X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=ce20997c84bcebefdc20dd2f90329155545cc505;hb=a5be11a965c2c38442e4e069874e7e21b5b43a5c;hp=42be8227e95e57ebd0481e5b500566b1fd6cafb4;hpb=2f99a801dfb1668d7efda4eb8a3f5638d2fe20d2;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index 42be8227e..ce20997c8 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -1,19 +1,20 @@ /* - Copyright (C) 2012-2014 Carl Hetherington + Copyright (C) 2012-2018 Carl Hetherington - This program is free software; you can redistribute it and/or modify + This file is part of DCP-o-matic. + + DCP-o-matic is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. - This program is distributed in the hope that it will be useful, + DCP-o-matic is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License - along with this program; if not, write to the Free Software - Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + along with DCP-o-matic. If not, see . */ @@ -21,104 +22,136 @@ * @brief A decoder using FFmpeg to decode content. */ -#include -#include -#include -#include -#include -#include -#include -extern "C" { -#include -#include -} #include "filter.h" #include "exceptions.h" #include "image.h" #include "util.h" #include "log.h" +#include "dcpomatic_log.h" #include "ffmpeg_decoder.h" +#include "text_decoder.h" #include "ffmpeg_audio_stream.h" #include "ffmpeg_subtitle_stream.h" -#include "filter_graph.h" +#include "video_filter_graph.h" #include "audio_buffers.h" #include "ffmpeg_content.h" -#include "image_proxy.h" +#include "raw_image_proxy.h" +#include "video_decoder.h" #include "film.h" +#include "audio_decoder.h" +#include "compose.hpp" +#include "text_content.h" +#include "audio_content.h" +#include +#include +#include +#include +extern "C" { +#include +#include +} +#include +#include +#include +#include +#include +#include #include "i18n.h" -#define LOG_GENERAL(...) _video_content->film()->log()->log (String::compose (__VA_ARGS__), Log::TYPE_GENERAL); -#define LOG_ERROR(...) _video_content->film()->log()->log (String::compose (__VA_ARGS__), Log::TYPE_ERROR); -#define LOG_WARNING(...) _video_content->film()->log()->log (__VA_ARGS__, Log::TYPE_WARNING); using std::cout; using std::string; using std::vector; -using std::stringstream; using std::list; using std::min; using std::pair; +using std::max; +using std::map; using boost::shared_ptr; +using boost::is_any_of; +using boost::split; using boost::optional; using boost::dynamic_pointer_cast; using dcp::Size; -FFmpegDecoder::FFmpegDecoder (shared_ptr c, shared_ptr log) - : VideoDecoder (c) - , AudioDecoder (c) - , SubtitleDecoder (c) - , FFmpeg (c) - , _log (log) +FFmpegDecoder::FFmpegDecoder (shared_ptr film, shared_ptr c, bool fast) + : FFmpeg (c) + , Decoder (film) + , _have_current_subtitle (false) { - /* Audio and video frame PTS values may not start with 0. We want - to fiddle them so that: - - 1. One of them starts at time 0. - 2. The first video PTS value ends up on a frame boundary. - - Then we remove big initial gaps in PTS and we allow our - insertion of black frames to work. - - We will do pts_to_use = pts_from_ffmpeg + pts_offset; - */ - - bool const have_video = c->first_video(); - bool const have_audio = c->audio_stream () && c->audio_stream()->first_audio; - - /* First, make one of them start at 0 */ + if (c->video) { + video.reset (new VideoDecoder (this, c)); + _pts_offset = pts_offset (c->ffmpeg_audio_streams(), c->first_video(), c->active_video_frame_rate(film)); + /* It doesn't matter what size or pixel format this is, it just needs to be black */ + _black_image.reset (new Image (AV_PIX_FMT_RGB24, dcp::Size (128, 128), true)); + _black_image->make_black (); + } else { + _pts_offset = ContentTime (); + } - if (have_audio && have_video) { - _pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get()); - } else if (have_video) { - _pts_offset = - c->first_video().get(); - } else if (have_audio) { - _pts_offset = - c->audio_stream()->first_audio.get(); + if (c->audio) { + audio.reset (new AudioDecoder (this, c->audio, fast)); } - /* Now adjust both so that the video pts starts on a frame */ - if (have_video && have_audio) { - ContentTime first_video = c->first_video().get() + _pts_offset; - ContentTime const old_first_video = first_video; - _pts_offset += first_video.round_up (c->video_frame_rate ()) - old_first_video; + if (c->only_text()) { + /* XXX: this time here should be the time of the first subtitle, not 0 */ + text.push_back (shared_ptr (new TextDecoder (this, c->only_text(), ContentTime()))); } + + _next_time.resize (_format_context->nb_streams); } void FFmpegDecoder::flush () { /* Get any remaining frames */ - + _packet.data = 0; _packet.size = 0; - + /* XXX: should we reset _packet.data and size after each *_decode_* call? */ - - while (decode_video_packet ()) {} - - if (_ffmpeg_content->audio_stream()) { + + while (video && decode_video_packet()) {} + + if (audio) { decode_audio_packet (); - AudioDecoder::flush (); + } + + /* Make sure all streams are the same length and round up to the next video frame */ + + FrameRateChange const frc = film()->active_frame_rate_change(_ffmpeg_content->position()); + ContentTime full_length (_ffmpeg_content->full_length(film()), frc); + full_length = full_length.ceil (frc.source); + if (video) { + double const vfr = _ffmpeg_content->video_frame_rate().get(); + Frame const f = full_length.frames_round (vfr); + Frame v = video->position(film()).frames_round(vfr) + 1; + while (v < f) { + video->emit (film(), shared_ptr (new RawImageProxy (_black_image)), v); + ++v; + } + } + + BOOST_FOREACH (shared_ptr i, _ffmpeg_content->ffmpeg_audio_streams ()) { + ContentTime a = audio->stream_position(film(), i); + /* Unfortunately if a is 0 that really means that we don't know the stream position since + there has been no data on it since the last seek. In this case we'll just do nothing + here. I'm not sure if that's the right idea. + */ + if (a > ContentTime()) { + while (a < full_length) { + ContentTime to_do = min (full_length - a, ContentTime::from_seconds (0.1)); + shared_ptr silence (new AudioBuffers (i->channels(), to_do.frames_ceil (i->frame_rate()))); + silence->make_silent (); + audio->emit (film(), i, silence, a); + a += to_do; + } + } + } + + if (audio) { + audio->flush (); } } @@ -127,12 +160,16 @@ FFmpegDecoder::pass () { int r = av_read_frame (_format_context, &_packet); - if (r < 0) { + /* AVERROR_INVALIDDATA can apparently be returned sometimes even when av_read_frame + has pretty-much succeeded (and hence generated data which should be processed). + Hence it makes sense to continue here in that case. + */ + if (r < 0 && r != AVERROR_INVALIDDATA) { if (r != AVERROR_EOF) { /* Maybe we should fail here, but for now we'll just finish off instead */ char buf[256]; av_strerror (r, buf, sizeof(buf)); - LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), buf, r); + LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), &buf[0], r); } flush (); @@ -140,16 +177,17 @@ FFmpegDecoder::pass () } int const si = _packet.stream_index; - - if (si == _video_stream) { + shared_ptr fc = _ffmpeg_content; + + if (_video_stream && si == _video_stream.get() && !video->ignore()) { decode_video_packet (); - } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, si)) { - decode_audio_packet (); - } else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si)) { + } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index(_format_context, si) && !only_text()->ignore()) { decode_subtitle_packet (); + } else { + decode_audio_packet (); } - av_free_packet (&_packet); + av_packet_unref (&_packet); return false; } @@ -157,47 +195,53 @@ FFmpegDecoder::pass () * Only the first buffer will be used for non-planar data, otherwise there will be one per channel. */ shared_ptr -FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) +FFmpegDecoder::deinterleave_audio (shared_ptr stream) const { - assert (_ffmpeg_content->audio_channels()); - assert (bytes_per_audio_sample()); + DCPOMATIC_ASSERT (bytes_per_audio_sample (stream)); - /* Deinterleave and convert to float */ + int const size = av_samples_get_buffer_size ( + 0, stream->stream(_format_context)->codec->channels, _frame->nb_samples, audio_sample_format (stream), 1 + ); - assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0); + /* Deinterleave and convert to float */ - int const total_samples = size / bytes_per_audio_sample(); - int const frames = total_samples / _ffmpeg_content->audio_channels(); - shared_ptr audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames)); + /* total_samples and frames will be rounded down here, so if there are stray samples at the end + of the block that do not form a complete sample or frame they will be dropped. + */ + int const total_samples = size / bytes_per_audio_sample (stream); + int const channels = stream->channels(); + int const frames = total_samples / channels; + shared_ptr audio (new AudioBuffers (channels, frames)); + float** data = audio->data(); - switch (audio_sample_format()) { + switch (audio_sample_format (stream)) { case AV_SAMPLE_FMT_U8: { - uint8_t* p = reinterpret_cast (data[0]); + uint8_t* p = reinterpret_cast (_frame->data[0]); int sample = 0; int channel = 0; for (int i = 0; i < total_samples; ++i) { - audio->data(channel)[sample] = float(*p++) / (1 << 23); + data[channel][sample] = float(*p++) / (1 << 23); ++channel; - if (channel == _ffmpeg_content->audio_channels()) { + if (channel == channels) { channel = 0; ++sample; } } } break; - + case AV_SAMPLE_FMT_S16: { - int16_t* p = reinterpret_cast (data[0]); + int16_t* p = reinterpret_cast (_frame->data[0]); int sample = 0; int channel = 0; for (int i = 0; i < total_samples; ++i) { - audio->data(channel)[sample] = float(*p++) / (1 << 15); + data[channel][sample] = float(*p++) / (1 << 15); ++channel; - if (channel == _ffmpeg_content->audio_channels()) { + if (channel == channels) { channel = 0; ++sample; } @@ -207,25 +251,25 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) case AV_SAMPLE_FMT_S16P: { - int16_t** p = reinterpret_cast (data); - for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) { + int16_t** p = reinterpret_cast (_frame->data); + for (int i = 0; i < channels; ++i) { for (int j = 0; j < frames; ++j) { - audio->data(i)[j] = static_cast(p[i][j]) / (1 << 15); + data[i][j] = static_cast(p[i][j]) / (1 << 15); } } } break; - + case AV_SAMPLE_FMT_S32: { - int32_t* p = reinterpret_cast (data[0]); + int32_t* p = reinterpret_cast (_frame->data[0]); int sample = 0; int channel = 0; for (int i = 0; i < total_samples; ++i) { - audio->data(channel)[sample] = static_cast(*p++) / (1 << 31); + data[channel][sample] = static_cast(*p++) / 2147483648; ++channel; - if (channel == _ffmpeg_content->audio_channels()) { + if (channel == channels) { channel = 0; ++sample; } @@ -233,215 +277,209 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) } break; + case AV_SAMPLE_FMT_S32P: + { + int32_t** p = reinterpret_cast (_frame->data); + for (int i = 0; i < channels; ++i) { + for (int j = 0; j < frames; ++j) { + data[i][j] = static_cast(p[i][j]) / 2147483648; + } + } + } + break; + case AV_SAMPLE_FMT_FLT: { - float* p = reinterpret_cast (data[0]); + float* p = reinterpret_cast (_frame->data[0]); int sample = 0; int channel = 0; for (int i = 0; i < total_samples; ++i) { - audio->data(channel)[sample] = *p++; + data[channel][sample] = *p++; ++channel; - if (channel == _ffmpeg_content->audio_channels()) { + if (channel == channels) { channel = 0; ++sample; } } } break; - + case AV_SAMPLE_FMT_FLTP: { - float** p = reinterpret_cast (data); - for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) { - memcpy (audio->data(i), p[i], frames * sizeof(float)); + float** p = reinterpret_cast (_frame->data); + /* Sometimes there aren't as many channels in the _frame as in the stream */ + for (int i = 0; i < _frame->channels; ++i) { + memcpy (data[i], p[i], frames * sizeof(float)); + } + for (int i = _frame->channels; i < channels; ++i) { + audio->make_silent (i); } } break; default: - throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast (audio_sample_format()))); + throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast (audio_sample_format (stream)))); } return audio; } AVSampleFormat -FFmpegDecoder::audio_sample_format () const +FFmpegDecoder::audio_sample_format (shared_ptr stream) const { - if (!_ffmpeg_content->audio_stream()) { - return (AVSampleFormat) 0; - } - - return audio_codec_context()->sample_fmt; + return stream->stream (_format_context)->codec->sample_fmt; } int -FFmpegDecoder::bytes_per_audio_sample () const +FFmpegDecoder::bytes_per_audio_sample (shared_ptr stream) const { - return av_get_bytes_per_sample (audio_sample_format ()); + return av_get_bytes_per_sample (audio_sample_format (stream)); } -int -FFmpegDecoder::minimal_run (boost::function, optional, int)> finished) +void +FFmpegDecoder::seek (ContentTime time, bool accurate) { - int frames_read = 0; - optional last_video; - optional last_audio; - - while (!finished (last_video, last_audio, frames_read)) { - int r = av_read_frame (_format_context, &_packet); - if (r < 0) { - /* We should flush our decoders here, possibly yielding a few more frames, - but the consequence of having to do that is too hideous to contemplate. - Instead we give up and say that you can't seek too close to the end - of a file. - */ - return frames_read; - } + Decoder::seek (time, accurate); - ++frames_read; + /* If we are doing an `accurate' seek, we need to use pre-roll, as + we don't really know what the seek will give us. + */ - double const time_base = av_q2d (_format_context->streams[_packet.stream_index]->time_base); + ContentTime pre_roll = accurate ? ContentTime::from_seconds (2) : ContentTime (0); + time -= pre_roll; - if (_packet.stream_index == _video_stream) { + /* XXX: it seems debatable whether PTS should be used here... + http://www.mjbshaw.com/2012/04/seeking-in-ffmpeg-know-your-timestamp.html + */ - av_frame_unref (_frame); - - int got_picture = 0; - r = avcodec_decode_video2 (video_codec_context(), _frame, &got_picture, &_packet); - if (r >= 0 && got_picture) { - last_video = ContentTime::from_seconds (av_frame_get_best_effort_timestamp (_frame) * time_base) + _pts_offset; - } + optional stream; - } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, _packet.stream_index)) { - AVPacket copy_packet = _packet; - while (copy_packet.size > 0) { - - int got_frame; - r = avcodec_decode_audio4 (audio_codec_context(), _frame, &got_frame, &_packet); - if (r >= 0 && got_frame) { - last_audio = ContentTime::from_seconds (av_frame_get_best_effort_timestamp (_frame) * time_base) + _pts_offset; - } - - copy_packet.data += r; - copy_packet.size -= r; - } + if (_video_stream) { + stream = _video_stream; + } else { + shared_ptr s = dynamic_pointer_cast (_ffmpeg_content->audio->stream ()); + if (s) { + stream = s->index (_format_context); } - - av_free_packet (&_packet); } - return frames_read; -} + DCPOMATIC_ASSERT (stream); -bool -FFmpegDecoder::seek_overrun_finished (ContentTime seek, optional last_video, optional last_audio) const -{ - return (last_video && last_video.get() >= seek) || (last_audio && last_audio.get() >= seek); -} - -bool -FFmpegDecoder::seek_final_finished (int n, int done) const -{ - return n == done; -} - -void -FFmpegDecoder::seek_and_flush (ContentTime t) -{ - ContentTime const u = t - _pts_offset; - int64_t s = u.seconds() / av_q2d (_format_context->streams[_video_stream]->time_base); - - if (_ffmpeg_content->audio_stream ()) { - s = min ( - s, int64_t (u.seconds() / av_q2d (_ffmpeg_content->audio_stream()->stream(_format_context)->time_base)) - ); + ContentTime u = time - _pts_offset; + if (u < ContentTime ()) { + u = ContentTime (); } + av_seek_frame ( + _format_context, + stream.get(), + u.seconds() / av_q2d (_format_context->streams[stream.get()]->time_base), + AVSEEK_FLAG_BACKWARD + ); - /* Ridiculous empirical hack */ - s--; - if (s < 0) { - s = 0; + if (video_codec_context ()) { + avcodec_flush_buffers (video_codec_context()); } - av_seek_frame (_format_context, _video_stream, s, 0); - - avcodec_flush_buffers (video_codec_context()); - if (audio_codec_context ()) { - avcodec_flush_buffers (audio_codec_context ()); + BOOST_FOREACH (shared_ptr i, ffmpeg_content()->ffmpeg_audio_streams()) { + avcodec_flush_buffers (i->stream(_format_context)->codec); } + if (subtitle_codec_context ()) { avcodec_flush_buffers (subtitle_codec_context ()); } + + _have_current_subtitle = false; } void -FFmpegDecoder::seek (ContentTime time, bool accurate) +FFmpegDecoder::decode_audio_packet () { - VideoDecoder::seek (time, accurate); - AudioDecoder::seek (time, accurate); - - /* If we are doing an accurate seek, our initial shot will be 2s (2 being - a number plucked from the air) earlier than we want to end up. The loop below - will hopefully then step through to where we want to be. + /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4 + several times. */ - ContentTime pre_roll = accurate ? ContentTime::from_seconds (2) : ContentTime (0); - ContentTime initial_seek = time - pre_roll; - if (initial_seek < ContentTime (0)) { - initial_seek = ContentTime (0); - } - - /* Initial seek time in the video stream's timebase */ - - seek_and_flush (initial_seek); + AVPacket copy_packet = _packet; + int const stream_index = copy_packet.stream_index; - if (!accurate) { - /* That'll do */ - return; + /* XXX: inefficient */ + vector > streams = ffmpeg_content()->ffmpeg_audio_streams (); + vector >::const_iterator stream = streams.begin (); + while (stream != streams.end () && !(*stream)->uses_index (_format_context, stream_index)) { + ++stream; } - int const N = minimal_run (boost::bind (&FFmpegDecoder::seek_overrun_finished, this, time, _1, _2)); - - seek_and_flush (initial_seek); - if (N > 0) { - minimal_run (boost::bind (&FFmpegDecoder::seek_final_finished, this, N - 1, _3)); + if (stream == streams.end ()) { + /* The packet's stream may not be an audio one; just ignore it in this method if so */ + return; } -} -void -FFmpegDecoder::decode_audio_packet () -{ - /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4 - several times. - */ - - AVPacket copy_packet = _packet; - while (copy_packet.size > 0) { int frame_finished; - int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet); - + int decode_result = avcodec_decode_audio4 ((*stream)->stream (_format_context)->codec, _frame, &frame_finished, ©_packet); if (decode_result < 0) { - LOG_ERROR ("avcodec_decode_audio4 failed (%1)", decode_result); - return; + /* avcodec_decode_audio4 can sometimes return an error even though it has decoded + some valid data; for example dca_subframe_footer can return AVERROR_INVALIDDATA + if it overreads the auxiliary data. ffplay carries on if frame_finished is true, + even in the face of such an error, so I think we should too. + + Returning from the method here caused mantis #352. + */ + LOG_WARNING ("avcodec_decode_audio4 failed (%1)", decode_result); + + /* Fudge decode_result so that we come out of the while loop when + we've processed this data. + */ + decode_result = copy_packet.size; } if (frame_finished) { - ContentTime const ct = ContentTime::from_seconds ( - av_frame_get_best_effort_timestamp (_frame) * - av_q2d (_ffmpeg_content->audio_stream()->stream (_format_context)->time_base)) - + _pts_offset; - - int const data_size = av_samples_get_buffer_size ( - 0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1 - ); + shared_ptr data = deinterleave_audio (*stream); + + ContentTime ct; + if (_frame->pts == AV_NOPTS_VALUE && _next_time[stream_index]) { + /* In some streams we see not every frame coming through with a timestamp; for those + that have AV_NOPTS_VALUE we need to work out the timestamp ourselves. This is + particularly noticeable with TrueHD streams (see #1111). + */ + ct = *_next_time[stream_index]; + } else { + ct = ContentTime::from_seconds ( + av_frame_get_best_effort_timestamp (_frame) * + av_q2d ((*stream)->stream (_format_context)->time_base)) + + _pts_offset; + } + + _next_time[stream_index] = ct + ContentTime::from_frames(data->frames(), (*stream)->frame_rate()); - audio (deinterleave_audio (_frame->data, data_size), ct); + if (ct < ContentTime ()) { + /* Discard audio data that comes before time 0 */ + Frame const remove = min (int64_t (data->frames()), (-ct).frames_ceil(double((*stream)->frame_rate ()))); + data->move (data->frames() - remove, remove, 0); + data->set_frames (data->frames() - remove); + ct += ContentTime::from_frames (remove, (*stream)->frame_rate ()); + } + + if (ct < ContentTime()) { + LOG_WARNING ( + "Crazy timestamp %1 for %2 samples in stream %3 packet pts %4 (ts=%5 tb=%6, off=%7)", + to_string(ct), + data->frames(), + copy_packet.stream_index, + copy_packet.pts, + av_frame_get_best_effort_timestamp(_frame), + av_q2d((*stream)->stream(_format_context)->time_base), + to_string(_pts_offset) + ); + } + + /* Give this data provided there is some, and its time is sane */ + if (ct >= ContentTime() && data->frames() > 0) { + audio->emit (film(), *stream, data, ct); + } } - + copy_packet.data += decode_result; copy_packet.size -= decode_result; } @@ -450,6 +488,8 @@ FFmpegDecoder::decode_audio_packet () bool FFmpegDecoder::decode_video_packet () { + DCPOMATIC_ASSERT (_video_stream); + int frame_finished; if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) { return false; @@ -457,15 +497,17 @@ FFmpegDecoder::decode_video_packet () boost::mutex::scoped_lock lm (_filter_graphs_mutex); - shared_ptr graph; - - list >::iterator i = _filter_graphs.begin(); + shared_ptr graph; + + list >::iterator i = _filter_graphs.begin(); while (i != _filter_graphs.end() && !(*i)->can_process (dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) { ++i; } if (i == _filter_graphs.end ()) { - graph.reset (new FilterGraph (_ffmpeg_content, dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)); + dcp::Fraction vfr (lrint(_ffmpeg_content->video_frame_rate().get() * 1000), 1000); + graph.reset (new VideoFilterGraph (dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format, vfr)); + graph->setup (_ffmpeg_content->filters ()); _filter_graphs.push_back (graph); LOG_GENERAL (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format); } else { @@ -477,21 +519,23 @@ FFmpegDecoder::decode_video_packet () for (list, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) { shared_ptr image = i->first; - + if (i->second != AV_NOPTS_VALUE) { - double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset.seconds (); - video ( - shared_ptr (new RawImageProxy (image, _video_content->film()->log())), - rint (pts * _ffmpeg_content->video_frame_rate ()) + double const pts = i->second * av_q2d (_format_context->streams[_video_stream.get()]->time_base) + _pts_offset.seconds (); + + video->emit ( + film(), + shared_ptr (new RawImageProxy (image)), + llrint(pts * _ffmpeg_content->active_video_frame_rate(film())) ); } else { - LOG_WARNING ("Dropping frame without PTS"); + LOG_WARNING_NC ("Dropping frame without PTS"); } } return true; } - + void FFmpegDecoder::decode_subtitle_packet () { @@ -501,40 +545,101 @@ FFmpegDecoder::decode_subtitle_packet () return; } - /* Sometimes we get an empty AVSubtitle, which is used by some codecs to - indicate that the previous subtitle should stop. - */ + /* Stop any current subtitle, either at the time it was supposed to stop, or now if now is sooner */ + if (_have_current_subtitle) { + if (_current_subtitle_to) { + only_text()->emit_stop (min(*_current_subtitle_to, subtitle_period(sub).from + _pts_offset)); + } else { + only_text()->emit_stop (subtitle_period(sub).from + _pts_offset); + } + _have_current_subtitle = false; + } + if (sub.num_rects <= 0) { - image_subtitle (ContentTimePeriod (), shared_ptr (), dcpomatic::Rect ()); + /* Nothing new in this subtitle */ return; - } else if (sub.num_rects > 1) { - throw DecodeError (_("multi-part subtitles not yet supported")); } - + /* Subtitle PTS (within the source, not taking into account any of the - source that we may have chopped off for the DCP) + source that we may have chopped off for the DCP). */ - ContentTimePeriod period = subtitle_period (sub) + _pts_offset; + FFmpegSubtitlePeriod sub_period = subtitle_period (sub); + ContentTime from; + from = sub_period.from + _pts_offset; + if (sub_period.to) { + _current_subtitle_to = *sub_period.to + _pts_offset; + } else { + _current_subtitle_to = optional(); + _have_current_subtitle = true; + } - AVSubtitleRect const * rect = sub.rects[0]; + for (unsigned int i = 0; i < sub.num_rects; ++i) { + AVSubtitleRect const * rect = sub.rects[i]; + + switch (rect->type) { + case SUBTITLE_NONE: + break; + case SUBTITLE_BITMAP: + decode_bitmap_subtitle (rect, from); + break; + case SUBTITLE_TEXT: + cout << "XXX: SUBTITLE_TEXT " << rect->text << "\n"; + break; + case SUBTITLE_ASS: + decode_ass_subtitle (rect->ass, from); + break; + } + } - if (rect->type != SUBTITLE_BITMAP) { - /* XXX */ - // throw DecodeError (_("non-bitmap subtitles not yet supported")); - return; + if (_current_subtitle_to) { + only_text()->emit_stop (*_current_subtitle_to); } - /* Note RGBA is expressed little-endian, so the first byte in the word is R, second - G, third B, fourth A. + avsubtitle_free (&sub); +} + +void +FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime from) +{ + /* Note BGRA is expressed little-endian, so the first byte in the word is B, second + G, third R, fourth A. */ - shared_ptr image (new Image (PIX_FMT_RGBA, dcp::Size (rect->w, rect->h), true)); + shared_ptr image (new Image (AV_PIX_FMT_BGRA, dcp::Size (rect->w, rect->h), true)); +#ifdef DCPOMATIC_HAVE_AVSUBTITLERECT_PICT /* Start of the first line in the subtitle */ uint8_t* sub_p = rect->pict.data[0]; /* sub_p looks up into a BGRA palette which is here (i.e. first byte B, second G, third R, fourth A) */ uint32_t const * palette = (uint32_t *) rect->pict.data[1]; +#else + /* Start of the first line in the subtitle */ + uint8_t* sub_p = rect->data[0]; + /* sub_p looks up into a BGRA palette which is here + (i.e. first byte B, second G, third R, fourth A) + */ + uint32_t const * palette = (uint32_t *) rect->data[1]; +#endif + /* And the stream has a map of those palette colours to colours + chosen by the user; created a `mapped' palette from those settings. + */ + map colour_map = ffmpeg_content()->subtitle_stream()->colours (); + vector mapped_palette (rect->nb_colors); + for (int i = 0; i < rect->nb_colors; ++i) { + RGBA c ((palette[i] & 0xff0000) >> 16, (palette[i] & 0xff00) >> 8, palette[i] & 0xff, (palette[i] & 0xff000000) >> 24); + map::const_iterator j = colour_map.find (c); + if (j != colour_map.end ()) { + mapped_palette[i] = j->second; + } else { + /* This colour was not found in the FFmpegSubtitleStream's colour map; probably because + it is from a project that was created before this stuff was added. Just use the + colour straight from the original palette. + */ + mapped_palette[i] = c; + } + } + /* Start of the output data */ uint32_t* out_p = (uint32_t *) image->data()[0]; @@ -542,31 +647,71 @@ FFmpegDecoder::decode_subtitle_packet () uint8_t* sub_line_p = sub_p; uint32_t* out_line_p = out_p; for (int x = 0; x < rect->w; ++x) { - uint32_t const p = palette[*sub_line_p++]; - *out_line_p++ = ((p & 0xff) << 16) | (p & 0xff00) | ((p & 0xff0000) >> 16) | (p & 0xff000000); + RGBA const p = mapped_palette[*sub_line_p++]; + /* XXX: this seems to be wrong to me (isn't the output image BGRA?) but it looks right on screen */ + *out_line_p++ = (p.a << 24) | (p.b << 16) | (p.g << 8) | p.r; } +#ifdef DCPOMATIC_HAVE_AVSUBTITLERECT_PICT sub_p += rect->pict.linesize[0]; +#else + sub_p += rect->linesize[0]; +#endif out_p += image->stride()[0] / sizeof (uint32_t); } - dcp::Size const vs = _ffmpeg_content->video_size (); - - image_subtitle ( - period, - image, - dcpomatic::Rect ( - static_cast (rect->x) / vs.width, - static_cast (rect->y) / vs.height, - static_cast (rect->w) / vs.width, - static_cast (rect->h) / vs.height - ) + int target_width = subtitle_codec_context()->width; + if (target_width == 0 && video_codec_context()) { + /* subtitle_codec_context()->width == 0 has been seen in the wild but I don't + know if it's supposed to mean something from FFmpeg's point of view. + */ + target_width = video_codec_context()->width; + } + int target_height = subtitle_codec_context()->height; + if (target_height == 0 && video_codec_context()) { + target_height = video_codec_context()->height; + } + DCPOMATIC_ASSERT (target_width); + DCPOMATIC_ASSERT (target_height); + dcpomatic::Rect const scaled_rect ( + static_cast (rect->x) / target_width, + static_cast (rect->y) / target_height, + static_cast (rect->w) / target_width, + static_cast (rect->h) / target_height ); - - avsubtitle_free (&sub); + + only_text()->emit_bitmap_start (from, image, scaled_rect); } -bool -FFmpegDecoder::has_subtitle_during (ContentTimePeriod p) const +void +FFmpegDecoder::decode_ass_subtitle (string ass, ContentTime from) { - return _ffmpeg_content->has_subtitle_during (p); + /* We have no styles and no Format: line, so I'm assuming that FFmpeg + produces a single format of Dialogue: lines... + */ + + int commas = 0; + string text; + for (size_t i = 0; i < ass.length(); ++i) { + if (commas < 9 && ass[i] == ',') { + ++commas; + } else if (commas == 9) { + text += ass[i]; + } + } + + if (text.empty ()) { + return; + } + + sub::RawSubtitle base; + list raw = sub::SSAReader::parse_line ( + base, + text, + _ffmpeg_content->video->size().width, + _ffmpeg_content->video->size().height + ); + + BOOST_FOREACH (sub::Subtitle const & i, sub::collect > (raw)) { + only_text()->emit_plain_start (from, i); + } }