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

   1 #include "../../stdafx.h"\r
   2 \r
   3 #include "util.h"\r
   4 \r
   5 #include "format/flv.h"\r
   6 \r
   7 #include <tbb/concurrent_unordered_map.h>\r
   8 #include <tbb/concurrent_queue.h>\r
   9 \r
  10 #include <core/producer/frame/frame_transform.h>\r
  11 #include <core/producer/frame/frame_factory.h>\r
  12 #include <core/producer/frame_producer.h>\r
  13 #include <core/mixer/write_frame.h>\r
  14 \r
  15 #include <common/exception/exceptions.h>\r
  16 \r
  17 #include <tbb/parallel_for.h>\r
  18 \r
  19 #include <boost/filesystem.hpp>\r
  20 #include <boost/lexical_cast.hpp>\r
  21 \r
  22 #if defined(_MSC_VER)\r
  23 #pragma warning (push)\r
  24 #pragma warning (disable : 4244)\r
  25 #endif\r
  26 extern "C" \r
  27 {\r
  28         #include <libswscale/swscale.h>\r
  29         #include <libavcodec/avcodec.h>\r
  30         #include <libavformat/avformat.h>\r
  31 }\r
  32 #if defined(_MSC_VER)\r
  33 #pragma warning (pop)\r
  34 #endif\r
  35 \r
  36 namespace caspar { namespace ffmpeg {\r
  37 \r
  38 core::field_mode::type get_mode(AVFrame& frame)\r
  39 {\r
  40         if(!frame.interlaced_frame)\r
  41                 return core::field_mode::progressive;\r
  42 \r
  43         return frame.top_field_first ? core::field_mode::upper : core::field_mode::lower;\r
  44 }\r
  45 \r
  46 core::pixel_format::type get_pixel_format(PixelFormat pix_fmt)\r
  47 {\r
  48         switch(pix_fmt)\r
  49         {\r
  50         case CASPAR_PIX_FMT_LUMA:       return core::pixel_format::luma;\r
  51         case PIX_FMT_GRAY8:                     return core::pixel_format::gray;\r
  52         case PIX_FMT_BGRA:                      return core::pixel_format::bgra;\r
  53         case PIX_FMT_ARGB:                      return core::pixel_format::argb;\r
  54         case PIX_FMT_RGBA:                      return core::pixel_format::rgba;\r
  55         case PIX_FMT_ABGR:                      return core::pixel_format::abgr;\r
  56         case PIX_FMT_YUV444P:           return core::pixel_format::ycbcr;\r
  57         case PIX_FMT_YUV422P:           return core::pixel_format::ycbcr;\r
  58         case PIX_FMT_YUV420P:           return core::pixel_format::ycbcr;\r
  59         case PIX_FMT_YUV411P:           return core::pixel_format::ycbcr;\r
  60         case PIX_FMT_YUV410P:           return core::pixel_format::ycbcr;\r
  61         case PIX_FMT_YUVA420P:          return core::pixel_format::ycbcra;\r
  62         default:                                        return core::pixel_format::invalid;\r
  63         }\r
  64 }\r
  65 \r
  66 core::pixel_format_desc get_pixel_format_desc(PixelFormat pix_fmt, size_t width, size_t height)\r
  67 {\r
  68         // Get linesizes\r
  69         AVPicture dummy_pict;   \r
  70         avpicture_fill(&dummy_pict, nullptr, pix_fmt == CASPAR_PIX_FMT_LUMA ? PIX_FMT_GRAY8 : pix_fmt, width, height);\r
  71 \r
  72         core::pixel_format_desc desc;\r
  73         desc.pix_fmt = get_pixel_format(pix_fmt);\r
  74                 \r
  75         switch(desc.pix_fmt)\r
  76         {\r
  77         case core::pixel_format::gray:\r
  78         case core::pixel_format::luma:\r
  79                 {\r
  80                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));                                               \r
  81                         return desc;\r
  82                 }\r
  83         case core::pixel_format::bgra:\r
  84         case core::pixel_format::argb:\r
  85         case core::pixel_format::rgba:\r
  86         case core::pixel_format::abgr:\r
  87                 {\r
  88                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0]/4, height, 4));                                             \r
  89                         return desc;\r
  90                 }\r
  91         case core::pixel_format::ycbcr:\r
  92         case core::pixel_format::ycbcra:\r
  93                 {               \r
  94                         // Find chroma height\r
  95                         size_t size2 = dummy_pict.data[2] - dummy_pict.data[1];\r
  96                         size_t h2 = size2/dummy_pict.linesize[1];                       \r
  97 \r
  98                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));\r
  99                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[1], h2, 1));\r
 100                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[2], h2, 1));\r
 101 \r
 102                         if(desc.pix_fmt == core::pixel_format::ycbcra)                                          \r
 103                                 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[3], height, 1));       \r
 104                         return desc;\r
 105                 }               \r
 106         default:                \r
 107                 desc.pix_fmt = core::pixel_format::invalid;\r
 108                 return desc;\r
 109         }\r
 110 }\r
 111 \r
 112 int make_alpha_format(int format)\r
 113 {\r
 114         switch(get_pixel_format(static_cast<PixelFormat>(format)))\r
 115         {\r
 116         case core::pixel_format::luma:\r
 117         case core::pixel_format::gray:\r
 118         case core::pixel_format::invalid:\r
 119                 return format;\r
 120         case core::pixel_format::ycbcr:\r
 121         case core::pixel_format::ycbcra:\r
 122                 return CASPAR_PIX_FMT_LUMA;\r
 123         default:\r
 124         return PIX_FMT_GRAY8;\r
 125         }\r
 126 }\r
 127 \r
 128 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
 129 {                       \r
 130         static tbb::concurrent_unordered_map<size_t, tbb::concurrent_queue<std::shared_ptr<SwsContext>>> sws_contexts_;\r
 131         \r
 132         const auto width  = decoded_frame->width;\r
 133         const auto height = decoded_frame->height;\r
 134         auto desc                 = get_pixel_format_desc(static_cast<PixelFormat>(decoded_frame->format), width, height);\r
 135         \r
 136         if(hints & core::frame_producer::ALPHA_HINT)\r
 137                 desc = get_pixel_format_desc(static_cast<PixelFormat>(make_alpha_format(decoded_frame->format)), width, height);\r
 138 \r
 139         if(desc.pix_fmt == core::pixel_format::invalid)\r
 140         {\r
 141                 auto pix_fmt = static_cast<PixelFormat>(decoded_frame->format);\r
 142 \r
 143                 auto write = frame_factory->create_frame(tag, get_pixel_format_desc(PIX_FMT_BGRA, width, height));\r
 144                 write->set_type(get_mode(*decoded_frame));\r
 145 \r
 146                 std::shared_ptr<SwsContext> sws_context;\r
 147 \r
 148                 //CASPAR_LOG(warning) << "Hardware accelerated color transform not supported.";\r
 149 \r
 150                 size_t key = width << 20 | height << 8 | pix_fmt;\r
 151                         \r
 152                 auto& pool = sws_contexts_[key];\r
 153                                                 \r
 154                 if(!pool.try_pop(sws_context))\r
 155                 {\r
 156                         double param;\r
 157                         sws_context.reset(sws_getContext(width, height, pix_fmt, width, height, PIX_FMT_BGRA, SWS_BILINEAR, nullptr, nullptr, &param), sws_freeContext);\r
 158                 }\r
 159                         \r
 160                 if(!sws_context)\r
 161                 {\r
 162                         BOOST_THROW_EXCEPTION(operation_failed() << msg_info("Could not create software scaling context.") << \r
 163                                                                         boost::errinfo_api_function("sws_getContext"));\r
 164                 }       \r
 165 \r
 166                 // Use sws_scale when provided colorspace has no hw-accel.\r
 167                 safe_ptr<AVFrame> av_frame(avcodec_alloc_frame(), av_free);     \r
 168                 avcodec_get_frame_defaults(av_frame.get());                     \r
 169                 avpicture_fill(reinterpret_cast<AVPicture*>(av_frame.get()), write->image_data().begin(), PIX_FMT_BGRA, width, height);\r
 170                  \r
 171                 sws_scale(sws_context.get(), decoded_frame->data, decoded_frame->linesize, 0, height, av_frame->data, av_frame->linesize);      \r
 172                 pool.push(sws_context);\r
 173 \r
 174                 write->commit();\r
 175 \r
 176                 return write;\r
 177         }\r
 178         else\r
 179         {\r
 180                 auto write = frame_factory->create_frame(tag, desc);\r
 181                 write->set_type(get_mode(*decoded_frame));\r
 182 \r
 183                 for(int n = 0; n < static_cast<int>(desc.planes.size()); ++n)\r
 184                 {\r
 185                         auto plane            = desc.planes[n];\r
 186                         auto result           = write->image_data(n).begin();\r
 187                         auto decoded          = decoded_frame->data[n];\r
 188                         auto decoded_linesize = decoded_frame->linesize[n];\r
 189                                 \r
 190                         // Copy line by line since ffmpeg sometimes pads each line.\r
 191                         tbb::affinity_partitioner ap;\r
 192                         tbb::parallel_for(tbb::blocked_range<size_t>(0, static_cast<int>(desc.planes[n].height)), [&](const tbb::blocked_range<size_t>& r)\r
 193                         {\r
 194                                 for(size_t y = r.begin(); y != r.end(); ++y)\r
 195                                         memcpy(result + y*plane.linesize, decoded + y*decoded_linesize, plane.linesize);\r
 196                         }, ap);\r
 197 \r
 198                         write->commit(n);\r
 199                 }\r
 200                 \r
 201                 return write;\r
 202         }\r
 203 }\r
 204 \r
 205 bool is_sane_fps(AVRational time_base)\r
 206 {\r
 207         double fps = static_cast<double>(time_base.den) / static_cast<double>(time_base.num);\r
 208         return fps > 20.0 && fps < 65.0;\r
 209 }\r
 210 \r
 211 void fix_meta_data(AVFormatContext& context)\r
 212 {\r
 213         auto video_index = av_find_best_stream(&context, AVMEDIA_TYPE_VIDEO, -1, -1, 0, 0);\r
 214         auto audio_index = av_find_best_stream(&context, AVMEDIA_TYPE_AUDIO, -1, -1, 0, 0);\r
 215 \r
 216         if(video_index < 0)\r
 217                 return;\r
 218 \r
 219         auto& video_context = *context.streams[video_index]->codec;\r
 220         auto& video_stream  = *context.streams[video_index];\r
 221                                                 \r
 222         if(boost::filesystem2::path(context.filename).extension() == ".flv")\r
 223         {\r
 224                 try\r
 225                 {\r
 226                         auto meta = read_flv_meta_info(context.filename);\r
 227                         double fps = boost::lexical_cast<double>(meta["framerate"]);\r
 228                         video_context.time_base.num = 1000000;\r
 229                         video_context.time_base.den = static_cast<int>(fps*1000000.0);\r
 230                         video_stream.nb_frames = static_cast<int64_t>(boost::lexical_cast<double>(meta["duration"])*fps);\r
 231                 }\r
 232                 catch(...){}\r
 233         }\r
 234         else\r
 235         {\r
 236                 if(video_stream.nb_frames == 0)\r
 237                         video_stream.nb_frames = video_stream.duration;\r
 238 \r
 239                 if(!is_sane_fps(video_context.time_base))\r
 240                 {                       \r
 241                         if(video_context.time_base.num == 1)\r
 242                                 video_context.time_base.num = static_cast<int>(std::pow(10.0, static_cast<int>(std::log10(static_cast<float>(video_context.time_base.den)))-1));        \r
 243 \r
 244                         if(!is_sane_fps(video_context.time_base) && audio_index > -1)\r
 245                         {\r
 246                                 auto& audio_context = *context.streams[audio_index]->codec;\r
 247                                 auto& audio_stream  = *context.streams[audio_index];\r
 248 \r
 249                                 double duration_sec = audio_stream.duration / static_cast<double>(audio_context.sample_rate);\r
 250                                                                 \r
 251                                 video_context.time_base.num = static_cast<int>(duration_sec*100000.0);\r
 252                                 video_context.time_base.den = static_cast<int>(video_stream.nb_frames*100000);\r
 253                         }\r
 254                 }\r
 255 \r
 256                 if(audio_index > -1) // Check for invalid double frame-rate\r
 257                 {\r
 258                         auto& audio_context = *context.streams[audio_index]->codec;\r
 259                         auto& audio_stream  = *context.streams[audio_index];\r
 260 \r
 261                         double duration_sec = audio_stream.duration / static_cast<double>(audio_context.sample_rate);\r
 262                         double fps = static_cast<double>(video_context.time_base.den) / static_cast<double>(video_context.time_base.num);\r
 263 \r
 264                         double fps_nb_frames    = static_cast<double>(duration_sec*fps);\r
 265                         double stream_nb_frames =  static_cast<double>(video_stream.nb_frames);\r
 266                         double diff = std::abs(fps_nb_frames - stream_nb_frames*2.0);\r
 267                         if(diff < fps_nb_frames*0.05)\r
 268                                 video_context.time_base.num *= 2;\r
 269                 }\r
 270         }\r
 271 \r
 272         double fps = static_cast<double>(video_context.time_base.den) / static_cast<double>(video_context.time_base.num);\r
 273 \r
 274         double closest_fps = 0.0;\r
 275         for(int n = 0; n < core::video_format::count; ++n)\r
 276         {\r
 277                 auto format = core::video_format_desc::get(static_cast<core::video_format::type>(n));\r
 278 \r
 279                 double diff1 = std::abs(format.fps - fps);\r
 280                 double diff2 = std::abs(closest_fps - fps);\r
 281 \r
 282                 if(diff1 < diff2)\r
 283                         closest_fps = format.fps;\r
 284         }\r
 285         \r
 286         video_context.time_base.num = 1000000;\r
 287         video_context.time_base.den = static_cast<int>(closest_fps*1000000.0);\r
 288 }\r
 289 \r
 290 }}