[MPlayer-dev-eng] Lots of stuff for NUT [new simplification, lower overhead]

[Michael Niedermayer]

Hi

On Thu, Jan 05, 2006 at 03:22:05PM -0500, Rich Felker wrote:
> OK instead of just flaming I'm going to try to be useful. 

:(((((((


> Here is a
> proposal to reduce the size and complexity of the index:
> 
> syncpoint_rel_pos                       v
> keyframe_bitmask                        v
> for (i=0; i<nstreams; i++)
>     if (keyframe_bitmask & (1<<i))
>         stream_pts                      v
> 
> Feel free to incorporate existing methods of compressing stream_pts.
> The basic idea is that back_ptr is redundant information and need not
> be included in the index. It's sufficient to know, for each stream,
> whether that stream has had a keyframe since the last syncpoint
> (indicated by keyframe_bitmask). From this information you can
> reconstruct back_ptrs for each stream as long as you've already read
> the earlier part of the index.
> 
> IIRC this is similar to something Michael or maybe Ivan proposed a
> long time ago, or maybe the same.

hmm, i dont remember i would say it was your or ivans idea

one problem with this is that you will have problems with >64 streams

[...]


> There may be a way we can eliminate per-stream pts in the
> syncpoints/index and still get the perfect seeking I want but I'm not
> sure. Would a solution that possibly adds rules for syncpoint
> placement or frame ordering be acceptable, provided it doesn't
> increase overhead?

depends upon the rule i guess ...

[...]
-- 
Michael