X-Git-Url: https://main.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=700e2983681750b17ecefcf07cd23a0037eac4ed;hp=68dcd518618eccfb10eac0467e8218e6ff65a3cd;hb=c93d0271dad86dccdbe518dbbe6e2175adce6804;hpb=19f94521139aac13ef8fb4eaa55855b2ada307b4 diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index 68dcd5186..700e29836 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -1,19 +1,20 @@ /* - Copyright (C) 2012-2016 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 . */ @@ -26,16 +27,22 @@ #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 "video_filter_graph.h" #include "audio_buffers.h" #include "ffmpeg_content.h" #include "raw_image_proxy.h" +#include "video_decoder.h" #include "film.h" -#include "md5_digester.h" +#include "audio_decoder.h" #include "compose.hpp" +#include "text_content.h" +#include "audio_content.h" +#include "frame_interval_checker.h" #include #include #include @@ -44,7 +51,6 @@ extern "C" { #include #include } -#include #include #include #include @@ -53,11 +59,6 @@ extern "C" { #include "i18n.h" -#define LOG_GENERAL(...) _log->log (String::compose (__VA_ARGS__), LogEntry::TYPE_GENERAL); -#define LOG_ERROR(...) _log->log (String::compose (__VA_ARGS__), LogEntry::TYPE_ERROR); -#define LOG_WARNING_NC(...) _log->log (__VA_ARGS__, LogEntry::TYPE_WARNING); -#define LOG_WARNING(...) _log->log (String::compose (__VA_ARGS__), LogEntry::TYPE_WARNING); - using std::cout; using std::string; using std::vector; @@ -66,41 +67,126 @@ using std::min; using std::pair; using std::max; using std::map; -using boost::shared_ptr; +using std::shared_ptr; +using std::make_shared; +using std::make_pair; using boost::is_any_of; using boost::split; +using boost::optional; +using std::dynamic_pointer_cast; using dcp::Size; +using namespace dcpomatic; + -FFmpegDecoder::FFmpegDecoder (shared_ptr c, shared_ptr log, bool fast) - : VideoDecoder (c, log) - , AudioDecoder (c->audio, fast, log) - , SubtitleDecoder (c->subtitle) - , FFmpeg (c) - , _log (log) - , _pts_offset (pts_offset (c->ffmpeg_audio_streams(), c->first_video(), c->active_video_frame_rate())) +FFmpegDecoder::FFmpegDecoder (shared_ptr film, shared_ptr c, bool fast) + : FFmpeg (c) + , Decoder (film) { + if (c->video && c->video->use()) { + video = make_shared(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 = {}; + } + + if (c->audio) { + audio = make_shared(this, c->audio, fast); + } + + if (c->only_text()) { + /* XXX: this time here should be the time of the first subtitle, not 0 */ + text.push_back (make_shared(this, c->only_text(), ContentTime())); + } + + for (auto i: c->ffmpeg_audio_streams()) { + _next_time[i] = boost::optional(); + } } -void + +bool FFmpegDecoder::flush () { - /* Get any remaining frames */ + /* Flush video and audio once */ + + bool did_something = false; + if (video) { + AVPacket packet; + av_init_packet (&packet); + packet.data = nullptr; + packet.size = 0; + if (decode_and_process_video_packet(&packet)) { + did_something = true; + } + } - _packet.data = 0; - _packet.size = 0; + for (auto i: ffmpeg_content()->ffmpeg_audio_streams()) { + AVPacket packet; + av_init_packet (&packet); + packet.data = nullptr; + packet.size = 0; + auto result = decode_audio_packet (i, &packet); + if (result.second) { + process_audio_frame (i); + did_something = true; + } + } + + if (did_something) { + /* We want to be called again */ + return false; + } - /* XXX: should we reset _packet.data and size after each *_decode_* call? */ + /* Make sure all streams are the same length and round up to the next video frame */ + + auto 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(); + auto const f = full_length.frames_round (vfr); + auto v = video->position(film()).get_value_or(ContentTime()).frames_round(vfr) + 1; + while (v < f) { + video->emit (film(), shared_ptr (new RawImageProxy (_black_image)), v); + ++v; + } + } + + for (auto i: _ffmpeg_content->ffmpeg_audio_streams ()) { + auto 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) { + auto to_do = min (full_length - a, ContentTime::from_seconds (0.1)); + auto silence = make_shared(i->channels(), to_do.frames_ceil (i->frame_rate())); + silence->make_silent (); + audio->emit (film(), i, silence, a, true); + a += to_do; + } + } + } - while (decode_video_packet ()) {} + if (audio) { + audio->flush (); + } - decode_audio_packet (); - AudioDecoder::flush (); + return true; } + bool -FFmpegDecoder::pass (PassReason reason, bool accurate) +FFmpegDecoder::pass () { - int r = av_read_frame (_format_context, &_packet); + auto packet = av_packet_alloc(); + DCPOMATIC_ASSERT (packet); + + int r = av_read_frame (_format_context, packet); /* AVERROR_INVALIDDATA can apparently be returned sometimes even when av_read_frame has pretty-much succeeded (and hence generated data which should be processed). @@ -111,28 +197,29 @@ FFmpegDecoder::pass (PassReason reason, bool accurate) /* 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 (); - return true; + av_packet_free (&packet); + return flush (); } - int const si = _packet.stream_index; - shared_ptr fc = _ffmpeg_content; + int const si = packet->stream_index; + auto fc = _ffmpeg_content; - if (si == _video_stream && !_ignore_video && (accurate || reason != PASS_REASON_SUBTITLE)) { - decode_video_packet (); - } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index (_format_context, si)) { - decode_subtitle_packet (); - } else if (accurate || reason != PASS_REASON_SUBTITLE) { - decode_audio_packet (); + if (_video_stream && si == _video_stream.get() && video && !video->ignore()) { + decode_and_process_video_packet (packet); + } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index(_format_context, si) && !only_text()->ignore()) { + decode_and_process_subtitle_packet (packet); + } else { + decode_and_process_audio_packet (packet); } - av_packet_unref (&_packet); + av_packet_free (&packet); return false; } + /** @param data pointer to array of pointers to buffers. * Only the first buffer will be used for non-planar data, otherwise there will be one per channel. */ @@ -142,8 +229,12 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream) const DCPOMATIC_ASSERT (bytes_per_audio_sample (stream)); int const size = av_samples_get_buffer_size ( - 0, stream->stream(_format_context)->codec->channels, _frame->nb_samples, audio_sample_format (stream), 1 + 0, stream->stream(_format_context)->codecpar->channels, _frame->nb_samples, audio_sample_format (stream), 1 ); + DCPOMATIC_ASSERT (size >= 0); + + /* XXX: can't we just use _frame->nb_samples directly here? */ + /* XXX: can't we use swr_convert() to do the format conversion? */ /* Deinterleave and convert to float */ @@ -151,8 +242,10 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream) const 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 frames = total_samples / stream->channels(); - shared_ptr audio (new AudioBuffers (stream->channels(), frames)); + int const channels = stream->channels(); + int const frames = total_samples / channels; + auto audio = make_shared(channels, frames); + auto data = audio->data(); switch (audio_sample_format (stream)) { case AV_SAMPLE_FMT_U8: @@ -161,10 +254,10 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream) const 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 == stream->channels()) { + if (channel == channels) { channel = 0; ++sample; } @@ -178,10 +271,10 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream) const 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 == stream->channels()) { + if (channel == channels) { channel = 0; ++sample; } @@ -192,9 +285,9 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream) const case AV_SAMPLE_FMT_S16P: { int16_t** p = reinterpret_cast (_frame->data); - for (int i = 0; i < stream->channels(); ++i) { + 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); } } } @@ -206,10 +299,10 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream) const 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 == stream->channels()) { + if (channel == channels) { channel = 0; ++sample; } @@ -220,9 +313,9 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream) const case AV_SAMPLE_FMT_S32P: { int32_t** p = reinterpret_cast (_frame->data); - for (int i = 0; i < stream->channels(); ++i) { + for (int i = 0; i < channels; ++i) { for (int j = 0; j < frames; ++j) { - audio->data(i)[j] = static_cast(p[i][j]) / (1 << 31); + data[i][j] = static_cast(p[i][j]) / 2147483648; } } } @@ -234,10 +327,10 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream) const 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 == stream->channels()) { + if (channel == channels) { channel = 0; ++sample; } @@ -248,11 +341,12 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream) const case AV_SAMPLE_FMT_FLTP: { float** p = reinterpret_cast (_frame->data); + DCPOMATIC_ASSERT (_frame->channels <= channels); /* Sometimes there aren't as many channels in the _frame as in the stream */ for (int i = 0; i < _frame->channels; ++i) { - memcpy (audio->data(i), p[i], frames * sizeof(float)); + memcpy (data[i], p[i], frames * sizeof(float)); } - for (int i = _frame->channels; i < stream->channels(); ++i) { + for (int i = _frame->channels; i < channels; ++i) { audio->make_silent (i); } } @@ -265,137 +359,244 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream) const return audio; } + AVSampleFormat FFmpegDecoder::audio_sample_format (shared_ptr stream) const { - return stream->stream (_format_context)->codec->sample_fmt; + return static_cast(stream->stream(_format_context)->codecpar->format); } + int FFmpegDecoder::bytes_per_audio_sample (shared_ptr stream) const { return av_get_bytes_per_sample (audio_sample_format (stream)); } + void FFmpegDecoder::seek (ContentTime time, bool accurate) { - VideoDecoder::seek (time, accurate); - AudioDecoder::seek (time, accurate); - SubtitleDecoder::seek (time, accurate); + Decoder::seek (time, accurate); /* 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. */ - ContentTime pre_roll = accurate ? ContentTime::from_seconds (2) : ContentTime (0); + auto pre_roll = accurate ? ContentTime::from_seconds (2) : ContentTime (0); time -= pre_roll; /* XXX: it seems debatable whether PTS should be used here... http://www.mjbshaw.com/2012/04/seeking-in-ffmpeg-know-your-timestamp.html */ - ContentTime u = time - _pts_offset; + optional stream; + + if (_video_stream) { + stream = _video_stream; + } else { + DCPOMATIC_ASSERT (_ffmpeg_content->audio); + auto s = dynamic_pointer_cast(_ffmpeg_content->audio->stream()); + if (s) { + stream = s->index (_format_context); + } + } + + DCPOMATIC_ASSERT (stream); + + auto u = time - _pts_offset; if (u < ContentTime ()) { u = ContentTime (); } - av_seek_frame (_format_context, _video_stream, u.seconds() / av_q2d (_format_context->streams[_video_stream]->time_base), AVSEEK_FLAG_BACKWARD); + av_seek_frame ( + _format_context, + stream.get(), + u.seconds() / av_q2d (_format_context->streams[stream.get()]->time_base), + AVSEEK_FLAG_BACKWARD + ); - avcodec_flush_buffers (video_codec_context()); + { + /* Force re-creation of filter graphs to reset them and hence to make sure + they don't have any pre-seek frames knocking about. + */ + boost::mutex::scoped_lock lm (_filter_graphs_mutex); + _filter_graphs.clear (); + } - /* XXX: should be flushing audio buffers? */ + if (video_codec_context ()) { + avcodec_flush_buffers (video_codec_context()); + } + +DCPOMATIC_DISABLE_WARNINGS + for (auto i: ffmpeg_content()->ffmpeg_audio_streams()) { + avcodec_flush_buffers (i->stream(_format_context)->codec); + } +DCPOMATIC_ENABLE_WARNINGS if (subtitle_codec_context ()) { avcodec_flush_buffers (subtitle_codec_context ()); } -} -void -FFmpegDecoder::decode_audio_packet () -{ - /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4 - several times. - */ + _have_current_subtitle = false; + + for (auto& i: _next_time) { + i.second = boost::optional(); + } +} - AVPacket copy_packet = _packet; +shared_ptr +FFmpegDecoder::audio_stream_from_index (int index) const +{ /* XXX: inefficient */ - vector > streams = ffmpeg_content()->ffmpeg_audio_streams (); - vector >::const_iterator stream = streams.begin (); - while (stream != streams.end () && !(*stream)->uses_index (_format_context, copy_packet.stream_index)) { + auto streams = ffmpeg_content()->ffmpeg_audio_streams(); + auto stream = streams.begin(); + while (stream != streams.end() && !(*stream)->uses_index(_format_context, index)) { ++stream; } if (stream == streams.end ()) { - /* The packet's stream may not be an audio one; just ignore it in this method if so */ + return {}; + } + + return *stream; +} + + +void +FFmpegDecoder::process_audio_frame (shared_ptr stream) +{ + auto data = deinterleave_audio (stream); + + ContentTime ct; + if (_frame->pts == AV_NOPTS_VALUE) { + /* 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). + */ + if (_next_time[stream]) { + ct = *_next_time[stream]; + } + } else { + ct = ContentTime::from_seconds ( + _frame->best_effort_timestamp * + av_q2d (stream->stream(_format_context)->time_base)) + + _pts_offset; + } + + _next_time[stream] = ct + ContentTime::from_frames(data->frames(), stream->frame_rate()); + + if (ct < ContentTime()) { + /* Discard audio data that comes before time 0 */ + auto 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 (ts=%4 tb=%5, off=%6)", + to_string(ct), + data->frames(), + stream->id(), + _frame->best_effort_timestamp, + 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); + } +} + + +pair +FFmpegDecoder::decode_audio_packet (shared_ptr stream, AVPacket* packet) +{ + int frame_finished; + DCPOMATIC_DISABLE_WARNINGS + int decode_result = avcodec_decode_audio4 (stream->stream(_format_context)->codec, _frame, &frame_finished, packet); + DCPOMATIC_ENABLE_WARNINGS + if (decode_result < 0) { + /* 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); + } + return make_pair(decode_result, frame_finished); +} + + +void +FFmpegDecoder::decode_and_process_audio_packet (AVPacket* packet) +{ + auto stream = audio_stream_from_index (packet->stream_index); + if (!stream) { return; } - while (copy_packet.size > 0) { + /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4 + several times. Make a simple copy so we can alter data and size. + */ + AVPacket copy_packet = *packet; - int frame_finished; - int decode_result = avcodec_decode_audio4 ((*stream)->stream (_format_context)->codec, _frame, &frame_finished, ©_packet); - if (decode_result < 0) { + while (copy_packet.size > 0) { + auto result = decode_audio_packet (stream, ©_packet); + if (result.first < 0) { /* 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, + 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 ct = ContentTime::from_seconds ( - av_frame_get_best_effort_timestamp (_frame) * - av_q2d ((*stream)->stream (_format_context)->time_base)) - + _pts_offset; - - shared_ptr data = deinterleave_audio (*stream); - - 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 (remove, 0, data->frames() - remove); - data->set_frames (data->frames() - remove); - ct += ContentTime::from_frames (remove, (*stream)->frame_rate ()); - } + if (result.second) { + process_audio_frame (stream); + } - if (data->frames() > 0) { - audio (*stream, data, ct); - } + if (result.first) { + break; } - copy_packet.data += decode_result; - copy_packet.size -= decode_result; + copy_packet.data += result.first; + copy_packet.size -= result.first; } } + bool -FFmpegDecoder::decode_video_packet () +FFmpegDecoder::decode_and_process_video_packet (AVPacket* packet) { + DCPOMATIC_ASSERT (_video_stream); + int frame_finished; - if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) { +DCPOMATIC_DISABLE_WARNINGS + if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, packet) < 0 || !frame_finished) { return false; } +DCPOMATIC_ENABLE_WARNINGS boost::mutex::scoped_lock lm (_filter_graphs_mutex); shared_ptr graph; - list >::iterator i = _filter_graphs.begin(); + auto 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 VideoFilterGraph (dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)); + dcp::Fraction vfr (lrint(_ffmpeg_content->video_frame_rate().get() * 1000), 1000); + graph = make_shared(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); @@ -403,17 +604,19 @@ FFmpegDecoder::decode_video_packet () graph = *i; } - list, int64_t> > images = graph->process (_frame); + auto images = graph->process (_frame); - for (list, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) { + for (auto const& i: images) { - shared_ptr image = i->first; + auto 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)), - llrint (pts * _ffmpeg_content->active_video_frame_rate ()) + if (i.second != AV_NOPTS_VALUE) { + double const pts = i.second * av_q2d(_format_context->streams[_video_stream.get()]->time_base) + _pts_offset.seconds(); + + video->emit ( + film(), + make_shared(image), + llrint(pts * _ffmpeg_content->active_video_frame_rate(film())) ); } else { LOG_WARNING_NC ("Dropping frame without PTS"); @@ -423,100 +626,101 @@ FFmpegDecoder::decode_video_packet () return true; } + void -FFmpegDecoder::decode_subtitle_packet () +FFmpegDecoder::decode_and_process_subtitle_packet (AVPacket* packet) { int got_subtitle; AVSubtitle sub; - if (avcodec_decode_subtitle2 (subtitle_codec_context(), &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) { + if (avcodec_decode_subtitle2 (subtitle_codec_context(), &sub, &got_subtitle, packet) < 0 || !got_subtitle) { return; } + /* 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) { - /* Sometimes we get an empty AVSubtitle, which is used by some codecs to - indicate that the previous subtitle should stop. We can ignore it here. - */ + /* Nothing new in this subtitle */ return; } /* Subtitle PTS (within the source, not taking into account any of the source that we may have chopped off for the DCP). */ - FFmpegSubtitlePeriod sub_period = subtitle_period (sub); - ContentTimePeriod period; - period.from = sub_period.from + _pts_offset; + auto sub_period = subtitle_period (sub); + ContentTime from; + from = sub_period.from + _pts_offset; if (sub_period.to) { - /* We already know the subtitle period `to' time */ - period.to = sub_period.to.get() + _pts_offset; + _current_subtitle_to = *sub_period.to + _pts_offset; } else { - /* We have to look up the `to' time in the stream's records */ - period.to = ffmpeg_content()->subtitle_stream()->find_subtitle_to (subtitle_id (sub)); + _current_subtitle_to = optional(); + _have_current_subtitle = true; } for (unsigned int i = 0; i < sub.num_rects; ++i) { - AVSubtitleRect const * rect = sub.rects[i]; + auto const rect = sub.rects[i]; switch (rect->type) { case SUBTITLE_NONE: break; case SUBTITLE_BITMAP: - decode_bitmap_subtitle (rect, period); + process_bitmap_subtitle (rect, from); break; case SUBTITLE_TEXT: cout << "XXX: SUBTITLE_TEXT " << rect->text << "\n"; break; case SUBTITLE_ASS: - decode_ass_subtitle (rect->ass, period); + process_ass_subtitle (rect->ass, from); break; } } - avsubtitle_free (&sub); -} + if (_current_subtitle_to) { + only_text()->emit_stop (*_current_subtitle_to); + } -list -FFmpegDecoder::image_subtitles_during (ContentTimePeriod p, bool starting) const -{ - return _ffmpeg_content->image_subtitles_during (p, starting); + avsubtitle_free (&sub); } -list -FFmpegDecoder::text_subtitles_during (ContentTimePeriod p, bool starting) const -{ - return _ffmpeg_content->text_subtitles_during (p, starting); -} void -FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTimePeriod period) +FFmpegDecoder::process_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime from) { - /* Note RGBA is expressed little-endian, so the first byte in the word is R, second - G, third B, fourth A. + /* 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 (AV_PIX_FMT_RGBA, dcp::Size (rect->w, rect->h), true)); + auto image = make_shared(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 + auto sub_p = rect->pict.data[0]; + /* sub_p looks up into a BGRA palette which is at rect->pict.data[1]; (i.e. first byte B, second G, third R, fourth A) */ - uint32_t const * palette = (uint32_t *) rect->pict.data[1]; + auto const palette = 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) + auto sub_p = rect->data[0]; + /* sub_p looks up into a BGRA palette which is at rect->data[1]. + (first byte B, second G, third R, fourth A) */ - uint32_t const * palette = (uint32_t *) rect->data[1]; + auto const* palette = 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 (); + auto 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); + RGBA c (palette[2], palette[1], palette[0], palette[3]); + auto j = colour_map.find (c); if (j != colour_map.end ()) { mapped_palette[i] = j->second; } else { @@ -526,101 +730,84 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTimeP */ mapped_palette[i] = c; } + palette += 4; } /* Start of the output data */ - uint32_t* out_p = (uint32_t *) image->data()[0]; + auto out_p = image->data()[0]; for (int y = 0; y < rect->h; ++y) { - uint8_t* sub_line_p = sub_p; - uint32_t* out_line_p = out_p; + auto sub_line_p = sub_p; + auto out_line_p = out_p; for (int x = 0; x < rect->w; ++x) { - RGBA const p = mapped_palette[*sub_line_p++]; - /* XXX: this seems to be wrong to me (isn't the output image RGBA?) but it looks right on screen */ - *out_line_p++ = (p.a << 24) | (p.r << 16) | (p.g << 8) | p.b; + auto const p = mapped_palette[*sub_line_p++]; + *out_line_p++ = p.b; + *out_line_p++ = p.g; + *out_line_p++ = p.r; + *out_line_p++ = p.a; } #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); + out_p += image->stride()[0]; } - dcp::Size const vs = _ffmpeg_content->video->size (); + 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) / vs.width, - static_cast (rect->y) / vs.height, - static_cast (rect->w) / vs.width, - static_cast (rect->h) / vs.height + static_cast(rect->x) / target_width, + static_cast(rect->y) / target_height, + static_cast(rect->w) / target_width, + static_cast(rect->h) / target_height ); - image_subtitle (period, image, scaled_rect); + only_text()->emit_bitmap_start (from, image, scaled_rect); } + void -FFmpegDecoder::decode_ass_subtitle (string ass, ContentTimePeriod period) +FFmpegDecoder::process_ass_subtitle (string ass, ContentTime from) { /* We have no styles and no Format: line, so I'm assuming that FFmpeg produces a single format of Dialogue: lines... */ - vector bits; - split (bits, ass, is_any_of (",")); - if (bits.size() < 10) { + 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, bits[9]); - list subs = sub::collect > (raw); - - /* XXX: lots of this is copied from TextSubtitle; there should probably be some sharing */ - - /* Highest line index in this subtitle */ - int highest = 0; - BOOST_FOREACH (sub::Subtitle i, subs) { - BOOST_FOREACH (sub::Line j, i.lines) { - DCPOMATIC_ASSERT (j.vertical_position.reference && j.vertical_position.reference.get() == sub::TOP_OF_SUBTITLE); - DCPOMATIC_ASSERT (j.vertical_position.line); - highest = max (highest, j.vertical_position.line.get()); - } - } + auto raw = sub::SSAReader::parse_line ( + base, + text, + _ffmpeg_content->video->size().width, + _ffmpeg_content->video->size().height + ); - list ss; - - BOOST_FOREACH (sub::Subtitle i, sub::collect > (sub::SSAReader::parse_line (base, bits[9]))) { - BOOST_FOREACH (sub::Line j, i.lines) { - BOOST_FOREACH (sub::Block k, j.blocks) { - ss.push_back ( - dcp::SubtitleString ( - boost::optional (), - k.italic, - k.bold, - dcp::Colour (255, 255, 255), - /* 48pt is 1/22nd of the screen height */ - 48, - 1, - dcp::Time (i.from.seconds(), 1000), - dcp::Time (i.to.seconds(), 1000), - 0, - dcp::HALIGN_CENTER, - /* This 1.015 is an arbitrary value to lift the bottom sub off the bottom - of the screen a bit to a pleasing degree. - */ - 1.015 - ((1 + highest - j.vertical_position.line.get()) * 1.5 / 22), - dcp::VALIGN_TOP, - dcp::DIRECTION_LTR, - k.text, - static_cast (0), - dcp::Colour (255, 255, 255), - dcp::Time (), - dcp::Time () - ) - ); - } - } + for (auto const& i: sub::collect>(raw)) { + only_text()->emit_plain_start (from, i); } - - text_subtitle (period, ss); }