On Wed, Nov 07, 2012 at 03:00:39PM +0000, David Griffiths wrote: > Hi all, > > I've pushed some more changes to fluxa, work on the temporal recursion, > but also a big fix for the internals of the sound engine. > > It turns out (as we suspected) that there was a bottleneck in the synth > command ringbuffer for the realtime thread - it was trying to allocate > 4k per message out of 2.6Mb of ringbuffer, so max 64 messages at any one > time.
Dave, I ruled some stuff on my side out; It's not in Alsa's insistence to convert up and down between 44.1K and 4*K as though that were some recreational activity It's also not in my go-flux function which adds MIDI sync; I replaced that with the current Master version and the symptom's persist. What I'm seeing at running musical sketches I did before is that the music starts fine, then after a while there will be some glitches and after that the music tends to get condensed to a single sound, for example just the kick drum or just highhats. Editing the seq function tends to give weird stuff too; the transition comes rather late and sometimes it goes back to the old loop for a few measures after a bit. (play-now) seems fine in all cases. I'm inclined to blame the ringbuffer, given these, but your last gig sounded fine in the posted video. Could you verify you committed and pushed everything at your end? I'm also on a 32bit install, could it be that you are on 64 bit and we have some sort of number in a state of architectural confusion? Finally; did you test these with the plain sequencing method as well or just with SLUB-sync? Yours, Kas.
