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)
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();
99 } else if (have_audio) {
100 _audio_pts_offset = - c->audio_stream()->first_audio.get();
103 /* Now adjust both so that the video pts starts on a frame */
104 if (have_video && have_audio) {
105 double first_video = c->first_video().get() + _video_pts_offset;
106 double const old_first_video = first_video;
108 /* Round the first video up to a frame boundary */
109 if (fabs (rint (first_video * c->video_frame_rate()) - first_video * c->video_frame_rate()) > 1e-6) {
110 first_video = ceil (first_video * c->video_frame_rate()) / c->video_frame_rate ();
113 _video_pts_offset += first_video - old_first_video;
114 _audio_pts_offset += first_video - old_first_video;
118 FFmpegDecoder::~FFmpegDecoder ()
120 boost::mutex::scoped_lock lm (_mutex);
122 if (_subtitle_codec_context) {
123 avcodec_close (_subtitle_codec_context);
128 FFmpegDecoder::flush ()
130 /* Get any remaining frames */
135 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
138 while (decode_video_packet ()) {}
141 if (_ffmpeg_content->audio_stream() && _decode_audio) {
142 decode_audio_packet ();
145 /* Stop us being asked for any more data */
146 _video_position = _ffmpeg_content->video_length ();
147 _audio_position = _ffmpeg_content->audio_length ();
151 FFmpegDecoder::pass ()
153 int r = av_read_frame (_format_context, &_packet);
156 if (r != AVERROR_EOF) {
157 /* Maybe we should fail here, but for now we'll just finish off instead */
159 av_strerror (r, buf, sizeof(buf));
160 shared_ptr<const Film> film = _film.lock ();
162 film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
169 avcodec_get_frame_defaults (_frame);
171 shared_ptr<const Film> film = _film.lock ();
174 int const si = _packet.stream_index;
176 if (si == _video_stream && _decode_video) {
177 decode_video_packet ();
178 } else if (_ffmpeg_content->audio_stream() && si == _ffmpeg_content->audio_stream()->index (_format_context) && _decode_audio) {
179 decode_audio_packet ();
180 } else if (_ffmpeg_content->subtitle_stream() && si == _ffmpeg_content->subtitle_stream()->index (_format_context) && film->with_subtitles ()) {
181 decode_subtitle_packet ();
184 av_free_packet (&_packet);
187 /** @param data pointer to array of pointers to buffers.
188 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
190 shared_ptr<AudioBuffers>
191 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
193 assert (_ffmpeg_content->audio_channels());
194 assert (bytes_per_audio_sample());
196 /* Deinterleave and convert to float */
198 assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0);
200 int const total_samples = size / bytes_per_audio_sample();
201 int const frames = total_samples / _ffmpeg_content->audio_channels();
202 shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
204 switch (audio_sample_format()) {
205 case AV_SAMPLE_FMT_S16:
207 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
210 for (int i = 0; i < total_samples; ++i) {
211 audio->data(channel)[sample] = float(*p++) / (1 << 15);
214 if (channel == _ffmpeg_content->audio_channels()) {
222 case AV_SAMPLE_FMT_S16P:
224 int16_t** p = reinterpret_cast<int16_t **> (data);
225 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
226 for (int j = 0; j < frames; ++j) {
227 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
233 case AV_SAMPLE_FMT_S32:
235 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
238 for (int i = 0; i < total_samples; ++i) {
239 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
242 if (channel == _ffmpeg_content->audio_channels()) {
250 case AV_SAMPLE_FMT_FLT:
252 float* p = reinterpret_cast<float*> (data[0]);
255 for (int i = 0; i < total_samples; ++i) {
256 audio->data(channel)[sample] = *p++;
259 if (channel == _ffmpeg_content->audio_channels()) {
267 case AV_SAMPLE_FMT_FLTP:
269 float** p = reinterpret_cast<float**> (data);
270 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
271 memcpy (audio->data(i), p[i], frames * sizeof(float));
277 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
284 FFmpegDecoder::audio_sample_format () const
286 if (!_ffmpeg_content->audio_stream()) {
287 return (AVSampleFormat) 0;
290 return audio_codec_context()->sample_fmt;
294 FFmpegDecoder::bytes_per_audio_sample () const
296 return av_get_bytes_per_sample (audio_sample_format ());
300 FFmpegDecoder::minimal_run (boost::function<bool (int)> finished)
304 while (!finished (frames_read)) {
305 int r = av_read_frame (_format_context, &_packet);
312 double const time_base = av_q2d (_format_context->streams[_packet.stream_index]->time_base);
314 if (_packet.stream_index == _video_stream) {
316 avcodec_get_frame_defaults (_frame);
319 r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
320 if (r >= 0 && finished) {
321 _video_position = rint (
322 (av_frame_get_best_effort_timestamp (_frame) * time_base + _video_pts_offset) * _ffmpeg_content->video_frame_rate()
326 } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->index (_format_context)) {
328 AVPacket copy_packet = _packet;
330 while (copy_packet.size > 0) {
333 r = avcodec_decode_audio4 (audio_codec_context(), _frame, &finished, ©_packet);
334 if (r >= 0 && finished) {
335 _audio_position = rint (
336 (av_frame_get_best_effort_timestamp (_frame) * time_base + _audio_pts_offset) *
337 _ffmpeg_content->audio_stream()->frame_rate
341 copy_packet.data += r;
342 copy_packet.size -= r;
346 av_free_packet (&_packet);
353 FFmpegDecoder::seek_overrun_finished (DCPTime seek) const
356 _video_position >= _ffmpeg_content->time_to_content_video_frames (seek) ||
357 _audio_position >= _ffmpeg_content->time_to_content_audio_frames (seek, _ffmpeg_content->position())
362 FFmpegDecoder::seek_final_finished (int n, int done) const
368 FFmpegDecoder::seek_and_flush (DCPTime t)
370 int64_t const initial_v = ((_ffmpeg_content->time_to_content_video_frames (t) / _ffmpeg_content->video_frame_rate()) - _video_pts_offset) /
371 av_q2d (_format_context->streams[_video_stream]->time_base);
373 av_seek_frame (_format_context, _video_stream, initial_v, AVSEEK_FLAG_BACKWARD);
375 shared_ptr<FFmpegAudioStream> as = _ffmpeg_content->audio_stream ();
377 int64_t initial_a = ((_ffmpeg_content->time_to_content_audio_frames (t, t) / as->frame_rate) - _audio_pts_offset) /
378 av_q2d (as->stream(_format_context)->time_base);
380 av_seek_frame (_format_context, as->index (_format_context), initial_a, AVSEEK_FLAG_BACKWARD);
383 avcodec_flush_buffers (video_codec_context());
384 if (audio_codec_context ()) {
385 avcodec_flush_buffers (audio_codec_context ());
387 if (_subtitle_codec_context) {
388 avcodec_flush_buffers (_subtitle_codec_context);
391 _video_position = _ffmpeg_content->time_to_content_video_frames (t);
392 _audio_position = _ffmpeg_content->time_to_content_audio_frames (t, t);
396 FFmpegDecoder::seek (DCPTime time, bool accurate)
398 /* If we are doing an accurate seek, our initial shot will be 200ms (200 being
399 a number plucked from the air) earlier than we want to end up. The loop below
400 will hopefully then step through to where we want to be.
403 DCPTime pre_roll = accurate ? (0.2 * TIME_HZ) : 0;
404 DCPTime initial_seek = time - pre_roll;
405 if (initial_seek < 0) {
409 /* Initial seek time in the video stream's timebase */
411 seek_and_flush (initial_seek);
413 if (time == 0 || !accurate) {
414 /* We're already there, or we're as close as we need to be */
418 int const N = minimal_run (boost::bind (&FFmpegDecoder::seek_overrun_finished, this, time));
420 seek_and_flush (initial_seek);
422 minimal_run (boost::bind (&FFmpegDecoder::seek_final_finished, this, N - 1, _1));
427 FFmpegDecoder::decode_audio_packet ()
429 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
433 AVPacket copy_packet = _packet;
435 while (copy_packet.size > 0) {
438 int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet);
440 if (decode_result < 0) {
441 shared_ptr<const Film> film = _film.lock ();
443 film->log()->log (String::compose ("avcodec_decode_audio4 failed (%1)", decode_result));
447 if (frame_finished) {
448 Time const t = (av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
449 * av_frame_get_best_effort_timestamp(_frame) + _audio_pts_offset) * TIME_HZ;
451 int const data_size = av_samples_get_buffer_size (
452 0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1
455 audio (deinterleave_audio (_frame->data, data_size), t);
458 copy_packet.data += decode_result;
459 copy_packet.size -= decode_result;
464 FFmpegDecoder::decode_video_packet ()
467 if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
471 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
473 shared_ptr<FilterGraph> graph;
475 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
476 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
480 if (i == _filter_graphs.end ()) {
481 shared_ptr<const Film> film = _film.lock ();
484 graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
485 _filter_graphs.push_back (graph);
487 film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
492 list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
494 string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second;
496 for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
498 shared_ptr<Image> image = i->first;
499 if (!post_process.empty ()) {
500 image = image->post_process (post_process, true);
503 if (i->second != AV_NOPTS_VALUE) {
504 Time const t = (i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _video_pts_offset) * TIME_HZ;
505 video (image, false, t);
507 shared_ptr<const Film> film = _film.lock ();
509 film->log()->log ("Dropping frame without PTS");
518 FFmpegDecoder::setup_subtitle ()
520 boost::mutex::scoped_lock lm (_mutex);
522 if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->index (_format_context) >= int (_format_context->nb_streams)) {
526 _subtitle_codec_context = _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec;
527 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
529 if (_subtitle_codec == 0) {
530 throw DecodeError (_("could not find subtitle decoder"));
533 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
534 throw DecodeError (N_("could not open subtitle decoder"));
539 FFmpegDecoder::done () const
541 bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length());
542 bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length());
547 FFmpegDecoder::decode_subtitle_packet ()
551 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
555 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
556 indicate that the previous subtitle should stop.
558 if (sub.num_rects <= 0) {
559 subtitle (shared_ptr<Image> (), dcpomatic::Rect<double> (), 0, 0);
561 } else if (sub.num_rects > 1) {
562 throw DecodeError (_("multi-part subtitles not yet supported"));
565 /* Subtitle PTS in seconds (within the source, not taking into account any of the
566 source that we may have chopped off for the DCP)
568 double const packet_time = static_cast<double> (sub.pts) / AV_TIME_BASE;
570 /* hence start time for this sub */
571 DCPTime const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ;
572 DCPTime const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ;
574 AVSubtitleRect const * rect = sub.rects[0];
576 if (rect->type != SUBTITLE_BITMAP) {
577 throw DecodeError (_("non-bitmap subtitles not yet supported"));
580 shared_ptr<Image> image (new Image (PIX_FMT_RGBA, libdcp::Size (rect->w, rect->h), true));
582 /* Start of the first line in the subtitle */
583 uint8_t* sub_p = rect->pict.data[0];
584 /* sub_p looks up into a RGB palette which is here */
585 uint32_t const * palette = (uint32_t *) rect->pict.data[1];
586 /* Start of the output data */
587 uint32_t* out_p = (uint32_t *) image->data()[0];
589 for (int y = 0; y < rect->h; ++y) {
590 uint8_t* sub_line_p = sub_p;
591 uint32_t* out_line_p = out_p;
592 for (int x = 0; x < rect->w; ++x) {
593 *out_line_p++ = palette[*sub_line_p++];
595 sub_p += rect->pict.linesize[0];
596 out_p += image->stride()[0] / sizeof (uint32_t);
599 libdcp::Size const vs = _ffmpeg_content->video_size ();
603 dcpomatic::Rect<double> (
604 static_cast<double> (rect->x) / vs.width,
605 static_cast<double> (rect->y) / vs.height,
606 static_cast<double> (rect->w) / vs.width,
607 static_cast<double> (rect->h) / vs.height
614 avsubtitle_free (&sub);