[PD] crazy weird stuff happening.

derek holzer derek at x-i.net
Mon Nov 1 01:38:03 CET 2004


Hello Tom,

don't lose the faith! There's an explaination for everything!

Tom Dunstan wrote:

> i just discovered little problem that i cannot figure out at all. 
> it runs fine for about 20 minutes 
> then all of a sudden the whole system goes into a melt down. 

 > p4 2.6 (800fsb)

OK. Since you are running Linux on a Pentium 4, I wouldn't rule out the 
possibility that this is a denormal numbers related problem. Usually, 
however, you see a huge spike in CPU usage with this, however...

Try running your patch on another CPU, OS or both. Does it run fine on 
OSX? What about with Linux on an Athlon? How about windoze?

Search the PD list archives for more on denormals, but basically the 
solution that is offered by other listers is to check out the CVS 
version of 0.37 or 0.37_devel and configure it with

-enable-optimize=pentium4 --enable-simd

You'll also need to do this with every external. These two config 
options, IIRC, insert the following flags into the makefile, which you 
will need to use with all your externals:

-march=pentium4 -mfpmath=sse

[correct me here, gang!]

Tim Blechmann has done some work with P4 optimizations by setting a 
Float To Zero switch in the devel branch. I didn't have any luck with 
it, but my P4 is too old to support this switch...

Another suggestion which came my way was to use something like 
KCacheGrind to find out what is eating up all your memory or CPU cycles. 
KCachegrind uses Calltree, and is part of KDE 3.3 (package kdesdk). See:
http://kcachegrind.sourceforge.net/

This will also help you figure out if you are simply running out of RAM, 
as Andy suggests ;-)

Or this:

Tim Blechmann wrote:

> thomas musil told me about the approach he used ... adding noise~ *~1e-14 to every feedback loop, filter ...

good luck,
d.

-- 
derek holzer ::: http://www.umatic.nl
---Oblique Strategy # 200:
"How would you explain this to your parents?"




More information about the Pd-list mailing list