LinHES Forums
http://forums.linhes.org/

XvMC, openchrome and bob
http://forums.linhes.org/viewtopic.php?f=6&t=18151		 Page 1 of 1
Author:  tuatara [ Mon Apr 07, 2008 6:26 pm ]
Post subject:  XvMC, openchrome and bob
Just wondering if anyone else is running KM on an EPIA box - specifically one with a CLE266 (such as MII1x000).

I've just installed R5F27 (forced on me due to HDD crash of my previous R5E50), and I'm not sure openchrome is working right yet.
In R5E50, with unichrome, cpu load was 20-40% during playback and I used the "bob" deinterlace filter to smooth motion.

In R5F27, with openchrome, cpu load is 50%+ during playback. Enabling "bob" makes the picture stutter and OSD flicker like crazy.
Its more pronounced with playback of DVB-S SD - same resolution but double the bitrate. cpu load during playback is 70-80%.

It all works ok with bob turned off, except for the high CPU load and awful motion artifacts.

I've checked myth and X logs and dont see any errors.

Any ideas would be most welcome.
Paul.
Author:  vabiro [ Mon Apr 07, 2008 10:00 pm ]
Post subject: 
Paul,

Have you selected "Via XvMC" in the settings|TV settings|Playback MPEG2 Decoder?

Also, check in your /var/log/Xorg.0.log to see if XvMC and DRI are actually working.

Hope this helps.
Victor
Author:  tuatara [ Tue Apr 08, 2008 1:57 am ]
Post subject:  Checking...
Hi Victor,
Thanks for getting back to me. I have checked those, but it was good to go back and confirm.

a) Yes - Via XvMC is selected in TV playback.

b) in the "Modules" section of xorg.conf :
Code:
Load  "dri" # This shouldn't be available choice if user has selected dri

in the "Device" section:
Code:
#       Driver      "via" # commented out to replace with openchrome below
        Driver      "openchrome"

Checking the X log file.
Code:
> grep -i dri Xorg.0.log
....
(II) LoadModule: "dri"
(II) Loading /usr/lib/xorg/modules/extensions/libdri.so
...
(II) Loading extension XFree86-DRI
(==) CHROME(0): DRI IRQ will be enabled if DRI is enabled.
(**) CHROME(0): AGP DMA will be enabled if DRI is enabled.
(==) CHROME(0): PCI DMA will be used for XV image transfer if DRI is enabled.
(==) CHROME(0): Will not impose a limit on video RAM reserved for DRI.
(II) CHROME(0): [DRI] installation complete
Fulfilled via DRI at 6638400
(II) GLX: Initialized DRI GL provider for screen 0

You can see dri loading, and OpenCrome seems to be using it.

Thanks
Paul.
Author:  mhawkshaw [ Tue Apr 08, 2008 12:08 pm ]
Post subject: 
I had a problem after installing R5F27 on a Via EPIA SP13000 - blank screen after reboot. The instructions in this thread fixed it for me, and video playback works fine: http://mysettopbox.tv/phpBB2/viewtopic.php?t=16408&start=15 might help you?
Author:  tuatara [ Tue Apr 08, 2008 4:07 pm ]
Post subject: 
mhawkshaw wrote:
I had a problem after installing R5F27 on a Via EPIA SP13000 - blank screen after reboot. The instructions in this thread fixed it for me, and video playback works fine: http://mysettopbox.tv/phpBB2/viewtopic.php?t=16408&start=15 might help you?


Yeah - I had that little beauty too, and that article was a gem. I have a fully working system except for this annoying problem with being unable to enable the bob deinterlacer, which makes any onscreen motion hard to watch.

You use a different northbridge to me. I've got the CLE266 whereas you have the CN400. Still - you probably switched to openchrome per the article. Are you seeing any change in CPU load? Do you use bob?

Thanks for the response.
Paul.
Author:  tjc [ Tue Apr 08, 2008 6:11 pm ]
Post subject: 
Bob2x _does_ double the frame rate required, so if your system isn't fast enough it may not work for you. The trick it uses is that rather than than trying to merge the two frames as other deinterlacers do, it just shows them both in succession, and lets persistence of vision and phosphors do the rest.

BTW - On this topic saw a great investigative science show the other day that actually debunked (almost unintentionally) the so called "flying rods" phenomena. It turns out to be a camera artifact from really fast flying birds and insects that move through the field of view with wing beats faster than the cameras frame rate. The researchers were _so_ depressed that the explanation was so mundane but the evidence was just inescapable. ;-) Somewhere The Amazing Randi is smiling... :D
Author:  alien [ Wed Apr 09, 2008 1:23 am ]
Post subject: 
Bob Stuttering/Shaking: Have you set an overscan in the mythtv settings (not the nvidia, but in the mythtv setup)? I found this caused really bad shaking. I ended up abandoning bob and going with kernel deinterlacing because I wanted the overscan. I think onefield deint works with XvMC if you absolutely need XvMC.

XvMC: You might try running with it off and kernel deint. The savings with the deinterlacer might be enough to run without XvMC. I find things work better without XvMC if there is enough CPU. And besides.... the OSD looks better :wink:
Author:  mhawkshaw [ Wed Apr 09, 2008 2:27 pm ]
Post subject: 
tuatara wrote:
You use a different northbridge to me. I've got the CLE266 whereas you have the CN400. Still - you probably switched to openchrome per the article. Are you seeing any change in CPU load? Do you use bob?


Paul, I did switch to Openchrome as per that article. I didn't notice any change in CPU load, but then I wasn't using bob. To be honest, I have a working R5F1 machine, but I wanted to try installing R5F27 on a different hard disk to see how it would go. I didn't use it that long, and I went back to my original hard disk (R5F1) as it had all my recordings on. As there will be another release in a few weeks or so, I thought I'd wait until then before trying the upgrade process again. Sorry I can't be of more help!
Author:  tuatara [ Wed Apr 09, 2008 6:04 pm ]
Post subject: 
alien wrote:
Bob Stuttering/Shaking: Have you set an overscan in the mythtv settings (not the nvidia, but in the mythtv setup)? I found this caused really bad shaking. I ended up abandoning bob and going with kernel deinterlacing because I wanted the overscan. I think onefield deint works with XvMC if you absolutely need XvMC.

XvMC: You might try running with it off and kernel deint. The savings with the deinterlacer might be enough to run without XvMC. I find things work better without XvMC if there is enough CPU. And besides.... the OSD looks better :wink:


Hi Alien,

I'm using an Via EPIA mobo with the CLE266 rather than NVidia. To answer your questions:
a) I have not set overscan in the mythtv settings.
b) I dont have the processing power to run without XvMC. Right now its actually at 60-70% playing back DVB SD programs using XvMC.

(NOTE: this is made much worse by the DVB/EIT grab that seems to be happening very often. When it does, mysql and mythbackend processor util goes to 20-30% each, and the front end stutters. You can delay it for up to 2 hours after the backend last recorded, but thats it).

I suspect that Openchrome is not doing as good a job as Unichrome.
Thanks for your feedback through.

Cheers.
Paul
Author:  tuatara [ Wed Apr 09, 2008 6:10 pm ]
Post subject: 
Hi,

I wish I didn't have to switch, but hard drive failure forced the issue. Otherwise I'd be hanging out for the imminent next release.

R5D1 wasn't as stable as I liked (crashed every few days), but otherwise worked very well.

Thanks for the feedback.

Cheers.
Paul.

mhawkshaw wrote:
tuatara wrote:
You use a different northbridge to me. I've got the CLE266 whereas you have the CN400. Still - you probably switched to openchrome per the article. Are you seeing any change in CPU load? Do you use bob?


Paul, I did switch to Openchrome as per that article. I didn't notice any change in CPU load, but then I wasn't using bob. To be honest, I have a working R5F1 machine, but I wanted to try installing R5F27 on a different hard disk to see how it would go. I didn't use it that long, and I went back to my original hard disk (R5F1) as it had all my recordings on. As there will be another release in a few weeks or so, I thought I'd wait until then before trying the upgrade process again. Sorry I can't be of more help!
Author:  cecil [ Wed Apr 09, 2008 8:46 pm ]
Post subject: 
Anyone want to test Openchrome in R5.5? If so, PM me.
Author:  tuatara [ Thu Apr 10, 2008 2:38 pm ]
Post subject:  FYI - openchrome bug logged
I've logged an openchrome bug on this one.

http://www.openchrome.org/trac/ticket/170#preview

Perhaps they can shed some light.
Author:  tuatara [ Thu Apr 10, 2008 2:39 pm ]
Post subject: 
One further thing...

Increasing resolution to 800x600 from 720x576Noscale removed the need for deinterlacing. Picture quality improved markedly (although CPU load didn't change).
Author:  tjc [ Thu Apr 10, 2008 6:06 pm ]
Post subject: 
This may be very dependent on how the video chips do TV output, or what they consider "native" output resolutions. (For example for a long time the ATi drivers would get a huge performance boost for 1024x768 because 1024 was a "natural" (power of 2) row size.) My guess is that 720x576 isn't handled as well as conventional computer display sizes.
Author:  tuatara [ Fri Apr 11, 2008 2:41 am ]
Post subject: 
tjc wrote:
This may be very dependent on how the video chips do TV output, or what they consider "native" output resolutions. (For example for a long time the ATi drivers would get a huge performance boost for 1024x768 because 1024 was a "natural" (power of 2) row size.) My guess is that 720x576 isn't handled as well as conventional computer display sizes.


A review of some of the openchrome code revealed that 720x576noscale is actually not a native resolution and is implemented inside the openchrome driver. I'm not sure I fully understand the difference between Noscale and scale yet. However the chipset supports a number of PAL and NTSC resolutions, such as 720x480, 720x576, 1024x768 etc etc.

Cheers.
Paul.
Page 1 of 1 All times are UTC - 6 hours
Powered by phpBB® Forum Software © phpBB Group
http://www.phpbb.com/