On Mon, Apr 14, 2014 at 11:44:08AM +0200, Niels Möller wrote: > In xll (lossless) data, samples can be split into msb and lsb parts, > coded separately. The spec is a bit unclear on exactly when these parts > are reassembled. > > Should they be reassembled immediately after they are parsed, before the > crunching for inverse prediction and inverse channel decorrelation? Or > should this crunching be applied to the msb samples only, with the lsb > data (which supposedly is more or less uniform random) added in at the > end?
At the very end of reconstructing channel. LSBs are not compressible and thus stored uncompressed and should not be involved in prediction and such. _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
