[FFmpeg-devel] [PATCH] h264: Support multi-field closed captions by using AVBufferRef and not resetting per field

[James Almer]

On 8/16/2018 7:52 AM, joshdk at ob-encoder.com wrote:
> From: Kieran Kunhya <kierank at obe.tv>
> 
> ---
> This is also useful when CC is fully contained in the first field.
> 
>  libavcodec/h264_sei.c   | 15 ++++++++-------
>  libavcodec/h264_sei.h   |  3 +--
>  libavcodec/h264_slice.c | 15 ++++++++++-----
>  libavcodec/h264dec.c    |  5 +++--
>  4 files changed, 22 insertions(+), 16 deletions(-)
> 
> diff --git a/libavcodec/h264_sei.c b/libavcodec/h264_sei.c
> index 6499086210..43593d34d2 100644
> --- a/libavcodec/h264_sei.c
> +++ b/libavcodec/h264_sei.c
> @@ -51,8 +51,7 @@ void ff_h264_sei_uninit(H264SEIContext *h)
>      h->display_orientation.present = 0;
>      h->afd.present                 =  0;
>  
> -    h->a53_caption.a53_caption_size = 0;
> -    av_freep(&h->a53_caption.a53_caption);
> +    av_buffer_unref(&h->a53_caption.buf_ref);
>  }
>  
>  static int decode_picture_timing(H264SEIPictureTiming *h, GetBitContext *gb,
> @@ -169,7 +168,8 @@ static int decode_registered_user_data_closed_caption(H264SEIA53Caption *h,
>              size -= 2;
>  
>              if (cc_count && size >= cc_count * 3) {
> -                const uint64_t new_size = (h->a53_caption_size + cc_count
> +                int old_size = h->buf_ref ? h->buf_ref->size : 0;
> +                const uint64_t new_size = (old_size + cc_count
>                                             * UINT64_C(3));
>                  int i, ret;
>  
> @@ -177,14 +177,15 @@ static int decode_registered_user_data_closed_caption(H264SEIA53Caption *h,
>                      return AVERROR(EINVAL);
>  
>                  /* Allow merging of the cc data from two fields. */
> -                ret = av_reallocp(&h->a53_caption, new_size);
> +                ret = av_buffer_realloc(&h->buf_ref, new_size);
>                  if (ret < 0)
>                      return ret;
>  
> +                /* Use of av_buffer_realloc assumes buffer is writeable */
>                  for (i = 0; i < cc_count; i++) {
> -                    h->a53_caption[h->a53_caption_size++] = get_bits(gb, 8);
> -                    h->a53_caption[h->a53_caption_size++] = get_bits(gb, 8);
> -                    h->a53_caption[h->a53_caption_size++] = get_bits(gb, 8);
> +                    h->buf_ref->data[old_size++] = get_bits(gb, 8);
> +                    h->buf_ref->data[old_size++] = get_bits(gb, 8);
> +                    h->buf_ref->data[old_size++] = get_bits(gb, 8);
>                  }
>  
>                  skip_bits(gb, 8);   // marker_bits
> diff --git a/libavcodec/h264_sei.h b/libavcodec/h264_sei.h
> index a169a10e49..5b7c8ef9d8 100644
> --- a/libavcodec/h264_sei.h
> +++ b/libavcodec/h264_sei.h
> @@ -95,8 +95,7 @@ typedef struct H264SEIAFD {
>  } H264SEIAFD;
>  
>  typedef struct H264SEIA53Caption {
> -    int a53_caption_size;
> -    uint8_t *a53_caption;
> +    AVBufferRef *buf_ref;

Nit: buf is shorter.

>  } H264SEIA53Caption;
>  
>  typedef struct H264SEIUnregistered {
> diff --git a/libavcodec/h264_slice.c b/libavcodec/h264_slice.c
> index ede9a1a6ea..7bbe65878f 100644
> --- a/libavcodec/h264_slice.c
> +++ b/libavcodec/h264_slice.c
> @@ -429,6 +429,12 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
>                         MAX_DELAYED_PIC_COUNT + 2, h, h1);
>  
>      h->frame_recovered       = h1->frame_recovered;
> +    if (h1->sei.a53_caption.buf_ref) {
> +        h->sei.a53_caption.buf_ref = av_buffer_ref(h1->sei.a53_caption.buf_ref);

Missing check for failure.

> +        av_buffer_unref(&h1->sei.a53_caption.buf_ref);

Does the reference in h1 need to be freed? h1 does not seem to be
modified in this function anywhere else.

> +    }
> +    else
> +        h->sei.a53_caption.buf_ref = NULL;

Is it not NULL already in this case?

>  
>      if (!h->cur_pic_ptr)
>          return 0;
> @@ -1269,15 +1275,14 @@ static int h264_export_frame_props(H264Context *h)
>          }
>      }
>  
> -    if (h->sei.a53_caption.a53_caption) {
> +    if (h->sei.a53_caption.buf_ref) {
>          H264SEIA53Caption *a53 = &h->sei.a53_caption;
>          AVFrameSideData *sd = av_frame_new_side_data(cur->f,
>                                                       AV_FRAME_DATA_A53_CC,
> -                                                     a53->a53_caption_size);
> +                                                     a53->buf_ref->size);
>          if (sd)
> -            memcpy(sd->data, a53->a53_caption, a53->a53_caption_size);
> -        av_freep(&a53->a53_caption);
> -        a53->a53_caption_size = 0;
> +            memcpy(sd->data, a53->buf_ref->data, a53->buf_ref->size);
> +        av_buffer_unref(&a53->buf_ref);

Use av_frame_new_side_data_from_buf() instead. It takes ownership of the
provided AVBufferRef instead of creating a new one for itself, so it's
ideal for this scenario. Saves you one memcpy.

>          h->avctx->properties |= FF_CODEC_PROPERTY_CLOSED_CAPTIONS;
>      }
>  
> diff --git a/libavcodec/h264dec.c b/libavcodec/h264dec.c
> index 7494c7a8f2..8d115fa040 100644
> --- a/libavcodec/h264dec.c
> +++ b/libavcodec/h264dec.c
> @@ -609,9 +609,10 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size)
>  
>      if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS)) {
>          h->current_slice = 0;
> -        if (!h->first_field)
> +        if (!h->first_field) {
>              h->cur_pic_ptr = NULL;
> -        ff_h264_sei_uninit(&h->sei);
> +            ff_h264_sei_uninit(&h->sei);
> +        }
>      }
>  
>      if (h->nal_length_size == 4) {
>