]> git.proxmox.com Git - ceph.git/blob - ceph/src/boost/boost/compute/random/discrete_distribution.hpp
update sources to v12.2.3
[ceph.git] / ceph / src / boost / boost / compute / random / discrete_distribution.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_RANDOM_DISCRETE_DISTRIBUTION_HPP
12 #define BOOST_COMPUTE_RANDOM_DISCRETE_DISTRIBUTION_HPP
13
14 #include <numeric>
15
16 #include <boost/config.hpp>
17 #include <boost/type_traits.hpp>
18 #include <boost/static_assert.hpp>
19
20 #include <boost/compute/command_queue.hpp>
21 #include <boost/compute/function.hpp>
22 #include <boost/compute/algorithm/accumulate.hpp>
23 #include <boost/compute/algorithm/copy.hpp>
24 #include <boost/compute/algorithm/transform.hpp>
25 #include <boost/compute/detail/literal.hpp>
26 #include <boost/compute/types/fundamental.hpp>
27
28 namespace boost {
29 namespace compute {
30
31 /// \class discrete_distribution
32 /// \brief Produces random integers on the interval [0, n), where
33 /// probability of each integer is given by the weight of the ith
34 /// integer divided by the sum of all weights.
35 ///
36 /// The following example shows how to setup a discrete distribution to
37 /// produce 0 and 1 with equal probability
38 ///
39 /// \snippet test/test_discrete_distribution.cpp generate
40 ///
41 template<class IntType = uint_>
42 class discrete_distribution
43 {
44 public:
45 typedef IntType result_type;
46
47 /// Creates a new discrete distribution with a single weight p = { 1 }.
48 /// This distribution produces only zeroes.
49 discrete_distribution()
50 : m_probabilities(1, double(1)),
51 m_scanned_probabilities(1, double(1))
52 {
53
54 }
55
56 /// Creates a new discrete distribution with weights given by
57 /// the range [\p first, \p last).
58 template<class InputIterator>
59 discrete_distribution(InputIterator first, InputIterator last)
60 : m_probabilities(first, last),
61 m_scanned_probabilities(std::distance(first, last))
62 {
63 if(first != last) {
64 // after this m_scanned_probabilities.back() is a sum of all
65 // weights from the range [first, last)
66 std::partial_sum(first, last, m_scanned_probabilities.begin());
67
68 std::vector<double>::iterator i = m_probabilities.begin();
69 std::vector<double>::iterator j = m_scanned_probabilities.begin();
70 for(; i != m_probabilities.end(); ++i, ++j)
71 {
72 // dividing each weight by sum of all weights to
73 // get probabilities
74 *i = *i / m_scanned_probabilities.back();
75 // dividing each partial sum of weights by sum of
76 // all weights to get partial sums of probabilities
77 *j = *j / m_scanned_probabilities.back();
78 }
79 }
80 else {
81 m_probabilities.push_back(double(1));
82 m_scanned_probabilities.push_back(double(1));
83 }
84 }
85
86 /// Destroys the discrete_distribution object.
87 ~discrete_distribution()
88 {
89 }
90
91 /// Returns the probabilities
92 ::std::vector<double> probabilities() const
93 {
94 return m_probabilities;
95 }
96
97 /// Returns the minimum potentially generated value.
98 result_type min BOOST_PREVENT_MACRO_SUBSTITUTION () const
99 {
100 return result_type(0);
101 }
102
103 /// Returns the maximum potentially generated value.
104 result_type max BOOST_PREVENT_MACRO_SUBSTITUTION () const
105 {
106 size_t type_max = static_cast<size_t>(
107 (std::numeric_limits<result_type>::max)()
108 );
109 if(m_probabilities.size() - 1 > type_max) {
110 return (std::numeric_limits<result_type>::max)();
111 }
112 return static_cast<result_type>(m_probabilities.size() - 1);
113 }
114
115 /// Generates uniformly distributed integers and stores
116 /// them to the range [\p first, \p last).
117 template<class OutputIterator, class Generator>
118 void generate(OutputIterator first,
119 OutputIterator last,
120 Generator &generator,
121 command_queue &queue)
122 {
123 std::string source = "inline IntType scale_random(uint x)\n";
124
125 source = source +
126 "{\n" +
127 "float rno = convert_float(x) / UINT_MAX;\n";
128 for(size_t i = 0; i < m_scanned_probabilities.size() - 1; i++)
129 {
130 source = source +
131 "if(rno <= " + detail::make_literal<float>(m_scanned_probabilities[i]) + ")\n" +
132 " return " + detail::make_literal(i) + ";\n";
133 }
134
135 source = source +
136 "return " + detail::make_literal(m_scanned_probabilities.size() - 1) + ";\n" +
137 "}\n";
138
139 BOOST_COMPUTE_FUNCTION(IntType, scale_random, (const uint_ x), {});
140
141 scale_random.set_source(source);
142 scale_random.define("IntType", type_name<IntType>());
143
144 generator.generate(first, last, scale_random, queue);
145 }
146
147 private:
148 ::std::vector<double> m_probabilities;
149 ::std::vector<double> m_scanned_probabilities;
150
151 BOOST_STATIC_ASSERT_MSG(
152 boost::is_integral<IntType>::value,
153 "Template argument must be integral"
154 );
155 };
156
157 } // end compute namespace
158 } // end boost namespace
159
160 #endif // BOOST_COMPUTE_RANDOM_UNIFORM_INT_DISTRIBUTION_HPP