[FFmpeg-devel] [PATCH] avcodec/ffv1: Support slice coding mode 1 with golomb rice

Michael Niedermayer michael at niedermayer.cc
Sat Oct 26 01:47:19 EEST 2024


On Fri, Oct 25, 2024 at 04:24:04AM +0200, Lynne via ffmpeg-devel wrote:
> On 25/10/2024 03:57, Michael Niedermayer wrote:
> > Sponsored-by: Sovereign Tech Fund
> > Signed-off-by: Michael Niedermayer <michael at niedermayer.cc>
> > ---
> >   libavcodec/ffv1dec.c          | 20 ++++++++++----------
> >   libavcodec/ffv1dec_template.c |  3 +++
> >   libavcodec/ffv1enc.c          | 23 ++++++++++++-----------
> >   3 files changed, 25 insertions(+), 21 deletions(-)
> > 
> > diff --git a/libavcodec/ffv1dec.c b/libavcodec/ffv1dec.c
> > index ac8cc8a31dc..de5abfe9b41 100644
> > --- a/libavcodec/ffv1dec.c
> > +++ b/libavcodec/ffv1dec.c
> > @@ -120,9 +120,8 @@ static int is_input_end(RangeCoder *c, GetBitContext *gb, int ac)
> >   static int decode_plane(FFV1Context *f, FFV1SliceContext *sc,
> >                           GetBitContext *gb,
> >                           uint8_t *src, int w, int h, int stride, int plane_index,
> > -                         int pixel_stride)
> > +                        int pixel_stride, int ac)
> >   {
> > -    const int ac = f->ac;
> >       int x, y;
> >       int16_t *sample[2];
> >       sample[0] = sc->sample_buffer + 3;
> > @@ -273,6 +272,7 @@ static int decode_slice(AVCodecContext *c, void *arg)
> >       AVFrame * const p = f->picture.f;
> >       const int      si = sc - f->slices;
> >       GetBitContext gb;
> > +    int ac = f->ac || sc->slice_coding_mode == 1;
> >       if (!(p->flags & AV_FRAME_FLAG_KEY) && f->last_picture.f)
> >           ff_progress_frame_await(&f->last_picture, si);
> > @@ -305,7 +305,7 @@ static int decode_slice(AVCodecContext *c, void *arg)
> >       x      = sc->slice_x;
> >       y      = sc->slice_y;
> > -    if (f->ac == AC_GOLOMB_RICE) {
> > +    if (ac == AC_GOLOMB_RICE) {
> >           if (f->version == 3 && f->micro_version > 1 || f->version > 3)
> >               get_rac(&sc->c, (uint8_t[]) { 129 });
> >           sc->ac_byte_count = f->version > 2 || (!x && !y) ? sc->c.bytestream - sc->c.bytestream_start - 1 : 0;
> > @@ -320,17 +320,17 @@ static int decode_slice(AVCodecContext *c, void *arg)
> >           const int chroma_height = AV_CEIL_RSHIFT(height, f->chroma_v_shift);
> >           const int cx            = x >> f->chroma_h_shift;
> >           const int cy            = y >> f->chroma_v_shift;
> > -        decode_plane(f, sc, &gb, p->data[0] + ps*x + y*p->linesize[0], width, height, p->linesize[0], 0, 1);
> > +        decode_plane(f, sc, &gb, p->data[0] + ps*x + y*p->linesize[0], width, height, p->linesize[0], 0, 1, ac);
> >           if (f->chroma_planes) {
> > -            decode_plane(f, sc, &gb, p->data[1] + ps*cx+cy*p->linesize[1], chroma_width, chroma_height, p->linesize[1], 1, 1);
> > -            decode_plane(f, sc, &gb, p->data[2] + ps*cx+cy*p->linesize[2], chroma_width, chroma_height, p->linesize[2], 1, 1);
> > +            decode_plane(f, sc, &gb, p->data[1] + ps*cx+cy*p->linesize[1], chroma_width, chroma_height, p->linesize[1], 1, 1, ac);
> > +            decode_plane(f, sc, &gb, p->data[2] + ps*cx+cy*p->linesize[2], chroma_width, chroma_height, p->linesize[2], 1, 1, ac);
> >           }
> >           if (f->transparency)
> > -            decode_plane(f, sc, &gb, p->data[3] + ps*x + y*p->linesize[3], width, height, p->linesize[3], (f->version >= 4 && !f->chroma_planes) ? 1 : 2, 1);
> > +            decode_plane(f, sc, &gb, p->data[3] + ps*x + y*p->linesize[3], width, height, p->linesize[3], (f->version >= 4 && !f->chroma_planes) ? 1 : 2, 1, ac);
> >       } else if (f->colorspace == 0) {
> > -         decode_plane(f, sc, &gb, p->data[0] + ps*x + y*p->linesize[0]    , width, height, p->linesize[0], 0, 2);
> > -         decode_plane(f, sc, &gb, p->data[0] + ps*x + y*p->linesize[0] + 1, width, height, p->linesize[0], 1, 2);
> > +         decode_plane(f, sc, &gb, p->data[0] + ps*x + y*p->linesize[0]    , width, height, p->linesize[0], 0, 2, ac);
> > +         decode_plane(f, sc, &gb, p->data[0] + ps*x + y*p->linesize[0] + 1, width, height, p->linesize[0], 1, 2, ac);
> >       } else if (f->use32bit) {
> >           uint8_t *planes[4] = { p->data[0] + ps * x + y * p->linesize[0],
> >                                  p->data[1] + ps * x + y * p->linesize[1],
> > @@ -344,7 +344,7 @@ static int decode_slice(AVCodecContext *c, void *arg)
> >                                  p->data[3] + ps * x + y * p->linesize[3] };
> >           decode_rgb_frame(f, sc, &gb, planes, width, height, p->linesize);
> >       }
> > -    if (f->ac != AC_GOLOMB_RICE && f->version > 2) {
> > +    if (ac != AC_GOLOMB_RICE && f->version > 2) {
> >           int v;
> >           get_rac(&sc->c, (uint8_t[]) { 129 });
> >           v = sc->c.bytestream_end - sc->c.bytestream - 2 - 5*!!f->ec;
> > diff --git a/libavcodec/ffv1dec_template.c b/libavcodec/ffv1dec_template.c
> > index 2da6bd935dc..e983d1ba648 100644
> > --- a/libavcodec/ffv1dec_template.c
> > +++ b/libavcodec/ffv1dec_template.c
> > @@ -143,6 +143,9 @@ static int RENAME(decode_rgb_frame)(FFV1Context *f, FFV1SliceContext *sc,
> >       int transparency = f->transparency;
> >       int ac = f->ac;
> > +    if (sc->slice_coding_mode == 1)
> > +        ac = 1;
> > +
> >       for (x = 0; x < 4; x++) {
> >           sample[x][0] = RENAME(sc->sample_buffer) +  x * 2      * (w + 6) + 3;
> >           sample[x][1] = RENAME(sc->sample_buffer) + (x * 2 + 1) * (w + 6) + 3;
> > diff --git a/libavcodec/ffv1enc.c b/libavcodec/ffv1enc.c
> > index a32059886c0..547fb598d6d 100644
> > --- a/libavcodec/ffv1enc.c
> > +++ b/libavcodec/ffv1enc.c
> > @@ -271,10 +271,9 @@ static inline void put_vlc_symbol(PutBitContext *pb, VlcState *const state,
> >   static int encode_plane(FFV1Context *f, FFV1SliceContext *sc,
> >                           const uint8_t *src, int w, int h,
> > -                         int stride, int plane_index, int pixel_stride)
> > +                         int stride, int plane_index, int pixel_stride, int ac)
> >   {
> >       int x, y, i, ret;
> > -    const int ac = f->ac;
> >       const int pass1 = !!(f->avctx->flags & AV_CODEC_FLAG_PASS1);
> >       const int ring_size = f->context_model ? 3 : 2;
> >       int16_t *sample[3];
> > @@ -1068,6 +1067,7 @@ static int encode_slice(AVCodecContext *c, void *arg)
> >                                   p->data[1] ? p->data[1] + ps*x + y*p->linesize[1] : NULL,
> >                                   p->data[2] ? p->data[2] + ps*x + y*p->linesize[2] : NULL,
> >                                   p->data[3] ? p->data[3] + ps*x + y*p->linesize[3] : NULL};
> > +    int ac = f->ac;
> >       sc->slice_coding_mode = 0;
> >       if (f->version > 3 && f->colorspace == 1) {
> > @@ -1083,7 +1083,7 @@ retry:
> >       if (f->version > 2) {
> >           encode_slice_header(f, sc);
> >       }
> > -    if (f->ac == AC_GOLOMB_RICE) {
> > +    if (ac == AC_GOLOMB_RICE) {
> >           sc->ac_byte_count = f->version > 2 || (!x && !y) ? ff_rac_terminate(&sc->c, f->version > 2) : 0;
> >           init_put_bits(&sc->pb,
> >                         sc->c.bytestream_start + sc->ac_byte_count,
> > @@ -1096,24 +1096,24 @@ retry:
> >           const int cx            = x >> f->chroma_h_shift;
> >           const int cy            = y >> f->chroma_v_shift;
> > -        ret = encode_plane(f, sc, p->data[0] + ps*x + y*p->linesize[0], width, height, p->linesize[0], 0, 1);
> > +        ret = encode_plane(f, sc, p->data[0] + ps*x + y*p->linesize[0], width, height, p->linesize[0], 0, 1, ac);
> >           if (f->chroma_planes) {
> > -            ret |= encode_plane(f, sc, p->data[1] + ps*cx+cy*p->linesize[1], chroma_width, chroma_height, p->linesize[1], 1, 1);
> > -            ret |= encode_plane(f, sc, p->data[2] + ps*cx+cy*p->linesize[2], chroma_width, chroma_height, p->linesize[2], 1, 1);
> > +            ret |= encode_plane(f, sc, p->data[1] + ps*cx+cy*p->linesize[1], chroma_width, chroma_height, p->linesize[1], 1, 1, ac);
> > +            ret |= encode_plane(f, sc, p->data[2] + ps*cx+cy*p->linesize[2], chroma_width, chroma_height, p->linesize[2], 1, 1, ac);
> >           }
> >           if (f->transparency)
> > -            ret |= encode_plane(f, sc, p->data[3] + ps*x + y*p->linesize[3], width, height, p->linesize[3], 2, 1);
> > +            ret |= encode_plane(f, sc, p->data[3] + ps*x + y*p->linesize[3], width, height, p->linesize[3], 2, 1, ac);
> >       } else if (c->pix_fmt == AV_PIX_FMT_YA8) {
> > -        ret  = encode_plane(f, sc, p->data[0] +     ps*x + y*p->linesize[0], width, height, p->linesize[0], 0, 2);
> > -        ret |= encode_plane(f, sc, p->data[0] + 1 + ps*x + y*p->linesize[0], width, height, p->linesize[0], 1, 2);
> > +        ret  = encode_plane(f, sc, p->data[0] +     ps*x + y*p->linesize[0], width, height, p->linesize[0], 0, 2, ac);
> > +        ret |= encode_plane(f, sc, p->data[0] + 1 + ps*x + y*p->linesize[0], width, height, p->linesize[0], 1, 2, ac);
> >       } else if (f->use32bit) {
> >           ret = encode_rgb_frame32(f, sc, planes, width, height, p->linesize);
> >       } else {
> >           ret = encode_rgb_frame(f, sc, planes, width, height, p->linesize);
> >       }
> > -    if (f->ac != AC_GOLOMB_RICE) {
> > +    if (ac != AC_GOLOMB_RICE) {
> >           sc->ac_byte_count = ff_rac_terminate(&sc->c, 1);
> >       } else {
> >           flush_put_bits(&sc->pb); // FIXME: nicer padding
> > @@ -1122,11 +1122,12 @@ retry:
> >       if (ret < 0) {
> >           av_assert0(sc->slice_coding_mode == 0);
> > -        if (f->version < 4 || !f->ac) {
> > +        if (f->version < 4) {
> >               av_log(c, AV_LOG_ERROR, "Buffer too small\n");
> >               return ret;
> >           }
> >           av_log(c, AV_LOG_DEBUG, "Coding slice as PCM\n");
> > +        ac = 1;
> >           sc->slice_coding_mode = 1;
> >           sc->c = c_bak;
> >           goto retry;
> 
> Is there a reason to support this with Golomb? Being a fallback-only mode
> for range coding, I don't think there's much point in supporting Golomb
> coding for it.

The reason behind slice_coding_mode is to ensure that the maximum size
of the bitstream can be bound in a reasonable way.
This will allows decoders and encoders to use a known and limited amount of
memory.
This is of course not a problem with normal videos but one could
craft a video that pushes each context state to one side and then
follows that up with crafted data where each is opposit thus generating
the worst case. Its a bigger issue for the range coder than golomb, yes

thx

[...]

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Those who are too smart to engage in politics are punished by being
governed by those who are dumber. -- Plato 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 195 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20241026/dedcbcfd/attachment.sig>


More information about the ffmpeg-devel mailing list