Merge branch '1.0' into 1.0-seek
[dcpomatic.git] / src / lib / ffmpeg_decoder.cc
index ceb14bacc3b442056cd21c096f814fd9e6e08a0e..b2e8cc44a0d56b1c9fe1aa30f10ab122c0fa15d7 100644 (file)
@@ -42,6 +42,7 @@ extern "C" {
 #include "ffmpeg_decoder.h"
 #include "filter_graph.h"
 #include "audio_buffers.h"
+#include "ffmpeg_content.h"
 
 #include "i18n.h"
 
@@ -60,7 +61,7 @@ using libdcp::Size;
 FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio)
        : Decoder (f)
        , VideoDecoder (f, c)
-       , AudioDecoder (f)
+       , AudioDecoder (f, c)
        , SubtitleDecoder (f)
        , FFmpeg (c)
        , _subtitle_codec_context (0)
@@ -69,7 +70,6 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegC
        , _decode_audio (audio)
        , _video_pts_offset (0)
        , _audio_pts_offset (0)
-       , _just_sought (false)
 {
        setup_subtitle ();
 
@@ -88,7 +88,7 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegC
        */
 
        bool const have_video = video && c->first_video();
-       bool const have_audio = audio && c->audio_stream() && c->audio_stream()->first_audio;
+       bool const have_audio = _decode_audio && c->audio_stream () && c->audio_stream()->first_audio;
 
        /* First, make one of them start at 0 */
 
@@ -96,6 +96,8 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegC
                _video_pts_offset = _audio_pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get());
        } else if (have_video) {
                _video_pts_offset = - c->first_video().get();
+       } else if (have_audio) {
+               _audio_pts_offset = - c->audio_stream()->first_audio.get();
        }
 
        /* Now adjust both so that the video pts starts on a frame */
@@ -140,12 +142,15 @@ FFmpegDecoder::flush ()
                decode_audio_packet ();
        }
 
+#if 0  
+       /* XXX */
        /* Stop us being asked for any more data */
        _video_position = _ffmpeg_content->video_length ();
        _audio_position = _ffmpeg_content->audio_length ();
+#endif 
 }
 
-void
+bool
 FFmpegDecoder::pass ()
 {
        int r = av_read_frame (_format_context, &_packet);
@@ -161,23 +166,26 @@ FFmpegDecoder::pass ()
                }
 
                flush ();
-               return;
+               return true;
        }
 
        avcodec_get_frame_defaults (_frame);
 
        shared_ptr<const Film> film = _film.lock ();
        assert (film);
+
+       int const si = _packet.stream_index;
        
-       if (_packet.stream_index == _video_stream && _decode_video) {
+       if (si == _video_stream && _decode_video) {
                decode_video_packet ();
-       } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) {
+       } else if (_ffmpeg_content->audio_stream() && si == _ffmpeg_content->audio_stream()->index (_format_context) && _decode_audio) {
                decode_audio_packet ();
-       } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id && film->with_subtitles ()) {
+       } else if (_ffmpeg_content->subtitle_stream() && si == _ffmpeg_content->subtitle_stream()->index (_format_context) && film->with_subtitles ()) {
                decode_subtitle_packet ();
        }
 
        av_free_packet (&_packet);
+       return false;
 }
 
 /** @param data pointer to array of pointers to buffers.
@@ -292,69 +300,131 @@ FFmpegDecoder::bytes_per_audio_sample () const
        return av_get_bytes_per_sample (audio_sample_format ());
 }
 
-void
-FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
+int
+FFmpegDecoder::minimal_run (boost::function<bool (ContentTime, ContentTime, int)> finished)
 {
-       double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base);
+       int frames_read = 0;
+       ContentTime last_video = 0;
+       ContentTime last_audio = 0;
+       bool flushing = false;
 
-       /* If we are doing an accurate seek, our initial shot will be 5 frames (5 being
-          a number plucked from the air) earlier than we want to end up.  The loop below
-          will hopefully then step through to where we want to be.
-       */
-       int initial = frame;
+       while (!finished (last_video, last_audio, frames_read)) {
+               int r = av_read_frame (_format_context, &_packet);
+               if (r < 0) {
+                       /* We should flush our decoders here, possibly yielding a few more frames,
+                          but the consequence of having to do that is too hideous to contemplate.
+                          Instead we give up and say that you can't seek too close to the end
+                          of a file.
+                       */
+                       return frames_read;
+               }
 
-       if (accurate) {
-               initial -= 5;
-       }
+               ++frames_read;
+
+               double const time_base = av_q2d (_format_context->streams[_packet.stream_index]->time_base);
+
+               if (_packet.stream_index == _video_stream) {
 
-       if (initial < 0) {
-               initial = 0;
+                       avcodec_get_frame_defaults (_frame);
+                       
+                       int finished = 0;
+                       r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
+                       if (r >= 0 && finished) {
+                               last_video = rint (
+                                       (av_frame_get_best_effort_timestamp (_frame) * time_base + _video_pts_offset) * TIME_HZ
+                                       );
+                       }
+                               
+               } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->index (_format_context)) {
+                       AVPacket copy_packet = _packet;
+                       while (copy_packet.size > 0) {
+
+                               int finished;
+                               r = avcodec_decode_audio4 (audio_codec_context(), _frame, &finished, &_packet);
+                               if (r >= 0 && finished) {
+                                       last_audio = rint (
+                                               (av_frame_get_best_effort_timestamp (_frame) * time_base + _audio_pts_offset) * TIME_HZ
+                                               );
+                               }
+                                       
+                               copy_packet.data += r;
+                               copy_packet.size -= r;
+                       }
+               }
+               
+               av_free_packet (&_packet);
        }
 
-       /* Initial seek time in the stream's timebase */
-       int64_t const initial_vt = ((initial / _ffmpeg_content->video_frame_rate()) - _video_pts_offset) / time_base;
+       return frames_read;
+}
+
+bool
+FFmpegDecoder::seek_overrun_finished (ContentTime seek, ContentTime last_video, ContentTime last_audio) const
+{
+       return last_video >= seek || last_audio >= seek;
+}
 
-       av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD);
+bool
+FFmpegDecoder::seek_final_finished (int n, int done) const
+{
+       return n == done;
+}
+
+void
+FFmpegDecoder::seek_and_flush (ContentTime t)
+{
+       int64_t const initial_v = ((double (t) / TIME_HZ) - _video_pts_offset) /
+               av_q2d (_format_context->streams[_video_stream]->time_base);
+
+       av_seek_frame (_format_context, _video_stream, initial_v, AVSEEK_FLAG_BACKWARD);
+
+       shared_ptr<FFmpegAudioStream> as = _ffmpeg_content->audio_stream ();
+       if (as) {
+               int64_t initial_a = ((double (t) / TIME_HZ) - _audio_pts_offset) /
+                       av_q2d (as->stream(_format_context)->time_base);
+
+               av_seek_frame (_format_context, as->index (_format_context), initial_a, AVSEEK_FLAG_BACKWARD);
+       }
 
        avcodec_flush_buffers (video_codec_context());
+       if (audio_codec_context ()) {
+               avcodec_flush_buffers (audio_codec_context ());
+       }
        if (_subtitle_codec_context) {
                avcodec_flush_buffers (_subtitle_codec_context);
        }
+}
 
-       _just_sought = true;
-       _video_position = frame;
+void
+FFmpegDecoder::seek (ContentTime time, bool accurate)
+{
+       Decoder::seek (time, accurate);
        
-       if (frame == 0 || !accurate) {
-               /* We're already there, or we're as close as we need to be */
-               return;
+       /* If we are doing an accurate seek, our initial shot will be 200ms (200 being
+          a number plucked from the air) earlier than we want to end up.  The loop below
+          will hopefully then step through to where we want to be.
+       */
+
+       ContentTime pre_roll = accurate ? (0.2 * TIME_HZ) : 0;
+       ContentTime initial_seek = time - pre_roll;
+       if (initial_seek < 0) {
+               initial_seek = 0;
        }
+       
+       /* Initial seek time in the video stream's timebase */
 
-       while (1) {
-               int r = av_read_frame (_format_context, &_packet);
-               if (r < 0) {
-                       return;
-               }
+       seek_and_flush (initial_seek);
 
-               if (_packet.stream_index != _video_stream) {
-                       continue;
-               }
-               
-               avcodec_get_frame_defaults (_frame);
-               
-               int finished = 0;
-               r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
-               if (r >= 0 && finished) {
-                       _video_position = rint (
-                               (av_frame_get_best_effort_timestamp (_frame) * time_base + _video_pts_offset) * _ffmpeg_content->video_frame_rate()
-                               );
+       if (time == 0 || !accurate) {
+               /* We're already there, or we're as close as we need to be */
+               return;
+       }
 
-                       if (_video_position >= (frame - 1)) {
-                               av_free_packet (&_packet);
-                               break;
-                       }
-               }
-               
-               av_free_packet (&_packet);
+       int const N = minimal_run (boost::bind (&FFmpegDecoder::seek_overrun_finished, this, time, _1, _2));
+       
+       seek_and_flush (initial_seek);
+       if (N > 0) {
+               minimal_run (boost::bind (&FFmpegDecoder::seek_final_finished, this, N - 1, _3));
        }
 }
 
@@ -371,6 +441,7 @@ FFmpegDecoder::decode_audio_packet ()
 
                int frame_finished;
                int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, &copy_packet);
+
                if (decode_result < 0) {
                        shared_ptr<const Film> film = _film.lock ();
                        assert (film);
@@ -379,31 +450,16 @@ FFmpegDecoder::decode_audio_packet ()
                }
 
                if (frame_finished) {
-                       
-                       if (_audio_position == 0) {
-                               /* Where we are in the source, in seconds */
-                               double const pts = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
-                                       * av_frame_get_best_effort_timestamp(_frame) + _audio_pts_offset;
-                               
-                               if (pts > 0) {
-                                       /* Emit some silence */
-                                       shared_ptr<AudioBuffers> silence (
-                                               new AudioBuffers (
-                                                       _ffmpeg_content->audio_channels(),
-                                                       pts * _ffmpeg_content->content_audio_frame_rate()
-                                                       )
-                                               );
-                                       
-                                       silence->make_silent ();
-                                       audio (silence, _audio_position);
-                               }
-                       }
-                       
+                       ContentTime const t = rint (
+                               (av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
+                                * av_frame_get_best_effort_timestamp(_frame) + _audio_pts_offset) * TIME_HZ
+                               );
+
                        int const data_size = av_samples_get_buffer_size (
                                0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1
                                );
                        
-                       audio (deinterleave_audio (_frame->data, data_size), _audio_position);
+                       audio (deinterleave_audio (_frame->data, data_size), t);
                }
                        
                copy_packet.data += decode_result;
@@ -452,45 +508,8 @@ FFmpegDecoder::decode_video_packet ()
                }
                
                if (i->second != AV_NOPTS_VALUE) {
-
-                       double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _video_pts_offset;
-
-                       if (_just_sought) {
-                               /* We just did a seek, so disable any attempts to correct for where we
-                                  are / should be.
-                               */
-                               _video_position = rint (pts * _ffmpeg_content->video_frame_rate ());
-                               _just_sought = false;
-                       }
-
-                       double const next = _video_position / _ffmpeg_content->video_frame_rate();
-                       double const one_frame = 1 / _ffmpeg_content->video_frame_rate ();
-                       double delta = pts - next;
-
-                       while (delta > one_frame) {
-                               /* This PTS is more than one frame forward in time of where we think we should be; emit
-                                  a black frame.
-                               */
-
-                               /* XXX: I think this should be a copy of the last frame... */
-                               boost::shared_ptr<Image> black (
-                                       new Image (
-                                               static_cast<AVPixelFormat> (_frame->format),
-                                               libdcp::Size (video_codec_context()->width, video_codec_context()->height),
-                                               true
-                                               )
-                                       );
-                               
-                               black->make_black ();
-                               video (image, false, _video_position);
-                               delta -= one_frame;
-                       }
-
-                       if (delta > -one_frame) {
-                               /* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */
-                               video (image, false, _video_position);
-                       }
-                               
+                       ContentTime const t = rint ((i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _video_pts_offset) * TIME_HZ);
+                       video (image, false, t);
                } else {
                        shared_ptr<const Film> film = _film.lock ();
                        assert (film);
@@ -507,11 +526,11 @@ FFmpegDecoder::setup_subtitle ()
 {
        boost::mutex::scoped_lock lm (_mutex);
        
-       if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
+       if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->index (_format_context) >= int (_format_context->nb_streams)) {
                return;
        }
 
-       _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
+       _subtitle_codec_context = _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec;
        _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
 
        if (_subtitle_codec == 0) {
@@ -523,14 +542,6 @@ FFmpegDecoder::setup_subtitle ()
        }
 }
 
-bool
-FFmpegDecoder::done () const
-{
-       bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length());
-       bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length());
-       return vd && ad;
-}
-       
 void
 FFmpegDecoder::decode_subtitle_packet ()
 {
@@ -556,8 +567,8 @@ FFmpegDecoder::decode_subtitle_packet ()
        double const packet_time = static_cast<double> (sub.pts) / AV_TIME_BASE;
        
        /* hence start time for this sub */
-       Time const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ;
-       Time const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ;
+       DCPTime const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ;
+       DCPTime const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ;
 
        AVSubtitleRect const * rect = sub.rects[0];