git.sesse.net Git - casparcg/blob - modules/ffmpeg/producer/frame_muxer.cpp

   1 #include "../StdAfx.h"\r
   2 \r
   3 #include "frame_muxer.h"\r
   4 \r
   5 #include "filter/filter.h"\r
   6 \r
   7 #include "util.h"\r
   8 \r
   9 #include <core/producer/frame/basic_frame.h>\r
  10 #include <core/producer/frame/image_transform.h>\r
  11 #include <core/producer/frame/pixel_format.h>\r
  12 #include <core/producer/frame/frame_factory.h>\r
  13 #include <core/mixer/write_frame.h>\r
  14 \r
  15 #include <common/env.h>\r
  16 #include <common/log/log.h>\r
  17 \r
  18 #include <boost/range/algorithm_ext/push_back.hpp>\r
  19 \r
  20 using namespace caspar::core;\r
  21 \r
  22 namespace caspar {\r
  23 \r
  24 struct display_mode\r
  25 {\r
  26         enum type\r
  27         {\r
  28                 simple,\r
  29                 duplicate,\r
  30                 half,\r
  31                 interlace,\r
  32                 deinterlace_bob,\r
  33                 deinterlace,\r
  34                 count,\r
  35                 invalid\r
  36         };\r
  37 \r
  38         static std::wstring print(display_mode::type value)\r
  39         {\r
  40                 switch(value)\r
  41                 {\r
  42                         case simple:\r
  43                                 return L"simple";\r
  44                         case duplicate:\r
  45                                 return L"duplicate";\r
  46                         case half:\r
  47                                 return L"half";\r
  48                         case interlace:\r
  49                                 return L"interlace";\r
  50                         case deinterlace_bob:\r
  51                                 return L"deinterlace_bob";\r
  52                         case deinterlace:\r
  53                                 return L"deinterlace";\r
  54                         default:\r
  55                                 return L"invalid";\r
  56                 }\r
  57         }\r
  58 };\r
  59 \r
  60 display_mode::type get_display_mode(const core::video_mode::type in_mode, double in_fps, const core::video_mode::type out_mode, double out_fps)\r
  61 {               \r
  62         if(in_mode == core::video_mode::invalid || out_mode == core::video_mode::invalid)\r
  63                 return display_mode::invalid;\r
  64 \r
  65         static const auto epsilon = 2.0;\r
  66 \r
  67         if(std::abs(in_fps - out_fps) < epsilon)\r
  68         {\r
  69                 if(in_mode != core::video_mode::progressive && out_mode == core::video_mode::progressive)\r
  70                         return display_mode::deinterlace;\r
  71                 //else if(in_mode == core::video_mode::progressive && out_mode != core::video_mode::progressive)\r
  72                 //      simple(); // interlace_duplicate();\r
  73                 else\r
  74                         return display_mode::simple;\r
  75         }\r
  76         else if(std::abs(in_fps/2.0 - out_fps) < epsilon)\r
  77         {\r
  78                 if(in_mode != core::video_mode::progressive)\r
  79                         return display_mode::invalid;\r
  80 \r
  81                 if(out_mode != core::video_mode::progressive)\r
  82                         return display_mode::interlace;\r
  83                 else\r
  84                         return display_mode::half;\r
  85         }\r
  86         else if(std::abs(in_fps - out_fps/2.0) < epsilon)\r
  87         {\r
  88                 if(out_mode != core::video_mode::progressive)\r
  89                         return display_mode::invalid;\r
  90 \r
  91                 if(in_mode != core::video_mode::progressive)\r
  92                         return display_mode::deinterlace_bob;\r
  93                 else\r
  94                         return display_mode::duplicate;\r
  95         }\r
  96 \r
  97         return display_mode::invalid;\r
  98 }\r
  99 \r
 100 struct frame_muxer::implementation : boost::noncopyable\r
 101 {       \r
 102         std::deque<std::queue<safe_ptr<write_frame>>>   video_streams_;\r
 103         std::deque<std::vector<int16_t>>                                audio_streams_;\r
 104         std::deque<safe_ptr<basic_frame>>                               frame_buffer_;\r
 105         display_mode::type                                                              display_mode_;\r
 106         const double                                                                    in_fps_;\r
 107         const video_format_desc                                                 format_desc_;\r
 108         bool                                                                                    auto_mode_;\r
 109 \r
 110         size_t                                                                                  audio_sample_count_;\r
 111         size_t                                                                                  video_frame_count_;\r
 112                 \r
 113         size_t                                                                                  processed_audio_sample_count_;\r
 114         size_t                                                                                  processed_video_frame_count_;\r
 115 \r
 116         std::unique_ptr<filter>                                                 filter_;\r
 117         safe_ptr<core::frame_factory>                                   frame_factory_;\r
 118                 \r
 119         implementation(double in_fps, const video_format_desc& format_desc, const safe_ptr<core::frame_factory>& frame_factory)\r
 120                 : display_mode_(display_mode::invalid)\r
 121                 , in_fps_(in_fps)\r
 122                 , format_desc_(format_desc)\r
 123                 , auto_mode_(env::properties().get("configuration.auto-mode", false))\r
 124                 , audio_sample_count_(0)\r
 125                 , video_frame_count_(0)\r
 126                 , frame_factory_(frame_factory)\r
 127                 , video_streams_(1)\r
 128                 , audio_streams_(1)\r
 129         {\r
 130         }\r
 131 \r
 132         void push(const std::shared_ptr<write_frame>& video_frame)\r
 133         {               \r
 134                 if(!video_frame)\r
 135                 {       \r
 136                         CASPAR_LOG(debug) << L"video-frame-count: " << static_cast<float>(video_frame_count_);\r
 137                         video_frame_count_ = 0;\r
 138                         video_streams_.push_back(std::queue<safe_ptr<write_frame>>());\r
 139                         return;\r
 140                 }\r
 141                 \r
 142                 if(display_mode_ == display_mode::invalid)\r
 143                 {\r
 144                         display_mode_ = auto_mode_ ? get_display_mode(video_frame->get_type(), in_fps_, format_desc_.mode, format_desc_.fps) : display_mode::simple;\r
 145                         CASPAR_LOG(info) << L"display_mode: " << display_mode::print(display_mode_);\r
 146                 }               \r
 147 \r
 148                 ++video_frame_count_;\r
 149 \r
 150                 // Fix field-order if needed\r
 151                 if(video_frame->get_type() == core::video_mode::lower && format_desc_.mode == core::video_mode::upper)\r
 152                         video_frame->get_image_transform().set_fill_translation(0.0f, 0.5/static_cast<double>(video_frame->get_pixel_format_desc().planes[0].height));\r
 153                 else if(video_frame->get_type() == core::video_mode::upper && format_desc_.mode == core::video_mode::lower)\r
 154                         video_frame->get_image_transform().set_fill_translation(0.0f, -0.5/static_cast<double>(video_frame->get_pixel_format_desc().planes[0].height));\r
 155 \r
 156                 video_streams_.back().push(make_safe(video_frame));\r
 157 \r
 158                 process(frame_buffer_);\r
 159 \r
 160                 video_frame->commit();\r
 161         }\r
 162 \r
 163         void push(const std::shared_ptr<std::vector<int16_t>>& audio_samples)\r
 164         {\r
 165                 if(!audio_samples)      \r
 166                 {\r
 167                         CASPAR_LOG(debug) << L"audio-chunk-count: " << audio_sample_count_/format_desc_.audio_samples_per_frame;\r
 168                         audio_streams_.push_back(std::vector<int16_t>());\r
 169                         audio_sample_count_ = 0;\r
 170                         return;\r
 171                 }\r
 172 \r
 173                 audio_sample_count_ += audio_samples->size();\r
 174 \r
 175                 boost::range::push_back(audio_streams_.back(), *audio_samples);\r
 176                 process(frame_buffer_);\r
 177         }\r
 178 \r
 179         safe_ptr<basic_frame> pop()\r
 180         {               \r
 181                 auto frame = frame_buffer_.front();\r
 182                 frame_buffer_.pop_front();              \r
 183                 return frame;\r
 184         }\r
 185 \r
 186         size_t size() const\r
 187         {\r
 188                 return frame_buffer_.size();\r
 189         }\r
 190 \r
 191         safe_ptr<core::write_frame> pop_video()\r
 192         {\r
 193                 auto frame = video_streams_.front().front();\r
 194                 video_streams_.front().pop();\r
 195                 \r
 196                 return frame;\r
 197         }\r
 198 \r
 199         std::vector<int16_t> pop_audio()\r
 200         {\r
 201                 CASPAR_VERIFY(audio_streams_.front().size() >= format_desc_.audio_samples_per_frame);\r
 202 \r
 203                 auto begin = audio_streams_.front().begin();\r
 204                 auto end   = begin + format_desc_.audio_samples_per_frame;\r
 205 \r
 206                 auto samples = std::vector<int16_t>(begin, end);\r
 207                 audio_streams_.front().erase(begin, end);\r
 208 \r
 209                 return samples;\r
 210         }\r
 211 \r
 212         bool video_ready() const\r
 213         {\r
 214                 return video_frames() > 1 && video_streams_.size() >= audio_streams_.size();\r
 215         }\r
 216         \r
 217         bool audio_ready() const\r
 218         {\r
 219                 return audio_chunks() > 1 && audio_streams_.size() >= video_streams_.size();\r
 220         }\r
 221 \r
 222         size_t video_frames() const\r
 223         {\r
 224                 return video_streams_.back().size();\r
 225         }\r
 226 \r
 227         size_t audio_chunks() const\r
 228         {\r
 229                 return audio_streams_.back().size() / format_desc_.audio_samples_per_frame;\r
 230         }\r
 231         \r
 232         void process(std::deque<safe_ptr<basic_frame>>& dest)\r
 233         {\r
 234                 if(video_streams_.size() > 1 && audio_streams_.size() > 1 &&\r
 235                         (video_streams_.front().empty() || audio_streams_.front().empty()))\r
 236                 {\r
 237                         if(!video_streams_.front().empty() || !audio_streams_.front().empty())\r
 238                                 CASPAR_LOG(debug) << " Truncating: " << video_streams_.front().size() << L" video-frames, " << audio_streams_.front().size() << L" audio-samples.";\r
 239 \r
 240                         video_streams_.pop_front();\r
 241                         audio_streams_.pop_front();\r
 242                 }\r
 243 \r
 244                 if(video_streams_.front().empty() || audio_streams_.front().size() < format_desc_.audio_samples_per_frame)\r
 245                         return;\r
 246                 \r
 247                 switch(display_mode_)\r
 248                 {\r
 249                 case display_mode::simple:                                              return simple(dest);\r
 250                 case display_mode::duplicate:                                   return duplicate(dest);\r
 251                 case display_mode::half:                                                return half(dest);\r
 252                 case display_mode::interlace:                                   return interlace(dest);\r
 253                 case display_mode::deinterlace_bob:                             return deinterlace_bob(dest);\r
 254                 case display_mode::deinterlace:                                 return deinterlace(dest);\r
 255                 default:                                                                                BOOST_THROW_EXCEPTION(invalid_operation());\r
 256                 }\r
 257         }\r
 258 \r
 259         void simple(std::deque<safe_ptr<basic_frame>>& dest)\r
 260         {\r
 261                 if(video_streams_.front().empty() || audio_streams_.front().size() < format_desc_.audio_samples_per_frame)\r
 262                         return;\r
 263                 \r
 264                 if(video_streams_.front().front()->get_type() != core::video_mode::progressive && format_desc_.mode != core::video_mode::progressive &&\r
 265                         video_streams_.front().front()->get_pixel_format_desc().planes.at(0).height != format_desc_.height )\r
 266                 { // The frame will most likely be scaled, we need to deinterlace->reinterlace\r
 267                         if(!filter_)\r
 268                                 filter_.reset(new filter(L"YADIF=1:-1"));\r
 269                 \r
 270                         auto frame = pop_video();\r
 271 \r
 272                         auto av_frames = filter_->execute(as_av_frame(frame));\r
 273 \r
 274                         if(av_frames.size() < 2)\r
 275                                 return;\r
 276 \r
 277                         auto frame1 = make_write_frame(frame->tag(), av_frames[0], frame_factory_);\r
 278                         frame1->commit();\r
 279                         auto frame2 = make_write_frame(frame->tag(), av_frames[1], frame_factory_);\r
 280                         frame2->commit();\r
 281 \r
 282                         frame1->audio_data() = pop_audio();\r
 283                         dest.push_back(core::basic_frame::interlace(frame1, frame2, format_desc_.mode));                \r
 284                 }\r
 285                 else\r
 286                 {\r
 287                         auto frame1 = pop_video();\r
 288                         frame1->commit();\r
 289                         frame1->audio_data() = pop_audio();\r
 290 \r
 291                         dest.push_back(frame1);\r
 292                 }\r
 293         }\r
 294 \r
 295         void duplicate(std::deque<safe_ptr<basic_frame>>& dest)\r
 296         {               \r
 297                 if(video_streams_.front().empty() || audio_streams_.front().size()/2 < format_desc_.audio_samples_per_frame)\r
 298                         return;\r
 299 \r
 300                 auto frame = pop_video();\r
 301                 frame->commit();\r
 302 \r
 303                 auto frame1 = make_safe<core::write_frame>(*frame); // make a copy\r
 304                 frame1->audio_data() = pop_audio();\r
 305 \r
 306                 auto frame2 = frame;\r
 307                 frame2->audio_data() = pop_audio();\r
 308 \r
 309                 dest.push_back(frame1);\r
 310                 dest.push_back(frame2);\r
 311         }\r
 312 \r
 313         void half(std::deque<safe_ptr<basic_frame>>& dest)\r
 314         {       \r
 315                 if(video_streams_.front().size() < 2 || audio_streams_.front().size() < format_desc_.audio_samples_per_frame)\r
 316                         return;\r
 317                                                 \r
 318                 auto frame1 = pop_video();\r
 319                 frame1->commit();\r
 320                 frame1->audio_data() = pop_audio();\r
 321                                 \r
 322                 video_streams_.front().pop(); // Throw away\r
 323 \r
 324                 dest.push_back(frame1);\r
 325         }\r
 326         \r
 327         void interlace(std::deque<safe_ptr<basic_frame>>& dest)\r
 328         {               \r
 329                 if(video_streams_.front().size() < 2 || audio_streams_.front().size() < format_desc_.audio_samples_per_frame)\r
 330                         return;\r
 331                 \r
 332                 auto frame1 = pop_video();\r
 333                 frame1->commit();\r
 334 \r
 335                 frame1->audio_data() = pop_audio();\r
 336                                 \r
 337                 auto frame2 = pop_video();\r
 338 \r
 339                 dest.push_back(core::basic_frame::interlace(frame1, frame2, format_desc_.mode));                \r
 340         }\r
 341                 \r
 342         void deinterlace_bob(std::deque<safe_ptr<basic_frame>>& dest)\r
 343         {\r
 344                 if(video_streams_.front().empty() || audio_streams_.front().size()/2 < format_desc_.audio_samples_per_frame)\r
 345                         return;\r
 346 \r
 347                 if(!filter_)\r
 348                         filter_.reset(new filter(L"YADIF=1:-1"));\r
 349                 \r
 350                 auto frame = pop_video();\r
 351 \r
 352                 auto av_frames = filter_->execute(as_av_frame(frame));\r
 353 \r
 354                 if(av_frames.size() < 2)\r
 355                         return;\r
 356 \r
 357                 auto frame1 = make_write_frame(frame->tag(), av_frames.at(0), frame_factory_);\r
 358                 frame1->commit();\r
 359                 frame1->audio_data() = pop_audio();\r
 360                 \r
 361                 auto frame2 = make_write_frame(frame->tag(), av_frames.at(1), frame_factory_);\r
 362                 frame2->commit();\r
 363                 frame2->audio_data() = pop_audio();\r
 364                 \r
 365                 dest.push_back(frame1);\r
 366                 dest.push_back(frame2);\r
 367         }\r
 368 \r
 369         void deinterlace(std::deque<safe_ptr<basic_frame>>& dest)\r
 370         {\r
 371                 if(video_streams_.front().empty() || audio_streams_.front().size() < format_desc_.audio_samples_per_frame)\r
 372                         return;\r
 373 \r
 374                 if(!filter_)\r
 375                         filter_.reset(new filter(L"YADIF=0:-1"));\r
 376                 \r
 377                 auto frame = pop_video();\r
 378                                 \r
 379                 auto av_frames = filter_->execute(as_av_frame(frame));\r
 380 \r
 381                 if(av_frames.empty())\r
 382                         return;\r
 383 \r
 384                 auto frame1 = make_write_frame(frame->tag(), av_frames.at(0), frame_factory_);\r
 385                 frame1->commit();\r
 386                 frame1->audio_data() = pop_audio();\r
 387                                 \r
 388                 dest.push_back(frame1);\r
 389         }\r
 390 };\r
 391 \r
 392 frame_muxer::frame_muxer(double in_fps, const video_format_desc& format_desc, const safe_ptr<core::frame_factory>& frame_factory)\r
 393         : impl_(new implementation(in_fps, format_desc, frame_factory)){}\r
 394 void frame_muxer::push(const std::shared_ptr<write_frame>& video_frame){impl_->push(video_frame);}\r
 395 void frame_muxer::push(const std::shared_ptr<std::vector<int16_t>>& audio_samples){return impl_->push(audio_samples);}\r
 396 safe_ptr<basic_frame> frame_muxer::pop(){return impl_->pop();}\r
 397 size_t frame_muxer::size() const {return impl_->size();}\r
 398 bool frame_muxer::empty() const {return impl_->size() == 0;}\r
 399 bool frame_muxer::video_ready() const{return impl_->video_ready();}\r
 400 bool frame_muxer::audio_ready() const{return impl_->audio_ready();}\r
 401 \r
 402 }