X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=668cab88530d1b385821056aeb837d4b2a011ff0;hb=c88a8a6ec6b396dc90d40a4843160d616a45db76;hp=dc2a5590b450f2226ef55ef0ffab005022aa899c;hpb=cab9a1d569396065a6e9eb39386736908564d6b4;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index dc2a5590b..668cab885 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 @@ -39,12 +38,21 @@ extern "C" { #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 "audio_buffers.h" #include "ffmpeg_content.h" +#include "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; @@ -52,25 +60,19 @@ 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 dcp::Size; -FFmpegDecoder::FFmpegDecoder (shared_ptr c, shared_ptr log, bool video, bool audio, bool subtitles) +FFmpegDecoder::FFmpegDecoder (shared_ptr c, shared_ptr log) : VideoDecoder (c) , AudioDecoder (c) + , SubtitleDecoder (c) , FFmpeg (c) , _log (log) - , _subtitle_codec_context (0) - , _subtitle_codec (0) - , _decode_video (video) - , _decode_audio (audio) - , _decode_subtitles (subtitles) - , _pts_offset (0) { - setup_subtitle (); - /* Audio and video frame PTS values may not start with 0. We want to fiddle them so that: @@ -83,8 +85,8 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr c, shared_ptr We will do pts_to_use = pts_from_ffmpeg + pts_offset; */ - bool const have_video = video && c->first_video(); - bool const have_audio = _decode_audio && c->audio_stream () && c->audio_stream()->first_audio; + 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 */ @@ -104,15 +106,6 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr c, shared_ptr } } -FFmpegDecoder::~FFmpegDecoder () -{ - boost::mutex::scoped_lock lm (_mutex); - - if (_subtitle_codec_context) { - avcodec_close (_subtitle_codec_context); - } -} - void FFmpegDecoder::flush () { @@ -123,11 +116,9 @@ FFmpegDecoder::flush () /* XXX: should we reset _packet.data and size after each *_decode_* call? */ - if (_decode_video) { - while (decode_video_packet ()) {} - } + while (decode_video_packet ()) {} - if (_ffmpeg_content->audio_stream() && _decode_audio) { + if (_ffmpeg_content->audio_stream()) { decode_audio_packet (); AudioDecoder::flush (); } @@ -143,7 +134,7 @@ 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)); - _log->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r)); + LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), buf, r); } flush (); @@ -151,12 +142,12 @@ FFmpegDecoder::pass () } int const si = _packet.stream_index; - - if (si == _video_stream && _decode_video) { + + if (si == _video_stream) { decode_video_packet (); - } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, si) && _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() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si) && _decode_subtitles) { + } else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si)) { decode_subtitle_packet (); } @@ -182,6 +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]); @@ -276,75 +284,23 @@ FFmpegDecoder::bytes_per_audio_sample () const return av_get_bytes_per_sample (audio_sample_format ()); } -int -FFmpegDecoder::minimal_run (boost::function, optional, int)> finished) +void +FFmpegDecoder::seek (ContentTime time, bool accurate) { - int frames_read = 0; - optional last_video; - optional last_audio; - - while (!finished (last_video, last_audio, frames_read)) { - int r = av_read_frame (_format_context, &_packet); - if (r < 0) { - /* We should flush our decoders here, possibly yielding a few more frames, - but the consequence of having to do that is too hideous to contemplate. - Instead we give up and say that you can't seek too close to the end - of a file. - */ - return frames_read; - } - - ++frames_read; - - double const time_base = av_q2d (_format_context->streams[_packet.stream_index]->time_base); - - if (_packet.stream_index == _video_stream) { - - avcodec_get_frame_defaults (_frame); - - int finished = 0; - r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet); - if (r >= 0 && finished) { - last_video = ContentTime::from_seconds (av_frame_get_best_effort_timestamp (_frame) * time_base) + _pts_offset; - } + 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. + */ - } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, _packet.stream_index)) { - AVPacket copy_packet = _packet; - while (copy_packet.size > 0) { - - int finished; - r = avcodec_decode_audio4 (audio_codec_context(), _frame, &finished, &_packet); - if (r >= 0 && finished) { - last_audio = ContentTime::from_seconds (av_frame_get_best_effort_timestamp (_frame) * time_base) + _pts_offset; - } - - copy_packet.data += r; - copy_packet.size -= r; - } - } - - av_free_packet (&_packet); + ContentTime pre_roll = accurate ? ContentTime::from_seconds (2) : ContentTime (0); + time -= pre_roll; + if (time < ContentTime (0)) { + time = ContentTime (0); } - return frames_read; -} - -bool -FFmpegDecoder::seek_overrun_finished (ContentTime seek, optional last_video, optional last_audio) const -{ - return (last_video && last_video.get() >= seek) || (last_audio && last_audio.get() >= seek); -} - -bool -FFmpegDecoder::seek_final_finished (int n, int done) const -{ - return n == done; -} - -void -FFmpegDecoder::seek_and_flush (ContentTime t) -{ - ContentTime const u = t - _pts_offset; + ContentTime const u = time - _pts_offset; int64_t s = u.seconds() / av_q2d (_format_context->streams[_video_stream]->time_base); if (_ffmpeg_content->audio_stream ()) { @@ -365,44 +321,8 @@ FFmpegDecoder::seek_and_flush (ContentTime t) if (audio_codec_context ()) { avcodec_flush_buffers (audio_codec_context ()); } - if (_subtitle_codec_context) { - avcodec_flush_buffers (_subtitle_codec_context); - } -} - -void -FFmpegDecoder::seek (ContentTime time, bool accurate) -{ - Decoder::seek (time, accurate); - if (_decode_audio) { - AudioDecoder::seek (time, accurate); - } - - /* If we are doing an accurate seek, our initial shot will be 200ms (200 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. - */ - - ContentTime pre_roll = accurate ? ContentTime::from_seconds (0.2) : ContentTime (0); - ContentTime initial_seek = time - pre_roll; - if (initial_seek < ContentTime (0)) { - initial_seek = ContentTime (0); - } - - /* Initial seek time in the video stream's timebase */ - - seek_and_flush (initial_seek); - - if (!accurate) { - /* That'll do */ - return; - } - - int const N = minimal_run (boost::bind (&FFmpegDecoder::seek_overrun_finished, this, time, _1, _2)); - - seek_and_flush (initial_seek); - if (N > 0) { - minimal_run (boost::bind (&FFmpegDecoder::seek_final_finished, this, N - 1, _3)); + if (subtitle_codec_context ()) { + avcodec_flush_buffers (subtitle_codec_context ()); } } @@ -421,7 +341,7 @@ FFmpegDecoder::decode_audio_packet () int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet); if (decode_result < 0) { - _log->log (String::compose ("avcodec_decode_audio4 failed (%1)", decode_result)); + LOG_ERROR ("avcodec_decode_audio4 failed (%1)", decode_result); return; } @@ -463,64 +383,37 @@ FFmpegDecoder::decode_video_packet () if (i == _filter_graphs.end ()) { graph.reset (new FilterGraph (_ffmpeg_content, dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)); _filter_graphs.push_back (graph); - _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, int64_t> > images = graph->process (_frame); - string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second; - for (list, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) { shared_ptr image = i->first; - if (!post_process.empty ()) { - image = image->post_process (post_process, true); - } if (i->second != AV_NOPTS_VALUE) { - video (image, false, ContentTime::from_seconds (i->second * 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.seconds (); + video ( + shared_ptr (new RawImageProxy (image, _video_content->film()->log())), + rint (pts * _ffmpeg_content->video_frame_rate ()) + ); } else { - _log->log ("Dropping frame without PTS"); + LOG_WARNING ("Dropping frame without PTS"); } } return true; } - -void -FFmpegDecoder::setup_subtitle () -{ - boost::mutex::scoped_lock lm (_mutex); - - if (!_ffmpeg_content->subtitle_stream()) { - return; - } - - _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 (N_("could not find subtitle decoder")); - } - - if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) { - throw DecodeError (N_("could not open subtitle decoder")); - } -} - void FFmpegDecoder::decode_subtitle_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; } @@ -528,7 +421,7 @@ FFmpegDecoder::decode_subtitle_packet () indicate that the previous subtitle should stop. */ if (sub.num_rects <= 0) { - image_subtitle (shared_ptr (), dcpomatic::Rect (), ContentTime (), ContentTime ()); + image_subtitle (ContentTimePeriod (), shared_ptr (), dcpomatic::Rect ()); return; } else if (sub.num_rects > 1) { throw DecodeError (_("multi-part subtitles not yet supported")); @@ -537,16 +430,14 @@ FFmpegDecoder::decode_subtitle_packet () /* Subtitle PTS (within the source, not taking into account any of the source that we may have chopped off for the DCP) */ - ContentTime packet_time = ContentTime::from_seconds (static_cast (sub.pts) / AV_TIME_BASE) + _pts_offset; - - /* hence start time for this sub */ - ContentTime const from = packet_time + ContentTime::from_seconds (sub.start_display_time / 1e3); - ContentTime const to = packet_time + ContentTime::from_seconds (sub.end_display_time / 1e3); + ContentTimePeriod period = subtitle_period (sub) + _pts_offset; AVSubtitleRect const * rect = sub.rects[0]; if (rect->type != SUBTITLE_BITMAP) { - throw DecodeError (_("non-bitmap subtitles not yet supported")); + /* XXX */ + // throw DecodeError (_("non-bitmap subtitles not yet supported")); + return; } /* Note RGBA is expressed little-endian, so the first byte in the word is R, second @@ -577,17 +468,21 @@ FFmpegDecoder::decode_subtitle_packet () 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 - ), - from, - to + ) ); - avsubtitle_free (&sub); } + +list +FFmpegDecoder::subtitles_during (ContentTimePeriod p) const +{ + return _ffmpeg_content->subtitles_during (p); +}