On Friday 20 April 2007 01:33, James Richard Tyrer wrote:
> Paul Brook wrote:
> > [Taking offlist]
> >
> >> Would using RT speed up the graphics board?
> >
> > Unlikely. Graphics don't tend to have very demanding latency
> > requirements. The regular process scheduler is generally sufficient for
> > graphical tasks on desktop class hardware/OS. You only have to display a
> > frame every 10ms or so, so you've a lot of slack to play with.
>
> I think you better lop off the 0 if you are talking about response time.
>
> The sync interrupt comes at the start of the vertical sync pulse and not
> only the ISR but the task which it starts must finish before the end of
> the vertical blanking interval.  This is a little over 1 ms on a fast
> monitor.

*shrug*, that's still millions of bus cycles. I don't see any evidence that 
that polling the card adds significantly to interrupt latency.

In my experience you only need to care about interrupt latency at this level 
when e.g. you're trying to manually sample a 400kHz signal on a 20MHz CPU.  
In that case your whole system has to be designed with this in mind.

> So, it would be much better if the sync interrupt could be serviced with
> RT.

You really think that people are going to want/need to replace their OS kernel 
and run high-end hardware just to sync with a video refresh?

I really don't expect syncing with the it to be an issue. It's a relatively 
low-frequency event that doesn't have particularly stringent latency 
requirements. Any problems are going to come from elsewhere (eg. the kernel 
disabling interrupts altogether for long periods of time).

Look at it another way: You want to vertical sync so that you can program the 
card without the state changing in the middle of a frame. A typical example 
may be changing the palette in 8-bit modes. This is probably going to take 
several dozen transactions to accomplish, so an extra read on the start isn't 
going to make a whole lot of difference.

I remember writing applications that synced with the vertical refresh on 
low-end i386 hardware, and definitely didn't have to do anything special.
I find it totally implausible that any modern hardware needs any special 
treatment to achieve this.

AFAIK, every other graphics card manages with a single IRQ line. This is a 
fairly strong indication that the problem you're trying to solve simply 
doesn't exist.

Paul
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to