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 #include <common/memory/memcpy.h>\r
  42 \r
  43 #include <tbb/parallel_for.h>\r
  44 \r
  45 #include <boost/filesystem.hpp>\r
  46 #include <boost/lexical_cast.hpp>\r
  47 \r
  48 #if defined(_MSC_VER)\r
  49 #pragma warning (push)\r
  50 #pragma warning (disable : 4244)\r
  51 #endif\r
  52 extern "C" \r
  53 {\r
  54         #include <libswscale/swscale.h>\r
  55         #include <libavcodec/avcodec.h>\r
  56         #include <libavformat/avformat.h>\r
  57 }\r
  58 #if defined(_MSC_VER)\r
  59 #pragma warning (pop)\r
  60 #endif\r
  61 \r
  62 namespace caspar { namespace ffmpeg {\r
  63                 \r
  64 std::shared_ptr<core::audio_buffer> flush_audio()\r
  65 {\r
  66         static std::shared_ptr<core::audio_buffer> audio(new core::audio_buffer());\r
  67         return audio;\r
  68 }\r
  69 \r
  70 std::shared_ptr<core::audio_buffer> empty_audio()\r
  71 {\r
  72         static std::shared_ptr<core::audio_buffer> audio(new core::audio_buffer());\r
  73         return audio;\r
  74 }\r
  75 \r
  76 std::shared_ptr<AVFrame>                        flush_video()\r
  77 {\r
  78         static std::shared_ptr<AVFrame> video(avcodec_alloc_frame(), av_free);\r
  79         return video;\r
  80 }\r
  81 \r
  82 std::shared_ptr<AVFrame>                        empty_video()\r
  83 {\r
  84         static std::shared_ptr<AVFrame> video(avcodec_alloc_frame(), av_free);\r
  85         return video;\r
  86 }\r
  87 \r
  88 core::field_mode::type get_mode(const AVFrame& frame)\r
  89 {\r
  90         if(!frame.interlaced_frame)\r
  91                 return core::field_mode::progressive;\r
  92 \r
  93         return frame.top_field_first ? core::field_mode::upper : core::field_mode::lower;\r
  94 }\r
  95 \r
  96 core::pixel_format::type get_pixel_format(PixelFormat pix_fmt)\r
  97 {\r
  98         switch(pix_fmt)\r
  99         {\r
 100         case CASPAR_PIX_FMT_LUMA:       return core::pixel_format::luma;\r
 101         case PIX_FMT_GRAY8:                     return core::pixel_format::gray;\r
 102         case PIX_FMT_BGRA:                      return core::pixel_format::bgra;\r
 103         case PIX_FMT_ARGB:                      return core::pixel_format::argb;\r
 104         case PIX_FMT_RGBA:                      return core::pixel_format::rgba;\r
 105         case PIX_FMT_ABGR:                      return core::pixel_format::abgr;\r
 106         case PIX_FMT_YUV444P:           return core::pixel_format::ycbcr;\r
 107         case PIX_FMT_YUV422P:           return core::pixel_format::ycbcr;\r
 108         case PIX_FMT_YUV420P:           return core::pixel_format::ycbcr;\r
 109         case PIX_FMT_YUV411P:           return core::pixel_format::ycbcr;\r
 110         case PIX_FMT_YUV410P:           return core::pixel_format::ycbcr;\r
 111         case PIX_FMT_YUVA420P:          return core::pixel_format::ycbcra;\r
 112         default:                                        return core::pixel_format::invalid;\r
 113         }\r
 114 }\r
 115 \r
 116 core::pixel_format_desc get_pixel_format_desc(PixelFormat pix_fmt, size_t width, size_t height)\r
 117 {\r
 118         // Get linesizes\r
 119         AVPicture dummy_pict;   \r
 120         avpicture_fill(&dummy_pict, nullptr, pix_fmt == CASPAR_PIX_FMT_LUMA ? PIX_FMT_GRAY8 : pix_fmt, width, height);\r
 121 \r
 122         core::pixel_format_desc desc;\r
 123         desc.pix_fmt = get_pixel_format(pix_fmt);\r
 124                 \r
 125         switch(desc.pix_fmt)\r
 126         {\r
 127         case core::pixel_format::gray:\r
 128         case core::pixel_format::luma:\r
 129                 {\r
 130                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));                                               \r
 131                         return desc;\r
 132                 }\r
 133         case core::pixel_format::bgra:\r
 134         case core::pixel_format::argb:\r
 135         case core::pixel_format::rgba:\r
 136         case core::pixel_format::abgr:\r
 137                 {\r
 138                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0]/4, height, 4));                                             \r
 139                         return desc;\r
 140                 }\r
 141         case core::pixel_format::ycbcr:\r
 142         case core::pixel_format::ycbcra:\r
 143                 {               \r
 144                         // Find chroma height\r
 145                         size_t size2 = dummy_pict.data[2] - dummy_pict.data[1];\r
 146                         size_t h2 = size2/dummy_pict.linesize[1];                       \r
 147 \r
 148                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));\r
 149                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[1], h2, 1));\r
 150                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[2], h2, 1));\r
 151 \r
 152                         if(desc.pix_fmt == core::pixel_format::ycbcra)                                          \r
 153                                 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[3], height, 1));       \r
 154                         return desc;\r
 155                 }               \r
 156         default:                \r
 157                 desc.pix_fmt = core::pixel_format::invalid;\r
 158                 return desc;\r
 159         }\r
 160 }\r
 161 \r
 162 int make_alpha_format(int format)\r
 163 {\r
 164         switch(get_pixel_format(static_cast<PixelFormat>(format)))\r
 165         {\r
 166         case core::pixel_format::ycbcr:\r
 167         case core::pixel_format::ycbcra:\r
 168                 return CASPAR_PIX_FMT_LUMA;\r
 169         default:\r
 170                 return format;\r
 171         }\r
 172 }\r
 173 \r
 174 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
 175 {                       \r
 176         static tbb::concurrent_unordered_map<size_t, tbb::concurrent_queue<std::shared_ptr<SwsContext>>> sws_contexts_;\r
 177         \r
 178         if(decoded_frame->width < 1 || decoded_frame->height < 1)\r
 179                 return make_safe<core::write_frame>(tag);\r
 180 \r
 181         const auto width  = decoded_frame->width;\r
 182         const auto height = decoded_frame->height;\r
 183         auto desc                 = get_pixel_format_desc(static_cast<PixelFormat>(decoded_frame->format), width, height);\r
 184         \r
 185         if(hints & core::frame_producer::ALPHA_HINT)\r
 186                 desc = get_pixel_format_desc(static_cast<PixelFormat>(make_alpha_format(decoded_frame->format)), width, height);\r
 187 \r
 188         std::shared_ptr<core::write_frame> write;\r
 189 \r
 190         if(desc.pix_fmt == core::pixel_format::invalid)\r
 191         {\r
 192                 auto pix_fmt = static_cast<PixelFormat>(decoded_frame->format);\r
 193 \r
 194                 write = frame_factory->create_frame(tag, get_pixel_format_desc(PIX_FMT_BGRA, width, height));\r
 195                 write->set_type(get_mode(*decoded_frame));\r
 196 \r
 197                 std::shared_ptr<SwsContext> sws_context;\r
 198 \r
 199                 //CASPAR_LOG(warning) << "Hardware accelerated color transform not supported.";\r
 200 \r
 201                 size_t key = width << 20 | height << 8 | pix_fmt;\r
 202                         \r
 203                 auto& pool = sws_contexts_[key];\r
 204                                                 \r
 205                 if(!pool.try_pop(sws_context))\r
 206                 {\r
 207                         double param;\r
 208                         sws_context.reset(sws_getContext(width, height, pix_fmt, width, height, PIX_FMT_BGRA, SWS_BILINEAR, nullptr, nullptr, &param), sws_freeContext);\r
 209                 }\r
 210                         \r
 211                 if(!sws_context)\r
 212                 {\r
 213                         BOOST_THROW_EXCEPTION(operation_failed() << msg_info("Could not create software scaling context.") << \r
 214                                                                         boost::errinfo_api_function("sws_getContext"));\r
 215                 }       \r
 216 \r
 217                 // Use sws_scale when provided colorspace has no hw-accel.\r
 218                 safe_ptr<AVFrame> av_frame(avcodec_alloc_frame(), av_free);     \r
 219                 avcodec_get_frame_defaults(av_frame.get());                     \r
 220                 auto size = avpicture_fill(reinterpret_cast<AVPicture*>(av_frame.get()), write->image_data().begin(), PIX_FMT_BGRA, width, height);\r
 221                 CASPAR_VERIFY(size == write->image_data().size()); \r
 222 \r
 223                 sws_scale(sws_context.get(), decoded_frame->data, decoded_frame->linesize, 0, height, av_frame->data, av_frame->linesize);      \r
 224                 pool.push(sws_context);\r
 225 \r
 226                 write->commit();\r
 227         }\r
 228         else\r
 229         {\r
 230                 write = frame_factory->create_frame(tag, desc);\r
 231                 write->set_type(get_mode(*decoded_frame));\r
 232 \r
 233                 for(int n = 0; n < static_cast<int>(desc.planes.size()); ++n)\r
 234                 {\r
 235                         auto plane            = desc.planes[n];\r
 236                         auto result           = write->image_data(n).begin();\r
 237                         auto decoded          = decoded_frame->data[n];\r
 238                         auto decoded_linesize = decoded_frame->linesize[n];\r
 239                         \r
 240                         CASPAR_ASSERT(decoded);\r
 241                         CASPAR_ASSERT(write->image_data(n).begin());\r
 242 \r
 243                         if(decoded_linesize != static_cast<int>(plane.width))\r
 244                         {\r
 245                                 // Copy line by line since ffmpeg sometimes pads each line.\r
 246                                 tbb::parallel_for<size_t>(0, desc.planes[n].height, [&](size_t y)\r
 247                                 {\r
 248                                         fast_memcpy(result + y*plane.linesize, decoded + y*decoded_linesize, plane.linesize);\r
 249                                 });\r
 250                         }\r
 251                         else\r
 252                         {\r
 253                                 fast_memcpy(result, decoded, plane.size);\r
 254                         }\r
 255 \r
 256                         write->commit(n);\r
 257                 }\r
 258         }\r
 259 \r
 260         if(decoded_frame->height == 480) // NTSC DV\r
 261         {\r
 262                 write->get_frame_transform().fill_translation[1] += 2.0/static_cast<double>(frame_factory->get_video_format_desc().height);\r
 263                 write->get_frame_transform().fill_scale[1] = 1.0 - 6.0*1.0/static_cast<double>(frame_factory->get_video_format_desc().height);\r
 264         }\r
 265         \r
 266         // Fix field-order if needed\r
 267         if(write->get_type() == core::field_mode::lower && frame_factory->get_video_format_desc().field_mode == core::field_mode::upper)\r
 268                 write->get_frame_transform().fill_translation[1] += 1.0/static_cast<double>(frame_factory->get_video_format_desc().height);\r
 269         else if(write->get_type() == core::field_mode::upper && frame_factory->get_video_format_desc().field_mode == core::field_mode::lower)\r
 270                 write->get_frame_transform().fill_translation[1] -= 1.0/static_cast<double>(frame_factory->get_video_format_desc().height);\r
 271 \r
 272         return make_safe_ptr(write);\r
 273 }\r
 274 \r
 275 bool is_sane_fps(AVRational time_base)\r
 276 {\r
 277         double fps = static_cast<double>(time_base.den) / static_cast<double>(time_base.num);\r
 278         return fps > 20.0 && fps < 65.0;\r
 279 }\r
 280 \r
 281 AVRational fix_time_base(AVRational time_base)\r
 282 {\r
 283         if(time_base.num == 1)\r
 284                 time_base.num = static_cast<int>(std::pow(10.0, static_cast<int>(std::log10(static_cast<float>(time_base.den)))-1));    \r
 285                         \r
 286         if(!is_sane_fps(time_base))\r
 287         {\r
 288                 auto tmp = time_base;\r
 289                 tmp.den /= 2;\r
 290                 if(is_sane_fps(tmp))\r
 291                         time_base = tmp;\r
 292         }\r
 293 \r
 294         return time_base;\r
 295 }\r
 296 \r
 297 double read_fps(AVFormatContext& context, double fail_value)\r
 298 {                                               \r
 299         auto video_index = av_find_best_stream(&context, AVMEDIA_TYPE_VIDEO, -1, -1, 0, 0);\r
 300         auto audio_index = av_find_best_stream(&context, AVMEDIA_TYPE_AUDIO, -1, -1, 0, 0);\r
 301         \r
 302         if(video_index > -1)\r
 303         {\r
 304                 const auto video_context = context.streams[video_index]->codec;\r
 305                 const auto video_stream  = context.streams[video_index];\r
 306                                                 \r
 307                 AVRational time_base = video_context->time_base;\r
 308 \r
 309                 if(boost::filesystem2::path(context.filename).extension() == ".flv")\r
 310                 {\r
 311                         try\r
 312                         {\r
 313                                 auto meta = read_flv_meta_info(context.filename);\r
 314                                 return boost::lexical_cast<double>(meta["framerate"]);\r
 315                         }\r
 316                         catch(...)\r
 317                         {\r
 318                                 return 0.0;\r
 319                         }\r
 320                 }\r
 321                 else\r
 322                 {\r
 323                         time_base.num *= video_context->ticks_per_frame;\r
 324 \r
 325                         if(!is_sane_fps(time_base))\r
 326                         {                       \r
 327                                 time_base = fix_time_base(time_base);\r
 328 \r
 329                                 if(!is_sane_fps(time_base) && audio_index > -1)\r
 330                                 {\r
 331                                         auto& audio_context = *context.streams[audio_index]->codec;\r
 332                                         auto& audio_stream  = *context.streams[audio_index];\r
 333 \r
 334                                         double duration_sec = audio_stream.duration / static_cast<double>(audio_context.sample_rate);\r
 335                                                                 \r
 336                                         time_base.num = static_cast<int>(duration_sec*100000.0);\r
 337                                         time_base.den = static_cast<int>(video_stream->nb_frames*100000);\r
 338                                 }\r
 339                         }\r
 340                 }\r
 341                 \r
 342                 double fps = static_cast<double>(time_base.den) / static_cast<double>(time_base.num);\r
 343 \r
 344                 double closest_fps = 0.0;\r
 345                 for(int n = 0; n < core::video_format::count; ++n)\r
 346                 {\r
 347                         auto format = core::video_format_desc::get(static_cast<core::video_format::type>(n));\r
 348 \r
 349                         double diff1 = std::abs(format.fps - fps);\r
 350                         double diff2 = std::abs(closest_fps - fps);\r
 351 \r
 352                         if(diff1 < diff2)\r
 353                                 closest_fps = format.fps;\r
 354                 }\r
 355         \r
 356                 return closest_fps;\r
 357         }\r
 358 \r
 359         return fail_value;      \r
 360 }\r
 361 \r
 362 void fix_meta_data(AVFormatContext& context)\r
 363 {\r
 364         auto video_index = av_find_best_stream(&context, AVMEDIA_TYPE_VIDEO, -1, -1, 0, 0);\r
 365 \r
 366         if(video_index > -1)\r
 367         {\r
 368                 auto video_stream   = context.streams[video_index];\r
 369                 auto video_context  = context.streams[video_index]->codec;\r
 370                                                 \r
 371                 if(boost::filesystem2::path(context.filename).extension() == ".flv")\r
 372                 {\r
 373                         try\r
 374                         {\r
 375                                 auto meta = read_flv_meta_info(context.filename);\r
 376                                 double fps = boost::lexical_cast<double>(meta["framerate"]);\r
 377                                 video_stream->nb_frames = static_cast<int64_t>(boost::lexical_cast<double>(meta["duration"])*fps);\r
 378                         }\r
 379                         catch(...){}\r
 380                 }\r
 381                 else\r
 382                 {\r
 383                         auto stream_time = video_stream->time_base;\r
 384                         auto duration    = video_stream->duration;\r
 385                         auto codec_time  = video_context->time_base;\r
 386                         auto ticks               = video_context->ticks_per_frame;\r
 387 \r
 388                         if(video_stream->nb_frames == 0)\r
 389                                 video_stream->nb_frames = (duration*stream_time.num*codec_time.den)/(stream_time.den*codec_time.num*ticks);     \r
 390                 }\r
 391         }\r
 392 }\r
 393 \r
 394 safe_ptr<AVPacket> create_packet()\r
 395 {\r
 396         safe_ptr<AVPacket> packet(new AVPacket, [](AVPacket* p)\r
 397         {\r
 398                 av_free_packet(p);\r
 399                 delete p;\r
 400         });\r
 401         \r
 402         av_init_packet(packet.get());\r
 403         return packet;\r
 404 }\r
 405 \r
 406 safe_ptr<AVCodecContext> open_codec(AVFormatContext& context, enum AVMediaType type, int& index)\r
 407 {       \r
 408         AVCodec* decoder;\r
 409         index = THROW_ON_ERROR2(av_find_best_stream(&context, type, -1, -1, &decoder, 0), "");\r
 410         //if(strcmp(decoder->name, "prores") == 0 && decoder->next && strcmp(decoder->next->name, "prores_lgpl") == 0)\r
 411         //      decoder = decoder->next;\r
 412 \r
 413         THROW_ON_ERROR2(tbb_avcodec_open(context.streams[index]->codec, decoder), "");\r
 414         return safe_ptr<AVCodecContext>(context.streams[index]->codec, tbb_avcodec_close);\r
 415 }\r
 416 \r
 417 safe_ptr<AVFormatContext> open_input(const std::wstring& filename)\r
 418 {\r
 419         AVFormatContext* weak_context = nullptr;\r
 420         THROW_ON_ERROR2(avformat_open_input(&weak_context, narrow(filename).c_str(), nullptr, nullptr), filename);\r
 421         safe_ptr<AVFormatContext> context(weak_context, av_close_input_file);                   \r
 422         THROW_ON_ERROR2(avformat_find_stream_info(weak_context, nullptr), filename);\r
 423         fix_meta_data(*context);\r
 424         return context;\r
 425 }\r
 426 \r
 427 std::wstring print_mode(size_t width, size_t height, double fps, bool interlaced)\r
 428 {\r
 429         std::wostringstream fps_ss;\r
 430         fps_ss << std::fixed << std::setprecision(2) << (!interlaced ? fps : 2.0 * fps);\r
 431 \r
 432         return boost::lexical_cast<std::wstring>(width) + L"x" + boost::lexical_cast<std::wstring>(height) + (!interlaced ? L"p" : L"i") + fps_ss.str();\r
 433 }\r
 434 //\r
 435 //void av_dup_frame(AVFrame* frame)\r
 436 //{\r
 437 //      AVFrame* new_frame = avcodec_alloc_frame();\r
 438 //\r
 439 //\r
 440 //      const uint8_t *src_data[4] = {0};\r
 441 //      memcpy(const_cast<uint8_t**>(&src_data[0]), frame->data, 4);\r
 442 //      const int src_linesizes[4] = {0};\r
 443 //      memcpy(const_cast<int*>(&src_linesizes[0]), frame->linesize, 4);\r
 444 //\r
 445 //      av_image_alloc(new_frame->data, new_frame->linesize, new_frame->width, new_frame->height, frame->format, 16);\r
 446 //\r
 447 //      av_image_copy(new_frame->data, new_frame->linesize, src_data, src_linesizes, frame->format, new_frame->width, new_frame->height);\r
 448 //\r
 449 //      frame =\r
 450 //}\r
 451 \r
 452 }}