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"
29 #include <dcp/warnings.h>
30 LIBDCP_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/channel_layout.h>
37 #include <libavutil/display.h>
38 #include <libavutil/eval.h>
40 LIBDCP_ENABLE_WARNINGS
47 using std::make_shared;
49 using std::shared_ptr;
52 using boost::optional;
53 using namespace dcpomatic;
56 /* This is how many frames from the start of any video that we will examine to see if we
57 * can spot soft 2:3 pull-down ("telecine").
59 static const int PULLDOWN_CHECK_FRAMES = 16;
62 /** @param job job that the examiner is operating in, or 0 */
63 FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Job> job)
66 , _need_video_length (false)
69 /* Find audio and subtitle streams */
71 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
72 auto s = _format_context->streams[i];
73 auto codec = _codec_context[i] ? _codec_context[i]->codec : nullptr;
74 if (s->codecpar->codec_type == AVMEDIA_TYPE_AUDIO && codec) {
76 DCPOMATIC_ASSERT (_format_context->duration != AV_NOPTS_VALUE);
77 DCPOMATIC_ASSERT (codec->name);
79 _audio_streams.push_back (
80 make_shared<FFmpegAudioStream>(
84 s->codecpar->sample_rate,
85 llrint ((double(_format_context->duration) / AV_TIME_BASE) * s->codecpar->sample_rate),
86 s->codecpar->ch_layout.nb_channels
90 } else if (s->codecpar->codec_type == AVMEDIA_TYPE_SUBTITLE) {
91 _subtitle_streams.push_back (make_shared<FFmpegSubtitleStream>(subtitle_stream_name (s), s->id));
96 /* See if the header has duration information in it */
97 _need_video_length = _format_context->duration == AV_NOPTS_VALUE;
98 if (!_need_video_length) {
99 _video_length = llrint ((double (_format_context->duration) / AV_TIME_BASE) * video_frame_rate().get());
103 if (job && _need_video_length) {
104 job->sub (_("Finding length"));
107 /* Run through until we find:
109 * - the first audio for each stream.
110 * - the top-field-first and repeat-first-frame values ("temporal_reference") for the first PULLDOWN_CHECK_FRAMES video frames.
113 int64_t const len = _file_group.length ();
114 /* A string which we build up to describe the top-field-first and repeat-first-frame values for the first few frames.
115 * It would be nicer to use something like vector<bool> here but we want to search the array for a pattern later,
116 * and a string seems a reasonably neat way to do that.
118 string temporal_reference;
120 auto packet = av_packet_alloc ();
121 DCPOMATIC_ASSERT (packet);
122 int r = av_read_frame (_format_context, packet);
124 av_packet_free (&packet);
130 job->set_progress (float (_format_context->pb->pos) / len);
132 job->set_progress_unknown ();
136 auto context = _codec_context[packet->stream_index];
138 if (_video_stream && packet->stream_index == _video_stream.get()) {
139 video_packet (context, temporal_reference, packet);
142 bool got_all_audio = true;
144 for (size_t i = 0; i < _audio_streams.size(); ++i) {
145 if (_audio_streams[i]->uses_index(_format_context, packet->stream_index)) {
146 audio_packet (context, _audio_streams[i], packet);
148 if (!_audio_streams[i]->first_audio) {
149 got_all_audio = false;
153 av_packet_free (&packet);
155 if (_first_video && got_all_audio && temporal_reference.size() >= (PULLDOWN_CHECK_FRAMES * 2)) {
162 auto context = _codec_context[_video_stream.get()];
163 while (video_packet(context, temporal_reference, nullptr)) {}
166 for (auto i: _audio_streams) {
167 auto context = _codec_context[i->index(_format_context)];
168 audio_packet(context, i, nullptr);
172 /* This code taken from get_rotation() in ffmpeg:cmdutils.c */
173 auto stream = _format_context->streams[*_video_stream];
174 auto rotate_tag = av_dict_get (stream->metadata, "rotate", 0, 0);
175 uint8_t* displaymatrix = av_stream_get_side_data (stream, AV_PKT_DATA_DISPLAYMATRIX, 0);
178 if (rotate_tag && *rotate_tag->value && strcmp(rotate_tag->value, "0")) {
180 _rotation = av_strtod (rotate_tag->value, &tail);
186 if (displaymatrix && !_rotation) {
187 _rotation = - av_display_rotation_get ((int32_t*) displaymatrix);
190 _rotation = *_rotation - 360 * floor (*_rotation / 360 + 0.9 / 360);
193 LOG_GENERAL("Temporal reference was %1", temporal_reference);
194 if (temporal_reference.find("T2T3B2B3T2T3B2B3") != string::npos || temporal_reference.find("B2B3T2T3B2B3T2T3") != string::npos) {
195 /* The magical sequence (taken from mediainfo) suggests that 2:3 pull-down is in use */
197 LOG_GENERAL_NC("Suggest that this may be 2:3 pull-down (soft telecine)");
202 /** @param temporal_reference A string to which we should add two characters per frame;
203 * the first is T or B depending on whether it's top- or bottom-field first,
204 * the second is 3 or 2 depending on whether "repeat_pict" is true or not.
205 * @return true if some video was decoded, otherwise false.
208 FFmpegExaminer::video_packet (AVCodecContext* context, string& temporal_reference, AVPacket* packet)
210 DCPOMATIC_ASSERT (_video_stream);
212 if (_first_video && !_need_video_length && temporal_reference.size() >= (PULLDOWN_CHECK_FRAMES * 2)) {
216 bool pending = false;
218 int r = avcodec_send_packet (context, packet);
220 LOG_WARNING("avcodec_send_packet returned %1 for a video packet", r);
223 /* EAGAIN means we should call avcodec_receive_frame and then re-send the same packet */
224 pending = r == AVERROR(EAGAIN);
226 r = avcodec_receive_frame (context, _video_frame);
227 if (r == AVERROR(EAGAIN)) {
228 /* More input is required */
230 } else if (r == AVERROR_EOF) {
231 /* No more output is coming */
237 _first_video = frame_time (_video_frame, _format_context->streams[_video_stream.get()]);
239 if (_need_video_length) {
240 _video_length = frame_time (
242 _format_context->streams[_video_stream.get()]
243 ).get_value_or (ContentTime ()).frames_round (video_frame_rate().get ());
245 if (temporal_reference.size() < (PULLDOWN_CHECK_FRAMES * 2)) {
246 temporal_reference += (_video_frame->top_field_first ? "T" : "B");
247 temporal_reference += (_video_frame->repeat_pict ? "3" : "2");
255 FFmpegExaminer::audio_packet (AVCodecContext* context, shared_ptr<FFmpegAudioStream> stream, AVPacket* packet)
257 if (stream->first_audio) {
261 int r = avcodec_send_packet (context, packet);
263 LOG_WARNING("avcodec_send_packet returned %1 for an audio packet", r);
267 auto frame = audio_frame (stream);
269 if (avcodec_receive_frame (context, frame) < 0) {
273 stream->first_audio = frame_time (frame, stream->stream(_format_context));
277 optional<ContentTime>
278 FFmpegExaminer::frame_time (AVFrame* frame, AVStream* stream) const
280 optional<ContentTime> t;
282 int64_t const bet = frame->best_effort_timestamp;
283 if (bet != AV_NOPTS_VALUE) {
284 t = ContentTime::from_seconds (bet * av_q2d(stream->time_base));
292 FFmpegExaminer::video_frame_rate () const
294 DCPOMATIC_ASSERT (_video_stream);
295 return av_q2d(av_guess_frame_rate(_format_context, _format_context->streams[_video_stream.get()], 0));
300 FFmpegExaminer::video_size () const
302 return dcp::Size (video_codec_context()->width, video_codec_context()->height);
306 /** @return Length according to our content's header */
308 FFmpegExaminer::video_length () const
310 return max (Frame (1), _video_length);
315 FFmpegExaminer::sample_aspect_ratio () const
317 DCPOMATIC_ASSERT (_video_stream);
318 auto sar = av_guess_sample_aspect_ratio (_format_context, _format_context->streams[_video_stream.get()], 0);
320 /* I assume this means that we don't know */
323 return double (sar.num) / sar.den;
328 FFmpegExaminer::subtitle_stream_name (AVStream* s) const
330 auto n = stream_name (s);
341 FFmpegExaminer::stream_name (AVStream* s) const
346 auto const lang = av_dict_get (s->metadata, "language", 0, 0);
351 auto const title = av_dict_get (s->metadata, "title", 0, 0);
365 FFmpegExaminer::bits_per_pixel () const
367 if (video_codec_context()->pix_fmt == -1) {
371 auto const d = av_pix_fmt_desc_get (video_codec_context()->pix_fmt);
372 DCPOMATIC_ASSERT (d);
373 return av_get_bits_per_pixel (d);
378 FFmpegExaminer::yuv () const
380 switch (video_codec_context()->pix_fmt) {
381 case AV_PIX_FMT_YUV420P:
382 case AV_PIX_FMT_YUYV422:
383 case AV_PIX_FMT_YUV422P:
384 case AV_PIX_FMT_YUV444P:
385 case AV_PIX_FMT_YUV410P:
386 case AV_PIX_FMT_YUV411P:
387 case AV_PIX_FMT_YUVJ420P:
388 case AV_PIX_FMT_YUVJ422P:
389 case AV_PIX_FMT_YUVJ444P:
390 case AV_PIX_FMT_UYVY422:
391 case AV_PIX_FMT_UYYVYY411:
392 case AV_PIX_FMT_NV12:
393 case AV_PIX_FMT_NV21:
394 case AV_PIX_FMT_YUV440P:
395 case AV_PIX_FMT_YUVJ440P:
396 case AV_PIX_FMT_YUVA420P:
397 case AV_PIX_FMT_YUV420P16LE:
398 case AV_PIX_FMT_YUV420P16BE:
399 case AV_PIX_FMT_YUV422P16LE:
400 case AV_PIX_FMT_YUV422P16BE:
401 case AV_PIX_FMT_YUV444P16LE:
402 case AV_PIX_FMT_YUV444P16BE:
403 case AV_PIX_FMT_YUV420P9BE:
404 case AV_PIX_FMT_YUV420P9LE:
405 case AV_PIX_FMT_YUV420P10BE:
406 case AV_PIX_FMT_YUV420P10LE:
407 case AV_PIX_FMT_YUV422P10BE:
408 case AV_PIX_FMT_YUV422P10LE:
409 case AV_PIX_FMT_YUV444P9BE:
410 case AV_PIX_FMT_YUV444P9LE:
411 case AV_PIX_FMT_YUV444P10BE:
412 case AV_PIX_FMT_YUV444P10LE:
413 case AV_PIX_FMT_YUV422P9BE:
414 case AV_PIX_FMT_YUV422P9LE:
415 case AV_PIX_FMT_YUVA420P9BE:
416 case AV_PIX_FMT_YUVA420P9LE:
417 case AV_PIX_FMT_YUVA422P9BE:
418 case AV_PIX_FMT_YUVA422P9LE:
419 case AV_PIX_FMT_YUVA444P9BE:
420 case AV_PIX_FMT_YUVA444P9LE:
421 case AV_PIX_FMT_YUVA420P10BE:
422 case AV_PIX_FMT_YUVA420P10LE:
423 case AV_PIX_FMT_YUVA422P10BE:
424 case AV_PIX_FMT_YUVA422P10LE:
425 case AV_PIX_FMT_YUVA444P10BE:
426 case AV_PIX_FMT_YUVA444P10LE:
427 case AV_PIX_FMT_YUVA420P16BE:
428 case AV_PIX_FMT_YUVA420P16LE:
429 case AV_PIX_FMT_YUVA422P16BE:
430 case AV_PIX_FMT_YUVA422P16LE:
431 case AV_PIX_FMT_YUVA444P16BE:
432 case AV_PIX_FMT_YUVA444P16LE:
433 case AV_PIX_FMT_NV16:
434 case AV_PIX_FMT_NV20LE:
435 case AV_PIX_FMT_NV20BE:
436 case AV_PIX_FMT_YVYU422:
437 case AV_PIX_FMT_YUVA444P:
438 case AV_PIX_FMT_YUVA422P:
439 case AV_PIX_FMT_YUV420P12BE:
440 case AV_PIX_FMT_YUV420P12LE:
441 case AV_PIX_FMT_YUV420P14BE:
442 case AV_PIX_FMT_YUV420P14LE:
443 case AV_PIX_FMT_YUV422P12BE:
444 case AV_PIX_FMT_YUV422P12LE:
445 case AV_PIX_FMT_YUV422P14BE:
446 case AV_PIX_FMT_YUV422P14LE:
447 case AV_PIX_FMT_YUV444P12BE:
448 case AV_PIX_FMT_YUV444P12LE:
449 case AV_PIX_FMT_YUV444P14BE:
450 case AV_PIX_FMT_YUV444P14LE:
451 case AV_PIX_FMT_YUVJ411P:
460 FFmpegExaminer::has_video () const
462 return static_cast<bool>(_video_stream);
467 FFmpegExaminer::range () const
469 switch (color_range()) {
470 case AVCOL_RANGE_MPEG:
471 case AVCOL_RANGE_UNSPECIFIED:
472 return VideoRange::VIDEO;
473 case AVCOL_RANGE_JPEG:
475 return VideoRange::FULL;
481 FFmpegExaminer::pixel_quanta () const
483 auto const desc = av_pix_fmt_desc_get(video_codec_context()->pix_fmt);
484 DCPOMATIC_ASSERT (desc);
485 return { 1 << desc->log2_chroma_w, 1 << desc->log2_chroma_h };