[FFmpeg-devel] [PATCH 2/2] mips: Optimization of AAC psychoacoustic model functions

[Michael Niedermayer]

On Tue, Nov 27, 2012 at 04:08:35PM +0000, Zivkovic, Bojan (c) wrote:
> Hello!
> 
> > First, our AAC encoder and its psy model are in not so good shape and
> > its likely that at some point they will change to get them into better
> > shape and produce better quality ...
> 
> We have already optimized the AAC floating point encoder - there are 2 more
> patches with MIPS assembly, remaining for sending. We are willing to
> try to integrate the optimizations into FFmpeg.
> 
> > But independant of that, optimiztaions need to have a clean API
> > for example one can write a FIR filter that takes pointers
> > to input, output, filter coefficients and the number of such values.
> > and write an optimized version for that.
> > 
> > The patch though would just take a random chunk of code and optimize
> > that, thats very unflexible.
> > It cant be reused anywhere and it will break with the tiniest change
> > to the code.
> > I also see that none of the named constants like PSY_LAME_FIR_LEN
> > that are used in the C code appear in the MIPS code so any change
> > to them will break it and this will not even be detected as no
> > assert or anything check for these assumtations.
> 
> The main reason of extracting parts of functions into new ones was to
> avoid unecessary copy-pasting of mainline code in the mips tree. I am not
> sure if I have understood you correctly what were you suggesting to make
> these optimizations acceptable - should we change the API of the extracted
> functions, or something else?

The API should be simple, clean and free of hidden dependancies

The c code will certainly change over time and with the interfaces as
it is the MIPS code will not last long but break quickly from such
changes.

[...]

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

While the State exists there can be no freedom; when there is freedom there
will be no State. -- Vladimir Lenin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20121128/b0791999/attachment.asc>