delwrite/vd question

[Miller Puckette]

Hi all,

There is a recurring problem in Pd whose symptom is that CPU usage
sometimes shoots upward unexpectedly.  The cause is usually
arithmetic underflows in tilde objects.  I've tried to find ways
to prevent this but it looks like I haven't got it quite right yet.

For example, in a delay feedback loop, if there's input followed by
a stream of zeros, eventually the recirculating values drop below 1E-38
or so.  Each time a calculation does this the processor receives a 
hardware interrupt!

I believe the I386 offers a way to disable this interrupt but I
haven't been able to find out how to do it in Linux.  Does anyone
on the list know about this???

My workaround has been to fix delwrite~ to bash its input to zero whenever
it drops below 1E-20.  Apparently this doesn't work in every case.  If
anyone would like to e-mail me an offending patch I'll be happy to try
to figure out where the underflows are occurring and fix Pd so they will
stop.

cheers
Miller

Does anyone on the list know if 
On Tue, Jun 13, 2000 at 04:16:48PM +0200, sfogar at libero.it wrote:
> Hi everybody,
> 
> I'm developing my new installation with pd under Linux but I notice a 
> strange behaviour.
> 
> I'll try to explain it...
> 
> When, in patchers which use delwrite~+vd~, there is no audio flowing, 
> the cpu utilization increases. 
> 
> For example, you have a patcher which reads a sample, plays it through 
> a chorus or a reverb, the audio goes to a dac~, when the sample is 
> played, the cpu utilization is at a certain level, when the sample is 
> not played, the cpu utilization increases.
> 
> Is this normal ? Am I wrong ? Is there a possible workaround ?
> 
> I'm using an RME Hammerfall 9636.
> 
> Regards
> 
> Alessandro Fogar
> 
> 
>