I dug into CRuby and YJIT to better understand a recent, popular synthetic benchmark run against various languages. It was alot of fun to investigate how YJIT and CRuby work together! jpcamara.com/2024/12/01/s... @k0kubun.com @timtilberg.bsky.social - ThreadSky

jpcamara.com • 95 days ago

I dug into CRuby and YJIT to better understand a recent, popular synthetic benchmark run against various languages. It was alot of fun to investigate how YJIT and CRuby work together!
https://jpcamara.com/2024/12/01/speeding-up-ruby.html
@k0kubun.com @timtilberg.bsky.social

Comments

csabahenk.bsky.social•66 days ago

> I have no idea why the Primitive.cexpr! is used to return the enumerator, or what value Primitive.attr! :inline_block provides. Please comment if you have insights there!

By blaming "def times; ...; end" in numeric.rb, you can find out about the rationales .

...

csabahenk.bsky.social•66 days ago

The 'inline block' "pragma" is brought in by 27c1dd8, and AFAIU it's for enabling some inlinging for the YJIT compiler. (Yet, at the EOD, it's just a guess on my part).

...

csabahenk.bsky.social•66 days ago

As for the ' Primitive.cexpr!' thingy, 3dccb71 brings this in, and the intent is pretty clear: do not rely on method calls that the user might redefine.

csabahenk.bsky.social•66 days ago

@jpcamara.com ^^

jpcamara.com•66 days ago

Thanks @csabahenk.bsky.social ! Great pointers! It's interesting to see a compile.c specialization for Primitive.attr. Very relevant to https://jpcamara.com/2024/12/27/peephole-optimizations-adding-optrespondto-to.html

I think the core of how their internals exactly work will take a few reads to fully understand, but this is great

daniel-morrison.bsky.social•95 days ago

great post!

jpcamara.com•95 days ago

Thanks @daniel-morrison.bsky.social !

onghu.com•95 days ago

@jpcamara.com
Since I use Windows most of the time and YJIT doesn't work on Windows, it bothers me a bit that rewriting these in Ruby may be good for YJIT but may be slower without... it's conjecture, of course!

jpcamara.com•95 days ago

Yea It's a fair concern. I looked at many of the rewrite PRs though, and they would usually verify that the Ruby code ran as fast, or negligibly slower. I think it's the reason `with_yjit` was introduced, to keep the existing C versions if there are any perf regressions on non-YJIT

onghu.com•95 days ago

Ya, it's just a concern yet but fwiw, the numbers on Ruby 3.4 (and 3.3) are worse than Ruby 3.2 in your post 🙂

jpcamara.com•95 days ago

Good point ☠️ It might be nice for my repo https://github.com/jpcamara/ruby_microbench to become something that gets run periodically as these YJIT changes are made. we'll see.

mediafinger.com•94 days ago

Are you writing Ruby or also actually running Ruby on Windows?

onghu.com•94 days ago

I use Ruby on Windows for a bunch of stuff from data manipulation, prototypes, automation, etc. - it's my go-to language for doing stuff & Windows is where I spend most of my time. I personally do some Rails but mainly during development.

onghu.com•94 days ago

I've had to move away from super-active programming due to work commitments (read: management) but this is my escape since I get more done in Ruby (compared with C or C++) in the little time I have.

mediafinger.com•94 days ago

If you only run scripts on your machine, then performance differences usually don't matter much. Sure, you might feel a difference when crushing loads of data... but you could always run those from the DSL. 😉

onghu.com•94 days ago

Btw, I also, more than once, built something in RSpec to do validation testing of our deliverables (incl 1 with JRuby that was also shipped to our customer for them to use in their environment). Those were fun!

onghu.com•94 days ago

Yes, I've been in this game long enough, and I know it. I think I need to write mg post about why Ruby on Windows matters 🙂

jpcamara.com•95 days ago

@eregon.bsky.social truffleruby leads the pack in perf on this, by a lot. Amazing work

https://github.com/jpcamara/ruby_microbench

I’m gonna submit a PR to add it to the original languages repo as well

jpcamara.com•95 days ago

It gets pretty deep near the end - come down the YJIT rabbit hole with me 😲

lucianghinda.com•95 days ago

Great article!

jpcamara.com•95 days ago

Thanks @lucianghinda.com !

bradgessler.com•95 days ago

I’m happy that more Ruby will be written in Ruby, which means I will be more inclined to hack on Ruby.

lucianghinda.com•94 days ago

Yes, I think the same. The more Ruby is written in Ruby the more we can contribute to it or at least think this is possible

Comments

Posting Rules

Reply