 @@ void _ReadWriteBarrier(void);
  * For more information on these semantics, take a look at the blog post:
  * http://preshing.com/20120913/acquire-and-release-semantics
  */
+extern DECLSPEC void SDLCALL SDL_MemoryBarrierReleaseFunction(void);
+extern DECLSPEC void SDLCALL SDL_MemoryBarrierAcquireFunction(void);
+
 #if defined(__GNUC__) && (defined(__powerpc__) || defined(__ppc__))
 #define SDL_MemoryBarrierRelease()   __asm__ __volatile__ ("lwsync" : : : "memory")
 #define SDL_MemoryBarrierAcquire()   __asm__ __volatile__ ("lwsync" : : : "memory")
@@ -156,11 +159,11 @@ void _ReadWriteBarrier(void);
 #if defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7EM__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__)
 #define SDL_MemoryBarrierRelease()   __asm__ __volatile__ ("dmb ish" : : : "memory")
 #define SDL_MemoryBarrierAcquire()   __asm__ __volatile__ ("dmb ish" : : : "memory")
-#elif defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6T2__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__)
+#elif defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6T2__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_5TE__)
 #ifdef __thumb__
 /* The mcr instruction isn't available in thumb mode, use real functions */
-extern DECLSPEC void SDLCALL SDL_MemoryBarrierRelease(void);
-extern DECLSPEC void SDLCALL SDL_MemoryBarrierAcquire(void);
+#define SDL_MemoryBarrierRelease()   SDL_MemoryBarrierReleaseFunction()
+#define SDL_MemoryBarrierAcquire()   SDL_MemoryBarrierAcquireFunction()
 #else
 #define SDL_MemoryBarrierRelease()   __asm__ __volatile__ ("mcr p15, 0, %0, c7, c10, 5" : : "r"(0) : "memory")
 #define SDL_MemoryBarrierAcquire()   __asm__ __volatile__ ("mcr p15, 0, %0, c7, c10, 5" : : "r"(0) : "memory")
diff --git a/src/atomic/SDL_atomic.c b/src/atomic/SDL_atomic.c
index e3c17134d..828efee76 100644
--- a/src/atomic/SDL_atomic.c
+++ b/src/atomic/SDL_atomic.c
@@ -228,19 +228,16 @@ SDL_AtomicGetPtr(void **a)
     return value;
 }
 
-#ifdef __thumb__
-#if defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6T2__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__)
-__asm__(
-"   .align 2\n"
-"   .globl _SDL_MemoryBarrierRelease\n"
-"   .globl _SDL_MemoryBarrierAcquire\n"
-"_SDL_MemoryBarrierRelease:\n"
-"_SDL_MemoryBarrierAcquire:\n"
-"   mov r0, #0\n"
-"   mcr p15, 0, r0, c7, c10, 5\n"
-"   bx lr\n"
-);
-#endif
-#endif
+void
+SDL_MemoryBarrierReleaseFunction(void)
+{
+    SDL_MemoryBarrierRelease();
+}
+
+void
+SDL_MemoryBarrierAcquireFunction(void)
+{
+    SDL_MemoryBarrierAcquire();
+}
 
 /* vi: set ts=4 sw=4 expandtab: */
diff --git a/src/dynapi/SDL_dynapi_overrides.h b/src/dynapi/SDL_dynapi_overrides.h
index 4e4acecd6..942737aa7 100644
--- a/src/dynapi/SDL_dynapi_overrides.h
+++ b/src/dynapi/SDL_dynapi_overrides.h
@@ -27,12 +27,6 @@
 #error You should not be here.
 #endif
 
-/* so annoying. */
-#if defined(__thumb__) && (defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6T2__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__))
-#define SDL_MemoryBarrierRelease SDL_MemoryBarrierRelease_REAL
-#define SDL_MemoryBarrierAcquire SDL_MemoryBarrierAcquire_REAL
-#endif
-
 #define SDL_SetError SDL_SetError_REAL
 #define SDL_Log SDL_Log_REAL
 #define SDL_LogVerbose SDL_LogVerbose_REAL
@@ -627,3 +621,5 @@
 #define SDL_JoystickGetAxisInitialState SDL_JoystickGetAxisInitialState_REAL
 #define SDL_JoystickGetDeviceType SDL_JoystickGetDeviceType_REAL
 #define SDL_JoystickGetType SDL_JoystickGetType_REAL
+#define SDL_MemoryBarrierReleaseFunction SDL_MemoryBarrierReleaseFunction_REAL
+#define SDL_MemoryBarrierAcquireFunction SDL_MemoryBarrierAcquireFunction_REAL
diff --git a/src/dynapi/SDL_dynapi_procs.h b/src/dynapi/SDL_dynapi_procs.h
index bec62672a..b54cdfff4 100644
--- a/src/dynapi/SDL_dynapi_procs.h
+++ b/src/dynapi/SDL_dynapi_procs.h
@@ -60,12 +60,6 @@ SDL_DYNAPI_PROC(SDL_RWops*,SDL_RWFromFP,(FILE *a, SDL_bool b),(a,b),return)
 SDL_DYNAPI_PROC(SDL_RWops*,SDL_RWFromFP,(void *a, SDL_bool b),(a,b),return)
 #endif
 
-/* so annoying. */
-#if defined(__thumb__) && (defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6T2__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__))
-SDL_DYNAPI_PROC(void,SDL_MemoryBarrierRelease,(void),(),)
-SDL_DYNAPI_PROC(void,SDL_MemoryBarrierAcquire,(void),(),)
-#endif
-
 #ifdef __WIN32__
 SDL_DYNAPI_PROC(int,SDL_RegisterApp,(char *a, Uint32 b, void *c),(a,b,c),return)
 SDL_DYNAPI_PROC(void,SDL_UnregisterApp,(void),(),)
@@ -659,3 +653,5 @@ SDL_DYNAPI_PROC(char*,SDL_GameControllerMappingForIndex,(int a),(a),return)
 SDL_DYNAPI_PROC(SDL_bool,SDL_JoystickGetAxisInitialState,(SDL_Joystick *a, int b, Sint16 *c),(a,b,c),return)
 SDL_DYNAPI_PROC(SDL_JoystickType,SDL_JoystickGetDeviceType,(int a),(a),return)
 SDL_DYNAPI_PROC(SDL_JoystickType,SDL_JoystickGetType,(SDL_Joystick *a),(a),return)
+SDL_DYNAPI_PROC(void,SDL_MemoryBarrierReleaseFunction,(void),(),)
+SDL_DYNAPI_PROC(void,SDL_MemoryBarrierAcquireFunction,(void),(),)