From:	sandmann AT clio DOT rice DOT edu (Charles Sandmann)
Message-Id:	<10303101722.AA20067@clio.rice.edu>
Subject:	Re: Example uclock() code
To:	djgpp-workers AT delorie DOT com
Date:	Mon, 10 Mar 2003 11:22:03 -0600 (CST)
In-Reply-To:	<0eec01c2e70c$1f78cef0$0600000a@broadpark.no> from "Gisle Vanem" at Mar 10, 2003 02:51:17 PM
X-Mailer:	ELM [version 2.5 PL2]
Mime-Version:	1.0
Reply-To:	djgpp-workers AT delorie DOT com
Errors-To:	nobody AT delorie DOT com
X-Mailing-List:	djgpp-workers AT delorie DOT com
X-Unsubscribes-To:	listserv AT delorie DOT com

> > +      while ( (tics = _farnspeekl(0x46c)) == otics);
> 
> A debugger will propably hang indefinitely here. Maybe add 
> a __asm__("sti").

Debuggers should not (?), but you can't CTRL-C out of it - it
must complete before you get the signal.  

> Instead of counting the TSC over CALIBRATION_TICS, it's 
> better to *average* the TSC over CALIBRATION_TICS samples.
> This because of the jitter in the while-wait loops. It doesn't
> (as you said in original message) really wait N*ticks.

If the tics in the middle are used for both start and end,
you get the same answer (it averages out).  If you don't,
and get a new start tic each time - then it takes twice as
long elapsed (the current 5 TIC wait can take up to 6 tics
waiting for the first tic roll-over).  If you did 5 
calibrations of 1 TIC long it would typically take 10 tics.
It appears a single calibration of 10 tics would be more
accurate than averaging 5 samples, based on the error
distributions.

> BTW. the accuracy will depend on DOS-box setting of
> "Program | Advanced | Emulation of timekeeper" (or what
> ever it's called in English Windows). Reading the PIT is 
> not very accurate even when set on.

I can't find such a setting on Windows 2000.  The PIT is
completely unreliable and uncoordinated with the timer 
tic (usually... that's a long story) so I'm calibrating
using the timer tic alone (which is also flakey ...).

- Raw text -