bpo-37207: Use vertorcall for list() #18928
Conversation
Misc/NEWS.d/next/Core and Builtins/2019-06-09-10-54-31.bpo-37207.bLjgLS.rst
Outdated
Show resolved
Hide resolved
Co-Authored-By: Dong-hee Na <donghee.na92@gmail.com>
Objects/listobject.c
Outdated
| list_vectorcall(PyObject *type, PyObject * const*args, | ||
| size_t nargsf, PyObject *kwnames) | ||
| { | ||
| if (kwnames && PyTuple_GET_SIZE(kwnames) != 0) { |
There was a problem hiding this comment.
Since #18980 is merged.
We can replace the logic into
if (!_PyArg_NoKwnames("list", kwnames)) {
return NULL;
}
Objects/listobject.c
Outdated
| return NULL; | ||
| } | ||
| Py_ssize_t nargs = PyVectorcall_NARGS(nargsf); | ||
| if (nargs > 1) { |
There was a problem hiding this comment.
We can replace it into
if (!_PyArg_CheckPositional("list", nargs, 0, 1)) {
return NULL;
}| return NULL; | ||
| } | ||
|
|
||
| assert(PyType_Check(type)); |
There was a problem hiding this comment.
Can you move this assert into the top of the function body since this parameter is the 1st parameter?
There was a problem hiding this comment.
I don't see why that's necessary.
|
@corona10 @encukou: Can one of you please run a benchmark? @markshannon wrote "As an initial step, we can speed up calls to range, list and dict by about 30%." at https://bugs.python.org/issue37207#msg345077 but I would prefer to see directly results of a microbenchmarks. |
|
@vstinner It shows a similar enhancement to the collections to which this feature is applied. |
|
Ok, 1.15x faster (-13%) and -118 nanoseconds looks worth it to me. list(iterable) is commonly used. Again, LGTM ;-) |
|
I merged the PR, thanks @markshannon, @encukou and @corona10 ;-) There was an interesting typo: "vertorcall call" :-) |
This continues the
list()part of #13930. The complete pull request is stalled on discussions around dicts, butlist()should not be controversial. (Range was done in #18464 and I plan to open PRs for other parts if this is merged.)I did two small changes on top of Mark's code.
https://bugs.python.org/issue37207