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