xref: /aosp_15_r20/external/angle/third_party/abseil-cpp/absl/random/gaussian_distribution.h (revision 8975f5c5ed3d1c378011245431ada316dfb6f244)
1 // Copyright 2017 The Abseil Authors.
2 //
3 // Licensed under the Apache License, Version 2.0 (the "License");
4 // you may not use this file except in compliance with the License.
5 // You may obtain a copy of the License at
6 //
7 //      https://www.apache.org/licenses/LICENSE-2.0
8 //
9 // Unless required by applicable law or agreed to in writing, software
10 // distributed under the License is distributed on an "AS IS" BASIS,
11 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 // See the License for the specific language governing permissions and
13 // limitations under the License.
14 
15 #ifndef ABSL_RANDOM_GAUSSIAN_DISTRIBUTION_H_
16 #define ABSL_RANDOM_GAUSSIAN_DISTRIBUTION_H_
17 
18 // absl::gaussian_distribution implements the Ziggurat algorithm
19 // for generating random gaussian numbers.
20 //
21 // Implementation based on "The Ziggurat Method for Generating Random Variables"
22 // by George Marsaglia and Wai Wan Tsang: http://www.jstatsoft.org/v05/i08/
23 //
24 
25 #include <cmath>
26 #include <cstdint>
27 #include <istream>
28 #include <limits>
29 #include <ostream>
30 #include <type_traits>
31 
32 #include "absl/base/config.h"
33 #include "absl/random/internal/fast_uniform_bits.h"
34 #include "absl/random/internal/generate_real.h"
35 #include "absl/random/internal/iostream_state_saver.h"
36 
37 namespace absl {
38 ABSL_NAMESPACE_BEGIN
39 namespace random_internal {
40 
41 // absl::gaussian_distribution_base implements the underlying ziggurat algorithm
42 // using the ziggurat tables generated by the gaussian_distribution_gentables
43 // binary.
44 //
45 // The specific algorithm has some of the improvements suggested by the
46 // 2005 paper, "An Improved Ziggurat Method to Generate Normal Random Samples",
47 // Jurgen A Doornik.  (https://www.doornik.com/research/ziggurat.pdf)
48 class ABSL_DLL gaussian_distribution_base {
49  public:
50   template <typename URBG>
51   inline double zignor(URBG& g);  // NOLINT(runtime/references)
52 
53  private:
54   friend class TableGenerator;
55 
56   template <typename URBG>
57   inline double zignor_fallback(URBG& g,  // NOLINT(runtime/references)
58                                 bool neg);
59 
60   // Constants used for the gaussian distribution.
61   static constexpr double kR = 3.442619855899;          // Start of the tail.
62   static constexpr double kRInv = 0.29047645161474317;  // ~= (1.0 / kR) .
63   static constexpr double kV = 9.91256303526217e-3;
64   static constexpr uint64_t kMask = 0x07f;
65 
66   // The ziggurat tables store the pdf(f) and inverse-pdf(x) for equal-area
67   // points on one-half of the normal distribution, where the pdf function,
68   // pdf = e ^ (-1/2 *x^2), assumes that the mean = 0 & stddev = 1.
69   //
70   // These tables are just over 2kb in size; larger tables might improve the
71   // distributions, but also lead to more cache pollution.
72   //
73   // x = {3.71308, 3.44261, 3.22308, ..., 0}
74   // f = {0.00101, 0.00266, 0.00554, ..., 1}
75   struct Tables {
76     double x[kMask + 2];
77     double f[kMask + 2];
78   };
79   static const Tables zg_;
80   random_internal::FastUniformBits<uint64_t> fast_u64_;
81 };
82 
83 }  // namespace random_internal
84 
85 // absl::gaussian_distribution:
86 // Generates a number conforming to a Gaussian distribution.
87 template <typename RealType = double>
88 class gaussian_distribution : random_internal::gaussian_distribution_base {
89  public:
90   using result_type = RealType;
91 
92   class param_type {
93    public:
94     using distribution_type = gaussian_distribution;
95 
96     explicit param_type(result_type mean = 0, result_type stddev = 1)
mean_(mean)97         : mean_(mean), stddev_(stddev) {}
98 
99     // Returns the mean distribution parameter.  The mean specifies the location
100     // of the peak.  The default value is 0.0.
mean()101     result_type mean() const { return mean_; }
102 
103     // Returns the deviation distribution parameter.  The default value is 1.0.
stddev()104     result_type stddev() const { return stddev_; }
105 
106     friend bool operator==(const param_type& a, const param_type& b) {
107       return a.mean_ == b.mean_ && a.stddev_ == b.stddev_;
108     }
109 
110     friend bool operator!=(const param_type& a, const param_type& b) {
111       return !(a == b);
112     }
113 
114    private:
115     result_type mean_;
116     result_type stddev_;
117 
118     static_assert(
119         std::is_floating_point<RealType>::value,
120         "Class-template absl::gaussian_distribution<> must be parameterized "
121         "using a floating-point type.");
122   };
123 
gaussian_distribution()124   gaussian_distribution() : gaussian_distribution(0) {}
125 
126   explicit gaussian_distribution(result_type mean, result_type stddev = 1)
param_(mean,stddev)127       : param_(mean, stddev) {}
128 
gaussian_distribution(const param_type & p)129   explicit gaussian_distribution(const param_type& p) : param_(p) {}
130 
reset()131   void reset() {}
132 
133   // Generating functions
134   template <typename URBG>
operator()135   result_type operator()(URBG& g) {  // NOLINT(runtime/references)
136     return (*this)(g, param_);
137   }
138 
139   template <typename URBG>
140   result_type operator()(URBG& g,  // NOLINT(runtime/references)
141                          const param_type& p);
142 
param()143   param_type param() const { return param_; }
param(const param_type & p)144   void param(const param_type& p) { param_ = p; }
145 
result_type(min)146   result_type(min)() const {
147     return -std::numeric_limits<result_type>::infinity();
148   }
result_type(max)149   result_type(max)() const {
150     return std::numeric_limits<result_type>::infinity();
151   }
152 
mean()153   result_type mean() const { return param_.mean(); }
stddev()154   result_type stddev() const { return param_.stddev(); }
155 
156   friend bool operator==(const gaussian_distribution& a,
157                          const gaussian_distribution& b) {
158     return a.param_ == b.param_;
159   }
160   friend bool operator!=(const gaussian_distribution& a,
161                          const gaussian_distribution& b) {
162     return a.param_ != b.param_;
163   }
164 
165  private:
166   param_type param_;
167 };
168 
169 // --------------------------------------------------------------------------
170 // Implementation details only below
171 // --------------------------------------------------------------------------
172 
173 template <typename RealType>
174 template <typename URBG>
175 typename gaussian_distribution<RealType>::result_type
operator()176 gaussian_distribution<RealType>::operator()(
177     URBG& g,  // NOLINT(runtime/references)
178     const param_type& p) {
179   return p.mean() + p.stddev() * static_cast<result_type>(zignor(g));
180 }
181 
182 template <typename CharT, typename Traits, typename RealType>
183 std::basic_ostream<CharT, Traits>& operator<<(
184     std::basic_ostream<CharT, Traits>& os,  // NOLINT(runtime/references)
185     const gaussian_distribution<RealType>& x) {
186   auto saver = random_internal::make_ostream_state_saver(os);
187   os.precision(random_internal::stream_precision_helper<RealType>::kPrecision);
188   os << x.mean() << os.fill() << x.stddev();
189   return os;
190 }
191 
192 template <typename CharT, typename Traits, typename RealType>
193 std::basic_istream<CharT, Traits>& operator>>(
194     std::basic_istream<CharT, Traits>& is,  // NOLINT(runtime/references)
195     gaussian_distribution<RealType>& x) {   // NOLINT(runtime/references)
196   using result_type = typename gaussian_distribution<RealType>::result_type;
197   using param_type = typename gaussian_distribution<RealType>::param_type;
198 
199   auto saver = random_internal::make_istream_state_saver(is);
200   auto mean = random_internal::read_floating_point<result_type>(is);
201   if (is.fail()) return is;
202   auto stddev = random_internal::read_floating_point<result_type>(is);
203   if (!is.fail()) {
204     x.param(param_type(mean, stddev));
205   }
206   return is;
207 }
208 
209 namespace random_internal {
210 
211 template <typename URBG>
zignor_fallback(URBG & g,bool neg)212 inline double gaussian_distribution_base::zignor_fallback(URBG& g, bool neg) {
213   using random_internal::GeneratePositiveTag;
214   using random_internal::GenerateRealFromBits;
215 
216   // This fallback path happens approximately 0.05% of the time.
217   double x, y;
218   do {
219     // kRInv = 1/r, U(0, 1)
220     x = kRInv *
221         std::log(GenerateRealFromBits<double, GeneratePositiveTag, false>(
222             fast_u64_(g)));
223     y = -std::log(
224         GenerateRealFromBits<double, GeneratePositiveTag, false>(fast_u64_(g)));
225   } while ((y + y) < (x * x));
226   return neg ? (x - kR) : (kR - x);
227 }
228 
229 template <typename URBG>
zignor(URBG & g)230 inline double gaussian_distribution_base::zignor(
231     URBG& g) {  // NOLINT(runtime/references)
232   using random_internal::GeneratePositiveTag;
233   using random_internal::GenerateRealFromBits;
234   using random_internal::GenerateSignedTag;
235 
236   while (true) {
237     // We use a single uint64_t to generate both a double and a strip.
238     // These bits are unused when the generated double is > 1/2^5.
239     // This may introduce some bias from the duplicated low bits of small
240     // values (those smaller than 1/2^5, which all end up on the left tail).
241     uint64_t bits = fast_u64_(g);
242     int i = static_cast<int>(bits & kMask);  // pick a random strip
243     double j = GenerateRealFromBits<double, GenerateSignedTag, false>(
244         bits);  // U(-1, 1)
245     const double x = j * zg_.x[i];
246 
247     // Retangular box. Handles >97% of all cases.
248     // For any given box, this handles between 75% and 99% of values.
249     // Equivalent to U(01) < (x[i+1] / x[i]), and when i == 0, ~93.5%
250     if (std::abs(x) < zg_.x[i + 1]) {
251       return x;
252     }
253 
254     // i == 0: Base box. Sample using a ratio of uniforms.
255     if (i == 0) {
256       // This path happens about 0.05% of the time.
257       return zignor_fallback(g, j < 0);
258     }
259 
260     // i > 0: Wedge samples using precomputed values.
261     double v = GenerateRealFromBits<double, GeneratePositiveTag, false>(
262         fast_u64_(g));  // U(0, 1)
263     if ((zg_.f[i + 1] + v * (zg_.f[i] - zg_.f[i + 1])) <
264         std::exp(-0.5 * x * x)) {
265       return x;
266     }
267 
268     // The wedge was missed; reject the value and try again.
269   }
270 }
271 
272 }  // namespace random_internal
273 ABSL_NAMESPACE_END
274 }  // namespace absl
275 
276 #endif  // ABSL_RANDOM_GAUSSIAN_DISTRIBUTION_H_
277