2 Copyright (C) 2017-2021 Carl Hetherington <cth@carlh.net>
4 This file is part of DCP-o-matic.
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.
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.
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/>.
22 #include "compose.hpp"
24 #include "ffmpeg_encoder.h"
25 #include "ffmpeg_wrapper.h"
31 #include "player_video.h"
33 #include <libavutil/channel_layout.h>
41 using std::make_shared;
42 using std::shared_ptr;
45 using namespace dcpomatic;
46 #if BOOST_VERSION >= 106100
47 using namespace boost::placeholders;
51 int FFmpegFileEncoder::_video_stream_index = 0;
52 int FFmpegFileEncoder::_audio_stream_index_base = 1;
55 class ExportAudioStream
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)
62 _codec = avcodec_find_encoder_by_name (codec_name.c_str());
64 throw EncodeError (String::compose("avcodec_find_encoder_by_name failed for %1", codec_name));
67 _codec_context = avcodec_alloc_context3 (_codec);
68 if (!_codec_context) {
69 throw std::bad_alloc ();
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;
79 int r = avcodec_open2 (_codec_context, _codec, 0);
81 throw EncodeError (N_("avcodec_open2"), N_("ExportAudioStream::ExportAudioStream"), r);
84 _stream = avformat_new_stream (format_context, _codec);
86 throw EncodeError (N_("avformat_new_stream"), N_("ExportAudioStream::ExportAudioStream"));
89 _stream->id = stream_index;
90 _stream->disposition |= AV_DISPOSITION_DEFAULT;
91 r = avcodec_parameters_from_context (_stream->codecpar, _codec_context);
93 throw EncodeError (N_("avcodec_parameters_from_context"), N_("ExportAudioStream::ExportAudioStream"), r);
99 avcodec_close (_codec_context);
102 ExportAudioStream (ExportAudioStream const&) = delete;
103 ExportAudioStream& operator= (ExportAudioStream const&) = delete;
105 int frame_size () const {
106 return _codec_context->frame_size;
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);
117 ffmpeg::Packet packet;
118 r = avcodec_receive_packet (_codec_context, packet.get());
119 if (r == AVERROR_EOF) {
122 throw EncodeError (N_("avcodec_receive_packet"), N_("ExportAudioStream::flush"), r);
125 packet->stream_index = _stream_index;
126 av_interleaved_write_frame (_format_context, packet.get());
130 void write (int size, int channel_offset, int channels, float** data, int64_t sample_offset)
132 DCPOMATIC_ASSERT (size);
134 auto frame = av_frame_alloc ();
135 DCPOMATIC_ASSERT (frame);
137 int const buffer_size = av_samples_get_buffer_size (0, channels, size, _codec_context->sample_fmt, 0);
138 DCPOMATIC_ASSERT (buffer_size >= 0);
140 auto samples = av_malloc (buffer_size);
141 DCPOMATIC_ASSERT (samples);
143 frame->nb_samples = size;
144 frame->format = _codec_context->sample_fmt;
145 frame->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);
149 switch (_codec_context->sample_fmt) {
150 case AV_SAMPLE_FMT_S16:
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;
160 case AV_SAMPLE_FMT_S32:
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;
170 case AV_SAMPLE_FMT_FLTP:
172 float* q = reinterpret_cast<float*> (samples);
173 for (int i = 0; i < channels; ++i) {
174 memcpy (q, data[i + channel_offset], sizeof(float) * size);
180 DCPOMATIC_ASSERT (false);
183 DCPOMATIC_ASSERT (_codec_context->time_base.num == 1);
184 frame->pts = sample_offset * _codec_context->time_base.den / _codec_context->sample_rate;
186 r = avcodec_send_frame (_codec_context, frame);
188 av_frame_free (&frame);
190 throw EncodeError (N_("avcodec_send_frame"), N_("ExportAudioStream::write"), r);
193 ffmpeg::Packet packet;
194 r = avcodec_receive_packet (_codec_context, packet.get());
195 if (r < 0 && r != AVERROR(EAGAIN)) {
196 throw EncodeError (N_("avcodec_receive_packet"), N_("ExportAudioStream::write"), r);
198 packet->stream_index = _stream_index;
199 av_interleaved_write_frame (_format_context, packet.get());
204 AVFormatContext* _format_context;
205 AVCodec const * _codec;
206 AVCodecContext* _codec_context;
212 FFmpegFileEncoder::FFmpegFileEncoder (
213 dcp::Size video_frame_size,
214 int video_frame_rate,
215 int audio_frame_rate,
218 bool audio_stream_per_channel,
220 boost::filesystem::path output
222 : _audio_stream_per_channel (audio_stream_per_channel)
223 , _audio_channels (channels)
225 , _video_frame_size (video_frame_size)
226 , _video_frame_rate (video_frame_rate)
227 , _audio_frame_rate (audio_frame_rate)
229 _pixel_format = pixel_format (format);
232 case ExportFormat::PRORES:
233 _sample_format = AV_SAMPLE_FMT_S16;
234 _video_codec_name = "prores_ks";
235 _audio_codec_name = "pcm_s16le";
236 av_dict_set (&_video_options, "profile", "3", 0);
237 av_dict_set (&_video_options, "threads", "auto", 0);
239 case ExportFormat::H264_AAC:
240 _sample_format = AV_SAMPLE_FMT_FLTP;
241 _video_codec_name = "libx264";
242 _audio_codec_name = "aac";
243 av_dict_set_int (&_video_options, "crf", x264_crf, 0);
245 case ExportFormat::H264_PCM:
246 _sample_format = AV_SAMPLE_FMT_S32;
247 _video_codec_name = "libx264";
248 _audio_codec_name = "pcm_s24le";
249 av_dict_set_int (&_video_options, "crf", x264_crf, 0);
252 DCPOMATIC_ASSERT (false);
255 int r = avformat_alloc_output_context2 (&_format_context, 0, 0, _output.string().c_str());
256 if (!_format_context) {
257 throw EncodeError (N_("avformat_alloc_output_context2"), "FFmpegFileEncoder::FFmpegFileEncoder", r);
263 r = avio_open_boost (&_format_context->pb, _output, AVIO_FLAG_WRITE);
265 throw EncodeError (String::compose(_("Could not open output file %1 (%2)"), _output.string(), r));
268 AVDictionary* options = nullptr;
270 r = avformat_write_header (_format_context, &options);
272 throw EncodeError (N_("avformat_write_header"), N_("FFmpegFileEncoder::FFmpegFileEncoder"), r);
275 _pending_audio.reset (new AudioBuffers(channels, 0));
279 FFmpegFileEncoder::~FFmpegFileEncoder ()
281 _audio_streams.clear ();
282 avcodec_close (_video_codec_context);
283 avformat_free_context (_format_context);
288 FFmpegFileEncoder::pixel_format (ExportFormat format)
291 case ExportFormat::PRORES:
292 return AV_PIX_FMT_YUV422P10;
293 case ExportFormat::H264_AAC:
294 case ExportFormat::H264_PCM:
295 return AV_PIX_FMT_YUV420P;
297 DCPOMATIC_ASSERT (false);
300 return AV_PIX_FMT_YUV422P10;
305 FFmpegFileEncoder::setup_video ()
307 _video_codec = avcodec_find_encoder_by_name (_video_codec_name.c_str());
309 throw EncodeError (String::compose("avcodec_find_encoder_by_name failed for %1", _video_codec_name));
312 _video_codec_context = avcodec_alloc_context3 (_video_codec);
313 if (!_video_codec_context) {
314 throw std::bad_alloc ();
317 /* Variable quantisation */
318 _video_codec_context->global_quality = 0;
319 _video_codec_context->width = _video_frame_size.width;
320 _video_codec_context->height = _video_frame_size.height;
321 _video_codec_context->time_base = (AVRational) { 1, _video_frame_rate };
322 _video_codec_context->pix_fmt = _pixel_format;
323 _video_codec_context->flags |= AV_CODEC_FLAG_QSCALE | AV_CODEC_FLAG_GLOBAL_HEADER;
325 if (avcodec_open2 (_video_codec_context, _video_codec, &_video_options) < 0) {
326 throw EncodeError (N_("avcodec_open"), N_("FFmpegFileEncoder::setup_video"));
329 _video_stream = avformat_new_stream (_format_context, _video_codec);
330 if (!_video_stream) {
331 throw EncodeError (N_("avformat_new_stream"), N_("FFmpegFileEncoder::setup_video"));
334 _video_stream->id = _video_stream_index;
335 int r = avcodec_parameters_from_context (_video_stream->codecpar, _video_codec_context);
337 throw EncodeError (N_("avcodec_parameters_from_context"), N_("FFmpegFileEncoder::setup_video"), r);
343 FFmpegFileEncoder::setup_audio ()
345 int const streams = _audio_stream_per_channel ? _audio_channels : 1;
346 int const channels_per_stream = _audio_stream_per_channel ? 1 : _audio_channels;
348 for (int i = 0; i < streams; ++i) {
349 _audio_streams.push_back(
350 make_shared<ExportAudioStream>(
351 _audio_codec_name, channels_per_stream, _audio_frame_rate, _sample_format, _format_context, _audio_stream_index_base + i
359 FFmpegFileEncoder::flush ()
361 if (_pending_audio->frames() > 0) {
362 audio_frame (_pending_audio->frames ());
365 bool flushed_video = false;
366 bool flushed_audio = false;
368 while (!flushed_video || !flushed_audio) {
369 int r = avcodec_send_frame (_video_codec_context, nullptr);
370 if (r < 0 && r != AVERROR_EOF) {
371 /* We get EOF if we've already flushed the stream once */
372 throw EncodeError (N_("avcodec_send_frame"), N_("FFmpegFileEncoder::flush"), r);
375 ffmpeg::Packet packet;
376 r = avcodec_receive_packet (_video_codec_context, packet.get());
377 if (r == AVERROR_EOF) {
378 flushed_video = true;
380 throw EncodeError (N_("avcodec_receive_packet"), N_("FFmpegFileEncoder::flush"), r);
382 packet->stream_index = _video_stream_index;
383 av_interleaved_write_frame (_format_context, packet.get());
386 flushed_audio = true;
387 for (auto i: _audio_streams) {
389 flushed_audio = false;
394 av_write_trailer (_format_context);
399 FFmpegFileEncoder::video (shared_ptr<PlayerVideo> video, DCPTime time)
401 /* All our output formats are video range at the moment */
402 auto image = video->image (
403 bind (&PlayerVideo::force, _1, _pixel_format),
408 auto frame = av_frame_alloc ();
409 DCPOMATIC_ASSERT (frame);
412 boost::mutex::scoped_lock lm (_pending_images_mutex);
413 _pending_images[image->data()[0]] = image;
416 for (int i = 0; i < 3; ++i) {
417 auto buffer = av_buffer_create(image->data()[i], image->stride()[i] * image->size().height, &buffer_free, this, 0);
418 frame->buf[i] = av_buffer_ref (buffer);
419 frame->data[i] = buffer->data;
420 frame->linesize[i] = image->stride()[i];
421 av_buffer_unref (&buffer);
424 frame->width = image->size().width;
425 frame->height = image->size().height;
426 frame->format = _pixel_format;
427 DCPOMATIC_ASSERT (_video_stream->time_base.num == 1);
428 frame->pts = time.get() * _video_stream->time_base.den / DCPTime::HZ;
430 int r = avcodec_send_frame (_video_codec_context, frame);
431 av_frame_free (&frame);
433 throw EncodeError (N_("avcodec_send_frame"), N_("FFmpegFileEncoder::video"), r);
436 ffmpeg::Packet packet;
437 r = avcodec_receive_packet (_video_codec_context, packet.get());
438 if (r < 0 && r != AVERROR(EAGAIN)) {
439 throw EncodeError (N_("avcodec_receive_packet"), N_("FFmpegFileEncoder::video"), r);
441 packet->stream_index = _video_stream_index;
442 av_interleaved_write_frame (_format_context, packet.get());
447 /** Called when the player gives us some audio */
449 FFmpegFileEncoder::audio (shared_ptr<AudioBuffers> audio)
451 _pending_audio->append (audio);
453 DCPOMATIC_ASSERT (!_audio_streams.empty());
454 int frame_size = _audio_streams[0]->frame_size();
455 if (frame_size == 0) {
456 /* codec has AV_CODEC_CAP_VARIABLE_FRAME_SIZE */
457 frame_size = _audio_frame_rate / _video_frame_rate;
460 while (_pending_audio->frames() >= frame_size) {
461 audio_frame (frame_size);
467 FFmpegFileEncoder::audio_frame (int size)
469 if (_audio_stream_per_channel) {
471 for (auto i: _audio_streams) {
472 i->write (size, offset, 1, _pending_audio->data(), _audio_frames);
476 DCPOMATIC_ASSERT (!_audio_streams.empty());
477 DCPOMATIC_ASSERT (_pending_audio->channels());
478 _audio_streams[0]->write (size, 0, _pending_audio->channels(), _pending_audio->data(), _audio_frames);
481 _pending_audio->trim_start (size);
482 _audio_frames += size;
487 FFmpegFileEncoder::subtitle (PlayerText, DCPTimePeriod)
494 FFmpegFileEncoder::buffer_free (void* opaque, uint8_t* data)
496 reinterpret_cast<FFmpegFileEncoder*>(opaque)->buffer_free2(data);
501 FFmpegFileEncoder::buffer_free2 (uint8_t* data)
503 boost::mutex::scoped_lock lm (_pending_images_mutex);
504 if (_pending_images.find(data) != _pending_images.end()) {
505 _pending_images.erase (data);