Mon, 04 Dec 2000 Victor Iannello wrote:
> We have a periodic real time task that typically executes in about 40 us on
> a Pentium 266 MHz processor. We find that when the calculation results in
> one or more underflows, the execution time may increase to 60 us or more due
> to denormalization of the floating point numbers. We can avoid this by
> checking the operands of the floating point multiplication before performing
> the multiplication, but we would prefer that this is handled implicitly. Is
> there a compiler flag or other method that can be used to set small numbers
> to zero rather than the slower, gradual underflow?
I *think* I just got some information that indicates that the slowdown is due
to the denormalization being handled by an exception. In user space, you just
have to disable that exception to avoid the problem.
However, first, I might be mixing this up with other platforms (PPC also suffer
from denormalization overhead - well known problem among audio plugin
programmers), and second, I don't know how this would apply to kernel space
code. (IIRC, the exceptions are turned off, as they would do nasty things if
triggered from kernel context. Might be missing something, though.)
//David
..- M A I A -------------------------------------------------.
| Multimedia Application Integration Architecture |
| A Free/Open Source Plugin API for Professional Multimedia |
`----------------------> http://www.linuxaudiodev.com/maia -'
..- David Olofson -------------------------------------------.
| Audio Hacker - Open Source Advocate - Singer - Songwriter |
`--------------------------------------> [EMAIL PROTECTED] -'
-- [rtl] ---
To unsubscribe:
echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR
echo "unsubscribe rtl <Your_email>" | mail [EMAIL PROTECTED]
---
For more information on Real-Time Linux see:
http://www.rtlinux.org/rtlinux/
