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>
32 #include <libavcodec/avcodec.h>
33 #include <libavformat/avformat.h>
34 #include <libswscale/swscale.h>
35 #include <libpostproc/postprocess.h>
40 #include "transcoder.h"
44 #include "exceptions.h"
48 #include "ffmpeg_decoder.h"
49 #include "filter_graph.h"
57 using std::stringstream;
59 using boost::shared_ptr;
60 using boost::optional;
61 using boost::dynamic_pointer_cast;
64 boost::mutex FFmpegDecoder::_mutex;
66 FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, DecodeOptions o)
73 , _video_codec_context (0)
75 , _audio_codec_context (0)
77 , _subtitle_codec_context (0)
86 FFmpegDecoder::~FFmpegDecoder ()
88 boost::mutex::scoped_lock lm (_mutex);
90 if (_audio_codec_context) {
91 avcodec_close (_audio_codec_context);
94 if (_video_codec_context) {
95 avcodec_close (_video_codec_context);
98 if (_subtitle_codec_context) {
99 avcodec_close (_subtitle_codec_context);
104 avformat_close_input (&_format_context);
108 FFmpegDecoder::setup_general ()
112 if (avformat_open_input (&_format_context, _film->content_path().c_str(), 0, 0) < 0) {
113 throw OpenFileError (_film->content_path ());
116 if (avformat_find_stream_info (_format_context, 0) < 0) {
117 throw DecodeError (_("could not find stream information"));
120 /* Find video, audio and subtitle streams and choose the first of each */
122 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
123 AVStream* s = _format_context->streams[i];
124 if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
126 } else if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
128 /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
129 so bodge it here. No idea why we should have to do this.
132 if (s->codec->channel_layout == 0) {
133 s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
136 _audio_streams.push_back (
137 shared_ptr<AudioStream> (
138 new FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channel_layout)
142 } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
143 _subtitle_streams.push_back (
144 shared_ptr<SubtitleStream> (
145 new SubtitleStream (stream_name (s), i)
151 if (_video_stream < 0) {
152 throw DecodeError (N_("could not find video stream"));
155 _frame = avcodec_alloc_frame ();
157 throw DecodeError (N_("could not allocate frame"));
162 FFmpegDecoder::setup_video ()
164 boost::mutex::scoped_lock lm (_mutex);
166 _video_codec_context = _format_context->streams[_video_stream]->codec;
167 _video_codec = avcodec_find_decoder (_video_codec_context->codec_id);
169 if (_video_codec == 0) {
170 throw DecodeError (_("could not find video decoder"));
173 if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) {
174 throw DecodeError (N_("could not open video decoder"));
179 FFmpegDecoder::setup_audio ()
181 boost::mutex::scoped_lock lm (_mutex);
183 if (!_audio_stream) {
187 shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
190 _audio_codec_context = _format_context->streams[ffa->id()]->codec;
191 _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id);
193 if (_audio_codec == 0) {
194 throw DecodeError (_("could not find audio decoder"));
197 if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
198 throw DecodeError (N_("could not open audio decoder"));
203 FFmpegDecoder::setup_subtitle ()
205 boost::mutex::scoped_lock lm (_mutex);
207 if (!_subtitle_stream || _subtitle_stream->id() >= int (_format_context->nb_streams)) {
211 _subtitle_codec_context = _format_context->streams[_subtitle_stream->id()]->codec;
212 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
214 if (_subtitle_codec == 0) {
215 throw DecodeError (_("could not find subtitle decoder"));
218 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
219 throw DecodeError (N_("could not open subtitle decoder"));
225 FFmpegDecoder::pass ()
227 int r = av_read_frame (_format_context, &_packet);
230 if (r != AVERROR_EOF) {
231 /* Maybe we should fail here, but for now we'll just finish off instead */
233 av_strerror (r, buf, sizeof(buf));
234 _film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
237 /* Get any remaining frames */
242 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
246 if (_opt.decode_video) {
247 while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
248 filter_and_emit_video ();
252 if (_audio_stream && _opt.decode_audio) {
253 decode_audio_packet ();
259 avcodec_get_frame_defaults (_frame);
261 shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
263 if (_packet.stream_index == _video_stream && _opt.decode_video) {
266 int const r = avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet);
267 if (r >= 0 && frame_finished) {
269 if (r != _packet.size) {
270 _film->log()->log (String::compose (N_("Used only %1 bytes of %2 in packet"), r, _packet.size));
273 filter_and_emit_video ();
276 } else if (ffa && _packet.stream_index == ffa->id() && _opt.decode_audio) {
277 decode_audio_packet ();
278 } else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt.decode_subtitles) {
282 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) {
283 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
284 indicate that the previous subtitle should stop.
286 if (sub.num_rects > 0) {
287 shared_ptr<TimedSubtitle> ts;
289 emit_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
291 /* some problem with the subtitle; we probably didn't understand it */
294 emit_subtitle (shared_ptr<TimedSubtitle> ());
296 avsubtitle_free (&sub);
300 av_free_packet (&_packet);
304 /** @param data pointer to array of pointers to buffers.
305 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
307 shared_ptr<AudioBuffers>
308 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
310 assert (_film->audio_channels());
311 assert (bytes_per_audio_sample());
313 shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
316 /* Deinterleave and convert to float */
318 assert ((size % (bytes_per_audio_sample() * ffa->channels())) == 0);
320 int const total_samples = size / bytes_per_audio_sample();
321 int const frames = total_samples / _film->audio_channels();
322 shared_ptr<AudioBuffers> audio (new AudioBuffers (ffa->channels(), frames));
324 switch (audio_sample_format()) {
325 case AV_SAMPLE_FMT_S16:
327 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
330 for (int i = 0; i < total_samples; ++i) {
331 audio->data(channel)[sample] = float(*p++) / (1 << 15);
334 if (channel == _film->audio_channels()) {
342 case AV_SAMPLE_FMT_S16P:
344 int16_t** p = reinterpret_cast<int16_t **> (data);
345 for (int i = 0; i < _film->audio_channels(); ++i) {
346 for (int j = 0; j < frames; ++j) {
347 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
353 case AV_SAMPLE_FMT_S32:
355 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
358 for (int i = 0; i < total_samples; ++i) {
359 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
362 if (channel == _film->audio_channels()) {
370 case AV_SAMPLE_FMT_FLT:
372 float* p = reinterpret_cast<float*> (data[0]);
375 for (int i = 0; i < total_samples; ++i) {
376 audio->data(channel)[sample] = *p++;
379 if (channel == _film->audio_channels()) {
387 case AV_SAMPLE_FMT_FLTP:
389 float** p = reinterpret_cast<float**> (data);
390 for (int i = 0; i < _film->audio_channels(); ++i) {
391 memcpy (audio->data(i), p[i], frames * sizeof(float));
397 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
404 FFmpegDecoder::frames_per_second () const
406 AVStream* s = _format_context->streams[_video_stream];
408 if (s->avg_frame_rate.num && s->avg_frame_rate.den) {
409 return av_q2d (s->avg_frame_rate);
412 return av_q2d (s->r_frame_rate);
416 FFmpegDecoder::audio_sample_format () const
418 if (_audio_codec_context == 0) {
419 return (AVSampleFormat) 0;
422 return _audio_codec_context->sample_fmt;
426 FFmpegDecoder::native_size () const
428 return libdcp::Size (_video_codec_context->width, _video_codec_context->height);
432 FFmpegDecoder::pixel_format () const
434 return _video_codec_context->pix_fmt;
438 FFmpegDecoder::time_base_numerator () const
440 return _video_codec_context->time_base.num;
444 FFmpegDecoder::time_base_denominator () const
446 return _video_codec_context->time_base.den;
450 FFmpegDecoder::sample_aspect_ratio_numerator () const
452 return _video_codec_context->sample_aspect_ratio.num;
456 FFmpegDecoder::sample_aspect_ratio_denominator () const
458 return _video_codec_context->sample_aspect_ratio.den;
462 FFmpegDecoder::stream_name (AVStream* s) const
467 AVDictionaryEntry const * lang = av_dict_get (s->metadata, N_("language"), 0, 0);
472 AVDictionaryEntry const * title = av_dict_get (s->metadata, N_("title"), 0, 0);
474 if (!n.str().empty()) {
481 if (n.str().empty()) {
489 FFmpegDecoder::bytes_per_audio_sample () const
491 return av_get_bytes_per_sample (audio_sample_format ());
495 FFmpegDecoder::set_audio_stream (shared_ptr<AudioStream> s)
497 AudioDecoder::set_audio_stream (s);
502 FFmpegDecoder::set_subtitle_stream (shared_ptr<SubtitleStream> s)
504 VideoDecoder::set_subtitle_stream (s);
510 FFmpegDecoder::filter_and_emit_video ()
512 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
513 if (bet == AV_NOPTS_VALUE) {
514 _film->log()->log ("Dropping frame without PTS");
518 shared_ptr<FilterGraph> graph;
521 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
523 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
524 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
528 if (i == _filter_graphs.end ()) {
529 graph = filter_graph_factory (_film, this, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format);
530 _filter_graphs.push_back (graph);
531 _film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
537 list<shared_ptr<Image> > images = graph->process (_frame);
539 for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
540 emit_video (*i, false, bet * av_q2d (_format_context->streams[_video_stream]->time_base));
545 FFmpegDecoder::seek (double p)
547 return do_seek (p, false, false);
551 FFmpegDecoder::seek_to_last ()
553 /* 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
554 (used when we change decoder parameters and want to re-fetch the frame) we end up going forwards rather than
555 staying in the same place.
557 return do_seek (last_source_time(), true, false);
561 FFmpegDecoder::seek_back ()
563 do_seek (last_source_time() - 2.5 / frames_per_second (), true, true);
567 FFmpegDecoder::seek_forward ()
569 do_seek (last_source_time() - 0.5 / frames_per_second(), true, true);
573 FFmpegDecoder::do_seek (double p, bool backwards, bool accurate)
575 int64_t const vt = p / av_q2d (_format_context->streams[_video_stream]->time_base);
577 int const r = av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0);
579 avcodec_flush_buffers (_video_codec_context);
580 if (_subtitle_codec_context) {
581 avcodec_flush_buffers (_subtitle_codec_context);
586 int r = av_read_frame (_format_context, &_packet);
591 avcodec_get_frame_defaults (_frame);
593 if (_packet.stream_index == _video_stream) {
595 int const r = avcodec_decode_video2 (_video_codec_context, _frame, &finished, &_packet);
596 if (r >= 0 && finished) {
597 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
604 av_free_packet (&_packet);
611 shared_ptr<FFmpegAudioStream>
612 FFmpegAudioStream::create (string t, optional<int> v)
615 /* version < 1; no type in the string, and there's only FFmpeg streams anyway */
616 return shared_ptr<FFmpegAudioStream> (new FFmpegAudioStream (t, v));
622 if (type != N_("ffmpeg")) {
623 return shared_ptr<FFmpegAudioStream> ();
626 return shared_ptr<FFmpegAudioStream> (new FFmpegAudioStream (t, v));
629 FFmpegAudioStream::FFmpegAudioStream (string t, optional<int> version)
637 n >> _id >> channels;
638 _channel_layout = av_get_default_channel_layout (channels);
642 /* Current (marked version 1) */
643 n >> type >> _id >> _sample_rate >> _channel_layout;
644 assert (type == N_("ffmpeg"));
647 for (int i = 0; i < name_index; ++i) {
648 size_t const s = t.find (' ');
649 if (s != string::npos) {
650 t = t.substr (s + 1);
658 FFmpegAudioStream::to_string () const
660 return String::compose (N_("ffmpeg %1 %2 %3 %4"), _id, _sample_rate, _channel_layout, _name);
664 FFmpegDecoder::film_changed (Film::Property p)
670 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
671 _filter_graphs.clear ();
681 /** @return Length (in video frames) according to our content's header */
683 FFmpegDecoder::length () const
685 return (double(_format_context->duration) / AV_TIME_BASE) * frames_per_second();
689 FFmpegDecoder::decode_audio_packet ()
691 shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
694 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
698 AVPacket copy_packet = _packet;
700 while (copy_packet.size > 0) {
703 int const decode_result = avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, ©_packet);
704 if (decode_result < 0) {
709 if (frame_finished) {
711 /* Where we are in the source, in seconds */
712 double const source_pts_seconds = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
713 * av_frame_get_best_effort_timestamp(_frame);
715 int const data_size = av_samples_get_buffer_size (
716 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
719 assert (_audio_codec_context->channels == _film->audio_channels());
720 Audio (deinterleave_audio (_frame->data, data_size), source_pts_seconds);
723 copy_packet.data += decode_result;
724 copy_packet.size -= decode_result;