On Wed, Jul 11, 2012 at 07:36:43AM -0700, Jason Garrett-Glaser wrote:
> On Tue, Jul 10, 2012 at 11:00 PM, Kostya Shishkov
> <[email protected]> wrote:
> > On Tue, Jul 10, 2012 at 06:01:39PM +0200, Kostya Shishkov wrote:
> >> On Tue, Jul 10, 2012 at 02:25:03PM +0200, Kostya Shishkov wrote:
> >> > On Mon, Jul 09, 2012 at 08:06:53PM +0200, Kostya Shishkov wrote:
> >> > > On Sat, Jul 07, 2012 at 09:38:53AM -0700, Jason Garrett-Glaser wrote:
> >> > > [...]
> >> > >
> >> > > This should address most of your and Diego's comments.
> >> >
> >> > ping?
> >>
> >> New version with more of Jason's comments addressed.
> >
> > Even newer version with newer Jason's comments addressed (i.e. now it stores
> > only one row of previous DCs).
> 
> I don't understand the purpose of dc_cache or update_dc_cache.  Can't
> you just leave everything in the main dc data structure and not add so
> many layers of complexity?  That is, store and read directly from
> prev_dc in the DC prediction and DCT decoding.
> 
> prev_dc is [plane][row][mb_x].
> 
> TOP is just prev_dc[p][(i&1)^1][mb_x].
> LEFT is just prev_dc[p][(i&1)][mb_x-1].
> TOPLEFT is just prev_dc[p][(i&1)^1][mb_x-1].
> 
> After decoding an mb, we store it in prev_dc[p][i&1][mb_x].
> 
> I don't understand why it needs to be so much more complicated than this?

So it uses only one row instead of two.
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to