gh-134584: Eliminate redundant refcounting from _CALL_BUILTION_O (GH-142695)

Co-authored-by: Ken Jin <kenjin4096@gmail.com>
This commit is contained in:
PuQing 2025-12-14 18:28:02 +08:00 committed by GitHub
parent 11aef219f5
commit a2a400af1e
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
9 changed files with 75 additions and 39 deletions

View file

@ -1111,7 +1111,7 @@ const struct opcode_metadata _PyOpcode_opcode_metadata[267] = {
[CALL_BUILTIN_CLASS] = { true, INSTR_FMT_IBC00, HAS_ARG_FLAG | HAS_EVAL_BREAK_FLAG | HAS_DEOPT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
[CALL_BUILTIN_FAST] = { true, INSTR_FMT_IBC00, HAS_ARG_FLAG | HAS_EVAL_BREAK_FLAG | HAS_DEOPT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
[CALL_BUILTIN_FAST_WITH_KEYWORDS] = { true, INSTR_FMT_IBC00, HAS_ARG_FLAG | HAS_EVAL_BREAK_FLAG | HAS_DEOPT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
[CALL_BUILTIN_O] = { true, INSTR_FMT_IBC00, HAS_ARG_FLAG | HAS_EVAL_BREAK_FLAG | HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
[CALL_BUILTIN_O] = { true, INSTR_FMT_IBC00, HAS_ARG_FLAG | HAS_EVAL_BREAK_FLAG | HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG },
[CALL_FUNCTION_EX] = { true, INSTR_FMT_IX, HAS_EVAL_BREAK_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG | HAS_SYNC_SP_FLAG | HAS_NEEDS_GUARD_IP_FLAG },
[CALL_INTRINSIC_1] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
[CALL_INTRINSIC_2] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
@ -1359,7 +1359,7 @@ _PyOpcode_macro_expansion[256] = {
[CALL_BUILTIN_CLASS] = { .nuops = 2, .uops = { { _CALL_BUILTIN_CLASS, OPARG_SIMPLE, 3 }, { _CHECK_PERIODIC_AT_END, OPARG_REPLACED, 3 } } },
[CALL_BUILTIN_FAST] = { .nuops = 2, .uops = { { _CALL_BUILTIN_FAST, OPARG_SIMPLE, 3 }, { _CHECK_PERIODIC_AT_END, OPARG_REPLACED, 3 } } },
[CALL_BUILTIN_FAST_WITH_KEYWORDS] = { .nuops = 2, .uops = { { _CALL_BUILTIN_FAST_WITH_KEYWORDS, OPARG_SIMPLE, 3 }, { _CHECK_PERIODIC_AT_END, OPARG_REPLACED, 3 } } },
[CALL_BUILTIN_O] = { .nuops = 2, .uops = { { _CALL_BUILTIN_O, OPARG_SIMPLE, 3 }, { _CHECK_PERIODIC_AT_END, OPARG_REPLACED, 3 } } },
[CALL_BUILTIN_O] = { .nuops = 4, .uops = { { _CALL_BUILTIN_O, OPARG_SIMPLE, 3 }, { _POP_TOP, OPARG_SIMPLE, 3 }, { _POP_TOP, OPARG_SIMPLE, 3 }, { _CHECK_PERIODIC_AT_END, OPARG_REPLACED, 3 } } },
[CALL_INTRINSIC_1] = { .nuops = 1, .uops = { { _CALL_INTRINSIC_1, OPARG_SIMPLE, 0 } } },
[CALL_INTRINSIC_2] = { .nuops = 1, .uops = { { _CALL_INTRINSIC_2, OPARG_SIMPLE, 0 } } },
[CALL_ISINSTANCE] = { .nuops = 3, .uops = { { _GUARD_THIRD_NULL, OPARG_SIMPLE, 3 }, { _GUARD_CALLABLE_ISINSTANCE, OPARG_SIMPLE, 3 }, { _CALL_ISINSTANCE, OPARG_SIMPLE, 3 } } },

View file

@ -426,7 +426,7 @@ extern "C" {
#define _CALL_BUILTIN_CLASS_r01 619
#define _CALL_BUILTIN_FAST_r01 620
#define _CALL_BUILTIN_FAST_WITH_KEYWORDS_r01 621
#define _CALL_BUILTIN_O_r01 622
#define _CALL_BUILTIN_O_r03 622
#define _CALL_INTRINSIC_1_r11 623
#define _CALL_INTRINSIC_2_r21 624
#define _CALL_ISINSTANCE_r31 625

View file

@ -281,7 +281,7 @@ const uint32_t _PyUop_Flags[MAX_UOP_ID+1] = {
[_CREATE_INIT_FRAME] = HAS_ARG_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG | HAS_SYNC_SP_FLAG,
[_EXIT_INIT_CHECK] = HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG,
[_CALL_BUILTIN_CLASS] = HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG,
[_CALL_BUILTIN_O] = HAS_ARG_FLAG | HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG,
[_CALL_BUILTIN_O] = HAS_ARG_FLAG | HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG,
[_CALL_BUILTIN_FAST] = HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG,
[_CALL_BUILTIN_FAST_WITH_KEYWORDS] = HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG,
[_GUARD_CALLABLE_LEN] = HAS_DEOPT_FLAG,
@ -2585,7 +2585,7 @@ const _PyUopCachingInfo _PyUop_Caching[MAX_UOP_ID+1] = {
[_CALL_BUILTIN_O] = {
.best = { 0, 0, 0, 0 },
.entries = {
{ 1, 0, _CALL_BUILTIN_O_r01 },
{ 3, 0, _CALL_BUILTIN_O_r03 },
{ -1, -1, -1 },
{ -1, -1, -1 },
{ -1, -1, -1 },
@ -3744,7 +3744,7 @@ const uint16_t _PyUop_Uncached[MAX_UOP_REGS_ID+1] = {
[_CREATE_INIT_FRAME_r01] = _CREATE_INIT_FRAME,
[_EXIT_INIT_CHECK_r10] = _EXIT_INIT_CHECK,
[_CALL_BUILTIN_CLASS_r01] = _CALL_BUILTIN_CLASS,
[_CALL_BUILTIN_O_r01] = _CALL_BUILTIN_O,
[_CALL_BUILTIN_O_r03] = _CALL_BUILTIN_O,
[_CALL_BUILTIN_FAST_r01] = _CALL_BUILTIN_FAST,
[_CALL_BUILTIN_FAST_WITH_KEYWORDS_r01] = _CALL_BUILTIN_FAST_WITH_KEYWORDS,
[_GUARD_CALLABLE_LEN_r03] = _GUARD_CALLABLE_LEN,
@ -4032,7 +4032,7 @@ const char *const _PyOpcode_uop_name[MAX_UOP_REGS_ID+1] = {
[_CALL_BUILTIN_FAST_WITH_KEYWORDS] = "_CALL_BUILTIN_FAST_WITH_KEYWORDS",
[_CALL_BUILTIN_FAST_WITH_KEYWORDS_r01] = "_CALL_BUILTIN_FAST_WITH_KEYWORDS_r01",
[_CALL_BUILTIN_O] = "_CALL_BUILTIN_O",
[_CALL_BUILTIN_O_r01] = "_CALL_BUILTIN_O_r01",
[_CALL_BUILTIN_O_r03] = "_CALL_BUILTIN_O_r03",
[_CALL_INTRINSIC_1] = "_CALL_INTRINSIC_1",
[_CALL_INTRINSIC_1_r11] = "_CALL_INTRINSIC_1_r11",
[_CALL_INTRINSIC_2] = "_CALL_INTRINSIC_2",

View file

@ -2133,6 +2133,21 @@ class TestUopsOptimization(unittest.TestCase):
self.assertNotIn("_COMPARE_OP_INT", uops)
self.assertNotIn("_GUARD_IS_TRUE_POP", uops)
def test_call_builtin_o(self):
def testfunc(n):
x = 0
for _ in range(n):
y = abs(1)
x += y
return x
res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, TIER2_THRESHOLD)
self.assertIsNotNone(ex)
uops = get_opnames(ex)
self.assertIn("_CALL_BUILTIN_O", uops)
self.assertIn("_POP_TOP", uops)
def test_get_len_with_const_tuple(self):
def testfunc(n):
x = 0.0

View file

@ -0,0 +1 @@
Eliminate redundant refcounting from ``_CALL_BUILTIN_O``.

View file

@ -4186,7 +4186,7 @@ dummy_func(
_CALL_BUILTIN_CLASS +
_CHECK_PERIODIC_AT_END;
op(_CALL_BUILTIN_O, (callable, self_or_null, args[oparg] -- res)) {
op(_CALL_BUILTIN_O, (callable, self_or_null, args[oparg] -- res, a, c)) {
/* Builtin METH_O functions */
PyObject *callable_o = PyStackRef_AsPyObjectBorrow(callable);
@ -4206,12 +4206,12 @@ dummy_func(
PyObject *res_o = _PyCFunction_TrampolineCall(cfunc, PyCFunction_GET_SELF(callable_o), PyStackRef_AsPyObjectBorrow(arg));
_Py_LeaveRecursiveCallTstate(tstate);
assert((res_o != NULL) ^ (_PyErr_Occurred(tstate) != NULL));
PyStackRef_CLOSE(arg);
DEAD(args);
DEAD(self_or_null);
PyStackRef_CLOSE(callable);
ERROR_IF(res_o == NULL);
if (res_o == NULL) {
ERROR_NO_POP();
}
a = arg;
c = callable;
INPUTS_DEAD();
res = PyStackRef_FromPyObjectSteal(res_o);
}
@ -4219,6 +4219,8 @@ dummy_func(
unused/1 +
unused/2 +
_CALL_BUILTIN_O +
POP_TOP +
POP_TOP +
_CHECK_PERIODIC_AT_END;
op(_CALL_BUILTIN_FAST, (callable, self_or_null, args[oparg] -- res)) {

View file

@ -13101,13 +13101,15 @@
break;
}
case _CALL_BUILTIN_O_r01: {
case _CALL_BUILTIN_O_r03: {
CHECK_CURRENT_CACHED_VALUES(0);
assert(WITHIN_STACK_BOUNDS_WITH_CACHE());
_PyStackRef *args;
_PyStackRef self_or_null;
_PyStackRef callable;
_PyStackRef res;
_PyStackRef a;
_PyStackRef c;
oparg = CURRENT_OPARG();
args = &stack_pointer[-oparg];
self_or_null = stack_pointer[-1 - oparg];
@ -13146,23 +13148,19 @@
stack_pointer = _PyFrame_GetStackPointer(frame);
_Py_LeaveRecursiveCallTstate(tstate);
assert((res_o != NULL) ^ (_PyErr_Occurred(tstate) != NULL));
_PyFrame_SetStackPointer(frame, stack_pointer);
PyStackRef_CLOSE(arg);
stack_pointer = _PyFrame_GetStackPointer(frame);
stack_pointer += -2 - oparg;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
_PyFrame_SetStackPointer(frame, stack_pointer);
PyStackRef_CLOSE(callable);
stack_pointer = _PyFrame_GetStackPointer(frame);
if (res_o == NULL) {
SET_CURRENT_CACHED_VALUES(0);
JUMP_TO_ERROR();
}
a = arg;
c = callable;
res = PyStackRef_FromPyObjectSteal(res_o);
_tos_cache2 = c;
_tos_cache1 = a;
_tos_cache0 = res;
_tos_cache1 = PyStackRef_ZERO_BITS;
_tos_cache2 = PyStackRef_ZERO_BITS;
SET_CURRENT_CACHED_VALUES(1);
SET_CURRENT_CACHED_VALUES(3);
stack_pointer += -2 - oparg;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
assert(WITHIN_STACK_BOUNDS_WITH_CACHE());
break;
}

View file

@ -2249,6 +2249,9 @@
_PyStackRef self_or_null;
_PyStackRef *args;
_PyStackRef res;
_PyStackRef a;
_PyStackRef c;
_PyStackRef value;
/* Skip 1 cache entry */
/* Skip 2 cache entries */
// _CALL_BUILTIN_O
@ -2290,24 +2293,35 @@
stack_pointer = _PyFrame_GetStackPointer(frame);
_Py_LeaveRecursiveCallTstate(tstate);
assert((res_o != NULL) ^ (_PyErr_Occurred(tstate) != NULL));
_PyFrame_SetStackPointer(frame, stack_pointer);
PyStackRef_CLOSE(arg);
stack_pointer = _PyFrame_GetStackPointer(frame);
stack_pointer += -2 - oparg;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
_PyFrame_SetStackPointer(frame, stack_pointer);
PyStackRef_CLOSE(callable);
stack_pointer = _PyFrame_GetStackPointer(frame);
if (res_o == NULL) {
JUMP_TO_LABEL(error);
}
a = arg;
c = callable;
res = PyStackRef_FromPyObjectSteal(res_o);
}
// _POP_TOP
{
value = c;
stack_pointer[-2 - oparg] = res;
stack_pointer[-1 - oparg] = a;
stack_pointer += -oparg;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
_PyFrame_SetStackPointer(frame, stack_pointer);
PyStackRef_XCLOSE(value);
stack_pointer = _PyFrame_GetStackPointer(frame);
}
// _POP_TOP
{
value = a;
stack_pointer += -1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
_PyFrame_SetStackPointer(frame, stack_pointer);
PyStackRef_XCLOSE(value);
stack_pointer = _PyFrame_GetStackPointer(frame);
}
// _CHECK_PERIODIC_AT_END
{
stack_pointer[0] = res;
stack_pointer += 1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
_PyFrame_SetStackPointer(frame, stack_pointer);
int err = check_periodics(tstate);
stack_pointer = _PyFrame_GetStackPointer(frame);

View file

@ -2964,10 +2964,16 @@
case _CALL_BUILTIN_O: {
JitOptRef res;
JitOptRef a;
JitOptRef c;
res = sym_new_not_null(ctx);
CHECK_STACK_BOUNDS(-1 - oparg);
a = sym_new_not_null(ctx);
c = sym_new_not_null(ctx);
CHECK_STACK_BOUNDS(1 - oparg);
stack_pointer[-2 - oparg] = res;
stack_pointer += -1 - oparg;
stack_pointer[-1 - oparg] = a;
stack_pointer[-oparg] = c;
stack_pointer += 1 - oparg;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
break;
}