Supporters update.
[dcpomatic.git] / src / lib / audio_decoder.cc
index 6a795f3ac1906c3bf0979c9623424ae36a588e41..61ff5d265c526e6de31e44d5bfb7eb6972883fba 100644 (file)
@@ -1,5 +1,5 @@
 /*
-    Copyright (C) 2012-2016 Carl Hetherington <cth@carlh.net>
+    Copyright (C) 2012-2021 Carl Hetherington <cth@carlh.net>
 
     This file is part of DCP-o-matic.
 
 
 */
 
+
 #include "audio_decoder.h"
 #include "audio_buffers.h"
-#include "audio_decoder_stream.h"
 #include "audio_content.h"
+#include "dcpomatic_log.h"
 #include "log.h"
+#include "resampler.h"
 #include "compose.hpp"
-#include <boost/foreach.hpp>
 #include <iostream>
 
 #include "i18n.h"
 
+
 using std::cout;
-using std::map;
-using boost::shared_ptr;
+using std::shared_ptr;
+using std::make_shared;
 using boost::optional;
+using namespace dcpomatic;
+
 
-AudioDecoder::AudioDecoder (Decoder* parent, shared_ptr<const AudioContent> content, shared_ptr<Log> log)
-       : DecoderPart (parent, log)
+AudioDecoder::AudioDecoder (Decoder* parent, shared_ptr<const AudioContent> content, bool fast)
+       : DecoderPart (parent)
+       , _content (content)
+       , _fast (fast)
 {
-       BOOST_FOREACH (AudioStreamPtr i, content->streams ()) {
-               _streams[i] = shared_ptr<AudioDecoderStream> (new AudioDecoderStream (content, i, parent, this, log));
+       /* Set up _positions so that we have one for each stream */
+       for (auto i: content->streams ()) {
+               _positions[i] = 0;
        }
 }
 
+
+/** @param time_already_delayed true if the delay should not be added to time */
 void
-AudioDecoder::emit (AudioStreamPtr stream, shared_ptr<const AudioBuffers> data, ContentTime time)
+AudioDecoder::emit(shared_ptr<const Film> film, AudioStreamPtr stream, shared_ptr<const AudioBuffers> data, ContentTime time, bool flushing)
 {
        if (ignore ()) {
                return;
        }
 
-       if (_streams.find (stream) == _streams.end ()) {
+       int const resampled_rate = _content->resampled_frame_rate(film);
+       if (!flushing) {
+               time += ContentTime::from_seconds (_content->delay() / 1000.0);
+       }
 
-               /* This method can be called with an unknown stream during the following sequence:
-                  - Add KDM to some DCP content.
-                  - Content gets re-examined.
-                  - SingleStreamAudioContent::take_from_audio_examiner creates a new stream.
-                  - Some content property change signal is delivered so Player::Changed is emitted.
-                  - Film viewer to re-gets the frame.
-                  - Player calls DCPDecoder pass which calls this method on the new stream.
+       /* Amount of error we will tolerate on audio timestamps; see comment below.
+        * We'll use 1 24fps video frame as this seems to be roughly how ffplay does it.
+        */
+       Frame const slack_frames = resampled_rate / 24;
+
+       /* first_since_seek is set to true if this is the first data we have
+          received since initialisation or seek.  We'll set the position based
+          on the ContentTime that was given.  After this first time we just
+          count samples unless the timestamp is more than slack_frames away
+          from where we think it should be.  This is because ContentTimes seem
+          to be slightly unreliable from FFmpegDecoder (i.e.  not sample
+          accurate), but we still need to obey them sometimes otherwise we get
+          sync problems such as #1833.
+       */
+
+       auto const first_since_seek = _positions[stream] == 0;
+       auto const need_reset = !first_since_seek && (std::abs(_positions[stream] - time.frames_round(resampled_rate)) > slack_frames);
+
+       if (need_reset) {
+               LOG_GENERAL (
+                       "Reset audio position: was %1, new data at %2, slack: %3 frames",
+                       _positions[stream],
+                       time.frames_round(resampled_rate),
+                       std::abs(_positions[stream] - time.frames_round(resampled_rate))
+                       );
+       }
+
+       if (first_since_seek || need_reset) {
+               _positions[stream] = time.frames_round (resampled_rate);
+       }
 
-                  At this point the AudioDecoder does not know about the new stream.
+       if (first_since_seek && _content->delay() > 0) {
+               silence (stream, _content->delay());
+       }
 
-                  Then
-                  - Some other property change signal is delivered which marks the player's pieces invalid.
-                  - Film viewer re-gets again.
-                  - Everything is OK.
+       shared_ptr<Resampler> resampler;
+       auto i = _resamplers.find(stream);
+       if (i != _resamplers.end()) {
+               resampler = i->second;
+       } else {
+               if (stream->frame_rate() != resampled_rate) {
+                       LOG_GENERAL (
+                               "Creating new resampler from %1 to %2 with %3 channels",
+                               stream->frame_rate(),
+                               resampled_rate,
+                               stream->channels()
+                               );
+
+                       resampler = make_shared<Resampler>(stream->frame_rate(), resampled_rate, stream->channels());
+                       if (_fast) {
+                               resampler->set_fast ();
+                       }
+                       _resamplers[stream] = resampler;
+               }
+       }
 
-                  In this situation it is fine for us to silently drop the audio.
-               */
+       if (resampler && !flushing) {
+               /* It can be the the data here has a different number of channels than the stream
+                * it comes from (e.g. the files decoded by FFmpegDecoder sometimes have a random
+                * frame, often at the end, with more channels).  Insert silence or discard channels
+                * here.
+                */
+               if (resampler->channels() != data->channels()) {
+                       LOG_WARNING("Received audio data with an unexpected channel count of %1 instead of %2", data->channels(), resampler->channels());
+                       auto data_copy = data->clone();
+                       data_copy->set_channels(resampler->channels());
+                       data = resampler->run(data_copy);
+               } else {
+                       data = resampler->run(data);
+               }
 
-               return;
+               if (data->frames() == 0) {
+                       return;
+               }
        }
 
-       _streams[stream]->audio (data, time);
+       Data(stream, ContentAudio (data, _positions[stream]));
+       _positions[stream] += data->frames();
 }
 
-void
-AudioDecoder::flush ()
+
+/** @return Time just after the last thing that was emitted from a given stream */
+ContentTime
+AudioDecoder::stream_position (shared_ptr<const Film> film, AudioStreamPtr stream) const
+{
+       auto i = _positions.find (stream);
+       DCPOMATIC_ASSERT (i != _positions.end ());
+       return ContentTime::from_frames (i->second, _content->resampled_frame_rate(film));
+}
+
+
+boost::optional<ContentTime>
+AudioDecoder::position (shared_ptr<const Film> film) const
 {
-       for (StreamMap::const_iterator i = _streams.begin(); i != _streams.end(); ++i) {
-               i->second->flush ();
+       optional<ContentTime> p;
+       for (auto i: _positions) {
+               auto const ct = stream_position (film, i.first);
+               if (!p || ct < *p) {
+                       p = ct;
+               }
        }
+
+       return p;
 }
 
+
 void
-AudioDecoder::set_fast ()
+AudioDecoder::seek ()
 {
-       for (StreamMap::const_iterator i = _streams.begin(); i != _streams.end(); ++i) {
-               i->second->set_fast ();
+       for (auto i: _resamplers) {
+               i.second->flush ();
+               i.second->reset ();
+       }
+
+       for (auto& i: _positions) {
+               i.second = 0;
        }
 }
 
-optional<ContentTime>
-AudioDecoder::position () const
+
+void
+AudioDecoder::flush ()
 {
-       optional<ContentTime> p;
-       for (map<AudioStreamPtr, ContentTime>::const_iterator i = _positions.begin(); i != _positions.end(); ++i) {
-               if (!p || i->second < *p) {
-                       p = i->second;
+       for (auto const& i: _resamplers) {
+               auto ro = i.second->flush ();
+               if (ro->frames() > 0) {
+                       Data (i.first, ContentAudio (ro, _positions[i.first]));
+                       _positions[i.first] += ro->frames();
                }
        }
 
-       return p;
+       if (_content->delay() < 0) {
+               /* Finish off with the gap caused by the delay */
+               for (auto stream: _content->streams()) {
+                       silence (stream, -_content->delay());
+               }
+       }
 }
 
+
 void
-AudioDecoder::set_position (AudioStreamPtr stream, ContentTime time)
+AudioDecoder::silence (AudioStreamPtr stream, int milliseconds)
 {
-       _positions[stream] = time;
+       int const samples = ContentTime::from_seconds(milliseconds / 1000.0).frames_round(stream->frame_rate());
+       auto silence = make_shared<AudioBuffers>(stream->channels(), samples);
+       silence->make_silent ();
+       Data (stream, ContentAudio(silence, _positions[stream]));
 }