[LAD] [ANN] IR: version 1.2

[torbenh]

On Wed, Jan 26, 2011 at 01:29:17PM +0100, Tom Szilagyi wrote:
> 2011/1/26 Jörn Nettingsmeier <nettings at folkwang-hochschule.de>:
> >
> > since it's based on zita-convolver: did you also use the multi-threaded
> > partitioned convolution approach from jconvolver?
> 
> Hi Jörn,
> 
> I'm not sure about the mentioned approach, since I'm not familiar with
> the jconvolver source, only zita-convolver (and only from a user
> p.o.v; haven't modified it). So I don't know that this multi-threaded
> approach you mention is something jconvolver actually does, or is it
> just the usual workings of zita-convolver (which I use in a really
> simple manner).
> 
> I will look at the jconvolver source to educate myself about said
> approach when I have a bit of free time.

http://www.mail-archive.com/linux-audio-dev@lists.linuxaudio.org/msg06888.html

the approach in the patch is not 100% RT clean.
however, the comments indicate where an atomic integer needs to be
applied to make it clean.

however... its highly unlikely that the RT thread blocks on that mutex.
and even if thats the case... its pretty unlikely that this would cause
an xrun.

please get this stuff fixed.
for IR, we probably need an lv2 extension which switches plugins into 
freewheeling mode.

i am unsure, how you determine the correct priority for the threads.
we need an extension for this also.

should be very trivial.


> 
> > if so, how does IR behave in freewheeling mode?
> 
> As usual: it works for me 10 out of 10 times (at least for what I do
> and how I work), but YMMV. I must admit that I don't use freewheeling
> that much, but some test passes with various projects (some simple and
> some big) on two machines, one older single processor and one fairly
> recent quad core (both 64 bit), indicated no problems. Of course this
> means nothing at all in this case.
> 
> >
> > i'm asking because jconvolver is known to have problems with the timing of
> > its non-jack helper threads in freewheeling mode, which has finally bitten
> > me a few days ago (haven't had problems with it before). jconvolver is nice
> > enough to halt the entire graph (at least with tschack), so it's always
> > obvious there was a problem.
> > i just want to make sure that *if* IR fails in freewheeling, it will fail in
> > some obvious way.
> 
> That would be definitely nice but currently missing. I would say that
> if there is a problem with jconvolver, IR most probably also has this
> problem.
> 
> I tend to listen to what I commit, though, so I bounce in realtime.
> 
> Tom
> _______________________________________________
> Linux-audio-dev mailing list
> Linux-audio-dev at lists.linuxaudio.org
> http://lists.linuxaudio.org/listinfo/linux-audio-dev

-- 
torben Hohn