On Tue, May 24, 2022 at 08:50:51PM +0200, Andreas Rheinhardt wrote:
> Michael Niedermayer:
> > Fixes: CID1435168
> > 
> > Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
> > ---
> >  libavcodec/vc1_loopfilter.c | 5 +----
> >  1 file changed, 1 insertion(+), 4 deletions(-)
> > 
> > diff --git a/libavcodec/vc1_loopfilter.c b/libavcodec/vc1_loopfilter.c
> > index 0f990cccef..ee694ede28 100644
> > --- a/libavcodec/vc1_loopfilter.c
> > +++ b/libavcodec/vc1_loopfilter.c
> > @@ -1125,10 +1125,7 @@ static av_always_inline void vc1_b_h_intfi_loop_filter(VC1Context *v, uint8_t *d
> >          dst = dest + (block_num & 2) * 4 * s->linesize + (block_num & 1) * 8;
> >  
> >      if (!(flags & RIGHT_EDGE) || !(block_num & 5)) {
> > -        if (block_num > 3)
> > -            v->vc1dsp.vc1_h_loop_filter8(dst + 8, linesize, pq);
> > -        else
> > -            v->vc1dsp.vc1_h_loop_filter8(dst + 8, linesize, pq);
> > +        v->vc1dsp.vc1_h_loop_filter8(dst + 8, linesize, pq);
> >      }
> >  
> >      tt = ttblk[0] >> (block_num * 4) & 0xf;
> 
> Are you certain that the current code is actually correct or whether
> something else was intended?

iam not certain no but the vc1_b_v_intfi_loop_filter() function also
treats luma and chroma the same

    if(!(flags & BOTTOM_EDGE) || block_num < 2)
        v->vc1dsp.vc1_v_loop_filter8(dst + 8 * linesize, linesize, pq);

this is just the matching function for horizontal instead of vertical

also whan looking at the 4x4 4x8 and 8x4 cases i see no difference
between horizontal and vertical for intfi. But ultimately yes of course
it could be buggy and something else could be correct.

but if i replace this by an abort()
make: *** [fate-vc1_sa10143] Error 134
make: *** [fate-vc1_ilaced_twomv] Error 134

Fate fails, so it should be tested

thx



[...]

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

You can kill me, but you cannot change the truth.