[PD] Parallel Audio Processing (Was: How's Pd limited?)

[Jonathan Wilkes]

I'd say a solution would deserve the nobel prize in computers. :)

-Jonathan


    On Wednesday, February 24, 2016 11:40 AM, david medine <dmedine at ucsd.edu> wrote:
 

  Reading a full inbox in a non-parallelized fashion leads to cross posting. Sorry for the repost, but this belongs on this thread.
 
 Yeah, but the problem here (automatic compute resource distribution) isn't just with the actual distribution. Control timing is a huge issue too. If you have multiple voices of the same synth on different threads, good luck playing a chord. It will frequently be an arpeggio. If there are very strict, predictable rules about the order of when each voice computes and when it sleeps in order to wait for new samples and control signals, this problem vanishes, but then you are no longer  computing in parallel, and you might as well have everything on one thread anyway. 
 
 This is a really interesting problem. If someone can solve it, she deserves the nobel prize in computer music.
 
 On 2/24/16 5:59 AM, Christof Ressi wrote:
  
 Another possibility for at least some degree of parallel audio processing, I guess, is to use streaming objects + several instances of Pd.  I tried out [udpsend~] and [udpreceive~] (taken from the "net" library in Pd extended) and they seem to work reliably. Of course there is some latency envolved and it's no practical solution for our future 100 core machines :-p.

What are in your experience the best methods for sharing audio via different Pd instances?
And does anyone know the current status of "Audio over OSC"? I found this paper from 2010 http://iem.kug.ac.at/fileadmin/media/iem/projects/2010/jaeger.pdf 
 
Christof



 
 Gesendet: Mittwoch, 24. Februar 2016 um 11:12 Uhr
Von: Jack <jack at rybn.org>
An: pd-list at lists.iem.at
Betreff: Re: [PD] How's Pd limited?

Hello,

Le 24/02/2016 00:19, Matt Barber a écrit :
 
 Can anyone explain more why [pd~] doesn't fulfill the desire for
parallel processing, and maybe provide an example of something outside
of Pd that does? I don't feel like I have a great handle on the design.
As Jonathan said, it seems like Pd's determinism constraint is a big
hurdle to clear, though it's already relaxed a bit with netsend/receive.
What are the main differences between running an instance of Pd as a
[pd~] slave to another instance, and running two instances that
communicate via netsend/receive and jack?
 
 I think, the main difference is :
- with [pd~] your communication is synchronous
- with [netsend]/[netreceive] your communication is asynchronous

So (if i am right), if something is heavy to compute (more than 100% of
your CPU) in your subprocess with [pd~], your parent have to wait the
end of this computation. This is not the case with [netsend]/[netreceive].
++

Jack


 
 On Tue, Feb 23, 2016 at 5:45 PM, David Medine <dmedine at ucsd.edu
<mailto:dmedine at ucsd.edu>> wrote:

    I think we all need to learn more about multi-threading if we want
    to run real-time, modular, digital signal processing algorithms on
    multi-core machines. I, for one, can not think of any general,
    robust way to do this. In that sense, Pd's adherence to single
    threading is actually a very elegant solution to the problem.


    On 2/23/2016 12:25 PM, martin brinkmann wrote:

        On 22/02/16 02:49, Matti Viljamaa wrote:

            How do you think Pure Data is limited?

        for me the only real and important (i can think of at the
        moment) limitation is the block-based audio processing.
        to me this seems quite unnatural and inconvenient when dealing with
        digital audio. it kept me for a couple of years from using pd,
        though it
        is only a 'showstopper' in rather few cases, i found out.
        feedback in large/complex patches for example, since it
        is not very practical (or possible at all) to re-block
        everything to 1...

        what i tried but couldn't (yet): build a decent piano-roll
        editor (vanilla).

        and i believe too, pd has to 'learn' better multithreading to run
        adequately on our future machines with hundreds or even thousands of
        arm-cores...

        _______________________________________________
        Pd-list at lists.iem.at <mailto:Pd-list at lists.iem.at> mailing list
        UNSUBSCRIBE and account-management ->
        http://lists.puredata.info/listinfo/pd-list



    _______________________________________________
    Pd-list at lists.iem.at <mailto:Pd-list at lists.iem.at> mailing list
    UNSUBSCRIBE and account-management ->
    http://lists.puredata.info/listinfo/pd-list




_______________________________________________
Pd-list at lists.iem.at mailing list
UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list

 
 _______________________________________________
Pd-list at lists.iem.at mailing list
UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list

 
 _______________________________________________
Pd-list at lists.iem.at mailing list
UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
 
 
 
_______________________________________________
Pd-list at lists.iem.at mailing list
UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list


  
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.puredata.info/pipermail/pd-list/attachments/20160224/38703a34/attachment-0001.html>