d645c4c6b854eb56df0b3bb2ef01b3afaa7b5186
[dcpomatic.git] / src / lib / ffmpeg_file_encoder.cc
1 /*
2     Copyright (C) 2017-2021 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
22 #include "compose.hpp"
23 #include "cross.h"
24 #include "ffmpeg_encoder.h"
25 #include "ffmpeg_wrapper.h"
26 #include "film.h"
27 #include "image.h"
28 #include "job.h"
29 #include "log.h"
30 #include "player.h"
31 #include "player_video.h"
32 extern "C" {
33 #include <libavutil/channel_layout.h>
34 }
35 #include <iostream>
36
37 #include "i18n.h"
38
39
40 using std::cout;
41 using std::make_shared;
42 using std::shared_ptr;
43 using std::string;
44 using boost::bind;
45 using namespace dcpomatic;
46 #if BOOST_VERSION >= 106100
47 using namespace boost::placeholders;
48 #endif
49
50
51 int FFmpegFileEncoder::_video_stream_index = 0;
52 int FFmpegFileEncoder::_audio_stream_index_base = 1;
53
54
55 class ExportAudioStream
56 {
57 public:
58         ExportAudioStream (string codec_name, int channels, int frame_rate, AVSampleFormat sample_format, AVFormatContext* format_context, int stream_index)
59                 : _format_context (format_context)
60                 , _stream_index (stream_index)
61         {
62                 _codec = avcodec_find_encoder_by_name (codec_name.c_str());
63                 if (!_codec) {
64                         throw EncodeError (String::compose("avcodec_find_encoder_by_name failed for %1", codec_name));
65                 }
66
67                 _codec_context = avcodec_alloc_context3 (_codec);
68                 if (!_codec_context) {
69                         throw std::bad_alloc ();
70                 }
71
72                 /* XXX: configurable */
73                 _codec_context->bit_rate = channels * 128 * 1024;
74                 _codec_context->sample_fmt = sample_format;
75                 _codec_context->sample_rate = frame_rate;
76                 _codec_context->channel_layout = av_get_default_channel_layout (channels);
77                 _codec_context->channels = channels;
78
79                 int r = avcodec_open2 (_codec_context, _codec, 0);
80                 if (r < 0) {
81                         throw EncodeError (N_("avcodec_open2"), N_("ExportAudioStream::ExportAudioStream"), r);
82                 }
83
84                 _stream = avformat_new_stream (format_context, _codec);
85                 if (!_stream) {
86                         throw EncodeError (N_("avformat_new_stream"), N_("ExportAudioStream::ExportAudioStream"));
87                 }
88
89                 _stream->id = stream_index;
90                 _stream->disposition |= AV_DISPOSITION_DEFAULT;
91                 r = avcodec_parameters_from_context (_stream->codecpar, _codec_context);
92                 if (r < 0) {
93                         throw EncodeError (N_("avcodec_parameters_from_context"), N_("ExportAudioStream::ExportAudioStream"), r);
94                 }
95         }
96
97         ~ExportAudioStream ()
98         {
99                 avcodec_close (_codec_context);
100         }
101
102         ExportAudioStream (ExportAudioStream const&) = delete;
103         ExportAudioStream& operator= (ExportAudioStream const&) = delete;
104
105         int frame_size () const {
106                 return _codec_context->frame_size;
107         }
108
109         bool flush ()
110         {
111                 int r = avcodec_send_frame (_codec_context, nullptr);
112                 if (r < 0 && r != AVERROR_EOF) {
113                         /* We get EOF if we've already flushed the stream once */
114                         throw EncodeError (N_("avcodec_send_frame"), N_("ExportAudioStream::flush"), r);
115                 }
116
117                 ffmpeg::Packet packet;
118                 r = avcodec_receive_packet (_codec_context, packet.get());
119                 if (r == AVERROR_EOF) {
120                         return true;
121                 } else if (r < 0) {
122                         throw EncodeError (N_("avcodec_receive_packet"), N_("ExportAudioStream::flush"), r);
123                 }
124
125                 packet->stream_index = _stream_index;
126                 av_interleaved_write_frame (_format_context, packet.get());
127                 return false;
128         }
129
130         void write (int size, int channel_offset, int channels, float* const* data, int64_t sample_offset)
131         {
132                 DCPOMATIC_ASSERT (size);
133
134                 auto frame = av_frame_alloc ();
135                 DCPOMATIC_ASSERT (frame);
136
137                 int line_size;
138                 int const buffer_size = av_samples_get_buffer_size (&line_size, channels, size, _codec_context->sample_fmt, 0);
139                 DCPOMATIC_ASSERT (buffer_size >= 0);
140
141                 auto samples = av_malloc (buffer_size);
142                 DCPOMATIC_ASSERT (samples);
143
144                 frame->nb_samples = size;
145                 frame->format = _codec_context->sample_fmt;
146                 frame->channels = channels;
147                 int r = avcodec_fill_audio_frame (frame, channels, _codec_context->sample_fmt, (const uint8_t *) samples, buffer_size, 0);
148                 DCPOMATIC_ASSERT (r >= 0);
149
150                 switch (_codec_context->sample_fmt) {
151                 case AV_SAMPLE_FMT_S16:
152                 {
153                         int16_t* q = reinterpret_cast<int16_t*> (samples);
154                         for (int i = 0; i < size; ++i) {
155                                 for (int j = 0; j < channels; ++j) {
156                                         *q++ = data[j + channel_offset][i] * 32767;
157                                 }
158                         }
159                         break;
160                 }
161                 case AV_SAMPLE_FMT_S32:
162                 {
163                         int32_t* q = reinterpret_cast<int32_t*> (samples);
164                         for (int i = 0; i < size; ++i) {
165                                 for (int j = 0; j < channels; ++j) {
166                                         *q++ = data[j + channel_offset][i] * 2147483647;
167                                 }
168                         }
169                         break;
170                 }
171                 case AV_SAMPLE_FMT_FLTP:
172                 {
173                         for (int i = 0; i < channels; ++i) {
174                                 memcpy (reinterpret_cast<float*>(static_cast<uint8_t*>(samples) + i * line_size), data[i + channel_offset], sizeof(float) * size);
175                         }
176                         break;
177                 }
178                 default:
179                         DCPOMATIC_ASSERT (false);
180                 }
181
182                 DCPOMATIC_ASSERT (_codec_context->time_base.num == 1);
183                 frame->pts = sample_offset * _codec_context->time_base.den / _codec_context->sample_rate;
184
185                 r = avcodec_send_frame (_codec_context, frame);
186                 av_free (samples);
187                 av_frame_free (&frame);
188                 if (r < 0) {
189                         throw EncodeError (N_("avcodec_send_frame"), N_("ExportAudioStream::write"), r);
190                 }
191
192                 ffmpeg::Packet packet;
193                 r = avcodec_receive_packet (_codec_context, packet.get());
194                 if (r < 0 && r != AVERROR(EAGAIN)) {
195                         throw EncodeError (N_("avcodec_receive_packet"), N_("ExportAudioStream::write"), r);
196                 } else if (r >= 0) {
197                         packet->stream_index = _stream_index;
198                         av_interleaved_write_frame (_format_context, packet.get());
199                 }
200         }
201
202 private:
203         AVFormatContext* _format_context;
204         AVCodec const * _codec;
205         AVCodecContext* _codec_context;
206         AVStream* _stream;
207         int _stream_index;
208 };
209
210
211 FFmpegFileEncoder::FFmpegFileEncoder (
212         dcp::Size video_frame_size,
213         int video_frame_rate,
214         int audio_frame_rate,
215         int channels,
216         ExportFormat format,
217         bool audio_stream_per_channel,
218         int x264_crf,
219         boost::filesystem::path output
220         )
221         : _audio_stream_per_channel (audio_stream_per_channel)
222         , _audio_channels (channels)
223         , _output (output)
224         , _video_frame_size (video_frame_size)
225         , _video_frame_rate (video_frame_rate)
226         , _audio_frame_rate (audio_frame_rate)
227 {
228         _pixel_format = pixel_format (format);
229
230         switch (format) {
231         case ExportFormat::PRORES:
232                 _sample_format = AV_SAMPLE_FMT_S16;
233                 _video_codec_name = "prores_ks";
234                 _audio_codec_name = "pcm_s16le";
235                 av_dict_set (&_video_options, "profile", "3", 0);
236                 av_dict_set (&_video_options, "threads", "auto", 0);
237                 break;
238         case ExportFormat::H264_AAC:
239                 _sample_format = AV_SAMPLE_FMT_FLTP;
240                 _video_codec_name = "libx264";
241                 _audio_codec_name = "aac";
242                 av_dict_set_int (&_video_options, "crf", x264_crf, 0);
243                 break;
244         default:
245                 DCPOMATIC_ASSERT (false);
246         }
247
248         int r = avformat_alloc_output_context2 (&_format_context, 0, 0, _output.string().c_str());
249         if (!_format_context) {
250                 throw EncodeError (N_("avformat_alloc_output_context2"), "FFmpegFileEncoder::FFmpegFileEncoder", r);
251         }
252
253         setup_video ();
254         setup_audio ();
255
256         r = avio_open_boost (&_format_context->pb, _output, AVIO_FLAG_WRITE);
257         if (r < 0) {
258                 throw EncodeError (String::compose(_("Could not open output file %1 (%2)"), _output.string(), r));
259         }
260
261         AVDictionary* options = nullptr;
262
263         r = avformat_write_header (_format_context, &options);
264         if (r < 0) {
265                 throw EncodeError (N_("avformat_write_header"), N_("FFmpegFileEncoder::FFmpegFileEncoder"), r);
266         }
267
268         _pending_audio = make_shared<AudioBuffers>(channels, 0);
269 }
270
271
272 FFmpegFileEncoder::~FFmpegFileEncoder ()
273 {
274         _audio_streams.clear ();
275         avcodec_close (_video_codec_context);
276         avformat_free_context (_format_context);
277 }
278
279
280 AVPixelFormat
281 FFmpegFileEncoder::pixel_format (ExportFormat format)
282 {
283         switch (format) {
284         case ExportFormat::PRORES:
285                 return AV_PIX_FMT_YUV422P10;
286         case ExportFormat::H264_AAC:
287                 return AV_PIX_FMT_YUV420P;
288         default:
289                 DCPOMATIC_ASSERT (false);
290         }
291
292         return AV_PIX_FMT_YUV422P10;
293 }
294
295
296 void
297 FFmpegFileEncoder::setup_video ()
298 {
299         _video_codec = avcodec_find_encoder_by_name (_video_codec_name.c_str());
300         if (!_video_codec) {
301                 throw EncodeError (String::compose("avcodec_find_encoder_by_name failed for %1", _video_codec_name));
302         }
303
304         _video_codec_context = avcodec_alloc_context3 (_video_codec);
305         if (!_video_codec_context) {
306                 throw std::bad_alloc ();
307         }
308
309         /* Variable quantisation */
310         _video_codec_context->global_quality = 0;
311         _video_codec_context->width = _video_frame_size.width;
312         _video_codec_context->height = _video_frame_size.height;
313         _video_codec_context->time_base = (AVRational) { 1, _video_frame_rate };
314         _video_codec_context->pix_fmt = _pixel_format;
315         _video_codec_context->flags |= AV_CODEC_FLAG_QSCALE | AV_CODEC_FLAG_GLOBAL_HEADER;
316
317         if (avcodec_open2 (_video_codec_context, _video_codec, &_video_options) < 0) {
318                 throw EncodeError (N_("avcodec_open"), N_("FFmpegFileEncoder::setup_video"));
319         }
320
321         _video_stream = avformat_new_stream (_format_context, _video_codec);
322         if (!_video_stream) {
323                 throw EncodeError (N_("avformat_new_stream"), N_("FFmpegFileEncoder::setup_video"));
324         }
325
326         _video_stream->id = _video_stream_index;
327         int r = avcodec_parameters_from_context (_video_stream->codecpar, _video_codec_context);
328         if (r < 0) {
329                 throw EncodeError (N_("avcodec_parameters_from_context"), N_("FFmpegFileEncoder::setup_video"), r);
330         }
331 }
332
333
334 void
335 FFmpegFileEncoder::setup_audio ()
336 {
337         int const streams = _audio_stream_per_channel ? _audio_channels : 1;
338         int const channels_per_stream = _audio_stream_per_channel ? 1 : _audio_channels;
339
340         for (int i = 0; i < streams; ++i) {
341                 _audio_streams.push_back(
342                         make_shared<ExportAudioStream>(
343                                 _audio_codec_name, channels_per_stream, _audio_frame_rate, _sample_format, _format_context, _audio_stream_index_base + i
344                                 )
345                         );
346         }
347 }
348
349
350 void
351 FFmpegFileEncoder::flush ()
352 {
353         if (_pending_audio->frames() > 0) {
354                 audio_frame (_pending_audio->frames ());
355         }
356
357         bool flushed_video = false;
358         bool flushed_audio = false;
359
360         while (!flushed_video || !flushed_audio) {
361                 int r = avcodec_send_frame (_video_codec_context, nullptr);
362                 if (r < 0 && r != AVERROR_EOF) {
363                         /* We get EOF if we've already flushed the stream once */
364                         throw EncodeError (N_("avcodec_send_frame"), N_("FFmpegFileEncoder::flush"), r);
365                 }
366
367                 ffmpeg::Packet packet;
368                 r = avcodec_receive_packet (_video_codec_context, packet.get());
369                 if (r == AVERROR_EOF) {
370                         flushed_video = true;
371                 } else if (r < 0) {
372                         throw EncodeError (N_("avcodec_receive_packet"), N_("FFmpegFileEncoder::flush"), r);
373                 } else {
374                         packet->stream_index = _video_stream_index;
375                         av_interleaved_write_frame (_format_context, packet.get());
376                 }
377
378                 flushed_audio = true;
379                 for (auto i: _audio_streams) {
380                         if (!i->flush()) {
381                                 flushed_audio = false;
382                         }
383                 }
384         }
385
386         av_write_trailer (_format_context);
387 }
388
389
390 void
391 FFmpegFileEncoder::video (shared_ptr<PlayerVideo> video, DCPTime time)
392 {
393         /* All our output formats are video range at the moment */
394         auto image = video->image (
395                 bind (&PlayerVideo::force, _pixel_format),
396                 VideoRange::VIDEO,
397                 false
398                 );
399
400         auto frame = av_frame_alloc ();
401         DCPOMATIC_ASSERT (frame);
402
403         {
404                 boost::mutex::scoped_lock lm (_pending_images_mutex);
405                 _pending_images[image->data()[0]] = image;
406         }
407
408         for (int i = 0; i < 3; ++i) {
409                 auto buffer = av_buffer_create(image->data()[i], image->stride()[i] * image->size().height, &buffer_free, this, 0);
410                 frame->buf[i] = av_buffer_ref (buffer);
411                 frame->data[i] = buffer->data;
412                 frame->linesize[i] = image->stride()[i];
413                 av_buffer_unref (&buffer);
414         }
415
416         frame->width = image->size().width;
417         frame->height = image->size().height;
418         frame->format = _pixel_format;
419         DCPOMATIC_ASSERT (_video_stream->time_base.num == 1);
420         frame->pts = time.get() * _video_stream->time_base.den / DCPTime::HZ;
421
422         int r = avcodec_send_frame (_video_codec_context, frame);
423         av_frame_free (&frame);
424         if (r < 0) {
425                 throw EncodeError (N_("avcodec_send_frame"), N_("FFmpegFileEncoder::video"), r);
426         }
427
428         ffmpeg::Packet packet;
429         r = avcodec_receive_packet (_video_codec_context, packet.get());
430         if (r < 0 && r != AVERROR(EAGAIN)) {
431                 throw EncodeError (N_("avcodec_receive_packet"), N_("FFmpegFileEncoder::video"), r);
432         } else if (r >= 0) {
433                 packet->stream_index = _video_stream_index;
434                 av_interleaved_write_frame (_format_context, packet.get());
435         }
436 }
437
438
439 /** Called when the player gives us some audio */
440 void
441 FFmpegFileEncoder::audio (shared_ptr<AudioBuffers> audio)
442 {
443         _pending_audio->append (audio);
444
445         DCPOMATIC_ASSERT (!_audio_streams.empty());
446         int frame_size = _audio_streams[0]->frame_size();
447         if (frame_size == 0) {
448                 /* codec has AV_CODEC_CAP_VARIABLE_FRAME_SIZE */
449                 frame_size = _audio_frame_rate / _video_frame_rate;
450         }
451
452         while (_pending_audio->frames() >= frame_size) {
453                 audio_frame (frame_size);
454         }
455 }
456
457
458 void
459 FFmpegFileEncoder::audio_frame (int size)
460 {
461         if (_audio_stream_per_channel) {
462                 int offset = 0;
463                 for (auto i: _audio_streams) {
464                         i->write (size, offset, 1, _pending_audio->data(), _audio_frames);
465                         ++offset;
466                 }
467         } else {
468                 DCPOMATIC_ASSERT (!_audio_streams.empty());
469                 DCPOMATIC_ASSERT (_pending_audio->channels());
470                 _audio_streams[0]->write (size, 0, _pending_audio->channels(), _pending_audio->data(), _audio_frames);
471         }
472
473         _pending_audio->trim_start (size);
474         _audio_frames += size;
475 }
476
477
478 void
479 FFmpegFileEncoder::subtitle (PlayerText, DCPTimePeriod)
480 {
481
482 }
483
484
485 void
486 FFmpegFileEncoder::buffer_free (void* opaque, uint8_t* data)
487 {
488         reinterpret_cast<FFmpegFileEncoder*>(opaque)->buffer_free2(data);
489 }
490
491
492 void
493 FFmpegFileEncoder::buffer_free2 (uint8_t* data)
494 {
495         boost::mutex::scoped_lock lm (_pending_images_mutex);
496         if (_pending_images.find(data) != _pending_images.end()) {
497                 _pending_images.erase (data);
498         }
499 }