git.sesse.net Git - casparcg/blob - core/mixer/read_frame.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 "read_frame.h"\r
  23 \r
  24 #include "gpu/fence.h"\r
  25 #include "gpu/host_buffer.h"    \r
  26 #include "gpu/ogl_device.h"\r
  27 \r
  28 namespace caspar { namespace core {\r
  29                                                                                                                                                                                                                                                                                                                         \r
  30 struct read_frame::implementation : boost::noncopyable\r
  31 {\r
  32         ogl_device&                                             ogl_;\r
  33         safe_ptr<host_buffer>                   image_data_;\r
  34         std::vector<int16_t>                    audio_data_;\r
  35 \r
  36 public:\r
  37         implementation(ogl_device& ogl, safe_ptr<host_buffer>&& image_data, std::vector<int16_t>&& audio_data) \r
  38                 : ogl_(ogl)\r
  39                 , image_data_(std::move(image_data))\r
  40                 , audio_data_(std::move(audio_data)){}  \r
  41         \r
  42         const boost::iterator_range<const uint8_t*> image_data()\r
  43         {\r
  44                 if(!image_data_->data())\r
  45                         image_data_->map(ogl_);\r
  46 \r
  47                 auto ptr = static_cast<const uint8_t*>(image_data_->data());\r
  48                 return boost::iterator_range<const uint8_t*>(ptr, ptr + image_data_->size());\r
  49         }\r
  50         const boost::iterator_range<const int16_t*> audio_data()\r
  51         {\r
  52                 return boost::iterator_range<const int16_t*>(audio_data_.data(), audio_data_.data() + audio_data_.size());\r
  53         }\r
  54 };\r
  55 \r
  56 read_frame::read_frame(ogl_device& ogl, safe_ptr<host_buffer>&& image_data, std::vector<int16_t>&& audio_data) \r
  57         : impl_(new implementation(ogl, std::move(image_data), std::move(audio_data))){}\r
  58 read_frame::read_frame(){}\r
  59 const boost::iterator_range<const uint8_t*> read_frame::image_data()\r
  60 {\r
  61         return impl_ ? impl_->image_data() : boost::iterator_range<const uint8_t*>();\r
  62 }\r
  63 \r
  64 const boost::iterator_range<const int16_t*> read_frame::audio_data()\r
  65 {\r
  66         return impl_ ? impl_->audio_data() : boost::iterator_range<const int16_t*>();\r
  67 }\r
  68 \r
  69 size_t read_frame::image_size() const{return impl_->image_data_->size();}\r
  70 \r
  71 //#include <tbb/scalable_allocator.h>\r
  72 //#include <tbb/parallel_for.h>\r
  73 //#include <tbb/enumerable_thread_specific.h>\r
  74 //#define               CACHED_BUFFER_SIZE      4096    \r
  75 //typedef               unsigned int            UINT;\r
  76 //\r
  77 //struct cache_buffer\r
  78 //{\r
  79 //      cache_buffer() : data(scalable_aligned_malloc(CACHED_BUFFER_SIZE, 64)){}\r
  80 //      ~cache_buffer() {scalable_aligned_free(data);}\r
  81 //      void* data;\r
  82 //};\r
  83 //\r
  84 //void  CopyFrame( void * pSrc, void * pDest, UINT width, UINT height, UINT pitch );\r
  85 //\r
  86 //void* copy_frame(void* dest, const safe_ptr<read_frame>& frame)\r
  87 //{\r
  88 //      auto src                = frame->image_data().begin();\r
  89 //      auto height             = 720;\r
  90 //      auto width4             = frame->image_data().size()/height;\r
  91 //\r
  92 //      CASPAR_ASSERT(frame->image_data().size() % height == 0);\r
  93 //                      \r
  94 //      tbb::affinity_partitioner ap;\r
  95 //      tbb::parallel_for(tbb::blocked_range<size_t>(0, height), [&](tbb::blocked_range<size_t>& r)\r
  96 //      {\r
  97 //              CopyFrame(const_cast<uint8_t*>(src)+r.begin()*width4, reinterpret_cast<uint8_t*>(dest)+r.begin()*width4, width4, r.size(), width4);\r
  98 //      }, ap);\r
  99 //\r
 100 //      return dest;\r
 101 //}\r
 102 //\r
 103 ////  CopyFrame( )\r
 104 ////\r
 105 ////  COPIES VIDEO FRAMES FROM USWC MEMORY TO WB SYSTEM MEMORY VIA CACHED BUFFER\r
 106 ////    ASSUMES PITCH IS A MULTIPLE OF 64B CACHE LINE SIZE, WIDTH MAY NOT BE\r
 107 //// http://software.intel.com/en-us/articles/copying-accelerated-video-decode-frame-buffers/\r
 108 //void CopyFrame( void * pSrc, void * pDest, UINT width, UINT height, UINT pitch )\r
 109 //{\r
 110 //      tbb::enumerable_thread_specific<cache_buffer> cache_buffers;\r
 111 //\r
 112 //      void *          pCacheBlock = cache_buffers.local().data;\r
 113 //\r
 114 //      __m128i         x0, x1, x2, x3;\r
 115 //      __m128i         *pLoad;\r
 116 //      __m128i         *pStore;\r
 117 //      __m128i         *pCache;\r
 118 //      UINT            x, y, yLoad, yStore;\r
 119 //      UINT            rowsPerBlock;\r
 120 //      UINT            width64;\r
 121 //      UINT            extraPitch;     \r
 122 //\r
 123 //      rowsPerBlock = CACHED_BUFFER_SIZE / pitch;\r
 124 //      width64 = (width + 63) & ~0x03f;\r
 125 //      extraPitch = (pitch - width64) / 16;\r
 126 //\r
 127 //      pLoad  = (__m128i *)pSrc;\r
 128 //      pStore = (__m128i *)pDest;\r
 129 //\r
 130 //      //  COPY THROUGH 4KB CACHED BUFFER\r
 131 //      for( y = 0; y < height; y += rowsPerBlock  )\r
 132 //      {\r
 133 //              //  ROWS LEFT TO COPY AT END\r
 134 //              if( y + rowsPerBlock > height )\r
 135 //                      rowsPerBlock = height - y;\r
 136 //\r
 137 //              pCache = (__m128i *)pCacheBlock;\r
 138 //\r
 139 //              _mm_mfence();                           \r
 140 //              \r
 141 //              // LOAD ROWS OF PITCH WIDTH INTO CACHED BLOCK\r
 142 //              for( yLoad = 0; yLoad < rowsPerBlock; yLoad++ )\r
 143 //              {\r
 144 //                      // COPY A ROW, CACHE LINE AT A TIME\r
 145 //                      for( x = 0; x < pitch; x +=64 )\r
 146 //                      {\r
 147 //                              x0 = _mm_stream_load_si128( pLoad +0 );\r
 148 //                              x1 = _mm_stream_load_si128( pLoad +1 );\r
 149 //                              x2 = _mm_stream_load_si128( pLoad +2 );\r
 150 //                              x3 = _mm_stream_load_si128( pLoad +3 );\r
 151 //\r
 152 //                              _mm_store_si128( pCache +0,     x0 );\r
 153 //                              _mm_store_si128( pCache +1, x1 );\r
 154 //                              _mm_store_si128( pCache +2, x2 );\r
 155 //                              _mm_store_si128( pCache +3, x3 );\r
 156 //\r
 157 //                              pCache += 4;\r
 158 //                              pLoad += 4;\r
 159 //                      }\r
 160 //              }\r
 161 //\r
 162 //              _mm_mfence();\r
 163 //\r
 164 //              pCache = (__m128i *)pCacheBlock;\r
 165 //\r
 166 //              // STORE ROWS OF FRAME WIDTH FROM CACHED BLOCK\r
 167 //              for( yStore = 0; yStore < rowsPerBlock; yStore++ )\r
 168 //              {\r
 169 //                      // copy a row, cache line at a time\r
 170 //                      for( x = 0; x < width64; x +=64 )\r
 171 //                      {\r
 172 //                              x0 = _mm_load_si128( pCache );\r
 173 //                              x1 = _mm_load_si128( pCache +1 );\r
 174 //                              x2 = _mm_load_si128( pCache +2 );\r
 175 //                              x3 = _mm_load_si128( pCache +3 );\r
 176 //\r
 177 //                              _mm_stream_si128( pStore,       x0 );\r
 178 //                              _mm_stream_si128( pStore +1, x1 );\r
 179 //                              _mm_stream_si128( pStore +2, x2 );\r
 180 //                              _mm_stream_si128( pStore +3, x3 );\r
 181 //\r
 182 //                              pCache += 4;\r
 183 //                              pStore += 4;\r
 184 //                      }\r
 185 //\r
 186 //                      pCache += extraPitch;\r
 187 //                      pStore += extraPitch;\r
 188 //              }\r
 189 //      }\r
 190 //}\r
 191 \r
 192 }}