New acceleration architecture

Eric Anholt eta at lclark.edu
Wed Jun 29 17:09:50 PDT 2005


On Wed, 2005-06-29 at 19:54 -0400, Adam Jackson wrote:
> On Wednesday 29 June 2005 19:07, Thomas Winischhofer wrote:
> > Adam Jackson wrote:
> > > Technically you don't need to care about doing it per-screen.  The
> > > server's single threaded, you're not going to get your state corrupted
> > > between the prepare and body calls.
> > >
> > > Not that that's pretty, or amenable to server rearchitecting.  I'd prefer
> > > that the body calls have a pScreen too.
> >
> > Basically, my concern wasn't saving the accel engine bits somewhere, but
> > getting a pointer to my command ringbuffer and the MMIO area. Can't do
> > without my pSiS (where the offset to the current command queue position
> > is stored)...
> 
> Hrm.  Yeah, that is major.  Anyone have an objection to adding ScreenPtrs to 
> the ExaAccelInfo functions?

There's no need for that in the Prepare hooks, because you've got the
PixmapPtr and can pull the ScreenPtr out of that.  I would probably just
keep passing the dst pixmap in to Copy, DoneCopy, Solid, and DoneSolid
as well.

-- 
Eric Anholt                                     eta at lclark.edu
http://people.freebsd.org/~anholt/              anholt at FreeBSD.org



More information about the xorg mailing list