ceph/src/boost/libs/compute/include/boost/compute/algorithm/set_intersection.hpp

   1 //---------------------------------------------------------------------------//
   2 // Copyright (c) 2014 Roshan <thisisroshansmail@gmail.com>
   3 //
   4 // Distributed under the Boost Software License, Version 1.0
   5 // See accompanying file LICENSE_1_0.txt or copy at
   6 // http://www.boost.org/LICENSE_1_0.txt
   7 //
   8 // See http://boostorg.github.com/compute for more information.
   9 //---------------------------------------------------------------------------//
  10
  11 #ifndef BOOST_COMPUTE_ALGORITHM_SET_INTERSECTION_HPP
  12 #define BOOST_COMPUTE_ALGORITHM_SET_INTERSECTION_HPP
  13
  14 #include <iterator>
  15
  16 #include <boost/compute/algorithm/detail/compact.hpp>
  17 #include <boost/compute/algorithm/detail/balanced_path.hpp>
  18 #include <boost/compute/algorithm/exclusive_scan.hpp>
  19 #include <boost/compute/algorithm/fill_n.hpp>
  20 #include <boost/compute/container/vector.hpp>
  21 #include <boost/compute/detail/iterator_range_size.hpp>
  22 #include <boost/compute/detail/meta_kernel.hpp>
  23 #include <boost/compute/system.hpp>
  24
  25 namespace boost {
  26 namespace compute {
  27 namespace detail {
  28
  29 ///
  30 /// \brief Serial set intersection kernel class
  31 ///
  32 /// Subclass of meta_kernel to perform serial set intersection after tiling
  33 ///
  34 class serial_set_intersection_kernel : meta_kernel
  35 {
  36 public:
  37     unsigned int tile_size;
  38
  39     serial_set_intersection_kernel() : meta_kernel("set_intersection")
  40     {
  41         tile_size = 4;
  42     }
  43
  44     template<class InputIterator1, class InputIterator2,
  45              class InputIterator3, class InputIterator4,
  46              class OutputIterator1, class OutputIterator2>
  47     void set_range(InputIterator1 first1,
  48                     InputIterator2 first2,
  49                     InputIterator3 tile_first1,
  50                     InputIterator3 tile_last1,
  51                     InputIterator4 tile_first2,
  52                     OutputIterator1 result,
  53                     OutputIterator2 counts)
  54     {
  55         m_count = iterator_range_size(tile_first1, tile_last1) - 1;
  56
  57         *this <<
  58         "uint i = get_global_id(0);\n" <<
  59         "uint start1 = " << tile_first1[expr<uint_>("i")] << ";\n" <<
  60         "uint end1 = " << tile_first1[expr<uint_>("i+1")] << ";\n" <<
  61         "uint start2 = " << tile_first2[expr<uint_>("i")] << ";\n" <<
  62         "uint end2 = " << tile_first2[expr<uint_>("i+1")] << ";\n" <<
  63         "uint index = i*" << tile_size << ";\n" <<
  64         "uint count = 0;\n" <<
  65         "while(start1<end1 && start2<end2)\n" <<
  66         "{\n" <<
  67         "   if(" << first1[expr<uint_>("start1")] << " == " <<
  68                     first2[expr<uint_>("start2")] << ")\n" <<
  69         "   {\n" <<
  70                 result[expr<uint_>("index")] <<
  71                     " = " << first1[expr<uint_>("start1")] << ";\n" <<
  72         "       index++; count++;\n" <<
  73         "       start1++; start2++;\n" <<
  74         "   }\n" <<
  75         "   else if(" << first1[expr<uint_>("start1")] << " < " <<
  76                         first2[expr<uint_>("start2")] << ")\n" <<
  77         "       start1++;\n" <<
  78         "   else start2++;\n" <<
  79         "}\n" <<
  80         counts[expr<uint_>("i")] << " = count;\n";
  81     }
  82
  83     event exec(command_queue &queue)
  84     {
  85         if(m_count == 0) {
  86             return event();
  87         }
  88
  89         return exec_1d(queue, 0, m_count);
  90     }
  91
  92 private:
  93     size_t m_count;
  94 };
  95
  96 } //end detail namespace
  97
  98 ///
  99 /// \brief Set intersection algorithm
 100 ///
 101 /// Finds the intersection of the sorted range [first1, last1) with the sorted
 102 /// range [first2, last2) and stores it in range starting at result
 103 /// \return Iterator pointing to end of intersection
 104 ///
 105 /// \param first1 Iterator pointing to start of first set
 106 /// \param last1 Iterator pointing to end of first set
 107 /// \param first2 Iterator pointing to start of second set
 108 /// \param last2 Iterator pointing to end of second set
 109 /// \param result Iterator pointing to start of range in which the intersection
 110 /// will be stored
 111 /// \param queue Queue on which to execute
 112 ///
 113 template<class InputIterator1, class InputIterator2, class OutputIterator>
 114 inline OutputIterator set_intersection(InputIterator1 first1,
 115                                        InputIterator1 last1,
 116                                        InputIterator2 first2,
 117                                        InputIterator2 last2,
 118                                        OutputIterator result,
 119                                        command_queue &queue = system::default_queue())
 120 {
 121     typedef typename std::iterator_traits<InputIterator1>::value_type value_type;
 122
 123     int tile_size = 1024;
 124
 125     int count1 = detail::iterator_range_size(first1, last1);
 126     int count2 = detail::iterator_range_size(first2, last2);
 127
 128     vector<uint_> tile_a((count1+count2+tile_size-1)/tile_size+1, queue.get_context());
 129     vector<uint_> tile_b((count1+count2+tile_size-1)/tile_size+1, queue.get_context());
 130
 131     // Tile the sets
 132     detail::balanced_path_kernel tiling_kernel;
 133     tiling_kernel.tile_size = tile_size;
 134     tiling_kernel.set_range(first1, last1, first2, last2,
 135                             tile_a.begin()+1, tile_b.begin()+1);
 136     fill_n(tile_a.begin(), 1, 0, queue);
 137     fill_n(tile_b.begin(), 1, 0, queue);
 138     tiling_kernel.exec(queue);
 139
 140     fill_n(tile_a.end()-1, 1, count1, queue);
 141     fill_n(tile_b.end()-1, 1, count2, queue);
 142
 143     vector<value_type> temp_result(count1+count2, queue.get_context());
 144     vector<uint_> counts((count1+count2+tile_size-1)/tile_size + 1, queue.get_context());
 145     fill_n(counts.end()-1, 1, 0, queue);
 146
 147     // Find individual intersections
 148     detail::serial_set_intersection_kernel intersection_kernel;
 149     intersection_kernel.tile_size = tile_size;
 150     intersection_kernel.set_range(first1, first2, tile_a.begin(), tile_a.end(),
 151                                   tile_b.begin(), temp_result.begin(), counts.begin());
 152
 153     intersection_kernel.exec(queue);
 154
 155     exclusive_scan(counts.begin(), counts.end(), counts.begin(), queue);
 156
 157     // Compact the results
 158     detail::compact_kernel compact_kernel;
 159     compact_kernel.tile_size = tile_size;
 160     compact_kernel.set_range(temp_result.begin(), counts.begin(), counts.end(), result);
 161
 162     compact_kernel.exec(queue);
 163
 164     return result + (counts.end() - 1).read(queue);
 165 }
 166
 167 } //end compute namespace
 168 } //end boost namespace
 169
 170 #endif // BOOST_COMPUTE_ALGORITHM_SET_INTERSECTION_HPP