X-Git-Url: https://main.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=ba96d71ff5da78a9fe084d1bd56bbb2d82be56d4;hp=db88562ea09b2bb05001d14b2c031ad018c19f6f;hb=HEAD;hpb=1516214cdc7970797b79bca06b46a2eed16a1da3 diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index db88562ea..e0d9918b2 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -18,31 +18,33 @@ */ + /** @file src/ffmpeg_decoder.cc * @brief A decoder using FFmpeg to decode content. */ -#include "filter.h" -#include "exceptions.h" -#include "image.h" -#include "util.h" -#include "log.h" + +#include "audio_buffers.h" +#include "audio_content.h" +#include "audio_decoder.h" +#include "compose.hpp" #include "dcpomatic_log.h" -#include "ffmpeg_decoder.h" -#include "text_decoder.h" +#include "exceptions.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 "raw_image_proxy.h" -#include "video_decoder.h" +#include "ffmpeg_decoder.h" +#include "ffmpeg_subtitle_stream.h" #include "film.h" -#include "audio_decoder.h" -#include "compose.hpp" -#include "text_content.h" -#include "audio_content.h" +#include "filter.h" #include "frame_interval_checker.h" +#include "image.h" +#include "log.h" +#include "raw_image_proxy.h" +#include "text_content.h" +#include "text_decoder.h" +#include "util.h" +#include "video_decoder.h" +#include "video_filter_graph.h" #include #include #include @@ -52,28 +54,22 @@ extern "C" { #include } #include -#include #include #include +#include #include #include "i18n.h" + using std::cout; -using std::string; -using std::vector; -using std::list; +using std::dynamic_pointer_cast; +using std::make_shared; using std::min; -using std::pair; -using std::max; -using std::map; using std::shared_ptr; -using std::make_shared; -using std::make_pair; -using boost::is_any_of; -using boost::split; +using std::string; +using std::vector; using boost::optional; -using std::dynamic_pointer_cast; using dcp::Size; using namespace dcpomatic; @@ -81,24 +77,27 @@ using namespace dcpomatic; FFmpegDecoder::FFmpegDecoder (shared_ptr film, shared_ptr c, bool fast) : FFmpeg (c) , Decoder (film) + , _filter_graphs(c->filters(), dcp::Fraction(lrint(_ffmpeg_content->video_frame_rate().get_value_or(24) * 1000), 1000)) { if (c->video && c->video->use()) { video = make_shared(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_shared(AV_PIX_FMT_RGB24, dcp::Size (128, 128), Image::Alignment::PADDED); _black_image->make_black (); } else { _pts_offset = {}; } - if (c->audio) { + if (c->audio && !c->audio->mapping().mapped_output_channels().empty()) { audio = make_shared(this, c->audio, fast); } if (c->only_text()) { - /* XXX: this time here should be the time of the first subtitle, not 0 */ - text.push_back (make_shared(this, c->only_text(), ContentTime())); + text.push_back (make_shared(this, c->only_text())); + /* XXX: we should be calling maybe_set_position() on this TextDecoder, but we can't easily find + * the time of the first subtitle at this point. + */ } for (auto i: c->ffmpeg_audio_streams()) { @@ -107,11 +106,41 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr film, shared_ptr(_flush_state)); + + switch (_flush_state) { + case FlushState::CODECS: + if (flush_codecs() == FlushResult::DONE) { + LOG_DEBUG_PLAYER_NC("Finished flushing codecs"); + _flush_state = FlushState::AUDIO_DECODER; + } + break; + case FlushState::AUDIO_DECODER: + if (audio) { + audio->flush(); + } + LOG_DEBUG_PLAYER_NC("Finished flushing audio decoder"); + _flush_state = FlushState::FILL; + break; + case FlushState::FILL: + if (flush_fill() == FlushResult::DONE) { + LOG_DEBUG_PLAYER_NC("Finished flushing fills"); + return FlushResult::DONE; + } + break; + } + + return FlushResult::AGAIN; +} + +/** @return true if we have finished flushing the codecs */ +FFmpegDecoder::FlushResult +FFmpegDecoder::flush_codecs() +{ bool did_something = false; if (video) { if (decode_and_process_video_packet(nullptr)) { @@ -126,55 +155,56 @@ FFmpegDecoder::flush () /* EOF can happen if we've already sent a flush packet */ throw DecodeError (N_("avcodec_send_packet"), N_("FFmpegDecoder::flush"), r); } - r = avcodec_receive_frame (context, _frame); + r = avcodec_receive_frame (context, audio_frame(i)); if (r >= 0) { process_audio_frame (i); did_something = true; } } - if (did_something) { - /* We want to be called again */ - return false; - } + return did_something ? FlushResult::AGAIN : FlushResult::DONE; +} + +FFmpegDecoder::FlushResult +FFmpegDecoder::flush_fill() +{ /* Make sure all streams are the same length and round up to the next video frame */ + bool did_something = false; + auto 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) { + if (video && !video->ignore()) { double const vfr = _ffmpeg_content->video_frame_rate().get(); auto const f = full_length.frames_round (vfr); - auto v = video->position(film()).get_value_or(ContentTime()).frames_round(vfr) + 1; - while (v < f) { - video->emit (film(), shared_ptr (new RawImageProxy (_black_image)), v); - ++v; + auto const v = video->position(film()).get_value_or(ContentTime()).frames_round(vfr) + 1; + if (v < f) { + video->emit(film(), make_shared(_black_image), v); + did_something = true; } } - for (auto i: _ffmpeg_content->ffmpeg_audio_streams ()) { - auto 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) { + if (audio && !audio->ignore()) { + for (auto i: _ffmpeg_content->ffmpeg_audio_streams ()) { + auto const 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() && a < full_length) { + LOG_DEBUG_PLAYER("Flush inserts silence at %1", to_string(a)); auto to_do = min (full_length - a, ContentTime::from_seconds (0.1)); auto silence = make_shared(i->channels(), to_do.frames_ceil (i->frame_rate())); silence->make_silent (); audio->emit (film(), i, silence, a, true); - a += to_do; + did_something = true; } } } - if (audio) { - audio->flush (); - } - - return true; + return did_something ? FlushResult::AGAIN : FlushResult::DONE; } @@ -191,6 +221,7 @@ FFmpegDecoder::pass () Hence it makes sense to continue here in that case. */ if (r < 0 && r != AVERROR_INVALIDDATA) { + LOG_DEBUG_PLAYER("FFpmegDecoder::pass flushes because av_read_frame returned %1", r); if (r != AVERROR_EOF) { /* Maybe we should fail here, but for now we'll just finish off instead */ char buf[256]; @@ -199,7 +230,7 @@ FFmpegDecoder::pass () } av_packet_free (&packet); - return flush (); + return flush() == FlushResult::DONE; } int const si = packet->stream_index; @@ -209,7 +240,7 @@ FFmpegDecoder::pass () decode_and_process_video_packet (packet); } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index(_format_context, si) && !only_text()->ignore()) { decode_and_process_subtitle_packet (packet); - } else { + } else if (audio) { decode_and_process_audio_packet (packet); } @@ -221,8 +252,9 @@ FFmpegDecoder::pass () /** @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. */ +static shared_ptr -FFmpegDecoder::deinterleave_audio (AVFrame* frame) +deinterleave_audio(AVFrame* frame) { auto format = static_cast(frame->format); @@ -234,6 +266,10 @@ FFmpegDecoder::deinterleave_audio (AVFrame* frame) auto audio = make_shared(channels, frames); auto data = audio->data(); + if (frames == 0) { + return audio; + } + switch (format) { case AV_SAMPLE_FMT_U8: { @@ -328,14 +364,9 @@ FFmpegDecoder::deinterleave_audio (AVFrame* frame) case AV_SAMPLE_FMT_FLTP: { auto 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) { + for (int i = 0; i < channels; ++i) { memcpy (data[i], p[i], frames * sizeof(float)); } - for (int i = frame->channels; i < channels; ++i) { - audio->make_silent (i); - } } break; @@ -402,13 +433,10 @@ 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 (); - } + /* Force re-creation of filter graphs to reset them and hence to make sure + they don't have any pre-seek frames knocking about. + */ + _filter_graphs.clear(); if (video_codec_context ()) { avcodec_flush_buffers (video_codec_context()); @@ -451,10 +479,13 @@ FFmpegDecoder::audio_stream_from_index (int index) const void FFmpegDecoder::process_audio_frame (shared_ptr stream) { - auto data = deinterleave_audio (_frame); + auto frame = audio_frame (stream); + auto data = deinterleave_audio(frame); + + auto const time_base = stream->stream(_format_context)->time_base; ContentTime ct; - if (_frame->pts == AV_NOPTS_VALUE) { + 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). @@ -464,9 +495,17 @@ FFmpegDecoder::process_audio_frame (shared_ptr stream) } } else { ct = ContentTime::from_seconds ( - _frame->best_effort_timestamp * - av_q2d (stream->stream(_format_context)->time_base)) + frame->best_effort_timestamp * + av_q2d(time_base)) + _pts_offset; + LOG_DEBUG_PLAYER( + "Process audio with timestamp %1 (BET %2, timebase %3/%4, (PTS offset %5)", + to_string(ct), + frame->best_effort_timestamp, + time_base.num, + time_base.den, + to_string(_pts_offset) + ); } _next_time[stream] = ct + ContentTime::from_frames(data->frames(), stream->frame_rate()); @@ -485,8 +524,8 @@ FFmpegDecoder::process_audio_frame (shared_ptr stream) to_string(ct), data->frames(), stream->id(), - _frame->best_effort_timestamp, - av_q2d(stream->stream(_format_context)->time_base), + frame->best_effort_timestamp, + av_q2d(time_base), to_string(_pts_offset) ); } @@ -507,19 +546,18 @@ FFmpegDecoder::decode_and_process_audio_packet (AVPacket* packet) } auto context = _codec_context[stream->index(_format_context)]; + auto frame = audio_frame (stream); + LOG_DEBUG_PLAYER("Send audio packet on stream %1", stream->index(_format_context)); int r = avcodec_send_packet (context, packet); if (r < 0) { - /* We could cope with AVERROR(EAGAIN) and re-send the packet but I think it should never happen. - * Likewise I think AVERROR_EOF should not happen. - */ - throw DecodeError (N_("avcodec_send_packet"), N_("FFmpegDecoder::decode_and_process_audio_packet"), r); + LOG_WARNING("avcodec_send_packet returned %1 for an audio packet", r); } - while (r >= 0) { - r = avcodec_receive_frame (context, _frame); + r = avcodec_receive_frame (context, frame); if (r == AVERROR(EAGAIN)) { /* More input is required */ + LOG_DEBUG_PLAYER_NC("EAGAIN after trying to receive audio frame"); return; } @@ -539,42 +577,40 @@ FFmpegDecoder::decode_and_process_video_packet (AVPacket* packet) auto context = video_codec_context(); - int r = avcodec_send_packet (context, packet); - if (r < 0 && !(r == AVERROR_EOF && !packet)) { - /* We could cope with AVERROR(EAGAIN) and re-send the packet but I think it should never happen. - * AVERROR_EOF can happen during flush if we've already sent a flush packet. - */ - throw DecodeError (N_("avcodec_send_packet"), N_("FFmpegDecoder::decode_and_process_video_packet"), r); - } - - r = avcodec_receive_frame (context, _frame); - if (r == AVERROR(EAGAIN) || r == AVERROR_EOF) { - /* More input is required, or no more frames are coming */ - return false; - } - - /* We assume we'll only get one frame here, which I think is safe */ + bool pending = false; + do { + int r = avcodec_send_packet (context, packet); + if (r < 0) { + LOG_WARNING("avcodec_send_packet returned %1 for a video packet", r); + } - boost::mutex::scoped_lock lm (_filter_graphs_mutex); + /* EAGAIN means we should call avcodec_receive_frame and then re-send the same packet */ + pending = r == AVERROR(EAGAIN); + + while (true) { + r = avcodec_receive_frame (context, _video_frame); + if (r == AVERROR(EAGAIN) || r == AVERROR_EOF || (r < 0 && !packet)) { + /* More input is required, no more frames are coming, or we are flushing and there was + * some error which we just want to ignore. + */ + return false; + } else if (r < 0) { + throw DecodeError (N_("avcodec_receive_frame"), N_("FFmpeg::decode_and_process_video_packet"), r); + } - shared_ptr graph; + process_video_frame (); + } + } while (pending); - auto i = _filter_graphs.begin(); - while (i != _filter_graphs.end() && !(*i)->can_process (dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) { - ++i; - } + return true; +} - if (i == _filter_graphs.end ()) { - dcp::Fraction vfr (lrint(_ffmpeg_content->video_frame_rate().get() * 1000), 1000); - graph = make_shared(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); - } else { - graph = *i; - } - auto images = graph->process (_frame); +void +FFmpegDecoder::process_video_frame () +{ + auto graph = _filter_graphs.get(dcp::Size(_video_frame->width, _video_frame->height), static_cast(_video_frame->format)); + auto images = graph->process (_video_frame); for (auto const& i: images) { @@ -592,48 +628,54 @@ FFmpegDecoder::decode_and_process_video_packet (AVPacket* packet) LOG_WARNING_NC ("Dropping frame without PTS"); } } - - return true; } void FFmpegDecoder::decode_and_process_subtitle_packet (AVPacket* packet) { + auto context = subtitle_codec_context(); + if (!context) { + return; + } + int got_subtitle; AVSubtitle sub; - if (avcodec_decode_subtitle2 (subtitle_codec_context(), &sub, &got_subtitle, packet) < 0 || !got_subtitle) { + if (avcodec_decode_subtitle2(context, &sub, &got_subtitle, packet) < 0 || !got_subtitle) { return; } + auto sub_period = subtitle_period (packet, ffmpeg_content()->subtitle_stream()->stream(_format_context), sub); + /* 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) { - only_text()->emit_stop (min(*_current_subtitle_to, subtitle_period(sub).from + _pts_offset)); + only_text()->emit_stop (min(*_current_subtitle_to, sub_period.from + _pts_offset)); } else { - only_text()->emit_stop (subtitle_period(sub).from + _pts_offset); + only_text()->emit_stop (sub_period.from + _pts_offset); } _have_current_subtitle = false; } if (sub.num_rects <= 0) { /* Nothing new in this subtitle */ + avsubtitle_free (&sub); return; } /* Subtitle PTS (within the source, not taking into account any of the source that we may have chopped off for the DCP). */ - auto 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; } + ContentBitmapText bitmap_text(from); for (unsigned int i = 0; i < sub.num_rects; ++i) { auto const rect = sub.rects[i]; @@ -641,7 +683,7 @@ FFmpegDecoder::decode_and_process_subtitle_packet (AVPacket* packet) case SUBTITLE_NONE: break; case SUBTITLE_BITMAP: - process_bitmap_subtitle (rect, from); + bitmap_text.subs.push_back(process_bitmap_subtitle(rect)); break; case SUBTITLE_TEXT: cout << "XXX: SUBTITLE_TEXT " << rect->text << "\n"; @@ -652,6 +694,10 @@ FFmpegDecoder::decode_and_process_subtitle_packet (AVPacket* packet) } } + if (!bitmap_text.subs.empty()) { + only_text()->emit_bitmap_start(bitmap_text); + } + if (_current_subtitle_to) { only_text()->emit_stop (*_current_subtitle_to); } @@ -660,13 +706,13 @@ FFmpegDecoder::decode_and_process_subtitle_packet (AVPacket* packet) } -void -FFmpegDecoder::process_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime from) +BitmapText +FFmpegDecoder::process_bitmap_subtitle (AVSubtitleRect const * rect) { /* Note BGRA is expressed little-endian, so the first byte in the word is B, second G, third R, fourth A. */ - auto image = make_shared(AV_PIX_FMT_BGRA, dcp::Size (rect->w, rect->h), true); + auto image = make_shared(AV_PIX_FMT_BGRA, dcp::Size (rect->w, rect->h), Image::Alignment::PADDED); #ifdef DCPOMATIC_HAVE_AVSUBTITLERECT_PICT /* Start of the first line in the subtitle */ @@ -674,7 +720,7 @@ FFmpegDecoder::process_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime /* 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) */ - auto const palette = rect->pict.data[1]; + auto const* palette = rect->pict.data[1]; #else /* Start of the first line in the subtitle */ auto sub_p = rect->data[0]; @@ -735,16 +781,28 @@ FFmpegDecoder::process_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime if (target_height == 0 && video_codec_context()) { target_height = video_codec_context()->height; } - DCPOMATIC_ASSERT (target_width); - DCPOMATIC_ASSERT (target_height); + + int x_offset = 0; + int y_offset = 0; + if (_ffmpeg_content->video && _ffmpeg_content->video->use()) { + auto const crop = _ffmpeg_content->video->actual_crop(); + target_width -= crop.left + crop.right; + target_height -= crop.top + crop.bottom; + x_offset = -crop.left; + y_offset = -crop.top; + } + + DCPOMATIC_ASSERT(target_width > 0); + DCPOMATIC_ASSERT(target_height > 0); + dcpomatic::Rect const scaled_rect ( - static_cast(rect->x) / target_width, - static_cast(rect->y) / target_height, + static_cast(rect->x + x_offset) / target_width, + static_cast(rect->y + y_offset) / target_height, static_cast(rect->w) / target_width, static_cast(rect->h) / target_height ); - only_text()->emit_bitmap_start (from, image, scaled_rect); + return { image, scaled_rect }; } @@ -770,11 +828,15 @@ FFmpegDecoder::process_ass_subtitle (string ass, ContentTime from) } sub::RawSubtitle base; + auto video_size = _ffmpeg_content->video->size(); + DCPOMATIC_ASSERT(video_size); + auto raw = sub::SSAReader::parse_line ( base, text, - _ffmpeg_content->video->size().width, - _ffmpeg_content->video->size().height + video_size->width, + video_size->height, + sub::Colour(1, 1, 1) ); for (auto const& i: sub::collect>(raw)) {