Log message when resampling audio.
[dcpomatic.git] / src / lib / j2k_wav_encoder.cc
index e6a1b285e1ba4d4e01af731f115b28bf043c7f60..24163940078f0d36bd78302c271ddf4a9349690b 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"
 #include "server.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)
+#ifdef HAVE_SWRESAMPLE   
+       , _swr_context (0)
+#endif   
        , _deinterleave_buffer_size (8192)
        , _deinterleave_buffer (0)
        , _process_end (false)
@@ -53,7 +58,7 @@ J2KWAVEncoder::J2KWAVEncoder (shared_ptr<const FilmState> s, shared_ptr<const Op
        */
        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;
@@ -125,11 +130,13 @@ J2KWAVEncoder::process_video (shared_ptr<Image> yuv, int frame)
                                          ));
                
                _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
@@ -173,7 +180,7 @@ J2KWAVEncoder::encoder_thread (Server* server)
                                        remote_backoff += 10;
                                }
                                stringstream s;
-                               s << "Remote encode on " << server->host_name() << " failed (" << e.what() << "); thread sleeping for " << remote_backoff << "s.";
+                               s << "Remote encode of " << vf->frame() << " on " << server->host_name() << " failed (" << e.what() << "); thread sleeping for " << remote_backoff << "s.";
                                _log->log (s.str ());
                        }
                                
@@ -189,7 +196,7 @@ J2KWAVEncoder::encoder_thread (Server* server)
 
                if (encoded) {
                        encoded->write (_opt, vf->frame ());
-                       frame_done ();
+                       frame_done (vf->frame ());
                } else {
                        lock.lock ();
                        _queue.push_front (vf);
@@ -197,7 +204,7 @@ J2KWAVEncoder::encoder_thread (Server* server)
                }
 
                if (remote_backoff > 0) {
-                       sleep (remote_backoff);
+                       dvdomatic_sleep (remote_backoff);
                }
 
                lock.lock ();
@@ -206,15 +213,43 @@ 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 != dcp_audio_sample_rate (_fs->audio_sample_rate)) || (rint (_fs->frames_per_second) != _fs->frames_per_second)) {
+#ifdef HAVE_SWRESAMPLE
+
+               stringstream s;
+               s << "Will resample audio from " << _fs->audio_sample_rate << " to " << target_sample_rate();
+               _log->log (s.str ());
+               
+               _swr_context = swr_alloc_set_opts (
+                       0,
+                       audio_channel_layout,
+                       audio_sample_format,
+                       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)));
                }
@@ -226,8 +261,11 @@ 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);
        }
@@ -235,6 +273,61 @@ 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) {
+               stringstream s;
+               s << "Encode left-over frame " << (*i)->frame();
+               _log->log (s.str ());
+               try {
+                       shared_ptr<EncodedData> e = (*i)->encode_locally ();
+                       e->write (_opt, (*i)->frame ());
+                       frame_done ((*i)->frame ());
+               } catch (std::exception& e) {
+                       stringstream s;
+                       s << "Local encode failed " << e.what() << ".";
+                       _log->log (s.str ());
+               }
+       }
+
+#if HAVE_SWRESAMPLE    
+       if (_swr_context) {
+
+               int mop = 0;
+               while (1) {
+                       uint8_t buffer[256 * _fs->bytes_per_sample() * _fs->audio_channels];
+                       uint8_t* out[1] = {
+                               buffer
+                       };
+
+                       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;
+                       }
+
+                       mop += frames;
+                       write_audio (buffer, frames);
+               }
+
+               swr_free (&_swr_context);
+       }
+#endif 
+       
        close_sound_files ();
 
        /* Rename .wav.tmp files to .wav */
@@ -247,11 +340,64 @@ J2KWAVEncoder::process_end ()
 }
 
 void
-J2KWAVEncoder::process_audio (uint8_t* data, int data_size)
+J2KWAVEncoder::process_audio (uint8_t* data, int size)
+{
+       /* This is a buffer we might use if we are sample-rate converting;
+          it will need freeing if so.
+       */
+       uint8_t* out_buffer = 0;
+       
+       /* 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 * 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, size);
+               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)
 {
        /* 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.
        */
@@ -259,7 +405,7 @@ J2KWAVEncoder::process_audio (uint8_t* data, int data_size)
        /* 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) {
@@ -279,7 +425,7 @@ J2KWAVEncoder::process_audio (uint8_t* data, int data_size)
                                sf_write_short (_sound_files[i], (const short *) _deinterleave_buffer, this_time / sample_size);
                                break;
                        default:
-                               throw DecodeError ("unknown audio sample format");
+                               throw EncodeError ("unknown audio sample format");
                        }
                }
                
@@ -287,3 +433,23 @@ J2KWAVEncoder::process_audio (uint8_t* data, int data_size)
                remaining -= this_time * _fs->audio_channels;
        }
 }
+
+int
+J2KWAVEncoder::target_sample_rate () const
+{
+       double t = dcp_audio_sample_rate (_fs->audio_sample_rate);
+       if (rint (_fs->frames_per_second) != _fs->frames_per_second) {
+               if (_fs->frames_per_second == 23.976) {
+                       /* 24fps drop-frame ie 24 * 1000 / 1001 frames per second;
+                          hence we need to resample the audio to dcp_audio_sample_rate * 1000 / 1001
+                          so that when we play it back at dcp_audio_sample_rate it is sped up
+                          by the same amount that the video is
+                       */
+                       t *= double(1000) / 1001;
+               } else {
+                       throw EncodeError ("unknown fractional frame rate");
+               }
+       }
+
+       return rint (t);
+}