X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=4ca06329b116ae9561ddb3c61fa11a17416af9c2;hb=b6aa4eaa6e4335c3b5a7144c0ab126eb8dbe94ec;hp=7c846349f466d225cd190601446a917bd0e02eab;hpb=54c0378137091de604b8c0d6d98959b55b0cddb6;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index 7c846349f..4ca06329b 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -1,5 +1,5 @@ /* - Copyright (C) 2012-2014 Carl Hetherington + Copyright (C) 2012-2015 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 @@ -21,16 +21,6 @@ * @brief A decoder using FFmpeg to decode content. */ -#include -#include -#include -#include -#include -#include -extern "C" { -#include -#include -} #include "filter.h" #include "exceptions.h" #include "image.h" @@ -45,6 +35,17 @@ extern "C" { #include "raw_image_proxy.h" #include "film.h" #include "timer.h" +extern "C" { +#include +#include +} +#include +#include +#include +#include +#include +#include +#include #include "i18n.h" @@ -60,6 +61,7 @@ using std::list; using std::min; using std::pair; using std::make_pair; +using std::max; using boost::shared_ptr; using boost::optional; using boost::dynamic_pointer_cast; @@ -81,20 +83,25 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr c, shared_ptr 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; + 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 = 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(); + vector > streams = c->ffmpeg_audio_streams (); + + _pts_offset = ContentTime::min (); + + if (c->first_video ()) { + _pts_offset = - c->first_video().get (); + } + + BOOST_FOREACH (shared_ptr i, streams) { + if (i->first_audio) { + _pts_offset = max (_pts_offset, - i->first_audio.get ()); + } } /* If _pts_offset is positive we would be pushing things from a -ve PTS to be played. @@ -105,8 +112,8 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr c, shared_ptr _pts_offset = ContentTime (); } - /* Now adjust both so that the video pts starts on a frame */ - if (have_video && have_audio) { + /* Now adjust so that the video pts starts on a frame */ + if (c->first_video ()) { 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; @@ -125,10 +132,8 @@ FFmpegDecoder::flush () while (decode_video_packet ()) {} - if (_ffmpeg_content->audio_stream()) { - decode_audio_packet (); - AudioDecoder::flush (); - } + decode_audio_packet (); + AudioDecoder::flush (); } bool @@ -153,13 +158,14 @@ FFmpegDecoder::pass () } int const si = _packet.stream_index; + shared_ptr fc = _ffmpeg_content; if (si == _video_stream && !_ignore_video) { decode_video_packet (); - } 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)) { + } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index (_format_context, si)) { decode_subtitle_packet (); + } else { + decode_audio_packet (); } av_free_packet (&_packet); @@ -170,21 +176,20 @@ FFmpegDecoder::pass () * Only the first buffer will be used for non-planar data, otherwise there will be one per channel. */ shared_ptr -FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) +FFmpegDecoder::deinterleave_audio (shared_ptr stream, uint8_t** data, int size) { - DCPOMATIC_ASSERT (_ffmpeg_content->audio_channels()); - DCPOMATIC_ASSERT (bytes_per_audio_sample()); + DCPOMATIC_ASSERT (bytes_per_audio_sample (stream)); /* Deinterleave and convert to float */ /* total_samples and frames will be rounded down here, so if there are stray samples at the end of the block that do not form a complete sample or frame they will be dropped. */ - int const total_samples = size / bytes_per_audio_sample(); - int const frames = total_samples / _ffmpeg_content->audio_channels(); - shared_ptr audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames)); + int const total_samples = size / bytes_per_audio_sample (stream); + int const frames = total_samples / stream->channels(); + shared_ptr audio (new AudioBuffers (stream->channels(), frames)); - switch (audio_sample_format()) { + switch (audio_sample_format (stream)) { case AV_SAMPLE_FMT_U8: { uint8_t* p = reinterpret_cast (data[0]); @@ -194,7 +199,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) audio->data(channel)[sample] = float(*p++) / (1 << 23); ++channel; - if (channel == _ffmpeg_content->audio_channels()) { + if (channel == stream->channels()) { channel = 0; ++sample; } @@ -211,7 +216,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) audio->data(channel)[sample] = float(*p++) / (1 << 15); ++channel; - if (channel == _ffmpeg_content->audio_channels()) { + if (channel == stream->channels()) { channel = 0; ++sample; } @@ -222,7 +227,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) case AV_SAMPLE_FMT_S16P: { int16_t** p = reinterpret_cast (data); - for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) { + for (int i = 0; i < stream->channels(); ++i) { for (int j = 0; j < frames; ++j) { audio->data(i)[j] = static_cast(p[i][j]) / (1 << 15); } @@ -239,7 +244,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) audio->data(channel)[sample] = static_cast(*p++) / (1 << 31); ++channel; - if (channel == _ffmpeg_content->audio_channels()) { + if (channel == stream->channels()) { channel = 0; ++sample; } @@ -256,7 +261,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) audio->data(channel)[sample] = *p++; ++channel; - if (channel == _ffmpeg_content->audio_channels()) { + if (channel == stream->channels()) { channel = 0; ++sample; } @@ -267,33 +272,29 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) case AV_SAMPLE_FMT_FLTP: { float** p = reinterpret_cast (data); - for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) { + for (int i = 0; i < stream->channels(); ++i) { memcpy (audio->data(i), p[i], frames * sizeof(float)); } } break; default: - throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast (audio_sample_format()))); + throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast (audio_sample_format (stream)))); } return audio; } AVSampleFormat -FFmpegDecoder::audio_sample_format () const +FFmpegDecoder::audio_sample_format (shared_ptr stream) const { - if (!_ffmpeg_content->audio_stream()) { - return (AVSampleFormat) 0; - } - - return audio_codec_context()->sample_fmt; + return stream->stream (_format_context)->codec->sample_fmt; } int -FFmpegDecoder::bytes_per_audio_sample () const +FFmpegDecoder::bytes_per_audio_sample (shared_ptr stream) const { - return av_get_bytes_per_sample (audio_sample_format ()); + return av_get_bytes_per_sample (audio_sample_format (stream)); } void @@ -301,7 +302,8 @@ FFmpegDecoder::seek (ContentTime time, bool accurate) { VideoDecoder::seek (time, accurate); AudioDecoder::seek (time, accurate); - + SubtitleDecoder::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. */ @@ -313,21 +315,16 @@ FFmpegDecoder::seek (ContentTime time, bool accurate) http://www.mjbshaw.com/2012/04/seeking-in-ffmpeg-know-your-timestamp.html */ - 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 ()) { - s = min ( - s, int64_t (u.seconds() / av_q2d (_ffmpeg_content->audio_stream()->stream(_format_context)->time_base)) - ); + ContentTime u = time - _pts_offset; + if (u < ContentTime ()) { + u = ContentTime (); } - - av_seek_frame (_format_context, _video_stream, s, 0); + av_seek_frame (_format_context, _video_stream, u.seconds() / av_q2d (_format_context->streams[_video_stream]->time_base), AVSEEK_FLAG_BACKWARD); avcodec_flush_buffers (video_codec_context()); - if (audio_codec_context ()) { - avcodec_flush_buffers (audio_codec_context ()); - } + + /* XXX: should be flushing audio buffers? */ + if (subtitle_codec_context ()) { avcodec_flush_buffers (subtitle_codec_context ()); } @@ -341,11 +338,23 @@ FFmpegDecoder::decode_audio_packet () */ AVPacket copy_packet = _packet; + + /* XXX: inefficient */ + vector > streams = ffmpeg_content()->ffmpeg_audio_streams (); + vector >::const_iterator stream = streams.begin (); + while (stream != streams.end () && !(*stream)->uses_index (_format_context, copy_packet.stream_index)) { + ++stream; + } + + if (stream == streams.end ()) { + /* The packet's stream may not be an audio one; just ignore it in this method if so */ + return; + } while (copy_packet.size > 0) { int frame_finished; - int decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet); + int decode_result = avcodec_decode_audio4 ((*stream)->stream (_format_context)->codec, _frame, &frame_finished, ©_packet); if (decode_result < 0) { /* avcodec_decode_audio4 can sometimes return an error even though it has decoded some valid data; for example dca_subframe_footer can return AVERROR_INVALIDDATA @@ -365,14 +374,14 @@ FFmpegDecoder::decode_audio_packet () 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)) + av_q2d ((*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 + 0, (*stream)->stream(_format_context)->codec->channels, _frame->nb_samples, audio_sample_format (*stream), 1 ); - audio (deinterleave_audio (_frame->data, data_size), ct); + audio (*stream, deinterleave_audio (*stream, _frame->data, data_size), ct); } copy_packet.data += decode_result; @@ -433,35 +442,69 @@ 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) { - image_subtitle (ContentTimePeriod (), shared_ptr (), dcpomatic::Rect ()); + /* Sometimes we get an empty AVSubtitle, which is used by some codecs to + indicate that the previous subtitle should stop. We can ignore it here. + */ 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) + source that we may have chopped off for the DCP). */ - ContentTimePeriod period = subtitle_period (sub) + _pts_offset; - + FFmpegSubtitlePeriod sub_period = subtitle_period (sub); + ContentTimePeriod period; + period.from = sub_period.from + _pts_offset; + if (sub_period.to) { + /* We already know the subtitle period `to' time */ + period.to = sub_period.to.get() + _pts_offset; + } else { + /* We have to look up the `to' time in the stream's records */ + period.to = ffmpeg_content()->subtitle_stream()->find_subtitle_to (sub_period.from); + } + AVSubtitleRect const * rect = sub.rects[0]; - if (rect->type != SUBTITLE_BITMAP) { - /* XXX */ - // throw DecodeError (_("non-bitmap subtitles not yet supported")); - return; + switch (rect->type) { + case SUBTITLE_NONE: + break; + case SUBTITLE_BITMAP: + decode_bitmap_subtitle (rect, period); + break; + case SUBTITLE_TEXT: + cout << "XXX: SUBTITLE_TEXT " << rect->text << "\n"; + break; + case SUBTITLE_ASS: + cout << "XXX: SUBTITLE_ASS " << rect->ass << "\n"; + break; } + + avsubtitle_free (&sub); +} + +list +FFmpegDecoder::image_subtitles_during (ContentTimePeriod p, bool starting) const +{ + return _ffmpeg_content->subtitles_during (p, starting); +} + +list +FFmpegDecoder::text_subtitles_during (ContentTimePeriod, bool) const +{ + return list (); +} +void +FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTimePeriod period) +{ /* 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 @@ -470,7 +513,7 @@ FFmpegDecoder::decode_subtitle_packet () 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; @@ -481,25 +524,15 @@ FFmpegDecoder::decode_subtitle_packet () 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 - ) + dcpomatic::Rect const scaled_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); + image_subtitle (period, image, scaled_rect); } -list -FFmpegDecoder::subtitles_during (ContentTimePeriod p, bool starting) const -{ - return _ffmpeg_content->subtitles_during (p, starting); -}