Improve Atomic.h:

- For GCC, use intrinsics that will work on ARM.
- Add AtomicExchangeAcquire.
- Make Atomic{Load,LoadAcquire,Store,StoreRelease} work for any suitable type.

Source/Core/Common/Src/Atomic_GCC.h

@@ -40,27 +40,45 @@ inline void AtomicIncrement(volatile u32& target) {
 	__sync_add_and_fetch(&target, 1);
 }
 
-inline u32 AtomicLoad(volatile u32& src) {
-	return src; // 32-bit reads are always atomic.
-}
-inline u32 AtomicLoadAcquire(volatile u32& src) {
-	//keep the compiler from caching any memory references
-	u32 result = src; // 32-bit reads are always atomic.
-	//__sync_synchronize(); // TODO: May not be necessary.
-	// Compiler instruction only. x86 loads always have acquire semantics.
-	__asm__ __volatile__ ( "":::"memory" );
-	return result;
-}
-
 inline void AtomicOr(volatile u32& target, u32 value) {
 	__sync_or_and_fetch(&target, value);
 }
 
-inline void AtomicStore(volatile u32& dest, u32 value) {
+#ifdef __clang__
-	dest = value; // 32-bit writes are always atomic.
+template <typename T>
+_Atomic(T)* ToC11Atomic(volatile T* loc)
+{
+	return (_Atomic(T)*) loc;
 }
-inline void AtomicStoreRelease(volatile u32& dest, u32 value) {
+
-	__sync_lock_test_and_set(&dest, value); // TODO: Wrong! This function is has acquire semantics.
+#define __atomic_load_n(p, m) __c11_atomic_load(ToC11Atomic(p), m)
+#define __atomic_store_n(p, v, m) __c11_atomic_store(ToC11Atomic(p), v, m)
+#define __atomic_exchange_n(p, v, m) __c11_atomic_exchange(ToC11Atomic(p), v, m)
+#endif
+
+template <typename T>
+inline T AtomicLoad(volatile T& src) {
+	return __atomic_load_n(&src, __ATOMIC_RELAXED);
+}
+
+template <typename T>
+inline T AtomicLoadAcquire(volatile T& src) {
+	return __atomic_load_n(&src, __ATOMIC_ACQUIRE);
+}
+
+template <typename T, typename U>
+inline void AtomicStore(volatile T& dest, U value) {
+	__atomic_store_n(&dest, value, __ATOMIC_RELAXED);
+}
+
+template <typename T, typename U>
+inline void AtomicStoreRelease(volatile T& dest, U value) {
+	__atomic_store_n(&dest, value, __ATOMIC_RELEASE);
+}
+
+template <typename T, typename U>
+inline T* AtomicExchangeAcquire(T* volatile& loc, U newval) {
+	return __atomic_exchange_n(&loc, newval, __ATOMIC_ACQ_REL);
 }
 
 }
@@ -74,9 +92,9 @@ LONG SyncInterlockedIncrement(LONG *Dest)
 #else
   register int result;
   __asm__ __volatile__("lock; xadd %0,%1"
-                       : "=r" (result), "=m" (*Dest)
+					   : "=r" (result), "=m" (*Dest)
-                       : "0" (1), "m" (*Dest)
+					   : "0" (1), "m" (*Dest)
-                       : "memory");
+					   : "memory");
   return result;
 #endif
 }
@@ -88,9 +106,9 @@ LONG SyncInterlockedExchangeAdd(LONG *Dest, LONG Val)
 #else
   register int result;
   __asm__ __volatile__("lock; xadd %0,%1"
-                       : "=r" (result), "=m" (*Dest)
+					   : "=r" (result), "=m" (*Dest)
-                       : "0" (Val), "m" (*Dest)
+					   : "0" (Val), "m" (*Dest)
-                       : "memory");
+					   : "memory");
   return result;
 #endif
 }
@@ -102,9 +120,9 @@ LONG SyncInterlockedExchange(LONG *Dest, LONG Val)
 #else
   register int result;
   __asm__ __volatile__("lock; xchg %0,%1"
-                       : "=r" (result), "=m" (*Dest)
+					   : "=r" (result), "=m" (*Dest)
-                       : "0" (Val), "m" (*Dest)
+					   : "0" (Val), "m" (*Dest)
-                       : "memory");
+					   : "memory");
   return result;
 #endif
 }

Source/Core/Common/Src/Atomic_Win32.h

@@ -31,7 +31,7 @@ namespace Common
 {
 
 inline void AtomicAdd(volatile u32& target, u32 value) {
-	InterlockedExchangeAdd((volatile LONG*)&target, (LONG)value);
+	_InterlockedExchangeAdd((volatile LONG*)&target, (LONG)value);
 }
 
 inline void AtomicAnd(volatile u32& target, u32 value) {
@@ -39,32 +39,43 @@ inline void AtomicAnd(volatile u32& target, u32 value) {
 }
 
 inline void AtomicIncrement(volatile u32& target) {
-	InterlockedIncrement((volatile LONG*)&target);
+	_InterlockedIncrement((volatile LONG*)&target);
 }
 
 inline void AtomicDecrement(volatile u32& target) {
-	InterlockedDecrement((volatile LONG*)&target);
+	_InterlockedDecrement((volatile LONG*)&target);
-}
-
-inline u32 AtomicLoad(volatile u32& src) {
-	return src; // 32-bit reads are always atomic.
-}
-inline u32 AtomicLoadAcquire(volatile u32& src) {
-	u32 result = src; // 32-bit reads are always atomic.
-	_ReadBarrier(); // Compiler instruction only. x86 loads always have acquire semantics.
-	return result;
 }
 
 inline void AtomicOr(volatile u32& target, u32 value) {
 	_InterlockedOr((volatile LONG*)&target, (LONG)value);
 }
 
-inline void AtomicStore(volatile u32& dest, u32 value) {
+template <typename T>
-	dest = value; // 32-bit writes are always atomic.
+inline T AtomicLoad(volatile T& src) {
+	return src; // 32-bit reads are always atomic.
 }
-inline void AtomicStoreRelease(volatile u32& dest, u32 value) {
+
+template <typename T>
+inline T AtomicLoadAcquire(volatile T& src) {
+	T result = src; // 32-bit reads are always atomic.
+	_ReadBarrier(); // Compiler instruction only. x86 loads always have acquire semantics.
+	return result;
+}
+
+template <typename T, typename U>
+inline void AtomicStore(volatile T& dest, U value) {
+	dest = (T) value; // 32-bit writes are always atomic.
+}
+
+template <typename T, typename U>
+inline void AtomicStoreRelease(volatile T& dest, U value) {
 	_WriteBarrier(); // Compiler instruction only. x86 stores always have release semantics.
-	dest = value; // 32-bit writes are always atomic.
+	dest = (T) value; // 32-bit writes are always atomic.
+}
+
+template <typename T, typename U>
+inline T* AtomicExchangeAcquire(T* volatile& loc, U newval) {
+	return (T*) _InterlockedExchangePointer_acq((void* volatile*) &loc, (void*) newval);
 }
 
 }