[FFmpeg-devel] [PATCH] Channels correlation

Michael Niedermayer michaelni
Fri Oct 30 12:27:38 CET 2009


On Fri, Oct 30, 2009 at 12:02:31PM +0100, Nicolas George wrote:
> Le nonidi 9 brumaire, an CCXVIII, Michael Niedermayer a ?crit?:
> > coeff= 1/coeff;
> > ...*= coeff
> > is faster
> 
> Changed.
> 
> > k can start with a value > 0 as the first few columns should already be 0
> > except the main diagonal elements
> 
> Done. And that is really k: this particular reduction is done on the matrix
> columns, not lines.

yes, your "transposed from what i expect" matrix confused me (thats my
problem i know). But isnt it faster the other way around? that loop reimar
complained about would then be along consecutive elements.


> 
> > btw, isnt libavutil/lls.c/h useable?
> 
> I tried to see a way, but it is just not designed for the same situation.
> The code to adapt the input data to the lls.h API would be quite ugly, and
> probably significantly slower.

What concerns me is that your code and lls do the same thing, and i dont
see why we would need another implementation of this code.
If the lls API is poor, maybe it can be improved rather than another lls
implementation being added inside a muxer?

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Opposition brings concord. Out of discord comes the fairest harmony.
-- Heraclitus
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20091030/c8c816f8/attachment.pgp>



More information about the ffmpeg-devel mailing list