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 FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, DecodeOptions o)
71 , _video_codec_context (0)
73 , _audio_codec_context (0)
75 , _subtitle_codec_context (0)
84 FFmpegDecoder::~FFmpegDecoder ()
86 if (_audio_codec_context) {
87 avcodec_close (_audio_codec_context);
90 if (_video_codec_context) {
91 avcodec_close (_video_codec_context);
94 if (_subtitle_codec_context) {
95 avcodec_close (_subtitle_codec_context);
100 avformat_close_input (&_format_context);
104 FFmpegDecoder::setup_general ()
108 if (avformat_open_input (&_format_context, _film->content_path().c_str(), 0, 0) < 0) {
109 throw OpenFileError (_film->content_path ());
112 if (avformat_find_stream_info (_format_context, 0) < 0) {
113 throw DecodeError (_("could not find stream information"));
116 /* Find video, audio and subtitle streams and choose the first of each */
118 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
119 AVStream* s = _format_context->streams[i];
120 if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
122 } else if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
124 /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
125 so bodge it here. No idea why we should have to do this.
128 if (s->codec->channel_layout == 0) {
129 s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
132 _audio_streams.push_back (
133 shared_ptr<AudioStream> (
134 new FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channel_layout)
138 } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
139 _subtitle_streams.push_back (
140 shared_ptr<SubtitleStream> (
141 new SubtitleStream (stream_name (s), i)
147 if (_video_stream < 0) {
148 throw DecodeError (N_("could not find video stream"));
151 _frame = avcodec_alloc_frame ();
153 throw DecodeError (N_("could not allocate frame"));
158 FFmpegDecoder::setup_video ()
160 _video_codec_context = _format_context->streams[_video_stream]->codec;
161 _video_codec = avcodec_find_decoder (_video_codec_context->codec_id);
163 if (_video_codec == 0) {
164 throw DecodeError (_("could not find video decoder"));
167 if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) {
168 throw DecodeError (N_("could not open video decoder"));
173 FFmpegDecoder::setup_audio ()
175 if (!_audio_stream) {
179 shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
182 _audio_codec_context = _format_context->streams[ffa->id()]->codec;
183 _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id);
185 if (_audio_codec == 0) {
186 throw DecodeError (_("could not find audio decoder"));
189 if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
190 throw DecodeError (N_("could not open audio decoder"));
195 FFmpegDecoder::setup_subtitle ()
197 if (!_subtitle_stream || _subtitle_stream->id() >= int (_format_context->nb_streams)) {
201 _subtitle_codec_context = _format_context->streams[_subtitle_stream->id()]->codec;
202 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
204 if (_subtitle_codec == 0) {
205 throw DecodeError (_("could not find subtitle decoder"));
208 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
209 throw DecodeError (N_("could not open subtitle decoder"));
215 FFmpegDecoder::pass ()
217 int r = av_read_frame (_format_context, &_packet);
220 if (r != AVERROR_EOF) {
221 /* Maybe we should fail here, but for now we'll just finish off instead */
223 av_strerror (r, buf, sizeof(buf));
224 _film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
227 /* Get any remaining frames */
232 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
236 if (_opt.decode_video) {
237 while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
238 filter_and_emit_video ();
242 if (_audio_stream && _opt.decode_audio) {
243 decode_audio_packet ();
249 avcodec_get_frame_defaults (_frame);
251 shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
253 if (_packet.stream_index == _video_stream && _opt.decode_video) {
256 int const r = avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet);
257 if (r >= 0 && frame_finished) {
259 if (r != _packet.size) {
260 _film->log()->log (String::compose (N_("Used only %1 bytes of %2 in packet"), r, _packet.size));
263 filter_and_emit_video ();
266 } else if (ffa && _packet.stream_index == ffa->id() && _opt.decode_audio) {
267 decode_audio_packet ();
268 } else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt.decode_subtitles) {
272 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) {
273 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
274 indicate that the previous subtitle should stop.
276 if (sub.num_rects > 0) {
277 shared_ptr<TimedSubtitle> ts;
279 emit_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
281 /* some problem with the subtitle; we probably didn't understand it */
284 emit_subtitle (shared_ptr<TimedSubtitle> ());
286 avsubtitle_free (&sub);
290 av_free_packet (&_packet);
294 /** @param data pointer to array of pointers to buffers.
295 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
297 shared_ptr<AudioBuffers>
298 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
300 assert (_film->audio_channels());
301 assert (bytes_per_audio_sample());
303 shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
306 /* Deinterleave and convert to float */
308 assert ((size % (bytes_per_audio_sample() * ffa->channels())) == 0);
310 int const total_samples = size / bytes_per_audio_sample();
311 int const frames = total_samples / _film->audio_channels();
312 shared_ptr<AudioBuffers> audio (new AudioBuffers (ffa->channels(), frames));
314 switch (audio_sample_format()) {
315 case AV_SAMPLE_FMT_S16:
317 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
320 for (int i = 0; i < total_samples; ++i) {
321 audio->data(channel)[sample] = float(*p++) / (1 << 15);
324 if (channel == _film->audio_channels()) {
332 case AV_SAMPLE_FMT_S16P:
334 int16_t** p = reinterpret_cast<int16_t **> (data);
335 for (int i = 0; i < _film->audio_channels(); ++i) {
336 for (int j = 0; j < frames; ++j) {
337 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
343 case AV_SAMPLE_FMT_S32:
345 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
348 for (int i = 0; i < total_samples; ++i) {
349 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
352 if (channel == _film->audio_channels()) {
360 case AV_SAMPLE_FMT_FLT:
362 float* p = reinterpret_cast<float*> (data[0]);
365 for (int i = 0; i < total_samples; ++i) {
366 audio->data(channel)[sample] = *p++;
369 if (channel == _film->audio_channels()) {
377 case AV_SAMPLE_FMT_FLTP:
379 float** p = reinterpret_cast<float**> (data);
380 for (int i = 0; i < _film->audio_channels(); ++i) {
381 memcpy (audio->data(i), p[i], frames * sizeof(float));
387 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
394 FFmpegDecoder::frames_per_second () const
396 AVStream* s = _format_context->streams[_video_stream];
398 if (s->avg_frame_rate.num && s->avg_frame_rate.den) {
399 return av_q2d (s->avg_frame_rate);
402 return av_q2d (s->r_frame_rate);
406 FFmpegDecoder::audio_sample_format () const
408 if (_audio_codec_context == 0) {
409 return (AVSampleFormat) 0;
412 return _audio_codec_context->sample_fmt;
416 FFmpegDecoder::native_size () const
418 return libdcp::Size (_video_codec_context->width, _video_codec_context->height);
422 FFmpegDecoder::pixel_format () const
424 return _video_codec_context->pix_fmt;
428 FFmpegDecoder::time_base_numerator () const
430 return _video_codec_context->time_base.num;
434 FFmpegDecoder::time_base_denominator () const
436 return _video_codec_context->time_base.den;
440 FFmpegDecoder::sample_aspect_ratio_numerator () const
442 return _video_codec_context->sample_aspect_ratio.num;
446 FFmpegDecoder::sample_aspect_ratio_denominator () const
448 return _video_codec_context->sample_aspect_ratio.den;
452 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()) {
471 if (n.str().empty()) {
479 FFmpegDecoder::bytes_per_audio_sample () const
481 return av_get_bytes_per_sample (audio_sample_format ());
485 FFmpegDecoder::set_audio_stream (shared_ptr<AudioStream> s)
487 AudioDecoder::set_audio_stream (s);
492 FFmpegDecoder::set_subtitle_stream (shared_ptr<SubtitleStream> s)
494 VideoDecoder::set_subtitle_stream (s);
500 FFmpegDecoder::filter_and_emit_video ()
502 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
503 if (bet == AV_NOPTS_VALUE) {
504 _film->log()->log ("Dropping frame without PTS");
508 shared_ptr<FilterGraph> graph;
511 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
513 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
514 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
518 if (i == _filter_graphs.end ()) {
519 graph = filter_graph_factory (_film, this, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format);
520 _filter_graphs.push_back (graph);
521 _film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
527 list<shared_ptr<Image> > images = graph->process (_frame);
529 for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
530 emit_video (*i, false, bet * av_q2d (_format_context->streams[_video_stream]->time_base));
535 FFmpegDecoder::seek (double p)
537 return do_seek (p, false, false);
541 FFmpegDecoder::seek_to_last ()
543 /* 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
544 (used when we change decoder parameters and want to re-fetch the frame) we end up going forwards rather than
545 staying in the same place.
547 return do_seek (last_source_time(), true, false);
551 FFmpegDecoder::seek_back ()
553 do_seek (last_source_time() - 2.5 / frames_per_second (), true, true);
557 FFmpegDecoder::seek_forward ()
559 do_seek (last_source_time() - 0.5 / frames_per_second(), true, true);
563 FFmpegDecoder::do_seek (double p, bool backwards, bool accurate)
565 int64_t const vt = p / av_q2d (_format_context->streams[_video_stream]->time_base);
567 int const r = av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0);
569 avcodec_flush_buffers (_video_codec_context);
570 if (_subtitle_codec_context) {
571 avcodec_flush_buffers (_subtitle_codec_context);
576 int r = av_read_frame (_format_context, &_packet);
581 avcodec_get_frame_defaults (_frame);
583 if (_packet.stream_index == _video_stream) {
585 int const r = avcodec_decode_video2 (_video_codec_context, _frame, &finished, &_packet);
586 if (r >= 0 && finished) {
587 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
594 av_free_packet (&_packet);
601 shared_ptr<FFmpegAudioStream>
602 FFmpegAudioStream::create (string t, optional<int> v)
605 /* version < 1; no type in the string, and there's only FFmpeg streams anyway */
606 return shared_ptr<FFmpegAudioStream> (new FFmpegAudioStream (t, v));
612 if (type != N_("ffmpeg")) {
613 return shared_ptr<FFmpegAudioStream> ();
616 return shared_ptr<FFmpegAudioStream> (new FFmpegAudioStream (t, v));
619 FFmpegAudioStream::FFmpegAudioStream (string t, optional<int> version)
627 n >> _id >> channels;
628 _channel_layout = av_get_default_channel_layout (channels);
632 /* Current (marked version 1) */
633 n >> type >> _id >> _sample_rate >> _channel_layout;
634 assert (type == N_("ffmpeg"));
637 for (int i = 0; i < name_index; ++i) {
638 size_t const s = t.find (' ');
639 if (s != string::npos) {
640 t = t.substr (s + 1);
648 FFmpegAudioStream::to_string () const
650 return String::compose (N_("ffmpeg %1 %2 %3 %4"), _id, _sample_rate, _channel_layout, _name);
654 FFmpegDecoder::film_changed (Film::Property p)
660 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
661 _filter_graphs.clear ();
671 /** @return Length (in video frames) according to our content's header */
673 FFmpegDecoder::length () const
675 return (double(_format_context->duration) / AV_TIME_BASE) * frames_per_second();
679 FFmpegDecoder::decode_audio_packet ()
681 shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
684 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
688 AVPacket copy_packet = _packet;
690 while (copy_packet.size > 0) {
693 int const decode_result = avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, ©_packet);
694 if (decode_result >= 0) {
695 if (frame_finished) {
697 /* Where we are in the source, in seconds */
698 double const source_pts_seconds = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
699 * av_frame_get_best_effort_timestamp(_frame);
701 int const data_size = av_samples_get_buffer_size (
702 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
705 assert (_audio_codec_context->channels == _film->audio_channels());
706 Audio (deinterleave_audio (_frame->data, data_size), source_pts_seconds);
709 copy_packet.data += decode_result;
710 copy_packet.size -= decode_result;