gh-106581: Split CALL_BOUND_METHOD_EXACT_ARGS into uops (#108462)

Instead of using `GO_TO_INSTRUCTION(CALL_PY_EXACT_ARGS)` we just add the macro elements of the latter to the macro for the former. This requires lengthening the uops array in struct opcode_macro_expansion. (It also required changes to stacking.py that were merged already.)

Author: gvanrossum

Include/internal/pycore_opcode_metadata.h

@@ -2940,19 +2940,18 @@ dummy_func(
             CHECK_EVAL_BREAKER();
         }
 
-        // Start out with [NULL, bound_method, arg1, arg2, ...]
-        // Transform to [callable, self, arg1, arg2, ...]
-        // Then fall through to CALL_PY_EXACT_ARGS
-        inst(CALL_BOUND_METHOD_EXACT_ARGS, (unused/1, unused/2, callable, null, unused[oparg] -- unused)) {
+        op(_CHECK_CALL_BOUND_METHOD_EXACT_ARGS, (callable, null, unused[oparg] -- callable, null, unused[oparg])) {
             DEOPT_IF(null != NULL, CALL);
             DEOPT_IF(Py_TYPE(callable) != &PyMethod_Type, CALL);
+        }
+
+        op(_INIT_CALL_BOUND_METHOD_EXACT_ARGS, (callable, unused, unused[oparg] -- func, self, unused[oparg])) {
             STAT_INC(CALL, hit);
-            PyObject *self = ((PyMethodObject *)callable)->im_self;
-            PEEK(oparg + 1) = Py_NewRef(self);  // self_or_null
-            PyObject *meth = ((PyMethodObject *)callable)->im_func;
-            PEEK(oparg + 2) = Py_NewRef(meth);  // callable
+            self = Py_NewRef(((PyMethodObject *)callable)->im_self);
+            stack_pointer[-1 - oparg] = self;  // Patch stack as it is used by _INIT_CALL_PY_EXACT_ARGS
+            func = Py_NewRef(((PyMethodObject *)callable)->im_func);
+            stack_pointer[-2 - oparg] = func;  // This is used by CALL, upon deoptimization
             Py_DECREF(callable);
-            GO_TO_INSTRUCTION(CALL_PY_EXACT_ARGS);
         }
 
         op(_CHECK_PEP_523, (--)) {
@@ -3010,6 +3009,18 @@ dummy_func(
             #endif
         }
 
+        macro(CALL_BOUND_METHOD_EXACT_ARGS) =
+            unused/1 + // Skip over the counter
+            _CHECK_PEP_523 +
+            _CHECK_CALL_BOUND_METHOD_EXACT_ARGS +
+            _INIT_CALL_BOUND_METHOD_EXACT_ARGS +
+            _CHECK_FUNCTION_EXACT_ARGS +
+            _CHECK_STACK_SPACE +
+            _INIT_CALL_PY_EXACT_ARGS +
+            SAVE_IP +  // Tier 2 only; special-cased oparg
+            SAVE_CURRENT_IP +  // Sets frame->prev_instr
+            _PUSH_FRAME;
+
         macro(CALL_PY_EXACT_ARGS) =
             unused/1 + // Skip over the counter
             _CHECK_PEP_523 +

Python/abstract_interp_cases.c.h

@@ -444,7 +444,7 @@ def write_metadata(self, metadata_filename: str, pymetadata_filename: str) -> No
             with self.out.block("struct opcode_macro_expansion", ";"):
                 self.out.emit("int nuops;")
                 self.out.emit(
-                    "struct { int16_t uop; int8_t size; int8_t offset; } uops[8];"
+                    "struct { int16_t uop; int8_t size; int8_t offset; } uops[12];"
                 )
             self.out.emit("")