[FFmpeg-devel] [PATCH] lavc/vvc: Only read split_cu_flag if a split is allowed

[Nuo Mi]

On Wed, Apr 3, 2024 at 5:59 PM Frank Plowman <post at frankplowman.com> wrote:

> On 02/04/2024 22:48, Frank Plowman wrote:
> > Add a check to ensure some split is possible before reading the
> > split_cu_flag.  This is present in the spec, in VVCv3 section 7.3.11.4.
> > Its omission could lead to infinite loops and ultimately crashing due to
> > stack overflow.
> > ---
> >   libavcodec/vvc/vvc_ctu.c | 7 ++++++-
> >   1 file changed, 6 insertions(+), 1 deletion(-)
> >
> > diff --git a/libavcodec/vvc/vvc_ctu.c b/libavcodec/vvc/vvc_ctu.c
> > index 8ba12c8d9f..32d8bc8f5c 100644
> > --- a/libavcodec/vvc/vvc_ctu.c
> > +++ b/libavcodec/vvc/vvc_ctu.c
> > @@ -2095,6 +2095,7 @@ static int hls_coding_tree(VVCLocalContext *lc,
> >       const int ch_type               = tree_type_curr ==
> DUAL_TREE_CHROMA;
> >       int ret;
> >       VVCAllowedSplit allowed;
> > +    int split_cu_flag;
> >
> >       if (pps->r->pps_cu_qp_delta_enabled_flag && qg_on_y && cb_sub_div
> <= sh->cu_qp_delta_subdiv) {
> >           lc->parse.is_cu_qp_delta_coded = 0;
> > @@ -2109,7 +2110,11 @@ static int hls_coding_tree(VVCLocalContext *lc,
> >
> >       can_split(lc, x0, y0, cb_width, cb_height, mtt_depth,
> depth_offset, part_idx,
> >           last_split_mode, tree_type_curr, mode_type_curr, &allowed);
> > -    if (ff_vvc_split_cu_flag(lc, x0, y0, cb_width, cb_height, ch_type,
> &allowed)) {
> > +    if (allowed.btv || allowed.bth || allowed.ttv || allowed.tth ||
> allowed.qt)
> > +        split_cu_flag = ff_vvc_split_cu_flag(lc, x0, y0, cb_width,
> cb_height, ch_type, &allowed);
> > +    else
> > +        split_cu_flag = 0;
> > +    if (split_cu_flag) {
> >           VVCSplitMode split      = ff_vvc_split_mode(lc, x0, y0,
> cb_width, cb_height, cqt_depth, mtt_depth, ch_type, &allowed);
> >           VVCModeType mode_type   = mode_type_decode(lc, x0, y0,
> cb_width, cb_height, split, ch_type, mode_type_curr);
> >
>
> Retracting this patch as I missed that this logic is in fact
> implemented, just elsewhere.  There is still a bug here, but it seems
> the condition to trigger it is more complex that I thought.  Should have
> an alternative patch soon.
>
Hi Frank,
Thanks for your patch.
Please forward the clip to me as well. I'll use it to test your patch
during the review.

> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
>