File: normal_distribution_base.h

package info (click to toggle)
libthrust 1.17.2-2
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 10,900 kB
  • sloc: ansic: 29,519; cpp: 23,989; python: 1,421; sh: 811; perl: 460; makefile: 112
file content (148 lines) | stat: -rw-r--r-- 4,200 bytes parent folder | download
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
/*
 *  Copyright 2008-2013 NVIDIA Corporation
 *
 *  Licensed under the Apache License, Version 2.0 (the "License");
 *  you may not use this file except in compliance with the License.
 *  You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 *  Unless required by applicable law or agreed to in writing, software
 *  distributed under the License is distributed on an "AS IS" BASIS,
 *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 *  See the License for the specific language governing permissions and
 *  limitations under the License.
 */

/*
 * Copyright Jens Maurer 2000-2001
 * Distributed under the Boost Software License, Version 1.0. (See
 * accompanying file LICENSE_1_0.txt or copy at
 * http://www.boost.org/LICENSE_1_0.txt)
 */

#pragma once

#include <thrust/detail/config.h>
#include <thrust/pair.h>
#include <thrust/random/uniform_real_distribution.h>
#include <limits>
#include <cmath>

THRUST_NAMESPACE_BEGIN
namespace random
{
namespace detail
{

// this version samples the normal distribution directly
// and uses the non-standard math function erfcinv
template<typename RealType>
  class normal_distribution_nvcc
{
  protected:
    template<typename UniformRandomNumberGenerator>
    __host__ __device__
    RealType sample(UniformRandomNumberGenerator &urng, const RealType mean, const RealType stddev)
    {
      using uint_type = typename UniformRandomNumberGenerator::result_type;
      constexpr uint_type urng_range = UniformRandomNumberGenerator::max - UniformRandomNumberGenerator::min;

      // Constants for conversion
      constexpr RealType S1 = static_cast<RealType>(1. / static_cast<double>(urng_range));
      constexpr RealType S2 = S1 / 2;

      RealType S3 = static_cast<RealType>(-1.4142135623730950488016887242097); // -sqrt(2)

      // Get the integer value
      uint_type u = urng() - UniformRandomNumberGenerator::min;

      // Ensure the conversion to float will give a value in the range [0,0.5)
      if(u > (urng_range / 2))
      {
        u = urng_range - u;
        S3 = -S3;
      }

      // Convert to floating point in [0,0.5)
      RealType p = u*S1 + S2;

      // Apply inverse error function
      return mean + stddev * S3 * erfcinv(2 * p);
    }

    // no-op
    __host__ __device__
    void reset() {}
};

// this version samples the normal distribution using
// Marsaglia's "polar method"
template<typename RealType>
  class normal_distribution_portable
{
  protected:
    normal_distribution_portable()
      : m_r1(), m_r2(), m_cached_rho(), m_valid(false)
    {}

    normal_distribution_portable(const normal_distribution_portable &other)
      : m_r1(other.m_r1), m_r2(other.m_r2), m_cached_rho(other.m_cached_rho), m_valid(other.m_valid)
    {}

    void reset()
    {
      m_valid = false;
    }

    // note that we promise to call this member function with the same mean and stddev
    template<typename UniformRandomNumberGenerator>
    __host__ __device__
    RealType sample(UniformRandomNumberGenerator &urng, const RealType mean, const RealType stddev)
    {
      // implementation from Boost
      // allow for Koenig lookup
      using std::sqrt; using std::log; using std::sin; using std::cos;

      if(!m_valid)
      {
        uniform_real_distribution<RealType> u01;
        m_r1 = u01(urng);
        m_r2 = u01(urng);
        m_cached_rho = sqrt(-RealType(2) * log(RealType(1)-m_r2));

        m_valid = true;
      }
      else
      {
        m_valid = false;
      }

      const RealType pi = RealType(3.14159265358979323846);

      RealType result = m_cached_rho * (m_valid ?
                          cos(RealType(2)*pi*m_r1) :
                          sin(RealType(2)*pi*m_r1));

      return mean + stddev * result;
    }

  private:
    RealType m_r1, m_r2, m_cached_rho;
    bool m_valid;
};

template<typename RealType>
  struct normal_distribution_base
{
#if THRUST_DEVICE_COMPILER == THRUST_DEVICE_COMPILER_NVCC && !defined(_NVHPC_CUDA)
  typedef normal_distribution_nvcc<RealType> type;
#else
  typedef normal_distribution_portable<RealType> type;
#endif
};

} // end detail
} // end random
THRUST_NAMESPACE_END