Re: [Xorg-driver-geode] Update for geode driver

2010-08-16 Thread Mart Raudsepp
On E, 2010-08-16 at 02:42 -0400, Bernie Innocenti wrote:
 El Mon, 16-08-2010 a las 07:43 +0300, Mart Raudsepp escribió:
  I don't suggest using NoAccel though, as it loses XVideo and many other
  non-acceleration (read non-deceleration) related things too.
 
 That was just to test whether a particular ExA performance problem would
 disappear. We have an application called Turtle Art that renders a lot
 of rounded shapes using Cairo:
 
   http://wiki.sugarlabs.org/go/File:TAclock4.png
 
 When you drag around a dozen of those bricks on the XO-1 (Geode LX), you
 get one frame every two seconds. On the XO-1.5 (VIA based), you get
 fluid 10-20 fps motion. The difference is way too big to be justified by
 the difference in performance of the CPU or the graphics processor.
 
 Any idea?
 
  Have you tested rotation (and rotation + Xv) with current GIT code on
  the older xserver-1.6 found in FC-11 and XO-1 builds?
 
 Yes, it worked perfectly now. 90 degrees modes were broken in 2.11.8.

What happens after you go back to normal (no) rotation? That is broken
for me per
http://lists.x.org/archives/xorg-driver-geode/2010-August/000995.html

Though we are looking into that already right now, maybe you'd like to
come to our IRC again ;)


Regards,
Mart

___
Devel mailing list
Devel@lists.laptop.org
http://lists.laptop.org/listinfo/devel


Re: [Xorg-driver-geode] Update for geode driver

2010-08-15 Thread Mart Raudsepp
On E, 2010-08-16 at 08:41 +0800, Huang, FrankR wrote:
 Bernie,
 
   I am very glad to get the feedback from you. Still some patches need to 
 be committed later(although no new bug is reported still).
   Hope you can report anything you think incorrect or abnormal after 
 using that new driver.
   For NoAccel segfault, I am not sure if you have tried Mart's patch in 
 the below link:
   https://bugs.freedesktop.org/attachment.cgi?id=35326
   I have given a try and it can nicely solve this bug. It is caused by 
 the zero pointer use. Could you give a try? I think Mart know more about that.
 
 Mart,
 
   Can you make this issue more clear for Bernie?

I can't reproduce the issue with no display until VT switch anymore, so
clearly I was mistaken or it's been solved by a different commit
meanwhile, or it just doesn't happen all that often and the idea of
crash being better than confusing picture for most people doesn't hold
anymore. So I've pushed this trivial segfault fix to GIT now, to be part
of 2.11.9.

I don't suggest using NoAccel though, as it loses XVideo and many other
non-acceleration (read non-deceleration) related things too.

Have you tested rotation (and rotation + Xv) with current GIT code on
the older xserver-1.6 found in FC-11 and XO-1 builds?

Regards,
Mart


 Thanks,
 Frank
 
  -Original Message-
  From: Bernie Innocenti [mailto:ber...@codewiz.org]
  Sent: 2010年8月15日 10:13
  To: Huang, FrankR
  Cc: OLPC Devel
  Subject: Re: Update for geode driver
  
  El Fri, 13-08-2010 a las 10:35 +0800, Huang, FrankR escribió:
   Bernie,
  
 You can git clone the lastest code from freedesktop for geode driver
   to use. The most outstanding rendering bug has been fixed. And some
   performance patch has been applied.
  
  I tested the driver on the XO-1 and couldn't spot any rendering bug
  neither in Sugar nor in Gnome, with full acceleration enabled and no
  quirks needed in xorg.conf.
  
  However, a long-standing bug is still there: setting Option NoAccel
  1 makes the driver segfault during startup.
  
  Thank you very much for taking the time to do this neat work, Frank!
  (pls forward my notes the geode list, I'm not subscribed)
  
  --
 // Bernie Innocenti - http://codewiz.org/
   \X/  Sugar Labs   - http://sugarlabs.org/
  
 
 ___
 Xorg-driver-geode mailing list
 xorg-driver-ge...@lists.x.org
 http://lists.x.org/mailman/listinfo/xorg-driver-geode

___
Devel mailing list
Devel@lists.laptop.org
http://lists.laptop.org/listinfo/devel


Re: New XO-1 10.1.2 build 300

2010-07-09 Thread Mart Raudsepp
On N, 2010-07-08 at 16:07 -0400, Martin Langhoff wrote:
 On Thu, Jul 8, 2010 at 3:48 PM, Daniel Drake d...@laptop.org wrote:
  It's particularly significant because (when the release goes final) it
  means we have synchronized software between XO-1 and XO-1.5.
 
 Are we aiming for a synchronised 10.1.2 for XO-1 and XO-1.5?
 
 If this is just for xo1, 10.1.1-xo1 might be a better moniker...
 
 [ I am still eyeing some bugs -- ie: screen rotation -- for a 10.1.2 ... ]

Please work on Xorg (driver) bugs and tasks together with the wider
community of Geode graphics. For example AMD is already looking into
screen rotation problems with xorg-server-1.7 and newer, and doing it
more cleanly in general for all versions.
We are not aware of any specific issues with the 1.6 server version used
in Fedora Core 11 in regards to screen rotation. So, is there?

http://www.x.org/wiki/GeodeDriver


Thank you,
Mart Raudsepp
xf86-video-geode developer
Mail: l...@gentoo.org

___
Devel mailing list
Devel@lists.laptop.org
http://lists.laptop.org/listinfo/devel


Cairo's slow rendering on XO-1 (was: olpcgames - mainloop bug help)

2010-06-27 Thread Mart Raudsepp
On R, 2010-06-25 at 22:49 -0400, Wade Brainerd wrote:
 That said I see in a later post that you switched to GTK/Cairo which
 is great - so long as you don't get too bogged down by Cairo's slow
 rendering on XO-1.

Any such slowness should be fixed at the root - the xf86-video-geode
driver - instead of avoiding GTK/cairo specifically due to that.
Unless heavily targeting existing system roll-outs without upgrading of
the system, I suppose.


While we do already have our work ahead of us in optimizing the standard
cairo benchmarks (found in the cairo-traces repository), concrete cases
that would benefit the applications used specifically on the XO-1 would
be appreciated. Fixing things that can potentially make 1.5 million kids
happier can be a great motivator.

Ideally this would be presented as a cairo-trace recording as per
http://cairographics.org/FAQ/#profiling but I do have a OLPC XO-1 and
other Geode LX systems to generate that myself if that's too complex
(currently that cairo-trace is only in the development version, and not
installed outside build tree), given a description of what
program/activity exactly is way too slow and while doing what exactly in
that program.

Such reports can be sent to freedesktop bugzilla under the Driver/geode
component of xorg product, or the xorg-driver-geode mailing list over at
http://lists.x.org/mailman/listinfo/xorg-driver-geode
Personally to me is also fine, and maybe also good as I can apply some
filtering and identify the slow routines beforehand.
We also hang out on FreeNode IRC channel #geode
More information on http://wiki.x.org/wiki/GeodeDriver


Currently we (primarily two AMD employees, not so much me) are
concentrating on fixing some of the awful bugs (many of which get
triggered only by a newer xorg-server version), such as misrendering
with HwAccel and rotation issues. After those are hopefully fixed soon,
some attention will probably start to go on hardware acceleration
performance, as the current situation is indeed rather sad:
http://people.freedesktop.org/~leio/geode/perf/


-- 
Mart Raudsepp
Gentoo Developer, xf86-video-geode contributor
Mail: l...@gentoo.org

___
Devel mailing list
Devel@lists.laptop.org
http://lists.laptop.org/listinfo/devel