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 <tbb/concurrent_unordered_map.h>\r
   6 #include <tbb/concurrent_queue.h>\r
   7 \r
   8 #include <core/producer/frame/image_transform.h>\r
   9 #include <core/producer/frame/frame_factory.h>\r
  10 #include <core/producer/frame_producer.h>\r
  11 #include <core/mixer/write_frame.h>\r
  12 \r
  13 #include <common/exception/exceptions.h>\r
  14 \r
  15 #include <tbb/parallel_for.h>\r
  16 \r
  17 #if defined(_MSC_VER)\r
  18 #pragma warning (push)\r
  19 #pragma warning (disable : 4244)\r
  20 #endif\r
  21 extern "C" \r
  22 {\r
  23         #include <libswscale/swscale.h>\r
  24         #include <libavcodec/avcodec.h>\r
  25 }\r
  26 #if defined(_MSC_VER)\r
  27 #pragma warning (pop)\r
  28 #endif\r
  29 \r
  30 namespace caspar {\r
  31 \r
  32 core::video_mode::type get_mode(AVFrame& frame)\r
  33 {\r
  34         if(!frame.interlaced_frame)\r
  35                 return core::video_mode::progressive;\r
  36 \r
  37         return frame.top_field_first ? core::video_mode::upper : core::video_mode::lower;\r
  38 }\r
  39 \r
  40 core::pixel_format::type get_pixel_format(PixelFormat pix_fmt)\r
  41 {\r
  42         switch(pix_fmt)\r
  43         {\r
  44         case CASPAR_PIX_FMT_LUMA:       return core::pixel_format::luma;\r
  45         case PIX_FMT_GRAY8:                     return core::pixel_format::gray;\r
  46         case PIX_FMT_BGRA:                      return core::pixel_format::bgra;\r
  47         case PIX_FMT_ARGB:                      return core::pixel_format::argb;\r
  48         case PIX_FMT_RGBA:                      return core::pixel_format::rgba;\r
  49         case PIX_FMT_ABGR:                      return core::pixel_format::abgr;\r
  50         case PIX_FMT_YUV444P:           return core::pixel_format::ycbcr;\r
  51         case PIX_FMT_YUV422P:           return core::pixel_format::ycbcr;\r
  52         case PIX_FMT_YUV420P:           return core::pixel_format::ycbcr;\r
  53         case PIX_FMT_YUV411P:           return core::pixel_format::ycbcr;\r
  54         case PIX_FMT_YUV410P:           return core::pixel_format::ycbcr;\r
  55         case PIX_FMT_YUVA420P:          return core::pixel_format::ycbcra;\r
  56         default:                                        return core::pixel_format::invalid;\r
  57         }\r
  58 }\r
  59 \r
  60 core::pixel_format_desc get_pixel_format_desc(PixelFormat pix_fmt, size_t width, size_t height)\r
  61 {\r
  62         // Get linesizes\r
  63         AVPicture dummy_pict;   \r
  64         avpicture_fill(&dummy_pict, nullptr, pix_fmt == CASPAR_PIX_FMT_LUMA ? PIX_FMT_GRAY8 : pix_fmt, width, height);\r
  65 \r
  66         core::pixel_format_desc desc;\r
  67         desc.pix_fmt = get_pixel_format(pix_fmt);\r
  68                 \r
  69         switch(desc.pix_fmt)\r
  70         {\r
  71         case core::pixel_format::gray:\r
  72         case core::pixel_format::luma:\r
  73                 {\r
  74                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));                                               \r
  75                         return desc;\r
  76                 }\r
  77         case core::pixel_format::bgra:\r
  78         case core::pixel_format::argb:\r
  79         case core::pixel_format::rgba:\r
  80         case core::pixel_format::abgr:\r
  81                 {\r
  82                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0]/4, height, 4));                                             \r
  83                         return desc;\r
  84                 }\r
  85         case core::pixel_format::ycbcr:\r
  86         case core::pixel_format::ycbcra:\r
  87                 {               \r
  88                         // Find chroma height\r
  89                         size_t size2 = dummy_pict.data[2] - dummy_pict.data[1];\r
  90                         size_t h2 = size2/dummy_pict.linesize[1];                       \r
  91 \r
  92                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));\r
  93                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[1], h2, 1));\r
  94                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[2], h2, 1));\r
  95 \r
  96                         if(desc.pix_fmt == core::pixel_format::ycbcra)                                          \r
  97                                 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[3], height, 1));       \r
  98                         return desc;\r
  99                 }               \r
 100         default:                \r
 101                 desc.pix_fmt = core::pixel_format::invalid;\r
 102                 return desc;\r
 103         }\r
 104 }\r
 105 \r
 106 int make_alpha_format(int format)\r
 107 {\r
 108         switch(get_pixel_format(static_cast<PixelFormat>(format)))\r
 109         {\r
 110         case core::pixel_format::luma:\r
 111         case core::pixel_format::gray:\r
 112         case core::pixel_format::invalid:\r
 113                 return format;\r
 114         case core::pixel_format::ycbcr:\r
 115         case core::pixel_format::ycbcra:\r
 116                 return CASPAR_PIX_FMT_LUMA;\r
 117         default:\r
 118         return PIX_FMT_GRAY8;\r
 119         }\r
 120 }\r
 121 \r
 122 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
 123 {                       \r
 124         static tbb::concurrent_unordered_map<size_t, tbb::concurrent_queue<std::shared_ptr<SwsContext>>> sws_contexts_;\r
 125         \r
 126         const auto width  = decoded_frame->width;\r
 127         const auto height = decoded_frame->height;\r
 128         auto desc                 = get_pixel_format_desc(static_cast<PixelFormat>(decoded_frame->format), width, height);\r
 129         \r
 130         if(hints & core::frame_producer::ALPHA_HINT)\r
 131                 desc = get_pixel_format_desc(static_cast<PixelFormat>(make_alpha_format(decoded_frame->format)), width, height);\r
 132 \r
 133         if(desc.pix_fmt == core::pixel_format::invalid)\r
 134         {\r
 135                 auto pix_fmt = static_cast<PixelFormat>(decoded_frame->format);\r
 136 \r
 137                 auto write = frame_factory->create_frame(tag, desc.pix_fmt != core::pixel_format::invalid ? desc : get_pixel_format_desc(PIX_FMT_BGRA, width, height));\r
 138                 write->set_type(get_mode(*decoded_frame));\r
 139 \r
 140                 std::shared_ptr<SwsContext> sws_context;\r
 141 \r
 142                 //CASPAR_LOG(warning) << "Hardware accelerated color transform not supported.";\r
 143 \r
 144                 size_t key = width << 20 | height << 8 | pix_fmt;\r
 145                         \r
 146                 auto& pool = sws_contexts_[key];\r
 147                                                 \r
 148                 if(!pool.try_pop(sws_context))\r
 149                 {\r
 150                         double param;\r
 151                         sws_context.reset(sws_getContext(width, height, pix_fmt, width, height, PIX_FMT_BGRA, SWS_BILINEAR, nullptr, nullptr, &param), sws_freeContext);\r
 152                 }\r
 153                         \r
 154                 if(!sws_context)\r
 155                 {\r
 156                         BOOST_THROW_EXCEPTION(operation_failed() << msg_info("Could not create software scaling context.") << \r
 157                                                                         boost::errinfo_api_function("sws_getContext"));\r
 158                 }       \r
 159 \r
 160                 // Use sws_scale when provided colorspace has no hw-accel.\r
 161                 safe_ptr<AVFrame> av_frame(avcodec_alloc_frame(), av_free);     \r
 162                 avcodec_get_frame_defaults(av_frame.get());                     \r
 163                 avpicture_fill(reinterpret_cast<AVPicture*>(av_frame.get()), write->image_data().begin(), PIX_FMT_BGRA, width, height);\r
 164                  \r
 165                 sws_scale(sws_context.get(), decoded_frame->data, decoded_frame->linesize, 0, height, av_frame->data, av_frame->linesize);      \r
 166                 pool.push(sws_context);\r
 167 \r
 168                 write->commit();\r
 169 \r
 170                 return write;\r
 171         }\r
 172         else\r
 173         {\r
 174                 auto write = frame_factory->create_frame(tag, desc);\r
 175                 write->set_type(get_mode(*decoded_frame));\r
 176 \r
 177                 for(int n = 0; n < static_cast<int>(desc.planes.size()); ++n)\r
 178                 {\r
 179                         auto plane            = desc.planes[n];\r
 180                         auto result           = write->image_data(n).begin();\r
 181                         auto decoded          = decoded_frame->data[n];\r
 182                         auto decoded_linesize = decoded_frame->linesize[n];\r
 183                                 \r
 184                         // Copy line by line since ffmpeg sometimes pads each line.\r
 185                         tbb::affinity_partitioner ap;\r
 186                         tbb::parallel_for(tbb::blocked_range<size_t>(0, static_cast<int>(desc.planes[n].height)), [&](const tbb::blocked_range<size_t>& r)\r
 187                         {\r
 188                                 for(size_t y = r.begin(); y != r.end(); ++y)\r
 189                                         memcpy(result + y*plane.linesize, decoded + y*decoded_linesize, plane.linesize);\r
 190                         }, ap);\r
 191 \r
 192                         write->commit(n);\r
 193                 }\r
 194 \r
 195                 //for(int n = 0; n < static_cast<int>(desc.planes.size()); ++n)\r
 196                 //{\r
 197                 //      auto plane            = desc.planes[n];\r
 198                 //      auto result           = write->image_data(n).begin();\r
 199                 //      auto decoded          = decoded_frame->data[n];\r
 200                 //      auto decoded_linesize = decoded_frame->linesize[n];\r
 201                 //              \r
 202                 //      for(size_t y = 0; y < static_cast<int>(desc.planes[n].height); ++y)\r
 203                 //              fast_memcpy(result + y*plane.linesize, decoded + y*decoded_linesize, plane.linesize);\r
 204 \r
 205                 //      write->commit(n);\r
 206                 //}\r
 207 \r
 208                 return write;\r
 209         }\r
 210 }\r
 211 \r
 212 }