Nouveau - Jun 2009 - [Nouveau-cvs] xf86-video-nv: Branch 'master' - 2 commits

I just realised that this mode of tiling is quite different from the
one i observed, this one is linear inside the tile, not recursively
tiled.

Also, one optimisation that you can do without fixing the xserver
(which should be done anyway), is to set pread and pwrite to the
linear passthrough functions if none of the currently prepared pixmaps
are tiled.

Maarten.

On Wed, 2009-06-10 at 12:00 +0200, Maarten Maathuis
wrote:
> I just realised that this mode of tiling is quite different from the
> one i observed, this one is linear inside the tile, not recursively
> tiled.
Ah right, yeah, I should've mentioned that earlier.  The GPU deals with
all the weirdness, and it becomes regular old tiling from the CPU's
point of view.
> 
> Also, one optimisation that you can do without fixing the xserver
> (which should be done anyway), is to set pread and pwrite to the
> linear passthrough functions if none of the currently prepared pixmaps
> are tiled.
Yup, that thought occurred to me, didn't get around to it yet though.  I
took a bit of a look at what's required to fix things server-side too,
and it's not going to be even remotely straight-forward.  There's even
cases where the accessors get called on buffers that haven't had
setupWrap called on them etc.. It looks like quite a rewrite of the fb
layer (and I'm not even sure pixman's quite up to the task, but
didn't
spend too much time looking there yet) would be in order..

Ben.
> 
> Maarten.