X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=201fac8aa5a74dabd82ee4e3b577de4b9e239cd1;hb=c719feb3cc4a5627827f8576a9b34756f701d59d;hp=3714c1542952a22c3d57e0ee0447d0f6eddaf3e5;hpb=611f2241c6732c2c38d87e129e51cf9d8d7a08b8;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index 3714c1542..201fac8aa 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -41,7 +41,6 @@ extern "C" { #include "log.h" #include "ffmpeg_decoder.h" #include "filter_graph.h" -#include "subtitle.h" #include "audio_buffers.h" #include "i18n.h" @@ -52,6 +51,7 @@ using std::vector; using std::stringstream; using std::list; using std::min; +using std::pair; using boost::shared_ptr; using boost::optional; using boost::dynamic_pointer_cast; @@ -67,28 +67,50 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptraudio_stream() && c->first_video() && c->audio_stream()->first_audio) { - _pts_offset = compute_pts_offset (c->first_video().get(), c->audio_stream()->first_audio.get(), c->video_frame_rate()); - } -} + /* Audio and video frame PTS values may not start with 0. We want + to fiddle them so that: -double -FFmpegDecoder::compute_pts_offset (double first_video, double first_audio, float video_frame_rate) -{ - double const old_first_video = first_video; - - /* Round the first video to a frame boundary */ - if (fabs (rint (first_video * video_frame_rate) - first_video * video_frame_rate) > 1e-6) { - first_video = ceil (first_video * video_frame_rate) / video_frame_rate; + 1. One of them starts at time 0. + 2. The first video PTS value ends up on a frame boundary. + + Then we remove big initial gaps in PTS and we allow our + insertion of black frames to work. + + We will do: + audio_pts_to_use = audio_pts_from_ffmpeg + audio_pts_offset; + video_pts_to_use = video_pts_from_ffmpeg + video_pts_offset; + */ + + bool const have_video = video && c->first_video(); + bool const have_audio = audio && c->audio_stream() && c->audio_stream()->first_audio; + + /* First, make one of them start at 0 */ + + if (have_audio && have_video) { + _video_pts_offset = _audio_pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get()); + } else if (have_video) { + _video_pts_offset = - c->first_video().get(); } - /* Compute the required offset (also removing any common start delay) */ - return first_video - old_first_video - min (first_video, first_audio); + /* Now adjust both so that the video pts starts on a frame */ + if (have_video && have_audio) { + double first_video = c->first_video().get() + _video_pts_offset; + double const old_first_video = first_video; + + /* Round the first video up to a frame boundary */ + if (fabs (rint (first_video * c->video_frame_rate()) - first_video * c->video_frame_rate()) > 1e-6) { + first_video = ceil (first_video * c->video_frame_rate()) / c->video_frame_rate (); + } + + _video_pts_offset += first_video - old_first_video; + _audio_pts_offset += first_video - old_first_video; + } } FFmpegDecoder::~FFmpegDecoder () @@ -265,8 +287,26 @@ void FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate) { double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base); - int64_t const vt = frame / (_ffmpeg_content->video_frame_rate() * time_base); - av_seek_frame (_format_context, _video_stream, vt, AVSEEK_FLAG_BACKWARD); + + /* If we are doing an accurate seek, our initial shot will be 5 frames (5 being + a number plucked from the air) earlier than we want to end up. The loop below + will hopefully then step through to where we want to be. + */ + int initial = frame; + if (accurate) { + initial -= 5; + } + + if (initial < 0) { + initial = 0; + } + + /* Initial seek time in the stream's timebase */ + int64_t const initial_vt = ((initial / _ffmpeg_content->video_frame_rate()) - _video_pts_offset) / time_base; + /* Wanted final seek time in the stream's timebase */ + int64_t const final_vt = ((frame / _ffmpeg_content->video_frame_rate()) - _video_pts_offset) / time_base; + + av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD); avcodec_flush_buffers (video_codec_context()); if (_subtitle_codec_context) { @@ -294,9 +334,9 @@ FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate) int const r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet); if (r >= 0 && finished) { int64_t const bet = av_frame_get_best_effort_timestamp (_frame); - if (bet >= vt) { + if (bet >= final_vt) { _video_position = rint ( - (bet * time_base + _pts_offset) * _ffmpeg_content->video_frame_rate() + (bet * time_base + _video_pts_offset) * _ffmpeg_content->video_frame_rate() ); av_free_packet (&_packet); break; @@ -328,7 +368,7 @@ FFmpegDecoder::decode_audio_packet () if (_audio_position == 0) { /* Where we are in the source, in seconds */ double const pts = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base) - * av_frame_get_best_effort_timestamp(_frame) - _pts_offset; + * av_frame_get_best_effort_timestamp(_frame) + _audio_pts_offset; if (pts > 0) { /* Emit some silence */ @@ -380,21 +420,20 @@ FFmpegDecoder::decode_video_packet () graph = *i; } - list > images = graph->process (_frame); + list, int64_t> > images = graph->process (_frame); string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second; - for (list >::iterator i = images.begin(); i != images.end(); ++i) { + for (list, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) { - shared_ptr image = *i; + shared_ptr image = i->first; if (!post_process.empty ()) { image = image->post_process (post_process, true); } - int64_t const bet = av_frame_get_best_effort_timestamp (_frame); - if (bet != AV_NOPTS_VALUE) { + if (i->second != AV_NOPTS_VALUE) { - double const pts = bet * av_q2d (_format_context->streams[_video_stream]->time_base) - _pts_offset; + double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _video_pts_offset; if (_just_sought) { /* We just did a seek, so disable any attempts to correct for where we @@ -413,7 +452,7 @@ FFmpegDecoder::decode_video_packet () a black frame. */ boost::shared_ptr black ( - new SimpleImage ( + new Image ( static_cast (_frame->format), libdcp::Size (video_codec_context()->width, video_codec_context()->height), true @@ -478,20 +517,65 @@ FFmpegDecoder::decode_subtitle_packet () if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) { return; } - + /* Sometimes we get an empty AVSubtitle, which is used by some codecs to indicate that the previous subtitle should stop. */ - if (sub.num_rects > 0) { - shared_ptr ts; - try { - subtitle (shared_ptr (new TimedSubtitle (sub))); - } catch (...) { - /* some problem with the subtitle; we probably didn't understand it */ + if (sub.num_rects <= 0) { + subtitle (shared_ptr (), dcpomatic::Rect (), 0, 0); + return; + } else if (sub.num_rects > 1) { + throw DecodeError (_("multi-part subtitles not yet supported")); + } + + /* Subtitle PTS in seconds (within the source, not taking into account any of the + source that we may have chopped off for the DCP) + */ + double const packet_time = static_cast (sub.pts) / AV_TIME_BASE; + + /* hence start time for this sub */ + Time const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ; + Time const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ; + + AVSubtitleRect const * rect = sub.rects[0]; + + if (rect->type != SUBTITLE_BITMAP) { + throw DecodeError (_("non-bitmap subtitles not yet supported")); + } + + shared_ptr image (new Image (PIX_FMT_RGBA, libdcp::Size (rect->w, rect->h), true)); + + /* Start of the first line in the subtitle */ + uint8_t* sub_p = rect->pict.data[0]; + /* sub_p looks up into a RGB palette which is here */ + uint32_t const * palette = (uint32_t *) rect->pict.data[1]; + /* Start of the output data */ + uint32_t* out_p = (uint32_t *) 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; + for (int x = 0; x < rect->w; ++x) { + *out_line_p++ = palette[*sub_line_p++]; } - } else { - subtitle (shared_ptr ()); + sub_p += rect->pict.linesize[0]; + out_p += image->stride()[0] / sizeof (uint32_t); } + + libdcp::Size const vs = _ffmpeg_content->video_size (); + + subtitle ( + image, + dcpomatic::Rect ( + static_cast (rect->x) / vs.width, + static_cast (rect->y) / vs.height, + static_cast (rect->w) / vs.width, + static_cast (rect->h) / vs.height + ), + from, + to + ); + avsubtitle_free (&sub); }