X-Git-Url: https://git.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=347d8cc08c86b0489795747a621c034fd9a8c74d;hb=6bc33bf8bbf13ab9aee0d3e455f62d98f9e527f4;hp=8da607e7eb62861812d035d88461f39fc83c1bcd;hpb=a4e7f2fe00fcec204de6562d0b8eba9ca80207cd;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index 8da607e7e..347d8cc08 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -61,15 +61,14 @@ using libdcp::Size; FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptr c, bool video, bool audio) : Decoder (f) , VideoDecoder (f, c) - , AudioDecoder (f) + , AudioDecoder (f, c) , SubtitleDecoder (f) , FFmpeg (c) , _subtitle_codec_context (0) , _subtitle_codec (0) , _decode_video (video) , _decode_audio (audio) - , _video_pts_offset (0) - , _audio_pts_offset (0) + , _pts_offset (0) , _just_sought (false) { setup_subtitle (); @@ -84,8 +83,8 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptrfirst_video(); @@ -94,16 +93,16 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptrfirst_video().get(), c->audio_stream()->first_audio.get()); + _pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get()); } else if (have_video) { - _video_pts_offset = - c->first_video().get(); + _pts_offset = - c->first_video().get(); } else if (have_audio) { - _audio_pts_offset = - c->audio_stream()->first_audio.get(); + _pts_offset = - c->audio_stream()->first_audio.get(); } /* Now adjust both so that the video pts starts on a frame */ if (have_video && have_audio) { - double first_video = c->first_video().get() + _video_pts_offset; + 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 */ @@ -111,8 +110,7 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptrvideo_frame_rate()) / c->video_frame_rate (); } - _video_pts_offset += first_video - old_first_video; - _audio_pts_offset += first_video - old_first_video; + _pts_offset += first_video - old_first_video; } } @@ -167,16 +165,16 @@ FFmpegDecoder::pass () return; } - avcodec_get_frame_defaults (_frame); - shared_ptr film = _film.lock (); assert (film); + + int const si = _packet.stream_index; - if (_packet.stream_index == _video_stream && _decode_video) { + 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 && film->with_subtitles ()) { + } else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si) && film->with_subtitles ()) { decode_subtitle_packet (); } @@ -201,6 +199,23 @@ 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]); @@ -315,7 +330,7 @@ FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate) } /* Initial seek time in the stream's timebase */ - int64_t const initial_vt = ((initial / _ffmpeg_content->video_frame_rate()) - _video_pts_offset) / time_base; + 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); @@ -324,7 +339,16 @@ FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate) avcodec_flush_buffers (_subtitle_codec_context); } - _just_sought = true; + /* 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) { @@ -339,16 +363,15 @@ FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate) } if (_packet.stream_index != _video_stream) { + av_free_packet (&_packet); continue; } - avcodec_get_frame_defaults (_frame); - 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 + _video_pts_offset) * _ffmpeg_content->video_frame_rate() + (av_frame_get_best_effort_timestamp (_frame) * time_base + _pts_offset) * _ffmpeg_content->video_frame_rate() ); if (_video_position >= (frame - 1)) { @@ -386,7 +409,7 @@ FFmpegDecoder::decode_audio_packet () 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) + _audio_pts_offset; + * av_frame_get_best_effort_timestamp(_frame) + _pts_offset; if (pts > 0) { /* Emit some silence */ @@ -456,7 +479,7 @@ FFmpegDecoder::decode_video_packet () if (i->second != AV_NOPTS_VALUE) { - double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _video_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 @@ -510,15 +533,19 @@ 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) { @@ -556,8 +583,8 @@ FFmpegDecoder::decode_subtitle_packet () /* 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; - + 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; @@ -567,21 +594,27 @@ FFmpegDecoder::decode_subtitle_packet () 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 RGB palette which is here */ + /* 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) { - *out_line_p++ = palette[*sub_line_p++]; + 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);