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

   1 /*\r
   2 * copyright (c) 2010 Sveriges Television AB <info@casparcg.com>\r
   3 *\r
   4 *  This file is part of CasparCG.\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 */\r
  20 #include "../../stdafx.h"\r
  21 \r
  22 #include "video_decoder.h"\r
  23 #include "pix_fmt.h"\r
  24 \r
  25 #include "../../ffmpeg_error.h"\r
  26 #include "../filter/filter.h"\r
  27 \r
  28 #include <common/memory/memcpy.h>\r
  29 \r
  30 #include <core/video_format.h>\r
  31 #include <core/producer/frame/basic_frame.h>\r
  32 #include <core/mixer/write_frame.h>\r
  33 #include <core/producer/frame/image_transform.h>\r
  34 #include <core/producer/frame/pixel_format.h>\r
  35 #include <core/producer/frame/frame_factory.h>\r
  36 \r
  37 #include <tbb/parallel_for.h>\r
  38 \r
  39 #include <boost/range/algorithm_ext.hpp>\r
  40 \r
  41 #if defined(_MSC_VER)\r
  42 #pragma warning (push)\r
  43 #pragma warning (disable : 4244)\r
  44 #endif\r
  45 extern "C" \r
  46 {\r
  47         #define __STDC_CONSTANT_MACROS\r
  48         #define __STDC_LIMIT_MACROS\r
  49         #include <libswscale/swscale.h>\r
  50         #include <libavformat/avformat.h>\r
  51         #include <libavcodec/avcodec.h>\r
  52 }\r
  53 #if defined(_MSC_VER)\r
  54 #pragma warning (pop)\r
  55 #endif\r
  56 \r
  57 namespace caspar {\r
  58         \r
  59 struct video_decoder::implementation : boost::noncopyable\r
  60 {\r
  61         input& input_;\r
  62         std::shared_ptr<SwsContext>                                     sws_context_;\r
  63         const std::shared_ptr<core::frame_factory>      frame_factory_;\r
  64         AVCodecContext&                                                         codec_context_;\r
  65         size_t                                                                          frame_number_;\r
  66 \r
  67         std::shared_ptr<filter>                                         filter_;\r
  68         size_t                                                                          filter_delay_;\r
  69         int                                                                                     eof_count_;\r
  70 \r
  71         std::string                                                                     filter_str_;\r
  72 \r
  73 public:\r
  74         explicit implementation(input& input, const safe_ptr<core::frame_factory>& frame_factory, const std::string& filter_str) \r
  75                 : input_(input)\r
  76                 , frame_factory_(frame_factory)\r
  77                 , codec_context_(*input_.get_video_codec_context())\r
  78                 , frame_number_(0)\r
  79                 , filter_(filter_str.empty() ? nullptr : new filter(filter_str))\r
  80                 , filter_delay_(1)\r
  81                 , filter_str_(filter_str)\r
  82                 , eof_count_(std::numeric_limits<int>::max())\r
  83         {\r
  84         }\r
  85 \r
  86         std::deque<std::pair<int, safe_ptr<core::write_frame>>> receive()\r
  87         {\r
  88                 std::deque<std::pair<int, safe_ptr<core::write_frame>>> result;\r
  89                 \r
  90                 std::shared_ptr<AVPacket> pkt;\r
  91                 for(int n = 0; n < 32 && result.empty() && input_.try_pop_video_packet(pkt); ++n)       \r
  92                         boost::range::push_back(result, decode(pkt));\r
  93 \r
  94                 return result;\r
  95         }\r
  96 \r
  97         std::deque<std::pair<int, safe_ptr<core::write_frame>>> decode(const std::shared_ptr<AVPacket>& video_packet)\r
  98         {                       \r
  99                 if(!video_packet) // eof\r
 100                         return flush();\r
 101                 \r
 102                 std::deque<std::pair<int, safe_ptr<core::write_frame>>> result;\r
 103                 \r
 104                 frame_number_ = frame_number_ % eof_count_;\r
 105                 \r
 106                 if(filter_)\r
 107                 {                                                       \r
 108                         std::shared_ptr<AVFrame> frame;\r
 109 \r
 110                         tbb::parallel_invoke(\r
 111                         [&]\r
 112                         {\r
 113                                 frame = decode_frame(video_packet);\r
 114                         },\r
 115                         [&]\r
 116                         {\r
 117                                 if(filter_->is_ready())\r
 118                                 {                               \r
 119                                         boost::range::transform(filter_->poll(), std::back_inserter(result), [&](const safe_ptr<AVFrame>& frame)\r
 120                                         {\r
 121                                                 return std::make_pair(frame_number_, make_write_frame(frame));\r
 122                                         });\r
 123                 \r
 124                                         if(!result.empty())\r
 125                                                 ++frame_number_;\r
 126                                         else            \r
 127                                                 ++filter_delay_;\r
 128                                 }\r
 129                         });             \r
 130 \r
 131                         if(frame)\r
 132                                 filter_->push(make_safe(frame));\r
 133                 }\r
 134                 else\r
 135                 {\r
 136                         auto frame = decode_frame(video_packet);\r
 137                         \r
 138                         if(frame)\r
 139                                 result.push_back(std::make_pair(frame_number_++, make_write_frame(make_safe(frame))));\r
 140                 }\r
 141 \r
 142                 return result;\r
 143         }\r
 144 \r
 145         std::deque<std::pair<int, safe_ptr<core::write_frame>>> flush()\r
 146         {\r
 147                 std::deque<std::pair<int, safe_ptr<core::write_frame>>> result;\r
 148 \r
 149                 eof_count_ = frame_number_;\r
 150 \r
 151                 if(filter_)\r
 152                         eof_count_ += filter_delay_;            \r
 153 \r
 154                 avcodec_flush_buffers(&codec_context_);\r
 155 \r
 156                 return result;\r
 157         }\r
 158                 \r
 159         std::shared_ptr<AVFrame> decode_frame(const std::shared_ptr<AVPacket>& video_packet)\r
 160         {\r
 161                 std::shared_ptr<AVFrame> decoded_frame(avcodec_alloc_frame(), av_free);\r
 162 \r
 163                 int frame_finished = 0;\r
 164                 const int errn = avcodec_decode_video2(&codec_context_, decoded_frame.get(), &frame_finished, video_packet.get());\r
 165                 \r
 166                 if(errn < 0)\r
 167                 {\r
 168                         BOOST_THROW_EXCEPTION(\r
 169                                 invalid_operation() <<\r
 170                                 msg_info(av_error_str(errn)) <<\r
 171                                 boost::errinfo_api_function("avcodec_decode_video") <<\r
 172                                 boost::errinfo_errno(AVUNERROR(errn)));\r
 173                 }\r
 174 \r
 175                 if(frame_finished == 0)         \r
 176                         decoded_frame = nullptr;\r
 177 \r
 178                 return decoded_frame;\r
 179         }\r
 180 \r
 181         safe_ptr<core::write_frame> make_write_frame(const safe_ptr<AVFrame>& decoded_frame)\r
 182         {                       \r
 183                 // We don't know what the filter output might give until we received the first frame. Initialize everything on first frame.\r
 184                 auto width   = decoded_frame->width;\r
 185                 auto height  = decoded_frame->height;\r
 186                 auto pix_fmt = static_cast<PixelFormat>(decoded_frame->format);\r
 187                 auto desc        = get_pixel_format_desc(pix_fmt, width, height);\r
 188                         \r
 189                 if(desc.pix_fmt == core::pixel_format::invalid)\r
 190                 {\r
 191                         CASPAR_VERIFY(!sws_context_); // Initialize only once. Nothing should change while running;\r
 192                         CASPAR_LOG(warning) << "Hardware accelerated color transform not supported.";\r
 193 \r
 194                         desc = get_pixel_format_desc(PIX_FMT_BGRA, width, height);\r
 195                         double param;\r
 196                         sws_context_.reset(sws_getContext(width, height, pix_fmt, width, height, PIX_FMT_BGRA, SWS_BILINEAR, nullptr, nullptr, &param), sws_freeContext);\r
 197                         if(!sws_context_)\r
 198                                 BOOST_THROW_EXCEPTION(operation_failed() <<\r
 199                                                                           msg_info("Could not create software scaling context.") << \r
 200                                                                           boost::errinfo_api_function("sws_getContext"));\r
 201                 }\r
 202 \r
 203                 auto write = frame_factory_->create_frame(this, desc);\r
 204                 write->set_is_interlaced(decoded_frame->interlaced_frame != 0);\r
 205 \r
 206                 if(sws_context_ == nullptr)\r
 207                 {\r
 208                         tbb::parallel_for(0, static_cast<int>(desc.planes.size()), 1, [&](int n)\r
 209                         {\r
 210                                 auto plane            = desc.planes[n];\r
 211                                 auto result           = write->image_data(n).begin();\r
 212                                 auto decoded          = decoded_frame->data[n];\r
 213                                 auto decoded_linesize = decoded_frame->linesize[n];\r
 214                                 \r
 215                                 // Copy line by line since ffmpeg sometimes pads each line.\r
 216                                 tbb::parallel_for(tbb::blocked_range<size_t>(0, static_cast<int>(desc.planes[n].height)), [&](const tbb::blocked_range<size_t>& r)\r
 217                                 {\r
 218                                         for(size_t y = r.begin(); y != r.end(); ++y)\r
 219                                                 memcpy(result + y*plane.linesize, decoded + y*decoded_linesize, plane.linesize);\r
 220                                 });\r
 221 \r
 222                                 write->commit(n);\r
 223                         });\r
 224                 }\r
 225                 else\r
 226                 {\r
 227                         // Use sws_scale when provided colorspace has no hw-accel.\r
 228                         safe_ptr<AVFrame> av_frame(avcodec_alloc_frame(), av_free);     \r
 229                         avcodec_get_frame_defaults(av_frame.get());                     \r
 230                         avpicture_fill(reinterpret_cast<AVPicture*>(av_frame.get()), write->image_data().begin(), PIX_FMT_BGRA, width, height);\r
 231                  \r
 232                         sws_scale(sws_context_.get(), decoded_frame->data, decoded_frame->linesize, 0, height, av_frame->data, av_frame->linesize);     \r
 233 \r
 234                         write->commit();\r
 235                 }       \r
 236 \r
 237                 // Fix field-order if needed. DVVIDEO is in lower field. Make it upper field if needed.\r
 238                 if(decoded_frame->interlaced_frame)\r
 239                 {\r
 240                         switch(frame_factory_->get_video_format_desc().mode)\r
 241                         {\r
 242                         case core::video_mode::upper:\r
 243                                 if(!decoded_frame->top_field_first)\r
 244                                         write->get_image_transform().set_fill_translation(0.0f, 0.5/static_cast<double>(height));\r
 245                                 break;\r
 246                         case core::video_mode::lower:\r
 247                                 if(decoded_frame->top_field_first)\r
 248                                         write->get_image_transform().set_fill_translation(0.0f, -0.5/static_cast<double>(height));\r
 249                                 break;\r
 250                         }\r
 251                 }\r
 252 \r
 253                 return write;\r
 254         }\r
 255 };\r
 256 \r
 257 video_decoder::video_decoder(input& input, const safe_ptr<core::frame_factory>& frame_factory, const std::string& filter_str) : impl_(new implementation(input, frame_factory, filter_str)){}\r
 258 std::deque<std::pair<int, safe_ptr<core::write_frame>>> video_decoder::receive(){return impl_->receive();}\r
 259 \r
 260 }