Fix Windows atomic ops to return pre-operation value (match Linux/C11 semantics) (#812)

* Initial plan

* fix: use InterlockedExchangeAdd for Windows atomic ops to return old value

Replace InterlockedAdd/InterlockedIncrement/InterlockedDecrement with
InterlockedExchangeAdd which returns the value before the operation,
matching the behavior of __sync_fetch_and_add (GCC) and
atomic_fetch_add (C11).

Also fix ATOMIC_SUB macro parenthesization: -(n) instead of -n.

Co-authored-by: ithewei <26049660+ithewei@users.noreply.github.com>

* fix: prefer GCC builtins over Interlocked for MinGW compatibility

Reorder #ifdef checks so GCC-based compilers (including MinGW) use
__sync_fetch_and_add builtins via a direct __GNUC__ version check,
instead of relying on __GNUC_PREREQ which is broken on MinGW.

Only pure MSVC now uses InterlockedExchangeAdd for atomic operations.
This completely avoids any MinGW compatibility concerns with Windows
Interlocked functions.

Co-authored-by: ithewei <26049660+ithewei@users.noreply.github.com>

---------

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: ithewei <26049660+ithewei@users.noreply.github.com>

Author: Copilot

base/hatomic.h

@@ -44,30 +44,30 @@ typedef volatile size_t             atomic_size_t;
 
 typedef struct atomic_flag { atomic_long _Value; } atomic_flag;
 
-#ifdef _WIN32
+#if defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 1))
 
 #define ATOMIC_FLAG_TEST_AND_SET    atomic_flag_test_and_set
 static inline bool atomic_flag_test_and_set(atomic_flag* p) {
-    // return InterlockedIncrement((LONG*)&p->_Value, 1);
-    return InterlockedCompareExchange(&p->_Value, 1, 0);
+    return !__sync_bool_compare_and_swap(&p->_Value, 0, 1);
 }
 
-#define ATOMIC_ADD          InterlockedAdd
-#define ATOMIC_SUB(p, n)    InterlockedAdd(p, -n)
-#define ATOMIC_INC          InterlockedIncrement
-#define ATOMIC_DEC          InterlockedDecrement
+#define ATOMIC_ADD          __sync_fetch_and_add
+#define ATOMIC_SUB          __sync_fetch_and_sub
+#define ATOMIC_INC(p)       ATOMIC_ADD(p, 1)
+#define ATOMIC_DEC(p)       ATOMIC_SUB(p, 1)
 
-#elif __GNUC_PREREQ(4, 1)
+#elif defined(_WIN32)
 
 #define ATOMIC_FLAG_TEST_AND_SET    atomic_flag_test_and_set
 static inline bool atomic_flag_test_and_set(atomic_flag* p) {
-    return !__sync_bool_compare_and_swap(&p->_Value, 0, 1);
+    // return InterlockedIncrement((LONG*)&p->_Value, 1);
+    return InterlockedCompareExchange(&p->_Value, 1, 0);
 }
 
-#define ATOMIC_ADD          __sync_fetch_and_add
-#define ATOMIC_SUB          __sync_fetch_and_sub
-#define ATOMIC_INC(p)       ATOMIC_ADD(p, 1)
-#define ATOMIC_DEC(p)       ATOMIC_SUB(p, 1)
+#define ATOMIC_ADD          InterlockedExchangeAdd
+#define ATOMIC_SUB(p, n)    InterlockedExchangeAdd(p, -(n))
+#define ATOMIC_INC(p)       InterlockedExchangeAdd(p, 1)
+#define ATOMIC_DEC(p)       InterlockedExchangeAdd(p, -1)
 
 #endif