[draft] Register machine - Add tmps in frame and compiler. Add RETURN_VALUE_R.

Include/cpython/code.h

@@ -54,6 +54,7 @@ typedef struct {
      * - co_posonlyargcount                                                    \
      * - co_kwonlyargcount                                                     \
      * - co_nlocals                                                            \
+     * - co_ntmps
      * - co_stacksize                                                          \
      * - co_flags                                                              \
      * - co_firstlineno                                                        \
@@ -79,6 +80,7 @@ typedef struct {
     int co_argcount;              /* #arguments, except *args */               \
     int co_posonlyargcount;       /* #positional only arguments */             \
     int co_kwonlyargcount;        /* #keyword only arguments */                \
+    int co_ntmps;                 /* number of tmps needed for evaluation */   \
     int co_stacksize;             /* #entries needed for evaluation stack */   \
     int co_firstlineno;           /* first source line number */               \
                                                                                \
@@ -166,13 +168,13 @@ static inline int PyCode_GetFirstFree(PyCodeObject *op) {
 
 /* Public interface */
 PyAPI_FUNC(PyCodeObject *) PyCode_New(
-        int, int, int, int, int, PyObject *, PyObject *,
+        int, int, int, int, int, int, PyObject *, PyObject *,
         PyObject *, PyObject *, PyObject *, PyObject *,
         PyObject *, PyObject *, PyObject *, int, PyObject *,
         PyObject *);
 
 PyAPI_FUNC(PyCodeObject *) PyCode_NewWithPosOnlyArgs(
-        int, int, int, int, int, int, PyObject *, PyObject *,
+        int, int, int, int, int, int, int, PyObject *, PyObject *,
         PyObject *, PyObject *, PyObject *, PyObject *,
         PyObject *, PyObject *, PyObject *, int, PyObject *,
         PyObject *);

Include/internal/pycore_code.h

@@ -171,6 +171,7 @@ struct _PyCodeConstructor {
     int kwonlyargcount;
 
     /* needed to create the frame */
+    int ntmps;
     int stacksize;
 
     /* used by the eval loop */
@@ -466,6 +467,7 @@ _PyCode_LineNumberFromArray(PyCodeObject *co, int index)
 typedef struct _PyShimCodeDef {
     const uint8_t *code;
     int codelen;
+    int ntmps;
     int stacksize;
     const char *cname;
 } _PyShimCodeDef;

Include/internal/pycore_frame.h

@@ -71,18 +71,23 @@ typedef struct _PyInterpreterFrame {
 #define _PyInterpreterFrame_LASTI(IF) \
     ((int)((IF)->prev_instr - _PyCode_CODE((IF)->f_code)))
 
+static inline int _PyFrame_StackbaseIndex(_PyInterpreterFrame *f) {
+    PyCodeObject *code = f->f_code;
+    return code->co_nlocalsplus + code->co_ntmps;
+}
+
 static inline PyObject **_PyFrame_Stackbase(_PyInterpreterFrame *f) {
-    return f->localsplus + f->f_code->co_nlocalsplus;
+    return f->localsplus + _PyFrame_StackbaseIndex(f);
 }
 
 static inline PyObject *_PyFrame_StackPeek(_PyInterpreterFrame *f) {
-    assert(f->stacktop > f->f_code->co_nlocalsplus);
+    assert(f->stacktop > _PyFrame_StackbaseIndex(f));
     assert(f->localsplus[f->stacktop-1] != NULL);
     return f->localsplus[f->stacktop-1];
 }
 
 static inline PyObject *_PyFrame_StackPop(_PyInterpreterFrame *f) {
-    assert(f->stacktop > f->f_code->co_nlocalsplus);
+    assert(f->stacktop > _PyFrame_StackbaseIndex(f));
     f->stacktop--;
     return f->localsplus[f->stacktop];
 }
@@ -97,10 +102,10 @@ static inline void _PyFrame_StackPush(_PyInterpreterFrame *f, PyObject *value) {
 static inline int
 _PyFrame_NumSlotsForCodeObject(PyCodeObject *code)
 {
-    /* This function needs to remain in sync with the calculation of
-     * co_framesize in Tools/build/deepfreeze.py */
     assert(code->co_framesize >= FRAME_SPECIALS_SIZE);
-    return code->co_framesize - FRAME_SPECIALS_SIZE;
+    int nslots = code->co_framesize - FRAME_SPECIALS_SIZE;
+    assert(nslots == code->co_nlocalsplus + code->co_ntmps + code->co_stacksize);
+    return nslots;
 }
 
 void _PyFrame_Copy(_PyInterpreterFrame *src, _PyInterpreterFrame *dest);
@@ -119,13 +124,14 @@ _PyFrame_Initialize(
     frame->f_builtins = func->func_builtins;
     frame->f_globals = func->func_globals;
     frame->f_locals = locals;
-    frame->stacktop = code->co_nlocalsplus;
     frame->frame_obj = NULL;
     frame->prev_instr = _PyCode_CODE(code) - 1;
     frame->yield_offset = 0;
     frame->owner = FRAME_OWNED_BY_THREAD;
 
-    for (int i = null_locals_from; i < code->co_nlocalsplus; i++) {
+    int stack_start = _PyFrame_StackbaseIndex(frame);
+    frame->stacktop = stack_start;
+    for (int i = null_locals_from; i < stack_start; i++) {
         frame->localsplus[i] = NULL;
     }
 }
@@ -142,7 +148,7 @@ _PyFrame_GetLocalsArray(_PyInterpreterFrame *frame)
 static inline PyObject**
 _PyFrame_GetStackPointer(_PyInterpreterFrame *frame)
 {
-    return frame->localsplus+frame->stacktop;
+    return frame->localsplus + frame->stacktop;
 }
 
 static inline void

Include/internal/pycore_global_strings.h

@@ -324,6 +324,7 @@ struct _Py_global_strings {
         STRUCT_FOR_ID(co_name)
         STRUCT_FOR_ID(co_names)
         STRUCT_FOR_ID(co_nlocals)
+        STRUCT_FOR_ID(co_ntmps)
         STRUCT_FOR_ID(co_posonlyargcount)
         STRUCT_FOR_ID(co_qualname)
         STRUCT_FOR_ID(co_stacksize)

Lib/importlib/_bootstrap_external.py

@@ -430,6 +430,8 @@ def _write_atomic(path, data, mode=0o666):
 #     Python 3.12a1 3514 (Remove ASYNC_GEN_WRAP, LIST_TO_TUPLE, and UNARY_POSITIVE)
 #     Python 3.12a1 3515 (Embed jump mask in COMPARE_OP oparg)
 #     Python 3.12a1 3516 (Add COMAPRE_AND_BRANCH instruction)
+#     Python 3.12a1 3517 (Add ntmps to code object)
+#     Python 3.12a1 3518 (Add RETURN_VALUE_R)
 
 #     Python 3.13 will start with 3550
 
@@ -442,7 +444,7 @@ 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 = (3516).to_bytes(2, 'little') + b'\r\n'
+MAGIC_NUMBER = (3518).to_bytes(2, 'little') + b'\r\n'
 
 _RAW_MAGIC_NUMBER = int.from_bytes(MAGIC_NUMBER, 'little')  # For import.c
 

Lib/opcode.py

@@ -197,6 +197,8 @@ def pseudo_op(name, op, real_ops):
 
 def_op('CALL_FUNCTION_EX', 142)  # Flags
 
+def_op('RETURN_VALUE_R', 143)
+
 def_op('EXTENDED_ARG', 144)
 EXTENDED_ARG = 144
 def_op('LIST_APPEND', 145)

Lib/test/test_call.py

@@ -934,6 +934,16 @@ def c_py_recurse(m):
         finally:
             sys.setrecursionlimit(depth)
 
+class TestFunctionWithManyArgs(unittest.TestCase):
+    def test_function_with_many_args(self):
+        for N in (10, 500, 1000):
+            with self.subTest(N=N):
+                args = ",".join([f"a{i}" for i in range(N)])
+                src = f"def f({args}) : return a{N//2}"
+                l = {}
+                exec(src, {}, l)
+                self.assertEqual(l['f'](*range(N)), N//2)
+
 
 if __name__ == "__main__":
     unittest.main()

Lib/test/test_code.py

@@ -225,6 +225,7 @@ def func(): pass
                         co.co_posonlyargcount,
                         co.co_kwonlyargcount,
                         co.co_nlocals,
+                        co.co_ntmps,
                         co.co_stacksize,
                         co.co_flags,
                         co.co_code,
@@ -264,6 +265,7 @@ def func2():
             ("co_posonlyargcount", 0),
             ("co_kwonlyargcount", 0),
             ("co_nlocals", 1),
+            ("co_ntmps", 13),
             ("co_stacksize", 0),
             ("co_flags", code.co_flags | inspect.CO_COROUTINE),
             ("co_firstlineno", 100),
@@ -302,6 +304,7 @@ def func():
                          co.co_kwonlyargcount,
                          # This is the only change.
                          co.co_nlocals + diff,
+                         co.co_ntmps,
                          co.co_stacksize,
                          co.co_flags,
                          co.co_code,