Re-work FilmState / Film relationship a bit; Film now inherits from FilmState and...
[dcpomatic.git] / src / lib / j2k_wav_encoder.cc
index b0881daf25296848eaf37562b723d7e607cae6e5..e5d120ad66795d89d725a6101f093be135ae0945 100644 (file)
@@ -27,6 +27,7 @@
 #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"
@@ -45,6 +46,9 @@ using namespace boost;
 
 J2KWAVEncoder::J2KWAVEncoder (shared_ptr<const FilmState> s, shared_ptr<const Options> o, Log* l)
        : Encoder (s, o, l)
+#ifdef HAVE_SWRESAMPLE   
+       , _swr_context (0)
+#endif   
        , _deinterleave_buffer_size (8192)
        , _deinterleave_buffer (0)
        , _process_end (false)
@@ -52,9 +56,9 @@ J2KWAVEncoder::J2KWAVEncoder (shared_ptr<const FilmState> s, shared_ptr<const Op
        /* 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) {
+       for (int i = 0; i < _fs->audio_channels(); ++i) {
                SF_INFO sf_info;
-               sf_info.samplerate = _fs->audio_sample_rate;
+               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;
@@ -101,13 +105,15 @@ J2KWAVEncoder::close_sound_files ()
 }      
 
 void
-J2KWAVEncoder::process_video (shared_ptr<Image> yuv, int frame)
+J2KWAVEncoder::process_video (shared_ptr<Image> yuv, int frame, 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) {
@@ -116,21 +122,25 @@ 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);
+               pair<string, string> const s = Filter::ffmpeg_strings (_fs->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, _fs->subtitle_offset(), _fs->subtitle_scale(),
+                                                 _fs->scaler(), frame, _fs->frames_per_second(), s.second,
                                                  Config::instance()->colour_lut_index (), Config::instance()->j2k_bandwidth (),
                                                  _log
                                                  )
                                          ));
                
                _worker_condition.notify_all ();
+       } else {
+               frame_skipped ();
        }
 }
 
 void
-J2KWAVEncoder::encoder_thread (Server* server)
+J2KWAVEncoder::encoder_thread (ServerDescription* server)
 {
        /* Number of seconds that we currently wait between attempts
           to connect to the server; not relevant for localhost
@@ -139,6 +149,8 @@ J2KWAVEncoder::encoder_thread (Server* 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);
@@ -148,7 +160,9 @@ J2KWAVEncoder::encoder_thread (Server* 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 ();
+               _log->log (String::compose ("Encoder thread %1 pops frame %2 from queue", boost::this_thread::get_id(), vf->frame()));
                _queue.pop_front ();
                
                lock.unlock ();
@@ -160,9 +174,7 @@ J2KWAVEncoder::encoder_thread (Server* 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 ());
+                                       _log->log (String::compose ("%1 was lost, but now she is found; removing backoff", server->host_name ()));
                                }
                                
                                /* This job succeeded, so remove any backoff */
@@ -173,26 +185,29 @@ J2KWAVEncoder::encoder_thread (Server* server)
                                        /* back off more */
                                        remote_backoff += 10;
                                }
-                               stringstream s;
-                               s << "Remote encode on " << server->host_name() << " failed (" << e.what() << "); thread sleeping for " << remote_backoff << "s.";
-                               _log->log (s.str ());
+                               _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 ());
+                               _log->log (String::compose ("Local encode failed (%1)", e.what ()));
                        }
                }
 
                if (encoded) {
                        encoded->write (_opt, vf->frame ());
-                       frame_done ();
+                       frame_done (vf->frame ());
                } else {
                        lock.lock ();
+                       _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 ();
                }
@@ -207,15 +222,43 @@ J2KWAVEncoder::encoder_thread (Server* server)
 }
 
 void
-J2KWAVEncoder::process_begin ()
+J2KWAVEncoder::process_begin (int64_t audio_channel_layout, AVSampleFormat audio_sample_format)
 {
+       if (_fs->audio_sample_rate() != _fs->target_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 ());
+               
+               _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(),
+                       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, (Server *) 0)));
+               _worker_threads.push_back (new boost::thread (boost::bind (&J2KWAVEncoder::encoder_thread, this, (ServerDescription *) 0)));
        }
 
-       vector<Server*> servers = Config::instance()->servers ();
+       vector<ServerDescription*> servers = Config::instance()->servers ();
 
-       for (vector<Server*>::iterator i = servers.begin(); i != servers.end(); ++i) {
+       for (vector<ServerDescription*>::iterator i = servers.begin(); i != servers.end(); ++i) {
                for (int j = 0; j < (*i)->threads (); ++j) {
                        _worker_threads.push_back (new boost::thread (boost::bind (&J2KWAVEncoder::encoder_thread, this, *i)));
                }
@@ -227,8 +270,11 @@ J2KWAVEncoder::process_end ()
 {
        boost::mutex::scoped_lock lock (_worker_mutex);
 
+       _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 ()));
                _worker_condition.notify_all ();
                _worker_condition.wait (lock);
        }
@@ -236,10 +282,60 @@ J2KWAVEncoder::process_end ()
        lock.unlock ();
        
        terminate_worker_threads ();
+
+       _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
+            3. the remote worker fails to encode the image and puts it back on the queue
+            4. the remote worker is then terminated by terminate_worker_threads
+
+            So just mop up anything left in the queue here.
+       */
+
+       for (list<shared_ptr<DCPVideoFrame> >::iterator i = _queue.begin(); i != _queue.end(); ++i) {
+               _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 ());
+               } catch (std::exception& e) {
+                       _log->log (String::compose ("Local encode failed (%1)", e.what ()));
+               }
+       }
+
+#if HAVE_SWRESAMPLE    
+       if (_swr_context) {
+
+               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);
+
+                       if (frames < 0) {
+                               throw EncodeError ("could not run sample-rate converter");
+                       }
+
+                       if (frames == 0) {
+                               break;
+                       }
+
+                       write_audio (buffer, frames * _fs->bytes_per_sample() * _fs->audio_channels());
+               }
+
+               swr_free (&_swr_context);
+       }
+#endif 
+       
        close_sound_files ();
 
        /* Rename .wav.tmp files to .wav */
-       for (int i = 0; i < _fs->audio_channels; ++i) {
+       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));
                }
@@ -248,43 +344,97 @@ J2KWAVEncoder::process_end ()
 }
 
 void
-J2KWAVEncoder::process_audio (uint8_t* data, int data_size)
+J2KWAVEncoder::process_audio (uint8_t* data, int size)
 {
-       /* Size of a sample in bytes */
-       int const sample_size = 2;
+       /* This is a buffer we might use if we are sample-rate converting;
+          it will need freeing if so.
+       */
+       uint8_t* out_buffer = 0;
        
-       /* 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.
+       /* 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];
+
+               uint8_t* out[2] = {
+                       out_buffer, 
+                       0
+               };
+
+               /* Resample audio */
+               int out_frames = swr_convert (_swr_context, out, out_buffer_size_frames, in, frames);
+               if (out_frames < 0) {
+                       throw EncodeError ("could not run sample-rate converter");
+               }
+
+               /* And point our variables at the resampled audio */
+               data = out_buffer;
+               size = out_frames * _fs->audio_channels() * _fs->bytes_per_sample();
+       }
+#endif
+
+       write_audio (data, size);
+
+       /* Delete the sample-rate conversion buffer, if it exists */
+       delete[] out_buffer;
+}
+
+void
+J2KWAVEncoder::write_audio (uint8_t* data, int size)
+{
+       /* 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 = data_size;
+       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 += sample_size) {
-                               for (int k = 0; k < sample_size; ++k) {
+               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 * sample_size) + (j * _fs->audio_channels) + 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) {
+                       switch (_fs->audio_sample_format()) {
                        case AV_SAMPLE_FMT_S16:
-                               sf_write_short (_sound_files[i], (const short *) _deinterleave_buffer, this_time / sample_size);
+                               sf_write_short (_sound_files[i], (const short *) _deinterleave_buffer, this_time / _fs->bytes_per_sample());
                                break;
                        default:
-                               throw DecodeError ("unknown audio sample format");
+                               throw EncodeError ("unknown audio sample format");
                        }
                }
                
                position += this_time;
-               remaining -= this_time * _fs->audio_channels;
+               remaining -= this_time * _fs->audio_channels();
        }
 }
+