[FFmpeg-devel] [PATCH 4/5] lavc: skip initial silence when requested

[Nicolas George]

Le septidi 17 messidor, an CCXX, Michael Niedermayer a écrit :
> Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
> ---
>  libavcodec/internal.h |    5 +++++
>  libavcodec/utils.c    |   18 ++++++++++++++++++
>  2 files changed, 23 insertions(+)
> 
> diff --git a/libavcodec/internal.h b/libavcodec/internal.h
> index 49c1a18..063aeca 100644
> --- a/libavcodec/internal.h
> +++ b/libavcodec/internal.h
> @@ -84,6 +84,11 @@ typedef struct AVCodecInternal {
>      unsigned int byte_buffer_size;
>  
>      void *frame_thread_encoder;
> +
> +    /**
> +     * Number of audio samples to skip at in the decoder

I can not parse this phrase. "Number of audio samples to skip at the start
of the next decoded frame"?

> +     */
> +    int skip_samples;
>  } AVCodecInternal;
>  
>  struct AVCodecDefault {
> diff --git a/libavcodec/utils.c b/libavcodec/utils.c
> index b3af77d..86fa3a8 100644
> --- a/libavcodec/utils.c
> +++ b/libavcodec/utils.c
> @@ -1611,6 +1611,8 @@ int attribute_align_arg avcodec_decode_audio4(AVCodecContext *avctx,
>      }
>  
>      if ((avctx->codec->capabilities & CODEC_CAP_DELAY) || avpkt->size) {
> +        uint8_t *side;
> +        int side_size;
>          // copy to ensure we do not change avpkt
>          AVPacket tmp = *avpkt;
>          int did_split = av_packet_split_side_data(&tmp);
> @@ -1629,6 +1631,22 @@ int attribute_align_arg avcodec_decode_audio4(AVCodecContext *avctx,
>                  frame->sample_rate = avctx->sample_rate;
>          }
>  
> +        side= av_packet_get_side_data(avctx->pkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
> +        if(side && side_size>=10) {
> +            avctx->internal->skip_samples = AV_RL32(side);
> +        }
> +        if (avctx->internal->skip_samples) {

Correct me if I am wrong: you designed the API so that
internal->skip_samples can also be set by another component, codec->decode
for example?

> +            if(frame->nb_samples <= avctx->internal->skip_samples){
> +                *got_frame_ptr = 0;
> +                avctx->internal->skip_samples -= frame->nb_samples;
> +            } else {
> +                av_samples_copy(frame->extended_data, frame->extended_data, 0, avctx->internal->skip_samples,
> +                                frame->nb_samples - avctx->internal->skip_samples, avctx->channels, frame->format);

av_samples_copy uses memcpy, not memmove. If skip_samples < nb_samples/2,
the source and destination will overlap.

> +                frame->nb_samples -= avctx->internal->skip_samples;
> +                avctx->internal->skip_samples = 0;

What about timestamps? If skip_samples is set by the decoder rather than the
demuxer (possibly for convergence reasons after a call to flush),
frame->pkt_pts will be the time of the first discarded sample.

By the way, I notice that best_effort_timestamp is not set for audio. That
may be just the solution.

> +            }
> +        }
> +
>          avctx->pkt = NULL;
>          if (did_split) {
>              ff_packet_free_side_data(&tmp);

Regards,

-- 
  Nicolas George
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20120705/d7526cf6/attachment.asc>