[MPlayer-dev-eng] Really nasty quick hack for making mp3lib faster...

[Thomas Orgis]

Am Sun, 30 May 2010 23:24:40 +0200
schrieb Thomas Orgis <thomas-forum at orgis.org>: 

> the discussion about the upcoming mpg123 patch brought some funky
> performance issues to the light

I have to correct the numbers. In those comparisons I had a variant of
mpg123 installed that contained more accurate code for 16 bit output
(--enable-int-quality configure option).
That variant gives something along that:

shell$ time src/mpg123 -q -w /dev/null ../../../convergence_-_points_of_view/*.mp3

real	0m6.018s
user	0m5.892s
sys	0m0.113s

A normal build gives that:

shell$ time src/mpg123 -q -w /dev/null ../../convergence_-_points_of_view/*.mp3

real	0m5.223s
user	0m5.074s
sys	0m0.125s

That definitely is faster than mplayer (with the forced alignment to 64
bits in mp3lib)... 

shell$ for i in mpg123 mp3; do echo $i;  time ./mplayer -ao pcm:file=/dev/null  -quiet -ac $i ../../convergence_-_points_of_view/*.mp3 > /dev/null ; done mpg123

real	0m5.479s
user	0m5.399s
sys	0m0.057s
mp3

real	0m6.084s
user	0m6.011s
sys	0m0.055s


And yes... that mplayer mp3lib time is larger than before ... I suspect
that it performs differently now with recent changes in MPlayer svn. It
changes every time...

Important point for me: On this Core2 system, both mpg123 and mplayer
with libmpg123 are faster, but there is a distinct margin added in
mplayer using libmpg123. That should be analyzed.

On my K6-3+, this margin is what makes mplayer -ac mpg123 slower than
mplayer -ac mp3. Perhaps Diego can gather
some supporting profiling on his machine? That time must go somewhere.

shell$  time ./mplayer -ac mp3 -quiet -ao pcm:file=/dev/null ~/convergence_-_points_of_view/*.mp3 >/dev/null

real	0m57.147s
user	0m55.430s
sys	0m1.130s
shell$  time ./mplayer -ac mpg123 -quiet -ao pcm:file=/dev/null ~/convergence_-_points_of_view/*.mp3 >/dev/null

real	1m0.599s
user	0m59.600s
sys	0m0.950s
shell$ time mpg123 -q -w /dev/null ~/convergence_-_points_of_view/*.mp3

real	0m55.469s
user	0m54.600s
sys	0m0.850s


Alrighty then,

Thomas.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 197 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/mplayer-dev-eng/attachments/20100607/2faa03db/attachment.pgp>