X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=c6e4217b6740959dc2ac23cf3dd2cfb91d2fc2e6;hb=63f506966cc4e9c459b3bcb430a11481f9d3a24a;hp=bcd1e738a46bbff94976afbc2fe178b8f3ff2cd3;hpb=c37d4e41da190144806463387fdbe64e65196730;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index bcd1e738a..dd47d306a 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -1,5 +1,5 @@ /* - Copyright (C) 2012 Carl Hetherington + Copyright (C) 2012-2014 Carl Hetherington This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by @@ -27,79 +27,104 @@ #include #include #include -#include #include extern "C" { #include #include } -#include "film.h" #include "filter.h" #include "exceptions.h" #include "image.h" #include "util.h" #include "log.h" #include "ffmpeg_decoder.h" +#include "ffmpeg_audio_stream.h" +#include "ffmpeg_subtitle_stream.h" #include "filter_graph.h" -#include "subtitle.h" #include "audio_buffers.h" +#include "ffmpeg_content.h" +#include "raw_image_proxy.h" +#include "film.h" +#include "timer.h" #include "i18n.h" +#define LOG_GENERAL(...) _video_content->film()->log()->log (String::compose (__VA_ARGS__), Log::TYPE_GENERAL); +#define LOG_ERROR(...) _video_content->film()->log()->log (String::compose (__VA_ARGS__), Log::TYPE_ERROR); +#define LOG_WARNING(...) _video_content->film()->log()->log (__VA_ARGS__, Log::TYPE_WARNING); + using std::cout; using std::string; using std::vector; using std::stringstream; using std::list; using std::min; +using std::pair; +using std::make_pair; using boost::shared_ptr; using boost::optional; using boost::dynamic_pointer_cast; -using libdcp::Size; +using dcp::Size; -FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptr c, bool video, bool audio) - : Decoder (f) - , VideoDecoder (f) - , AudioDecoder (f) +FFmpegDecoder::FFmpegDecoder (shared_ptr c, shared_ptr log) + : VideoDecoder (c) + , AudioDecoder (c) + , SubtitleDecoder (c) , FFmpeg (c) - , _subtitle_codec_context (0) - , _subtitle_codec (0) - , _decode_video (video) - , _decode_audio (audio) - , _pts_offset (0) - , _just_sought (false) + , _log (log) { - setup_subtitle (); + /* Audio and video frame PTS values may not start with 0. We want + to fiddle them so that: - if (video && audio && c->audio_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()); - } -} + 1. One of them starts at time 0. + 2. The first video PTS value ends up on a frame boundary. -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; + Then we remove big initial gaps in PTS and we allow our + insertion of black frames to work. + + We will do pts_to_use = pts_from_ffmpeg + pts_offset; + */ + + bool const have_video = c->first_video(); + bool const have_audio = c->audio_stream () && c->audio_stream()->first_audio; + + /* First, make one of them start at 0 */ + + if (have_audio && have_video) { + _pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get()); + } else if (have_video) { + _pts_offset = - c->first_video().get(); + } else if (have_audio) { + _pts_offset = - c->audio_stream()->first_audio.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) { + ContentTime first_video = c->first_video().get() + _pts_offset; + ContentTime const old_first_video = first_video; + _pts_offset += first_video.round_up (c->video_frame_rate ()) - old_first_video; + } } -FFmpegDecoder::~FFmpegDecoder () +void +FFmpegDecoder::flush () { - boost::mutex::scoped_lock lm (_mutex); - - if (_subtitle_codec_context) { - avcodec_close (_subtitle_codec_context); + /* Get any remaining frames */ + + _packet.data = 0; + _packet.size = 0; + + /* XXX: should we reset _packet.data and size after each *_decode_* call? */ + + while (decode_video_packet ()) {} + + if (_ffmpeg_content->audio_stream()) { + decode_audio_packet (); + AudioDecoder::flush (); } -} +} -void +bool FFmpegDecoder::pass () { int r = av_read_frame (_format_context, &_packet); @@ -109,63 +134,25 @@ FFmpegDecoder::pass () /* Maybe we should fail here, but for now we'll just finish off instead */ char buf[256]; av_strerror (r, buf, sizeof(buf)); - shared_ptr film = _film.lock (); - assert (film); - film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r)); - } - - /* Get any remaining frames */ - - _packet.data = 0; - _packet.size = 0; - - /* XXX: should we reset _packet.data and size after each *_decode_* call? */ - - if (_decode_video) { - while (decode_video_packet ()); - } - - if (_ffmpeg_content->audio_stream() && _decode_audio) { - decode_audio_packet (); + LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), buf, r); } - /* Stop us being asked for any more data */ - _video_position = _ffmpeg_content->video_length (); - _audio_position = _ffmpeg_content->audio_length (); - return; + flush (); + return true; } - avcodec_get_frame_defaults (_frame); + int const si = _packet.stream_index; - if (_packet.stream_index == _video_stream && _decode_video) { + if (si == _video_stream) { decode_video_packet (); - } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) { + } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, si)) { decode_audio_packet (); - } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id) { -#if 0 - - int got_subtitle; - AVSubtitle sub; - if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) { - /* 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 */ - } - } else { - subtitle (shared_ptr ()); - } - avsubtitle_free (&sub); - } -#endif + } else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si)) { + decode_subtitle_packet (); } av_free_packet (&_packet); + return false; } /** @param data pointer to array of pointers to buffers. @@ -186,6 +173,23 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) shared_ptr audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames)); switch (audio_sample_format()) { + case AV_SAMPLE_FMT_U8: + { + uint8_t* p = reinterpret_cast (data[0]); + int sample = 0; + int channel = 0; + for (int i = 0; i < total_samples; ++i) { + audio->data(channel)[sample] = float(*p++) / (1 << 23); + + ++channel; + if (channel == _ffmpeg_content->audio_channels()) { + channel = 0; + ++sample; + } + } + } + break; + case AV_SAMPLE_FMT_S16: { int16_t* p = reinterpret_cast (data[0]); @@ -281,45 +285,44 @@ FFmpegDecoder::bytes_per_audio_sample () const } void -FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate) +FFmpegDecoder::seek (ContentTime time, 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); + VideoDecoder::seek (time, accurate); + AudioDecoder::seek (time, accurate); + + /* If we are doing an `accurate' seek, we need to use pre-roll, as + we don't really know what the seek will give us. + */ - avcodec_flush_buffers (video_codec_context()); - if (_subtitle_codec_context) { - avcodec_flush_buffers (_subtitle_codec_context); + ContentTime pre_roll = accurate ? ContentTime::from_seconds (2) : ContentTime (0); + time -= pre_roll; + if (time < ContentTime (0)) { + time = ContentTime (0); } - _just_sought = true; + ContentTime const u = time - _pts_offset; + int64_t s = u.seconds() / av_q2d (_format_context->streams[_video_stream]->time_base); - if (accurate) { - while (1) { - int r = av_read_frame (_format_context, &_packet); - if (r < 0) { - return; - } - - avcodec_get_frame_defaults (_frame); - - if (_packet.stream_index == _video_stream) { - int finished = 0; - 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) { - _video_position = rint ( - (bet * time_base + _pts_offset) * _ffmpeg_content->video_frame_rate() - ); - av_free_packet (&_packet); - break; - } - } - } - - av_free_packet (&_packet); - } + if (_ffmpeg_content->audio_stream ()) { + s = min ( + s, int64_t (u.seconds() / av_q2d (_ffmpeg_content->audio_stream()->stream(_format_context)->time_base)) + ); + } + + /* Ridiculous empirical hack */ + s--; + if (s < 0) { + s = 0; + } + + av_seek_frame (_format_context, _video_stream, s, 0); + + avcodec_flush_buffers (video_codec_context()); + if (audio_codec_context ()) { + avcodec_flush_buffers (audio_codec_context ()); + } + if (subtitle_codec_context ()) { + avcodec_flush_buffers (subtitle_codec_context ()); } } @@ -331,37 +334,32 @@ FFmpegDecoder::decode_audio_packet () */ AVPacket copy_packet = _packet; - + while (copy_packet.size > 0) { int frame_finished; int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet); - if (decode_result >= 0) { - if (frame_finished) { - - 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; - - if (pts > 0) { - /* Emit some silence */ - shared_ptr silence ( - new AudioBuffers ( - _ffmpeg_content->audio_channels(), - pts * _ffmpeg_content->content_audio_frame_rate() - ) - ); - - silence->make_silent (); - audio (silence, _audio_position); - } - } - - copy_packet.data += decode_result; - copy_packet.size -= decode_result; - } + + if (decode_result < 0) { + LOG_ERROR ("avcodec_decode_audio4 failed (%1)", decode_result); + return; } + + if (frame_finished) { + ContentTime const ct = ContentTime::from_seconds ( + av_frame_get_best_effort_timestamp (_frame) * + av_q2d (_ffmpeg_content->audio_stream()->stream (_format_context)->time_base)) + + _pts_offset; + + int const data_size = av_samples_get_buffer_size ( + 0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1 + ); + + audio (deinterleave_audio (_frame->data, data_size), ct); + } + + copy_packet.data += decode_result; + copy_packet.size -= decode_result; } } @@ -372,115 +370,119 @@ FFmpegDecoder::decode_video_packet () if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) { return false; } - + boost::mutex::scoped_lock lm (_filter_graphs_mutex); shared_ptr graph; list >::iterator i = _filter_graphs.begin(); - while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) { + while (i != _filter_graphs.end() && !(*i)->can_process (dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) { ++i; } if (i == _filter_graphs.end ()) { - shared_ptr film = _film.lock (); - assert (film); - - graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)); + graph.reset (new FilterGraph (_ffmpeg_content, dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)); _filter_graphs.push_back (graph); - - film->log()->log (String::compose (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"), _frame->width, _frame->height, _frame->format); } else { 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; - if (!post_process.empty ()) { - image = image->post_process (post_process, true); - } + shared_ptr image = i->first; - int64_t const bet = av_frame_get_best_effort_timestamp (_frame); - if (bet != AV_NOPTS_VALUE) { - - double const pts = bet * av_q2d (_format_context->streams[_video_stream]->time_base) - _pts_offset; - - if (_just_sought) { - /* We just did a seek, so disable any attempts to correct for where we - are / should be. - */ - _video_position = rint (pts * _ffmpeg_content->video_frame_rate ()); - _just_sought = false; - } - - double const next = _video_position / _ffmpeg_content->video_frame_rate(); - double const one_frame = 1 / _ffmpeg_content->video_frame_rate (); - double delta = pts - next; - - while (delta > one_frame) { - /* This PTS is more than one frame forward in time of where we think we should be; emit - a black frame. - */ - boost::shared_ptr black ( - new SimpleImage ( - static_cast (_frame->format), - libdcp::Size (video_codec_context()->width, video_codec_context()->height), - true - ) - ); - - black->make_black (); - video (image, false, _video_position); - delta -= one_frame; - } - - if (delta > -one_frame) { - /* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */ - video (image, false, _video_position); - } - + if (i->second != AV_NOPTS_VALUE) { + double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset.seconds (); + video ( + shared_ptr (new RawImageProxy (image, _video_content->film()->log())), + rint (pts * _ffmpeg_content->video_frame_rate ()) + ); } else { - shared_ptr film = _film.lock (); - assert (film); - film->log()->log ("Dropping frame without PTS"); + LOG_WARNING ("Dropping frame without PTS"); } } return true; } - void -FFmpegDecoder::setup_subtitle () +FFmpegDecoder::decode_subtitle_packet () { - boost::mutex::scoped_lock lm (_mutex); - - if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) { + int got_subtitle; + AVSubtitle sub; + 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) { + image_subtitle (ContentTimePeriod (), shared_ptr (), dcpomatic::Rect ()); return; + } else if (sub.num_rects > 1) { + throw DecodeError (_("multi-part subtitles not yet supported")); } + + /* Subtitle PTS (within the source, not taking into account any of the + source that we may have chopped off for the DCP) + */ + ContentTimePeriod period = subtitle_period (sub) + _pts_offset; - _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec; - _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id); + AVSubtitleRect const * rect = sub.rects[0]; - if (_subtitle_codec == 0) { - throw DecodeError (_("could not find subtitle decoder")); + if (rect->type != SUBTITLE_BITMAP) { + /* XXX */ + // throw DecodeError (_("non-bitmap subtitles not yet supported")); + return; } - - if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) { - throw DecodeError (N_("could not open subtitle decoder")); + + /* Note RGBA is expressed little-endian, so the first byte in the word is R, second + G, third B, fourth A. + */ + shared_ptr image (new Image (PIX_FMT_RGBA, dcp::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 BGRA palette which is here + (i.e. first byte B, second G, third R, fourth A) + */ + 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) { + uint32_t const p = palette[*sub_line_p++]; + *out_line_p++ = ((p & 0xff) << 16) | (p & 0xff00) | ((p & 0xff0000) >> 16) | (p & 0xff000000); + } + sub_p += rect->pict.linesize[0]; + out_p += image->stride()[0] / sizeof (uint32_t); } + + dcp::Size const vs = _ffmpeg_content->video_size (); + + image_subtitle ( + period, + 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 + ) + ); + + avsubtitle_free (&sub); } -bool -FFmpegDecoder::done () const +list +FFmpegDecoder::subtitles_during (ContentTimePeriod p, bool starting) const { - bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length()); - bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length()); - return vd && ad; + return _ffmpeg_content->subtitles_during (p, starting); } -