2 Copyright (C) 2012 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.
20 /** @file src/ffmpeg_decoder.cc
21 * @brief A decoder using FFmpeg to decode content.
30 #include <boost/lexical_cast.hpp>
33 #include <libavcodec/avcodec.h>
34 #include <libavformat/avformat.h>
38 #include "exceptions.h"
42 #include "ffmpeg_decoder.h"
43 #include "filter_graph.h"
44 #include "audio_buffers.h"
51 using std::stringstream;
55 using boost::shared_ptr;
56 using boost::optional;
57 using boost::dynamic_pointer_cast;
60 FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio)
66 , _subtitle_codec_context (0)
68 , _decode_video (video)
69 , _decode_audio (audio)
70 , _video_pts_offset (0)
71 , _audio_pts_offset (0)
72 , _just_sought (false)
76 /* Audio and video frame PTS values may not start with 0. We want
77 to fiddle them so that:
79 1. One of them starts at time 0.
80 2. The first video PTS value ends up on a frame boundary.
82 Then we remove big initial gaps in PTS and we allow our
83 insertion of black frames to work.
86 audio_pts_to_use = audio_pts_from_ffmpeg + audio_pts_offset;
87 video_pts_to_use = video_pts_from_ffmpeg + video_pts_offset;
90 bool const have_video = video && c->first_video();
91 bool const have_audio = audio && c->audio_stream() && c->audio_stream()->first_audio;
93 /* First, make one of them start at 0 */
95 if (have_audio && have_video) {
96 _video_pts_offset = _audio_pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get());
97 } else if (have_video) {
98 _video_pts_offset = - c->first_video().get();
101 /* Now adjust both so that the video pts starts on a frame */
102 if (have_video && have_audio) {
103 double first_video = c->first_video().get() + _video_pts_offset;
104 double const old_first_video = first_video;
106 /* Round the first video up to a frame boundary */
107 if (fabs (rint (first_video * c->video_frame_rate()) - first_video * c->video_frame_rate()) > 1e-6) {
108 first_video = ceil (first_video * c->video_frame_rate()) / c->video_frame_rate ();
111 _video_pts_offset += first_video - old_first_video;
112 _audio_pts_offset += first_video - old_first_video;
116 FFmpegDecoder::~FFmpegDecoder ()
118 boost::mutex::scoped_lock lm (_mutex);
120 if (_subtitle_codec_context) {
121 avcodec_close (_subtitle_codec_context);
126 FFmpegDecoder::flush ()
128 /* Get any remaining frames */
133 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
136 while (decode_video_packet ()) {}
139 if (_ffmpeg_content->audio_stream() && _decode_audio) {
140 decode_audio_packet ();
143 /* Stop us being asked for any more data */
144 _video_position = _ffmpeg_content->video_length ();
145 _audio_position = _ffmpeg_content->audio_length ();
149 FFmpegDecoder::pass ()
151 int r = av_read_frame (_format_context, &_packet);
154 if (r != AVERROR_EOF) {
155 /* Maybe we should fail here, but for now we'll just finish off instead */
157 av_strerror (r, buf, sizeof(buf));
158 shared_ptr<const Film> film = _film.lock ();
160 film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
167 avcodec_get_frame_defaults (_frame);
169 if (_packet.stream_index == _video_stream && _decode_video) {
170 decode_video_packet ();
171 } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) {
172 decode_audio_packet ();
173 } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id) {
174 decode_subtitle_packet ();
177 av_free_packet (&_packet);
180 /** @param data pointer to array of pointers to buffers.
181 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
183 shared_ptr<AudioBuffers>
184 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
186 assert (_ffmpeg_content->audio_channels());
187 assert (bytes_per_audio_sample());
189 /* Deinterleave and convert to float */
191 assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0);
193 int const total_samples = size / bytes_per_audio_sample();
194 int const frames = total_samples / _ffmpeg_content->audio_channels();
195 shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
197 switch (audio_sample_format()) {
198 case AV_SAMPLE_FMT_S16:
200 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
203 for (int i = 0; i < total_samples; ++i) {
204 audio->data(channel)[sample] = float(*p++) / (1 << 15);
207 if (channel == _ffmpeg_content->audio_channels()) {
215 case AV_SAMPLE_FMT_S16P:
217 int16_t** p = reinterpret_cast<int16_t **> (data);
218 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
219 for (int j = 0; j < frames; ++j) {
220 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
226 case AV_SAMPLE_FMT_S32:
228 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
231 for (int i = 0; i < total_samples; ++i) {
232 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
235 if (channel == _ffmpeg_content->audio_channels()) {
243 case AV_SAMPLE_FMT_FLT:
245 float* p = reinterpret_cast<float*> (data[0]);
248 for (int i = 0; i < total_samples; ++i) {
249 audio->data(channel)[sample] = *p++;
252 if (channel == _ffmpeg_content->audio_channels()) {
260 case AV_SAMPLE_FMT_FLTP:
262 float** p = reinterpret_cast<float**> (data);
263 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
264 memcpy (audio->data(i), p[i], frames * sizeof(float));
270 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
277 FFmpegDecoder::audio_sample_format () const
279 if (!_ffmpeg_content->audio_stream()) {
280 return (AVSampleFormat) 0;
283 return audio_codec_context()->sample_fmt;
287 FFmpegDecoder::bytes_per_audio_sample () const
289 return av_get_bytes_per_sample (audio_sample_format ());
293 FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
295 double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base);
297 /* If we are doing an accurate seek, our initial shot will be 5 frames (5 being
298 a number plucked from the air) earlier than we want to end up. The loop below
299 will hopefully then step through to where we want to be.
311 /* Initial seek time in the stream's timebase */
312 int64_t const initial_vt = ((initial / _ffmpeg_content->video_frame_rate()) - _video_pts_offset) / time_base;
314 av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD);
316 avcodec_flush_buffers (video_codec_context());
317 if (_subtitle_codec_context) {
318 avcodec_flush_buffers (_subtitle_codec_context);
322 _video_position = frame;
324 if (frame == 0 || !accurate) {
325 /* We're already there, or we're as close as we need to be */
330 int r = av_read_frame (_format_context, &_packet);
335 if (_packet.stream_index != _video_stream) {
339 avcodec_get_frame_defaults (_frame);
342 r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
343 if (r >= 0 && finished) {
344 _video_position = rint (
345 (av_frame_get_best_effort_timestamp (_frame) * time_base + _video_pts_offset) * _ffmpeg_content->video_frame_rate()
348 if (_video_position >= (frame - 1)) {
349 av_free_packet (&_packet);
354 av_free_packet (&_packet);
359 FFmpegDecoder::decode_audio_packet ()
361 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
365 AVPacket copy_packet = _packet;
367 while (copy_packet.size > 0) {
370 int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet);
371 if (decode_result < 0) {
372 shared_ptr<const Film> film = _film.lock ();
374 film->log()->log (String::compose ("avcodec_decode_audio4 failed (%1)", decode_result));
378 if (frame_finished) {
380 if (_audio_position == 0) {
381 /* Where we are in the source, in seconds */
382 double const pts = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
383 * av_frame_get_best_effort_timestamp(_frame) + _audio_pts_offset;
386 /* Emit some silence */
387 shared_ptr<AudioBuffers> silence (
389 _ffmpeg_content->audio_channels(),
390 pts * _ffmpeg_content->content_audio_frame_rate()
394 silence->make_silent ();
395 audio (silence, _audio_position);
399 int const data_size = av_samples_get_buffer_size (
400 0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1
403 audio (deinterleave_audio (_frame->data, data_size), _audio_position);
406 copy_packet.data += decode_result;
407 copy_packet.size -= decode_result;
412 FFmpegDecoder::decode_video_packet ()
415 if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
419 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
421 shared_ptr<FilterGraph> graph;
423 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
424 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
428 if (i == _filter_graphs.end ()) {
429 shared_ptr<const Film> film = _film.lock ();
432 graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
433 _filter_graphs.push_back (graph);
435 film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
440 list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
442 string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second;
444 for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
446 shared_ptr<Image> image = i->first;
447 if (!post_process.empty ()) {
448 image = image->post_process (post_process, true);
451 if (i->second != AV_NOPTS_VALUE) {
453 double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _video_pts_offset;
456 /* We just did a seek, so disable any attempts to correct for where we
459 _video_position = rint (pts * _ffmpeg_content->video_frame_rate ());
460 _just_sought = false;
463 double const next = _video_position / _ffmpeg_content->video_frame_rate();
464 double const one_frame = 1 / _ffmpeg_content->video_frame_rate ();
465 double delta = pts - next;
467 while (delta > one_frame) {
468 /* This PTS is more than one frame forward in time of where we think we should be; emit
472 /* XXX: I think this should be a copy of the last frame... */
473 boost::shared_ptr<Image> black (
475 static_cast<AVPixelFormat> (_frame->format),
476 libdcp::Size (video_codec_context()->width, video_codec_context()->height),
481 black->make_black ();
482 video (image, false, _video_position);
486 if (delta > -one_frame) {
487 /* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */
488 video (image, false, _video_position);
492 shared_ptr<const Film> film = _film.lock ();
494 film->log()->log ("Dropping frame without PTS");
503 FFmpegDecoder::setup_subtitle ()
505 boost::mutex::scoped_lock lm (_mutex);
507 if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
511 _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
512 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
514 if (_subtitle_codec == 0) {
515 throw DecodeError (_("could not find subtitle decoder"));
518 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
519 throw DecodeError (N_("could not open subtitle decoder"));
524 FFmpegDecoder::done () const
526 bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length());
527 bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length());
532 FFmpegDecoder::decode_subtitle_packet ()
536 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
540 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
541 indicate that the previous subtitle should stop.
543 if (sub.num_rects <= 0) {
544 subtitle (shared_ptr<Image> (), dcpomatic::Rect<double> (), 0, 0);
546 } else if (sub.num_rects > 1) {
547 throw DecodeError (_("multi-part subtitles not yet supported"));
550 /* Subtitle PTS in seconds (within the source, not taking into account any of the
551 source that we may have chopped off for the DCP)
553 double const packet_time = static_cast<double> (sub.pts) / AV_TIME_BASE;
555 /* hence start time for this sub */
556 Time const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ;
557 Time const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ;
559 AVSubtitleRect const * rect = sub.rects[0];
561 if (rect->type != SUBTITLE_BITMAP) {
562 throw DecodeError (_("non-bitmap subtitles not yet supported"));
565 shared_ptr<Image> image (new Image (PIX_FMT_RGBA, libdcp::Size (rect->w, rect->h), true));
567 /* Start of the first line in the subtitle */
568 uint8_t* sub_p = rect->pict.data[0];
569 /* sub_p looks up into a RGB palette which is here */
570 uint32_t const * palette = (uint32_t *) rect->pict.data[1];
571 /* Start of the output data */
572 uint32_t* out_p = (uint32_t *) image->data()[0];
574 for (int y = 0; y < rect->h; ++y) {
575 uint8_t* sub_line_p = sub_p;
576 uint32_t* out_line_p = out_p;
577 for (int x = 0; x < rect->w; ++x) {
578 *out_line_p++ = palette[*sub_line_p++];
580 sub_p += rect->pict.linesize[0];
581 out_p += image->stride()[0] / sizeof (uint32_t);
584 libdcp::Size const vs = _ffmpeg_content->video_size ();
588 dcpomatic::Rect<double> (
589 static_cast<double> (rect->x) / vs.width,
590 static_cast<double> (rect->y) / vs.height,
591 static_cast<double> (rect->w) / vs.width,
592 static_cast<double> (rect->h) / vs.height
599 avsubtitle_free (&sub);