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 "exceptions.h"
44 #include "ffmpeg_decoder.h"
45 #include "filter_graph.h"
47 #include "audio_buffers.h"
54 using std::stringstream;
57 using boost::shared_ptr;
58 using boost::optional;
59 using boost::dynamic_pointer_cast;
62 boost::mutex FFmpegDecoder::_mutex;
64 FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio)
72 , _video_codec_context (0)
74 , _audio_codec_context (0)
76 , _subtitle_codec_context (0)
78 , _decode_video (video)
79 , _decode_audio (audio)
87 FFmpegDecoder::~FFmpegDecoder ()
89 boost::mutex::scoped_lock lm (_mutex);
91 if (_audio_codec_context) {
92 avcodec_close (_audio_codec_context);
95 if (_video_codec_context) {
96 avcodec_close (_video_codec_context);
99 if (_subtitle_codec_context) {
100 avcodec_close (_subtitle_codec_context);
105 avformat_close_input (&_format_context);
109 FFmpegDecoder::setup_general ()
113 if (avformat_open_input (&_format_context, _ffmpeg_content->file().string().c_str(), 0, 0) < 0) {
114 throw OpenFileError (_ffmpeg_content->file().string ());
117 if (avformat_find_stream_info (_format_context, 0) < 0) {
118 throw DecodeError (_("could not find stream information"));
121 /* Find video, audio and subtitle streams */
123 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
124 AVStream* s = _format_context->streams[i];
125 if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
127 } else if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
129 /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
130 so bodge it here. No idea why we should have to do this.
133 if (s->codec->channel_layout == 0) {
134 s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
137 _audio_streams.push_back (
138 shared_ptr<FFmpegAudioStream> (
139 new FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channels)
143 } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
144 _subtitle_streams.push_back (shared_ptr<FFmpegSubtitleStream> (new FFmpegSubtitleStream (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 boost::mutex::scoped_lock lm (_mutex);
163 _video_codec_context = _format_context->streams[_video_stream]->codec;
164 _video_codec = avcodec_find_decoder (_video_codec_context->codec_id);
166 if (_video_codec == 0) {
167 throw DecodeError (_("could not find video decoder"));
170 if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) {
171 throw DecodeError (N_("could not open video decoder"));
176 FFmpegDecoder::setup_audio ()
178 boost::mutex::scoped_lock lm (_mutex);
180 if (!_ffmpeg_content->audio_stream ()) {
184 _audio_codec_context = _format_context->streams[_ffmpeg_content->audio_stream()->id]->codec;
185 _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id);
187 if (_audio_codec == 0) {
188 throw DecodeError (_("could not find audio decoder"));
191 if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
192 throw DecodeError (N_("could not open audio decoder"));
197 FFmpegDecoder::setup_subtitle ()
199 boost::mutex::scoped_lock lm (_mutex);
201 if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
205 _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
206 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
208 if (_subtitle_codec == 0) {
209 throw DecodeError (_("could not find subtitle decoder"));
212 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
213 throw DecodeError (N_("could not open subtitle decoder"));
219 FFmpegDecoder::pass ()
221 int r = av_read_frame (_format_context, &_packet);
224 if (r != AVERROR_EOF) {
225 /* Maybe we should fail here, but for now we'll just finish off instead */
227 av_strerror (r, buf, sizeof(buf));
228 shared_ptr<const Film> film = _film.lock ();
230 film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
233 /* Get any remaining frames */
238 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
241 while (decode_video_packet ());
244 if (_ffmpeg_content->audio_stream() && _decode_audio) {
245 decode_audio_packet ();
248 /* Stop us being asked for any more data */
249 _next_video = _next_audio = _ffmpeg_content->length ();
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) {
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 subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
272 /* some problem with the subtitle; we probably didn't understand it */
275 subtitle (shared_ptr<TimedSubtitle> ());
277 avsubtitle_free (&sub);
281 av_free_packet (&_packet);
284 /** @param data pointer to array of pointers to buffers.
285 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
287 shared_ptr<AudioBuffers>
288 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
290 assert (_ffmpeg_content->audio_channels());
291 assert (bytes_per_audio_sample());
293 /* Deinterleave and convert to float */
295 assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0);
297 int const total_samples = size / bytes_per_audio_sample();
298 int const frames = total_samples / _ffmpeg_content->audio_channels();
299 shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
301 switch (audio_sample_format()) {
302 case AV_SAMPLE_FMT_S16:
304 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
307 for (int i = 0; i < total_samples; ++i) {
308 audio->data(channel)[sample] = float(*p++) / (1 << 15);
311 if (channel == _ffmpeg_content->audio_channels()) {
319 case AV_SAMPLE_FMT_S16P:
321 int16_t** p = reinterpret_cast<int16_t **> (data);
322 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
323 for (int j = 0; j < frames; ++j) {
324 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
330 case AV_SAMPLE_FMT_S32:
332 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
335 for (int i = 0; i < total_samples; ++i) {
336 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
339 if (channel == _ffmpeg_content->audio_channels()) {
347 case AV_SAMPLE_FMT_FLT:
349 float* p = reinterpret_cast<float*> (data[0]);
352 for (int i = 0; i < total_samples; ++i) {
353 audio->data(channel)[sample] = *p++;
356 if (channel == _ffmpeg_content->audio_channels()) {
364 case AV_SAMPLE_FMT_FLTP:
366 float** p = reinterpret_cast<float**> (data);
367 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
368 memcpy (audio->data(i), p[i], frames * sizeof(float));
374 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
381 FFmpegDecoder::video_frame_rate () const
383 AVStream* s = _format_context->streams[_video_stream];
385 if (s->avg_frame_rate.num && s->avg_frame_rate.den) {
386 return av_q2d (s->avg_frame_rate);
389 return av_q2d (s->r_frame_rate);
393 FFmpegDecoder::audio_sample_format () const
395 if (_audio_codec_context == 0) {
396 return (AVSampleFormat) 0;
399 return _audio_codec_context->sample_fmt;
403 FFmpegDecoder::video_size () const
405 return libdcp::Size (_video_codec_context->width, _video_codec_context->height);
409 FFmpegDecoder::stream_name (AVStream* s) const
414 AVDictionaryEntry const * lang = av_dict_get (s->metadata, N_("language"), 0, 0);
419 AVDictionaryEntry const * title = av_dict_get (s->metadata, N_("title"), 0, 0);
421 if (!n.str().empty()) {
428 if (n.str().empty()) {
436 FFmpegDecoder::bytes_per_audio_sample () const
438 return av_get_bytes_per_sample (audio_sample_format ());
442 FFmpegDecoder::seek (Time t)
444 do_seek (t, false, false);
445 VideoDecoder::seek (t);
449 FFmpegDecoder::seek_back ()
451 if (position() < (2.5 * TIME_HZ / video_frame_rate())) {
455 do_seek (position() - 2.5 * TIME_HZ / video_frame_rate(), true, true);
456 VideoDecoder::seek_back ();
460 FFmpegDecoder::seek_forward ()
462 if (position() >= (_ffmpeg_content->length() - 0.5 * TIME_HZ / video_frame_rate())) {
466 do_seek (position() - 0.5 * TIME_HZ / video_frame_rate(), true, true);
467 VideoDecoder::seek_forward ();
471 FFmpegDecoder::do_seek (Time t, bool backwards, bool accurate)
473 int64_t const vt = t / (av_q2d (_format_context->streams[_video_stream]->time_base) * TIME_HZ);
474 av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0);
476 avcodec_flush_buffers (_video_codec_context);
477 if (_subtitle_codec_context) {
478 avcodec_flush_buffers (_subtitle_codec_context);
483 int r = av_read_frame (_format_context, &_packet);
488 avcodec_get_frame_defaults (_frame);
490 if (_packet.stream_index == _video_stream) {
492 int const r = avcodec_decode_video2 (_video_codec_context, _frame, &finished, &_packet);
493 if (r >= 0 && finished) {
494 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
501 av_free_packet (&_packet);
508 /** @return Length (in video frames) according to our content's header */
510 FFmpegDecoder::video_length () const
512 return (double(_format_context->duration) / AV_TIME_BASE) * video_frame_rate();
516 FFmpegDecoder::decode_audio_packet ()
518 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
522 AVPacket copy_packet = _packet;
524 while (copy_packet.size > 0) {
527 int const decode_result = avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, ©_packet);
528 if (decode_result >= 0) {
529 if (frame_finished) {
531 /* Where we are in the source, in seconds */
532 double const source_pts_seconds = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
533 * av_frame_get_best_effort_timestamp(_frame);
535 int const data_size = av_samples_get_buffer_size (
536 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
539 assert (_audio_codec_context->channels == _ffmpeg_content->audio_channels());
540 audio (deinterleave_audio (_frame->data, data_size), source_pts_seconds * TIME_HZ);
543 copy_packet.data += decode_result;
544 copy_packet.size -= decode_result;
550 FFmpegDecoder::decode_video_packet ()
553 if (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
557 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
559 shared_ptr<FilterGraph> graph;
561 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
562 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
566 if (i == _filter_graphs.end ()) {
567 shared_ptr<const Film> film = _film.lock ();
570 graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
571 _filter_graphs.push_back (graph);
573 film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
578 list<shared_ptr<Image> > images = graph->process (_frame);
580 string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second;
582 for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
584 shared_ptr<Image> image = *i;
585 if (!post_process.empty ()) {
586 image = image->post_process (post_process, true);
589 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
590 if (bet != AV_NOPTS_VALUE) {
591 /* XXX: may need to insert extra frames / remove frames here ...
594 Time const t = bet * av_q2d (_format_context->streams[_video_stream]->time_base) * TIME_HZ;
595 video (image, false, t);
597 shared_ptr<const Film> film = _film.lock ();
599 film->log()->log ("Dropping frame without PTS");
607 FFmpegDecoder::position () const
609 if (_decode_video && _decode_audio && _audio_codec_context) {
610 return min (_next_video, _next_audio);
613 if (_decode_audio && _audio_codec_context) {
621 FFmpegDecoder::done () const
623 return (!_decode_audio || !_audio_codec_context || audio_done()) && (!_decode_video || video_done());