[iiwusynth-devel] Re: Fluiwusynth

[Josh Green]

On Tue, 2002-10-22 at 23:13, M. Nentwig wrote:
> Hei,
> 
> Sorry for the subject line, I just couldn't restrain myself :-)
> 

Ahh, you reveal too much :) I don't think the name change has been
posted to the list yet. Peter, are you going to do the honors of
unveiling the new name?

> I just committed my changes to CVS. The problems Josh discovered are not
> yet fixed, coming soon...
> Modulating loop points has been addressed. Also the loop volume
> detection for the turnoff condition is run only once per sample, that
> should save some time.
> I have also disabled real-time priority for the MIDI thread, because it
> helped to get rid of dropouts.
> 

Which MIDI driver were you experiencing drop outs with? Which reminds
me, there are a couple of bugs with the ALSA sequencer driver that I
have been meaning to fix. Looks like the ALSA sequencer port isn't being
released when the iiwusynth context is destroyed. The other problem is
with treating SIGPIPE as an error when reading sequencer events, rather
than a temporary condition (basically an XRUN I think).

> 
> Before the voice loop is recoded in assembler, we should streamline the
> algorithm as much as possible. I'll add soon:
> - a condition to turn off interpolation at root pitch and normal phase
> - a condition to turn off the filter, if fc is high enough.
> - a condition to turn off reverb send and chorus send each, if the send
> level is 0
> 

So you've decided to add the root pitch/normal phase optimization?

> 
> > > Now if only the filter was less scratchy when ...
> 
> The problem could still be, that the filter is updated only once per
> buffer. This theory is easy to verify - set a buffer size of 4 or so,
> recompile the synth and compare the results. But CPU usage will go
> through the roof, because recomputing the filter is so expensive.
> 

I'll see if I can test this, if someone else does let me know the
results :) If this is what is causing the problem what could be done?
Perhaps its because the filter cutoff value is jumping too fast between
values? If that were true maybe some interpolation of the FilterFC
modulation is in order?

> 
> Cheers
> 
> Markus

Lates.
        Josh