Re: [avrdude-dev] Win 32 port

[Jan-Hinnerk Reichert]

On Tuesday 17 February 2004 16:23, Lou Cypher wrote:
> On 17 Feb 2004 at 13:46, Jan-Hinnerk Reichert wrote:
> > They are used in the CVS-code in usecPerfDelay() in "ppiwin.c".
> > They seem to work very stable. The only problem is, that they
> > might not be available on all hardware.
>
> Maybe it largely depends on the kind of hardware: maybe with NT
> embedded on a PC-104 form factor board, while they should be on any
> desktop system.

Sorry, if I was unclear. I was referring to the counter-hardware, not 
to the functions. I thought about 386 and 486-systems. AFAIK, there 
might be some old Pentium I without time-stamp-register, too.

> If the target is for older systems, it just can use a double
> methode for delays, checking first if that timer exists.

This is already done in avrdude. However, Martin replaced the usleep 
with something different in his version.

> If the system has no reliable delay, under a minimal amount of time
> (i.e. Windows 95 under a heavy load), then it can only be "slow".
>
> In case Sleep() isn't reliable on Windows 2000, maybe you could
> roll your own, using that timeGetTime() and a proper loop -- but if

Yes, that should work. However, I really want to know if the Sleep() 
fails in the way I expect; I'm just curious ;-)

For the resulting programming time it shouldn't make a difference how 
we implement the fallback. 

BTW: For devices with paged flash it doesn't really matter, if the 
delay is 10ms too long. However, it would quite hurt performance for 
programming Classic-AVR (AT90Sxxxx).

Cheers
  Jan-Hinnerk