Herb Peyerl wrote:

> The mpg123 I have, I can't distribute however I'm told that another NetBSD
> sociallite in .au has fixed-pointed mpg123 for the shark and made it
> work... I'm not sure if he plans or is able to distribute source.

Hmm, that would be me.  I've got permission from the author of mpg123
to redistribute my code, but after getting that permission I found
a couple of bugs.  First, it doesn't seem to like some .mp3 samples
I found on the 'net - these were 64kbit/s 22050Hz samples.  Second,
every now and then I get a nasty bit of noise - in particular "Kitty"
by "The Presidents Of The U.S.A." and "Like To Get To Know You Well"
by "Howard Jones" both encoded at 192kbit/s with bladeenc have a
reproducable problem.  Playing back with the FP version (at about 1/3rd
real time!) doesn't show the problem, so it's definitely related to my
implementation.  I've listened to about 20ish full CDs and random tracks
off another 80 or so CDs and have only had those two problems...  It's
also sort of usable when the machines is doing other things without too
many rememberable problems - I just did a test (using mpg123 -t) and it
takes about 33 seconds to decode a 2:23 sample (and 10:01 with the FP
version).

I was going to instrument the mpg123 decoder to see if some values are
falling out of the fixed point range I'm using, but haven't had a chance
to yet.  If there's any budding fixed-point-fixers, there's code (and a
binary) in ftp://melanoma.cs.rmit.edu.au/pub/simonb/.  The binary was
compiled with -mtune=strongarm to use the 32x32->64 bit multiply.

Simon.