x1250 horizontal tearing problems

[Michel Dänzer]

On Wed, 2008-04-30 at 14:39 -0400, Alex Deucher wrote: 
> On Wed, Apr 30, 2008 at 3:06 AM, Michel Dänzer
> <michel at tungstengraphics.com> wrote:
> > On Tue, 2008-04-29 at 16:52 -0400, Alex Deucher wrote:
> >  > On Tue, Apr 29, 2008 at 4:34 PM, Alex Rades <alerades at gmail.com> wrote:
> >  > > Hi,
> >  > > when playing videos (either in xv or plain x11) on my x1250, I always see
> >  > > horizontal (not diagonal, which are now fixed) tearing problems. They seem
> >  > > related to vertical sync problems. Do you have suggestions?
> >  >
> >  > We need sync to vblank support for textured video to properly deal
> >  > with that.  this untested hack may help, but it's not optimal:
> >  > http://www.botchco.com/alex/xorg/texvid_wait_vsync.diff
> >
> >  Good to see this getting tackled. Here's what I think is missing:
> >
> >       * Set up the CRTC*_GUI_TRIG_VLINE register such that it waits for
> >         scanout to be outside of the destination vertical range.
> 
> yeah, makes sense.
> 
> >       * Only wait if the window isn't redirected (backing pixmap is the
> >         screen pixmap)
> 
> yeah, as you said, any rendering to the front buffer should wait for
> vblank.  

Technically not for vblank, but for scanout to be outside of the
affected vertical screen range.

> I may play around with it a bit if I have time, but unfortunately, I
> seem to be unable to notice tearing generally.  Do you have any good
> tips or content that would make it easier to notice?

Generally, I find it most noticeable with horizontal animation or with
large solid areas that continuously change colour.

> The other issue is that the current IB scheme doesn't really lend
> itself to this.  Ideally we'd queue up everything and then pre-pend a
> wait_until vblank when we submit the IB.

I don't think one wait per IB is sufficient anyway, as any given IB will
likely have both on- and offscreen operations. It could be tricky to
find a good granularity for the waits.

> Additionally, we have the issue of multiple crtcs for regular
> rendering as well. In that case, we really need shatter.

Right, or again pick the CRTC with larger visibility and/or let the user
choose somehow.


-- 
Earthling Michel Dänzer           |          http://tungstengraphics.com
Libre software enthusiast         |          Debian, X and DRI developer