2 Copyright (C) 2013-2016 Carl Hetherington <cth@carlh.net>
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation; either version 2 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 #include "ffmpeg_content.h"
23 #include "exceptions.h"
25 #include "raw_convert.h"
27 #include "ffmpeg_subtitle_stream.h"
28 #include "ffmpeg_audio_stream.h"
29 #include "md5_digester.h"
30 #include "compose.hpp"
32 #include <libavcodec/avcodec.h>
33 #include <libavformat/avformat.h>
34 #include <libswscale/swscale.h>
36 #include <boost/algorithm/string.hpp>
37 #include <boost/foreach.hpp>
46 using boost::shared_ptr;
47 using boost::optional;
49 boost::mutex FFmpeg::_mutex;
50 boost::weak_ptr<Log> FFmpeg::_ffmpeg_log;
52 FFmpeg::FFmpeg (boost::shared_ptr<const FFmpegContent> c)
55 , _avio_buffer_size (4096)
66 boost::mutex::scoped_lock lm (_mutex);
68 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
69 avcodec_close (_format_context->streams[i]->codec);
72 av_frame_free (&_frame);
73 avformat_close_input (&_format_context);
77 avio_read_wrapper (void* data, uint8_t* buffer, int amount)
79 return reinterpret_cast<FFmpeg*>(data)->avio_read (buffer, amount);
83 avio_seek_wrapper (void* data, int64_t offset, int whence)
85 return reinterpret_cast<FFmpeg*>(data)->avio_seek (offset, whence);
89 FFmpeg::ffmpeg_log_callback (void* ptr, int level, const char* fmt, va_list vl)
91 if (level > AV_LOG_WARNING) {
96 static int prefix = 0;
97 av_log_format_line (ptr, level, fmt, vl, line, sizeof (line), &prefix);
98 shared_ptr<Log> log = _ffmpeg_log.lock ();
101 boost::algorithm::trim (str);
102 log->log (String::compose ("FFmpeg: %1", str), LogEntry::TYPE_GENERAL);
109 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 if (avformat_open_input (&_format_context, 0, 0, &options) < 0) {
133 throw OpenFileError (_ffmpeg_content->path(0).string ());
136 if (avformat_find_stream_info (_format_context, 0) < 0) {
137 throw DecodeError (_("could not find stream information"));
140 /* Find video stream */
142 optional<int> video_stream_undefined_frame_rate;
144 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
145 AVStream* s = _format_context->streams[i];
146 if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
147 if (s->avg_frame_rate.num > 0 && s->avg_frame_rate.den > 0) {
148 /* This is definitely our video stream */
151 /* This is our video stream if we don't get a better offer */
152 video_stream_undefined_frame_rate = i;
157 /* Files from iTunes sometimes have two video streams, one with the avg_frame_rate.num and .den set
158 to zero. Only use such a stream if there is no alternative.
160 if (!_video_stream && video_stream_undefined_frame_rate) {
161 _video_stream = video_stream_undefined_frame_rate.get();
164 if (!_video_stream) {
165 throw DecodeError (N_("could not find video stream"));
168 /* Hack: if the AVStreams have duplicate IDs, replace them with our
169 own. We use the IDs so that we can cope with VOBs, in which streams
170 move about in index but remain with the same ID in different
171 VOBs. However, some files have duplicate IDs, hence this hack.
174 bool duplicates = false;
175 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
176 for (uint32_t j = i + 1; j < _format_context->nb_streams; ++j) {
177 if (_format_context->streams[i]->id == _format_context->streams[j]->id) {
184 /* Put in our own IDs */
185 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
186 _format_context->streams[i]->id = i;
190 _frame = av_frame_alloc ();
192 throw DecodeError (N_("could not allocate frame"));
197 FFmpeg::setup_decoders ()
199 boost::mutex::scoped_lock lm (_mutex);
201 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
202 AVCodecContext* context = _format_context->streams[i]->codec;
204 AVCodec* codec = avcodec_find_decoder (context->codec_id);
207 /* This option disables decoding of DCA frame footers in our patched version
208 of FFmpeg. I believe these footers are of no use to us, and they can cause
209 problems when FFmpeg fails to decode them (mantis #352).
211 AVDictionary* options = 0;
212 av_dict_set (&options, "disable_footer", "1", 0);
214 if (avcodec_open2 (context, codec, &options) < 0) {
215 throw DecodeError (N_("could not open decoder"));
219 /* We are silently ignoring any failures to find suitable decoders here */
224 FFmpeg::video_codec_context () const
226 DCPOMATIC_ASSERT (_video_stream);
227 return _format_context->streams[_video_stream.get()]->codec;
231 FFmpeg::subtitle_codec_context () const
233 if (!_ffmpeg_content->subtitle_stream ()) {
237 return _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec;
241 FFmpeg::avio_read (uint8_t* buffer, int const amount)
243 return _file_group.read (buffer, amount);
247 FFmpeg::avio_seek (int64_t const pos, int whence)
249 if (whence == AVSEEK_SIZE) {
250 return _file_group.length ();
253 return _file_group.seek (pos, whence);
257 FFmpeg::subtitle_period (AVSubtitle const & sub)
259 ContentTime const packet_time = ContentTime::from_seconds (static_cast<double> (sub.pts) / AV_TIME_BASE);
261 if (sub.end_display_time == static_cast<uint32_t> (-1)) {
262 /* End time is not known */
263 return FFmpegSubtitlePeriod (packet_time + ContentTime::from_seconds (sub.start_display_time / 1e3));
266 return FFmpegSubtitlePeriod (
267 packet_time + ContentTime::from_seconds (sub.start_display_time / 1e3),
268 packet_time + ContentTime::from_seconds (sub.end_display_time / 1e3)
273 FFmpeg::subtitle_id (AVSubtitle const & sub)
275 MD5Digester digester;
276 digester.add (sub.pts);
277 for (unsigned int i = 0; i < sub.num_rects; ++i) {
278 AVSubtitleRect* rect = sub.rects[i];
279 digester.add (rect->x);
280 digester.add (rect->y);
281 digester.add (rect->w);
282 digester.add (rect->h);
283 #ifdef DCPOMATIC_HAVE_AVSUBTITLERECT_PICT
284 int const line = rect->pict.linesize[0];
285 for (int j = 0; j < rect->h; ++j) {
286 digester.add (rect->pict.data[0] + j * line, line);
289 int const line = rect->linesize[0];
290 for (int j = 0; j < rect->h; ++j) {
291 digester.add (rect->data[0] + j * line, line);
295 return digester.get ();
298 /** @return true if sub starts a new image subtitle */
300 FFmpeg::subtitle_starts_image (AVSubtitle const & sub)
305 for (unsigned int i = 0; i < sub.num_rects; ++i) {
306 switch (sub.rects[i]->type) {
307 case SUBTITLE_BITMAP:
319 /* We can't cope with mixed image/text in one AVSubtitle */
320 DCPOMATIC_ASSERT (!image || !text);
325 /** Compute the pts offset to use given a set of audio streams and some video details.
326 * Sometimes these parameters will have just been determined by an Examiner, sometimes
327 * they will have been retrieved from a piece of Content, hence the need for this method
331 FFmpeg::pts_offset (vector<shared_ptr<FFmpegAudioStream> > audio_streams, optional<ContentTime> first_video, double video_frame_rate) const
333 /* Audio and video frame PTS values may not start with 0. We want
334 to fiddle them so that:
336 1. One of them starts at time 0.
337 2. The first video PTS value ends up on a frame boundary.
339 Then we remove big initial gaps in PTS and we allow our
340 insertion of black frames to work.
343 audio_pts_to_use = audio_pts_from_ffmpeg + pts_offset;
344 video_pts_to_use = video_pts_from_ffmpeg + pts_offset;
347 /* First, make one of them start at 0 */
349 ContentTime po = ContentTime::min ();
352 po = - first_video.get ();
355 BOOST_FOREACH (shared_ptr<FFmpegAudioStream> i, audio_streams) {
356 if (i->first_audio) {
357 po = max (po, - i->first_audio.get ());
361 /* If the offset is positive we would be pushing things from a -ve PTS to be played.
362 I don't think we ever want to do that, as it seems things at -ve PTS are not meant
363 to be seen (use for alignment bars etc.); see mantis #418.
365 if (po > ContentTime ()) {
369 /* Now adjust so that the video pts starts on a frame */
371 ContentTime const fvc = first_video.get() + po;
372 po += fvc.round_up (video_frame_rate) - fvc;