Hi all, two weeks ago we gave a talk about the current status of Pyston, including some technical details about our JIT strategy and its motivation. I’m glad many of you could make it! For those who couldn’t, we now have the talk uploaded, check it out:
[Update] Slides
Nice overview. How could one get the slides?
LikeLike
Ok, just updated the post with the link 🙂
LikeLike
Nice talk! I’m intrigued by the various JITs that are now trying tracing on a much smaller scale. HHVM does it on a basic block level, Pyston on a bytecode level (with, as far as I understand it, plans to extend it to a somewhat larger scale).
Is the benchmark that tries more and more megamorphic attribute lookups available somewhere?
LikeLike
Sorry for the delay — the benchmark is available here. For what it’s worth, I put together a similar benchmark that stresses a different “problem difficulty” axis (control flow diversity) which I didn’t end up presenting but has a similar performance graph.
LikeLike
You might be also interested in pyparallel?
LikeLike
I saw that, looks a really interesting project. It’s almost unique in being an open source project that’s not using .NET and is targeted (for the moment) at Windows!
LikeLike