[Openchrome-users] Performance with OpenChrome+viafb / workaround DDC-issue

Philipp Wolter wolter
Tue Aug 31 07:35:09 PDT 2010


Hey there,

on my way getting Chrome 9 HC3 and DVI to work, I may found out some 
relations between Xorg performance, viafb and the OpenChrome driver.

Preconditions here:
Linux 2.6.35, Xorg 7.5-6, Openchrome 858-1 and UniChrome DRI 7.8.2-1
Hardware platform is a HP T5545; VX800/VX820 Chrome 9 HC3 (rev 11)

To give Xorg-"performance" a value, I'll use the Xorg memcopy throughput 
benchmark done while it starts up.

Observations:

1. Output:RGB, no viafb, no xorg.conf --> 418 MiB/s
2. Output:DVI, no viafb, no xorg.conf --> no picture in Xorg

My intention was to use DVI, because with RGB the picture wasn't that 
crisp on FullHD resolution. But out of the box, I just got a flickering 
and fast panning 640x480 at 60Hz output. My first solution was to load 
viafb on startup with kind of 'modprobe viafb viafb_active_dev=DVI 
viafb_refresh=50'. Due to DDC issues when using DVI, I additionally 
wrote a small xorg.conf with a correct modeline.

Observation:

3. Output:DVI, with viafb, with xorg.conf --> 88 MiB/s

I fiddled around with the Driver parameters in Xorg, but the performance 
stayed decreased. Also, the Xorg-logfiles of
a) DVI with viafb
b) DVI without viafb
showed no differences, as loading viafb has no relevant impact on Xorgs 
decisions. To get rid of viafb and have DVI working without DDC anyway, 
I had to set the following two driver-parameters in the xorg.conf 
'Device'-section:

	Option "PanelSize" "1920x1080"
	Option "ForcePanel" "true"

The important thing here is, that 'PanelSize' had no effect if not used 
in combination with 'ForcePanel'. Having a correct Modeline, and even 
additionally a 'Virtual' screensize in xorg.conf, was not enough. I had 
to tell the driver (!) the resolution, just as I told it viafb when 
using it. Last observation then:

4. Output:DVI, no viafb, with corrected xorg.conf --> 418 MiB/s

Conclusions:
1st: viafb might reduce performance of xorg massively (about factor 5)
2st: viafb so is no proper way to make DVI work
3st: DDC issues with DVI can better be solved with a proper xorg.conf

Suggestions:
a) mention the existence and effects of viafb in the documentation
b) give a hint in manpage that 'PanelSize' might be useless unless 
'ForcePanel' is set


If these observations are affirmed somehow (especially if performance is 
reduced with RGB and viafb too), I'd like to add these information to 
the manpage/documentation, if desired.


Best regards,
Philipp Wolter
-------------- n?chster Teil --------------
Ein Dateianhang mit Bin?rdaten wurde abgetrennt...
Dateiname   : Xorg_logs.gz
Dateityp    : application/gzip
Dateigr??e  : 24853 bytes
Beschreibung: nicht verf?gbar
URL         : http://wiki.openchrome.org/pipermail/openchrome-users/attachments/20100831/68934862/Xorg_logs-0001.bin
-------------- n?chster Teil --------------
Ein Dateianhang mit Bin?rdaten wurde abgetrennt...
Dateiname   : 20-openchromeDVI.conf_short.gz
Dateityp    : application/gzip
Dateigr??e  : 314 bytes
Beschreibung: nicht verf?gbar
URL         : http://wiki.openchrome.org/pipermail/openchrome-users/attachments/20100831/68934862/20-openchromeDVI.conf_short-0001.bin



More information about the Openchrome-users mailing list