[PlanetCCRMA] Denormal problem with TAP Reverb

[Fernando Lopez-Lezcano]

On Fri, 2006-11-24 at 21:56 +0100, Nigel Henry wrote:
> On Friday 24 November 2006 09:29, Ken Smith wrote:
> > nicholas manojlovic wrote:
> > > Ken Smith wrote:
> > >> Hi, I seem to be getting what looks like a denormal problem.
> > >>
> > >> The setup is Intel P4 HT, FC5 2.6.17 CCRMA RDT Kernel SMP, Ardour
> > >> ardour-0.99.3-2.rhfc5.ccrma
> > >> <http://ccrma.stanford.edu/planetccrma/mirror/fedora/linux/planetccrma/5
> > >>/i386/repodata/repoview/ardour-0-0.99.3-2.rhfc5.ccrma.html>:, TAP Plugins
> > >> tap-plugins-0.7.0-1.rhfc5.ccrma
> > >> <http://ccrma.stanford.edu/planetccrma/mirror/fedora/linux/planetccrma/5
> > >>/i386/repodata/repoview/tap-plugins-0-0.7.0-1.rhfc5.ccrma.html>:
> > >>
> > >>
> > >> The project has 3 instances of the TAP Reverb on three busses. If I
> > >> reduce the input to the busses to zero, by soloing something for
> > >> example, or by lowering a fader to zero, the CPU load pretty quickly
> > >> maxes out.
> > >>
> > >> Is anyone else seeing this?
> > >>
> > >> Thanks in advance
> > >>
> > >> Ken
> > >
> > > Yes, definitely - many times. Which is a shame as I find its one of
> > > the few usable plugins around. Ultimately though, the denormal
> > > problems have forced me to stop using it.
> > >
> > > Nic
> >
> > Thaks Nic, I'll try the external TAP Reverb Editor to see if its any
> > different, but somhow I suspect the same algorithm will be in there. I
> > agree the TAP reverb is a pretty useful plugin. Maybe it's time to hack
> > some code.... ;-)  Ken
> 
> This link was on the list a while ago. I think Fernando provided it. 
> http://ardour.org/node/139
> 
> Most of it is info about what denormals are, but there are a couple of 
> workarounds. I havn't tried them, but the dithering one seems interesting, 
> where you introduce a background sound low enough to be not detected by the 
> human ear, but high enough to prevent the CPU going into denormal mode.

I did that with the freeverb plugins in Planet CCRMA and it solved the
denormal problem. For some reason the same trick did not work in
freqtweak, most probably I was not inserting the "noise" in the right
place. 

-- Fernando