bpo-47127: Specialize calls for fastcall c methods with keywords (GH-32125)

* add PRECALL_METHOD_DESCRIPTOR_FAST_WITH_KEYWORDS
This commit is contained in:
Kumar Aditya 2022-03-28 01:23:25 +05:30 committed by GitHub
parent 785cc67705
commit 58448cbd96
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
6 changed files with 77 additions and 37 deletions

54
Include/opcode.h generated
View file

@ -157,32 +157,33 @@ extern "C" {
#define PRECALL_BOUND_METHOD 62 #define PRECALL_BOUND_METHOD 62
#define PRECALL_BUILTIN_CLASS 63 #define PRECALL_BUILTIN_CLASS 63
#define PRECALL_BUILTIN_FAST_WITH_KEYWORDS 64 #define PRECALL_BUILTIN_FAST_WITH_KEYWORDS 64
#define PRECALL_NO_KW_BUILTIN_FAST 65 #define PRECALL_METHOD_DESCRIPTOR_FAST_WITH_KEYWORDS 65
#define PRECALL_NO_KW_BUILTIN_O 66 #define PRECALL_NO_KW_BUILTIN_FAST 66
#define PRECALL_NO_KW_ISINSTANCE 67 #define PRECALL_NO_KW_BUILTIN_O 67
#define PRECALL_NO_KW_LEN 72 #define PRECALL_NO_KW_ISINSTANCE 72
#define PRECALL_NO_KW_LIST_APPEND 73 #define PRECALL_NO_KW_LEN 73
#define PRECALL_NO_KW_METHOD_DESCRIPTOR_FAST 76 #define PRECALL_NO_KW_LIST_APPEND 76
#define PRECALL_NO_KW_METHOD_DESCRIPTOR_NOARGS 77 #define PRECALL_NO_KW_METHOD_DESCRIPTOR_FAST 77
#define PRECALL_NO_KW_METHOD_DESCRIPTOR_O 78 #define PRECALL_NO_KW_METHOD_DESCRIPTOR_NOARGS 78
#define PRECALL_NO_KW_STR_1 79 #define PRECALL_NO_KW_METHOD_DESCRIPTOR_O 79
#define PRECALL_NO_KW_TUPLE_1 80 #define PRECALL_NO_KW_STR_1 80
#define PRECALL_NO_KW_TYPE_1 81 #define PRECALL_NO_KW_TUPLE_1 81
#define PRECALL_PYFUNC 140 #define PRECALL_NO_KW_TYPE_1 140
#define RESUME_QUICK 141 #define PRECALL_PYFUNC 141
#define STORE_ATTR_ADAPTIVE 143 #define RESUME_QUICK 143
#define STORE_ATTR_INSTANCE_VALUE 150 #define STORE_ATTR_ADAPTIVE 150
#define STORE_ATTR_SLOT 153 #define STORE_ATTR_INSTANCE_VALUE 153
#define STORE_ATTR_WITH_HINT 154 #define STORE_ATTR_SLOT 154
#define STORE_FAST__LOAD_FAST 158 #define STORE_ATTR_WITH_HINT 158
#define STORE_FAST__STORE_FAST 159 #define STORE_FAST__LOAD_FAST 159
#define STORE_SUBSCR_ADAPTIVE 161 #define STORE_FAST__STORE_FAST 161
#define STORE_SUBSCR_DICT 167 #define STORE_SUBSCR_ADAPTIVE 167
#define STORE_SUBSCR_LIST_INT 168 #define STORE_SUBSCR_DICT 168
#define UNPACK_SEQUENCE_ADAPTIVE 169 #define STORE_SUBSCR_LIST_INT 169
#define UNPACK_SEQUENCE_LIST 170 #define UNPACK_SEQUENCE_ADAPTIVE 170
#define UNPACK_SEQUENCE_TUPLE 173 #define UNPACK_SEQUENCE_LIST 173
#define UNPACK_SEQUENCE_TWO_TUPLE 174 #define UNPACK_SEQUENCE_TUPLE 174
#define UNPACK_SEQUENCE_TWO_TUPLE 175
#define DO_TRACING 255 #define DO_TRACING 255
extern const uint8_t _PyOpcode_Caches[256]; extern const uint8_t _PyOpcode_Caches[256];
@ -347,6 +348,7 @@ const uint8_t _PyOpcode_Deopt[256] = {
[PRECALL_BOUND_METHOD] = PRECALL, [PRECALL_BOUND_METHOD] = PRECALL,
[PRECALL_BUILTIN_CLASS] = PRECALL, [PRECALL_BUILTIN_CLASS] = PRECALL,
[PRECALL_BUILTIN_FAST_WITH_KEYWORDS] = PRECALL, [PRECALL_BUILTIN_FAST_WITH_KEYWORDS] = PRECALL,
[PRECALL_METHOD_DESCRIPTOR_FAST_WITH_KEYWORDS] = PRECALL,
[PRECALL_NO_KW_BUILTIN_FAST] = PRECALL, [PRECALL_NO_KW_BUILTIN_FAST] = PRECALL,
[PRECALL_NO_KW_BUILTIN_O] = PRECALL, [PRECALL_NO_KW_BUILTIN_O] = PRECALL,
[PRECALL_NO_KW_ISINSTANCE] = PRECALL, [PRECALL_NO_KW_ISINSTANCE] = PRECALL,

View file

@ -294,6 +294,7 @@ _specializations = {
"PRECALL_BOUND_METHOD", "PRECALL_BOUND_METHOD",
"PRECALL_BUILTIN_CLASS", "PRECALL_BUILTIN_CLASS",
"PRECALL_BUILTIN_FAST_WITH_KEYWORDS", "PRECALL_BUILTIN_FAST_WITH_KEYWORDS",
"PRECALL_METHOD_DESCRIPTOR_FAST_WITH_KEYWORDS",
"PRECALL_NO_KW_BUILTIN_FAST", "PRECALL_NO_KW_BUILTIN_FAST",
"PRECALL_NO_KW_BUILTIN_O", "PRECALL_NO_KW_BUILTIN_O",
"PRECALL_NO_KW_ISINSTANCE", "PRECALL_NO_KW_ISINSTANCE",

View file

@ -0,0 +1 @@
Speed up calls to c functions with keyword arguments by 25% with specialization. Patch by Kumar Aditya.

View file

@ -5090,6 +5090,38 @@ handle_eval_breaker:
DISPATCH(); DISPATCH();
} }
TARGET(PRECALL_METHOD_DESCRIPTOR_FAST_WITH_KEYWORDS) {
int is_meth = is_method(stack_pointer, oparg);
int total_args = oparg + is_meth;
PyObject *callable = PEEK(total_args + 1);
DEOPT_IF(!Py_IS_TYPE(callable, &PyMethodDescr_Type), PRECALL);
PyMethodDef *meth = ((PyMethodDescrObject *)callable)->d_method;
DEOPT_IF(meth->ml_flags != (METH_FASTCALL|METH_KEYWORDS), PRECALL);
STAT_INC(PRECALL, hit);
SKIP_CALL();
int nargs = total_args-1;
STACK_SHRINK(nargs);
_PyCFunctionFastWithKeywords cfunc = (_PyCFunctionFastWithKeywords)(void(*)(void))meth->ml_meth;
PyObject *self = TOP();
PyObject *res = cfunc(self, stack_pointer, nargs - KWNAMES_LEN(), call_shape.kwnames);
assert((res != NULL) ^ (_PyErr_Occurred(tstate) != NULL));
call_shape.kwnames = NULL;
/* Free the arguments. */
for (int i = 0; i < nargs; i++) {
Py_DECREF(stack_pointer[i]);
}
Py_DECREF(self);
STACK_SHRINK(2-is_meth);
SET_TOP(res);
Py_DECREF(callable);
if (res == NULL) {
goto error;
}
CHECK_EVAL_BREAKER();
DISPATCH();
}
TARGET(PRECALL_NO_KW_METHOD_DESCRIPTOR_NOARGS) { TARGET(PRECALL_NO_KW_METHOD_DESCRIPTOR_NOARGS) {
assert(call_shape.kwnames == NULL); assert(call_shape.kwnames == NULL);
assert(oparg == 0 || oparg == 1); assert(oparg == 0 || oparg == 1);

View file

@ -64,23 +64,23 @@ static void *opcode_targets[256] = {
&&TARGET_PRECALL_BOUND_METHOD, &&TARGET_PRECALL_BOUND_METHOD,
&&TARGET_PRECALL_BUILTIN_CLASS, &&TARGET_PRECALL_BUILTIN_CLASS,
&&TARGET_PRECALL_BUILTIN_FAST_WITH_KEYWORDS, &&TARGET_PRECALL_BUILTIN_FAST_WITH_KEYWORDS,
&&TARGET_PRECALL_METHOD_DESCRIPTOR_FAST_WITH_KEYWORDS,
&&TARGET_PRECALL_NO_KW_BUILTIN_FAST, &&TARGET_PRECALL_NO_KW_BUILTIN_FAST,
&&TARGET_PRECALL_NO_KW_BUILTIN_O, &&TARGET_PRECALL_NO_KW_BUILTIN_O,
&&TARGET_PRECALL_NO_KW_ISINSTANCE,
&&TARGET_GET_ITER, &&TARGET_GET_ITER,
&&TARGET_GET_YIELD_FROM_ITER, &&TARGET_GET_YIELD_FROM_ITER,
&&TARGET_PRINT_EXPR, &&TARGET_PRINT_EXPR,
&&TARGET_LOAD_BUILD_CLASS, &&TARGET_LOAD_BUILD_CLASS,
&&TARGET_PRECALL_NO_KW_ISINSTANCE,
&&TARGET_PRECALL_NO_KW_LEN, &&TARGET_PRECALL_NO_KW_LEN,
&&TARGET_PRECALL_NO_KW_LIST_APPEND,
&&TARGET_LOAD_ASSERTION_ERROR, &&TARGET_LOAD_ASSERTION_ERROR,
&&TARGET_RETURN_GENERATOR, &&TARGET_RETURN_GENERATOR,
&&TARGET_PRECALL_NO_KW_LIST_APPEND,
&&TARGET_PRECALL_NO_KW_METHOD_DESCRIPTOR_FAST, &&TARGET_PRECALL_NO_KW_METHOD_DESCRIPTOR_FAST,
&&TARGET_PRECALL_NO_KW_METHOD_DESCRIPTOR_NOARGS, &&TARGET_PRECALL_NO_KW_METHOD_DESCRIPTOR_NOARGS,
&&TARGET_PRECALL_NO_KW_METHOD_DESCRIPTOR_O, &&TARGET_PRECALL_NO_KW_METHOD_DESCRIPTOR_O,
&&TARGET_PRECALL_NO_KW_STR_1, &&TARGET_PRECALL_NO_KW_STR_1,
&&TARGET_PRECALL_NO_KW_TUPLE_1, &&TARGET_PRECALL_NO_KW_TUPLE_1,
&&TARGET_PRECALL_NO_KW_TYPE_1,
&&TARGET_LIST_TO_TUPLE, &&TARGET_LIST_TO_TUPLE,
&&TARGET_RETURN_VALUE, &&TARGET_RETURN_VALUE,
&&TARGET_IMPORT_STAR, &&TARGET_IMPORT_STAR,
@ -139,39 +139,40 @@ static void *opcode_targets[256] = {
&&TARGET_LOAD_DEREF, &&TARGET_LOAD_DEREF,
&&TARGET_STORE_DEREF, &&TARGET_STORE_DEREF,
&&TARGET_DELETE_DEREF, &&TARGET_DELETE_DEREF,
&&TARGET_PRECALL_NO_KW_TYPE_1,
&&TARGET_PRECALL_PYFUNC, &&TARGET_PRECALL_PYFUNC,
&&TARGET_RESUME_QUICK,
&&TARGET_CALL_FUNCTION_EX, &&TARGET_CALL_FUNCTION_EX,
&&TARGET_STORE_ATTR_ADAPTIVE, &&TARGET_RESUME_QUICK,
&&TARGET_EXTENDED_ARG, &&TARGET_EXTENDED_ARG,
&&TARGET_LIST_APPEND, &&TARGET_LIST_APPEND,
&&TARGET_SET_ADD, &&TARGET_SET_ADD,
&&TARGET_MAP_ADD, &&TARGET_MAP_ADD,
&&TARGET_LOAD_CLASSDEREF, &&TARGET_LOAD_CLASSDEREF,
&&TARGET_COPY_FREE_VARS, &&TARGET_COPY_FREE_VARS,
&&TARGET_STORE_ATTR_INSTANCE_VALUE, &&TARGET_STORE_ATTR_ADAPTIVE,
&&TARGET_RESUME, &&TARGET_RESUME,
&&TARGET_MATCH_CLASS, &&TARGET_MATCH_CLASS,
&&TARGET_STORE_ATTR_INSTANCE_VALUE,
&&TARGET_STORE_ATTR_SLOT, &&TARGET_STORE_ATTR_SLOT,
&&TARGET_STORE_ATTR_WITH_HINT,
&&TARGET_FORMAT_VALUE, &&TARGET_FORMAT_VALUE,
&&TARGET_BUILD_CONST_KEY_MAP, &&TARGET_BUILD_CONST_KEY_MAP,
&&TARGET_BUILD_STRING, &&TARGET_BUILD_STRING,
&&TARGET_STORE_ATTR_WITH_HINT,
&&TARGET_STORE_FAST__LOAD_FAST, &&TARGET_STORE_FAST__LOAD_FAST,
&&TARGET_STORE_FAST__STORE_FAST,
&&TARGET_LOAD_METHOD, &&TARGET_LOAD_METHOD,
&&TARGET_STORE_SUBSCR_ADAPTIVE, &&TARGET_STORE_FAST__STORE_FAST,
&&TARGET_LIST_EXTEND, &&TARGET_LIST_EXTEND,
&&TARGET_SET_UPDATE, &&TARGET_SET_UPDATE,
&&TARGET_DICT_MERGE, &&TARGET_DICT_MERGE,
&&TARGET_DICT_UPDATE, &&TARGET_DICT_UPDATE,
&&TARGET_PRECALL, &&TARGET_PRECALL,
&&TARGET_STORE_SUBSCR_ADAPTIVE,
&&TARGET_STORE_SUBSCR_DICT, &&TARGET_STORE_SUBSCR_DICT,
&&TARGET_STORE_SUBSCR_LIST_INT, &&TARGET_STORE_SUBSCR_LIST_INT,
&&TARGET_UNPACK_SEQUENCE_ADAPTIVE, &&TARGET_UNPACK_SEQUENCE_ADAPTIVE,
&&TARGET_UNPACK_SEQUENCE_LIST,
&&TARGET_CALL, &&TARGET_CALL,
&&TARGET_KW_NAMES, &&TARGET_KW_NAMES,
&&TARGET_UNPACK_SEQUENCE_LIST,
&&TARGET_UNPACK_SEQUENCE_TUPLE, &&TARGET_UNPACK_SEQUENCE_TUPLE,
&&TARGET_UNPACK_SEQUENCE_TWO_TUPLE, &&TARGET_UNPACK_SEQUENCE_TWO_TUPLE,
&&_unknown_opcode, &&_unknown_opcode,
@ -253,6 +254,5 @@ static void *opcode_targets[256] = {
&&_unknown_opcode, &&_unknown_opcode,
&&_unknown_opcode, &&_unknown_opcode,
&&_unknown_opcode, &&_unknown_opcode,
&&_unknown_opcode,
&&TARGET_DO_TRACING &&TARGET_DO_TRACING
}; };

View file

@ -1446,6 +1446,10 @@ specialize_method_descriptor(PyMethodDescrObject *descr, _Py_CODEUNIT *instr,
_Py_SET_OPCODE(*instr, PRECALL_NO_KW_METHOD_DESCRIPTOR_FAST); _Py_SET_OPCODE(*instr, PRECALL_NO_KW_METHOD_DESCRIPTOR_FAST);
return 0; return 0;
} }
case METH_FASTCALL|METH_KEYWORDS: {
_Py_SET_OPCODE(*instr, PRECALL_METHOD_DESCRIPTOR_FAST_WITH_KEYWORDS);
return 0;
}
} }
SPECIALIZATION_FAIL(PRECALL, builtin_call_fail_kind(descr->d_method->ml_flags)); SPECIALIZATION_FAIL(PRECALL, builtin_call_fail_kind(descr->d_method->ml_flags));
return -1; return -1;