[FFmpeg-devel] [PATCH 04/12] Add vector_fmul_matrix to dsputil

Reimar Döffinger Reimar.Doeffinger
Thu Oct 22 08:47:02 CEST 2009


On Wed, Oct 21, 2009 at 08:03:58PM +0100, M?ns Rullg?rd wrote:
> Reimar D?ffinger <Reimar.Doeffinger at gmx.de> writes:
> 
> > On Wed, Oct 21, 2009 at 07:22:57PM +0100, M?ns Rullg?rd wrote:
> >> Michael Niedermayer <michaelni at gmx.at> writes:
> >> 
> >> > On Mon, Oct 19, 2009 at 01:00:24AM +0100, M?ns Rullg?rd wrote:
> >> >> We are doing len [1 x w]*[w x w] matrix multiplications.
> >> >
> >> > you can also see it as a single [len x w][w x w] matrix multiplication
> >> > and that may also allow faster matrix multiplication algos to be used ...
> >> 
> >> Do you know a way of multiplying matrixes with fewer operations?
> >
> > I'm tempted to say "you don't want to go there". I'd have to read up
> > again, but yes, it is possible.
> 
> Why didn't they tell us that in linear algebra class?  Any pointers?

It doesn't belong into linear algebra (at least at Karlsruhe the actual
calculation is not part of that).
They should have mentioned it algorithms courses, namely the Strassen
algorithm (better O(...) performance but basically a mess in almost
every other aspect including real-world performance in almost all
cases and numeric stability) and possible Coppersmith-Winograd (completely
useless today and only of theoretical interest I think).
As usual while Wikipedia has articles on them, as with all math
stuff I heavily recommend not use it, whoever writes those manages
to make the simplest mathematical stuff incomprehensively confusing.



More information about the ffmpeg-devel mailing list