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

   1 /*\r
   2 * Copyright (c) 2011 Sveriges Television AB <info@casparcg.com>\r
   3 *\r
   4 * This file is part of CasparCG (www.casparcg.com).\r
   5 *\r
   6 * CasparCG is free software: you can redistribute it and/or modify\r
   7 * it under the terms of the GNU General Public License as published by\r
   8 * the Free Software Foundation, either version 3 of the License, or\r
   9 * (at your option) any later version.\r
  10 *\r
  11 * CasparCG is distributed in the hope that it will be useful,\r
  12 * but WITHOUT ANY WARRANTY; without even the implied warranty of\r
  13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the\r
  14 * GNU General Public License for more details.\r
  15 *\r
  16 * You should have received a copy of the GNU General Public License\r
  17 * along with CasparCG. If not, see <http://www.gnu.org/licenses/>.\r
  18 *\r
  19 * Author: Robert Nagy, ronag89@gmail.com\r
  20 */\r
  21 \r
  22 #include "../../stdafx.h"\r
  23 \r
  24 #include "util.h"\r
  25 \r
  26 #include "flv.h"\r
  27 \r
  28 #include "../tbb_avcodec.h"\r
  29 #include "../../ffmpeg_error.h"\r
  30 \r
  31 #include <tbb/concurrent_unordered_map.h>\r
  32 #include <tbb/concurrent_queue.h>\r
  33 \r
  34 #include <core/producer/frame/frame_transform.h>\r
  35 #include <core/producer/frame/frame_factory.h>\r
  36 #include <core/producer/frame_producer.h>\r
  37 #include <core/mixer/write_frame.h>\r
  38 \r
  39 #include <common/exception/exceptions.h>\r
  40 #include <common/utility/assert.h>\r
  41 \r
  42 #include <tbb/parallel_for.h>\r
  43 \r
  44 #include <boost/filesystem.hpp>\r
  45 #include <boost/lexical_cast.hpp>\r
  46 \r
  47 #if defined(_MSC_VER)\r
  48 #pragma warning (push)\r
  49 #pragma warning (disable : 4244)\r
  50 #endif\r
  51 extern "C" \r
  52 {\r
  53         #include <libswscale/swscale.h>\r
  54         #include <libavcodec/avcodec.h>\r
  55         #include <libavformat/avformat.h>\r
  56 }\r
  57 #if defined(_MSC_VER)\r
  58 #pragma warning (pop)\r
  59 #endif\r
  60 \r
  61 namespace caspar { namespace ffmpeg {\r
  62                 \r
  63 std::shared_ptr<core::audio_buffer> flush_audio()\r
  64 {\r
  65         static std::shared_ptr<core::audio_buffer> audio(new core::audio_buffer());\r
  66         return audio;\r
  67 }\r
  68 \r
  69 std::shared_ptr<core::audio_buffer> empty_audio()\r
  70 {\r
  71         static std::shared_ptr<core::audio_buffer> audio(new core::audio_buffer());\r
  72         return audio;\r
  73 }\r
  74 \r
  75 std::shared_ptr<AVFrame>                        flush_video()\r
  76 {\r
  77         static std::shared_ptr<AVFrame> video(avcodec_alloc_frame(), av_free);\r
  78         return video;\r
  79 }\r
  80 \r
  81 std::shared_ptr<AVFrame>                        empty_video()\r
  82 {\r
  83         static std::shared_ptr<AVFrame> video(avcodec_alloc_frame(), av_free);\r
  84         return video;\r
  85 }\r
  86 \r
  87 core::field_mode::type get_mode(const AVFrame& frame)\r
  88 {\r
  89         if(!frame.interlaced_frame)\r
  90                 return core::field_mode::progressive;\r
  91 \r
  92         return frame.top_field_first ? core::field_mode::upper : core::field_mode::lower;\r
  93 }\r
  94 \r
  95 core::pixel_format::type get_pixel_format(PixelFormat pix_fmt)\r
  96 {\r
  97         switch(pix_fmt)\r
  98         {\r
  99         case CASPAR_PIX_FMT_LUMA:       return core::pixel_format::luma;\r
 100         case PIX_FMT_GRAY8:                     return core::pixel_format::gray;\r
 101         case PIX_FMT_BGRA:                      return core::pixel_format::bgra;\r
 102         case PIX_FMT_ARGB:                      return core::pixel_format::argb;\r
 103         case PIX_FMT_RGBA:                      return core::pixel_format::rgba;\r
 104         case PIX_FMT_ABGR:                      return core::pixel_format::abgr;\r
 105         case PIX_FMT_YUV444P:           return core::pixel_format::ycbcr;\r
 106         case PIX_FMT_YUV422P:           return core::pixel_format::ycbcr;\r
 107         case PIX_FMT_YUV420P:           return core::pixel_format::ycbcr;\r
 108         case PIX_FMT_YUV411P:           return core::pixel_format::ycbcr;\r
 109         case PIX_FMT_YUV410P:           return core::pixel_format::ycbcr;\r
 110         case PIX_FMT_YUVA420P:          return core::pixel_format::ycbcra;\r
 111         default:                                        return core::pixel_format::invalid;\r
 112         }\r
 113 }\r
 114 \r
 115 core::pixel_format_desc get_pixel_format_desc(PixelFormat pix_fmt, int width, int height)\r
 116 {\r
 117         // Get linesizes\r
 118         AVPicture dummy_pict;   \r
 119         avpicture_fill(&dummy_pict, nullptr, pix_fmt == CASPAR_PIX_FMT_LUMA ? PIX_FMT_GRAY8 : pix_fmt, width, height);\r
 120 \r
 121         core::pixel_format_desc desc;\r
 122         desc.pix_fmt = get_pixel_format(pix_fmt);\r
 123                 \r
 124         switch(desc.pix_fmt)\r
 125         {\r
 126         case core::pixel_format::gray:\r
 127         case core::pixel_format::luma:\r
 128                 {\r
 129                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));                                               \r
 130                         return desc;\r
 131                 }\r
 132         case core::pixel_format::bgra:\r
 133         case core::pixel_format::argb:\r
 134         case core::pixel_format::rgba:\r
 135         case core::pixel_format::abgr:\r
 136                 {\r
 137                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0]/4, height, 4));                                             \r
 138                         return desc;\r
 139                 }\r
 140         case core::pixel_format::ycbcr:\r
 141         case core::pixel_format::ycbcra:\r
 142                 {               \r
 143                         // Find chroma height\r
 144                         int size2 = static_cast<int>(dummy_pict.data[2] - dummy_pict.data[1]);\r
 145                         int h2 = size2/dummy_pict.linesize[1];                  \r
 146 \r
 147                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));\r
 148                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[1], h2, 1));\r
 149                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[2], h2, 1));\r
 150 \r
 151                         if(desc.pix_fmt == core::pixel_format::ycbcra)                                          \r
 152                                 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[3], height, 1));       \r
 153                         return desc;\r
 154                 }               \r
 155         default:                \r
 156                 desc.pix_fmt = core::pixel_format::invalid;\r
 157                 return desc;\r
 158         }\r
 159 }\r
 160 \r
 161 int make_alpha_format(int format)\r
 162 {\r
 163         switch(get_pixel_format(static_cast<PixelFormat>(format)))\r
 164         {\r
 165         case core::pixel_format::ycbcr:\r
 166         case core::pixel_format::ycbcra:\r
 167                 return CASPAR_PIX_FMT_LUMA;\r
 168         default:\r
 169                 return format;\r
 170         }\r
 171 }\r
 172 \r
 173 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 flags)\r
 174 {                       \r
 175         static tbb::concurrent_unordered_map<int, tbb::concurrent_queue<std::shared_ptr<SwsContext>>> sws_contexts_;\r
 176         \r
 177         if(decoded_frame->width < 1 || decoded_frame->height < 1)\r
 178                 return make_safe<core::write_frame>(tag);\r
 179 \r
 180         const auto width  = decoded_frame->width;\r
 181         const auto height = decoded_frame->height;\r
 182         auto desc                 = get_pixel_format_desc(static_cast<PixelFormat>(decoded_frame->format), width, height);\r
 183         \r
 184         if(flags & core::frame_producer::ALPHA_ONLY_FLAG)\r
 185                 desc = get_pixel_format_desc(static_cast<PixelFormat>(make_alpha_format(decoded_frame->format)), width, height);\r
 186 \r
 187         std::shared_ptr<core::write_frame> write;\r
 188 \r
 189         if(desc.pix_fmt == core::pixel_format::invalid)\r
 190         {\r
 191                 auto pix_fmt = static_cast<PixelFormat>(decoded_frame->format);\r
 192                 auto target_pix_fmt = PIX_FMT_BGRA;\r
 193 \r
 194                 if(pix_fmt == PIX_FMT_UYVY422)\r
 195                         target_pix_fmt = PIX_FMT_YUV422P;\r
 196                 else if(pix_fmt == PIX_FMT_YUYV422)\r
 197                         target_pix_fmt = PIX_FMT_YUV422P;\r
 198                 else if(pix_fmt == PIX_FMT_UYYVYY411)\r
 199                         target_pix_fmt = PIX_FMT_YUV411P;\r
 200                 else if(pix_fmt == PIX_FMT_YUV420P10)\r
 201                         target_pix_fmt = PIX_FMT_YUV420P;\r
 202                 else if(pix_fmt == PIX_FMT_YUV422P10)\r
 203                         target_pix_fmt = PIX_FMT_YUV422P;\r
 204                 else if(pix_fmt == PIX_FMT_YUV444P10)\r
 205                         target_pix_fmt = PIX_FMT_YUV444P;\r
 206                 \r
 207                 auto target_desc = get_pixel_format_desc(target_pix_fmt, width, height);\r
 208 \r
 209                 write = frame_factory->create_frame(tag, target_desc);\r
 210                 write->set_type(get_mode(*decoded_frame));\r
 211 \r
 212                 std::shared_ptr<SwsContext> sws_context;\r
 213 \r
 214                 //CASPAR_LOG(warning) << "Hardware accelerated color transform not supported.";\r
 215 \r
 216                 int key = ((width << 22) & 0xFFC00000) | ((height << 6) & 0x003FC000) | ((pix_fmt << 7) & 0x00007F00) | ((target_pix_fmt << 0) & 0x0000007F);\r
 217                         \r
 218                 auto& pool = sws_contexts_[key];\r
 219                                                 \r
 220                 if(!pool.try_pop(sws_context))\r
 221                 {\r
 222                         double param;\r
 223                         sws_context.reset(sws_getContext(width, height, pix_fmt, width, height, target_pix_fmt, SWS_BILINEAR, nullptr, nullptr, &param), sws_freeContext);\r
 224                 }\r
 225                         \r
 226                 if(!sws_context)\r
 227                 {\r
 228                         BOOST_THROW_EXCEPTION(operation_failed() << msg_info("Could not create software scaling context.") << \r
 229                                                                         boost::errinfo_api_function("sws_getContext"));\r
 230                 }       \r
 231                 \r
 232                 safe_ptr<AVFrame> av_frame(avcodec_alloc_frame(), av_free);     \r
 233                 avcodec_get_frame_defaults(av_frame.get());                     \r
 234                 if(target_pix_fmt == PIX_FMT_BGRA)\r
 235                 {\r
 236                         auto size = avpicture_fill(reinterpret_cast<AVPicture*>(av_frame.get()), write->image_data().begin(), PIX_FMT_BGRA, width, height);\r
 237                         CASPAR_VERIFY(size == write->image_data().size()); \r
 238                 }\r
 239                 else\r
 240                 {\r
 241                         av_frame->width  = width;\r
 242                         av_frame->height = height;\r
 243                         for(int n = 0; n < target_desc.planes.size(); ++n)\r
 244                         {\r
 245                                 av_frame->data[n]               = write->image_data(n).begin();\r
 246                                 av_frame->linesize[n]   = target_desc.planes[n].linesize;\r
 247                         }\r
 248                 }\r
 249 \r
 250                 sws_scale(sws_context.get(), decoded_frame->data, decoded_frame->linesize, 0, height, av_frame->data, av_frame->linesize);      \r
 251                 pool.push(sws_context);\r
 252 \r
 253                 write->commit();                \r
 254         }\r
 255         else\r
 256         {\r
 257                 write = frame_factory->create_frame(tag, desc);\r
 258                 write->set_type(get_mode(*decoded_frame));\r
 259 \r
 260                 for(int n = 0; n < static_cast<int>(desc.planes.size()); ++n)\r
 261                 {\r
 262                         auto plane            = desc.planes[n];\r
 263                         auto result           = write->image_data(n).begin();\r
 264                         auto decoded          = decoded_frame->data[n];\r
 265                         auto decoded_linesize = decoded_frame->linesize[n];\r
 266                         \r
 267                         CASPAR_ASSERT(decoded);\r
 268                         CASPAR_ASSERT(write->image_data(n).begin());\r
 269 \r
 270                         // Copy line by line since ffmpeg sometimes pads each line.\r
 271                         tbb::parallel_for<int>(0, desc.planes[n].height, [&](int y)\r
 272                         {\r
 273                                 memcpy(result + y*plane.linesize, decoded + y*decoded_linesize, plane.linesize);\r
 274                         });\r
 275 \r
 276                         write->commit(n);\r
 277                 }\r
 278         }\r
 279 \r
 280         if(decoded_frame->height == 480) // NTSC DV\r
 281         {\r
 282                 write->get_frame_transform().fill_translation[1] += 2.0/static_cast<double>(frame_factory->get_video_format_desc().height);\r
 283                 write->get_frame_transform().fill_scale[1] = 1.0 - 6.0*1.0/static_cast<double>(frame_factory->get_video_format_desc().height);\r
 284         }\r
 285         \r
 286         // Fix field-order if needed\r
 287         if(write->get_type() == core::field_mode::lower && frame_factory->get_video_format_desc().field_mode == core::field_mode::upper)\r
 288                 write->get_frame_transform().fill_translation[1] += 1.0/static_cast<double>(frame_factory->get_video_format_desc().height);\r
 289         else if(write->get_type() == core::field_mode::upper && frame_factory->get_video_format_desc().field_mode == core::field_mode::lower)\r
 290                 write->get_frame_transform().fill_translation[1] -= 1.0/static_cast<double>(frame_factory->get_video_format_desc().height);\r
 291 \r
 292         return make_safe_ptr(write);\r
 293 }\r
 294 \r
 295 bool is_sane_fps(AVRational time_base)\r
 296 {\r
 297         double fps = static_cast<double>(time_base.den) / static_cast<double>(time_base.num);\r
 298         return fps > 20.0 && fps < 65.0;\r
 299 }\r
 300 \r
 301 AVRational fix_time_base(AVRational time_base)\r
 302 {\r
 303         if(time_base.num == 1)\r
 304                 time_base.num = static_cast<int>(std::pow(10.0, static_cast<int>(std::log10(static_cast<float>(time_base.den)))-1));    \r
 305                         \r
 306         if(!is_sane_fps(time_base))\r
 307         {\r
 308                 auto tmp = time_base;\r
 309                 tmp.den /= 2;\r
 310                 if(is_sane_fps(tmp))\r
 311                         time_base = tmp;\r
 312         }\r
 313 \r
 314         return time_base;\r
 315 }\r
 316 \r
 317 double read_fps(AVFormatContext& context, double fail_value)\r
 318 {                                               \r
 319         auto video_index = av_find_best_stream(&context, AVMEDIA_TYPE_VIDEO, -1, -1, 0, 0);\r
 320         auto audio_index = av_find_best_stream(&context, AVMEDIA_TYPE_AUDIO, -1, -1, 0, 0);\r
 321         \r
 322         if(video_index > -1)\r
 323         {\r
 324                 const auto video_context = context.streams[video_index]->codec;\r
 325                 const auto video_stream  = context.streams[video_index];\r
 326                                                 \r
 327                 AVRational time_base = video_context->time_base;\r
 328 \r
 329                 if(boost::filesystem::path(context.filename).extension().string() == ".flv")\r
 330                 {\r
 331                         try\r
 332                         {\r
 333                                 auto meta = read_flv_meta_info(context.filename);\r
 334                                 return boost::lexical_cast<double>(meta["framerate"]);\r
 335                         }\r
 336                         catch(...)\r
 337                         {\r
 338                                 return 0.0;\r
 339                         }\r
 340                 }\r
 341                 else\r
 342                 {\r
 343                         time_base.num *= video_context->ticks_per_frame;\r
 344 \r
 345                         if(!is_sane_fps(time_base))\r
 346                         {                       \r
 347                                 time_base = fix_time_base(time_base);\r
 348 \r
 349                                 if(!is_sane_fps(time_base) && audio_index > -1)\r
 350                                 {\r
 351                                         auto& audio_context = *context.streams[audio_index]->codec;\r
 352                                         auto& audio_stream  = *context.streams[audio_index];\r
 353 \r
 354                                         double duration_sec = audio_stream.duration / static_cast<double>(audio_context.sample_rate);\r
 355                                                                 \r
 356                                         time_base.num = static_cast<int>(duration_sec*100000.0);\r
 357                                         time_base.den = static_cast<int>(video_stream->nb_frames*100000);\r
 358                                 }\r
 359                         }\r
 360                 }\r
 361                 \r
 362                 double fps = static_cast<double>(time_base.den) / static_cast<double>(time_base.num);\r
 363 \r
 364                 double closest_fps = 0.0;\r
 365                 for(int n = 0; n < core::video_format::count; ++n)\r
 366                 {\r
 367                         auto format = core::video_format_desc::get(static_cast<core::video_format::type>(n));\r
 368 \r
 369                         double diff1 = std::abs(format.fps - fps);\r
 370                         double diff2 = std::abs(closest_fps - fps);\r
 371 \r
 372                         if(diff1 < diff2)\r
 373                                 closest_fps = format.fps;\r
 374                 }\r
 375         \r
 376                 return closest_fps;\r
 377         }\r
 378 \r
 379         return fail_value;      \r
 380 }\r
 381 \r
 382 void fix_meta_data(AVFormatContext& context)\r
 383 {\r
 384         auto video_index = av_find_best_stream(&context, AVMEDIA_TYPE_VIDEO, -1, -1, 0, 0);\r
 385 \r
 386         if(video_index > -1)\r
 387         {\r
 388                 auto video_stream   = context.streams[video_index];\r
 389                 auto video_context  = context.streams[video_index]->codec;\r
 390                                                 \r
 391                 if(boost::filesystem::path(context.filename).extension().string() == ".flv")\r
 392                 {\r
 393                         try\r
 394                         {\r
 395                                 auto meta = read_flv_meta_info(context.filename);\r
 396                                 double fps = boost::lexical_cast<double>(meta["framerate"]);\r
 397                                 video_stream->nb_frames = static_cast<int64_t>(boost::lexical_cast<double>(meta["duration"])*fps);\r
 398                         }\r
 399                         catch(...){}\r
 400                 }\r
 401                 else\r
 402                 {\r
 403                         auto stream_time = video_stream->time_base;\r
 404                         auto duration    = video_stream->duration;\r
 405                         auto codec_time  = video_context->time_base;\r
 406                         auto ticks               = video_context->ticks_per_frame;\r
 407 \r
 408                         if(video_stream->nb_frames == 0)\r
 409                                 video_stream->nb_frames = (duration*stream_time.num*codec_time.den)/(stream_time.den*codec_time.num*ticks);     \r
 410                 }\r
 411         }\r
 412 }\r
 413 \r
 414 safe_ptr<AVPacket> create_packet()\r
 415 {\r
 416         safe_ptr<AVPacket> packet(new AVPacket, [](AVPacket* p)\r
 417         {\r
 418                 av_free_packet(p);\r
 419                 delete p;\r
 420         });\r
 421         \r
 422         av_init_packet(packet.get());\r
 423         return packet;\r
 424 }\r
 425 \r
 426 safe_ptr<AVCodecContext> open_codec(AVFormatContext& context, enum AVMediaType type, int& index)\r
 427 {       \r
 428         AVCodec* decoder;\r
 429         index = THROW_ON_ERROR2(av_find_best_stream(&context, type, -1, -1, &decoder, 0), "");\r
 430         //if(strcmp(decoder->name, "prores") == 0 && decoder->next && strcmp(decoder->next->name, "prores_lgpl") == 0)\r
 431         //      decoder = decoder->next;\r
 432 \r
 433         THROW_ON_ERROR2(tbb_avcodec_open(context.streams[index]->codec, decoder), "");\r
 434         return safe_ptr<AVCodecContext>(context.streams[index]->codec, tbb_avcodec_close);\r
 435 }\r
 436 \r
 437 safe_ptr<AVFormatContext> open_input(const std::wstring& filename)\r
 438 {\r
 439         AVFormatContext* weak_context = nullptr;\r
 440         THROW_ON_ERROR2(avformat_open_input(&weak_context, u8(filename).c_str(), nullptr, nullptr), filename);\r
 441         safe_ptr<AVFormatContext> context(weak_context, av_close_input_file);                   \r
 442         THROW_ON_ERROR2(avformat_find_stream_info(weak_context, nullptr), filename);\r
 443         fix_meta_data(*context);\r
 444         return context;\r
 445 }\r
 446 \r
 447 std::wstring print_mode(int width, int height, double fps, bool interlaced)\r
 448 {\r
 449         std::wostringstream fps_ss;\r
 450         fps_ss << std::fixed << std::setprecision(2) << (!interlaced ? fps : 2.0 * fps);\r
 451 \r
 452         return boost::lexical_cast<std::wstring>(width) + L"x" + boost::lexical_cast<std::wstring>(height) + (!interlaced ? L"p" : L"i") + fps_ss.str();\r
 453 }\r
 454 \r
 455 bool is_valid_file(const std::wstring filename)\r
 456 {                       \r
 457         std::ifstream file(filename);\r
 458 \r
 459         std::vector<unsigned char> buf;\r
 460         for(auto file_it = std::istreambuf_iterator<char>(file); file_it != std::istreambuf_iterator<char>() && buf.size() < 2048; ++file_it)\r
 461                 buf.push_back(*file_it);\r
 462 \r
 463         if(buf.empty())\r
 464                 return nullptr;\r
 465 \r
 466         AVProbeData pb;\r
 467         pb.filename = u8(filename).c_str();\r
 468         pb.buf          = buf.data();\r
 469         pb.buf_size = static_cast<int>(buf.size());\r
 470 \r
 471         int score = 0;\r
 472         return av_probe_input_format2(&pb, true, &score) != nullptr;\r
 473 }\r
 474 \r
 475 std::wstring probe_stem(const std::wstring stem)\r
 476 {\r
 477         auto stem2 = boost::filesystem::path(stem);\r
 478         auto dir = stem2.parent_path();\r
 479         for(auto it = boost::filesystem::directory_iterator(dir); it != boost::filesystem::directory_iterator(); ++it)\r
 480         {\r
 481                 if(boost::iequals(it->path().stem().wstring(), stem2.filename().wstring()) && is_valid_file(it->path().wstring()))\r
 482                         return it->path().wstring();\r
 483         }\r
 484         return L"";\r
 485 }\r
 486 //\r
 487 //void av_dup_frame(AVFrame* frame)\r
 488 //{\r
 489 //      AVFrame* new_frame = avcodec_alloc_frame();\r
 490 //\r
 491 //\r
 492 //      const uint8_t *src_data[4] = {0};\r
 493 //      memcpy(const_cast<uint8_t**>(&src_data[0]), frame->data, 4);\r
 494 //      const int src_linesizes[4] = {0};\r
 495 //      memcpy(const_cast<int*>(&src_linesizes[0]), frame->linesize, 4);\r
 496 //\r
 497 //      av_image_alloc(new_frame->data, new_frame->linesize, new_frame->width, new_frame->height, frame->format, 16);\r
 498 //\r
 499 //      av_image_copy(new_frame->data, new_frame->linesize, src_data, src_linesizes, frame->format, new_frame->width, new_frame->height);\r
 500 //\r
 501 //      frame =\r
 502 //}\r
 503 \r
 504 }}