[Openchrome-users] HW mpeg2 decoding artefacts

[Andrey Liakhovets]

I've got 16x16 square artefacts while playing MPEG2 video
on EPIA SP8000 with `mplayer -vo xvmc -vc ffmpeg12mc'.
There are no artefacts when I play it with `mplayer -vo
xv'.

These are video fragments (856x768, 33 frames, 1.2 Mb
each):

        http://rapidshare.de/files/13628854/01_33.avi.html
        http://rapidshare.de/files/13629424/04_36.avi.html
        http://rapidshare.de/files/13629742/07_39.avi.html
        http://rapidshare.de/files/13629988/10_42.avi.html
        http://rapidshare.de/files/13630257/11_43.avi.html
        http://rapidshare.de/files/13630931/13_45.avi.html

Brief description:
-----------------------------------------------------------
  MPEG2     Frames with  Approximate artefacts coordinates
  video     artefacts    (relative to 856x768 window)
-----------------------------------------------------------
 01_33.avi    22-31     X=480..495, Y=656..671, then
                         minor shifting and transformations
                         and additional squares at
                         X=496..511, Y=624..639 and
                         X=496..511, Y=640..655
 04_36.avi    10-16     X=480..495, Y=480..495
                        X=464..479, Y=528..543
                        X=464..479, Y=544..559, then
                         minor shifting and transformations
                         and additional squares
 07_39.avi    10-16     X=464..479, Y=592..607, then
                         minor shifting and transformations
                         and additional squares
 10_42.avi    14-16     X=512..527, Y=576..591, then
                         minor shifting and transformations
 11_43.avi     5-16     X=432..448, Y=672..687, then
                         minor shifting and transformations
                         and additional square at
                         X=400..415, Y=688..703
 13_45.avi     4-16     X=448..463, Y=640..655, then
                         minor shifting and transformations
-----------------------------------------------------------
Nearly all artefact squares look like
some 16x16 part of a frame.
It seems that artefacts disappear at (or after?) key frame.

Can somebody confirm this?

For me, this looks like a hardware motion compensation bug.
Am I right?

If somebody is interested,
these MPEG2s were made from 45 frames at
     http://rapidshare.de/files/13632732/frames.tar.gz.html
each MPEG is made from the 33 frames indicated in its name
with the command:
  mencoder mf://*.jpg -mf fps=30 -ovc lavc -lavcopts \
        vcodec=mpeg2video:\
        vrc_buf_size=1835:vrc_maxrate=9800:vbitrate=8000:\
        keyint=15:trell:mbd=2:precmp=10:subcmp=10:cmp=10:\
        dia=-10:predia=-10:cbp:mv0:vqmin=1:lmin=1:dc=10

I used MPlayer-cvs20051214 with
`mplayer-uni-20050926.diff',
Xorg 6.9.0, kernel 2.6.14.
Xorg logs are attached (I've tried 1024x768x24 and x16).

I've got the same results with VIA's driver & VeMP:
`CN-CLE-PMXF40056-kernel-src_20050124.tgz',
`VeMP-patch-MPlayer1.0-pre5_v1.2_20050616',
XFree86-4.3.0, kernel 2.4.26.

Andrey Liakhovets
-------------- next part --------------
A non-text attachment was scrubbed...
Name: X_log.tar.gz
Type: application/x-gzip-compressed
Size: 16940 bytes
Desc: not available
Url : http://wiki.openchrome.org/pipermail/openchrome-users/attachments/20060219/8790e9ac/X_log.tar.bin