>From: "Michael Kabot" <[EMAIL PROTECTED]>
>The runtime of the floating point section varies from less than 50 us
> to 500 us in rare occasions. Its unbelievable but that's what I can see on
my
> scope. (And in the error detection of the software :-)
If by "runtime," you mean the time actually to do the calculations as
opposed
to other delays such as interrupt latencies, and you are sure the calcs are
not pre-empted by a handler and/or higher priority task, I would search
elsewhere from a FPU switch. For example, we've found that numerical
underflows were the cause for a sporadic increase in thread execution times.
When an underflow occurs in x86 hardware, a process calls "denormalization"
occurs that increases the bits assigned to the exponent. This is all handled
by an exception, which eats up clock cycles. To avoid this, test the
operands before the FP multiplication.
Victor Iannello
Synchrony Inc.
-- [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/
