X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=2f46f17c649c21b32a97ebcd3936bbec5d612750;hb=f17c1575ec037e3ac2ec95f2725335ca306fd700;hp=a51b521d0fe22918010a6976c5ae754dd9d8e051;hpb=7e690d21278df14b113f3602cbbd43f6214fd614;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index a51b521d0..2f46f17c6 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) 2012-2016 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 . */ @@ -21,215 +22,223 @@ * @brief A decoder using FFmpeg to decode content. */ -#include -#include -#include -#include -#include -#include -#include -extern "C" { -#include -#include -} -#include "film.h" #include "filter.h" #include "exceptions.h" #include "image.h" #include "util.h" #include "log.h" #include "ffmpeg_decoder.h" -#include "filter_graph.h" +#include "subtitle_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 "image_proxy.h" +#include "raw_image_proxy.h" +#include "video_decoder.h" +#include "film.h" +#include "audio_decoder.h" +#include "compose.hpp" +#include "subtitle_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(...) film->log()->log (String::compose (__VA_ARGS__), Log::TYPE_GENERAL); -#define LOG_ERROR(...) film->log()->log (String::compose (__VA_ARGS__), Log::TYPE_ERROR); -#define LOG_WARNING(...) film->log()->log (__VA_ARGS__, Log::TYPE_WARNING); +#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; -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 libdcp::Size; - -FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptr c, bool video, bool audio) - : Decoder (f) - , VideoDecoder (f, c) - , AudioDecoder (f, c) - , SubtitleDecoder (f) - , FFmpeg (c) - , _subtitle_codec_context (0) - , _subtitle_codec (0) - , _decode_video (video) - , _decode_audio (audio) - , _pts_offset (0) - , _just_sought (false) -{ - setup_subtitle (); - - /* 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. +using dcp::Size; - We will do: - audio_pts_to_use = audio_pts_from_ffmpeg + pts_offset; - video_pts_to_use = video_pts_from_ffmpeg + pts_offset; - */ - - bool const have_video = video && c->first_video(); - bool const have_audio = audio && c->audio_stream() && c->audio_stream()->first_audio; - - /* First, make one of them start at 0 */ - - 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(); +FFmpegDecoder::FFmpegDecoder (shared_ptr c, shared_ptr log, bool fast) + : FFmpeg (c) + , _log (log) + , _have_current_subtitle (false) +{ + if (c->video) { + video.reset (new VideoDecoder (this, c, log)); + _pts_offset = pts_offset (c->ffmpeg_audio_streams(), c->first_video(), c->active_video_frame_rate()); + /* 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 (); } - /* Now adjust both so that the video pts starts on a frame */ - if (have_video && have_audio) { - double first_video = c->first_video().get() + _pts_offset; - double const old_first_video = first_video; - - /* Round the first video up to a frame boundary */ - if (fabs (rint (first_video * c->video_frame_rate()) - first_video * c->video_frame_rate()) > 1e-6) { - first_video = ceil (first_video * c->video_frame_rate()) / c->video_frame_rate (); - } - - _pts_offset += first_video - old_first_video; + if (c->audio) { + audio.reset (new AudioDecoder (this, c->audio, log, fast)); } -} -FFmpegDecoder::~FFmpegDecoder () -{ - boost::mutex::scoped_lock lm (_mutex); - - if (_subtitle_codec_context) { - avcodec_close (_subtitle_codec_context); + if (c->subtitle) { + /* XXX: this time here should be the time of the first subtitle, not 0 */ + subtitle.reset (new SubtitleDecoder (this, c->subtitle, log, 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? */ - - if (_decode_video) { - while (decode_video_packet ()) {} - } - - if (_ffmpeg_content->audio_stream() && _decode_audio) { + + while (video && decode_video_packet ()) {} + + if (audio) { decode_audio_packet (); } - /* Stop us being asked for any more data */ - _video_position = _ffmpeg_content->video_length_after_3d_combine (); - _audio_position = _ffmpeg_content->audio_length (); + /* Make sure all streams are the same length and round up to the next video frame */ + + FrameRateChange const frc = _ffmpeg_content->film()->active_frame_rate_change(_ffmpeg_content->position()); + ContentTime full_length (_ffmpeg_content->full_length(), 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().frames_round (vfr); + while (v < f) { + video->emit (shared_ptr (new RawImageProxy (_black_image)), v); + ++v; + } + } + + BOOST_FOREACH (shared_ptr i, _ffmpeg_content->ffmpeg_audio_streams ()) { + ContentTime a = audio->stream_position(i); + 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 (i, silence, a); + a += to_do; + } + } + + if (audio) { + audio->flush (); + } } -void +bool 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)); - shared_ptr film = _film.lock (); - assert (film); - 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; + return true; } - shared_ptr film = _film.lock (); - assert (film); - int const si = _packet.stream_index; - - if (si == _video_stream && _decode_video) { + 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) { - decode_audio_packet (); - } else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si) && film->with_subtitles ()) { + } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index(_format_context, si) && !subtitle->ignore()) { decode_subtitle_packet (); + } else { + decode_audio_packet (); } - av_free_packet (&_packet); + av_packet_unref (&_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. */ 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; } @@ -239,25 +248,25 @@ 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; } @@ -265,132 +274,119 @@ 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)); } void -FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate) +FFmpegDecoder::seek (ContentTime time, bool accurate) { - double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base); + Decoder::seek (time, accurate); - /* If we are doing an accurate seek, our initial shot will be 5 frames (5 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. + /* 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. */ - int initial = frame; - - if (accurate) { - initial -= 5; - } - if (initial < 0) { - initial = 0; - } + ContentTime pre_roll = accurate ? ContentTime::from_seconds (2) : ContentTime (0); + time -= pre_roll; - /* Initial seek time in the stream's timebase */ - int64_t const initial_vt = ((initial / _ffmpeg_content->video_frame_rate()) - _pts_offset) / time_base; + /* XXX: it seems debatable whether PTS should be used here... + http://www.mjbshaw.com/2012/04/seeking-in-ffmpeg-know-your-timestamp.html + */ - av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD); + optional stream; - avcodec_flush_buffers (video_codec_context()); - if (_subtitle_codec_context) { - avcodec_flush_buffers (_subtitle_codec_context); + if (_video_stream) { + stream = _video_stream; + } else { + shared_ptr s = dynamic_pointer_cast (_ffmpeg_content->audio->stream ()); + if (s) { + stream = s->index (_format_context); + } } - /* This !accurate is piling hack upon hack; setting _just_sought to true - even with accurate == true defeats our attempt to align the start - of the video and audio. Here we disable that defeat when accurate == true - i.e. when we are making a DCP rather than just previewing one. - Ewww. This should be gone in 2.0. - */ - if (!accurate) { - _just_sought = true; - } - - _video_position = frame; - - if (frame == 0 || !accurate) { - /* We're already there, or we're as close as we need to be */ - return; + DCPOMATIC_ASSERT (stream); + + 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 + ); - while (1) { - int r = av_read_frame (_format_context, &_packet); - if (r < 0) { - return; - } + if (video_codec_context ()) { + avcodec_flush_buffers (video_codec_context()); + } - if (_packet.stream_index != _video_stream) { - av_free_packet (&_packet); - continue; - } - - int finished = 0; - r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet); - if (r >= 0 && finished) { - _video_position = rint ( - (av_frame_get_best_effort_timestamp (_frame) * time_base + _pts_offset) * _ffmpeg_content->video_frame_rate() - ); + BOOST_FOREACH (shared_ptr i, ffmpeg_content()->ffmpeg_audio_streams()) { + avcodec_flush_buffers (i->stream(_format_context)->codec); + } - if (_video_position >= (frame - 1)) { - av_free_packet (&_packet); - break; - } - } - - av_free_packet (&_packet); + if (subtitle_codec_context ()) { + avcodec_flush_buffers (subtitle_codec_context ()); } - /* _video_position should be the next thing to be emitted, which will the one after the thing - we just saw. - */ - _video_position++; + _have_current_subtitle = false; } void @@ -399,51 +395,88 @@ 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; - + int const stream_index = copy_packet.stream_index; + + /* 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; + } + + if (stream == streams.end ()) { + /* The packet's stream may not be an audio one; just ignore it in this method if so */ + return; + } + 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) { - shared_ptr film = _film.lock (); - assert (film); - 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) { - - if (_audio_position == 0) { - /* Where we are in the source, in seconds */ - double const pts = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base) - * av_frame_get_best_effort_timestamp(_frame) + _pts_offset; - - if (pts > 0) { - /* Emit some silence */ - int64_t frames = pts * _ffmpeg_content->content_audio_frame_rate (); - while (frames > 0) { - int64_t const this_time = min (frames, (int64_t) _ffmpeg_content->content_audio_frame_rate() / 2); - - shared_ptr silence ( - new AudioBuffers (_ffmpeg_content->audio_channels(), this_time) - ); - - silence->make_silent (); - audio (silence, _audio_position); - frames -= this_time; - } - } + 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()); + + 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 (*stream, data, ct); } - - int const data_size = av_samples_get_buffer_size ( - 0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1 - ); - - audio (deinterleave_audio (_frame->data, data_size), _audio_position); } - + copy_packet.data += decode_result; copy_packet.size -= decode_result; } @@ -452,6 +485,8 @@ 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; @@ -459,20 +494,17 @@ FFmpegDecoder::decode_video_packet () boost::mutex::scoped_lock lm (_filter_graphs_mutex); - shared_ptr graph; - - list >::iterator i = _filter_graphs.begin(); - while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) { + 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 ()) { - shared_ptr film = _film.lock (); - assert (film); - - graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)); + graph.reset (new VideoFilterGraph (dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)); + 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 { graph = *i; @@ -480,143 +512,128 @@ FFmpegDecoder::decode_video_packet () list, int64_t> > images = graph->process (_frame); - shared_ptr film = _film.lock (); - assert (film); - 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; - - if (_just_sought) { - /* We just did a seek, so disable any attempts to correct for where we - are / should be. - */ - _video_position = rint (pts * _ffmpeg_content->video_frame_rate ()); - _just_sought = false; - } - - double const next = _video_position / _ffmpeg_content->video_frame_rate(); - double const one_frame = 1 / _ffmpeg_content->video_frame_rate (); - double delta = pts - next; - - while (delta > one_frame) { - /* This PTS is more than one frame forward in time of where we think we should be; emit - a black frame. - */ - - /* XXX: I think this should be a copy of the last frame... */ - boost::shared_ptr black ( - new Image ( - static_cast (_frame->format), - libdcp::Size (video_codec_context()->width, video_codec_context()->height), - true - ) - ); - - shared_ptr film = _film.lock (); - assert (film); - - black->make_black (); - video (shared_ptr (new RawImageProxy (image, film->log())), false, _video_position); - delta -= one_frame; - } - if (delta > -one_frame) { - /* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */ - video (shared_ptr (new RawImageProxy (image, film->log())), false, _video_position); - } - + 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 ( + shared_ptr (new RawImageProxy (image)), + llrint (pts * _ffmpeg_content->active_video_frame_rate ()) + ); } else { - LOG_WARNING ("Dropping frame without PTS"); + LOG_WARNING_NC ("Dropping frame without PTS"); } } return true; } - -void -FFmpegDecoder::setup_subtitle () -{ - boost::mutex::scoped_lock lm (_mutex); - - if (!_ffmpeg_content->subtitle_stream()) { - return; - } - - _subtitle_codec_context = _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec; - if (_subtitle_codec_context == 0) { - throw DecodeError (N_("could not find subtitle stream")); - } - - _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id); - - if (_subtitle_codec == 0) { - throw DecodeError (N_("could not find subtitle decoder")); - } - - if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) { - throw DecodeError (N_("could not open subtitle decoder")); - } -} - -bool -FFmpegDecoder::done () const -{ - bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length()); - bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length()); - return vd && ad; -} - void FFmpegDecoder::decode_subtitle_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; } - /* 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) { + subtitle->emit_stop (min(*_current_subtitle_to, subtitle_period(sub).from + _pts_offset)); + } else { + subtitle->emit_stop (subtitle_period(sub).from + _pts_offset); + } + _have_current_subtitle = false; + } + if (sub.num_rects <= 0) { - subtitle (shared_ptr (), dcpomatic::Rect (), 0, 0); + /* Nothing new in this subtitle */ return; - } else if (sub.num_rects > 1) { - throw DecodeError (_("multi-part subtitles not yet supported")); } - - /* Subtitle PTS in seconds (within the source, not taking into account any of the - source that we may have chopped off for the DCP) - */ - double const packet_time = (static_cast (sub.pts ) / AV_TIME_BASE) + _pts_offset; - - /* hence start time for this sub */ - Time const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ; - Time const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ; - AVSubtitleRect const * rect = sub.rects[0]; + /* 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); + ContentTime from; + from = sub_period.from + _pts_offset; + _have_current_subtitle = true; + if (sub_period.to) { + _current_subtitle_to = *sub_period.to + _pts_offset; + } else { + _current_subtitle_to = optional(); + } + + 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) { - throw DecodeError (_("non-bitmap subtitles not yet supported")); + if (_current_subtitle_to) { + subtitle->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, libdcp::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]; @@ -624,27 +641,52 @@ 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); } - libdcp::Size const vs = _ffmpeg_content->video_size (); + int const target_width = subtitle_codec_context()->width; + int const target_height = subtitle_codec_context()->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 + ); + + subtitle->emit_image_start (from, image, scaled_rect); +} + +void +FFmpegDecoder::decode_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... + */ - subtitle ( - 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 - ), - from, - to + vector bits; + split (bits, ass, is_any_of (",")); + if (bits.size() < 10) { + return; + } + + sub::RawSubtitle base; + list raw = sub::SSAReader::parse_line ( + base, + bits[9], + _ffmpeg_content->video->size().width, + _ffmpeg_content->video->size().height ); - - - avsubtitle_free (&sub); + + BOOST_FOREACH (sub::Subtitle const & i, sub::collect > (raw)) { + subtitle->emit_text_start (from, i); + } }