On 8/31/17, Nicolas George <geo...@nsup.org> wrote:
> L'octidi 8 fructidor, an CCXXV, Paul B Mahol a ecrit :
>> + if ((ret = ff_inlink_consume_frame(ctx->inputs[0], &in[0])) > 0) {
>> + av_audio_fifo_write(s->fifo[0], (void **)in[0]->extended_data,
>> + in[0]->nb_samples);
>> + av_frame_free(&in[0]);
>> + }
>> + if (ret < 0)
>> + return ret;
>> + if ((ret = ff_inlink_consume_frame(ctx->inputs[1], &in[1])) > 0) {
>> + av_audio_fifo_write(s->fifo[1], (void **)in[1]->extended_data,
>> + in[1]->nb_samples);
>> + av_frame_free(&in[1]);
>> + }
>> + if (ret < 0)
>> + return ret;
>
> Just to be sure all is explained correctly:
>
> With this patch, you are certainly enhancing the scheduling, but it is
> not taking full advantage of the new design: with activate(), all links
> have a frame FIFO with unlimited capacity (and plans to add options to
> manage that capacity globally). In this snippet of code, you move the
> frames from this FIFO into another FIFO built in the private context. In
> other words, you are nullifying the benefit of the link's built-in FIFO.
>
> A complete conversion to activate() would require working from the
> link's FIFO directly, and getting rid of the code with AVAudioFifo
> altogether.
>
> It is more work, and since it works that way it is fine to leave it for
> later and reap the easy benefits immediately. But I wanted to make sure
> you were aware of that.
>
> The same applies to af_sidechaincompress.
I did it because it started hanging like overlay filter.
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
