GH-100110: Specialize FOR_ITER for tuples (GH-100109)

* Specialize FOR_ITER for tuples
This commit is contained in:
Ken Jin 2022-12-09 18:27:01 +08:00 committed by GitHub
parent 0448deac70
commit 748c6c0921
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
10 changed files with 122 additions and 67 deletions

View file

@ -125,6 +125,7 @@ const uint8_t _PyOpcode_Deopt[256] = {
[FOR_ITER_GEN] = FOR_ITER,
[FOR_ITER_LIST] = FOR_ITER,
[FOR_ITER_RANGE] = FOR_ITER,
[FOR_ITER_TUPLE] = FOR_ITER,
[GET_AITER] = GET_AITER,
[GET_ANEXT] = GET_ANEXT,
[GET_AWAITABLE] = GET_AWAITABLE,
@ -293,31 +294,31 @@ static const char *const _PyOpcode_OpName[263] = {
[FOR_ITER_LIST] = "FOR_ITER_LIST",
[STORE_SUBSCR] = "STORE_SUBSCR",
[DELETE_SUBSCR] = "DELETE_SUBSCR",
[FOR_ITER_RANGE] = "FOR_ITER_RANGE",
[FOR_ITER_TUPLE] = "FOR_ITER_TUPLE",
[STOPITERATION_ERROR] = "STOPITERATION_ERROR",
[FOR_ITER_RANGE] = "FOR_ITER_RANGE",
[FOR_ITER_GEN] = "FOR_ITER_GEN",
[LOAD_ATTR_CLASS] = "LOAD_ATTR_CLASS",
[LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN] = "LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN",
[LOAD_ATTR_INSTANCE_VALUE] = "LOAD_ATTR_INSTANCE_VALUE",
[GET_ITER] = "GET_ITER",
[GET_YIELD_FROM_ITER] = "GET_YIELD_FROM_ITER",
[PRINT_EXPR] = "PRINT_EXPR",
[LOAD_BUILD_CLASS] = "LOAD_BUILD_CLASS",
[LOAD_ATTR_INSTANCE_VALUE] = "LOAD_ATTR_INSTANCE_VALUE",
[LOAD_ATTR_MODULE] = "LOAD_ATTR_MODULE",
[LOAD_ATTR_PROPERTY] = "LOAD_ATTR_PROPERTY",
[LOAD_ASSERTION_ERROR] = "LOAD_ASSERTION_ERROR",
[RETURN_GENERATOR] = "RETURN_GENERATOR",
[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_DICT] = "LOAD_ATTR_METHOD_WITH_DICT",
[LOAD_ATTR_METHOD_WITH_VALUES] = "LOAD_ATTR_METHOD_WITH_VALUES",
[LIST_TO_TUPLE] = "LIST_TO_TUPLE",
[RETURN_VALUE] = "RETURN_VALUE",
[IMPORT_STAR] = "IMPORT_STAR",
[SETUP_ANNOTATIONS] = "SETUP_ANNOTATIONS",
[LOAD_CONST__LOAD_FAST] = "LOAD_CONST__LOAD_FAST",
[LOAD_ATTR_METHOD_WITH_VALUES] = "LOAD_ATTR_METHOD_WITH_VALUES",
[ASYNC_GEN_WRAP] = "ASYNC_GEN_WRAP",
[PREP_RERAISE_STAR] = "PREP_RERAISE_STAR",
[POP_EXCEPT] = "POP_EXCEPT",
@ -344,7 +345,7 @@ static const char *const _PyOpcode_OpName[263] = {
[JUMP_FORWARD] = "JUMP_FORWARD",
[JUMP_IF_FALSE_OR_POP] = "JUMP_IF_FALSE_OR_POP",
[JUMP_IF_TRUE_OR_POP] = "JUMP_IF_TRUE_OR_POP",
[LOAD_FAST__LOAD_CONST] = "LOAD_FAST__LOAD_CONST",
[LOAD_CONST__LOAD_FAST] = "LOAD_CONST__LOAD_FAST",
[POP_JUMP_IF_FALSE] = "POP_JUMP_IF_FALSE",
[POP_JUMP_IF_TRUE] = "POP_JUMP_IF_TRUE",
[LOAD_GLOBAL] = "LOAD_GLOBAL",
@ -352,7 +353,7 @@ static const char *const _PyOpcode_OpName[263] = {
[CONTAINS_OP] = "CONTAINS_OP",
[RERAISE] = "RERAISE",
[COPY] = "COPY",
[LOAD_FAST__LOAD_FAST] = "LOAD_FAST__LOAD_FAST",
[LOAD_FAST__LOAD_CONST] = "LOAD_FAST__LOAD_CONST",
[BINARY_OP] = "BINARY_OP",
[SEND] = "SEND",
[LOAD_FAST] = "LOAD_FAST",
@ -372,9 +373,9 @@ static const char *const _PyOpcode_OpName[263] = {
[STORE_DEREF] = "STORE_DEREF",
[DELETE_DEREF] = "DELETE_DEREF",
[JUMP_BACKWARD] = "JUMP_BACKWARD",
[LOAD_GLOBAL_BUILTIN] = "LOAD_GLOBAL_BUILTIN",
[LOAD_FAST__LOAD_FAST] = "LOAD_FAST__LOAD_FAST",
[CALL_FUNCTION_EX] = "CALL_FUNCTION_EX",
[LOAD_GLOBAL_MODULE] = "LOAD_GLOBAL_MODULE",
[LOAD_GLOBAL_BUILTIN] = "LOAD_GLOBAL_BUILTIN",
[EXTENDED_ARG] = "EXTENDED_ARG",
[LIST_APPEND] = "LIST_APPEND",
[SET_ADD] = "SET_ADD",
@ -384,24 +385,24 @@ static const char *const _PyOpcode_OpName[263] = {
[YIELD_VALUE] = "YIELD_VALUE",
[RESUME] = "RESUME",
[MATCH_CLASS] = "MATCH_CLASS",
[LOAD_GLOBAL_MODULE] = "LOAD_GLOBAL_MODULE",
[STORE_ATTR_INSTANCE_VALUE] = "STORE_ATTR_INSTANCE_VALUE",
[STORE_ATTR_SLOT] = "STORE_ATTR_SLOT",
[FORMAT_VALUE] = "FORMAT_VALUE",
[BUILD_CONST_KEY_MAP] = "BUILD_CONST_KEY_MAP",
[BUILD_STRING] = "BUILD_STRING",
[STORE_ATTR_SLOT] = "STORE_ATTR_SLOT",
[STORE_ATTR_WITH_HINT] = "STORE_ATTR_WITH_HINT",
[STORE_FAST__LOAD_FAST] = "STORE_FAST__LOAD_FAST",
[STORE_FAST__STORE_FAST] = "STORE_FAST__STORE_FAST",
[STORE_SUBSCR_DICT] = "STORE_SUBSCR_DICT",
[LIST_EXTEND] = "LIST_EXTEND",
[SET_UPDATE] = "SET_UPDATE",
[DICT_MERGE] = "DICT_MERGE",
[DICT_UPDATE] = "DICT_UPDATE",
[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",
[UNPACK_SEQUENCE_TWO_TUPLE] = "UNPACK_SEQUENCE_TWO_TUPLE",
[170] = "<170>",
[CALL] = "CALL",
[KW_NAMES] = "KW_NAMES",
[173] = "<173>",
@ -498,7 +499,6 @@ static const char *const _PyOpcode_OpName[263] = {
#endif
#define EXTRA_CASES \
case 170: \
case 173: \
case 174: \
case 175: \

View file

@ -67,6 +67,13 @@ struct _Py_tuple_state {
extern PyObject *_PyTuple_FromArray(PyObject *const *, Py_ssize_t);
extern PyObject *_PyTuple_FromArraySteal(PyObject *const *, Py_ssize_t);
typedef struct {
PyObject_HEAD
Py_ssize_t it_index;
PyTupleObject *it_seq; /* Set to NULL when iterator is exhausted */
} _PyTupleIterObject;
#ifdef __cplusplus
}
#endif

57
Include/opcode.h generated
View file

@ -162,34 +162,35 @@ extern "C" {
#define COMPARE_OP_INT_JUMP 57
#define COMPARE_OP_STR_JUMP 58
#define FOR_ITER_LIST 59
#define FOR_ITER_RANGE 62
#define FOR_ITER_GEN 64
#define LOAD_ATTR_CLASS 65
#define LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN 66
#define LOAD_ATTR_INSTANCE_VALUE 67
#define LOAD_ATTR_MODULE 72
#define LOAD_ATTR_PROPERTY 73
#define LOAD_ATTR_SLOT 76
#define LOAD_ATTR_WITH_HINT 77
#define LOAD_ATTR_METHOD_LAZY_DICT 78
#define LOAD_ATTR_METHOD_NO_DICT 79
#define LOAD_ATTR_METHOD_WITH_DICT 80
#define LOAD_ATTR_METHOD_WITH_VALUES 81
#define LOAD_CONST__LOAD_FAST 86
#define LOAD_FAST__LOAD_CONST 113
#define LOAD_FAST__LOAD_FAST 121
#define LOAD_GLOBAL_BUILTIN 141
#define LOAD_GLOBAL_MODULE 143
#define STORE_ATTR_INSTANCE_VALUE 153
#define STORE_ATTR_SLOT 154
#define STORE_ATTR_WITH_HINT 158
#define STORE_FAST__LOAD_FAST 159
#define STORE_FAST__STORE_FAST 160
#define STORE_SUBSCR_DICT 161
#define STORE_SUBSCR_LIST_INT 166
#define UNPACK_SEQUENCE_LIST 167
#define UNPACK_SEQUENCE_TUPLE 168
#define UNPACK_SEQUENCE_TWO_TUPLE 169
#define FOR_ITER_TUPLE 62
#define FOR_ITER_RANGE 64
#define FOR_ITER_GEN 65
#define LOAD_ATTR_CLASS 66
#define LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN 67
#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_DICT 81
#define LOAD_ATTR_METHOD_WITH_VALUES 86
#define LOAD_CONST__LOAD_FAST 113
#define LOAD_FAST__LOAD_CONST 121
#define LOAD_FAST__LOAD_FAST 141
#define LOAD_GLOBAL_BUILTIN 143
#define LOAD_GLOBAL_MODULE 153
#define STORE_ATTR_INSTANCE_VALUE 154
#define STORE_ATTR_SLOT 158
#define STORE_ATTR_WITH_HINT 159
#define STORE_FAST__LOAD_FAST 160
#define STORE_FAST__STORE_FAST 161
#define STORE_SUBSCR_DICT 166
#define STORE_SUBSCR_LIST_INT 167
#define UNPACK_SEQUENCE_LIST 168
#define UNPACK_SEQUENCE_TUPLE 169
#define UNPACK_SEQUENCE_TWO_TUPLE 170
#define DO_TRACING 255
#define HAS_ARG(op) ((((op) >= HAVE_ARGUMENT) && (!IS_PSEUDO_OPCODE(op)))\

View file

@ -320,6 +320,7 @@ def pseudo_op(name, op, real_ops):
],
"FOR_ITER": [
"FOR_ITER_LIST",
"FOR_ITER_TUPLE",
"FOR_ITER_RANGE",
"FOR_ITER_GEN",
],

View file

@ -0,0 +1 @@
Specialize ``FOR_ITER`` for tuples.

View file

@ -995,14 +995,9 @@ _PyTuple_ClearFreeList(PyInterpreterState *interp)
/*********************** Tuple Iterator **************************/
typedef struct {
PyObject_HEAD
Py_ssize_t it_index;
PyTupleObject *it_seq; /* Set to NULL when iterator is exhausted */
} tupleiterobject;
static void
tupleiter_dealloc(tupleiterobject *it)
tupleiter_dealloc(_PyTupleIterObject *it)
{
_PyObject_GC_UNTRACK(it);
Py_XDECREF(it->it_seq);
@ -1010,14 +1005,14 @@ tupleiter_dealloc(tupleiterobject *it)
}
static int
tupleiter_traverse(tupleiterobject *it, visitproc visit, void *arg)
tupleiter_traverse(_PyTupleIterObject *it, visitproc visit, void *arg)
{
Py_VISIT(it->it_seq);
return 0;
}
static PyObject *
tupleiter_next(tupleiterobject *it)
tupleiter_next(_PyTupleIterObject *it)
{
PyTupleObject *seq;
PyObject *item;
@ -1040,7 +1035,7 @@ tupleiter_next(tupleiterobject *it)
}
static PyObject *
tupleiter_len(tupleiterobject *it, PyObject *Py_UNUSED(ignored))
tupleiter_len(_PyTupleIterObject *it, PyObject *Py_UNUSED(ignored))
{
Py_ssize_t len = 0;
if (it->it_seq)
@ -1051,7 +1046,7 @@ tupleiter_len(tupleiterobject *it, PyObject *Py_UNUSED(ignored))
PyDoc_STRVAR(length_hint_doc, "Private method returning an estimate of len(list(it)).");
static PyObject *
tupleiter_reduce(tupleiterobject *it, PyObject *Py_UNUSED(ignored))
tupleiter_reduce(_PyTupleIterObject *it, PyObject *Py_UNUSED(ignored))
{
if (it->it_seq)
return Py_BuildValue("N(O)n", _PyEval_GetBuiltin(&_Py_ID(iter)),
@ -1061,7 +1056,7 @@ tupleiter_reduce(tupleiterobject *it, PyObject *Py_UNUSED(ignored))
}
static PyObject *
tupleiter_setstate(tupleiterobject *it, PyObject *state)
tupleiter_setstate(_PyTupleIterObject *it, PyObject *state)
{
Py_ssize_t index = PyLong_AsSsize_t(state);
if (index == -1 && PyErr_Occurred())
@ -1089,7 +1084,7 @@ static PyMethodDef tupleiter_methods[] = {
PyTypeObject PyTupleIter_Type = {
PyVarObject_HEAD_INIT(&PyType_Type, 0)
"tuple_iterator", /* tp_name */
sizeof(tupleiterobject), /* tp_basicsize */
sizeof(_PyTupleIterObject), /* tp_basicsize */
0, /* tp_itemsize */
/* methods */
(destructor)tupleiter_dealloc, /* tp_dealloc */
@ -1122,13 +1117,13 @@ PyTypeObject PyTupleIter_Type = {
static PyObject *
tuple_iter(PyObject *seq)
{
tupleiterobject *it;
_PyTupleIterObject *it;
if (!PyTuple_Check(seq)) {
PyErr_BadInternalCall();
return NULL;
}
it = PyObject_GC_New(tupleiterobject, &PyTupleIter_Type);
it = PyObject_GC_New(_PyTupleIterObject, &PyTupleIter_Type);
if (it == NULL)
return NULL;
it->it_index = 0;

View file

@ -2523,6 +2523,29 @@ dummy_func(
end_for_iter_list:
}
// stack effect: ( -- __0)
inst(FOR_ITER_TUPLE) {
assert(cframe.use_tracing == 0);
_PyTupleIterObject *it = (_PyTupleIterObject *)TOP();
DEOPT_IF(Py_TYPE(it) != &PyTupleIter_Type, FOR_ITER);
STAT_INC(FOR_ITER, hit);
PyTupleObject *seq = it->it_seq;
if (seq) {
if (it->it_index < PyTuple_GET_SIZE(seq)) {
PyObject *next = PyTuple_GET_ITEM(seq, it->it_index++);
PUSH(Py_NewRef(next));
JUMPBY(INLINE_CACHE_ENTRIES_FOR_ITER);
goto end_for_iter_tuple; // End of this instruction
}
it->it_seq = NULL;
Py_DECREF(seq);
}
STACK_SHRINK(1);
Py_DECREF(it);
JUMPBY(INLINE_CACHE_ENTRIES_FOR_ITER + oparg + 1);
end_for_iter_tuple:
}
// stack effect: ( -- __0)
inst(FOR_ITER_RANGE) {
assert(cframe.use_tracing == 0);

View file

@ -2759,6 +2759,29 @@
DISPATCH();
}
TARGET(FOR_ITER_TUPLE) {
assert(cframe.use_tracing == 0);
_PyTupleIterObject *it = (_PyTupleIterObject *)TOP();
DEOPT_IF(Py_TYPE(it) != &PyTupleIter_Type, FOR_ITER);
STAT_INC(FOR_ITER, hit);
PyTupleObject *seq = it->it_seq;
if (seq) {
if (it->it_index < PyTuple_GET_SIZE(seq)) {
PyObject *next = PyTuple_GET_ITEM(seq, it->it_index++);
PUSH(Py_NewRef(next));
JUMPBY(INLINE_CACHE_ENTRIES_FOR_ITER);
goto end_for_iter_tuple; // End of this instruction
}
it->it_seq = NULL;
Py_DECREF(seq);
}
STACK_SHRINK(1);
Py_DECREF(it);
JUMPBY(INLINE_CACHE_ENTRIES_FOR_ITER + oparg + 1);
end_for_iter_tuple:
DISPATCH();
}
TARGET(FOR_ITER_RANGE) {
assert(cframe.use_tracing == 0);
_PyRangeIterObject *r = (_PyRangeIterObject *)TOP();

View file

@ -61,31 +61,31 @@ static void *opcode_targets[256] = {
&&TARGET_FOR_ITER_LIST,
&&TARGET_STORE_SUBSCR,
&&TARGET_DELETE_SUBSCR,
&&TARGET_FOR_ITER_RANGE,
&&TARGET_FOR_ITER_TUPLE,
&&TARGET_STOPITERATION_ERROR,
&&TARGET_FOR_ITER_RANGE,
&&TARGET_FOR_ITER_GEN,
&&TARGET_LOAD_ATTR_CLASS,
&&TARGET_LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN,
&&TARGET_LOAD_ATTR_INSTANCE_VALUE,
&&TARGET_GET_ITER,
&&TARGET_GET_YIELD_FROM_ITER,
&&TARGET_PRINT_EXPR,
&&TARGET_LOAD_BUILD_CLASS,
&&TARGET_LOAD_ATTR_INSTANCE_VALUE,
&&TARGET_LOAD_ATTR_MODULE,
&&TARGET_LOAD_ATTR_PROPERTY,
&&TARGET_LOAD_ASSERTION_ERROR,
&&TARGET_RETURN_GENERATOR,
&&TARGET_LOAD_ATTR_PROPERTY,
&&TARGET_LOAD_ATTR_SLOT,
&&TARGET_LOAD_ATTR_WITH_HINT,
&&TARGET_LOAD_ATTR_METHOD_LAZY_DICT,
&&TARGET_LOAD_ATTR_METHOD_NO_DICT,
&&TARGET_LOAD_ATTR_METHOD_WITH_DICT,
&&TARGET_LOAD_ATTR_METHOD_WITH_VALUES,
&&TARGET_LIST_TO_TUPLE,
&&TARGET_RETURN_VALUE,
&&TARGET_IMPORT_STAR,
&&TARGET_SETUP_ANNOTATIONS,
&&TARGET_LOAD_CONST__LOAD_FAST,
&&TARGET_LOAD_ATTR_METHOD_WITH_VALUES,
&&TARGET_ASYNC_GEN_WRAP,
&&TARGET_PREP_RERAISE_STAR,
&&TARGET_POP_EXCEPT,
@ -112,7 +112,7 @@ static void *opcode_targets[256] = {
&&TARGET_JUMP_FORWARD,
&&TARGET_JUMP_IF_FALSE_OR_POP,
&&TARGET_JUMP_IF_TRUE_OR_POP,
&&TARGET_LOAD_FAST__LOAD_CONST,
&&TARGET_LOAD_CONST__LOAD_FAST,
&&TARGET_POP_JUMP_IF_FALSE,
&&TARGET_POP_JUMP_IF_TRUE,
&&TARGET_LOAD_GLOBAL,
@ -120,7 +120,7 @@ static void *opcode_targets[256] = {
&&TARGET_CONTAINS_OP,
&&TARGET_RERAISE,
&&TARGET_COPY,
&&TARGET_LOAD_FAST__LOAD_FAST,
&&TARGET_LOAD_FAST__LOAD_CONST,
&&TARGET_BINARY_OP,
&&TARGET_SEND,
&&TARGET_LOAD_FAST,
@ -140,9 +140,9 @@ static void *opcode_targets[256] = {
&&TARGET_STORE_DEREF,
&&TARGET_DELETE_DEREF,
&&TARGET_JUMP_BACKWARD,
&&TARGET_LOAD_GLOBAL_BUILTIN,
&&TARGET_LOAD_FAST__LOAD_FAST,
&&TARGET_CALL_FUNCTION_EX,
&&TARGET_LOAD_GLOBAL_MODULE,
&&TARGET_LOAD_GLOBAL_BUILTIN,
&&TARGET_EXTENDED_ARG,
&&TARGET_LIST_APPEND,
&&TARGET_SET_ADD,
@ -152,24 +152,24 @@ static void *opcode_targets[256] = {
&&TARGET_YIELD_VALUE,
&&TARGET_RESUME,
&&TARGET_MATCH_CLASS,
&&TARGET_LOAD_GLOBAL_MODULE,
&&TARGET_STORE_ATTR_INSTANCE_VALUE,
&&TARGET_STORE_ATTR_SLOT,
&&TARGET_FORMAT_VALUE,
&&TARGET_BUILD_CONST_KEY_MAP,
&&TARGET_BUILD_STRING,
&&TARGET_STORE_ATTR_SLOT,
&&TARGET_STORE_ATTR_WITH_HINT,
&&TARGET_STORE_FAST__LOAD_FAST,
&&TARGET_STORE_FAST__STORE_FAST,
&&TARGET_STORE_SUBSCR_DICT,
&&TARGET_LIST_EXTEND,
&&TARGET_SET_UPDATE,
&&TARGET_DICT_MERGE,
&&TARGET_DICT_UPDATE,
&&TARGET_STORE_SUBSCR_DICT,
&&TARGET_STORE_SUBSCR_LIST_INT,
&&TARGET_UNPACK_SEQUENCE_LIST,
&&TARGET_UNPACK_SEQUENCE_TUPLE,
&&TARGET_UNPACK_SEQUENCE_TWO_TUPLE,
&&_unknown_opcode,
&&TARGET_CALL,
&&TARGET_KW_NAMES,
&&_unknown_opcode,

View file

@ -2132,6 +2132,10 @@ _Py_Specialize_ForIter(PyObject *iter, _Py_CODEUNIT *instr, int oparg)
_Py_SET_OPCODE(*instr, FOR_ITER_LIST);
goto success;
}
else if (tp == &PyTupleIter_Type) {
_Py_SET_OPCODE(*instr, FOR_ITER_TUPLE);
goto success;
}
else if (tp == &PyRangeIter_Type && next_op == STORE_FAST) {
_Py_SET_OPCODE(*instr, FOR_ITER_RANGE);
goto success;