Bump ffmpeg to 5.1.2 "Riemann"
[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                 av_channel_layout_default(&_codec_context->ch_layout, channels);
77
78                 int r = avcodec_open2 (_codec_context, _codec, 0);
79                 if (r < 0) {
80                         throw EncodeError (N_("avcodec_open2"), N_("ExportAudioStream::ExportAudioStream"), r);
81                 }
82
83                 _stream = avformat_new_stream (format_context, _codec);
84                 if (!_stream) {
85                         throw EncodeError (N_("avformat_new_stream"), N_("ExportAudioStream::ExportAudioStream"));
86                 }
87
88                 _stream->id = stream_index;
89                 _stream->disposition |= AV_DISPOSITION_DEFAULT;
90                 r = avcodec_parameters_from_context (_stream->codecpar, _codec_context);
91                 if (r < 0) {
92                         throw EncodeError (N_("avcodec_parameters_from_context"), N_("ExportAudioStream::ExportAudioStream"), r);
93                 }
94         }
95
96         ~ExportAudioStream ()
97         {
98                 avcodec_close (_codec_context);
99         }
100
101         ExportAudioStream (ExportAudioStream const&) = delete;
102         ExportAudioStream& operator= (ExportAudioStream const&) = delete;
103
104         int frame_size () const {
105                 return _codec_context->frame_size;
106         }
107
108         bool flush ()
109         {
110                 int r = avcodec_send_frame (_codec_context, nullptr);
111                 if (r < 0 && r != AVERROR_EOF) {
112                         /* We get EOF if we've already flushed the stream once */
113                         throw EncodeError (N_("avcodec_send_frame"), N_("ExportAudioStream::flush"), r);
114                 }
115
116                 ffmpeg::Packet packet;
117                 r = avcodec_receive_packet (_codec_context, packet.get());
118                 if (r == AVERROR_EOF) {
119                         return true;
120                 } else if (r < 0) {
121                         throw EncodeError (N_("avcodec_receive_packet"), N_("ExportAudioStream::flush"), r);
122                 }
123
124                 packet->stream_index = _stream_index;
125                 av_interleaved_write_frame (_format_context, packet.get());
126                 return false;
127         }
128
129         void write (int size, int channel_offset, int channels, float* const* data, int64_t sample_offset)
130         {
131                 DCPOMATIC_ASSERT (size);
132
133                 auto frame = av_frame_alloc ();
134                 DCPOMATIC_ASSERT (frame);
135
136                 int line_size;
137                 int const buffer_size = av_samples_get_buffer_size (&line_size, channels, size, _codec_context->sample_fmt, 0);
138                 DCPOMATIC_ASSERT (buffer_size >= 0);
139
140                 auto samples = av_malloc (buffer_size);
141                 DCPOMATIC_ASSERT (samples);
142
143                 frame->nb_samples = size;
144                 frame->format = _codec_context->sample_fmt;
145                 frame->ch_layout.nb_channels = channels;
146                 int r = avcodec_fill_audio_frame (frame, channels, _codec_context->sample_fmt, (const uint8_t *) samples, buffer_size, 0);
147                 DCPOMATIC_ASSERT (r >= 0);
148
149                 switch (_codec_context->sample_fmt) {
150                 case AV_SAMPLE_FMT_S16:
151                 {
152                         int16_t* q = reinterpret_cast<int16_t*> (samples);
153                         for (int i = 0; i < size; ++i) {
154                                 for (int j = 0; j < channels; ++j) {
155                                         *q++ = data[j + channel_offset][i] * 32767;
156                                 }
157                         }
158                         break;
159                 }
160                 case AV_SAMPLE_FMT_S32:
161                 {
162                         int32_t* q = reinterpret_cast<int32_t*> (samples);
163                         for (int i = 0; i < size; ++i) {
164                                 for (int j = 0; j < channels; ++j) {
165                                         *q++ = data[j + channel_offset][i] * 2147483647;
166                                 }
167                         }
168                         break;
169                 }
170                 case AV_SAMPLE_FMT_FLTP:
171                 {
172                         for (int i = 0; i < channels; ++i) {
173                                 memcpy (reinterpret_cast<float*>(static_cast<uint8_t*>(samples) + i * line_size), data[i + channel_offset], sizeof(float) * size);
174                         }
175                         break;
176                 }
177                 default:
178                         DCPOMATIC_ASSERT (false);
179                 }
180
181                 DCPOMATIC_ASSERT (_codec_context->time_base.num == 1);
182                 frame->pts = sample_offset * _codec_context->time_base.den / _codec_context->sample_rate;
183
184                 r = avcodec_send_frame (_codec_context, frame);
185                 av_free (samples);
186                 av_frame_free (&frame);
187                 if (r < 0) {
188                         throw EncodeError (N_("avcodec_send_frame"), N_("ExportAudioStream::write"), r);
189                 }
190
191                 ffmpeg::Packet packet;
192                 r = avcodec_receive_packet (_codec_context, packet.get());
193                 if (r < 0 && r != AVERROR(EAGAIN)) {
194                         throw EncodeError (N_("avcodec_receive_packet"), N_("ExportAudioStream::write"), r);
195                 } else if (r >= 0) {
196                         packet->stream_index = _stream_index;
197                         av_interleaved_write_frame (_format_context, packet.get());
198                 }
199         }
200
201 private:
202         AVFormatContext* _format_context;
203         AVCodec const * _codec;
204         AVCodecContext* _codec_context;
205         AVStream* _stream;
206         int _stream_index;
207 };
208
209
210 FFmpegFileEncoder::FFmpegFileEncoder (
211         dcp::Size video_frame_size,
212         int video_frame_rate,
213         int audio_frame_rate,
214         int channels,
215         ExportFormat format,
216         bool audio_stream_per_channel,
217         int x264_crf,
218         boost::filesystem::path output
219         )
220         : _audio_stream_per_channel (audio_stream_per_channel)
221         , _audio_channels (channels)
222         , _output (output)
223         , _video_frame_size (video_frame_size)
224         , _video_frame_rate (video_frame_rate)
225         , _audio_frame_rate (audio_frame_rate)
226 {
227         _pixel_format = pixel_format (format);
228
229         switch (format) {
230         case ExportFormat::PRORES_4444:
231                 _sample_format = AV_SAMPLE_FMT_S16;
232                 _video_codec_name = "prores_ks";
233                 _audio_codec_name = "pcm_s16le";
234                 av_dict_set(&_video_options, "profile", "4", 0);
235                 av_dict_set(&_video_options, "threads", "auto", 0);
236                 break;
237         case ExportFormat::PRORES_HQ:
238                 _sample_format = AV_SAMPLE_FMT_S16;
239                 _video_codec_name = "prores_ks";
240                 _audio_codec_name = "pcm_s16le";
241                 av_dict_set (&_video_options, "profile", "3", 0);
242                 av_dict_set (&_video_options, "threads", "auto", 0);
243                 break;
244         case ExportFormat::H264_AAC:
245                 _sample_format = AV_SAMPLE_FMT_FLTP;
246                 _video_codec_name = "libx264";
247                 _audio_codec_name = "aac";
248                 av_dict_set_int (&_video_options, "crf", x264_crf, 0);
249                 break;
250         default:
251                 DCPOMATIC_ASSERT (false);
252         }
253
254         int r = avformat_alloc_output_context2 (&_format_context, 0, 0, _output.string().c_str());
255         if (!_format_context) {
256                 throw EncodeError (N_("avformat_alloc_output_context2"), "FFmpegFileEncoder::FFmpegFileEncoder", r);
257         }
258
259         setup_video ();
260         setup_audio ();
261
262         r = avio_open_boost (&_format_context->pb, _output, AVIO_FLAG_WRITE);
263         if (r < 0) {
264                 throw EncodeError (String::compose(_("Could not open output file %1 (%2)"), _output.string(), r));
265         }
266
267         AVDictionary* options = nullptr;
268
269         r = avformat_write_header (_format_context, &options);
270         if (r < 0) {
271                 throw EncodeError (N_("avformat_write_header"), N_("FFmpegFileEncoder::FFmpegFileEncoder"), r);
272         }
273
274         _pending_audio = make_shared<AudioBuffers>(channels, 0);
275 }
276
277
278 FFmpegFileEncoder::~FFmpegFileEncoder ()
279 {
280         _audio_streams.clear ();
281         avcodec_close (_video_codec_context);
282         avio_close (_format_context->pb);
283         avformat_free_context (_format_context);
284 }
285
286
287 AVPixelFormat
288 FFmpegFileEncoder::pixel_format (ExportFormat format)
289 {
290         switch (format) {
291         case ExportFormat::PRORES_4444:
292                 return AV_PIX_FMT_YUV444P10;
293         case ExportFormat::PRORES_HQ:
294                 return AV_PIX_FMT_YUV422P10;
295         case ExportFormat::H264_AAC:
296                 return AV_PIX_FMT_YUV420P;
297         default:
298                 DCPOMATIC_ASSERT (false);
299         }
300
301         return AV_PIX_FMT_YUV422P10;
302 }
303
304
305 void
306 FFmpegFileEncoder::setup_video ()
307 {
308         _video_codec = avcodec_find_encoder_by_name (_video_codec_name.c_str());
309         if (!_video_codec) {
310                 throw EncodeError (String::compose("avcodec_find_encoder_by_name failed for %1", _video_codec_name));
311         }
312
313         _video_codec_context = avcodec_alloc_context3 (_video_codec);
314         if (!_video_codec_context) {
315                 throw std::bad_alloc ();
316         }
317
318         /* Variable quantisation */
319         _video_codec_context->global_quality = 0;
320         _video_codec_context->width = _video_frame_size.width;
321         _video_codec_context->height = _video_frame_size.height;
322         _video_codec_context->time_base = (AVRational) { 1, _video_frame_rate };
323         _video_codec_context->pix_fmt = _pixel_format;
324         _video_codec_context->flags |= AV_CODEC_FLAG_QSCALE | AV_CODEC_FLAG_GLOBAL_HEADER;
325
326         if (avcodec_open2 (_video_codec_context, _video_codec, &_video_options) < 0) {
327                 throw EncodeError (N_("avcodec_open"), N_("FFmpegFileEncoder::setup_video"));
328         }
329
330         _video_stream = avformat_new_stream (_format_context, _video_codec);
331         if (!_video_stream) {
332                 throw EncodeError (N_("avformat_new_stream"), N_("FFmpegFileEncoder::setup_video"));
333         }
334
335         _video_stream->id = _video_stream_index;
336         int r = avcodec_parameters_from_context (_video_stream->codecpar, _video_codec_context);
337         if (r < 0) {
338                 throw EncodeError (N_("avcodec_parameters_from_context"), N_("FFmpegFileEncoder::setup_video"), r);
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                         make_shared<ExportAudioStream>(
352                                 _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                 int r = avcodec_send_frame (_video_codec_context, nullptr);
371                 if (r < 0 && r != AVERROR_EOF) {
372                         /* We get EOF if we've already flushed the stream once */
373                         throw EncodeError (N_("avcodec_send_frame"), N_("FFmpegFileEncoder::flush"), r);
374                 }
375
376                 ffmpeg::Packet packet;
377                 r = avcodec_receive_packet (_video_codec_context, packet.get());
378                 if (r == AVERROR_EOF) {
379                         flushed_video = true;
380                 } else if (r < 0) {
381                         throw EncodeError (N_("avcodec_receive_packet"), N_("FFmpegFileEncoder::flush"), r);
382                 } else {
383                         packet->stream_index = _video_stream_index;
384                         av_interleaved_write_frame (_format_context, packet.get());
385                 }
386
387                 flushed_audio = true;
388                 for (auto i: _audio_streams) {
389                         if (!i->flush()) {
390                                 flushed_audio = false;
391                         }
392                 }
393         }
394
395         av_write_trailer (_format_context);
396 }
397
398
399 void
400 FFmpegFileEncoder::video (shared_ptr<PlayerVideo> video, DCPTime time)
401 {
402         /* All our output formats are video range at the moment */
403         auto image = video->image (
404                 bind (&PlayerVideo::force, _pixel_format),
405                 VideoRange::VIDEO,
406                 false
407                 );
408
409         auto 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                 auto 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         int r = avcodec_send_frame (_video_codec_context, frame);
432         av_frame_free (&frame);
433         if (r < 0) {
434                 throw EncodeError (N_("avcodec_send_frame"), N_("FFmpegFileEncoder::video"), r);
435         }
436
437         ffmpeg::Packet packet;
438         r = avcodec_receive_packet (_video_codec_context, packet.get());
439         if (r < 0 && r != AVERROR(EAGAIN)) {
440                 throw EncodeError (N_("avcodec_receive_packet"), N_("FFmpegFileEncoder::video"), r);
441         } else if (r >= 0) {
442                 packet->stream_index = _video_stream_index;
443                 av_interleaved_write_frame (_format_context, packet.get());
444         }
445 }
446
447
448 /** Called when the player gives us some audio */
449 void
450 FFmpegFileEncoder::audio (shared_ptr<AudioBuffers> audio)
451 {
452         _pending_audio->append (audio);
453
454         DCPOMATIC_ASSERT (!_audio_streams.empty());
455         int frame_size = _audio_streams[0]->frame_size();
456         if (frame_size == 0) {
457                 /* codec has AV_CODEC_CAP_VARIABLE_FRAME_SIZE */
458                 frame_size = _audio_frame_rate / _video_frame_rate;
459         }
460
461         while (_pending_audio->frames() >= frame_size) {
462                 audio_frame (frame_size);
463         }
464 }
465
466
467 void
468 FFmpegFileEncoder::audio_frame (int size)
469 {
470         if (_audio_stream_per_channel) {
471                 int offset = 0;
472                 for (auto i: _audio_streams) {
473                         i->write (size, offset, 1, _pending_audio->data(), _audio_frames);
474                         ++offset;
475                 }
476         } else {
477                 DCPOMATIC_ASSERT (!_audio_streams.empty());
478                 DCPOMATIC_ASSERT (_pending_audio->channels());
479                 _audio_streams[0]->write (size, 0, _pending_audio->channels(), _pending_audio->data(), _audio_frames);
480         }
481
482         _pending_audio->trim_start (size);
483         _audio_frames += size;
484 }
485
486
487 void
488 FFmpegFileEncoder::subtitle (PlayerText, DCPTimePeriod)
489 {
490
491 }
492
493
494 void
495 FFmpegFileEncoder::buffer_free (void* opaque, uint8_t* data)
496 {
497         reinterpret_cast<FFmpegFileEncoder*>(opaque)->buffer_free2(data);
498 }
499
500
501 void
502 FFmpegFileEncoder::buffer_free2 (uint8_t* data)
503 {
504         boost::mutex::scoped_lock lm (_pending_images_mutex);
505         if (_pending_images.find(data) != _pending_images.end()) {
506                 _pending_images.erase (data);
507         }
508 }