diff --git a/Include/internal/pycore_code.h b/Include/internal/pycore_code.h
@@ -307,6 +307,7 @@ int _Py_Specialize_StoreAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *nam
int _Py_Specialize_LoadGlobal(PyObject *globals, PyObject *builtins, _Py_CODEUNIT *instr, PyObject *name, SpecializedCacheEntry *cache);
int _Py_Specialize_LoadMethod(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, SpecializedCacheEntry *cache);
int _Py_Specialize_BinarySubscr(PyObject *sub, PyObject *container, _Py_CODEUNIT *instr);

#define PRINT_SPECIALIZATION_STATS 0
#define PRINT_SPECIALIZATION_STATS_DETAILED 0

diff --git a/Include/internal/pycore_long.h b/Include/internal/pycore_long.h
@@ -34,6 +34,8 @@ static inline PyObject* _PyLong_GetZero(void)
static inline PyObject* _PyLong_GetOne(void)
{ return __PyLong_GetSmallInt_internal(1); }

#ifdef __cplusplus
}
#endif

diff --git a/Include/opcode.h b/Include/opcode.h
diff --git a/Lib/opcode.py b/Lib/opcode.py
@@ -220,6 +220,11 @@ def jabs_op(name, op):
del def_op, name_op, jrel_op, jabs_op

_specialized_instructions = [
    "BINARY_SUBSCR_ADAPTIVE",
    "BINARY_SUBSCR_LIST_INT",
    "BINARY_SUBSCR_TUPLE_INT",

diff --git a/Misc/NEWS.d/next/Core and Builtins/2021-08-18-11-14-38.bpo-44945.CO3s77.rst b/Misc/NEWS.d/next/Core and Builtins/2021-08-18-11-14-38.bpo-44945.CO3s77.rst
@@ -0,0 +1,7 @@
@@ -1435,6 +1435,12 @@ eval_frame_handle_pending(PyThreadState *tstate)

#define UPDATE_PREV_INSTR_OPARG(instr, oparg) ((uint8_t*)(instr))[-1] = (oparg)

#define GLOBALS() frame->f_globals
#define BUILTINS() frame->f_builtins
#define LOCALS() frame->f_locals
@@ -1980,28 +1986,120 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
        }

        TARGET(BINARY_ADD): {
            PyObject *right = POP();
            PyObject *left = TOP();
-            PyObject *sum;
-            /* NOTE(vstinner): Please don't try to micro-optimize int+int on
-               CPython using bytecode, it is simply worthless.
-               See http://bugs.python.org/issue21955 and
-               http://bugs.python.org/issue10044 for the discussion. In short,
-               no patch shown any impact on a realistic benchmark, only a minor
-               speedup on microbenchmarks. */
-            if (PyUnicode_CheckExact(left) &&
-                     PyUnicode_CheckExact(right)) {
-                sum = unicode_concatenate(tstate, left, right, frame, next_instr);
-                /* unicode_concatenate consumed the ref to left */
            }
            else {
-                sum = PyNumber_Add(left, right);
-                Py_DECREF(left);
            }
            Py_DECREF(right);
-            SET_TOP(sum);
-            if (sum == NULL)
                goto error;
            DISPATCH();
        }

@@ -4761,6 +4859,7 @@ MISS_WITH_CACHE(STORE_ATTR)
MISS_WITH_CACHE(LOAD_GLOBAL)
MISS_WITH_CACHE(LOAD_METHOD)
MISS_WITH_OPARG_COUNTER(BINARY_SUBSCR)

binary_subscr_dict_error:
        {
-Original file line number
+Diff line change
@@ Expand Up / @@ -34,6 +34,8 @@ static inline PyObject* _PyLong_GetZero(void) @@
     static inline PyObject* _PyLong_GetOne(void)
     { return __PyLong_GetSmallInt_internal(1); }
     #ifdef __cplusplus
     }
     #endif
@@ Expand Down @@