X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=9076464ac73aaa318566a1d1cb8f089be8b1173b;hb=92f024ea58c7279b8096e5e9f60f9cb2613e8a91;hp=a09eab68e12e11763436cd83db1378d710a632cc;hpb=366910025ce80231f1192662efe79f76d78ff572;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index a09eab68e..9076464ac 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. @@ -28,7 +28,7 @@ #include "util.h" #include "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" @@ -39,7 +39,7 @@ #include "film.h" #include "audio_decoder.h" #include "compose.hpp" -#include "subtitle_content.h" +#include "text_content.h" #include "audio_content.h" #include #include @@ -78,7 +78,7 @@ using boost::optional; using boost::dynamic_pointer_cast; using dcp::Size; -FFmpegDecoder::FFmpegDecoder (shared_ptr c, shared_ptr log) +FFmpegDecoder::FFmpegDecoder (shared_ptr c, shared_ptr log, bool fast) : FFmpeg (c) , _log (log) , _have_current_subtitle (false) @@ -94,12 +94,15 @@ 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, log, 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(), log, ContentTime()))); } + + _next_time.resize (_format_context->nb_streams); } void @@ -126,7 +129,7 @@ FFmpegDecoder::flush () 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().frames_round (vfr) + 1; while (v < f) { video->emit (shared_ptr (new RawImageProxy (_black_image)), v); ++v; @@ -135,12 +138,18 @@ FFmpegDecoder::flush () 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; + /* 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 (i, silence, a); + a += to_do; + } } } @@ -175,7 +184,7 @@ FFmpegDecoder::pass () if (_video_stream && si == _video_stream.get() && !video->ignore()) { decode_video_packet (); - } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index(_format_context, si) && !subtitle->ignore()) { + } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index(_format_context, si) && !only_text()->ignore()) { decode_subtitle_packet (); } else { decode_audio_packet (); @@ -203,8 +212,10 @@ 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 +224,10 @@ 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 +241,10 @@ 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 +255,9 @@ 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 +269,10 @@ 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 +283,9 @@ 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 +297,10 @@ 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; } @@ -302,9 +313,9 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream) const 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)); + 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); } } @@ -392,11 +403,12 @@ 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; } @@ -426,13 +438,24 @@ 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 && _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 ()))); @@ -442,7 +465,16 @@ 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) + ); } /* Give this data provided there is some, and its time is sane */ @@ -476,7 +508,8 @@ 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 +525,10 @@ 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 ( shared_ptr (new RawImageProxy (image)), - llrint (pts * _ffmpeg_content->active_video_frame_rate ()) + llrint(pts * _ffmpeg_content->active_video_frame_rate()) ); } else { LOG_WARNING_NC ("Dropping frame without PTS"); @@ -516,9 +550,9 @@ 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 +568,11 @@ 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) { @@ -557,16 +593,20 @@ 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 */ @@ -610,8 +650,8 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime uint32_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; + /* 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]; @@ -630,7 +670,7 @@ 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 +680,29 @@ 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); + only_text()->emit_plain_start (from, i); } }