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>
35 #include <libavcodec/avcodec.h>
36 #include <libavformat/avformat.h>
37 #include <libswscale/swscale.h>
38 #include <libpostproc/postprocess.h>
43 #include "transcoder.h"
46 #include "exceptions.h"
50 #include "ffmpeg_decoder.h"
51 #include "filter_graph.h"
53 #include "audio_buffers.h"
60 using std::stringstream;
63 using boost::shared_ptr;
64 using boost::optional;
65 using boost::dynamic_pointer_cast;
68 boost::mutex FFmpegDecoder::_mutex;
70 FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio, bool subtitles)
78 , _video_codec_context (0)
80 , _audio_codec_context (0)
82 , _subtitle_codec_context (0)
84 , _decode_video (video)
85 , _decode_audio (audio)
86 , _decode_subtitles (subtitles)
94 FFmpegDecoder::~FFmpegDecoder ()
96 boost::mutex::scoped_lock lm (_mutex);
98 if (_audio_codec_context) {
99 avcodec_close (_audio_codec_context);
102 if (_video_codec_context) {
103 avcodec_close (_video_codec_context);
106 if (_subtitle_codec_context) {
107 avcodec_close (_subtitle_codec_context);
112 avformat_close_input (&_format_context);
116 FFmpegDecoder::setup_general ()
120 if (avformat_open_input (&_format_context, _ffmpeg_content->file().string().c_str(), 0, 0) < 0) {
121 throw OpenFileError (_ffmpeg_content->file().string ());
124 if (avformat_find_stream_info (_format_context, 0) < 0) {
125 throw DecodeError (_("could not find stream information"));
128 /* Find video, audio and subtitle streams */
130 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
131 AVStream* s = _format_context->streams[i];
132 if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
134 } else if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
136 /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
137 so bodge it here. No idea why we should have to do this.
140 if (s->codec->channel_layout == 0) {
141 s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
144 _audio_streams.push_back (
145 shared_ptr<FFmpegAudioStream> (
146 new FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channels)
150 } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
151 _subtitle_streams.push_back (shared_ptr<FFmpegSubtitleStream> (new FFmpegSubtitleStream (stream_name (s), i)));
155 if (_video_stream < 0) {
156 throw DecodeError (N_("could not find video stream"));
159 _frame = avcodec_alloc_frame ();
161 throw DecodeError (N_("could not allocate frame"));
166 FFmpegDecoder::setup_video ()
168 boost::mutex::scoped_lock lm (_mutex);
170 _video_codec_context = _format_context->streams[_video_stream]->codec;
171 _video_codec = avcodec_find_decoder (_video_codec_context->codec_id);
173 if (_video_codec == 0) {
174 throw DecodeError (_("could not find video decoder"));
177 if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) {
178 throw DecodeError (N_("could not open video decoder"));
183 FFmpegDecoder::setup_audio ()
185 boost::mutex::scoped_lock lm (_mutex);
187 if (!_ffmpeg_content->audio_stream ()) {
191 _audio_codec_context = _format_context->streams[_ffmpeg_content->audio_stream()->id]->codec;
192 _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id);
194 if (_audio_codec == 0) {
195 throw DecodeError (_("could not find audio decoder"));
198 if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
199 throw DecodeError (N_("could not open audio decoder"));
204 FFmpegDecoder::setup_subtitle ()
206 boost::mutex::scoped_lock lm (_mutex);
208 if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
212 _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
213 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
215 if (_subtitle_codec == 0) {
216 throw DecodeError (_("could not find subtitle decoder"));
219 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
220 throw DecodeError (N_("could not open subtitle decoder"));
226 FFmpegDecoder::pass ()
228 int r = av_read_frame (_format_context, &_packet);
231 if (r != AVERROR_EOF) {
232 /* Maybe we should fail here, but for now we'll just finish off instead */
234 av_strerror (r, buf, sizeof(buf));
235 shared_ptr<const Film> film = _film.lock ();
237 film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
240 /* Get any remaining frames */
245 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
248 while (decode_video_packet ());
251 if (_ffmpeg_content->audio_stream() && _decode_audio) {
252 decode_audio_packet ();
258 avcodec_get_frame_defaults (_frame);
260 if (_packet.stream_index == _video_stream && _decode_video) {
261 decode_video_packet ();
262 } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) {
263 decode_audio_packet ();
264 } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id && _decode_subtitles) {
268 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) {
269 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
270 indicate that the previous subtitle should stop.
272 if (sub.num_rects > 0) {
273 shared_ptr<TimedSubtitle> ts;
275 subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
277 /* some problem with the subtitle; we probably didn't understand it */
280 subtitle (shared_ptr<TimedSubtitle> ());
282 avsubtitle_free (&sub);
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);
453 FFmpegDecoder::seek_back ()
455 if (next() < (2.5 * TIME_HZ / video_frame_rate())) {
459 do_seek (next() - 2.5 * TIME_HZ / video_frame_rate(), true, true);
463 FFmpegDecoder::seek_forward ()
465 if (next() >= (_ffmpeg_content->length() - 0.5 * TIME_HZ / video_frame_rate())) {
469 do_seek (next() - 0.5 * TIME_HZ / video_frame_rate(), true, true);
473 FFmpegDecoder::do_seek (Time t, bool backwards, bool accurate)
475 int64_t const vt = t / (av_q2d (_format_context->streams[_video_stream]->time_base) * TIME_HZ);
476 av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0);
478 avcodec_flush_buffers (_video_codec_context);
479 if (_subtitle_codec_context) {
480 avcodec_flush_buffers (_subtitle_codec_context);
485 int r = av_read_frame (_format_context, &_packet);
490 avcodec_get_frame_defaults (_frame);
492 if (_packet.stream_index == _video_stream) {
494 int const r = avcodec_decode_video2 (_video_codec_context, _frame, &finished, &_packet);
495 if (r >= 0 && finished) {
496 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
503 av_free_packet (&_packet);
510 /** @return Length (in video frames) according to our content's header */
512 FFmpegDecoder::video_length () const
514 return (double(_format_context->duration) / AV_TIME_BASE) * video_frame_rate();
518 FFmpegDecoder::decode_audio_packet ()
520 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
524 AVPacket copy_packet = _packet;
526 while (copy_packet.size > 0) {
529 int const decode_result = avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, ©_packet);
530 if (decode_result >= 0) {
531 if (frame_finished) {
533 /* Where we are in the source, in seconds */
534 double const source_pts_seconds = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
535 * av_frame_get_best_effort_timestamp(_frame);
537 int const data_size = av_samples_get_buffer_size (
538 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
541 assert (_audio_codec_context->channels == _ffmpeg_content->audio_channels());
542 Audio (deinterleave_audio (_frame->data, data_size), source_pts_seconds);
545 copy_packet.data += decode_result;
546 copy_packet.size -= decode_result;
552 FFmpegDecoder::decode_video_packet ()
555 if (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
559 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
561 shared_ptr<FilterGraph> graph;
563 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
564 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
568 if (i == _filter_graphs.end ()) {
569 shared_ptr<const Film> film = _film.lock ();
572 graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
573 _filter_graphs.push_back (graph);
575 film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
580 list<shared_ptr<Image> > images = graph->process (_frame);
582 string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second;
584 for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
586 shared_ptr<Image> image = *i;
587 if (!post_process.empty ()) {
588 image = image->post_process (post_process, true);
591 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
592 if (bet != AV_NOPTS_VALUE) {
593 /* XXX: may need to insert extra frames / remove frames here ...
596 Time const t = bet * av_q2d (_format_context->streams[_video_stream]->time_base) * TIME_HZ;
597 video (image, false, t);
599 shared_ptr<const Film> film = _film.lock ();
601 film->log()->log ("Dropping frame without PTS");
609 FFmpegDecoder::next () const
611 if (_decode_video && _decode_audio) {
612 return min (_next_video, _next_audio);