Specialize __setitem__ dunder method for STORE_SUBSCR

This is precursuor for JIT improvements but also results in
standard improvements due to specializing the bytecode

Author: johng

Include/internal/pycore_call.h

@@ -39,6 +39,13 @@ extern PyObject* _PyObject_Call_Prepend(
     PyObject *args,
     PyObject *kwargs);
 
+PyAPI_FUNC(int) _PyObject_CallSetItemDunder(
+    PyThreadState *tstate,
+    PyObject *func,
+    PyObject *self,
+    PyObject *key,
+    PyObject *value);
+
 extern PyObject* _PyObject_VectorcallDictTstate(
     PyThreadState *tstate,
     PyObject *callable,

Include/internal/pycore_code.h

@@ -131,6 +131,7 @@ typedef struct {
 
 typedef struct {
     _Py_BackoffCounter counter;
+    uint16_t version[2];
 } _PyStoreSubscrCache;
 
 #define INLINE_CACHE_ENTRIES_STORE_SUBSCR CACHE_ENTRIES(_PyStoreSubscrCache)

Include/internal/pycore_magic_number.h

@@ -303,6 +303,7 @@ Known values:
     Python 3.16a1 3703 (Replace DELETE_GLOBAL with PUSH_NULL; STORE_GLOBAL)
     Python 3.16a1 3704 (Replace DELETE_ATTR with PUSH_NULL; STORE_ATTR)
     Python 3.16a1 3705 (Add INTRINSIC_ADD_CONDITIONAL_ANNOTATION)
+    Python 3.16a1 3706 (Add STORE_SUBSCR_PY_DUNDER)
 
     Python 3.17 will start with 3750
 
@@ -312,7 +313,7 @@ Known values:
 
 */
 
-#define PYC_MAGIC_NUMBER 3705
+#define PYC_MAGIC_NUMBER 3706
 /* This is equivalent to converting PYC_MAGIC_NUMBER to 2 bytes
    (little-endian) and then appending b'\r\n'. */
 #define PYC_MAGIC_NUMBER_TOKEN \

Include/internal/pycore_opcode_metadata.h

@@ -102,6 +102,7 @@
     ),
     STORE_SUBSCR=frozendict(
         counter=1,
+        version=2,
     ),
     SEND=frozendict(
         counter=1,

Include/internal/pycore_uop_ids.h

@@ -691,6 +691,36 @@ def write(items):
         opname = "BINARY_OP_SUBSCR_GETITEM"
         self.assert_races_do_not_crash(opname, get_items, read, write)
 
+    @requires_specialization
+    def test_store_subscr_py_dunder(self):
+        def get_items():
+            class C:
+                __setitem__ = lambda self, item, value: None
+
+            items = []
+            for _ in range(self.ITEMS):
+                item = C()
+                items.append(item)
+            return items
+
+        def read(items):
+            for item in items:
+                try:
+                    item[None] = None
+                except TypeError:
+                    pass
+
+        def write(items):
+            for item in items:
+                try:
+                    del item.__setitem__
+                except AttributeError:
+                    pass
+                type(item).__setitem__ = lambda self, item, value: None
+
+        opname = "STORE_SUBSCR_PY_DUNDER"
+        self.assert_races_do_not_crash(opname, get_items, read, write)
+
     @requires_specialization
     def test_binary_subscr_list_int(self):
         def get_items():