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"
57 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 FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channels)
144 } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
145 _subtitle_streams.push_back (FFmpegSubtitleStream (stream_name (s), i));
149 if (_video_stream < 0) {
150 throw DecodeError (N_("could not find video stream"));
153 _frame = avcodec_alloc_frame ();
155 throw DecodeError (N_("could not allocate frame"));
160 FFmpegDecoder::setup_video ()
162 boost::mutex::scoped_lock lm (_mutex);
164 _video_codec_context = _format_context->streams[_video_stream]->codec;
165 _video_codec = avcodec_find_decoder (_video_codec_context->codec_id);
167 if (_video_codec == 0) {
168 throw DecodeError (_("could not find video decoder"));
171 if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) {
172 throw DecodeError (N_("could not open video decoder"));
177 FFmpegDecoder::setup_audio ()
179 boost::mutex::scoped_lock lm (_mutex);
181 if (!_ffmpeg_content->audio_stream ()) {
185 _audio_codec_context = _format_context->streams[_ffmpeg_content->audio_stream()->id]->codec;
186 _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id);
188 if (_audio_codec == 0) {
189 throw DecodeError (_("could not find audio decoder"));
192 if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
193 throw DecodeError (N_("could not open audio decoder"));
198 FFmpegDecoder::setup_subtitle ()
200 boost::mutex::scoped_lock lm (_mutex);
202 if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
206 _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
207 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
209 if (_subtitle_codec == 0) {
210 throw DecodeError (_("could not find subtitle decoder"));
213 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
214 throw DecodeError (N_("could not open subtitle decoder"));
220 FFmpegDecoder::pass ()
222 int r = av_read_frame (_format_context, &_packet);
225 if (r != AVERROR_EOF) {
226 /* Maybe we should fail here, but for now we'll just finish off instead */
228 av_strerror (r, buf, sizeof(buf));
229 _film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
232 /* Get any remaining frames */
237 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
242 while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
243 filter_and_emit_video ();
247 if (_ffmpeg_content->audio_stream() && _decode_audio) {
248 decode_audio_packet ();
254 avcodec_get_frame_defaults (_frame);
256 if (_packet.stream_index == _video_stream && _decode_video) {
259 int const r = avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet);
260 if (r >= 0 && frame_finished) {
262 if (r != _packet.size) {
263 _film->log()->log (String::compose (N_("Used only %1 bytes of %2 in packet"), r, _packet.size));
266 filter_and_emit_video ();
269 } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) {
270 decode_audio_packet ();
271 } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id && _decode_subtitles) {
275 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) {
276 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
277 indicate that the previous subtitle should stop.
279 if (sub.num_rects > 0) {
280 shared_ptr<TimedSubtitle> ts;
282 emit_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
284 /* some problem with the subtitle; we probably didn't understand it */
287 emit_subtitle (shared_ptr<TimedSubtitle> ());
289 avsubtitle_free (&sub);
293 av_free_packet (&_packet);
297 /** @param data pointer to array of pointers to buffers.
298 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
300 shared_ptr<AudioBuffers>
301 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
303 assert (_ffmpeg_content->audio_channels());
304 assert (bytes_per_audio_sample());
306 /* Deinterleave and convert to float */
308 assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0);
310 int const total_samples = size / bytes_per_audio_sample();
311 int const frames = total_samples / _ffmpeg_content->audio_channels();
312 shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_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 == _ffmpeg_content->audio_channels()) {
332 case AV_SAMPLE_FMT_S16P:
334 int16_t** p = reinterpret_cast<int16_t **> (data);
335 for (int i = 0; i < _ffmpeg_content->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 == _ffmpeg_content->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 == _ffmpeg_content->audio_channels()) {
377 case AV_SAMPLE_FMT_FLTP:
379 float** p = reinterpret_cast<float**> (data);
380 for (int i = 0; i < _ffmpeg_content->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::video_frame_rate () 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::filter_and_emit_video ()
487 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
489 shared_ptr<FilterGraph> graph;
491 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
492 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
496 if (i == _filter_graphs.end ()) {
497 graph.reset (new FilterGraph (_film, this, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
498 _filter_graphs.push_back (graph);
499 _film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
504 list<shared_ptr<Image> > images = graph->process (_frame);
506 for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
507 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
508 if (bet != AV_NOPTS_VALUE) {
509 emit_video (*i, false, bet * av_q2d (_format_context->streams[_video_stream]->time_base));
511 _film->log()->log ("Dropping frame without PTS");
517 FFmpegDecoder::seek (double p)
519 return do_seek (p, false, false);
523 FFmpegDecoder::seek_back ()
525 do_seek (last_content_time() - 2.5 / video_frame_rate(), true, true);
529 FFmpegDecoder::seek_forward ()
531 do_seek (last_content_time() - 0.5 / video_frame_rate(), true, true);
535 FFmpegDecoder::do_seek (double p, bool backwards, bool accurate)
537 int64_t const vt = p / av_q2d (_format_context->streams[_video_stream]->time_base);
539 int const r = av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0);
541 avcodec_flush_buffers (_video_codec_context);
542 if (_subtitle_codec_context) {
543 avcodec_flush_buffers (_subtitle_codec_context);
548 int r = av_read_frame (_format_context, &_packet);
553 avcodec_get_frame_defaults (_frame);
555 if (_packet.stream_index == _video_stream) {
557 int const r = avcodec_decode_video2 (_video_codec_context, _frame, &finished, &_packet);
558 if (r >= 0 && finished) {
559 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
566 av_free_packet (&_packet);
574 FFmpegDecoder::film_changed (Film::Property p)
580 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
581 _filter_graphs.clear ();
590 /** @return Length (in video frames) according to our content's header */
592 FFmpegDecoder::video_length () const
594 return (double(_format_context->duration) / AV_TIME_BASE) * video_frame_rate();
598 FFmpegDecoder::decode_audio_packet ()
600 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
604 AVPacket copy_packet = _packet;
606 while (copy_packet.size > 0) {
609 int const decode_result = avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, ©_packet);
610 if (decode_result >= 0) {
611 if (frame_finished) {
613 /* Where we are in the source, in seconds */
614 double const source_pts_seconds = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
615 * av_frame_get_best_effort_timestamp(_frame);
617 int const data_size = av_samples_get_buffer_size (
618 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
621 assert (_audio_codec_context->channels == _ffmpeg_content->audio_channels());
622 Audio (deinterleave_audio (_frame->data, data_size), source_pts_seconds);
625 copy_packet.data += decode_result;
626 copy_packet.size -= decode_result;