[NUT-devel] Split index per stream?

[Michael Niedermayer]

Hi

On Sat, Mar 11, 2006 at 04:49:16PM +0200, Oded Shimon wrote:
> On Sat, Mar 11, 2006 at 03:29:46PM +0100, Michael Niedermayer wrote:
> > Hi
> > 
> > On Fri, Mar 10, 2006 at 11:28:38PM +0200, Oded Shimon wrote:
> > > On Fri, Mar 10, 2006 at 02:42:00PM +0100, Michael Niedermayer wrote:
> > > > Hi
> > > > 
> > > > one index for syncpoint pos, one for each stream?
> > > 
> > > Hmm, not sure it is worth it.. How do you even tell between them (I guess a 
> > > stream_id_plus1 in the begginning). Doesn't add much for robustness (the 
> > > stream stuff is tiny, 1-4kb of 80kb), and brings back the old index_ptr and 
> > > several index complexity issues.. I'm against.
> > 
> > ok
> > 
> > btw, how much space would we gain/loose if we use
> > - sp[i-2] + 2*sp[i-1] as sycpoint position predictor instead of sp[i-1]?
> > or even
> > sp[i-1] + median(sp[i-1] - sp[i-2], sp[i-2] - sp[i-3], sp[i-3] - sp[i-4])
> 
> You mean, for the syncpoint positions?.. I actually have tried that already 
> (only the former suggestion), quite a while ago... and found it to be 
> completely useless... At least for my test video. The numbers were pretty 
> much out of /dev/urandom ...
> 
> Actually, hmm, I tried something different, iirc it was '65536 - sp[i-1]' 
> or something like that... I'll look into this...
> 
> BTW, whats median() ?... 

median(a,b,c) = a+b+c - max(a,b,c) - min(a,b,c)

[...]
-- 
Michael