src/lib/ffmpeg_examiner.cc

   1 /*
   2     Copyright (C) 2013-2021 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
  22 #include "dcpomatic_log.h"
  23 #include "ffmpeg_examiner.h"
  24 #include "ffmpeg_content.h"
  25 #include "job.h"
  26 #include "ffmpeg_audio_stream.h"
  27 #include "ffmpeg_subtitle_stream.h"
  28 #include "util.h"
  29 #include "warnings.h"
  30 DCPOMATIC_DISABLE_WARNINGS
  31 extern "C" {
  32 #include <libavcodec/avcodec.h>
  33 #include <libavformat/avformat.h>
  34 #include <libavutil/pixfmt.h>
  35 #include <libavutil/pixdesc.h>
  36 #include <libavutil/eval.h>
  37 #include <libavutil/display.h>
  38 }
  39 DCPOMATIC_ENABLE_WARNINGS
  40 #include <iostream>
  41
  42 #include "i18n.h"
  43
  44
  45 using std::cout;
  46 using std::make_shared;
  47 using std::max;
  48 using std::shared_ptr;
  49 using std::string;
  50 using std::vector;
  51 using boost::optional;
  52 using namespace dcpomatic;
  53
  54
  55 /* This is how many frames from the start of any video that we will examine to see if we
  56  * can spot soft 2:3 pull-down ("telecine").
  57  */
  58 static const int PULLDOWN_CHECK_FRAMES = 16;
  59
  60
  61 /** @param job job that the examiner is operating in, or 0 */
  62 FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Job> job)
  63         : FFmpeg (c)
  64         , _video_length (0)
  65         , _need_video_length (false)
  66         , _pulldown (false)
  67 {
  68         /* Find audio and subtitle streams */
  69
  70         for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
  71                 auto s = _format_context->streams[i];
  72 DCPOMATIC_DISABLE_WARNINGS
  73                 if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
  74
  75                         /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
  76                            so bodge it here.  No idea why we should have to do this.
  77                         */
  78
  79                         if (s->codec->channel_layout == 0) {
  80                                 s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
  81                         }
  82
  83                         DCPOMATIC_ASSERT (_format_context->duration != AV_NOPTS_VALUE);
  84                         DCPOMATIC_ASSERT (s->codec->codec);
  85                         DCPOMATIC_ASSERT (s->codec->codec->name);
  86
  87                         _audio_streams.push_back (
  88                                 make_shared<FFmpegAudioStream>(
  89                                         stream_name (s),
  90                                         s->codec->codec->name,
  91                                         s->id,
  92                                         s->codec->sample_rate,
  93                                         llrint ((double(_format_context->duration) / AV_TIME_BASE) * s->codec->sample_rate),
  94                                         s->codec->channels
  95                                         )
  96                                 );
  97
  98                 } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
  99                         _subtitle_streams.push_back (shared_ptr<FFmpegSubtitleStream> (new FFmpegSubtitleStream (subtitle_stream_name (s), s->id)));
 100                 }
 101         }
 102
 103         if (has_video ()) {
 104                 /* See if the header has duration information in it */
 105                 _need_video_length = _format_context->duration == AV_NOPTS_VALUE;
 106                 if (!_need_video_length) {
 107                         _video_length = llrint ((double (_format_context->duration) / AV_TIME_BASE) * video_frame_rate().get());
 108                 }
 109         }
 110
 111         if (job && _need_video_length) {
 112                 job->sub (_("Finding length"));
 113         }
 114
 115         /* Run through until we find:
 116          *   - the first video.
 117          *   - the first audio for each stream.
 118          *   - the top-field-first and repeat-first-frame values ("temporal_reference") for the first PULLDOWN_CHECK_FRAMES video frames.
 119          */
 120
 121         int64_t const len = _file_group.length ();
 122         /* A string which we build up to describe the top-field-first and repeat-first-frame values for the first few frames.
 123          * It would be nicer to use something like vector<bool> here but we want to search the array for a pattern later,
 124          * and a string seems a reasonably neat way to do that.
 125          */
 126         string temporal_reference;
 127         while (true) {
 128                 auto packet = av_packet_alloc ();
 129                 DCPOMATIC_ASSERT (packet);
 130                 int r = av_read_frame (_format_context, packet);
 131                 if (r < 0) {
 132                         av_packet_free (&packet);
 133                         break;
 134                 }
 135
 136                 if (job) {
 137                         if (len > 0) {
 138                                 job->set_progress (float (_format_context->pb->pos) / len);
 139                         } else {
 140                                 job->set_progress_unknown ();
 141                         }
 142                 }
 143
 144                 auto context = _format_context->streams[packet->stream_index]->codec;
 145 DCPOMATIC_ENABLE_WARNINGS
 146
 147                 if (_video_stream && packet->stream_index == _video_stream.get()) {
 148                         video_packet (context, temporal_reference, packet);
 149                 }
 150
 151                 bool got_all_audio = true;
 152
 153                 for (size_t i = 0; i < _audio_streams.size(); ++i) {
 154                         if (_audio_streams[i]->uses_index(_format_context, packet->stream_index)) {
 155                                 audio_packet (context, _audio_streams[i], packet);
 156                         }
 157                         if (!_audio_streams[i]->first_audio) {
 158                                 got_all_audio = false;
 159                         }
 160                 }
 161
 162                 av_packet_free (&packet);
 163
 164                 if (_first_video && got_all_audio && temporal_reference.size() >= (PULLDOWN_CHECK_FRAMES * 2)) {
 165                         /* All done */
 166                         break;
 167                 }
 168         }
 169
 170         if (_video_stream) {
 171                 AVPacket packet;
 172                 av_init_packet (&packet);
 173                 packet.data = nullptr;
 174                 packet.size = 0;
 175 DCPOMATIC_DISABLE_WARNINGS
 176                 auto context = _format_context->streams[*_video_stream]->codec;
 177 DCPOMATIC_ENABLE_WARNINGS
 178                 while (video_packet(context, temporal_reference, &packet)) {}
 179         }
 180
 181         for (auto i: _audio_streams) {
 182                 AVPacket packet;
 183                 av_init_packet (&packet);
 184                 packet.data = nullptr;
 185                 packet.size = 0;
 186 DCPOMATIC_DISABLE_WARNINGS
 187                 audio_packet (i->stream(_format_context)->codec, i, &packet);
 188 DCPOMATIC_ENABLE_WARNINGS
 189         }
 190
 191         if (_video_stream) {
 192                 /* This code taken from get_rotation() in ffmpeg:cmdutils.c */
 193                 auto stream = _format_context->streams[*_video_stream];
 194                 auto rotate_tag = av_dict_get (stream->metadata, "rotate", 0, 0);
 195                 uint8_t* displaymatrix = av_stream_get_side_data (stream, AV_PKT_DATA_DISPLAYMATRIX, 0);
 196                 _rotation = 0;
 197
 198                 if (rotate_tag && *rotate_tag->value && strcmp(rotate_tag->value, "0")) {
 199                         char *tail;
 200                         _rotation = av_strtod (rotate_tag->value, &tail);
 201                         if (*tail) {
 202                                 _rotation = 0;
 203                         }
 204                 }
 205
 206                 if (displaymatrix && !_rotation) {
 207                         _rotation = - av_display_rotation_get ((int32_t*) displaymatrix);
 208                 }
 209
 210                 _rotation = *_rotation - 360 * floor (*_rotation / 360 + 0.9 / 360);
 211         }
 212
 213         LOG_GENERAL("Temporal reference was %1", temporal_reference);
 214         if (temporal_reference.find("T2T3B2B3T2T3B2B3") != string::npos || temporal_reference.find("B2B3T2T3B2B3T2T3") != string::npos) {
 215                 /* The magical sequence (taken from mediainfo) suggests that 2:3 pull-down is in use */
 216                 _pulldown = true;
 217                 LOG_GENERAL_NC("Suggest that this may be 2:3 pull-down (soft telecine)");
 218         }
 219 }
 220
 221
 222 /** @param temporal_reference A string to which we should add two characters per frame;
 223  *  the first   is T or B depending on whether it's top- or bottom-field first,
 224  *  the second  is 3 or 2 depending on whether "repeat_pict" is true or not.
 225  *  @return true if some video was decoded, otherwise false.
 226  */
 227 bool
 228 FFmpegExaminer::video_packet (AVCodecContext* context, string& temporal_reference, AVPacket* packet)
 229 {
 230         DCPOMATIC_ASSERT (_video_stream);
 231
 232         if (_first_video && !_need_video_length && temporal_reference.size() >= (PULLDOWN_CHECK_FRAMES * 2)) {
 233                 return false;
 234         }
 235
 236         int frame_finished;
 237 DCPOMATIC_DISABLE_WARNINGS
 238         if (avcodec_decode_video2 (context, _frame, &frame_finished, packet) < 0 || !frame_finished) {
 239                 return false;
 240         }
 241 DCPOMATIC_ENABLE_WARNINGS
 242
 243         if (!_first_video) {
 244                 _first_video = frame_time (_format_context->streams[_video_stream.get()]);
 245         }
 246         if (_need_video_length) {
 247                 _video_length = frame_time (
 248                         _format_context->streams[_video_stream.get()]
 249                         ).get_value_or (ContentTime ()).frames_round (video_frame_rate().get ());
 250         }
 251         if (temporal_reference.size() < (PULLDOWN_CHECK_FRAMES * 2)) {
 252                 temporal_reference += (_frame->top_field_first ? "T" : "B");
 253                 temporal_reference += (_frame->repeat_pict ? "3" : "2");
 254         }
 255
 256         return true;
 257 }
 258
 259
 260 void
 261 FFmpegExaminer::audio_packet (AVCodecContext* context, shared_ptr<FFmpegAudioStream> stream, AVPacket* packet)
 262 {
 263         if (stream->first_audio) {
 264                 return;
 265         }
 266
 267         int frame_finished;
 268 DCPOMATIC_DISABLE_WARNINGS
 269         if (avcodec_decode_audio4 (context, _frame, &frame_finished, packet) >= 0 && frame_finished) {
 270 DCPOMATIC_ENABLE_WARNINGS
 271                 stream->first_audio = frame_time (stream->stream (_format_context));
 272         }
 273 }
 274
 275
 276 optional<ContentTime>
 277 FFmpegExaminer::frame_time (AVStream* s) const
 278 {
 279         optional<ContentTime> t;
 280
 281         int64_t const bet = _frame->best_effort_timestamp;
 282         if (bet != AV_NOPTS_VALUE) {
 283                 t = ContentTime::from_seconds (bet * av_q2d (s->time_base));
 284         }
 285
 286         return t;
 287 }
 288
 289
 290 optional<double>
 291 FFmpegExaminer::video_frame_rate () const
 292 {
 293         DCPOMATIC_ASSERT (_video_stream);
 294         return av_q2d(av_guess_frame_rate(_format_context, _format_context->streams[_video_stream.get()], 0));
 295 }
 296
 297
 298 dcp::Size
 299 FFmpegExaminer::video_size () const
 300 {
 301         return dcp::Size (video_codec_context()->width, video_codec_context()->height);
 302 }
 303
 304
 305 /** @return Length according to our content's header */
 306 Frame
 307 FFmpegExaminer::video_length () const
 308 {
 309         return max (Frame (1), _video_length);
 310 }
 311
 312
 313 optional<double>
 314 FFmpegExaminer::sample_aspect_ratio () const
 315 {
 316         DCPOMATIC_ASSERT (_video_stream);
 317         auto sar = av_guess_sample_aspect_ratio (_format_context, _format_context->streams[_video_stream.get()], 0);
 318         if (sar.num == 0) {
 319                 /* I assume this means that we don't know */
 320                 return {};
 321         }
 322         return double (sar.num) / sar.den;
 323 }
 324
 325
 326 string
 327 FFmpegExaminer::subtitle_stream_name (AVStream* s) const
 328 {
 329         auto n = stream_name (s);
 330
 331         if (n.empty()) {
 332                 n = _("unknown");
 333         }
 334
 335         return n;
 336 }
 337
 338
 339 string
 340 FFmpegExaminer::stream_name (AVStream* s) const
 341 {
 342         string n;
 343
 344         if (s->metadata) {
 345                 auto const lang = av_dict_get (s->metadata, "language", 0, 0);
 346                 if (lang) {
 347                         n = lang->value;
 348                 }
 349
 350                 auto const title = av_dict_get (s->metadata, "title", 0, 0);
 351                 if (title) {
 352                         if (!n.empty()) {
 353                                 n += " ";
 354                         }
 355                         n += title->value;
 356                 }
 357         }
 358
 359         return n;
 360 }
 361
 362
 363 optional<int>
 364 FFmpegExaminer::bits_per_pixel () const
 365 {
 366         if (video_codec_context()->pix_fmt == -1) {
 367                 return {};
 368         }
 369
 370         auto const d = av_pix_fmt_desc_get (video_codec_context()->pix_fmt);
 371         DCPOMATIC_ASSERT (d);
 372         return av_get_bits_per_pixel (d);
 373 }
 374
 375
 376 bool
 377 FFmpegExaminer::yuv () const
 378 {
 379         switch (video_codec_context()->pix_fmt) {
 380         case AV_PIX_FMT_YUV420P:
 381         case AV_PIX_FMT_YUYV422:
 382         case AV_PIX_FMT_YUV422P:
 383         case AV_PIX_FMT_YUV444P:
 384         case AV_PIX_FMT_YUV410P:
 385         case AV_PIX_FMT_YUV411P:
 386         case AV_PIX_FMT_YUVJ420P:
 387         case AV_PIX_FMT_YUVJ422P:
 388         case AV_PIX_FMT_YUVJ444P:
 389         case AV_PIX_FMT_UYVY422:
 390         case AV_PIX_FMT_UYYVYY411:
 391         case AV_PIX_FMT_NV12:
 392         case AV_PIX_FMT_NV21:
 393         case AV_PIX_FMT_YUV440P:
 394         case AV_PIX_FMT_YUVJ440P:
 395         case AV_PIX_FMT_YUVA420P:
 396         case AV_PIX_FMT_YUV420P16LE:
 397         case AV_PIX_FMT_YUV420P16BE:
 398         case AV_PIX_FMT_YUV422P16LE:
 399         case AV_PIX_FMT_YUV422P16BE:
 400         case AV_PIX_FMT_YUV444P16LE:
 401         case AV_PIX_FMT_YUV444P16BE:
 402         case AV_PIX_FMT_YUV420P9BE:
 403         case AV_PIX_FMT_YUV420P9LE:
 404         case AV_PIX_FMT_YUV420P10BE:
 405         case AV_PIX_FMT_YUV420P10LE:
 406         case AV_PIX_FMT_YUV422P10BE:
 407         case AV_PIX_FMT_YUV422P10LE:
 408         case AV_PIX_FMT_YUV444P9BE:
 409         case AV_PIX_FMT_YUV444P9LE:
 410         case AV_PIX_FMT_YUV444P10BE:
 411         case AV_PIX_FMT_YUV444P10LE:
 412         case AV_PIX_FMT_YUV422P9BE:
 413         case AV_PIX_FMT_YUV422P9LE:
 414         case AV_PIX_FMT_YUVA420P9BE:
 415         case AV_PIX_FMT_YUVA420P9LE:
 416         case AV_PIX_FMT_YUVA422P9BE:
 417         case AV_PIX_FMT_YUVA422P9LE:
 418         case AV_PIX_FMT_YUVA444P9BE:
 419         case AV_PIX_FMT_YUVA444P9LE:
 420         case AV_PIX_FMT_YUVA420P10BE:
 421         case AV_PIX_FMT_YUVA420P10LE:
 422         case AV_PIX_FMT_YUVA422P10BE:
 423         case AV_PIX_FMT_YUVA422P10LE:
 424         case AV_PIX_FMT_YUVA444P10BE:
 425         case AV_PIX_FMT_YUVA444P10LE:
 426         case AV_PIX_FMT_YUVA420P16BE:
 427         case AV_PIX_FMT_YUVA420P16LE:
 428         case AV_PIX_FMT_YUVA422P16BE:
 429         case AV_PIX_FMT_YUVA422P16LE:
 430         case AV_PIX_FMT_YUVA444P16BE:
 431         case AV_PIX_FMT_YUVA444P16LE:
 432         case AV_PIX_FMT_NV16:
 433         case AV_PIX_FMT_NV20LE:
 434         case AV_PIX_FMT_NV20BE:
 435         case AV_PIX_FMT_YVYU422:
 436         case AV_PIX_FMT_YUVA444P:
 437         case AV_PIX_FMT_YUVA422P:
 438         case AV_PIX_FMT_YUV420P12BE:
 439         case AV_PIX_FMT_YUV420P12LE:
 440         case AV_PIX_FMT_YUV420P14BE:
 441         case AV_PIX_FMT_YUV420P14LE:
 442         case AV_PIX_FMT_YUV422P12BE:
 443         case AV_PIX_FMT_YUV422P12LE:
 444         case AV_PIX_FMT_YUV422P14BE:
 445         case AV_PIX_FMT_YUV422P14LE:
 446         case AV_PIX_FMT_YUV444P12BE:
 447         case AV_PIX_FMT_YUV444P12LE:
 448         case AV_PIX_FMT_YUV444P14BE:
 449         case AV_PIX_FMT_YUV444P14LE:
 450         case AV_PIX_FMT_YUVJ411P:
 451                 return true;
 452         default:
 453                 return false;
 454         }
 455 }
 456
 457
 458 bool
 459 FFmpegExaminer::has_video () const
 460 {
 461         return static_cast<bool>(_video_stream);
 462 }
 463
 464
 465 VideoRange
 466 FFmpegExaminer::range () const
 467 {
 468         switch (color_range()) {
 469         case AVCOL_RANGE_MPEG:
 470         case AVCOL_RANGE_UNSPECIFIED:
 471                 return VideoRange::VIDEO;
 472         case AVCOL_RANGE_JPEG:
 473         default:
 474                 return VideoRange::FULL;
 475         }
 476 }