Misc #10754: Tiny optimisation of Set#include?

• Status: Closed
• Assignee: -

[ismael (Ismael Abreu)]

Hi!

I'm not sure if this a relevant optimisation but here it goes. A simple way to improve Set#include?
I don't think there should be any side effects with this change.

Here are some dummy benchmarks https://gist.github.com/ismaelga/68bb3ea51b4742f65699
They are not consistent, but that should be because of garbage collection or something. Right?

I've also sent a Pull Request: https://github.com/ruby/ruby/pull/810

[normalperson (Eric Wong)]

This speedup probably happens because Hash#[] has an optimized dispatch
VM instruction (opt_aref in insns.def) while Hash#include? does not.

Seems like an OK change to make, I'll commit in a few days unless others
raise objections.

[ismael (Ismael Abreu)]

With the benchmarks I've run I noticed that sometimes rarely, the old include? was faster.
Do you have any idea why is that? The only cause I can think of is the GC.

[normalperson (Eric Wong)]

I didn't notice the old code being faster, and I don't see your
benchmark triggering GC. Perhaps it is related to memory size
(swapping or CPU cache misses) or power management
(CPU clock frequency adjustment, "turbo boost", etc...)
Modern hardware is tricky to benchmark :/

[ismael (Ismael Abreu)]

Any update on this?

[normalperson (Eric Wong)]

Thanks for the reminder, I got distracted with other stuff :x
Committed r49568

[ismael (Ismael Abreu)]

Awesome! :)

Shouldn't this issue be closed also?

[marcandre (Marc-Andre Lafortune)]

• Status changed from Open to Closed