bpo-45829: Specialize BINARY_SUBSCR for __getitem__ implemented in Python. (GH-29592)

3 files changed, 107 insertions, 56 deletions

diff --git a/Python/ceval.c b/Python/ceval.c
index e579ede..2b7b31c 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -2140,21 +2140,21 @@ check_eval_breaker:
         }
 
         TARGET(BINARY_SUBSCR_ADAPTIVE) {
-            if (oparg == 0) {
+            SpecializedCacheEntry *cache = GET_CACHE();
+            if (cache->adaptive.counter == 0) {
                 PyObject *sub = TOP();
                 PyObject *container = SECOND();
                 next_instr--;
-                if (_Py_Specialize_BinarySubscr(container, sub, next_instr) < 0) {
+                if (_Py_Specialize_BinarySubscr(container, sub, next_instr, cache) < 0) {
                     goto error;
                 }
                 DISPATCH();
             }
             else {
                 STAT_INC(BINARY_SUBSCR, deferred);
-                // oparg is the adaptive cache counter
-                UPDATE_PREV_INSTR_OPARG(next_instr, oparg - 1);
-                assert(_Py_OPCODE(next_instr[-1]) == BINARY_SUBSCR_ADAPTIVE);
-                assert(_Py_OPARG(next_instr[-1]) == oparg - 1);
+                cache->adaptive.counter--;
+                assert(cache->adaptive.original_oparg == 0);
+                /* No need to set oparg here; it isn't used by BINARY_SUBSCR */
                 STAT_DEC(BINARY_SUBSCR, unquickened);
                 JUMP_TO_INSTRUCTION(BINARY_SUBSCR);
             }
@@ -2223,6 +2223,37 @@ check_eval_breaker:
             DISPATCH();
         }
 
+        TARGET(BINARY_SUBSCR_GETITEM) {
+            PyObject *sub = TOP();
+            PyObject *container = SECOND();
+            SpecializedCacheEntry *caches = GET_CACHE();
+            _PyAdaptiveEntry *cache0 = &caches[0].adaptive;
+            _PyObjectCache *cache1 = &caches[-1].obj;
+            PyFunctionObject *getitem = (PyFunctionObject *)cache1->obj;
+            DEOPT_IF(Py_TYPE(container)->tp_version_tag != cache0->version, BINARY_SUBSCR);
+            DEOPT_IF(getitem->func_version != cache0->index, BINARY_SUBSCR);
+            PyCodeObject *code = (PyCodeObject *)getitem->func_code;
+            size_t size = code->co_nlocalsplus + code->co_stacksize + FRAME_SPECIALS_SIZE;
+            assert(code->co_argcount == 2);
+            InterpreterFrame *new_frame = _PyThreadState_BumpFramePointer(tstate, size);
+            if (new_frame == NULL) {
+                goto error;
+            }
+            _PyFrame_InitializeSpecials(new_frame, PyFunction_AS_FRAME_CONSTRUCTOR(getitem),
+                                        NULL, code->co_nlocalsplus);
+            STACK_SHRINK(2);
+            new_frame->localsplus[0] = container;
+            new_frame->localsplus[1] = sub;
+            for (int i = 2; i < code->co_nlocalsplus; i++) {
+                new_frame->localsplus[i] = NULL;
+            }
+            _PyFrame_SetStackPointer(frame, stack_pointer);
+            new_frame->previous = frame;
+            frame = cframe.current_frame = new_frame;
+            new_frame->depth = frame->depth + 1;
+            goto start_frame;
+        }
+
         TARGET(LIST_APPEND) {
             PyObject *v = POP();
             PyObject *list = PEEK(oparg);
@@ -4878,29 +4909,13 @@ opname ## _miss: \
         JUMP_TO_INSTRUCTION(opname); \
     }
 
-#define MISS_WITH_OPARG_COUNTER(opname) \
-opname ## _miss: \
-    { \
-        STAT_INC(opname, miss); \
-        uint8_t oparg = _Py_OPARG(next_instr[-1])-1; \
-        UPDATE_PREV_INSTR_OPARG(next_instr, oparg); \
-        assert(_Py_OPARG(next_instr[-1]) == oparg); \
-        if (oparg == 0) /* too many cache misses */ { \
-            oparg = ADAPTIVE_CACHE_BACKOFF; \
-            next_instr[-1] = _Py_MAKECODEUNIT(opname ## _ADAPTIVE, oparg); \
-            STAT_INC(opname, deopt); \
-        } \
-        STAT_DEC(opname, unquickened); \
-        JUMP_TO_INSTRUCTION(opname); \
-    }
-
 MISS_WITH_CACHE(LOAD_ATTR)
 MISS_WITH_CACHE(STORE_ATTR)
 MISS_WITH_CACHE(LOAD_GLOBAL)
 MISS_WITH_CACHE(LOAD_METHOD)
 MISS_WITH_CACHE(CALL_FUNCTION)
 MISS_WITH_CACHE(BINARY_OP)
-MISS_WITH_OPARG_COUNTER(BINARY_SUBSCR)
+MISS_WITH_CACHE(BINARY_SUBSCR)
 
 binary_subscr_dict_error:
         {
diff --git a/Python/opcode_targets.h b/Python/opcode_targets.h
index 07852d1..4703255 100644
--- a/Python/opcode_targets.h
+++ b/Python/opcode_targets.h
@@ -21,22 +21,23 @@ static void *opcode_targets[256] = {
     &&TARGET_BINARY_OP_SUBTRACT_INT,
     &&TARGET_BINARY_OP_SUBTRACT_FLOAT,
     &&TARGET_BINARY_SUBSCR_ADAPTIVE,
+    &&TARGET_BINARY_SUBSCR_GETITEM,
     &&TARGET_BINARY_SUBSCR_LIST_INT,
     &&TARGET_BINARY_SUBSCR_TUPLE_INT,
-    &&TARGET_BINARY_SUBSCR_DICT,
     &&TARGET_BINARY_SUBSCR,
+    &&TARGET_BINARY_SUBSCR_DICT,
     &&TARGET_CALL_FUNCTION_ADAPTIVE,
     &&TARGET_CALL_FUNCTION_BUILTIN_O,
     &&TARGET_CALL_FUNCTION_BUILTIN_FAST,
-    &&TARGET_CALL_FUNCTION_LEN,
     &&TARGET_GET_LEN,
     &&TARGET_MATCH_MAPPING,
     &&TARGET_MATCH_SEQUENCE,
     &&TARGET_MATCH_KEYS,
-    &&TARGET_CALL_FUNCTION_ISINSTANCE,
+    &&TARGET_CALL_FUNCTION_LEN,
     &&TARGET_PUSH_EXC_INFO,
-    &&TARGET_CALL_FUNCTION_PY_SIMPLE,
+    &&TARGET_CALL_FUNCTION_ISINSTANCE,
     &&TARGET_POP_EXCEPT_AND_RERAISE,
+    &&TARGET_CALL_FUNCTION_PY_SIMPLE,
     &&TARGET_JUMP_ABSOLUTE_QUICK,
     &&TARGET_LOAD_ATTR_ADAPTIVE,
     &&TARGET_LOAD_ATTR_INSTANCE_VALUE,
@@ -47,26 +48,25 @@ static void *opcode_targets[256] = {
     &&TARGET_LOAD_GLOBAL_MODULE,
     &&TARGET_LOAD_GLOBAL_BUILTIN,
     &&TARGET_LOAD_METHOD_ADAPTIVE,
-    &&TARGET_LOAD_METHOD_CACHED,
     &&TARGET_WITH_EXCEPT_START,
     &&TARGET_GET_AITER,
     &&TARGET_GET_ANEXT,
     &&TARGET_BEFORE_ASYNC_WITH,
     &&TARGET_BEFORE_WITH,
     &&TARGET_END_ASYNC_FOR,
+    &&TARGET_LOAD_METHOD_CACHED,
     &&TARGET_LOAD_METHOD_CLASS,
     &&TARGET_LOAD_METHOD_MODULE,
     &&TARGET_LOAD_METHOD_NO_DICT,
     &&TARGET_STORE_ATTR_ADAPTIVE,
-    &&TARGET_STORE_ATTR_INSTANCE_VALUE,
     &&TARGET_STORE_SUBSCR,
     &&TARGET_DELETE_SUBSCR,
+    &&TARGET_STORE_ATTR_INSTANCE_VALUE,
     &&TARGET_STORE_ATTR_SLOT,
     &&TARGET_STORE_ATTR_WITH_HINT,
     &&TARGET_LOAD_FAST__LOAD_FAST,
     &&TARGET_STORE_FAST__LOAD_FAST,
     &&TARGET_LOAD_FAST__LOAD_CONST,
-    &&TARGET_LOAD_CONST__LOAD_FAST,
     &&TARGET_GET_ITER,
     &&TARGET_GET_YIELD_FROM_ITER,
     &&TARGET_PRINT_EXPR,
@@ -74,13 +74,13 @@ static void *opcode_targets[256] = {
     &&TARGET_YIELD_FROM,
     &&TARGET_GET_AWAITABLE,
     &&TARGET_LOAD_ASSERTION_ERROR,
+    &&TARGET_LOAD_CONST__LOAD_FAST,
     &&TARGET_STORE_FAST__STORE_FAST,
     &&_unknown_opcode,
     &&_unknown_opcode,
     &&_unknown_opcode,
     &&_unknown_opcode,
     &&_unknown_opcode,
-    &&_unknown_opcode,
     &&TARGET_LIST_TO_TUPLE,
     &&TARGET_RETURN_VALUE,
     &&TARGET_IMPORT_STAR,
diff --git a/Python/specialize.c b/Python/specialize.c
index dd15de7..06b0764 100644
--- a/Python/specialize.c
+++ b/Python/specialize.c
@@ -243,7 +243,7 @@ static uint8_t cache_requirements[256] = {
     [LOAD_ATTR] = 2, /* _PyAdaptiveEntry and _PyAttrCache */
     [LOAD_GLOBAL] = 2, /* _PyAdaptiveEntry and _PyLoadGlobalCache */
     [LOAD_METHOD] = 3, /* _PyAdaptiveEntry, _PyAttrCache and _PyObjectCache */
-    [BINARY_SUBSCR] = 0,
+    [BINARY_SUBSCR] = 2, /* _PyAdaptiveEntry, _PyObjectCache */
     [CALL_FUNCTION] = 2, /* _PyAdaptiveEntry and _PyObjectCache/_PyCallCache */
     [STORE_ATTR] = 2, /* _PyAdaptiveEntry and _PyAttrCache */
     [BINARY_OP] = 1,  // _PyAdaptiveEntry
@@ -1100,7 +1100,7 @@ success:
 
 #if COLLECT_SPECIALIZATION_STATS_DETAILED
 static int
-binary_subscr_faiL_kind(PyTypeObject *container_type, PyObject *sub)
+binary_subscr_fail_kind(PyTypeObject *container_type, PyObject *sub)
 {
     if (container_type == &PyUnicode_Type) {
         if (PyLong_CheckExact(sub)) {
@@ -1138,14 +1138,37 @@ binary_subscr_faiL_kind(PyTypeObject *container_type, PyObject *sub)
 }
 #endif
 
+_Py_IDENTIFIER(__getitem__);
+
+#define SIMPLE_FUNCTION 0
+
+static int
+function_kind(PyCodeObject *code) {
+    int flags = code->co_flags;
+    if (flags & (CO_GENERATOR | CO_COROUTINE | CO_ASYNC_GENERATOR)) {
+        return SPEC_FAIL_GENERATOR;
+    }
+    if ((flags & (CO_VARKEYWORDS | CO_VARARGS)) || code->co_kwonlyargcount) {
+        return SPEC_FAIL_COMPLEX_PARAMETERS;
+    }
+    if ((flags & CO_OPTIMIZED) == 0) {
+        return SPEC_FAIL_CO_NOT_OPTIMIZED;
+    }
+    if (code->co_nfreevars) {
+        return SPEC_FAIL_FREE_VARS;
+    }
+    return SIMPLE_FUNCTION;
+}
+
 int
 _Py_Specialize_BinarySubscr(
-     PyObject *container, PyObject *sub, _Py_CODEUNIT *instr)
+     PyObject *container, PyObject *sub, _Py_CODEUNIT *instr, SpecializedCacheEntry *cache)
 {
+    _PyAdaptiveEntry *cache0 = &cache->adaptive;
     PyTypeObject *container_type = Py_TYPE(container);
     if (container_type == &PyList_Type) {
         if (PyLong_CheckExact(sub)) {
-            *instr = _Py_MAKECODEUNIT(BINARY_SUBSCR_LIST_INT, initial_counter_value());
+            *instr = _Py_MAKECODEUNIT(BINARY_SUBSCR_LIST_INT, _Py_OPARG(*instr));
             goto success;
         }
         SPECIALIZATION_FAIL(BINARY_SUBSCR,
@@ -1154,7 +1177,7 @@ _Py_Specialize_BinarySubscr(
     }
     if (container_type == &PyTuple_Type) {
         if (PyLong_CheckExact(sub)) {
-            *instr = _Py_MAKECODEUNIT(BINARY_SUBSCR_TUPLE_INT, initial_counter_value());
+            *instr = _Py_MAKECODEUNIT(BINARY_SUBSCR_TUPLE_INT, _Py_OPARG(*instr));
             goto success;
         }
         SPECIALIZATION_FAIL(BINARY_SUBSCR,
@@ -1162,20 +1185,46 @@ _Py_Specialize_BinarySubscr(
         goto fail;
     }
     if (container_type == &PyDict_Type) {
-        *instr = _Py_MAKECODEUNIT(BINARY_SUBSCR_DICT, initial_counter_value());
+        *instr = _Py_MAKECODEUNIT(BINARY_SUBSCR_DICT, _Py_OPARG(*instr));
+        goto success;
+    }
+    PyTypeObject *cls = Py_TYPE(container);
+    PyObject *descriptor = _PyType_LookupId(cls, &PyId___getitem__);
+    if (descriptor && Py_TYPE(descriptor) == &PyFunction_Type) {
+        PyFunctionObject *func = (PyFunctionObject *)descriptor;
+        PyCodeObject *code = (PyCodeObject *)func->func_code;
+        int kind = function_kind(code);
+        if (kind != SIMPLE_FUNCTION) {
+            SPECIALIZATION_FAIL(BINARY_SUBSCR, kind);
+            goto fail;
+        }
+        if (code->co_argcount != 2) {
+            SPECIALIZATION_FAIL(BINARY_SUBSCR, SPEC_FAIL_WRONG_NUMBER_ARGUMENTS);
+            goto fail;
+        }
+        assert(cls->tp_version_tag != 0);
+        cache0->version = cls->tp_version_tag;
+        int version = _PyFunction_GetVersionForCurrentState(func);
+        if (version == 0) {
+            SPECIALIZATION_FAIL(BINARY_SUBSCR, SPEC_FAIL_OUT_OF_VERSIONS);
+            goto fail;
+        }
+        cache0->index = version;
+        cache[-1].obj.obj = descriptor;
+        *instr = _Py_MAKECODEUNIT(BINARY_SUBSCR_GETITEM, _Py_OPARG(*instr));
         goto success;
     }
     SPECIALIZATION_FAIL(BINARY_SUBSCR,
-                        binary_subscr_faiL_kind(container_type, sub));
-    goto fail;
+                        binary_subscr_fail_kind(container_type, sub));
 fail:
     STAT_INC(BINARY_SUBSCR, specialization_failure);
     assert(!PyErr_Occurred());
-    *instr = _Py_MAKECODEUNIT(_Py_OPCODE(*instr), ADAPTIVE_CACHE_BACKOFF);
+    cache_backoff(cache0);
     return 0;
 success:
     STAT_INC(BINARY_SUBSCR, specialization_success);
     assert(!PyErr_Occurred());
+    cache0->counter = initial_counter_value();
     return 0;
 }
 
@@ -1194,23 +1243,10 @@ specialize_py_call(
     int nargs, SpecializedCacheEntry *cache)
 {
     _PyCallCache *cache1 = &cache[-1].call;
-    /* Exclude generator or coroutines for now */
     PyCodeObject *code = (PyCodeObject *)func->func_code;
-    int flags = code->co_flags;
-    if (flags & (CO_GENERATOR | CO_COROUTINE | CO_ASYNC_GENERATOR)) {
-        SPECIALIZATION_FAIL(CALL_FUNCTION, SPEC_FAIL_GENERATOR);
-        return -1;
-    }
-    if ((flags & (CO_VARKEYWORDS | CO_VARARGS)) || code->co_kwonlyargcount) {
-        SPECIALIZATION_FAIL(CALL_FUNCTION, SPEC_FAIL_COMPLEX_PARAMETERS);
-        return -1;
-    }
-    if ((flags & CO_OPTIMIZED) == 0) {
-        SPECIALIZATION_FAIL(CALL_FUNCTION, SPEC_FAIL_CO_NOT_OPTIMIZED);
-        return -1;
-    }
-    if (code->co_nfreevars) {
-        SPECIALIZATION_FAIL(CALL_FUNCTION, SPEC_FAIL_FREE_VARS);
+    int kind = function_kind(code);
+    if (kind != SIMPLE_FUNCTION) {
+        SPECIALIZATION_FAIL(CALL_FUNCTION, kind);
         return -1;
     }
     int argcount = code->co_argcount;