On 2018-09-19, Antoine Pitrou wrote:
Thanks for doing this and thanks for sharing the benchmark numbers. This helps bring reality into the discussion ;-)
It seems to me that CPython has reached a sort of "local maximum" in terms of optimization. Change the implementation in any serious way and it is going to get slower. There are all these small optimization tricks that only work well based on the particulars of the current design. So, I don't feel too bad amount my performance numbers. Computing with fixed size integers seems to matter not so much for modern programs. So, I didn't gain much from stuff like fixedint + fixedint. However, there is a whole new set of tricks the tagged pointer support will allow now. Endless fun. ;-P
Cheers,
Neil