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/static_assert.hpp>
17
18 #include <boost/compute/algorithm/detail/compact.hpp>
19 #include <boost/compute/algorithm/detail/balanced_path.hpp>
20 #include <boost/compute/algorithm/exclusive_scan.hpp>
21 #include <boost/compute/algorithm/fill_n.hpp>
22 #include <boost/compute/container/vector.hpp>
23 #include <boost/compute/detail/iterator_range_size.hpp>
24 #include <boost/compute/detail/meta_kernel.hpp>
25 #include <boost/compute/system.hpp>
26 #include <boost/compute/type_traits/is_device_iterator.hpp>
27
28 namespace boost {
29 namespace compute {
30 namespace detail {
31
32 ///
33 /// \brief Serial set intersection kernel class
34 ///
35 /// Subclass of meta_kernel to perform serial set intersection after tiling
36 ///
37 class serial_set_intersection_kernel : meta_kernel
38 {
39 public:
40 unsigned int tile_size;
41
serial_set_intersection_kernel()42 serial_set_intersection_kernel() : meta_kernel("set_intersection")
43 {
44 tile_size = 4;
45 }
46
47 template<class InputIterator1, class InputIterator2,
48 class InputIterator3, class InputIterator4,
49 class OutputIterator1, class OutputIterator2>
set_range(InputIterator1 first1,InputIterator2 first2,InputIterator3 tile_first1,InputIterator3 tile_last1,InputIterator4 tile_first2,OutputIterator1 result,OutputIterator2 counts)50 void set_range(InputIterator1 first1,
51 InputIterator2 first2,
52 InputIterator3 tile_first1,
53 InputIterator3 tile_last1,
54 InputIterator4 tile_first2,
55 OutputIterator1 result,
56 OutputIterator2 counts)
57 {
58 m_count = iterator_range_size(tile_first1, tile_last1) - 1;
59
60 *this <<
61 "uint i = get_global_id(0);\n" <<
62 "uint start1 = " << tile_first1[expr<uint_>("i")] << ";\n" <<
63 "uint end1 = " << tile_first1[expr<uint_>("i+1")] << ";\n" <<
64 "uint start2 = " << tile_first2[expr<uint_>("i")] << ";\n" <<
65 "uint end2 = " << tile_first2[expr<uint_>("i+1")] << ";\n" <<
66 "uint index = i*" << tile_size << ";\n" <<
67 "uint count = 0;\n" <<
68 "while(start1<end1 && start2<end2)\n" <<
69 "{\n" <<
70 " if(" << first1[expr<uint_>("start1")] << " == " <<
71 first2[expr<uint_>("start2")] << ")\n" <<
72 " {\n" <<
73 result[expr<uint_>("index")] <<
74 " = " << first1[expr<uint_>("start1")] << ";\n" <<
75 " index++; count++;\n" <<
76 " start1++; start2++;\n" <<
77 " }\n" <<
78 " else if(" << first1[expr<uint_>("start1")] << " < " <<
79 first2[expr<uint_>("start2")] << ")\n" <<
80 " start1++;\n" <<
81 " else start2++;\n" <<
82 "}\n" <<
83 counts[expr<uint_>("i")] << " = count;\n";
84 }
85
exec(command_queue & queue)86 event exec(command_queue &queue)
87 {
88 if(m_count == 0) {
89 return event();
90 }
91
92 return exec_1d(queue, 0, m_count);
93 }
94
95 private:
96 size_t m_count;
97 };
98
99 } //end detail namespace
100
101 ///
102 /// \brief Set intersection algorithm
103 ///
104 /// Finds the intersection of the sorted range [first1, last1) with the sorted
105 /// range [first2, last2) and stores it in range starting at result
106 /// \return Iterator pointing to end of intersection
107 ///
108 /// \param first1 Iterator pointing to start of first set
109 /// \param last1 Iterator pointing to end of first set
110 /// \param first2 Iterator pointing to start of second set
111 /// \param last2 Iterator pointing to end of second set
112 /// \param result Iterator pointing to start of range in which the intersection
113 /// will be stored
114 /// \param queue Queue on which to execute
115 ///
116 /// Space complexity:
117 /// \Omega(2(distance(\p first1, \p last1) + distance(\p first2, \p last2)))
118 template<class InputIterator1, class InputIterator2, class OutputIterator>
set_intersection(InputIterator1 first1,InputIterator1 last1,InputIterator2 first2,InputIterator2 last2,OutputIterator result,command_queue & queue=system::default_queue ())119 inline OutputIterator set_intersection(InputIterator1 first1,
120 InputIterator1 last1,
121 InputIterator2 first2,
122 InputIterator2 last2,
123 OutputIterator result,
124 command_queue &queue = system::default_queue())
125 {
126 BOOST_STATIC_ASSERT(is_device_iterator<InputIterator1>::value);
127 BOOST_STATIC_ASSERT(is_device_iterator<InputIterator2>::value);
128 BOOST_STATIC_ASSERT(is_device_iterator<OutputIterator>::value);
129
130 typedef typename std::iterator_traits<InputIterator1>::value_type value_type;
131
132 int tile_size = 1024;
133
134 int count1 = detail::iterator_range_size(first1, last1);
135 int count2 = detail::iterator_range_size(first2, last2);
136
137 vector<uint_> tile_a((count1+count2+tile_size-1)/tile_size+1, queue.get_context());
138 vector<uint_> tile_b((count1+count2+tile_size-1)/tile_size+1, queue.get_context());
139
140 // Tile the sets
141 detail::balanced_path_kernel tiling_kernel;
142 tiling_kernel.tile_size = tile_size;
143 tiling_kernel.set_range(first1, last1, first2, last2,
144 tile_a.begin()+1, tile_b.begin()+1);
145 fill_n(tile_a.begin(), 1, 0, queue);
146 fill_n(tile_b.begin(), 1, 0, queue);
147 tiling_kernel.exec(queue);
148
149 fill_n(tile_a.end()-1, 1, count1, queue);
150 fill_n(tile_b.end()-1, 1, count2, queue);
151
152 vector<value_type> temp_result(count1+count2, queue.get_context());
153 vector<uint_> counts((count1+count2+tile_size-1)/tile_size + 1, queue.get_context());
154 fill_n(counts.end()-1, 1, 0, queue);
155
156 // Find individual intersections
157 detail::serial_set_intersection_kernel intersection_kernel;
158 intersection_kernel.tile_size = tile_size;
159 intersection_kernel.set_range(first1, first2, tile_a.begin(), tile_a.end(),
160 tile_b.begin(), temp_result.begin(), counts.begin());
161
162 intersection_kernel.exec(queue);
163
164 exclusive_scan(counts.begin(), counts.end(), counts.begin(), queue);
165
166 // Compact the results
167 detail::compact_kernel compact_kernel;
168 compact_kernel.tile_size = tile_size;
169 compact_kernel.set_range(temp_result.begin(), counts.begin(), counts.end(), result);
170
171 compact_kernel.exec(queue);
172
173 return result + (counts.end() - 1).read(queue);
174 }
175
176 } //end compute namespace
177 } //end boost namespace
178
179 #endif // BOOST_COMPUTE_ALGORITHM_SET_INTERSECTION_HPP
180