| 1 | // (C) Copyright John Maddock 2008. |
| 2 | // Use, modification and distribution are subject to the |
| 3 | // Boost Software License, Version 1.0. (See accompanying file |
| 4 | // LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt) |
| 5 | |
| 6 | #ifndef BOOST_MATH_SPECIAL_NEXT_HPP |
| 7 | #define BOOST_MATH_SPECIAL_NEXT_HPP |
| 8 | |
| 9 | #ifdef _MSC_VER |
| 10 | #pragma once |
| 11 | #endif |
| 12 | |
| 13 | #include <boost/math/special_functions/math_fwd.hpp> |
| 14 | #include <boost/math/policies/error_handling.hpp> |
| 15 | #include <boost/math/special_functions/fpclassify.hpp> |
| 16 | #include <boost/math/special_functions/sign.hpp> |
| 17 | #include <boost/math/special_functions/trunc.hpp> |
| 18 | #include <boost/math/tools/traits.hpp> |
| 19 | #include <type_traits> |
| 20 | #include <cfloat> |
| 21 | |
| 22 | |
| 23 | #if !defined(_CRAYC) && !defined(__CUDACC__) && (!defined(__GNUC__) || (__GNUC__ > 3) || ((__GNUC__ == 3) && (__GNUC_MINOR__ > 3))) |
| 24 | #if (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || defined(__SSE2__) |
| 25 | #include "xmmintrin.h" |
| 26 | #define BOOST_MATH_CHECK_SSE2 |
| 27 | #endif |
| 28 | #endif |
| 29 | |
| 30 | namespace boost{ namespace math{ |
| 31 | |
| 32 | namespace concepts { |
| 33 | |
| 34 | class real_concept; |
| 35 | class std_real_concept; |
| 36 | |
| 37 | } |
| 38 | |
| 39 | namespace detail{ |
| 40 | |
| 41 | template <class T> |
| 42 | struct has_hidden_guard_digits; |
| 43 | template <> |
| 44 | struct has_hidden_guard_digits<float> : public std::false_type {}; |
| 45 | template <> |
| 46 | struct has_hidden_guard_digits<double> : public std::false_type {}; |
| 47 | template <> |
| 48 | struct has_hidden_guard_digits<long double> : public std::false_type {}; |
| 49 | #ifdef BOOST_HAS_FLOAT128 |
| 50 | template <> |
| 51 | struct has_hidden_guard_digits<__float128> : public std::false_type {}; |
| 52 | #endif |
| 53 | template <> |
| 54 | struct has_hidden_guard_digits<boost::math::concepts::real_concept> : public std::false_type {}; |
| 55 | template <> |
| 56 | struct has_hidden_guard_digits<boost::math::concepts::std_real_concept> : public std::false_type {}; |
| 57 | |
| 58 | template <class T, bool b> |
| 59 | struct has_hidden_guard_digits_10 : public std::false_type {}; |
| 60 | template <class T> |
| 61 | struct has_hidden_guard_digits_10<T, true> : public std::integral_constant<bool, (std::numeric_limits<T>::digits10 != std::numeric_limits<T>::max_digits10)> {}; |
| 62 | |
| 63 | template <class T> |
| 64 | struct has_hidden_guard_digits |
| 65 | : public has_hidden_guard_digits_10<T, |
| 66 | std::numeric_limits<T>::is_specialized |
| 67 | && (std::numeric_limits<T>::radix == 10) > |
| 68 | {}; |
| 69 | |
| 70 | template <class T> |
| 71 | inline const T& normalize_value(const T& val, const std::false_type&) { return val; } |
| 72 | template <class T> |
| 73 | inline T normalize_value(const T& val, const std::true_type&) |
| 74 | { |
| 75 | static_assert(std::numeric_limits<T>::is_specialized, "Type T must be specialized." ); |
| 76 | static_assert(std::numeric_limits<T>::radix != 2, "Type T must be specialized." ); |
| 77 | |
| 78 | std::intmax_t shift = (std::intmax_t)std::numeric_limits<T>::digits - (std::intmax_t)ilogb(val) - 1; |
| 79 | T result = scalbn(val, shift); |
| 80 | result = round(result); |
| 81 | return scalbn(result, -shift); |
| 82 | } |
| 83 | |
| 84 | template <class T> |
| 85 | inline T get_smallest_value(std::true_type const&) |
| 86 | { |
| 87 | // |
| 88 | // numeric_limits lies about denorms being present - particularly |
| 89 | // when this can be turned on or off at runtime, as is the case |
| 90 | // when using the SSE2 registers in DAZ or FTZ mode. |
| 91 | // |
| 92 | static const T m = std::numeric_limits<T>::denorm_min(); |
| 93 | #ifdef BOOST_MATH_CHECK_SSE2 |
| 94 | return (_mm_getcsr() & (_MM_FLUSH_ZERO_ON | 0x40)) ? tools::min_value<T>() : m; |
| 95 | #else |
| 96 | return ((tools::min_value<T>() / 2) == 0) ? tools::min_value<T>() : m; |
| 97 | #endif |
| 98 | } |
| 99 | |
| 100 | template <class T> |
| 101 | inline T get_smallest_value(std::false_type const&) |
| 102 | { |
| 103 | return tools::min_value<T>(); |
| 104 | } |
| 105 | |
| 106 | template <class T> |
| 107 | inline T get_smallest_value() |
| 108 | { |
| 109 | #if defined(BOOST_MSVC) && (BOOST_MSVC <= 1310) |
| 110 | return get_smallest_value<T>(std::integral_constant<bool, std::numeric_limits<T>::is_specialized && (std::numeric_limits<T>::has_denorm == 1)>()); |
| 111 | #else |
| 112 | return get_smallest_value<T>(std::integral_constant<bool, std::numeric_limits<T>::is_specialized && (std::numeric_limits<T>::has_denorm == std::denorm_present)>()); |
| 113 | #endif |
| 114 | } |
| 115 | |
| 116 | // |
| 117 | // Returns the smallest value that won't generate denorms when |
| 118 | // we calculate the value of the least-significant-bit: |
| 119 | // |
| 120 | template <class T> |
| 121 | T get_min_shift_value(); |
| 122 | |
| 123 | template <class T> |
| 124 | struct min_shift_initializer |
| 125 | { |
| 126 | struct init |
| 127 | { |
| 128 | init() |
| 129 | { |
| 130 | do_init(); |
| 131 | } |
| 132 | static void do_init() |
| 133 | { |
| 134 | get_min_shift_value<T>(); |
| 135 | } |
| 136 | void force_instantiate()const{} |
| 137 | }; |
| 138 | static const init initializer; |
| 139 | static void force_instantiate() |
| 140 | { |
| 141 | initializer.force_instantiate(); |
| 142 | } |
| 143 | }; |
| 144 | |
| 145 | template <class T> |
| 146 | const typename min_shift_initializer<T>::init min_shift_initializer<T>::initializer; |
| 147 | |
| 148 | template <class T> |
| 149 | inline T calc_min_shifted(const std::true_type&) |
| 150 | { |
| 151 | BOOST_MATH_STD_USING |
| 152 | return ldexp(tools::min_value<T>(), tools::digits<T>() + 1); |
| 153 | } |
| 154 | template <class T> |
| 155 | inline T calc_min_shifted(const std::false_type&) |
| 156 | { |
| 157 | static_assert(std::numeric_limits<T>::is_specialized, "Type T must be specialized." ); |
| 158 | static_assert(std::numeric_limits<T>::radix != 2, "Type T must be specialized." ); |
| 159 | |
| 160 | return scalbn(tools::min_value<T>(), std::numeric_limits<T>::digits + 1); |
| 161 | } |
| 162 | |
| 163 | |
| 164 | template <class T> |
| 165 | inline T get_min_shift_value() |
| 166 | { |
| 167 | static const T val = calc_min_shifted<T>(std::integral_constant<bool, !std::numeric_limits<T>::is_specialized || std::numeric_limits<T>::radix == 2>()); |
| 168 | min_shift_initializer<T>::force_instantiate(); |
| 169 | |
| 170 | return val; |
| 171 | } |
| 172 | |
| 173 | template <class T, bool b = boost::math::tools::detail::has_backend_type<T>::value> |
| 174 | struct exponent_type |
| 175 | { |
| 176 | typedef int type; |
| 177 | }; |
| 178 | |
| 179 | template <class T> |
| 180 | struct exponent_type<T, true> |
| 181 | { |
| 182 | typedef typename T::backend_type::exponent_type type; |
| 183 | }; |
| 184 | |
| 185 | template <class T, class Policy> |
| 186 | T float_next_imp(const T& val, const std::true_type&, const Policy& pol) |
| 187 | { |
| 188 | typedef typename exponent_type<T>::type exponent_type; |
| 189 | |
| 190 | BOOST_MATH_STD_USING |
| 191 | exponent_type expon; |
| 192 | static const char* function = "float_next<%1%>(%1%)" ; |
| 193 | |
| 194 | int fpclass = (boost::math::fpclassify)(val); |
| 195 | |
| 196 | if((fpclass == (int)FP_NAN) || (fpclass == (int)FP_INFINITE)) |
| 197 | { |
| 198 | if(val < 0) |
| 199 | return -tools::max_value<T>(); |
| 200 | return policies::raise_domain_error<T>( |
| 201 | function, |
| 202 | "Argument must be finite, but got %1%" , val, pol); |
| 203 | } |
| 204 | |
| 205 | if(val >= tools::max_value<T>()) |
| 206 | return policies::raise_overflow_error<T>(function, nullptr, pol); |
| 207 | |
| 208 | if(val == 0) |
| 209 | return detail::get_smallest_value<T>(); |
| 210 | |
| 211 | if((fpclass != (int)FP_SUBNORMAL) && (fpclass != (int)FP_ZERO) && (fabs(val) < detail::get_min_shift_value<T>()) && (val != -tools::min_value<T>())) |
| 212 | { |
| 213 | // |
| 214 | // Special case: if the value of the least significant bit is a denorm, and the result |
| 215 | // would not be a denorm, then shift the input, increment, and shift back. |
| 216 | // This avoids issues with the Intel SSE2 registers when the FTZ or DAZ flags are set. |
| 217 | // |
| 218 | return ldexp(float_next(T(ldexp(val, 2 * tools::digits<T>())), pol), -2 * tools::digits<T>()); |
| 219 | } |
| 220 | |
| 221 | if(-0.5f == frexp(val, &expon)) |
| 222 | --expon; // reduce exponent when val is a power of two, and negative. |
| 223 | T diff = ldexp(T(1), expon - tools::digits<T>()); |
| 224 | if(diff == 0) |
| 225 | diff = detail::get_smallest_value<T>(); |
| 226 | return val + diff; |
| 227 | } // float_next_imp |
| 228 | // |
| 229 | // Special version for some base other than 2: |
| 230 | // |
| 231 | template <class T, class Policy> |
| 232 | T float_next_imp(const T& val, const std::false_type&, const Policy& pol) |
| 233 | { |
| 234 | typedef typename exponent_type<T>::type exponent_type; |
| 235 | |
| 236 | static_assert(std::numeric_limits<T>::is_specialized, "Type T must be specialized." ); |
| 237 | static_assert(std::numeric_limits<T>::radix != 2, "Type T must be specialized." ); |
| 238 | |
| 239 | BOOST_MATH_STD_USING |
| 240 | exponent_type expon; |
| 241 | static const char* function = "float_next<%1%>(%1%)" ; |
| 242 | |
| 243 | int fpclass = (boost::math::fpclassify)(val); |
| 244 | |
| 245 | if((fpclass == (int)FP_NAN) || (fpclass == (int)FP_INFINITE)) |
| 246 | { |
| 247 | if(val < 0) |
| 248 | return -tools::max_value<T>(); |
| 249 | return policies::raise_domain_error<T>( |
| 250 | function, |
| 251 | "Argument must be finite, but got %1%" , val, pol); |
| 252 | } |
| 253 | |
| 254 | if(val >= tools::max_value<T>()) |
| 255 | return policies::raise_overflow_error<T>(function, nullptr, pol); |
| 256 | |
| 257 | if(val == 0) |
| 258 | return detail::get_smallest_value<T>(); |
| 259 | |
| 260 | if((fpclass != (int)FP_SUBNORMAL) && (fpclass != (int)FP_ZERO) && (fabs(val) < detail::get_min_shift_value<T>()) && (val != -tools::min_value<T>())) |
| 261 | { |
| 262 | // |
| 263 | // Special case: if the value of the least significant bit is a denorm, and the result |
| 264 | // would not be a denorm, then shift the input, increment, and shift back. |
| 265 | // This avoids issues with the Intel SSE2 registers when the FTZ or DAZ flags are set. |
| 266 | // |
| 267 | return scalbn(float_next(T(scalbn(val, 2 * std::numeric_limits<T>::digits)), pol), -2 * std::numeric_limits<T>::digits); |
| 268 | } |
| 269 | |
| 270 | expon = 1 + ilogb(val); |
| 271 | if(-1 == scalbn(val, -expon) * std::numeric_limits<T>::radix) |
| 272 | --expon; // reduce exponent when val is a power of base, and negative. |
| 273 | T diff = scalbn(T(1), expon - std::numeric_limits<T>::digits); |
| 274 | if(diff == 0) |
| 275 | diff = detail::get_smallest_value<T>(); |
| 276 | return val + diff; |
| 277 | } // float_next_imp |
| 278 | |
| 279 | } // namespace detail |
| 280 | |
| 281 | template <class T, class Policy> |
| 282 | inline typename tools::promote_args<T>::type float_next(const T& val, const Policy& pol) |
| 283 | { |
| 284 | typedef typename tools::promote_args<T>::type result_type; |
| 285 | return detail::float_next_imp(detail::normalize_value(static_cast<result_type>(val), typename detail::has_hidden_guard_digits<result_type>::type()), std::integral_constant<bool, !std::numeric_limits<result_type>::is_specialized || (std::numeric_limits<result_type>::radix == 2)>(), pol); |
| 286 | } |
| 287 | |
| 288 | #if 0 //def BOOST_MSVC |
| 289 | // |
| 290 | // We used to use ::_nextafter here, but doing so fails when using |
| 291 | // the SSE2 registers if the FTZ or DAZ flags are set, so use our own |
| 292 | // - albeit slower - code instead as at least that gives the correct answer. |
| 293 | // |
| 294 | template <class Policy> |
| 295 | inline double float_next(const double& val, const Policy& pol) |
| 296 | { |
| 297 | static const char* function = "float_next<%1%>(%1%)" ; |
| 298 | |
| 299 | if(!(boost::math::isfinite)(val) && (val > 0)) |
| 300 | return policies::raise_domain_error<double>( |
| 301 | function, |
| 302 | "Argument must be finite, but got %1%" , val, pol); |
| 303 | |
| 304 | if(val >= tools::max_value<double>()) |
| 305 | return policies::raise_overflow_error<double>(function, nullptr, pol); |
| 306 | |
| 307 | return ::_nextafter(val, tools::max_value<double>()); |
| 308 | } |
| 309 | #endif |
| 310 | |
| 311 | template <class T> |
| 312 | inline typename tools::promote_args<T>::type float_next(const T& val) |
| 313 | { |
| 314 | return float_next(val, policies::policy<>()); |
| 315 | } |
| 316 | |
| 317 | namespace detail{ |
| 318 | |
| 319 | template <class T, class Policy> |
| 320 | T float_prior_imp(const T& val, const std::true_type&, const Policy& pol) |
| 321 | { |
| 322 | typedef typename exponent_type<T>::type exponent_type; |
| 323 | |
| 324 | BOOST_MATH_STD_USING |
| 325 | exponent_type expon; |
| 326 | static const char* function = "float_prior<%1%>(%1%)" ; |
| 327 | |
| 328 | int fpclass = (boost::math::fpclassify)(val); |
| 329 | |
| 330 | if((fpclass == (int)FP_NAN) || (fpclass == (int)FP_INFINITE)) |
| 331 | { |
| 332 | if(val > 0) |
| 333 | return tools::max_value<T>(); |
| 334 | return policies::raise_domain_error<T>( |
| 335 | function, |
| 336 | "Argument must be finite, but got %1%" , val, pol); |
| 337 | } |
| 338 | |
| 339 | if(val <= -tools::max_value<T>()) |
| 340 | return -policies::raise_overflow_error<T>(function, nullptr, pol); |
| 341 | |
| 342 | if(val == 0) |
| 343 | return -detail::get_smallest_value<T>(); |
| 344 | |
| 345 | if((fpclass != (int)FP_SUBNORMAL) && (fpclass != (int)FP_ZERO) && (fabs(val) < detail::get_min_shift_value<T>()) && (val != tools::min_value<T>())) |
| 346 | { |
| 347 | // |
| 348 | // Special case: if the value of the least significant bit is a denorm, and the result |
| 349 | // would not be a denorm, then shift the input, increment, and shift back. |
| 350 | // This avoids issues with the Intel SSE2 registers when the FTZ or DAZ flags are set. |
| 351 | // |
| 352 | return ldexp(float_prior(T(ldexp(val, 2 * tools::digits<T>())), pol), -2 * tools::digits<T>()); |
| 353 | } |
| 354 | |
| 355 | T remain = frexp(val, &expon); |
| 356 | if(remain == 0.5f) |
| 357 | --expon; // when val is a power of two we must reduce the exponent |
| 358 | T diff = ldexp(T(1), expon - tools::digits<T>()); |
| 359 | if(diff == 0) |
| 360 | diff = detail::get_smallest_value<T>(); |
| 361 | return val - diff; |
| 362 | } // float_prior_imp |
| 363 | // |
| 364 | // Special version for bases other than 2: |
| 365 | // |
| 366 | template <class T, class Policy> |
| 367 | T float_prior_imp(const T& val, const std::false_type&, const Policy& pol) |
| 368 | { |
| 369 | typedef typename exponent_type<T>::type exponent_type; |
| 370 | |
| 371 | static_assert(std::numeric_limits<T>::is_specialized, "Type T must be specialized." ); |
| 372 | static_assert(std::numeric_limits<T>::radix != 2, "Type T must be specialized." ); |
| 373 | |
| 374 | BOOST_MATH_STD_USING |
| 375 | exponent_type expon; |
| 376 | static const char* function = "float_prior<%1%>(%1%)" ; |
| 377 | |
| 378 | int fpclass = (boost::math::fpclassify)(val); |
| 379 | |
| 380 | if((fpclass == (int)FP_NAN) || (fpclass == (int)FP_INFINITE)) |
| 381 | { |
| 382 | if(val > 0) |
| 383 | return tools::max_value<T>(); |
| 384 | return policies::raise_domain_error<T>( |
| 385 | function, |
| 386 | "Argument must be finite, but got %1%" , val, pol); |
| 387 | } |
| 388 | |
| 389 | if(val <= -tools::max_value<T>()) |
| 390 | return -policies::raise_overflow_error<T>(function, nullptr, pol); |
| 391 | |
| 392 | if(val == 0) |
| 393 | return -detail::get_smallest_value<T>(); |
| 394 | |
| 395 | if((fpclass != (int)FP_SUBNORMAL) && (fpclass != (int)FP_ZERO) && (fabs(val) < detail::get_min_shift_value<T>()) && (val != tools::min_value<T>())) |
| 396 | { |
| 397 | // |
| 398 | // Special case: if the value of the least significant bit is a denorm, and the result |
| 399 | // would not be a denorm, then shift the input, increment, and shift back. |
| 400 | // This avoids issues with the Intel SSE2 registers when the FTZ or DAZ flags are set. |
| 401 | // |
| 402 | return scalbn(float_prior(T(scalbn(val, 2 * std::numeric_limits<T>::digits)), pol), -2 * std::numeric_limits<T>::digits); |
| 403 | } |
| 404 | |
| 405 | expon = 1 + ilogb(val); |
| 406 | T remain = scalbn(val, -expon); |
| 407 | if(remain * std::numeric_limits<T>::radix == 1) |
| 408 | --expon; // when val is a power of two we must reduce the exponent |
| 409 | T diff = scalbn(T(1), expon - std::numeric_limits<T>::digits); |
| 410 | if(diff == 0) |
| 411 | diff = detail::get_smallest_value<T>(); |
| 412 | return val - diff; |
| 413 | } // float_prior_imp |
| 414 | |
| 415 | } // namespace detail |
| 416 | |
| 417 | template <class T, class Policy> |
| 418 | inline typename tools::promote_args<T>::type float_prior(const T& val, const Policy& pol) |
| 419 | { |
| 420 | typedef typename tools::promote_args<T>::type result_type; |
| 421 | return detail::float_prior_imp(detail::normalize_value(static_cast<result_type>(val), typename detail::has_hidden_guard_digits<result_type>::type()), std::integral_constant<bool, !std::numeric_limits<result_type>::is_specialized || (std::numeric_limits<result_type>::radix == 2)>(), pol); |
| 422 | } |
| 423 | |
| 424 | #if 0 //def BOOST_MSVC |
| 425 | // |
| 426 | // We used to use ::_nextafter here, but doing so fails when using |
| 427 | // the SSE2 registers if the FTZ or DAZ flags are set, so use our own |
| 428 | // - albeit slower - code instead as at least that gives the correct answer. |
| 429 | // |
| 430 | template <class Policy> |
| 431 | inline double float_prior(const double& val, const Policy& pol) |
| 432 | { |
| 433 | static const char* function = "float_prior<%1%>(%1%)" ; |
| 434 | |
| 435 | if(!(boost::math::isfinite)(val) && (val < 0)) |
| 436 | return policies::raise_domain_error<double>( |
| 437 | function, |
| 438 | "Argument must be finite, but got %1%" , val, pol); |
| 439 | |
| 440 | if(val <= -tools::max_value<double>()) |
| 441 | return -policies::raise_overflow_error<double>(function, nullptr, pol); |
| 442 | |
| 443 | return ::_nextafter(val, -tools::max_value<double>()); |
| 444 | } |
| 445 | #endif |
| 446 | |
| 447 | template <class T> |
| 448 | inline typename tools::promote_args<T>::type float_prior(const T& val) |
| 449 | { |
| 450 | return float_prior(val, policies::policy<>()); |
| 451 | } |
| 452 | |
| 453 | template <class T, class U, class Policy> |
| 454 | inline typename tools::promote_args<T, U>::type nextafter(const T& val, const U& direction, const Policy& pol) |
| 455 | { |
| 456 | typedef typename tools::promote_args<T, U>::type result_type; |
| 457 | return val < direction ? boost::math::float_next<result_type>(val, pol) : val == direction ? val : boost::math::float_prior<result_type>(val, pol); |
| 458 | } |
| 459 | |
| 460 | template <class T, class U> |
| 461 | inline typename tools::promote_args<T, U>::type nextafter(const T& val, const U& direction) |
| 462 | { |
| 463 | return nextafter(val, direction, policies::policy<>()); |
| 464 | } |
| 465 | |
| 466 | namespace detail{ |
| 467 | |
| 468 | template <class T, class Policy> |
| 469 | T float_distance_imp(const T& a, const T& b, const std::true_type&, const Policy& pol) |
| 470 | { |
| 471 | BOOST_MATH_STD_USING |
| 472 | // |
| 473 | // Error handling: |
| 474 | // |
| 475 | static const char* function = "float_distance<%1%>(%1%, %1%)" ; |
| 476 | if(!(boost::math::isfinite)(a)) |
| 477 | return policies::raise_domain_error<T>( |
| 478 | function, |
| 479 | "Argument a must be finite, but got %1%" , a, pol); |
| 480 | if(!(boost::math::isfinite)(b)) |
| 481 | return policies::raise_domain_error<T>( |
| 482 | function, |
| 483 | "Argument b must be finite, but got %1%" , b, pol); |
| 484 | // |
| 485 | // Special cases: |
| 486 | // |
| 487 | if(a > b) |
| 488 | return -float_distance(b, a, pol); |
| 489 | if(a == b) |
| 490 | return T(0); |
| 491 | if(a == 0) |
| 492 | return 1 + fabs(float_distance(static_cast<T>((b < 0) ? T(-detail::get_smallest_value<T>()) : detail::get_smallest_value<T>()), b, pol)); |
| 493 | if(b == 0) |
| 494 | return 1 + fabs(float_distance(static_cast<T>((a < 0) ? T(-detail::get_smallest_value<T>()) : detail::get_smallest_value<T>()), a, pol)); |
| 495 | if(boost::math::sign(a) != boost::math::sign(b)) |
| 496 | return 2 + fabs(float_distance(static_cast<T>((b < 0) ? T(-detail::get_smallest_value<T>()) : detail::get_smallest_value<T>()), b, pol)) |
| 497 | + fabs(float_distance(static_cast<T>((a < 0) ? T(-detail::get_smallest_value<T>()) : detail::get_smallest_value<T>()), a, pol)); |
| 498 | // |
| 499 | // By the time we get here, both a and b must have the same sign, we want |
| 500 | // b > a and both positive for the following logic: |
| 501 | // |
| 502 | if(a < 0) |
| 503 | return float_distance(static_cast<T>(-b), static_cast<T>(-a), pol); |
| 504 | |
| 505 | BOOST_MATH_ASSERT(a >= 0); |
| 506 | BOOST_MATH_ASSERT(b >= a); |
| 507 | |
| 508 | int expon; |
| 509 | // |
| 510 | // Note that if a is a denorm then the usual formula fails |
| 511 | // because we actually have fewer than tools::digits<T>() |
| 512 | // significant bits in the representation: |
| 513 | // |
| 514 | (void)frexp(((boost::math::fpclassify)(a) == (int)FP_SUBNORMAL) ? tools::min_value<T>() : a, &expon); |
| 515 | T upper = ldexp(T(1), expon); |
| 516 | T result = T(0); |
| 517 | // |
| 518 | // If b is greater than upper, then we *must* split the calculation |
| 519 | // as the size of the ULP changes with each order of magnitude change: |
| 520 | // |
| 521 | if(b > upper) |
| 522 | { |
| 523 | int expon2; |
| 524 | (void)frexp(b, &expon2); |
| 525 | T upper2 = ldexp(T(0.5), expon2); |
| 526 | result = float_distance(upper2, b); |
| 527 | result += (expon2 - expon - 1) * ldexp(T(1), tools::digits<T>() - 1); |
| 528 | } |
| 529 | // |
| 530 | // Use compensated double-double addition to avoid rounding |
| 531 | // errors in the subtraction: |
| 532 | // |
| 533 | expon = tools::digits<T>() - expon; |
| 534 | T mb, x, y, z; |
| 535 | if(((boost::math::fpclassify)(a) == (int)FP_SUBNORMAL) || (b - a < tools::min_value<T>())) |
| 536 | { |
| 537 | // |
| 538 | // Special case - either one end of the range is a denormal, or else the difference is. |
| 539 | // The regular code will fail if we're using the SSE2 registers on Intel and either |
| 540 | // the FTZ or DAZ flags are set. |
| 541 | // |
| 542 | T a2 = ldexp(a, tools::digits<T>()); |
| 543 | T b2 = ldexp(b, tools::digits<T>()); |
| 544 | mb = -(std::min)(T(ldexp(upper, tools::digits<T>())), b2); |
| 545 | x = a2 + mb; |
| 546 | z = x - a2; |
| 547 | y = (a2 - (x - z)) + (mb - z); |
| 548 | |
| 549 | expon -= tools::digits<T>(); |
| 550 | } |
| 551 | else |
| 552 | { |
| 553 | mb = -(std::min)(upper, b); |
| 554 | x = a + mb; |
| 555 | z = x - a; |
| 556 | y = (a - (x - z)) + (mb - z); |
| 557 | } |
| 558 | if(x < 0) |
| 559 | { |
| 560 | x = -x; |
| 561 | y = -y; |
| 562 | } |
| 563 | result += ldexp(x, expon) + ldexp(y, expon); |
| 564 | // |
| 565 | // Result must be an integer: |
| 566 | // |
| 567 | BOOST_MATH_ASSERT(result == floor(result)); |
| 568 | return result; |
| 569 | } // float_distance_imp |
| 570 | // |
| 571 | // Special versions for bases other than 2: |
| 572 | // |
| 573 | template <class T, class Policy> |
| 574 | T float_distance_imp(const T& a, const T& b, const std::false_type&, const Policy& pol) |
| 575 | { |
| 576 | static_assert(std::numeric_limits<T>::is_specialized, "Type T must be specialized." ); |
| 577 | static_assert(std::numeric_limits<T>::radix != 2, "Type T must be specialized." ); |
| 578 | |
| 579 | BOOST_MATH_STD_USING |
| 580 | // |
| 581 | // Error handling: |
| 582 | // |
| 583 | static const char* function = "float_distance<%1%>(%1%, %1%)" ; |
| 584 | if(!(boost::math::isfinite)(a)) |
| 585 | return policies::raise_domain_error<T>( |
| 586 | function, |
| 587 | "Argument a must be finite, but got %1%" , a, pol); |
| 588 | if(!(boost::math::isfinite)(b)) |
| 589 | return policies::raise_domain_error<T>( |
| 590 | function, |
| 591 | "Argument b must be finite, but got %1%" , b, pol); |
| 592 | // |
| 593 | // Special cases: |
| 594 | // |
| 595 | if(a > b) |
| 596 | return -float_distance(b, a, pol); |
| 597 | if(a == b) |
| 598 | return T(0); |
| 599 | if(a == 0) |
| 600 | return 1 + fabs(float_distance(static_cast<T>((b < 0) ? T(-detail::get_smallest_value<T>()) : detail::get_smallest_value<T>()), b, pol)); |
| 601 | if(b == 0) |
| 602 | return 1 + fabs(float_distance(static_cast<T>((a < 0) ? T(-detail::get_smallest_value<T>()) : detail::get_smallest_value<T>()), a, pol)); |
| 603 | if(boost::math::sign(a) != boost::math::sign(b)) |
| 604 | return 2 + fabs(float_distance(static_cast<T>((b < 0) ? T(-detail::get_smallest_value<T>()) : detail::get_smallest_value<T>()), b, pol)) |
| 605 | + fabs(float_distance(static_cast<T>((a < 0) ? T(-detail::get_smallest_value<T>()) : detail::get_smallest_value<T>()), a, pol)); |
| 606 | // |
| 607 | // By the time we get here, both a and b must have the same sign, we want |
| 608 | // b > a and both positive for the following logic: |
| 609 | // |
| 610 | if(a < 0) |
| 611 | return float_distance(static_cast<T>(-b), static_cast<T>(-a), pol); |
| 612 | |
| 613 | BOOST_MATH_ASSERT(a >= 0); |
| 614 | BOOST_MATH_ASSERT(b >= a); |
| 615 | |
| 616 | std::intmax_t expon; |
| 617 | // |
| 618 | // Note that if a is a denorm then the usual formula fails |
| 619 | // because we actually have fewer than tools::digits<T>() |
| 620 | // significant bits in the representation: |
| 621 | // |
| 622 | expon = 1 + ilogb(((boost::math::fpclassify)(a) == (int)FP_SUBNORMAL) ? tools::min_value<T>() : a); |
| 623 | T upper = scalbn(T(1), expon); |
| 624 | T result = T(0); |
| 625 | // |
| 626 | // If b is greater than upper, then we *must* split the calculation |
| 627 | // as the size of the ULP changes with each order of magnitude change: |
| 628 | // |
| 629 | if(b > upper) |
| 630 | { |
| 631 | std::intmax_t expon2 = 1 + ilogb(b); |
| 632 | T upper2 = scalbn(T(1), expon2 - 1); |
| 633 | result = float_distance(upper2, b); |
| 634 | result += (expon2 - expon - 1) * scalbn(T(1), std::numeric_limits<T>::digits - 1); |
| 635 | } |
| 636 | // |
| 637 | // Use compensated double-double addition to avoid rounding |
| 638 | // errors in the subtraction: |
| 639 | // |
| 640 | expon = std::numeric_limits<T>::digits - expon; |
| 641 | T mb, x, y, z; |
| 642 | if(((boost::math::fpclassify)(a) == (int)FP_SUBNORMAL) || (b - a < tools::min_value<T>())) |
| 643 | { |
| 644 | // |
| 645 | // Special case - either one end of the range is a denormal, or else the difference is. |
| 646 | // The regular code will fail if we're using the SSE2 registers on Intel and either |
| 647 | // the FTZ or DAZ flags are set. |
| 648 | // |
| 649 | T a2 = scalbn(a, std::numeric_limits<T>::digits); |
| 650 | T b2 = scalbn(b, std::numeric_limits<T>::digits); |
| 651 | mb = -(std::min)(T(scalbn(upper, std::numeric_limits<T>::digits)), b2); |
| 652 | x = a2 + mb; |
| 653 | z = x - a2; |
| 654 | y = (a2 - (x - z)) + (mb - z); |
| 655 | |
| 656 | expon -= std::numeric_limits<T>::digits; |
| 657 | } |
| 658 | else |
| 659 | { |
| 660 | mb = -(std::min)(upper, b); |
| 661 | x = a + mb; |
| 662 | z = x - a; |
| 663 | y = (a - (x - z)) + (mb - z); |
| 664 | } |
| 665 | if(x < 0) |
| 666 | { |
| 667 | x = -x; |
| 668 | y = -y; |
| 669 | } |
| 670 | result += scalbn(x, expon) + scalbn(y, expon); |
| 671 | // |
| 672 | // Result must be an integer: |
| 673 | // |
| 674 | BOOST_MATH_ASSERT(result == floor(result)); |
| 675 | return result; |
| 676 | } // float_distance_imp |
| 677 | |
| 678 | } // namespace detail |
| 679 | |
| 680 | template <class T, class U, class Policy> |
| 681 | inline typename tools::promote_args<T, U>::type float_distance(const T& a, const U& b, const Policy& pol) |
| 682 | { |
| 683 | // |
| 684 | // We allow ONE of a and b to be an integer type, otherwise both must be the SAME type. |
| 685 | // |
| 686 | static_assert( |
| 687 | (std::is_same<T, U>::value |
| 688 | || (std::is_integral<T>::value && !std::is_integral<U>::value) |
| 689 | || (!std::is_integral<T>::value && std::is_integral<U>::value) |
| 690 | || (std::numeric_limits<T>::is_specialized && std::numeric_limits<U>::is_specialized |
| 691 | && (std::numeric_limits<T>::digits == std::numeric_limits<U>::digits) |
| 692 | && (std::numeric_limits<T>::radix == std::numeric_limits<U>::radix) |
| 693 | && !std::numeric_limits<T>::is_integer && !std::numeric_limits<U>::is_integer)), |
| 694 | "Float distance between two different floating point types is undefined." ); |
| 695 | |
| 696 | BOOST_IF_CONSTEXPR (!std::is_same<T, U>::value) |
| 697 | { |
| 698 | BOOST_IF_CONSTEXPR(std::is_integral<T>::value) |
| 699 | { |
| 700 | return float_distance(static_cast<U>(a), b, pol); |
| 701 | } |
| 702 | else |
| 703 | { |
| 704 | return float_distance(a, static_cast<T>(b), pol); |
| 705 | } |
| 706 | } |
| 707 | else |
| 708 | { |
| 709 | typedef typename tools::promote_args<T, U>::type result_type; |
| 710 | return detail::float_distance_imp(detail::normalize_value(static_cast<result_type>(a), typename detail::has_hidden_guard_digits<result_type>::type()), detail::normalize_value(static_cast<result_type>(b), typename detail::has_hidden_guard_digits<result_type>::type()), std::integral_constant<bool, !std::numeric_limits<result_type>::is_specialized || (std::numeric_limits<result_type>::radix == 2)>(), pol); |
| 711 | } |
| 712 | } |
| 713 | |
| 714 | template <class T, class U> |
| 715 | typename tools::promote_args<T, U>::type float_distance(const T& a, const U& b) |
| 716 | { |
| 717 | return boost::math::float_distance(a, b, policies::policy<>()); |
| 718 | } |
| 719 | |
| 720 | namespace detail{ |
| 721 | |
| 722 | template <class T, class Policy> |
| 723 | T float_advance_imp(T val, int distance, const std::true_type&, const Policy& pol) |
| 724 | { |
| 725 | BOOST_MATH_STD_USING |
| 726 | // |
| 727 | // Error handling: |
| 728 | // |
| 729 | static const char* function = "float_advance<%1%>(%1%, int)" ; |
| 730 | |
| 731 | int fpclass = (boost::math::fpclassify)(val); |
| 732 | |
| 733 | if((fpclass == (int)FP_NAN) || (fpclass == (int)FP_INFINITE)) |
| 734 | return policies::raise_domain_error<T>( |
| 735 | function, |
| 736 | "Argument val must be finite, but got %1%" , val, pol); |
| 737 | |
| 738 | if(val < 0) |
| 739 | return -float_advance(-val, -distance, pol); |
| 740 | if(distance == 0) |
| 741 | return val; |
| 742 | if(distance == 1) |
| 743 | return float_next(val, pol); |
| 744 | if(distance == -1) |
| 745 | return float_prior(val, pol); |
| 746 | |
| 747 | if(fabs(val) < detail::get_min_shift_value<T>()) |
| 748 | { |
| 749 | // |
| 750 | // Special case: if the value of the least significant bit is a denorm, |
| 751 | // implement in terms of float_next/float_prior. |
| 752 | // This avoids issues with the Intel SSE2 registers when the FTZ or DAZ flags are set. |
| 753 | // |
| 754 | if(distance > 0) |
| 755 | { |
| 756 | do{ val = float_next(val, pol); } while(--distance); |
| 757 | } |
| 758 | else |
| 759 | { |
| 760 | do{ val = float_prior(val, pol); } while(++distance); |
| 761 | } |
| 762 | return val; |
| 763 | } |
| 764 | |
| 765 | int expon; |
| 766 | (void)frexp(val, &expon); |
| 767 | T limit = ldexp((distance < 0 ? T(0.5f) : T(1)), expon); |
| 768 | if(val <= tools::min_value<T>()) |
| 769 | { |
| 770 | limit = sign(T(distance)) * tools::min_value<T>(); |
| 771 | } |
| 772 | T limit_distance = float_distance(val, limit); |
| 773 | while(fabs(limit_distance) < abs(distance)) |
| 774 | { |
| 775 | distance -= itrunc(limit_distance); |
| 776 | val = limit; |
| 777 | if(distance < 0) |
| 778 | { |
| 779 | limit /= 2; |
| 780 | expon--; |
| 781 | } |
| 782 | else |
| 783 | { |
| 784 | limit *= 2; |
| 785 | expon++; |
| 786 | } |
| 787 | limit_distance = float_distance(val, limit); |
| 788 | if(distance && (limit_distance == 0)) |
| 789 | { |
| 790 | return policies::raise_evaluation_error<T>(function, "Internal logic failed while trying to increment floating point value %1%: most likely your FPU is in non-IEEE conforming mode." , val, pol); |
| 791 | } |
| 792 | } |
| 793 | if((0.5f == frexp(val, &expon)) && (distance < 0)) |
| 794 | --expon; |
| 795 | T diff = 0; |
| 796 | if(val != 0) |
| 797 | diff = distance * ldexp(T(1), expon - tools::digits<T>()); |
| 798 | if(diff == 0) |
| 799 | diff = distance * detail::get_smallest_value<T>(); |
| 800 | return val += diff; |
| 801 | } // float_advance_imp |
| 802 | // |
| 803 | // Special version for bases other than 2: |
| 804 | // |
| 805 | template <class T, class Policy> |
| 806 | T float_advance_imp(T val, int distance, const std::false_type&, const Policy& pol) |
| 807 | { |
| 808 | static_assert(std::numeric_limits<T>::is_specialized, "Type T must be specialized." ); |
| 809 | static_assert(std::numeric_limits<T>::radix != 2, "Type T must be specialized." ); |
| 810 | |
| 811 | BOOST_MATH_STD_USING |
| 812 | // |
| 813 | // Error handling: |
| 814 | // |
| 815 | static const char* function = "float_advance<%1%>(%1%, int)" ; |
| 816 | |
| 817 | int fpclass = (boost::math::fpclassify)(val); |
| 818 | |
| 819 | if((fpclass == (int)FP_NAN) || (fpclass == (int)FP_INFINITE)) |
| 820 | return policies::raise_domain_error<T>( |
| 821 | function, |
| 822 | "Argument val must be finite, but got %1%" , val, pol); |
| 823 | |
| 824 | if(val < 0) |
| 825 | return -float_advance(-val, -distance, pol); |
| 826 | if(distance == 0) |
| 827 | return val; |
| 828 | if(distance == 1) |
| 829 | return float_next(val, pol); |
| 830 | if(distance == -1) |
| 831 | return float_prior(val, pol); |
| 832 | |
| 833 | if(fabs(val) < detail::get_min_shift_value<T>()) |
| 834 | { |
| 835 | // |
| 836 | // Special case: if the value of the least significant bit is a denorm, |
| 837 | // implement in terms of float_next/float_prior. |
| 838 | // This avoids issues with the Intel SSE2 registers when the FTZ or DAZ flags are set. |
| 839 | // |
| 840 | if(distance > 0) |
| 841 | { |
| 842 | do{ val = float_next(val, pol); } while(--distance); |
| 843 | } |
| 844 | else |
| 845 | { |
| 846 | do{ val = float_prior(val, pol); } while(++distance); |
| 847 | } |
| 848 | return val; |
| 849 | } |
| 850 | |
| 851 | std::intmax_t expon = 1 + ilogb(val); |
| 852 | T limit = scalbn(T(1), distance < 0 ? expon - 1 : expon); |
| 853 | if(val <= tools::min_value<T>()) |
| 854 | { |
| 855 | limit = sign(T(distance)) * tools::min_value<T>(); |
| 856 | } |
| 857 | T limit_distance = float_distance(val, limit); |
| 858 | while(fabs(limit_distance) < abs(distance)) |
| 859 | { |
| 860 | distance -= itrunc(limit_distance); |
| 861 | val = limit; |
| 862 | if(distance < 0) |
| 863 | { |
| 864 | limit /= std::numeric_limits<T>::radix; |
| 865 | expon--; |
| 866 | } |
| 867 | else |
| 868 | { |
| 869 | limit *= std::numeric_limits<T>::radix; |
| 870 | expon++; |
| 871 | } |
| 872 | limit_distance = float_distance(val, limit); |
| 873 | if(distance && (limit_distance == 0)) |
| 874 | { |
| 875 | return policies::raise_evaluation_error<T>(function, "Internal logic failed while trying to increment floating point value %1%: most likely your FPU is in non-IEEE conforming mode." , val, pol); |
| 876 | } |
| 877 | } |
| 878 | /*expon = 1 + ilogb(val); |
| 879 | if((1 == scalbn(val, 1 + expon)) && (distance < 0)) |
| 880 | --expon;*/ |
| 881 | T diff = 0; |
| 882 | if(val != 0) |
| 883 | diff = distance * scalbn(T(1), expon - std::numeric_limits<T>::digits); |
| 884 | if(diff == 0) |
| 885 | diff = distance * detail::get_smallest_value<T>(); |
| 886 | return val += diff; |
| 887 | } // float_advance_imp |
| 888 | |
| 889 | } // namespace detail |
| 890 | |
| 891 | template <class T, class Policy> |
| 892 | inline typename tools::promote_args<T>::type float_advance(T val, int distance, const Policy& pol) |
| 893 | { |
| 894 | typedef typename tools::promote_args<T>::type result_type; |
| 895 | return detail::float_advance_imp(detail::normalize_value(static_cast<result_type>(val), typename detail::has_hidden_guard_digits<result_type>::type()), distance, std::integral_constant<bool, !std::numeric_limits<result_type>::is_specialized || (std::numeric_limits<result_type>::radix == 2)>(), pol); |
| 896 | } |
| 897 | |
| 898 | template <class T> |
| 899 | inline typename tools::promote_args<T>::type float_advance(const T& val, int distance) |
| 900 | { |
| 901 | return boost::math::float_advance(val, distance, policies::policy<>()); |
| 902 | } |
| 903 | |
| 904 | }} // boost math namespaces |
| 905 | |
| 906 | #endif // BOOST_MATH_SPECIAL_NEXT_HPP |
| 907 | |