X-Git-Url: https://main.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=4d10f8ef7400069332ccdde4c08be5172ce642f2;hp=df6b7416b35a2fc521846ed91ec490e6faaef45d;hb=d7ac100c0eb1b5efdcfbec59be870fd869252840;hpb=a306df9145d16046e51e8b7ff5222e341e98fdbd diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index df6b7416b..4d10f8ef7 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 +39,8 @@ #include "film.h" #include "audio_decoder.h" #include "compose.hpp" -#include "subtitle_content.h" +#include "text_content.h" +#include "audio_content.h" #include #include #include @@ -73,33 +74,35 @@ using std::map; using boost::shared_ptr; using boost::is_any_of; using boost::split; +using boost::optional; +using boost::dynamic_pointer_cast; using dcp::Size; 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 (); } if (c->audio) { - audio.reset (new AudioDecoder (this, c->audio, fast, log)); + audio.reset (new AudioDecoder (this, c->audio, log, fast)); } if (c->subtitle) { - subtitle.reset ( - new SubtitleDecoder ( - this, - c->subtitle, - bind (&FFmpegDecoder::image_subtitles_during, this, _1, _2), - bind (&FFmpegDecoder::text_subtitles_during, this, _1, _2) - ) - ); + /* XXX: this time here should be the time of the first subtitle, not 0 */ + subtitle.reset (new TextDecoder (this, c->subtitle, log, ContentTime())); } + + _next_time.resize (_format_context->nb_streams); } void @@ -116,12 +119,47 @@ FFmpegDecoder::flush () if (audio) { decode_audio_packet (); + } + + /* 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) + 1; + 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); + /* 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; + } + } + } + + if (audio) { audio->flush (); } } bool -FFmpegDecoder::pass (PassReason reason, bool accurate) +FFmpegDecoder::pass () { int r = av_read_frame (_format_context, &_packet); @@ -134,7 +172,7 @@ 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 (); @@ -144,11 +182,11 @@ FFmpegDecoder::pass (PassReason reason, bool accurate) int const si = _packet.stream_index; shared_ptr fc = _ffmpeg_content; - if (_video_stream && si == _video_stream.get() && !video->ignore() && (accurate || reason != PASS_REASON_SUBTITLE)) { + 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)) { + } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index(_format_context, si) && !subtitle->ignore()) { decode_subtitle_packet (); - } else if (accurate || reason != PASS_REASON_SUBTITLE) { + } else { decode_audio_packet (); } @@ -174,8 +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: @@ -184,10 +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; } @@ -201,10 +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; } @@ -215,9 +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); } } } @@ -229,10 +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; } @@ -243,9 +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; } } } @@ -257,10 +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; } @@ -273,9 +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); } } @@ -303,17 +343,7 @@ FFmpegDecoder::bytes_per_audio_sample (shared_ptr stream) con void FFmpegDecoder::seek (ContentTime time, bool accurate) { - if (video) { - video->seek (time, accurate); - } - - if (audio) { - audio->seek (time, accurate); - } - - if (subtitle) { - subtitle->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. @@ -326,7 +356,18 @@ FFmpegDecoder::seek (ContentTime time, bool accurate) http://www.mjbshaw.com/2012/04/seeking-in-ffmpeg-know-your-timestamp.html */ - DCPOMATIC_ASSERT (_video_stream); + optional stream; + + if (_video_stream) { + stream = _video_stream; + } else { + shared_ptr s = dynamic_pointer_cast (_ffmpeg_content->audio->stream ()); + if (s) { + stream = s->index (_format_context); + } + } + + DCPOMATIC_ASSERT (stream); ContentTime u = time - _pts_offset; if (u < ContentTime ()) { @@ -334,18 +375,24 @@ FFmpegDecoder::seek (ContentTime time, bool accurate) } av_seek_frame ( _format_context, - _video_stream.get(), - u.seconds() / av_q2d (_format_context->streams[_video_stream.get()]->time_base), + stream.get(), + u.seconds() / av_q2d (_format_context->streams[stream.get()]->time_base), AVSEEK_FLAG_BACKWARD ); - avcodec_flush_buffers (video_codec_context()); + if (video_codec_context ()) { + avcodec_flush_buffers (video_codec_context()); + } - /* XXX: should be flushing audio buffers? */ + BOOST_FOREACH (shared_ptr i, ffmpeg_content()->ffmpeg_audio_streams()) { + avcodec_flush_buffers (i->stream(_format_context)->codec); + } if (subtitle_codec_context ()) { avcodec_flush_buffers (subtitle_codec_context ()); } + + _have_current_subtitle = false; } void @@ -356,11 +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; } @@ -376,7 +424,7 @@ FFmpegDecoder::decode_audio_packet () 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, + 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. @@ -390,23 +438,48 @@ 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 ()))); - data->move (remove, 0, data->frames() - remove); + data->move (data->frames() - remove, remove, 0); data->set_frames (data->frames() - remove); ct += ContentTime::from_frames (remove, (*stream)->frame_rate ()); } - if (data->frames() > 0) { - audio->give (*stream, data, ct); + 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); } } @@ -435,7 +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); @@ -451,7 +525,7 @@ 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->give ( + video->emit ( shared_ptr (new RawImageProxy (image)), llrint (pts * _ffmpeg_content->active_video_frame_rate ()) ); @@ -472,10 +546,18 @@ FFmpegDecoder::decode_subtitle_packet () 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) { + 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) { - /* 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; } @@ -483,14 +565,13 @@ FFmpegDecoder::decode_subtitle_packet () 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; + 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) { @@ -500,39 +581,31 @@ FFmpegDecoder::decode_subtitle_packet () case SUBTITLE_NONE: break; case SUBTITLE_BITMAP: - decode_bitmap_subtitle (rect, period); + 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, period); + decode_ass_subtitle (rect->ass, from); break; } } - avsubtitle_free (&sub); -} - -list -FFmpegDecoder::image_subtitles_during (ContentTimePeriod p, bool starting) const -{ - return _ffmpeg_content->image_subtitles_during (p, starting); -} + if (_current_subtitle_to) { + subtitle->emit_stop (*_current_subtitle_to); + } -list -FFmpegDecoder::text_subtitles_during (ContentTimePeriod p, bool starting) const -{ - return _ffmpeg_content->text_subtitles_during (p, starting); + avsubtitle_free (&sub); } void -FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTimePeriod period) +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 */ @@ -576,8 +649,8 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTimeP 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]; @@ -587,34 +660,48 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTimeP out_p += image->stride()[0] / sizeof (uint32_t); } - dcp::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) / 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 ); - subtitle->give_image (period, image, scaled_rect); + subtitle->emit_image_start (from, image, scaled_rect); } void -FFmpegDecoder::decode_ass_subtitle (string ass, ContentTimePeriod period) +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... */ - 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->give_text (period, i); + subtitle->emit_text_start (from, i); } }