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"
44 #include "exceptions.h"
48 #include "ffmpeg_decoder.h"
49 #include "filter_graph.h"
51 #include "audio_buffers.h"
58 using std::stringstream;
60 using boost::shared_ptr;
61 using boost::optional;
62 using boost::dynamic_pointer_cast;
65 boost::mutex FFmpegDecoder::_mutex;
67 FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio, bool subtitles)
75 , _video_codec_context (0)
77 , _audio_codec_context (0)
79 , _subtitle_codec_context (0)
81 , _decode_video (video)
82 , _decode_audio (audio)
83 , _decode_subtitles (subtitles)
91 FFmpegDecoder::~FFmpegDecoder ()
93 boost::mutex::scoped_lock lm (_mutex);
95 if (_audio_codec_context) {
96 avcodec_close (_audio_codec_context);
99 if (_video_codec_context) {
100 avcodec_close (_video_codec_context);
103 if (_subtitle_codec_context) {
104 avcodec_close (_subtitle_codec_context);
109 avformat_close_input (&_format_context);
113 FFmpegDecoder::setup_general ()
117 if (avformat_open_input (&_format_context, _ffmpeg_content->file().string().c_str(), 0, 0) < 0) {
118 throw OpenFileError (_ffmpeg_content->file().string ());
121 if (avformat_find_stream_info (_format_context, 0) < 0) {
122 throw DecodeError (_("could not find stream information"));
125 /* Find video, audio and subtitle streams */
127 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
128 AVStream* s = _format_context->streams[i];
129 if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
131 } else if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
133 /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
134 so bodge it here. No idea why we should have to do this.
137 if (s->codec->channel_layout == 0) {
138 s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
141 _audio_streams.push_back (
142 FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channels)
145 } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
146 _subtitle_streams.push_back (FFmpegSubtitleStream (stream_name (s), i));
150 if (_video_stream < 0) {
151 throw DecodeError (N_("could not find video stream"));
154 _frame = avcodec_alloc_frame ();
156 throw DecodeError (N_("could not allocate frame"));
161 FFmpegDecoder::setup_video ()
163 boost::mutex::scoped_lock lm (_mutex);
165 _video_codec_context = _format_context->streams[_video_stream]->codec;
166 _video_codec = avcodec_find_decoder (_video_codec_context->codec_id);
168 if (_video_codec == 0) {
169 throw DecodeError (_("could not find video decoder"));
172 if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) {
173 throw DecodeError (N_("could not open video decoder"));
178 FFmpegDecoder::setup_audio ()
180 boost::mutex::scoped_lock lm (_mutex);
182 if (!_ffmpeg_content->audio_stream ()) {
186 _audio_codec_context = _format_context->streams[_ffmpeg_content->audio_stream()->id]->codec;
187 _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id);
189 if (_audio_codec == 0) {
190 throw DecodeError (_("could not find audio decoder"));
193 if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
194 throw DecodeError (N_("could not open audio decoder"));
199 FFmpegDecoder::setup_subtitle ()
201 boost::mutex::scoped_lock lm (_mutex);
203 if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
207 _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
208 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
210 if (_subtitle_codec == 0) {
211 throw DecodeError (_("could not find subtitle decoder"));
214 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
215 throw DecodeError (N_("could not open subtitle decoder"));
221 FFmpegDecoder::pass ()
223 cout << "FFmpeg::pass\n";
225 int r = av_read_frame (_format_context, &_packet);
228 if (r != AVERROR_EOF) {
229 /* Maybe we should fail here, but for now we'll just finish off instead */
231 av_strerror (r, buf, sizeof(buf));
232 _film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
235 /* Get any remaining frames */
240 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
243 decode_video_packet ();
246 if (_ffmpeg_content->audio_stream() && _decode_audio) {
247 decode_audio_packet ();
253 avcodec_get_frame_defaults (_frame);
255 if (_packet.stream_index == _video_stream && _decode_video) {
256 decode_video_packet ();
257 } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) {
258 decode_audio_packet ();
259 } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id && _decode_subtitles) {
263 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) {
264 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
265 indicate that the previous subtitle should stop.
267 if (sub.num_rects > 0) {
268 shared_ptr<TimedSubtitle> ts;
270 emit_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
272 /* some problem with the subtitle; we probably didn't understand it */
275 emit_subtitle (shared_ptr<TimedSubtitle> ());
277 avsubtitle_free (&sub);
281 av_free_packet (&_packet);
285 /** @param data pointer to array of pointers to buffers.
286 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
288 shared_ptr<AudioBuffers>
289 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
291 assert (_ffmpeg_content->audio_channels());
292 assert (bytes_per_audio_sample());
294 /* Deinterleave and convert to float */
296 assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0);
298 int const total_samples = size / bytes_per_audio_sample();
299 int const frames = total_samples / _ffmpeg_content->audio_channels();
300 shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
302 switch (audio_sample_format()) {
303 case AV_SAMPLE_FMT_S16:
305 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
308 for (int i = 0; i < total_samples; ++i) {
309 audio->data(channel)[sample] = float(*p++) / (1 << 15);
312 if (channel == _ffmpeg_content->audio_channels()) {
320 case AV_SAMPLE_FMT_S16P:
322 int16_t** p = reinterpret_cast<int16_t **> (data);
323 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
324 for (int j = 0; j < frames; ++j) {
325 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
331 case AV_SAMPLE_FMT_S32:
333 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
336 for (int i = 0; i < total_samples; ++i) {
337 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
340 if (channel == _ffmpeg_content->audio_channels()) {
348 case AV_SAMPLE_FMT_FLT:
350 float* p = reinterpret_cast<float*> (data[0]);
353 for (int i = 0; i < total_samples; ++i) {
354 audio->data(channel)[sample] = *p++;
357 if (channel == _ffmpeg_content->audio_channels()) {
365 case AV_SAMPLE_FMT_FLTP:
367 float** p = reinterpret_cast<float**> (data);
368 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
369 memcpy (audio->data(i), p[i], frames * sizeof(float));
375 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
382 FFmpegDecoder::video_frame_rate () const
384 AVStream* s = _format_context->streams[_video_stream];
386 if (s->avg_frame_rate.num && s->avg_frame_rate.den) {
387 return av_q2d (s->avg_frame_rate);
390 return av_q2d (s->r_frame_rate);
394 FFmpegDecoder::audio_sample_format () const
396 if (_audio_codec_context == 0) {
397 return (AVSampleFormat) 0;
400 return _audio_codec_context->sample_fmt;
404 FFmpegDecoder::native_size () const
406 return libdcp::Size (_video_codec_context->width, _video_codec_context->height);
410 FFmpegDecoder::pixel_format () const
412 return _video_codec_context->pix_fmt;
416 FFmpegDecoder::time_base_numerator () const
418 return _video_codec_context->time_base.num;
422 FFmpegDecoder::time_base_denominator () const
424 return _video_codec_context->time_base.den;
428 FFmpegDecoder::sample_aspect_ratio_numerator () const
430 return _video_codec_context->sample_aspect_ratio.num;
434 FFmpegDecoder::sample_aspect_ratio_denominator () const
436 return _video_codec_context->sample_aspect_ratio.den;
440 FFmpegDecoder::stream_name (AVStream* s) const
445 AVDictionaryEntry const * lang = av_dict_get (s->metadata, N_("language"), 0, 0);
450 AVDictionaryEntry const * title = av_dict_get (s->metadata, N_("title"), 0, 0);
452 if (!n.str().empty()) {
459 if (n.str().empty()) {
467 FFmpegDecoder::bytes_per_audio_sample () const
469 return av_get_bytes_per_sample (audio_sample_format ());
473 FFmpegDecoder::seek (double p)
475 return do_seek (p, false, false);
479 FFmpegDecoder::seek_back ()
481 if (last_content_time() < 2.5) {
485 return do_seek (last_content_time() - 2.5 / video_frame_rate(), true, true);
489 FFmpegDecoder::seek_forward ()
491 if (last_content_time() >= (video_length() - video_frame_rate())) {
495 return do_seek (last_content_time() - 0.5 / video_frame_rate(), true, true);
499 FFmpegDecoder::do_seek (double p, bool backwards, bool accurate)
501 int64_t const vt = p / av_q2d (_format_context->streams[_video_stream]->time_base);
503 int const r = av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0);
505 avcodec_flush_buffers (_video_codec_context);
506 if (_subtitle_codec_context) {
507 avcodec_flush_buffers (_subtitle_codec_context);
512 int r = av_read_frame (_format_context, &_packet);
517 avcodec_get_frame_defaults (_frame);
519 if (_packet.stream_index == _video_stream) {
521 int const r = avcodec_decode_video2 (_video_codec_context, _frame, &finished, &_packet);
522 if (r >= 0 && finished) {
523 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
530 av_free_packet (&_packet);
538 FFmpegDecoder::film_changed (Film::Property p)
544 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
545 _filter_graphs.clear ();
554 /** @return Length (in video frames) according to our content's header */
556 FFmpegDecoder::video_length () const
558 return (double(_format_context->duration) / AV_TIME_BASE) * video_frame_rate();
562 FFmpegDecoder::decode_audio_packet ()
564 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
568 AVPacket copy_packet = _packet;
570 while (copy_packet.size > 0) {
573 int const decode_result = avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, ©_packet);
574 if (decode_result >= 0) {
575 if (frame_finished) {
577 /* Where we are in the source, in seconds */
578 double const source_pts_seconds = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
579 * av_frame_get_best_effort_timestamp(_frame);
581 int const data_size = av_samples_get_buffer_size (
582 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
585 assert (_audio_codec_context->channels == _ffmpeg_content->audio_channels());
586 Audio (deinterleave_audio (_frame->data, data_size), source_pts_seconds);
589 copy_packet.data += decode_result;
590 copy_packet.size -= decode_result;
596 FFmpegDecoder::decode_video_packet ()
599 while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
600 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
602 shared_ptr<FilterGraph> graph;
604 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
605 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
609 if (i == _filter_graphs.end ()) {
610 graph.reset (new FilterGraph (_film, this, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
611 _filter_graphs.push_back (graph);
612 _film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
617 list<shared_ptr<Image> > images = graph->process (_frame);
619 for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
620 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
621 if (bet != AV_NOPTS_VALUE) {
622 /* XXX: may need to insert extra frames / remove frames here ...
625 emit_video (*i, false, bet * av_q2d (_format_context->streams[_video_stream]->time_base));
627 _film->log()->log ("Dropping frame without PTS");