Tim Blechmann wrote:
pd used to have problems with denormals until version 0.36 ... since 0.37, denormal handling have been implemented ... the devel_0_37 branch is using assembler coded simd instruction that are not that critical concerning denormal numbers ...
hmmmmm.... my last year of using PD on a P4 had been *nothing* but denormal problems! I can pretty much guarentee that any complex patch I make will crash on my 1.6 Ghz laptop and run flawlessly on my 1.8Ghz AMD shuttle.
In fact, this problem is not only with PD, but also with many other core Linux sound apps, such as Jamin, and a boatload of LADSPA plugins. It's gotten to the point where my options are either to get a new Athlon-based laptop or to find a kernel-patch which deals with the problme globally instead of waiting around for the application developers [who must be working on Celerons or AMDs!] to fix it.
on some p4 cpus it's even possible to switch on DAZ which causes the cpu to handle denormals as zeros in simd instructions ...
Can you show me some documentation on this? I'd like to check this out before I have to go shopping for a new computer ;-)
i'm not sure how different operating systems handle this, but i _suppose_ this explains problems i had when switching from suse to gentoo and got stuck in denormal crashes ...
I think we chatted about this once... Debian, Red Hat and Gnetoo all give me the same denormal CPU leaps and freezes. SuSE *must* be doing something special to their kernel. So perhaps it's available as a kernel patch somewhere. Either that, or they compile everything with a proprietary Intel compiler. Using an Intel compiler on PD alone didn't really give me any big gains, so it must be done on the level of the kernel.
anyway, a machine that is not affected by the denormal problem is not as important as it was about one year ago if you are on pd ...
*Cough!* Then it won't be so important when I send you the bill for my next, LinuxPD-friendly laptop ;-)
d.