Revert "[libc] Use best-fit binary trie to make malloc logarithmic" (#117065)
[llvm-project.git] / libcxx / include / __random / poisson_distribution.h
blob61a092ef9dd4dd299b3d1a11d054b0062d127e8a
1 //===----------------------------------------------------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
9 #ifndef _LIBCPP___RANDOM_POISSON_DISTRIBUTION_H
10 #define _LIBCPP___RANDOM_POISSON_DISTRIBUTION_H
12 #include <__config>
13 #include <__random/clamp_to_integral.h>
14 #include <__random/exponential_distribution.h>
15 #include <__random/is_valid.h>
16 #include <__random/normal_distribution.h>
17 #include <__random/uniform_real_distribution.h>
18 #include <cmath>
19 #include <iosfwd>
20 #include <limits>
22 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
23 # pragma GCC system_header
24 #endif
26 _LIBCPP_PUSH_MACROS
27 #include <__undef_macros>
29 _LIBCPP_BEGIN_NAMESPACE_STD
31 template <class _IntType = int>
32 class _LIBCPP_TEMPLATE_VIS poisson_distribution {
33 static_assert(__libcpp_random_is_valid_inttype<_IntType>::value, "IntType must be a supported integer type");
35 public:
36 // types
37 typedef _IntType result_type;
39 class _LIBCPP_TEMPLATE_VIS param_type {
40 double __mean_;
41 double __s_;
42 double __d_;
43 double __l_;
44 double __omega_;
45 double __c0_;
46 double __c1_;
47 double __c2_;
48 double __c3_;
49 double __c_;
51 public:
52 typedef poisson_distribution distribution_type;
54 _LIBCPP_HIDE_FROM_ABI explicit param_type(double __mean = 1.0);
56 _LIBCPP_HIDE_FROM_ABI double mean() const { return __mean_; }
58 friend _LIBCPP_HIDE_FROM_ABI bool operator==(const param_type& __x, const param_type& __y) {
59 return __x.__mean_ == __y.__mean_;
61 friend _LIBCPP_HIDE_FROM_ABI bool operator!=(const param_type& __x, const param_type& __y) { return !(__x == __y); }
63 friend class poisson_distribution;
66 private:
67 param_type __p_;
69 public:
70 // constructors and reset functions
71 #ifndef _LIBCPP_CXX03_LANG
72 _LIBCPP_HIDE_FROM_ABI poisson_distribution() : poisson_distribution(1.0) {}
73 _LIBCPP_HIDE_FROM_ABI explicit poisson_distribution(double __mean) : __p_(__mean) {}
74 #else
75 _LIBCPP_HIDE_FROM_ABI explicit poisson_distribution(double __mean = 1.0) : __p_(__mean) {}
76 #endif
77 _LIBCPP_HIDE_FROM_ABI explicit poisson_distribution(const param_type& __p) : __p_(__p) {}
78 _LIBCPP_HIDE_FROM_ABI void reset() {}
80 // generating functions
81 template <class _URNG>
82 _LIBCPP_HIDE_FROM_ABI result_type operator()(_URNG& __g) {
83 return (*this)(__g, __p_);
85 template <class _URNG>
86 _LIBCPP_HIDE_FROM_ABI result_type operator()(_URNG& __g, const param_type& __p);
88 // property functions
89 _LIBCPP_HIDE_FROM_ABI double mean() const { return __p_.mean(); }
91 _LIBCPP_HIDE_FROM_ABI param_type param() const { return __p_; }
92 _LIBCPP_HIDE_FROM_ABI void param(const param_type& __p) { __p_ = __p; }
94 _LIBCPP_HIDE_FROM_ABI result_type min() const { return 0; }
95 _LIBCPP_HIDE_FROM_ABI result_type max() const { return numeric_limits<result_type>::max(); }
97 friend _LIBCPP_HIDE_FROM_ABI bool operator==(const poisson_distribution& __x, const poisson_distribution& __y) {
98 return __x.__p_ == __y.__p_;
100 friend _LIBCPP_HIDE_FROM_ABI bool operator!=(const poisson_distribution& __x, const poisson_distribution& __y) {
101 return !(__x == __y);
105 template <class _IntType>
106 poisson_distribution<_IntType>::param_type::param_type(double __mean)
107 // According to the standard `inf` is a valid input, but it causes the
108 // distribution to hang, so we replace it with the maximum representable
109 // mean.
110 : __mean_(isinf(__mean) ? numeric_limits<double>::max() : __mean) {
111 if (__mean_ < 10) {
112 __s_ = 0;
113 __d_ = 0;
114 __l_ = std::exp(-__mean_);
115 __omega_ = 0;
116 __c3_ = 0;
117 __c2_ = 0;
118 __c1_ = 0;
119 __c0_ = 0;
120 __c_ = 0;
121 } else {
122 __s_ = std::sqrt(__mean_);
123 __d_ = 6 * __mean_ * __mean_;
124 __l_ = std::trunc(__mean_ - 1.1484);
125 __omega_ = .3989423 / __s_;
126 double __b1 = .4166667E-1 / __mean_;
127 double __b2 = .3 * __b1 * __b1;
128 __c3_ = .1428571 * __b1 * __b2;
129 __c2_ = __b2 - 15. * __c3_;
130 __c1_ = __b1 - 6. * __b2 + 45. * __c3_;
131 __c0_ = 1. - __b1 + 3. * __b2 - 15. * __c3_;
132 __c_ = .1069 / __mean_;
136 template <class _IntType>
137 template <class _URNG>
138 _IntType poisson_distribution<_IntType>::operator()(_URNG& __urng, const param_type& __pr) {
139 static_assert(__libcpp_random_is_valid_urng<_URNG>::value, "");
140 double __tx;
141 uniform_real_distribution<double> __urd;
142 if (__pr.__mean_ < 10) {
143 __tx = 0;
144 for (double __p = __urd(__urng); __p > __pr.__l_; ++__tx)
145 __p *= __urd(__urng);
146 } else {
147 double __difmuk;
148 double __g = __pr.__mean_ + __pr.__s_ * normal_distribution<double>()(__urng);
149 double __u;
150 if (__g > 0) {
151 __tx = std::trunc(__g);
152 if (__tx >= __pr.__l_)
153 return std::__clamp_to_integral<result_type>(__tx);
154 __difmuk = __pr.__mean_ - __tx;
155 __u = __urd(__urng);
156 if (__pr.__d_ * __u >= __difmuk * __difmuk * __difmuk)
157 return std::__clamp_to_integral<result_type>(__tx);
159 exponential_distribution<double> __edist;
160 for (bool __using_exp_dist = false; true; __using_exp_dist = true) {
161 double __e;
162 if (__using_exp_dist || __g <= 0) {
163 double __t;
164 do {
165 __e = __edist(__urng);
166 __u = __urd(__urng);
167 __u += __u - 1;
168 __t = 1.8 + (__u < 0 ? -__e : __e);
169 } while (__t <= -.6744);
170 __tx = std::trunc(__pr.__mean_ + __pr.__s_ * __t);
171 __difmuk = __pr.__mean_ - __tx;
172 __using_exp_dist = true;
174 double __px;
175 double __py;
176 if (__tx < 10 && __tx >= 0) {
177 const double __fac[] = {1, 1, 2, 6, 24, 120, 720, 5040, 40320, 362880};
178 __px = -__pr.__mean_;
179 __py = std::pow(__pr.__mean_, (double)__tx) / __fac[static_cast<int>(__tx)];
180 } else {
181 double __del = .8333333E-1 / __tx;
182 __del -= 4.8 * __del * __del * __del;
183 double __v = __difmuk / __tx;
184 if (std::abs(__v) > 0.25)
185 __px = __tx * std::log(1 + __v) - __difmuk - __del;
186 else
187 __px = __tx * __v * __v *
188 (((((((.1250060 * __v + -.1384794) * __v + .1421878) * __v + -.1661269) * __v + .2000118) * __v +
189 -.2500068) *
190 __v +
191 .3333333) *
192 __v +
193 -.5) -
194 __del;
195 __py = .3989423 / std::sqrt(__tx);
197 double __r = (0.5 - __difmuk) / __pr.__s_;
198 double __r2 = __r * __r;
199 double __fx = -0.5 * __r2;
200 double __fy = __pr.__omega_ * (((__pr.__c3_ * __r2 + __pr.__c2_) * __r2 + __pr.__c1_) * __r2 + __pr.__c0_);
201 if (__using_exp_dist) {
202 if (__pr.__c_ * std::abs(__u) <= __py * std::exp(__px + __e) - __fy * std::exp(__fx + __e))
203 break;
204 } else {
205 if (__fy - __u * __fy <= __py * std::exp(__px - __fx))
206 break;
210 return std::__clamp_to_integral<result_type>(__tx);
213 template <class _CharT, class _Traits, class _IntType>
214 _LIBCPP_HIDE_FROM_ABI basic_ostream<_CharT, _Traits>&
215 operator<<(basic_ostream<_CharT, _Traits>& __os, const poisson_distribution<_IntType>& __x) {
216 __save_flags<_CharT, _Traits> __lx(__os);
217 typedef basic_ostream<_CharT, _Traits> _OStream;
218 __os.flags(_OStream::dec | _OStream::left | _OStream::fixed | _OStream::scientific);
219 return __os << __x.mean();
222 template <class _CharT, class _Traits, class _IntType>
223 _LIBCPP_HIDE_FROM_ABI basic_istream<_CharT, _Traits>&
224 operator>>(basic_istream<_CharT, _Traits>& __is, poisson_distribution<_IntType>& __x) {
225 typedef poisson_distribution<_IntType> _Eng;
226 typedef typename _Eng::param_type param_type;
227 __save_flags<_CharT, _Traits> __lx(__is);
228 typedef basic_istream<_CharT, _Traits> _Istream;
229 __is.flags(_Istream::dec | _Istream::skipws);
230 double __mean;
231 __is >> __mean;
232 if (!__is.fail())
233 __x.param(param_type(__mean));
234 return __is;
237 _LIBCPP_END_NAMESPACE_STD
239 _LIBCPP_POP_MACROS
241 #endif // _LIBCPP___RANDOM_POISSON_DISTRIBUTION_H