[vorbis] Performance tests

[Kenneth Arnold]

On Tue, Oct 16, 2001 at 07:50:04PM -0400, Greg Wooledge wrote:
> volsung at asu.edu (volsung at asu.edu) wrote:
> 
> > time ogg123 -q file.ogg
> > 
> > That way CPU usage is being limited by the playback audio device (forcing CPU
> > usage to be approximately realtime)
> 
> Good point.  Let me do one of those.  And while we wait...

But not strictly necessary, see below.

> > and CPU time is not being wasted on
> > printing the status line.
> 
> I think it's safe to say that the amount of time spent on that particular
> part of the code is negligible compared to the actual time spent decoding
> any realistically-sized Ogg file. :-)

You'd be suprised. ogg123 rc2 is quite stupid in this regard, fetching
instantaneous bitrate from libvorbis (not instantaneous in CPU though)
and updating the status line every 4096 bytes of decoded output. My
branch in CVS is better, but I do need to get status update
throttling.

> Strangely enough, this is less CPU time than was required to play that
> same file to /dev/null.  Perhaps the fact that it was cached in RAM has
> more influence than I suspected.  (And this box has more than enough RAM.)

If you really want to be accurate with CPU usage, I'd use the
following procedure:

$ cat test.ogg > /dev/null ; time xiph-cvs/vorbis/examples/decoder_example < test.ogg > /dev/null
$ cat test.mp3 > /dev/null ; time mpg123 <decode and write to /dev/null, not just test decode> test.mp3
$ cat test.mp3 > /dev/null ; time mpg321 <decode and write to /dev/null, not just test decode> test.mp3

Testing on my Athlon 1.3 (DDR) shows 13.830 seconds to decode a 737.9
second song (Nebularia - Ultima Ratio). I don't have any songs to
compare adequately for mp3 (should really be the same song for best accuracy).


-- 
Kenneth Arnold <ken at arnoldnet.net>
- "Know thyself."



<HR NOSHADE>
<UL>
<LI>application/pgp-signature attachment: stored
</UL>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: part
Type: application/octet-stream
Size: 190 bytes
Desc: not available
Url : http://lists.xiph.org/pipermail/vorbis/attachments/20011016/c7164ab0/part.obj