[FFmpeg-devel] [PATCH 3/3] avcodec/h264_slice: use ff_h264_replace_picture when updating thread contexts

[Andreas Rheinhardt]

James Almer:
> Signed-off-by: James Almer <jamrial at gmail.com>
> ---
>  libavcodec/h264_slice.c | 14 +++++---------
>  1 file changed, 5 insertions(+), 9 deletions(-)
> 
> diff --git a/libavcodec/h264_slice.c b/libavcodec/h264_slice.c
> index a31e804620..0d7107d455 100644
> --- a/libavcodec/h264_slice.c
> +++ b/libavcodec/h264_slice.c
> @@ -379,19 +379,15 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
>      h->droppable            = h1->droppable;
>  
>      for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
> -        ff_h264_unref_picture(h, &h->DPB[i]);
> -        if (h1->DPB[i].f->buf[0] &&
> -            (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
> +        ret = ff_h264_replace_picture(h, &h->DPB[i], &h1->DPB[i]);
> +        if (ret < 0)
>              return ret;
>      }
>  
>      h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
> -    ff_h264_unref_picture(h, &h->cur_pic);
> -    if (h1->cur_pic.f->buf[0]) {
> -        ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
> -        if (ret < 0)
> -            return ret;
> -    }
> +    ret = ff_h264_replace_picture(h, &h->cur_pic, &h1->cur_pic);
> +    if (ret < 0)
> +        return ret;
>  
>      h->enable_er       = h1->enable_er;
>      h->workaround_bugs = h1->workaround_bugs;
> 
I like what you are doing here; I once had the same idea, but IIRC it
was way uglier.
Can you share some numbers on how many allocations are avoided by this?
(I guess that avoiding the allocation will be the normal case for at
least the PPS buffer.) How many would it be with this patch:
https://ffmpeg.org/pipermail/ffmpeg-devel/2020-November/272508.html?

- Andreas