Merge branch 'main' into refactor-optimize-structs

Author: markshannon

Include/internal/pycore_dict.h

@@ -347,8 +347,7 @@ PyDictObject *_PyObject_MaterializeManagedDict_LockHeld(PyObject *);
 static inline Py_ssize_t
 _PyDict_UniqueId(PyDictObject *mp)
 {
-    // Offset by one so that _ma_watcher_tag=0 represents an unassigned id
-    return (Py_ssize_t)(mp->_ma_watcher_tag >> DICT_UNIQUE_ID_SHIFT) - 1;
+    return (Py_ssize_t)(mp->_ma_watcher_tag >> DICT_UNIQUE_ID_SHIFT);
 }
 
 static inline void

Include/internal/pycore_object.h

@@ -336,20 +336,20 @@ _Py_THREAD_INCREF_OBJECT(PyObject *obj, Py_ssize_t unique_id)
 {
     _PyThreadStateImpl *tstate = (_PyThreadStateImpl *)_PyThreadState_GET();
 
-    // Unsigned comparison so that `unique_id=-1`, which indicates that
-    // per-thread refcounting has been disabled on this object, is handled by
-    // the "else".
-    if ((size_t)unique_id < (size_t)tstate->refcounts.size) {
+    // The table index is `unique_id - 1` because 0 is not a valid unique id.
+    // Unsigned comparison so that `idx=-1` is handled by the "else".
+    size_t idx = (size_t)(unique_id - 1);
+    if (idx < (size_t)tstate->refcounts.size) {
 #  ifdef Py_REF_DEBUG
         _Py_INCREF_IncRefTotal();
 #  endif
         _Py_INCREF_STAT_INC();
-        tstate->refcounts.values[unique_id]++;
+        tstate->refcounts.values[idx]++;
     }
     else {
         // The slow path resizes the per-thread refcount array if necessary.
-        // It handles the unique_id=-1 case to keep the inlinable function smaller.
-        _PyObject_ThreadIncrefSlow(obj, unique_id);
+        // It handles the unique_id=0 case to keep the inlinable function smaller.
+        _PyObject_ThreadIncrefSlow(obj, idx);
     }
 }
 
@@ -386,15 +386,15 @@ _Py_THREAD_DECREF_OBJECT(PyObject *obj, Py_ssize_t unique_id)
 {
     _PyThreadStateImpl *tstate = (_PyThreadStateImpl *)_PyThreadState_GET();
 
-    // Unsigned comparison so that `unique_id=-1`, which indicates that
-    // per-thread refcounting has been disabled on this object, is handled by
-    // the "else".
-    if ((size_t)unique_id < (size_t)tstate->refcounts.size) {
+    // The table index is `unique_id - 1` because 0 is not a valid unique id.
+    // Unsigned comparison so that `idx=-1` is handled by the "else".
+    size_t idx = (size_t)(unique_id - 1);
+    if (idx < (size_t)tstate->refcounts.size) {
 #  ifdef Py_REF_DEBUG
         _Py_DECREF_DecRefTotal();
 #  endif
         _Py_DECREF_STAT_INC();
-        tstate->refcounts.values[unique_id]--;
+        tstate->refcounts.values[idx]--;
     }
     else {
         // Directly decref the object if the id is not assigned or if

Include/internal/pycore_optimizer.h

@@ -98,11 +98,6 @@ struct _PyOptimizerObject {
 };
 
 /** Test support **/
-typedef struct {
-    _PyOptimizerObject base;
-    int64_t count;
-} _PyCounterOptimizerObject;
-
 _PyOptimizerObject *_Py_SetOptimizer(PyInterpreterState *interp, _PyOptimizerObject* optimizer);
 
 
@@ -119,7 +114,6 @@ PyAPI_FUNC(void) _Py_Executor_DependsOn(_PyExecutorObject *executor, void *obj);
 // Export for '_testinternalcapi' shared extension.
 PyAPI_FUNC(_PyOptimizerObject *) _Py_GetOptimizer(void);
 PyAPI_FUNC(int) _Py_SetTier2Optimizer(_PyOptimizerObject* optimizer);
-PyAPI_FUNC(PyObject *) _PyOptimizer_NewCounter(void);
 PyAPI_FUNC(PyObject *) _PyOptimizer_NewUOpOptimizer(void);
 
 #define _Py_MAX_ALLOWED_BUILTINS_MODIFICATIONS 3
@@ -150,8 +144,6 @@ int _Py_uop_analyze_and_optimize(struct _PyInterpreterFrame *frame,
     _PyUOpInstruction *trace, int trace_len, int curr_stackentries,
     _PyBloomFilter *dependencies);
 
-extern PyTypeObject _PyCounterExecutor_Type;
-extern PyTypeObject _PyCounterOptimizer_Type;
 extern PyTypeObject _PyDefaultOptimizer_Type;
 extern PyTypeObject _PyUOpExecutor_Type;
 extern PyTypeObject _PyUOpOptimizer_Type;

Include/internal/pycore_uniqueid.h

@@ -16,7 +16,7 @@ extern "C" {
 // Per-thread reference counting is used along with deferred reference
 // counting to avoid scaling bottlenecks due to reference count contention.
 //
-// An id of -1 is used to indicate that an object doesn't use per-thread
+// An id of 0 is used to indicate that an object doesn't use per-thread
 // refcounting. This value is used when the object is finalized by the GC
 // and during interpreter shutdown to allow the object to be
 // deallocated promptly when the object's refcount reaches zero.
@@ -45,6 +45,8 @@ struct _Py_unique_id_pool {
     Py_ssize_t size;
 };
 
+#define _Py_INVALID_UNIQUE_ID 0
+
 // Assigns the next id from the pool of ids.
 extern Py_ssize_t _PyObject_AssignUniqueId(PyObject *obj);
 
@@ -65,7 +67,7 @@ extern void _PyObject_FinalizePerThreadRefcounts(_PyThreadStateImpl *tstate);
 extern void _PyObject_FinalizeUniqueIdPool(PyInterpreterState *interp);
 
 // Increfs the object, resizing the thread-local refcount array if necessary.
-PyAPI_FUNC(void) _PyObject_ThreadIncrefSlow(PyObject *obj, Py_ssize_t unique_id);
+PyAPI_FUNC(void) _PyObject_ThreadIncrefSlow(PyObject *obj, size_t idx);
 
 #endif   /* Py_GIL_DISABLED */
 

Include/internal/pycore_uop_ids.h

@@ -11,7 +11,7 @@ execution state.
 
 A generator object resumes execution in its frame when its `send()`
 method is called. This is analogous to a function executing in its own
-fram when it is called, but a function returns to the calling frame only once,
+frame when it is called, but a function returns to the calling frame only once,
 while a generator "returns" execution to the caller's frame every time
 it emits a new item with a
 [`yield` expression](https://docs.python.org/dev/reference/expressions.html#yield-expressions).