[linux-audio-dev] P4 kernel patch/option against denormals

[derek holzer]

Hi all,

I have had non-stop CPU spikes and freezes related to denormal numbers 
on my Pentium 4 1.6GHz laptop ever since I switch to Linux, going from 
kernel 2.4.18 up to 2.6.4.

The main applications which seem to be affected are Freqtweak, Jamin and 
Pure Data, along with a handful of LADSPA plugins, which can sometimes 
make my Ardour sessions a real PITA. Does everyone develop on Celerons 
or AMDs these days?!? ;-)

I have pursued the problem in particular with Pure Data, but have been 
assured that there is already denormal handling in the program... It has 
gotten to the point where either I find a kernel-level solution for the 
problem or I buy a new laptop!

Tim Blechman reported on the PD list that his denormal problems only 
started when he switched from SuSE to Gentoo. My suspicion is that the 
SuSE kernel developers either cooked up a very good patch against 
denormals, or they are using a proprietary Intel compiler for their 
kernels which somehow alleviates the problem. For my own purposes, the 
former is more attractive than the latter, of course.

Tim also mentioned that some P4s have a Denormals-As-Zero switch to 
handle the problem [although I suspect that my CPU might be too old for 
this feature...]. Can anyone explain how such a DAZ switch can be tested 
for and toggled?

Can anybody shed some light on the situation? I would hate to have to 
ditch a decent computer over this issue.

best wishes,
Derek
-- 
derek holzer ::: http://www.umatic.nl
---Oblique Strategy # 127:
"Openly resist change"