gh-124218: Refactor per-thread reference counting (#124844)

Currently, we only use per-thread reference counting for heap type objects and
the naming reflects that. We will extend it to a few additional types in an
upcoming change to avoid scaling bottlenecks when creating nested functions.

Rename some of the files and functions in preparation for this change.

5 files changed, 83 insertions, 86 deletions

diff --git a/Include/internal/pycore_interp.h b/Include/internal/pycore_interp.h
index a1898d9..ade69be 100644
--- a/Include/internal/pycore_interp.h
+++ b/Include/internal/pycore_interp.h
@@ -35,7 +35,7 @@ extern "C" {
 #include "pycore_qsbr.h"          // struct _qsbr_state
 #include "pycore_tstate.h"        // _PyThreadStateImpl
 #include "pycore_tuple.h"         // struct _Py_tuple_state
-#include "pycore_typeid.h"        // struct _Py_type_id_pool
+#include "pycore_uniqueid.h"      // struct _Py_unique_id_pool
 #include "pycore_typeobject.h"    // struct types_state
 #include "pycore_unicodeobject.h" // struct _Py_unicode_state
 #include "pycore_warnings.h"      // struct _warnings_runtime_state
@@ -221,7 +221,7 @@ struct _is {
 #if defined(Py_GIL_DISABLED)
     struct _mimalloc_interp_state mimalloc;
     struct _brc_state brc;  // biased reference counting state
-    struct _Py_type_id_pool type_ids;
+    struct _Py_unique_id_pool unique_ids;  // object ids for per-thread refcounts
     PyMutex weakref_locks[NUM_WEAKREF_LIST_LOCKS];
 #endif
 
diff --git a/Include/internal/pycore_object.h b/Include/internal/pycore_object.h
index 80b5888..0af13b1 100644
--- a/Include/internal/pycore_object.h
+++ b/Include/internal/pycore_object.h
@@ -14,7 +14,7 @@ extern "C" {
 #include "pycore_interp.h"        // PyInterpreterState.gc
 #include "pycore_pyatomic_ft_wrappers.h"  // FT_ATOMIC_STORE_PTR_RELAXED
 #include "pycore_pystate.h"       // _PyInterpreterState_GET()
-#include "pycore_typeid.h"        // _PyType_IncrefSlow
+#include "pycore_uniqueid.h"      // _PyType_IncrefSlow
 
 
 #define _Py_IMMORTAL_REFCNT_LOOSE ((_Py_IMMORTAL_REFCNT >> 1) + 1)
@@ -335,12 +335,12 @@ _Py_INCREF_TYPE(PyTypeObject *type)
     // Unsigned comparison so that `unique_id=-1`, which indicates that
     // per-thread refcounting has been disabled on this type, is handled by
     // the "else".
-    if ((size_t)ht->unique_id < (size_t)tstate->types.size) {
+    if ((size_t)ht->unique_id < (size_t)tstate->refcounts.size) {
 #  ifdef Py_REF_DEBUG
         _Py_INCREF_IncRefTotal();
 #  endif
         _Py_INCREF_STAT_INC();
-        tstate->types.refcounts[ht->unique_id]++;
+        tstate->refcounts.values[ht->unique_id]++;
     }
     else {
         // The slow path resizes the thread-local refcount array if necessary.
@@ -368,12 +368,12 @@ _Py_DECREF_TYPE(PyTypeObject *type)
     // Unsigned comparison so that `unique_id=-1`, which indicates that
     // per-thread refcounting has been disabled on this type, is handled by
     // the "else".
-    if ((size_t)ht->unique_id < (size_t)tstate->types.size) {
+    if ((size_t)ht->unique_id < (size_t)tstate->refcounts.size) {
 #  ifdef Py_REF_DEBUG
         _Py_DECREF_DecRefTotal();
 #  endif
         _Py_DECREF_STAT_INC();
-        tstate->types.refcounts[ht->unique_id]--;
+        tstate->refcounts.values[ht->unique_id]--;
     }
     else {
         // Directly decref the type if the type id is not assigned or if
diff --git a/Include/internal/pycore_tstate.h b/Include/internal/pycore_tstate.h
index f681b64..a72ef44 100644
--- a/Include/internal/pycore_tstate.h
+++ b/Include/internal/pycore_tstate.h
@@ -32,15 +32,15 @@ typedef struct _PyThreadStateImpl {
     struct _Py_freelists freelists;
     struct _brc_thread_state brc;
     struct {
-        // The thread-local refcounts for heap type objects
-        Py_ssize_t *refcounts;
+        // The per-thread refcounts
+        Py_ssize_t *values;
 
         // Size of the refcounts array.
         Py_ssize_t size;
 
-        // If set, don't use thread-local refcounts
+        // If set, don't use per-thread refcounts
         int is_finalized;
-    } types;
+    } refcounts;
 #endif
 
 #if defined(Py_REF_DEBUG) && defined(Py_GIL_DISABLED)
diff --git a/Include/internal/pycore_typeid.h b/Include/internal/pycore_typeid.h
deleted file mode 100644
index e64d144..0000000
--- a/Include/internal/pycore_typeid.h
+++ /dev/null
@@ -1,75 +0,0 @@
-#ifndef Py_INTERNAL_TYPEID_H
-#define Py_INTERNAL_TYPEID_H
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-#ifndef Py_BUILD_CORE
-#  error "this header requires Py_BUILD_CORE define"
-#endif
-
-#ifdef Py_GIL_DISABLED
-
-// This contains code for allocating unique ids to heap type objects
-// and re-using those ids when the type is deallocated.
-//
-// The type ids are used to implement per-thread reference counts of
-// heap type objects to avoid contention on the reference count fields
-// of heap type objects. Static type objects are immortal, so contention
-// is not an issue for those types.
-//
-// Type id of -1 is used to indicate a type doesn't use thread-local
-// refcounting. This value is used when a type object is finalized by the GC
-// and during interpreter shutdown to allow the type object to be
-// deallocated promptly when the object's refcount reaches zero.
-//
-// Each entry implicitly represents a type id based on it's offset in the
-// table. Non-allocated entries form a free-list via the 'next' pointer.
-// Allocated entries store the corresponding PyTypeObject.
-typedef union _Py_type_id_entry {
-    // Points to the next free type id, when part of the freelist
-    union _Py_type_id_entry *next;
-
-    // Stores the type object when the id is assigned
-    PyHeapTypeObject *type;
-} _Py_type_id_entry;
-
-struct _Py_type_id_pool {
-    PyMutex mutex;
-
-    // combined table of types with allocated type ids and unallocated
-    // type ids.
-    _Py_type_id_entry *table;
-
-    // Next entry to allocate inside 'table' or NULL
-    _Py_type_id_entry *freelist;
-
-    // size of 'table'
-    Py_ssize_t size;
-};
-
-// Assigns the next id from the pool of type ids.
-extern void _PyType_AssignId(PyHeapTypeObject *type);
-
-// Releases the allocated type id back to the pool.
-extern void _PyType_ReleaseId(PyHeapTypeObject *type);
-
-// Merges the thread-local reference counts into the corresponding types.
-extern void _PyType_MergeThreadLocalRefcounts(_PyThreadStateImpl *tstate);
-
-// Like _PyType_MergeThreadLocalRefcounts, but also frees the thread-local
-// array of refcounts.
-extern void _PyType_FinalizeThreadLocalRefcounts(_PyThreadStateImpl *tstate);
-
-// Frees the interpreter's pool of type ids.
-extern void _PyType_FinalizeIdPool(PyInterpreterState *interp);
-
-// Increfs the type, resizing the thread-local refcount array if necessary.
-PyAPI_FUNC(void) _PyType_IncrefSlow(PyHeapTypeObject *type);
-
-#endif   /* Py_GIL_DISABLED */
-
-#ifdef __cplusplus
-}
-#endif
-#endif   /* !Py_INTERNAL_TYPEID_H */
diff --git a/Include/internal/pycore_uniqueid.h b/Include/internal/pycore_uniqueid.h
new file mode 100644
index 0000000..8f3b441
--- /dev/null
+++ b/Include/internal/pycore_uniqueid.h
@@ -0,0 +1,72 @@
+#ifndef Py_INTERNAL_UNIQUEID_H
+#define Py_INTERNAL_UNIQUEID_H
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef Py_BUILD_CORE
+#  error "this header requires Py_BUILD_CORE define"
+#endif
+
+#ifdef Py_GIL_DISABLED
+
+// This contains code for allocating unique ids to objects for per-thread
+// reference counting.
+//
+// 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
+// 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.
+//
+// Each entry implicitly represents a unique id based on its offset in the
+// table. Non-allocated entries form a free-list via the 'next' pointer.
+// Allocated entries store the corresponding PyObject.
+typedef union _Py_unique_id_entry {
+    // Points to the next free type id, when part of the freelist
+    union _Py_unique_id_entry *next;
+
+    // Stores the object when the id is assigned
+    PyObject *obj;
+} _Py_unique_id_entry;
+
+struct _Py_unique_id_pool {
+    PyMutex mutex;
+
+    // combined table of object with allocated unique ids and unallocated ids.
+    _Py_unique_id_entry *table;
+
+    // Next entry to allocate inside 'table' or NULL
+    _Py_unique_id_entry *freelist;
+
+    // size of 'table'
+    Py_ssize_t size;
+};
+
+// Assigns the next id from the pool of ids.
+extern Py_ssize_t _PyObject_AssignUniqueId(PyObject *obj);
+
+// Releases the allocated id back to the pool.
+extern void _PyObject_ReleaseUniqueId(Py_ssize_t unique_id);
+
+// Merges the per-thread reference counts into the corresponding objects.
+extern void _PyObject_MergePerThreadRefcounts(_PyThreadStateImpl *tstate);
+
+// Like _PyObject_MergePerThreadRefcounts, but also frees the per-thread
+// array of refcounts.
+extern void _PyObject_FinalizePerThreadRefcounts(_PyThreadStateImpl *tstate);
+
+// Frees the interpreter's pool of type ids.
+extern void _PyObject_FinalizeUniqueIdPool(PyInterpreterState *interp);
+
+// Increfs the type, resizing the per-thread refcount array if necessary.
+PyAPI_FUNC(void) _PyType_IncrefSlow(PyHeapTypeObject *type);
+
+#endif   /* Py_GIL_DISABLED */
+
+#ifdef __cplusplus
+}
+#endif
+#endif   /* !Py_INTERNAL_UNIQUEID_H */