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"
26 #include "raw_convert.h"
28 #include "ffmpeg_subtitle_stream.h"
29 #include "ffmpeg_audio_stream.h"
31 #include "compose.hpp"
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;
50 boost::mutex FFmpeg::_mutex;
51 boost::weak_ptr<Log> FFmpeg::_ffmpeg_log;
53 FFmpeg::FFmpeg (boost::shared_ptr<const FFmpegContent> c)
56 , _avio_buffer_size (4096)
67 boost::mutex::scoped_lock lm (_mutex);
69 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
70 avcodec_close (_format_context->streams[i]->codec);
73 av_frame_free (&_frame);
74 avformat_close_input (&_format_context);
78 avio_read_wrapper (void* data, uint8_t* buffer, int amount)
80 return reinterpret_cast<FFmpeg*>(data)->avio_read (buffer, amount);
84 avio_seek_wrapper (void* data, int64_t offset, int whence)
86 return reinterpret_cast<FFmpeg*>(data)->avio_seek (offset, whence);
90 FFmpeg::ffmpeg_log_callback (void* ptr, int level, const char* fmt, va_list vl)
92 if (level > AV_LOG_WARNING) {
97 static int prefix = 0;
98 av_log_format_line (ptr, level, fmt, vl, line, sizeof (line), &prefix);
99 shared_ptr<Log> log = _ffmpeg_log.lock ();
102 boost::algorithm::trim (str);
103 log->log (String::compose ("FFmpeg: %1", str), LogEntry::TYPE_GENERAL);
110 FFmpeg::setup_general ()
114 /* This might not work too well in some cases of multiple FFmpeg decoders,
115 but it's probably good enough.
117 _ffmpeg_log = _ffmpeg_content->film()->log ();
118 av_log_set_callback (FFmpeg::ffmpeg_log_callback);
120 _file_group.set_paths (_ffmpeg_content->paths ());
121 _avio_buffer = static_cast<uint8_t*> (wrapped_av_malloc (_avio_buffer_size));
122 _avio_context = avio_alloc_context (_avio_buffer, _avio_buffer_size, 0, this, avio_read_wrapper, 0, avio_seek_wrapper);
123 _format_context = avformat_alloc_context ();
124 _format_context->pb = _avio_context;
126 AVDictionary* options = 0;
127 /* These durations are in microseconds, and represent how far into the content file
128 we will look for streams.
130 av_dict_set (&options, "analyzeduration", raw_convert<string> (5 * 60 * 1000000).c_str(), 0);
131 av_dict_set (&options, "probesize", raw_convert<string> (5 * 60 * 1000000).c_str(), 0);
133 if (avformat_open_input (&_format_context, 0, 0, &options) < 0) {
134 throw OpenFileError (_ffmpeg_content->path(0).string ());
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 /* This option disables decoding of DCA frame footers in our patched version
205 of FFmpeg. I believe these footers are of no use to us, and they can cause
206 problems when FFmpeg fails to decode them (mantis #352).
208 AVDictionary* options = 0;
209 av_dict_set (&options, "disable_footer", "1", 0);
211 if (avcodec_open2 (context, codec, &options) < 0) {
212 throw DecodeError (N_("could not open decoder"));
216 /* We are silently ignoring any failures to find suitable decoders here */
221 FFmpeg::video_codec_context () const
223 DCPOMATIC_ASSERT (_video_stream);
224 return _format_context->streams[_video_stream.get()]->codec;
228 FFmpeg::subtitle_codec_context () const
230 if (!_ffmpeg_content->subtitle_stream ()) {
234 return _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec;
238 FFmpeg::avio_read (uint8_t* buffer, int const amount)
240 return _file_group.read (buffer, amount);
244 FFmpeg::avio_seek (int64_t const pos, int whence)
246 if (whence == AVSEEK_SIZE) {
247 return _file_group.length ();
250 return _file_group.seek (pos, whence);
254 FFmpeg::subtitle_period (AVSubtitle const & sub)
256 ContentTime const packet_time = ContentTime::from_seconds (static_cast<double> (sub.pts) / AV_TIME_BASE);
258 if (sub.end_display_time == static_cast<uint32_t> (-1)) {
259 /* End time is not known */
260 return FFmpegSubtitlePeriod (packet_time + ContentTime::from_seconds (sub.start_display_time / 1e3));
263 return FFmpegSubtitlePeriod (
264 packet_time + ContentTime::from_seconds (sub.start_display_time / 1e3),
265 packet_time + ContentTime::from_seconds (sub.end_display_time / 1e3)
270 FFmpeg::subtitle_id (AVSubtitle const & sub)
273 digester.add (sub.pts);
274 for (unsigned int i = 0; i < sub.num_rects; ++i) {
275 AVSubtitleRect* rect = sub.rects[i];
276 digester.add (rect->x);
277 digester.add (rect->y);
278 digester.add (rect->w);
279 digester.add (rect->h);
280 #ifdef DCPOMATIC_HAVE_AVSUBTITLERECT_PICT
281 int const line = rect->pict.linesize[0];
282 for (int j = 0; j < rect->h; ++j) {
283 digester.add (rect->pict.data[0] + j * line, line);
286 int const line = rect->linesize[0];
287 for (int j = 0; j < rect->h; ++j) {
288 digester.add (rect->data[0] + j * line, line);
292 return digester.get ();
295 /** @return true if sub starts a new image subtitle */
297 FFmpeg::subtitle_starts_image (AVSubtitle const & sub)
302 for (unsigned int i = 0; i < sub.num_rects; ++i) {
303 switch (sub.rects[i]->type) {
304 case SUBTITLE_BITMAP:
316 /* We can't cope with mixed image/text in one AVSubtitle */
317 DCPOMATIC_ASSERT (!image || !text);
322 /** Compute the pts offset to use given a set of audio streams and some video details.
323 * Sometimes these parameters will have just been determined by an Examiner, sometimes
324 * they will have been retrieved from a piece of Content, hence the need for this method
328 FFmpeg::pts_offset (vector<shared_ptr<FFmpegAudioStream> > audio_streams, optional<ContentTime> first_video, double video_frame_rate) const
330 /* Audio and video frame PTS values may not start with 0. We want
331 to fiddle them so that:
333 1. One of them starts at time 0.
334 2. The first video PTS value ends up on a frame boundary.
336 Then we remove big initial gaps in PTS and we allow our
337 insertion of black frames to work.
340 audio_pts_to_use = audio_pts_from_ffmpeg + pts_offset;
341 video_pts_to_use = video_pts_from_ffmpeg + pts_offset;
344 /* First, make one of them start at 0 */
346 ContentTime po = ContentTime::min ();
349 po = - first_video.get ();
352 BOOST_FOREACH (shared_ptr<FFmpegAudioStream> i, audio_streams) {
353 if (i->first_audio) {
354 po = max (po, - i->first_audio.get ());
358 /* If the offset is positive we would be pushing things from a -ve PTS to be played.
359 I don't think we ever want to do that, as it seems things at -ve PTS are not meant
360 to be seen (use for alignment bars etc.); see mantis #418.
362 if (po > ContentTime ()) {
366 /* Now adjust so that the video pts starts on a frame */
368 ContentTime const fvc = first_video.get() + po;
369 po += fvc.round_up (video_frame_rate) - fvc;