libmad library.
Ported to MS-DOS/OpenWatcom by Jonathan Campbell.


Notes:

- The Sound Blaster MP3 player uses the same codebase as the test.exe
  executable under hw/sndsb, except that some modes have been removed
  and the code rewritten to decode and play back MP3 data.

- If your DOS machine has no sound card, but does have the traditional
  "PC Speaker", you can use playspkr.exe to play the MP3 through the
  internal beeper (using the old DPCM trick). However there may be
  slight audio glitches depending on background processes,
  interrupts, and disk I/O. Make sure your DOS extender and CPU are
  fast enough to service interrupts at a high timer tick rate.
  Playback will also not sound very good if the "PC speaker" is not
  an actual speaker, such as the piezoelectric ones on laptops.

- If you have a Gravis Ultrasound, use the playgus.exe MP3 player.
  The Sound Blaster version does NOT work with SBOS or MEGA-EM despite
  my best efforts.

- If you have a LPT parallel port, you can use the LPT DAC version,
  playlpt.exe. Though playback up to 44.1KHz is supported, it may
  not be possibly depending on whether or not the parallel port
  delays I/O for compatability reasons, or any other sources of
  latency.

- Experimental: If you have a Covox Sound Master, try playsm.exe.
  Since I don't actually have a Covox Sound Master, I don't know whether
  this works on real hardware.



TODO LIST:
 - 16-bit real-mode MP3 decoding. Currently, libmad seems to compile properly
   only for 32-bit flat protected mode. I got it to "sort of" compile for
   16-bit large memory model builds, but it doesn't work properly and it seems
   to cause memory corruption as well.

   Specification for 16-bit real mode: Take libmad, valgrind it under Linux to
   catch all possible problems, then incorporate the validated code into this
   source tree. That will be the 32-bit build. Then, take the libmad source
   code, put it under a separate directory (libmad16), modify it to use
   14-bit precision and math that favors 16-bit integers (what will fit into
   real-mode 8086 CPU registers), and incorporate that into /etc/libmad16.
   Copy over the test programs here and compile them with it, to produce the
   16-bit real mode builds. Make the makefile for libmad16 so that it only
   produces binaries for 16-bit real mode builds. If the differences are
   minimal enough, you could merge the two source trees into one project
   that can produce a fast high quality decode for 386+ systems, and a faster
   (less precise) decoder for systems as low as an 8086.