Try to fix seeking with FFmpeg.
[dcpomatic.git] / src / lib / ffmpeg_decoder.cc
1 /*
2     Copyright (C) 2012 Carl Hetherington <cth@carlh.net>
3
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.
8
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.
13
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.
17
18 */
19
20 /** @file  src/ffmpeg_decoder.cc
21  *  @brief A decoder using FFmpeg to decode content.
22  */
23
24 #include <stdexcept>
25 #include <vector>
26 #include <sstream>
27 #include <iomanip>
28 #include <iostream>
29 #include <stdint.h>
30 #include <boost/lexical_cast.hpp>
31 #include <sndfile.h>
32 extern "C" {
33 #include <libavcodec/avcodec.h>
34 #include <libavformat/avformat.h>
35 }
36 #include "film.h"
37 #include "filter.h"
38 #include "exceptions.h"
39 #include "image.h"
40 #include "util.h"
41 #include "log.h"
42 #include "ffmpeg_decoder.h"
43 #include "filter_graph.h"
44 #include "subtitle.h"
45 #include "audio_buffers.h"
46
47 #include "i18n.h"
48
49 using std::cout;
50 using std::string;
51 using std::vector;
52 using std::stringstream;
53 using std::list;
54 using std::min;
55 using boost::shared_ptr;
56 using boost::optional;
57 using boost::dynamic_pointer_cast;
58 using libdcp::Size;
59
60 FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio)
61         : Decoder (f)
62         , VideoDecoder (f)
63         , AudioDecoder (f)
64         , FFmpeg (c)
65         , _subtitle_codec_context (0)
66         , _subtitle_codec (0)
67         , _decode_video (video)
68         , _decode_audio (audio)
69         , _pts_offset (0)
70 {
71         setup_subtitle ();
72
73         if (video && audio && c->audio_stream() && c->first_video() && c->audio_stream()->first_audio) {
74                 _pts_offset = compute_pts_offset (c->first_video().get(), c->audio_stream()->first_audio.get(), c->video_frame_rate());
75         }
76 }
77
78 double
79 FFmpegDecoder::compute_pts_offset (double first_video, double first_audio, float video_frame_rate)
80 {
81         assert (first_video >= 0);
82         assert (first_audio >= 0);
83         
84         double const old_first_video = first_video;
85         
86         /* Round the first video to a frame boundary */
87         if (fabs (rint (first_video * video_frame_rate) - first_video * video_frame_rate) > 1e-6) {
88                 first_video = ceil (first_video * video_frame_rate) / video_frame_rate;
89         }
90
91         /* Compute the required offset (also removing any common start delay) */
92         return first_video - old_first_video - min (first_video, first_audio);
93 }
94
95 FFmpegDecoder::~FFmpegDecoder ()
96 {
97         if (_subtitle_codec_context) {
98                 avcodec_close (_subtitle_codec_context);
99         }
100 }       
101
102 void
103 FFmpegDecoder::pass ()
104 {
105         int r = av_read_frame (_format_context, &_packet);
106
107         if (r < 0) {
108                 if (r != AVERROR_EOF) {
109                         /* Maybe we should fail here, but for now we'll just finish off instead */
110                         char buf[256];
111                         av_strerror (r, buf, sizeof(buf));
112                         shared_ptr<const Film> film = _film.lock ();
113                         assert (film);
114                         film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
115                 }
116
117                 /* Get any remaining frames */
118                 
119                 _packet.data = 0;
120                 _packet.size = 0;
121                 
122                 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
123                 
124                 if (_decode_video) {
125                         while (decode_video_packet ());
126                 }
127
128                 if (_ffmpeg_content->audio_stream() && _decode_audio) {
129                         decode_audio_packet ();
130                 }
131
132                 /* Stop us being asked for any more data */
133                 _video_position = _ffmpeg_content->video_length ();
134                 _audio_position = _ffmpeg_content->audio_length ();
135                 return;
136         }
137
138         avcodec_get_frame_defaults (_frame);
139
140         if (_packet.stream_index == _video_stream && _decode_video) {
141                 decode_video_packet ();
142         } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) {
143                 decode_audio_packet ();
144         } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id) {
145 #if 0           
146
147                 int got_subtitle;
148                 AVSubtitle sub;
149                 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) {
150                         /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
151                            indicate that the previous subtitle should stop.
152                         */
153                         if (sub.num_rects > 0) {
154                                 shared_ptr<TimedSubtitle> ts;
155                                 try {
156                                         subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
157                                 } catch (...) {
158                                         /* some problem with the subtitle; we probably didn't understand it */
159                                 }
160                         } else {
161                                 subtitle (shared_ptr<TimedSubtitle> ());
162                         }
163                         avsubtitle_free (&sub);
164                 }
165 #endif          
166         }
167
168         av_free_packet (&_packet);
169 }
170
171 /** @param data pointer to array of pointers to buffers.
172  *  Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
173  */
174 shared_ptr<AudioBuffers>
175 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
176 {
177         assert (_ffmpeg_content->audio_channels());
178         assert (bytes_per_audio_sample());
179
180         /* Deinterleave and convert to float */
181
182         assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0);
183
184         int const total_samples = size / bytes_per_audio_sample();
185         int const frames = total_samples / _ffmpeg_content->audio_channels();
186         shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
187
188         switch (audio_sample_format()) {
189         case AV_SAMPLE_FMT_S16:
190         {
191                 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
192                 int sample = 0;
193                 int channel = 0;
194                 for (int i = 0; i < total_samples; ++i) {
195                         audio->data(channel)[sample] = float(*p++) / (1 << 15);
196
197                         ++channel;
198                         if (channel == _ffmpeg_content->audio_channels()) {
199                                 channel = 0;
200                                 ++sample;
201                         }
202                 }
203         }
204         break;
205
206         case AV_SAMPLE_FMT_S16P:
207         {
208                 int16_t** p = reinterpret_cast<int16_t **> (data);
209                 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
210                         for (int j = 0; j < frames; ++j) {
211                                 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
212                         }
213                 }
214         }
215         break;
216         
217         case AV_SAMPLE_FMT_S32:
218         {
219                 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
220                 int sample = 0;
221                 int channel = 0;
222                 for (int i = 0; i < total_samples; ++i) {
223                         audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
224
225                         ++channel;
226                         if (channel == _ffmpeg_content->audio_channels()) {
227                                 channel = 0;
228                                 ++sample;
229                         }
230                 }
231         }
232         break;
233
234         case AV_SAMPLE_FMT_FLT:
235         {
236                 float* p = reinterpret_cast<float*> (data[0]);
237                 int sample = 0;
238                 int channel = 0;
239                 for (int i = 0; i < total_samples; ++i) {
240                         audio->data(channel)[sample] = *p++;
241
242                         ++channel;
243                         if (channel == _ffmpeg_content->audio_channels()) {
244                                 channel = 0;
245                                 ++sample;
246                         }
247                 }
248         }
249         break;
250                 
251         case AV_SAMPLE_FMT_FLTP:
252         {
253                 float** p = reinterpret_cast<float**> (data);
254                 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
255                         memcpy (audio->data(i), p[i], frames * sizeof(float));
256                 }
257         }
258         break;
259
260         default:
261                 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
262         }
263
264         return audio;
265 }
266
267 AVSampleFormat
268 FFmpegDecoder::audio_sample_format () const
269 {
270         if (!_ffmpeg_content->audio_stream()) {
271                 return (AVSampleFormat) 0;
272         }
273         
274         return audio_codec_context()->sample_fmt;
275 }
276
277 int
278 FFmpegDecoder::bytes_per_audio_sample () const
279 {
280         return av_get_bytes_per_sample (audio_sample_format ());
281 }
282
283 void
284 FFmpegDecoder::seek (VideoContent::Frame frame)
285 {
286         do_seek (frame, false, false);
287 }
288
289 void
290 FFmpegDecoder::seek_back ()
291 {
292         if (_video_position == 0) {
293                 return;
294         }
295         
296         do_seek (_video_position - 1, true, true);
297 }
298
299 void
300 FFmpegDecoder::do_seek (VideoContent::Frame frame, bool backwards, bool accurate)
301 {
302         int64_t const vt = frame * _ffmpeg_content->video_frame_rate() / av_q2d (_format_context->streams[_video_stream]->time_base);
303         av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0);
304         _video_position = frame;
305
306         avcodec_flush_buffers (video_codec_context());
307         if (_subtitle_codec_context) {
308                 avcodec_flush_buffers (_subtitle_codec_context);
309         }
310
311         if (accurate) {
312                 while (1) {
313                         int r = av_read_frame (_format_context, &_packet);
314                         if (r < 0) {
315                                 return;
316                         }
317                         
318                         avcodec_get_frame_defaults (_frame);
319                         
320                         if (_packet.stream_index == _video_stream) {
321                                 int finished = 0;
322                                 int const r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
323                                 if (r >= 0 && finished) {
324                                         int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
325                                         if (bet > vt) {
326                                                 _video_position = (bet * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset)
327                                                         * _ffmpeg_content->video_frame_rate();
328                                                 break;
329                                         }
330                                 }
331                         }
332                         
333                         av_free_packet (&_packet);
334                 }
335         }
336 }
337
338 void
339 FFmpegDecoder::decode_audio_packet ()
340 {
341         /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
342            several times.
343         */
344         
345         AVPacket copy_packet = _packet;
346
347         while (copy_packet.size > 0) {
348
349                 int frame_finished;
350                 int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, &copy_packet);
351                 if (decode_result >= 0) {
352                         if (frame_finished) {
353
354                                 if (_audio_position == 0) {
355                                         /* Where we are in the source, in seconds */
356                                         double const pts = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
357                                                 * av_frame_get_best_effort_timestamp(_frame) - _pts_offset;
358
359                                         if (pts > 0) {
360                                                 /* Emit some silence */
361                                                 shared_ptr<AudioBuffers> silence (
362                                                         new AudioBuffers (
363                                                                 _ffmpeg_content->audio_channels(),
364                                                                 pts * _ffmpeg_content->content_audio_frame_rate()
365                                                                 )
366                                                         );
367                                                 
368                                                 silence->make_silent ();
369                                                 audio (silence, _audio_position);
370                                         }
371                                 }
372                                         
373                                 
374                                 int const data_size = av_samples_get_buffer_size (
375                                         0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1
376                                         );
377                                 
378                                 assert (audio_codec_context()->channels == _ffmpeg_content->audio_channels());
379                                 audio (deinterleave_audio (_frame->data, data_size), _audio_position);
380                         }
381                         
382                         copy_packet.data += decode_result;
383                         copy_packet.size -= decode_result;
384                 }
385         }
386 }
387
388 bool
389 FFmpegDecoder::decode_video_packet ()
390 {
391         int frame_finished;
392         if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
393                 return false;
394         }
395                 
396         boost::mutex::scoped_lock lm (_filter_graphs_mutex);
397
398         shared_ptr<FilterGraph> graph;
399         
400         list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
401         while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
402                 ++i;
403         }
404
405         if (i == _filter_graphs.end ()) {
406                 shared_ptr<const Film> film = _film.lock ();
407                 assert (film);
408
409                 graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
410                 _filter_graphs.push_back (graph);
411
412                 film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
413         } else {
414                 graph = *i;
415         }
416
417         list<shared_ptr<Image> > images = graph->process (_frame);
418
419         string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second;
420         
421         for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
422
423                 shared_ptr<Image> image = *i;
424                 if (!post_process.empty ()) {
425                         image = image->post_process (post_process, true);
426                 }
427                 
428                 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
429                 if (bet != AV_NOPTS_VALUE) {
430
431                         double const pts = bet * av_q2d (_format_context->streams[_video_stream]->time_base) - _pts_offset;
432                         double const next = _video_position / _ffmpeg_content->video_frame_rate();
433                         double const one_frame = 1 / _ffmpeg_content->video_frame_rate ();
434                         double delta = pts - next;
435
436                         while (delta > one_frame) {
437                                 /* This PTS is more than one frame forward in time of where we think we should be; emit
438                                    a black frame.
439                                 */
440                                 boost::shared_ptr<Image> black (
441                                         new SimpleImage (
442                                                 static_cast<AVPixelFormat> (_frame->format),
443                                                 libdcp::Size (video_codec_context()->width, video_codec_context()->height),
444                                                 true
445                                                 )
446                                         );
447                                 
448                                 black->make_black ();
449                                 video (image, false, _video_position);
450                                 delta -= one_frame;
451                         }
452
453                         if (delta > -one_frame) {
454                                 /* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */
455                                 video (image, false, _video_position);
456                         }
457                 } else {
458                         shared_ptr<const Film> film = _film.lock ();
459                         assert (film);
460                         film->log()->log ("Dropping frame without PTS");
461                 }
462         }
463
464         return true;
465 }
466
467         
468 void
469 FFmpegDecoder::setup_subtitle ()
470 {
471         boost::mutex::scoped_lock lm (_mutex);
472         
473         if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
474                 return;
475         }
476
477         _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
478         _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
479
480         if (_subtitle_codec == 0) {
481                 throw DecodeError (_("could not find subtitle decoder"));
482         }
483         
484         if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
485                 throw DecodeError (N_("could not open subtitle decoder"));
486         }
487 }
488
489 bool
490 FFmpegDecoder::done () const
491 {
492         bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length());
493         bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length());
494         return vd && ad;
495 }
496