2 Copyright (C) 2013-2021 Carl Hetherington <cth@carlh.net>
4 This file is part of DCP-o-matic.
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.
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.
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/>.
22 #include "dcpomatic_log.h"
23 #include "ffmpeg_examiner.h"
24 #include "ffmpeg_content.h"
26 #include "ffmpeg_audio_stream.h"
27 #include "ffmpeg_subtitle_stream.h"
30 DCPOMATIC_DISABLE_WARNINGS
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>
39 DCPOMATIC_ENABLE_WARNINGS
46 using std::make_shared;
48 using std::shared_ptr;
51 using boost::optional;
52 using namespace dcpomatic;
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").
58 static const int PULLDOWN_CHECK_FRAMES = 16;
61 /** @param job job that the examiner is operating in, or 0 */
62 FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Job> job)
65 , _need_video_length (false)
68 /* Find audio and subtitle streams */
70 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
71 auto s = _format_context->streams[i];
72 if (s->codecpar->codec_type == AVMEDIA_TYPE_AUDIO) {
74 /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
75 so bodge it here. No idea why we should have to do this.
78 if (s->codecpar->channel_layout == 0) {
79 s->codecpar->channel_layout = av_get_default_channel_layout (s->codecpar->channels);
82 auto codec = _codec_context[i]->codec;
84 DCPOMATIC_ASSERT (_format_context->duration != AV_NOPTS_VALUE);
85 DCPOMATIC_ASSERT (codec);
86 DCPOMATIC_ASSERT (codec->name);
88 _audio_streams.push_back (
89 make_shared<FFmpegAudioStream>(
93 s->codecpar->sample_rate,
94 llrint ((double(_format_context->duration) / AV_TIME_BASE) * s->codecpar->sample_rate),
99 } else if (s->codecpar->codec_type == AVMEDIA_TYPE_SUBTITLE) {
100 _subtitle_streams.push_back (make_shared<FFmpegSubtitleStream>(subtitle_stream_name (s), s->id));
105 /* See if the header has duration information in it */
106 _need_video_length = _format_context->duration == AV_NOPTS_VALUE;
107 if (!_need_video_length) {
108 _video_length = llrint ((double (_format_context->duration) / AV_TIME_BASE) * video_frame_rate().get());
112 if (job && _need_video_length) {
113 job->sub (_("Finding length"));
116 /* Run through until we find:
118 * - the first audio for each stream.
119 * - the top-field-first and repeat-first-frame values ("temporal_reference") for the first PULLDOWN_CHECK_FRAMES video frames.
122 int64_t const len = _file_group.length ();
123 /* A string which we build up to describe the top-field-first and repeat-first-frame values for the first few frames.
124 * It would be nicer to use something like vector<bool> here but we want to search the array for a pattern later,
125 * and a string seems a reasonably neat way to do that.
127 string temporal_reference;
129 auto packet = av_packet_alloc ();
130 DCPOMATIC_ASSERT (packet);
131 int r = av_read_frame (_format_context, packet);
133 av_packet_free (&packet);
139 job->set_progress (float (_format_context->pb->pos) / len);
141 job->set_progress_unknown ();
145 auto context = _codec_context[packet->stream_index];
147 if (_video_stream && packet->stream_index == _video_stream.get()) {
148 video_packet (context, temporal_reference, packet);
151 bool got_all_audio = true;
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);
157 if (!_audio_streams[i]->first_audio) {
158 got_all_audio = false;
162 av_packet_free (&packet);
164 if (_first_video && got_all_audio && temporal_reference.size() >= (PULLDOWN_CHECK_FRAMES * 2)) {
171 auto context = _codec_context[_video_stream.get()];
172 while (video_packet(context, temporal_reference, nullptr)) {}
175 for (auto i: _audio_streams) {
176 auto context = _codec_context[i->index(_format_context)];
177 audio_packet(context, i, nullptr);
181 /* This code taken from get_rotation() in ffmpeg:cmdutils.c */
182 auto stream = _format_context->streams[*_video_stream];
183 auto rotate_tag = av_dict_get (stream->metadata, "rotate", 0, 0);
184 uint8_t* displaymatrix = av_stream_get_side_data (stream, AV_PKT_DATA_DISPLAYMATRIX, 0);
187 if (rotate_tag && *rotate_tag->value && strcmp(rotate_tag->value, "0")) {
189 _rotation = av_strtod (rotate_tag->value, &tail);
195 if (displaymatrix && !_rotation) {
196 _rotation = - av_display_rotation_get ((int32_t*) displaymatrix);
199 _rotation = *_rotation - 360 * floor (*_rotation / 360 + 0.9 / 360);
202 LOG_GENERAL("Temporal reference was %1", temporal_reference);
203 if (temporal_reference.find("T2T3B2B3T2T3B2B3") != string::npos || temporal_reference.find("B2B3T2T3B2B3T2T3") != string::npos) {
204 /* The magical sequence (taken from mediainfo) suggests that 2:3 pull-down is in use */
206 LOG_GENERAL_NC("Suggest that this may be 2:3 pull-down (soft telecine)");
211 /** @param temporal_reference A string to which we should add two characters per frame;
212 * the first is T or B depending on whether it's top- or bottom-field first,
213 * the second is 3 or 2 depending on whether "repeat_pict" is true or not.
214 * @return true if some video was decoded, otherwise false.
217 FFmpegExaminer::video_packet (AVCodecContext* context, string& temporal_reference, AVPacket* packet)
219 DCPOMATIC_ASSERT (_video_stream);
221 if (_first_video && !_need_video_length && temporal_reference.size() >= (PULLDOWN_CHECK_FRAMES * 2)) {
225 int r = avcodec_send_packet (context, packet);
226 if (r < 0 && !(r == AVERROR_EOF && !packet)) {
227 /* We could cope with AVERROR(EAGAIN) and re-send the packet but I think it should never happen.
228 * AVERROR_EOF can happen during flush if we've already sent a flush packet.
230 throw DecodeError (N_("avcodec_send_packet"), N_("FFmpegExaminer::video_packet"), r);
233 r = avcodec_receive_frame (context, _frame);
234 if (r == AVERROR(EAGAIN)) {
235 /* More input is required */
237 } else if (r == AVERROR_EOF) {
238 /* No more output is coming */
243 _first_video = frame_time (_format_context->streams[_video_stream.get()]);
245 if (_need_video_length) {
246 _video_length = frame_time (
247 _format_context->streams[_video_stream.get()]
248 ).get_value_or (ContentTime ()).frames_round (video_frame_rate().get ());
250 if (temporal_reference.size() < (PULLDOWN_CHECK_FRAMES * 2)) {
251 temporal_reference += (_frame->top_field_first ? "T" : "B");
252 temporal_reference += (_frame->repeat_pict ? "3" : "2");
260 FFmpegExaminer::audio_packet (AVCodecContext* context, shared_ptr<FFmpegAudioStream> stream, AVPacket* packet)
262 if (stream->first_audio) {
266 int r = avcodec_send_packet (context, packet);
267 if (r < 0 && !(r == AVERROR_EOF && !packet) && r != AVERROR(EAGAIN)) {
268 /* We could cope with AVERROR(EAGAIN) and re-send the packet but I think it should never happen.
269 * AVERROR_EOF can happen during flush if we've already sent a flush packet.
270 * EAGAIN means we need to do avcodec_receive_frame, so just carry on and do that.
272 throw DecodeError (N_("avcodec_send_packet"), N_("FFmpegExaminer::audio_packet"), r);
275 if (avcodec_receive_frame (context, _frame) < 0) {
279 stream->first_audio = frame_time (stream->stream(_format_context));
283 optional<ContentTime>
284 FFmpegExaminer::frame_time (AVStream* s) const
286 optional<ContentTime> t;
288 int64_t const bet = _frame->best_effort_timestamp;
289 if (bet != AV_NOPTS_VALUE) {
290 t = ContentTime::from_seconds (bet * av_q2d (s->time_base));
298 FFmpegExaminer::video_frame_rate () const
300 DCPOMATIC_ASSERT (_video_stream);
301 return av_q2d(av_guess_frame_rate(_format_context, _format_context->streams[_video_stream.get()], 0));
306 FFmpegExaminer::video_size () const
308 return dcp::Size (video_codec_context()->width, video_codec_context()->height);
312 /** @return Length according to our content's header */
314 FFmpegExaminer::video_length () const
316 return max (Frame (1), _video_length);
321 FFmpegExaminer::sample_aspect_ratio () const
323 DCPOMATIC_ASSERT (_video_stream);
324 auto sar = av_guess_sample_aspect_ratio (_format_context, _format_context->streams[_video_stream.get()], 0);
326 /* I assume this means that we don't know */
329 return double (sar.num) / sar.den;
334 FFmpegExaminer::subtitle_stream_name (AVStream* s) const
336 auto n = stream_name (s);
347 FFmpegExaminer::stream_name (AVStream* s) const
352 auto const lang = av_dict_get (s->metadata, "language", 0, 0);
357 auto const title = av_dict_get (s->metadata, "title", 0, 0);
371 FFmpegExaminer::bits_per_pixel () const
373 if (video_codec_context()->pix_fmt == -1) {
377 auto const d = av_pix_fmt_desc_get (video_codec_context()->pix_fmt);
378 DCPOMATIC_ASSERT (d);
379 return av_get_bits_per_pixel (d);
384 FFmpegExaminer::yuv () const
386 switch (video_codec_context()->pix_fmt) {
387 case AV_PIX_FMT_YUV420P:
388 case AV_PIX_FMT_YUYV422:
389 case AV_PIX_FMT_YUV422P:
390 case AV_PIX_FMT_YUV444P:
391 case AV_PIX_FMT_YUV410P:
392 case AV_PIX_FMT_YUV411P:
393 case AV_PIX_FMT_YUVJ420P:
394 case AV_PIX_FMT_YUVJ422P:
395 case AV_PIX_FMT_YUVJ444P:
396 case AV_PIX_FMT_UYVY422:
397 case AV_PIX_FMT_UYYVYY411:
398 case AV_PIX_FMT_NV12:
399 case AV_PIX_FMT_NV21:
400 case AV_PIX_FMT_YUV440P:
401 case AV_PIX_FMT_YUVJ440P:
402 case AV_PIX_FMT_YUVA420P:
403 case AV_PIX_FMT_YUV420P16LE:
404 case AV_PIX_FMT_YUV420P16BE:
405 case AV_PIX_FMT_YUV422P16LE:
406 case AV_PIX_FMT_YUV422P16BE:
407 case AV_PIX_FMT_YUV444P16LE:
408 case AV_PIX_FMT_YUV444P16BE:
409 case AV_PIX_FMT_YUV420P9BE:
410 case AV_PIX_FMT_YUV420P9LE:
411 case AV_PIX_FMT_YUV420P10BE:
412 case AV_PIX_FMT_YUV420P10LE:
413 case AV_PIX_FMT_YUV422P10BE:
414 case AV_PIX_FMT_YUV422P10LE:
415 case AV_PIX_FMT_YUV444P9BE:
416 case AV_PIX_FMT_YUV444P9LE:
417 case AV_PIX_FMT_YUV444P10BE:
418 case AV_PIX_FMT_YUV444P10LE:
419 case AV_PIX_FMT_YUV422P9BE:
420 case AV_PIX_FMT_YUV422P9LE:
421 case AV_PIX_FMT_YUVA420P9BE:
422 case AV_PIX_FMT_YUVA420P9LE:
423 case AV_PIX_FMT_YUVA422P9BE:
424 case AV_PIX_FMT_YUVA422P9LE:
425 case AV_PIX_FMT_YUVA444P9BE:
426 case AV_PIX_FMT_YUVA444P9LE:
427 case AV_PIX_FMT_YUVA420P10BE:
428 case AV_PIX_FMT_YUVA420P10LE:
429 case AV_PIX_FMT_YUVA422P10BE:
430 case AV_PIX_FMT_YUVA422P10LE:
431 case AV_PIX_FMT_YUVA444P10BE:
432 case AV_PIX_FMT_YUVA444P10LE:
433 case AV_PIX_FMT_YUVA420P16BE:
434 case AV_PIX_FMT_YUVA420P16LE:
435 case AV_PIX_FMT_YUVA422P16BE:
436 case AV_PIX_FMT_YUVA422P16LE:
437 case AV_PIX_FMT_YUVA444P16BE:
438 case AV_PIX_FMT_YUVA444P16LE:
439 case AV_PIX_FMT_NV16:
440 case AV_PIX_FMT_NV20LE:
441 case AV_PIX_FMT_NV20BE:
442 case AV_PIX_FMT_YVYU422:
443 case AV_PIX_FMT_YUVA444P:
444 case AV_PIX_FMT_YUVA422P:
445 case AV_PIX_FMT_YUV420P12BE:
446 case AV_PIX_FMT_YUV420P12LE:
447 case AV_PIX_FMT_YUV420P14BE:
448 case AV_PIX_FMT_YUV420P14LE:
449 case AV_PIX_FMT_YUV422P12BE:
450 case AV_PIX_FMT_YUV422P12LE:
451 case AV_PIX_FMT_YUV422P14BE:
452 case AV_PIX_FMT_YUV422P14LE:
453 case AV_PIX_FMT_YUV444P12BE:
454 case AV_PIX_FMT_YUV444P12LE:
455 case AV_PIX_FMT_YUV444P14BE:
456 case AV_PIX_FMT_YUV444P14LE:
457 case AV_PIX_FMT_YUVJ411P:
466 FFmpegExaminer::has_video () const
468 return static_cast<bool>(_video_stream);
473 FFmpegExaminer::range () const
475 switch (color_range()) {
476 case AVCOL_RANGE_MPEG:
477 case AVCOL_RANGE_UNSPECIFIED:
478 return VideoRange::VIDEO;
479 case AVCOL_RANGE_JPEG:
481 return VideoRange::FULL;