[XviD-devel] Changes to get_pmv2

Tom Jacobs T.R.Jacobs at lboro.ac.uk
Wed Sep 22 16:38:48 CEST 2004


hi

my work on mpeg2 has only been simulated in an ideal SimpleScalar simulator.
so no realworld results. but it did show a 96% complexity reduction for 20
processors on a 320x240 image. im going to do a cycle actuate model of it
later this year and then make me make an rtl model after that. it intended
for SoC.   well that the plan anyway, have got 2 more years to do something
and write my thesis.

to clear up a few things while im writtin this. is there a single search
function (i no it sound stupid) where the subtraction takes place? if i did
stop it from calcuating difference in search would i have to do it in every
search method (havent looked in depth at the different searchs)?

pmvs[*] and mv16 are the location for the difference MV, correct? or is it
mvs[*] ?

thanks for all your help

Tom

Quoting Christoph Lampert <chl at math.uni-bonn.de>:

> On Wed, 22 Sep 2004, Tom Jacobs wrote:
>
> > thanks
> >
> > i think i got the prediction and MV bit now, and yes i see how the
> > prediction need to be correct since they are recalculated in the
> decoder.
> >
> > so if i set it up so that the main searches are done in parallel with
> the
> > wrong predictions i can then, in serial, add the wrong predictions to
> the
> > differential MV and then subtract the correct prediction. would this
> work?
>
> It should, yes. I guess, it would be cleanst to modify the search then,
> such that it stores the vector itself (not the differential), and add a
> little code that before (or while) writing the bitstream, calculates the
> right predictors and subtracts them.
> That's a perfectly reasonable way to do ME, and we only chose to directly
> store the differential, because it saves a few CPU cycles by not having
> to
> calculate the predictor again.
>
> > thanskd for the other suggestions and i am not just ignoring them but i
> am
> > trying to fine grain parall xvid, ive already done it with mpeg2.
>
> That's interesting. What encoder, and what were your results?
>
> gruel
>
> _______________________________________________
> XviD-devel mailing list
> XviD-devel at xvid.org
> http://list.xvid.org/mailman/listinfo/xvid-devel
>





More information about the XviD-devel mailing list