On Mon, Jul 14, 2008 at 5:21 PM, Jason Tackaberry <[EMAIL PROTECTED]> wrote:
> On Mon, 2008-07-14 at 10:19 -0700, Chandan Pitta wrote:
> > This is interesting. I have a high end nVidia graphics card (8800 GT)
> > and yet when using gl my cpu hovers around 75% (on a 2.4 GHz AMD X2)
> > when playing 1080i, so chances of being able to play blu-ray are
> > pretty slim. Jason can you send me the command-line you are using. I
> > have been searching forums for a long time to understand the problem.
>
> It's not really comparable to using the player's gl vo. It uses a very
> different pipeline, which works like this:
>
> 1. kaa.candy maps a new window and then redirects it using
> XComposite (via clutter).
> 2. We spawn the media player (MPlayer, Xine, etc.) and have it draw
> to the wid from step 1, which is redirected. The player uses
> Xv, and so it follows the Xv path to get the frame into video
> memory, which typically means hardware colorspace conversion.
> 3. Clutter handles Damage events for the redirected video window,
> and uses GLX_EXT_texture_from_pixmap to take the redirected
> window and put into a ClutterTexture.
>
Your explanation is very good. I understand what you mean now. So video
decoding is still accelerated via xv but only the output is redirected and
transformed using gl.
>
> If GLX_EXT_texture_from_pixmap isn't available (or isn't working [huzzah
> Intel!]) there is a slower X11 fallback, which should work with 720p if
> the CPU is fast enough. Dischi and I are expecting that Intel can't
> suck forever, so eventually the GLX approach will work with it. I have
> no idea how it works with ATI, but I fully expect it's even more broken
> than Intel. (Anyone who uses an ATI card for their HTPC deserves what
> they get anyway.)
>
> For systems whose Xv uses video texture (instead of video overlay), step
> 3 should be extremely fast. (Because the work is basically already
> done.)
>
> Now Nvidia uses video textures and has a working
> GLX_EXT_texture_from_pixmap. This means that the above pipeline is very
> close in performance to using unredirected Xv. On both of my systems
> (E6600 with Nvidia 7100GS, and E6750 with Nvidia 8600GT), for a
> 1920x1080 video there is 2-4% overhead in cpu (primarily the Xorg
> process).
>
> If you'd like to test this out, try the attached file, clutter-test.c.
> Compile it using:
>
> gcc clutter-test.c `pkg-config clutter-0.7 --cflags --libs` -o
> clutter-test
>
> It requires clutter 0.7 or 0.8 (change pkg-config above accordingly).
>
> Make sure you do not have a compositing manager running. If you do,
> stop it. In fact, you should probably restart X if you were running a
> compmgr. (I had some issues even after switching from compiz to
> metacity until I restarted X.) Start up an mplayer:
>
> mplayer video.avi
>
> Find out its window id using xwininfo. Let's say it's 0xa00001. Then
> run:
>
> ./clutter-test 0xa00001
>
> It will open a 640x480 window and begin redirecting MPlayer onto a
> rotated texture on the clutter stage. (Note that it uses the
> antialiasing settings on your card, so if it looks a bit ugly, increase
> the AA with nvidia-settings. With an 8800GT you should have no trouble
> with 8x or better.)
>
This is cool!!! I like how clutter applies the transformations. The edges
look ugly. I did not change AA but I take your word for it. What kind of
problems did you see when using compiz? I ran the test with compiz-fusion
and it works well for me at least. However the only difference I saw was
that both mplayer window and the clutter window have the video with compiz,
while with kwin only the clutter window got updated. And yes my cpu is low
(just slightly higher than plain xv but not as high as gl). Thanks for the
info Jason. Very helpful.
>
> With nvidia, you should find it performs quite well. With Intel, it
> almost certainly won't work, but you can edit clutter-test.c and replace
> clutter_glx_texture_pixmap_new_with_window() with
> clutter_x11_texture_pixmap_new_with_window() to use the slower X11
> fallback.
>
> Cheers,
> Jason.
>
>
> -------------------------------------------------------------------------
> This SF.Net email is sponsored by the Moblin Your Move Developer's
> challenge
> Build the coolest Linux based applications with Moblin SDK & win great
> prizes
> Grand prize is a trip for two to an Open Source event anywhere in the world
> http://moblin-contest.org/redirect.php?banner_id=100&url=/
> _______________________________________________
> Freevo-devel mailing list
> Freevo-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/freevo-devel
>
>
-------------------------------------------------------------------------
This SF.Net email is sponsored by the Moblin Your Move Developer's challenge
Build the coolest Linux based applications with Moblin SDK & win great prizes
Grand prize is a trip for two to an Open Source event anywhere in the world
http://moblin-contest.org/redirect.php?banner_id=100&url=/
_______________________________________________
Freevo-devel mailing list
Freevo-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/freevo-devel
