Merge branch 'master' into speed-up
[dcpomatic.git] / src / lib / j2k_wav_encoder.cc
index ef1e8edc1b1728285276163c0f4f2c129769387c..134d74623d7087cf3d508778d6ae4cfcc096d40e 100644 (file)
 #include <iostream>
 #include <boost/thread.hpp>
 #include <boost/filesystem.hpp>
+#include <boost/lexical_cast.hpp>
 #include <sndfile.h>
 #include <openjpeg.h>
 #include "j2k_wav_encoder.h"
 #include "config.h"
-#include "film_state.h"
 #include "options.h"
 #include "exceptions.h"
 #include "dcp_video_frame.h"
 #include "filter.h"
 #include "log.h"
 #include "cross.h"
-
-using namespace std;
-using namespace boost;
-
-J2KWAVEncoder::J2KWAVEncoder (shared_ptr<const FilmState> s, shared_ptr<const Options> o, Log* l)
-       : Encoder (s, o, l)
-       , _deinterleave_buffer_size (8192)
-       , _deinterleave_buffer (0)
+#include "film.h"
+
+using std::string;
+using std::stringstream;
+using std::list;
+using std::vector;
+using std::pair;
+using std::cout;
+using boost::shared_ptr;
+using boost::thread;
+using boost::lexical_cast;
+
+J2KWAVEncoder::J2KWAVEncoder (shared_ptr<const Film> f, shared_ptr<const Options> o)
+       : Encoder (f, o)
+#ifdef HAVE_SWRESAMPLE   
+       , _swr_context (0)
+#endif   
+       , _audio_frames_written (0)
        , _process_end (false)
 {
-       /* Create sound output files with .tmp suffixes; we will rename
-          them if and when we complete.
-       */
-       for (int i = 0; i < _fs->audio_channels; ++i) {
-               SF_INFO sf_info;
-               sf_info.samplerate = dcp_audio_sample_rate (_fs->audio_sample_rate);
-               /* We write mono files */
-               sf_info.channels = 1;
-               sf_info.format = SF_FORMAT_WAV | SF_FORMAT_PCM_24;
-               SNDFILE* f = sf_open (_opt->multichannel_audio_out_path (i, true).c_str (), SFM_WRITE, &sf_info);
-               if (f == 0) {
-                       throw CreateFileError (_opt->multichannel_audio_out_path (i, true));
+       if (_film->audio_stream()) {
+               /* Create sound output files with .tmp suffixes; we will rename
+                  them if and when we complete.
+               */
+               for (int i = 0; i < _film->audio_channels(); ++i) {
+                       SF_INFO sf_info;
+                       sf_info.samplerate = dcp_audio_sample_rate (_film->audio_stream()->sample_rate());
+                       /* We write mono files */
+                       sf_info.channels = 1;
+                       sf_info.format = SF_FORMAT_WAV | SF_FORMAT_PCM_24;
+                       SNDFILE* f = sf_open (_opt->multichannel_audio_out_path (i, true).c_str (), SFM_WRITE, &sf_info);
+                       if (f == 0) {
+                               throw CreateFileError (_opt->multichannel_audio_out_path (i, true));
+                       }
+                       _sound_files.push_back (f);
                }
-               _sound_files.push_back (f);
        }
-
-       /* Create buffer for deinterleaving audio */
-       _deinterleave_buffer = new uint8_t[_deinterleave_buffer_size];
 }
 
 J2KWAVEncoder::~J2KWAVEncoder ()
 {
        terminate_worker_threads ();
-       delete[] _deinterleave_buffer;
        close_sound_files ();
 }
 
@@ -101,13 +109,15 @@ J2KWAVEncoder::close_sound_files ()
 }      
 
 void
-J2KWAVEncoder::process_video (shared_ptr<Image> yuv, int frame)
+J2KWAVEncoder::do_process_video (shared_ptr<Image> yuv, shared_ptr<Subtitle> sub)
 {
        boost::mutex::scoped_lock lock (_worker_mutex);
 
        /* Wait until the queue has gone down a bit */
        while (_queue.size() >= _worker_threads.size() * 2 && !_process_end) {
+               TIMING ("decoder sleeps with queue of %1", _queue.size());
                _worker_condition.wait (lock);
+               TIMING ("decoder wakes with queue of %1", _queue.size());
        }
 
        if (_process_end) {
@@ -115,13 +125,15 @@ J2KWAVEncoder::process_video (shared_ptr<Image> yuv, int frame)
        }
 
        /* Only do the processing if we don't already have a file for this frame */
-       if (!boost::filesystem::exists (_opt->frame_out_path (frame, false))) {
-               pair<string, string> const s = Filter::ffmpeg_strings (_fs->filters);
+       if (!boost::filesystem::exists (_opt->frame_out_path (_video_frame, false))) {
+               pair<string, string> const s = Filter::ffmpeg_strings (_film->filters());
+               TIMING ("adding to queue of %1", _queue.size ());
                _queue.push_back (boost::shared_ptr<DCPVideoFrame> (
                                          new DCPVideoFrame (
-                                                 yuv, _opt->out_size, _opt->padding, _fs->scaler, frame, _fs->frames_per_second, s.second,
+                                                 yuv, sub, _opt->out_size, _opt->padding, _film->subtitle_offset(), _film->subtitle_scale(),
+                                                 _film->scaler(), _video_frame, _film->frames_per_second(), s.second,
                                                  Config::instance()->colour_lut_index (), Config::instance()->j2k_bandwidth (),
-                                                 _log
+                                                 _film->log()
                                                  )
                                          ));
                
@@ -141,6 +153,8 @@ J2KWAVEncoder::encoder_thread (ServerDescription* server)
        int remote_backoff = 0;
        
        while (1) {
+
+               TIMING ("encoder thread %1 sleeps", boost::this_thread::get_id());
                boost::mutex::scoped_lock lock (_worker_mutex);
                while (_queue.empty () && !_process_end) {
                        _worker_condition.wait (lock);
@@ -150,7 +164,9 @@ J2KWAVEncoder::encoder_thread (ServerDescription* server)
                        return;
                }
 
+               TIMING ("encoder thread %1 wakes with queue of %2", boost::this_thread::get_id(), _queue.size());
                boost::shared_ptr<DCPVideoFrame> vf = _queue.front ();
+               _film->log()->log (String::compose ("Encoder thread %1 pops frame %2 from queue", boost::this_thread::get_id(), vf->frame()), Log::VERBOSE);
                _queue.pop_front ();
                
                lock.unlock ();
@@ -162,9 +178,7 @@ J2KWAVEncoder::encoder_thread (ServerDescription* server)
                                encoded = vf->encode_remotely (server);
 
                                if (remote_backoff > 0) {
-                                       stringstream s;
-                                       s << server->host_name() << " was lost, but now she is found; removing backoff";
-                                       _log->log (s.str ());
+                                       _film->log()->log (String::compose ("%1 was lost, but now she is found; removing backoff", server->host_name ()));
                                }
                                
                                /* This job succeeded, so remove any backoff */
@@ -175,26 +189,31 @@ J2KWAVEncoder::encoder_thread (ServerDescription* server)
                                        /* back off more */
                                        remote_backoff += 10;
                                }
-                               stringstream s;
-                               s << "Remote encode of " << vf->frame() << " on " << server->host_name() << " failed (" << e.what() << "); thread sleeping for " << remote_backoff << "s.";
-                               _log->log (s.str ());
+                               _film->log()->log (
+                                       String::compose (
+                                               "Remote encode of %1 on %2 failed (%3); thread sleeping for %4s",
+                                               vf->frame(), server->host_name(), e.what(), remote_backoff)
+                                       );
                        }
                                
                } else {
                        try {
+                               TIMING ("encoder thread %1 begins local encode of %2", boost::this_thread::get_id(), vf->frame());
                                encoded = vf->encode_locally ();
+                               TIMING ("encoder thread %1 finishes local encode of %2", boost::this_thread::get_id(), vf->frame());
                        } catch (std::exception& e) {
-                               stringstream s;
-                               s << "Local encode failed " << e.what() << ".";
-                               _log->log (s.str ());
+                               _film->log()->log (String::compose ("Local encode failed (%1)", e.what ()));
                        }
                }
 
                if (encoded) {
                        encoded->write (_opt, vf->frame ());
-                       frame_done (vf->frame ());
+                       frame_done ();
                } else {
                        lock.lock ();
+                       _film->log()->log (
+                               String::compose ("Encoder thread %1 pushes frame %2 back onto queue after failure", boost::this_thread::get_id(), vf->frame())
+                               );
                        _queue.push_front (vf);
                        lock.unlock ();
                }
@@ -211,6 +230,35 @@ J2KWAVEncoder::encoder_thread (ServerDescription* server)
 void
 J2KWAVEncoder::process_begin ()
 {
+       if (_film->audio_stream() && _film->audio_stream()->sample_rate() != _film->target_audio_sample_rate()) {
+#ifdef HAVE_SWRESAMPLE
+
+               stringstream s;
+               s << "Will resample audio from " << _film->audio_stream()->sample_rate() << " to " << _film->target_audio_sample_rate();
+               _film->log()->log (s.str ());
+
+               /* We will be using planar float data when we call the resampler */
+               _swr_context = swr_alloc_set_opts (
+                       0,
+                       _film->audio_stream()->channel_layout(),
+                       AV_SAMPLE_FMT_FLTP,
+                       _film->target_audio_sample_rate(),
+                       _film->audio_stream()->channel_layout(),
+                       AV_SAMPLE_FMT_FLTP,
+                       _film->audio_stream()->sample_rate(),
+                       0, 0
+                       );
+               
+               swr_init (_swr_context);
+#else
+               throw EncodeError ("Cannot resample audio as libswresample is not present");
+#endif
+       } else {
+#ifdef HAVE_SWRESAMPLE
+               _swr_context = 0;
+#endif         
+       }
+       
        for (int i = 0; i < Config::instance()->num_local_encoding_threads (); ++i) {
                _worker_threads.push_back (new boost::thread (boost::bind (&J2KWAVEncoder::encoder_thread, this, (ServerDescription *) 0)));
        }
@@ -229,8 +277,11 @@ J2KWAVEncoder::process_end ()
 {
        boost::mutex::scoped_lock lock (_worker_mutex);
 
+       _film->log()->log ("Clearing queue of " + lexical_cast<string> (_queue.size ()));
+
        /* Keep waking workers until the queue is empty */
        while (!_queue.empty ()) {
+               _film->log()->log ("Waking with " + lexical_cast<string> (_queue.size ()), Log::VERBOSE);
                _worker_condition.notify_all ();
                _worker_condition.wait (lock);
        }
@@ -239,6 +290,8 @@ J2KWAVEncoder::process_end ()
        
        terminate_worker_threads ();
 
+       _film->log()->log ("Mopping up " + lexical_cast<string> (_queue.size()));
+
        /* The following sequence of events can occur in the above code:
             1. a remote worker takes the last image off the queue
             2. the loop above terminates
@@ -249,69 +302,93 @@ J2KWAVEncoder::process_end ()
        */
 
        for (list<shared_ptr<DCPVideoFrame> >::iterator i = _queue.begin(); i != _queue.end(); ++i) {
-               stringstream s;
-               s << "Encode left-over frame " << (*i)->frame();
-               _log->log (s.str ());
+               _film->log()->log (String::compose ("Encode left-over frame %1", (*i)->frame ()));
                try {
                        shared_ptr<EncodedData> e = (*i)->encode_locally ();
                        e->write (_opt, (*i)->frame ());
-                       frame_done ((*i)->frame ());
+                       frame_done ();
                } catch (std::exception& e) {
-                       stringstream s;
-                       s << "Local encode failed " << e.what() << ".";
-                       _log->log (s.str ());
+                       _film->log()->log (String::compose ("Local encode failed (%1)", e.what ()));
                }
        }
-       
-       close_sound_files ();
 
-       /* Rename .wav.tmp files to .wav */
-       for (int i = 0; i < _fs->audio_channels; ++i) {
-               if (boost::filesystem::exists (_opt->multichannel_audio_out_path (i, false))) {
-                       boost::filesystem::remove (_opt->multichannel_audio_out_path (i, false));
+#if HAVE_SWRESAMPLE    
+       if (_film->audio_stream() && _swr_context) {
+
+               shared_ptr<AudioBuffers> out (new AudioBuffers (_film->audio_stream()->channels(), 256));
+                       
+               while (1) {
+                       int const frames = swr_convert (_swr_context, (uint8_t **) out->data(), 256, 0, 0);
+
+                       if (frames < 0) {
+                               throw EncodeError ("could not run sample-rate converter");
+                       }
+
+                       if (frames == 0) {
+                               break;
+                       }
+
+                       out->set_frames (frames);
+                       write_audio (out);
+               }
+
+               swr_free (&_swr_context);
+       }
+#endif
+
+       if (_film->audio_stream()) {
+               close_sound_files ();
+               
+               /* Rename .wav.tmp files to .wav */
+               for (int i = 0; i < _film->audio_channels(); ++i) {
+                       if (boost::filesystem::exists (_opt->multichannel_audio_out_path (i, false))) {
+                               boost::filesystem::remove (_opt->multichannel_audio_out_path (i, false));
+                       }
+                       boost::filesystem::rename (_opt->multichannel_audio_out_path (i, true), _opt->multichannel_audio_out_path (i, false));
                }
-               boost::filesystem::rename (_opt->multichannel_audio_out_path (i, true), _opt->multichannel_audio_out_path (i, false));
        }
 }
 
 void
-J2KWAVEncoder::process_audio (uint8_t* data, int data_size)
+J2KWAVEncoder::do_process_audio (shared_ptr<AudioBuffers> audio)
 {
-       /* Size of a sample in bytes */
-       int const sample_size = 2;
-       
-       /* XXX: we are assuming that sample_size is right, the _deinterleave_buffer_size is a multiple
-          of the sample size and that data_size is a multiple of _fs->audio_channels * sample_size.
-       */
+       shared_ptr<AudioBuffers> resampled;
        
-       /* XXX: this code is very tricksy and it must be possible to make it simpler ... */
-       
-       /* Number of bytes left to read this time */
-       int remaining = data_size;
-       /* Our position in the output buffers, in bytes */
-       int position = 0;
-       while (remaining > 0) {
-               /* How many bytes of the deinterleaved data to do this time */
-               int this_time = min (remaining / _fs->audio_channels, _deinterleave_buffer_size);
-               for (int i = 0; i < _fs->audio_channels; ++i) {
-                       for (int j = 0; j < this_time; j += sample_size) {
-                               for (int k = 0; k < sample_size; ++k) {
-                                       int const to = j + k;
-                                       int const from = position + (i * sample_size) + (j * _fs->audio_channels) + k;
-                                       _deinterleave_buffer[to] = data[from];
-                               }
-                       }
-                       
-                       switch (_fs->audio_sample_format) {
-                       case AV_SAMPLE_FMT_S16:
-                               sf_write_short (_sound_files[i], (const short *) _deinterleave_buffer, this_time / sample_size);
-                               break;
-                       default:
-                               throw DecodeError ("unknown audio sample format");
-                       }
+#if HAVE_SWRESAMPLE
+       /* Maybe sample-rate convert */
+       if (_swr_context) {
+
+               /* Compute the resampled frames count and add 32 for luck */
+               int const max_resampled_frames = ceil ((int64_t) audio->frames() * _film->target_audio_sample_rate() / _film->audio_stream()->sample_rate()) + 32;
+
+               resampled.reset (new AudioBuffers (_film->audio_stream()->channels(), max_resampled_frames));
+
+               /* Resample audio */
+               int const resampled_frames = swr_convert (
+                       _swr_context, (uint8_t **) resampled->data(), max_resampled_frames, (uint8_t const **) audio->data(), audio->frames()
+                       );
+               
+               if (resampled_frames < 0) {
+                       throw EncodeError ("could not run sample-rate converter");
                }
+
+               resampled->set_frames (resampled_frames);
                
-               position += this_time;
-               remaining -= this_time * _fs->audio_channels;
+               /* And point our variables at the resampled audio */
+               audio = resampled;
        }
+#endif
+
+       write_audio (audio);
 }
+
+void
+J2KWAVEncoder::write_audio (shared_ptr<const AudioBuffers> audio)
+{
+       for (int i = 0; i < _film->audio_channels(); ++i) {
+               sf_write_float (_sound_files[i], audio->data(i), audio->frames());
+       }
+
+       _audio_frames_written += audio->frames ();
+}
+