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 \r
  70         std::shared_ptr<AVFrame>                                        last_frame_;\r
  71         std::shared_ptr<AVFrame>                                        decoded_frame_;\r
  72         std::string                                                                     filter_str_;\r
  73 \r
  74 public:\r
  75         explicit implementation(input& input, const safe_ptr<core::frame_factory>& frame_factory, const std::string& filter_str) \r
  76                 : input_(input)\r
  77                 , frame_factory_(frame_factory)\r
  78                 , codec_context_(*input_.get_video_codec_context())\r
  79                 , frame_number_(0)\r
  80                 , filter_(filter_str.empty() ? nullptr : new filter(filter_str))\r
  81                 , filter_delay_(0)\r
  82                 , last_frame_(avcodec_alloc_frame(), av_free)\r
  83                 , filter_str_(filter_str)\r
  84         {\r
  85         }\r
  86 \r
  87         std::deque<std::pair<int, safe_ptr<core::write_frame>>> receive()\r
  88         {\r
  89                 std::deque<std::pair<int, safe_ptr<core::write_frame>>> result;\r
  90                 \r
  91                 std::shared_ptr<AVPacket> pkt;\r
  92                 for(int n = 0; n < 32 && result.empty() && input_.try_pop_video_packet(pkt); ++n)       \r
  93                         boost::range::push_back(result, decode(pkt));\r
  94 \r
  95                 return result;\r
  96         }\r
  97 \r
  98         std::deque<std::pair<int, safe_ptr<core::write_frame>>> decode(const std::shared_ptr<AVPacket>& video_packet)\r
  99         {                       \r
 100                 std::deque<std::pair<int, safe_ptr<core::write_frame>>> result;\r
 101                 \r
 102                 if(!video_packet) // eof\r
 103                         return flush();\r
 104 \r
 105                 if(filter_)\r
 106                 {\r
 107                         if(decoded_frame_)\r
 108                                 push_filter_frames(make_safe(decoded_frame_));  \r
 109                         \r
 110                         std::shared_ptr<AVFrame> frame;\r
 111 \r
 112                         tbb::parallel_invoke(\r
 113                         [&]\r
 114                         {\r
 115                                 frame = decode_frame(video_packet);\r
 116                         },\r
 117                         [&]\r
 118                         {\r
 119                                 if(decoded_frame_)\r
 120                                         result = poll_filter_frames();\r
 121                         });             \r
 122 \r
 123                         decoded_frame_ = frame;\r
 124                         if(frame)\r
 125                                 last_frame_ = frame;                    \r
 126                 }\r
 127                 else\r
 128                 {\r
 129                         auto frame = decode_frame(video_packet);\r
 130                         \r
 131                         if(frame)\r
 132                                 result.push_back(std::make_pair(frame_number_++, make_write_frame(make_safe(frame))));\r
 133                 }\r
 134 \r
 135                 return result;\r
 136         }\r
 137 \r
 138         std::deque<std::pair<int, safe_ptr<core::write_frame>>> flush()\r
 139         {\r
 140                 std::deque<std::pair<int, safe_ptr<core::write_frame>>> result;\r
 141 \r
 142                 if(filter_)\r
 143                 {\r
 144                         // Get all buffered frames\r
 145                         if(decoded_frame_)\r
 146                                 push_filter_frames(make_safe(decoded_frame_));  \r
 147                         boost::range::push_back(result, poll_filter_frames());\r
 148 \r
 149                         for(size_t n = 0; n < filter_delay_; ++n)\r
 150                         {                                       \r
 151                                 push_filter_frames(make_safe(last_frame_));     \r
 152                                 boost::range::push_back(result, poll_filter_frames());\r
 153                         }\r
 154 \r
 155                         // FIXME: Unnecessary reinitialization\r
 156                         filter_.reset(filter_str_.empty() ? nullptr : new filter(filter_str_));\r
 157                         filter_delay_ = 0;\r
 158                         decoded_frame_ = nullptr;\r
 159                 }\r
 160 \r
 161                 frame_number_ = 0;\r
 162                 avcodec_flush_buffers(&codec_context_);\r
 163 \r
 164                 return result;\r
 165         }\r
 166 \r
 167         void push_filter_frames(const safe_ptr<AVFrame>& frame)\r
 168         {               \r
 169                 filter_->push(frame);           \r
 170         }\r
 171 \r
 172         std::deque<std::pair<int, safe_ptr<core::write_frame>>> poll_filter_frames()\r
 173         {\r
 174                 std::deque<std::pair<int, safe_ptr<core::write_frame>>> result;\r
 175 \r
 176                 auto frames = filter_->poll(); \r
 177 \r
 178                 boost::range::transform(frames, std::back_inserter(result), [&](const safe_ptr<AVFrame>& frame)\r
 179                 {\r
 180                         return std::make_pair(frame_number_, make_write_frame(frame));\r
 181                 });\r
 182 \r
 183                 if(!frames.empty())\r
 184                         ++frame_number_;\r
 185                 else            \r
 186                         ++filter_delay_;\r
 187                 \r
 188                 return result;\r
 189         }\r
 190         \r
 191         std::shared_ptr<AVFrame> decode_frame(const std::shared_ptr<AVPacket>& video_packet)\r
 192         {\r
 193                 std::shared_ptr<AVFrame> decoded_frame(avcodec_alloc_frame(), av_free);\r
 194 \r
 195                 int frame_finished = 0;\r
 196                 const int errn = avcodec_decode_video2(&codec_context_, decoded_frame.get(), &frame_finished, video_packet.get());\r
 197                 \r
 198                 if(errn < 0)\r
 199                 {\r
 200                         BOOST_THROW_EXCEPTION(\r
 201                                 invalid_operation() <<\r
 202                                 msg_info(av_error_str(errn)) <<\r
 203                                 boost::errinfo_api_function("avcodec_decode_video") <<\r
 204                                 boost::errinfo_errno(AVUNERROR(errn)));\r
 205                 }\r
 206 \r
 207                 if(frame_finished == 0)         \r
 208                         decoded_frame = nullptr;\r
 209 \r
 210                 return decoded_frame;\r
 211         }\r
 212 \r
 213         safe_ptr<core::write_frame> make_write_frame(safe_ptr<AVFrame> decoded_frame)\r
 214         {                       \r
 215                 // We don't know what the filter output might give until we received the first frame. Initialize everything on first frame.\r
 216                 auto width   = decoded_frame->width;\r
 217                 auto height  = decoded_frame->height;\r
 218                 auto pix_fmt = static_cast<PixelFormat>(decoded_frame->format);\r
 219                 auto desc        = get_pixel_format_desc(pix_fmt, width, height);\r
 220                         \r
 221                 if(desc.pix_fmt == core::pixel_format::invalid)\r
 222                 {\r
 223                         CASPAR_VERIFY(!sws_context_); // Initialize only once. Nothing should change while running;\r
 224                         CASPAR_LOG(warning) << "Hardware accelerated color transform not supported.";\r
 225 \r
 226                         desc = get_pixel_format_desc(PIX_FMT_BGRA, width, height);\r
 227                         double param;\r
 228                         sws_context_.reset(sws_getContext(width, height, pix_fmt, width, height, PIX_FMT_BGRA, SWS_BILINEAR, nullptr, nullptr, &param), sws_freeContext);\r
 229                         if(!sws_context_)\r
 230                                 BOOST_THROW_EXCEPTION(operation_failed() <<\r
 231                                                                           msg_info("Could not create software scaling context.") << \r
 232                                                                           boost::errinfo_api_function("sws_getContext"));\r
 233                 }\r
 234 \r
 235                 auto write = frame_factory_->create_frame(this, desc);\r
 236                 write->set_is_interlaced(decoded_frame->interlaced_frame != 0);\r
 237 \r
 238                 if(sws_context_ == nullptr)\r
 239                 {\r
 240                         tbb::parallel_for(0, static_cast<int>(desc.planes.size()), 1, [&](int n)\r
 241                         {\r
 242                                 auto plane            = desc.planes[n];\r
 243                                 auto result           = write->image_data(n).begin();\r
 244                                 auto decoded          = decoded_frame->data[n];\r
 245                                 auto decoded_linesize = decoded_frame->linesize[n];\r
 246                                 \r
 247                                 // Copy line by line since ffmpeg sometimes pads each line.\r
 248                                 tbb::parallel_for(tbb::blocked_range<size_t>(0, static_cast<int>(desc.planes[n].height)), [&](const tbb::blocked_range<size_t>& r)\r
 249                                 {\r
 250                                         for(size_t y = r.begin(); y != r.end(); ++y)\r
 251                                                 memcpy(result + y*plane.linesize, decoded + y*decoded_linesize, plane.linesize);\r
 252                                 });\r
 253 \r
 254                                 write->commit(n);\r
 255                         });\r
 256                 }\r
 257                 else\r
 258                 {\r
 259                         // Use sws_scale when provided colorspace has no hw-accel.\r
 260                         safe_ptr<AVFrame> av_frame(avcodec_alloc_frame(), av_free);     \r
 261                         avcodec_get_frame_defaults(av_frame.get());                     \r
 262                         avpicture_fill(reinterpret_cast<AVPicture*>(av_frame.get()), write->image_data().begin(), PIX_FMT_BGRA, width, height);\r
 263                  \r
 264                         sws_scale(sws_context_.get(), decoded_frame->data, decoded_frame->linesize, 0, height, av_frame->data, av_frame->linesize);     \r
 265 \r
 266                         write->commit();\r
 267                 }       \r
 268 \r
 269                 // Fix field-order if needed. DVVIDEO is in lower field. Make it upper field if needed.\r
 270                 if(decoded_frame->interlaced_frame)\r
 271                 {\r
 272                         switch(frame_factory_->get_video_format_desc().mode)\r
 273                         {\r
 274                         case core::video_mode::upper:\r
 275                                 if(!decoded_frame->top_field_first)\r
 276                                         write->get_image_transform().set_fill_translation(0.0f, 0.5/static_cast<double>(height));\r
 277                                 break;\r
 278                         case core::video_mode::lower:\r
 279                                 if(decoded_frame->top_field_first)\r
 280                                         write->get_image_transform().set_fill_translation(0.0f, -0.5/static_cast<double>(height));\r
 281                                 break;\r
 282                         }\r
 283                 }\r
 284 \r
 285                 return write;\r
 286         }\r
 287 };\r
 288 \r
 289 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
 290 std::deque<std::pair<int, safe_ptr<core::write_frame>>> video_decoder::receive(){return impl_->receive();}\r
 291 \r
 292 }