[Theora-dev] FPGA implementation/ players speed?

[Ivan Popov]

On Wed, Mar 09, 2005 at 01:56:25AM -0700, Andrey Filippov wrote:
> > Pity that it turns out being hard to decode them.
> 
> There is little sense of having a camera too fast for the regular PC. It

Hyperthreaded/multicore CPUs become a commonplace, so a threaded player
is quite natural thing to have. I guess this year most of new computers
will be able to play your data, given suitable software.

> I'm not a profi in PC software design, but what about various SIMD
> extensions in CPUs? Are they already used in the library?

Thanks for the question, I forgot I do not have them in the default setup.

With player MMX-patches from VP3HoSwiYO, i get about 5.6 fps at .73 GHz,
which means about 4 GHz to be able to play this clip at full speed.

It can certainly be improved, but I doubt you can easily double the speed,
while threading is a more certain gain. Applied together they surely
make your high resolution movies playable on most new PCs.

> What about memory vs. speed trade-offs? Is it possible to maximize the
> speed using more memory for the decoder?

I wonder if it is possible to decimate the frames without full decoding
to reduce fps to say 25 without a perceptible loss of quality.
It seems it would make it playable on a 3.4 GHz, if you are lucky.

Which platform are you going to decode the movies on?

My tests are with mplayer on *nix.

Regards,
--
Ivan