Add static inline _Py_NewReferenceInline() function

Author: vstinner

Include/internal/pycore_freelist.h

@@ -13,6 +13,7 @@ extern "C" {
 #include "pycore_pyatomic_ft_wrappers.h" // FT_ATOMIC_STORE_PTR_RELAXED()
 #include "pycore_pystate.h"             // _PyThreadState_GET
 #include "pycore_stats.h"               // OBJECT_STAT_INC
+#include "pycore_object.h"              // _Py_NewReferenceInline()
 
 static inline struct _Py_freelists *
 _Py_freelists_GET(void)
@@ -89,7 +90,7 @@ _PyFreeList_Pop(struct _Py_freelist *fl)
     PyObject *op = _PyFreeList_PopNoStats(fl);
     if (op != NULL) {
         OBJECT_STAT_INC(from_freelist);
-        _Py_NewReference(op);
+        _Py_NewReferenceInline(op);
     }
     return op;
 }

Include/internal/pycore_object.h

@@ -19,6 +19,13 @@ extern "C" {
 #include <stdbool.h>              // bool
 
 
+#ifdef Py_TRACE_REFS
+extern void _Py_AddToAllObjects(PyObject *op);
+extern void _Py_PrintReferences(PyInterpreterState *, FILE *);
+extern void _Py_PrintReferenceAddresses(PyInterpreterState *, FILE *);
+#endif
+
+
 // This value is added to `ob_ref_shared` for objects that use deferred
 // reference counting so that they are not immediately deallocated when the
 // non-deferred reference count drops to zero.
@@ -468,6 +475,52 @@ static inline void Py_DECREF_MORTAL_SPECIALIZED(PyObject *op, destructor destruc
 # define Py_DECREF_MORTAL_SPECIALIZED(op, destruct) Py_DECREF(op)
 #endif
 
+static inline void
+_Py_NewReferenceNoTotalInline(PyObject *op)
+{
+    // Skip the immortal object check in Py_SET_REFCNT; always set refcnt to 1
+#if !defined(Py_GIL_DISABLED)
+    // With GIL
+#if SIZEOF_VOID_P > 4
+    op->ob_refcnt_full = 1;
+    assert(op->ob_refcnt == 1);
+    assert(op->ob_flags == 0);
+#else
+    op->ob_refcnt = 1;
+#endif
+#else
+    // Free Threading
+    op->ob_flags = 0;
+    op->ob_mutex = (PyMutex){ 0 };
+#ifdef _Py_THREAD_SANITIZER
+    _Py_atomic_store_uintptr_relaxed(&op->ob_tid, _Py_ThreadId());
+    _Py_atomic_store_uint8_relaxed(&op->ob_gc_bits, 0);
+    _Py_atomic_store_uint32_relaxed(&op->ob_ref_local, 1);
+    _Py_atomic_store_ssize_relaxed(&op->ob_ref_shared, 0);
+#else
+    op->ob_tid = _Py_ThreadId();
+    op->ob_gc_bits = 0;
+    op->ob_ref_local = 1;
+    op->ob_ref_shared = 0;
+#endif
+#endif
+
+#ifdef Py_TRACE_REFS
+    _Py_AddToAllObjects(op);
+#endif
+
+    _PyReftracerTrack(op, PyRefTracer_CREATE);
+}
+
+static inline void
+_Py_NewReferenceInline(PyObject *op)
+{
+#ifdef Py_REF_DEBUG
+    _Py_IncRefTotal(_PyThreadState_GET());
+#endif
+    _Py_NewReferenceNoTotalInline(op);
+}
+
 /* Inline functions trading binary compatibility for speed:
    _PyObject_Init() is the fast version of PyObject_Init(), and
    _PyObject_InitVar() is the fast version of PyObject_InitVar().
@@ -480,7 +533,7 @@ _PyObject_Init(PyObject *op, PyTypeObject *typeobj)
     Py_SET_TYPE(op, typeobj);
     assert(_PyType_HasFeature(typeobj, Py_TPFLAGS_HEAPTYPE) || _Py_IsImmortal(typeobj));
     _Py_INCREF_TYPE(typeobj);
-    _Py_NewReference(op);
+    _Py_NewReferenceInline(op);
 }
 
 static inline void
@@ -764,12 +817,6 @@ extern void _Py_FinalizeRefTotal(_PyRuntimeState *);
 extern void _PyDebug_PrintTotalRefs(void);
 #endif
 
-#ifdef Py_TRACE_REFS
-extern void _Py_AddToAllObjects(PyObject *op);
-extern void _Py_PrintReferences(PyInterpreterState *, FILE *);
-extern void _Py_PrintReferenceAddresses(PyInterpreterState *, FILE *);
-#endif
-
 
 /* Return the *address* of the object's weaklist.  The address may be
  * dereferenced to get the current head of the weaklist.  This is useful

Modules/_testcapimodule.c

@@ -2889,6 +2889,29 @@ uptime_bsd(PyObject *Py_UNUSED(self), PyObject *Py_UNUSED(args))
 #endif
 
 
+static PyObject *
+bench_writer(PyObject *self, PyObject *args)
+{
+    Py_ssize_t size, loops;
+    if (!PyArg_ParseTuple(args, "nn", &size, &loops)) {
+        return NULL;
+    }
+
+    PyTime_t t1, t2;
+    (void)PyTime_PerfCounterRaw(&t1);
+    for (Py_ssize_t i=0; i < loops; i++) {
+        PyObject *bytes = PyBytes_FromString("abc");
+        if (bytes == NULL) {
+            return NULL;
+        }
+        Py_DECREF(bytes);
+    }
+    (void)PyTime_PerfCounterRaw(&t2);
+
+    return PyFloat_FromDouble(PyTime_AsSecondsDouble(t2 - t1));
+}
+
+
 static PyMethodDef TestMethods[] = {
     {"set_errno",               set_errno,                       METH_VARARGS},
     {"test_config",             test_config,                     METH_NOARGS},
@@ -2988,6 +3011,7 @@ static PyMethodDef TestMethods[] = {
 #ifdef HAVE_SYSCTLBYNAME
     {"uptime_bsd", uptime_bsd, METH_NOARGS},
 #endif
+    {"bench_writer", bench_writer, METH_VARARGS},
     {NULL, NULL} /* sentinel */
 };
 

Objects/bytesobject.c

@@ -3438,7 +3438,7 @@ bytes_resize_inplace(PyObject **pv, Py_ssize_t newsize)
 
     *pv = result;
     v = result;
-    _Py_NewReferenceNoTotal(v);
+    _Py_NewReferenceNoTotalInline(v);
     PyBytesObject *sv = (PyBytesObject *)v;
     Py_SET_SIZE(sv, newsize);
     sv->ob_sval[newsize] = '\0';

Objects/object.c

@@ -2738,52 +2738,19 @@ _PyTypes_FiniTypes(PyInterpreterState *interp)
 }
 
 
-static inline void
-new_reference(PyObject *op)
-{
-    // Skip the immortal object check in Py_SET_REFCNT; always set refcnt to 1
-#if !defined(Py_GIL_DISABLED)
-#if SIZEOF_VOID_P > 4
-    op->ob_refcnt_full = 1;
-    assert(op->ob_refcnt == 1);
-    assert(op->ob_flags == 0);
-#else
-    op->ob_refcnt = 1;
-#endif
-#else
-    op->ob_flags = 0;
-    op->ob_mutex = (PyMutex){ 0 };
-#ifdef _Py_THREAD_SANITIZER
-    _Py_atomic_store_uintptr_relaxed(&op->ob_tid, _Py_ThreadId());
-    _Py_atomic_store_uint8_relaxed(&op->ob_gc_bits, 0);
-    _Py_atomic_store_uint32_relaxed(&op->ob_ref_local, 1);
-    _Py_atomic_store_ssize_relaxed(&op->ob_ref_shared, 0);
-#else
-    op->ob_tid = _Py_ThreadId();
-    op->ob_gc_bits = 0;
-    op->ob_ref_local = 1;
-    op->ob_ref_shared = 0;
-#endif
-#endif
-#ifdef Py_TRACE_REFS
-    _Py_AddToAllObjects(op);
-#endif
-    _PyReftracerTrack(op, PyRefTracer_CREATE);
-}
-
 void
 _Py_NewReference(PyObject *op)
 {
 #ifdef Py_REF_DEBUG
     _Py_IncRefTotal(_PyThreadState_GET());
 #endif
-    new_reference(op);
+    _Py_NewReferenceNoTotalInline(op);
 }
 
 void
 _Py_NewReferenceNoTotal(PyObject *op)
 {
-    new_reference(op);
+    _Py_NewReferenceNoTotalInline(op);
 }
 
 void

Objects/tupleobject.c

@@ -1069,7 +1069,7 @@ _PyTuple_Resize(PyObject **pv, Py_ssize_t newsize)
         PyObject_GC_Del(v);
         return -1;
     }
-    _Py_NewReferenceNoTotal((PyObject *) sv);
+    _Py_NewReferenceNoTotalInline((PyObject *) sv);
     /* Zero out items added by growing */
     if (newsize > oldsize)
         memset(&sv->ob_item[oldsize], 0,

Objects/unicodeobject.c

@@ -1124,12 +1124,12 @@ _PyUnicode_ResizeCompact(PyObject *unicode, Py_ssize_t length)
 
     new_unicode = (PyObject *)PyObject_Realloc(unicode, new_size);
     if (new_unicode == NULL) {
-        _Py_NewReferenceNoTotal(unicode);
+        _Py_NewReferenceNoTotalInline(unicode);
         PyErr_NoMemory();
         return NULL;
     }
     unicode = new_unicode;
-    _Py_NewReferenceNoTotal(unicode);
+    _Py_NewReferenceNoTotalInline(unicode);
 
     _PyUnicode_LENGTH(unicode) = length;
 #ifdef Py_DEBUG