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;
54 using boost::shared_ptr;
55 using boost::optional;
56 using boost::dynamic_pointer_cast;
59 FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio)
65 , _subtitle_codec_context (0)
67 , _decode_video (video)
68 , _decode_audio (audio)
70 , _just_sought (false)
74 if (video && audio && c->audio_stream() && c->first_video() && c->audio_stream()->first_audio) {
75 _pts_offset = compute_pts_offset (c->first_video().get(), c->audio_stream()->first_audio.get(), c->video_frame_rate());
80 FFmpegDecoder::compute_pts_offset (double first_video, double first_audio, float video_frame_rate)
82 double const old_first_video = first_video;
84 /* Round the first video to a frame boundary */
85 if (fabs (rint (first_video * video_frame_rate) - first_video * video_frame_rate) > 1e-6) {
86 first_video = ceil (first_video * video_frame_rate) / video_frame_rate;
89 /* Compute the required offset (also removing any common start delay) */
90 return first_video - old_first_video - min (first_video, first_audio);
93 FFmpegDecoder::~FFmpegDecoder ()
95 boost::mutex::scoped_lock lm (_mutex);
97 if (_subtitle_codec_context) {
98 avcodec_close (_subtitle_codec_context);
103 FFmpegDecoder::pass ()
105 int r = av_read_frame (_format_context, &_packet);
108 if (r != AVERROR_EOF) {
109 /* Maybe we should fail here, but for now we'll just finish off instead */
111 av_strerror (r, buf, sizeof(buf));
112 shared_ptr<const Film> film = _film.lock ();
114 film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
117 /* Get any remaining frames */
122 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
125 while (decode_video_packet ()) {}
128 if (_ffmpeg_content->audio_stream() && _decode_audio) {
129 decode_audio_packet ();
132 /* Stop us being asked for any more data */
133 _video_position = _ffmpeg_content->video_length ();
134 _audio_position = _ffmpeg_content->audio_length ();
138 avcodec_get_frame_defaults (_frame);
140 if (_packet.stream_index == _video_stream && _decode_video) {
141 decode_video_packet ();
142 } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) {
143 decode_audio_packet ();
144 } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id) {
145 decode_subtitle_packet ();
148 av_free_packet (&_packet);
151 /** @param data pointer to array of pointers to buffers.
152 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
154 shared_ptr<AudioBuffers>
155 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
157 assert (_ffmpeg_content->audio_channels());
158 assert (bytes_per_audio_sample());
160 /* Deinterleave and convert to float */
162 assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0);
164 int const total_samples = size / bytes_per_audio_sample();
165 int const frames = total_samples / _ffmpeg_content->audio_channels();
166 shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
168 switch (audio_sample_format()) {
169 case AV_SAMPLE_FMT_S16:
171 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
174 for (int i = 0; i < total_samples; ++i) {
175 audio->data(channel)[sample] = float(*p++) / (1 << 15);
178 if (channel == _ffmpeg_content->audio_channels()) {
186 case AV_SAMPLE_FMT_S16P:
188 int16_t** p = reinterpret_cast<int16_t **> (data);
189 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
190 for (int j = 0; j < frames; ++j) {
191 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
197 case AV_SAMPLE_FMT_S32:
199 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
202 for (int i = 0; i < total_samples; ++i) {
203 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
206 if (channel == _ffmpeg_content->audio_channels()) {
214 case AV_SAMPLE_FMT_FLT:
216 float* p = reinterpret_cast<float*> (data[0]);
219 for (int i = 0; i < total_samples; ++i) {
220 audio->data(channel)[sample] = *p++;
223 if (channel == _ffmpeg_content->audio_channels()) {
231 case AV_SAMPLE_FMT_FLTP:
233 float** p = reinterpret_cast<float**> (data);
234 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
235 memcpy (audio->data(i), p[i], frames * sizeof(float));
241 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
248 FFmpegDecoder::audio_sample_format () const
250 if (!_ffmpeg_content->audio_stream()) {
251 return (AVSampleFormat) 0;
254 return audio_codec_context()->sample_fmt;
258 FFmpegDecoder::bytes_per_audio_sample () const
260 return av_get_bytes_per_sample (audio_sample_format ());
264 FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
266 double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base);
267 int64_t const vt = frame / (_ffmpeg_content->video_frame_rate() * time_base);
268 av_seek_frame (_format_context, _video_stream, vt, AVSEEK_FLAG_BACKWARD);
270 avcodec_flush_buffers (video_codec_context());
271 if (_subtitle_codec_context) {
272 avcodec_flush_buffers (_subtitle_codec_context);
278 /* We're already there; from here on we can only seek non-zero amounts */
284 int r = av_read_frame (_format_context, &_packet);
289 avcodec_get_frame_defaults (_frame);
291 if (_packet.stream_index == _video_stream) {
293 int const r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
294 if (r >= 0 && finished) {
295 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
297 _video_position = rint (
298 (bet * time_base + _pts_offset) * _ffmpeg_content->video_frame_rate()
300 av_free_packet (&_packet);
306 av_free_packet (&_packet);
312 FFmpegDecoder::decode_audio_packet ()
314 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
318 AVPacket copy_packet = _packet;
320 while (copy_packet.size > 0) {
323 int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet);
324 if (decode_result >= 0) {
325 if (frame_finished) {
327 if (_audio_position == 0) {
328 /* Where we are in the source, in seconds */
329 double const pts = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
330 * av_frame_get_best_effort_timestamp(_frame) - _pts_offset;
333 /* Emit some silence */
334 shared_ptr<AudioBuffers> silence (
336 _ffmpeg_content->audio_channels(),
337 pts * _ffmpeg_content->content_audio_frame_rate()
341 silence->make_silent ();
342 audio (silence, _audio_position);
346 copy_packet.data += decode_result;
347 copy_packet.size -= decode_result;
354 FFmpegDecoder::decode_video_packet ()
357 if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
361 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
363 shared_ptr<FilterGraph> graph;
365 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
366 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
370 if (i == _filter_graphs.end ()) {
371 shared_ptr<const Film> film = _film.lock ();
374 graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
375 _filter_graphs.push_back (graph);
377 film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
382 list<shared_ptr<Image> > images = graph->process (_frame);
384 string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second;
386 for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
388 shared_ptr<Image> image = *i;
389 if (!post_process.empty ()) {
390 image = image->post_process (post_process, true);
393 int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
394 if (bet != AV_NOPTS_VALUE) {
396 double const pts = bet * av_q2d (_format_context->streams[_video_stream]->time_base) - _pts_offset;
399 /* We just did a seek, so disable any attempts to correct for where we
402 _video_position = rint (pts * _ffmpeg_content->video_frame_rate ());
403 _just_sought = false;
406 double const next = _video_position / _ffmpeg_content->video_frame_rate();
407 double const one_frame = 1 / _ffmpeg_content->video_frame_rate ();
408 double delta = pts - next;
410 while (delta > one_frame) {
411 /* This PTS is more than one frame forward in time of where we think we should be; emit
414 boost::shared_ptr<Image> black (
416 static_cast<AVPixelFormat> (_frame->format),
417 libdcp::Size (video_codec_context()->width, video_codec_context()->height),
422 black->make_black ();
423 video (image, false, _video_position);
427 if (delta > -one_frame) {
428 /* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */
429 video (image, false, _video_position);
433 shared_ptr<const Film> film = _film.lock ();
435 film->log()->log ("Dropping frame without PTS");
444 FFmpegDecoder::setup_subtitle ()
446 boost::mutex::scoped_lock lm (_mutex);
448 if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
452 _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
453 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
455 if (_subtitle_codec == 0) {
456 throw DecodeError (_("could not find subtitle decoder"));
459 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
460 throw DecodeError (N_("could not open subtitle decoder"));
465 FFmpegDecoder::done () const
467 bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length());
468 bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length());
473 FFmpegDecoder::decode_subtitle_packet ()
477 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
481 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
482 indicate that the previous subtitle should stop.
484 if (sub.num_rects <= 0) {
485 subtitle (shared_ptr<Image> (), dcpomatic::Rect<double> (), 0, 0);
487 } else if (sub.num_rects > 1) {
488 throw DecodeError (_("multi-part subtitles not yet supported"));
491 /* Subtitle PTS in seconds (within the source, not taking into account any of the
492 source that we may have chopped off for the DCP)
494 double const packet_time = static_cast<double> (sub.pts) / AV_TIME_BASE;
496 /* hence start time for this sub */
497 Time const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ;
498 Time const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ;
500 AVSubtitleRect const * rect = sub.rects[0];
502 if (rect->type != SUBTITLE_BITMAP) {
503 throw DecodeError (_("non-bitmap subtitles not yet supported"));
506 shared_ptr<Image> image (new SimpleImage (PIX_FMT_RGBA, libdcp::Size (rect->w, rect->h), true));
508 /* Start of the first line in the subtitle */
509 uint8_t* sub_p = rect->pict.data[0];
510 /* sub_p looks up into a RGB palette which is here */
511 uint32_t const * palette = (uint32_t *) rect->pict.data[1];
512 /* Start of the output data */
513 uint32_t* out_p = (uint32_t *) image->data()[0];
515 for (int y = 0; y < rect->h; ++y) {
516 uint8_t* sub_line_p = sub_p;
517 uint32_t* out_line_p = out_p;
518 for (int x = 0; x < rect->w; ++x) {
519 *out_line_p++ = palette[*sub_line_p++];
521 sub_p += rect->pict.linesize[0];
522 out_p += image->stride()[0] / sizeof (uint32_t);
525 libdcp::Size const vs = _ffmpeg_content->video_size ();
529 dcpomatic::Rect<double> (
530 static_cast<double> (rect->x) / vs.width,
531 static_cast<double> (rect->y) / vs.height,
532 static_cast<double> (rect->w) / vs.width,
533 static_cast<double> (rect->h) / vs.height
540 avsubtitle_free (&sub);