Don't try to make decoders when there is no content.
[dcpomatic.git] / src / lib / ffmpeg_decoder.cc
index fd522a5ac374bc10116bbd916c6a27dce0f611ea..1f11f70a0a462d9d5ac265c8b6e3961879bd46c7 100644 (file)
@@ -58,11 +58,12 @@ using std::list;
 using boost::shared_ptr;
 using boost::optional;
 using boost::dynamic_pointer_cast;
+using libdcp::Size;
 
-FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, shared_ptr<const Options> o, Job* j)
-       : Decoder (f, o, j)
-       , VideoDecoder (f, o, j)
-       , AudioDecoder (f, o, j)
+FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, DecodeOptions o)
+       : Decoder (f, o)
+       , VideoDecoder (f, o)
+       , AudioDecoder (f, o)
        , _format_context (0)
        , _video_stream (-1)
        , _frame (0)
@@ -77,6 +78,10 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, shared_ptr<const Options> o, J
        setup_video ();
        setup_audio ();
        setup_subtitle ();
+
+       if (!o.video_sync) {
+               _first_video = 0;
+       }
 }
 
 FFmpegDecoder::~FFmpegDecoder ()
@@ -101,11 +106,9 @@ FFmpegDecoder::~FFmpegDecoder ()
 void
 FFmpegDecoder::setup_general ()
 {
-       int r;
-       
        av_register_all ();
 
-       if ((r = avformat_open_input (&_format_context, _film->content_path().c_str(), 0, 0)) != 0) {
+       if (avformat_open_input (&_format_context, _film->content_path().c_str(), 0, 0) < 0) {
                throw OpenFileError (_film->content_path ());
        }
 
@@ -120,11 +123,21 @@ FFmpegDecoder::setup_general ()
                if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
                        _video_stream = i;
                } else if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
+
+                       /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
+                          so bodge it here.  No idea why we should have to do this.
+                       */
+                       
+                       if (s->codec->channel_layout == 0) {
+                               s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
+                       }
+                       
                        _audio_streams.push_back (
                                shared_ptr<AudioStream> (
                                        new FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channel_layout)
                                        )
                                );
+                       
                } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
                        _subtitle_streams.push_back (
                                shared_ptr<SubtitleStream> (
@@ -179,14 +192,6 @@ FFmpegDecoder::setup_audio ()
        if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
                throw DecodeError ("could not open audio decoder");
        }
-
-       /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
-          so bodge it here.  No idea why we should have to do this.
-       */
-
-       if (_audio_codec_context->channel_layout == 0) {
-               _audio_codec_context->channel_layout = av_get_default_channel_layout (ffa->channels());
-       }
 }
 
 void
@@ -235,7 +240,7 @@ FFmpegDecoder::pass ()
                        filter_and_emit_video (_frame);
                }
 
-               if (_audio_stream && _opt->decode_audio) {
+               if (_audio_stream && _opt.decode_audio) {
                        while (avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
                                int const data_size = av_samples_get_buffer_size (
                                        0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
@@ -252,7 +257,7 @@ FFmpegDecoder::pass ()
        avcodec_get_frame_defaults (_frame);
 
        shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
-       
+
        if (_packet.stream_index == _video_stream) {
 
                int frame_finished;
@@ -263,50 +268,14 @@ FFmpegDecoder::pass ()
                                _film->log()->log (String::compose ("Used only %1 bytes of %2 in packet", r, _packet.size));
                        }
 
-                       /* Where we are in the output, in seconds */
-                       double const out_pts_seconds = video_frame() / frames_per_second();
-
-                       /* Where we are in the source, in seconds */
-                       double const source_pts_seconds = av_q2d (_format_context->streams[_packet.stream_index]->time_base)
-                               * av_frame_get_best_effort_timestamp(_frame);
-
-                       _film->log()->log (
-                               String::compose ("Source video frame ready; source at %1, output at %2", source_pts_seconds, out_pts_seconds),
-                               Log::VERBOSE
-                               );
-
-                       if (!_first_video) {
-                               _first_video = source_pts_seconds;
-                       }
-
-                       /* Difference between where we are and where we should be */
-                       double const delta = source_pts_seconds - _first_video.get() - out_pts_seconds;
-                       double const one_frame = 1 / frames_per_second();
-
-                       /* Insert frames if required to get out_pts_seconds up to pts_seconds */
-                       if (delta > one_frame) {
-                               int const extra = rint (delta / one_frame);
-                               for (int i = 0; i < extra; ++i) {
-                                       repeat_last_video ();
-                                       _film->log()->log (
-                                               String::compose (
-                                                       "Extra video frame inserted at %1s; source frame %2, source PTS %3 (at %4 fps)",
-                                                       out_pts_seconds, video_frame(), source_pts_seconds, frames_per_second()
-                                                       )
-                                               );
-                               }
-                       }
-
-                       if (delta > -one_frame) {
-                               /* Process this frame */
-                               filter_and_emit_video (_frame);
+                       if (_opt.video_sync) {
+                               out_with_sync ();
                        } else {
-                               /* Otherwise we are omitting a frame to keep things right */
-                               _film->log()->log (String::compose ("Frame removed at %1s", out_pts_seconds));
+                               filter_and_emit_video (_frame);
                        }
                }
 
-       } else if (ffa && _packet.stream_index == ffa->id() && _opt->decode_audio) {
+       } else if (ffa && _packet.stream_index == ffa->id() && _opt.decode_audio) {
 
                int frame_finished;
                if (avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
@@ -355,7 +324,7 @@ FFmpegDecoder::pass ()
                        }
                }
                        
-       } else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt->decode_subtitles && _first_video) {
+       } else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt.decode_subtitles && _first_video) {
 
                int got_subtitle;
                AVSubtitle sub;
@@ -366,7 +335,7 @@ FFmpegDecoder::pass ()
                        if (sub.num_rects > 0) {
                                shared_ptr<TimedSubtitle> ts;
                                try {
-                                       emit_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub, _first_video.get())));
+                                       emit_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
                                } catch (...) {
                                        /* some problem with the subtitle; we probably didn't understand it */
                                }
@@ -401,7 +370,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size)
        switch (audio_sample_format()) {
        case AV_SAMPLE_FMT_S16:
        {
-               int16_t* p = (int16_t *) data;
+               int16_t* p = reinterpret_cast<int16_t *> (data);
                int sample = 0;
                int channel = 0;
                for (int i = 0; i < total_samples; ++i) {
@@ -416,13 +385,24 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size)
        }
        break;
 
+       case AV_SAMPLE_FMT_S16P:
+       {
+               int16_t* p = reinterpret_cast<int16_t *> (data);
+               for (int i = 0; i < _film->audio_channels(); ++i) {
+                       for (int j = 0; j < frames; ++j) {
+                               audio->data(i)[j] = static_cast<float>(*p++) / (1 << 15);
+                       }
+               }
+       }
+       break;
+       
        case AV_SAMPLE_FMT_S32:
        {
-               int32_t* p = (int32_t *) data;
+               int32_t* p = reinterpret_cast<int32_t *> (data);
                int sample = 0;
                int channel = 0;
                for (int i = 0; i < total_samples; ++i) {
-                       audio->data(channel)[sample] = float(*p++) / (1 << 31);
+                       audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
 
                        ++channel;
                        if (channel == _film->audio_channels()) {
@@ -431,7 +411,25 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size)
                        }
                }
        }
+       break;
+
+       case AV_SAMPLE_FMT_FLT:
+       {
+               float* p = reinterpret_cast<float*> (data);
+               int sample = 0;
+               int channel = 0;
+               for (int i = 0; i < total_samples; ++i) {
+                       audio->data(channel)[sample] = *p++;
 
+                       ++channel;
+                       if (channel == _film->audio_channels()) {
+                               channel = 0;
+                               ++sample;
+                       }
+               }
+       }
+       break;
+               
        case AV_SAMPLE_FMT_FLTP:
        {
                float* p = reinterpret_cast<float*> (data);
@@ -443,7 +441,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size)
        break;
 
        default:
-               assert (false);
+               throw DecodeError (String::compose ("Unrecognised audio sample format (%1)", static_cast<int> (audio_sample_format())));
        }
 
        return audio;
@@ -471,10 +469,10 @@ FFmpegDecoder::audio_sample_format () const
        return _audio_codec_context->sample_fmt;
 }
 
-Size
+libdcp::Size
 FFmpegDecoder::native_size () const
 {
-       return Size (_video_codec_context->width, _video_codec_context->height);
+       return libdcp::Size (_video_codec_context->width, _video_codec_context->height);
 }
 
 PixelFormat
@@ -550,20 +548,23 @@ FFmpegDecoder::set_subtitle_stream (shared_ptr<SubtitleStream> s)
 {
        VideoDecoder::set_subtitle_stream (s);
        setup_subtitle ();
+       OutputChanged ();
 }
 
 void
 FFmpegDecoder::filter_and_emit_video (AVFrame* frame)
 {
+       boost::mutex::scoped_lock lm (_filter_graphs_mutex);
+       
        shared_ptr<FilterGraph> graph;
 
        list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
-       while (i != _filter_graphs.end() && !(*i)->can_process (Size (frame->width, frame->height), (AVPixelFormat) frame->format)) {
+       while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (frame->width, frame->height), (AVPixelFormat) frame->format)) {
                ++i;
        }
 
        if (i == _filter_graphs.end ()) {
-               graph.reset (new FilterGraph (_film, this, _opt->apply_crop, Size (frame->width, frame->height), (AVPixelFormat) frame->format));
+               graph.reset (new FilterGraph (_film, this, libdcp::Size (frame->width, frame->height), (AVPixelFormat) frame->format));
                _filter_graphs.push_back (graph);
                _film->log()->log (String::compose ("New graph for %1x%2, pixel format %3", frame->width, frame->height, frame->format));
        } else {
@@ -573,10 +574,41 @@ FFmpegDecoder::filter_and_emit_video (AVFrame* frame)
        list<shared_ptr<Image> > images = graph->process (frame);
 
        for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
-               emit_video (*i);
+               emit_video (*i, frame_time ());
        }
 }
 
+bool
+FFmpegDecoder::seek (double p)
+{
+       return do_seek (p, false);
+}
+
+bool
+FFmpegDecoder::seek_to_last ()
+{
+       /* This AVSEEK_FLAG_BACKWARD in do_seek is a bit of a hack; without it, if we ask for a seek to the same place as last time
+          (used when we change decoder parameters and want to re-fetch the frame) we end up going forwards rather than
+          staying in the same place.
+       */
+       return do_seek (last_source_time(), true);
+}
+
+bool
+FFmpegDecoder::do_seek (double p, bool backwards)
+{
+       int64_t const vt = p / av_q2d (_format_context->streams[_video_stream]->time_base);
+
+       int const r = av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0);
+       
+       avcodec_flush_buffers (_video_codec_context);
+       if (_subtitle_codec_context) {
+               avcodec_flush_buffers (_subtitle_codec_context);
+       }
+       
+       return r < 0;
+}
+
 shared_ptr<FFmpegAudioStream>
 FFmpegAudioStream::create (string t, optional<int> v)
 {
@@ -629,9 +661,80 @@ FFmpegAudioStream::to_string () const
        return String::compose ("ffmpeg %1 %2 %3 %4", _id, _sample_rate, _channel_layout, _name);
 }
 
+void
+FFmpegDecoder::out_with_sync ()
+{
+       /* Where we are in the output, in seconds */
+       double const out_pts_seconds = video_frame() / frames_per_second();
+       
+       /* Where we are in the source, in seconds */
+       double const source_pts_seconds = av_q2d (_format_context->streams[_packet.stream_index]->time_base)
+               * av_frame_get_best_effort_timestamp(_frame);
+       
+       _film->log()->log (
+               String::compose ("Source video frame ready; source at %1, output at %2", source_pts_seconds, out_pts_seconds),
+               Log::VERBOSE
+               );
+       
+       if (!_first_video) {
+               _first_video = source_pts_seconds;
+       }
+       
+       /* Difference between where we are and where we should be */
+       double const delta = source_pts_seconds - _first_video.get() - out_pts_seconds;
+       double const one_frame = 1 / frames_per_second();
+       
+       /* Insert frames if required to get out_pts_seconds up to pts_seconds */
+       if (delta > one_frame) {
+               int const extra = rint (delta / one_frame);
+               for (int i = 0; i < extra; ++i) {
+                       repeat_last_video ();
+                       _film->log()->log (
+                               String::compose (
+                                       "Extra video frame inserted at %1s; source frame %2, source PTS %3 (at %4 fps)",
+                                       out_pts_seconds, video_frame(), source_pts_seconds, frames_per_second()
+                                       )
+                               );
+               }
+       }
+       
+       if (delta > -one_frame) {
+               /* Process this frame */
+               filter_and_emit_video (_frame);
+       } else {
+               /* Otherwise we are omitting a frame to keep things right */
+               _film->log()->log (String::compose ("Frame removed at %1s", out_pts_seconds));
+       }
+}
+
+void
+FFmpegDecoder::film_changed (Film::Property p)
+{
+       switch (p) {
+       case Film::CROP:
+       case Film::FILTERS:
+       {
+               boost::mutex::scoped_lock lm (_filter_graphs_mutex);
+               _filter_graphs.clear ();
+       }
+       OutputChanged ();
+       break;
+
+       default:
+               break;
+       }
+}
+
 /** @return Length (in video frames) according to our content's header */
 SourceFrame
 FFmpegDecoder::length () const
 {
        return (double(_format_context->duration) / AV_TIME_BASE) * frames_per_second();
 }
+
+double
+FFmpegDecoder::frame_time () const
+{
+       return av_frame_get_best_effort_timestamp(_frame) * av_q2d (_format_context->streams[_video_stream]->time_base);
+}
+