[mythtv] Mpeg2 Decoding Optimization For HDTV

[Anthony Vito]

 A nice discussion on the users list about hardware requirements for
HDTV decoding was had a bit ago. We touched on some of the effects of
low level factors on the decoders performance.
http://www.gossamer-threads.com/lists/mythtv/users/93418?search_string=Anthony%20HDTV;#93418
In my experience the MythTV mpeg2 decoder is slower then the other
common mpeg2 decoders available for linux. This could be do to factors
other then the lowest level decoding of frames ( sound/video sync ..
etc ). However, I do believe that the decoder can be improved. In my
googling of the topic I came across this paper.
http://www.cs.princeton.edu/~chenhan/papers/icme2002t.pdf
In which optimizations are put forth to speed real time playing of
mpeg2 streams, specifically HDTV size frames. The paper claims that
720p decoding was achieved at 62fps on a PIII 933 machine with AGP 4x.
This seems like a significant improvement over current decoders, even
if AC3 decoding is thrown on top.

My questions are.
1) The mpeg12.c file .. I assume from the ffmpeg project.. has the low
level mpeg2 decoding algorithm... is decoding isolated to here?
2) What is the general structure of frames, and path of data flow
through the functions during mpeg2 decoding?
3) Does the current design allow for the optimizations presented in
the paper, or would significant restructuring be needed?
4) Do these optimizations, and the paper's statements in general, seem
realistic?

I believe that MythTV ( and open source in general ) can significantly
benefit from having the best mpeg2 software decoder possible. HDTV
isn't going anywhere soon. My dream is an all in one HDTV MythTV box
with an undervolted, underclocked Athlon.... Who wants to wait for the
hardware to catch up?... honestly...

-- 
Anthony Vito
anthony.vito at gmail.com