git.sesse.net Git - casparcg/blob - accelerator/cpu/image/image_mixer.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 "image_mixer.h"\r
  25 \r
  26 #include "../util/write_frame.h"\r
  27 #include "../util/blend.h"\r
  28 \r
  29 #include <common/assert.h>\r
  30 #include <common/gl/gl_check.h>\r
  31 #include <common/concurrency/async.h>\r
  32 #include <common/memory/memcpy.h>\r
  33 \r
  34 #include <core/frame/write_frame.h>\r
  35 #include <core/frame/frame_transform.h>\r
  36 #include <core/frame/pixel_format.h>\r
  37 #include <core/video_format.h>\r
  38 \r
  39 #include <modules/ffmpeg/producer/util/util.h>\r
  40 \r
  41 #include <asmlib.h>\r
  42 \r
  43 #include <gl/glew.h>\r
  44 \r
  45 #include <tbb/cache_aligned_allocator.h>\r
  46 #include <tbb/parallel_for_each.h>\r
  47 \r
  48 #include <boost/assign.hpp>\r
  49 #include <boost/foreach.hpp>\r
  50 #include <boost/range.hpp>\r
  51 #include <boost/range/algorithm_ext/erase.hpp>\r
  52 #include <boost/thread/future.hpp>\r
  53 \r
  54 #include <algorithm>\r
  55 #include <vector>\r
  56 \r
  57 #if defined(_MSC_VER)\r
  58 #pragma warning (push)\r
  59 #pragma warning (disable : 4244)\r
  60 #endif\r
  61 extern "C" \r
  62 {\r
  63         #include <libswscale/swscale.h>\r
  64         #include <libavcodec/avcodec.h>\r
  65         #include <libavformat/avformat.h>\r
  66 }\r
  67 #if defined(_MSC_VER)\r
  68 #pragma warning (pop)\r
  69 #endif\r
  70 \r
  71 namespace caspar { namespace accelerator { namespace cpu {\r
  72                 \r
  73 struct item\r
  74 {\r
  75         core::pixel_format_desc                                         pix_desc;\r
  76         std::vector<spl::shared_ptr<host_buffer>>       buffers;\r
  77         core::frame_transform                                           transform;\r
  78 \r
  79         item()\r
  80                 : pix_desc(core::pixel_format::invalid)\r
  81         {\r
  82         }\r
  83 };\r
  84 \r
  85 bool operator==(const item& lhs, const item& rhs)\r
  86 {\r
  87         return lhs.buffers == rhs.buffers && lhs.transform == rhs.transform;\r
  88 }\r
  89 \r
  90 bool operator!=(const item& lhs, const item& rhs)\r
  91 {\r
  92         return !(lhs == rhs);\r
  93 }\r
  94 \r
  95 struct layer\r
  96 {\r
  97         std::vector<item>       items;\r
  98 \r
  99         layer()\r
 100         {\r
 101         }\r
 102 \r
 103         layer(std::vector<item> items)\r
 104                 : items(std::move(items))\r
 105         {\r
 106         }\r
 107 };\r
 108 \r
 109 bool operator==(const layer& lhs, const layer& rhs)\r
 110 {\r
 111         return lhs.items == rhs.items;\r
 112 }\r
 113 \r
 114 bool operator!=(const layer& lhs, const layer& rhs)\r
 115 {\r
 116         return !(lhs == rhs);\r
 117 }\r
 118 \r
 119 class image_renderer\r
 120 {\r
 121         std::pair<std::vector<layer>, boost::shared_future<boost::iterator_range<const uint8_t*>>>      last_image_;\r
 122         std::map<int, std::shared_ptr<SwsContext>>                                                                                                      sws_contexts_;\r
 123 public: \r
 124         boost::shared_future<boost::iterator_range<const uint8_t*>> operator()(std::vector<layer> layers, const core::video_format_desc& format_desc)\r
 125         {       \r
 126                 if(last_image_.first == layers && last_image_.second.has_value())\r
 127                         return last_image_.second;\r
 128 \r
 129                 auto image      = render(layers, format_desc);\r
 130                 last_image_ = std::make_pair(std::move(layers), image);\r
 131                 return image;\r
 132         }\r
 133 \r
 134 private:\r
 135         boost::shared_future<boost::iterator_range<const uint8_t*>> render(std::vector<layer> layers, const core::video_format_desc& format_desc)\r
 136         {\r
 137                 static const auto empty = spl::make_shared<const std::vector<uint8_t, tbb::cache_aligned_allocator<uint8_t>>>(2048*2048*4, 0);\r
 138                 CASPAR_VERIFY(empty->size() >= format_desc.size);\r
 139                 \r
 140                 std::vector<item> items;\r
 141                 BOOST_FOREACH(auto& layer, layers)\r
 142                         items.insert(items.end(), layer.items.begin(), layer.items.end());\r
 143 \r
 144                 if(items.empty())\r
 145                 {\r
 146                         return async(launch_policy::deferred, [=]\r
 147                         {\r
 148                                 return boost::iterator_range<const uint8_t*>(empty->data(), empty->data() + format_desc.size);\r
 149                         });             \r
 150                 }\r
 151 \r
 152                 convert(items.begin(), items.end(), format_desc);                       \r
 153                 blend(items.begin(), items.end());\r
 154                 \r
 155                 auto buffer = items.front().buffers.at(0);\r
 156                 return async(launch_policy::deferred, [=]\r
 157                 {\r
 158                         return boost::iterator_range<const uint8_t*>(buffer->data(), buffer->data() + format_desc.size);\r
 159                 });             \r
 160         }\r
 161 \r
 162         template<typename I>\r
 163         void blend(I begin, I end)\r
 164         {\r
 165                 for(auto it = begin + 1; it != end; ++it)\r
 166                 {\r
 167                         auto size    = begin->buffers.at(0)->size();\r
 168                         auto dest    = begin->buffers.at(0)->data();\r
 169                         auto source2 = it->buffers.at(0)->data();\r
 170                         cpu::blend(dest, dest, source2, size);\r
 171                 }\r
 172         }\r
 173         \r
 174         template<typename I>\r
 175         void convert(I begin, I end, const core::video_format_desc& format_desc)\r
 176         {\r
 177                 tbb::parallel_for_each(begin, end, [&](item& item)\r
 178                 {\r
 179                         if(item.pix_desc.format == core::pixel_format::bgra)\r
 180                                 return;\r
 181 \r
 182                         auto input_av_frame = ffmpeg::make_av_frame(item.buffers, item.pix_desc);\r
 183                                                                 \r
 184                         int key = ((input_av_frame->width << 22) & 0xFFC00000) | ((input_av_frame->height << 6) & 0x003FC000) | ((input_av_frame->format << 7) & 0x00007F00);\r
 185                                                                         \r
 186                         auto& sws_context = sws_contexts_[key];\r
 187                         if(!sws_context)\r
 188                         {\r
 189                                 double param;\r
 190                                 sws_context.reset(sws_getContext(input_av_frame->width, input_av_frame->height, static_cast<PixelFormat>(input_av_frame->format), format_desc.width, format_desc.height, PIX_FMT_BGRA, SWS_BILINEAR, nullptr, nullptr, &param), sws_freeContext);\r
 191                         }\r
 192                         \r
 193                         if(!sws_context)                                \r
 194                                 BOOST_THROW_EXCEPTION(operation_failed() << msg_info("Could not create software scaling context.") << boost::errinfo_api_function("sws_getContext"));                           \r
 195                 \r
 196                         auto dest = spl::make_shared<host_buffer>(format_desc.size);\r
 197 \r
 198                         spl::shared_ptr<AVFrame> av_frame(avcodec_alloc_frame(), av_free);      \r
 199                         avcodec_get_frame_defaults(av_frame.get());                     \r
 200                         avpicture_fill(reinterpret_cast<AVPicture*>(av_frame.get()), dest->data(), PIX_FMT_BGRA, format_desc.width, format_desc.height);\r
 201                                 \r
 202                         sws_scale(sws_context.get(), input_av_frame->data, input_av_frame->linesize, 0, input_av_frame->height, av_frame->data, av_frame->linesize);    \r
 203 \r
 204                         item.buffers.clear();\r
 205                         item.buffers.push_back(dest);\r
 206                         item.pix_desc = core::pixel_format_desc(core::pixel_format::bgra);\r
 207                         item.pix_desc.planes.clear();\r
 208                         item.pix_desc.planes.push_back(core::pixel_format_desc::plane(format_desc.width, format_desc.height, 4));\r
 209                 });\r
 210         }\r
 211 };\r
 212                 \r
 213 struct image_mixer::impl : boost::noncopyable\r
 214 {       \r
 215         image_renderer                                          renderer_;\r
 216         std::vector<core::frame_transform>      transform_stack_;\r
 217         std::vector<layer>                                      layers_; // layer/stream/items\r
 218 public:\r
 219         impl() \r
 220                 : transform_stack_(1)   \r
 221         {\r
 222                 CASPAR_LOG(info) << L"Initialized CPU Accelerated Image Mixer";\r
 223         }\r
 224 \r
 225         void begin_layer(core::blend_mode blend_mode)\r
 226         {\r
 227                 layers_.push_back(layer(std::vector<item>()));\r
 228         }\r
 229                 \r
 230         void push(core::frame_transform& transform)\r
 231         {\r
 232                 transform_stack_.push_back(transform_stack_.back()*transform);\r
 233         }\r
 234                 \r
 235         void visit(core::data_frame& frame2)\r
 236         {                       \r
 237                 write_frame* frame = dynamic_cast<write_frame*>(&frame2);\r
 238                 if(frame == nullptr)\r
 239                         return;\r
 240 \r
 241                 if(frame->get_pixel_format_desc().format == core::pixel_format::invalid)\r
 242                         return;\r
 243 \r
 244                 if(frame->get_buffers().empty())\r
 245                         return;\r
 246 \r
 247                 if(transform_stack_.back().field_mode == core::field_mode::empty)\r
 248                         return;\r
 249 \r
 250                 item item;\r
 251                 item.pix_desc                   = frame->get_pixel_format_desc();\r
 252                 item.buffers                    = frame->get_buffers();                         \r
 253                 item.transform                  = transform_stack_.back();\r
 254                 item.transform.volume   = core::frame_transform().volume; // Set volume to default since we don't care about it here.\r
 255 \r
 256                 layers_.back().items.push_back(item);\r
 257         }\r
 258 \r
 259         void pop()\r
 260         {\r
 261                 transform_stack_.pop_back();\r
 262         }\r
 263 \r
 264         void end_layer()\r
 265         {               \r
 266         }\r
 267         \r
 268         boost::shared_future<boost::iterator_range<const uint8_t*>> render(const core::video_format_desc& format_desc)\r
 269         {\r
 270                 // Remove empty layers.\r
 271                 boost::range::remove_erase_if(layers_, [](const layer& layer)\r
 272                 {\r
 273                         return layer.items.empty();\r
 274                 });\r
 275 \r
 276                 return renderer_(std::move(layers_), format_desc);\r
 277         }\r
 278         \r
 279         virtual spl::shared_ptr<cpu::write_frame> create_frame(const void* tag, const core::pixel_format_desc& desc)\r
 280         {\r
 281                 return spl::make_shared<cpu::write_frame>(tag, desc);\r
 282         }\r
 283 };\r
 284 \r
 285 image_mixer::image_mixer() : impl_(new impl()){}\r
 286 void image_mixer::push(core::frame_transform& transform){impl_->push(transform);}\r
 287 void image_mixer::visit(core::data_frame& frame){impl_->visit(frame);}\r
 288 void image_mixer::pop(){impl_->pop();}\r
 289 boost::shared_future<boost::iterator_range<const uint8_t*>> image_mixer::operator()(const core::video_format_desc& format_desc){return impl_->render(format_desc);}\r
 290 void image_mixer::begin_layer(core::blend_mode blend_mode){impl_->begin_layer(blend_mode);}\r
 291 void image_mixer::end_layer(){impl_->end_layer();}\r
 292 spl::shared_ptr<core::write_frame> image_mixer::create_frame(const void* tag, const core::pixel_format_desc& desc) {return impl_->create_frame(tag, desc);}\r
 293 \r
 294 }}}