99637564abd557d523153c9385f04da374ad95bf
[dcpomatic.git] / src / lib / ffmpeg_file_encoder.cc
1 /*
2     Copyright (C) 2017-2018 Carl Hetherington <cth@carlh.net>
3
4     This file is part of DCP-o-matic.
5
6     DCP-o-matic is free software; you can redistribute it and/or modify
7     it under the terms of the GNU General Public License as published by
8     the Free Software Foundation; either version 2 of the License, or
9     (at your option) any later version.
10
11     DCP-o-matic is distributed in the hope that it will be useful,
12     but WITHOUT ANY WARRANTY; without even the implied warranty of
13     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14     GNU General Public License for more details.
15
16     You should have received a copy of the GNU General Public License
17     along with DCP-o-matic.  If not, see <http://www.gnu.org/licenses/>.
18
19 */
20
21 #include "ffmpeg_encoder.h"
22 #include "film.h"
23 #include "job.h"
24 #include "player.h"
25 #include "player_video.h"
26 #include "log.h"
27 #include "image.h"
28 #include "cross.h"
29 #include "compose.hpp"
30 #include <iostream>
31
32 #include "i18n.h"
33
34 using std::string;
35 using std::runtime_error;
36 using std::cout;
37 using std::pair;
38 using std::shared_ptr;
39 using boost::bind;
40 using std::weak_ptr;
41 using boost::optional;
42 using namespace dcpomatic;
43 #if BOOST_VERSION >= 106100
44 using namespace boost::placeholders;
45 #endif
46
47 int FFmpegFileEncoder::_video_stream_index = 0;
48 int FFmpegFileEncoder::_audio_stream_index_base = 1;
49
50
51 class ExportAudioStream : public boost::noncopyable
52 {
53 public:
54         ExportAudioStream (string codec_name, int channels, int frame_rate, AVSampleFormat sample_format, AVFormatContext* format_context, int stream_index)
55                 : _format_context (format_context)
56                 , _stream_index (stream_index)
57         {
58                 _codec = avcodec_find_encoder_by_name (codec_name.c_str());
59                 if (!_codec) {
60                         throw runtime_error (String::compose("could not find FFmpeg encoder %1", codec_name));
61                 }
62
63                 _codec_context = avcodec_alloc_context3 (_codec);
64                 if (!_codec_context) {
65                         throw runtime_error ("could not allocate FFmpeg audio context");
66                 }
67
68                 avcodec_get_context_defaults3 (_codec_context, _codec);
69
70                 /* XXX: configurable */
71                 _codec_context->bit_rate = channels * 128 * 1024;
72                 _codec_context->sample_fmt = sample_format;
73                 _codec_context->sample_rate = frame_rate;
74                 _codec_context->channel_layout = av_get_default_channel_layout (channels);
75                 _codec_context->channels = channels;
76
77                 _stream = avformat_new_stream (format_context, _codec);
78                 if (!_stream) {
79                         throw runtime_error ("could not create FFmpeg output audio stream");
80                 }
81
82                 _stream->id = stream_index;
83                 _stream->disposition |= AV_DISPOSITION_DEFAULT;
84 DCPOMATIC_DISABLE_WARNINGS
85                 _stream->codec = _codec_context;
86 DCPOMATIC_ENABLE_WARNINGS
87
88                 int r = avcodec_open2 (_codec_context, _codec, 0);
89                 if (r < 0) {
90                         char buffer[256];
91                         av_strerror (r, buffer, sizeof(buffer));
92                         throw runtime_error (String::compose("could not open FFmpeg audio codec (%1)", buffer));
93                 }
94         }
95
96         ~ExportAudioStream ()
97         {
98                 avcodec_close (_codec_context);
99         }
100
101         int frame_size () const {
102                 return _codec_context->frame_size;
103         }
104
105         bool flush ()
106         {
107                 AVPacket packet;
108                 av_init_packet (&packet);
109                 packet.data = 0;
110                 packet.size = 0;
111                 bool flushed = false;
112
113                 int got_packet;
114 DCPOMATIC_DISABLE_WARNINGS
115                 avcodec_encode_audio2 (_codec_context, &packet, 0, &got_packet);
116 DCPOMATIC_ENABLE_WARNINGS
117                 if (got_packet) {
118                         packet.stream_index = 0;
119                         av_interleaved_write_frame (_format_context, &packet);
120                 } else {
121                         flushed = true;
122                 }
123                 av_packet_unref (&packet);
124                 return flushed;
125         }
126
127         void write (int size, int channel_offset, int channels, float** data, int64_t sample_offset)
128         {
129                 DCPOMATIC_ASSERT (size);
130
131                 AVFrame* frame = av_frame_alloc ();
132                 DCPOMATIC_ASSERT (frame);
133
134                 int const buffer_size = av_samples_get_buffer_size (0, channels, size, _codec_context->sample_fmt, 0);
135                 DCPOMATIC_ASSERT (buffer_size >= 0);
136
137                 void* samples = av_malloc (buffer_size);
138                 DCPOMATIC_ASSERT (samples);
139
140                 frame->nb_samples = size;
141                 int r = avcodec_fill_audio_frame (frame, channels, _codec_context->sample_fmt, (const uint8_t *) samples, buffer_size, 0);
142                 DCPOMATIC_ASSERT (r >= 0);
143
144                 switch (_codec_context->sample_fmt) {
145                 case AV_SAMPLE_FMT_S16:
146                 {
147                         int16_t* q = reinterpret_cast<int16_t*> (samples);
148                         for (int i = 0; i < size; ++i) {
149                                 for (int j = 0; j < channels; ++j) {
150                                         *q++ = data[j + channel_offset][i] * 32767;
151                                 }
152                         }
153                         break;
154                 }
155                 case AV_SAMPLE_FMT_S32:
156                 {
157                         int32_t* q = reinterpret_cast<int32_t*> (samples);
158                         for (int i = 0; i < size; ++i) {
159                                 for (int j = 0; j < channels; ++j) {
160                                         *q++ = data[j + channel_offset][i] * 2147483647;
161                                 }
162                         }
163                         break;
164                 }
165                 case AV_SAMPLE_FMT_FLTP:
166                 {
167                         float* q = reinterpret_cast<float*> (samples);
168                         for (int i = 0; i < channels; ++i) {
169                                 memcpy (q, data[i + channel_offset], sizeof(float) * size);
170                                 q += size;
171                         }
172                         break;
173                 }
174                 default:
175                         DCPOMATIC_ASSERT (false);
176                 }
177
178                 DCPOMATIC_ASSERT (_codec_context->time_base.num == 1);
179                 frame->pts = sample_offset * _codec_context->time_base.den / _codec_context->sample_rate;
180
181                 AVPacket packet;
182                 av_init_packet (&packet);
183                 packet.data = 0;
184                 packet.size = 0;
185                 int got_packet;
186
187                 DCPOMATIC_DISABLE_WARNINGS
188                 if (avcodec_encode_audio2 (_codec_context, &packet, frame, &got_packet) < 0) {
189                         throw EncodeError ("FFmpeg audio encode failed");
190                 }
191                 DCPOMATIC_ENABLE_WARNINGS
192
193                 if (got_packet && packet.size) {
194                         packet.stream_index = _stream_index;
195                         av_interleaved_write_frame (_format_context, &packet);
196                         av_packet_unref (&packet);
197                 }
198
199                 av_free (samples);
200                 av_frame_free (&frame);
201         }
202
203 private:
204         AVFormatContext* _format_context;
205         AVCodec* _codec;
206         AVCodecContext* _codec_context;
207         AVStream* _stream;
208         int _stream_index;
209 };
210
211
212 FFmpegFileEncoder::FFmpegFileEncoder (
213         dcp::Size video_frame_size,
214         int video_frame_rate,
215         int audio_frame_rate,
216         int channels,
217         ExportFormat format,
218         bool audio_stream_per_channel,
219         int x264_crf,
220         boost::filesystem::path output
221         )
222         : _audio_stream_per_channel (audio_stream_per_channel)
223         , _video_options (0)
224         , _audio_channels (channels)
225         , _output (output)
226         , _video_frame_size (video_frame_size)
227         , _video_frame_rate (video_frame_rate)
228         , _audio_frame_rate (audio_frame_rate)
229         , _audio_frames (0)
230 {
231         _pixel_format = pixel_format (format);
232
233         switch (format) {
234         case EXPORT_FORMAT_PRORES:
235                 _sample_format = AV_SAMPLE_FMT_S16;
236                 _video_codec_name = "prores_ks";
237                 _audio_codec_name = "pcm_s16le";
238                 av_dict_set (&_video_options, "profile", "3", 0);
239                 av_dict_set (&_video_options, "threads", "auto", 0);
240                 break;
241         case EXPORT_FORMAT_H264_AAC:
242                 _sample_format = AV_SAMPLE_FMT_FLTP;
243                 _video_codec_name = "libx264";
244                 _audio_codec_name = "aac";
245                 av_dict_set_int (&_video_options, "crf", x264_crf, 0);
246                 break;
247         case EXPORT_FORMAT_H264_PCM:
248                 _sample_format = AV_SAMPLE_FMT_S32;
249                 _video_codec_name = "libx264";
250                 _audio_codec_name = "pcm_s24le";
251                 av_dict_set_int (&_video_options, "crf", x264_crf, 0);
252                 break;
253         default:
254                 DCPOMATIC_ASSERT (false);
255         }
256
257         int r = avformat_alloc_output_context2 (&_format_context, 0, 0, _output.string().c_str());
258         if (!_format_context) {
259                 throw runtime_error (String::compose("could not allocate FFmpeg format context (%1)", r));
260         }
261
262         setup_video ();
263         setup_audio ();
264
265         r = avio_open_boost (&_format_context->pb, _output, AVIO_FLAG_WRITE);
266         if (r < 0) {
267                 throw runtime_error (String::compose("could not open FFmpeg output file %1 (%2)", _output.string(), r));
268         }
269
270         AVDictionary* options = 0;
271
272         if (avformat_write_header (_format_context, &options) < 0) {
273                 throw runtime_error ("could not write header to FFmpeg output file");
274         }
275
276         _pending_audio.reset (new AudioBuffers(channels, 0));
277 }
278
279
280 FFmpegFileEncoder::~FFmpegFileEncoder ()
281 {
282         _audio_streams.clear ();
283         avcodec_close (_video_codec_context);
284         avformat_free_context (_format_context);
285 }
286
287
288 AVPixelFormat
289 FFmpegFileEncoder::pixel_format (ExportFormat format)
290 {
291         switch (format) {
292         case EXPORT_FORMAT_PRORES:
293                 return AV_PIX_FMT_YUV422P10;
294         case EXPORT_FORMAT_H264_AAC:
295         case EXPORT_FORMAT_H264_PCM:
296                 return AV_PIX_FMT_YUV420P;
297         default:
298                 DCPOMATIC_ASSERT (false);
299         }
300
301         return AV_PIX_FMT_YUV422P10;
302 }
303
304 void
305 FFmpegFileEncoder::setup_video ()
306 {
307         _video_codec = avcodec_find_encoder_by_name (_video_codec_name.c_str());
308         if (!_video_codec) {
309                 throw runtime_error (String::compose ("could not find FFmpeg encoder %1", _video_codec_name));
310         }
311
312         _video_codec_context = avcodec_alloc_context3 (_video_codec);
313         if (!_video_codec_context) {
314                 throw runtime_error ("could not allocate FFmpeg video context");
315         }
316
317         avcodec_get_context_defaults3 (_video_codec_context, _video_codec);
318
319         /* Variable quantisation */
320         _video_codec_context->global_quality = 0;
321         _video_codec_context->width = _video_frame_size.width;
322         _video_codec_context->height = _video_frame_size.height;
323         _video_codec_context->time_base = (AVRational) { 1, _video_frame_rate };
324         _video_codec_context->pix_fmt = _pixel_format;
325         _video_codec_context->flags |= AV_CODEC_FLAG_QSCALE | AV_CODEC_FLAG_GLOBAL_HEADER;
326
327         _video_stream = avformat_new_stream (_format_context, _video_codec);
328         if (!_video_stream) {
329                 throw runtime_error ("could not create FFmpeg output video stream");
330         }
331
332 DCPOMATIC_DISABLE_WARNINGS
333         _video_stream->id = _video_stream_index;
334         _video_stream->codec = _video_codec_context;
335 DCPOMATIC_ENABLE_WARNINGS
336
337         if (avcodec_open2 (_video_codec_context, _video_codec, &_video_options) < 0) {
338                 throw runtime_error ("could not open FFmpeg video codec");
339         }
340 }
341
342
343 void
344 FFmpegFileEncoder::setup_audio ()
345 {
346         int const streams = _audio_stream_per_channel ? _audio_channels : 1;
347         int const channels_per_stream = _audio_stream_per_channel ? 1 : _audio_channels;
348
349         for (int i = 0; i < streams; ++i) {
350                 _audio_streams.push_back(
351                         shared_ptr<ExportAudioStream>(
352                                 new ExportAudioStream(_audio_codec_name, channels_per_stream, _audio_frame_rate, _sample_format, _format_context, _audio_stream_index_base + i)
353                                 )
354                         );
355         }
356 }
357
358
359 void
360 FFmpegFileEncoder::flush ()
361 {
362         if (_pending_audio->frames() > 0) {
363                 audio_frame (_pending_audio->frames ());
364         }
365
366         bool flushed_video = false;
367         bool flushed_audio = false;
368
369         while (!flushed_video || !flushed_audio) {
370                 AVPacket packet;
371                 av_init_packet (&packet);
372                 packet.data = 0;
373                 packet.size = 0;
374
375                 int got_packet;
376 DCPOMATIC_DISABLE_WARNINGS
377                 avcodec_encode_video2 (_video_codec_context, &packet, 0, &got_packet);
378 DCPOMATIC_ENABLE_WARNINGS
379                 if (got_packet) {
380                         packet.stream_index = 0;
381                         av_interleaved_write_frame (_format_context, &packet);
382                 } else {
383                         flushed_video = true;
384                 }
385                 av_packet_unref (&packet);
386
387                 flushed_audio = true;
388                 BOOST_FOREACH (shared_ptr<ExportAudioStream> i, _audio_streams) {
389                         if (!i->flush()) {
390                                 flushed_audio = false;
391                         }
392                 }
393         }
394
395         av_write_trailer (_format_context);
396 }
397
398 void
399 FFmpegFileEncoder::video (shared_ptr<PlayerVideo> video, DCPTime time)
400 {
401         /* All our output formats are video range at the moment */
402         shared_ptr<Image> image = video->image (
403                 bind (&PlayerVideo::force, _1, _pixel_format),
404                 VIDEO_RANGE_VIDEO,
405                 true,
406                 false
407                 );
408
409         AVFrame* frame = av_frame_alloc ();
410         DCPOMATIC_ASSERT (frame);
411
412         {
413                 boost::mutex::scoped_lock lm (_pending_images_mutex);
414                 _pending_images[image->data()[0]] = image;
415         }
416
417         for (int i = 0; i < 3; ++i) {
418                 AVBufferRef* buffer = av_buffer_create(image->data()[i], image->stride()[i] * image->size().height, &buffer_free, this, 0);
419                 frame->buf[i] = av_buffer_ref (buffer);
420                 frame->data[i] = buffer->data;
421                 frame->linesize[i] = image->stride()[i];
422                 av_buffer_unref (&buffer);
423         }
424
425         frame->width = image->size().width;
426         frame->height = image->size().height;
427         frame->format = _pixel_format;
428         DCPOMATIC_ASSERT (_video_stream->time_base.num == 1);
429         frame->pts = time.get() * _video_stream->time_base.den / DCPTime::HZ;
430
431         AVPacket packet;
432         av_init_packet (&packet);
433         packet.data = 0;
434         packet.size = 0;
435
436         int got_packet;
437 DCPOMATIC_DISABLE_WARNINGS
438         if (avcodec_encode_video2 (_video_codec_context, &packet, frame, &got_packet) < 0) {
439                 throw EncodeError ("FFmpeg video encode failed");
440         }
441 DCPOMATIC_ENABLE_WARNINGS
442
443         if (got_packet && packet.size) {
444                 packet.stream_index = _video_stream_index;
445                 av_interleaved_write_frame (_format_context, &packet);
446                 av_packet_unref (&packet);
447         }
448
449         av_frame_free (&frame);
450
451 }
452
453 /** Called when the player gives us some audio */
454 void
455 FFmpegFileEncoder::audio (shared_ptr<AudioBuffers> audio)
456 {
457         _pending_audio->append (audio);
458
459         DCPOMATIC_ASSERT (!_audio_streams.empty());
460         int frame_size = _audio_streams[0]->frame_size();
461         if (frame_size == 0) {
462                 /* codec has AV_CODEC_CAP_VARIABLE_FRAME_SIZE */
463                 frame_size = _audio_frame_rate / _video_frame_rate;
464         }
465
466         while (_pending_audio->frames() >= frame_size) {
467                 audio_frame (frame_size);
468         }
469 }
470
471 void
472 FFmpegFileEncoder::audio_frame (int size)
473 {
474         if (_audio_stream_per_channel) {
475                 int offset = 0;
476                 BOOST_FOREACH (shared_ptr<ExportAudioStream> i, _audio_streams) {
477                         i->write (size, offset, 1, _pending_audio->data(), _audio_frames);
478                         ++offset;
479                 }
480         } else {
481                 DCPOMATIC_ASSERT (!_audio_streams.empty());
482                 DCPOMATIC_ASSERT (_pending_audio->channels());
483                 _audio_streams[0]->write (size, 0, _pending_audio->channels(), _pending_audio->data(), _audio_frames);
484         }
485
486         _pending_audio->trim_start (size);
487         _audio_frames += size;
488 }
489
490 void
491 FFmpegFileEncoder::subtitle (PlayerText, DCPTimePeriod)
492 {
493
494 }
495
496 void
497 FFmpegFileEncoder::buffer_free (void* opaque, uint8_t* data)
498 {
499         reinterpret_cast<FFmpegFileEncoder*>(opaque)->buffer_free2(data);
500 }
501
502 void
503 FFmpegFileEncoder::buffer_free2 (uint8_t* data)
504 {
505         boost::mutex::scoped_lock lm (_pending_images_mutex);
506         if (_pending_images.find(data) != _pending_images.end()) {
507                 _pending_images.erase (data);
508         }
509 }