Fix free-threading scaling bottleneck in sys.intern and `PyObject_SetAttr` by
avoiding the interpreter-wide lock when the string is already interned and
immortalized.
Add special cases for classmethod and staticmethod descriptors in
_PyObject_GetMethodStackRef() to avoid calling tp_descr_get, which
avoids reference count contention on the bound method and underlying
callable. This improves scaling when calling classmethods and
staticmethods from multiple threads.
Also refactor method_vectorcall in classobject.c into a new _PyObject_VectorcallPrepend() helper so that it can be used by
PyObject_VectorcallMethod as well.
Add `_Py_type_getattro_stackref`, a variant of type attribute lookup
that returns `_PyStackRef` instead of `PyObject*`. This allows returning
deferred references in the free-threaded build, reducing reference count
contention when accessing type attributes.
This significantly improves scaling of namedtuple instantiation across
multiple threads.
* Add blurb
* Rename PyObject_GetAttrStackRef to _PyObject_GetAttrStackRef
* Apply suggestion from @vstinner
Co-authored-by: Victor Stinner <vstinner@python.org>
* Apply suggestion from @vstinner
Co-authored-by: Victor Stinner <vstinner@python.org>
* format
* Update Include/internal/pycore_function.h
Co-authored-by: Victor Stinner <vstinner@python.org>
---------
Co-authored-by: Victor Stinner <vstinner@python.org>
If we are specializing to `LOAD_GLOBAL_MODULE` or `LOAD_ATTR_MODULE`, try
to enable deferred reference counting for the value, if the object is owned by
a different thread. This applies to the free-threaded build only and should
improve scaling of multi-threaded programs.
The dataclasses `__init__` function is generated dynamically by a call to `exec()` and so doesn't have deferred reference counting enabled. Enable deferred reference counting on functions when assigned as an attribute to type objects to avoid reference count contention when creating dataclass instances.
Adds `_PyObject_GetMethodStackRef` which uses stackrefs and takes advantage of deferred reference counting in free-threading while calling method objects in vectorcall.
Add a separate benchmark that measures the effect of
`_PyObject_LookupSpecial()` on scaling.
In the process of cleaning up the scaling benchmarks for inclusion, I
unintentionally changed the "cmodule_function" benchmark to pass an
`int` to `math.floor()` instead of a `float`, which causes it to use the
`_PyObject_LookupSpecial()` code path. `_PyObject_LookupSpecial()` has
its own scaling issues that we want to measure separately from calling a
function on a C module.
These consist of a number of short snippets that help identify scaling
bottlenecks in the free threaded interpreter.
The current bottlenecks are in calling functions in benchmarks that call
functions (due to `LOAD_ATTR` not yet using deferred reference counting)
and when accessing thread-local data.
