X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=a51b521d0fe22918010a6976c5ae754dd9d8e051;hb=7e690d21278df14b113f3602cbbd43f6214fd614;hp=785a805f4f5b2aba6c41615dae79d8616df678ea;hpb=50ce388189f9c930ea0ba80cfb20237cfa464a3f;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index 785a805f4..a51b521d0 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 @@ -27,7 +27,6 @@ #include #include #include -#include #include extern "C" { #include @@ -41,17 +40,23 @@ extern "C" { #include "log.h" #include "ffmpeg_decoder.h" #include "filter_graph.h" -#include "subtitle.h" #include "audio_buffers.h" +#include "ffmpeg_content.h" +#include "image_proxy.h" #include "i18n.h" +#define LOG_GENERAL(...) film->log()->log (String::compose (__VA_ARGS__), Log::TYPE_GENERAL); +#define LOG_ERROR(...) film->log()->log (String::compose (__VA_ARGS__), Log::TYPE_ERROR); +#define LOG_WARNING(...) 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 boost::shared_ptr; using boost::optional; using boost::dynamic_pointer_cast; @@ -59,8 +64,9 @@ using libdcp::Size; FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptr c, bool video, bool audio) : Decoder (f) - , VideoDecoder (f) - , AudioDecoder (f) + , VideoDecoder (f, c) + , AudioDecoder (f, c) + , SubtitleDecoder (f) , FFmpeg (c) , _subtitle_codec_context (0) , _subtitle_codec (0) @@ -71,23 +77,45 @@ 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 + pts_offset; + video_pts_to_use = video_pts_from_ffmpeg + 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) { + _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) { + double first_video = c->first_video().get() + _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 (); + } + + _pts_offset += first_video - old_first_video; + } } FFmpegDecoder::~FFmpegDecoder () @@ -97,7 +125,30 @@ FFmpegDecoder::~FFmpegDecoder () if (_subtitle_codec_context) { avcodec_close (_subtitle_codec_context); } -} +} + +void +FFmpegDecoder::flush () +{ + /* 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 (); + } + + /* Stop us being asked for any more data */ + _video_position = _ffmpeg_content->video_length_after_3d_combine (); + _audio_position = _ffmpeg_content->audio_length (); +} void FFmpegDecoder::pass () @@ -111,58 +162,24 @@ FFmpegDecoder::pass () 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 ()); + LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), buf, r); } - if (_ffmpeg_content->audio_stream() && _decode_audio) { - decode_audio_packet (); - } - - /* Stop us being asked for any more data */ - _video_position = _ffmpeg_content->video_length (); - _audio_position = _ffmpeg_content->audio_length (); + flush (); return; } - avcodec_get_frame_defaults (_frame); + shared_ptr film = _film.lock (); + assert (film); - if (_packet.stream_index == _video_stream && _decode_video) { + int const si = _packet.stream_index; + + if (si == _video_stream && _decode_video) { 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) { 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) && film->with_subtitles ()) { + decode_subtitle_packet (); } av_free_packet (&_packet); @@ -186,6 +203,23 @@ 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]); @@ -284,48 +318,79 @@ 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()) - _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) { avcodec_flush_buffers (_subtitle_codec_context); } - _just_sought = true; - - if (frame == 0) { - /* We're already there; from here on we can only seek non-zero amounts */ + /* This !accurate is piling hack upon hack; setting _just_sought to true + even with accurate == true defeats our attempt to align the start + of the video and audio. Here we disable that defeat when accurate == true + i.e. when we are making a DCP rather than just previewing one. + Ewww. This should be gone in 2.0. + */ + if (!accurate) { + _just_sought = true; + } + + _video_position = frame; + + if (frame == 0 || !accurate) { + /* We're already there, or we're as close as we need to be */ return; } - 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; - } - } - } - + while (1) { + int r = av_read_frame (_format_context, &_packet); + if (r < 0) { + return; + } + + if (_packet.stream_index != _video_stream) { av_free_packet (&_packet); + continue; } + + int finished = 0; + r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet); + if (r >= 0 && finished) { + _video_position = rint ( + (av_frame_get_best_effort_timestamp (_frame) * time_base + _pts_offset) * _ffmpeg_content->video_frame_rate() + ); + + if (_video_position >= (frame - 1)) { + av_free_packet (&_packet); + break; + } + } + + av_free_packet (&_packet); } + + /* _video_position should be the next thing to be emitted, which will the one after the thing + we just saw. + */ + _video_position++; } void @@ -336,37 +401,51 @@ 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 (decode_result < 0) { + shared_ptr film = _film.lock (); + assert (film); + LOG_ERROR ("avcodec_decode_audio4 failed (%1)", decode_result); + return; + } - if (pts > 0) { - /* Emit some silence */ + 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 */ + int64_t frames = pts * _ffmpeg_content->content_audio_frame_rate (); + while (frames > 0) { + int64_t const this_time = min (frames, (int64_t) _ffmpeg_content->content_audio_frame_rate() / 2); + shared_ptr silence ( - new AudioBuffers ( - _ffmpeg_content->audio_channels(), - pts * _ffmpeg_content->content_audio_frame_rate() - ) + new AudioBuffers (_ffmpeg_content->audio_channels(), this_time) ); - + silence->make_silent (); audio (silence, _audio_position); + frames -= this_time; } } - - copy_packet.data += decode_result; - copy_packet.size -= decode_result; } + + 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), _audio_position); } + + copy_packet.data += decode_result; + copy_packet.size -= decode_result; } } @@ -394,26 +473,23 @@ FFmpegDecoder::decode_video_packet () graph.reset (new FilterGraph (_ffmpeg_content, libdcp::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) { + shared_ptr film = _film.lock (); + assert (film); - shared_ptr image = *i; - if (!post_process.empty ()) { - image = image->post_process (post_process, true); - } + for (list, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) { + + shared_ptr image = i->first; - 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) + _pts_offset; if (_just_sought) { /* We just did a seek, so disable any attempts to correct for where we @@ -431,28 +507,31 @@ FFmpegDecoder::decode_video_packet () /* This PTS is more than one frame forward in time of where we think we should be; emit a black frame. */ + + /* XXX: I think this should be a copy of the last frame... */ boost::shared_ptr black ( - new SimpleImage ( + new Image ( static_cast (_frame->format), libdcp::Size (video_codec_context()->width, video_codec_context()->height), true ) ); + shared_ptr film = _film.lock (); + assert (film); + black->make_black (); - video (image, false, _video_position); + video (shared_ptr (new RawImageProxy (image, film->log())), 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); + video (shared_ptr (new RawImageProxy (image, film->log())), false, _video_position); } } else { - shared_ptr film = _film.lock (); - assert (film); - film->log()->log ("Dropping frame without PTS"); + LOG_WARNING ("Dropping frame without PTS"); } } @@ -465,15 +544,19 @@ FFmpegDecoder::setup_subtitle () { boost::mutex::scoped_lock lm (_mutex); - if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) { + if (!_ffmpeg_content->subtitle_stream()) { return; } - _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec; + _subtitle_codec_context = _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec; + if (_subtitle_codec_context == 0) { + throw DecodeError (N_("could not find subtitle stream")); + } + _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id); if (_subtitle_codec == 0) { - throw DecodeError (_("could not find subtitle decoder")); + throw DecodeError (N_("could not find subtitle decoder")); } if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) { @@ -489,3 +572,79 @@ FFmpegDecoder::done () const return vd && ad; } +void +FFmpegDecoder::decode_subtitle_packet () +{ + 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) { + 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) + _pts_offset; + + /* 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")); + } + + /* 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, 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 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); + } + + 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); +}