P4 problemos

Sveiki,

Tiesiog norėjau nuraminti / perspėti visus tuos kurie jau turi arba planuoja artimiausiu metu įsigyti sistemas su P4 procesoriais - būsite (arba jau esate...) nemaloniai nustebinti sistemos silpnumu:(((

Viena iš problemų - didžiulis procesoriaus apkrovimas apdorojant tylius (artimus 0) audio failus. Problema yra žinoma, bet kol kas neaišku, kada ją visi softo gamintojai bandys taisyti:(

Skaitykite čia:
http://linux1723.dn.net/forum/Forum2/HTML/025691.html




Kad lengviau būtų - cituoju problemos detalų išaišknimą:

" tobybear
K-v-R Regular
Member # 629

posted 10 May 2002 19:56
--------------------------------------------------------------------------------
Okay, I did some research.

Problem is quite simple and has been posted already: if floating point values (that are commonly used in audio programs) get very very small, the Pentium CPU enters a special "high-precision mode", which takes much longer to calculate (eats up more cpu power).
Since almost all plugins have filters or other algorithms with feedback path, small numbers occur automatically sometime, the more complex the algorithm is, the more time the cpu spends in the so called "denormal" mode, which results in the feared "spikes".

There is currently no way to switch off this "denomal mode" behaviour, actually we developer knew about it as it was already present on the previous Pentium CPUs, only the P4 seems to have a higher threshold, so the old fixes will normally not work.

Solution:
It seems all developers have to update their plugins to "support" the new P4 (well, rather include a P4 workaround). It is actually quite simple to fix it, I picked this stuff from Laurent de Soras text and it seems to make my Deconstructor VSTi work on a P4 reliably:
*** P4 fix ***
On every location in the code where denormalization might occur, just add 1.0e-24 to the float value, then subtract it again, that should fix it.

Cheers

Toby
http://www.tobybear.de/ "



Dalius