X-Git-Url: https://main.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=f64ccbd3b6305249c4522ee69d175dee6bb26849;hp=54d7981a08d11430cb0aba7172ac1fe8656483c1;hb=84039db657c3cc72719706a136c90741170c765a;hpb=fd542231f5f8a4933fb1d3fa9285c1a92d9833c0 diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index 54d7981a0..f64ccbd3b 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 @@ -52,27 +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 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; @@ -80,13 +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 = {}; @@ -101,24 +97,41 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr film, shared_ptr(this, c->only_text(), ContentTime())); } - _next_time.resize (_format_context->nb_streams); + for (auto i: c->ffmpeg_audio_streams()) { + _next_time[i] = boost::optional(); + } } -void +bool FFmpegDecoder::flush () { - /* Get any remaining frames */ - - _packet.data = 0; - _packet.size = 0; + /* Flush video and audio once */ - /* XXX: should we reset _packet.data and size after each *_decode_* call? */ + bool did_something = false; + if (video) { + if (decode_and_process_video_packet(nullptr)) { + did_something = true; + } + } - while (video && decode_video_packet()) {} + for (auto i: ffmpeg_content()->ffmpeg_audio_streams()) { + 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; + } + } - if (audio) { - decode_audio_packet (); + if (did_something) { + /* We want to be called again */ + return false; } /* Make sure all streams are the same length and round up to the next video frame */ @@ -156,13 +169,18 @@ FFmpegDecoder::flush () if (audio) { audio->flush (); } + + return true; } bool FFmpegDecoder::pass () { - int r = av_read_frame (_format_context, &_packet); + auto packet = av_packet_alloc(); + DCPOMATIC_ASSERT (packet); + + int r = av_read_frame (_format_context, packet); /* AVERROR_INVALIDDATA can apparently be returned sometimes even when av_read_frame has pretty-much succeeded (and hence generated data which should be processed). @@ -176,22 +194,22 @@ FFmpegDecoder::pass () LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), &buf[0], r); } - flush (); - return true; + av_packet_free (&packet); + return flush (); } - int const si = _packet.stream_index; + int const si = packet->stream_index; auto fc = _ffmpeg_content; if (_video_stream && si == _video_stream.get() && video && !video->ignore()) { - decode_video_packet (); + decode_and_process_video_packet (packet); } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index(_format_context, si) && !only_text()->ignore()) { - decode_subtitle_packet (); + decode_and_process_subtitle_packet (packet); } else { - decode_audio_packet (); + decode_and_process_audio_packet (packet); } - av_packet_unref (&_packet); + av_packet_free (&packet); return false; } @@ -200,34 +218,22 @@ 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 + 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) { @@ -244,7 +250,7 @@ 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) { @@ -261,7 +267,7 @@ 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); @@ -272,7 +278,7 @@ 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) { @@ -289,7 +295,7 @@ 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; @@ -300,7 +306,7 @@ 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) { @@ -317,20 +323,20 @@ 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; @@ -340,9 +346,7 @@ 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); } @@ -406,11 +410,9 @@ 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 ()); @@ -419,7 +421,7 @@ DCPOMATIC_ENABLE_WARNINGS _have_current_subtitle = false; for (auto& i: _next_time) { - i = optional(); + i.second = boost::optional(); } } @@ -443,129 +445,140 @@ FFmpegDecoder::audio_stream_from_index (int index) const void -FFmpegDecoder::decode_audio_packet () +FFmpegDecoder::process_audio_frame (shared_ptr stream) { - /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4 - several times. - */ - - AVPacket copy_packet = _packet; - int const stream_index = copy_packet.stream_index; - - auto stream = audio_stream_from_index (stream_index); - if (!stream) { - /* The packet's stream may not be an audio one; just ignore it in this method if so */ - return; + auto frame = audio_frame (stream); + auto data = deinterleave_audio (frame); + + 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]) { + ct = *_next_time[stream]; + } + } else { + ct = ContentTime::from_seconds ( + frame->best_effort_timestamp * + av_q2d (stream->stream(_format_context)->time_base)) + + _pts_offset; } -DCPOMATIC_DISABLE_WARNINGS - while (copy_packet.size > 0) { + _next_time[stream] = ct + ContentTime::from_frames(data->frames(), stream->frame_rate()); - int frame_finished; - int decode_result = avcodec_decode_audio4 (stream->stream(_format_context)->codec, _frame, &frame_finished, ©_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, - 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); + if (ct < ContentTime()) { + /* Discard audio data that comes before time 0 */ + auto const remove = min (int64_t(data->frames()), (-ct).frames_ceil(double(stream->frame_rate()))); + data->move (data->frames() - remove, remove, 0); + data->set_frames (data->frames() - remove); + ct += ContentTime::from_frames (remove, stream->frame_rate()); + } - /* Fudge decode_result so that we come out of the while loop when - we've processed this data. - */ - decode_result = copy_packet.size; - } + if (ct < ContentTime()) { + LOG_WARNING ( + "Crazy timestamp %1 for %2 samples in stream %3 (ts=%4 tb=%5, off=%6)", + to_string(ct), + data->frames(), + stream->id(), + frame->best_effort_timestamp, + av_q2d(stream->stream(_format_context)->time_base), + to_string(_pts_offset) + ); + } - if (frame_finished) { - auto 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; - } + /* Give this data provided there is some, and its time is sane */ + if (ct >= ContentTime() && data->frames() > 0) { + audio->emit (film(), stream, data, ct); + } +} - _next_time[stream_index] = ct + ContentTime::from_frames(data->frames(), stream->frame_rate()); - if (ct < ContentTime()) { - /* Discard audio data that comes before time 0 */ - auto const remove = min (int64_t(data->frames()), (-ct).frames_ceil(double(stream->frame_rate()))); - data->move (data->frames() - remove, remove, 0); - data->set_frames (data->frames() - remove); - ct += ContentTime::from_frames (remove, stream->frame_rate()); - } +void +FFmpegDecoder::decode_and_process_audio_packet (AVPacket* packet) +{ + auto stream = audio_stream_from_index (packet->stream_index); + if (!stream) { + return; + } - 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) - ); - } -DCPOMATIC_ENABLE_WARNINGS + auto context = _codec_context[stream->index(_format_context)]; + auto frame = audio_frame (stream); - /* Give this data provided there is some, and its time is sane */ - if (ct >= ContentTime() && data->frames() > 0) { - audio->emit (film(), stream, data, ct); - } + 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 */ + return; } - copy_packet.data += decode_result; - copy_packet.size -= decode_result; + /* 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); } } bool -FFmpegDecoder::decode_video_packet () +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) { + auto context = video_codec_context(); + + int r = avcodec_send_packet (context, packet); + if (r < 0) { + LOG_WARNING("avcodec_send_packet returned %1 for a video packet", r); + } + + 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); } -DCPOMATIC_ENABLE_WARNINGS + + process_video_frame (); + + return true; +} + + +void +FFmpegDecoder::process_video_frame () +{ + /* We assume we'll only get one frame here, which I think is safe */ 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) { @@ -583,39 +596,39 @@ DCPOMATIC_ENABLE_WARNINGS LOG_WARNING_NC ("Dropping frame without PTS"); } } - - return true; } void -FFmpegDecoder::decode_subtitle_packet () +FFmpegDecoder::decode_and_process_subtitle_packet (AVPacket* packet) { int got_subtitle; AVSubtitle sub; - if (avcodec_decode_subtitle2 (subtitle_codec_context(), &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) { + if (avcodec_decode_subtitle2 (subtitle_codec_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; if (sub_period.to) { @@ -632,13 +645,13 @@ FFmpegDecoder::decode_subtitle_packet () case SUBTITLE_NONE: break; case SUBTITLE_BITMAP: - decode_bitmap_subtitle (rect, from); + process_bitmap_subtitle (rect, from); break; case SUBTITLE_TEXT: cout << "XXX: SUBTITLE_TEXT " << rect->text << "\n"; break; case SUBTITLE_ASS: - decode_ass_subtitle (rect->ass, from); + process_ass_subtitle (rect->ass, from); break; } } @@ -652,12 +665,12 @@ FFmpegDecoder::decode_subtitle_packet () void -FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime from) +FFmpegDecoder::process_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime from) { /* 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 */ @@ -740,7 +753,7 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime void -FFmpegDecoder::decode_ass_subtitle (string ass, ContentTime from) +FFmpegDecoder::process_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...