Specialize CALL_FUNCTION_EX

Fidget-Spinner · Apr 30, 2023 · Oct 6, 2022 · Oct 6, 2022 · Oct 6, 2022 · Oct 6, 2022
commit 830084cf59e4e422424663a78677b9c1f4940197
diff --git a/Include/internal/pycore_call.h b/Include/internal/pycore_call.h
@@ -116,6 +116,16 @@ _PyObject_FastCallTstate(PyThreadState *tstate, PyObject *func, PyObject *const
     return _PyObject_VectorcallTstate(tstate, func, args, (size_t)nargs, NULL);
 }
 
+PyObject *const *
+_PyStack_UnpackDict(PyThreadState *tstate,
+    PyObject *const *args, Py_ssize_t nargs,
+    PyObject *kwargs, PyObject **p_kwnames);
+
+void
+_PyStack_UnpackDict_Free(PyObject *const *stack, Py_ssize_t nargs,
+    PyObject *kwnames);
+
+void _PyStack_UnpackDict_FreeNoDecRef(PyObject *const *stack, PyObject *kwnames);
 
 #ifdef __cplusplus
 }

diff --git a/Include/internal/pycore_code.h b/Include/internal/pycore_code.h
@@ -244,6 +244,7 @@ extern int _Py_Specialize_BinarySubscr(PyObject *sub, PyObject *container, _Py_C
 extern int _Py_Specialize_StoreSubscr(PyObject *container, PyObject *sub, _Py_CODEUNIT *instr);
 extern int _Py_Specialize_Call(PyObject *callable, _Py_CODEUNIT *instr,
                                int nargs, PyObject *kwnames);
+extern int _Py_Specialize_CallEx(PyObject *callable, _Py_CODEUNIT *instr, PyObject *args, PyObject *kwargs);
 extern void _Py_Specialize_BinaryOp(PyObject *lhs, PyObject *rhs, _Py_CODEUNIT *instr,
                                     int oparg, PyObject **locals);
 extern void _Py_Specialize_CompareOp(PyObject *lhs, PyObject *rhs,

diff --git a/Include/internal/pycore_opcode.h b/Include/internal/pycore_opcode.h
diff --git a/Include/opcode.h b/Include/opcode.h
diff --git a/Lib/importlib/_bootstrap_external.py b/Lib/importlib/_bootstrap_external.py
@@ -413,6 +413,8 @@ def _write_atomic(path, data, mode=0o666):
 #     Python 3.12a1 3507 (Set lineno of module's RESUME to 0)
 #     Python 3.12a1 3508 (Add CLEANUP_THROW)
 #     Python 3.12a1 3509 (Conditional jumps only jump forward)
+#     Python 3.12a1 3510 (Specialization/Cache for CALL_FUNCTION_EX)
+
 
 #     Python 3.13 will start with 3550
 
@@ -425,7 +427,7 @@ def _write_atomic(path, data, mode=0o666):
 # Whenever MAGIC_NUMBER is changed, the ranges in the magic_values array
 # in PC/launcher.c must also be updated.
 
-MAGIC_NUMBER = (3509).to_bytes(2, 'little') + b'\r\n'
+MAGIC_NUMBER = (3510).to_bytes(2, 'little') + b'\r\n'
 
 _RAW_MAGIC_NUMBER = int.from_bytes(MAGIC_NUMBER, 'little')  # For import.c
 

diff --git a/Lib/opcode.py b/Lib/opcode.py
@@ -311,6 +311,12 @@ def pseudo_op(name, op, real_ops):
         "CALL_NO_KW_TUPLE_1",
         "CALL_NO_KW_TYPE_1",
     ],
+    "CALL_FUNCTION_EX": [
+        "CALL_FUNCTION_EX_ADAPTIVE",
+        "CALL_FUNCTION_EX_PY_KWARGS",
+        "CALL_FUNCTION_EX_PY_NO_KWARGS",
+        "CALL_FUNCTION_EX_BUILTIN_PYCFUNCTIONWITHKEYWORDS",
+    ],
     "COMPARE_OP": [
         "COMPARE_OP_ADAPTIVE",
         "COMPARE_OP_FLOAT_JUMP",
@@ -434,6 +440,11 @@ def pseudo_op(name, op, real_ops):
         "func_version": 2,
         "min_args": 1,
     },
+    "CALL_FUNCTION_EX": {
+        "counter": 1,
+        "func_version": 2,
+        "min_args": 1,
+    },
     "STORE_SUBSCR": {
         "counter": 1,
     },

@@ -7,16 +7,6 @@
 #include "pycore_tuple.h"         // _PyTuple_ITEMS()
 
 
-static PyObject *const *
-_PyStack_UnpackDict(PyThreadState *tstate,
-                    PyObject *const *args, Py_ssize_t nargs,
-                    PyObject *kwargs, PyObject **p_kwnames);
-
-static void
-_PyStack_UnpackDict_Free(PyObject *const *stack, Py_ssize_t nargs,
-                         PyObject *kwnames);
-
-
 static PyObject *
 null_error(PyThreadState *tstate)
 {
@@ -964,7 +954,7 @@ _PyStack_AsDict(PyObject *const *values, PyObject *kwnames)
    The newly allocated argument vector supports PY_VECTORCALL_ARGUMENTS_OFFSET.
 
    When done, you must call _PyStack_UnpackDict_Free(stack, nargs, kwnames) */
-static PyObject *const *
+PyObject *const *
 _PyStack_UnpackDict(PyThreadState *tstate,
                     PyObject *const *args, Py_ssize_t nargs,
                     PyObject *kwargs, PyObject **p_kwnames)
@@ -1036,14 +1026,20 @@ _PyStack_UnpackDict(PyThreadState *tstate,
     return stack;
 }
 
-static void
+void
 _PyStack_UnpackDict_Free(PyObject *const *stack, Py_ssize_t nargs,
                          PyObject *kwnames)
 {
     Py_ssize_t n = PyTuple_GET_SIZE(kwnames) + nargs;
     for (Py_ssize_t i = 0; i < n; i++) {
         Py_DECREF(stack[i]);
     }
+    _PyStack_UnpackDict_FreeNoDecRef(stack, kwnames);
+}
+
+void
+_PyStack_UnpackDict_FreeNoDecRef(PyObject *const *stack, PyObject *kwnames)
+{
     PyMem_Free((PyObject **)stack - 1);
     Py_DECREF(kwnames);
 }