Re: LuaJIT FFI and the FFMPEG library

[Daurnimator <quae@...>]

On 20 February 2011 06:46, Michal Kottman <k0mpjut0r@gmail.com> wrote:
> Hi,
>
> I wanted to see, how LuaJIT & FFI deal with sound and image
> processing, having seen great performance with number crunching. To do
> that, you first need to load the data. For the greatest flexibility
> w.r.t. various codecs, I wanted to try out the FFI with the FFMPEG
> libraries (libavformat, libavcodec). To my surprise, getting the
> libraries to Lua using the FFI was easy and quick - it took about 5
> minutes along with thinking and preparation :)
>
> All I had to do was to preprocess the headers using gcc.I created a
> file tmp.c, in which I wrote several #include lines, and then ran "gcc
> -E tmp.c > ffmpeg.h". In Lua, all I have to then do is:
>
>    avcodec = ffi.load('avcodec-52')
>    avformat = ffi.load('avformat-52')
>    ffi.cdef(assert(io.open("ffmpeg.h")):read('*a'))
>
> And then, "everything-just-works". You only need to know/guess, in
> which module each function lies, and call it appropriately (like
> avcodec.avcodec_decode_audio3(...)). I spent the next several hours
> learning how to decode audio with FFMPEG, which was a PITA because of
> lack of good documentation - I had to look up relevant calls in ffmpeg
> source files... Fortunately, I've done the work for you, so if you
> want to do load audio using LuaJIT and FFMPEG, head over to the
> following page and get inspired (consider it public domain):
>
> https://github.com/mkottman/ffi_fun/blob/master/ffmpeg_audio.lua
>
> It loads any file with audio stream that FFMPEG understands, decodes
> it, and produces a FFI array of samples for the whole file, which can
> be further processed. The process is really fast, a 10MB MP3 file is
> processed in ~3 seconds (this involves everything, including library
> loading, cdef parsing, decoding and writing the 60MB of raw samples to
> disk).
>
>

Very nice, I may actually use this soon.
First to try and shoehorn it into ALSA though :P
Seems ALSA is the only audio framework that doesn't require callbacks.