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.
32 #include <libavcodec/avcodec.h>
33 #include <libavformat/avformat.h>
36 #include "exceptions.h"
40 #include "ffmpeg_decoder.h"
41 #include "ffmpeg_audio_stream.h"
42 #include "ffmpeg_subtitle_stream.h"
43 #include "filter_graph.h"
44 #include "audio_buffers.h"
45 #include "ffmpeg_content.h"
46 #include "raw_image_proxy.h"
52 #define LOG_GENERAL(...) _video_content->film()->log()->log (String::compose (__VA_ARGS__), Log::TYPE_GENERAL);
53 #define LOG_ERROR(...) _video_content->film()->log()->log (String::compose (__VA_ARGS__), Log::TYPE_ERROR);
54 #define LOG_WARNING(...) _video_content->film()->log()->log (__VA_ARGS__, Log::TYPE_WARNING);
59 using std::stringstream;
64 using boost::shared_ptr;
65 using boost::optional;
66 using boost::dynamic_pointer_cast;
69 FFmpegDecoder::FFmpegDecoder (shared_ptr<const FFmpegContent> c, shared_ptr<Log> log)
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.
85 We will do pts_to_use = pts_from_ffmpeg + pts_offset;
88 bool const have_video = c->first_video();
89 bool const have_audio = c->audio_stream () && c->audio_stream()->first_audio;
91 /* First, make one of them start at 0 */
93 if (have_audio && have_video) {
94 _pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get());
95 } else if (have_video) {
96 _pts_offset = - c->first_video().get();
97 } else if (have_audio) {
98 _pts_offset = - c->audio_stream()->first_audio.get();
101 /* Now adjust both so that the video pts starts on a frame */
102 if (have_video && have_audio) {
103 ContentTime first_video = c->first_video().get() + _pts_offset;
104 ContentTime const old_first_video = first_video;
105 _pts_offset += first_video.round_up (c->video_frame_rate ()) - old_first_video;
110 FFmpegDecoder::flush ()
112 /* Get any remaining frames */
117 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
119 while (decode_video_packet ()) {}
121 if (_ffmpeg_content->audio_stream()) {
122 decode_audio_packet ();
123 AudioDecoder::flush ();
128 FFmpegDecoder::pass ()
130 int r = av_read_frame (_format_context, &_packet);
133 if (r != AVERROR_EOF) {
134 /* Maybe we should fail here, but for now we'll just finish off instead */
136 av_strerror (r, buf, sizeof(buf));
137 LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), buf, r);
144 int const si = _packet.stream_index;
146 if (si == _video_stream) {
147 decode_video_packet ();
148 } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, si)) {
149 decode_audio_packet ();
150 } else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si)) {
151 decode_subtitle_packet ();
154 av_free_packet (&_packet);
158 /** @param data pointer to array of pointers to buffers.
159 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
161 shared_ptr<AudioBuffers>
162 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
164 assert (_ffmpeg_content->audio_channels());
165 assert (bytes_per_audio_sample());
167 /* Deinterleave and convert to float */
169 assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0);
171 int const total_samples = size / bytes_per_audio_sample();
172 int const frames = total_samples / _ffmpeg_content->audio_channels();
173 shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
175 switch (audio_sample_format()) {
176 case AV_SAMPLE_FMT_U8:
178 uint8_t* p = reinterpret_cast<uint8_t *> (data[0]);
181 for (int i = 0; i < total_samples; ++i) {
182 audio->data(channel)[sample] = float(*p++) / (1 << 23);
185 if (channel == _ffmpeg_content->audio_channels()) {
193 case AV_SAMPLE_FMT_S16:
195 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
198 for (int i = 0; i < total_samples; ++i) {
199 audio->data(channel)[sample] = float(*p++) / (1 << 15);
202 if (channel == _ffmpeg_content->audio_channels()) {
210 case AV_SAMPLE_FMT_S16P:
212 int16_t** p = reinterpret_cast<int16_t **> (data);
213 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
214 for (int j = 0; j < frames; ++j) {
215 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
221 case AV_SAMPLE_FMT_S32:
223 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
226 for (int i = 0; i < total_samples; ++i) {
227 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
230 if (channel == _ffmpeg_content->audio_channels()) {
238 case AV_SAMPLE_FMT_FLT:
240 float* p = reinterpret_cast<float*> (data[0]);
243 for (int i = 0; i < total_samples; ++i) {
244 audio->data(channel)[sample] = *p++;
247 if (channel == _ffmpeg_content->audio_channels()) {
255 case AV_SAMPLE_FMT_FLTP:
257 float** p = reinterpret_cast<float**> (data);
258 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
259 memcpy (audio->data(i), p[i], frames * sizeof(float));
265 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
272 FFmpegDecoder::audio_sample_format () const
274 if (!_ffmpeg_content->audio_stream()) {
275 return (AVSampleFormat) 0;
278 return audio_codec_context()->sample_fmt;
282 FFmpegDecoder::bytes_per_audio_sample () const
284 return av_get_bytes_per_sample (audio_sample_format ());
288 FFmpegDecoder::seek (ContentTime time, bool accurate)
290 VideoDecoder::seek (time, accurate);
291 AudioDecoder::seek (time, accurate);
293 /* If we are doing an `accurate' seek, we need to use pre-roll, as
294 we don't really know what the seek will give us.
297 ContentTime pre_roll = accurate ? ContentTime::from_seconds (2) : ContentTime (0);
299 if (time < ContentTime (0)) {
300 time = ContentTime (0);
303 /* XXX: it seems debatable whether PTS should be used here... */
305 ContentTime const u = time - _pts_offset;
306 int64_t s = u.seconds() / av_q2d (_format_context->streams[_video_stream]->time_base);
308 if (_ffmpeg_content->audio_stream ()) {
310 s, int64_t (u.seconds() / av_q2d (_ffmpeg_content->audio_stream()->stream(_format_context)->time_base))
314 /* Ridiculous empirical hack */
320 av_seek_frame (_format_context, _video_stream, s, 0);
322 avcodec_flush_buffers (video_codec_context());
323 if (audio_codec_context ()) {
324 avcodec_flush_buffers (audio_codec_context ());
326 if (subtitle_codec_context ()) {
327 avcodec_flush_buffers (subtitle_codec_context ());
332 FFmpegDecoder::decode_audio_packet ()
334 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
338 AVPacket copy_packet = _packet;
340 while (copy_packet.size > 0) {
343 int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet);
345 if (decode_result < 0) {
346 LOG_ERROR ("avcodec_decode_audio4 failed (%1)", decode_result);
350 if (frame_finished) {
351 ContentTime const ct = ContentTime::from_seconds (
352 av_frame_get_best_effort_timestamp (_frame) *
353 av_q2d (_ffmpeg_content->audio_stream()->stream (_format_context)->time_base))
356 int const data_size = av_samples_get_buffer_size (
357 0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1
360 audio (deinterleave_audio (_frame->data, data_size), ct);
363 copy_packet.data += decode_result;
364 copy_packet.size -= decode_result;
369 FFmpegDecoder::decode_video_packet ()
372 if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
376 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
378 shared_ptr<FilterGraph> graph;
380 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
381 while (i != _filter_graphs.end() && !(*i)->can_process (dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
385 if (i == _filter_graphs.end ()) {
386 graph.reset (new FilterGraph (_ffmpeg_content, dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
387 _filter_graphs.push_back (graph);
388 LOG_GENERAL (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format);
393 list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
395 for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
397 shared_ptr<Image> image = i->first;
399 if (i->second != AV_NOPTS_VALUE) {
400 double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset.seconds ();
402 shared_ptr<ImageProxy> (new RawImageProxy (image, _video_content->film()->log())),
403 rint (pts * _ffmpeg_content->video_frame_rate ())
406 LOG_WARNING ("Dropping frame without PTS");
414 FFmpegDecoder::decode_subtitle_packet ()
418 if (avcodec_decode_subtitle2 (subtitle_codec_context(), &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
422 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
423 indicate that the previous subtitle should stop.
425 if (sub.num_rects <= 0) {
426 image_subtitle (ContentTimePeriod (), shared_ptr<Image> (), dcpomatic::Rect<double> ());
428 } else if (sub.num_rects > 1) {
429 throw DecodeError (_("multi-part subtitles not yet supported"));
432 /* Subtitle PTS (within the source, not taking into account any of the
433 source that we may have chopped off for the DCP)
435 ContentTimePeriod period = subtitle_period (sub) + _pts_offset;
437 AVSubtitleRect const * rect = sub.rects[0];
439 if (rect->type != SUBTITLE_BITMAP) {
441 // throw DecodeError (_("non-bitmap subtitles not yet supported"));
445 /* Note RGBA is expressed little-endian, so the first byte in the word is R, second
446 G, third B, fourth A.
448 shared_ptr<Image> image (new Image (PIX_FMT_RGBA, dcp::Size (rect->w, rect->h), true));
450 /* Start of the first line in the subtitle */
451 uint8_t* sub_p = rect->pict.data[0];
452 /* sub_p looks up into a BGRA palette which is here
453 (i.e. first byte B, second G, third R, fourth A)
455 uint32_t const * palette = (uint32_t *) rect->pict.data[1];
456 /* Start of the output data */
457 uint32_t* out_p = (uint32_t *) image->data()[0];
459 for (int y = 0; y < rect->h; ++y) {
460 uint8_t* sub_line_p = sub_p;
461 uint32_t* out_line_p = out_p;
462 for (int x = 0; x < rect->w; ++x) {
463 uint32_t const p = palette[*sub_line_p++];
464 *out_line_p++ = ((p & 0xff) << 16) | (p & 0xff00) | ((p & 0xff0000) >> 16) | (p & 0xff000000);
466 sub_p += rect->pict.linesize[0];
467 out_p += image->stride()[0] / sizeof (uint32_t);
470 dcp::Size const vs = _ffmpeg_content->video_size ();
475 dcpomatic::Rect<double> (
476 static_cast<double> (rect->x) / vs.width,
477 static_cast<double> (rect->y) / vs.height,
478 static_cast<double> (rect->w) / vs.width,
479 static_cast<double> (rect->h) / vs.height
483 avsubtitle_free (&sub);
486 list<ContentTimePeriod>
487 FFmpegDecoder::subtitles_during (ContentTimePeriod p, bool starting) const
489 return _ffmpeg_content->subtitles_during (p, starting);