Merge branch 'master' into speed-up
[dcpomatic.git] / src / lib / j2k_wav_encoder.cc
index 87514bf14eff5999a9dc78e8eb995f6ed039c849..134d74623d7087cf3d508778d6ae4cfcc096d40e 100644 (file)
@@ -32,7 +32,6 @@
 #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)
+#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   
-       , _deinterleave_buffer_size (8192)
-       , _deinterleave_buffer (0)
+       , _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 ();
 }
 
@@ -105,7 +109,7 @@ 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);
 
@@ -121,14 +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()
                                                  )
                                          ));
                
@@ -149,7 +154,7 @@ J2KWAVEncoder::encoder_thread (ServerDescription* server)
        
        while (1) {
 
-               TIMING ("encoder thread %1 sleeps", pthread_self ());
+               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);
@@ -159,8 +164,9 @@ J2KWAVEncoder::encoder_thread (ServerDescription* server)
                        return;
                }
 
-               TIMING ("encoder thread %1 wakes with queue of %2", pthread_self(), _queue.size());
+               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 ();
@@ -172,7 +178,7 @@ J2KWAVEncoder::encoder_thread (ServerDescription* server)
                                encoded = vf->encode_remotely (server);
 
                                if (remote_backoff > 0) {
-                                       _log->log (String::compose ("%1 was lost, but now she is found; removing backoff", server->host_name ()));
+                                       _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 */
@@ -183,7 +189,7 @@ J2KWAVEncoder::encoder_thread (ServerDescription* server)
                                        /* back off more */
                                        remote_backoff += 10;
                                }
-                               _log->log (
+                               _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)
@@ -192,19 +198,22 @@ J2KWAVEncoder::encoder_thread (ServerDescription* server)
                                
                } else {
                        try {
-                               TIMING ("encoder thread %1 begins local encode of %2", pthread_self(), vf->frame());
+                               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", pthread_self(), vf->frame());
+                               TIMING ("encoder thread %1 finishes local encode of %2", boost::this_thread::get_id(), vf->frame());
                        } catch (std::exception& e) {
-                               _log->log (String::compose ("Local encode failed (%1)", e.what ()));
+                               _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 ();
                }
@@ -219,23 +228,24 @@ J2KWAVEncoder::encoder_thread (ServerDescription* server)
 }
 
 void
-J2KWAVEncoder::process_begin (int64_t audio_channel_layout, AVSampleFormat audio_sample_format)
+J2KWAVEncoder::process_begin ()
 {
-       if ((_fs->audio_sample_rate != dcp_audio_sample_rate (_fs->audio_sample_rate)) || (rint (_fs->frames_per_second) != _fs->frames_per_second)) {
+       if (_film->audio_stream() && _film->audio_stream()->sample_rate() != _film->target_audio_sample_rate()) {
 #ifdef HAVE_SWRESAMPLE
 
                stringstream s;
-               s << "Will resample audio from " << _fs->audio_sample_rate << " to " << _fs->target_sample_rate();
-               _log->log (s.str ());
-               
+               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,
-                       audio_channel_layout,
-                       audio_sample_format,
-                       _fs->target_sample_rate(),
-                       audio_channel_layout,
-                       audio_sample_format,
-                       _fs->audio_sample_rate,
+                       _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
                        );
                
@@ -267,11 +277,11 @@ J2KWAVEncoder::process_end ()
 {
        boost::mutex::scoped_lock lock (_worker_mutex);
 
-       _log->log ("Clearing queue of " + lexical_cast<string> (_queue.size ()));
+       _film->log()->log ("Clearing queue of " + lexical_cast<string> (_queue.size ()));
 
        /* Keep waking workers until the queue is empty */
        while (!_queue.empty ()) {
-               _log->log ("Waking with " + lexical_cast<string> (_queue.size ()));
+               _film->log()->log ("Waking with " + lexical_cast<string> (_queue.size ()), Log::VERBOSE);
                _worker_condition.notify_all ();
                _worker_condition.wait (lock);
        }
@@ -280,7 +290,7 @@ J2KWAVEncoder::process_end ()
        
        terminate_worker_threads ();
 
-       _log->log ("Mopping up " + lexical_cast<string> (_queue.size()));
+       _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
@@ -292,27 +302,23 @@ J2KWAVEncoder::process_end ()
        */
 
        for (list<shared_ptr<DCPVideoFrame> >::iterator i = _queue.begin(); i != _queue.end(); ++i) {
-               _log->log (String::compose ("Encode left-over frame %1", (*i)->frame ()));
+               _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) {
-                       _log->log (String::compose ("Local encode failed (%1)", e.what ()));
+                       _film->log()->log (String::compose ("Local encode failed (%1)", e.what ()));
                }
        }
 
 #if HAVE_SWRESAMPLE    
-       if (_swr_context) {
+       if (_film->audio_stream() && _swr_context) {
 
+               shared_ptr<AudioBuffers> out (new AudioBuffers (_film->audio_stream()->channels(), 256));
+                       
                while (1) {
-                       uint8_t buffer[256 * _fs->bytes_per_sample() * _fs->audio_channels];
-                       uint8_t* out[2] = {
-                               buffer,
-                               0
-                       };
-
-                       int const frames = swr_convert (_swr_context, out, 256, 0, 0);
+                       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");
@@ -322,116 +328,67 @@ J2KWAVEncoder::process_end ()
                                break;
                        }
 
-                       write_audio (buffer, frames * _fs->bytes_per_sample() * _fs->audio_channels);
+                       out->set_frames (frames);
+                       write_audio (out);
                }
 
                swr_free (&_swr_context);
        }
-#endif 
-       
-       close_sound_files ();
+#endif
 
-       /* 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 (_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 size)
+J2KWAVEncoder::do_process_audio (shared_ptr<AudioBuffers> audio)
 {
-       /* This is a buffer we might use if we are sample-rate converting;
-          it will need freeing if so.
-       */
-       uint8_t* out_buffer = 0;
+       shared_ptr<AudioBuffers> resampled;
        
+#if HAVE_SWRESAMPLE
        /* Maybe sample-rate convert */
-#if HAVE_SWRESAMPLE    
        if (_swr_context) {
 
-               uint8_t const * in[2] = {
-                       data,
-                       0
-               };
-
-               /* Here's samples per channel */
-               int const samples = size / _fs->bytes_per_sample();
-               
-               /* And here's frames (where 1 frame is a collection of samples, 1 for each channel,
-                  so for 5.1 a frame would be 6 samples)
-               */
-               int const frames = samples / _fs->audio_channels;
-
-               /* Compute the resampled frame count and add 32 for luck */
-               int const out_buffer_size_frames = ceil (frames * _fs->target_sample_rate() / _fs->audio_sample_rate) + 32;
-               int const out_buffer_size_bytes = out_buffer_size_frames * _fs->audio_channels * _fs->bytes_per_sample();
-               out_buffer = new uint8_t[out_buffer_size_bytes];
+               /* 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;
 
-               uint8_t* out[2] = {
-                       out_buffer, 
-                       0
-               };
+               resampled.reset (new AudioBuffers (_film->audio_stream()->channels(), max_resampled_frames));
 
                /* Resample audio */
-               int out_frames = swr_convert (_swr_context, out, out_buffer_size_frames, in, frames);
-               if (out_frames < 0) {
+               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);
+               
                /* And point our variables at the resampled audio */
-               data = out_buffer;
-               size = out_frames * _fs->audio_channels * _fs->bytes_per_sample();
+               audio = resampled;
        }
 #endif
 
-       write_audio (data, size);
-
-       /* Delete the sample-rate conversion buffer, if it exists */
-       delete[] out_buffer;
+       write_audio (audio);
 }
 
 void
-J2KWAVEncoder::write_audio (uint8_t* data, int size)
+J2KWAVEncoder::write_audio (shared_ptr<const AudioBuffers> audio)
 {
-       /* XXX: we are assuming that the _deinterleave_buffer_size is a multiple
-          of the sample size and that size is a multiple of _fs->audio_channels * sample_size.
-       */
-
-       assert ((size % (_fs->audio_channels * _fs->bytes_per_sample())) == 0);
-       assert ((_deinterleave_buffer_size % _fs->bytes_per_sample()) == 0);
-       
-       /* 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 = 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 += _fs->bytes_per_sample()) {
-                               for (int k = 0; k < _fs->bytes_per_sample(); ++k) {
-                                       int const to = j + k;
-                                       int const from = position + (i * _fs->bytes_per_sample()) + (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 / _fs->bytes_per_sample());
-                               break;
-                       default:
-                               throw EncodeError ("unknown audio sample format");
-                       }
-               }
-               
-               position += this_time;
-               remaining -= this_time * _fs->audio_channels;
+       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 ();
 }