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