[PATCH] damage: Only track extents where possible

Adam Jackson ajax at redhat.com
Tue Mar 3 11:46:57 PST 2015


On Thu, 2015-02-26 at 22:01 +0000, Chris Wilson wrote:
> On Thu, Feb 26, 2015 at 02:58:45PM -0500, Adam Jackson wrote:
> > So I guess the question is whether automatic windows should take the
> > bounding box snap too, and I guess whichever method makes x11perf look
> > better when 'xcompmgr -a' is good enough for me.
> 
> x11perf -aa10text certainly prefers extents-only tracking here. I don't
> imagine any other test would regress, but I can check for completeness.

That's a fine bellweather for core rendering performance, I'm convinced.

> As for DamageReportNonEmpty, exa has to pessimise that pixel data is slow
> to readback and so should track rectangles. xwayland/glamor already
> ignore the rectangles. composite however doesn't know whether it will be
> faster to ignore rectangles (because the copy is hwaccel) or to track
> rectangles... Which is where we can make a decision based on xvfb +
> xcompgr -a (and also check with the major ddx).

The composite case I'm worried about is firefox, where the animations
are a spinner and a banner ad and they're not near each other;
regardless of hardware acceleration, at some point two blits are cheaper
than one if the one moves much more data.

- ajax



More information about the xorg-devel mailing list