2 Copyright (C) 2013-2016 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 "ffmpeg_content.h"
24 #include "exceptions.h"
27 #include "ffmpeg_subtitle_stream.h"
28 #include "ffmpeg_audio_stream.h"
30 #include "compose.hpp"
31 #include <dcp/raw_convert.h>
33 #include <libavcodec/avcodec.h>
34 #include <libavformat/avformat.h>
35 #include <libswscale/swscale.h>
37 #include <boost/algorithm/string.hpp>
38 #include <boost/foreach.hpp>
47 using boost::shared_ptr;
48 using boost::optional;
49 using dcp::raw_convert;
51 boost::mutex FFmpeg::_mutex;
52 boost::weak_ptr<Log> FFmpeg::_ffmpeg_log;
54 FFmpeg::FFmpeg (boost::shared_ptr<const FFmpegContent> c)
57 , _avio_buffer_size (4096)
68 boost::mutex::scoped_lock lm (_mutex);
70 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
71 avcodec_close (_format_context->streams[i]->codec);
74 av_frame_free (&_frame);
75 avformat_close_input (&_format_context);
79 avio_read_wrapper (void* data, uint8_t* buffer, int amount)
81 return reinterpret_cast<FFmpeg*>(data)->avio_read (buffer, amount);
85 avio_seek_wrapper (void* data, int64_t offset, int whence)
87 return reinterpret_cast<FFmpeg*>(data)->avio_seek (offset, whence);
91 FFmpeg::ffmpeg_log_callback (void* ptr, int level, const char* fmt, va_list vl)
93 if (level > AV_LOG_WARNING) {
98 static int prefix = 0;
99 av_log_format_line (ptr, level, fmt, vl, line, sizeof (line), &prefix);
100 shared_ptr<Log> log = _ffmpeg_log.lock ();
103 boost::algorithm::trim (str);
104 log->log (String::compose ("FFmpeg: %1", str), LogEntry::TYPE_GENERAL);
111 FFmpeg::setup_general ()
113 /* This might not work too well in some cases of multiple FFmpeg decoders,
114 but it's probably good enough.
116 _ffmpeg_log = _ffmpeg_content->film()->log ();
117 av_log_set_callback (FFmpeg::ffmpeg_log_callback);
119 _file_group.set_paths (_ffmpeg_content->paths ());
120 _avio_buffer = static_cast<uint8_t*> (wrapped_av_malloc (_avio_buffer_size));
121 _avio_context = avio_alloc_context (_avio_buffer, _avio_buffer_size, 0, this, avio_read_wrapper, 0, avio_seek_wrapper);
122 _format_context = avformat_alloc_context ();
123 _format_context->pb = _avio_context;
125 AVDictionary* options = 0;
126 /* These durations are in microseconds, and represent how far into the content file
127 we will look for streams.
129 av_dict_set (&options, "analyzeduration", raw_convert<string> (5 * 60 * 1000000).c_str(), 0);
130 av_dict_set (&options, "probesize", raw_convert<string> (5 * 60 * 1000000).c_str(), 0);
132 int e = avformat_open_input (&_format_context, 0, 0, &options);
134 throw OpenFileError (_ffmpeg_content->path(0).string(), e, true);
137 if (avformat_find_stream_info (_format_context, 0) < 0) {
138 throw DecodeError (_("could not find stream information"));
141 /* Find video stream */
143 optional<int> video_stream_undefined_frame_rate;
145 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
146 AVStream* s = _format_context->streams[i];
147 if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
148 if (s->avg_frame_rate.num > 0 && s->avg_frame_rate.den > 0) {
149 /* This is definitely our video stream */
152 /* This is our video stream if we don't get a better offer */
153 video_stream_undefined_frame_rate = i;
158 /* Files from iTunes sometimes have two video streams, one with the avg_frame_rate.num and .den set
159 to zero. Only use such a stream if there is no alternative.
161 if (!_video_stream && video_stream_undefined_frame_rate) {
162 _video_stream = video_stream_undefined_frame_rate.get();
165 /* Hack: if the AVStreams have duplicate IDs, replace them with our
166 own. We use the IDs so that we can cope with VOBs, in which streams
167 move about in index but remain with the same ID in different
168 VOBs. However, some files have duplicate IDs, hence this hack.
171 bool duplicates = false;
172 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
173 for (uint32_t j = i + 1; j < _format_context->nb_streams; ++j) {
174 if (_format_context->streams[i]->id == _format_context->streams[j]->id) {
181 /* Put in our own IDs */
182 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
183 _format_context->streams[i]->id = i;
187 _frame = av_frame_alloc ();
189 throw DecodeError (N_("could not allocate frame"));
194 FFmpeg::setup_decoders ()
196 boost::mutex::scoped_lock lm (_mutex);
198 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
199 AVCodecContext* context = _format_context->streams[i]->codec;
201 AVCodec* codec = avcodec_find_decoder (context->codec_id);
204 AVDictionary* options = 0;
205 /* This option disables decoding of DCA frame footers in our patched version
206 of FFmpeg. I believe these footers are of no use to us, and they can cause
207 problems when FFmpeg fails to decode them (mantis #352).
209 av_dict_set (&options, "disable_footer", "1", 0);
210 /* This allows decoding of some DNxHR 444 and HQX files; see
211 https://trac.ffmpeg.org/ticket/5681
213 av_dict_set_int (&options, "strict", FF_COMPLIANCE_EXPERIMENTAL, 0);
215 if (avcodec_open2 (context, codec, &options) < 0) {
216 throw DecodeError (N_("could not open decoder"));
220 /* We are silently ignoring any failures to find suitable decoders here */
225 FFmpeg::video_codec_context () const
227 if (!_video_stream) {
231 return _format_context->streams[_video_stream.get()]->codec;
235 FFmpeg::subtitle_codec_context () const
237 if (!_ffmpeg_content->subtitle_stream ()) {
241 return _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec;
245 FFmpeg::avio_read (uint8_t* buffer, int const amount)
247 return _file_group.read (buffer, amount);
251 FFmpeg::avio_seek (int64_t const pos, int whence)
253 if (whence == AVSEEK_SIZE) {
254 return _file_group.length ();
257 return _file_group.seek (pos, whence);
261 FFmpeg::subtitle_period (AVSubtitle const & sub)
263 ContentTime const packet_time = ContentTime::from_seconds (static_cast<double> (sub.pts) / AV_TIME_BASE);
265 if (sub.end_display_time == static_cast<uint32_t> (-1)) {
266 /* End time is not known */
267 return FFmpegSubtitlePeriod (packet_time + ContentTime::from_seconds (sub.start_display_time / 1e3));
270 return FFmpegSubtitlePeriod (
271 packet_time + ContentTime::from_seconds (sub.start_display_time / 1e3),
272 packet_time + ContentTime::from_seconds (sub.end_display_time / 1e3)
277 FFmpeg::subtitle_id (AVSubtitle const & sub)
280 digester.add (sub.pts);
281 for (unsigned int i = 0; i < sub.num_rects; ++i) {
282 AVSubtitleRect* rect = sub.rects[i];
283 if (rect->type == SUBTITLE_BITMAP) {
284 digester.add (rect->x);
285 digester.add (rect->y);
286 digester.add (rect->w);
287 digester.add (rect->h);
288 #ifdef DCPOMATIC_HAVE_AVSUBTITLERECT_PICT
289 int const line = rect->pict.linesize[0];
290 for (int j = 0; j < rect->h; ++j) {
291 digester.add (rect->pict.data[0] + j * line, line);
294 int const line = rect->linesize[0];
295 for (int j = 0; j < rect->h; ++j) {
296 digester.add (rect->data[0] + j * line, line);
299 } else if (rect->type == SUBTITLE_TEXT) {
300 digester.add (string (rect->text));
301 } else if (rect->type == SUBTITLE_ASS) {
302 digester.add (string (rect->ass));
305 return digester.get ();
308 /** @return true if sub starts a new image subtitle */
310 FFmpeg::subtitle_starts_image (AVSubtitle const & sub)
315 for (unsigned int i = 0; i < sub.num_rects; ++i) {
316 switch (sub.rects[i]->type) {
317 case SUBTITLE_BITMAP:
329 /* We can't cope with mixed image/text in one AVSubtitle */
330 DCPOMATIC_ASSERT (!image || !text);
335 /** Compute the pts offset to use given a set of audio streams and some video details.
336 * Sometimes these parameters will have just been determined by an Examiner, sometimes
337 * they will have been retrieved from a piece of Content, hence the need for this method
341 FFmpeg::pts_offset (vector<shared_ptr<FFmpegAudioStream> > audio_streams, optional<ContentTime> first_video, double video_frame_rate) const
343 /* Audio and video frame PTS values may not start with 0. We want
344 to fiddle them so that:
346 1. One of them starts at time 0.
347 2. The first video PTS value ends up on a frame boundary.
349 Then we remove big initial gaps in PTS and we allow our
350 insertion of black frames to work.
353 audio_pts_to_use = audio_pts_from_ffmpeg + pts_offset;
354 video_pts_to_use = video_pts_from_ffmpeg + pts_offset;
357 /* First, make one of them start at 0 */
359 ContentTime po = ContentTime::min ();
362 po = - first_video.get ();
365 BOOST_FOREACH (shared_ptr<FFmpegAudioStream> i, audio_streams) {
366 if (i->first_audio) {
367 po = max (po, - i->first_audio.get ());
371 /* If the offset is positive we would be pushing things from a -ve PTS to be played.
372 I don't think we ever want to do that, as it seems things at -ve PTS are not meant
373 to be seen (use for alignment bars etc.); see mantis #418.
375 if (po > ContentTime ()) {
379 /* Now adjust so that the video pts starts on a frame */
381 ContentTime const fvc = first_video.get() + po;
382 po += fvc.ceil (video_frame_rate) - fvc;