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_bob_reinterlace,\r
  34                 deinterlace,\r
  35                 count,\r
  36                 invalid\r
  37         };\r
  38 \r
  39         static std::wstring print(display_mode::type value)\r
  40         {\r
  41                 switch(value)\r
  42                 {\r
  43                         case simple:\r
  44                                 return L"simple";\r
  45                         case duplicate:\r
  46                                 return L"duplicate";\r
  47                         case half:\r
  48                                 return L"half";\r
  49                         case interlace:\r
  50                                 return L"interlace";\r
  51                         case deinterlace_bob:\r
  52                                 return L"deinterlace_bob";\r
  53                         case deinterlace_bob_reinterlace:\r
  54                                 return L"deinterlace_bob_reinterlace";\r
  55                         case deinterlace:\r
  56                                 return L"deinterlace";\r
  57                         default:\r
  58                                 return L"invalid";\r
  59                 }\r
  60         }\r
  61 };\r
  62 \r
  63 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
  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<AVFrame>& 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(video_frame->data[0] == nullptr)\r
 143                 {\r
 144                         video_streams_.back().push(make_safe<core::write_frame>(this));\r
 145                         ++video_frame_count_;\r
 146                         process(frame_buffer_);\r
 147                         return;\r
 148                 }\r
 149                 \r
 150                 if(display_mode_ == display_mode::invalid)\r
 151                 {\r
 152                         auto in_mode = get_mode(*video_frame);\r
 153                         display_mode_ = auto_mode_ ? get_display_mode(in_mode, in_fps_, format_desc_.mode, format_desc_.fps) : display_mode::simple;\r
 154                         \r
 155                         if(display_mode_ == display_mode::simple && in_mode != core::video_mode::progressive && format_desc_.mode != core::video_mode::progressive && video_frame->height != static_cast<int>(format_desc_.height))\r
 156                                 display_mode_ = display_mode::deinterlace_bob_reinterlace; // The frame will most likely be scaled, we need to deinterlace->reinterlace         \r
 157                 }\r
 158                 \r
 159                 std::vector<safe_ptr<core::write_frame>> frames;\r
 160 \r
 161                 if(display_mode_ == display_mode::deinterlace)\r
 162                 {\r
 163                         if(!filter_)\r
 164                                 filter_.reset(new filter(L"YADIF=0:-1"));\r
 165 \r
 166                         BOOST_FOREACH(auto& frame, filter_->execute(video_frame))\r
 167                                 frames.push_back(make_write_frame(this, frame, frame_factory_));\r
 168 \r
 169                 }\r
 170                 else if(display_mode_ == display_mode::deinterlace_bob || display_mode_ == display_mode::deinterlace_bob_reinterlace)\r
 171                 {\r
 172                         if(!filter_)\r
 173                                 filter_.reset(new filter(L"YADIF=1:-1"));\r
 174 \r
 175                         BOOST_FOREACH(auto& frame, filter_->execute(video_frame))\r
 176                                 frames.push_back(make_write_frame(this, frame, frame_factory_));\r
 177                 }\r
 178                 else\r
 179                         frames.push_back(make_write_frame(this, make_safe(video_frame), frame_factory_));\r
 180 \r
 181                 BOOST_FOREACH(auto& frame, frames)\r
 182                 {\r
 183                         // Fix field-order if needed\r
 184                         if(frame->get_type() == core::video_mode::lower && format_desc_.mode == core::video_mode::upper)\r
 185                                 frame->get_image_transform().set_fill_translation(0.0f, 0.5/static_cast<double>(frame->get_pixel_format_desc().planes[0].height));\r
 186                         else if(frame->get_type() == core::video_mode::upper && format_desc_.mode == core::video_mode::lower)\r
 187                                 frame->get_image_transform().set_fill_translation(0.0f, -0.5/static_cast<double>(frame->get_pixel_format_desc().planes[0].height));\r
 188 \r
 189                         video_streams_.back().push(frame);\r
 190                         ++video_frame_count_;\r
 191 \r
 192                         process(frame_buffer_);\r
 193                 }\r
 194         }\r
 195 \r
 196         void push(const std::shared_ptr<std::vector<int16_t>>& audio_samples)\r
 197         {\r
 198                 if(!audio_samples)      \r
 199                 {\r
 200                         CASPAR_LOG(debug) << L"audio-chunk-count: " << audio_sample_count_/format_desc_.audio_samples_per_frame;\r
 201                         audio_streams_.push_back(std::vector<int16_t>());\r
 202                         audio_sample_count_ = 0;\r
 203                         return;\r
 204                 }\r
 205 \r
 206                 audio_sample_count_ += audio_samples->size();\r
 207 \r
 208                 boost::range::push_back(audio_streams_.back(), *audio_samples);\r
 209                 process(frame_buffer_);\r
 210         }\r
 211 \r
 212         safe_ptr<basic_frame> pop()\r
 213         {               \r
 214                 auto frame = frame_buffer_.front();\r
 215                 frame_buffer_.pop_front();              \r
 216                 return frame;\r
 217         }\r
 218 \r
 219         size_t size() const\r
 220         {\r
 221                 return frame_buffer_.size();\r
 222         }\r
 223 \r
 224         safe_ptr<core::write_frame> pop_video()\r
 225         {\r
 226                 auto frame = video_streams_.front().front();\r
 227                 video_streams_.front().pop();\r
 228                 \r
 229                 return frame;\r
 230         }\r
 231 \r
 232         std::vector<int16_t> pop_audio()\r
 233         {\r
 234                 CASPAR_VERIFY(audio_streams_.front().size() >= format_desc_.audio_samples_per_frame);\r
 235 \r
 236                 auto begin = audio_streams_.front().begin();\r
 237                 auto end   = begin + format_desc_.audio_samples_per_frame;\r
 238 \r
 239                 auto samples = std::vector<int16_t>(begin, end);\r
 240                 audio_streams_.front().erase(begin, end);\r
 241 \r
 242                 return samples;\r
 243         }\r
 244 \r
 245         bool video_ready() const\r
 246         {\r
 247                 return video_frames() > 1 && video_streams_.size() >= audio_streams_.size();\r
 248         }\r
 249         \r
 250         bool audio_ready() const\r
 251         {\r
 252                 return audio_chunks() > 1 && audio_streams_.size() >= video_streams_.size();\r
 253         }\r
 254 \r
 255         size_t video_frames() const\r
 256         {\r
 257                 return video_streams_.back().size();\r
 258         }\r
 259 \r
 260         size_t audio_chunks() const\r
 261         {\r
 262                 return audio_streams_.back().size() / format_desc_.audio_samples_per_frame;\r
 263         }\r
 264         \r
 265         void process(std::deque<safe_ptr<basic_frame>>& dest)\r
 266         {\r
 267                 if(video_streams_.size() > 1 && audio_streams_.size() > 1 &&\r
 268                         (video_streams_.front().empty() || audio_streams_.front().empty()))\r
 269                 {\r
 270                         if(!video_streams_.front().empty() || !audio_streams_.front().empty())\r
 271                                 CASPAR_LOG(debug) << "Truncating: " << video_streams_.front().size() << L" video-frames, " << audio_streams_.front().size() << L" audio-samples.";\r
 272 \r
 273                         video_streams_.pop_front();\r
 274                         audio_streams_.pop_front();\r
 275                 }\r
 276 \r
 277                 if(video_streams_.front().empty() || audio_streams_.front().size() < format_desc_.audio_samples_per_frame)\r
 278                         return;\r
 279                 \r
 280                 switch(display_mode_)\r
 281                 {\r
 282                 case display_mode::simple:                                              return simple(dest);\r
 283                 case display_mode::duplicate:                                   return duplicate(dest);\r
 284                 case display_mode::half:                                                return half(dest);\r
 285                 case display_mode::interlace:                                   return interlace(dest);\r
 286                 case display_mode::deinterlace_bob:                             return simple(dest);\r
 287                 case display_mode::deinterlace_bob_reinterlace: return interlace(dest);\r
 288                 case display_mode::deinterlace:                                 return simple(dest);\r
 289                 default:                                                                                BOOST_THROW_EXCEPTION(invalid_operation());\r
 290                 }\r
 291         }\r
 292 \r
 293         void simple(std::deque<safe_ptr<basic_frame>>& dest)\r
 294         {\r
 295                 if(video_streams_.front().empty() || audio_streams_.front().size() < format_desc_.audio_samples_per_frame)\r
 296                         return;\r
 297                 \r
 298                 auto frame1 = pop_video();\r
 299                 frame1->audio_data() = pop_audio();\r
 300 \r
 301                 dest.push_back(frame1);         \r
 302         }\r
 303 \r
 304         void duplicate(std::deque<safe_ptr<basic_frame>>& dest)\r
 305         {               \r
 306                 if(video_streams_.front().empty() || audio_streams_.front().size()/2 < format_desc_.audio_samples_per_frame)\r
 307                         return;\r
 308 \r
 309                 auto frame = pop_video();\r
 310 \r
 311                 auto frame1 = make_safe<core::write_frame>(*frame); // make a copy\r
 312                 frame1->audio_data() = pop_audio();\r
 313 \r
 314                 auto frame2 = frame;\r
 315                 frame2->audio_data() = pop_audio();\r
 316 \r
 317                 dest.push_back(frame1);\r
 318                 dest.push_back(frame2);\r
 319         }\r
 320 \r
 321         void half(std::deque<safe_ptr<basic_frame>>& dest)\r
 322         {       \r
 323                 if(video_streams_.front().size() < 2 || audio_streams_.front().size() < format_desc_.audio_samples_per_frame)\r
 324                         return;\r
 325                                                 \r
 326                 auto frame1 = pop_video();\r
 327                 frame1->audio_data() = pop_audio();\r
 328                                 \r
 329                 video_streams_.front().pop(); // Throw away\r
 330 \r
 331                 dest.push_back(frame1);\r
 332         }\r
 333         \r
 334         void interlace(std::deque<safe_ptr<basic_frame>>& dest)\r
 335         {               \r
 336                 if(video_streams_.front().size() < 2 || audio_streams_.front().size() < format_desc_.audio_samples_per_frame)\r
 337                         return;\r
 338                 \r
 339                 auto frame1 = pop_video();\r
 340                 frame1->audio_data() = pop_audio();\r
 341                                 \r
 342                 auto frame2 = pop_video();\r
 343 \r
 344                 dest.push_back(core::basic_frame::interlace(frame1, frame2, format_desc_.mode));                \r
 345         }\r
 346 };\r
 347 \r
 348 frame_muxer::frame_muxer(double in_fps, const video_format_desc& format_desc, const safe_ptr<core::frame_factory>& frame_factory)\r
 349         : impl_(new implementation(in_fps, format_desc, frame_factory)){}\r
 350 void frame_muxer::push(const std::shared_ptr<AVFrame>& video_frame){impl_->push(video_frame);}\r
 351 void frame_muxer::push(const std::shared_ptr<std::vector<int16_t>>& audio_samples){return impl_->push(audio_samples);}\r
 352 safe_ptr<basic_frame> frame_muxer::pop(){return impl_->pop();}\r
 353 size_t frame_muxer::size() const {return impl_->size();}\r
 354 bool frame_muxer::empty() const {return impl_->size() == 0;}\r
 355 bool frame_muxer::video_ready() const{return impl_->video_ready();}\r
 356 bool frame_muxer::audio_ready() const{return impl_->audio_ready();}\r
 357 \r
 358 }