Support more detailed horizontal positioning coming from libsub.
[dcpomatic.git] / src / lib / ffmpeg_decoder.cc
index a1d90b2ba99af04368b056bd84cede08a9790099..af309cdbe4aa99715eca949e88b1cb934f3699a5 100644 (file)
@@ -1,19 +1,20 @@
 /*
     Copyright (C) 2012-2016 Carl Hetherington <cth@carlh.net>
 
-    This program is free software; you can redistribute it and/or modify
+    This file is part of DCP-o-matic.
+
+    DCP-o-matic is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
     the Free Software Foundation; either version 2 of the License, or
     (at your option) any later version.
 
-    This program is distributed in the hope that it will be useful,
+    DCP-o-matic is distributed in the hope that it will be useful,
     but WITHOUT ANY WARRANTY; without even the implied warranty of
     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
     GNU General Public License for more details.
 
     You should have received a copy of the GNU General Public License
-    along with this program; if not, write to the Free Software
-    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+    along with DCP-o-matic.  If not, see <http://www.gnu.org/licenses/>.
 
 */
 
 #include "util.h"
 #include "log.h"
 #include "ffmpeg_decoder.h"
+#include "subtitle_decoder.h"
 #include "ffmpeg_audio_stream.h"
 #include "ffmpeg_subtitle_stream.h"
 #include "video_filter_graph.h"
 #include "audio_buffers.h"
 #include "ffmpeg_content.h"
 #include "raw_image_proxy.h"
+#include "video_decoder.h"
 #include "film.h"
-#include "md5_digester.h"
+#include "audio_decoder.h"
 #include "compose.hpp"
+#include "subtitle_content.h"
+#include "audio_content.h"
 #include <dcp/subtitle_string.h>
 #include <sub/ssa_reader.h>
 #include <sub/subtitle.h>
@@ -65,20 +70,36 @@ using std::list;
 using std::min;
 using std::pair;
 using std::max;
+using std::map;
 using boost::shared_ptr;
 using boost::is_any_of;
 using boost::split;
+using boost::optional;
+using boost::dynamic_pointer_cast;
 using dcp::Size;
 
-FFmpegDecoder::FFmpegDecoder (shared_ptr<const FFmpegContent> c, shared_ptr<Log> log, bool fast)
-       : VideoDecoder (c)
-       , AudioDecoder (c, fast)
-       , SubtitleDecoder (c)
-       , FFmpeg (c)
+FFmpegDecoder::FFmpegDecoder (shared_ptr<const FFmpegContent> c, shared_ptr<Log> log)
+       : FFmpeg (c)
        , _log (log)
-       , _pts_offset (pts_offset (c->ffmpeg_audio_streams(), c->first_video(), c->video_frame_rate()))
+       , _have_current_subtitle (false)
 {
+       if (c->video) {
+               video.reset (new VideoDecoder (this, c, log));
+               _pts_offset = pts_offset (c->ffmpeg_audio_streams(), c->first_video(), c->active_video_frame_rate());
+               /* It doesn't matter what size or pixel format this is, it just needs to be black */
+               _black_image.reset (new Image (AV_PIX_FMT_RGB24, dcp::Size (128, 128), true));
+               _black_image->make_black ();
+       } else {
+               _pts_offset = ContentTime ();
+       }
+
+       if (c->audio) {
+               audio.reset (new AudioDecoder (this, c->audio, log));
+       }
 
+       if (c->subtitle) {
+               subtitle.reset (new SubtitleDecoder (this, c->subtitle, log));
+       }
 }
 
 void
@@ -91,14 +112,45 @@ FFmpegDecoder::flush ()
 
        /* XXX: should we reset _packet.data and size after each *_decode_* call? */
 
-       while (decode_video_packet ()) {}
+       while (video && decode_video_packet ()) {}
+
+       if (audio) {
+               decode_audio_packet ();
+       }
 
-       decode_audio_packet ();
-       AudioDecoder::flush ();
+       /* Make sure all streams are the same length and round up to the next video frame */
+
+       FrameRateChange const frc = _ffmpeg_content->film()->active_frame_rate_change(_ffmpeg_content->position());
+       ContentTime full_length (_ffmpeg_content->full_length(), frc);
+       full_length = full_length.ceil (frc.source);
+       if (video) {
+               double const vfr = _ffmpeg_content->video_frame_rate().get();
+               Frame const f = full_length.frames_round (vfr);
+               Frame v = video->position().frames_round (vfr);
+               while (v < f) {
+                       video->emit (shared_ptr<const ImageProxy> (new RawImageProxy (_black_image)), v);
+                       ++v;
+               }
+       }
+
+       BOOST_FOREACH (shared_ptr<FFmpegAudioStream> i, _ffmpeg_content->ffmpeg_audio_streams ()) {
+               ContentTime a = audio->stream_position(i);
+               while (a < full_length) {
+                       ContentTime to_do = min (full_length - a, ContentTime::from_seconds (0.1));
+                       shared_ptr<AudioBuffers> silence (new AudioBuffers (i->channels(), to_do.frames_ceil (i->frame_rate())));
+                       silence->make_silent ();
+                       audio->emit (i, silence, a);
+                       a += to_do;
+               }
+       }
+
+       if (audio) {
+               audio->flush ();
+       }
 }
 
 bool
-FFmpegDecoder::pass (PassReason reason, bool accurate)
+FFmpegDecoder::pass ()
 {
        int r = av_read_frame (_format_context, &_packet);
 
@@ -111,7 +163,7 @@ FFmpegDecoder::pass (PassReason reason, bool accurate)
                        /* Maybe we should fail here, but for now we'll just finish off instead */
                        char buf[256];
                        av_strerror (r, buf, sizeof(buf));
-                       LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), buf, r);
+                       LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), &buf[0], r);
                }
 
                flush ();
@@ -121,15 +173,15 @@ FFmpegDecoder::pass (PassReason reason, bool accurate)
        int const si = _packet.stream_index;
        shared_ptr<const FFmpegContent> fc = _ffmpeg_content;
 
-       if (si == _video_stream && !_ignore_video && (accurate || reason != PASS_REASON_SUBTITLE)) {
+       if (_video_stream && si == _video_stream.get() && !video->ignore()) {
                decode_video_packet ();
-       } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index (_format_context, si)) {
+       } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index(_format_context, si) && !subtitle->ignore()) {
                decode_subtitle_packet ();
-       } else if (accurate || reason != PASS_REASON_SUBTITLE) {
+       } else {
                decode_audio_packet ();
        }
 
-       av_free_packet (&_packet);
+       av_packet_unref (&_packet);
        return false;
 }
 
@@ -206,7 +258,7 @@ FFmpegDecoder::deinterleave_audio (shared_ptr<FFmpegAudioStream> stream) const
                int sample = 0;
                int channel = 0;
                for (int i = 0; i < total_samples; ++i) {
-                       audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
+                       audio->data(channel)[sample] = static_cast<float>(*p++) / 2147483648;
 
                        ++channel;
                        if (channel == stream->channels()) {
@@ -217,6 +269,17 @@ FFmpegDecoder::deinterleave_audio (shared_ptr<FFmpegAudioStream> stream) const
        }
        break;
 
+       case AV_SAMPLE_FMT_S32P:
+       {
+               int32_t** p = reinterpret_cast<int32_t **> (_frame->data);
+               for (int i = 0; i < stream->channels(); ++i) {
+                       for (int j = 0; j < frames; ++j) {
+                               audio->data(i)[j] = static_cast<float>(p[i][j]) / 2147483648;
+                       }
+               }
+       }
+       break;
+
        case AV_SAMPLE_FMT_FLT:
        {
                float* p = reinterpret_cast<float*> (_frame->data[0]);
@@ -269,9 +332,7 @@ FFmpegDecoder::bytes_per_audio_sample (shared_ptr<FFmpegAudioStream> stream) con
 void
 FFmpegDecoder::seek (ContentTime time, bool accurate)
 {
-       VideoDecoder::seek (time, accurate);
-       AudioDecoder::seek (time, accurate);
-       SubtitleDecoder::seek (time, accurate);
+       Decoder::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.
@@ -284,19 +345,43 @@ FFmpegDecoder::seek (ContentTime time, bool accurate)
           http://www.mjbshaw.com/2012/04/seeking-in-ffmpeg-know-your-timestamp.html
        */
 
+       optional<int> stream;
+
+       if (_video_stream) {
+               stream = _video_stream;
+       } else {
+               shared_ptr<FFmpegAudioStream> s = dynamic_pointer_cast<FFmpegAudioStream> (_ffmpeg_content->audio->stream ());
+               if (s) {
+                       stream = s->index (_format_context);
+               }
+       }
+
+       DCPOMATIC_ASSERT (stream);
+
        ContentTime u = time - _pts_offset;
        if (u < ContentTime ()) {
                u = ContentTime ();
        }
-       av_seek_frame (_format_context, _video_stream, u.seconds() / av_q2d (_format_context->streams[_video_stream]->time_base), AVSEEK_FLAG_BACKWARD);
+       av_seek_frame (
+               _format_context,
+               stream.get(),
+               u.seconds() / av_q2d (_format_context->streams[stream.get()]->time_base),
+               AVSEEK_FLAG_BACKWARD
+               );
 
-       avcodec_flush_buffers (video_codec_context());
+       if (video_codec_context ()) {
+               avcodec_flush_buffers (video_codec_context());
+       }
 
-       /* XXX: should be flushing audio buffers? */
+       BOOST_FOREACH (shared_ptr<FFmpegAudioStream> i, ffmpeg_content()->ffmpeg_audio_streams()) {
+               avcodec_flush_buffers (i->stream(_format_context)->codec);
+       }
 
        if (subtitle_codec_context ()) {
                avcodec_flush_buffers (subtitle_codec_context ());
        }
+
+       _have_current_subtitle = false;
 }
 
 void
@@ -327,7 +412,7 @@ FFmpegDecoder::decode_audio_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
-                          if it overreads the auxiliary data.  ffplay carries on if frame_finished is true,
+                          if it overreads the auxiliary data.  ffplay carries on if frame_finished is true,
                           even in the face of such an error, so I think we should too.
 
                           Returning from the method here caused mantis #352.
@@ -351,13 +436,18 @@ FFmpegDecoder::decode_audio_packet ()
                        if (ct < ContentTime ()) {
                                /* Discard audio data that comes before time 0 */
                                Frame const remove = min (int64_t (data->frames()), (-ct).frames_ceil(double((*stream)->frame_rate ())));
-                               data->move (remove, 0, data->frames() - remove);
+                               data->move (data->frames() - remove, remove, 0);
                                data->set_frames (data->frames() - remove);
                                ct += ContentTime::from_frames (remove, (*stream)->frame_rate ());
                        }
 
-                       if (data->frames() > 0) {
-                               audio (*stream, data, ct);
+                       if (ct < ContentTime()) {
+                               LOG_WARNING ("Crazy timestamp %1", to_string (ct));
+                       }
+
+                       /* Give this data provided there is some, and its time is sane */
+                       if (ct >= ContentTime() && data->frames() > 0) {
+                               audio->emit (*stream, data, ct);
                        }
                }
 
@@ -369,6 +459,8 @@ FFmpegDecoder::decode_audio_packet ()
 bool
 FFmpegDecoder::decode_video_packet ()
 {
+       DCPOMATIC_ASSERT (_video_stream);
+
        int frame_finished;
        if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
                return false;
@@ -399,10 +491,10 @@ FFmpegDecoder::decode_video_packet ()
                shared_ptr<Image> image = i->first;
 
                if (i->second != AV_NOPTS_VALUE) {
-                       double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset.seconds ();
-                       video (
+                       double const pts = i->second * av_q2d (_format_context->streams[_video_stream.get()]->time_base) + _pts_offset.seconds ();
+                       video->emit (
                                shared_ptr<ImageProxy> (new RawImageProxy (image)),
-                               llrint (pts * _ffmpeg_content->video_frame_rate ())
+                               llrint (pts * _ffmpeg_content->active_video_frame_rate ())
                                );
                } else {
                        LOG_WARNING_NC ("Dropping frame without PTS");
@@ -421,10 +513,18 @@ FFmpegDecoder::decode_subtitle_packet ()
                return;
        }
 
+       /* Stop any current subtitle, either at the time it was supposed to stop, or now if now is sooner */
+       if (_have_current_subtitle) {
+               if (_current_subtitle_to) {
+                       subtitle->emit_stop (min(*_current_subtitle_to, subtitle_period(sub).from + _pts_offset));
+               } else {
+                       subtitle->emit_stop (subtitle_period(sub).from + _pts_offset);
+               }
+               _have_current_subtitle = false;
+       }
+
        if (sub.num_rects <= 0) {
-               /* 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.
-               */
+               /* Nothing new in this subtitle */
                return;
        }
 
@@ -432,14 +532,11 @@ FFmpegDecoder::decode_subtitle_packet ()
           source that we may have chopped off for the DCP).
        */
        FFmpegSubtitlePeriod sub_period = subtitle_period (sub);
-       ContentTimePeriod period;
-       period.from = sub_period.from + _pts_offset;
+       ContentTime from;
+       from = sub_period.from + _pts_offset;
+       _have_current_subtitle = true;
        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 (subtitle_id (sub));
+               _current_subtitle_to = *sub_period.to + _pts_offset;
        }
 
        for (unsigned int i = 0; i < sub.num_rects; ++i) {
@@ -449,13 +546,13 @@ FFmpegDecoder::decode_subtitle_packet ()
                case SUBTITLE_NONE:
                        break;
                case SUBTITLE_BITMAP:
-                       decode_bitmap_subtitle (rect, period);
+                       decode_bitmap_subtitle (rect, from);
                        break;
                case SUBTITLE_TEXT:
                        cout << "XXX: SUBTITLE_TEXT " << rect->text << "\n";
                        break;
                case SUBTITLE_ASS:
-                       decode_ass_subtitle (rect->ass, period);
+                       decode_ass_subtitle (rect->ass, from);
                        break;
                }
        }
@@ -463,32 +560,48 @@ FFmpegDecoder::decode_subtitle_packet ()
        avsubtitle_free (&sub);
 }
 
-list<ContentTimePeriod>
-FFmpegDecoder::image_subtitles_during (ContentTimePeriod p, bool starting) const
-{
-       return _ffmpeg_content->image_subtitles_during (p, starting);
-}
-
-list<ContentTimePeriod>
-FFmpegDecoder::text_subtitles_during (ContentTimePeriod p, bool starting) const
-{
-       return _ffmpeg_content->text_subtitles_during (p, starting);
-}
-
 void
-FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTimePeriod period)
+FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime from)
 {
        /* Note RGBA is expressed little-endian, so the first byte in the word is R, second
           G, third B, fourth A.
        */
        shared_ptr<Image> image (new Image (AV_PIX_FMT_RGBA, dcp::Size (rect->w, rect->h), true));
 
+#ifdef DCPOMATIC_HAVE_AVSUBTITLERECT_PICT
        /* 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];
+#else
+       /* Start of the first line in the subtitle */
+       uint8_t* sub_p = rect->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->data[1];
+#endif
+       /* And the stream has a map of those palette colours to colours
+          chosen by the user; created a `mapped' palette from those settings.
+       */
+       map<RGBA, RGBA> colour_map = ffmpeg_content()->subtitle_stream()->colours ();
+       vector<RGBA> mapped_palette (rect->nb_colors);
+       for (int i = 0; i < rect->nb_colors; ++i) {
+               RGBA c ((palette[i] & 0xff0000) >> 16, (palette[i] & 0xff00) >> 8, palette[i] & 0xff, (palette[i] & 0xff000000) >> 24);
+               map<RGBA, RGBA>::const_iterator j = colour_map.find (c);
+               if (j != colour_map.end ()) {
+                       mapped_palette[i] = j->second;
+               } else {
+                       /* This colour was not found in the FFmpegSubtitleStream's colour map; probably because
+                          it is from a project that was created before this stuff was added.  Just use the
+                          colour straight from the original palette.
+                       */
+                       mapped_palette[i] = c;
+               }
+       }
+
        /* Start of the output data */
        uint32_t* out_p = (uint32_t *) image->data()[0];
 
@@ -496,26 +609,32 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTimeP
                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);
+                       RGBA const p = mapped_palette[*sub_line_p++];
+                       /* XXX: this seems to be wrong to me (isn't the output image RGBA?) but it looks right on screen */
+                       *out_line_p++ = (p.a << 24) | (p.r << 16) | (p.g << 8) | p.b;
                }
+#ifdef DCPOMATIC_HAVE_AVSUBTITLERECT_PICT
                sub_p += rect->pict.linesize[0];
+#else
+               sub_p += rect->linesize[0];
+#endif
                out_p += image->stride()[0] / sizeof (uint32_t);
        }
 
-       dcp::Size const vs = _ffmpeg_content->video_size ();
+       int const target_width = subtitle_codec_context()->width;
+       int const target_height = subtitle_codec_context()->height;
        dcpomatic::Rect<double> const scaled_rect (
-               static_cast<double> (rect->x) / vs.width,
-               static_cast<double> (rect->y) / vs.height,
-               static_cast<double> (rect->w) / vs.width,
-               static_cast<double> (rect->h) / vs.height
+               static_cast<double> (rect->x) / target_width,
+               static_cast<double> (rect->y) / target_height,
+               static_cast<double> (rect->w) / target_width,
+               static_cast<double> (rect->h) / target_height
                );
 
-       image_subtitle (period, image, scaled_rect);
+       subtitle->emit_image_start (from, image, scaled_rect);
 }
 
 void
-FFmpegDecoder::decode_ass_subtitle (string ass, ContentTimePeriod period)
+FFmpegDecoder::decode_ass_subtitle (string ass, ContentTime from)
 {
        /* We have no styles and no Format: line, so I'm assuming that FFmpeg
           produces a single format of Dialogue: lines...
@@ -528,52 +647,14 @@ FFmpegDecoder::decode_ass_subtitle (string ass, ContentTimePeriod period)
        }
 
        sub::RawSubtitle base;
-       list<sub::RawSubtitle> raw = sub::SSAReader::parse_line (base, bits[9]);
-       list<sub::Subtitle> subs = sub::collect<list<sub::Subtitle> > (raw);
-
-       /* XXX: lots of this is copied from TextSubtitle; there should probably be some sharing */
-
-       /* Highest line index in this subtitle */
-       int highest = 0;
-       BOOST_FOREACH (sub::Subtitle i, subs) {
-               BOOST_FOREACH (sub::Line j, i.lines) {
-                       DCPOMATIC_ASSERT (j.vertical_position.reference && j.vertical_position.reference.get() == sub::TOP_OF_SUBTITLE);
-                       DCPOMATIC_ASSERT (j.vertical_position.line);
-                       highest = max (highest, j.vertical_position.line.get());
-               }
-       }
+       list<sub::RawSubtitle> raw = sub::SSAReader::parse_line (
+               base,
+               bits[9],
+               _ffmpeg_content->video->size().width,
+               _ffmpeg_content->video->size().height
+               );
 
-       list<dcp::SubtitleString> ss;
-
-       BOOST_FOREACH (sub::Subtitle i, sub::collect<list<sub::Subtitle> > (sub::SSAReader::parse_line (base, bits[9]))) {
-               BOOST_FOREACH (sub::Line j, i.lines) {
-                       BOOST_FOREACH (sub::Block k, j.blocks) {
-                               ss.push_back (
-                                       dcp::SubtitleString (
-                                               boost::optional<string> (),
-                                               k.italic,
-                                               dcp::Colour (255, 255, 255),
-                                               60,
-                                               1,
-                                               dcp::Time (i.from.seconds(), 1000),
-                                               dcp::Time (i.to.seconds(), 1000),
-                                               0,
-                                               dcp::HALIGN_CENTER,
-                                               /* This 1.015 is an arbitrary value to lift the bottom sub off the bottom
-                                                  of the screen a bit to a pleasing degree.
-                                               */
-                                               1.015 - ((1 + highest - j.vertical_position.line.get()) * 1.5 / 22),
-                                               dcp::VALIGN_TOP,
-                                               k.text,
-                                               static_cast<dcp::Effect> (0),
-                                               dcp::Colour (255, 255, 255),
-                                               dcp::Time (),
-                                               dcp::Time ()
-                                               )
-                                       );
-                       }
-               }
+       BOOST_FOREACH (sub::Subtitle const & i, sub::collect<list<sub::Subtitle> > (raw)) {
+               subtitle->emit_text_start (from, i);
        }
-
-       text_subtitle (period, ss);
 }