gh-87729: improve hit rate of LOAD_SUPER_ATTR specialization (#104270)

4 files changed, 42 insertions, 46 deletions

diff --git a/Include/internal/pycore_code.h b/Include/internal/pycore_code.h
index c1f017f..75a23f3 100644
--- a/Include/internal/pycore_code.h
+++ b/Include/internal/pycore_code.h
@@ -53,9 +53,6 @@ typedef struct {
 
 typedef struct {
     uint16_t counter;
-    uint16_t class_version[2];
-    uint16_t self_type_version[2];
-    uint16_t method[4];
 } _PySuperAttrCache;
 
 #define INLINE_CACHE_ENTRIES_LOAD_SUPER_ATTR CACHE_ENTRIES(_PySuperAttrCache)
@@ -227,8 +224,8 @@ extern int _PyLineTable_PreviousAddressRange(PyCodeAddressRange *range);
 
 /* Specialization functions */
 
-extern void _Py_Specialize_LoadSuperAttr(PyObject *global_super, PyObject *cls, PyObject *self,
-                                         _Py_CODEUNIT *instr, PyObject *name, int load_method);
+extern void _Py_Specialize_LoadSuperAttr(PyObject *global_super, PyObject *cls,
+                                         _Py_CODEUNIT *instr, int load_method);
 extern void _Py_Specialize_LoadAttr(PyObject *owner, _Py_CODEUNIT *instr,
                                     PyObject *name);
 extern void _Py_Specialize_StoreAttr(PyObject *owner, _Py_CODEUNIT *instr,
diff --git a/Include/internal/pycore_opcode.h b/Include/internal/pycore_opcode.h
index 52ee70a..e823e1b 100644
--- a/Include/internal/pycore_opcode.h
+++ b/Include/internal/pycore_opcode.h
@@ -42,7 +42,7 @@ const uint8_t _PyOpcode_Caches[256] = {
     [LOAD_GLOBAL] = 4,
     [BINARY_OP] = 1,
     [SEND] = 1,
-    [LOAD_SUPER_ATTR] = 9,
+    [LOAD_SUPER_ATTR] = 1,
     [CALL] = 3,
 };
 
@@ -182,6 +182,7 @@ const uint8_t _PyOpcode_Deopt[256] = {
     [LOAD_GLOBAL_MODULE] = LOAD_GLOBAL,
     [LOAD_NAME] = LOAD_NAME,
     [LOAD_SUPER_ATTR] = LOAD_SUPER_ATTR,
+    [LOAD_SUPER_ATTR_ATTR] = LOAD_SUPER_ATTR,
     [LOAD_SUPER_ATTR_METHOD] = LOAD_SUPER_ATTR,
     [MAKE_CELL] = MAKE_CELL,
     [MAKE_FUNCTION] = MAKE_FUNCTION,
@@ -307,29 +308,29 @@ static const char *const _PyOpcode_OpName[267] = {
     [FOR_ITER_TUPLE] = "FOR_ITER_TUPLE",
     [FOR_ITER_RANGE] = "FOR_ITER_RANGE",
     [FOR_ITER_GEN] = "FOR_ITER_GEN",
+    [LOAD_SUPER_ATTR_ATTR] = "LOAD_SUPER_ATTR_ATTR",
     [LOAD_SUPER_ATTR_METHOD] = "LOAD_SUPER_ATTR_METHOD",
-    [LOAD_ATTR_CLASS] = "LOAD_ATTR_CLASS",
     [GET_ITER] = "GET_ITER",
     [GET_YIELD_FROM_ITER] = "GET_YIELD_FROM_ITER",
-    [LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN] = "LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN",
+    [LOAD_ATTR_CLASS] = "LOAD_ATTR_CLASS",
     [LOAD_BUILD_CLASS] = "LOAD_BUILD_CLASS",
+    [LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN] = "LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN",
     [LOAD_ATTR_INSTANCE_VALUE] = "LOAD_ATTR_INSTANCE_VALUE",
-    [LOAD_ATTR_MODULE] = "LOAD_ATTR_MODULE",
     [LOAD_ASSERTION_ERROR] = "LOAD_ASSERTION_ERROR",
     [RETURN_GENERATOR] = "RETURN_GENERATOR",
+    [LOAD_ATTR_MODULE] = "LOAD_ATTR_MODULE",
     [LOAD_ATTR_PROPERTY] = "LOAD_ATTR_PROPERTY",
     [LOAD_ATTR_SLOT] = "LOAD_ATTR_SLOT",
     [LOAD_ATTR_WITH_HINT] = "LOAD_ATTR_WITH_HINT",
     [LOAD_ATTR_METHOD_LAZY_DICT] = "LOAD_ATTR_METHOD_LAZY_DICT",
     [LOAD_ATTR_METHOD_NO_DICT] = "LOAD_ATTR_METHOD_NO_DICT",
     [LOAD_ATTR_METHOD_WITH_VALUES] = "LOAD_ATTR_METHOD_WITH_VALUES",
-    [LOAD_CONST__LOAD_FAST] = "LOAD_CONST__LOAD_FAST",
     [RETURN_VALUE] = "RETURN_VALUE",
-    [LOAD_FAST__LOAD_CONST] = "LOAD_FAST__LOAD_CONST",
+    [LOAD_CONST__LOAD_FAST] = "LOAD_CONST__LOAD_FAST",
     [SETUP_ANNOTATIONS] = "SETUP_ANNOTATIONS",
+    [LOAD_FAST__LOAD_CONST] = "LOAD_FAST__LOAD_CONST",
     [LOAD_FAST__LOAD_FAST] = "LOAD_FAST__LOAD_FAST",
     [LOAD_GLOBAL_BUILTIN] = "LOAD_GLOBAL_BUILTIN",
-    [LOAD_GLOBAL_MODULE] = "LOAD_GLOBAL_MODULE",
     [POP_EXCEPT] = "POP_EXCEPT",
     [STORE_NAME] = "STORE_NAME",
     [DELETE_NAME] = "DELETE_NAME",
@@ -352,9 +353,9 @@ static const char *const _PyOpcode_OpName[267] = {
     [IMPORT_NAME] = "IMPORT_NAME",
     [IMPORT_FROM] = "IMPORT_FROM",
     [JUMP_FORWARD] = "JUMP_FORWARD",
+    [LOAD_GLOBAL_MODULE] = "LOAD_GLOBAL_MODULE",
     [STORE_ATTR_INSTANCE_VALUE] = "STORE_ATTR_INSTANCE_VALUE",
     [STORE_ATTR_SLOT] = "STORE_ATTR_SLOT",
-    [STORE_ATTR_WITH_HINT] = "STORE_ATTR_WITH_HINT",
     [POP_JUMP_IF_FALSE] = "POP_JUMP_IF_FALSE",
     [POP_JUMP_IF_TRUE] = "POP_JUMP_IF_TRUE",
     [LOAD_GLOBAL] = "LOAD_GLOBAL",
@@ -394,22 +395,22 @@ static const char *const _PyOpcode_OpName[267] = {
     [YIELD_VALUE] = "YIELD_VALUE",
     [RESUME] = "RESUME",
     [MATCH_CLASS] = "MATCH_CLASS",
+    [STORE_ATTR_WITH_HINT] = "STORE_ATTR_WITH_HINT",
     [STORE_FAST__LOAD_FAST] = "STORE_FAST__LOAD_FAST",
-    [STORE_FAST__STORE_FAST] = "STORE_FAST__STORE_FAST",
     [FORMAT_VALUE] = "FORMAT_VALUE",
     [BUILD_CONST_KEY_MAP] = "BUILD_CONST_KEY_MAP",
     [BUILD_STRING] = "BUILD_STRING",
+    [STORE_FAST__STORE_FAST] = "STORE_FAST__STORE_FAST",
     [STORE_SUBSCR_DICT] = "STORE_SUBSCR_DICT",
     [STORE_SUBSCR_LIST_INT] = "STORE_SUBSCR_LIST_INT",
     [UNPACK_SEQUENCE_LIST] = "UNPACK_SEQUENCE_LIST",
-    [UNPACK_SEQUENCE_TUPLE] = "UNPACK_SEQUENCE_TUPLE",
     [LIST_EXTEND] = "LIST_EXTEND",
     [SET_UPDATE] = "SET_UPDATE",
     [DICT_MERGE] = "DICT_MERGE",
     [DICT_UPDATE] = "DICT_UPDATE",
+    [UNPACK_SEQUENCE_TUPLE] = "UNPACK_SEQUENCE_TUPLE",
     [UNPACK_SEQUENCE_TWO_TUPLE] = "UNPACK_SEQUENCE_TWO_TUPLE",
     [SEND_GEN] = "SEND_GEN",
-    [168] = "<168>",
     [169] = "<169>",
     [170] = "<170>",
     [CALL] = "CALL",
@@ -512,7 +513,6 @@ static const char *const _PyOpcode_OpName[267] = {
 #endif
 
 #define EXTRA_CASES \
-    case 168: \
     case 169: \
     case 170: \
     case 175: \
diff --git a/Include/internal/pycore_typeobject.h b/Include/internal/pycore_typeobject.h
index f42f8f6..c7d0000 100644
--- a/Include/internal/pycore_typeobject.h
+++ b/Include/internal/pycore_typeobject.h
@@ -142,8 +142,6 @@ PyAPI_DATA(PyTypeObject) _PyBufferWrapper_Type;
 
 PyObject *
 _PySuper_Lookup(PyTypeObject *su_type, PyObject *su_obj, PyObject *name, int *meth_found);
-PyObject *
-_PySuper_LookupDescr(PyTypeObject *su_type, PyObject *su_obj, PyObject *name);
 
 #ifdef __cplusplus
 }
diff --git a/Include/opcode.h b/Include/opcode.h
index f6f4af8..c3b6e5d 100644
--- a/Include/opcode.h
+++ b/Include/opcode.h
@@ -185,33 +185,34 @@ extern "C" {
 #define FOR_ITER_TUPLE                          63
 #define FOR_ITER_RANGE                          64
 #define FOR_ITER_GEN                            65
-#define LOAD_SUPER_ATTR_METHOD                  66
-#define LOAD_ATTR_CLASS                         67
-#define LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN       70
-#define LOAD_ATTR_INSTANCE_VALUE                72
-#define LOAD_ATTR_MODULE                        73
-#define LOAD_ATTR_PROPERTY                      76
-#define LOAD_ATTR_SLOT                          77
-#define LOAD_ATTR_WITH_HINT                     78
-#define LOAD_ATTR_METHOD_LAZY_DICT              79
-#define LOAD_ATTR_METHOD_NO_DICT                80
-#define LOAD_ATTR_METHOD_WITH_VALUES            81
-#define LOAD_CONST__LOAD_FAST                   82
-#define LOAD_FAST__LOAD_CONST                   84
-#define LOAD_FAST__LOAD_FAST                    86
-#define LOAD_GLOBAL_BUILTIN                     87
-#define LOAD_GLOBAL_MODULE                      88
-#define STORE_ATTR_INSTANCE_VALUE              111
-#define STORE_ATTR_SLOT                        112
-#define STORE_ATTR_WITH_HINT                   113
-#define STORE_FAST__LOAD_FAST                  153
-#define STORE_FAST__STORE_FAST                 154
-#define STORE_SUBSCR_DICT                      158
-#define STORE_SUBSCR_LIST_INT                  159
-#define UNPACK_SEQUENCE_LIST                   160
-#define UNPACK_SEQUENCE_TUPLE                  161
-#define UNPACK_SEQUENCE_TWO_TUPLE              166
-#define SEND_GEN                               167
+#define LOAD_SUPER_ATTR_ATTR                    66
+#define LOAD_SUPER_ATTR_METHOD                  67
+#define LOAD_ATTR_CLASS                         70
+#define LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN       72
+#define LOAD_ATTR_INSTANCE_VALUE                73
+#define LOAD_ATTR_MODULE                        76
+#define LOAD_ATTR_PROPERTY                      77
+#define LOAD_ATTR_SLOT                          78
+#define LOAD_ATTR_WITH_HINT                     79
+#define LOAD_ATTR_METHOD_LAZY_DICT              80
+#define LOAD_ATTR_METHOD_NO_DICT                81
+#define LOAD_ATTR_METHOD_WITH_VALUES            82
+#define LOAD_CONST__LOAD_FAST                   84
+#define LOAD_FAST__LOAD_CONST                   86
+#define LOAD_FAST__LOAD_FAST                    87
+#define LOAD_GLOBAL_BUILTIN                     88
+#define LOAD_GLOBAL_MODULE                     111
+#define STORE_ATTR_INSTANCE_VALUE              112
+#define STORE_ATTR_SLOT                        113
+#define STORE_ATTR_WITH_HINT                   153
+#define STORE_FAST__LOAD_FAST                  154
+#define STORE_FAST__STORE_FAST                 158
+#define STORE_SUBSCR_DICT                      159
+#define STORE_SUBSCR_LIST_INT                  160
+#define UNPACK_SEQUENCE_LIST                   161
+#define UNPACK_SEQUENCE_TUPLE                  166
+#define UNPACK_SEQUENCE_TWO_TUPLE              167
+#define SEND_GEN                               168
 
 #define HAS_ARG(op) ((((op) >= HAVE_ARGUMENT) && (!IS_PSEUDO_OPCODE(op)))\
     || ((op) == JUMP) \