gh-121459: Deferred LOAD_GLOBAL (GH-123128)

Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
Co-authored-by: Sam Gross <655866+colesbury@users.noreply.github.com>

Include/internal/pycore_ceval.h

@@ -271,7 +271,7 @@ PyAPI_FUNC(PyObject **) _PyObjectArray_FromStackRefArray(_PyStackRef *input, Py_
 PyAPI_FUNC(void) _PyObjectArray_Free(PyObject **array, PyObject **scratch);
 
 PyAPI_FUNC(PyObject *) _PyEval_GetANext(PyObject *aiter);
-PyAPI_FUNC(PyObject *) _PyEval_LoadGlobal(PyObject *globals, PyObject *builtins, PyObject *name);
+PyAPI_FUNC(void) _PyEval_LoadGlobalStackRef(PyObject *globals, PyObject *builtins, PyObject *name, _PyStackRef *writeto);
 PyAPI_FUNC(PyObject *) _PyEval_GetAwaitable(PyObject *iterable, int oparg);
 PyAPI_FUNC(PyObject *) _PyEval_LoadName(PyThreadState *tstate, _PyInterpreterFrame *frame, PyObject *name);
 

Include/internal/pycore_dict.h

@@ -10,6 +10,7 @@ extern "C" {
 
 #include "pycore_object.h"               // PyManagedDictPointer
 #include "pycore_pyatomic_ft_wrappers.h" // FT_ATOMIC_LOAD_SSIZE_ACQUIRE
+#include "pycore_stackref.h"             // _PyStackRef
 
 // Unsafe flavor of PyDict_GetItemWithError(): no error checking
 extern PyObject* _PyDict_GetItemWithError(PyObject *dp, PyObject *key);
@@ -100,10 +101,12 @@ extern void _PyDictKeys_DecRef(PyDictKeysObject *keys);
  */
 extern Py_ssize_t _Py_dict_lookup(PyDictObject *mp, PyObject *key, Py_hash_t hash, PyObject **value_addr);
 extern Py_ssize_t _Py_dict_lookup_threadsafe(PyDictObject *mp, PyObject *key, Py_hash_t hash, PyObject **value_addr);
+extern Py_ssize_t _Py_dict_lookup_threadsafe_stackref(PyDictObject *mp, PyObject *key, Py_hash_t hash, _PyStackRef *value_addr);
 
 extern Py_ssize_t _PyDict_LookupIndex(PyDictObject *, PyObject *);
 extern Py_ssize_t _PyDictKeys_StringLookup(PyDictKeysObject* dictkeys, PyObject *key);
 PyAPI_FUNC(PyObject *)_PyDict_LoadGlobal(PyDictObject *, PyDictObject *, PyObject *);
+PyAPI_FUNC(void) _PyDict_LoadGlobalStackRef(PyDictObject *, PyDictObject *, PyObject *, _PyStackRef *);
 
 /* Consumes references to key and value */
 PyAPI_FUNC(int) _PyDict_SetItem_Take2(PyDictObject *op, PyObject *key, PyObject *value);

Objects/dictobject.c

@@ -1496,6 +1496,45 @@ read_failed:
     return ix;
 }
 
+Py_ssize_t
+_Py_dict_lookup_threadsafe_stackref(PyDictObject *mp, PyObject *key, Py_hash_t hash, _PyStackRef *value_addr)
+{
+    PyDictKeysObject *dk = _Py_atomic_load_ptr(&mp->ma_keys);
+    if (dk->dk_kind == DICT_KEYS_UNICODE && PyUnicode_CheckExact(key)) {
+        Py_ssize_t ix = unicodekeys_lookup_unicode_threadsafe(dk, key, hash);
+        if (ix == DKIX_EMPTY) {
+            *value_addr = PyStackRef_NULL;
+            return ix;
+        }
+        else if (ix >= 0) {
+            PyObject **addr_of_value = &DK_UNICODE_ENTRIES(dk)[ix].me_value;
+            PyObject *value = _Py_atomic_load_ptr(addr_of_value);
+            if (value == NULL) {
+                *value_addr = PyStackRef_NULL;
+                return DKIX_EMPTY;
+            }
+            if (_Py_IsImmortal(value) || _PyObject_HasDeferredRefcount(value)) {
+                *value_addr =  (_PyStackRef){ .bits = (uintptr_t)value | Py_TAG_DEFERRED };
+                return ix;
+            }
+            if (_Py_TryIncrefCompare(addr_of_value, value)) {
+                *value_addr = PyStackRef_FromPyObjectSteal(value);
+                return ix;
+            }
+        }
+    }
+
+    PyObject *obj;
+    Py_ssize_t ix = _Py_dict_lookup_threadsafe(mp, key, hash, &obj);
+    if (ix >= 0 && obj != NULL) {
+        *value_addr = PyStackRef_FromPyObjectSteal(obj);
+    }
+    else {
+        *value_addr = PyStackRef_NULL;
+    }
+    return ix;
+}
+
 #else   // Py_GIL_DISABLED
 
 Py_ssize_t
@@ -1506,6 +1545,15 @@ _Py_dict_lookup_threadsafe(PyDictObject *mp, PyObject *key, Py_hash_t hash, PyOb
     return ix;
 }
 
+Py_ssize_t
+_Py_dict_lookup_threadsafe_stackref(PyDictObject *mp, PyObject *key, Py_hash_t hash, _PyStackRef *value_addr)
+{
+    PyObject *val;
+    Py_ssize_t ix = _Py_dict_lookup(mp, key, hash, &val);
+	*value_addr = val == NULL ? PyStackRef_NULL : PyStackRef_FromPyObjectNew(val);
+    return ix;
+}
+
 #endif
 
 int
@@ -2420,6 +2468,32 @@ _PyDict_LoadGlobal(PyDictObject *globals, PyDictObject *builtins, PyObject *key)
     return value;
 }
 
+void
+_PyDict_LoadGlobalStackRef(PyDictObject *globals, PyDictObject *builtins, PyObject *key, _PyStackRef *res)
+{
+    Py_ssize_t ix;
+    Py_hash_t hash;
+
+    hash = _PyObject_HashFast(key);
+    if (hash == -1) {
+        *res = PyStackRef_NULL;
+        return;
+    }
+
+    /* namespace 1: globals */
+    ix = _Py_dict_lookup_threadsafe_stackref(globals, key, hash, res);
+    if (ix == DKIX_ERROR) {
+        *res = PyStackRef_NULL;
+    }
+    if (ix != DKIX_EMPTY && !PyStackRef_IsNull(*res)) {
+        return;
+    }
+
+    /* namespace 2: builtins */
+    ix = _Py_dict_lookup_threadsafe_stackref(builtins, key, hash, res);
+    assert(ix >= 0 || PyStackRef_IsNull(*res));
+}
+
 /* Consumes references to key and value */
 static int
 setitem_take2_lock_held(PyDictObject *mp, PyObject *key, PyObject *value)

Python/bytecodes.c

@@ -1526,12 +1526,12 @@ dummy_func(
             #endif  /* ENABLE_SPECIALIZATION */
         }
 
-        op(_LOAD_GLOBAL, ( -- res, null if (oparg & 1))) {
+        // res[1] because we need a pointer to res to pass it to _PyEval_LoadGlobalStackRef
+        op(_LOAD_GLOBAL, ( -- res[1], null if (oparg & 1))) {
             PyObject *name = GETITEM(FRAME_CO_NAMES, oparg>>1);
-            PyObject *res_o = _PyEval_LoadGlobal(GLOBALS(), BUILTINS(), name);
+            _PyEval_LoadGlobalStackRef(GLOBALS(), BUILTINS(), name, res);
-            ERROR_IF(res_o == NULL, error);
+            ERROR_IF(PyStackRef_IsNull(*res), error);
             null = PyStackRef_NULL;
-            res = PyStackRef_FromPyObjectSteal(res_o);
         }
 
         macro(LOAD_GLOBAL) =

Python/ceval.c

@@ -3110,15 +3110,14 @@ _PyEval_GetANext(PyObject *aiter)
     return awaitable;
 }
 
-PyObject *
+void
-_PyEval_LoadGlobal(PyObject *globals, PyObject *builtins, PyObject *name)
+_PyEval_LoadGlobalStackRef(PyObject *globals, PyObject *builtins, PyObject *name, _PyStackRef *writeto)
 {
-    PyObject *res;
     if (PyDict_CheckExact(globals) && PyDict_CheckExact(builtins)) {
-        res = _PyDict_LoadGlobal((PyDictObject *)globals,
+        _PyDict_LoadGlobalStackRef((PyDictObject *)globals,
                                     (PyDictObject *)builtins,
-                                    name);
+                                    name, writeto);
-        if (res == NULL && !PyErr_Occurred()) {
+        if (PyStackRef_IsNull(*writeto) && !PyErr_Occurred()) {
             /* _PyDict_LoadGlobal() returns NULL without raising
                 * an exception if the key doesn't exist */
             _PyEval_FormatExcCheckArg(PyThreadState_GET(), PyExc_NameError,
@@ -3128,13 +3127,16 @@ _PyEval_LoadGlobal(PyObject *globals, PyObject *builtins, PyObject *name)
     else {
         /* Slow-path if globals or builtins is not a dict */
         /* namespace 1: globals */
+        PyObject *res;
         if (PyMapping_GetOptionalItem(globals, name, &res) < 0) {
-            return NULL;
+            *writeto = PyStackRef_NULL;
+            return;
         }
         if (res == NULL) {
             /* namespace 2: builtins */
             if (PyMapping_GetOptionalItem(builtins, name, &res) < 0) {
-                return NULL;
+                *writeto = PyStackRef_NULL;
+                return;
             }
             if (res == NULL) {
                 _PyEval_FormatExcCheckArg(
@@ -3142,8 +3144,8 @@ _PyEval_LoadGlobal(PyObject *globals, PyObject *builtins, PyObject *name)
                             NAME_ERROR_MSG, name);
             }
         }
+        *writeto = PyStackRef_FromPyObjectSteal(res);
     }
-    return res;
 }
 
 PyObject *

Python/executor_cases.c.h

@@ -1661,15 +1661,14 @@
         }
 
         case _LOAD_GLOBAL: {
-            _PyStackRef res;
+            _PyStackRef *res;
             _PyStackRef null = PyStackRef_NULL;
             oparg = CURRENT_OPARG();
+            res = &stack_pointer[0];
             PyObject *name = GETITEM(FRAME_CO_NAMES, oparg>>1);
-            PyObject *res_o = _PyEval_LoadGlobal(GLOBALS(), BUILTINS(), name);
+            _PyEval_LoadGlobalStackRef(GLOBALS(), BUILTINS(), name, res);
-            if (res_o == NULL) JUMP_TO_ERROR();
+            if (PyStackRef_IsNull(*res)) JUMP_TO_ERROR();
             null = PyStackRef_NULL;
-            res = PyStackRef_FromPyObjectSteal(res_o);
-            stack_pointer[0] = res;
             if (oparg & 1) stack_pointer[1] = null;
             stack_pointer += 1 + (oparg & 1);
             assert(WITHIN_STACK_BOUNDS());

Python/generated_cases.c.h

@@ -5674,7 +5674,7 @@
             PREDICTED(LOAD_GLOBAL);
             _Py_CODEUNIT *this_instr = next_instr - 5;
             (void)this_instr;
-            _PyStackRef res;
+            _PyStackRef *res;
             _PyStackRef null = PyStackRef_NULL;
             // _SPECIALIZE_LOAD_GLOBAL
             {
@@ -5696,13 +5696,12 @@
             /* Skip 1 cache entry */
             // _LOAD_GLOBAL
             {
+                res = &stack_pointer[0];
                 PyObject *name = GETITEM(FRAME_CO_NAMES, oparg>>1);
-                PyObject *res_o = _PyEval_LoadGlobal(GLOBALS(), BUILTINS(), name);
+                _PyEval_LoadGlobalStackRef(GLOBALS(), BUILTINS(), name, res);
-                if (res_o == NULL) goto error;
+                if (PyStackRef_IsNull(*res)) goto error;
                 null = PyStackRef_NULL;
-                res = PyStackRef_FromPyObjectSteal(res_o);
             }
-            stack_pointer[0] = res;
             if (oparg & 1) stack_pointer[1] = null;
             stack_pointer += 1 + (oparg & 1);
             assert(WITHIN_STACK_BOUNDS());

Python/optimizer_cases.c.h

@@ -829,11 +829,13 @@
         }
 
         case _LOAD_GLOBAL: {
-            _Py_UopsSymbol *res;
+            _Py_UopsSymbol **res;
             _Py_UopsSymbol *null = NULL;
-            res = sym_new_not_null(ctx);
+            res = &stack_pointer[0];
+            for (int _i = 1; --_i >= 0;) {
+                res[_i] = sym_new_not_null(ctx);
+            }
             null = sym_new_null(ctx);
-            stack_pointer[0] = res;
             if (oparg & 1) stack_pointer[1] = null;
             stack_pointer += 1 + (oparg & 1);
             assert(WITHIN_STACK_BOUNDS());