Trying to create export audio encoders with between 9 and 15 channels
[dcpomatic.git] / src / lib / ffmpeg_encoder.cc
index 10d439f94f05b0f2954c39ef0799b98cf28e3454..a3d2ff86f3446d2429bc2ccb979a8d57cbbd7635 100644 (file)
@@ -1,5 +1,5 @@
 /*
-    Copyright (C) 2017 Carl Hetherington <cth@carlh.net>
+    Copyright (C) 2017-2018 Carl Hetherington <cth@carlh.net>
 
     This file is part of DCP-o-matic.
 
@@ -36,43 +36,32 @@ using std::string;
 using std::runtime_error;
 using std::cout;
 using std::pair;
+using std::list;
+using std::map;
 using boost::shared_ptr;
 using boost::bind;
 using boost::weak_ptr;
-
-int FFmpegEncoder::_video_stream_index = 0;
-int FFmpegEncoder::_audio_stream_index = 1;
-
-static AVPixelFormat
-force_pixel_format (AVPixelFormat, AVPixelFormat out)
-{
-       return out;
-}
-
-FFmpegEncoder::FFmpegEncoder (shared_ptr<const Film> film, weak_ptr<Job> job, boost::filesystem::path output, Format format, bool mixdown_to_stereo)
+using boost::optional;
+using namespace dcpomatic;
+
+/** @param key Key to use to encrypt MP4 outputs */
+FFmpegEncoder::FFmpegEncoder (
+       shared_ptr<const Film> film,
+       weak_ptr<Job> job,
+       boost::filesystem::path output,
+       ExportFormat format,
+       bool mixdown_to_stereo,
+       bool split_reels,
+       int x264_crf
+#ifdef DCPOMATIC_VARIANT_SWAROOP
+       , optional<dcp::Key> key
+       , optional<string> id
+#endif
+       )
        : Encoder (film, job)
-       , _video_options (0)
-       , _history (1000)
-       , _output (output)
+       , _history (200)
 {
-       switch (format) {
-       case FORMAT_PRORES:
-               _pixel_format = AV_PIX_FMT_YUV422P10;
-               _sample_format = AV_SAMPLE_FMT_S16;
-               _video_codec_name = "prores_ks";
-               _audio_codec_name = "pcm_s16le";
-               av_dict_set (&_video_options, "profile", "3", 0);
-               av_dict_set (&_video_options, "threads", "auto", 0);
-               break;
-       case FORMAT_H264:
-               _pixel_format = AV_PIX_FMT_YUV420P;
-               _sample_format = AV_SAMPLE_FMT_FLTP;
-               _video_codec_name = "libx264";
-               _audio_codec_name = "aac";
-               break;
-       }
-
-       _player->set_always_burn_captions (CAPTION_OPEN);
+       _player->set_always_burn_open_subtitles ();
        _player->set_play_referenced ();
 
        int const ch = film->audio_channels ();
@@ -83,131 +72,128 @@ FFmpegEncoder::FFmpegEncoder (shared_ptr<const Film> film, weak_ptr<Job> job, bo
                map = AudioMapping (ch, 2);
                float const overall_gain = 2 / (4 + sqrt(2));
                float const minus_3dB = 1 / sqrt(2);
-               map.set (dcp::LEFT,   0, overall_gain);
-               map.set (dcp::RIGHT,  1, overall_gain);
-               map.set (dcp::CENTRE, 0, overall_gain * minus_3dB);
-               map.set (dcp::CENTRE, 1, overall_gain * minus_3dB);
-               map.set (dcp::LS,     0, overall_gain);
-               map.set (dcp::RS,     1, overall_gain);
-               _pending_audio.reset (new AudioBuffers (2, 0));
+               if (ch == 2) {
+                       map.set (dcp::LEFT, 0, 1);
+                       map.set (dcp::RIGHT, 1, 1);
+               } else if (ch == 4) {
+                       map.set (dcp::LEFT,   0, overall_gain);
+                       map.set (dcp::RIGHT,  1, overall_gain);
+                       map.set (dcp::CENTRE, 0, overall_gain * minus_3dB);
+                       map.set (dcp::CENTRE, 1, overall_gain * minus_3dB);
+                       map.set (dcp::LS,     0, overall_gain);
+               } else if (ch >= 6) {
+                       map.set (dcp::LEFT,   0, overall_gain);
+                       map.set (dcp::RIGHT,  1, overall_gain);
+                       map.set (dcp::CENTRE, 0, overall_gain * minus_3dB);
+                       map.set (dcp::CENTRE, 1, overall_gain * minus_3dB);
+                       map.set (dcp::LS,     0, overall_gain);
+                       map.set (dcp::RS,     1, overall_gain);
+               }
+               /* XXX: maybe we should do something better for >6 channel DCPs */
        } else {
-               _output_audio_channels = ch;
-               map = AudioMapping (ch, ch);
-               _pending_audio.reset (new AudioBuffers (ch, 0));
+               /* Our encoders don't really want to encode any channel count between 9 and 15 inclusive,
+                * so let's just use 16 channel exports for any project with more than 8 channels.
+                */
+               _output_audio_channels = ch > 8 ? 16 : ch;
+               map = AudioMapping (ch, _output_audio_channels);
                for (int i = 0; i < ch; ++i) {
                        map.set (i, i, 1);
                }
        }
 
-       _butler.reset (new Butler (_player, film->log(), map, _output_audio_channels));
-}
+       _butler.reset (new Butler(_player, map, _output_audio_channels, bind(&PlayerVideo::force, _1, FFmpegFileEncoder::pixel_format(format)), true, false));
 
-void
-FFmpegEncoder::setup_video ()
-{
-       _video_codec = avcodec_find_encoder_by_name (_video_codec_name.c_str());
-       if (!_video_codec) {
-               throw runtime_error (String::compose ("could not find FFmpeg encoder %1", _video_codec_name));
-       }
-
-       _video_codec_context = avcodec_alloc_context3 (_video_codec);
-       if (!_video_codec_context) {
-               throw runtime_error ("could not allocate FFmpeg video context");
-       }
-
-       avcodec_get_context_defaults3 (_video_codec_context, _video_codec);
+       int const files = split_reels ? film->reels().size() : 1;
+       for (int i = 0; i < files; ++i) {
 
-       /* Variable quantisation */
-       _video_codec_context->global_quality = 0;
-       _video_codec_context->width = _film->frame_size().width;
-       _video_codec_context->height = _film->frame_size().height;
-       _video_codec_context->time_base = (AVRational) { 1, _film->video_frame_rate() };
-       _video_codec_context->pix_fmt = _pixel_format;
-       _video_codec_context->flags |= AV_CODEC_FLAG_QSCALE | AV_CODEC_FLAG_GLOBAL_HEADER;
-}
+               boost::filesystem::path filename = output;
+               string extension = boost::filesystem::extension (filename);
+               filename = boost::filesystem::change_extension (filename, "");
 
-void
-FFmpegEncoder::setup_audio ()
-{
-       _audio_codec = avcodec_find_encoder_by_name (_audio_codec_name.c_str());
-       if (!_audio_codec) {
-               throw runtime_error (String::compose ("could not find FFmpeg encoder %1", _audio_codec_name));
-       }
+               if (files > 1) {
+                       /// TRANSLATORS: _reel%1 here is to be added to an export filename to indicate
+                       /// which reel it is.  Preserve the %1; it will be replaced with the reel number.
+                       filename = filename.string() + String::compose(_("_reel%1"), i + 1);
+               }
 
-       _audio_codec_context = avcodec_alloc_context3 (_audio_codec);
-       if (!_audio_codec_context) {
-               throw runtime_error ("could not allocate FFmpeg audio context");
+               _file_encoders.push_back (
+                       FileEncoderSet (
+                               _film->frame_size(),
+                               _film->video_frame_rate(),
+                               _film->audio_frame_rate(),
+                               _output_audio_channels,
+                               format,
+                               x264_crf,
+                               _film->three_d(),
+                               filename,
+                               extension
+#ifdef DCPOMATIC_VARIANT_SWAROOP
+                               , key
+                               , id
+#endif
+                               )
+                       );
        }
-
-       avcodec_get_context_defaults3 (_audio_codec_context, _audio_codec);
-
-       /* XXX: configurable */
-       _audio_codec_context->bit_rate = 256 * 1024;
-       _audio_codec_context->sample_fmt = _sample_format;
-       _audio_codec_context->sample_rate = _film->audio_frame_rate ();
-       _audio_codec_context->channel_layout = av_get_default_channel_layout (_output_audio_channels);
-       _audio_codec_context->channels = _output_audio_channels;
 }
 
+
 void
 FFmpegEncoder::go ()
 {
-       setup_video ();
-       setup_audio ();
-
-       avformat_alloc_output_context2 (&_format_context, 0, 0, _output.string().c_str());
-       if (!_format_context) {
-               throw runtime_error ("could not allocate FFmpeg format context");
-       }
-
-       _video_stream = avformat_new_stream (_format_context, _video_codec);
-       if (!_video_stream) {
-               throw runtime_error ("could not create FFmpeg output video stream");
-       }
-
-       _audio_stream = avformat_new_stream (_format_context, _audio_codec);
-       if (!_audio_stream) {
-               throw runtime_error ("could not create FFmpeg output audio stream");
-       }
-
-       _video_stream->id = _video_stream_index;
-       _video_stream->codec = _video_codec_context;
-
-       _audio_stream->id = _audio_stream_index;
-       _audio_stream->codec = _audio_codec_context;
-
-       if (avcodec_open2 (_video_codec_context, _video_codec, &_video_options) < 0) {
-               throw runtime_error ("could not open FFmpeg video codec");
-       }
-
-       int r = avcodec_open2 (_audio_codec_context, _audio_codec, 0);
-       if (r < 0) {
-               char buffer[256];
-               av_strerror (r, buffer, sizeof(buffer));
-               throw runtime_error (String::compose ("could not open FFmpeg audio codec (%1)", buffer));
-       }
-
-       if (avio_open_boost (&_format_context->pb, _output, AVIO_FLAG_WRITE) < 0) {
-               throw runtime_error ("could not open FFmpeg output file");
-       }
-
-       if (avformat_write_header (_format_context, 0) < 0) {
-               throw runtime_error ("could not write header to FFmpeg output file");
-       }
-
        {
                shared_ptr<Job> job = _job.lock ();
                DCPOMATIC_ASSERT (job);
                job->sub (_("Encoding"));
        }
 
+       Waker waker;
+
+       list<DCPTimePeriod> reel_periods = _film->reels ();
+       list<DCPTimePeriod>::const_iterator reel = reel_periods.begin ();
+       list<FileEncoderSet>::iterator encoder = _file_encoders.begin ();
+
        DCPTime const video_frame = DCPTime::from_frames (1, _film->video_frame_rate ());
        int const audio_frames = video_frame.frames_round(_film->audio_frame_rate());
        float* interleaved = new float[_output_audio_channels * audio_frames];
        shared_ptr<AudioBuffers> deinterleaved (new AudioBuffers (_output_audio_channels, audio_frames));
+       int const gets_per_frame = _film->three_d() ? 2 : 1;
        for (DCPTime i; i < _film->length(); i += video_frame) {
-               pair<shared_ptr<PlayerVideo>, DCPTime> v = _butler->get_video ();
-               video (v.first, v.second);
+
+               if (_file_encoders.size() > 1 && !reel->contains(i)) {
+                       /* Next reel and file */
+                       ++reel;
+                       ++encoder;
+                       DCPOMATIC_ASSERT (reel != reel_periods.end());
+                       DCPOMATIC_ASSERT (encoder != _file_encoders.end());
+               }
+
+               for (int j = 0; j < gets_per_frame; ++j) {
+                       Butler::Error e;
+                       pair<shared_ptr<PlayerVideo>, DCPTime> v = _butler->get_video (true, &e);
+                       _butler->rethrow ();
+                       if (!v.first) {
+                               throw ProgrammingError(__FILE__, __LINE__, String::compose("butler returned no video; error was %1", static_cast<int>(e)));
+                       }
+                       shared_ptr<FFmpegFileEncoder> fe = encoder->get (v.first->eyes());
+                       if (fe) {
+                               fe->video(v.first, v.second);
+                       }
+               }
+
+               _history.event ();
+
+               {
+                       boost::mutex::scoped_lock lm (_mutex);
+                       _last_time = i;
+               }
+
+               shared_ptr<Job> job = _job.lock ();
+               if (job) {
+                       job->set_progress (float(i.get()) / _film->length().get());
+               }
+
+               waker.nudge ();
+
                _butler->get_audio (interleaved, audio_frames);
                /* XXX: inefficient; butler interleaves and we deinterleave again */
                float* p = interleaved;
@@ -216,227 +202,103 @@ FFmpegEncoder::go ()
                                deinterleaved->data(k)[j] = *p++;
                        }
                }
-               audio (deinterleaved);
+               encoder->audio (deinterleaved);
        }
        delete[] interleaved;
 
-       if (_pending_audio->frames() > 0) {
-               audio_frame (_pending_audio->frames ());
-       }
-
-       /* Flush */
-
-       bool flushed_video = false;
-       bool flushed_audio = false;
-
-       while (!flushed_video || !flushed_audio) {
-               AVPacket packet;
-               av_init_packet (&packet);
-               packet.data = 0;
-               packet.size = 0;
-
-               int got_packet;
-               avcodec_encode_video2 (_video_codec_context, &packet, 0, &got_packet);
-               if (got_packet) {
-                       packet.stream_index = 0;
-                       av_interleaved_write_frame (_format_context, &packet);
-               } else {
-                       flushed_video = true;
-               }
-               av_packet_unref (&packet);
-
-               av_init_packet (&packet);
-               packet.data = 0;
-               packet.size = 0;
-
-               avcodec_encode_audio2 (_audio_codec_context, &packet, 0, &got_packet);
-               if (got_packet) {
-                       packet.stream_index = 0;
-                       av_interleaved_write_frame (_format_context, &packet);
-               } else {
-                       flushed_audio = true;
-               }
-               av_packet_unref (&packet);
+       BOOST_FOREACH (FileEncoderSet i, _file_encoders) {
+               i.flush ();
        }
-
-       av_write_trailer (_format_context);
-
-       avcodec_close (_video_codec_context);
-       avcodec_close (_audio_codec_context);
-       avio_close (_format_context->pb);
-       avformat_free_context (_format_context);
 }
 
-void
-FFmpegEncoder::video (shared_ptr<PlayerVideo> video, DCPTime time)
+optional<float>
+FFmpegEncoder::current_rate () const
 {
-       shared_ptr<Image> image = video->image (
-               bind (&Log::dcp_log, _film->log().get(), _1, _2),
-               bind (&force_pixel_format, _1, _pixel_format),
-               true,
-               false
-               );
-
-       AVFrame* frame = av_frame_alloc ();
-       DCPOMATIC_ASSERT (frame);
-
-       _pending_images[image->data()[0]] = image;
-       for (int i = 0; i < 3; ++i) {
-               AVBufferRef* buffer = av_buffer_create(image->data()[i], image->stride()[i] * image->size().height, &buffer_free, this, 0);
-               frame->buf[i] = av_buffer_ref (buffer);
-               frame->data[i] = buffer->data;
-               frame->linesize[i] = image->stride()[i];
-               av_buffer_unref (&buffer);
-       }
-
-       frame->width = image->size().width;
-       frame->height = image->size().height;
-       frame->format = _pixel_format;
-       frame->pts = time.seconds() / av_q2d (_video_stream->time_base);
-
-       AVPacket packet;
-       av_init_packet (&packet);
-       packet.data = 0;
-       packet.size = 0;
-
-       int got_packet;
-       if (avcodec_encode_video2 (_video_codec_context, &packet, frame, &got_packet) < 0) {
-               throw EncodeError ("FFmpeg video encode failed");
-       }
-
-       if (got_packet && packet.size) {
-               packet.stream_index = _video_stream_index;
-               av_interleaved_write_frame (_format_context, &packet);
-               av_packet_unref (&packet);
-       }
-
-       av_frame_free (&frame);
-
-       _history.event ();
-
-       {
-               boost::mutex::scoped_lock lm (_mutex);
-               _last_time = time;
-       }
-
-       shared_ptr<Job> job = _job.lock ();
-       if (job) {
-               job->set_progress (float(time.get()) / _film->length().get());
-       }
+       return _history.rate ();
 }
 
-/** Called when the player gives us some audio */
-void
-FFmpegEncoder::audio (shared_ptr<AudioBuffers> audio)
+Frame
+FFmpegEncoder::frames_done () const
 {
-       _pending_audio->append (audio);
-
-       int frame_size = _audio_codec_context->frame_size;
-       if (frame_size == 0) {
-               /* codec has AV_CODEC_CAP_VARIABLE_FRAME_SIZE */
-               frame_size = _film->audio_frame_rate() / _film->video_frame_rate();
-       }
-
-       while (_pending_audio->frames() >= frame_size) {
-               audio_frame (frame_size);
-       }
+       boost::mutex::scoped_lock lm (_mutex);
+       return _last_time.frames_round (_film->video_frame_rate ());
 }
 
-void
-FFmpegEncoder::audio_frame (int size)
+FFmpegEncoder::FileEncoderSet::FileEncoderSet (
+       dcp::Size video_frame_size,
+       int video_frame_rate,
+       int audio_frame_rate,
+       int channels,
+       ExportFormat format,
+       int x264_crf,
+       bool three_d,
+       boost::filesystem::path output,
+       string extension
+#ifdef DCPOMATIC_VARIANT_SWAROOP
+       , optional<dcp::Key> key
+       , optional<string> id
+#endif
+       )
 {
-       DCPOMATIC_ASSERT (size);
-
-       AVFrame* frame = av_frame_alloc ();
-       DCPOMATIC_ASSERT (frame);
-
-       int const channels = _pending_audio->channels();
-       DCPOMATIC_ASSERT (channels);
-
-       int const buffer_size = av_samples_get_buffer_size (0, channels, size, _audio_codec_context->sample_fmt, 0);
-       DCPOMATIC_ASSERT (buffer_size >= 0);
-
-       void* samples = av_malloc (buffer_size);
-       DCPOMATIC_ASSERT (samples);
-
-       frame->nb_samples = size;
-       int r = avcodec_fill_audio_frame (frame, channels, _audio_codec_context->sample_fmt, (const uint8_t *) samples, buffer_size, 0);
-       DCPOMATIC_ASSERT (r >= 0);
-
-       float** p = _pending_audio->data ();
-       switch (_audio_codec_context->sample_fmt) {
-       case AV_SAMPLE_FMT_S16:
-       {
-               int16_t* q = reinterpret_cast<int16_t*> (samples);
-               for (int i = 0; i < size; ++i) {
-                       for (int j = 0; j < channels; ++j) {
-                               *q++ = p[j][i] * 32767;
-                       }
-               }
-               break;
-       }
-       case AV_SAMPLE_FMT_FLTP:
-       {
-               float* q = reinterpret_cast<float*> (samples);
-               for (int i = 0; i < channels; ++i) {
-                       memcpy (q, p[i], sizeof(float) * size);
-                       q += size;
-               }
-               break;
-       }
-       default:
-               DCPOMATIC_ASSERT (false);
-       }
-
-       AVPacket packet;
-       av_init_packet (&packet);
-       packet.data = 0;
-       packet.size = 0;
-
-       int got_packet;
-       if (avcodec_encode_audio2 (_audio_codec_context, &packet, frame, &got_packet) < 0) {
-               throw EncodeError ("FFmpeg audio encode failed");
-       }
-
-       if (got_packet && packet.size) {
-               packet.stream_index = _audio_stream_index;
-               av_interleaved_write_frame (_format_context, &packet);
-               av_packet_unref (&packet);
+       if (three_d) {
+               /// TRANSLATORS: L here is an abbreviation for "left", to indicate the left-eye part of a 3D export
+               _encoders[EYES_LEFT] = shared_ptr<FFmpegFileEncoder>(
+                       new FFmpegFileEncoder(video_frame_size, video_frame_rate, audio_frame_rate, channels, format, x264_crf, String::compose("%1_%2%3", output.string(), _("L"), extension)
+#ifdef DCPOMATIC_VARIANT_SWAROOP
+                                             , key, id
+#endif
+                               )
+                       );
+               /// TRANSLATORS: R here is an abbreviation for "right", to indicate the right-eye part of a 3D export
+               _encoders[EYES_RIGHT] = shared_ptr<FFmpegFileEncoder>(
+                       new FFmpegFileEncoder(video_frame_size, video_frame_rate, audio_frame_rate, channels, format, x264_crf, String::compose("%1_%2%3", output.string(), _("R"), extension)
+#ifdef DCPOMATIC_VARIANT_SWAROOP
+                                             , key, id
+#endif
+                               )
+                       );
+       } else {
+               _encoders[EYES_BOTH]  = shared_ptr<FFmpegFileEncoder>(
+                       new FFmpegFileEncoder(video_frame_size, video_frame_rate, audio_frame_rate, channels, format, x264_crf, String::compose("%1%2", output.string(), extension)
+#ifdef DCPOMATIC_VARIANT_SWAROOP
+                                             , key, id
+#endif
+                               )
+                       );
        }
-
-       av_free (samples);
-       av_frame_free (&frame);
-
-       _pending_audio->trim_start (size);
-}
-
-void
-FFmpegEncoder::subtitle (PlayerCaption, DCPTimePeriod)
-{
-
 }
 
-float
-FFmpegEncoder::current_rate () const
+shared_ptr<FFmpegFileEncoder>
+FFmpegEncoder::FileEncoderSet::get (Eyes eyes) const
 {
-       return _history.rate ();
-}
+       if (_encoders.size() == 1) {
+               /* We are doing a 2D export... */
+               if (eyes == EYES_LEFT) {
+                       /* ...but we got some 3D data; put the left eye into the output... */
+                       eyes = EYES_BOTH;
+               } else if (eyes == EYES_RIGHT) {
+                       /* ...and ignore the right eye.*/
+                       return shared_ptr<FFmpegFileEncoder>();
+               }
+       }
 
-Frame
-FFmpegEncoder::frames_done () const
-{
-       boost::mutex::scoped_lock lm (_mutex);
-       return _last_time.frames_round (_film->video_frame_rate ());
+       map<Eyes, boost::shared_ptr<FFmpegFileEncoder> >::const_iterator i = _encoders.find (eyes);
+       DCPOMATIC_ASSERT (i != _encoders.end());
+       return i->second;
 }
 
 void
-FFmpegEncoder::buffer_free (void* opaque, uint8_t* data)
+FFmpegEncoder::FileEncoderSet::flush ()
 {
-       reinterpret_cast<FFmpegEncoder*>(opaque)->buffer_free2(data);
+       for (map<Eyes, boost::shared_ptr<FFmpegFileEncoder> >::iterator i = _encoders.begin(); i != _encoders.end(); ++i) {
+               i->second->flush ();
+       }
 }
 
 void
-FFmpegEncoder::buffer_free2 (uint8_t* data)
+FFmpegEncoder::FileEncoderSet::audio (shared_ptr<AudioBuffers> a)
 {
-       _pending_images.erase (data);
+       for (map<Eyes, boost::shared_ptr<FFmpegFileEncoder> >::iterator i = _encoders.begin(); i != _encoders.end(); ++i) {
+               i->second->audio (a);
+       }
 }