[LAU] Zyn and the art of jack audio/midi

cal cal at graggrag.com
Sat Aug 1 05:04:35 EDT 2009


Lars-Erik Helander wrote:
> I would be happy to do this. Could you tell a bit more about the jack 
> design in Zyn, what is bad and what you have fixed.
> If I know what to expect as a difference it is easier to make 
> comparative tests.
> 
> /Lars
> 

Cool, thanks. As it stands, zyn's jack audio/midi is somewhat primitive. The
jack process callback works against a horrible, horrible hack in Master.cpp.
Also, the code attempts to piggyback jack midi on top of the audio's process
callback, complete with it's own need for a mutex lock.

I've split off the jack midi into a separate entity, which sets up it own
connection to jack, and thus gets it's own process callback that just deals
with midi. 

I've seriously reworked the jack audio driver. The process callback
now directly calls the Master.cpp function formerly known as AudioOut().
I've renamed it MasterAudio(), in line with it's accompanying comment -
"Master audio out (the final sound)".

Having to acquire the master processing lock in order for the callback to
pick up the samples is problematic. Rather than throwing in a ringbuffer,
I'm currently trying a little experiment. The audio callback makes just
three attempts to get a lock using trylock. If it gets it, everything is
fine.

If it doesn't acquire the lock, then it simply feeds jack with the resident
set of samples currently in the buffers. My theory there is that repeating
256 samples might be a softer error than feeding 256 silent samples to jack.
>From my observations, it doesn't actually miss the lock all that often.

Cpu usage seems light enough, and jack audio does seem to be improved. The
Alsa audio also seems promising. I'm keen to see someone else's assessment,
maybe I'm just kidding myself.

And I'm dealing with Linux, so don't expect good things with a Windows build
for example.

For what it actually adds, the patch is ridiculously huge. One of the very
real problems with the zyn code is the complete absence of whitespace and
rational indentation. 25 years ago I used to be able to read (and write)
code like that, but not any more. Nowadays, I can't even look at code like
that without compulsively adjusting the whitespace - one of my many little
ocd's :-(.

cheers, Cal




More information about the Linux-audio-user mailing list