diff --git a/Xcode/SDLTest/SDLTest.xcodeproj/project.pbxproj b/Xcode/SDLTest/SDLTest.xcodeproj/project.pbxproj
index bb95f7aae..c27f4a47d 100755
--- a/Xcode/SDLTest/SDLTest.xcodeproj/project.pbxproj
+++ b/Xcode/SDLTest/SDLTest.xcodeproj/project.pbxproj
@@ -3985,7 +3985,14 @@
 			isa = PBXProject;
 			buildConfigurationList = 001B5A0C08BDB826006539E9 /* Build configuration list for PBXProject "SDLTest" */;
 			compatibilityVersion = "Xcode 3.0";
+			developmentRegion = English;
 			hasScannedForEncodings = 1;
+			knownRegions = (
+				English,
+				Japanese,
+				French,
+				German,
+			);
 			mainGroup = 08FB7794FE84155DC02AAC07 /* SDLTest */;
 			projectDirPath = "";
 			projectReferences = (
diff --git a/configure.in b/configure.in
index 4c1ebae16..e682a4613 100644
--- a/configure.in
+++ b/configure.in
@@ -181,7 +181,7 @@ if test x$enable_dependency_tracking = xyes; then
         DEPENDENCY_TRACKING_OPTIONS="-MMD -MT \$@"
     fi
 fi
-    
+
 dnl See whether we are allowed to use the system C library
 AC_ARG_ENABLE(libc,
 AC_HELP_STRING([--enable-libc], [Use the system C library [[default=yes]]]),
@@ -285,6 +285,32 @@ if test x$have_inttypes != xyes; then
     AC_DEFINE(uintptr_t, unsigned long)
 fi
 
+dnl See whether we can use gcc atomic operations on this architecture
+AC_ARG_ENABLE(gcc-atomics,
+AC_HELP_STRING([--enable-gcc-atomics],
+               [Use gcc builtin atomics [[default=yes]]]),
+              , enable_gcc_atomics=yes)
+if test x$enable_gcc_atomics = xyes; then
+    have_gcc_atomics=no
+    AC_MSG_CHECKING(for GCC builtin atomic operations)
+    AC_TRY_LINK([
+    ],[
+    int a;
+    void *x, *y, *z;
+    __sync_lock_test_and_set(&a, 4);
+    __sync_fetch_and_add(&a, 1);
+    __sync_bool_compare_and_swap(&a, 5, 10);
+    __sync_bool_compare_and_swap(&x, y, z);
+    ],[
+    have_gcc_atomics=yes
+    ])
+    AC_MSG_RESULT($have_gcc_atomics)
+
+    if test x$have_gcc_mmd_mt = xyes; then
+        AC_DEFINE(HAVE_GCC_ATOMICS)
+    fi
+fi
+
 # Standard C sources
 SOURCES="$SOURCES $srcdir/src/*.c"
 SOURCES="$SOURCES $srcdir/src/audio/*.c"
diff --git a/include/SDL_atomic.h b/include/SDL_atomic.h
index df80c3c68..1c8cf28e6 100644
--- a/include/SDL_atomic.h
+++ b/include/SDL_atomic.h
@@ -108,7 +108,62 @@ extern DECLSPEC void SDLCALL SDL_AtomicUnlock(SDL_SpinLock *lock);
 /*@}*//*SDL AtomicLock*/
 
 /* Platform specific optimized versions of the atomic functions */
-/* None yet... */
+#if defined(__WIN32__)
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+#define SDL_AtomicSet(a, v)     InterlockedExchange(&(a)->value, v)
+#define SDL_AtomicGet(a)        ((a)->value)
+#define SDL_AtomicAdd(a, v)     InterlockedAdd(&(a)->value, v)
+#define SDL_AtomicCAS(a, oldval, newval) (InterlockedCompareExchange(&(a)->value, newval, oldval) == (oldval))
+#define SDL_AtomicSetPtr(a, v)  InterlockedExchangePointer(a, v)
+#define SDL_AtomicGetPtr(a)     (*(a))
+#define SDL_AtomicCASPtr(a, oldval, newval) (InterlockedCompareExchangePointer(a, newval, oldval) == (oldval))
+
+#elif defined(__MACOSX__)
+#include <libkern/OSAtomic.h>
+
+#define SDL_AtomicSet(a, v) \
+({                          \
+    int oldvalue;           \
+                            \
+    do {                    \
+        oldvalue = (a)->value; \
+    } while (!OSAtomicCompareAndSwap32Barrier(oldvalue, v, &(a)->value)); \
+                            \
+    oldvalue;               \
+})
+#define SDL_AtomicGet(a)        ((a)->value)
+#define SDL_AtomicAdd(a, v) \
+({                          \
+    int oldvalue;           \
+                            \
+    do {                    \
+        oldvalue = (a)->value; \
+    } while (!OSAtomicCompareAndSwap32Barrier(oldvalue, oldvalue+v, &(a)->value)); \
+                            \
+    oldvalue;               \
+})
+#define SDL_AtomicCAS(a, oldval, newval) OSAtomicCompareAndSwap32Barrier(oldval, newval, &(a)->value)
+#define SDL_AtomicSetPtr(a, v)  (*(a) = v, OSMemoryBarrier())
+#define SDL_AtomicGetPtr(a)     (*(a))
+#if SIZEOF_VOIDP == 4
+#define SDL_AtomicCASPtr(a, oldval, newval) OSAtomicCompareAndSwap32Barrier((int32_t)(oldval), (int32_t)(newval), (int32_t*)(a))
+#elif SIZEOF_VOIDP == 8
+#define SDL_AtomicCASPtr(a, oldval, newval) OSAtomicCompareAndSwap64Barrier((int64_t)(oldval), (int64_t)(newval), (int64_t*)(a))
+#endif
+
+#elif defined(HAVE_GCC_ATOMICS)
+
+#define SDL_AtomicSet(a, v)     __sync_lock_test_and_set(&(a)->value, v)
+#define SDL_AtomicGet(a)        ((a)->value)
+#define SDL_AtomicAdd(a, v)     __sync_fetch_and_add(&(a)->value, v)
+#define SDL_AtomicCAS(a, oldval, newval) __sync_bool_compare_and_swap(&(a)->value, oldval, newval)
+#define SDL_AtomicSetPtr(a, v)  (*(a) = v, __sync_synchronize())
+#define SDL_AtomicGetPtr(a)     (*(a))
+#define SDL_AtomicCASPtr(a, oldval, newval) __sync_bool_compare_and_swap(a, oldval, newval)
+
+#endif
 
 /**
  * \brief A type representing an atomic integer value.  It is a struct
@@ -163,12 +218,12 @@ extern DECLSPEC SDL_bool SDLCALL SDL_AtomicDecRef(SDL_atomic_t *a);
 /**
  * \brief Set an atomic variable to a new value if it is currently an old value.
  *
- * \return The previous value of the atomic variable
+ * \return SDL_TRUE if the atomic variable was set, SDL_FALSE otherwise.
  *
  * \note If you don't know what this function is for, you shouldn't use it!
 */
 #ifndef SDL_AtomicCAS
-extern DECLSPEC int SDLCALL SDL_AtomicCAS(SDL_atomic_t *a, int oldval, int newval);
+extern DECLSPEC SDL_bool SDLCALL SDL_AtomicCAS(SDL_atomic_t *a, int oldval, int newval);
 #endif
 
 /**
@@ -188,12 +243,12 @@ extern DECLSPEC void* SDLCALL SDL_AtomicGetPtr(void** a);
 /**
  * \brief Set a pointer to a new value if it is currently an old value.
  *
- * \return The previous value of the pointer
+ * \return SDL_TRUE if the pointer was set, SDL_FALSE otherwise.
  *
  * \note If you don't know what this function is for, you shouldn't use it!
 */
 #ifndef SDL_AtomicCASPtr
-extern DECLSPEC void* SDLCALL SDL_AtomicCASPtr(void **a, void *oldval, void *newval);
+extern DECLSPEC SDL_bool SDLCALL SDL_AtomicCASPtr(void **a, void *oldval, void *newval);
 #endif
 
 /* Ends C function definitions when using C++ */
diff --git a/include/SDL_config.h.in b/include/SDL_config.h.in
index f99ecb763..1cff0dc69 100644
--- a/include/SDL_config.h.in
+++ b/include/SDL_config.h.in
@@ -58,6 +58,7 @@
 
 #undef SIZEOF_VOIDP
 #undef SDL_HAS_64BIT_TYPE
+#undef HAVE_GCC_ATOMICS
 
 /* Comment this if you want to build without any C library requirements */
 #undef HAVE_LIBC
diff --git a/include/SDL_config_iphoneos.h b/include/SDL_config_iphoneos.h
index 5b27790c0..bc3d876fc 100644
--- a/include/SDL_config_iphoneos.h
+++ b/include/SDL_config_iphoneos.h
@@ -43,6 +43,8 @@ typedef unsigned long uintptr_t;
 
 #define SDL_HAS_64BIT_TYPE	1
 
+#define HAVE_GCC_ATOMICS	1
+
 #define HAVE_ALLOCA_H		1
 #define HAVE_SYS_TYPES_H	1
 #define HAVE_STDIO_H	1
diff --git a/src/atomic/SDL_atomic.c b/src/atomic/SDL_atomic.c
index 320c1dd71..50cc7da36 100644
--- a/src/atomic/SDL_atomic.c
+++ b/src/atomic/SDL_atomic.c
@@ -23,6 +23,12 @@
 
 #include "SDL_atomic.h"
 
+/* Note that we undefine the atomic operations here, in case they are
+   defined as compiler intrinsics while building SDL but the library user
+   doesn't have that compiler.  That way we always have a working set of
+   atomic operations built into the library.
+*/
+ 
 /* 
   If any of the operations are not provided then we must emulate some
   of them. That means we need a nice implementation of spin locks
@@ -51,20 +57,20 @@ static SDL_SpinLock locks[32];
 static __inline__ void
 enterLock(void *a)
 {
-   uintptr_t index = ((((uintptr_t)a) >> 3) & 0x1f);
+    uintptr_t index = ((((uintptr_t)a) >> 3) & 0x1f);
 
-   SDL_AtomicLock(&locks[index]);
+    SDL_AtomicLock(&locks[index]);
 }
 
 static __inline__ void
 leaveLock(void *a)
 {
-   uintptr_t index = ((((uintptr_t)a) >> 3) & 0x1f);
+    uintptr_t index = ((((uintptr_t)a) >> 3) & 0x1f);
 
-   SDL_AtomicUnlock(&locks[index]);
+    SDL_AtomicUnlock(&locks[index]);
 }
 
-#ifndef SDL_AtomicSet
+#undef SDL_AtomicSet
 int
 SDL_AtomicSet(SDL_atomic_t *a, int value)
 {
@@ -77,9 +83,8 @@ SDL_AtomicSet(SDL_atomic_t *a, int value)
 
     return oldvalue;
 }
-#endif
 
-#ifndef SDL_AtomicGet
+#undef SDL_AtomicGet
 int
 SDL_AtomicGet(SDL_atomic_t *a)
 {
@@ -88,9 +93,8 @@ SDL_AtomicGet(SDL_atomic_t *a)
     */
     return a->value;
 }
-#endif
 
-#ifndef SDL_AtomicAdd
+#undef SDL_AtomicAdd
 int
 SDL_AtomicAdd(SDL_atomic_t *a, int value)
 {
@@ -103,53 +107,48 @@ SDL_AtomicAdd(SDL_atomic_t *a, int value)
 
     return oldvalue;
 }
-#endif
 
-#ifndef SDL_AtomicIncRef
+#undef SDL_AtomicIncRef
 void
 SDL_AtomicIncRef(SDL_atomic_t *a)
 {
     SDL_AtomicAdd(a, 1);
 }
-#endif
 
-#ifndef SDL_AtomicDecRef
+#undef SDL_AtomicDecRef
 SDL_bool
 SDL_AtomicDecRef(SDL_atomic_t *a)
 {
     return SDL_AtomicAdd(a, -1) == 1;
 }
-#endif
 
-#ifndef SDL_AtomicCAS
-int
+#undef SDL_AtomicCAS
+SDL_bool
 SDL_AtomicCAS(SDL_atomic_t *a, int oldval, int newval)
 {
-    int prevval;
+    SDL_bool retval = SDL_FALSE;
 
     enterLock(a);
-    prevval = a->value;
-    if (prevval == oldval) {
+    if (a->value == oldval) {
         a->value = newval;
+        retval = SDL_TRUE;
     }
     leaveLock(a);
 
-    return prevval;
+    return retval;
 }
-#endif
 
-#ifndef SDL_AtomicSetPtr
+#undef SDL_AtomicSetPtr
 void
 SDL_AtomicSetPtr(void** a, void* value)
 {
     void *prevval;
     do {
         prevval = *a;
-    } while (SDL_AtomicCASPtr(a, prevval, value) != prevval);
+    } while (!SDL_AtomicCASPtr(a, prevval, value));
 }
-#endif
 
-#ifndef SDL_AtomicGetPtr
+#undef SDL_AtomicGetPtr
 void*
 SDL_AtomicGetPtr(void** a)
 {
@@ -158,22 +157,20 @@ SDL_AtomicGetPtr(void** a)
     */
     return *a;
 }
-#endif
 
-#ifndef SDL_AtomicCASPtr
-void* SDL_AtomicCASPtr(void **a, void *oldval, void *newval)
+#undef SDL_AtomicCASPtr
+SDL_bool SDL_AtomicCASPtr(void **a, void *oldval, void *newval)
 {
-    void *prevval;
+    SDL_bool retval = SDL_FALSE;
 
     enterLock(a);
-    prevval = *a;
     if (*a == oldval) {
         *a = newval;
+        retval = SDL_TRUE;
     }
     leaveLock(a);
 
-    return prevval;
+    return retval;
 }
-#endif
 
 /* vi: set ts=4 sw=4 expandtab: */
diff --git a/src/atomic/SDL_spinlock.c b/src/atomic/SDL_spinlock.c
index d0631f835..9fbbfd448 100644
--- a/src/atomic/SDL_spinlock.c
+++ b/src/atomic/SDL_spinlock.c
@@ -44,24 +44,22 @@ SDL_AtomicTryLock(SDL_SpinLock *lock)
 #elif defined(__MACOSX__)
     return OSAtomicCompareAndSwap32Barrier(0, 1, lock);
 
-#elif defined(__GNUC__)
-#if defined(__arm__)
-#ifdef __ARM_ARCH_5__
+#elif defined(HAVE_GCC_ATOMICS)
+    return (__sync_lock_test_and_set(lock, 1) == 0);
+
+#elif defined(__GNUC__) && defined(__arm__) && defined(__ARM_ARCH_5__)
     int result;
     __asm__ __volatile__ (
         "swp %0, %1, [%2]\n"
         : "=&r,&r" (result) : "r,0" (1), "r,r" (lock) : "memory");
     return (result == 0);
-#else
+
+#elif defined(__GNUC__) && defined(__arm__)
     int result;
     __asm__ __volatile__ (
         "ldrex %0, [%2]\nteq   %0, #0\nstrexeq %0, %1, [%2]"
         : "=&r" (result) : "r" (1), "r" (lock) : "cc", "memory");
     return (result == 0);
-#endif
-#else
-    return (__sync_lock_test_and_set(lock, 1) == 0);
-#endif
 
 #else
     /* Need CPU instructions for spinlock here! */
@@ -81,19 +79,8 @@ SDL_AtomicLock(SDL_SpinLock *lock)
 void
 SDL_AtomicUnlock(SDL_SpinLock *lock)
 {
-#if defined(__WIN32__)
+    /* Assuming atomic assignment operation and full memory barrier in lock */
     *lock = 0;
-
-#elif defined(__MACOSX__)
-    *lock = 0;
-
-#elif defined(__GNUC__) && !defined(__arm__)
-    __sync_lock_release(lock);
-
-#else
-    /* Assuming memory barrier in lock and integral assignment operation */
-    *lock = 0;
-#endif
 }
 
 /* vi: set ts=4 sw=4 expandtab: */
diff --git a/test/testatomic.c b/test/testatomic.c
index 85f7b567e..bf845d129 100644
--- a/test/testatomic.c
+++ b/test/testatomic.c
@@ -1,11 +1,13 @@
 #include <stdio.h>
 #include "SDL.h"
+#include "SDL_assert.h"
 
 /*
   Absolutely basic tests just to see if we get the expected value
   after calling each function.
 */
 
+static
 char *
 tf(SDL_bool tf)
 {
@@ -20,8 +22,8 @@ tf(SDL_bool tf)
     return f;
 }
 
-int
-main(int argc, char *argv[])
+static
+void RunBasicTest()
 {
     int value;
     SDL_SpinLock lock = 0;
@@ -58,11 +60,173 @@ main(int argc, char *argv[])
     printf("AtomicDecRef()       tfret=%s val=%"PRIu32"\n", tf(tfret), SDL_AtomicGet(&v));
 
     SDL_AtomicSet(&v, 10);
-    tfret = (SDL_AtomicCAS(&v, 0, 20) != 0);
+    tfret = (SDL_AtomicCAS(&v, 0, 20) == SDL_FALSE);
     printf("AtomicCAS()          tfret=%s val=%"PRIu32"\n", tf(tfret), SDL_AtomicGet(&v));
     value = SDL_AtomicGet(&v);
-    tfret = (SDL_AtomicCAS(&v, value, 20) == value);
+    tfret = (SDL_AtomicCAS(&v, value, 20) == SDL_TRUE);
     printf("AtomicCAS()          tfret=%s val=%"PRIu32"\n", tf(tfret), SDL_AtomicGet(&v));
+}
 
+/* Atomic operation test, adapted from code by Michael Davidsaver at:
+    http://bazaar.launchpad.net/~mdavidsaver/epics-base/atomic/revision/12105#src/libCom/test/epicsAtomicTest.c
+*/
+
+/* Tests semantics of atomic operations.  Also a stress test
+ * to see if they are really atomic.
+ *
+ * Serveral threads adding to the same variable.
+ * at the end the value is compared with the expected
+ * and with a non-atomic counter.
+ */
+ 
+/* Number of concurrent incrementers */
+#define NThreads 2
+#define CountInc 100
+#define VALBITS (sizeof(atomicValue)*8)
+ 
+#define atomicValue int
+#define CountTo ((atomicValue)((unsigned int)(1<<(VALBITS-1))-1))
+#define NInter (CountTo/CountInc/NThreads)
+#define Expect (CountTo-NInter*CountInc*NThreads)
+ 
+SDL_COMPILE_TIME_ASSERT(size, CountTo>0); /* check for rollover */
+ 
+static SDL_atomic_t good = { 42 };
+ 
+static atomicValue bad = 42;
+ 
+static SDL_atomic_t threadsRunning;
+
+static SDL_sem *threadDone;
+ 
+static
+int adder(void* junk)
+{
+    unsigned long N=NInter;
+    printf("Thread subtracting %d %lu times\n",CountInc,N);
+    while (N--) {
+        SDL_AtomicAdd(&good, -CountInc);
+        bad-=CountInc;
+    }
+    SDL_AtomicAdd(&threadsRunning, -1);
+    SDL_SemPost(threadDone);
+    return 0;
+}
+ 
+static
+void runAdder(void)
+{
+    Uint32 start, end;
+    int T=NThreads;
+ 
+    start = SDL_GetTicks();
+ 
+    threadDone = SDL_CreateSemaphore(0);
+
+    SDL_AtomicSet(&threadsRunning, NThreads);
+
+    while (T--)
+        SDL_CreateThread(adder, NULL);
+ 
+    while (SDL_AtomicGet(&threadsRunning) > 0)
+        SDL_SemWait(threadDone);
+ 
+    SDL_DestroySemaphore(threadDone);
+
+    end = SDL_GetTicks();
+ 
+    printf("Finished in %f sec\n", (end - start) / 1000.f);
+}
+ 
+static
+void RunEpicTest()
+{
+    int b;
+    atomicValue v;
+ 
+    printf("\nepic test---------------------------------------\n\n");
+
+    printf("Size asserted to be >= 32-bit\n");
+    SDL_assert(sizeof(atomicValue)>=4);
+ 
+    printf("Check static initializer\n");
+    v=SDL_AtomicGet(&good);
+    SDL_assert(v==42);
+ 
+    SDL_assert(bad==42);
+ 
+    printf("Test negative values\n");
+    SDL_AtomicSet(&good, -5);
+    v=SDL_AtomicGet(&good);
+    SDL_assert(v==-5);
+ 
+    printf("Verify maximum value\n");
+    SDL_AtomicSet(&good, CountTo);
+    v=SDL_AtomicGet(&good);
+    SDL_assert(v==CountTo);
+ 
+    printf("Test compare and exchange\n");
+ 
+    b=SDL_AtomicCAS(&good, 500, 43);
+    SDL_assert(!b); /* no swap since CountTo!=500 */
+    v=SDL_AtomicGet(&good);
+    SDL_assert(v==CountTo); /* ensure no swap */
+ 
+    b=SDL_AtomicCAS(&good, CountTo, 44);
+    SDL_assert(!!b); /* will swap */
+    v=SDL_AtomicGet(&good);
+    SDL_assert(v==44);
+ 
+    printf("Test Add\n");
+ 
+    v=SDL_AtomicAdd(&good, 1);
+    SDL_assert(v==44);
+    v=SDL_AtomicGet(&good);
+    SDL_assert(v==45);
+ 
+    v=SDL_AtomicAdd(&good, 10);
+    SDL_assert(v==45);
+    v=SDL_AtomicGet(&good);
+    SDL_assert(v==55);
+ 
+    printf("Test Add (Negative values)\n");
+ 
+    v=SDL_AtomicAdd(&good, -20);
+    SDL_assert(v==55);
+    v=SDL_AtomicGet(&good);
+    SDL_assert(v==35);
+ 
+    v=SDL_AtomicAdd(&good, -50); /* crossing zero down */
+    SDL_assert(v==35);
+    v=SDL_AtomicGet(&good);
+    SDL_assert(v==-15);
+ 
+    v=SDL_AtomicAdd(&good, 30); /* crossing zero up */
+    SDL_assert(v==-15);
+    v=SDL_AtomicGet(&good);
+    SDL_assert(v==15);
+ 
+    printf("Reset before count down test\n");
+    SDL_AtomicSet(&good, CountTo);
+    v=SDL_AtomicGet(&good);
+    SDL_assert(v==CountTo);
+ 
+    bad=CountTo;
+    SDL_assert(bad==CountTo);
+ 
+    printf("Counting down from %d, Expect %d remaining\n",CountTo,Expect);
+    runAdder();
+ 
+    v=SDL_AtomicGet(&good);
+    printf("Atomic %d Non-Atomic %d\n",v,bad);
+    SDL_assert(v==Expect);
+    SDL_assert(bad!=Expect);
+}
+
+int
+main(int argc, char *argv[])
+{
+    RunBasicTest();
+    RunEpicTest();
     return 0;
 }