[MPlayer-dev-eng] Lots of stuff for NUT

[Oded Shimon]

On Tue, Jan 03, 2006 at 08:31:57AM +0200, Oded Shimon wrote:
> 1. do the pts and back_ptr per stream in syncpoints. I want to AT LEAST try 
>    this out and really see how much overhead it is, and I'd like you guys 
>    to try to think up way to reduce overhead for it.
> 
> 
> I'm now gonna finally check how much overhead this idea takes.

OK, first of all, here's the spec:

syncpoint:
	syncpoint_startcode					u(64)
	coded_pts						v
	stream = coded_pts % stream_count
	global_pts = coded_pts / stream_count
	for (i=0; i<stream_count; i++) {
		if (i != stream) {
			coded_pts				v
			back_ptr_div8				v
			pts[i] = covert_ts(global_pts, timebase[stream], timebase[i])
			pts[i] -= coded_pts
		} else {
			back_ptr_div8				v
			pts[i] = global_pts
		}
	}


Now, here are the numbers:

just pts in global timebase:     Syncpoints: 27674 size: 315730
single back_ptr and pts:         Syncpoints: 27678 size: 391804
single pts, per stream back_ptr: Syncpoints: 27675 size: 441061
The above spec:                  Syncpoints: 27676 size: 459898


All of these are for a normal 2 stream 700mb audio video file.

Unless you proove otherwise Michael, back_ptr is essential, so the 
additional 70kb is not horribly bloating, and it meets every one of the 
goals. Also, assuming <200kb index, the total overhead for the entire 700mb 
file is less than 1mb, this is still far less than ANY container that 
exists today. (mkv with lacing is 1.5mb for this file, without lacing is 
2.5mb iirc)

- ods15