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>

5 files changed, 30 insertions, 28 deletions

diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index 078f06d..846404e 100644
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -1469,8 +1469,8 @@ dummy_func(
                     && PyDict_CheckExact(BUILTINS()))
                 {
                     v_o = _PyDict_LoadGlobal((PyDictObject *)GLOBALS(),
-                                            (PyDictObject *)BUILTINS(),
-                                            name);
+                                             (PyDictObject *)BUILTINS(),
+                                             name);
                     if (v_o == NULL) {
                         if (!_PyErr_Occurred(tstate)) {
                             /* _PyDict_LoadGlobal() returns NULL without raising
@@ -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);
-            ERROR_IF(res_o == NULL, error);
+            _PyEval_LoadGlobalStackRef(GLOBALS(), BUILTINS(), name, res);
+            ERROR_IF(PyStackRef_IsNull(*res), error);
             null = PyStackRef_NULL;
-            res = PyStackRef_FromPyObjectSteal(res_o);
         }
 
         macro(LOAD_GLOBAL) =
diff --git a/Python/ceval.c b/Python/ceval.c
index 252833a..6236c66 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -3110,15 +3110,14 @@ _PyEval_GetANext(PyObject *aiter)
     return awaitable;
 }
 
-PyObject *
-_PyEval_LoadGlobal(PyObject *globals, PyObject *builtins, PyObject *name)
+void
+_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);
-        if (res == NULL && !PyErr_Occurred()) {
+                                    name, writeto);
+        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 *
diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index b36fff9..93ab068 100644
--- a/Python/executor_cases.c.h
+++ b/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);
-            if (res_o == NULL) JUMP_TO_ERROR();
+            _PyEval_LoadGlobalStackRef(GLOBALS(), BUILTINS(), name, res);
+            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());
diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h
index 3e9f039..6d902e2 100644
--- a/Python/generated_cases.c.h
+++ b/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);
-                if (res_o == NULL) goto error;
+                _PyEval_LoadGlobalStackRef(GLOBALS(), BUILTINS(), name, res);
+                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());
diff --git a/Python/optimizer_cases.c.h b/Python/optimizer_cases.c.h
index e89c969..a6cfa27 100644
--- a/Python/optimizer_cases.c.h
+++ b/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());