1 /* -*- c-basic-offset: 8; default-tab-width: 8; -*- */
4 Copyright (C) 2012 Carl Hetherington <cth@carlh.net>
6 This program is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 This program is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with this program; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 /** @file src/ffmpeg_decoder.cc
23 * @brief A decoder using FFmpeg to decode content.
32 #include <boost/lexical_cast.hpp>
34 #include <libavcodec/avcodec.h>
35 #include <libavformat/avformat.h>
36 #include <libswscale/swscale.h>
37 #include <libpostproc/postprocess.h>
42 #include "exceptions.h"
46 #include "ffmpeg_decoder.h"
47 #include "filter_graph.h"
49 #include "audio_buffers.h"
56 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<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio, bool subtitles)
74 , _video_codec_context (0)
76 , _audio_codec_context (0)
78 , _subtitle_codec_context (0)
80 , _decode_video (video)
81 , _decode_audio (audio)
82 , _decode_subtitles (subtitles)
90 FFmpegDecoder::~FFmpegDecoder ()
92 boost::mutex::scoped_lock lm (_mutex);
94 if (_audio_codec_context) {
95 avcodec_close (_audio_codec_context);
98 if (_video_codec_context) {
99 avcodec_close (_video_codec_context);
102 if (_subtitle_codec_context) {
103 avcodec_close (_subtitle_codec_context);
108 avformat_close_input (&_format_context);
112 FFmpegDecoder::setup_general ()
116 if (avformat_open_input (&_format_context, _ffmpeg_content->file().string().c_str(), 0, 0) < 0) {
117 throw OpenFileError (_ffmpeg_content->file().string ());
120 if (avformat_find_stream_info (_format_context, 0) < 0) {
121 throw DecodeError (_("could not find stream information"));
124 /* Find video, audio and subtitle streams */
126 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
127 AVStream* s = _format_context->streams[i];
128 if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
130 } else if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
132 /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
133 so bodge it here. No idea why we should have to do this.
136 if (s->codec->channel_layout == 0) {
137 s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
140 _audio_streams.push_back (
141 shared_ptr<FFmpegAudioStream> (
142 new FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channels)
146 } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
147 _subtitle_streams.push_back (shared_ptr<FFmpegSubtitleStream> (new FFmpegSubtitleStream (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 (!_ffmpeg_content->audio_stream ()) {
187 _audio_codec_context = _format_context->streams[_ffmpeg_content->audio_stream()->id]->codec;
188 _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id);
190 if (_audio_codec == 0) {
191 throw DecodeError (_("could not find audio decoder"));
194 if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
195 throw DecodeError (N_("could not open audio decoder"));
200 FFmpegDecoder::setup_subtitle ()
202 boost::mutex::scoped_lock lm (_mutex);
204 if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
208 _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
209 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
211 if (_subtitle_codec == 0) {
212 throw DecodeError (_("could not find subtitle decoder"));
215 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
216 throw DecodeError (N_("could not open subtitle decoder"));
222 FFmpegDecoder::pass ()
224 int r = av_read_frame (_format_context, &_packet);
227 if (r != AVERROR_EOF) {
228 /* Maybe we should fail here, but for now we'll just finish off instead */
230 av_strerror (r, buf, sizeof(buf));
231 shared_ptr<const Film> film = _film.lock ();
233 film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
236 /* Get any remaining frames */
241 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
244 while (decode_video_packet ());
247 if (_ffmpeg_content->audio_stream() && _decode_audio) {
248 decode_audio_packet ();
251 /* Stop us being asked for any more data */
252 _next_video = _next_audio = _ffmpeg_content->length ();
256 avcodec_get_frame_defaults (_frame);
258 if (_packet.stream_index == _video_stream && _decode_video) {
259 decode_video_packet ();
260 } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) {
261 decode_audio_packet ();
262 } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id && _decode_subtitles) {
266 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) {
267 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
268 indicate that the previous subtitle should stop.
270 if (sub.num_rects > 0) {
271 shared_ptr<TimedSubtitle> ts;
273 subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
275 /* some problem with the subtitle; we probably didn't understand it */
278 subtitle (shared_ptr<TimedSubtitle> ());
280 avsubtitle_free (&sub);
283 cout << "[ffmpeg] other packet.\n";
286 av_free_packet (&_packet);
289 /** @param data pointer to array of pointers to buffers.
290 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
292 shared_ptr<AudioBuffers>
293 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
295 assert (_ffmpeg_content->audio_channels());
296 assert (bytes_per_audio_sample());
298 /* Deinterleave and convert to float */
300 assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0);
302 int const total_samples = size / bytes_per_audio_sample();
303 int const frames = total_samples / _ffmpeg_content->audio_channels();
304 shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
306 switch (audio_sample_format()) {
307 case AV_SAMPLE_FMT_S16:
309 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
312 for (int i = 0; i < total_samples; ++i) {
313 audio->data(channel)[sample] = float(*p++) / (1 << 15);
316 if (channel == _ffmpeg_content->audio_channels()) {
324 case AV_SAMPLE_FMT_S16P:
326 int16_t** p = reinterpret_cast<int16_t **> (data);
327 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
328 for (int j = 0; j < frames; ++j) {
329 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
335 case AV_SAMPLE_FMT_S32:
337 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
340 for (int i = 0; i < total_samples; ++i) {
341 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
344 if (channel == _ffmpeg_content->audio_channels()) {
352 case AV_SAMPLE_FMT_FLT:
354 float* p = reinterpret_cast<float*> (data[0]);
357 for (int i = 0; i < total_samples; ++i) {
358 audio->data(channel)[sample] = *p++;
361 if (channel == _ffmpeg_content->audio_channels()) {
369 case AV_SAMPLE_FMT_FLTP:
371 float** p = reinterpret_cast<float**> (data);
372 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
373 memcpy (audio->data(i), p[i], frames * sizeof(float));
379 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
386 FFmpegDecoder::video_frame_rate () const
388 AVStream* s = _format_context->streams[_video_stream];
390 if (s->avg_frame_rate.num && s->avg_frame_rate.den) {
391 return av_q2d (s->avg_frame_rate);
394 return av_q2d (s->r_frame_rate);
398 FFmpegDecoder::audio_sample_format () const
400 if (_audio_codec_context == 0) {
401 return (AVSampleFormat) 0;
404 return _audio_codec_context->sample_fmt;
408 FFmpegDecoder::video_size () const
410 return libdcp::Size (_video_codec_context->width, _video_codec_context->height);
414 FFmpegDecoder::stream_name (AVStream* s) const
419 AVDictionaryEntry const * lang = av_dict_get (s->metadata, N_("language"), 0, 0);
424 AVDictionaryEntry const * title = av_dict_get (s->metadata, N_("title"), 0, 0);
426 if (!n.str().empty()) {
433 if (n.str().empty()) {
441 FFmpegDecoder::bytes_per_audio_sample () const
443 return av_get_bytes_per_sample (audio_sample_format ());
447 FFmpegDecoder::seek (Time t)
449 do_seek (t, false, false);
450 VideoDecoder::seek (t);
454 FFmpegDecoder::seek_back ()
456 if (next() < (2.5 * TIME_HZ / video_frame_rate())) {
460 do_seek (next() - 2.5 * TIME_HZ / video_frame_rate(), true, true);
461 VideoDecoder::seek_back ();
465 FFmpegDecoder::seek_forward ()
467 if (next() >= (_ffmpeg_content->length() - 0.5 * TIME_HZ / video_frame_rate())) {
471 do_seek (next() - 0.5 * TIME_HZ / video_frame_rate(), true, true);
472 VideoDecoder::seek_forward ();
476 FFmpegDecoder::do_seek (Time t, bool backwards, bool accurate)
478 int64_t const vt = t / (av_q2d (_format_context->streams[_video_stream]->time_base) * TIME_HZ);
479 av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0);
481 avcodec_flush_buffers (_video_codec_context);
482 if (_subtitle_codec_context) {
483 avcodec_flush_buffers (_subtitle_codec_context);
488 int r = av_read_frame (_format_context, &_packet);
493 avcodec_get_frame_defaults (_frame);
495 if (_packet.stream_index == _video_stream) {
497 int const r = avcodec_decode_video2 (_video_codec_context, _frame, &finished, &_packet);
498 if (r >= 0 && finished) {
499 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
506 av_free_packet (&_packet);
513 /** @return Length (in video frames) according to our content's header */
515 FFmpegDecoder::video_length () const
517 return (double(_format_context->duration) / AV_TIME_BASE) * video_frame_rate();
521 FFmpegDecoder::decode_audio_packet ()
523 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
527 AVPacket copy_packet = _packet;
529 while (copy_packet.size > 0) {
532 int const decode_result = avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, ©_packet);
533 if (decode_result >= 0) {
534 if (frame_finished) {
536 /* Where we are in the source, in seconds */
537 double const source_pts_seconds = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
538 * av_frame_get_best_effort_timestamp(_frame);
540 int const data_size = av_samples_get_buffer_size (
541 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
544 assert (_audio_codec_context->channels == _ffmpeg_content->audio_channels());
545 audio (deinterleave_audio (_frame->data, data_size), source_pts_seconds);
548 copy_packet.data += decode_result;
549 copy_packet.size -= decode_result;
555 FFmpegDecoder::decode_video_packet ()
558 if (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
562 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
564 shared_ptr<FilterGraph> graph;
566 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
567 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
571 if (i == _filter_graphs.end ()) {
572 shared_ptr<const Film> film = _film.lock ();
575 graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
576 _filter_graphs.push_back (graph);
578 film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
583 list<shared_ptr<Image> > images = graph->process (_frame);
585 string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second;
587 for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
589 shared_ptr<Image> image = *i;
590 if (!post_process.empty ()) {
591 image = image->post_process (post_process, true);
594 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
595 if (bet != AV_NOPTS_VALUE) {
596 /* XXX: may need to insert extra frames / remove frames here ...
599 Time const t = bet * av_q2d (_format_context->streams[_video_stream]->time_base) * TIME_HZ;
600 video (image, false, t);
602 shared_ptr<const Film> film = _film.lock ();
604 film->log()->log ("Dropping frame without PTS");
612 FFmpegDecoder::next () const
614 if (_decode_video && _decode_audio && _audio_codec_context) {
615 return min (_next_video, _next_audio);
618 if (_decode_audio && _audio_codec_context) {
626 FFmpegDecoder::done () const
628 return (!_decode_audio || !_audio_codec_context || audio_done()) && (!_decode_video || video_done());