[FFmpeg-devel] [PATCH 2/2] lavc/videotoolboxenc: set extradata when opening codec

[Richard Kern]

> On Jun 8, 2016, at 5:48 AM, Michael Niedermayer <michael at niedermayer.cc> wrote:
> 
> On Thu, Jun 02, 2016 at 02:43:56AM -0400, Rick Kern wrote:
>> VideoToolbox doesn't supply parameter sets until the first frame is done
>> encoding. This spins up a temporary encoder and encodes a single frame to
>> get this data.
>> 
>> Signed-off-by: Rick Kern <kernrj at gmail.com>
>> ---
>> libavcodec/videotoolboxenc.c | 326 ++++++++++++++++++++++++++++++++-----------
>> 1 file changed, 241 insertions(+), 85 deletions(-)
> [...]
> 
>> @@ -1753,6 +1801,114 @@ end_nopkt:
>>     return status;
>> }
>> 
>> +static int vtenc_populate_extradata(AVCodecContext   *avctx,
>> +                                    CMVideoCodecType codec_type,
>> +                                    CFStringRef      profile_level,
>> +                                    CFNumberRef      gamma_level,
>> +                                    CFDictionaryRef  enc_info,
>> +                                    CFDictionaryRef  pixel_buffer_info)
>> +{
>> +    VTEncContext *vtctx = avctx->priv_data;
>> +    AVFrame *frame = av_frame_alloc();
>> +    int y_size = avctx->width * avctx->height;
>> +    int chroma_size = (avctx->width / 2) * (avctx->height / 2);
>> +    CMSampleBufferRef buf = NULL;
>> +    int status;
>> +
>> +    if (!frame)
>> +        return AVERROR(ENOMEM);
>> +
>> +    frame->buf[0] = av_buffer_alloc(y_size + 2 * chroma_size);
>> +
>> +    if(!frame->buf[0]){
>> +        status = AVERROR(ENOMEM);
>> +        goto pe_cleanup;
>> +    }
>> +
>> +    status = vtenc_create_encoder(avctx,
>> +                                  codec_type,
>> +                                  profile_level,
>> +                                  gamma_level,
>> +                                  enc_info,
>> +                                  pixel_buffer_info,
>> +                                  &vtctx->session);
>> +    if (status)
>> +        goto pe_cleanup;
>> +
>> +    frame->data[0] = frame->buf[0]->data;
>> +    memset(frame->data[0],   0,      y_size);
>> +
>> +    frame->data[1] = frame->buf[0]->data + y_size;
>> +    memset(frame->data[1], 128, chroma_size);
>> +
>> +
>> +    if (avctx->pix_fmt == AV_PIX_FMT_YUV420P) {
>> +        frame->data[2] = frame->buf[0]->data + y_size + chroma_size;
>> +        memset(frame->data[2], 128, chroma_size);
>> +    }
>> +
>> +    frame->linesize[0] = avctx->width;
>> +
>> +    if (avctx->pix_fmt == AV_PIX_FMT_YUV420P) {
>> +        frame->linesize[1] =
>> +        frame->linesize[2] = (avctx->width + 1) / 2;
>> +    } else {
>> +        frame->linesize[1] = (avctx->width + 1) / 2;
>> +    }
>> +
>> +    frame->format          = avctx->pix_fmt;
>> +    frame->width           = avctx->width;
>> +    frame->height          = avctx->height;
> 
>> +    frame->colorspace      = avctx->colorspace;
> 
> Iam not sure the docs fully match current intend but
> colorspace and color_range are documented to be only accessed
> through av_frame_get_color_range(), av_frame_set_color_range(),
> av_frame_get_colorspace() and av_frame_set_colorspace()

I’ll change these locally and push tomorrow if no one else reviews.

> 
> 
> [...]
> -- 
> Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
> 
> While the State exists there can be no freedom; when there is freedom there
> will be no State. -- Vladimir Lenin
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel