comparison DOCS/tech/codec-devel.txt @ 3862:74622f23bce5

initial commit of "Guide To Codec Hacking"

3862:74622f23bce5

A Guide To Developing MPlayer Codecs
by Mike Melanson (melanson at pcisys dot net)

Introduction
------------
I've developed a number of open source decoders for the MPlayer project,
for both audio and video data. As such, I feel I'm qualified to document a
few notes about developing new codecs for the codebase.

As always, the best way to learn how to incorporate a new codec is to
study a bunch of existing code. This document is supplementary material to
the code, meant to give some tips, pointers, and a general roadmap.

A note about terminology: "Codec" stands for coder/decoder (or
compressor/decompressor, if you prefer). The term refers to a module that
can both encode and decode data. However, this document focuses primarily
on incorporating decoders. Still, the terms "decoder" and "codec" are
often used interchangeably.

Necessary Materials
-------------------
So you've decided that you want to implement a new decoder for
MPlayer. There are a few things you will need:

- Knowledge of the codec to be implemented: You will need to know the data
format of the chunks that MPlayer will pass to you. You will need to know
how to take apart the data structures inside. You will need to know the
algorithmic operations that need to be performed on the data in order to
reconstruct the original media.

- Sample media: Preferably, lots of it. You will need media encoded in
your data format and stored in a media file format that MPlayer knows how
to parse (these include AVI, ASF, MOV, RM, VIVO, among others). If the
encoded data is stored in a media file format that MPlayer doesn't
understand, then you will either need to somehow convert the format to a
media file format that the program does understand, or write your own
MPlayer file demuxer that can handle the data. Writing a file demuxer
is beyond the scope of this document.
	Try to obtain media that stresses all possible modes of a
decoder. If an audio codec is known to work with both mono and stereo
data, search for sample media of both types. If a video codec is known to
work at 7 different bit depths, then, as painful as it may be, do what you
can to obtain sample media encoded for each of the 7 bit depths.

- Latest CVS snapshot: It's always useful to develop code for the very
latest development version of MPlayer. Be sure to update your local CVS
copy often.

- General programming knowledge, working Linux development environment: I
would hope that these items would go without saying, but you never know.

Typical Development Cycle
-------------------------
1) Set up basic infrastructure
First things first, there's a big song and dance to go through in order to
let the MPlayer program know that you have a new codec to incorporate.

MPlayer does not feature what some would term a "clean" codec plugin
architecture. Some log this as a complaint. Personally, I think it's
necessary to allow MPlayer the type of flexibility to incorporate so many
open- and closed-source codecs.

First, modify your local copy of codecs.conf. It may be system-shared or
in your home directory. Add a new entry for your codec. If it's an open
source codec, it would be a good idea to place the new entry with the rest
of the open source codecs. When you're confident that you have the entry
right, be sure to add it to etc/codecs.conf in your workspace. See the
file codecs.conf.txt for a detailed description of the format of this
file. Create a new audiocodec or videocodec block with the proper info,
FOURCCs/format numbers, output formats, and a unique driver name. Remember
the driver name.

Next, edit the file codec-cfg.h. You will find a list of #define's that
map names like AFM_MSADPCM and VFM_MSVIDC to numbers. The definitions that
begin with AFM_ are audio drivers. The definitions that begin with VFM_
are video drivers. If you want to implement a new audio driver, go to the
end of the AFM_ list and create a new AFM_ definition for your decoder
using the next available number in the list. If you want to make a new 
video decoder, do the same thing to the VFM_ list.

Next, edit the file codec-cfg.c. You will find an array of audio driver
names and video driver names. If you are implementing a new audio codec,
add your new driver name (the one you entered into the codecs.conf
file) between the last non-NULL entry and the NULL at the end of the
audio driver array. If you are implementing a new video codec, do the
same in the video driver array.

Next, create a new source file which contains the main decoding function
that MPlayer will call to decode data. Eventually, you may have multiple
files which comprise your decoder, but let's start simple here. Create the
skeleton function for your decoder. Since you will also have to write the
code to invoke the function, you can make the decoding function look
however you want, with whatever parameters you feel will be
necessary. Here's an example video decoder:

void some_video_decoder(
  char *encoded,	// buffer of encoded data
  int encoded_size,	// length of encoded buffer
  char *decoded,	// buffer where decoded data is written
  int width,		// width of decoded frame in pixels
  int height,		// height of decoded frame in pixels
  int bytes_per_pixel)	// bytes/pixel in output image

Here's an example audio decoder:

int some_audio_decoder(
  unsigned short *output,  // buffer where decoded 16-bit PCM samples go
  unsigned char *input,	   // encoded data
  int channels)	           // mono = 1, stereo = 2

Next, modify the Makefile so that it will compile your new source
file.

Next, modify either dec_audio.c or dec_video.c, depending on whether
you're writing an audio or video decoder. You'll probably put the new
decoder function header at the top of the file unless you've created a
header file to handle it, in which case, you'll include the new header
file. The dec_*.c functions are in charge of initializing codecs and then
passing encoded data to the proper decoder function. The init and decode
functions are big switch statements that key off of the codec definition
numbers from codec-cfg.h. Your best bet in here is to examine some simple
other simple decoders and clone relevant portions of the case blocks.

Next, compile the project and see if you have everything correct so far.

Next, you want to make sure that the encoded data is making it to your
decoding function in the first place. This may sound like a trivial
exercise, but there are a lot of things that can go wrong (and I've
watched most of them go wrong in my experience). At the beginning of your
skeleton decoder function, enter the following code:
  int i;
  for (i = 0; i < 16; i++)
    printf ("%02X ", input[i]);
  printf ("\n");
When you compile and run MPlayer, your decoder function will print the
first 16 bytes of each data chunk that it receives. Open the sample media
in a hex editor and reconcile what you see on the screen with what
you find in the file. If the decoder is printing the first 16 bytes of
each block, that's a good sign that you're ready to move on to step
2. Otherwise, you need to figure out why the data isn't getting to your
decoder. Is your decoder even being invoked? If not, why not?

2) Develop the decoder
Go for it. Remember to make it work, first, then make it work fast.

3) Debug and test the decoder
If you're extremely lucky, the decoder will work the first time. If you're
very lucky, it will work after you've reviewed your code a few times and
corrected a few obvious programming mistakes. Realistically, you will
write the decoder, review it many times and fix many obvious and subtle
programming errors, and still have to go through an elaborate debug
process in order to get the decoder to a minimally functional state.

Big hint: Ask for all warnings. You know, the -Wall option in
gcc? It's very useful to develop your codec while running in debug
mode. In order to compile MPlayer with debug support (which includes -Wall
for all gcc operations), use the --enable-debug option when configuring
the project. Pay attention to all warnings and make it a goal to get
rid of every single one. I'll never forget when the compiler warned me
that there was no point in clamping a signed 16-bit variable within a
signed 16-bit range (the calculation to be clamped was supposed to be
stored in a signed 32-bit variable and then stored in the signed 16-bit
variable). I sat stunned for a moment, feeling like I had just dodged a
bullet as I knew that would have taken me hours to debug that kind of
mistake.

4) Contribute decoder to codebase
Create a patch with the "diff -u" format and email it to the MPlayer
development team for approval. You will likely need to diff the following
files:
- Makefile
- etc/codecs.conf
- codec-cfg.c
- codec-cfg.h
- dec_audio.c -OR- dec_video.c
Of course, you will need to include your newly-created file(s). If you
contribute enough decoders, the development team may even grant you write
privileges to the CVS repository.

5) Wait for bug reports to start rolling in
You may think you're finished when you release the codec and if you're
extremely lucky, you will be right. However, it's more likely that people
will start throwing all kinds of oddball media at your decoder that it
never counted on. Cheer up; take comfort in knowing that people are
testing your code and attempting to use it as a real world 
application. Download the problem media that people upload to the MPlayer
FTP site and get back to work, implementing fixed code that addresses the
issues. Contribute more patches and encourage people to hammer on your
decoder even more. This is how you make your decoder rock-solid.

EOF