EXA and migration [was: shrink xrender featureset]

[Michel Dänzer]

On Tue, 2009-01-06 at 22:08 +0100, Clemens Eisserer wrote:
> Hi Michael,

Ahem. ;)

> > I can definitely see how diagonal lines would cause migration ping-pong,
> > but for gradients without blending, as of xserver 1.6 they should just
> > get rendered by the CPU to the system memory copy without any previous
> > migration necessary. If you're not seeing that, please provide more
> > information.
> 
> Is this also the case when a mask is used for composition?

Hmm, I think you've just made me realize a bug in
ExaCheckComposite(). :( The way it's currently implemented, mask or no
mask doesn't matter, only whether the composite operation itself reads
from the destination does. However, of course with a mask the result may
depend on the previous destination contents regardless, so the migration
would indeed be necessary in that case. Maybe this is behind some of the
EXA rendering corruption bug reports I haven't been able to reproduce...

> If this is the case, what would be the drawbacks to use a temporary
> pixmap to avoid migrating out of VRAM?

It's hard to say without trying it.


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