[FFmpeg-devel] [PATCH] libavfilter/af_atempo: offset output frames' pts by first_frame_pts / tempo

[Pavel Koshevoy]

On Thu, Jan 24, 2019 at 5:49 AM Paweł Wegner <pawel.wegner95 at gmail.com> wrote:
>
> Signed-off-by: Paweł Wegner <pawel.wegner95 at gmail.com>
> ---
>  libavfilter/af_atempo.c | 9 +++++++++
>  1 file changed, 9 insertions(+)
>
> diff --git a/libavfilter/af_atempo.c b/libavfilter/af_atempo.c
> index bfdad7d76b..1245eae8c1 100644
> --- a/libavfilter/af_atempo.c
> +++ b/libavfilter/af_atempo.c
> @@ -147,6 +147,8 @@ typedef struct ATempoContext {
>      uint8_t *dst_end;
>      uint64_t nsamples_in;
>      uint64_t nsamples_out;
> +
> +    int64_t first_frame_pts;
>  } ATempoContext;
>
>  #define YAE_ATEMPO_MIN 0.5
> @@ -994,6 +996,7 @@ static av_cold int init(AVFilterContext *ctx)
>      ATempoContext *atempo = ctx->priv;
>      atempo->format = AV_SAMPLE_FMT_NONE;
>      atempo->state  = YAE_LOAD_FRAGMENT;
> +    atempo->first_frame_pts = AV_NOPTS_VALUE;
>      return 0;
>  }
>
> @@ -1069,6 +1072,7 @@ static int push_samples(ATempoContext *atempo,
>
>      // adjust the PTS:
>      atempo->dst_buffer->pts =
> +        (atempo->first_frame_pts == AV_NOPTS_VALUE ? 0 : atempo->first_frame_pts / atempo->tempo) +
>          av_rescale_q(atempo->nsamples_out,
>                       (AVRational){ 1, outlink->sample_rate },
>                       outlink->time_base);
> @@ -1108,6 +1112,11 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *src_buffer)
>
>              atempo->dst = atempo->dst_buffer->data[0];
>              atempo->dst_end = atempo->dst + n_out * atempo->stride;
> +
> +            if (atempo->first_frame_pts == AV_NOPTS_VALUE)
> +                atempo->first_frame_pts = av_rescale_q(atempo->dst_buffer->pts,
> +                                                       inlink->time_base,
> +                                                       outlink->time_base);
>          }
>
>          yae_apply(atempo, &src, src_end, &atempo->dst, atempo->dst_end);
> --
> 2.17.1
>


Probably okay. The reason I didn't do this to begin with is because
this is an audio stream filter... and how the timeline of the stream
was transformed up to the 1st frame is unknown.  You are making the
assumption that it should have been transformed using the same tempo
parameter as current tempo, but (video) tempo can be varied at runtime
prior to 1st audio frame, so pts_0' = pts_0 / tempo could be wrong.

Anyway, I don't have a use case where this change would break
something, so if this fixes something for you then it's fine.

    Pavel.