RIP: Gateway 22" monitor

Well after at least 7yrs of service my Gateway 22" display died today. Off to shop for something new. I know I want something with at least 1920x1080 resolution, but not sure if I want to go larger than 23".

Such a tough decision...

Gnome MPlayer on Windows

One of the small tasks I've been working on is to get Gnome MPlayer to cross compile to Windows. A few people have asked for it, and while I admit it has not been high on my priority list, it has been something I've been interested in seeing how it was done.

The only way I have tested it so far is by cross compiling it on Fedora and then running it under Wine. I have to say that it actually works pretty decent. Notice in the screen shot that the VO is directx and the AO is dsound.

gnome-mplayer-win32 running under Wine on Fedora 14

To cross compile I installed the mingw32 packages on Fedora.
I checked out the gnome-mplayer source code
Ran configure like this  ./configure --host=i686-pc-mingw32
Ran make

I then ran the executable under wine. Probably the trickiest part was getting the gtk icons in the right spot

I ended up symlinking the system themes into the mingw32 GTK directories

cd /usr/i686-pc-mingw32/sys-root/mingw/share
ln -s /usr/share/icons/ .
ln -s /usr/share/themes/ .

After that was done, it seemed to work like normal, minus the dbus, gpod and musicbrainz integration.

So if you have a complete mingw32 environment setup on Windows, you should be able to cross compile gnome-mplayer and then point it to a windows mplayer executable and it should work.

Radeon 2d benchmarks

For this round of testing I'll be using kernel 2.6.28-rc5 and libdrm, mesa and xf86-drv-ati all from git as of Feb 16, 2011 8:30am GMT - 7. About 2 years ago I did some 2d tests of the Linux driver on the r600g and got some good results. The performance of gtkperf has went up and down over time, but generally has hovered around 33-38 seconds on my machine when run with gtkperf -c 500 -a.

The recent radeon patches have enabled linear aligned performance and according to the changelog


linear aligned is supposedly more performant, but more importantly, linear general only works on the CB without slices.  The texture blocks technically don't support    linear general although, I think linear general gets upgraded to linear aligned in the hw which is why it currently works.


This change required changes to the kernel module as well, and those changes appeared in the 2.6.38-rc5 kernel. At this point I have also enabled color tiling, but it is currently only 1d color tiling at this point. Alex Deucher has stated that he would like to have 2d color tiling working before color tiling is enabled by default on the r6xx chips.

So the results of the test are

GtkPerf 0.40 - Starting testing: Wed Feb 16 08:56:58 2011

GtkEntry - time:  0.09
GtkComboBox - time:  2.54
GtkComboBoxEntry - time:  2.13
GtkSpinButton - time:  0.79
GtkProgressBar - time:  0.61
GtkToggleButton - time:  0.48
GtkCheckButton - time:  0.35
GtkRadioButton - time:  0.56
GtkTextView - Add text - time:  3.48
GtkTextView - Scroll - time:  1.20
GtkDrawingArea - Lines - time:  5.16
GtkDrawingArea - Circles - time:  6.67
GtkDrawingArea - Text - time:  4.16
GtkDrawingArea - Pixbufs - time:  0.42
 ---
Total time: 28.63


I did this test a couple of times and always got results below 30 on my machine, which means that 2d performance has roughly increased by 10%. Please note that gtkperf is not a very good benchmark and so you can consider this test garbage if you want, but I have found the results useful at times.


gtkperf is very sensitive to other applications updating the screen, so be sure you have minimal applications running when performing this test.

r600g Benchmarks Round 3

In October I did some benchmarking of r600g performance on my machine. Since it has been about five months since that set of tests I decided to revisit the tests and see if any improvements are to be had.

As a recap for my hardware I'm using a Asus Silent Magic EAH3650 video card (rv635 chip). A Q6600 processor (2.4 Ghz x 4), 8GB of RAM.

The OS is Fedora 14 64bit, but with the 2.6.38-rc4 kernel and libdrm, mesa, and xf86-drv-ati all from git.

In the last round I had the following results

Testing results (October 2010)

• nexuiz-glx, normal settings, 1024x768 windowed, 30fps
• vdrift, 1024x768 windowed, 30-60fps
• openarena, 1024x768, windowed, 50-60fps
• etracer, 1024x768, 25-30fps

 
With the newer r600g code I have the following results (Febuary 2011)

• nexuiz-glx, normal settings, 1024x768 windowed, 40-45fps (improvement)
• vdrift, 1024x768 windowed, 60fps at the main screen, 15-30fps during the game. The visual quality has improved quite a bit even though the frame rate is a little lower
• openarena, 1024x768, windowed, 50-60fps (no change)
• etracer, 1024x768, 50-60fps (improvement)

Also, since the October test Savage2 and Heroes of Neworth have started working correctly. So more or less improvements all around. When you see the r600g report 60fps in a test it means the card is able to draw at the monitor refresh rate, which is good.

Heroes of Neworth on r600g

As a follow up to my Savage2 results. I decided to try Heroes of Neworth.

The game is a little slow on my 3650 card with the default settings, but I think I can lower some of the quality settings and get it working fine.

Below are some screen shots running the tutorial

Training Screen

Tutorial Screen

So the progress on the driver is really coming along.

Savage 2 on R600g

Thanks to a patch from Henri Verbeet, the r600g driver in Mesa is now playing Savage2 as of Feb 7, 2011. The only thing that needs to be disabled is shadows. Everything else seems to look very nice and run very fast even on my rv635 (3650) video card.

I'm using
Kernel 2.6.38-rc3
libdrm - git
mesa - git
xf86-drv-ati - git
Based on Fedora 14

Savage2 on r600g screenshot