Hello Reinhard,

I managed to build ffmpeg 0.6-2 on armel by forcing

has_vfp := 1
has_neon := 1

in debian/confflags. After that mplayer 1.0~rc4~try1.dsfg1-1 built
without problems and the bug is now gone, thanks!

I ran some benchmarks to ensure that we have no serious
regressions. The following is in the format

(command)
# (output with 2:1.0~rc3++final.dfsg1-1)
# (output with 1.0~rc4~try1.dsfg1-1)
# (comment)

so that it can easily be run as a script if somebody wants to
replicate my results. /tmp/theora/lib used by the last two commands
contains the (r17262 of the) theorarm branch of libtheora that has
ARM optimizations which are not in Debian yet:

wget -c 
http://ftp.acc.umu.se/pub/debian-meetings/2009/debconf9/low/1050_Lightning_talk_Redirecting_require.ogv
rm -fr ~/.mplayer

mplayer -quiet -benchmark -nosound -vo null 
1050_Lightning_talk_Redirecting_require.ogv
# BENCHMARKs: VC:  42.123s VO:   0.056s A:   0.000s Sys:   3.309s =   45.488s
# BENCHMARKs: VC:  42.410s VO:   0.057s A:   0.000s Sys:   3.711s =   46.178s

mplayer -quiet -benchmark -ao null -vo null 
1050_Lightning_talk_Redirecting_require.ogv
# BENCHMARKs: VC:  42.557s VO:   0.059s A:  29.162s Sys: 132.968s =  204.745s
# BENCHMARKs: VC:  42.573s VO:   0.059s A:  29.164s Sys: 132.922s =  204.718s

mplayer -quiet -benchmark -ao null -vo fbdev 
1050_Lightning_talk_Redirecting_require.ogv
# BENCHMARKs: VC: 100.571s VO:  62.395s A:  29.759s Sys:  11.450s =  204.176s
# BENCHMARKs: VC: 100.950s VO:  62.561s A:  29.855s Sys:  10.820s =  204.185s

mplayer -quiet -benchmark -ac -ffvorbis, -ao null -vo fbdev 
1050_Lightning_talk_Redirecting_require.ogv
# BENCHMARKs: VC: 100.338s VO:  62.138s A:  18.401s Sys:  23.125s =  204.002s
# BENCHMARKs: VC: 100.369s VO:  62.127s A:  18.408s Sys:  23.093s =  203.998s
# banning -ffvorbis makes mplayer use tremor which is faster

LD_LIBRARY_PATH=/tmp/theora/lib mplayer -quiet -benchmark -ac -ffvorbis, -ao 
null -vo fbdev 1050_Lightning_talk_Redirecting_require.ogv
# BENCHMARKs: VC: 100.330s VO:  62.173s A:  18.398s Sys:  23.073s =  203.974s
# BENCHMARKs: VC: 100.351s VO:  62.199s A:  18.434s Sys:  22.987s =  203.971s
# having libtheora in LD_LIBRARY_PATH does not help since mplayer prefers 
fftheora

LD_LIBRARY_PATH=/tmp/theora/lib mplayer -quiet -benchmark -ac -ffvorbis, -ao 
null -vc -fftheora, -vo fbdev 1050_Lightning_talk_Redirecting_require.ogv
# BENCHMARKs: VC:  46.867s VO:  85.455s A:  18.274s Sys:  54.063s =  204.659s
# BENCHMARKs: VC:  46.873s VO:  85.435s A:  18.291s Sys:  54.035s =  204.634s
# banning fftheora makes mplayer use libtheora which is faster (since it has 
ARM optimizations)

LD_LIBRARY_PATH=/tmp/theora/lib mplayer -endpos 00:30 -framedrop -ac -ffvorbis, 
-ao alsa -vc -fftheora, -vo fbdev 1050_Lightning_talk_Redirecting_require.ogv
# A:  30.0 V:  30.0 A-V:  0.021 ct: -0.008 750/750 27% 42% 14.4% 24 0
# A:  76.3 V:   0.4 A-V: 75.919 ct:  0.000   0/  0 ??% ??% ??,?% 4607 0
# REGRESSION: output is filled with "Too many buffered pts" and a huge
# number of frames are dropped. Google finds
# http://lists.mplayerhq.hu/pipermail/mplayer-advusers/2007-May/001650.html
# which suggests -nocorrect-pts. With that I get
# A:  30.0 V:  30.0 A-V:  0.025 ct:  0.000 751/751 27% 42% 14.6% 25 0


Should I file a separate bug report for this regression? Are you able
to reproduce it?



_______________________________________________
pkg-multimedia-maintainers mailing list
pkg-multimedia-maintainers@lists.alioth.debian.org
http://lists.alioth.debian.org/mailman/listinfo/pkg-multimedia-maintainers

Reply via email to