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 "compose.hpp"
24 #include "dcpomatic_log.h"
26 #include "exceptions.h"
28 #include "ffmpeg_audio_stream.h"
29 #include "ffmpeg_content.h"
30 #include "ffmpeg_subtitle_stream.h"
33 #include "memory_util.h"
35 #include <dcp/raw_convert.h>
37 #include <libavcodec/avcodec.h>
38 #include <libavformat/avformat.h>
39 #include <libswscale/swscale.h>
41 #include <boost/algorithm/string.hpp>
51 using std::shared_ptr;
52 using boost::optional;
53 using dcp::raw_convert;
54 using namespace dcpomatic;
57 boost::mutex FFmpeg::_mutex;
60 FFmpeg::FFmpeg (std::shared_ptr<const FFmpegContent> c)
70 boost::mutex::scoped_lock lm (_mutex);
72 for (auto& i: _codec_context) {
73 avcodec_free_context (&i);
76 av_frame_free (&_video_frame);
77 for (auto& audio_frame: _audio_frame) {
78 av_frame_free (&audio_frame.second);
81 avformat_close_input (&_format_context);
86 avio_read_wrapper (void* data, uint8_t* buffer, int amount)
88 return reinterpret_cast<FFmpeg*>(data)->avio_read (buffer, amount);
93 avio_seek_wrapper (void* data, int64_t offset, int whence)
95 return reinterpret_cast<FFmpeg*>(data)->avio_seek (offset, whence);
100 FFmpeg::setup_general ()
102 _file_group.set_paths (_ffmpeg_content->paths ());
103 _avio_buffer = static_cast<uint8_t*> (wrapped_av_malloc(_avio_buffer_size));
104 _avio_context = avio_alloc_context (_avio_buffer, _avio_buffer_size, 0, this, avio_read_wrapper, 0, avio_seek_wrapper);
105 if (!_avio_context) {
106 throw std::bad_alloc ();
108 _format_context = avformat_alloc_context ();
109 if (!_format_context) {
110 throw std::bad_alloc ();
112 _format_context->pb = _avio_context;
114 AVDictionary* options = nullptr;
115 int e = avformat_open_input (&_format_context, 0, 0, &options);
117 throw OpenFileError (_ffmpeg_content->path(0).string(), e, OpenFileError::READ);
120 if (avformat_find_stream_info (_format_context, 0) < 0) {
121 throw DecodeError (_("could not find stream information"));
124 /* Find video stream */
126 optional<int> video_stream_undefined_frame_rate;
128 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
129 auto stream = _format_context->streams[i];
130 if (stream->codecpar->codec_type == AVMEDIA_TYPE_VIDEO && avcodec_find_decoder(stream->codecpar->codec_id) && stream->disposition != AV_DISPOSITION_ATTACHED_PIC) {
131 auto const frame_rate = av_q2d(stream->avg_frame_rate);
132 if (frame_rate < 1 || frame_rate > 1000) {
133 /* Ignore video streams with crazy frame rates. These are usually things like album art on MP3s. */
136 if (stream->avg_frame_rate.num > 0 && stream->avg_frame_rate.den > 0) {
137 /* This is definitely our video stream */
140 /* This is our video stream if we don't get a better offer */
141 video_stream_undefined_frame_rate = i;
146 /* Files from iTunes sometimes have two video streams, one with the avg_frame_rate.num and .den set
147 to zero. Only use such a stream if there is no alternative.
149 if (!_video_stream && video_stream_undefined_frame_rate) {
150 _video_stream = video_stream_undefined_frame_rate.get();
153 /* Hack: if the AVStreams have duplicate IDs, replace them with our
154 own. We use the IDs so that we can cope with VOBs, in which streams
155 move about in index but remain with the same ID in different
156 VOBs. However, some files have duplicate IDs, hence this hack.
159 bool duplicates = false;
160 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
161 for (uint32_t j = i + 1; j < _format_context->nb_streams; ++j) {
162 if (_format_context->streams[i]->id == _format_context->streams[j]->id) {
169 /* Put in our own IDs */
170 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
171 _format_context->streams[i]->id = i;
175 _video_frame = av_frame_alloc ();
176 if (_video_frame == nullptr) {
177 throw std::bad_alloc ();
183 FFmpeg::setup_decoders ()
185 boost::mutex::scoped_lock lm (_mutex);
187 _codec_context.resize (_format_context->nb_streams);
188 for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
189 auto codec = avcodec_find_decoder (_format_context->streams[i]->codecpar->codec_id);
191 auto context = avcodec_alloc_context3 (codec);
193 throw std::bad_alloc ();
195 _codec_context[i] = context;
197 int r = avcodec_parameters_to_context (context, _format_context->streams[i]->codecpar);
199 throw DecodeError ("avcodec_parameters_to_context", "FFmpeg::setup_decoders", r);
202 context->thread_count = 8;
203 context->thread_type = FF_THREAD_FRAME | FF_THREAD_SLICE;
205 AVDictionary* options = nullptr;
206 /* This option disables decoding of DCA frame footers in our patched version
207 of FFmpeg. I believe these footers are of no use to us, and they can cause
208 problems when FFmpeg fails to decode them (mantis #352).
210 av_dict_set (&options, "disable_footer", "1", 0);
211 /* This allows decoding of some DNxHR 444 and HQX files; see
212 https://trac.ffmpeg.org/ticket/5681
214 av_dict_set_int (&options, "strict", FF_COMPLIANCE_EXPERIMENTAL, 0);
215 /* Enable following of links in files */
216 av_dict_set_int (&options, "enable_drefs", 1, 0);
218 r = avcodec_open2 (context, codec, &options);
220 throw DecodeError (N_("avcodec_open2"), N_("FFmpeg::setup_decoders"), r);
223 dcpomatic_log->log (String::compose ("No codec found for stream %1", i), LogEntry::TYPE_WARNING);
230 FFmpeg::video_codec_context () const
232 if (!_video_stream) {
236 return _codec_context[_video_stream.get()];
241 FFmpeg::subtitle_codec_context () const
243 auto str = _ffmpeg_content->subtitle_stream();
248 return _codec_context[str->index(_format_context)];
253 FFmpeg::avio_read (uint8_t* buffer, int const amount)
255 auto result = _file_group.read(buffer, amount);
256 if (result.eof && result.bytes_read == 0) {
259 return result.bytes_read;
264 FFmpeg::avio_seek (int64_t const pos, int whence)
266 if (whence == AVSEEK_SIZE) {
267 return _file_group.length ();
270 return _file_group.seek (pos, whence);
275 FFmpeg::subtitle_period (AVPacket const* packet, AVStream const* stream, AVSubtitle const & sub)
277 auto const packet_time = ContentTime::from_seconds (packet->pts * av_q2d(stream->time_base));
278 auto const start = packet_time + ContentTime::from_seconds(sub.start_display_time / 1e3);
280 if (sub.end_display_time == 0 || sub.end_display_time == static_cast<uint32_t>(-1)) {
281 /* End time is not in the AVSubtitle; perhaps we can use the AVPacket's duration */
282 if (packet->duration) {
283 return FFmpegSubtitlePeriod(start, start + ContentTime::from_seconds(packet->duration * av_q2d(stream->time_base)));
285 return FFmpegSubtitlePeriod(start);
289 return FFmpegSubtitlePeriod (start, packet_time + ContentTime::from_seconds(sub.end_display_time / 1e3));
293 /** Compute the pts offset to use given a set of audio streams and some video details.
294 * Sometimes these parameters will have just been determined by an Examiner, sometimes
295 * they will have been retrieved from a piece of Content, hence the need for this method
299 FFmpeg::pts_offset (vector<shared_ptr<FFmpegAudioStream>> audio_streams, optional<ContentTime> first_video, double video_frame_rate) const
301 /* Audio and video frame PTS values may not start with 0. We want
302 to fiddle them so that:
304 1. One of them starts at time 0.
305 2. The first video PTS value ends up on a frame boundary.
307 Then we remove big initial gaps in PTS and we allow our
308 insertion of black frames to work.
311 audio_pts_to_use = audio_pts_from_ffmpeg + pts_offset;
312 video_pts_to_use = video_pts_from_ffmpeg + pts_offset;
315 /* First, make one of them start at 0 */
317 auto po = ContentTime::min ();
320 po = - first_video.get ();
323 for (auto i: audio_streams) {
324 if (i->first_audio) {
325 po = max (po, - i->first_audio.get ());
329 /* If the offset is positive we would be pushing things from a -ve PTS to be played.
330 I don't think we ever want to do that, as it seems things at -ve PTS are not meant
331 to be seen (use for alignment bars etc.); see mantis #418.
333 if (po > ContentTime ()) {
337 /* Now adjust so that the video pts starts on a frame */
339 auto const fvc = first_video.get() + po;
340 po += fvc.ceil (video_frame_rate) - fvc;
348 FFmpeg::audio_frame (shared_ptr<const FFmpegAudioStream> stream)
350 auto iter = _audio_frame.find(stream);
351 if (iter != _audio_frame.end()) {
355 auto frame = av_frame_alloc ();
356 if (frame == nullptr) {
357 throw std::bad_alloc();
360 _audio_frame[stream] = frame;