X-Git-Url: https://git.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=ba96d71ff5da78a9fe084d1bd56bbb2d82be56d4;hp=147bb384f3eb1bad5a811adf9d624501b69eedf4;hb=182b9d2e2feb6545592868606aaf0f0146095481;hpb=c7573918f64657f720c7e9437df0a33e505c88bc diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index 147bb384f..ba96d71ff 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; @@ -81,13 +77,12 @@ using namespace dcpomatic; FFmpegDecoder::FFmpegDecoder (shared_ptr film, shared_ptr c, bool fast) : FFmpeg (c) , Decoder (film) - , _have_current_subtitle (false) { 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 = {}; @@ -98,8 +93,10 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr film, shared_ptronly_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()) { @@ -115,22 +112,20 @@ FFmpegDecoder::flush () bool did_something = false; if (video) { - AVPacket packet; - av_init_packet (&packet); - packet.data = nullptr; - packet.size = 0; - if (decode_and_process_video_packet(&packet)) { + if (decode_and_process_video_packet(nullptr)) { did_something = true; } } for (auto i: ffmpeg_content()->ffmpeg_audio_streams()) { - AVPacket packet; - av_init_packet (&packet); - packet.data = nullptr; - packet.size = 0; - auto result = decode_audio_packet (i, &packet); - if (result.second) { + auto context = _codec_context[i->index(_format_context)]; + int r = avcodec_send_packet (context, nullptr); + if (r < 0 && r != AVERROR_EOF) { + /* 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, audio_frame(i)); + if (r >= 0) { process_audio_frame (i); did_something = true; } @@ -151,7 +146,7 @@ FFmpegDecoder::flush () 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); + video->emit (film(), make_shared(_black_image), v); ++v; } } @@ -225,35 +220,22 @@ 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) const +FFmpegDecoder::deinterleave_audio (AVFrame* frame) { - 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 - DCPOMATIC_ASSERT (size >= 0); + auto format = static_cast(frame->format); - /* 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 */ - - /* total_samples and frames will be rounded down here, so if there are stray samples at the end - 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 channels = stream->channels(); - int const frames = total_samples / channels; + int const channels = frame->channels; + int const frames = frame->nb_samples; + int const total_samples = frames * channels; auto audio = make_shared(channels, frames); auto data = audio->data(); - switch (audio_sample_format (stream)) { + switch (format) { case AV_SAMPLE_FMT_U8: { - uint8_t* p = reinterpret_cast (_frame->data[0]); + auto p = reinterpret_cast (frame->data[0]); int sample = 0; int channel = 0; for (int i = 0; i < total_samples; ++i) { @@ -270,7 +252,7 @@ DCPOMATIC_ENABLE_WARNINGS case AV_SAMPLE_FMT_S16: { - int16_t* p = reinterpret_cast (_frame->data[0]); + auto p = reinterpret_cast (frame->data[0]); int sample = 0; int channel = 0; for (int i = 0; i < total_samples; ++i) { @@ -287,7 +269,7 @@ DCPOMATIC_ENABLE_WARNINGS case AV_SAMPLE_FMT_S16P: { - int16_t** p = reinterpret_cast (_frame->data); + auto p = reinterpret_cast (frame->data); for (int i = 0; i < channels; ++i) { for (int j = 0; j < frames; ++j) { data[i][j] = static_cast(p[i][j]) / (1 << 15); @@ -298,7 +280,7 @@ DCPOMATIC_ENABLE_WARNINGS case AV_SAMPLE_FMT_S32: { - int32_t* p = reinterpret_cast (_frame->data[0]); + auto p = reinterpret_cast (frame->data[0]); int sample = 0; int channel = 0; for (int i = 0; i < total_samples; ++i) { @@ -315,7 +297,7 @@ DCPOMATIC_ENABLE_WARNINGS case AV_SAMPLE_FMT_S32P: { - int32_t** p = reinterpret_cast (_frame->data); + auto p = reinterpret_cast (frame->data); for (int i = 0; i < channels; ++i) { for (int j = 0; j < frames; ++j) { data[i][j] = static_cast(p[i][j]) / 2147483648; @@ -326,7 +308,7 @@ DCPOMATIC_ENABLE_WARNINGS case AV_SAMPLE_FMT_FLT: { - float* p = reinterpret_cast (_frame->data[0]); + auto p = reinterpret_cast (frame->data[0]); int sample = 0; int channel = 0; for (int i = 0; i < total_samples; ++i) { @@ -343,20 +325,20 @@ DCPOMATIC_ENABLE_WARNINGS 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) { + 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) { memcpy (data[i], p[i], frames * sizeof(float)); } - for (int i = _frame->channels; i < channels; ++i) { + for (int i = frame->channels; i < channels; ++i) { audio->make_silent (i); } } break; default: - throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast (audio_sample_format (stream)))); + throw DecodeError (String::compose(_("Unrecognised audio sample format (%1)"), static_cast(format))); } return audio; @@ -366,9 +348,7 @@ DCPOMATIC_ENABLE_WARNINGS AVSampleFormat FFmpegDecoder::audio_sample_format (shared_ptr stream) const { -DCPOMATIC_DISABLE_WARNINGS - return stream->stream (_format_context)->codec->sample_fmt; -DCPOMATIC_ENABLE_WARNINGS + return static_cast(stream->stream(_format_context)->codecpar->format); } @@ -432,11 +412,9 @@ FFmpegDecoder::seek (ContentTime time, bool accurate) avcodec_flush_buffers (video_codec_context()); } -DCPOMATIC_DISABLE_WARNINGS for (auto i: ffmpeg_content()->ffmpeg_audio_streams()) { - avcodec_flush_buffers (i->stream(_format_context)->codec); + avcodec_flush_buffers (_codec_context[i->index(_format_context)]); } -DCPOMATIC_ENABLE_WARNINGS if (subtitle_codec_context ()) { avcodec_flush_buffers (subtitle_codec_context ()); @@ -471,10 +449,13 @@ FFmpegDecoder::audio_stream_from_index (int index) const void FFmpegDecoder::process_audio_frame (shared_ptr stream) { - auto data = deinterleave_audio (stream); + 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). @@ -484,9 +465,17 @@ 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()); @@ -505,8 +494,8 @@ 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) ); } @@ -518,27 +507,6 @@ FFmpegDecoder::process_audio_frame (shared_ptr stream) } -pair -FFmpegDecoder::decode_audio_packet (shared_ptr stream, AVPacket* packet) -{ - int frame_finished; - DCPOMATIC_DISABLE_WARNINGS - int decode_result = avcodec_decode_audio4 (stream->stream(_format_context)->codec, _frame, &frame_finished, packet); - DCPOMATIC_ENABLE_WARNINGS - 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, - even in the face of such an error, so I think we should too. - - Returning from the method here caused mantis #352. - */ - LOG_WARNING ("avcodec_decode_audio4 failed (%1)", decode_result); - } - return make_pair(decode_result, frame_finished); -} - - void FFmpegDecoder::decode_and_process_audio_packet (AVPacket* packet) { @@ -547,33 +515,27 @@ FFmpegDecoder::decode_and_process_audio_packet (AVPacket* packet) return; } - /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4 - several times. Make a simple copy so we can alter data and size. - */ - AVPacket copy_packet = *packet; - - while (copy_packet.size > 0) { - auto result = decode_audio_packet (stream, ©_packet); - if (result.first < 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, - even in the face of such an error, so I think we should too. - - Returning from the method here caused mantis #352. - */ - } - - if (result.second) { - process_audio_frame (stream); - } + auto context = _codec_context[stream->index(_format_context)]; + auto frame = audio_frame (stream); - if (result.first) { - break; + LOG_DEBUG_PLAYER("Send audio packet on stream %1", stream->index(_format_context)); + int r = avcodec_send_packet (context, packet); + if (r < 0) { + LOG_WARNING("avcodec_send_packet returned %1 for an audio packet", r); + } + while (r >= 0) { + 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; } - copy_packet.data += result.first; - copy_packet.size -= result.first; + /* We choose to be relaxed here about other errors; it seems that there may be valid + * data to decode even if an error occurred. #352 may be related (though this was + * when we were using an old version of the FFmpeg API). + */ + process_audio_frame (stream); } } @@ -583,33 +545,60 @@ FFmpegDecoder::decode_and_process_video_packet (AVPacket* 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 + auto context = video_codec_context(); + + 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); + } + + /* 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); + } + + process_video_frame (); + } + } while (pending); + + return true; +} + +void +FFmpegDecoder::process_video_frame () +{ boost::mutex::scoped_lock lm (_filter_graphs_mutex); shared_ptr graph; auto i = _filter_graphs.begin(); - while (i != _filter_graphs.end() && !(*i)->can_process (dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) { + while (i != _filter_graphs.end() && !(*i)->can_process(dcp::Size(_video_frame->width, _video_frame->height), (AVPixelFormat) _video_frame->format)) { ++i; } 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 = make_shared(dcp::Size(_video_frame->width, _video_frame->height), (AVPixelFormat) _video_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); + LOG_GENERAL (N_("New graph for %1x%2, pixel format %3"), _video_frame->width, _video_frame->height, _video_frame->format); } else { graph = *i; } - auto images = graph->process (_frame); + auto images = graph->process (_video_frame); for (auto const& i: images) { @@ -627,8 +616,6 @@ DCPOMATIC_ENABLE_WARNINGS LOG_WARNING_NC ("Dropping frame without PTS"); } } - - return true; } @@ -641,25 +628,27 @@ FFmpegDecoder::decode_and_process_subtitle_packet (AVPacket* packet) 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; if (sub_period.to) { @@ -669,6 +658,7 @@ FFmpegDecoder::decode_and_process_subtitle_packet (AVPacket* packet) _have_current_subtitle = true; } + ContentBitmapText bitmap_text(from); for (unsigned int i = 0; i < sub.num_rects; ++i) { auto const rect = sub.rects[i]; @@ -676,7 +666,7 @@ 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"; @@ -687,6 +677,10 @@ 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); } @@ -695,13 +689,13 @@ 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 */ @@ -779,7 +773,7 @@ FFmpegDecoder::process_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime static_cast(rect->h) / target_height ); - only_text()->emit_bitmap_start (from, image, scaled_rect); + return { image, scaled_rect }; } @@ -809,7 +803,8 @@ FFmpegDecoder::process_ass_subtitle (string ass, ContentTime from) base, text, _ffmpeg_content->video->size().width, - _ffmpeg_content->video->size().height + _ffmpeg_content->video->size().height, + sub::Colour(1, 1, 1) ); for (auto const& i: sub::collect>(raw)) {