[FFmpeg-devel] Optimizations for Cell Broadband Engine

[Guillaume POIRIER]

Hi,

On 5/7/07, Luca Barbato <lu_zero at gentoo.org> wrote:
> Loren Merritt wrote:
> >
> > Are row transforms even useful on Cell? Both SSE2 and Altivec can run
> > column/transpose/column faster than row/column.
>
> SPU vector instructions are more or less an evolution of altivec so it's
> more or less to:
>
> - make the structure in order to use the spus (maybe more than one) in
> an efficient way
> - port with minimal changes the altivec code
> - add newer

Let's not forget: split code so that you can feed the SPUs with all
the different parts of the computation, paying extra attention to the
size of the working set (SPUs have access to only their local memory,
there's no shared memory, at least, to the best of my knowledge).

Not easy to optimize for Cell....

Guillaume
-- 
Rich, you're forgetting one thing here: *everybody* except you is
stupid.
    M?ns Rullg?rd