X-Git-Url: https://main.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=b050fc594a76f97ab72833c38792f62daf4f100d;hp=a54ac96addbd7dd158ea6d13f02d784065fd2e2c;hb=a5d004b0773f633401528392fc28e66d70e13ac8;hpb=9824173a79ce723068296b3a44499101408c24f2 diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index a54ac96ad..b050fc594 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -1,5 +1,5 @@ /* - Copyright (C) 2012-2016 Carl Hetherington + Copyright (C) 2012-2018 Carl Hetherington This file is part of DCP-o-matic. @@ -27,8 +27,9 @@ #include "image.h" #include "util.h" #include "log.h" +#include "dcpomatic_log.h" #include "ffmpeg_decoder.h" -#include "subtitle_decoder.h" +#include "text_decoder.h" #include "ffmpeg_audio_stream.h" #include "ffmpeg_subtitle_stream.h" #include "video_filter_graph.h" @@ -39,8 +40,9 @@ #include "film.h" #include "audio_decoder.h" #include "compose.hpp" -#include "subtitle_content.h" +#include "text_content.h" #include "audio_content.h" +#include "frame_interval_checker.h" #include #include #include @@ -49,7 +51,6 @@ extern "C" { #include #include } -#include #include #include #include @@ -58,11 +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; @@ -71,21 +67,22 @@ using std::min; using std::pair; using std::max; using std::map; -using boost::shared_ptr; +using std::shared_ptr; using boost::is_any_of; using boost::split; using boost::optional; -using boost::dynamic_pointer_cast; +using std::dynamic_pointer_cast; using dcp::Size; +using namespace dcpomatic; -FFmpegDecoder::FFmpegDecoder (shared_ptr c, shared_ptr log) +FFmpegDecoder::FFmpegDecoder (shared_ptr film, shared_ptr c, bool fast) : FFmpeg (c) - , _log (log) + , Decoder (film) , _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()); + if (c->video && c->video->use()) { + 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 (); @@ -94,12 +91,15 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr c, shared_ptr } if (c->audio) { - audio.reset (new AudioDecoder (this, c->audio, log)); + audio.reset (new AudioDecoder (this, c->audio, fast)); } - if (c->subtitle) { - subtitle.reset (new SubtitleDecoder (this, c->subtitle, log)); + 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 @@ -112,7 +112,7 @@ FFmpegDecoder::flush () /* XXX: should we reset _packet.data and size after each *_decode_* call? */ - while (video && decode_video_packet ()) {} + while (video && decode_video_packet()) {} if (audio) { decode_audio_packet (); @@ -120,27 +120,33 @@ FFmpegDecoder::flush () /* 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); + 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().frames_round (vfr); + Frame v = video->position(film()).get_value_or(ContentTime()).frames_round(vfr) + 1; while (v < f) { - video->emit (shared_ptr (new RawImageProxy (_black_image)), v); + 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(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; + for (auto 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, true); + a += to_do; + } } } @@ -173,9 +179,9 @@ FFmpegDecoder::pass () int const si = _packet.stream_index; shared_ptr fc = _ffmpeg_content; - if (_video_stream && si == _video_stream.get() && !video->ignore()) { + if (_video_stream && si == _video_stream.get() && video && !video->ignore()) { decode_video_packet (); - } else if (fc->subtitle_stream() && fc->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 (); @@ -193,9 +199,14 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream) const { DCPOMATIC_ASSERT (bytes_per_audio_sample (stream)); +DCPOMATIC_DISABLE_WARNINGS int const size = av_samples_get_buffer_size ( 0, stream->stream(_format_context)->codec->channels, _frame->nb_samples, audio_sample_format (stream), 1 ); +DCPOMATIC_ENABLE_WARNINGS + + /* 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 */ @@ -203,8 +214,10 @@ 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; + shared_ptr audio (new AudioBuffers (channels, frames)); + float** data = audio->data(); switch (audio_sample_format (stream)) { case AV_SAMPLE_FMT_U8: @@ -213,10 +226,10 @@ 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; } @@ -230,10 +243,10 @@ 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; } @@ -244,9 +257,9 @@ 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); } } } @@ -258,10 +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] = static_cast(*p++) / 2147483648; + data[channel][sample] = static_cast(*p++) / 2147483648; ++channel; - if (channel == stream->channels()) { + if (channel == channels) { channel = 0; ++sample; } @@ -272,9 +285,9 @@ 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]) / 2147483648; + data[i][j] = static_cast(p[i][j]) / 2147483648; } } } @@ -286,10 +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] = *p++; + data[channel][sample] = *p++; ++channel; - if (channel == stream->channels()) { + if (channel == channels) { channel = 0; ++sample; } @@ -300,11 +313,12 @@ 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); } } @@ -320,7 +334,9 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream) const AVSampleFormat FFmpegDecoder::audio_sample_format (shared_ptr stream) const { +DCPOMATIC_DISABLE_WARNINGS return stream->stream (_format_context)->codec->sample_fmt; +DCPOMATIC_ENABLE_WARNINGS } int @@ -350,6 +366,7 @@ FFmpegDecoder::seek (ContentTime time, bool accurate) if (_video_stream) { stream = _video_stream; } else { + DCPOMATIC_ASSERT (_ffmpeg_content->audio); shared_ptr s = dynamic_pointer_cast (_ffmpeg_content->audio->stream ()); if (s) { stream = s->index (_format_context); @@ -369,19 +386,33 @@ FFmpegDecoder::seek (ContentTime time, bool accurate) AVSEEK_FLAG_BACKWARD ); + { + /* 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 (); + } + if (video_codec_context ()) { avcodec_flush_buffers (video_codec_context()); } - BOOST_FOREACH (shared_ptr i, ffmpeg_content()->ffmpeg_audio_streams()) { +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 ()); } _have_current_subtitle = false; + + for (auto& i: _next_time) { + i = optional(); + } } void @@ -392,11 +423,12 @@ FFmpegDecoder::decode_audio_packet () */ 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, copy_packet.stream_index)) { + while (stream != streams.end () && !(*stream)->uses_index (_format_context, stream_index)) { ++stream; } @@ -405,6 +437,7 @@ FFmpegDecoder::decode_audio_packet () return; } +DCPOMATIC_DISABLE_WARNINGS while (copy_packet.size > 0) { int frame_finished; @@ -426,13 +459,26 @@ FFmpegDecoder::decode_audio_packet () } 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); + 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_index]) { + 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 ()))); @@ -442,12 +488,22 @@ FFmpegDecoder::decode_audio_packet () } if (ct < ContentTime()) { - LOG_WARNING ("Crazy timestamp %1", to_string (ct)); + 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) + ); } +DCPOMATIC_ENABLE_WARNINGS /* Give this data provided there is some, and its time is sane */ if (ct >= ContentTime() && data->frames() > 0) { - audio->emit (*stream, data, ct); + audio->emit (film(), *stream, data, ct); } } @@ -462,9 +518,11 @@ FFmpegDecoder::decode_video_packet () DCPOMATIC_ASSERT (_video_stream); int 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); @@ -476,7 +534,8 @@ FFmpegDecoder::decode_video_packet () } 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.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); @@ -492,9 +551,11 @@ FFmpegDecoder::decode_video_packet () 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(), shared_ptr (new RawImageProxy (image)), - llrint (pts * _ffmpeg_content->active_video_frame_rate ()) + llrint(pts * _ffmpeg_content->active_video_frame_rate(film())) ); } else { LOG_WARNING_NC ("Dropping frame without PTS"); @@ -516,9 +577,9 @@ FFmpegDecoder::decode_subtitle_packet () /* 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)); + only_text()->emit_stop (min(*_current_subtitle_to, subtitle_period(sub).from + _pts_offset)); } else { - subtitle->emit_stop (subtitle_period(sub).from + _pts_offset); + only_text()->emit_stop (subtitle_period(sub).from + _pts_offset); } _have_current_subtitle = false; } @@ -534,9 +595,11 @@ FFmpegDecoder::decode_subtitle_packet () 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(); + _have_current_subtitle = true; } for (unsigned int i = 0; i < sub.num_rects; ++i) { @@ -557,31 +620,35 @@ FFmpegDecoder::decode_subtitle_packet () } } + if (_current_subtitle_to) { + only_text()->emit_stop (*_current_subtitle_to); + } + avsubtitle_free (&sub); } void FFmpegDecoder::decode_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)); + 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 + /* 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]; + uint8_t 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) + /* 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]; + uint8_t 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. @@ -589,7 +656,7 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime 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); + RGBA c (palette[2], palette[1], palette[0], palette[3]); map::const_iterator j = colour_map.find (c); if (j != colour_map.end ()) { mapped_palette[i] = j->second; @@ -600,29 +667,43 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime */ mapped_palette[i] = c; } + palette += 4; } /* Start of the output data */ - uint32_t* out_p = (uint32_t *) image->data()[0]; + uint8_t* 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; + uint8_t* 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; + *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]; } - int const target_width = subtitle_codec_context()->width; - int const target_height = subtitle_codec_context()->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, @@ -630,7 +711,7 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime static_cast (rect->h) / target_height ); - subtitle->emit_image_start (from, image, scaled_rect); + only_text()->emit_bitmap_start (from, image, scaled_rect); } void @@ -640,16 +721,29 @@ FFmpegDecoder::decode_ass_subtitle (string ass, ContentTime from) 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 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)) { - subtitle->emit_text_start (from, i); + for (auto const& i: sub::collect > (raw)) { + only_text()->emit_plain_start (from, i); } }