X-Git-Url: https://git.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=698078b5b4339eae552aa67d249d24d24bd82d9d;hb=e5eec6c5e7b96cc6e7697bbf42d3f27546c5ad52;hp=f8152b56e7629021d1a197b633cbe3d1559eaaff;hpb=2651ef8c2fd486332457630d459c1e620e626c1b;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index f8152b56e..698078b5b 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -1,5 +1,5 @@ /* - Copyright (C) 2012-2015 Carl Hetherington + Copyright (C) 2012-2016 Carl Hetherington This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by @@ -29,30 +29,34 @@ #include "ffmpeg_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 "raw_image_proxy.h" #include "film.h" -#include "timer.h" +#include "md5_digester.h" +#include "compose.hpp" +#include +#include +#include +#include extern "C" { #include #include } #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_NC(...) _video_content->film()->log()->log (__VA_ARGS__, Log::TYPE_WARNING); -#define LOG_WARNING(...) _video_content->film()->log()->log (String::compose (__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; @@ -60,84 +64,42 @@ using std::vector; using std::list; using std::min; using std::pair; -using std::make_pair; using std::max; +using std::map; using boost::shared_ptr; -using boost::optional; -using boost::dynamic_pointer_cast; +using boost::is_any_of; +using boost::split; using dcp::Size; -FFmpegDecoder::FFmpegDecoder (shared_ptr c, shared_ptr log) +FFmpegDecoder::FFmpegDecoder (shared_ptr c, shared_ptr log, bool fast) : VideoDecoder (c) - , AudioDecoder (c) + , AudioDecoder (c, fast) , SubtitleDecoder (c) , FFmpeg (c) , _log (log) + , _pts_offset (pts_offset (c->ffmpeg_audio_streams(), c->first_video(), c->video_frame_rate())) { - /* 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: - audio_pts_to_use = audio_pts_from_ffmpeg + pts_offset; - video_pts_to_use = video_pts_from_ffmpeg + pts_offset; - */ - - /* First, make one of them start at 0 */ - - vector > streams = c->ffmpeg_audio_streams (); - - _pts_offset = ContentTime::min (); - - if (c->first_video ()) { - _pts_offset = - c->first_video().get (); - } - - BOOST_FOREACH (shared_ptr i, streams) { - if (i->first_audio) { - _pts_offset = max (_pts_offset, - i->first_audio.get ()); - } - } - - /* If _pts_offset is positive we would be pushing things from a -ve PTS to be played. - I don't think we ever want to do that, as it seems things at -ve PTS are not meant - to be seen (use for alignment bars etc.); see mantis #418. - */ - if (_pts_offset > ContentTime ()) { - _pts_offset = ContentTime (); - } - - /* Now adjust so that the video pts starts on a frame */ - if (c->first_video ()) { - 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; - } } 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 ()) {} - + decode_audio_packet (); AudioDecoder::flush (); } bool -FFmpegDecoder::pass () +FFmpegDecoder::pass (PassReason reason, bool accurate) { int r = av_read_frame (_format_context, &_packet); @@ -152,7 +114,7 @@ FFmpegDecoder::pass () av_strerror (r, buf, sizeof(buf)); LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), buf, r); } - + flush (); return true; } @@ -160,15 +122,15 @@ FFmpegDecoder::pass () int const si = _packet.stream_index; shared_ptr fc = _ffmpeg_content; - if (si == _video_stream && !_ignore_video) { + 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 { + } else if (accurate || reason != PASS_REASON_SUBTITLE) { decode_audio_packet (); } - av_free_packet (&_packet); + av_packet_unref (&_packet); return false; } @@ -176,10 +138,14 @@ 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 (shared_ptr stream, uint8_t** data, int size) +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 + ); + /* Deinterleave and convert to float */ /* total_samples and frames will be rounded down here, so if there are stray samples at the end @@ -192,7 +158,7 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream, uint8_t 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) { @@ -206,10 +172,10 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream, uint8_t } } 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) { @@ -226,7 +192,7 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream, uint8_t case AV_SAMPLE_FMT_S16P: { - int16_t** p = reinterpret_cast (data); + int16_t** p = reinterpret_cast (_frame->data); for (int i = 0; i < stream->channels(); ++i) { for (int j = 0; j < frames; ++j) { audio->data(i)[j] = static_cast(p[i][j]) / (1 << 15); @@ -234,10 +200,10 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream, uint8_t } } 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) { @@ -252,9 +218,20 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream, uint8_t } break; + case AV_SAMPLE_FMT_S32P: + { + int32_t** p = reinterpret_cast (_frame->data); + for (int i = 0; i < stream->channels(); ++i) { + for (int j = 0; j < frames; ++j) { + audio->data(i)[j] = static_cast(p[i][j]) / (1 << 31); + } + } + } + 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) { @@ -268,13 +245,17 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream, uint8_t } } break; - + case AV_SAMPLE_FMT_FLTP: { - float** p = reinterpret_cast (data); - for (int i = 0; i < stream->channels(); ++i) { + 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 (audio->data(i), p[i], frames * sizeof(float)); } + for (int i = _frame->channels; i < stream->channels(); ++i) { + audio->make_silent (i); + } } break; @@ -314,14 +295,17 @@ FFmpegDecoder::seek (ContentTime time, bool accurate) /* XXX: it seems debatable whether PTS should be used here... http://www.mjbshaw.com/2012/04/seeking-in-ffmpeg-know-your-timestamp.html */ - - ContentTime const u = time - _pts_offset; + + ContentTime 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); avcodec_flush_buffers (video_codec_context()); /* XXX: should be flushing audio buffers? */ - + if (subtitle_codec_context ()) { avcodec_flush_buffers (subtitle_codec_context ()); } @@ -333,7 +317,7 @@ 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; /* XXX: inefficient */ @@ -347,7 +331,7 @@ FFmpegDecoder::decode_audio_packet () /* 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; @@ -369,18 +353,26 @@ FFmpegDecoder::decode_audio_packet () } if (frame_finished) { - ContentTime const ct = ContentTime::from_seconds ( + ContentTime ct = ContentTime::from_seconds ( av_frame_get_best_effort_timestamp (_frame) * av_q2d ((*stream)->stream (_format_context)->time_base)) + _pts_offset; - - int const data_size = av_samples_get_buffer_size ( - 0, (*stream)->stream(_format_context)->codec->channels, _frame->nb_samples, audio_sample_format (*stream), 1 - ); - audio (*stream, deinterleave_audio (*stream, _frame->data, data_size), ct); + 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 (data->frames() > 0) { + audio (*stream, data, ct); + } } - + copy_packet.data += decode_result; copy_packet.size -= decode_result; } @@ -396,15 +388,16 @@ 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)); + 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 { @@ -416,12 +409,12 @@ 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)), - rint (pts * _ffmpeg_content->video_frame_rate ()) + llrint (pts * _ffmpeg_content->video_frame_rate ()) ); } else { LOG_WARNING_NC ("Dropping frame without PTS"); @@ -430,7 +423,7 @@ FFmpegDecoder::decode_video_packet () return true; } - + void FFmpegDecoder::decode_subtitle_packet () { @@ -439,14 +432,12 @@ FFmpegDecoder::decode_subtitle_packet () if (avcodec_decode_subtitle2 (subtitle_codec_context(), &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) { return; } - + 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. */ 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 @@ -460,38 +451,40 @@ FFmpegDecoder::decode_subtitle_packet () period.to = sub_period.to.get() + _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 (sub_period.from); + period.to = ffmpeg_content()->subtitle_stream()->find_subtitle_to (subtitle_id (sub)); } - - AVSubtitleRect const * rect = sub.rects[0]; - - switch (rect->type) { - case SUBTITLE_NONE: - break; - case SUBTITLE_BITMAP: - decode_bitmap_subtitle (rect, period); - break; - case SUBTITLE_TEXT: - cout << "XXX: SUBTITLE_TEXT " << rect->text << "\n"; - break; - case SUBTITLE_ASS: - cout << "XXX: SUBTITLE_ASS " << rect->ass << "\n"; - break; + + 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, period); + break; + case SUBTITLE_TEXT: + cout << "XXX: SUBTITLE_TEXT " << rect->text << "\n"; + break; + case SUBTITLE_ASS: + decode_ass_subtitle (rect->ass, period); + break; + } } - + avsubtitle_free (&sub); } list FFmpegDecoder::image_subtitles_during (ContentTimePeriod p, bool starting) const { - return _ffmpeg_content->subtitles_during (p, starting); + return _ffmpeg_content->image_subtitles_during (p, starting); } list -FFmpegDecoder::text_subtitles_during (ContentTimePeriod, bool) const +FFmpegDecoder::text_subtitles_during (ContentTimePeriod p, bool starting) const { - return list (); + return _ffmpeg_content->text_subtitles_during (p, starting); } void @@ -500,28 +493,61 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTimeP /* Note RGBA is expressed little-endian, so the first byte in the word is R, second G, third B, 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_RGBA, 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]; - + for (int y = 0; y < rect->h; ++y) { 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 RGBA?) but it looks right on screen */ + *out_line_p++ = (p.a << 24) | (p.r << 16) | (p.g << 8) | p.b; } +#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 (); dcpomatic::Rect const scaled_rect ( static_cast (rect->x) / vs.width, @@ -529,7 +555,73 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTimeP static_cast (rect->w) / vs.width, static_cast (rect->h) / vs.height ); - + image_subtitle (period, image, scaled_rect); } +void +FFmpegDecoder::decode_ass_subtitle (string ass, ContentTimePeriod period) +{ + /* 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) { + 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()); + } + } + + 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 () + ) + ); + } + } + } + + text_subtitle (period, ss); +}