[issue37562] PEP 590 implementation may have introduced a performance regression

[Jeroen Demeyer]

Jeroen Demeyer <J.Demeyer at UGent.be> added the comment:

I did some benchmarks WITHOUT PGO (simply because it's much faster to compile and therefore easier to test things out).

The command I used for testing is
./python -m perf timeit --duplicate 200 -s 'f = len; x = ()' 'f(x)'

* d30da5dd9a8a965cf24a22bbaff8a5b1341c2944 (before PEP 590)
  Mean +- std dev: 25.1 ns +- 0.2 ns
* aacc77fbd77640a8f03638216fa09372cc21673d (first commit of PEP 590)
  Mean +- std dev: 29.3 ns +- 0.9 ns
* b8e198a5d09ca876b87baaf6efd2b2e7c9e3a0b3 (3.8 branch)
  Mean +- std dev: 29.5 ns +- 0.3 ns
* cd6e83b4810549c308ab2d7315dbab526e35ccf6 (master)
  Mean +- std dev: 25.1 ns +- 0.2 ns

So there does seem to be a regression with 3.8 but not with 3.9. I don't know what to do with this... can we backport more of the PEP 590 commits to 3.8? Or will the release manager complain?

----------

_______________________________________
Python tracker <report at bugs.python.org>
<https://bugs.python.org/issue37562>
_______________________________________