On Thu, 2007-11-22 at 18:25 +0100, Burkhard Plaum wrote: > > Maybe the codec libraries need some NLE extensions to support "sloppy" > > decoding, at least for the most common aquisition codecs. > > Probably not the codecs themselves, but the surrounding libraries (like > quicktime4linux). > > > For example, instead of asking the codec for a particular frame, one > > asks for a range, and gets the frame that is fastest to retrieve within > > that range. > > Or make the library cache entire GOPs. When reading the lastest qt4l > annoucement > at heroinewarrior.com: > > "Frames are cached when reading the first frame after a seek. > This allows faster reverse playback." > > Isn't that what we are talking about?
Yes, that helps. I was not talking only about caching frames for playback, but also caching frames that are needed for decoding other frames. > > The codec also needs to cache frames, or use Cinelerra's frame cache > > to avoid a lot of repeaded decoding. If the codec has to decode frame > > 2-4 to deliver frame 5, it should be able to reuse those for decoding > > frame 1 or frame 6 later. (memory or disk footprint permitting) > > Ok, regarding NLE I'm more a user and know little about the internals. > > So my question is, how often are frames needed out of order? Certainly > for reverse playback, that's clear. > But during rendering, you'll have mostly many small pieces, which are, > however, decoded sequentially (and thus don't need caching). Not meaning, > that advanced mechanisms are unneccesary, but to what extent are "real life" > situations affected by that? Mostly "scrubbing"; playback in both directions at varying speed, with as little lag as possible. Seeking around large DV files is really responsive in Cinelerra on a fast computer, and this means a great deal for the user experience. Quick and responsive scrubbing of large HDV files requires a bit more cunning than DV. Caching alone will not be enough, since the user often will be zipping around parts of the video files for the first time. This applies to the usecase of much footage that has to be edited quickly. I think Cinelerra ought to support that reasonably well, also for temporally compressed video. -- Herman Robak _______________________________________________ Cinelerra mailing list [email protected] https://init.linpro.no/mailman/skolelinux.no/listinfo/cinelerra
