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;
62 using boost::shared_ptr;
63 using boost::optional;
64 using boost::dynamic_pointer_cast;
67 boost::mutex FFmpegDecoder::_mutex;
69 FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio, bool subtitles)
77 , _video_codec_context (0)
79 , _audio_codec_context (0)
81 , _subtitle_codec_context (0)
83 , _decode_video (video)
84 , _decode_audio (audio)
85 , _decode_subtitles (subtitles)
93 FFmpegDecoder::~FFmpegDecoder ()
95 boost::mutex::scoped_lock lm (_mutex);
97 if (_audio_codec_context) {
98 avcodec_close (_audio_codec_context);
101 if (_video_codec_context) {
102 avcodec_close (_video_codec_context);
105 if (_subtitle_codec_context) {
106 avcodec_close (_subtitle_codec_context);
111 avformat_close_input (&_format_context);
115 FFmpegDecoder::setup_general ()
119 if (avformat_open_input (&_format_context, _ffmpeg_content->file().string().c_str(), 0, 0) < 0) {
120 throw OpenFileError (_ffmpeg_content->file().string ());
123 if (avformat_find_stream_info (_format_context, 0) < 0) {
124 throw DecodeError (_("could not find stream information"));
127 /* Find video, audio and subtitle streams */
129 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
130 AVStream* s = _format_context->streams[i];
131 if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
133 } else if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
135 /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
136 so bodge it here. No idea why we should have to do this.
139 if (s->codec->channel_layout == 0) {
140 s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
143 _audio_streams.push_back (
144 shared_ptr<FFmpegAudioStream> (
145 new FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channels)
149 } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
150 _subtitle_streams.push_back (shared_ptr<FFmpegSubtitleStream> (new FFmpegSubtitleStream (stream_name (s), i)));
154 if (_video_stream < 0) {
155 throw DecodeError (N_("could not find video stream"));
158 _frame = avcodec_alloc_frame ();
160 throw DecodeError (N_("could not allocate frame"));
165 FFmpegDecoder::setup_video ()
167 boost::mutex::scoped_lock lm (_mutex);
169 _video_codec_context = _format_context->streams[_video_stream]->codec;
170 _video_codec = avcodec_find_decoder (_video_codec_context->codec_id);
172 if (_video_codec == 0) {
173 throw DecodeError (_("could not find video decoder"));
176 if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) {
177 throw DecodeError (N_("could not open video decoder"));
182 FFmpegDecoder::setup_audio ()
184 boost::mutex::scoped_lock lm (_mutex);
186 if (!_ffmpeg_content->audio_stream ()) {
190 _audio_codec_context = _format_context->streams[_ffmpeg_content->audio_stream()->id]->codec;
191 _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id);
193 if (_audio_codec == 0) {
194 throw DecodeError (_("could not find audio decoder"));
197 if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
198 throw DecodeError (N_("could not open audio decoder"));
203 FFmpegDecoder::setup_subtitle ()
205 boost::mutex::scoped_lock lm (_mutex);
207 if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
211 _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
212 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
214 if (_subtitle_codec == 0) {
215 throw DecodeError (_("could not find subtitle decoder"));
218 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
219 throw DecodeError (N_("could not open subtitle decoder"));
225 FFmpegDecoder::pass ()
227 cout << "ffd pass.\n";
228 int r = av_read_frame (_format_context, &_packet);
229 cout << "A " << r << "\n";
232 if (r != AVERROR_EOF) {
233 /* Maybe we should fail here, but for now we'll just finish off instead */
235 av_strerror (r, buf, sizeof(buf));
236 _film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
239 /* Get any remaining frames */
244 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
247 while (decode_video_packet ());
250 if (_ffmpeg_content->audio_stream() && _decode_audio) {
251 decode_audio_packet ();
257 avcodec_get_frame_defaults (_frame);
259 if (_packet.stream_index == _video_stream && _decode_video) {
261 decode_video_packet ();
263 } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) {
264 decode_audio_packet ();
265 } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id && _decode_subtitles) {
269 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) {
270 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
271 indicate that the previous subtitle should stop.
273 if (sub.num_rects > 0) {
274 shared_ptr<TimedSubtitle> ts;
276 emit_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
278 /* some problem with the subtitle; we probably didn't understand it */
281 emit_subtitle (shared_ptr<TimedSubtitle> ());
283 avsubtitle_free (&sub);
288 av_free_packet (&_packet);
292 /** @param data pointer to array of pointers to buffers.
293 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
295 shared_ptr<AudioBuffers>
296 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
298 assert (_ffmpeg_content->audio_channels());
299 assert (bytes_per_audio_sample());
301 /* Deinterleave and convert to float */
303 assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0);
305 int const total_samples = size / bytes_per_audio_sample();
306 int const frames = total_samples / _ffmpeg_content->audio_channels();
307 shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
309 switch (audio_sample_format()) {
310 case AV_SAMPLE_FMT_S16:
312 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
315 for (int i = 0; i < total_samples; ++i) {
316 audio->data(channel)[sample] = float(*p++) / (1 << 15);
319 if (channel == _ffmpeg_content->audio_channels()) {
327 case AV_SAMPLE_FMT_S16P:
329 int16_t** p = reinterpret_cast<int16_t **> (data);
330 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
331 for (int j = 0; j < frames; ++j) {
332 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
338 case AV_SAMPLE_FMT_S32:
340 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
343 for (int i = 0; i < total_samples; ++i) {
344 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
347 if (channel == _ffmpeg_content->audio_channels()) {
355 case AV_SAMPLE_FMT_FLT:
357 float* p = reinterpret_cast<float*> (data[0]);
360 for (int i = 0; i < total_samples; ++i) {
361 audio->data(channel)[sample] = *p++;
364 if (channel == _ffmpeg_content->audio_channels()) {
372 case AV_SAMPLE_FMT_FLTP:
374 float** p = reinterpret_cast<float**> (data);
375 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
376 memcpy (audio->data(i), p[i], frames * sizeof(float));
382 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
389 FFmpegDecoder::video_frame_rate () const
391 AVStream* s = _format_context->streams[_video_stream];
393 if (s->avg_frame_rate.num && s->avg_frame_rate.den) {
394 return av_q2d (s->avg_frame_rate);
397 return av_q2d (s->r_frame_rate);
401 FFmpegDecoder::audio_sample_format () const
403 if (_audio_codec_context == 0) {
404 return (AVSampleFormat) 0;
407 return _audio_codec_context->sample_fmt;
411 FFmpegDecoder::native_size () const
413 return libdcp::Size (_video_codec_context->width, _video_codec_context->height);
417 FFmpegDecoder::pixel_format () const
419 return _video_codec_context->pix_fmt;
423 FFmpegDecoder::time_base_numerator () const
425 return _video_codec_context->time_base.num;
429 FFmpegDecoder::time_base_denominator () const
431 return _video_codec_context->time_base.den;
435 FFmpegDecoder::sample_aspect_ratio_numerator () const
437 return _video_codec_context->sample_aspect_ratio.num;
441 FFmpegDecoder::sample_aspect_ratio_denominator () const
443 return _video_codec_context->sample_aspect_ratio.den;
447 FFmpegDecoder::stream_name (AVStream* s) const
452 AVDictionaryEntry const * lang = av_dict_get (s->metadata, N_("language"), 0, 0);
457 AVDictionaryEntry const * title = av_dict_get (s->metadata, N_("title"), 0, 0);
459 if (!n.str().empty()) {
466 if (n.str().empty()) {
474 FFmpegDecoder::bytes_per_audio_sample () const
476 return av_get_bytes_per_sample (audio_sample_format ());
480 FFmpegDecoder::seek (Time t)
482 return do_seek (t, false, false);
486 FFmpegDecoder::seek_back ()
488 if (last_content_time() < 2.5) {
492 return do_seek (last_content_time() - 2.5 * TIME_HZ / video_frame_rate(), true, true);
496 FFmpegDecoder::seek_forward ()
498 if (last_content_time() >= (video_length() - video_frame_rate())) {
502 return do_seek (last_content_time() - 0.5 * TIME_HZ / video_frame_rate(), true, true);
506 FFmpegDecoder::do_seek (Time t, bool backwards, bool accurate)
508 int64_t const vt = t / (av_q2d (_format_context->streams[_video_stream]->time_base) * TIME_HZ);
510 cout << "seek to " << vt << " (acc=" << accurate << ") (sec " << (vt * av_q2d (_format_context->streams[_video_stream]->time_base)) << "\n";
512 int const r = av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0);
514 avcodec_flush_buffers (_video_codec_context);
515 if (_subtitle_codec_context) {
516 avcodec_flush_buffers (_subtitle_codec_context);
521 int r = av_read_frame (_format_context, &_packet);
526 avcodec_get_frame_defaults (_frame);
528 if (_packet.stream_index == _video_stream) {
530 int const r = avcodec_decode_video2 (_video_codec_context, _frame, &finished, &_packet);
531 if (r >= 0 && finished) {
532 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
539 av_free_packet (&_packet);
543 cout << "seek ok.\n";
548 FFmpegDecoder::film_changed (Film::Property p)
553 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
554 _filter_graphs.clear ();
563 /** @return Length (in video frames) according to our content's header */
565 FFmpegDecoder::video_length () const
567 return (double(_format_context->duration) / AV_TIME_BASE) * video_frame_rate();
571 FFmpegDecoder::decode_audio_packet ()
573 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
577 AVPacket copy_packet = _packet;
579 while (copy_packet.size > 0) {
582 int const decode_result = avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, ©_packet);
583 if (decode_result >= 0) {
584 if (frame_finished) {
586 /* Where we are in the source, in seconds */
587 double const source_pts_seconds = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
588 * av_frame_get_best_effort_timestamp(_frame);
590 int const data_size = av_samples_get_buffer_size (
591 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
594 assert (_audio_codec_context->channels == _ffmpeg_content->audio_channels());
595 Audio (deinterleave_audio (_frame->data, data_size), source_pts_seconds);
598 copy_packet.data += decode_result;
599 copy_packet.size -= decode_result;
605 FFmpegDecoder::decode_video_packet ()
608 cout << "avc decode v2\n";
609 if (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
612 cout << "done that.\n";
614 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
616 cout << "got lock.\n";
618 shared_ptr<FilterGraph> graph;
620 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
621 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
625 cout << "found graph.\n";
627 if (i == _filter_graphs.end ()) {
628 graph.reset (new FilterGraph (_film, this, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
629 _filter_graphs.push_back (graph);
630 _film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
636 cout << "pushed in.\n";
637 list<shared_ptr<Image> > images = graph->process (_frame);
638 cout << "got " << images.size() << "\n";
640 for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
641 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
642 if (bet != AV_NOPTS_VALUE) {
643 /* XXX: may need to insert extra frames / remove frames here ...
646 cout << "emitting.\n";
647 emit_video (*i, false, bet * av_q2d (_format_context->streams[_video_stream]->time_base) * TIME_HZ);
648 cout << "emitted.\n";
650 _film->log()->log ("Dropping frame without PTS");