gh-131757: allow lru_cache functions to execute concurrently

Include/internal/pycore_pyatomic_ft_wrappers.h

@@ -109,6 +109,8 @@
     _Py_atomic_store_ullong_relaxed(&value, new_value)
 #define FT_ATOMIC_LOAD_ULLONG_RELAXED(value) \
     _Py_atomic_load_ullong_relaxed(&value)
+#define FT_ATOMIC_ADD_SSIZE(value, new_value) \
+    _Py_atomic_add_ssize(&value, new_value)
 
 #else
 #define FT_ATOMIC_LOAD_PTR(value) value
@@ -156,6 +158,7 @@
 #define FT_ATOMIC_STORE_LLONG_RELAXED(value, new_value) value = new_value
 #define FT_ATOMIC_LOAD_ULLONG_RELAXED(value) value
 #define FT_ATOMIC_STORE_ULLONG_RELAXED(value, new_value) value = new_value
+#define FT_ATOMIC_ADD_SSIZE(value, new_value) (value += new_value) - new_value
 
 #endif
 

Modules/_functoolsmodule.c

@@ -41,13 +41,6 @@
     return (_functools_state *)state;
 }
 
-#ifdef Py_GIL_DISABLED
-#define FT_ATOMIC_ADD_SSIZE(value, new_value) \
-    _Py_atomic_add_ssize(&value, new_value)
-#else
-#define FT_ATOMIC_ADD_SSIZE(value, new_value) value += new_value
-#endif
-
 /* partial object **********************************************************/
 
 
@@ -1180,7 +1173,7 @@
 static PyObject *
 uncached_lru_cache_wrapper(lru_cache_object *self, PyObject *args, PyObject *kwds)
 {
     PyObject *result;
 
     FT_ATOMIC_ADD_SSIZE(self->misses, 1);
     result = PyObject_Call(self->func, args, kwds);
@@ -1202,7 +1195,7 @@
         Py_DECREF(key);
         return NULL;
     }
     int res = _PyDict_GetItemRef_KnownHash((PyDictObject *)self->cache, key, hash, &result);
     if (res > 0) {
         FT_ATOMIC_ADD_SSIZE(self->hits, 1);
         Py_DECREF(key);
@@ -1210,7 +1203,7 @@
     }
     if (res < 0) {
         Py_DECREF(key);
         return NULL;
     }
     FT_ATOMIC_ADD_SSIZE(self->misses, 1);
     result = PyObject_Call(self->func, args, kwds);