On 08/01/14 08:35, Luca Barbato wrote: > On 29/12/13 17:10, Hendrik Leppkes wrote: >> On Sun, Dec 29, 2013 at 4:54 PM, Luca Barbato <[email protected]> wrote: >>> The Xvid-derived idct is known to cause blurry/garbled results. >>> >>> Bug-Id: vlc/7411 >>> --- >>> libavcodec/mpeg4videodec.c | 10 ---------- >>> 1 file changed, 10 deletions(-) >>> >> >> I know you usually refuse to, but did you look at how the code looks >> in ffmpeg today, since ffmpeg works fine on the sample and still uses >> the xvid idct? > > I doubt it uses that idct on arm. > >> Judging from their changes, and this one, this is more of a guess then >> an actual fix, since it appears todays libav code would just not do >> the idct switch properly, instead of the idct actually being bad. > > Since you read the code already why you aren't providing a patch? > >> If something was encoded with the xvid dct, shouldn't the idct also match >> that? > > Should we import also the C version of xvid idct so we stay consistent?
So I spent 2 hours looking at what Michael did, why he did that and how brain damaged is that bit of code from start. The problem is due the fact we use the idct_permutation while reading matrixes (and possibly something else I stopped at that) and we do that before we reach the point we know the sample is supposedly xvid someversion. So then we switch idct and we have a mismatch between permutations. That said I'm still convinced my patch is saner. lu _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
