gh-155628: Use relaxed adds for the lru_cache hit/miss counters

Add an FT_ATOMIC_ADD_SSIZE_RELAXED wrapper and use it for the
lru_cache hits and misses counters, which are updated on every cached
call in the free-threaded build. The counters are pure statistics:
their only readers are cache_info() and cache_clear(), which already
use relaxed loads, so the adds need atomicity but no ordering.

Author: kddnewton

Include/internal/pycore_pyatomic_ft_wrappers.h

@@ -137,6 +137,8 @@ extern "C" {
     _Py_atomic_load_ullong_relaxed(&value)
 #define FT_ATOMIC_ADD_SSIZE(value, new_value) \
     (void)_Py_atomic_add_ssize(&value, new_value)
+#define FT_ATOMIC_ADD_SSIZE_RELAXED(value, new_value) \
+    (void)_Py_atomic_add_ssize_relaxed(&value, new_value)
 #define FT_MUTEX_LOCK(lock) PyMutex_Lock(lock)
 #define FT_MUTEX_LOCK_FLAGS(lock, flags) PyMutex_LockFlags(lock, flags)
 #define FT_MUTEX_UNLOCK(lock) PyMutex_Unlock(lock)
@@ -201,6 +203,7 @@ extern "C" {
 #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) (void)(value += new_value)
+#define FT_ATOMIC_ADD_SSIZE_RELAXED(value, new_value) (void)(value += new_value)
 #define FT_MUTEX_LOCK(lock) do {} while (0)
 #define FT_MUTEX_LOCK_FLAGS(lock, flags) do {} while (0)
 #define FT_MUTEX_UNLOCK(lock) do {} while (0)

Modules/_functoolsmodule.c

@@ -1299,7 +1299,7 @@ uncached_lru_cache_wrapper(lru_cache_object *self, PyObject *args, PyObject *kwd
 {
     PyObject *result;
 
-    FT_ATOMIC_ADD_SSIZE(self->misses, 1);
+    FT_ATOMIC_ADD_SSIZE_RELAXED(self->misses, 1);
     result = PyObject_Call(self->func, args, kwds);
     if (!result)
         return NULL;
@@ -1321,15 +1321,15 @@ infinite_lru_cache_wrapper(lru_cache_object *self, PyObject *args, PyObject *kwd
     }
     int res = _PyDict_GetItemRef_KnownHash((PyDictObject *)self->cache, key, hash, &result);
     if (res > 0) {
-        FT_ATOMIC_ADD_SSIZE(self->hits, 1);
+        FT_ATOMIC_ADD_SSIZE_RELAXED(self->hits, 1);
         Py_DECREF(key);
         return result;
     }
     if (res < 0) {
         Py_DECREF(key);
         return NULL;
     }
-    FT_ATOMIC_ADD_SSIZE(self->misses, 1);
+    FT_ATOMIC_ADD_SSIZE_RELAXED(self->misses, 1);
     result = PyObject_Call(self->func, args, kwds);
     if (!result) {
         Py_DECREF(key);
@@ -1425,7 +1425,7 @@ bounded_lru_cache_get_lock_held(lru_cache_object *self, PyObject *args, PyObject
         lru_cache_extract_link(link);
         lru_cache_append_link(self, link);
         *result = link->result;
-        FT_ATOMIC_ADD_SSIZE(self->hits, 1);
+        FT_ATOMIC_ADD_SSIZE_RELAXED(self->hits, 1);
         Py_INCREF(link->result);
         Py_DECREF(link);
         Py_DECREF(key_);
@@ -1435,7 +1435,7 @@ bounded_lru_cache_get_lock_held(lru_cache_object *self, PyObject *args, PyObject
         Py_DECREF(key_);
         return -1;
     }
-    FT_ATOMIC_ADD_SSIZE(self->misses, 1);
+    FT_ATOMIC_ADD_SSIZE_RELAXED(self->misses, 1);
     return 0;
 }