Skip to content

GH-116422: Modify a few uops so that they can be supported by tier 2 with hot/cold splitting #116832

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Mar 15, 2024
Merged

GH-116422: Modify a few uops so that they can be supported by tier 2 with hot/cold splitting #116832

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5bf0179
Make _CALL_BUILTIN_O tier 2 friendly
markshannon Mar 14, 2024
5720a6d
Make a few other uops tier2 friendly
markshannon Mar 14, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 7 additions & 2 deletions Include/internal/pycore_ceval.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -138,12 +138,12 @@ extern void _PyEval_DeactivateOpCache(void);
/* With USE_STACKCHECK macro defined, trigger stack checks in
_Py_CheckRecursiveCall() on every 64th call to _Py_EnterRecursiveCall. */
static inline int _Py_MakeRecCheck(PyThreadState *tstate) {
return (tstate->c_recursion_remaining-- <= 0
return (tstate->c_recursion_remaining-- < 0
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The reason this change is here is to make the guard tstate->c_recursion_remaining <= 0 instead of tstate->c_recursion_remaining <= 1. Comparing to zero is generally faster.

It also seems more correct. If c_recursion_remaining is 1, that suggest that there is 1 remaining call, not 0.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The actual number we compare to doesn't really matter, as long as we are consistent.

|| (tstate->c_recursion_remaining & 63) == 0);
}
#else
static inline int _Py_MakeRecCheck(PyThreadState *tstate) {
return tstate->c_recursion_remaining-- <= 0;
return tstate->c_recursion_remaining-- < 0;
}
#endif

Expand All @@ -161,6 +161,11 @@ static inline int _Py_EnterRecursiveCallTstate(PyThreadState *tstate,
return (_Py_MakeRecCheck(tstate) && _Py_CheckRecursiveCall(tstate, where));
}

static inline void _Py_EnterRecursiveCallTstateUnchecked(PyThreadState *tstate) {
assert(tstate->c_recursion_remaining > 0);
tstate->c_recursion_remaining--;
}

static inline int _Py_EnterRecursiveCall(const char *where) {
PyThreadState *tstate = _PyThreadState_GET();
return _Py_EnterRecursiveCallTstate(tstate, where);
Expand Down
34 changes: 15 additions & 19 deletions Python/bytecodes.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3376,14 +3376,12 @@ dummy_func(
DEOPT_IF(total_args != 1);
DEOPT_IF(!PyCFunction_CheckExact(callable));
DEOPT_IF(PyCFunction_GET_FLAGS(callable) != METH_O);
// CPython promises to check all non-vectorcall function calls.
DEOPT_IF(tstate->c_recursion_remaining <= 0);
STAT_INC(CALL, hit);
PyCFunction cfunc = PyCFunction_GET_FUNCTION(callable);
// This is slower but CPython promises to check all non-vectorcall
// function calls.
if (_Py_EnterRecursiveCallTstate(tstate, " while calling a Python object")) {
GOTO_ERROR(error);
}
PyObject *arg = args[0];
_Py_EnterRecursiveCallTstateUnchecked(tstate);
res = _PyCFunction_TrampolineCall(cfunc, PyCFunction_GET_SELF(callable), arg);
_Py_LeaveRecursiveCallTstate(tstate);
assert((res != NULL) ^ (_PyErr_Occurred(tstate) != NULL));
Expand Down Expand Up @@ -3480,10 +3478,11 @@ dummy_func(
}
res = PyLong_FromSsize_t(len_i);
assert((res != NULL) ^ (_PyErr_Occurred(tstate) != NULL));

if (res == NULL) {
GOTO_ERROR(error);
}
Py_DECREF(callable);
Py_DECREF(arg);
ERROR_IF(res == NULL, error);
}

inst(CALL_ISINSTANCE, (unused/1, unused/2, callable, self_or_null, args[oparg] -- res)) {
Expand All @@ -3505,11 +3504,12 @@ dummy_func(
}
res = PyBool_FromLong(retval);
assert((res != NULL) ^ (_PyErr_Occurred(tstate) != NULL));

if (res == NULL) {
GOTO_ERROR(error);
}
Py_DECREF(inst);
Py_DECREF(cls);
Py_DECREF(callable);
ERROR_IF(res == NULL, error);
}

// This is secretly a super-instruction
Expand Down Expand Up @@ -3543,16 +3543,14 @@ dummy_func(
DEOPT_IF(!Py_IS_TYPE(method, &PyMethodDescr_Type));
PyMethodDef *meth = method->d_method;
DEOPT_IF(meth->ml_flags != METH_O);
// CPython promises to check all non-vectorcall function calls.
DEOPT_IF(tstate->c_recursion_remaining <= 0);
PyObject *arg = args[1];
PyObject *self = args[0];
DEOPT_IF(!Py_IS_TYPE(self, method->d_common.d_type));
STAT_INC(CALL, hit);
PyCFunction cfunc = meth->ml_meth;
// This is slower but CPython promises to check all non-vectorcall
// function calls.
if (_Py_EnterRecursiveCallTstate(tstate, " while calling a Python object")) {
GOTO_ERROR(error);
}
_Py_EnterRecursiveCallTstateUnchecked(tstate);
res = _PyCFunction_TrampolineCall(cfunc, self, arg);
_Py_LeaveRecursiveCallTstate(tstate);
assert((res != NULL) ^ (_PyErr_Occurred(tstate) != NULL));
Expand Down Expand Up @@ -3616,13 +3614,11 @@ dummy_func(
PyObject *self = args[0];
DEOPT_IF(!Py_IS_TYPE(self, method->d_common.d_type));
DEOPT_IF(meth->ml_flags != METH_NOARGS);
// CPython promises to check all non-vectorcall function calls.
DEOPT_IF(tstate->c_recursion_remaining <= 0);
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The idea is that after DEOPT we will end up doing GOTO_ERROR in the original bytecode? (Is that what the comment is saying?)

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The comment refers to our obligation to check the C recursion limit for almost all builtin function calls.

We can fulfil that obligation more cheaply this way.

STAT_INC(CALL, hit);
PyCFunction cfunc = meth->ml_meth;
// This is slower but CPython promises to check all non-vectorcall
// function calls.
if (_Py_EnterRecursiveCallTstate(tstate, " while calling a Python object")) {
GOTO_ERROR(error);
}
_Py_EnterRecursiveCallTstateUnchecked(tstate);
res = _PyCFunction_TrampolineCall(cfunc, self, NULL);
_Py_LeaveRecursiveCallTstate(tstate);
assert((res != NULL) ^ (_PyErr_Occurred(tstate) != NULL));
Expand Down
32 changes: 15 additions & 17 deletions Python/executor_cases.c.h
View file
Open in desktop

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

32 changes: 15 additions & 17 deletions Python/generated_cases.c.h
View file
Open in desktop

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.