MPlayer codebase contains a copy of liba52 from a52dec, which has been heavily patched with various things. This includes usage of SSE, 3DNOW and AltiVec when possible for the IMDCT transformation, which appear to be much quicker than the generic C code or even the sci-libs/djbfft based software accelerated version. What is the story of a52dec upstream? It seems like a dead project from the outside - is there any movement there and could the mplayer patches be integrated into the system copy of a52dec, as for other projects to be able to benefit from the SSE, 3dnow! and AltiVec based accelerations? For example currently gst-plugins-a52dec happily sets all the CPU flags correctly for its a52_init() call, but our system liba52.so.0.0.0 doesn't have any code for CPU instructions based acceleration, like mplayer bundled copy does. Additionally if the MPlayer patches would be merged to the system a52dec library, we could start thinking of using the system version of liba52, instead of the bundled copy, which is also a good QA thing. If this is done, then djbfft as default perhaps doesn't make sense anymore, which is requested in bug 282412
I think this one's "dead" in the sense of it's complete, there's no more development to do. I'm hesitant to port patches from mplayer into a52dec, since it'd cause a new set of maintenance issues. If a user prefers mplayer's version, use that. :)
