[FFmpeg-devel] [PATCH v4] avcodec/libvpxenc: add VP9 temporal scalability encoding option

[James Zern]

On Fri, Jan 17, 2020 at 1:56 PM Wonkap Jang
<wonkap-at-google.com at ffmpeg.org> wrote:
>
> Hi James,
>
> On Fri, Jan 17, 2020 at 1:50 PM Wonkap Jang <wonkap at google.com> wrote:
>
> > This commit reuses the configuration options for VP8 that enables
> > temporal scalability for VP9. It also adds a way to enable three
> > preset temporal structures (refer to the documentation for more
> > detail) that can be used in offline encoding.
> > ---
> >  doc/encoders.texi      |  18 ++-
> >  libavcodec/libvpxenc.c | 250 +++++++++++++++++++++++++++++++++++++----
> >  2 files changed, 242 insertions(+), 26 deletions(-)
> >

I realized I missed a couple things previously.

> > [...]
> > +static int vpx_ts_param_parse(VPxContext *ctx, struct vpx_codec_enc_cfg
> > *enccfg,
> > +                              char *key, char *value, enum AVCodecID
> > codec_id)
> >  {
> >      size_t value_len = strlen(value);
> > +    int ts_layering_mode = 0;
> >
> >      if (!value_len)
> >          return -1;
> >
> >      if (!strcmp(key, "ts_number_layers"))

There isn't much validation here, libvpx should provide that, but can you try
some incompatible input? ts_layering_mode is defined up to 3, so
ts_number_layers > 3 should fail on init.

> > [...]
> > +
> > +#if (VPX_ENCODER_ABI_VERSION >= 12) && CONFIG_LIBVPX_VP9_ENCODER
> > +    enccfg->temporal_layering_mode = 1; // only bypass mode is supported

There's VP9E_TEMPORAL_LAYERING_MODE_BYPASS for this.

> > for now.
> > +    enccfg->ss_number_layers = 1; // TODO: add spatial scalability
> > support.
> > +#endif
> > +    if (ts_layering_mode) {
> > +        // make sure the ts_layering_mode comes at the end of the
> > ts_parameter string to ensure that
> > +        // correct configuration is done.
> > +        ctx->ts_layer_flags = av_malloc(sizeof(*ctx->ts_layer_flags) *
> > VPX_TS_MAX_PERIODICITY);

This would be better written as av_malloc_array. Looking at the use of the
array, it doesn't look like it needs to be zeroed, but there's av_mallocz_array
for that.