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.
32 #include <libavcodec/avcodec.h>
33 #include <libavformat/avformat.h>
37 #include "exceptions.h"
41 #include "ffmpeg_decoder.h"
42 #include "filter_graph.h"
43 #include "audio_buffers.h"
44 #include "ffmpeg_content.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)
71 , _just_sought (false)
75 /* Audio and video frame PTS values may not start with 0. We want
76 to fiddle them so that:
78 1. One of them starts at time 0.
79 2. The first video PTS value ends up on a frame boundary.
81 Then we remove big initial gaps in PTS and we allow our
82 insertion of black frames to work.
85 audio_pts_to_use = audio_pts_from_ffmpeg + pts_offset;
86 video_pts_to_use = video_pts_from_ffmpeg + pts_offset;
89 bool const have_video = video && c->first_video();
90 bool const have_audio = audio && c->audio_stream() && c->audio_stream()->first_audio;
92 /* First, make one of them start at 0 */
94 if (have_audio && have_video) {
95 _pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get());
96 } else if (have_video) {
97 _pts_offset = - c->first_video().get();
98 } else if (have_audio) {
99 _pts_offset = - c->audio_stream()->first_audio.get();
102 /* Now adjust both so that the video pts starts on a frame */
103 if (have_video && have_audio) {
104 double first_video = c->first_video().get() + _pts_offset;
105 double const old_first_video = first_video;
107 /* Round the first video up to a frame boundary */
108 if (fabs (rint (first_video * c->video_frame_rate()) - first_video * c->video_frame_rate()) > 1e-6) {
109 first_video = ceil (first_video * c->video_frame_rate()) / c->video_frame_rate ();
112 _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_after_3d_combine ();
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 shared_ptr<const Film> film = _film.lock ();
170 int const si = _packet.stream_index;
172 if (si == _video_stream && _decode_video) {
173 decode_video_packet ();
174 } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, si) && _decode_audio) {
175 decode_audio_packet ();
176 } else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si) && film->with_subtitles ()) {
177 decode_subtitle_packet ();
180 av_free_packet (&_packet);
183 /** @param data pointer to array of pointers to buffers.
184 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
186 shared_ptr<AudioBuffers>
187 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
189 assert (_ffmpeg_content->audio_channels());
190 assert (bytes_per_audio_sample());
192 /* Deinterleave and convert to float */
194 assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0);
196 int const total_samples = size / bytes_per_audio_sample();
197 int const frames = total_samples / _ffmpeg_content->audio_channels();
198 shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
200 switch (audio_sample_format()) {
201 case AV_SAMPLE_FMT_U8:
203 uint8_t* p = reinterpret_cast<uint8_t *> (data[0]);
206 for (int i = 0; i < total_samples; ++i) {
207 audio->data(channel)[sample] = float(*p++) / (1 << 23);
210 if (channel == _ffmpeg_content->audio_channels()) {
218 case AV_SAMPLE_FMT_S16:
220 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
223 for (int i = 0; i < total_samples; ++i) {
224 audio->data(channel)[sample] = float(*p++) / (1 << 15);
227 if (channel == _ffmpeg_content->audio_channels()) {
235 case AV_SAMPLE_FMT_S16P:
237 int16_t** p = reinterpret_cast<int16_t **> (data);
238 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
239 for (int j = 0; j < frames; ++j) {
240 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
246 case AV_SAMPLE_FMT_S32:
248 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
251 for (int i = 0; i < total_samples; ++i) {
252 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
255 if (channel == _ffmpeg_content->audio_channels()) {
263 case AV_SAMPLE_FMT_FLT:
265 float* p = reinterpret_cast<float*> (data[0]);
268 for (int i = 0; i < total_samples; ++i) {
269 audio->data(channel)[sample] = *p++;
272 if (channel == _ffmpeg_content->audio_channels()) {
280 case AV_SAMPLE_FMT_FLTP:
282 float** p = reinterpret_cast<float**> (data);
283 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
284 memcpy (audio->data(i), p[i], frames * sizeof(float));
290 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
297 FFmpegDecoder::audio_sample_format () const
299 if (!_ffmpeg_content->audio_stream()) {
300 return (AVSampleFormat) 0;
303 return audio_codec_context()->sample_fmt;
307 FFmpegDecoder::bytes_per_audio_sample () const
309 return av_get_bytes_per_sample (audio_sample_format ());
313 FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
315 double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base);
317 /* If we are doing an accurate seek, our initial shot will be 5 frames (5 being
318 a number plucked from the air) earlier than we want to end up. The loop below
319 will hopefully then step through to where we want to be.
331 /* Initial seek time in the stream's timebase */
332 int64_t const initial_vt = ((initial / _ffmpeg_content->video_frame_rate()) - _pts_offset) / time_base;
334 av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD);
336 avcodec_flush_buffers (video_codec_context());
337 if (_subtitle_codec_context) {
338 avcodec_flush_buffers (_subtitle_codec_context);
341 /* This !accurate is piling hack upon hack; setting _just_sought to true
342 even with accurate == true defeats our attempt to align the start
343 of the video and audio. Here we disable that defeat when accurate == true
344 i.e. when we are making a DCP rather than just previewing one.
345 Ewww. This should be gone in 2.0.
351 _video_position = frame;
353 if (frame == 0 || !accurate) {
354 /* We're already there, or we're as close as we need to be */
359 int r = av_read_frame (_format_context, &_packet);
364 if (_packet.stream_index != _video_stream) {
365 av_free_packet (&_packet);
370 r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
371 if (r >= 0 && finished) {
372 _video_position = rint (
373 (av_frame_get_best_effort_timestamp (_frame) * time_base + _pts_offset) * _ffmpeg_content->video_frame_rate()
376 if (_video_position >= (frame - 1)) {
377 av_free_packet (&_packet);
382 av_free_packet (&_packet);
387 FFmpegDecoder::decode_audio_packet ()
389 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
393 AVPacket copy_packet = _packet;
395 while (copy_packet.size > 0) {
398 int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet);
399 if (decode_result < 0) {
400 shared_ptr<const Film> film = _film.lock ();
402 film->log()->log (String::compose ("avcodec_decode_audio4 failed (%1)", decode_result));
406 if (frame_finished) {
408 if (_audio_position == 0) {
409 /* Where we are in the source, in seconds */
410 double const pts = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
411 * av_frame_get_best_effort_timestamp(_frame) + _pts_offset;
414 /* Emit some silence */
415 shared_ptr<AudioBuffers> silence (
417 _ffmpeg_content->audio_channels(),
418 pts * _ffmpeg_content->content_audio_frame_rate()
422 silence->make_silent ();
423 audio (silence, _audio_position);
427 int const data_size = av_samples_get_buffer_size (
428 0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1
431 audio (deinterleave_audio (_frame->data, data_size), _audio_position);
434 copy_packet.data += decode_result;
435 copy_packet.size -= decode_result;
440 FFmpegDecoder::decode_video_packet ()
443 if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
447 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
449 shared_ptr<FilterGraph> graph;
451 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
452 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
456 if (i == _filter_graphs.end ()) {
457 shared_ptr<const Film> film = _film.lock ();
460 graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
461 _filter_graphs.push_back (graph);
463 film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
468 list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
470 for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
472 shared_ptr<Image> image = i->first;
474 if (i->second != AV_NOPTS_VALUE) {
476 double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset;
479 /* We just did a seek, so disable any attempts to correct for where we
482 _video_position = rint (pts * _ffmpeg_content->video_frame_rate ());
483 _just_sought = false;
486 double const next = _video_position / _ffmpeg_content->video_frame_rate();
487 double const one_frame = 1 / _ffmpeg_content->video_frame_rate ();
488 double delta = pts - next;
490 while (delta > one_frame) {
491 /* This PTS is more than one frame forward in time of where we think we should be; emit
495 /* XXX: I think this should be a copy of the last frame... */
496 boost::shared_ptr<Image> black (
498 static_cast<AVPixelFormat> (_frame->format),
499 libdcp::Size (video_codec_context()->width, video_codec_context()->height),
504 black->make_black ();
505 video (image, false, _video_position);
509 if (delta > -one_frame) {
510 /* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */
511 video (image, false, _video_position);
515 shared_ptr<const Film> film = _film.lock ();
517 film->log()->log ("Dropping frame without PTS");
526 FFmpegDecoder::setup_subtitle ()
528 boost::mutex::scoped_lock lm (_mutex);
530 if (!_ffmpeg_content->subtitle_stream()) {
534 _subtitle_codec_context = _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec;
535 if (_subtitle_codec_context == 0) {
536 throw DecodeError (N_("could not find subtitle stream"));
539 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
541 if (_subtitle_codec == 0) {
542 throw DecodeError (N_("could not find subtitle decoder"));
545 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
546 throw DecodeError (N_("could not open subtitle decoder"));
551 FFmpegDecoder::done () const
553 bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length());
554 bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length());
559 FFmpegDecoder::decode_subtitle_packet ()
563 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
567 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
568 indicate that the previous subtitle should stop.
570 if (sub.num_rects <= 0) {
571 subtitle (shared_ptr<Image> (), dcpomatic::Rect<double> (), 0, 0);
573 } else if (sub.num_rects > 1) {
574 throw DecodeError (_("multi-part subtitles not yet supported"));
577 /* Subtitle PTS in seconds (within the source, not taking into account any of the
578 source that we may have chopped off for the DCP)
580 double const packet_time = (static_cast<double> (sub.pts ) / AV_TIME_BASE) + _pts_offset;
582 /* hence start time for this sub */
583 Time const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ;
584 Time const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ;
586 AVSubtitleRect const * rect = sub.rects[0];
588 if (rect->type != SUBTITLE_BITMAP) {
589 throw DecodeError (_("non-bitmap subtitles not yet supported"));
592 /* Note RGBA is expressed little-endian, so the first byte in the word is R, second
593 G, third B, fourth A.
595 shared_ptr<Image> image (new Image (PIX_FMT_RGBA, libdcp::Size (rect->w, rect->h), true));
597 /* Start of the first line in the subtitle */
598 uint8_t* sub_p = rect->pict.data[0];
599 /* sub_p looks up into a BGRA palette which is here
600 (i.e. first byte B, second G, third R, fourth A)
602 uint32_t const * palette = (uint32_t *) rect->pict.data[1];
603 /* Start of the output data */
604 uint32_t* out_p = (uint32_t *) image->data()[0];
606 for (int y = 0; y < rect->h; ++y) {
607 uint8_t* sub_line_p = sub_p;
608 uint32_t* out_line_p = out_p;
609 for (int x = 0; x < rect->w; ++x) {
610 uint32_t const p = palette[*sub_line_p++];
611 *out_line_p++ = ((p & 0xff) << 16) | (p & 0xff00) | ((p & 0xff0000) >> 16) | (p & 0xff000000);
613 sub_p += rect->pict.linesize[0];
614 out_p += image->stride()[0] / sizeof (uint32_t);
617 libdcp::Size const vs = _ffmpeg_content->video_size ();
621 dcpomatic::Rect<double> (
622 static_cast<double> (rect->x) / vs.width,
623 static_cast<double> (rect->y) / vs.height,
624 static_cast<double> (rect->w) / vs.width,
625 static_cast<double> (rect->h) / vs.height
632 avsubtitle_free (&sub);