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

   1 #include "../../stdafx.h"\r
   2 \r
   3 #include "util.h"\r
   4 \r
   5 #include "flv.h"\r
   6 \r
   7 #include "../tbb_avcodec.h"\r
   8 #include "../../ffmpeg_error.h"\r
   9 \r
  10 #include <tbb/concurrent_unordered_map.h>\r
  11 #include <tbb/concurrent_queue.h>\r
  12 \r
  13 #include <core/producer/frame/frame_transform.h>\r
  14 #include <core/producer/frame/frame_factory.h>\r
  15 #include <core/producer/frame_producer.h>\r
  16 #include <core/mixer/write_frame.h>\r
  17 \r
  18 #include <common/exception/exceptions.h>\r
  19 #include <common/utility/assert.h>\r
  20 #include <common/memory/memcpy.h>\r
  21 \r
  22 #include <tbb/parallel_for.h>\r
  23 \r
  24 #include <boost/filesystem.hpp>\r
  25 #include <boost/lexical_cast.hpp>\r
  26 \r
  27 #if defined(_MSC_VER)\r
  28 #pragma warning (push)\r
  29 #pragma warning (disable : 4244)\r
  30 #endif\r
  31 extern "C" \r
  32 {\r
  33         #include <libswscale/swscale.h>\r
  34         #include <libavcodec/avcodec.h>\r
  35         #include <libavformat/avformat.h>\r
  36 }\r
  37 #if defined(_MSC_VER)\r
  38 #pragma warning (pop)\r
  39 #endif\r
  40 \r
  41 namespace caspar { namespace ffmpeg {\r
  42                 \r
  43 std::shared_ptr<core::audio_buffer> flush_audio()\r
  44 {\r
  45         static std::shared_ptr<core::audio_buffer> audio(new core::audio_buffer());\r
  46         return audio;\r
  47 }\r
  48 \r
  49 std::shared_ptr<core::audio_buffer> empty_audio()\r
  50 {\r
  51         static std::shared_ptr<core::audio_buffer> audio(new core::audio_buffer());\r
  52         return audio;\r
  53 }\r
  54 \r
  55 std::shared_ptr<AVFrame>                        flush_video()\r
  56 {\r
  57         static std::shared_ptr<AVFrame> video(avcodec_alloc_frame(), av_free);\r
  58         return video;\r
  59 }\r
  60 \r
  61 std::shared_ptr<AVFrame>                        empty_video()\r
  62 {\r
  63         static std::shared_ptr<AVFrame> video(avcodec_alloc_frame(), av_free);\r
  64         return video;\r
  65 }\r
  66 \r
  67 core::field_mode::type get_mode(const AVFrame& frame)\r
  68 {\r
  69         if(!frame.interlaced_frame)\r
  70                 return core::field_mode::progressive;\r
  71 \r
  72         return frame.top_field_first ? core::field_mode::upper : core::field_mode::lower;\r
  73 }\r
  74 \r
  75 core::pixel_format::type get_pixel_format(PixelFormat pix_fmt)\r
  76 {\r
  77         switch(pix_fmt)\r
  78         {\r
  79         case CASPAR_PIX_FMT_LUMA:       return core::pixel_format::luma;\r
  80         case PIX_FMT_GRAY8:                     return core::pixel_format::gray;\r
  81         case PIX_FMT_BGRA:                      return core::pixel_format::bgra;\r
  82         case PIX_FMT_ARGB:                      return core::pixel_format::argb;\r
  83         case PIX_FMT_RGBA:                      return core::pixel_format::rgba;\r
  84         case PIX_FMT_ABGR:                      return core::pixel_format::abgr;\r
  85         case PIX_FMT_YUV444P:           return core::pixel_format::ycbcr;\r
  86         case PIX_FMT_YUV422P:           return core::pixel_format::ycbcr;\r
  87         case PIX_FMT_YUV420P:           return core::pixel_format::ycbcr;\r
  88         case PIX_FMT_YUV411P:           return core::pixel_format::ycbcr;\r
  89         case PIX_FMT_YUV410P:           return core::pixel_format::ycbcr;\r
  90         case PIX_FMT_YUVA420P:          return core::pixel_format::ycbcra;\r
  91         default:                                        return core::pixel_format::invalid;\r
  92         }\r
  93 }\r
  94 \r
  95 core::pixel_format_desc get_pixel_format_desc(PixelFormat pix_fmt, size_t width, size_t height)\r
  96 {\r
  97         // Get linesizes\r
  98         AVPicture dummy_pict;   \r
  99         avpicture_fill(&dummy_pict, nullptr, pix_fmt == CASPAR_PIX_FMT_LUMA ? PIX_FMT_GRAY8 : pix_fmt, width, height);\r
 100 \r
 101         core::pixel_format_desc desc;\r
 102         desc.pix_fmt = get_pixel_format(pix_fmt);\r
 103                 \r
 104         switch(desc.pix_fmt)\r
 105         {\r
 106         case core::pixel_format::gray:\r
 107         case core::pixel_format::luma:\r
 108                 {\r
 109                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));                                               \r
 110                         return desc;\r
 111                 }\r
 112         case core::pixel_format::bgra:\r
 113         case core::pixel_format::argb:\r
 114         case core::pixel_format::rgba:\r
 115         case core::pixel_format::abgr:\r
 116                 {\r
 117                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0]/4, height, 4));                                             \r
 118                         return desc;\r
 119                 }\r
 120         case core::pixel_format::ycbcr:\r
 121         case core::pixel_format::ycbcra:\r
 122                 {               \r
 123                         // Find chroma height\r
 124                         size_t size2 = dummy_pict.data[2] - dummy_pict.data[1];\r
 125                         size_t h2 = size2/dummy_pict.linesize[1];                       \r
 126 \r
 127                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));\r
 128                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[1], h2, 1));\r
 129                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[2], h2, 1));\r
 130 \r
 131                         if(desc.pix_fmt == core::pixel_format::ycbcra)                                          \r
 132                                 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[3], height, 1));       \r
 133                         return desc;\r
 134                 }               \r
 135         default:                \r
 136                 desc.pix_fmt = core::pixel_format::invalid;\r
 137                 return desc;\r
 138         }\r
 139 }\r
 140 \r
 141 int make_alpha_format(int format)\r
 142 {\r
 143         switch(get_pixel_format(static_cast<PixelFormat>(format)))\r
 144         {\r
 145         case core::pixel_format::ycbcr:\r
 146         case core::pixel_format::ycbcra:\r
 147                 return CASPAR_PIX_FMT_LUMA;\r
 148         default:\r
 149                 return format;\r
 150         }\r
 151 }\r
 152 \r
 153 safe_ptr<core::write_frame> make_write_frame(const void* tag, const safe_ptr<AVFrame>& decoded_frame, const safe_ptr<core::frame_factory>& frame_factory, int hints)\r
 154 {                       \r
 155         static tbb::concurrent_unordered_map<size_t, tbb::concurrent_queue<std::shared_ptr<SwsContext>>> sws_contexts_;\r
 156         \r
 157         if(decoded_frame->width < 1 || decoded_frame->height < 1)\r
 158                 return make_safe<core::write_frame>(tag);\r
 159 \r
 160         const auto width  = decoded_frame->width;\r
 161         const auto height = decoded_frame->height;\r
 162         auto desc                 = get_pixel_format_desc(static_cast<PixelFormat>(decoded_frame->format), width, height);\r
 163         \r
 164         if(hints & core::frame_producer::ALPHA_HINT)\r
 165                 desc = get_pixel_format_desc(static_cast<PixelFormat>(make_alpha_format(decoded_frame->format)), width, height);\r
 166 \r
 167         std::shared_ptr<core::write_frame> write;\r
 168 \r
 169         if(desc.pix_fmt == core::pixel_format::invalid)\r
 170         {\r
 171                 auto pix_fmt = static_cast<PixelFormat>(decoded_frame->format);\r
 172 \r
 173                 write = frame_factory->create_frame(tag, get_pixel_format_desc(PIX_FMT_BGRA, width, height));\r
 174                 write->set_type(get_mode(*decoded_frame));\r
 175 \r
 176                 std::shared_ptr<SwsContext> sws_context;\r
 177 \r
 178                 //CASPAR_LOG(warning) << "Hardware accelerated color transform not supported.";\r
 179 \r
 180                 size_t key = width << 20 | height << 8 | pix_fmt;\r
 181                         \r
 182                 auto& pool = sws_contexts_[key];\r
 183                                                 \r
 184                 if(!pool.try_pop(sws_context))\r
 185                 {\r
 186                         double param;\r
 187                         sws_context.reset(sws_getContext(width, height, pix_fmt, width, height, PIX_FMT_BGRA, SWS_BILINEAR, nullptr, nullptr, &param), sws_freeContext);\r
 188                 }\r
 189                         \r
 190                 if(!sws_context)\r
 191                 {\r
 192                         BOOST_THROW_EXCEPTION(operation_failed() << msg_info("Could not create software scaling context.") << \r
 193                                                                         boost::errinfo_api_function("sws_getContext"));\r
 194                 }       \r
 195 \r
 196                 // Use sws_scale when provided colorspace has no hw-accel.\r
 197                 safe_ptr<AVFrame> av_frame(avcodec_alloc_frame(), av_free);     \r
 198                 avcodec_get_frame_defaults(av_frame.get());                     \r
 199                 auto size = avpicture_fill(reinterpret_cast<AVPicture*>(av_frame.get()), write->image_data().begin(), PIX_FMT_BGRA, width, height);\r
 200                 CASPAR_VERIFY(size == write->image_data().size()); \r
 201 \r
 202                 sws_scale(sws_context.get(), decoded_frame->data, decoded_frame->linesize, 0, height, av_frame->data, av_frame->linesize);      \r
 203                 pool.push(sws_context);\r
 204 \r
 205                 write->commit();\r
 206         }\r
 207         else\r
 208         {\r
 209                 write = frame_factory->create_frame(tag, desc);\r
 210                 write->set_type(get_mode(*decoded_frame));\r
 211 \r
 212                 for(int n = 0; n < static_cast<int>(desc.planes.size()); ++n)\r
 213                 {\r
 214                         auto plane            = desc.planes[n];\r
 215                         auto result           = write->image_data(n).begin();\r
 216                         auto decoded          = decoded_frame->data[n];\r
 217                         auto decoded_linesize = decoded_frame->linesize[n];\r
 218                         \r
 219                         CASPAR_ASSERT(decoded);\r
 220                         CASPAR_ASSERT(write->image_data(n).begin());\r
 221 \r
 222                         if(decoded_linesize != static_cast<int>(plane.width))\r
 223                         {\r
 224                                 // Copy line by line since ffmpeg sometimes pads each line.\r
 225                                 tbb::parallel_for<size_t>(0, desc.planes[n].height, [&](size_t y)\r
 226                                 {\r
 227                                         fast_memcpy(result + y*plane.linesize, decoded + y*decoded_linesize, plane.linesize);\r
 228                                 });\r
 229                         }\r
 230                         else\r
 231                         {\r
 232                                 fast_memcpy(result, decoded, plane.size);\r
 233                         }\r
 234 \r
 235                         write->commit(n);\r
 236                 }\r
 237         }\r
 238 \r
 239         if(decoded_frame->height == 480) // NTSC DV\r
 240         {\r
 241                 write->get_frame_transform().fill_translation[1] += 2.0/static_cast<double>(frame_factory->get_video_format_desc().height);\r
 242                 write->get_frame_transform().fill_scale[1] = 1.0 - 6.0*1.0/static_cast<double>(frame_factory->get_video_format_desc().height);\r
 243         }\r
 244         \r
 245         // Fix field-order if needed\r
 246         if(write->get_type() == core::field_mode::lower && frame_factory->get_video_format_desc().field_mode == core::field_mode::upper)\r
 247                 write->get_frame_transform().fill_translation[1] += 1.0/static_cast<double>(frame_factory->get_video_format_desc().height);\r
 248         else if(write->get_type() == core::field_mode::upper && frame_factory->get_video_format_desc().field_mode == core::field_mode::lower)\r
 249                 write->get_frame_transform().fill_translation[1] -= 1.0/static_cast<double>(frame_factory->get_video_format_desc().height);\r
 250 \r
 251         return make_safe_ptr(write);\r
 252 }\r
 253 \r
 254 bool is_sane_fps(AVRational time_base)\r
 255 {\r
 256         double fps = static_cast<double>(time_base.den) / static_cast<double>(time_base.num);\r
 257         return fps > 20.0 && fps < 65.0;\r
 258 }\r
 259 \r
 260 AVRational fix_time_base(AVRational time_base)\r
 261 {\r
 262         if(time_base.num == 1)\r
 263                 time_base.num = static_cast<int>(std::pow(10.0, static_cast<int>(std::log10(static_cast<float>(time_base.den)))-1));    \r
 264                         \r
 265         if(!is_sane_fps(time_base))\r
 266         {\r
 267                 auto tmp = time_base;\r
 268                 tmp.den /= 2;\r
 269                 if(is_sane_fps(tmp))\r
 270                         time_base = tmp;\r
 271         }\r
 272 \r
 273         return time_base;\r
 274 }\r
 275 \r
 276 void fix_meta_data(AVFormatContext& context)\r
 277 {\r
 278         auto video_index = av_find_best_stream(&context, AVMEDIA_TYPE_VIDEO, -1, -1, 0, 0);\r
 279         auto audio_index = av_find_best_stream(&context, AVMEDIA_TYPE_AUDIO, -1, -1, 0, 0);\r
 280 \r
 281         if(video_index > -1)\r
 282         {\r
 283                 auto video_context = context.streams[video_index]->codec;\r
 284                 auto video_stream  = context.streams[video_index];\r
 285                                                 \r
 286                 if(boost::filesystem2::path(context.filename).extension() == ".flv")\r
 287                 {\r
 288                         try\r
 289                         {\r
 290                                 auto meta = read_flv_meta_info(context.filename);\r
 291                                 double fps = boost::lexical_cast<double>(meta["framerate"]);\r
 292                                 video_context->time_base.num = 1000000;\r
 293                                 video_context->time_base.den = static_cast<int>(fps*1000000.0);\r
 294                                 video_stream->nb_frames = static_cast<int64_t>(boost::lexical_cast<double>(meta["duration"])*fps);\r
 295                         }\r
 296                         catch(...){}\r
 297                 }\r
 298                 else\r
 299                 {\r
 300                         video_context->time_base.num *= video_context->ticks_per_frame;\r
 301 \r
 302                         if(!is_sane_fps(video_context->time_base))\r
 303                         {                       \r
 304                                 video_context->time_base = fix_time_base(video_context->time_base);\r
 305 \r
 306                                 if(!is_sane_fps(video_context->time_base) && audio_index > -1)\r
 307                                 {\r
 308                                         auto& audio_context = *context.streams[audio_index]->codec;\r
 309                                         auto& audio_stream  = *context.streams[audio_index];\r
 310 \r
 311                                         double duration_sec = audio_stream.duration / static_cast<double>(audio_context.sample_rate);\r
 312                                                                 \r
 313                                         video_context->time_base.num = static_cast<int>(duration_sec*100000.0);\r
 314                                         video_context->time_base.den = static_cast<int>(video_stream->nb_frames*100000);\r
 315                                 }\r
 316                         }\r
 317                                 \r
 318                         //if(audio_index > -1) // Check for invalid double frame-rate\r
 319                         //{\r
 320                         //      auto& audio_context             = *context.streams[audio_index]->codec;\r
 321                         //      auto& audio_stream              = *context.streams[audio_index];\r
 322                         //      \r
 323                         //      double duration_sec             = audio_stream.duration / static_cast<double>(audio_context.sample_rate);\r
 324                         //      double fps                              = static_cast<double>(video_context->time_base.den) / static_cast<double>(video_context->time_base.num);\r
 325 \r
 326                         //      double fps_nb_frames    = static_cast<double>(duration_sec*fps);\r
 327                         //      double stream_nb_frames = static_cast<double>(video_stream->nb_frames);\r
 328                         //      double diff                             = std::abs(fps_nb_frames - stream_nb_frames*2.0);\r
 329                         //      if(diff < fps_nb_frames*0.05)\r
 330                         //              video_context->time_base.num *= 2;\r
 331                         //}\r
 332                         //else\r
 333                         //{\r
 334                         //      video_context->time_base.den = video_stream->r_frame_rate.num;\r
 335                         //      video_context->time_base.num = video_stream->r_frame_rate.den;\r
 336                         //}\r
 337                 }\r
 338 \r
 339                 auto nb_frames = static_cast<double>(video_stream->duration*video_stream->time_base.num)/static_cast<double>(video_stream->time_base.den);\r
 340                 nb_frames = (nb_frames*video_context->time_base.den)/video_context->time_base.num;\r
 341                 video_stream->nb_frames = static_cast<int64_t>(nb_frames+0.5);\r
 342 \r
 343                 double fps = static_cast<double>(video_context->time_base.den) / static_cast<double>(video_context->time_base.num);\r
 344 \r
 345                 double closest_fps = 0.0;\r
 346                 for(int n = 0; n < core::video_format::count; ++n)\r
 347                 {\r
 348                         auto format = core::video_format_desc::get(static_cast<core::video_format::type>(n));\r
 349 \r
 350                         double diff1 = std::abs(format.fps - fps);\r
 351                         double diff2 = std::abs(closest_fps - fps);\r
 352 \r
 353                         if(diff1 < diff2)\r
 354                                 closest_fps = format.fps;\r
 355                 }\r
 356         \r
 357                 video_context->time_base.num = 1000000;\r
 358                 video_context->time_base.den = static_cast<int>(closest_fps*1000000.0);\r
 359         }\r
 360 }\r
 361 \r
 362 safe_ptr<AVPacket> create_packet()\r
 363 {\r
 364         safe_ptr<AVPacket> packet(new AVPacket, [](AVPacket* p)\r
 365         {\r
 366                 av_free_packet(p);\r
 367                 delete p;\r
 368         });\r
 369         \r
 370         av_init_packet(packet.get());\r
 371         return packet;\r
 372 }\r
 373 \r
 374 safe_ptr<AVCodecContext> open_codec(AVFormatContext& context, enum AVMediaType type, int& index)\r
 375 {       \r
 376         AVCodec* decoder;\r
 377         index = THROW_ON_ERROR2(av_find_best_stream(&context, type, -1, -1, &decoder, 0), "");\r
 378         //if(strcmp(decoder->name, "prores") == 0 && decoder->next && strcmp(decoder->next->name, "prores_lgpl") == 0)\r
 379         //      decoder = decoder->next;\r
 380 \r
 381         THROW_ON_ERROR2(tbb_avcodec_open(context.streams[index]->codec, decoder), "");\r
 382         return safe_ptr<AVCodecContext>(context.streams[index]->codec, tbb_avcodec_close);\r
 383 }\r
 384 \r
 385 safe_ptr<AVFormatContext> open_input(const std::wstring& filename)\r
 386 {\r
 387         AVFormatContext* weak_context = nullptr;\r
 388         THROW_ON_ERROR2(avformat_open_input(&weak_context, narrow(filename).c_str(), nullptr, nullptr), filename);\r
 389         safe_ptr<AVFormatContext> context(weak_context, av_close_input_file);                   \r
 390         THROW_ON_ERROR2(avformat_find_stream_info(weak_context, nullptr), filename);\r
 391         fix_meta_data(*context);\r
 392         return context;\r
 393 }\r
 394 //\r
 395 //void av_dup_frame(AVFrame* frame)\r
 396 //{\r
 397 //      AVFrame* new_frame = avcodec_alloc_frame();\r
 398 //\r
 399 //\r
 400 //      const uint8_t *src_data[4] = {0};\r
 401 //      memcpy(const_cast<uint8_t**>(&src_data[0]), frame->data, 4);\r
 402 //      const int src_linesizes[4] = {0};\r
 403 //      memcpy(const_cast<int*>(&src_linesizes[0]), frame->linesize, 4);\r
 404 //\r
 405 //      av_image_alloc(new_frame->data, new_frame->linesize, new_frame->width, new_frame->height, frame->format, 16);\r
 406 //\r
 407 //      av_image_copy(new_frame->data, new_frame->linesize, src_data, src_linesizes, frame->format, new_frame->width, new_frame->height);\r
 408 //\r
 409 //      frame =\r
 410 //}\r
 411 \r
 412 }}