bpo-37231: optimize calls of special methods

[jdemeyer]

https://bugs.python.org/issue37231

[mangrisano]

/cc @vstinner @methane

[methane]

LGTM. Have you run some microbench?
If there is measurable speedup, please write it in NEWS entry.

[jdemeyer]

LGTM. Have you run some microbench?

Yes. With this code

class X:
    def __mul__(self, other):
        pass

x = X()

and benchmarking x * x I get a speedup from

Mean +- std dev: 103 ns +- 1 ns

to

Mean +- std dev: 98.6 ns +- 2.2 ns

If there is measurable speedup, please write it in NEWS entry.

Done.

[jdemeyer]

I'm thinking that this vectorcall_method is useful enough to become official API. But that will be for a different PR.