2 Copyright (C) 2012-2014 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.
31 #include <libavcodec/avcodec.h>
32 #include <libavformat/avformat.h>
36 #include "exceptions.h"
40 #include "ffmpeg_decoder.h"
41 #include "filter_graph.h"
42 #include "audio_buffers.h"
43 #include "ffmpeg_content.h"
44 #include "image_proxy.h"
48 #define LOG_GENERAL(...) film->log()->log (String::compose (__VA_ARGS__), Log::TYPE_GENERAL);
49 #define LOG_ERROR(...) film->log()->log (String::compose (__VA_ARGS__), Log::TYPE_ERROR);
50 #define LOG_WARNING(...) film->log()->log (__VA_ARGS__, Log::TYPE_WARNING);
58 using boost::shared_ptr;
59 using boost::optional;
60 using boost::dynamic_pointer_cast;
63 FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio)
69 , _subtitle_codec_context (0)
71 , _decode_video (video)
72 , _decode_audio (audio)
74 , _just_sought (false)
78 /* Audio and video frame PTS values may not start with 0. We want
79 to fiddle them so that:
81 1. One of them starts at time 0.
82 2. The first video PTS value ends up on a frame boundary.
84 Then we remove big initial gaps in PTS and we allow our
85 insertion of black frames to work.
88 audio_pts_to_use = audio_pts_from_ffmpeg + pts_offset;
89 video_pts_to_use = video_pts_from_ffmpeg + pts_offset;
92 bool const have_video = video && c->first_video();
93 bool const have_audio = audio && c->audio_stream() && c->audio_stream()->first_audio;
95 /* First, make one of them start at 0 */
97 if (have_audio && have_video) {
98 _pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get());
99 } else if (have_video) {
100 _pts_offset = - c->first_video().get();
101 } else if (have_audio) {
102 _pts_offset = - c->audio_stream()->first_audio.get();
105 /* Now adjust both so that the video pts starts on a frame */
106 if (have_video && have_audio) {
107 double first_video = c->first_video().get() + _pts_offset;
108 double const old_first_video = first_video;
110 /* Round the first video up to a frame boundary */
111 if (fabs (rint (first_video * c->video_frame_rate()) - first_video * c->video_frame_rate()) > 1e-6) {
112 first_video = ceil (first_video * c->video_frame_rate()) / c->video_frame_rate ();
115 _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_after_3d_combine ();
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 LOG_ERROR (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() && _ffmpeg_content->audio_stream()->uses_index (_format_context, si) && _decode_audio) {
178 decode_audio_packet ();
179 } else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si) && 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_U8:
206 uint8_t* p = reinterpret_cast<uint8_t *> (data[0]);
209 for (int i = 0; i < total_samples; ++i) {
210 audio->data(channel)[sample] = float(*p++) / (1 << 23);
213 if (channel == _ffmpeg_content->audio_channels()) {
221 case AV_SAMPLE_FMT_S16:
223 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
226 for (int i = 0; i < total_samples; ++i) {
227 audio->data(channel)[sample] = float(*p++) / (1 << 15);
230 if (channel == _ffmpeg_content->audio_channels()) {
238 case AV_SAMPLE_FMT_S16P:
240 int16_t** p = reinterpret_cast<int16_t **> (data);
241 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
242 for (int j = 0; j < frames; ++j) {
243 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
249 case AV_SAMPLE_FMT_S32:
251 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
254 for (int i = 0; i < total_samples; ++i) {
255 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
258 if (channel == _ffmpeg_content->audio_channels()) {
266 case AV_SAMPLE_FMT_FLT:
268 float* p = reinterpret_cast<float*> (data[0]);
271 for (int i = 0; i < total_samples; ++i) {
272 audio->data(channel)[sample] = *p++;
275 if (channel == _ffmpeg_content->audio_channels()) {
283 case AV_SAMPLE_FMT_FLTP:
285 float** p = reinterpret_cast<float**> (data);
286 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
287 memcpy (audio->data(i), p[i], frames * sizeof(float));
293 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
300 FFmpegDecoder::audio_sample_format () const
302 if (!_ffmpeg_content->audio_stream()) {
303 return (AVSampleFormat) 0;
306 return audio_codec_context()->sample_fmt;
310 FFmpegDecoder::bytes_per_audio_sample () const
312 return av_get_bytes_per_sample (audio_sample_format ());
316 FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
318 double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base);
320 /* If we are doing an accurate seek, our initial shot will be 5 frames (5 being
321 a number plucked from the air) earlier than we want to end up. The loop below
322 will hopefully then step through to where we want to be.
334 /* Initial seek time in the stream's timebase */
335 int64_t const initial_vt = ((initial / _ffmpeg_content->original_video_frame_rate()) - _pts_offset) / time_base;
337 av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD);
339 avcodec_flush_buffers (video_codec_context());
340 if (_subtitle_codec_context) {
341 avcodec_flush_buffers (_subtitle_codec_context);
344 /* This !accurate is piling hack upon hack; setting _just_sought to true
345 even with accurate == true defeats our attempt to align the start
346 of the video and audio. Here we disable that defeat when accurate == true
347 i.e. when we are making a DCP rather than just previewing one.
348 Ewww. This should be gone in 2.0.
354 _video_position = frame;
356 if (frame == 0 || !accurate) {
357 /* We're already there, or we're as close as we need to be */
362 int r = av_read_frame (_format_context, &_packet);
367 if (_packet.stream_index != _video_stream) {
368 av_free_packet (&_packet);
373 r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
374 if (r >= 0 && finished) {
375 _video_position = rint (
376 (av_frame_get_best_effort_timestamp (_frame) * time_base + _pts_offset) * _ffmpeg_content->original_video_frame_rate()
379 if (_video_position >= (frame - 1)) {
380 av_free_packet (&_packet);
385 av_free_packet (&_packet);
388 /* _video_position should be the next thing to be emitted, which will the one after the thing
395 FFmpegDecoder::decode_audio_packet ()
397 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
401 AVPacket copy_packet = _packet;
403 while (copy_packet.size > 0) {
406 int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet);
407 if (decode_result < 0) {
408 shared_ptr<const Film> film = _film.lock ();
410 LOG_ERROR ("avcodec_decode_audio4 failed (%1)", decode_result);
414 if (frame_finished) {
416 if (_audio_position == 0) {
417 /* Where we are in the source, in seconds */
418 double const pts = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
419 * av_frame_get_best_effort_timestamp(_frame) + _pts_offset;
422 /* Emit some silence */
423 int64_t frames = pts * _ffmpeg_content->content_audio_frame_rate ();
425 int64_t const this_time = min (frames, (int64_t) _ffmpeg_content->content_audio_frame_rate() / 2);
427 shared_ptr<AudioBuffers> silence (
428 new AudioBuffers (_ffmpeg_content->audio_channels(), this_time)
431 silence->make_silent ();
432 audio (silence, _audio_position);
438 int const data_size = av_samples_get_buffer_size (
439 0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1
442 audio (deinterleave_audio (_frame->data, data_size), _audio_position);
445 copy_packet.data += decode_result;
446 copy_packet.size -= decode_result;
451 FFmpegDecoder::decode_video_packet ()
454 if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
458 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
460 shared_ptr<FilterGraph> graph;
462 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
463 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
467 if (i == _filter_graphs.end ()) {
468 shared_ptr<const Film> film = _film.lock ();
471 graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
472 _filter_graphs.push_back (graph);
474 LOG_GENERAL (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format);
479 list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
481 shared_ptr<const Film> film = _film.lock ();
484 for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
486 shared_ptr<Image> image = i->first;
488 if (i->second != AV_NOPTS_VALUE) {
490 double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset;
493 /* We just did a seek, so disable any attempts to correct for where we
496 _video_position = rint (pts * _ffmpeg_content->original_video_frame_rate ());
497 _just_sought = false;
500 double const next = _video_position / _ffmpeg_content->original_video_frame_rate();
501 double const one_frame = 1 / _ffmpeg_content->original_video_frame_rate ();
502 double delta = pts - next;
504 while (delta > one_frame) {
505 /* This PTS is more than one frame forward in time of where we think we should be; emit
509 /* XXX: I think this should be a copy of the last frame... */
510 boost::shared_ptr<Image> black (
512 static_cast<AVPixelFormat> (_frame->format),
513 libdcp::Size (video_codec_context()->width, video_codec_context()->height),
518 shared_ptr<const Film> film = _film.lock ();
521 black->make_black ();
522 video (shared_ptr<ImageProxy> (new RawImageProxy (image, film->log())), false, _video_position);
526 if (delta > -one_frame) {
527 /* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */
528 video (shared_ptr<ImageProxy> (new RawImageProxy (image, film->log())), false, _video_position);
532 LOG_WARNING ("Dropping frame without PTS");
541 FFmpegDecoder::setup_subtitle ()
543 boost::mutex::scoped_lock lm (_mutex);
545 if (!_ffmpeg_content->subtitle_stream()) {
549 _subtitle_codec_context = _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec;
550 if (_subtitle_codec_context == 0) {
551 throw DecodeError (N_("could not find subtitle stream"));
554 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
556 if (_subtitle_codec == 0) {
557 throw DecodeError (N_("could not find subtitle decoder"));
560 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
561 throw DecodeError (N_("could not open subtitle decoder"));
566 FFmpegDecoder::done () const
568 bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length());
569 bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length());
574 FFmpegDecoder::decode_subtitle_packet ()
578 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
582 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
583 indicate that the previous subtitle should stop.
585 if (sub.num_rects <= 0) {
586 subtitle (shared_ptr<Image> (), dcpomatic::Rect<double> (), 0, 0);
588 } else if (sub.num_rects > 1) {
589 throw DecodeError (_("multi-part subtitles not yet supported"));
592 /* Subtitle PTS in seconds (within the source, not taking into account any of the
593 source that we may have chopped off for the DCP)
595 double const packet_time = (static_cast<double> (sub.pts ) / AV_TIME_BASE) + _pts_offset;
597 /* hence start time for this sub */
598 Time const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ;
599 Time const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ;
601 AVSubtitleRect const * rect = sub.rects[0];
603 if (rect->type != SUBTITLE_BITMAP) {
604 throw DecodeError (_("non-bitmap subtitles not yet supported"));
607 /* Note RGBA is expressed little-endian, so the first byte in the word is R, second
608 G, third B, fourth A.
610 shared_ptr<Image> image (new Image (PIX_FMT_RGBA, libdcp::Size (rect->w, rect->h), true));
612 /* Start of the first line in the subtitle */
613 uint8_t* sub_p = rect->pict.data[0];
614 /* sub_p looks up into a BGRA palette which is here
615 (i.e. first byte B, second G, third R, fourth A)
617 uint32_t const * palette = (uint32_t *) rect->pict.data[1];
618 /* Start of the output data */
619 uint32_t* out_p = (uint32_t *) image->data()[0];
621 for (int y = 0; y < rect->h; ++y) {
622 uint8_t* sub_line_p = sub_p;
623 uint32_t* out_line_p = out_p;
624 for (int x = 0; x < rect->w; ++x) {
625 uint32_t const p = palette[*sub_line_p++];
626 *out_line_p++ = ((p & 0xff) << 16) | (p & 0xff00) | ((p & 0xff0000) >> 16) | (p & 0xff000000);
628 sub_p += rect->pict.linesize[0];
629 out_p += image->stride()[0] / sizeof (uint32_t);
632 libdcp::Size const vs = _ffmpeg_content->video_size ();
636 dcpomatic::Rect<double> (
637 static_cast<double> (rect->x) / vs.width,
638 static_cast<double> (rect->y) / vs.height,
639 static_cast<double> (rect->w) / vs.width,
640 static_cast<double> (rect->h) / vs.height
647 avsubtitle_free (&sub);