Zdenek Kabelac ([EMAIL PROTECTED]):

> > It works without any external interrupt driver?  What's mga_vid? :)
> 
> Well mga_vid could be only used with Matrox card owners - which is
> rather minority I would say - so it's really not a general solution
> and avifile is not just MGA video player :)
> 
> So yes it works very well even without interrupt driver with just poor
> linux scheduler - but of course  tearing is unvoidable in this case...

  Tearing is totally avoidable without a kernel vsync device!  Like I
mentioned in my email, the Xv drivers for nVidia cards, i810 chips, and
ATI cards all double buffer and don't flip until the next retrace!

> > Yes, I am working to avoid judder at refres rates less than 85hz!
> > For example, many CRT projectors can run at 72hz which is great for
> > 24fps video.  To ensure a 3-3-3-3 pattern we must have a vsync
> > device.
> 
> I would say we need better kernel :)

  Why?  One with vsync devices for all video cards or one with lower
scheduling latency?

> Again - if you have such device - avifile will use - but if you don't
> have the current solution from avifile couldn't be made better in my
> eyes...  (if you will enable  export AVIPLAY_DEBUG=1   you will see
> sync messages and you will also see how the linux scheduler behaves -
> from time to time it will miss the timer wakeup point by more then
> 15ms)

  You don't need to convince me that we need lower scheduling latencies,
which is why I noted that you can do better using /dev/rtc/SCHED_FIFO.
The solution sucks but it's one way to do a little better without a
vsync device.

> > Even worse, consider playback of 60fps material at 85hz.  Judder is
> 
> Well I do not have 60fps material :) I do use 29.97Hz movies...  I'm
> currious what do you mean by 60fps material - do you use some high
> speed video cammeras :)  in my eyes there is no such thing like 60fps
> video - movie has 23.97,  PAL 25,  NTSC 29.97 

  If you deinterlace a 29.97fps NTSC stream you get 59.94fps video, each
field gets expanded to a frame.

> > Dude, my sentence begins with 'Given we need a vsync device'.  I say
> > here that we need BOTH.  Don't you agree?
> 
> Well not just 'BOTH' of us - almost every video app needs such device
> :)

  I meant both as in we need a vsync device and all Xv drivers should
double buffer.

> > > odd/even frame for TV (might be used the HZ 4byte sign bit) but
> > > generaly I would expect XFree driver would be doing this itself if
> > > it possible...
> > 
> >   Do what itself?  Tell you if you're on an odd or even field?
> 
> No I mean - I'll give two following images and video card would be
> doing proper interlacing...

  I'm sorry, I don't know what you mean.

> > I was just suggesting.  LKM gets too much traffic, and this is
> > definitely a video issue.  I think you'll find more interested
> > people there.
> 
> Well there is no point to disuccess this on Xpert - as long as Linus
> will be mostly ignoring any improvements in scheduling (I think I do
> have somewhere his sentence where he said that 1.5GHz CPU should be
> fast for every video application thus we do not need to modify kernel
> :) to fix latency problems

  Do you have a link to him saying that?  A vsync device would be
sufficient, and Alan Cox is apparently willing to put it in the kernel
if the code is written.

> I think that the need for vsync device is know for a long time and yet
> there has nothing (except for my completely ignored attemp)

  How much did you publish your attempt?  I just learned that your
mga_vid had been modified.  Regardless, there is interest now I think we
just need to write the code.  I'm currently modifying your mga_vid
module to be a little more generic and I'll try to code support for
i810.

-- 
Billy Biggs
[EMAIL PROTECTED]

_______________________________________________
Avifile mailing list
[EMAIL PROTECTED]
http://prak.org/mailman/listinfo/avifile

Reply via email to