bpo-46329: Split calls into precall and call instructions. (GH-30855)

* Add PRECALL_FUNCTION opcode.

* Move 'call shape' varaibles into struct.

* Replace CALL_NO_KW and CALL_KW with KW_NAMES and CALL instructions.

* Specialize for builtin methods taking using the METH_FASTCALL | METH_KEYWORDS protocol.

* Allow kwnames for specialized calls to builtin types.

* Specialize calls to tuple(arg) and str(arg).

Author: markshannon

Doc/library/dis.rst

@@ -34,10 +34,13 @@ the following command can be used to display the disassembly of
 :func:`myfunc`::
 
    >>> dis.dis(myfunc)
-     2           0 LOAD_GLOBAL              0 (len)
-                 2 LOAD_FAST                0 (alist)
-                 4 CALL_NO_KW               1
-                 6 RETURN_VALUE
+     1           0 RESUME                   0
+
+     2           2 LOAD_GLOBAL              0 (len)
+                 4 LOAD_FAST                0 (alist)
+                 6 PRECALL_FUNCTION         1
+                 8 CALL                     0
+                10 RETURN_VALUE
 
 (The "2" is a line number).
 
@@ -102,9 +105,11 @@ Example::
     >>> for instr in bytecode:
     ...     print(instr.opname)
     ...
+    RESUME
     LOAD_GLOBAL
     LOAD_FAST
-    CALL_NO_KW
+    PRECALL_FUNCTION
+    CALL
     RETURN_VALUE
 
 
@@ -617,7 +622,7 @@ iterations of the loop.
 .. opcode:: LOAD_BUILD_CLASS
 
    Pushes :func:`builtins.__build_class__` onto the stack.  It is later called
-   by :opcode:`CALL_NO_KW` to construct a class.
+   to construct a class.
 
 
 .. opcode:: BEFORE_WITH (delta)
@@ -1058,30 +1063,19 @@ iterations of the loop.
      with ``__cause__`` set to ``TOS``)
 
 
-.. opcode:: CALL_NO_KW (argc)
-
-   Calls a callable object with positional arguments.
-   *argc* indicates the number of positional arguments.
-   The top of the stack contains positional arguments, with the right-most
-   argument on top.  Below the arguments is a callable object to call.
-   ``CALL_NO_KW`` pops all arguments and the callable object off the stack,
-   calls the callable object with those arguments, and pushes the return value
-   returned by the callable object.
-
-   .. versionadded:: 3.11
+.. opcode:: CALL (named)
 
+   Calls a callable object with the number of positional arguments specified by
+   the preceding :opcode:`PRECALL_FUNCTION` or :opcode:`PRECALL_METHOD` and
+   the named arguments specified by the preceding :opcode:`KW_NAMES`, if any.
+   *named* indicates the number of named arguments.
+   On the stack are (in ascending order):
 
-.. opcode:: CALL_KW (argc)
+   * The callable
+   * The positional arguments
+   * The named arguments
 
-   Calls a callable object with positional (if any) and keyword arguments.
-   *argc* indicates the total number of positional and keyword arguments.
-   The top element on the stack contains a tuple with the names of the
-   keyword arguments, which must be strings.
-   Below that are the values for the keyword arguments,
-   in the order corresponding to the tuple.
-   Below that are positional arguments, with the right-most parameter on
-   top.  Below the arguments is a callable object to call.
-   ``CALL_KW`` pops all arguments and the callable object off the stack,
+   ``CALL`` pops all arguments and the callable object off the stack,
    calls the callable object with those arguments, and pushes the return value
    returned by the callable object.
 
@@ -1108,7 +1102,7 @@ iterations of the loop.
    Loads a method named ``co_names[namei]`` from the TOS object. TOS is popped.
    This bytecode distinguishes two cases: if TOS has a method with the correct
    name, the bytecode pushes the unbound method and TOS. TOS will be used as
-   the first argument (``self``) by :opcode:`CALL_METHOD` when calling the
+   the first argument (``self``) by :opcode:`PRECALL_METHOD` when calling the
    unbound method. Otherwise, ``NULL`` and the object return by the attribute
    lookup are pushed.
 
@@ -1117,14 +1111,30 @@ iterations of the loop.
 
 .. opcode:: PRECALL_METHOD (argc)
 
-   Prefixes either :opcode:`CALL_NO_KW` or :opcode:`CALL_KW`.
+   Prefixes :opcode:`CALL` (possibly with an intervening ``KW_NAMES``).
    This opcode is designed to be used with :opcode:`LOAD_METHOD`.
-   Sets internal variables, so that :opcode:`CALL_NO_KW` or :opcode:`CALL_KW`
+   Sets internal variables, so that :opcode:`CALL`
    clean up after :opcode:`LOAD_METHOD` correctly.
 
    .. versionadded:: 3.11
 
 
+.. opcode:: PRECALL_FUNCTION (args)
+
+   Prefixes :opcode:`CALL` (possibly with an intervening ``KW_NAMES``).
+   Sets internal variables, so that :opcode:`CALL` can execute correctly.
+
+   .. versionadded:: 3.11
+
+
+.. opcode:: KW_NAMES (i)
+
+   Stores a reference to ``co_consts[consti]`` into an internal variable
+   for use by :opcode:`CALL`. ``co_consts[consti]`` must be a tuple of strings.
+
+   .. versionadded:: 3.11
+
+
 .. opcode:: MAKE_FUNCTION (flags)
 
    Pushes a new function object on the stack.  From bottom to top, the consumed

Doc/whatsnew/3.11.rst

@@ -400,7 +400,8 @@ CPython bytecode changes
 
 * Replaced the three call instructions: :opcode:`CALL_FUNCTION`,
   :opcode:`CALL_FUNCTION_KW` and :opcode:`CALL_METHOD` with
-  :opcode:`CALL_NO_KW`, :opcode:`CALL_KW` and :opcode:`PRECALL_METHOD`.
+  :opcode:`PRECALL_FUNCTION`, :opcode:`PRECALL_METHOD`, :opcode:`CALL`,
+  and :opcode:`KW_NAMES`.
   This decouples the argument shifting for methods from the handling of
   keyword arguments and allows better specialization of calls.
 

Include/internal/pycore_code.h

@@ -38,7 +38,7 @@ typedef struct {
 
 typedef struct {
     uint32_t func_version;
-    uint16_t defaults_start;
+    uint16_t min_args;
     uint16_t defaults_len;
 } _PyCallCache;
 
@@ -271,7 +271,8 @@ int _Py_Specialize_LoadGlobal(PyObject *globals, PyObject *builtins, _Py_CODEUNI
 int _Py_Specialize_LoadMethod(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, SpecializedCacheEntry *cache);
 int _Py_Specialize_BinarySubscr(PyObject *sub, PyObject *container, _Py_CODEUNIT *instr, SpecializedCacheEntry *cache);
 int _Py_Specialize_StoreSubscr(PyObject *container, PyObject *sub, _Py_CODEUNIT *instr);
-int _Py_Specialize_CallNoKw(PyObject *callable, _Py_CODEUNIT *instr, int nargs, SpecializedCacheEntry *cache, PyObject *builtins);
+int _Py_Specialize_CallNoKw(PyObject *callable, _Py_CODEUNIT *instr, int nargs,
+    PyObject *kwnames, SpecializedCacheEntry *cache, PyObject *builtins);
 void _Py_Specialize_BinaryOp(PyObject *lhs, PyObject *rhs, _Py_CODEUNIT *instr,
                              SpecializedCacheEntry *cache);
 void _Py_Specialize_CompareOp(PyObject *lhs, PyObject *rhs, _Py_CODEUNIT *instr, SpecializedCacheEntry *cache);

Include/internal/pycore_frame.h

@@ -189,6 +189,9 @@ _PyThreadState_BumpFramePointer(PyThreadState *tstate, size_t size)
 
 void _PyThreadState_PopFrame(PyThreadState *tstate, InterpreterFrame *frame);
 
+InterpreterFrame *
+_PyFrame_Push(PyThreadState *tstate, PyFunctionObject *func);
+
 #ifdef __cplusplus
 }
 #endif

Include/opcode.h

@@ -384,9 +384,14 @@ def _write_atomic(path, data, mode=0o666):
 #     Python 3.11a5 3476 (Add ASYNC_GEN_WRAP opcode)
 #     Python 3.11a5 3477 (Replace DUP_TOP/DUP_TOP_TWO with COPY and
 #                         ROT_TWO/ROT_THREE/ROT_FOUR/ROT_N with SWAP)
+#     Python 3.11a5 3478 (New CALL opcodes)
 
 #     Python 3.12 will start with magic number 3500
 
+
+#     Python 3.12 will start with magic number 3500
+
+
 #
 # MAGIC must change whenever the bytecode emitted by the compiler may no
 # longer be understood by older implementations of the eval loop (usually
@@ -397,7 +402,7 @@ 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 = (3477).to_bytes(2, 'little') + b'\r\n'
+MAGIC_NUMBER = (3478).to_bytes(2, 'little') + b'\r\n'
 _RAW_MAGIC_NUMBER = int.from_bytes(MAGIC_NUMBER, 'little')  # For import.c
 
 _PYCACHE = '__pycache__'

Lib/importlib/_bootstrap_external.py

@@ -188,9 +188,12 @@ def jabs_op(name, op):
 def_op('DICT_MERGE', 164)
 def_op('DICT_UPDATE', 165)
 
+def_op('PRECALL_FUNCTION', 167)
 def_op('PRECALL_METHOD', 168)
-def_op('CALL_NO_KW', 169)
-def_op('CALL_KW', 170)
+
+def_op('CALL', 171)
+def_op('KW_NAMES', 172)
+hasconst.append(172)
 
 del def_op, name_op, jrel_op, jabs_op
 
@@ -245,16 +248,21 @@ def jabs_op(name, op):
     "STORE_SUBSCR_ADAPTIVE",
     "STORE_SUBSCR_LIST_INT",
     "STORE_SUBSCR_DICT",
-    "CALL_NO_KW_ADAPTIVE",
+    "CALL_ADAPTIVE",
+    "CALL_BUILTIN_CLASS",
     "CALL_NO_KW_BUILTIN_O",
     "CALL_NO_KW_BUILTIN_FAST",
+    "CALL_BUILTIN_FAST_WITH_KEYWORDS",
     "CALL_NO_KW_LEN",
     "CALL_NO_KW_ISINSTANCE",
-    "CALL_NO_KW_PY_SIMPLE",
+    "CALL_PY_EXACT_ARGS",
+    "CALL_PY_WITH_DEFAULTS",
     "CALL_NO_KW_LIST_APPEND",
     "CALL_NO_KW_METHOD_DESCRIPTOR_O",
+    "CALL_NO_KW_METHOD_DESCRIPTOR_NOARGS",
+    "CALL_NO_KW_STR_1",
+    "CALL_NO_KW_TUPLE_1",
     "CALL_NO_KW_TYPE_1",
-    "CALL_NO_KW_BUILTIN_CLASS_1",
     "CALL_NO_KW_METHOD_DESCRIPTOR_FAST",
     "JUMP_ABSOLUTE_QUICK",
     "LOAD_ATTR_ADAPTIVE",

Lib/opcode.py

@@ -837,9 +837,8 @@ def foo(x):
                 opcodes = list(dis.get_instructions(func))
                 instructions = [opcode.opname for opcode in opcodes]
                 self.assertNotIn('LOAD_METHOD', instructions)
-                self.assertNotIn('CALL_METHOD', instructions)
                 self.assertIn('LOAD_ATTR', instructions)
-                self.assertIn('CALL_NO_KW', instructions)
+                self.assertIn('PRECALL_FUNCTION', instructions)
 
     def test_lineno_procedure_call(self):
         def call():
@@ -1096,7 +1095,7 @@ def test_multiline_expression(self):
 )
 """
         compiled_code, _ = self.check_positions_against_ast(snippet)
-        self.assertOpcodeSourcePositionIs(compiled_code, 'CALL_NO_KW',
+        self.assertOpcodeSourcePositionIs(compiled_code, 'CALL',
             line=1, end_line=3, column=0, end_column=1)
 
     def test_very_long_line_end_offset(self):
@@ -1106,7 +1105,7 @@ def test_very_long_line_end_offset(self):
         snippet = f"g('{long_string}')"
 
         compiled_code, _ = self.check_positions_against_ast(snippet)
-        self.assertOpcodeSourcePositionIs(compiled_code, 'CALL_NO_KW',
+        self.assertOpcodeSourcePositionIs(compiled_code, 'CALL',
             line=1, end_line=1, column=None, end_column=None)
 
     def test_complex_single_line_expression(self):