Finally, Intel G45 VA-API Support Is Available

[Rizzly]

I had the exact same problem, and though I'm not a programmer I took a quick look at i965_media.c line 348, and it looks like they forgot the one function call to actually have the new code used...

This little patch got the vaapi_h264 test working for me;

--- i965_drv_video/i965_media.c.orig	2011-05-17 19:59:00.126115745 +0100
+++ i965_drv_video/i965_media.c	2011-05-17 20:00:51.048731380 +0100
@@ -341,6 +341,9 @@
     case VAProfileH264Baseline:
     case VAProfileH264Main:
     case VAProfileH264High:
+        i965_media_h264_dec_context_init(ctx, media_context);
+        break;
+
     case VAProfileVC1Simple:
     case VAProfileVC1Main:
     case VAProfileVC1Advanced:

Now to find an mplayer ebuild with vaapi support...

It looks like this was updated before I made my install, and I still get the very same error.

[darkbasic]

Now to find an mplayer ebuild with vaapi support...

That's the easy part...
http://gpo.zugaina.org/media-video/mplayer

[Kano]

I added that patch to the q45-h264 branch, but it takes ages to see a decoded frame. It decodes something but it looks like it runs with slowmotion... I tested it with q45 and mplayer vaapi if that matters, not sure if it can decode it or not.

[BlackStar]

I added that patch to the q45-h264 branch, but it takes ages to see a decoded frame. It decodes something but it looks like it runs with slowmotion... I tested it with q45 and mplayer vaapi if that matters, not sure if it can decode it or not.

Well, it's a start. At least your system didn't hang.

[daveisadork]

I got all this running with mplayer-vaapi (using the patch posted on page 1), but it's extremely slow, like maybe 0.5fps. I tried a variety of different H.264 videos in various resolutions, containers, etc and all produced the same results.

MPEG2 acceleration seems to work a bit better, but higher bitrate videos don't maintain audio sync for even a few seconds.

Overall, VA-API on my hardware (glxinfo reports a Mobile Intel® GM45 Express Chipset) appears to be a step back in terms of performance for both H.264 and MPEG2. I've yet to find a single file that plays back better with it than without.

This is on Fedora 15 with the stock kernel (2.6.38.6-27) and Intel driver (2.15.0-3). I wonder if it would work better with a newer driver and/or kernel.

[maxwell_smart]

I notice that the check-in comment says that it is single-threaded, and that there is something called the sw_scoreboard that is disabled.

It seems that the sw_scoreboard is to coordinate multiple threads (the values seem to be 23 or 64) to work together to decode frames. This sw_scoreboard_vfe is set to 1 instead for this checkin.

Maybe once they figure out the sw-scoreboard, the patch will be faster, but until then, at least this check-in lets developers test their applications on G45 hardware.

[kbios]

It is just me or this is heavily obfuscated code? 90% of lines are just hex values...

[darkbasic]

Is there any news?

[Kano]

I am sure you can look yourself at

http://cgit.freedesktop.org/libva/

[darkbasic]

(character limit)