| 1 |
// (C) Copyright John Maddock 2006. |
| 2 |
// (C) Copyright Matt Borland 2024. |
| 3 |
// Use, modification and distribution are subject to the |
| 4 |
// Boost Software License, Version 1.0. (See accompanying file |
| 5 |
// LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt) |
| 6 |
|
| 7 |
#ifndef BOOST_MATH_SF_ERF_INV_HPP |
| 8 |
#define BOOST_MATH_SF_ERF_INV_HPP |
| 9 |
|
| 10 |
#ifdef _MSC_VER |
| 11 |
#pragma once |
| 12 |
#pragma warning(push) |
| 13 |
#pragma warning(disable:4127) // Conditional expression is constant |
| 14 |
#pragma warning(disable:4702) // Unreachable code: optimization warning |
| 15 |
#endif |
| 16 |
|
| 17 |
#include <boost/math/tools/config.hpp> |
| 18 |
|
| 19 |
#ifndef BOOST_MATH_HAS_NVRTC |
| 20 |
|
| 21 |
#include <type_traits> |
| 22 |
|
| 23 |
namespace boost{ namespace math{ |
| 24 |
|
| 25 |
namespace detail{ |
| 26 |
// |
| 27 |
// The inverse erf and erfc functions share a common implementation, |
| 28 |
// this version is for 80-bit long double's and smaller: |
| 29 |
// |
| 30 |
template <class T, class Policy> |
| 31 |
BOOST_MATH_GPU_ENABLED T erf_inv_imp(const T& p, const T& q, const Policy&, const std::integral_constant<int, 64>&) |
| 32 |
{ |
| 33 |
BOOST_MATH_STD_USING // for ADL of std names. |
| 34 |
|
| 35 |
T result = 0; |
| 36 |
|
| 37 |
if(p <= 0.5) |
| 38 |
{ |
| 39 |
// |
| 40 |
// Evaluate inverse erf using the rational approximation: |
| 41 |
// |
| 42 |
// x = p(p+10)(Y+R(p)) |
| 43 |
// |
| 44 |
// Where Y is a constant, and R(p) is optimised for a low |
| 45 |
// absolute error compared to |Y|. |
| 46 |
// |
| 47 |
// double: Max error found: 2.001849e-18 |
| 48 |
// long double: Max error found: 1.017064e-20 |
| 49 |
// Maximum Deviation Found (actual error term at infinite precision) 8.030e-21 |
| 50 |
// |
| 51 |
// LCOV_EXCL_START |
| 52 |
BOOST_MATH_STATIC_LOCAL_VARIABLE const float Y = 0.0891314744949340820313f; |
| 53 |
BOOST_MATH_STATIC const T P[] = { |
| 54 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.000508781949658280665617), |
| 55 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.00836874819741736770379), |
| 56 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.0334806625409744615033), |
| 57 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.0126926147662974029034), |
| 58 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.0365637971411762664006), |
| 59 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.0219878681111168899165), |
| 60 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.00822687874676915743155), |
| 61 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.00538772965071242932965) |
| 62 |
}; |
| 63 |
BOOST_MATH_STATIC const T Q[] = { |
| 64 |
BOOST_MATH_BIG_CONSTANT(T, 64, 1.0), |
| 65 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.970005043303290640362), |
| 66 |
BOOST_MATH_BIG_CONSTANT(T, 64, -1.56574558234175846809), |
| 67 |
BOOST_MATH_BIG_CONSTANT(T, 64, 1.56221558398423026363), |
| 68 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.662328840472002992063), |
| 69 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.71228902341542847553), |
| 70 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.0527396382340099713954), |
| 71 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.0795283687341571680018), |
| 72 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.00233393759374190016776), |
| 73 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.000886216390456424707504) |
| 74 |
}; |
| 75 |
// LCOV_EXCL_STOP |
| 76 |
T g = p * (p + 10); |
| 77 |
T r = tools::evaluate_polynomial(P, p) / tools::evaluate_polynomial(Q, p); |
| 78 |
result = g * Y + g * r; |
| 79 |
} |
| 80 |
else if(q >= 0.25) |
| 81 |
{ |
| 82 |
// |
| 83 |
// Rational approximation for 0.5 > q >= 0.25 |
| 84 |
// |
| 85 |
// x = sqrt(-2*log(q)) / (Y + R(q)) |
| 86 |
// |
| 87 |
// Where Y is a constant, and R(q) is optimised for a low |
| 88 |
// absolute error compared to Y. |
| 89 |
// |
| 90 |
// double : Max error found: 7.403372e-17 |
| 91 |
// long double : Max error found: 6.084616e-20 |
| 92 |
// Maximum Deviation Found (error term) 4.811e-20 |
| 93 |
// |
| 94 |
// LCOV_EXCL_START |
| 95 |
BOOST_MATH_STATIC_LOCAL_VARIABLE const float Y = 2.249481201171875f; |
| 96 |
BOOST_MATH_STATIC const T P[] = { |
| 97 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.202433508355938759655), |
| 98 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.105264680699391713268), |
| 99 |
BOOST_MATH_BIG_CONSTANT(T, 64, 8.37050328343119927838), |
| 100 |
BOOST_MATH_BIG_CONSTANT(T, 64, 17.6447298408374015486), |
| 101 |
BOOST_MATH_BIG_CONSTANT(T, 64, -18.8510648058714251895), |
| 102 |
BOOST_MATH_BIG_CONSTANT(T, 64, -44.6382324441786960818), |
| 103 |
BOOST_MATH_BIG_CONSTANT(T, 64, 17.445385985570866523), |
| 104 |
BOOST_MATH_BIG_CONSTANT(T, 64, 21.1294655448340526258), |
| 105 |
BOOST_MATH_BIG_CONSTANT(T, 64, -3.67192254707729348546) |
| 106 |
}; |
| 107 |
BOOST_MATH_STATIC const T Q[] = { |
| 108 |
BOOST_MATH_BIG_CONSTANT(T, 64, 1.0), |
| 109 |
BOOST_MATH_BIG_CONSTANT(T, 64, 6.24264124854247537712), |
| 110 |
BOOST_MATH_BIG_CONSTANT(T, 64, 3.9713437953343869095), |
| 111 |
BOOST_MATH_BIG_CONSTANT(T, 64, -28.6608180499800029974), |
| 112 |
BOOST_MATH_BIG_CONSTANT(T, 64, -20.1432634680485188801), |
| 113 |
BOOST_MATH_BIG_CONSTANT(T, 64, 48.5609213108739935468), |
| 114 |
BOOST_MATH_BIG_CONSTANT(T, 64, 10.8268667355460159008), |
| 115 |
BOOST_MATH_BIG_CONSTANT(T, 64, -22.6436933413139721736), |
| 116 |
BOOST_MATH_BIG_CONSTANT(T, 64, 1.72114765761200282724) |
| 117 |
}; |
| 118 |
// LCOV_EXCL_STOP |
| 119 |
T g = sqrt(-2 * log(q)); |
| 120 |
T xs = q - 0.25f; |
| 121 |
T r = tools::evaluate_polynomial(P, xs) / tools::evaluate_polynomial(Q, xs); |
| 122 |
result = g / (Y + r); |
| 123 |
} |
| 124 |
else |
| 125 |
{ |
| 126 |
// |
| 127 |
// For q < 0.25 we have a series of rational approximations all |
| 128 |
// of the general form: |
| 129 |
// |
| 130 |
// let: x = sqrt(-log(q)) |
| 131 |
// |
| 132 |
// Then the result is given by: |
| 133 |
// |
| 134 |
// x(Y+R(x-B)) |
| 135 |
// |
| 136 |
// where Y is a constant, B is the lowest value of x for which |
| 137 |
// the approximation is valid, and R(x-B) is optimised for a low |
| 138 |
// absolute error compared to Y. |
| 139 |
// |
| 140 |
// Note that almost all code will really go through the first |
| 141 |
// or maybe second approximation. After than we're dealing with very |
| 142 |
// small input values indeed: 80 and 128 bit long double's go all the |
| 143 |
// way down to ~ 1e-5000 so the "tail" is rather long... |
| 144 |
// |
| 145 |
T x = sqrt(-log(q)); |
| 146 |
if(x < 3) |
| 147 |
{ |
| 148 |
// LCOV_EXCL_START |
| 149 |
// Max error found: 1.089051e-20 |
| 150 |
BOOST_MATH_STATIC_LOCAL_VARIABLE const float Y = 0.807220458984375f; |
| 151 |
BOOST_MATH_STATIC const T P[] = { |
| 152 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.131102781679951906451), |
| 153 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.163794047193317060787), |
| 154 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.117030156341995252019), |
| 155 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.387079738972604337464), |
| 156 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.337785538912035898924), |
| 157 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.142869534408157156766), |
| 158 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.0290157910005329060432), |
| 159 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.00214558995388805277169), |
| 160 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.679465575181126350155e-6), |
| 161 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.285225331782217055858e-7), |
| 162 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.681149956853776992068e-9) |
| 163 |
}; |
| 164 |
BOOST_MATH_STATIC const T Q[] = { |
| 165 |
BOOST_MATH_BIG_CONSTANT(T, 64, 1.0), |
| 166 |
BOOST_MATH_BIG_CONSTANT(T, 64, 3.46625407242567245975), |
| 167 |
BOOST_MATH_BIG_CONSTANT(T, 64, 5.38168345707006855425), |
| 168 |
BOOST_MATH_BIG_CONSTANT(T, 64, 4.77846592945843778382), |
| 169 |
BOOST_MATH_BIG_CONSTANT(T, 64, 2.59301921623620271374), |
| 170 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.848854343457902036425), |
| 171 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.152264338295331783612), |
| 172 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.01105924229346489121) |
| 173 |
}; |
| 174 |
// LCOV_EXCL_STOP |
| 175 |
T xs = x - 1.125f; |
| 176 |
T R = tools::evaluate_polynomial(P, xs) / tools::evaluate_polynomial(Q, xs); |
| 177 |
result = Y * x + R * x; |
| 178 |
} |
| 179 |
else if(x < 6) |
| 180 |
{ |
| 181 |
// LCOV_EXCL_START |
| 182 |
// Max error found: 8.389174e-21 |
| 183 |
BOOST_MATH_STATIC_LOCAL_VARIABLE const float Y = 0.93995571136474609375f; |
| 184 |
BOOST_MATH_STATIC const T P[] = { |
| 185 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.0350353787183177984712), |
| 186 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.00222426529213447927281), |
| 187 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.0185573306514231072324), |
| 188 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.00950804701325919603619), |
| 189 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.00187123492819559223345), |
| 190 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.000157544617424960554631), |
| 191 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.460469890584317994083e-5), |
| 192 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.230404776911882601748e-9), |
| 193 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.266339227425782031962e-11) |
| 194 |
}; |
| 195 |
BOOST_MATH_STATIC const T Q[] = { |
| 196 |
BOOST_MATH_BIG_CONSTANT(T, 64, 1.0), |
| 197 |
BOOST_MATH_BIG_CONSTANT(T, 64, 1.3653349817554063097), |
| 198 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.762059164553623404043), |
| 199 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.220091105764131249824), |
| 200 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.0341589143670947727934), |
| 201 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.00263861676657015992959), |
| 202 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.764675292302794483503e-4) |
| 203 |
}; |
| 204 |
// LCOV_EXCL_STOP |
| 205 |
T xs = x - 3; |
| 206 |
T R = tools::evaluate_polynomial(P, xs) / tools::evaluate_polynomial(Q, xs); |
| 207 |
result = Y * x + R * x; |
| 208 |
} |
| 209 |
else if(x < 18) |
| 210 |
{ |
| 211 |
// LCOV_EXCL_START |
| 212 |
// Max error found: 1.481312e-19 |
| 213 |
BOOST_MATH_STATIC_LOCAL_VARIABLE const float Y = 0.98362827301025390625f; |
| 214 |
BOOST_MATH_STATIC const T P[] = { |
| 215 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.0167431005076633737133), |
| 216 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.00112951438745580278863), |
| 217 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.00105628862152492910091), |
| 218 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.000209386317487588078668), |
| 219 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.149624783758342370182e-4), |
| 220 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.449696789927706453732e-6), |
| 221 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.462596163522878599135e-8), |
| 222 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.281128735628831791805e-13), |
| 223 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.99055709973310326855e-16) |
| 224 |
}; |
| 225 |
BOOST_MATH_STATIC const T Q[] = { |
| 226 |
BOOST_MATH_BIG_CONSTANT(T, 64, 1.0), |
| 227 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.591429344886417493481), |
| 228 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.138151865749083321638), |
| 229 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.0160746087093676504695), |
| 230 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.000964011807005165528527), |
| 231 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.275335474764726041141e-4), |
| 232 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.282243172016108031869e-6) |
| 233 |
}; |
| 234 |
// LCOV_EXCL_STOP |
| 235 |
T xs = x - 6; |
| 236 |
T R = tools::evaluate_polynomial(P, xs) / tools::evaluate_polynomial(Q, xs); |
| 237 |
result = Y * x + R * x; |
| 238 |
} |
| 239 |
else if(x < 44) |
| 240 |
{ |
| 241 |
// LCOV_EXCL_START |
| 242 |
// Max error found: 5.697761e-20 |
| 243 |
BOOST_MATH_STATIC_LOCAL_VARIABLE const float Y = 0.99714565277099609375f; |
| 244 |
BOOST_MATH_STATIC const T P[] = { |
| 245 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.0024978212791898131227), |
| 246 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.779190719229053954292e-5), |
| 247 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.254723037413027451751e-4), |
| 248 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.162397777342510920873e-5), |
| 249 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.396341011304801168516e-7), |
| 250 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.411632831190944208473e-9), |
| 251 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.145596286718675035587e-11), |
| 252 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.116765012397184275695e-17) |
| 253 |
}; |
| 254 |
BOOST_MATH_STATIC const T Q[] = { |
| 255 |
BOOST_MATH_BIG_CONSTANT(T, 64, 1.0), |
| 256 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.207123112214422517181), |
| 257 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.0169410838120975906478), |
| 258 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.000690538265622684595676), |
| 259 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.145007359818232637924e-4), |
| 260 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.144437756628144157666e-6), |
| 261 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.509761276599778486139e-9) |
| 262 |
}; |
| 263 |
// LCOV_EXCL_STOP |
| 264 |
T xs = x - 18; |
| 265 |
T R = tools::evaluate_polynomial(P, xs) / tools::evaluate_polynomial(Q, xs); |
| 266 |
result = Y * x + R * x; |
| 267 |
} |
| 268 |
else |
| 269 |
{ |
| 270 |
// LCOV_EXCL_START |
| 271 |
// Max error found: 1.279746e-20 |
| 272 |
BOOST_MATH_STATIC_LOCAL_VARIABLE const float Y = 0.99941349029541015625f; |
| 273 |
BOOST_MATH_STATIC const T P[] = { |
| 274 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.000539042911019078575891), |
| 275 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.28398759004727721098e-6), |
| 276 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.899465114892291446442e-6), |
| 277 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.229345859265920864296e-7), |
| 278 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.225561444863500149219e-9), |
| 279 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.947846627503022684216e-12), |
| 280 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.135880130108924861008e-14), |
| 281 |
BOOST_MATH_BIG_CONSTANT(T, 64, -0.348890393399948882918e-21) |
| 282 |
}; |
| 283 |
BOOST_MATH_STATIC const T Q[] = { |
| 284 |
BOOST_MATH_BIG_CONSTANT(T, 64, 1.0), |
| 285 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.0845746234001899436914), |
| 286 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.00282092984726264681981), |
| 287 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.468292921940894236786e-4), |
| 288 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.399968812193862100054e-6), |
| 289 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.161809290887904476097e-8), |
| 290 |
BOOST_MATH_BIG_CONSTANT(T, 64, 0.231558608310259605225e-11) |
| 291 |
}; |
| 292 |
// LCOV_EXCL_STOP |
| 293 |
T xs = x - 44; |
| 294 |
T R = tools::evaluate_polynomial(P, xs) / tools::evaluate_polynomial(Q, xs); |
| 295 |
result = Y * x + R * x; |
| 296 |
} |
| 297 |
} |
| 298 |
return result; |
| 299 |
} |
| 300 |
|
| 301 |
template <class T, class Policy> |
| 302 |
struct erf_roots |
| 303 |
{ |
| 304 |
boost::math::tuple<T,T,T> operator()(const T& guess) |
| 305 |
{ |
| 306 |
BOOST_MATH_STD_USING |
| 307 |
T derivative = sign * (2 / sqrt(constants::pi<T>())) * exp(-(guess * guess)); |
| 308 |
T derivative2 = -2 * guess * derivative; |
| 309 |
return boost::math::make_tuple(((sign > 0) ? static_cast<T>(boost::math::erf(guess, Policy()) - target) : static_cast<T>(boost::math::erfc(guess, Policy())) - target), derivative, derivative2); |
| 310 |
} |
| 311 |
erf_roots(T z, int s) : target(z), sign(s) {} |
| 312 |
private: |
| 313 |
T target; |
| 314 |
int sign; |
| 315 |
}; |
| 316 |
|
| 317 |
template <class T, class Policy> |
| 318 |
T erf_inv_imp(const T& p, const T& q, const Policy& pol, const std::integral_constant<int, 0>&) |
| 319 |
{ |
| 320 |
// |
| 321 |
// Generic version, get a guess that's accurate to 64-bits (10^-19) |
| 322 |
// |
| 323 |
using tag_type = std::integral_constant<int, 64>; |
| 324 |
T guess = erf_inv_imp(p, q, pol, tag_type()); |
| 325 |
T result; |
| 326 |
// |
| 327 |
// If T has more bit's than 64 in it's mantissa then we need to iterate, |
| 328 |
// otherwise we can just return the result: |
| 329 |
// |
| 330 |
if(policies::digits<T, Policy>() > 64) |
| 331 |
{ |
| 332 |
std::uintmax_t max_iter = policies::get_max_root_iterations<Policy>(); |
| 333 |
if(p <= 0.5) |
| 334 |
{ |
| 335 |
result = tools::halley_iterate(detail::erf_roots<typename std::remove_cv<T>::type, Policy>(p, 1), guess, static_cast<T>(0), tools::max_value<T>(), (policies::digits<T, Policy>() * 2) / 3, max_iter); |
| 336 |
} |
| 337 |
else |
| 338 |
{ |
| 339 |
result = tools::halley_iterate(detail::erf_roots<typename std::remove_cv<T>::type, Policy>(q, -1), guess, static_cast<T>(0), tools::max_value<T>(), (policies::digits<T, Policy>() * 2) / 3, max_iter); |
| 340 |
} |
| 341 |
policies::check_root_iterations<T>("boost::math::erf_inv<%1%>", max_iter, pol); |
| 342 |
} |
| 343 |
else |
| 344 |
{ |
| 345 |
result = guess; |
| 346 |
} |
| 347 |
return result; |
| 348 |
} |
| 349 |
|
| 350 |
} // namespace detail |
| 351 |
|
| 352 |
template <class T, class Policy> |
| 353 |
BOOST_MATH_GPU_ENABLED typename tools::promote_args<T>::type erfc_inv(T z, const Policy& pol) |
| 354 |
{ |
| 355 |
typedef typename tools::promote_args<T>::type result_type; |
| 356 |
|
| 357 |
// |
| 358 |
// Begin by testing for domain errors, and other special cases: |
| 359 |
// |
| 360 |
constexpr auto function = "boost::math::erfc_inv<%1%>(%1%, %1%)"; |
| 361 |
if((z < 0) || (z > 2)) |
| 362 |
return policies::raise_domain_error<result_type>(function, "Argument outside range [0,2] in inverse erfc function (got p=%1%).", z, pol); |
| 363 |
if(z == 0) |
| 364 |
return policies::raise_overflow_error<result_type>(function, nullptr, pol); |
| 365 |
if(z == 2) |
| 366 |
return -policies::raise_overflow_error<result_type>(function, nullptr, pol); |
| 367 |
// |
| 368 |
// Normalise the input, so it's in the range [0,1], we will |
| 369 |
// negate the result if z is outside that range. This is a simple |
| 370 |
// application of the erfc reflection formula: erfc(-z) = 2 - erfc(z) |
| 371 |
// |
| 372 |
result_type p, q, s; |
| 373 |
if(z > 1) |
| 374 |
{ |
| 375 |
q = 2 - z; |
| 376 |
p = 1 - q; |
| 377 |
s = -1; |
| 378 |
} |
| 379 |
else |
| 380 |
{ |
| 381 |
p = 1 - z; |
| 382 |
q = z; |
| 383 |
s = 1; |
| 384 |
} |
| 385 |
// |
| 386 |
// A bit of meta-programming to figure out which implementation |
| 387 |
// to use, based on the number of bits in the mantissa of T: |
| 388 |
// |
| 389 |
typedef typename policies::precision<result_type, Policy>::type precision_type; |
| 390 |
typedef std::integral_constant<int, |
| 391 |
precision_type::value <= 0 ? 0 : |
| 392 |
precision_type::value <= 64 ? 64 : 0 |
| 393 |
> tag_type; |
| 394 |
// |
| 395 |
// Likewise use internal promotion, so we evaluate at a higher |
| 396 |
// precision internally if it's appropriate: |
| 397 |
// |
| 398 |
typedef typename policies::evaluation<result_type, Policy>::type eval_type; |
| 399 |
typedef typename policies::normalise< |
| 400 |
Policy, |
| 401 |
policies::promote_float<false>, |
| 402 |
policies::promote_double<false>, |
| 403 |
policies::discrete_quantile<>, |
| 404 |
policies::assert_undefined<> >::type forwarding_policy; |
| 405 |
|
| 406 |
// |
| 407 |
// And get the result, negating where required: |
| 408 |
// |
| 409 |
return s * policies::checked_narrowing_cast<result_type, forwarding_policy>( |
| 410 |
detail::erf_inv_imp(static_cast<eval_type>(p), static_cast<eval_type>(q), forwarding_policy(), tag_type()), function); |
| 411 |
} |
| 412 |
|
| 413 |
template <class T, class Policy> |
| 414 |
BOOST_MATH_GPU_ENABLED typename tools::promote_args<T>::type erf_inv(T z, const Policy& pol) |
| 415 |
{ |
| 416 |
typedef typename tools::promote_args<T>::type result_type; |
| 417 |
|
| 418 |
// |
| 419 |
// Begin by testing for domain errors, and other special cases: |
| 420 |
// |
| 421 |
constexpr auto function = "boost::math::erf_inv<%1%>(%1%, %1%)"; |
| 422 |
if((z < -1) || (z > 1)) |
| 423 |
return policies::raise_domain_error<result_type>(function, "Argument outside range [-1, 1] in inverse erf function (got p=%1%).", z, pol); |
| 424 |
if(z == 1) |
| 425 |
return policies::raise_overflow_error<result_type>(function, nullptr, pol); |
| 426 |
if(z == -1) |
| 427 |
return -policies::raise_overflow_error<result_type>(function, nullptr, pol); |
| 428 |
if(z == 0) |
| 429 |
return 0; |
| 430 |
// |
| 431 |
// Normalise the input, so it's in the range [0,1], we will |
| 432 |
// negate the result if z is outside that range. This is a simple |
| 433 |
// application of the erf reflection formula: erf(-z) = -erf(z) |
| 434 |
// |
| 435 |
result_type p, q, s; |
| 436 |
if(z < 0) |
| 437 |
{ |
| 438 |
p = -z; |
| 439 |
q = 1 - p; |
| 440 |
s = -1; |
| 441 |
} |
| 442 |
else |
| 443 |
{ |
| 444 |
p = z; |
| 445 |
q = 1 - z; |
| 446 |
s = 1; |
| 447 |
} |
| 448 |
// |
| 449 |
// A bit of meta-programming to figure out which implementation |
| 450 |
// to use, based on the number of bits in the mantissa of T: |
| 451 |
// |
| 452 |
typedef typename policies::precision<result_type, Policy>::type precision_type; |
| 453 |
typedef std::integral_constant<int, |
| 454 |
precision_type::value <= 0 ? 0 : |
| 455 |
precision_type::value <= 64 ? 64 : 0 |
| 456 |
> tag_type; |
| 457 |
// |
| 458 |
// Likewise use internal promotion, so we evaluate at a higher |
| 459 |
// precision internally if it's appropriate: |
| 460 |
// |
| 461 |
typedef typename policies::evaluation<result_type, Policy>::type eval_type; |
| 462 |
typedef typename policies::normalise< |
| 463 |
Policy, |
| 464 |
policies::promote_float<false>, |
| 465 |
policies::promote_double<false>, |
| 466 |
policies::discrete_quantile<>, |
| 467 |
policies::assert_undefined<> >::type forwarding_policy; |
| 468 |
// |
| 469 |
// Likewise use internal promotion, so we evaluate at a higher |
| 470 |
// precision internally if it's appropriate: |
| 471 |
// |
| 472 |
typedef typename policies::evaluation<result_type, Policy>::type eval_type; |
| 473 |
|
| 474 |
// |
| 475 |
// And get the result, negating where required: |
| 476 |
// |
| 477 |
return s * policies::checked_narrowing_cast<result_type, forwarding_policy>( |
| 478 |
detail::erf_inv_imp(static_cast<eval_type>(p), static_cast<eval_type>(q), forwarding_policy(), tag_type()), function); |
| 479 |
} |
| 480 |
|
| 481 |
template <class T> |
| 482 |
BOOST_MATH_GPU_ENABLED inline typename tools::promote_args<T>::type erfc_inv(T z) |
| 483 |
{ |
| 484 |
return erfc_inv(z, policies::policy<>()); |
| 485 |
} |
| 486 |
|
| 487 |
template <class T> |
| 488 |
BOOST_MATH_GPU_ENABLED inline typename tools::promote_args<T>::type erf_inv(T z) |
| 489 |
{ |
| 490 |
return erf_inv(z, policies::policy<>()); |
| 491 |
} |
| 492 |
|
| 493 |
} // namespace math |
| 494 |
} // namespace boost |
| 495 |
|
| 496 |
#else // Special handling for NVRTC |
| 497 |
|
| 498 |
namespace boost { |
| 499 |
namespace math { |
| 500 |
|
| 501 |
template <typename T> |
| 502 |
BOOST_MATH_GPU_ENABLED auto erf_inv(T x) |
| 503 |
{ |
| 504 |
return ::erfinv(x); |
| 505 |
} |
| 506 |
|
| 507 |
template <> |
| 508 |
BOOST_MATH_GPU_ENABLED auto erf_inv(float x) |
| 509 |
{ |
| 510 |
return ::erfinvf(x); |
| 511 |
} |
| 512 |
|
| 513 |
template <typename T, typename Policy> |
| 514 |
BOOST_MATH_GPU_ENABLED auto erf_inv(T x, const Policy&) |
| 515 |
{ |
| 516 |
return ::erfinv(x); |
| 517 |
} |
| 518 |
|
| 519 |
template <typename Policy> |
| 520 |
BOOST_MATH_GPU_ENABLED auto erf_inv(float x, const Policy&) |
| 521 |
{ |
| 522 |
return ::erfinvf(x); |
| 523 |
} |
| 524 |
|
| 525 |
template <typename T> |
| 526 |
BOOST_MATH_GPU_ENABLED auto erfc_inv(T x) |
| 527 |
{ |
| 528 |
return ::erfcinv(x); |
| 529 |
} |
| 530 |
|
| 531 |
template <> |
| 532 |
BOOST_MATH_GPU_ENABLED auto erfc_inv(float x) |
| 533 |
{ |
| 534 |
return ::erfcinvf(x); |
| 535 |
} |
| 536 |
|
| 537 |
template <typename T, typename Policy> |
| 538 |
BOOST_MATH_GPU_ENABLED auto erfc_inv(T x, const Policy&) |
| 539 |
{ |
| 540 |
return ::erfcinv(x); |
| 541 |
} |
| 542 |
|
| 543 |
template <typename Policy> |
| 544 |
BOOST_MATH_GPU_ENABLED auto erfc_inv(float x, const Policy&) |
| 545 |
{ |
| 546 |
return ::erfcinvf(x); |
| 547 |
} |
| 548 |
|
| 549 |
} // namespace math |
| 550 |
} // namespace boost |
| 551 |
|
| 552 |
#endif // BOOST_MATH_HAS_NVRTV |
| 553 |
|
| 554 |
#ifdef _MSC_VER |
| 555 |
#pragma warning(pop) |
| 556 |
#endif |
| 557 |
|
| 558 |
#endif // BOOST_MATH_SF_ERF_INV_HPP |
| 559 |
|