threads branch status update

[Andrew Whitworth]

I'm very excited about this branch, and I think there is a lot of
opportunity to both improve existing applications with new threading
support, but also to improve this implementation to further increase
performance and features over time.

I had promised to add the missing windows support for green threads,
but I have since lost access to the windows box I was using for
development and haven't been able to follow through with that promise.
I am going to try to set up a virtual machine and see what kind of
progress I can make there. If anybody else is able to help with this
work in any way, even just being available to answer the inevitable
questions about how the heck windows timers work, that would be great.

This is a pretty big deal, and we don't want to rush to merge it but
we also don't want to wait forever. We would like to start making
threading available, especially to Rakudo, soon. I think it's
reasonable to expect we can merge it by 4.6.0 or earlier, if people
look at it, test it, and like it.

Thanks,

--Andrew Whitworth



On Mon, Apr 16, 2012 at 10:50 AM, Stefan Seifert <nine at detonation.org> wrote:
> Dear parrot developers,
>
> I made a lot of progress on the threads branch these past few months.
> Threading works as outlined in my bachelor's thesis at:
> http://niner.name/Hybrid_Threads_for_the_Parrot_VM.pdf
>
> The branch is up to date with regards to the current master. It passes
> fulltest with gcc and g++ on my machine. It also passes Rakudo's spectest
> except for a failure in t/spec/S11-modules/require.t which I also see in
> master.
>
> Using moritz++' mandelbrot-color.pl on an optimized Parrot build as a
> benchmark I could not measure any performance degradation. In fact threads was
> even a bit faster than master, but this is surely within the error margin.
>
> I got a couple of test programs in examples/threads which demonstrate how to
> use the new threading capabilities.
>
> Though the threading support itself uses the old OS abstraction macros and
> thus should work on Windows as well, green threads still don't. And since
> green threads are used for supporting threading, there's still no progress
> there. On the other hand, there's not much that's missing: we just need
> replacements for POSIX alarms and pipes. In other words: some timer and a
> means for having threads block and wait for new work if idle so they don't
> burn CPU. Sounds like standard threading tutorial examples and a quick fix for
> someone with a Windows programming environment handy.
>
> So I guess it's time to throw a couple of more eyeballs at this branch. It
> will also offer plenty of opportunity for further improvements. It's got
> everything from a beginner's exercise like making the number of threads
> configurable or even auto-tuned to the number of available CPU cores to
> challenges like getting rid of the locks in the GC.
>
> And of course, simply using it in real programs will turn up some hidden bugs
> and limitations which might warrant some fixes as well ;)
>
> Regards,
> Stefan
> _______________________________________________
> http://lists.parrot.org/mailman/listinfo/parrot-dev
>