Re: [fluid-dev] JACK connection problems with fluidsynth

[David Henningsson]

On 2011-05-23 16:51, Shamus wrote:
> Hi David,
>
> I'm using JACK for audio and ALSA for MIDI, but it doesn't matter if I
> use JACK for both--the result is the same. If I run JACK in non-realtime
> mode there is no kick-out anywhere, but this is an unacceptable mode of
> operation as I use this setup for multi-track recording. I have an
> M-Audio Delta Audiophile 2496, which is an ICE1712 based card.
>
> The latency on JACK is 11.6 msec, according to QJackCtl, and this is
> already skating on the high side of things. I have it set to 256 frames,
> 2 periods/buffer at a sample rate of 44.1KHz.

So resetting would take more than ~ 5 ms then...that's pretty much, I 
guess. Would be interesting to make a profile of that and see what's 
taking so long.

> JACK version is 0.120.1.

As a side note, JACK2 is more forgiving towards applications not 
returning in time, but FS should work with both JACK versions.

> If fluidsynth is just a wrapper around libfluidsynth, then I'm not sure
> why Qsynth doesn't work correctly as it's basically just a GUI wrapper
> around libfluidsynth. But perhaps it's related to the problem with
> resetting fluidsynth from the command line: All the cases where kick-out
> happens haven't been addressed yet.
>
> Qsynth/fluidsynth used to work in the past with this configuration (I've
> been using it for a few years at least!), so I'm not sure why it doesn't
> work correctly anymore. :-/

It probably stopped working when we started caring about thread safety...

> But I *do* know that the patch that was
> posted to this list *does* work; apparently it just needs to be applied
> to some other code paths (like the "reset" code).

Unfortunately it is not that simple - applying the patch as it was 
applied to the initialisation can only be done when the audio engine is 
not running in parallel. When you do a system reset, the audio can get 
calls in parallel, which ruins thread safety.

Looking into fluid_synth_system_reset in particular, it seems it might 
do some time intensive tasks in fluid_rvoice_mixer_reset_fx that could 
cause the timeout. In this case I guess we might have to move parts of 
the reverb and chorus to the non-realtime thread or something :-/

// David

> BTW, Qsynth/fluidsynth are the only programs that exhibit this kick-out
> behavior. All of the other JACK aware programs that I use (Rosegarden,
> Ardour, Hydrogen) *do not* exhibit this kind of behavior.
>
> Warmest regards,
>
> -- Shamus
>
> On 05/22/11 07:32, David Henningsson wrote:
> > Hi Shamus,
> >
> > Being kicked out by Jack is essentially an optimization problem; Jack
> > kicks Fluidsynth out if it does not respond in time. I think this
> > timeout is configurable. Could it be that you're on the edge and just
> > need to adjust the timeout up a little bit, or increase the latency? In
> > general, it doesn't make sense that libfluidsynth would behave different
> > than the fluidsynth executable since the executable is just a wrapper
> > around libfluidsynth.
> >
> > Also, are you using Jack for MIDI as well as audio, or do you use Jack
> > audio + alsa MIDI?
> >
> > // David