[cairo] git "performance" patchset # 1 available + a few questions

[Vladimir Vukicevic]

On 6/22/06, David Turner <dturner at nds.com> wrote:
> - on my home machine (a Pentium M latop with an Intel 855G graphics
> "card"), the "cairogears" benchmark indicates that the Cairo image backend is about
> twice faster than the XRender one (except for the TEXT2 benchmark).
>
>   I'm a bit puzzled, but is this normal (using Ubuntu Dapper with latest
> X.Org server package) ?

Yep, this is pretty much normal for any currently shipping X server
(with the exception of Xgl).  Most, if not all RENDER operations are
done in software, requiring a read back into system memory of any data
that was placed in video memory for one; that doesn't explain all the
slowness, but that was one of the reasons given to me.  I saw much the
same results with my cairo-bench tool
(http://rig.vlad1.com/~vladimir/git/cairo-bench.git).  This situation
will supposedly improve as drivers switch over to supporting EXA;
however, I have no hardware on any of my machines for which there
exist EXA drivers (as of a few months ago, at least), so I haven't
been able to check that.

The cairo xlib surface goes to a lot of trouble to avoid calling into
RENDER in some cases (composite in particular); it would be nice if
those were unnecessary, because at some point those core X calls might
be slower than an actual XRenderComposite call.  It would be good to
be able to identify whether RENDER is really accelerated or not on a
given server.

   - Vlad