[FFmpeg-devel] [PATCH 1/2] lavd/alsa: fix timestamp calculation

[Lukasz M]

On 26 October 2013 11:46, Nicolas George <george at nsup.org> wrote:

> Le quintidi 5 brumaire, an CCXXII, Lukasz M a écrit :
> > From 3d7590ed1d012367f0e1408efa4b629d6343220c Mon Sep 17 00:00:00 2001
> > From: Lukasz Marek <lukasz.m.luki at gmail.com>
> > Date: Sat, 26 Oct 2013 01:19:31 +0200
> > Subject: [PATCH 1/2] lavd/alsa: fix timestamp calculation
> >
> > Current implementation didn't include duration of
> > last processed packet.
> > Device may return negative timestamps without
> > this correction.
> >
> > Signed-off-by: Lukasz Marek <lukasz.m.luki at gmail.com>
> > ---
> >  libavdevice/alsa-audio-enc.c |    9 ++++++++-
> >  libavdevice/alsa-audio.h     |    1 +
> >  2 files changed, 9 insertions(+), 1 deletion(-)
> >
> > diff --git a/libavdevice/alsa-audio-enc.c b/libavdevice/alsa-audio-enc.c
> > index 0f4e4a2..2239046 100644
> > --- a/libavdevice/alsa-audio-enc.c
> > +++ b/libavdevice/alsa-audio-enc.c
> > @@ -79,6 +79,13 @@ static int audio_write_packet(AVFormatContext *s1,
> AVPacket *pkt)
> >      int size     = pkt->size;
> >      uint8_t *buf = pkt->data;
> >
> > +    if (!(s->timestamp_diff = pkt->duration)) {
> > +        AVStream *st = s1->streams[0];
> > +        AVCodecContext *codec_ctx = st->codec;
> > +        /* XXX: no need to recalculate: 1/sample_rate ==
> avpriv_set_pts_info() */
>
> > +        s->timestamp_diff = pkt->size /
> (av_get_bytes_per_sample(codec_ctx->sample_fmt) * codec_ctx->channels);
>
> The number of bytes per sample is already in s->frame_size. In fact:
>
> > +    }
> > +
> >      size /= s->frame_size;
>
> ... the duration is precisely "size" now.
>
> Therefore, you could probably just write:
>
>     s->timestamp_diff = pkt->duration ? pkt->duration : size;
>
> (and curse C not to be able to write "pkt->duration || size").
>
> >      if (s->reorder_func) {
> >          if (size > s->reorder_buf_size)
> > @@ -112,7 +119,7 @@ audio_get_output_timestamp(AVFormatContext *s1, int
> stream,
> >      snd_pcm_sframes_t delay = 0;
> >      *wall = av_gettime();
> >      snd_pcm_delay(s->h, &delay);
> > -    *dts = s1->streams[0]->cur_dts - delay;
> > +    *dts = s1->streams[0]->cur_dts + s->timestamp_diff - delay;
> >  }
> >
> >  AVOutputFormat ff_alsa_muxer = {
> > diff --git a/libavdevice/alsa-audio.h b/libavdevice/alsa-audio.h
> > index 44b7c72..005cf83 100644
> > --- a/libavdevice/alsa-audio.h
> > +++ b/libavdevice/alsa-audio.h
> > @@ -57,6 +57,7 @@ typedef struct AlsaData {
> >      void (*reorder_func)(const void *, void *, int);
> >      void *reorder_buf;
> >      int reorder_buf_size; ///< in frames
> > +    int64_t timestamp_diff; ///< duration of last packet, needed to
> calculate timestamp
> >  } AlsaData;
> >
> >  /**
>
> Apart from that simplification, it looks right, thanks.
>

I will fix it as you pointed, but for now Michael pointed at similar patch
that st->cur_dts is not good idea to use.
I will resubmit new patch when it is discussed basing on pulse version.