forgot to cc the list ... forwarding. <br><br><div class="gmail_quote">---------- Forwarded message ----------<br>From: <b class="gmail_sendername">Justin Smith</b> <span dir="ltr"><<a href="mailto:justinsmith2009@gmail.com">justinsmith2009@gmail.com</a>></span><br>
Date: Thu, Jan 22, 2009 at 4:59 PM<br>Subject: Re: [Xorg-driver-geode] huge difference between hwaccel and no accel performance ??<br>To: Mart Raudsepp <<a href="mailto:mart.raudsepp@artecdesign.ee">mart.raudsepp@artecdesign.ee</a>><br>
<br><br>Hi Mart,<br><br>Thanks for your detailed response.<br><br>Another followup question I have is how to get the xv maximum image size from 1k x 1k to 2k x 2k. Is there a patch available already?<br>I have found couple of bugs, one in 2.11.0 and one in the last checkin that you have done - what is the best way to report bugs. <br>
<br>Thanks,<br><font color="#888888"><br>Justin</font><div><div></div><div class="Wj3C7c"><br><br><div class="gmail_quote">On Wed, Jan 21, 2009 at 4:02 AM, Mart Raudsepp <span dir="ltr"><<a href="mailto:mart.raudsepp@artecdesign.ee" target="_blank">mart.raudsepp@artecdesign.ee</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">Ühel kenal päeval, T, 2009-01-20 kell 17:09, kirjutas Justin Smith:<br>
<div><div></div><div>> Hi guys,<br>
><br>
> I am running some x11perf benchmarks and I see a huge performance<br>
> difference in the wrong direction. The no accel option is much faster<br>
> than the accelerated option.<br>
> Is the "No Accel" flag reversed, meaning "No Accel" is enabling the<br>
> acceleration?<br>
<br>
</div></div>No, it is not reversed. If NoAccel is set, then EXA isn't initialized<br>
and the acceleration architecture isn't really used.<br>
<br>
The most common reason of accelerated being slower than software is when<br>
the pixmap migration overhead exceeds the wins of acceleration. I'm not<br>
sure this is the case here, but that's what has plagued many EXA using<br>
drivers in the past.<br>
<br>
Do you have any concrete things that are slower and are actually a<br>
problem in real life modern applications?<br>
<br>
Note that it is also quite important to be using a recent xorg-server,<br>
as the EXA common code has received many improvements to it that improve<br>
the matter in xorg-server-1.4, even more in 1.5 and also more in what<br>
will become 1.6 (a RC2 or so should be out).<br>
<br>
<br>
However for information, and a possible call for contributions:<br>
<br>
One problem with the very latest xorg-server for geode seems to be the<br>
glyph cache found in trunk and server-1.6-branch. It is a huge win for<br>
many drivers, but it is my belief that its mask generation is causing it<br>
to be a major loss for geode at this time. This however needs some<br>
further verification.<br>
<br>
In concrete I have commit 54184110f6f3e5d7276d5431e739a4fcf0c3523e to<br>
xserver in mind. It adds a glyph cache and I believe it adds quite some<br>
Pict_A8 (alpha channel only) PictOpAdd operations to that hot path,<br>
which is currently not accelerated on geode, because the hardware does<br>
not have pure alpha pixmap support.<br>
A possible solution for that regression would be to make it possible for<br>
the driver to instruct the EXA glyph infrastructure to always use ARGB<br>
there, so we can accelerate the PictOpAdd composite operations, trading<br>
the overhead to four time larger caches (though it seems its also<br>
keeping ARGB caches as well anyway?). Another possibility could be to<br>
hack together a way to convert A8 to ARGB in the hardware side. Adam<br>
Jackson had some ideas in that area, but that could be a complementary<br>
to cover the cases where we can't make the pixmaps passed to not be A8<br>
in the first place - probably for other operations than glyph drawing.<br>
Once someone fixes that regression, we can reap the benefits of the<br>
glyph cache and end up with quicker glyph drawing than without glyph<br>
cache in <xorg-server-1.6, as for other drivers.<br>
All the other EXA changes in trunk and server-1-6-branch are probably a<br>
win, so possibly just glyph drawing would be slower.<br>
<br>
Note that some distributions have backported the EXA glyph changes and<br>
more to their xorg-server-1.5 as well. Gentoo and Fedora come to mind,<br>
iirc.<br>
<br>
<br>
Regards,<br>
Mart Raudsepp<br>
Artec Design LLC<br>
<br>
_______________________________________________<br>
Xorg-driver-geode mailing list<br>
<a href="mailto:Xorg-driver-geode@lists.x.org" target="_blank">Xorg-driver-geode@lists.x.org</a><br>
<a href="http://lists.x.org/mailman/listinfo/xorg-driver-geode" target="_blank">http://lists.x.org/mailman/listinfo/xorg-driver-geode</a><br>
</blockquote></div><br>
</div></div></div><br>