pythongh-122417: Implement per-thread heap type refcounts (python#122418

)

The free-threaded build partially stores heap type reference counts in
distributed manner in per-thread arrays. This avoids reference count
contention when creating or destroying instances.

Co-authored-by: Ken Jin <kenjin@python.org>

Include/cpython/object.h

@@ -270,6 +270,9 @@ typedef struct _heaptypeobject {
     PyObject *ht_module;
     char *_ht_tpname;  // Storage for "tp_name"; see PyType_FromModuleAndSpec
     struct _specialization_cache _spec_cache; // For use by the specializer.
+#ifdef Py_GIL_DISABLED
+    Py_ssize_t unique_id;  // ID used for thread-local refcounting
+#endif
     /* here are optional user slots, followed by the members. */
 } PyHeapTypeObject;
 

Include/internal/pycore_gc.h

@@ -381,10 +381,6 @@ extern void _PyGC_ClearAllFreeLists(PyInterpreterState *interp);
 extern void _Py_ScheduleGC(PyThreadState *tstate);
 extern void _Py_RunGC(PyThreadState *tstate);
 
-#ifdef Py_GIL_DISABLED
-// gh-117783: Immortalize objects that use deferred reference counting
-extern void _PyGC_ImmortalizeDeferredObjects(PyInterpreterState *interp);
-#endif
 
 #ifdef __cplusplus
 }

Include/internal/pycore_interp.h

@@ -35,6 +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_typeobject.h"    // struct types_state
 #include "pycore_unicodeobject.h" // struct _Py_unicode_state
 #include "pycore_warnings.h"      // struct _warnings_runtime_state
@@ -220,6 +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;
     PyMutex weakref_locks[NUM_WEAKREF_LIST_LOCKS];
 #endif
 

Include/internal/pycore_object.h

@@ -14,10 +14,19 @@ 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
 
 
 #define _Py_IMMORTAL_REFCNT_LOOSE ((_Py_IMMORTAL_REFCNT >> 1) + 1)
 
+// 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.
+//
+// The value is half the maximum shared refcount because the low two bits of
+// `ob_ref_shared` are used for flags.
+#define _Py_REF_DEFERRED (PY_SSIZE_T_MAX / 8)
+
 // gh-121528, gh-118997: Similar to _Py_IsImmortal() but be more loose when
 // comparing the reference count to stay compatible with C extensions built
 // with the stable ABI 3.11 or older. Such extensions implement INCREF/DECREF
@@ -280,6 +289,67 @@ extern PyStatus _PyObject_InitState(PyInterpreterState *interp);
 extern void _PyObject_FiniState(PyInterpreterState *interp);
 extern bool _PyRefchain_IsTraced(PyInterpreterState *interp, PyObject *obj);
 
+#ifndef Py_GIL_DISABLED
+#  define _Py_INCREF_TYPE Py_INCREF
+#  define _Py_DECREF_TYPE Py_DECREF
+#else
+static inline void
+_Py_INCREF_TYPE(PyTypeObject *type)
+{
+    if (!_PyType_HasFeature(type, Py_TPFLAGS_HEAPTYPE)) {
+        assert(_Py_IsImmortal(type));
+        return;
+    }
+
+    _PyThreadStateImpl *tstate = (_PyThreadStateImpl *)_PyThreadState_GET();
+    PyHeapTypeObject *ht = (PyHeapTypeObject *)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) {
+#  ifdef Py_REF_DEBUG
+        _Py_INCREF_IncRefTotal();
+#  endif
+        _Py_INCREF_STAT_INC();
+        tstate->types.refcounts[ht->unique_id]++;
+    }
+    else {
+        // The slow path resizes the thread-local refcount array if necessary.
+        // It handles the unique_id=-1 case to keep the inlinable function smaller.
+        _PyType_IncrefSlow(ht);
+    }
+}
+
+static inline void
+_Py_DECREF_TYPE(PyTypeObject *type)
+{
+    if (!_PyType_HasFeature(type, Py_TPFLAGS_HEAPTYPE)) {
+        assert(_Py_IsImmortal(type));
+        return;
+    }
+
+    _PyThreadStateImpl *tstate = (_PyThreadStateImpl *)_PyThreadState_GET();
+    PyHeapTypeObject *ht = (PyHeapTypeObject *)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) {
+#  ifdef Py_REF_DEBUG
+        _Py_DECREF_DecRefTotal();
+#  endif
+        _Py_DECREF_STAT_INC();
+        tstate->types.refcounts[ht->unique_id]--;
+    }
+    else {
+        // Directly decref the type if the type id is not assigned or if
+        // per-thread refcounting has been disabled on this type.
+        Py_DECREF(type);
+    }
+}
+#endif
+
 /* 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().
@@ -291,7 +361,7 @@ _PyObject_Init(PyObject *op, PyTypeObject *typeobj)
     assert(op != NULL);
     Py_SET_TYPE(op, typeobj);
     assert(_PyType_HasFeature(typeobj, Py_TPFLAGS_HEAPTYPE) || _Py_IsImmortalLoose(typeobj));
-    Py_INCREF(typeobj);
+    _Py_INCREF_TYPE(typeobj);
     _Py_NewReference(op);
 }
 

Include/internal/pycore_tstate.h

@@ -31,6 +31,16 @@ typedef struct _PyThreadStateImpl {
     struct _mimalloc_thread_state mimalloc;
     struct _Py_freelists freelists;
     struct _brc_thread_state brc;
+    struct {
+        // The thread-local refcounts for heap type objects
+        Py_ssize_t *refcounts;
+
+        // Size of the refcounts array.
+        Py_ssize_t size;
+
+        // If set, don't use thread-local refcounts
+        int is_finalized;
+    } types;
 #endif
 
 #if defined(Py_REF_DEBUG) && defined(Py_GIL_DISABLED)

Include/internal/pycore_typeid.h

@@ -0,0 +1,75 @@
+#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 */

Lib/test/test_sys.py

@@ -1710,6 +1710,7 @@ def delx(self): del self.__x
         fmt = 'P2nPI13Pl4Pn9Pn12PIPc'
         s = vsize(fmt)
         check(int, s)
+        typeid = 'n' if support.Py_GIL_DISABLED else ''
         # class
         s = vsize(fmt +                 # PyTypeObject
                   '4P'                  # PyAsyncMethods
@@ -1718,7 +1719,8 @@ def delx(self): del self.__x
                   '10P'                 # PySequenceMethods
                   '2P'                  # PyBufferProcs
                   '6P'
-                  '1PIP'                 # Specializer cache
+                  '1PIP'                # Specializer cache
+                  + typeid              # heap type id (free-threaded only)
                   )
         class newstyleclass(object): pass
         # Separate block for PyDictKeysObject with 8 keys and 5 entries

Makefile.pre.in

@@ -483,6 +483,7 @@ PYTHON_OBJS=	\
 		Python/thread.o \
 		Python/traceback.o \
 		Python/tracemalloc.o \
+		Python/typeid.o \
 		Python/getopt.o \
 		Python/pystrcmp.o \
 		Python/pystrtod.o \
@@ -1257,6 +1258,7 @@ PYTHON_HEADERS= \
 		$(srcdir)/Include/internal/pycore_tracemalloc.h \
 		$(srcdir)/Include/internal/pycore_tstate.h \
 		$(srcdir)/Include/internal/pycore_tuple.h \
+		$(srcdir)/Include/internal/pycore_typeid.h \
 		$(srcdir)/Include/internal/pycore_typeobject.h \
 		$(srcdir)/Include/internal/pycore_typevarobject.h \
 		$(srcdir)/Include/internal/pycore_ucnhash.h \

Misc/NEWS.d/next/Core_and_Builtins/2024-07-29-19-20-25.gh-issue-122417.NVgs0a.rst

@@ -0,0 +1,4 @@
+In the free-threaded build, the reference counts for heap type objects are now
+partially stored in a distributed manner in per-thread arrays.  This reduces
+contention on the heap type's reference count fields when creating or
+destroying instances of the same type from multiple threads concurrently.

Objects/object.c

@@ -2477,15 +2477,7 @@ _PyObject_SetDeferredRefcount(PyObject *op)
     assert(_Py_IsOwnedByCurrentThread(op));
     assert(op->ob_ref_shared == 0);
     _PyObject_SET_GC_BITS(op, _PyGC_BITS_DEFERRED);
-    PyInterpreterState *interp = _PyInterpreterState_GET();
-    if (_Py_atomic_load_int_relaxed(&interp->gc.immortalize) == 1) {
-        // gh-117696: immortalize objects instead of using deferred reference
-        // counting for now.
-        _Py_SetImmortal(op);
-        return;
-    }
-    op->ob_ref_local += 1;
-    op->ob_ref_shared = _Py_REF_QUEUED;
+    op->ob_ref_shared = _Py_REF_SHARED(_Py_REF_DEFERRED, 0);
 #endif
 }
 

Objects/typeobject.c

@@ -2452,7 +2452,7 @@ subtype_dealloc(PyObject *self)
            reference counting. Only decref if the base type is not already a heap
            allocated type. Otherwise, basedealloc should have decref'd it already */
         if (type_needs_decref) {
-            Py_DECREF(type);
+            _Py_DECREF_TYPE(type);
         }
 
         /* Done */
@@ -2562,7 +2562,7 @@ subtype_dealloc(PyObject *self)
        reference counting. Only decref if the base type is not already a heap
        allocated type. Otherwise, basedealloc should have decref'd it already */
     if (type_needs_decref) {
-        Py_DECREF(type);
+        _Py_DECREF_TYPE(type);
     }
 
   endlabel:
@@ -3913,7 +3913,9 @@ type_new_alloc(type_new_ctx *ctx)
     et->ht_module = NULL;
     et->_ht_tpname = NULL;
 
-    _PyObject_SetDeferredRefcount((PyObject *)et);
+#ifdef Py_GIL_DISABLED
+    _PyType_AssignId(et);
+#endif
 
     return type;
 }
@@ -4965,6 +4967,11 @@ _PyType_FromMetaclass_impl(
     type->tp_weaklistoffset = weaklistoffset;
     type->tp_dictoffset = dictoffset;
 
+#ifdef Py_GIL_DISABLED
+    // Assign a type id to enable thread-local refcounting
+    _PyType_AssignId(res);
+#endif
+
     /* Ready the type (which includes inheritance).
      *
      * After this call we should generally only touch up what's
@@ -5914,6 +5921,9 @@ type_dealloc(PyObject *self)
     }
     Py_XDECREF(et->ht_module);
     PyMem_Free(et->_ht_tpname);
+#ifdef Py_GIL_DISABLED
+    _PyType_ReleaseId(et);
+#endif
     Py_TYPE(type)->tp_free((PyObject *)type);
 }
 

PCbuild/_freeze_module.vcxproj

@@ -267,6 +267,7 @@
     <ClCompile Include="..\Python\thread.c" />
     <ClCompile Include="..\Python\traceback.c" />
     <ClCompile Include="..\Python\tracemalloc.c" />
+    <ClCompile Include="..\Python\typeid.c" />
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="..\PC\pyconfig.h.in" />

PCbuild/_freeze_module.vcxproj.filters

@@ -464,6 +464,9 @@
     <ClCompile Include="..\Python\tracemalloc.c">
       <Filter>Source Files</Filter>
     </ClCompile>
+    <ClCompile Include="..\Python\typeid.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
     <ClCompile Include="..\Objects\tupleobject.c">
       <Filter>Source Files</Filter>
     </ClCompile>

PCbuild/pythoncore.vcxproj

@@ -304,6 +304,7 @@
     <ClInclude Include="..\Include\internal\pycore_tracemalloc.h" />
     <ClInclude Include="..\Include\internal\pycore_tstate.h" />
     <ClInclude Include="..\Include\internal\pycore_tuple.h" />
+    <ClInclude Include="..\Include\internal\pycore_typeid.h" />
     <ClInclude Include="..\Include\internal\pycore_typeobject.h" />
     <ClInclude Include="..\Include\internal\pycore_typevarobject.h" />
     <ClInclude Include="..\Include\internal\pycore_ucnhash.h" />
@@ -643,6 +644,7 @@
     <ClCompile Include="..\Python\thread.c" />
     <ClCompile Include="..\Python\traceback.c" />
     <ClCompile Include="..\Python\tracemalloc.c" />
+    <ClCompile Include="..\Python\typeid.c" />
   </ItemGroup>
   <ItemGroup Condition="$(IncludeExternals)">
     <ClCompile Include="..\Modules\zlibmodule.c" />

PCbuild/pythoncore.vcxproj.filters

@@ -831,6 +831,9 @@
     <ClInclude Include="..\Include\internal\pycore_tuple.h">
       <Filter>Include\internal</Filter>
     </ClInclude>
+    <ClInclude Include="..\Include\internal\pycore_typeid.h">
+      <Filter>Include\internal</Filter>
+    </ClInclude>
     <ClInclude Include="..\Include\internal\pycore_typeobject.h">
       <Filter>Include\internal</Filter>
     </ClInclude>
@@ -1493,6 +1496,9 @@
     <ClCompile Include="..\Python\tracemalloc.c">
       <Filter>Python</Filter>
     </ClCompile>
+    <ClCompile Include="..\Python\typeid.c">
+      <Filter>Python</Filter>
+    </ClCompile>
     <ClCompile Include="..\Python\bootstrap_hash.c">
       <Filter>Python</Filter>
     </ClCompile>