On Tue, 2007-03-06 at 10:52 -0800, Eric Anholt wrote:
> On Tue, 2007-03-06 at 11:07 +0100, Michel Dänzer wrote:
> > I developed and tested this on i945 hardware. I've had reports of it
> > working on i915 as well but would be interested in hearing whether it
> > works on i830 and above.
Still haven't seen any reports about this unfortunately. I guess this
will be more likely to get tested with the branches merged. :) The new
functionality is disabled by default anyway.
> > The only significant outstanding issue I'm aware of is that moving a
> > page-flipped 3D window from being visible on pipe A only to being
> > visible on pipes A+B can result in a lockup if the flips on pipe B are
> > emitted 'too quickly'. I'm currently avoiding this by not syncing the
> > pages between pipes immediately in that situation, so it only seems to
> > happen with insane framerates such as glxgears', but this could probably
> > cause issues of its own with apps that render to the front buffer. It
> > doesn't seem to happen when moving from pipe B to pipes A+B, and the
> > command stream always waits for previous asynchronous flips to complete
> > before emitting a new flip, so I'm not sure what's up. Any ideas would
> > be appreciated, especially from the Intel folks.
> > Possibly related to this, it's been reported that there can be issues
> > with outputs that aren't handled quite correctly by the modesetting
> > driver (yet), e.g. LVDS on desktop boards.
I should have nailed these hopefully. Big thanks to Todd Merrill for
live testing feedback on IRC.
> > If you have issues with page flipping and xrandr 1.2 shows outputs that
> > shouldn't be there, try disabling them.
> Even better, file a bug with Xorg.0.log and the output of scanpci from
> libpciaccess, and we can blacklist the LVDS on your board hopefully.
> > If there are no objections, I'd like to merge these branches to the
> > respective master branches by the end of this week. Of course, if the
> > issues above can't be solved before the merge, I'll document them in the
> > intel.4 manpage.
There's nothing new to document, and I'm going to merge the branches
Earthling Michel Dänzer | http://tungstengraphics.com
Libre software enthusiast | Debian, X and DRI developer