src/lib/ffmpeg_examiner.cc

   1 /*
   2     Copyright (C) 2013-2015 Carl Hetherington <cth@carlh.net>
   3
   4     This file is part of DCP-o-matic.
   5
   6     DCP-o-matic 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.
  10
  11     DCP-o-matic 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.
  15
  16     You should have received a copy of the GNU General Public License
  17     along with DCP-o-matic.  If not, see <http://www.gnu.org/licenses/>.
  18
  19 */
  20
  21 #include "dcpomatic_log.h"
  22 #include "ffmpeg_examiner.h"
  23 #include "ffmpeg_content.h"
  24 #include "job.h"
  25 #include "ffmpeg_audio_stream.h"
  26 #include "ffmpeg_subtitle_stream.h"
  27 #include "util.h"
  28 #include "warnings.h"
  29 DCPOMATIC_DISABLE_WARNINGS
  30 extern "C" {
  31 #include <libavcodec/avcodec.h>
  32 #include <libavformat/avformat.h>
  33 #include <libavutil/pixfmt.h>
  34 #include <libavutil/pixdesc.h>
  35 #include <libavutil/eval.h>
  36 #include <libavutil/display.h>
  37 }
  38 DCPOMATIC_ENABLE_WARNINGS
  39 #include <iostream>
  40
  41 #include "i18n.h"
  42
  43 using std::string;
  44 using std::cout;
  45 using std::max;
  46 using std::vector;
  47 using std::shared_ptr;
  48 using boost::optional;
  49 using namespace dcpomatic;
  50
  51
  52 /* This is how many frames from the start of any video that we will examine to see if we
  53  * can spot soft 2:3 pull-down ("telecine").
  54  */
  55 static const int PULLDOWN_CHECK_FRAMES = 16;
  56
  57
  58 /** @param job job that the examiner is operating in, or 0 */
  59 FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Job> job)
  60         : FFmpeg (c)
  61         , _video_length (0)
  62         , _need_video_length (false)
  63         , _pulldown (false)
  64 {
  65         /* Find audio and subtitle streams */
  66
  67         for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
  68                 AVStream* s = _format_context->streams[i];
  69 DCPOMATIC_DISABLE_WARNINGS
  70                 if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
  71
  72                         /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
  73                            so bodge it here.  No idea why we should have to do this.
  74                         */
  75
  76                         if (s->codec->channel_layout == 0) {
  77                                 s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
  78                         }
  79
  80                         DCPOMATIC_ASSERT (_format_context->duration != AV_NOPTS_VALUE);
  81                         DCPOMATIC_ASSERT (s->codec->codec);
  82                         DCPOMATIC_ASSERT (s->codec->codec->name);
  83
  84                         _audio_streams.push_back (
  85                                 shared_ptr<FFmpegAudioStream> (
  86                                         new FFmpegAudioStream (
  87                                                 stream_name (s),
  88                                                 s->codec->codec->name,
  89                                                 s->id,
  90                                                 s->codec->sample_rate,
  91                                                 llrint ((double (_format_context->duration) / AV_TIME_BASE) * s->codec->sample_rate),
  92                                                 s->codec->channels
  93                                                 )
  94                                         )
  95                                 );
  96
  97                 } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
  98                         _subtitle_streams.push_back (shared_ptr<FFmpegSubtitleStream> (new FFmpegSubtitleStream (subtitle_stream_name (s), s->id)));
  99                 }
 100         }
 101
 102         if (has_video ()) {
 103                 /* See if the header has duration information in it */
 104                 _need_video_length = _format_context->duration == AV_NOPTS_VALUE;
 105                 if (!_need_video_length) {
 106                         _video_length = llrint ((double (_format_context->duration) / AV_TIME_BASE) * video_frame_rate().get());
 107                 }
 108         }
 109
 110         if (job && _need_video_length) {
 111                 job->sub (_("Finding length"));
 112         }
 113
 114         /* Run through until we find:
 115          *   - the first video.
 116          *   - the first audio for each stream.
 117          *   - the top-field-first and repeat-first-frame values ("temporal_reference") for the first PULLDOWN_CHECK_FRAMES video frames.
 118          */
 119
 120         int64_t const len = _file_group.length ();
 121         /* A string which we build up to describe the top-field-first and repeat-first-frame values for the first few frames.
 122          * It would be nicer to use something like vector<bool> here but we want to search the array for a pattern later,
 123          * and a string seems a reasonably neat way to do that.
 124          */
 125         string temporal_reference;
 126         while (true) {
 127                 int r = av_read_frame (_format_context, &_packet);
 128                 if (r < 0) {
 129                         break;
 130                 }
 131
 132                 if (job) {
 133                         if (len > 0) {
 134                                 job->set_progress (float (_format_context->pb->pos) / len);
 135                         } else {
 136                                 job->set_progress_unknown ();
 137                         }
 138                 }
 139
 140                 auto context = _format_context->streams[_packet.stream_index]->codec;
 141 DCPOMATIC_ENABLE_WARNINGS
 142
 143                 if (_video_stream && _packet.stream_index == _video_stream.get()) {
 144                         video_packet (context, temporal_reference);
 145                 }
 146
 147                 bool got_all_audio = true;
 148
 149                 for (size_t i = 0; i < _audio_streams.size(); ++i) {
 150                         if (_audio_streams[i]->uses_index (_format_context, _packet.stream_index)) {
 151                                 audio_packet (context, _audio_streams[i]);
 152                         }
 153                         if (!_audio_streams[i]->first_audio) {
 154                                 got_all_audio = false;
 155                         }
 156                 }
 157
 158                 av_packet_unref (&_packet);
 159
 160                 if (_first_video && got_all_audio && temporal_reference.size() >= (PULLDOWN_CHECK_FRAMES * 2)) {
 161                         /* All done */
 162                         break;
 163                 }
 164         }
 165
 166         _packet.data = nullptr;
 167         _packet.size = 0;
 168         /* XXX: I'm not sure this makes any sense: how does _packet.stream_index get the right value here? */
 169         auto context = _format_context->streams[_packet.stream_index]->codec;
 170         while (video_packet(context, temporal_reference)) {}
 171         for (size_t i = 0; i < _audio_streams.size(); ++i) {
 172                 if (_audio_streams[i]->uses_index (_format_context, _packet.stream_index)) {
 173                         audio_packet (context, _audio_streams[i]);
 174                 }
 175         }
 176
 177         if (_video_stream) {
 178                 /* This code taken from get_rotation() in ffmpeg:cmdutils.c */
 179                 AVStream* stream = _format_context->streams[*_video_stream];
 180                 AVDictionaryEntry* rotate_tag = av_dict_get (stream->metadata, "rotate", 0, 0);
 181                 uint8_t* displaymatrix = av_stream_get_side_data (stream, AV_PKT_DATA_DISPLAYMATRIX, 0);
 182                 _rotation = 0;
 183
 184                 if (rotate_tag && *rotate_tag->value && strcmp(rotate_tag->value, "0")) {
 185                         char *tail;
 186                         _rotation = av_strtod (rotate_tag->value, &tail);
 187                         if (*tail) {
 188                                 _rotation = 0;
 189                         }
 190                 }
 191
 192                 if (displaymatrix && !_rotation) {
 193                         _rotation = - av_display_rotation_get ((int32_t*) displaymatrix);
 194                 }
 195
 196                 _rotation = *_rotation - 360 * floor (*_rotation / 360 + 0.9 / 360);
 197         }
 198
 199         LOG_GENERAL("Temporal reference was %1", temporal_reference);
 200         if (temporal_reference.find("T2T3B2B3T2T3B2B3") != string::npos || temporal_reference.find("B2B3T2T3B2B3T2T3") != string::npos) {
 201                 /* The magical sequence (taken from mediainfo) suggests that 2:3 pull-down is in use */
 202                 _pulldown = true;
 203                 LOG_GENERAL_NC("Suggest that this may be 2:3 pull-down (soft telecine)");
 204         }
 205 }
 206
 207
 208 /** @param temporal_reference A string to which we should add two characters per frame;
 209  *  the first   is T or B depending on whether it's top- or bottom-field first,
 210  *  the second  is 3 or 2 depending on whether "repeat_pict" is true or not.
 211  *  @return true if some video was decoded, otherwise false.
 212  */
 213 bool
 214 FFmpegExaminer::video_packet (AVCodecContext* context, string& temporal_reference)
 215 {
 216         DCPOMATIC_ASSERT (_video_stream);
 217
 218         if (_first_video && !_need_video_length && temporal_reference.size() >= (PULLDOWN_CHECK_FRAMES * 2)) {
 219                 return false;
 220         }
 221
 222         int frame_finished;
 223 DCPOMATIC_DISABLE_WARNINGS
 224         if (avcodec_decode_video2 (context, _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
 225                 return false;
 226         }
 227 DCPOMATIC_ENABLE_WARNINGS
 228
 229         if (!_first_video) {
 230                 _first_video = frame_time (_format_context->streams[_video_stream.get()]);
 231         }
 232         if (_need_video_length) {
 233                 _video_length = frame_time (
 234                         _format_context->streams[_video_stream.get()]
 235                         ).get_value_or (ContentTime ()).frames_round (video_frame_rate().get ());
 236         }
 237         if (temporal_reference.size() < (PULLDOWN_CHECK_FRAMES * 2)) {
 238                 temporal_reference += (_frame->top_field_first ? "T" : "B");
 239                 temporal_reference += (_frame->repeat_pict ? "3" : "2");
 240         }
 241
 242         return true;
 243 }
 244
 245
 246 void
 247 FFmpegExaminer::audio_packet (AVCodecContext* context, shared_ptr<FFmpegAudioStream> stream)
 248 {
 249         if (stream->first_audio) {
 250                 return;
 251         }
 252
 253         int frame_finished;
 254 DCPOMATIC_DISABLE_WARNINGS
 255         if (avcodec_decode_audio4 (context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
 256 DCPOMATIC_ENABLE_WARNINGS
 257                 stream->first_audio = frame_time (stream->stream (_format_context));
 258         }
 259 }
 260
 261
 262 optional<ContentTime>
 263 FFmpegExaminer::frame_time (AVStream* s) const
 264 {
 265         optional<ContentTime> t;
 266
 267 DCPOMATIC_DISABLE_WARNINGS
 268         int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
 269 DCPOMATIC_ENABLE_WARNINGS
 270         if (bet != AV_NOPTS_VALUE) {
 271                 t = ContentTime::from_seconds (bet * av_q2d (s->time_base));
 272         }
 273
 274         return t;
 275 }
 276
 277
 278 optional<double>
 279 FFmpegExaminer::video_frame_rate () const
 280 {
 281         DCPOMATIC_ASSERT (_video_stream);
 282         return av_q2d(av_guess_frame_rate(_format_context, _format_context->streams[_video_stream.get()], 0));
 283 }
 284
 285 dcp::Size
 286 FFmpegExaminer::video_size () const
 287 {
 288         return dcp::Size (video_codec_context()->width, video_codec_context()->height);
 289 }
 290
 291 /** @return Length according to our content's header */
 292 Frame
 293 FFmpegExaminer::video_length () const
 294 {
 295         return max (Frame (1), _video_length);
 296 }
 297
 298 optional<double>
 299 FFmpegExaminer::sample_aspect_ratio () const
 300 {
 301         DCPOMATIC_ASSERT (_video_stream);
 302         AVRational sar = av_guess_sample_aspect_ratio (_format_context, _format_context->streams[_video_stream.get()], 0);
 303         if (sar.num == 0) {
 304                 /* I assume this means that we don't know */
 305                 return optional<double> ();
 306         }
 307         return double (sar.num) / sar.den;
 308 }
 309
 310 string
 311 FFmpegExaminer::subtitle_stream_name (AVStream* s) const
 312 {
 313         string n = stream_name (s);
 314
 315         if (n.empty()) {
 316                 n = _("unknown");
 317         }
 318
 319         return n;
 320 }
 321
 322 string
 323 FFmpegExaminer::stream_name (AVStream* s) const
 324 {
 325         string n;
 326
 327         if (s->metadata) {
 328                 AVDictionaryEntry const * lang = av_dict_get (s->metadata, "language", 0, 0);
 329                 if (lang) {
 330                         n = lang->value;
 331                 }
 332
 333                 AVDictionaryEntry const * title = av_dict_get (s->metadata, "title", 0, 0);
 334                 if (title) {
 335                         if (!n.empty()) {
 336                                 n += " ";
 337                         }
 338                         n += title->value;
 339                 }
 340         }
 341
 342         return n;
 343 }
 344
 345 optional<int>
 346 FFmpegExaminer::bits_per_pixel () const
 347 {
 348         if (video_codec_context()->pix_fmt == -1) {
 349                 return optional<int>();
 350         }
 351
 352         AVPixFmtDescriptor const * d = av_pix_fmt_desc_get (video_codec_context()->pix_fmt);
 353         DCPOMATIC_ASSERT (d);
 354         return av_get_bits_per_pixel (d);
 355 }
 356
 357 bool
 358 FFmpegExaminer::yuv () const
 359 {
 360         switch (video_codec_context()->pix_fmt) {
 361         case AV_PIX_FMT_YUV420P:
 362         case AV_PIX_FMT_YUYV422:
 363         case AV_PIX_FMT_YUV422P:
 364         case AV_PIX_FMT_YUV444P:
 365         case AV_PIX_FMT_YUV410P:
 366         case AV_PIX_FMT_YUV411P:
 367         case AV_PIX_FMT_YUVJ420P:
 368         case AV_PIX_FMT_YUVJ422P:
 369         case AV_PIX_FMT_YUVJ444P:
 370         case AV_PIX_FMT_UYVY422:
 371         case AV_PIX_FMT_UYYVYY411:
 372         case AV_PIX_FMT_NV12:
 373         case AV_PIX_FMT_NV21:
 374         case AV_PIX_FMT_YUV440P:
 375         case AV_PIX_FMT_YUVJ440P:
 376         case AV_PIX_FMT_YUVA420P:
 377         case AV_PIX_FMT_YUV420P16LE:
 378         case AV_PIX_FMT_YUV420P16BE:
 379         case AV_PIX_FMT_YUV422P16LE:
 380         case AV_PIX_FMT_YUV422P16BE:
 381         case AV_PIX_FMT_YUV444P16LE:
 382         case AV_PIX_FMT_YUV444P16BE:
 383         case AV_PIX_FMT_YUV420P9BE:
 384         case AV_PIX_FMT_YUV420P9LE:
 385         case AV_PIX_FMT_YUV420P10BE:
 386         case AV_PIX_FMT_YUV420P10LE:
 387         case AV_PIX_FMT_YUV422P10BE:
 388         case AV_PIX_FMT_YUV422P10LE:
 389         case AV_PIX_FMT_YUV444P9BE:
 390         case AV_PIX_FMT_YUV444P9LE:
 391         case AV_PIX_FMT_YUV444P10BE:
 392         case AV_PIX_FMT_YUV444P10LE:
 393         case AV_PIX_FMT_YUV422P9BE:
 394         case AV_PIX_FMT_YUV422P9LE:
 395         case AV_PIX_FMT_YUVA420P9BE:
 396         case AV_PIX_FMT_YUVA420P9LE:
 397         case AV_PIX_FMT_YUVA422P9BE:
 398         case AV_PIX_FMT_YUVA422P9LE:
 399         case AV_PIX_FMT_YUVA444P9BE:
 400         case AV_PIX_FMT_YUVA444P9LE:
 401         case AV_PIX_FMT_YUVA420P10BE:
 402         case AV_PIX_FMT_YUVA420P10LE:
 403         case AV_PIX_FMT_YUVA422P10BE:
 404         case AV_PIX_FMT_YUVA422P10LE:
 405         case AV_PIX_FMT_YUVA444P10BE:
 406         case AV_PIX_FMT_YUVA444P10LE:
 407         case AV_PIX_FMT_YUVA420P16BE:
 408         case AV_PIX_FMT_YUVA420P16LE:
 409         case AV_PIX_FMT_YUVA422P16BE:
 410         case AV_PIX_FMT_YUVA422P16LE:
 411         case AV_PIX_FMT_YUVA444P16BE:
 412         case AV_PIX_FMT_YUVA444P16LE:
 413         case AV_PIX_FMT_NV16:
 414         case AV_PIX_FMT_NV20LE:
 415         case AV_PIX_FMT_NV20BE:
 416         case AV_PIX_FMT_YVYU422:
 417         case AV_PIX_FMT_YUVA444P:
 418         case AV_PIX_FMT_YUVA422P:
 419         case AV_PIX_FMT_YUV420P12BE:
 420         case AV_PIX_FMT_YUV420P12LE:
 421         case AV_PIX_FMT_YUV420P14BE:
 422         case AV_PIX_FMT_YUV420P14LE:
 423         case AV_PIX_FMT_YUV422P12BE:
 424         case AV_PIX_FMT_YUV422P12LE:
 425         case AV_PIX_FMT_YUV422P14BE:
 426         case AV_PIX_FMT_YUV422P14LE:
 427         case AV_PIX_FMT_YUV444P12BE:
 428         case AV_PIX_FMT_YUV444P12LE:
 429         case AV_PIX_FMT_YUV444P14BE:
 430         case AV_PIX_FMT_YUV444P14LE:
 431         case AV_PIX_FMT_YUVJ411P:
 432                 return true;
 433         default:
 434                 return false;
 435         }
 436 }
 437
 438 bool
 439 FFmpegExaminer::has_video () const
 440 {
 441         return static_cast<bool> (_video_stream);
 442 }
 443
 444 VideoRange
 445 FFmpegExaminer::range () const
 446 {
 447         switch (color_range()) {
 448         case AVCOL_RANGE_MPEG:
 449         case AVCOL_RANGE_UNSPECIFIED:
 450                 return VideoRange::VIDEO;
 451         case AVCOL_RANGE_JPEG:
 452         default:
 453                 return VideoRange::FULL;
 454         }
 455 }