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"
45 #include "ffmpeg_content.h"
52 using std::stringstream;
56 using boost::shared_ptr;
57 using boost::optional;
58 using boost::dynamic_pointer_cast;
61 FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio)
67 , _subtitle_codec_context (0)
69 , _decode_video (video)
70 , _decode_audio (audio)
71 , _video_pts_offset (0)
72 , _audio_pts_offset (0)
73 , _just_sought (false)
77 /* Audio and video frame PTS values may not start with 0. We want
78 to fiddle them so that:
80 1. One of them starts at time 0.
81 2. The first video PTS value ends up on a frame boundary.
83 Then we remove big initial gaps in PTS and we allow our
84 insertion of black frames to work.
87 audio_pts_to_use = audio_pts_from_ffmpeg + audio_pts_offset;
88 video_pts_to_use = video_pts_from_ffmpeg + video_pts_offset;
91 bool const have_video = video && c->first_video();
92 bool const have_audio = audio && c->audio_stream() && c->audio_stream()->first_audio;
94 /* First, make one of them start at 0 */
96 if (have_audio && have_video) {
97 _video_pts_offset = _audio_pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get());
98 } else if (have_video) {
99 _video_pts_offset = - c->first_video().get();
100 } else if (have_audio) {
101 _audio_pts_offset = - c->audio_stream()->first_audio.get();
104 /* Now adjust both so that the video pts starts on a frame */
105 if (have_video && have_audio) {
106 double first_video = c->first_video().get() + _video_pts_offset;
107 double const old_first_video = first_video;
109 /* Round the first video up to a frame boundary */
110 if (fabs (rint (first_video * c->video_frame_rate()) - first_video * c->video_frame_rate()) > 1e-6) {
111 first_video = ceil (first_video * c->video_frame_rate()) / c->video_frame_rate ();
114 _video_pts_offset += first_video - old_first_video;
115 _audio_pts_offset += first_video - old_first_video;
119 FFmpegDecoder::~FFmpegDecoder ()
121 boost::mutex::scoped_lock lm (_mutex);
123 if (_subtitle_codec_context) {
124 avcodec_close (_subtitle_codec_context);
129 FFmpegDecoder::flush ()
131 /* Get any remaining frames */
136 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
139 while (decode_video_packet ()) {}
142 if (_ffmpeg_content->audio_stream() && _decode_audio) {
143 decode_audio_packet ();
146 /* Stop us being asked for any more data */
147 _video_position = _ffmpeg_content->video_length ();
148 _audio_position = _ffmpeg_content->audio_length ();
152 FFmpegDecoder::pass ()
154 int r = av_read_frame (_format_context, &_packet);
157 if (r != AVERROR_EOF) {
158 /* Maybe we should fail here, but for now we'll just finish off instead */
160 av_strerror (r, buf, sizeof(buf));
161 shared_ptr<const Film> film = _film.lock ();
163 film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
170 shared_ptr<const Film> film = _film.lock ();
173 int const si = _packet.stream_index;
175 if (si == _video_stream && _decode_video) {
176 decode_video_packet ();
177 } else if (_ffmpeg_content->audio_stream() && si == _ffmpeg_content->audio_stream()->index (_format_context) && _decode_audio) {
178 decode_audio_packet ();
179 } else if (_ffmpeg_content->subtitle_stream() && si == _ffmpeg_content->subtitle_stream()->index (_format_context) && film->with_subtitles ()) {
180 decode_subtitle_packet ();
183 av_free_packet (&_packet);
186 /** @param data pointer to array of pointers to buffers.
187 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
189 shared_ptr<AudioBuffers>
190 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
192 assert (_ffmpeg_content->audio_channels());
193 assert (bytes_per_audio_sample());
195 /* Deinterleave and convert to float */
197 assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0);
199 int const total_samples = size / bytes_per_audio_sample();
200 int const frames = total_samples / _ffmpeg_content->audio_channels();
201 shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
203 switch (audio_sample_format()) {
204 case AV_SAMPLE_FMT_S16:
206 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
209 for (int i = 0; i < total_samples; ++i) {
210 audio->data(channel)[sample] = float(*p++) / (1 << 15);
213 if (channel == _ffmpeg_content->audio_channels()) {
221 case AV_SAMPLE_FMT_S16P:
223 int16_t** p = reinterpret_cast<int16_t **> (data);
224 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
225 for (int j = 0; j < frames; ++j) {
226 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
232 case AV_SAMPLE_FMT_S32:
234 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
237 for (int i = 0; i < total_samples; ++i) {
238 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
241 if (channel == _ffmpeg_content->audio_channels()) {
249 case AV_SAMPLE_FMT_FLT:
251 float* p = reinterpret_cast<float*> (data[0]);
254 for (int i = 0; i < total_samples; ++i) {
255 audio->data(channel)[sample] = *p++;
258 if (channel == _ffmpeg_content->audio_channels()) {
266 case AV_SAMPLE_FMT_FLTP:
268 float** p = reinterpret_cast<float**> (data);
269 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
270 memcpy (audio->data(i), p[i], frames * sizeof(float));
276 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
283 FFmpegDecoder::audio_sample_format () const
285 if (!_ffmpeg_content->audio_stream()) {
286 return (AVSampleFormat) 0;
289 return audio_codec_context()->sample_fmt;
293 FFmpegDecoder::bytes_per_audio_sample () const
295 return av_get_bytes_per_sample (audio_sample_format ());
299 FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
301 double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base);
303 /* If we are doing an accurate seek, our initial shot will be 5 frames (5 being
304 a number plucked from the air) earlier than we want to end up. The loop below
305 will hopefully then step through to where we want to be.
317 /* Initial seek time in the stream's timebase */
318 int64_t const initial_vt = ((initial / _ffmpeg_content->video_frame_rate()) - _video_pts_offset) / time_base;
320 av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD);
322 avcodec_flush_buffers (video_codec_context());
323 if (_subtitle_codec_context) {
324 avcodec_flush_buffers (_subtitle_codec_context);
328 _video_position = frame;
330 if (frame == 0 || !accurate) {
331 /* We're already there, or we're as close as we need to be */
336 int r = av_read_frame (_format_context, &_packet);
341 if (_packet.stream_index != _video_stream) {
342 av_free_packet (&_packet);
347 r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
348 if (r >= 0 && finished) {
349 _video_position = rint (
350 (av_frame_get_best_effort_timestamp (_frame) * time_base + _video_pts_offset) * _ffmpeg_content->video_frame_rate()
353 if (_video_position >= (frame - 1)) {
354 av_free_packet (&_packet);
359 av_free_packet (&_packet);
364 FFmpegDecoder::decode_audio_packet ()
366 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
370 AVPacket copy_packet = _packet;
372 while (copy_packet.size > 0) {
375 int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet);
376 if (decode_result < 0) {
377 shared_ptr<const Film> film = _film.lock ();
379 film->log()->log (String::compose ("avcodec_decode_audio4 failed (%1)", decode_result));
383 if (frame_finished) {
385 if (_audio_position == 0) {
386 /* Where we are in the source, in seconds */
387 double const pts = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
388 * av_frame_get_best_effort_timestamp(_frame) + _audio_pts_offset;
391 /* Emit some silence */
392 shared_ptr<AudioBuffers> silence (
394 _ffmpeg_content->audio_channels(),
395 pts * _ffmpeg_content->content_audio_frame_rate()
399 silence->make_silent ();
400 audio (silence, _audio_position);
404 int const data_size = av_samples_get_buffer_size (
405 0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1
408 audio (deinterleave_audio (_frame->data, data_size), _audio_position);
411 copy_packet.data += decode_result;
412 copy_packet.size -= decode_result;
417 FFmpegDecoder::decode_video_packet ()
420 if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
424 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
426 shared_ptr<FilterGraph> graph;
428 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
429 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
433 if (i == _filter_graphs.end ()) {
434 shared_ptr<const Film> film = _film.lock ();
437 graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
438 _filter_graphs.push_back (graph);
440 film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
445 list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
447 string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second;
449 for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
451 shared_ptr<Image> image = i->first;
452 if (!post_process.empty ()) {
453 image = image->post_process (post_process, true);
456 if (i->second != AV_NOPTS_VALUE) {
458 double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _video_pts_offset;
461 /* We just did a seek, so disable any attempts to correct for where we
464 _video_position = rint (pts * _ffmpeg_content->video_frame_rate ());
465 _just_sought = false;
468 double const next = _video_position / _ffmpeg_content->video_frame_rate();
469 double const one_frame = 1 / _ffmpeg_content->video_frame_rate ();
470 double delta = pts - next;
472 while (delta > one_frame) {
473 /* This PTS is more than one frame forward in time of where we think we should be; emit
477 /* XXX: I think this should be a copy of the last frame... */
478 boost::shared_ptr<Image> black (
480 static_cast<AVPixelFormat> (_frame->format),
481 libdcp::Size (video_codec_context()->width, video_codec_context()->height),
486 black->make_black ();
487 video (image, false, _video_position);
491 if (delta > -one_frame) {
492 /* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */
493 video (image, false, _video_position);
497 shared_ptr<const Film> film = _film.lock ();
499 film->log()->log ("Dropping frame without PTS");
508 FFmpegDecoder::setup_subtitle ()
510 boost::mutex::scoped_lock lm (_mutex);
512 if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->index (_format_context) >= int (_format_context->nb_streams)) {
516 _subtitle_codec_context = _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec;
517 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
519 if (_subtitle_codec == 0) {
520 throw DecodeError (_("could not find subtitle decoder"));
523 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
524 throw DecodeError (N_("could not open subtitle decoder"));
529 FFmpegDecoder::done () const
531 bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length());
532 bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length());
537 FFmpegDecoder::decode_subtitle_packet ()
541 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
545 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
546 indicate that the previous subtitle should stop.
548 if (sub.num_rects <= 0) {
549 image_subtitle (shared_ptr<Image> (), dcpomatic::Rect<double> (), 0, 0);
551 } else if (sub.num_rects > 1) {
552 throw DecodeError (_("multi-part subtitles not yet supported"));
555 /* Subtitle PTS in seconds (within the source, not taking into account any of the
556 source that we may have chopped off for the DCP)
558 double const packet_time = (static_cast<double> (sub.pts ) / AV_TIME_BASE) + _video_pts_offset;
560 /* hence start time for this sub */
561 Time const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ;
562 Time const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ;
564 AVSubtitleRect const * rect = sub.rects[0];
566 if (rect->type != SUBTITLE_BITMAP) {
567 throw DecodeError (_("non-bitmap subtitles not yet supported"));
570 /* Note RGBA is expressed little-endian, so the first byte in the word is R, second
571 G, third B, fourth A.
573 shared_ptr<Image> image (new Image (PIX_FMT_RGBA, libdcp::Size (rect->w, rect->h), true));
575 /* Start of the first line in the subtitle */
576 uint8_t* sub_p = rect->pict.data[0];
577 /* sub_p looks up into a BGRA palette which is here
578 (i.e. first byte B, second G, third R, fourth A)
580 uint32_t const * palette = (uint32_t *) rect->pict.data[1];
581 /* Start of the output data */
582 uint32_t* out_p = (uint32_t *) image->data()[0];
584 for (int y = 0; y < rect->h; ++y) {
585 uint8_t* sub_line_p = sub_p;
586 uint32_t* out_line_p = out_p;
587 for (int x = 0; x < rect->w; ++x) {
588 uint32_t const p = palette[*sub_line_p++];
589 *out_line_p++ = ((p & 0xff) << 16) | (p & 0xff00) | ((p & 0xff0000) >> 16) | (p & 0xff000000);
591 sub_p += rect->pict.linesize[0];
592 out_p += image->stride()[0] / sizeof (uint32_t);
595 libdcp::Size const vs = _ffmpeg_content->video_size ();
599 dcpomatic::Rect<double> (
600 static_cast<double> (rect->x) / vs.width,
601 static_cast<double> (rect->y) / vs.height,
602 static_cast<double> (rect->w) / vs.width,
603 static_cast<double> (rect->h) / vs.height
610 avsubtitle_free (&sub);