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         , _just_sought (false)
  71 {
  72         setup_subtitle ();
  73
  74         if (video && audio && c->audio_stream() && c->first_video() && c->audio_stream()->first_audio) {
  75                 _pts_offset = compute_pts_offset (c->first_video().get(), c->audio_stream()->first_audio.get(), c->video_frame_rate());
  76         }
  77 }
  78
  79 double
  80 FFmpegDecoder::compute_pts_offset (double first_video, double first_audio, float video_frame_rate)
  81 {
  82         double const old_first_video = first_video;
  83
  84         /* Round the first video to a frame boundary */
  85         if (fabs (rint (first_video * video_frame_rate) - first_video * video_frame_rate) > 1e-6) {
  86                 first_video = ceil (first_video * video_frame_rate) / video_frame_rate;
  87         }
  88
  89         /* Compute the required offset (also removing any common start delay) */
  90         return first_video - old_first_video - min (first_video, first_audio);
  91 }
  92
  93 FFmpegDecoder::~FFmpegDecoder ()
  94 {
  95         boost::mutex::scoped_lock lm (_mutex);
  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, bool accurate)
 285 {
 286         double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base);
 287         int64_t const vt = frame / (_ffmpeg_content->video_frame_rate() * time_base);
 288         av_seek_frame (_format_context, _video_stream, vt, AVSEEK_FLAG_BACKWARD);
 289
 290         avcodec_flush_buffers (video_codec_context());
 291         if (_subtitle_codec_context) {
 292                 avcodec_flush_buffers (_subtitle_codec_context);
 293         }
 294
 295         _just_sought = true;
 296
 297         if (accurate) {
 298                 while (1) {
 299                         int r = av_read_frame (_format_context, &_packet);
 300                         if (r < 0) {
 301                                 return;
 302                         }
 303
 304                         avcodec_get_frame_defaults (_frame);
 305
 306                         if (_packet.stream_index == _video_stream) {
 307                                 int finished = 0;
 308                                 int const r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
 309                                 if (r >= 0 && finished) {
 310                                         int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
 311                                         if (bet >= vt) {
 312                                                 _video_position = rint (
 313                                                         (bet * time_base + _pts_offset) * _ffmpeg_content->video_frame_rate()
 314                                                         );
 315                                                 av_free_packet (&_packet);
 316                                                 break;
 317                                         }
 318                                 }
 319                         }
 320
 321                         av_free_packet (&_packet);
 322                 }
 323         }
 324 }
 325
 326 void
 327 FFmpegDecoder::decode_audio_packet ()
 328 {
 329         /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
 330            several times.
 331         */
 332
 333         AVPacket copy_packet = _packet;
 334
 335         while (copy_packet.size > 0) {
 336
 337                 int frame_finished;
 338                 int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, &copy_packet);
 339                 if (decode_result >= 0) {
 340                         if (frame_finished) {
 341
 342                                 if (_audio_position == 0) {
 343                                         /* Where we are in the source, in seconds */
 344                                         double const pts = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
 345                                                 * av_frame_get_best_effort_timestamp(_frame) - _pts_offset;
 346
 347                                         if (pts > 0) {
 348                                                 /* Emit some silence */
 349                                                 shared_ptr<AudioBuffers> silence (
 350                                                         new AudioBuffers (
 351                                                                 _ffmpeg_content->audio_channels(),
 352                                                                 pts * _ffmpeg_content->content_audio_frame_rate()
 353                                                                 )
 354                                                         );
 355
 356                                                 silence->make_silent ();
 357                                                 audio (silence, _audio_position);
 358                                         }
 359                                 }
 360
 361                                 copy_packet.data += decode_result;
 362                                 copy_packet.size -= decode_result;
 363                         }
 364                 }
 365         }
 366 }
 367
 368 bool
 369 FFmpegDecoder::decode_video_packet ()
 370 {
 371         int frame_finished;
 372         if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
 373                 return false;
 374         }
 375
 376         boost::mutex::scoped_lock lm (_filter_graphs_mutex);
 377
 378         shared_ptr<FilterGraph> graph;
 379
 380         list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
 381         while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
 382                 ++i;
 383         }
 384
 385         if (i == _filter_graphs.end ()) {
 386                 shared_ptr<const Film> film = _film.lock ();
 387                 assert (film);
 388
 389                 graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
 390                 _filter_graphs.push_back (graph);
 391
 392                 film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
 393         } else {
 394                 graph = *i;
 395         }
 396
 397         list<shared_ptr<Image> > images = graph->process (_frame);
 398
 399         string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second;
 400
 401         for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
 402
 403                 shared_ptr<Image> image = *i;
 404                 if (!post_process.empty ()) {
 405                         image = image->post_process (post_process, true);
 406                 }
 407
 408                 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
 409                 if (bet != AV_NOPTS_VALUE) {
 410
 411                         double const pts = bet * av_q2d (_format_context->streams[_video_stream]->time_base) - _pts_offset;
 412
 413                         if (_just_sought) {
 414                                 /* We just did a seek, so disable any attempts to correct for where we
 415                                    are / should be.
 416                                 */
 417                                 _video_position = rint (pts * _ffmpeg_content->video_frame_rate ());
 418                                 _just_sought = false;
 419                         }
 420
 421                         double const next = _video_position / _ffmpeg_content->video_frame_rate();
 422                         double const one_frame = 1 / _ffmpeg_content->video_frame_rate ();
 423                         double delta = pts - next;
 424
 425                         while (delta > one_frame) {
 426                                 /* This PTS is more than one frame forward in time of where we think we should be; emit
 427                                    a black frame.
 428                                 */
 429                                 boost::shared_ptr<Image> black (
 430                                         new SimpleImage (
 431                                                 static_cast<AVPixelFormat> (_frame->format),
 432                                                 libdcp::Size (video_codec_context()->width, video_codec_context()->height),
 433                                                 true
 434                                                 )
 435                                         );
 436
 437                                 black->make_black ();
 438                                 video (image, false, _video_position);
 439                                 delta -= one_frame;
 440                         }
 441
 442                         if (delta > -one_frame) {
 443                                 /* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */
 444                                 video (image, false, _video_position);
 445                         }
 446
 447                 } else {
 448                         shared_ptr<const Film> film = _film.lock ();
 449                         assert (film);
 450                         film->log()->log ("Dropping frame without PTS");
 451                 }
 452         }
 453
 454         return true;
 455 }
 456
 457
 458 void
 459 FFmpegDecoder::setup_subtitle ()
 460 {
 461         boost::mutex::scoped_lock lm (_mutex);
 462
 463         if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
 464                 return;
 465         }
 466
 467         _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
 468         _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
 469
 470         if (_subtitle_codec == 0) {
 471                 throw DecodeError (_("could not find subtitle decoder"));
 472         }
 473
 474         if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
 475                 throw DecodeError (N_("could not open subtitle decoder"));
 476         }
 477 }
 478
 479 bool
 480 FFmpegDecoder::done () const
 481 {
 482         bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length());
 483         bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length());
 484         return vd && ad;
 485 }
 486