X-Authentication-Warning:	delorie.com: mail set sender to djgpp-workers-bounces using -f
Date:	Wed, 28 Feb 2007 22:28:31 -0700
From:	Brian Inglis <Brian DOT Inglis AT SystematicSw DOT ab DOT ca>
Subject:	Re: Final version of gettimeofday() patch
In-reply-to:
	<OF8D61EBF3 DOT 4C147636-ON87257291 DOT 0003A12D-87257291 DOT 0003C9EF AT seagate DOT com>
To:	DJGPP-workers <DJGPP-workers AT delorie DOT com>
Message-id:	<u7mcu2tmm1p9q46ame58tas2bj73vm163g@4ax.com>
Organization:	Systematic Software
MIME-version:	1.0
X-Mailer:	Forte Agent 1.93/32.576 English (American)
References:	<200703010015 DOT l210FOi6004906 AT envy DOT delorie DOT com>
	<OF8D61EBF3 DOT 4C147636-ON87257291 DOT 0003A12D-87257291 DOT 0003C9EF AT seagate DOT com>
Reply-To:	djgpp-workers AT delorie DOT com

On Wed, 28 Feb 2007 17:41:23 -0700, Gordon DOT Schumacher AT seagate DOT com wrote:

>DJ Delorie <dj AT delorie DOT com> wrote on 02/28/2007 05:15:24 PM:
>
># > +      double secs;
>#
># We need to avoid floating point in libc, whenever possible.  Could
># this be done with suitably scaled integer math?  CLOCKS_PER_SEC is
># scaled by five to make it an integer.
>
>Hmm... let me ponder that.
>
>(I'm assuming this is to avoid issues on systems without FPUs?)
>
>The goal there is to be as precise as possible so as to avoid clock
>skew.  As you can see that frequency is not a very nice number...
>but perhaps working with a long long int we'd have enough precision?
>
>I'll have to think about that.

Can someone please check if this is correct and feasible. 
The exact clock rate is 7*5*3*10^6/11/2^19 == 105*10^6/88/2^16 Hz. 
Could we do:

typedef unsigned long long int clock_t
#define CLOCKS_PER_SEC 1000000
clock() == (rawclock() << 16)*88/105

This would give us a pretty good integer approximation but with unsigned
int we get a period of just less than 01:11:35 before clock() overflows.

-- 
Thanks. Take care, Brian Inglis 	Calgary, Alberta, Canada

- Raw text -