Fix reffing of the AVFrame output by the decoder.
[dcpomatic.git] / src / lib / ffmpeg_decoder.cc
index 58c7317ac49700c1f792a52f97ddbe409e0e9606..8e09810cb2cc75c3a6109139e0a0db8427f4eb8e 100644 (file)
@@ -50,6 +50,8 @@ extern "C" {
 #include "filter_graph.h"
 #include "subtitle.h"
 
+#include "i18n.h"
+
 using std::cout;
 using std::string;
 using std::vector;
@@ -78,10 +80,6 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, DecodeOptions o)
        setup_video ();
        setup_audio ();
        setup_subtitle ();
-
-       if (!o.video_sync) {
-               _first_video = 0;
-       }
 }
 
 FFmpegDecoder::~FFmpegDecoder ()
@@ -113,7 +111,7 @@ FFmpegDecoder::setup_general ()
        }
 
        if (avformat_find_stream_info (_format_context, 0) < 0) {
-               throw DecodeError ("could not find stream information");
+               throw DecodeError (_("could not find stream information"));
        }
 
        /* Find video, audio and subtitle streams and choose the first of each */
@@ -148,12 +146,12 @@ FFmpegDecoder::setup_general ()
        }
 
        if (_video_stream < 0) {
-               throw DecodeError ("could not find video stream");
+               throw DecodeError (N_("could not find video stream"));
        }
 
        _frame = avcodec_alloc_frame ();
        if (_frame == 0) {
-               throw DecodeError ("could not allocate frame");
+               throw DecodeError (N_("could not allocate frame"));
        }
 }
 
@@ -164,11 +162,11 @@ FFmpegDecoder::setup_video ()
        _video_codec = avcodec_find_decoder (_video_codec_context->codec_id);
 
        if (_video_codec == 0) {
-               throw DecodeError ("could not find video decoder");
+               throw DecodeError (_("could not find video decoder"));
        }
 
        if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) {
-               throw DecodeError ("could not open video decoder");
+               throw DecodeError (N_("could not open video decoder"));
        }
 }
 
@@ -186,18 +184,18 @@ FFmpegDecoder::setup_audio ()
        _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id);
 
        if (_audio_codec == 0) {
-               throw DecodeError ("could not find audio decoder");
+               throw DecodeError (_("could not find audio decoder"));
        }
 
        if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
-               throw DecodeError ("could not open audio decoder");
+               throw DecodeError (N_("could not open audio decoder"));
        }
 }
 
 void
 FFmpegDecoder::setup_subtitle ()
 {
-       if (!_subtitle_stream) {
+       if (!_subtitle_stream || _subtitle_stream->id() >= int (_format_context->nb_streams)) {
                return;
        }
 
@@ -205,11 +203,11 @@ FFmpegDecoder::setup_subtitle ()
        _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
 
        if (_subtitle_codec == 0) {
-               throw DecodeError ("could not find subtitle decoder");
+               throw DecodeError (_("could not find subtitle decoder"));
        }
        
        if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
-               throw DecodeError ("could not open subtitle decoder");
+               throw DecodeError (N_("could not open subtitle decoder"));
        }
 }
 
@@ -224,35 +222,28 @@ FFmpegDecoder::pass ()
                        /* Maybe we should fail here, but for now we'll just finish off instead */
                        char buf[256];
                        av_strerror (r, buf, sizeof(buf));
-                       _film->log()->log (String::compose ("error on av_read_frame (%1) (%2)", buf, r));
+                       _film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
                }
-               
+
                /* Get any remaining frames */
                
                _packet.data = 0;
                _packet.size = 0;
-
+               
                /* XXX: should we reset _packet.data and size after each *_decode_* call? */
-
+               
                int frame_finished;
-
+               
                if (_opt.decode_video) {
                        while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
-                               filter_and_emit_video (_frame);
+                               filter_and_emit_video ();
                        }
                }
-
+               
                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
-                                       );
-
-                               assert (_audio_codec_context->channels == _film->audio_channels());
-                               Audio (deinterleave_audio (_frame->data, data_size));
-                       }
+                       decode_audio_packet ();
                }
-
+                       
                return true;
        }
 
@@ -267,66 +258,15 @@ FFmpegDecoder::pass ()
                if (r >= 0 && frame_finished) {
 
                        if (r != _packet.size) {
-                               _film->log()->log (String::compose ("Used only %1 bytes of %2 in packet", r, _packet.size));
+                               _film->log()->log (String::compose (N_("Used only %1 bytes of %2 in packet"), r, _packet.size));
                        }
 
-                       if (_opt.video_sync) {
-                               out_with_sync ();
-                       } else {
-                               filter_and_emit_video (_frame);
-                       }
+                       filter_and_emit_video ();
                }
 
        } 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) {
-
-                       /* 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);
-
-                       /* We only decode audio if we've had our first video packet through, and if it
-                          was before this packet.  Until then audio is thrown away.
-                       */
-                               
-                       if ((_first_video && _first_video.get() <= source_pts_seconds) || !_opt.decode_video) {
-
-                               if (!_first_audio && _opt.decode_video) {
-                                       _first_audio = source_pts_seconds;
-                                       
-                                       /* This is our first audio frame, and if we've arrived here we must have had our
-                                          first video frame.  Push some silence to make up any gap between our first
-                                          video frame and our first audio.
-                                       */
-                       
-                                       /* frames of silence that we must push */
-                                       int const s = rint ((_first_audio.get() - _first_video.get()) * ffa->sample_rate ());
-                                       
-                                       _film->log()->log (
-                                               String::compose (
-                                                       "First video at %1, first audio at %2, pushing %3 audio frames of silence for %4 channels (%5 bytes per sample)",
-                                                       _first_video.get(), _first_audio.get(), s, ffa->channels(), bytes_per_audio_sample()
-                                                       )
-                                               );
-                                       
-                                       if (s) {
-                                               shared_ptr<AudioBuffers> audio (new AudioBuffers (ffa->channels(), s));
-                                               audio->make_silent ();
-                                               Audio (audio);
-                                       }
-                               }
-
-                               int const data_size = av_samples_get_buffer_size (
-                                       0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
-                                       );
-                               
-                               assert (_audio_codec_context->channels == _film->audio_channels());
-                               Audio (deinterleave_audio (_frame->data, data_size));
-                       }
-               }
-                       
-       } else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt.decode_subtitles && _first_video) {
+               decode_audio_packet ();
+       } else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt.decode_subtitles) {
 
                int got_subtitle;
                AVSubtitle sub;
@@ -445,7 +385,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
        break;
 
        default:
-               throw DecodeError (String::compose ("Unrecognised audio sample format (%1)", static_cast<int> (audio_sample_format())));
+               throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
        }
 
        return audio;
@@ -513,22 +453,24 @@ string
 FFmpegDecoder::stream_name (AVStream* s) const
 {
        stringstream n;
-       
-       AVDictionaryEntry const * lang = av_dict_get (s->metadata, "language", 0, 0);
-       if (lang) {
-               n << lang->value;
-       }
-       
-       AVDictionaryEntry const * title = av_dict_get (s->metadata, "title", 0, 0);
-       if (title) {
-               if (!n.str().empty()) {
-                       n << " ";
+
+       if (s->metadata) {
+               AVDictionaryEntry const * lang = av_dict_get (s->metadata, N_("language"), 0, 0);
+               if (lang) {
+                       n << lang->value;
+               }
+               
+               AVDictionaryEntry const * title = av_dict_get (s->metadata, N_("title"), 0, 0);
+               if (title) {
+                       if (!n.str().empty()) {
+                               n << N_(" ");
+                       }
+                       n << title->value;
                }
-               n << title->value;
        }
 
        if (n.str().empty()) {
-               n << "unknown";
+               n << N_("unknown");
        }
 
        return n.str ();
@@ -556,36 +498,41 @@ FFmpegDecoder::set_subtitle_stream (shared_ptr<SubtitleStream> s)
 }
 
 void
-FFmpegDecoder::filter_and_emit_video (AVFrame* frame)
+FFmpegDecoder::filter_and_emit_video ()
 {
        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 (libdcp::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, libdcp::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));
+               _film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
        } else {
                graph = *i;
        }
 
-       list<shared_ptr<Image> > images = graph->process (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, frame_time ());
+               int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
+               if (bet != AV_NOPTS_VALUE) {
+                       emit_video (*i, false, bet * av_q2d (_format_context->streams[_video_stream]->time_base));
+               } else {
+                       _film->log()->log ("Dropping frame without PTS");
+               }
        }
 }
 
 bool
 FFmpegDecoder::seek (double p)
 {
-       return do_seek (p, false);
+       return do_seek (p, false, false);
 }
 
 bool
@@ -595,21 +542,57 @@ FFmpegDecoder::seek_to_last ()
           (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);
+       return do_seek (last_source_time(), true, false);
+}
+
+void
+FFmpegDecoder::seek_back ()
+{
+       do_seek (last_source_time() - 2.5 / frames_per_second (), true, true);
+}
+
+void
+FFmpegDecoder::seek_forward ()
+{
+       do_seek (last_source_time() - 0.5 / frames_per_second(), true, true);
 }
 
 bool
-FFmpegDecoder::do_seek (double p, bool backwards)
+FFmpegDecoder::do_seek (double p, bool backwards, bool accurate)
 {
        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);
        }
-       
+
+       if (accurate) {
+               while (1) {
+                       int r = av_read_frame (_format_context, &_packet);
+                       if (r < 0) {
+                               return true;
+                       }
+                       
+                       avcodec_get_frame_defaults (_frame);
+                       
+                       if (_packet.stream_index == _video_stream) {
+                               int finished = 0;
+                               int const r = avcodec_decode_video2 (_video_codec_context, _frame, &finished, &_packet);
+                               if (r >= 0 && finished) {
+                                       int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
+                                       if (bet > vt) {
+                                               break;
+                                       }
+                               }
+                       }
+                       
+                       av_free_packet (&_packet);
+               }
+       }
+               
        return r < 0;
 }
 
@@ -624,7 +607,7 @@ FFmpegAudioStream::create (string t, optional<int> v)
        stringstream s (t);
        string type;
        s >> type;
-       if (type != "ffmpeg") {
+       if (type != N_("ffmpeg")) {
                return shared_ptr<FFmpegAudioStream> ();
        }
 
@@ -646,7 +629,7 @@ FFmpegAudioStream::FFmpegAudioStream (string t, optional<int> version)
                string type;
                /* Current (marked version 1) */
                n >> type >> _id >> _sample_rate >> _channel_layout;
-               assert (type == "ffmpeg");
+               assert (type == N_("ffmpeg"));
        }
 
        for (int i = 0; i < name_index; ++i) {
@@ -662,53 +645,7 @@ FFmpegAudioStream::FFmpegAudioStream (string t, optional<int> version)
 string
 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));
-       }
+       return String::compose (N_("ffmpeg %1 %2 %3 %4"), _id, _sample_rate, _channel_layout, _name);
 }
 
 void
@@ -736,9 +673,39 @@ FFmpegDecoder::length () const
        return (double(_format_context->duration) / AV_TIME_BASE) * frames_per_second();
 }
 
-double
-FFmpegDecoder::frame_time () const
+void
+FFmpegDecoder::decode_audio_packet ()
 {
-       return av_frame_get_best_effort_timestamp(_frame) * av_q2d (_format_context->streams[_video_stream]->time_base);
-}
+       shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
+       assert (ffa);
+
+       /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
+          several times.
+       */
+       
+       AVPacket copy_packet = _packet;
+
+       while (copy_packet.size > 0) {
 
+               int frame_finished;
+               int const decode_result = avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, &copy_packet);
+               if (decode_result >= 0) {
+                       if (frame_finished) {
+                       
+                               /* Where we are in the source, in seconds */
+                               double const source_pts_seconds = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
+                                       * av_frame_get_best_effort_timestamp(_frame);
+                               
+                               int const data_size = av_samples_get_buffer_size (
+                                       0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
+                                       );
+                               
+                               assert (_audio_codec_context->channels == _film->audio_channels());
+                               Audio (deinterleave_audio (_frame->data, data_size), source_pts_seconds);
+                       }
+                       
+                       copy_packet.data += decode_result;
+                       copy_packet.size -= decode_result;
+               }
+       }
+}