search

mathics / Mathics

This repository is for archival. Please see https://github.com/Mathics3/mathics-core
https://mathics.org
Other
2.08k stars 208 forks source link

Small performance improvements though rule rewriting #1576

Closed TiagoCavalcante closed 2 years ago

TiagoCavalcante commented 2 years ago

Last small tweaks at least for the next week.

image

I'll update mathics-benchmark so it breaks the query into multiple lines every x characters.

rocky commented 2 years ago

@TiagoCavalcante I would greatly appreciate it if you would break down this PR into the part that is crucial for the 10% speedup and put the other stuff in another PR.

We would fix the performance problems sooner and easier if we were to focus attention and have those changes giving big wins and isolate those a single Pull Request.

Sure, I understand there is all this other stuff that might be changed and that's okay. But please, keep the important stuff from all the other things that one might want to do.

Thanks.

TiagoCavalcante commented 2 years ago

Ok, of course, I'll do it too for my other PR too tomorrow.

TiagoCavalcante commented 2 years ago

@rocky benchmarking again in a better way (without using the machine at the same time), and split into groups I see all my changes except Minus and MantissaExponent have improvements: image image

How should I split this PR?

TiagoCavalcante commented 2 years ago

Please ignore SplitBy.

rocky commented 2 years ago

@rocky benchmarking again in a better way (without using the machine at the same time), and split into groups I see all my changes except Minus and MantissaExponent have improvements: image image

How should I split this PR?

I'm trying to figure out which is the part that is doing the most speedups. So I'd like to try to replicate the benchmark. Then I'll pare the PR down into the essential part so that it is in one PR. Stay tuned...

rocky commented 2 years ago

@TiagoCavalcante merging into a branch so I can duplicate the results and see if I can pinpoint what's what.