X-Git-Url: https://main.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=b050fc594a76f97ab72833c38792f62daf4f100d;hp=3d9e965a9d169ad3e66a014cb3cb4e0d7a750c1b;hb=a5d004b0773f633401528392fc28e66d70e13ac8;hpb=ad1ef39eda58b3a919ea3b7084401a0439409ec6 diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index 3d9e965a9..b050fc594 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -42,6 +42,7 @@ #include "compose.hpp" #include "text_content.h" #include "audio_content.h" +#include "frame_interval_checker.h" #include #include #include @@ -50,7 +51,6 @@ extern "C" { #include #include } -#include #include #include #include @@ -59,7 +59,6 @@ extern "C" { #include "i18n.h" - using std::cout; using std::string; using std::vector; @@ -68,19 +67,20 @@ using std::min; using std::pair; using std::max; using std::map; -using boost::shared_ptr; +using std::shared_ptr; using boost::is_any_of; using boost::split; using boost::optional; -using boost::dynamic_pointer_cast; +using std::dynamic_pointer_cast; using dcp::Size; +using namespace dcpomatic; FFmpegDecoder::FFmpegDecoder (shared_ptr film, shared_ptr c, bool fast) : FFmpeg (c) , Decoder (film) , _have_current_subtitle (false) { - if (c->video) { + if (c->video && c->video->use()) { video.reset (new VideoDecoder (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 */ @@ -126,14 +126,14 @@ FFmpegDecoder::flush () if (video) { double const vfr = _ffmpeg_content->video_frame_rate().get(); Frame const f = full_length.frames_round (vfr); - Frame v = video->position(film()).frames_round(vfr) + 1; + Frame 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; } } - BOOST_FOREACH (shared_ptr i, _ffmpeg_content->ffmpeg_audio_streams ()) { + for (auto i: _ffmpeg_content->ffmpeg_audio_streams ()) { ContentTime 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 @@ -144,7 +144,7 @@ FFmpegDecoder::flush () 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 (film(), i, silence, a); + audio->emit (film(), i, silence, a, true); a += to_do; } } @@ -179,7 +179,7 @@ FFmpegDecoder::pass () int const si = _packet.stream_index; shared_ptr fc = _ffmpeg_content; - if (_video_stream && si == _video_stream.get() && !video->ignore()) { + if (_video_stream && si == _video_stream.get() && video && !video->ignore()) { decode_video_packet (); } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index(_format_context, si) && !only_text()->ignore()) { decode_subtitle_packet (); @@ -199,9 +199,14 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream) const { 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 + + /* 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 */ @@ -308,6 +313,7 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream) const 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) { memcpy (data[i], p[i], frames * sizeof(float)); @@ -328,7 +334,9 @@ FFmpegDecoder::deinterleave_audio (shared_ptr stream) const AVSampleFormat FFmpegDecoder::audio_sample_format (shared_ptr stream) const { +DCPOMATIC_DISABLE_WARNINGS return stream->stream (_format_context)->codec->sample_fmt; +DCPOMATIC_ENABLE_WARNINGS } int @@ -358,6 +366,7 @@ FFmpegDecoder::seek (ContentTime time, bool accurate) if (_video_stream) { stream = _video_stream; } else { + DCPOMATIC_ASSERT (_ffmpeg_content->audio); shared_ptr s = dynamic_pointer_cast (_ffmpeg_content->audio->stream ()); if (s) { stream = s->index (_format_context); @@ -377,19 +386,33 @@ 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 (); + } + if (video_codec_context ()) { avcodec_flush_buffers (video_codec_context()); } - BOOST_FOREACH (shared_ptr i, ffmpeg_content()->ffmpeg_audio_streams()) { +DCPOMATIC_DISABLE_WARNINGS + for (auto i: ffmpeg_content()->ffmpeg_audio_streams()) { avcodec_flush_buffers (i->stream(_format_context)->codec); } +DCPOMATIC_ENABLE_WARNINGS if (subtitle_codec_context ()) { avcodec_flush_buffers (subtitle_codec_context ()); } _have_current_subtitle = false; + + for (auto& i: _next_time) { + i = optional(); + } } void @@ -414,6 +437,7 @@ FFmpegDecoder::decode_audio_packet () return; } +DCPOMATIC_DISABLE_WARNINGS while (copy_packet.size > 0) { int frame_finished; @@ -438,12 +462,14 @@ FFmpegDecoder::decode_audio_packet () shared_ptr data = deinterleave_audio (*stream); ContentTime ct; - if (_frame->pts == AV_NOPTS_VALUE && _next_time[stream_index]) { + 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). */ - ct = *_next_time[stream_index]; + if (_next_time[stream_index]) { + ct = *_next_time[stream_index]; + } } else { ct = ContentTime::from_seconds ( av_frame_get_best_effort_timestamp (_frame) * @@ -473,6 +499,7 @@ FFmpegDecoder::decode_audio_packet () to_string(_pts_offset) ); } +DCPOMATIC_ENABLE_WARNINGS /* Give this data provided there is some, and its time is sane */ if (ct >= ContentTime() && data->frames() > 0) { @@ -491,9 +518,11 @@ FFmpegDecoder::decode_video_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 boost::mutex::scoped_lock lm (_filter_graphs_mutex); @@ -609,17 +638,17 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime #ifdef DCPOMATIC_HAVE_AVSUBTITLERECT_PICT /* Start of the first line in the subtitle */ uint8_t* sub_p = rect->pict.data[0]; - /* sub_p looks up into a BGRA palette which is here + /* 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) */ - uint32_t const * palette = (uint32_t *) rect->pict.data[1]; + uint8_t const * palette = rect->pict.data[1]; #else /* Start of the first line in the subtitle */ uint8_t* sub_p = rect->data[0]; - /* sub_p looks up into a BGRA palette which is here - (i.e. first byte B, second G, third R, fourth A) + /* sub_p looks up into a BGRA palette which is at rect->data[1]. + (first byte B, second G, third R, fourth A) */ - uint32_t const * palette = (uint32_t *) rect->data[1]; + uint8_t const * palette = rect->data[1]; #endif /* And the stream has a map of those palette colours to colours chosen by the user; created a `mapped' palette from those settings. @@ -627,7 +656,7 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime map colour_map = ffmpeg_content()->subtitle_stream()->colours (); vector mapped_palette (rect->nb_colors); for (int i = 0; i < rect->nb_colors; ++i) { - RGBA c ((palette[i] & 0xff0000) >> 16, (palette[i] & 0xff00) >> 8, palette[i] & 0xff, (palette[i] & 0xff000000) >> 24); + RGBA c (palette[2], palette[1], palette[0], palette[3]); map::const_iterator j = colour_map.find (c); if (j != colour_map.end ()) { mapped_palette[i] = j->second; @@ -638,29 +667,43 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime */ mapped_palette[i] = c; } + palette += 4; } /* Start of the output data */ - uint32_t* out_p = (uint32_t *) image->data()[0]; + uint8_t* out_p = image->data()[0]; for (int y = 0; y < rect->h; ++y) { uint8_t* sub_line_p = sub_p; - uint32_t* out_line_p = out_p; + uint8_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 BGRA?) but it looks right on screen */ - *out_line_p++ = (p.a << 24) | (p.b << 16) | (p.g << 8) | p.r; + *out_line_p++ = p.b; + *out_line_p++ = p.g; + *out_line_p++ = p.r; + *out_line_p++ = p.a; } #ifdef DCPOMATIC_HAVE_AVSUBTITLERECT_PICT sub_p += rect->pict.linesize[0]; #else sub_p += rect->linesize[0]; #endif - out_p += image->stride()[0] / sizeof (uint32_t); + out_p += image->stride()[0]; } - int const target_width = subtitle_codec_context()->width; - int const target_height = subtitle_codec_context()->height; + int target_width = subtitle_codec_context()->width; + if (target_width == 0 && video_codec_context()) { + /* subtitle_codec_context()->width == 0 has been seen in the wild but I don't + know if it's supposed to mean something from FFmpeg's point of view. + */ + target_width = video_codec_context()->width; + } + int target_height = subtitle_codec_context()->height; + if (target_height == 0 && video_codec_context()) { + target_height = video_codec_context()->height; + } + DCPOMATIC_ASSERT (target_width); + DCPOMATIC_ASSERT (target_height); dcpomatic::Rect const scaled_rect ( static_cast (rect->x) / target_width, static_cast (rect->y) / target_height, @@ -700,7 +743,7 @@ FFmpegDecoder::decode_ass_subtitle (string ass, ContentTime from) _ffmpeg_content->video->size().height ); - BOOST_FOREACH (sub::Subtitle const & i, sub::collect > (raw)) { + for (auto const& i: sub::collect > (raw)) { only_text()->emit_plain_start (from, i); } }