From 3b12d1b9ccef8a2942dd37a92399ac55730aefd1 Mon Sep 17 00:00:00 2001
From: Fidget-Spinner <28750310+Fidget-Spinner@users.noreply.github.com>
Date: Sat, 29 Jan 2022 00:45:38 +0800
Subject: [PATCH 1/9] Near zero-cost super().meth() calls

---
 Include/internal/pycore_code.h                |  8 +-
 Include/internal/pycore_typeobject.h          |  5 ++
 Include/opcode.h                              |  2 +
 Lib/opcode.py                                 |  3 +
 .../2022-01-29-00-44-09.bpo-46564.MwLSHf.rst  |  5 ++
 Objects/typeobject.c                          | 81 +++++++++++------
 Python/ceval.c                                | 87 ++++++++++++++++++-
 Python/opcode_targets.h                       |  4 +-
 Python/specialize.c                           | 55 +++++++++++-
 9 files changed, 216 insertions(+), 34 deletions(-)
 create mode 100644 Misc/NEWS.d/next/Core and Builtins/2022-01-29-00-44-09.bpo-46564.MwLSHf.rst

diff --git a/Include/internal/pycore_code.h b/Include/internal/pycore_code.h
index bbf7a06189ba10..ca7a878d215504 100644
--- a/Include/internal/pycore_code.h
+++ b/Include/internal/pycore_code.h
@@ -263,6 +263,9 @@ cache_backoff(_PyAdaptiveEntry *entry) {
     entry->counter = ADAPTIVE_CACHE_BACKOFF;
 }
 
+/* _interpreter_frame is defined in pycore_frame.h */
+typedef struct _interpreter_frame InterpreterFrame;
+
 /* Specialization functions */
 
 int _Py_Specialize_LoadAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, SpecializedCacheEntry *cache);
@@ -271,7 +274,8 @@ 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, SpecializedCacheEntry *cache, PyObject *builtins, PyObject **stack_pointer,
+  InterpreterFrame *frame, PyObject *names);
 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);
@@ -279,6 +283,8 @@ void _Py_Specialize_CompareOp(PyObject *lhs, PyObject *rhs, _Py_CODEUNIT *instr,
 /* Deallocator function for static codeobjects used in deepfreeze.py */
 void _PyStaticCode_Dealloc(PyCodeObject *co);
 
+#define Py_STATS
+
 #ifdef Py_STATS
 
 #define SPECIALIZATION_FAILURE_KINDS 30
diff --git a/Include/internal/pycore_typeobject.h b/Include/internal/pycore_typeobject.h
index c480a3a57b436c..9b10398345768e 100644
--- a/Include/internal/pycore_typeobject.h
+++ b/Include/internal/pycore_typeobject.h
@@ -43,6 +43,11 @@ extern PyStatus _PyTypes_InitSlotDefs(void);
 
 extern void _PyStaticType_Dealloc(PyTypeObject *type);
 
+/* _interpreter_frame is defined in pycore_frame.h */
+typedef struct _interpreter_frame InterpreterFrame;
+
+PyObject *_PySuper_Lookup(PyTypeObject *, PyObject *, PyObject *, int *);
+int _PySuper_GetTypeArgs(InterpreterFrame *, PyCodeObject *, PyTypeObject **, PyObject **);
 
 #ifdef __cplusplus
 }
diff --git a/Include/opcode.h b/Include/opcode.h
index 02cdf42fe2443f..ea32d7e38b333f 100644
--- a/Include/opcode.h
+++ b/Include/opcode.h
@@ -167,6 +167,8 @@ extern "C" {
 #define LOAD_FAST__LOAD_CONST            80
 #define LOAD_CONST__LOAD_FAST            81
 #define STORE_FAST__STORE_FAST          131
+#define CALL_NO_KW_SUPER_0__LOAD_METHOD_CACHED 140
+#define CALL_NO_KW_SUPER_2__LOAD_METHOD_CACHED 141
 #define DO_TRACING                      255
 #ifdef NEED_OPCODE_JUMP_TABLES
 static uint32_t _PyOpcode_RelativeJump[8] = {
diff --git a/Lib/opcode.py b/Lib/opcode.py
index 37da05d76d734a..d0fc8b94565575 100644
--- a/Lib/opcode.py
+++ b/Lib/opcode.py
@@ -280,6 +280,9 @@ def jabs_op(name, op):
     "LOAD_FAST__LOAD_CONST",
     "LOAD_CONST__LOAD_FAST",
     "STORE_FAST__STORE_FAST",
+    # Specialized super instructions.
+    "CALL_NO_KW_SUPER_0__LOAD_METHOD_CACHED",
+    "CALL_NO_KW_SUPER_2__LOAD_METHOD_CACHED",
 ]
 _specialization_stats = [
     "success",
diff --git a/Misc/NEWS.d/next/Core and Builtins/2022-01-29-00-44-09.bpo-46564.MwLSHf.rst b/Misc/NEWS.d/next/Core and Builtins/2022-01-29-00-44-09.bpo-46564.MwLSHf.rst
new file mode 100644
index 00000000000000..3f2722c42cbfa6
--- /dev/null
+++ b/Misc/NEWS.d/next/Core and Builtins/2022-01-29-00-44-09.bpo-46564.MwLSHf.rst	
@@ -0,0 +1,5 @@
+Method calls on :class:`super` are sped up. The 2-argument form,
+``super(type, obj).meth()`` is now nearly as fast as an equivalent
+``self.meth()`` call. The 0-argument form, while still slower, is still
+faster than in previous versions of CPython. Patch by Ken Jin, with
+additional contributions by Vladimir Matveev.
diff --git a/Objects/typeobject.c b/Objects/typeobject.c
index 621ad9745d8448..cbc1b06adc3bca 100644
--- a/Objects/typeobject.c
+++ b/Objects/typeobject.c
@@ -8846,16 +8846,18 @@ super_repr(PyObject *self)
             "<super: <class '%s'>, NULL>",
             su->type ? su->type->tp_name : "NULL");
 }
+/* Forward */
+static PyTypeObject *supercheck(PyTypeObject *type, PyObject *obj);
 
 static PyObject *
-super_getattro(PyObject *self, PyObject *name)
+do_super_lookup(superobject *su, PyTypeObject *su_type, PyObject *su_obj,
+                PyTypeObject *su_obj_type, PyObject *name, int *meth_found)
 {
-    superobject *su = (superobject *)self;
     PyTypeObject *starttype;
     PyObject *mro;
     Py_ssize_t i, n;
 
-    starttype = su->obj_type;
+    starttype = su_obj_type;
     if (starttype == NULL)
         goto skip;
 
@@ -8875,7 +8877,7 @@ super_getattro(PyObject *self, PyObject *name)
 
     /* No need to check the last one: it's gonna be skipped anyway.  */
     for (i = 0; i+1 < n; i++) {
-        if ((PyObject *)(su->type) == PyTuple_GET_ITEM(mro, i))
+        if ((PyObject *)(su_type) == PyTuple_GET_ITEM(mro, i))
             break;
     }
     i++;  /* skip su->type (if any)  */
@@ -8893,17 +8895,25 @@ super_getattro(PyObject *self, PyObject *name)
         PyObject *res = PyDict_GetItemWithError(dict, name);
         if (res != NULL) {
             Py_INCREF(res);
-
-            descrgetfunc f = Py_TYPE(res)->tp_descr_get;
-            if (f != NULL) {
-                PyObject *res2;
-                res2 = f(res,
-                    /* Only pass 'obj' param if this is instance-mode super
-                       (See SF ID #743627)  */
-                    (su->obj == (PyObject *)starttype) ? NULL : su->obj,
-                    (PyObject *)starttype);
-                Py_DECREF(res);
-                res = res2;
+            if (meth_found &&
+                _PyType_HasFeature(Py_TYPE(res), Py_TPFLAGS_METHOD_DESCRIPTOR)) {
+                *meth_found = 1;
+            }
+            else {
+                if (meth_found) {
+                    *meth_found = 0;
+                }
+                descrgetfunc f = Py_TYPE(res)->tp_descr_get;
+                if (f != NULL) {
+                    PyObject *res2;
+                    res2 = f(res,
+                            /* Only pass 'obj' param if this is instance-mode super
+                            (See SF ID #743627)  */
+                            (su_obj == (PyObject *)starttype) ? NULL : su_obj,
+                            (PyObject *)starttype);
+                    Py_DECREF(res);
+                    res = res2;
+                }
             }
 
             Py_DECREF(mro);
@@ -8919,7 +8929,27 @@ super_getattro(PyObject *self, PyObject *name)
     Py_DECREF(mro);
 
   skip:
-    return PyObject_GenericGetAttr(self, name);
+    assert(su != NULL);
+    return PyObject_GenericGetAttr((PyObject *)su, name);
+}
+
+static PyObject *
+super_getattro(PyObject *self, PyObject *name)
+{
+    superobject *su = (superobject *)self;
+    return do_super_lookup(su, su->type, su->obj, su->obj_type, name, NULL);
+}
+
+PyObject *
+_PySuper_Lookup(PyTypeObject *su_type, PyObject *su_obj, PyObject *name, int *meth_found)
+{
+    PyTypeObject *starttype = supercheck(su_type, su_obj);
+    if (starttype == NULL) {
+        return NULL;
+    }
+    PyObject *res = do_super_lookup(NULL, su_type, su_obj, starttype, name, meth_found);
+    Py_DECREF(starttype);
+    return res;
 }
 
 static PyTypeObject *
@@ -9011,8 +9041,8 @@ super_descr_get(PyObject *self, PyObject *obj, PyObject *type)
     }
 }
 
-static int
-super_init_without_args(PyFrameObject *f, PyCodeObject *co,
+int
+_PySuper_GetTypeArgs(InterpreterFrame *f, PyCodeObject *co,
                         PyTypeObject **type_p, PyObject **obj_p)
 {
     if (co->co_argcount == 0) {
@@ -9021,13 +9051,13 @@ super_init_without_args(PyFrameObject *f, PyCodeObject *co,
         return -1;
     }
 
-    assert(f->f_frame->f_code->co_nlocalsplus > 0);
-    PyObject *firstarg = _PyFrame_GetLocalsArray(f->f_frame)[0];
+    assert(f->f_code->co_nlocalsplus > 0);
+    PyObject *firstarg = _PyFrame_GetLocalsArray(f)[0];
     // The first argument might be a cell.
     if (firstarg != NULL && (_PyLocals_GetKind(co->co_localspluskinds, 0) & CO_FAST_CELL)) {
         // "firstarg" is a cell here unless (very unlikely) super()
         // was called from the C-API before the first MAKE_CELL op.
-        if (f->f_frame->f_lasti >= 0) {
+        if (f->f_lasti >= 0) {
             assert(_Py_OPCODE(*co->co_firstinstr) == MAKE_CELL || _Py_OPCODE(*co->co_firstinstr) == COPY_FREE_VARS);
             assert(PyCell_Check(firstarg));
             firstarg = PyCell_GET(firstarg);
@@ -9047,7 +9077,7 @@ super_init_without_args(PyFrameObject *f, PyCodeObject *co,
         PyObject *name = PyTuple_GET_ITEM(co->co_localsplusnames, i);
         assert(PyUnicode_Check(name));
         if (_PyUnicode_EqualToASCIIId(name, &PyId___class__)) {
-            PyObject *cell = _PyFrame_GetLocalsArray(f->f_frame)[i];
+            PyObject *cell = _PyFrame_GetLocalsArray(f)[i];
             if (cell == NULL || !PyCell_Check(cell)) {
                 PyErr_SetString(PyExc_RuntimeError,
                   "super(): bad __class__ cell");
@@ -9096,17 +9126,14 @@ super_init(PyObject *self, PyObject *args, PyObject *kwds)
         /* Call super(), without args -- fill in from __class__
            and first local variable on the stack. */
         PyThreadState *tstate = _PyThreadState_GET();
-        PyFrameObject *frame = PyThreadState_GetFrame(tstate);
+        InterpreterFrame *frame = tstate->cframe->current_frame;
         if (frame == NULL) {
             PyErr_SetString(PyExc_RuntimeError,
                             "super(): no current frame");
             return -1;
         }
 
-        PyCodeObject *code = PyFrame_GetCode(frame);
-        int res = super_init_without_args(frame, code, &type, &obj);
-        Py_DECREF(frame);
-        Py_DECREF(code);
+        int res = _PySuper_GetTypeArgs(frame, frame->f_code, &type, &obj);
 
         if (res < 0) {
             return -1;
diff --git a/Python/ceval.c b/Python/ceval.c
index 106e4080840f44..3fddde7e2f66d1 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -4613,7 +4613,8 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
                 next_instr--;
                 int nargs = oparg+extra_args;
                 if (_Py_Specialize_CallNoKw(
-                    PEEK(nargs + 1), next_instr, nargs, cache, BUILTINS()) < 0) {
+                    PEEK(nargs + 1), next_instr, nargs, cache, BUILTINS(),
+                    stack_pointer, frame, names) < 0) {
                     goto error;
                 }
                 DISPATCH();
@@ -4930,6 +4931,90 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
             DISPATCH();
         }
 
+        TARGET(CALL_NO_KW_SUPER_0__LOAD_METHOD_CACHED) {
+            /* super().meth */
+            assert(_Py_OPCODE(next_instr[0]) == LOAD_METHOD_ADAPTIVE);
+            assert(_Py_OPCODE(next_instr[-2]) != PRECALL_METHOD);
+            SpecializedCacheEntry *caches = GET_CACHE();
+            _PyAdaptiveEntry *cache0 = &caches[0].adaptive;
+            _PyObjectCache *cache1 = &caches[-1].obj;
+            _PyAdaptiveEntry *lm_adaptive = &caches[-2].adaptive;
+            int nargs = cache0->original_oparg;
+            assert(nargs == 0);
+
+            /* CALL_NO_KW_SUPER */
+            PyObject *su_obj;
+            PyTypeObject *su_type;
+            PyObject *meth;
+            PyObject *super_callable = TOP();
+
+            DEOPT_IF(_PyType_CAST(super_callable) != &PySuper_Type, CALL_NO_KW);
+
+            /* super() - zero argument form */
+            if (_PySuper_GetTypeArgs(frame, frame->f_code, &su_type, &su_obj) < 0) {
+                PyErr_Clear();
+                DEOPT_IF(1, CALL_NO_KW);
+            }
+            assert(su_obj != NULL);
+            DEOPT_IF(lm_adaptive->version != Py_TYPE(su_obj)->tp_version_tag, CALL_NO_KW);
+            DEOPT_IF(cache0->version != su_type->tp_version_tag, CALL_NO_KW);
+
+            STAT_INC(CALL_NO_KW, hit);
+
+            /* LOAD_METHOD_CACHED */
+            meth = cache1->obj;
+            assert(meth != NULL && _PyType_HasFeature(Py_TYPE(meth), Py_TPFLAGS_METHOD_DESCRIPTOR));
+            Py_INCREF(meth);
+            SET_TOP(meth);
+            Py_INCREF(su_obj);
+            PUSH(su_obj);
+
+            Py_DECREF(super_callable);
+            next_instr++;
+            DISPATCH();
+        }
+
+        TARGET(CALL_NO_KW_SUPER_2__LOAD_METHOD_CACHED) {
+            /* super(type, obj).meth */
+            assert(_Py_OPCODE(next_instr[0]) == LOAD_METHOD_ADAPTIVE);
+            assert(_Py_OPCODE(next_instr[-2]) != PRECALL_METHOD);
+            SpecializedCacheEntry *caches = GET_CACHE();
+            _PyAdaptiveEntry *cache0 = &caches[0].adaptive;
+            _PyObjectCache *cache1 = &caches[-1].obj;
+            _PyAdaptiveEntry *lm_adaptive = &caches[-2].adaptive;
+            int nargs = cache0->original_oparg;
+            int lm_oparg = cache0->index;
+            assert(nargs == 2);
+
+            /* CALL_NO_KW_SUPER */
+            /* super(type, obj) - two argument form */
+            PyObject *su_obj = TOP();
+            PyTypeObject *su_type = _PyType_CAST(SECOND());
+            PyObject *super_callable = THIRD();
+            PyObject *meth;
+
+            DEOPT_IF(_PyType_CAST(super_callable) != &PySuper_Type, CALL_NO_KW);
+
+            assert(su_obj != NULL);
+            DEOPT_IF(lm_adaptive->version != Py_TYPE(su_obj)->tp_version_tag, CALL_NO_KW);
+            DEOPT_IF(cache0->version != su_type->tp_version_tag, CALL_NO_KW);
+
+            STAT_INC(CALL_NO_KW, hit);
+
+            (void)(POP());
+            /* LOAD_METHOD_CACHED */
+            meth = cache1->obj;
+            assert(meth != NULL && _PyType_HasFeature(Py_TYPE(meth), Py_TPFLAGS_METHOD_DESCRIPTOR));
+            Py_INCREF(meth);
+            SET_SECOND(meth);
+            SET_TOP(su_obj);
+
+            Py_DECREF(super_callable);
+            Py_DECREF(su_type);
+            next_instr++;
+            DISPATCH();
+        }
+
         TARGET(CALL_FUNCTION_EX) {
             PREDICTED(CALL_FUNCTION_EX);
             PyObject *func, *callargs, *kwargs = NULL, *result;
diff --git a/Python/opcode_targets.h b/Python/opcode_targets.h
index c6e6d826aee609..acfc6537f7eed1 100644
--- a/Python/opcode_targets.h
+++ b/Python/opcode_targets.h
@@ -139,8 +139,8 @@ static void *opcode_targets[256] = {
     &&TARGET_LOAD_DEREF,
     &&TARGET_STORE_DEREF,
     &&TARGET_DELETE_DEREF,
-    &&_unknown_opcode,
-    &&_unknown_opcode,
+    &&TARGET_CALL_NO_KW_SUPER_0__LOAD_METHOD_CACHED,
+    &&TARGET_CALL_NO_KW_SUPER_2__LOAD_METHOD_CACHED,
     &&TARGET_CALL_FUNCTION_EX,
     &&_unknown_opcode,
     &&TARGET_EXTENDED_ARG,
diff --git a/Python/specialize.c b/Python/specialize.c
index 44c006245ebfc9..8e32a3b11275a5 100644
--- a/Python/specialize.c
+++ b/Python/specialize.c
@@ -4,6 +4,7 @@
 #include "pycore_long.h"
 #include "pycore_moduleobject.h"
 #include "pycore_object.h"
+#include "pycore_frame.h"
 #include "opcode.h"
 #include "structmember.h"         // struct PyMemberDef, T_OFFSET_EX
 
@@ -1337,7 +1338,8 @@ _Py_Specialize_StoreSubscr(PyObject *container, PyObject *sub, _Py_CODEUNIT *ins
 static int
 specialize_class_call(
     PyObject *callable, _Py_CODEUNIT *instr,
-    int nargs, SpecializedCacheEntry *cache)
+    int nargs, SpecializedCacheEntry *cache, PyObject **stack_pointer,
+    InterpreterFrame *frame, PyObject *names)
 {
     PyTypeObject *tp = _PyType_CAST(callable);
     if (_Py_OPCODE(instr[-1]) == PRECALL_METHOD) {
@@ -1359,6 +1361,51 @@ specialize_class_call(
             return 0;
         }
     }
+
+    _Py_CODEUNIT next_instr = instr[1];
+    /* Adaptive super instruction of CALL_NO_KW andLOAD_METHOD_ADAPTIVE. */
+    if (tp == &PySuper_Type &&
+        _Py_OPCODE(next_instr) == LOAD_METHOD_ADAPTIVE &&
+        (nargs == 0 || nargs == 2)) {
+        /* Use load_method cache entries too. */
+        _PyAdaptiveEntry *lm_adaptive = &cache[-cache_requirements[CALL_NO_KW]].adaptive;
+        _PyObjectCache *cache1 = &cache[-1].obj;
+        PyObject *su_obj;
+        PyTypeObject *su_type;
+        PyObject *meth;
+        int meth_found;
+        PyObject *name = PyTuple_GET_ITEM(names, lm_adaptive->original_oparg);
+
+        /* Note (KJ): the following operations must not affect tp_version_tag. */
+        /* super() zero arg form. */
+        if (nargs == 0) {
+            if (_PySuper_GetTypeArgs(frame, frame->f_code, &su_type, &su_obj) < 0) {
+                PyErr_Clear();
+                goto fail;
+            }
+        }
+        /* super(su_type, su_obj) two arg form. */
+        else if (nargs == 2) {
+            su_type = _PyType_CAST(stack_pointer[-2]);
+            su_obj = stack_pointer[-1];
+        }
+        meth = _PySuper_Lookup(su_type, su_obj, name, &meth_found);
+        if (meth == NULL) {
+            assert(PyErr_Occurred());
+            PyErr_Clear();
+            goto fail;
+        }
+        if (!meth_found) {
+            goto fail;
+        }
+        cache->adaptive.version = su_type->tp_version_tag;
+        cache1->obj = meth;
+        lm_adaptive->version = Py_TYPE(su_obj)->tp_version_tag;
+        *instr = _Py_MAKECODEUNIT(nargs == 0 ? CALL_NO_KW_SUPER_0__LOAD_METHOD_CACHED
+            : CALL_NO_KW_SUPER_2__LOAD_METHOD_CACHED, _Py_OPARG(*instr));
+        return 0;
+    }
+fail:
     SPECIALIZATION_FAIL(CALL_NO_KW, SPEC_FAIL_CLASS);
     return -1;
 }
@@ -1560,7 +1607,8 @@ int
 _Py_Specialize_CallNoKw(
     PyObject *callable, _Py_CODEUNIT *instr,
     int nargs, SpecializedCacheEntry *cache,
-    PyObject *builtins)
+    PyObject *builtins, PyObject **stack_pointer, InterpreterFrame *frame,
+    PyObject *names)
 {
     int fail;
     if (PyCFunction_CheckExact(callable)) {
@@ -1570,7 +1618,8 @@ _Py_Specialize_CallNoKw(
         fail = specialize_py_call((PyFunctionObject *)callable, instr, nargs, cache);
     }
     else if (PyType_Check(callable)) {
-        fail = specialize_class_call(callable, instr, nargs, cache);
+        fail = specialize_class_call(callable, instr, nargs, cache, stack_pointer,
+            frame, names);
     }
     else if (Py_IS_TYPE(callable, &PyMethodDescr_Type)) {
         fail = specialize_method_descriptor(

From 5b7a98d77f27cbad654195c027bc597b64fdfb05 Mon Sep 17 00:00:00 2001
From: Fidget-Spinner <28750310+Fidget-Spinner@users.noreply.github.com>
Date: Sat, 29 Jan 2022 00:47:02 +0800
Subject: [PATCH 2/9] remove py_stats

---
 Include/internal/pycore_code.h | 1 -
 1 file changed, 1 deletion(-)

diff --git a/Include/internal/pycore_code.h b/Include/internal/pycore_code.h
index ca7a878d215504..682c6b4d868e2e 100644
--- a/Include/internal/pycore_code.h
+++ b/Include/internal/pycore_code.h
@@ -283,7 +283,6 @@ void _Py_Specialize_CompareOp(PyObject *lhs, PyObject *rhs, _Py_CODEUNIT *instr,
 /* Deallocator function for static codeobjects used in deepfreeze.py */
 void _PyStaticCode_Dealloc(PyCodeObject *co);
 
-#define Py_STATS
 
 #ifdef Py_STATS
 

From efb0ae37c4863b0c3bb19f8980424656d0343970 Mon Sep 17 00:00:00 2001
From: Fidget-Spinner <28750310+Fidget-Spinner@users.noreply.github.com>
Date: Sat, 29 Jan 2022 01:06:55 +0800
Subject: [PATCH 3/9] Add vladima to co-authors

Co-Authored-By: Vladimir Matveev <v2matveev@outlook.com>
---
 Include/internal/pycore_code.h | 1 -
 1 file changed, 1 deletion(-)

diff --git a/Include/internal/pycore_code.h b/Include/internal/pycore_code.h
index 682c6b4d868e2e..59e423a3d6b0ed 100644
--- a/Include/internal/pycore_code.h
+++ b/Include/internal/pycore_code.h
@@ -283,7 +283,6 @@ void _Py_Specialize_CompareOp(PyObject *lhs, PyObject *rhs, _Py_CODEUNIT *instr,
 /* Deallocator function for static codeobjects used in deepfreeze.py */
 void _PyStaticCode_Dealloc(PyCodeObject *co);
 
-
 #ifdef Py_STATS
 
 #define SPECIALIZATION_FAILURE_KINDS 30

From 9d64819e62a7aa6d5b7882bfb5e5b44c349b9058 Mon Sep 17 00:00:00 2001
From: Fidget-Spinner <28750310+Fidget-Spinner@users.noreply.github.com>
Date: Sat, 29 Jan 2022 01:38:12 +0800
Subject: [PATCH 4/9] fix merge errors

---
 Include/internal/pycore_code.h |  2 +-
 Include/opcode.h               | 77 ++++++++++++++++++----------------
 Programs/test_frozenmain.h     | 59 +++++++++++++-------------
 Python/ceval.c                 | 20 ++++-----
 Python/opcode_targets.h        |  6 +--
 Python/specialize.c            |  4 +-
 6 files changed, 85 insertions(+), 83 deletions(-)

diff --git a/Include/internal/pycore_code.h b/Include/internal/pycore_code.h
index b77bab7f7546b5..f81829733d94c5 100644
--- a/Include/internal/pycore_code.h
+++ b/Include/internal/pycore_code.h
@@ -275,7 +275,7 @@ int _Py_Specialize_LoadMethod(PyObject *owner, _Py_CODEUNIT *instr, PyObject *na
 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 *kwnames, PyObject *builtins,
+     PyObject *kwnames, SpecializedCacheEntry *cache, PyObject *builtins,
     PyObject **stack_pointer, InterpreterFrame *frame, PyObject *names);
 void _Py_Specialize_BinaryOp(PyObject *lhs, PyObject *rhs, _Py_CODEUNIT *instr,
                              SpecializedCacheEntry *cache);
diff --git a/Include/opcode.h b/Include/opcode.h
index c29d23c780a7ab..646c4baae010ae 100644
--- a/Include/opcode.h
+++ b/Include/opcode.h
@@ -134,42 +134,47 @@ extern "C" {
 #define STORE_SUBSCR_ADAPTIVE            26
 #define STORE_SUBSCR_LIST_INT            27
 #define STORE_SUBSCR_DICT                28
-#define CALL_NO_KW_ADAPTIVE              29
-#define CALL_NO_KW_BUILTIN_O             34
-#define CALL_NO_KW_BUILTIN_FAST          36
-#define CALL_NO_KW_LEN                   37
-#define CALL_NO_KW_ISINSTANCE            38
-#define CALL_NO_KW_PY_SIMPLE             39
-#define CALL_NO_KW_LIST_APPEND           40
-#define CALL_NO_KW_METHOD_DESCRIPTOR_O   41
-#define CALL_NO_KW_TYPE_1                42
-#define CALL_NO_KW_BUILTIN_CLASS_1       43
-#define CALL_NO_KW_METHOD_DESCRIPTOR_FAST  44
-#define JUMP_ABSOLUTE_QUICK              45
-#define LOAD_ATTR_ADAPTIVE               46
-#define LOAD_ATTR_INSTANCE_VALUE         47
-#define LOAD_ATTR_WITH_HINT              48
-#define LOAD_ATTR_SLOT                   55
-#define LOAD_ATTR_MODULE                 56
-#define LOAD_GLOBAL_ADAPTIVE             57
-#define LOAD_GLOBAL_MODULE               58
-#define LOAD_GLOBAL_BUILTIN              59
-#define LOAD_METHOD_ADAPTIVE             62
-#define LOAD_METHOD_CACHED               63
-#define LOAD_METHOD_CLASS                64
-#define LOAD_METHOD_MODULE               65
-#define LOAD_METHOD_NO_DICT              66
-#define STORE_ATTR_ADAPTIVE              67
-#define STORE_ATTR_INSTANCE_VALUE        72
-#define STORE_ATTR_SLOT                  76
-#define STORE_ATTR_WITH_HINT             77
-#define LOAD_FAST__LOAD_FAST             78
-#define STORE_FAST__LOAD_FAST            79
-#define LOAD_FAST__LOAD_CONST            80
-#define LOAD_CONST__LOAD_FAST            81
-#define STORE_FAST__STORE_FAST          131
-#define CALL_NO_KW_SUPER_0__LOAD_METHOD_CACHED 140
-#define CALL_NO_KW_SUPER_2__LOAD_METHOD_CACHED 141
+#define CALL_ADAPTIVE                    29
+#define CALL_BUILTIN_CLASS               34
+#define CALL_NO_KW_BUILTIN_O             36
+#define CALL_NO_KW_BUILTIN_FAST          37
+#define CALL_BUILTIN_FAST_WITH_KEYWORDS  38
+#define CALL_NO_KW_LEN                   39
+#define CALL_NO_KW_ISINSTANCE            40
+#define CALL_PY_EXACT_ARGS               41
+#define CALL_PY_WITH_DEFAULTS            42
+#define CALL_NO_KW_LIST_APPEND           43
+#define CALL_NO_KW_METHOD_DESCRIPTOR_O   44
+#define CALL_NO_KW_METHOD_DESCRIPTOR_NOARGS  45
+#define CALL_NO_KW_STR_1                 46
+#define CALL_NO_KW_TUPLE_1               47
+#define CALL_NO_KW_TYPE_1                48
+#define CALL_NO_KW_METHOD_DESCRIPTOR_FAST  55
+#define JUMP_ABSOLUTE_QUICK              56
+#define LOAD_ATTR_ADAPTIVE               57
+#define LOAD_ATTR_INSTANCE_VALUE         58
+#define LOAD_ATTR_WITH_HINT              59
+#define LOAD_ATTR_SLOT                   62
+#define LOAD_ATTR_MODULE                 63
+#define LOAD_GLOBAL_ADAPTIVE             64
+#define LOAD_GLOBAL_MODULE               65
+#define LOAD_GLOBAL_BUILTIN              66
+#define LOAD_METHOD_ADAPTIVE             67
+#define LOAD_METHOD_CACHED               72
+#define LOAD_METHOD_CLASS                76
+#define LOAD_METHOD_MODULE               77
+#define LOAD_METHOD_NO_DICT              78
+#define STORE_ATTR_ADAPTIVE              79
+#define STORE_ATTR_INSTANCE_VALUE        80
+#define STORE_ATTR_SLOT                  81
+#define STORE_ATTR_WITH_HINT            131
+#define LOAD_FAST__LOAD_FAST            140
+#define STORE_FAST__LOAD_FAST           141
+#define LOAD_FAST__LOAD_CONST           143
+#define LOAD_CONST__LOAD_FAST           150
+#define STORE_FAST__STORE_FAST          153
+#define CALL_NO_KW_SUPER_0__LOAD_METHOD_CACHED 154
+#define CALL_NO_KW_SUPER_2__LOAD_METHOD_CACHED 158
 #define DO_TRACING                      255
 #ifdef NEED_OPCODE_JUMP_TABLES
 static uint32_t _PyOpcode_RelativeJump[8] = {
diff --git a/Programs/test_frozenmain.h b/Programs/test_frozenmain.h
index 00f36ff9bd70fa..4052fc6c92632f 100644
--- a/Programs/test_frozenmain.h
+++ b/Programs/test_frozenmain.h
@@ -1,36 +1,35 @@
 // Auto-generated by Programs/freeze_test_frozenmain.py
 unsigned char M_test_frozenmain[] = {
     227,0,0,0,0,0,0,0,0,0,0,0,0,7,0,0,
-    0,0,0,0,0,115,96,0,0,0,151,0,100,0,100,1,
+    0,0,0,0,0,115,88,0,0,0,151,0,100,0,100,1,
     108,0,90,0,100,0,100,1,108,1,90,1,101,2,100,2,
-    167,1,171,0,1,0,101,2,100,3,101,0,106,3,167,2,
-    171,0,1,0,101,1,106,4,167,0,171,0,100,4,25,0,
-    90,5,100,5,68,0,93,15,90,6,101,2,100,6,101,6,
-    155,0,100,7,101,5,101,6,25,0,155,0,157,4,167,1,
-    171,0,1,0,113,30,100,1,83,0,41,8,233,0,0,0,
-    0,78,122,18,70,114,111,122,101,110,32,72,101,108,108,111,
-    32,87,111,114,108,100,122,8,115,121,115,46,97,114,103,118,
-    218,6,99,111,110,102,105,103,41,5,90,12,112,114,111,103,
-    114,97,109,95,110,97,109,101,218,10,101,120,101,99,117,116,
-    97,98,108,101,90,15,117,115,101,95,101,110,118,105,114,111,
-    110,109,101,110,116,90,17,99,111,110,102,105,103,117,114,101,
-    95,99,95,115,116,100,105,111,90,14,98,117,102,102,101,114,
-    101,100,95,115,116,100,105,111,122,7,99,111,110,102,105,103,
-    32,122,2,58,32,41,7,218,3,115,121,115,90,17,95,116,
-    101,115,116,105,110,116,101,114,110,97,108,99,97,112,105,218,
-    5,112,114,105,110,116,218,4,97,114,103,118,90,11,103,101,
-    116,95,99,111,110,102,105,103,115,114,2,0,0,0,218,3,
-    107,101,121,169,0,243,0,0,0,0,250,18,116,101,115,116,
-    95,102,114,111,122,101,110,109,97,105,110,46,112,121,218,8,
-    60,109,111,100,117,108,101,62,114,11,0,0,0,1,0,0,
-    0,115,18,0,0,0,2,128,8,3,8,1,10,2,14,1,
-    14,1,8,1,28,7,4,249,115,20,0,0,0,2,128,8,
-    3,8,1,10,2,14,1,14,1,2,7,4,1,2,249,32,
-    7,115,96,0,0,0,0,0,1,11,1,11,1,11,1,11,
-    1,25,1,25,1,25,1,25,1,6,7,27,1,28,1,28,
-    1,28,1,6,7,17,19,22,19,27,1,28,1,28,1,28,
-    10,27,10,39,10,41,10,41,42,50,10,51,1,7,12,2,
-    1,42,1,42,5,8,5,10,11,41,21,24,11,41,11,41,
-    28,34,35,38,28,39,11,41,11,41,5,42,5,42,5,42,
+    169,1,1,0,101,2,100,3,101,0,106,3,169,2,1,0,
+    101,1,106,4,169,0,100,4,25,0,90,5,100,5,68,0,
+    93,14,90,6,101,2,100,6,101,6,155,0,100,7,101,5,
+    101,6,25,0,155,0,157,4,169,1,1,0,113,27,100,1,
+    83,0,41,8,233,0,0,0,0,78,122,18,70,114,111,122,
+    101,110,32,72,101,108,108,111,32,87,111,114,108,100,122,8,
+    115,121,115,46,97,114,103,118,218,6,99,111,110,102,105,103,
+    41,5,90,12,112,114,111,103,114,97,109,95,110,97,109,101,
+    218,10,101,120,101,99,117,116,97,98,108,101,90,15,117,115,
+    101,95,101,110,118,105,114,111,110,109,101,110,116,90,17,99,
+    111,110,102,105,103,117,114,101,95,99,95,115,116,100,105,111,
+    90,14,98,117,102,102,101,114,101,100,95,115,116,100,105,111,
+    122,7,99,111,110,102,105,103,32,122,2,58,32,41,7,218,
+    3,115,121,115,90,17,95,116,101,115,116,105,110,116,101,114,
+    110,97,108,99,97,112,105,218,5,112,114,105,110,116,218,4,
+    97,114,103,118,90,11,103,101,116,95,99,111,110,102,105,103,
+    115,114,2,0,0,0,218,3,107,101,121,169,0,243,0,0,
+    0,0,250,18,116,101,115,116,95,102,114,111,122,101,110,109,
+    97,105,110,46,112,121,218,8,60,109,111,100,117,108,101,62,
+    114,11,0,0,0,1,0,0,0,115,18,0,0,0,2,128,
+    8,3,8,1,8,2,12,1,12,1,8,1,26,7,4,249,
+    115,20,0,0,0,2,128,8,3,8,1,8,2,12,1,12,
+    1,2,7,4,1,2,249,30,7,115,88,0,0,0,0,0,
+    1,11,1,11,1,11,1,11,1,25,1,25,1,25,1,25,
+    1,6,7,27,1,28,1,28,1,6,7,17,19,22,19,27,
+    1,28,1,28,10,27,10,39,10,41,42,50,10,51,1,7,
+    12,2,1,42,1,42,5,8,5,10,11,41,21,24,11,41,
+    11,41,28,34,35,38,28,39,11,41,11,41,5,42,5,42,
     5,42,1,42,1,42,114,9,0,0,0,
 };
diff --git a/Python/ceval.c b/Python/ceval.c
index 4110cdbdc76076..bc519e7de80b7c 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -4634,7 +4634,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
                 int err = _Py_Specialize_CallNoKw(
                     call_shape.callable, next_instr, nargs,
                     call_shape.kwnames, cache, BUILTINS(),
-                    tack_pointer, frame, names);
+                    stack_pointer, frame, names);
                 if (err < 0) {
                     goto error;
                 }
@@ -5087,18 +5087,18 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
             PyObject *meth;
             PyObject *super_callable = TOP();
 
-            DEOPT_IF(_PyType_CAST(super_callable) != &PySuper_Type, CALL_NO_KW);
+            DEOPT_IF(_PyType_CAST(super_callable) != &PySuper_Type, CALL);
 
             /* super() - zero argument form */
             if (_PySuper_GetTypeArgs(frame, frame->f_code, &su_type, &su_obj) < 0) {
                 PyErr_Clear();
-                DEOPT_IF(1, CALL_NO_KW);
+                DEOPT_IF(1, CALL);
             }
             assert(su_obj != NULL);
-            DEOPT_IF(lm_adaptive->version != Py_TYPE(su_obj)->tp_version_tag, CALL_NO_KW);
-            DEOPT_IF(cache0->version != su_type->tp_version_tag, CALL_NO_KW);
+            DEOPT_IF(lm_adaptive->version != Py_TYPE(su_obj)->tp_version_tag, CALL);
+            DEOPT_IF(cache0->version != su_type->tp_version_tag, CALL);
 
-            STAT_INC(CALL_NO_KW, hit);
+            STAT_INC(CALL, hit);
 
             /* LOAD_METHOD_CACHED */
             meth = cache1->obj;
@@ -5132,13 +5132,13 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
             PyObject *super_callable = THIRD();
             PyObject *meth;
 
-            DEOPT_IF(_PyType_CAST(super_callable) != &PySuper_Type, CALL_NO_KW);
+            DEOPT_IF(_PyType_CAST(super_callable) != &PySuper_Type, CALL);
 
             assert(su_obj != NULL);
-            DEOPT_IF(lm_adaptive->version != Py_TYPE(su_obj)->tp_version_tag, CALL_NO_KW);
-            DEOPT_IF(cache0->version != su_type->tp_version_tag, CALL_NO_KW);
+            DEOPT_IF(lm_adaptive->version != Py_TYPE(su_obj)->tp_version_tag, CALL);
+            DEOPT_IF(cache0->version != su_type->tp_version_tag, CALL);
 
-            STAT_INC(CALL_NO_KW, hit);
+            STAT_INC(CALL, hit);
 
             (void)(POP());
             /* LOAD_METHOD_CACHED */
diff --git a/Python/opcode_targets.h b/Python/opcode_targets.h
index c0076ef0dfa53e..a8a7649280e64c 100644
--- a/Python/opcode_targets.h
+++ b/Python/opcode_targets.h
@@ -139,8 +139,6 @@ static void *opcode_targets[256] = {
     &&TARGET_LOAD_DEREF,
     &&TARGET_STORE_DEREF,
     &&TARGET_DELETE_DEREF,
-    &&TARGET_CALL_NO_KW_SUPER_0__LOAD_METHOD_CACHED,
-    &&TARGET_CALL_NO_KW_SUPER_2__LOAD_METHOD_CACHED,
     &&TARGET_LOAD_FAST__LOAD_FAST,
     &&TARGET_STORE_FAST__LOAD_FAST,
     &&TARGET_CALL_FUNCTION_EX,
@@ -155,11 +153,11 @@ static void *opcode_targets[256] = {
     &&TARGET_RESUME,
     &&TARGET_MATCH_CLASS,
     &&TARGET_STORE_FAST__STORE_FAST,
-    &&_unknown_opcode,
+    &&TARGET_CALL_NO_KW_SUPER_0__LOAD_METHOD_CACHED,
     &&TARGET_FORMAT_VALUE,
     &&TARGET_BUILD_CONST_KEY_MAP,
     &&TARGET_BUILD_STRING,
-    &&_unknown_opcode,
+    &&TARGET_CALL_NO_KW_SUPER_2__LOAD_METHOD_CACHED,
     &&_unknown_opcode,
     &&TARGET_LOAD_METHOD,
     &&_unknown_opcode,
diff --git a/Python/specialize.c b/Python/specialize.c
index ebc6493b10f438..d67608338e4ce9 100644
--- a/Python/specialize.c
+++ b/Python/specialize.c
@@ -1385,12 +1385,12 @@ specialize_class_call(
     }
 
     _Py_CODEUNIT next_instr = instr[1];
-    /* Adaptive super instruction of CALL_NO_KW andLOAD_METHOD_ADAPTIVE. */
+    /* Adaptive super instruction of CALL and LOAD_METHOD_ADAPTIVE. */
     if (tp == &PySuper_Type &&
         _Py_OPCODE(next_instr) == LOAD_METHOD_ADAPTIVE &&
         (nargs == 0 || nargs == 2)) {
         /* Use load_method cache entries too. */
-        _PyAdaptiveEntry *lm_adaptive = &cache[-cache_requirements[CALL_NO_KW]].adaptive;
+        _PyAdaptiveEntry *lm_adaptive = &cache[-cache_requirements[CALL]].adaptive;
         _PyObjectCache *cache1 = &cache[-1].obj;
         PyObject *su_obj;
         PyTypeObject *su_type;

From ac19aa1c1955f9c639904a871c9a24e3f769b351 Mon Sep 17 00:00:00 2001
From: Fidget-Spinner <28750310+Fidget-Spinner@users.noreply.github.com>
Date: Sat, 29 Jan 2022 01:46:57 +0800
Subject: [PATCH 5/9] Update test_frozenmain.h

---
 Programs/test_frozenmain.h | 59 +++++++++++++++++++-------------------
 1 file changed, 30 insertions(+), 29 deletions(-)

diff --git a/Programs/test_frozenmain.h b/Programs/test_frozenmain.h
index 4052fc6c92632f..00f36ff9bd70fa 100644
--- a/Programs/test_frozenmain.h
+++ b/Programs/test_frozenmain.h
@@ -1,35 +1,36 @@
 // Auto-generated by Programs/freeze_test_frozenmain.py
 unsigned char M_test_frozenmain[] = {
     227,0,0,0,0,0,0,0,0,0,0,0,0,7,0,0,
-    0,0,0,0,0,115,88,0,0,0,151,0,100,0,100,1,
+    0,0,0,0,0,115,96,0,0,0,151,0,100,0,100,1,
     108,0,90,0,100,0,100,1,108,1,90,1,101,2,100,2,
-    169,1,1,0,101,2,100,3,101,0,106,3,169,2,1,0,
-    101,1,106,4,169,0,100,4,25,0,90,5,100,5,68,0,
-    93,14,90,6,101,2,100,6,101,6,155,0,100,7,101,5,
-    101,6,25,0,155,0,157,4,169,1,1,0,113,27,100,1,
-    83,0,41,8,233,0,0,0,0,78,122,18,70,114,111,122,
-    101,110,32,72,101,108,108,111,32,87,111,114,108,100,122,8,
-    115,121,115,46,97,114,103,118,218,6,99,111,110,102,105,103,
-    41,5,90,12,112,114,111,103,114,97,109,95,110,97,109,101,
-    218,10,101,120,101,99,117,116,97,98,108,101,90,15,117,115,
-    101,95,101,110,118,105,114,111,110,109,101,110,116,90,17,99,
-    111,110,102,105,103,117,114,101,95,99,95,115,116,100,105,111,
-    90,14,98,117,102,102,101,114,101,100,95,115,116,100,105,111,
-    122,7,99,111,110,102,105,103,32,122,2,58,32,41,7,218,
-    3,115,121,115,90,17,95,116,101,115,116,105,110,116,101,114,
-    110,97,108,99,97,112,105,218,5,112,114,105,110,116,218,4,
-    97,114,103,118,90,11,103,101,116,95,99,111,110,102,105,103,
-    115,114,2,0,0,0,218,3,107,101,121,169,0,243,0,0,
-    0,0,250,18,116,101,115,116,95,102,114,111,122,101,110,109,
-    97,105,110,46,112,121,218,8,60,109,111,100,117,108,101,62,
-    114,11,0,0,0,1,0,0,0,115,18,0,0,0,2,128,
-    8,3,8,1,8,2,12,1,12,1,8,1,26,7,4,249,
-    115,20,0,0,0,2,128,8,3,8,1,8,2,12,1,12,
-    1,2,7,4,1,2,249,30,7,115,88,0,0,0,0,0,
-    1,11,1,11,1,11,1,11,1,25,1,25,1,25,1,25,
-    1,6,7,27,1,28,1,28,1,6,7,17,19,22,19,27,
-    1,28,1,28,10,27,10,39,10,41,42,50,10,51,1,7,
-    12,2,1,42,1,42,5,8,5,10,11,41,21,24,11,41,
-    11,41,28,34,35,38,28,39,11,41,11,41,5,42,5,42,
+    167,1,171,0,1,0,101,2,100,3,101,0,106,3,167,2,
+    171,0,1,0,101,1,106,4,167,0,171,0,100,4,25,0,
+    90,5,100,5,68,0,93,15,90,6,101,2,100,6,101,6,
+    155,0,100,7,101,5,101,6,25,0,155,0,157,4,167,1,
+    171,0,1,0,113,30,100,1,83,0,41,8,233,0,0,0,
+    0,78,122,18,70,114,111,122,101,110,32,72,101,108,108,111,
+    32,87,111,114,108,100,122,8,115,121,115,46,97,114,103,118,
+    218,6,99,111,110,102,105,103,41,5,90,12,112,114,111,103,
+    114,97,109,95,110,97,109,101,218,10,101,120,101,99,117,116,
+    97,98,108,101,90,15,117,115,101,95,101,110,118,105,114,111,
+    110,109,101,110,116,90,17,99,111,110,102,105,103,117,114,101,
+    95,99,95,115,116,100,105,111,90,14,98,117,102,102,101,114,
+    101,100,95,115,116,100,105,111,122,7,99,111,110,102,105,103,
+    32,122,2,58,32,41,7,218,3,115,121,115,90,17,95,116,
+    101,115,116,105,110,116,101,114,110,97,108,99,97,112,105,218,
+    5,112,114,105,110,116,218,4,97,114,103,118,90,11,103,101,
+    116,95,99,111,110,102,105,103,115,114,2,0,0,0,218,3,
+    107,101,121,169,0,243,0,0,0,0,250,18,116,101,115,116,
+    95,102,114,111,122,101,110,109,97,105,110,46,112,121,218,8,
+    60,109,111,100,117,108,101,62,114,11,0,0,0,1,0,0,
+    0,115,18,0,0,0,2,128,8,3,8,1,10,2,14,1,
+    14,1,8,1,28,7,4,249,115,20,0,0,0,2,128,8,
+    3,8,1,10,2,14,1,14,1,2,7,4,1,2,249,32,
+    7,115,96,0,0,0,0,0,1,11,1,11,1,11,1,11,
+    1,25,1,25,1,25,1,25,1,6,7,27,1,28,1,28,
+    1,28,1,6,7,17,19,22,19,27,1,28,1,28,1,28,
+    10,27,10,39,10,41,10,41,42,50,10,51,1,7,12,2,
+    1,42,1,42,5,8,5,10,11,41,21,24,11,41,11,41,
+    28,34,35,38,28,39,11,41,11,41,5,42,5,42,5,42,
     5,42,1,42,1,42,114,9,0,0,0,
 };

From 1f9bb6cfb98be80c2841431557ad139413b2a0ac Mon Sep 17 00:00:00 2001
From: Fidget-Spinner <28750310+Fidget-Spinner@users.noreply.github.com>
Date: Sat, 29 Jan 2022 02:07:43 +0800
Subject: [PATCH 6/9] fix formatting, correct specialization

---
 Python/ceval.c      |  4 --
 Python/specialize.c | 89 +++++++++++++++++++++++----------------------
 2 files changed, 46 insertions(+), 47 deletions(-)

diff --git a/Python/ceval.c b/Python/ceval.c
index bc519e7de80b7c..2067f06fa4e4d7 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -5088,7 +5088,6 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
             PyObject *super_callable = TOP();
 
             DEOPT_IF(_PyType_CAST(super_callable) != &PySuper_Type, CALL);
-
             /* super() - zero argument form */
             if (_PySuper_GetTypeArgs(frame, frame->f_code, &su_type, &su_obj) < 0) {
                 PyErr_Clear();
@@ -5097,7 +5096,6 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
             assert(su_obj != NULL);
             DEOPT_IF(lm_adaptive->version != Py_TYPE(su_obj)->tp_version_tag, CALL);
             DEOPT_IF(cache0->version != su_type->tp_version_tag, CALL);
-
             STAT_INC(CALL, hit);
 
             /* LOAD_METHOD_CACHED */
@@ -5133,11 +5131,9 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
             PyObject *meth;
 
             DEOPT_IF(_PyType_CAST(super_callable) != &PySuper_Type, CALL);
-
             assert(su_obj != NULL);
             DEOPT_IF(lm_adaptive->version != Py_TYPE(su_obj)->tp_version_tag, CALL);
             DEOPT_IF(cache0->version != su_type->tp_version_tag, CALL);
-
             STAT_INC(CALL, hit);
 
             (void)(POP());
diff --git a/Python/specialize.c b/Python/specialize.c
index d67608338e4ce9..c655ddce2a9170 100644
--- a/Python/specialize.c
+++ b/Python/specialize.c
@@ -1375,6 +1375,52 @@ specialize_class_call(
                 return 0;
             }
         }
+        /* Adaptive super instruction of CALL and LOAD_METHOD_ADAPTIVE. */
+        if (tp == &PySuper_Type &&
+            _Py_OPCODE(instr[1]) == LOAD_METHOD_ADAPTIVE &&
+            _Py_OPCODE(instr[-1]) == PRECALL_FUNCTION &&
+            (nargs == 0 || nargs == 2)) {
+            /* Use load_method cache entries too. */
+            _PyAdaptiveEntry *lm_adaptive = &cache[-cache_requirements[CALL]].adaptive;
+            _PyObjectCache *cache1 = &cache[-1].obj;
+            PyObject *su_obj;
+            PyTypeObject *su_type;
+            PyObject *meth;
+            int meth_found;
+            PyObject *name = PyTuple_GET_ITEM(names, lm_adaptive->original_oparg);
+
+            /* Note (KJ): the following operations must not affect tp_version_tag. */
+            /* super() zero arg form. */
+            if (nargs == 0) {
+                if (_PySuper_GetTypeArgs(frame, frame->f_code, &su_type, &su_obj) < 0) {
+                    PyErr_Clear();
+                    SPECIALIZATION_FAIL(CALL, SPEC_FAIL_NOT_DESCRIPTOR);
+                    return -1;
+                }
+            }
+            /* super(su_type, su_obj) two arg form. */
+            else if (nargs == 2) {
+                su_type = _PyType_CAST(stack_pointer[-2]);
+                su_obj = stack_pointer[-1];
+            }
+            meth = _PySuper_Lookup(su_type, su_obj, name, &meth_found);
+            if (meth == NULL) {
+                assert(PyErr_Occurred());
+                PyErr_Clear();
+                SPECIALIZATION_FAIL(CALL, SPEC_FAIL_OTHER);
+                return -1;
+            }
+            if (!meth_found) {
+                SPECIALIZATION_FAIL(CALL, SPEC_FAIL_NOT_DESCRIPTOR);
+                return -1;
+            }
+            cache->adaptive.version = su_type->tp_version_tag;
+            cache1->obj = meth;
+            lm_adaptive->version = Py_TYPE(su_obj)->tp_version_tag;
+            *instr = _Py_MAKECODEUNIT(nargs == 0 ? CALL_NO_KW_SUPER_0__LOAD_METHOD_CACHED
+                : CALL_NO_KW_SUPER_2__LOAD_METHOD_CACHED, _Py_OPARG(*instr));
+            return 0;
+        }
         if (tp->tp_vectorcall != NULL) {
             *instr = _Py_MAKECODEUNIT(CALL_BUILTIN_CLASS, _Py_OPARG(*instr));
             return 0;
@@ -1384,49 +1430,6 @@ specialize_class_call(
         return -1;
     }
 
-    _Py_CODEUNIT next_instr = instr[1];
-    /* Adaptive super instruction of CALL and LOAD_METHOD_ADAPTIVE. */
-    if (tp == &PySuper_Type &&
-        _Py_OPCODE(next_instr) == LOAD_METHOD_ADAPTIVE &&
-        (nargs == 0 || nargs == 2)) {
-        /* Use load_method cache entries too. */
-        _PyAdaptiveEntry *lm_adaptive = &cache[-cache_requirements[CALL]].adaptive;
-        _PyObjectCache *cache1 = &cache[-1].obj;
-        PyObject *su_obj;
-        PyTypeObject *su_type;
-        PyObject *meth;
-        int meth_found;
-        PyObject *name = PyTuple_GET_ITEM(names, lm_adaptive->original_oparg);
-
-        /* Note (KJ): the following operations must not affect tp_version_tag. */
-        /* super() zero arg form. */
-        if (nargs == 0) {
-            if (_PySuper_GetTypeArgs(frame, frame->f_code, &su_type, &su_obj) < 0) {
-                PyErr_Clear();
-                goto fail;
-            }
-        }
-        /* super(su_type, su_obj) two arg form. */
-        else if (nargs == 2) {
-            su_type = _PyType_CAST(stack_pointer[-2]);
-            su_obj = stack_pointer[-1];
-        }
-        meth = _PySuper_Lookup(su_type, su_obj, name, &meth_found);
-        if (meth == NULL) {
-            assert(PyErr_Occurred());
-            PyErr_Clear();
-            goto fail;
-        }
-        if (!meth_found) {
-            goto fail;
-        }
-        cache->adaptive.version = su_type->tp_version_tag;
-        cache1->obj = meth;
-        lm_adaptive->version = Py_TYPE(su_obj)->tp_version_tag;
-        *instr = _Py_MAKECODEUNIT(nargs == 0 ? CALL_NO_KW_SUPER_0__LOAD_METHOD_CACHED
-            : CALL_NO_KW_SUPER_2__LOAD_METHOD_CACHED, _Py_OPARG(*instr));
-        return 0;
-    }
 fail:
     SPECIALIZATION_FAIL(CALL, SPEC_FAIL_CLASS_MUTABLE);
     return -1;

From f4cd3f9aa3f73b34c2bc3a85431d42b993938e70 Mon Sep 17 00:00:00 2001
From: Fidget-Spinner <28750310+Fidget-Spinner@users.noreply.github.com>
Date: Sat, 29 Jan 2022 02:22:52 +0800
Subject: [PATCH 7/9] work with new call convention

---
 Python/ceval.c      | 6 +++---
 Python/specialize.c | 2 +-
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/Python/ceval.c b/Python/ceval.c
index 2067f06fa4e4d7..31843b005ac8a4 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -5078,7 +5078,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
             _PyAdaptiveEntry *cache0 = &caches[0].adaptive;
             _PyObjectCache *cache1 = &caches[-1].obj;
             _PyAdaptiveEntry *lm_adaptive = &caches[-2].adaptive;
-            int nargs = cache0->original_oparg;
+            int nargs = call_shape.total_args;
             assert(nargs == 0);
 
             /* CALL_NO_KW_SUPER */
@@ -5119,9 +5119,9 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
             _PyAdaptiveEntry *cache0 = &caches[0].adaptive;
             _PyObjectCache *cache1 = &caches[-1].obj;
             _PyAdaptiveEntry *lm_adaptive = &caches[-2].adaptive;
-            int nargs = cache0->original_oparg;
-            int lm_oparg = cache0->index;
+            int nargs = call_shape.total_args;
             assert(nargs == 2);
+            assert(call_shape.kwnames == NULL);
 
             /* CALL_NO_KW_SUPER */
             /* super(type, obj) - two argument form */
diff --git a/Python/specialize.c b/Python/specialize.c
index c655ddce2a9170..4fe6a5a990f62e 100644
--- a/Python/specialize.c
+++ b/Python/specialize.c
@@ -1377,6 +1377,7 @@ specialize_class_call(
         }
         /* Adaptive super instruction of CALL and LOAD_METHOD_ADAPTIVE. */
         if (tp == &PySuper_Type &&
+            kwnames == NULL &&
             _Py_OPCODE(instr[1]) == LOAD_METHOD_ADAPTIVE &&
             _Py_OPCODE(instr[-1]) == PRECALL_FUNCTION &&
             (nargs == 0 || nargs == 2)) {
@@ -1430,7 +1431,6 @@ specialize_class_call(
         return -1;
     }
 
-fail:
     SPECIALIZATION_FAIL(CALL, SPEC_FAIL_CLASS_MUTABLE);
     return -1;
 }

From 19880a919812b51167981fef4e08d6ec64730028 Mon Sep 17 00:00:00 2001
From: Fidget-Spinner <28750310+Fidget-Spinner@users.noreply.github.com>
Date: Sat, 29 Jan 2022 21:27:04 +0800
Subject: [PATCH 8/9] partially address review: fix tests, add asserts, add
 cache modification guards

---
 Objects/typeobject.c |  6 +++++-
 Python/ceval.c       | 12 ++++++++----
 Python/specialize.c  | 11 ++++++++++-
 3 files changed, 23 insertions(+), 6 deletions(-)

diff --git a/Objects/typeobject.c b/Objects/typeobject.c
index cbc1b06adc3bca..8bcc90c4474ff7 100644
--- a/Objects/typeobject.c
+++ b/Objects/typeobject.c
@@ -8929,7 +8929,11 @@ do_super_lookup(superobject *su, PyTypeObject *su_type, PyObject *su_obj,
     Py_DECREF(mro);
 
   skip:
-    assert(su != NULL);
+    /* only happens when using manual _PySuper_Lookup, never happens in super_getattro */
+    if (su == NULL) {
+        PyErr_BadInternalCall();
+        return NULL;
+    }
     return PyObject_GenericGetAttr((PyObject *)su, name);
 }
 
diff --git a/Python/ceval.c b/Python/ceval.c
index 31843b005ac8a4..049f9cda1dc54a 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -1361,6 +1361,7 @@ eval_frame_handle_pending(PyThreadState *tstate)
 
 /* The integer overflow is checked by an assertion below. */
 #define INSTR_OFFSET() ((int)(next_instr - first_instr))
+#define NEXT_INSTR_OFFSET() ((int)(next_instr+1 - first_instr))
 #define NEXTOPARG()  do { \
         _Py_CODEUNIT word = *next_instr; \
         opcode = _Py_OPCODE(word); \
@@ -1486,6 +1487,9 @@ eval_frame_handle_pending(PyThreadState *tstate)
 #define GET_CACHE() \
     _GetSpecializedCacheEntryForInstruction(first_instr, INSTR_OFFSET(), oparg)
 
+# define GET_NEXT_INSTR_CACHE() \
+    _GetSpecializedCacheEntryForInstruction(first_instr, NEXT_INSTR_OFFSET(), \
+    _Py_OPARG(*next_instr))
 
 #define DEOPT_IF(cond, instname) if (cond) { goto instname ## _miss; }
 
@@ -5078,8 +5082,8 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
             _PyAdaptiveEntry *cache0 = &caches[0].adaptive;
             _PyObjectCache *cache1 = &caches[-1].obj;
             _PyAdaptiveEntry *lm_adaptive = &caches[-2].adaptive;
-            int nargs = call_shape.total_args;
-            assert(nargs == 0);
+            assert(lm_adaptive == &GET_NEXT_INSTR_CACHE()[0].adaptive);
+            assert(call_shape.total_args == 0);
 
             /* CALL_NO_KW_SUPER */
             PyObject *su_obj;
@@ -5119,8 +5123,8 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
             _PyAdaptiveEntry *cache0 = &caches[0].adaptive;
             _PyObjectCache *cache1 = &caches[-1].obj;
             _PyAdaptiveEntry *lm_adaptive = &caches[-2].adaptive;
-            int nargs = call_shape.total_args;
-            assert(nargs == 2);
+            assert(lm_adaptive == &GET_NEXT_INSTR_CACHE()[0].adaptive);
+            assert(call_shape.total_args == 2);
             assert(call_shape.kwnames == NULL);
 
             /* CALL_NO_KW_SUPER */
diff --git a/Python/specialize.c b/Python/specialize.c
index 4fe6a5a990f62e..104921de4cdd80 100644
--- a/Python/specialize.c
+++ b/Python/specialize.c
@@ -962,6 +962,12 @@ _Py_Specialize_LoadMethod(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name,
     _PyObjectCache *cache2 = &cache[-2].obj;
 
     PyTypeObject *owner_cls = Py_TYPE(owner);
+    _Py_CODEUNIT prev_instr = _Py_OPCODE(instr[-1]);
+    if (prev_instr == CALL_NO_KW_SUPER_0__LOAD_METHOD_CACHED ||
+        prev_instr == CALL_NO_KW_SUPER_2__LOAD_METHOD_CACHED) {
+        /* Our own cache entries are already being used by superinstructions. */
+        goto fail;
+    }
     if (PyModule_CheckExact(owner)) {
         int err = specialize_module_load_attr(owner, instr, name, cache0, cache1,
             LOAD_METHOD, LOAD_METHOD_MODULE);
@@ -1378,6 +1384,10 @@ specialize_class_call(
         /* Adaptive super instruction of CALL and LOAD_METHOD_ADAPTIVE. */
         if (tp == &PySuper_Type &&
             kwnames == NULL &&
+            /* Important: this also protects us from accidentally overriding a
+               the next specialized instruction's cache. We can only use the
+               subsequent LOAD_METHOD cache if it hasn't specialized yet.
+            */
             _Py_OPCODE(instr[1]) == LOAD_METHOD_ADAPTIVE &&
             _Py_OPCODE(instr[-1]) == PRECALL_FUNCTION &&
             (nargs == 0 || nargs == 2)) {
@@ -1430,7 +1440,6 @@ specialize_class_call(
             SPEC_FAIL_CALL_STR : SPEC_FAIL_CLASS_NO_VECTORCALL);
         return -1;
     }
-
     SPECIALIZATION_FAIL(CALL, SPEC_FAIL_CLASS_MUTABLE);
     return -1;
 }

From 696a0e8c3591ba276c6cc102573010aa228939ed Mon Sep 17 00:00:00 2001
From: Fidget-Spinner <28750310+Fidget-Spinner@users.noreply.github.com>
Date: Sat, 29 Jan 2022 21:50:46 +0800
Subject: [PATCH 9/9] fix refleak

---
 Python/specialize.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/Python/specialize.c b/Python/specialize.c
index 104921de4cdd80..1986a19df6a0d5 100644
--- a/Python/specialize.c
+++ b/Python/specialize.c
@@ -1421,12 +1421,13 @@ specialize_class_call(
                 SPECIALIZATION_FAIL(CALL, SPEC_FAIL_OTHER);
                 return -1;
             }
+            Py_DECREF(meth);
             if (!meth_found) {
                 SPECIALIZATION_FAIL(CALL, SPEC_FAIL_NOT_DESCRIPTOR);
                 return -1;
             }
             cache->adaptive.version = su_type->tp_version_tag;
-            cache1->obj = meth;
+            cache1->obj = meth; /* borrowed */
             lm_adaptive->version = Py_TYPE(su_obj)->tp_version_tag;
             *instr = _Py_MAKECODEUNIT(nargs == 0 ? CALL_NO_KW_SUPER_0__LOAD_METHOD_CACHED
                 : CALL_NO_KW_SUPER_2__LOAD_METHOD_CACHED, _Py_OPARG(*instr));