bpo-39465: Add pycore_atomic_funcs.h header (GH-20766)

Add pycore_atomic_funcs.h internal header file: similar to
pycore_atomic.h but don't require to declare variables as atomic.

Add _Py_atomic_size_get() and _Py_atomic_size_set() functions.

2 files changed, 97 insertions, 3 deletions

diff --git a/Include/internal/pycore_atomic.h b/Include/internal/pycore_atomic.h
index 1d5c5621677..3d42e54464c 100644
--- a/Include/internal/pycore_atomic.h
+++ b/Include/internal/pycore_atomic.h
@@ -11,8 +11,8 @@ extern "C" {
 #include "dynamic_annotations.h"   /* _Py_ANNOTATE_MEMORY_ORDER */
 #include "pyconfig.h"
 
-#if defined(HAVE_STD_ATOMIC)
-#include <stdatomic.h>
+#ifdef HAVE_STD_ATOMIC
+#  include <stdatomic.h>
 #endif
 
 
@@ -62,7 +62,7 @@ typedef struct _Py_atomic_int {
 #define _Py_atomic_load_explicit(ATOMIC_VAL, ORDER) \
     atomic_load_explicit(&((ATOMIC_VAL)->_value), ORDER)
 
-/* Use builtin atomic operations in GCC >= 4.7 */
+// Use builtin atomic operations in GCC >= 4.7 and clang
 #elif defined(HAVE_BUILTIN_ATOMIC)
 
 typedef enum _Py_memory_order {
diff --git a/Include/internal/pycore_atomic_funcs.h b/Include/internal/pycore_atomic_funcs.h
new file mode 100644
index 00000000000..a708789cea7
--- /dev/null
+++ b/Include/internal/pycore_atomic_funcs.h
@@ -0,0 +1,94 @@
+/* Atomic functions: similar to pycore_atomic.h, but don't need
+   to declare variables as atomic.
+
+   Py_ssize_t type:
+
+   * value = _Py_atomic_size_get(&var)
+   * _Py_atomic_size_set(&var, value)
+
+   Use sequentially-consistent ordering (__ATOMIC_SEQ_CST memory order):
+   enforce total ordering with all other atomic functions.
+*/
+#ifndef Py_ATOMIC_FUNC_H
+#define Py_ATOMIC_FUNC_H
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef Py_BUILD_CORE
+#  error "this header requires Py_BUILD_CORE define"
+#endif
+
+#if defined(_MSC_VER)
+#  include <intrin.h>             // _InterlockedExchange()
+#endif
+
+
+// Use builtin atomic operations in GCC >= 4.7 and clang
+#ifdef HAVE_BUILTIN_ATOMIC
+
+static inline Py_ssize_t _Py_atomic_size_get(Py_ssize_t *var)
+{
+    return __atomic_load_n(var, __ATOMIC_SEQ_CST);
+}
+
+static inline void _Py_atomic_size_set(Py_ssize_t *var, Py_ssize_t value)
+{
+    __atomic_store_n(var, value, __ATOMIC_SEQ_CST);
+}
+
+#elif defined(_MSC_VER)
+
+static inline Py_ssize_t _Py_atomic_size_get(Py_ssize_t *var)
+{
+#if SIZEOF_VOID_P == 8
+    Py_BUILD_ASSERT(sizeof(__int64) == sizeof(*var));
+    volatile __int64 *volatile_var = (volatile __int64 *)var;
+    __int64 old;
+    do {
+        old = *volatile_var;
+    } while(_InterlockedCompareExchange64(volatile_var, old, old) != old);
+#else
+    Py_BUILD_ASSERT(sizeof(long) == sizeof(*var));
+    volatile long *volatile_var = (volatile long *)var;
+    long old;
+    do {
+        old = *volatile_var;
+    } while(_InterlockedCompareExchange(volatile_var, old, old) != old);
+#endif
+    return old;
+}
+
+static inline void _Py_atomic_size_set(Py_ssize_t *var, Py_ssize_t value)
+{
+#if SIZEOF_VOID_P == 8
+    Py_BUILD_ASSERT(sizeof(__int64) == sizeof(*var));
+    volatile __int64 *volatile_var = (volatile __int64 *)var;
+    _InterlockedExchange64(volatile_var, value);
+#else
+    Py_BUILD_ASSERT(sizeof(long) == sizeof(*var));
+    volatile long *volatile_var = (volatile long *)var;
+    _InterlockedExchange(volatile_var, value);
+#endif
+}
+
+#else
+// Fallback implementation using volatile
+
+static inline Py_ssize_t _Py_atomic_size_get(Py_ssize_t *var)
+{
+    volatile Py_ssize_t *volatile_var = (volatile Py_ssize_t *)var;
+    return *volatile_var;
+}
+
+static inline void _Py_atomic_size_set(Py_ssize_t *var, Py_ssize_t value)
+{
+    volatile Py_ssize_t *volatile_var = (volatile Py_ssize_t *)var;
+    *volatile_var = value;
+}
+#endif
+
+#ifdef __cplusplus
+}
+#endif
+#endif  /* Py_ATOMIC_FUNC_H */