2 Copyright (C) 2012 Carl Hetherington <cth@carlh.net>
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation; either version 2 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 /** @file src/ffmpeg_decoder.cc
21 * @brief A decoder using FFmpeg to decode content.
30 #include <boost/lexical_cast.hpp>
33 #include <libavcodec/avcodec.h>
34 #include <libavformat/avformat.h>
35 #include <libswscale/swscale.h>
36 #include <libpostproc/postprocess.h>
41 #include "transcoder.h"
45 #include "exceptions.h"
49 #include "ffmpeg_decoder.h"
50 #include "filter_graph.h"
58 using std::stringstream;
60 using boost::shared_ptr;
61 using boost::optional;
62 using boost::dynamic_pointer_cast;
65 FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, DecodeOptions o)
72 , _video_codec_context (0)
74 , _audio_codec_context (0)
76 , _subtitle_codec_context (0)
85 FFmpegDecoder::~FFmpegDecoder ()
87 if (_audio_codec_context) {
88 avcodec_close (_audio_codec_context);
91 if (_video_codec_context) {
92 avcodec_close (_video_codec_context);
95 if (_subtitle_codec_context) {
96 avcodec_close (_subtitle_codec_context);
101 avformat_close_input (&_format_context);
105 FFmpegDecoder::setup_general ()
109 if (avformat_open_input (&_format_context, _film->content_path().c_str(), 0, 0) < 0) {
110 throw OpenFileError (_film->content_path ());
113 if (avformat_find_stream_info (_format_context, 0) < 0) {
114 throw DecodeError (_("could not find stream information"));
117 /* Find video, audio and subtitle streams and choose the first of each */
119 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
120 AVStream* s = _format_context->streams[i];
121 if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
123 } else if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
125 /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
126 so bodge it here. No idea why we should have to do this.
129 if (s->codec->channel_layout == 0) {
130 s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
133 _audio_streams.push_back (
134 shared_ptr<AudioStream> (
135 new FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channel_layout)
139 } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
140 _subtitle_streams.push_back (
141 shared_ptr<SubtitleStream> (
142 new SubtitleStream (stream_name (s), i)
148 if (_video_stream < 0) {
149 throw DecodeError (N_("could not find video stream"));
152 _frame = avcodec_alloc_frame ();
154 throw DecodeError (N_("could not allocate frame"));
159 FFmpegDecoder::setup_video ()
161 _video_codec_context = _format_context->streams[_video_stream]->codec;
162 _video_codec = avcodec_find_decoder (_video_codec_context->codec_id);
164 if (_video_codec == 0) {
165 throw DecodeError (_("could not find video decoder"));
168 if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) {
169 throw DecodeError (N_("could not open video decoder"));
174 FFmpegDecoder::setup_audio ()
176 if (!_audio_stream) {
180 shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
183 _audio_codec_context = _format_context->streams[ffa->id()]->codec;
184 _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id);
186 if (_audio_codec == 0) {
187 throw DecodeError (_("could not find audio decoder"));
190 if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
191 throw DecodeError (N_("could not open audio decoder"));
196 FFmpegDecoder::setup_subtitle ()
198 if (!_subtitle_stream || _subtitle_stream->id() >= int (_format_context->nb_streams)) {
202 _subtitle_codec_context = _format_context->streams[_subtitle_stream->id()]->codec;
203 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
205 if (_subtitle_codec == 0) {
206 throw DecodeError (_("could not find subtitle decoder"));
209 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
210 throw DecodeError (N_("could not open subtitle decoder"));
216 FFmpegDecoder::pass ()
218 int r = av_read_frame (_format_context, &_packet);
221 if (r != AVERROR_EOF) {
222 /* Maybe we should fail here, but for now we'll just finish off instead */
224 av_strerror (r, buf, sizeof(buf));
225 _film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
228 /* Get any remaining frames */
233 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
237 if (_opt.decode_video) {
238 while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
239 filter_and_emit_video ();
243 if (_audio_stream && _opt.decode_audio) {
244 decode_audio_packet ();
250 avcodec_get_frame_defaults (_frame);
252 shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
254 if (_packet.stream_index == _video_stream && _opt.decode_video) {
257 int const r = avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet);
258 if (r >= 0 && frame_finished) {
260 if (r != _packet.size) {
261 _film->log()->log (String::compose (N_("Used only %1 bytes of %2 in packet"), r, _packet.size));
264 filter_and_emit_video ();
267 } else if (ffa && _packet.stream_index == ffa->id() && _opt.decode_audio) {
268 decode_audio_packet ();
269 } else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt.decode_subtitles) {
273 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) {
274 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
275 indicate that the previous subtitle should stop.
277 if (sub.num_rects > 0) {
278 shared_ptr<TimedSubtitle> ts;
280 emit_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
282 /* some problem with the subtitle; we probably didn't understand it */
285 emit_subtitle (shared_ptr<TimedSubtitle> ());
287 avsubtitle_free (&sub);
291 av_free_packet (&_packet);
295 /** @param data pointer to array of pointers to buffers.
296 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
298 shared_ptr<AudioBuffers>
299 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
301 assert (_film->audio_channels());
302 assert (bytes_per_audio_sample());
304 shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
307 /* Deinterleave and convert to float */
309 assert ((size % (bytes_per_audio_sample() * ffa->channels())) == 0);
311 int const total_samples = size / bytes_per_audio_sample();
312 int const frames = total_samples / _film->audio_channels();
313 shared_ptr<AudioBuffers> audio (new AudioBuffers (ffa->channels(), frames));
315 switch (audio_sample_format()) {
316 case AV_SAMPLE_FMT_S16:
318 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
321 for (int i = 0; i < total_samples; ++i) {
322 audio->data(channel)[sample] = float(*p++) / (1 << 15);
325 if (channel == _film->audio_channels()) {
333 case AV_SAMPLE_FMT_S16P:
335 int16_t** p = reinterpret_cast<int16_t **> (data);
336 for (int i = 0; i < _film->audio_channels(); ++i) {
337 for (int j = 0; j < frames; ++j) {
338 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
344 case AV_SAMPLE_FMT_S32:
346 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
349 for (int i = 0; i < total_samples; ++i) {
350 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
353 if (channel == _film->audio_channels()) {
361 case AV_SAMPLE_FMT_FLT:
363 float* p = reinterpret_cast<float*> (data[0]);
366 for (int i = 0; i < total_samples; ++i) {
367 audio->data(channel)[sample] = *p++;
370 if (channel == _film->audio_channels()) {
378 case AV_SAMPLE_FMT_FLTP:
380 float** p = reinterpret_cast<float**> (data);
381 for (int i = 0; i < _film->audio_channels(); ++i) {
382 memcpy (audio->data(i), p[i], frames * sizeof(float));
388 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
395 FFmpegDecoder::frames_per_second () const
397 AVStream* s = _format_context->streams[_video_stream];
399 if (s->avg_frame_rate.num && s->avg_frame_rate.den) {
400 return av_q2d (s->avg_frame_rate);
403 return av_q2d (s->r_frame_rate);
407 FFmpegDecoder::audio_sample_format () const
409 if (_audio_codec_context == 0) {
410 return (AVSampleFormat) 0;
413 return _audio_codec_context->sample_fmt;
417 FFmpegDecoder::native_size () const
419 return libdcp::Size (_video_codec_context->width, _video_codec_context->height);
423 FFmpegDecoder::pixel_format () const
425 return _video_codec_context->pix_fmt;
429 FFmpegDecoder::time_base_numerator () const
431 return _video_codec_context->time_base.num;
435 FFmpegDecoder::time_base_denominator () const
437 return _video_codec_context->time_base.den;
441 FFmpegDecoder::sample_aspect_ratio_numerator () const
443 return _video_codec_context->sample_aspect_ratio.num;
447 FFmpegDecoder::sample_aspect_ratio_denominator () const
449 return _video_codec_context->sample_aspect_ratio.den;
453 FFmpegDecoder::stream_name (AVStream* s) const
457 AVDictionaryEntry const * lang = av_dict_get (s->metadata, N_("language"), 0, 0);
462 AVDictionaryEntry const * title = av_dict_get (s->metadata, N_("title"), 0, 0);
464 if (!n.str().empty()) {
470 if (n.str().empty()) {
478 FFmpegDecoder::bytes_per_audio_sample () const
480 return av_get_bytes_per_sample (audio_sample_format ());
484 FFmpegDecoder::set_audio_stream (shared_ptr<AudioStream> s)
486 AudioDecoder::set_audio_stream (s);
491 FFmpegDecoder::set_subtitle_stream (shared_ptr<SubtitleStream> s)
493 VideoDecoder::set_subtitle_stream (s);
499 FFmpegDecoder::filter_and_emit_video ()
501 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
503 shared_ptr<FilterGraph> graph;
505 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
506 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
510 if (i == _filter_graphs.end ()) {
511 graph.reset (new FilterGraph (_film, this, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
512 _filter_graphs.push_back (graph);
513 _film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
518 list<shared_ptr<Image> > images = graph->process (_frame);
520 for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
521 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
522 if (bet != AV_NOPTS_VALUE) {
523 emit_video (*i, false, bet * av_q2d (_format_context->streams[_video_stream]->time_base));
525 _film->log()->log ("Dropping frame without PTS");
531 FFmpegDecoder::seek (double p)
533 return do_seek (p, false);
537 FFmpegDecoder::seek_to_last ()
539 /* This AVSEEK_FLAG_BACKWARD in do_seek is a bit of a hack; without it, if we ask for a seek to the same place as last time
540 (used when we change decoder parameters and want to re-fetch the frame) we end up going forwards rather than
541 staying in the same place.
543 return do_seek (last_source_time(), true);
547 FFmpegDecoder::do_seek (double p, bool backwards)
549 int64_t const vt = p / av_q2d (_format_context->streams[_video_stream]->time_base);
551 int const r = av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0);
553 avcodec_flush_buffers (_video_codec_context);
554 if (_subtitle_codec_context) {
555 avcodec_flush_buffers (_subtitle_codec_context);
561 shared_ptr<FFmpegAudioStream>
562 FFmpegAudioStream::create (string t, optional<int> v)
565 /* version < 1; no type in the string, and there's only FFmpeg streams anyway */
566 return shared_ptr<FFmpegAudioStream> (new FFmpegAudioStream (t, v));
572 if (type != N_("ffmpeg")) {
573 return shared_ptr<FFmpegAudioStream> ();
576 return shared_ptr<FFmpegAudioStream> (new FFmpegAudioStream (t, v));
579 FFmpegAudioStream::FFmpegAudioStream (string t, optional<int> version)
587 n >> _id >> channels;
588 _channel_layout = av_get_default_channel_layout (channels);
592 /* Current (marked version 1) */
593 n >> type >> _id >> _sample_rate >> _channel_layout;
594 assert (type == N_("ffmpeg"));
597 for (int i = 0; i < name_index; ++i) {
598 size_t const s = t.find (' ');
599 if (s != string::npos) {
600 t = t.substr (s + 1);
608 FFmpegAudioStream::to_string () const
610 return String::compose (N_("ffmpeg %1 %2 %3 %4"), _id, _sample_rate, _channel_layout, _name);
614 FFmpegDecoder::film_changed (Film::Property p)
620 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
621 _filter_graphs.clear ();
631 /** @return Length (in video frames) according to our content's header */
633 FFmpegDecoder::length () const
635 return (double(_format_context->duration) / AV_TIME_BASE) * frames_per_second();
639 FFmpegDecoder::decode_audio_packet ()
641 shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
644 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
648 AVPacket copy_packet = _packet;
650 while (copy_packet.size > 0) {
653 int const decode_result = avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, ©_packet);
654 if (decode_result >= 0) {
655 if (frame_finished) {
657 /* Where we are in the source, in seconds */
658 double const source_pts_seconds = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
659 * av_frame_get_best_effort_timestamp(_frame);
661 int const data_size = av_samples_get_buffer_size (
662 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
665 assert (_audio_codec_context->channels == _film->audio_channels());
666 Audio (deinterleave_audio (_frame->data, data_size), source_pts_seconds);
669 copy_packet.data += decode_result;
670 copy_packet.size -= decode_result;