2 Copyright (C) 2012-2016 Carl Hetherington <cth@carlh.net>
4 This file is part of DCP-o-matic.
6 DCP-o-matic is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 DCP-o-matic is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with DCP-o-matic. If not, see <http://www.gnu.org/licenses/>.
21 /** @file src/ffmpeg_decoder.cc
22 * @brief A decoder using FFmpeg to decode content.
26 #include "exceptions.h"
30 #include "ffmpeg_decoder.h"
31 #include "subtitle_decoder.h"
32 #include "ffmpeg_audio_stream.h"
33 #include "ffmpeg_subtitle_stream.h"
34 #include "video_filter_graph.h"
35 #include "audio_buffers.h"
36 #include "ffmpeg_content.h"
37 #include "raw_image_proxy.h"
38 #include "video_decoder.h"
40 #include "audio_decoder.h"
41 #include "compose.hpp"
42 #include "subtitle_content.h"
43 #include "audio_content.h"
44 #include <dcp/subtitle_string.h>
45 #include <sub/ssa_reader.h>
46 #include <sub/subtitle.h>
47 #include <sub/collect.h>
49 #include <libavcodec/avcodec.h>
50 #include <libavformat/avformat.h>
52 #include <boost/foreach.hpp>
53 #include <boost/algorithm/string.hpp>
61 #define LOG_GENERAL(...) _log->log (String::compose (__VA_ARGS__), LogEntry::TYPE_GENERAL);
62 #define LOG_ERROR(...) _log->log (String::compose (__VA_ARGS__), LogEntry::TYPE_ERROR);
63 #define LOG_WARNING_NC(...) _log->log (__VA_ARGS__, LogEntry::TYPE_WARNING);
64 #define LOG_WARNING(...) _log->log (String::compose (__VA_ARGS__), LogEntry::TYPE_WARNING);
74 using boost::shared_ptr;
75 using boost::is_any_of;
77 using boost::optional;
78 using boost::dynamic_pointer_cast;
81 FFmpegDecoder::FFmpegDecoder (shared_ptr<const FFmpegContent> c, shared_ptr<Log> log, bool fast)
84 , _have_current_subtitle (false)
87 video.reset (new VideoDecoder (this, c, log));
88 _pts_offset = pts_offset (c->ffmpeg_audio_streams(), c->first_video(), c->active_video_frame_rate());
89 /* It doesn't matter what size or pixel format this is, it just needs to be black */
90 _black_image.reset (new Image (AV_PIX_FMT_RGB24, dcp::Size (128, 128), true));
91 _black_image->make_black ();
93 _pts_offset = ContentTime ();
97 audio.reset (new AudioDecoder (this, c->audio, log, fast));
101 /* XXX: this time here should be the time of the first subtitle, not 0 */
102 subtitle.reset (new SubtitleDecoder (this, c->subtitle, log, ContentTime()));
105 _next_time.resize (_format_context->nb_streams);
109 FFmpegDecoder::flush ()
111 /* Get any remaining frames */
116 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
118 while (video && decode_video_packet ()) {}
121 decode_audio_packet ();
124 /* Make sure all streams are the same length and round up to the next video frame */
126 FrameRateChange const frc = _ffmpeg_content->film()->active_frame_rate_change(_ffmpeg_content->position());
127 ContentTime full_length (_ffmpeg_content->full_length(), frc);
128 full_length = full_length.ceil (frc.source);
130 double const vfr = _ffmpeg_content->video_frame_rate().get();
131 Frame const f = full_length.frames_round (vfr);
132 Frame v = video->position().frames_round (vfr) + 1;
134 video->emit (shared_ptr<const ImageProxy> (new RawImageProxy (_black_image)), v);
139 BOOST_FOREACH (shared_ptr<FFmpegAudioStream> i, _ffmpeg_content->ffmpeg_audio_streams ()) {
140 ContentTime a = audio->stream_position(i);
141 while (a < full_length) {
142 ContentTime to_do = min (full_length - a, ContentTime::from_seconds (0.1));
143 shared_ptr<AudioBuffers> silence (new AudioBuffers (i->channels(), to_do.frames_ceil (i->frame_rate())));
144 silence->make_silent ();
145 audio->emit (i, silence, a);
156 FFmpegDecoder::pass ()
158 int r = av_read_frame (_format_context, &_packet);
160 /* AVERROR_INVALIDDATA can apparently be returned sometimes even when av_read_frame
161 has pretty-much succeeded (and hence generated data which should be processed).
162 Hence it makes sense to continue here in that case.
164 if (r < 0 && r != AVERROR_INVALIDDATA) {
165 if (r != AVERROR_EOF) {
166 /* Maybe we should fail here, but for now we'll just finish off instead */
168 av_strerror (r, buf, sizeof(buf));
169 LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), &buf[0], r);
176 int const si = _packet.stream_index;
177 shared_ptr<const FFmpegContent> fc = _ffmpeg_content;
179 if (_video_stream && si == _video_stream.get() && !video->ignore()) {
180 decode_video_packet ();
181 } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index(_format_context, si) && !subtitle->ignore()) {
182 decode_subtitle_packet ();
184 decode_audio_packet ();
187 av_packet_unref (&_packet);
191 /** @param data pointer to array of pointers to buffers.
192 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
194 shared_ptr<AudioBuffers>
195 FFmpegDecoder::deinterleave_audio (shared_ptr<FFmpegAudioStream> stream) const
197 DCPOMATIC_ASSERT (bytes_per_audio_sample (stream));
199 int const size = av_samples_get_buffer_size (
200 0, stream->stream(_format_context)->codec->channels, _frame->nb_samples, audio_sample_format (stream), 1
203 /* Deinterleave and convert to float */
205 /* total_samples and frames will be rounded down here, so if there are stray samples at the end
206 of the block that do not form a complete sample or frame they will be dropped.
208 int const total_samples = size / bytes_per_audio_sample (stream);
209 int const channels = stream->channels();
210 int const frames = total_samples / channels;
211 shared_ptr<AudioBuffers> audio (new AudioBuffers (channels, frames));
212 float** data = audio->data();
214 switch (audio_sample_format (stream)) {
215 case AV_SAMPLE_FMT_U8:
217 uint8_t* p = reinterpret_cast<uint8_t *> (_frame->data[0]);
220 for (int i = 0; i < total_samples; ++i) {
221 data[channel][sample] = float(*p++) / (1 << 23);
224 if (channel == channels) {
232 case AV_SAMPLE_FMT_S16:
234 int16_t* p = reinterpret_cast<int16_t *> (_frame->data[0]);
237 for (int i = 0; i < total_samples; ++i) {
238 data[channel][sample] = float(*p++) / (1 << 15);
241 if (channel == channels) {
249 case AV_SAMPLE_FMT_S16P:
251 int16_t** p = reinterpret_cast<int16_t **> (_frame->data);
252 for (int i = 0; i < channels; ++i) {
253 for (int j = 0; j < frames; ++j) {
254 data[i][j] = static_cast<float>(p[i][j]) / (1 << 15);
260 case AV_SAMPLE_FMT_S32:
262 int32_t* p = reinterpret_cast<int32_t *> (_frame->data[0]);
265 for (int i = 0; i < total_samples; ++i) {
266 data[channel][sample] = static_cast<float>(*p++) / 2147483648;
269 if (channel == channels) {
277 case AV_SAMPLE_FMT_S32P:
279 int32_t** p = reinterpret_cast<int32_t **> (_frame->data);
280 for (int i = 0; i < channels; ++i) {
281 for (int j = 0; j < frames; ++j) {
282 data[i][j] = static_cast<float>(p[i][j]) / 2147483648;
288 case AV_SAMPLE_FMT_FLT:
290 float* p = reinterpret_cast<float*> (_frame->data[0]);
293 for (int i = 0; i < total_samples; ++i) {
294 data[channel][sample] = *p++;
297 if (channel == channels) {
305 case AV_SAMPLE_FMT_FLTP:
307 float** p = reinterpret_cast<float**> (_frame->data);
308 /* Sometimes there aren't as many channels in the _frame as in the stream */
309 for (int i = 0; i < _frame->channels; ++i) {
310 memcpy (data[i], p[i], frames * sizeof(float));
312 for (int i = _frame->channels; i < channels; ++i) {
313 audio->make_silent (i);
319 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format (stream))));
326 FFmpegDecoder::audio_sample_format (shared_ptr<FFmpegAudioStream> stream) const
328 return stream->stream (_format_context)->codec->sample_fmt;
332 FFmpegDecoder::bytes_per_audio_sample (shared_ptr<FFmpegAudioStream> stream) const
334 return av_get_bytes_per_sample (audio_sample_format (stream));
338 FFmpegDecoder::seek (ContentTime time, bool accurate)
340 Decoder::seek (time, accurate);
342 /* If we are doing an `accurate' seek, we need to use pre-roll, as
343 we don't really know what the seek will give us.
346 ContentTime pre_roll = accurate ? ContentTime::from_seconds (2) : ContentTime (0);
349 /* XXX: it seems debatable whether PTS should be used here...
350 http://www.mjbshaw.com/2012/04/seeking-in-ffmpeg-know-your-timestamp.html
353 optional<int> stream;
356 stream = _video_stream;
358 shared_ptr<FFmpegAudioStream> s = dynamic_pointer_cast<FFmpegAudioStream> (_ffmpeg_content->audio->stream ());
360 stream = s->index (_format_context);
364 DCPOMATIC_ASSERT (stream);
366 ContentTime u = time - _pts_offset;
367 if (u < ContentTime ()) {
373 u.seconds() / av_q2d (_format_context->streams[stream.get()]->time_base),
377 if (video_codec_context ()) {
378 avcodec_flush_buffers (video_codec_context());
381 BOOST_FOREACH (shared_ptr<FFmpegAudioStream> i, ffmpeg_content()->ffmpeg_audio_streams()) {
382 avcodec_flush_buffers (i->stream(_format_context)->codec);
385 if (subtitle_codec_context ()) {
386 avcodec_flush_buffers (subtitle_codec_context ());
389 _have_current_subtitle = false;
393 FFmpegDecoder::decode_audio_packet ()
395 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
399 AVPacket copy_packet = _packet;
400 int const stream_index = copy_packet.stream_index;
402 /* XXX: inefficient */
403 vector<shared_ptr<FFmpegAudioStream> > streams = ffmpeg_content()->ffmpeg_audio_streams ();
404 vector<shared_ptr<FFmpegAudioStream> >::const_iterator stream = streams.begin ();
405 while (stream != streams.end () && !(*stream)->uses_index (_format_context, stream_index)) {
409 if (stream == streams.end ()) {
410 /* The packet's stream may not be an audio one; just ignore it in this method if so */
414 while (copy_packet.size > 0) {
417 int decode_result = avcodec_decode_audio4 ((*stream)->stream (_format_context)->codec, _frame, &frame_finished, ©_packet);
418 if (decode_result < 0) {
419 /* avcodec_decode_audio4 can sometimes return an error even though it has decoded
420 some valid data; for example dca_subframe_footer can return AVERROR_INVALIDDATA
421 if it overreads the auxiliary data. ffplay carries on if frame_finished is true,
422 even in the face of such an error, so I think we should too.
424 Returning from the method here caused mantis #352.
426 LOG_WARNING ("avcodec_decode_audio4 failed (%1)", decode_result);
428 /* Fudge decode_result so that we come out of the while loop when
429 we've processed this data.
431 decode_result = copy_packet.size;
434 if (frame_finished) {
435 shared_ptr<AudioBuffers> data = deinterleave_audio (*stream);
438 if (_frame->pts == AV_NOPTS_VALUE && _next_time[stream_index]) {
439 /* In some streams we see not every frame coming through with a timestamp; for those
440 that have AV_NOPTS_VALUE we need to work out the timestamp ourselves. This is
441 particularly noticeable with TrueHD streams (see #1111).
443 ct = *_next_time[stream_index];
445 ct = ContentTime::from_seconds (
446 av_frame_get_best_effort_timestamp (_frame) *
447 av_q2d ((*stream)->stream (_format_context)->time_base))
451 _next_time[stream_index] = ct + ContentTime::from_frames(data->frames(), (*stream)->frame_rate());
453 if (ct < ContentTime ()) {
454 /* Discard audio data that comes before time 0 */
455 Frame const remove = min (int64_t (data->frames()), (-ct).frames_ceil(double((*stream)->frame_rate ())));
456 data->move (data->frames() - remove, remove, 0);
457 data->set_frames (data->frames() - remove);
458 ct += ContentTime::from_frames (remove, (*stream)->frame_rate ());
461 if (ct < ContentTime()) {
463 "Crazy timestamp %1 for %2 samples in stream %3 packet pts %4 (ts=%5 tb=%6, off=%7)",
466 copy_packet.stream_index,
468 av_frame_get_best_effort_timestamp(_frame),
469 av_q2d((*stream)->stream(_format_context)->time_base),
470 to_string(_pts_offset)
474 /* Give this data provided there is some, and its time is sane */
475 if (ct >= ContentTime() && data->frames() > 0) {
476 audio->emit (*stream, data, ct);
480 copy_packet.data += decode_result;
481 copy_packet.size -= decode_result;
486 FFmpegDecoder::decode_video_packet ()
488 DCPOMATIC_ASSERT (_video_stream);
491 if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
495 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
497 shared_ptr<VideoFilterGraph> graph;
499 list<shared_ptr<VideoFilterGraph> >::iterator i = _filter_graphs.begin();
500 while (i != _filter_graphs.end() && !(*i)->can_process (dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
504 if (i == _filter_graphs.end ()) {
505 graph.reset (new VideoFilterGraph (dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
506 graph->setup (_ffmpeg_content->filters ());
507 _filter_graphs.push_back (graph);
508 LOG_GENERAL (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format);
513 list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
515 for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
517 shared_ptr<Image> image = i->first;
519 if (i->second != AV_NOPTS_VALUE) {
520 double const pts = i->second * av_q2d (_format_context->streams[_video_stream.get()]->time_base) + _pts_offset.seconds ();
522 shared_ptr<ImageProxy> (new RawImageProxy (image)),
523 llrint (pts * _ffmpeg_content->active_video_frame_rate ())
526 LOG_WARNING_NC ("Dropping frame without PTS");
534 FFmpegDecoder::decode_subtitle_packet ()
538 if (avcodec_decode_subtitle2 (subtitle_codec_context(), &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
542 /* Stop any current subtitle, either at the time it was supposed to stop, or now if now is sooner */
543 if (_have_current_subtitle) {
544 if (_current_subtitle_to) {
545 subtitle->emit_stop (min(*_current_subtitle_to, subtitle_period(sub).from + _pts_offset));
547 subtitle->emit_stop (subtitle_period(sub).from + _pts_offset);
549 _have_current_subtitle = false;
552 if (sub.num_rects <= 0) {
553 /* Nothing new in this subtitle */
557 /* Subtitle PTS (within the source, not taking into account any of the
558 source that we may have chopped off for the DCP).
560 FFmpegSubtitlePeriod sub_period = subtitle_period (sub);
562 from = sub_period.from + _pts_offset;
563 _have_current_subtitle = true;
565 _current_subtitle_to = *sub_period.to + _pts_offset;
567 _current_subtitle_to = optional<ContentTime>();
570 for (unsigned int i = 0; i < sub.num_rects; ++i) {
571 AVSubtitleRect const * rect = sub.rects[i];
573 switch (rect->type) {
576 case SUBTITLE_BITMAP:
577 decode_bitmap_subtitle (rect, from);
580 cout << "XXX: SUBTITLE_TEXT " << rect->text << "\n";
583 decode_ass_subtitle (rect->ass, from);
588 if (_current_subtitle_to) {
589 subtitle->emit_stop (*_current_subtitle_to);
592 avsubtitle_free (&sub);
596 FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime from)
598 /* Note BGRA is expressed little-endian, so the first byte in the word is B, second
599 G, third R, fourth A.
601 shared_ptr<Image> image (new Image (AV_PIX_FMT_BGRA, dcp::Size (rect->w, rect->h), true));
603 #ifdef DCPOMATIC_HAVE_AVSUBTITLERECT_PICT
604 /* Start of the first line in the subtitle */
605 uint8_t* sub_p = rect->pict.data[0];
606 /* sub_p looks up into a BGRA palette which is here
607 (i.e. first byte B, second G, third R, fourth A)
609 uint32_t const * palette = (uint32_t *) rect->pict.data[1];
611 /* Start of the first line in the subtitle */
612 uint8_t* sub_p = rect->data[0];
613 /* sub_p looks up into a BGRA palette which is here
614 (i.e. first byte B, second G, third R, fourth A)
616 uint32_t const * palette = (uint32_t *) rect->data[1];
618 /* And the stream has a map of those palette colours to colours
619 chosen by the user; created a `mapped' palette from those settings.
621 map<RGBA, RGBA> colour_map = ffmpeg_content()->subtitle_stream()->colours ();
622 vector<RGBA> mapped_palette (rect->nb_colors);
623 for (int i = 0; i < rect->nb_colors; ++i) {
624 RGBA c ((palette[i] & 0xff0000) >> 16, (palette[i] & 0xff00) >> 8, palette[i] & 0xff, (palette[i] & 0xff000000) >> 24);
625 map<RGBA, RGBA>::const_iterator j = colour_map.find (c);
626 if (j != colour_map.end ()) {
627 mapped_palette[i] = j->second;
629 /* This colour was not found in the FFmpegSubtitleStream's colour map; probably because
630 it is from a project that was created before this stuff was added. Just use the
631 colour straight from the original palette.
633 mapped_palette[i] = c;
637 /* Start of the output data */
638 uint32_t* out_p = (uint32_t *) image->data()[0];
640 for (int y = 0; y < rect->h; ++y) {
641 uint8_t* sub_line_p = sub_p;
642 uint32_t* out_line_p = out_p;
643 for (int x = 0; x < rect->w; ++x) {
644 RGBA const p = mapped_palette[*sub_line_p++];
645 /* XXX: this seems to be wrong to me (isn't the output image BGRA?) but it looks right on screen */
646 *out_line_p++ = (p.a << 24) | (p.b << 16) | (p.g << 8) | p.r;
648 #ifdef DCPOMATIC_HAVE_AVSUBTITLERECT_PICT
649 sub_p += rect->pict.linesize[0];
651 sub_p += rect->linesize[0];
653 out_p += image->stride()[0] / sizeof (uint32_t);
656 int const target_width = subtitle_codec_context()->width;
657 int const target_height = subtitle_codec_context()->height;
658 dcpomatic::Rect<double> const scaled_rect (
659 static_cast<double> (rect->x) / target_width,
660 static_cast<double> (rect->y) / target_height,
661 static_cast<double> (rect->w) / target_width,
662 static_cast<double> (rect->h) / target_height
665 subtitle->emit_image_start (from, image, scaled_rect);
669 FFmpegDecoder::decode_ass_subtitle (string ass, ContentTime from)
671 /* We have no styles and no Format: line, so I'm assuming that FFmpeg
672 produces a single format of Dialogue: lines...
676 split (bits, ass, is_any_of (","));
677 if (bits.size() < 10) {
681 sub::RawSubtitle base;
682 list<sub::RawSubtitle> raw = sub::SSAReader::parse_line (
685 _ffmpeg_content->video->size().width,
686 _ffmpeg_content->video->size().height
689 BOOST_FOREACH (sub::Subtitle const & i, sub::collect<list<sub::Subtitle> > (raw)) {
690 subtitle->emit_text_start (from, i);