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"
45 #include "exceptions.h"
49 #include "ffmpeg_decoder.h"
50 #include "filter_graph.h"
56 using std::stringstream;
58 using boost::shared_ptr;
59 using boost::optional;
60 using boost::dynamic_pointer_cast;
62 FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, shared_ptr<const Options> o, Job* j)
64 , VideoDecoder (f, o, j)
65 , AudioDecoder (f, o, j)
69 , _video_codec_context (0)
71 , _audio_codec_context (0)
73 , _subtitle_codec_context (0)
82 FFmpegDecoder::~FFmpegDecoder ()
84 if (_audio_codec_context) {
85 avcodec_close (_audio_codec_context);
88 if (_video_codec_context) {
89 avcodec_close (_video_codec_context);
92 if (_subtitle_codec_context) {
93 avcodec_close (_subtitle_codec_context);
98 avformat_close_input (&_format_context);
102 FFmpegDecoder::setup_general ()
108 if ((r = avformat_open_input (&_format_context, _film->content_path().c_str(), 0, 0)) != 0) {
109 throw OpenFileError (_film->content_path ());
112 if (avformat_find_stream_info (_format_context, 0) < 0) {
113 throw DecodeError ("could not find stream information");
116 /* Find video, audio and subtitle streams and choose the first of each */
118 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
119 AVStream* s = _format_context->streams[i];
120 if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
122 } else if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
124 /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
125 so bodge it here. No idea why we should have to do this.
128 if (s->codec->channel_layout == 0) {
129 s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
132 _audio_streams.push_back (
133 shared_ptr<AudioStream> (
134 new FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channel_layout)
138 } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
139 _subtitle_streams.push_back (
140 shared_ptr<SubtitleStream> (
141 new SubtitleStream (stream_name (s), i)
147 if (_video_stream < 0) {
148 throw DecodeError ("could not find video stream");
151 _frame = avcodec_alloc_frame ();
153 throw DecodeError ("could not allocate frame");
158 FFmpegDecoder::setup_video ()
160 _video_codec_context = _format_context->streams[_video_stream]->codec;
161 _video_codec = avcodec_find_decoder (_video_codec_context->codec_id);
163 if (_video_codec == 0) {
164 throw DecodeError ("could not find video decoder");
167 if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) {
168 throw DecodeError ("could not open video decoder");
173 FFmpegDecoder::setup_audio ()
175 if (!_audio_stream) {
179 shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
182 _audio_codec_context = _format_context->streams[ffa->id()]->codec;
183 _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id);
185 if (_audio_codec == 0) {
186 throw DecodeError ("could not find audio decoder");
189 if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
190 throw DecodeError ("could not open audio decoder");
195 FFmpegDecoder::setup_subtitle ()
197 if (!_subtitle_stream) {
201 _subtitle_codec_context = _format_context->streams[_subtitle_stream->id()]->codec;
202 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
204 if (_subtitle_codec == 0) {
205 throw DecodeError ("could not find subtitle decoder");
208 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
209 throw DecodeError ("could not open subtitle decoder");
215 FFmpegDecoder::pass ()
217 int r = av_read_frame (_format_context, &_packet);
220 if (r != AVERROR_EOF) {
221 /* Maybe we should fail here, but for now we'll just finish off instead */
223 av_strerror (r, buf, sizeof(buf));
224 _film->log()->log (String::compose ("error on av_read_frame (%1) (%2)", buf, r));
227 /* Get any remaining frames */
232 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
236 while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
237 filter_and_emit_video (_frame);
240 if (_audio_stream && _opt->decode_audio) {
241 while (avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
242 int const data_size = av_samples_get_buffer_size (
243 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
246 assert (_audio_codec_context->channels == _film->audio_channels());
247 Audio (deinterleave_audio (_frame->data[0], data_size));
254 avcodec_get_frame_defaults (_frame);
256 shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
258 if (_packet.stream_index == _video_stream) {
261 int const r = avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet);
262 if (r >= 0 && frame_finished) {
264 if (r != _packet.size) {
265 _film->log()->log (String::compose ("Used only %1 bytes of %2 in packet", r, _packet.size));
268 /* Where we are in the output, in seconds */
269 double const out_pts_seconds = video_frame() / frames_per_second();
271 /* Where we are in the source, in seconds */
272 double const source_pts_seconds = av_q2d (_format_context->streams[_packet.stream_index]->time_base)
273 * av_frame_get_best_effort_timestamp(_frame);
276 String::compose ("Source video frame ready; source at %1, output at %2", source_pts_seconds, out_pts_seconds),
281 _first_video = source_pts_seconds;
284 /* Difference between where we are and where we should be */
285 double const delta = source_pts_seconds - _first_video.get() - out_pts_seconds;
286 double const one_frame = 1 / frames_per_second();
288 /* Insert frames if required to get out_pts_seconds up to pts_seconds */
289 if (delta > one_frame) {
290 int const extra = rint (delta / one_frame);
291 for (int i = 0; i < extra; ++i) {
292 repeat_last_video ();
295 "Extra video frame inserted at %1s; source frame %2, source PTS %3 (at %4 fps)",
296 out_pts_seconds, video_frame(), source_pts_seconds, frames_per_second()
302 if (delta > -one_frame) {
303 /* Process this frame */
304 filter_and_emit_video (_frame);
306 /* Otherwise we are omitting a frame to keep things right */
307 _film->log()->log (String::compose ("Frame removed at %1s", out_pts_seconds));
311 } else if (ffa && _packet.stream_index == ffa->id() && _opt->decode_audio) {
314 if (avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
316 /* Where we are in the source, in seconds */
317 double const source_pts_seconds = av_q2d (_format_context->streams[_packet.stream_index]->time_base)
318 * av_frame_get_best_effort_timestamp(_frame);
320 /* We only decode audio if we've had our first video packet through, and if it
321 was before this packet. Until then audio is thrown away.
324 if (_first_video && _first_video.get() <= source_pts_seconds) {
327 _first_audio = source_pts_seconds;
329 /* This is our first audio frame, and if we've arrived here we must have had our
330 first video frame. Push some silence to make up any gap between our first
331 video frame and our first audio.
334 /* frames of silence that we must push */
335 int const s = rint ((_first_audio.get() - _first_video.get()) * ffa->sample_rate ());
339 "First video at %1, first audio at %2, pushing %3 audio frames of silence for %4 channels (%5 bytes per sample)",
340 _first_video.get(), _first_audio.get(), s, ffa->channels(), bytes_per_audio_sample()
345 shared_ptr<AudioBuffers> audio (new AudioBuffers (ffa->channels(), s));
346 audio->make_silent ();
351 int const data_size = av_samples_get_buffer_size (
352 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
355 assert (_audio_codec_context->channels == _film->audio_channels());
356 Audio (deinterleave_audio (_frame->data[0], data_size));
360 } else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt->decode_subtitles && _first_video) {
364 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) {
365 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
366 indicate that the previous subtitle should stop.
368 if (sub.num_rects > 0) {
369 shared_ptr<TimedSubtitle> ts;
371 emit_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub, _first_video.get())));
373 /* some problem with the subtitle; we probably didn't understand it */
376 emit_subtitle (shared_ptr<TimedSubtitle> ());
378 avsubtitle_free (&sub);
382 av_free_packet (&_packet);
386 shared_ptr<AudioBuffers>
387 FFmpegDecoder::deinterleave_audio (uint8_t* data, int size)
389 assert (_film->audio_channels());
390 assert (bytes_per_audio_sample());
392 shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
395 /* Deinterleave and convert to float */
397 assert ((size % (bytes_per_audio_sample() * ffa->channels())) == 0);
399 int const total_samples = size / bytes_per_audio_sample();
400 int const frames = total_samples / _film->audio_channels();
401 shared_ptr<AudioBuffers> audio (new AudioBuffers (ffa->channels(), frames));
403 switch (audio_sample_format()) {
404 case AV_SAMPLE_FMT_S16:
406 int16_t* p = (int16_t *) data;
409 for (int i = 0; i < total_samples; ++i) {
410 audio->data(channel)[sample] = float(*p++) / (1 << 15);
413 if (channel == _film->audio_channels()) {
421 case AV_SAMPLE_FMT_S32:
423 int32_t* p = (int32_t *) data;
426 for (int i = 0; i < total_samples; ++i) {
427 audio->data(channel)[sample] = float(*p++) / (1 << 31);
430 if (channel == _film->audio_channels()) {
437 case AV_SAMPLE_FMT_FLTP:
439 float* p = reinterpret_cast<float*> (data);
440 for (int i = 0; i < _film->audio_channels(); ++i) {
441 memcpy (audio->data(i), p, frames * sizeof(float));
455 FFmpegDecoder::frames_per_second () const
457 AVStream* s = _format_context->streams[_video_stream];
459 if (s->avg_frame_rate.num && s->avg_frame_rate.den) {
460 return av_q2d (s->avg_frame_rate);
463 return av_q2d (s->r_frame_rate);
467 FFmpegDecoder::audio_sample_format () const
469 if (_audio_codec_context == 0) {
470 return (AVSampleFormat) 0;
473 return _audio_codec_context->sample_fmt;
477 FFmpegDecoder::native_size () const
479 return Size (_video_codec_context->width, _video_codec_context->height);
483 FFmpegDecoder::pixel_format () const
485 return _video_codec_context->pix_fmt;
489 FFmpegDecoder::time_base_numerator () const
491 return _video_codec_context->time_base.num;
495 FFmpegDecoder::time_base_denominator () const
497 return _video_codec_context->time_base.den;
501 FFmpegDecoder::sample_aspect_ratio_numerator () const
503 return _video_codec_context->sample_aspect_ratio.num;
507 FFmpegDecoder::sample_aspect_ratio_denominator () const
509 return _video_codec_context->sample_aspect_ratio.den;
513 FFmpegDecoder::stream_name (AVStream* s) const
517 AVDictionaryEntry const * lang = av_dict_get (s->metadata, "language", 0, 0);
522 AVDictionaryEntry const * title = av_dict_get (s->metadata, "title", 0, 0);
524 if (!n.str().empty()) {
530 if (n.str().empty()) {
538 FFmpegDecoder::bytes_per_audio_sample () const
540 return av_get_bytes_per_sample (audio_sample_format ());
544 FFmpegDecoder::set_audio_stream (shared_ptr<AudioStream> s)
546 AudioDecoder::set_audio_stream (s);
551 FFmpegDecoder::set_subtitle_stream (shared_ptr<SubtitleStream> s)
553 VideoDecoder::set_subtitle_stream (s);
558 FFmpegDecoder::filter_and_emit_video (AVFrame* frame)
560 shared_ptr<FilterGraph> graph;
562 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
563 while (i != _filter_graphs.end() && !(*i)->can_process (Size (frame->width, frame->height), (AVPixelFormat) frame->format)) {
567 if (i == _filter_graphs.end ()) {
568 graph.reset (new FilterGraph (_film, this, _opt->apply_crop, Size (frame->width, frame->height), (AVPixelFormat) frame->format));
569 _filter_graphs.push_back (graph);
570 _film->log()->log (String::compose ("New graph for %1x%2, pixel format %3", frame->width, frame->height, frame->format));
575 list<shared_ptr<Image> > images = graph->process (frame);
577 for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
582 shared_ptr<FFmpegAudioStream>
583 FFmpegAudioStream::create (string t, optional<int> v)
586 /* version < 1; no type in the string, and there's only FFmpeg streams anyway */
587 return shared_ptr<FFmpegAudioStream> (new FFmpegAudioStream (t, v));
593 if (type != "ffmpeg") {
594 return shared_ptr<FFmpegAudioStream> ();
597 return shared_ptr<FFmpegAudioStream> (new FFmpegAudioStream (t, v));
600 FFmpegAudioStream::FFmpegAudioStream (string t, optional<int> version)
608 n >> _id >> channels;
609 _channel_layout = av_get_default_channel_layout (channels);
613 /* Current (marked version 1) */
614 n >> type >> _id >> _sample_rate >> _channel_layout;
615 assert (type == "ffmpeg");
618 for (int i = 0; i < name_index; ++i) {
619 size_t const s = t.find (' ');
620 if (s != string::npos) {
621 t = t.substr (s + 1);
629 FFmpegAudioStream::to_string () const
631 return String::compose ("ffmpeg %1 %2 %3 %4", _id, _sample_rate, _channel_layout, _name);
634 /** @return Length (in video frames) according to our content's header */
636 FFmpegDecoder::length () const
638 return (double(_format_context->duration) / AV_TIME_BASE) * frames_per_second();