Triple monitor AGP+PCI, mergedfb and xvideo

Alex Deucher alexdeucher at gmail.com
Tue Oct 3 08:24:12 PDT 2006


On 10/3/06, Aefron <aefron_ml at yahoo.fr> wrote:
> Hello,
>
> I'm sending on this mailing-list to try to investigate an annoying
> problem I encounter with my system.
>
> My layout is quite unusual, but here is what I try to set up:
>
>     *
>
>       MergedFB 1600x1200-1280x1024 dual head using an AGP Radeon with
>       the free-driver (overall in order to benefit from double-buffer
>       and xvideo extensions, which I lose benefit from if I try to
>       activate Xinerama)
>
>     *
>
>       800x600 isolated X-screen on my TV using a PCI Radeon with the
>       free-driver (I use a VGA-to-S-Video converter, which I find more
>       convenient than patching my X.org with the Gatos-like driver)
>
>
> Everything works fine until I decide to launch a fullscreen video using
> mPlayer and xv output on the TV X-display (namely :0.1)... the video is
> displayed quite incorrectly, as I am only able to see a small part of
> it, overzoomed and truncated.
>
> What seems to happen is that fullscreen XVideo output on the PCI overlay
> is doing as if it tried to map over a surface which size equals the AGP
> overlay's one that is much bigger. Indeed, if I try to pass the
> -geometry 800x600+0+0 parameter to mPlayer, the video fulls the TV
> screen in its totality (though it's deformed, just being a resized
> X-window, in which video aspect ratio is not being conserved... tough,
> xvideo, even windowed, seems to works, relative to the very low CPU
> consommation)... Worse... If I try to pass the -geometry 800x600-0-0
> parameter to mPlayer, the video is not centered on my monitor and gets
> truncated...
>
> Of course, it could be a mPlayer bug, but what makes me think MergedFB
> is in cause is that passing it in clone-mode on both AGP outputs
> (instead of 1600x1200-1280x1024) is enough to make the PCI card display
> the fullscreen Xvideo like it should. Only dual-headed MergedFB makes
> thing behaves strangely (and as Xinerama disables xv and dbe extensions,
> this is not an option to me either).
>

Xv actually works fine with xinerama.

>
> I think I've tried out every single documented option in man xorg.conf,
> have used well-supported 9200 Radeon, as well as
> experimentally-supported 9600 Radeon (as well in AGP as in PCI), tried
> not to use the VGA-converter and use a third computer CRT instead, did
> not have seen any change using xvattr and the XV_SWITCHCRT attribute
> (though I am not sure to understand how it works) and do not know how to
> change the option Videokey mentioned in the free-driver radeon
> manpage... Nothing has changed...
>
> Nobody seeming to know how the VideoAdaptors section works in xorg.conf
> doesn't relieve me either...
>
> I also have seen on the X.org 7.2 release plans
> (http://xorg.freedesktop.org/wiki/ChangesForX11R72) that this release
> should fix some PCI bugs (I have been forced to upgrade to X.org 7.1 on
> my Gentoo because 7.0 made the PCI card crash on startx) and that
> MergedFB-like pseudo-xineramas setups are gonna be unified in that they
> will have to use the Xinerama extension. Does anybody knows wether it
> may solve my problem or not? Has this problem already been encoutered or
> reported (information about triple-screen configs with X.org is quite
> rare... ooohhh yeeah...) ?
>
> Another question that hangs around my mind concerns any method to verify
> that the PCI fullscreen overlay may indeed be of a size it should not
> (xvinfo does not seem to...).... Any suggestion?
>

I suspect what is happening is that mplayer is querying screen0 for
dimensions and such regardless of what screen you ask it to run on.
Do other video player also have problems?

Alex

>
> Now for a résumé :
>
>
> => Config
>
> 1 PCI Radeon 9200 and 1 AGP Radeon 9600 (or 9200... tried both, nothing
> changed)
>
> Linux 2.6.17-r4
>
> X.org 7.1
>
> xorg.conf, Xorg.0.log and xvinfo output in attachment
>
>
> ___________________________________________________________
>
> Aefron
>
>



More information about the xorg mailing list