X-Git-Url: https://git.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=147bb384f3eb1bad5a811adf9d624501b69eedf4;hb=a7a8cd74f2f32de8b708c78cc8bc9c0cf17d60f5;hp=ecbd8d00931be7e4505417ce3b607ad87b962dbc;hpb=ded500fef3fcae71f03594c2be5ea9196d21b039;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index ecbd8d009..147bb384f 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -69,6 +69,7 @@ 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 boost::optional; @@ -101,24 +102,43 @@ 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 */ + /* Flush video and audio once */ - _packet.data = 0; - _packet.size = 0; - - /* XXX: should we reset _packet.data and size after each *_decode_* call? */ + 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)) { + did_something = true; + } + } - while (video && decode_video_packet()) {} + 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) { + 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 +176,18 @@ 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 +201,22 @@ 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; } @@ -209,6 +234,7 @@ DCPOMATIC_DISABLE_WARNINGS 0, stream->stream(_format_context)->codec->channels, _frame->nb_samples, audio_sample_format (stream), 1 ); DCPOMATIC_ENABLE_WARNINGS + DCPOMATIC_ASSERT (size >= 0); /* XXX: can't we just use _frame->nb_samples directly here? */ /* XXX: can't we use swr_convert() to do the format conversion? */ @@ -419,39 +445,116 @@ DCPOMATIC_ENABLE_WARNINGS _have_current_subtitle = false; for (auto& i: _next_time) { - i = optional(); + i.second = boost::optional(); } } -void -FFmpegDecoder::decode_audio_packet () +shared_ptr +FFmpegDecoder::audio_stream_from_index (int index) const { - /* 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; - /* XXX: inefficient */ auto streams = ffmpeg_content()->ffmpeg_audio_streams(); - auto stream = streams.begin (); - while (stream != streams.end () && !(*stream)->uses_index (_format_context, stream_index)) { + auto stream = streams.begin(); + while (stream != streams.end() && !(*stream)->uses_index(_format_context, index)) { ++stream; } if (stream == streams.end ()) { - /* The packet's stream may not be an audio one; just ignore it in this method if so */ + return {}; + } + + return *stream; +} + + +void +FFmpegDecoder::process_audio_frame (shared_ptr stream) +{ + 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]) { + ct = *_next_time[stream]; + } + } else { + ct = ContentTime::from_seconds ( + _frame->best_effort_timestamp * + av_q2d (stream->stream(_format_context)->time_base)) + + _pts_offset; + } + + _next_time[stream] = 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()); + } + + 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) + ); + } + + /* Give this data provided there is some, and its time is sane */ + if (ct >= ContentTime() && data->frames() > 0) { + audio->emit (film(), stream, data, ct); + } +} + + +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) +{ + auto stream = audio_stream_from_index (packet->stream_index); + if (!stream) { return; } -DCPOMATIC_DISABLE_WARNINGS - while (copy_packet.size > 0) { + /* 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; - int frame_finished; - int decode_result = avcodec_decode_audio4 ((*stream)->stream (_format_context)->codec, _frame, &frame_finished, ©_packet); - if (decode_result < 0) { + 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, @@ -459,77 +562,30 @@ DCPOMATIC_DISABLE_WARNINGS Returning from the method here caused mantis #352. */ - LOG_WARNING ("avcodec_decode_audio4 failed (%1)", decode_result); - - /* Fudge decode_result so that we come out of the while loop when - we've processed this data. - */ - decode_result = copy_packet.size; } - if (frame_finished) { - shared_ptr 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; - } - - _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 ()); - } - - 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 + if (result.second) { + process_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); - } + if (result.first) { + break; } - copy_packet.data += decode_result; - copy_packet.size -= decode_result; + copy_packet.data += result.first; + copy_packet.size -= result.first; } } 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) { + if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, packet) < 0 || !frame_finished) { return false; } DCPOMATIC_ENABLE_WARNINGS @@ -577,11 +633,11 @@ DCPOMATIC_ENABLE_WARNINGS 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; } @@ -620,13 +676,13 @@ 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; } } @@ -640,7 +696,7 @@ 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. @@ -728,7 +784,7 @@ 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...