| 1 | /////////////////////////////////////////////////////////////////////////////// |
| 2 | // Copyright 2013 John Maddock |
| 3 | // Distributed under the Boost |
| 4 | // 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_BERNOULLI_DETAIL_HPP |
| 8 | #define BOOST_MATH_BERNOULLI_DETAIL_HPP |
| 9 | |
| 10 | #include <boost/config.hpp> |
| 11 | #include <boost/detail/lightweight_mutex.hpp> |
| 12 | #include <boost/math/tools/atomic.hpp> |
| 13 | #include <boost/utility/enable_if.hpp> |
| 14 | #include <boost/math/tools/toms748_solve.hpp> |
| 15 | #include <boost/math/tools/cxx03_warn.hpp> |
| 16 | #include <vector> |
| 17 | |
| 18 | namespace boost{ namespace math{ namespace detail{ |
| 19 | // |
| 20 | // Asymptotic expansion for B2n due to |
| 21 | // Luschny LogB3 formula (http://www.luschny.de/math/primes/bernincl.html) |
| 22 | // |
| 23 | template <class T, class Policy> |
| 24 | T b2n_asymptotic(int n) |
| 25 | { |
| 26 | BOOST_MATH_STD_USING |
| 27 | const T nx = static_cast<T>(n); |
| 28 | const T nx2(nx * nx); |
| 29 | |
| 30 | const T approximate_log_of_bernoulli_bn = |
| 31 | ((boost::math::constants::half<T>() + nx) * log(nx)) |
| 32 | + ((boost::math::constants::half<T>() - nx) * log(boost::math::constants::pi<T>())) |
| 33 | + (((T(3) / 2) - nx) * boost::math::constants::ln_two<T>()) |
| 34 | + ((nx * (T(2) - (nx2 * 7) * (1 + ((nx2 * 30) * ((nx2 * 12) - 1))))) / (((nx2 * nx2) * nx2) * 2520)); |
| 35 | return ((n / 2) & 1 ? 1 : -1) * (approximate_log_of_bernoulli_bn > tools::log_max_value<T>() |
| 36 | ? policies::raise_overflow_error<T>("boost::math::bernoulli_b2n<%1%>(std::size_t)" , 0, nx, Policy()) |
| 37 | : static_cast<T>(exp(approximate_log_of_bernoulli_bn))); |
| 38 | } |
| 39 | |
| 40 | template <class T, class Policy> |
| 41 | T t2n_asymptotic(int n) |
| 42 | { |
| 43 | BOOST_MATH_STD_USING |
| 44 | // Just get B2n and convert to a Tangent number: |
| 45 | T t2n = fabs(b2n_asymptotic<T, Policy>(2 * n)) / (2 * n); |
| 46 | T p2 = ldexp(T(1), n); |
| 47 | if(tools::max_value<T>() / p2 < t2n) |
| 48 | return policies::raise_overflow_error<T>("boost::math::tangent_t2n<%1%>(std::size_t)" , 0, T(n), Policy()); |
| 49 | t2n *= p2; |
| 50 | p2 -= 1; |
| 51 | if(tools::max_value<T>() / p2 < t2n) |
| 52 | return policies::raise_overflow_error<T>("boost::math::tangent_t2n<%1%>(std::size_t)" , 0, Policy()); |
| 53 | t2n *= p2; |
| 54 | return t2n; |
| 55 | } |
| 56 | // |
| 57 | // We need to know the approximate value of /n/ which will |
| 58 | // cause bernoulli_b2n<T>(n) to return infinity - this allows |
| 59 | // us to elude a great deal of runtime checking for values below |
| 60 | // n, and only perform the full overflow checks when we know that we're |
| 61 | // getting close to the point where our calculations will overflow. |
| 62 | // We use Luschny's LogB3 formula (http://www.luschny.de/math/primes/bernincl.html) |
| 63 | // to find the limit, and since we're dealing with the log of the Bernoulli numbers |
| 64 | // we need only perform the calculation at double precision and not with T |
| 65 | // (which may be a multiprecision type). The limit returned is within 1 of the true |
| 66 | // limit for all the types tested. Note that although the code below is basically |
| 67 | // the same as b2n_asymptotic above, it has been recast as a continuous real-valued |
| 68 | // function as this makes the root finding go smoother/faster. It also omits the |
| 69 | // sign of the Bernoulli number. |
| 70 | // |
| 71 | struct max_bernoulli_root_functor |
| 72 | { |
| 73 | max_bernoulli_root_functor(ulong_long_type t) : target(static_cast<double>(t)) {} |
| 74 | double operator()(double n) |
| 75 | { |
| 76 | BOOST_MATH_STD_USING |
| 77 | |
| 78 | // Luschny LogB3(n) formula. |
| 79 | |
| 80 | const double nx2(n * n); |
| 81 | |
| 82 | const double approximate_log_of_bernoulli_bn |
| 83 | = ((boost::math::constants::half<double>() + n) * log(x: n)) |
| 84 | + ((boost::math::constants::half<double>() - n) * log(x: boost::math::constants::pi<double>())) |
| 85 | + (((double(3) / 2) - n) * boost::math::constants::ln_two<double>()) |
| 86 | + ((n * (2 - (nx2 * 7) * (1 + ((nx2 * 30) * ((nx2 * 12) - 1))))) / (((nx2 * nx2) * nx2) * 2520)); |
| 87 | |
| 88 | return approximate_log_of_bernoulli_bn - target; |
| 89 | } |
| 90 | private: |
| 91 | double target; |
| 92 | }; |
| 93 | |
| 94 | template <class T, class Policy> |
| 95 | inline std::size_t find_bernoulli_overflow_limit(const boost::false_type&) |
| 96 | { |
| 97 | // Set a limit on how large the result can ever be: |
| 98 | static const double max_result = static_cast<double>((std::numeric_limits<std::size_t>::max)() - 1000u); |
| 99 | |
| 100 | ulong_long_type t = lltrunc(boost::math::tools::log_max_value<T>()); |
| 101 | max_bernoulli_root_functor fun(t); |
| 102 | boost::math::tools::equal_floor tol; |
| 103 | boost::uintmax_t max_iter = boost::math::policies::get_max_root_iterations<Policy>(); |
| 104 | double result = boost::math::tools::toms748_solve(f: fun, ax: sqrt(x: double(t)), bx: double(t), tol, max_iter).first / 2; |
| 105 | if (result > max_result) |
| 106 | result = max_result; |
| 107 | |
| 108 | return static_cast<std::size_t>(result); |
| 109 | } |
| 110 | |
| 111 | template <class T, class Policy> |
| 112 | inline std::size_t find_bernoulli_overflow_limit(const boost::true_type&) |
| 113 | { |
| 114 | return max_bernoulli_index<bernoulli_imp_variant<T>::value>::value; |
| 115 | } |
| 116 | |
| 117 | template <class T, class Policy> |
| 118 | std::size_t b2n_overflow_limit() |
| 119 | { |
| 120 | // This routine is called at program startup if it's called at all: |
| 121 | // that guarantees safe initialization of the static variable. |
| 122 | typedef boost::integral_constant<bool, (bernoulli_imp_variant<T>::value >= 1) && (bernoulli_imp_variant<T>::value <= 3)> tag_type; |
| 123 | static const std::size_t lim = find_bernoulli_overflow_limit<T, Policy>(tag_type()); |
| 124 | return lim; |
| 125 | } |
| 126 | |
| 127 | // |
| 128 | // The tangent numbers grow larger much more rapidly than the Bernoulli numbers do.... |
| 129 | // so to compute the Bernoulli numbers from the tangent numbers, we need to avoid spurious |
| 130 | // overflow in the calculation, we can do this by scaling all the tangent number by some scale factor: |
| 131 | // |
| 132 | template <class T> |
| 133 | inline typename enable_if_c<std::numeric_limits<T>::is_specialized && (std::numeric_limits<T>::radix == 2), T>::type tangent_scale_factor() |
| 134 | { |
| 135 | BOOST_MATH_STD_USING |
| 136 | return ldexp(T(1), std::numeric_limits<T>::min_exponent + 5); |
| 137 | } |
| 138 | template <class T> |
| 139 | inline typename disable_if_c<std::numeric_limits<T>::is_specialized && (std::numeric_limits<T>::radix == 2), T>::type tangent_scale_factor() |
| 140 | { |
| 141 | return tools::min_value<T>() * 16; |
| 142 | } |
| 143 | // |
| 144 | // Initializer: ensure all our constants are initialized prior to the first call of main: |
| 145 | // |
| 146 | template <class T, class Policy> |
| 147 | struct bernoulli_initializer |
| 148 | { |
| 149 | struct init |
| 150 | { |
| 151 | init() |
| 152 | { |
| 153 | // |
| 154 | // We call twice, once to initialize our static table, and once to |
| 155 | // initialize our dymanic table: |
| 156 | // |
| 157 | boost::math::bernoulli_b2n<T>(2, Policy()); |
| 158 | #ifndef BOOST_NO_EXCEPTIONS |
| 159 | try{ |
| 160 | #endif |
| 161 | boost::math::bernoulli_b2n<T>(max_bernoulli_b2n<T>::value + 1, Policy()); |
| 162 | #ifndef BOOST_NO_EXCEPTIONS |
| 163 | } catch(const std::overflow_error&){} |
| 164 | #endif |
| 165 | boost::math::tangent_t2n<T>(2, Policy()); |
| 166 | } |
| 167 | void force_instantiate()const{} |
| 168 | }; |
| 169 | static const init initializer; |
| 170 | static void force_instantiate() |
| 171 | { |
| 172 | initializer.force_instantiate(); |
| 173 | } |
| 174 | }; |
| 175 | |
| 176 | template <class T, class Policy> |
| 177 | const typename bernoulli_initializer<T, Policy>::init bernoulli_initializer<T, Policy>::initializer; |
| 178 | |
| 179 | // |
| 180 | // We need something to act as a cache for our calculated Bernoulli numbers. In order to |
| 181 | // ensure both fast access and thread safety, we need a stable table which may be extended |
| 182 | // in size, but which never reallocates: that way values already calculated may be accessed |
| 183 | // concurrently with another thread extending the table with new values. |
| 184 | // |
| 185 | // Very very simple vector class that will never allocate more than once, we could use |
| 186 | // boost::container::static_vector here, but that allocates on the stack, which may well |
| 187 | // cause issues for the amount of memory we want in the extreme case... |
| 188 | // |
| 189 | template <class T> |
| 190 | struct fixed_vector : private std::allocator<T> |
| 191 | { |
| 192 | typedef unsigned size_type; |
| 193 | typedef T* iterator; |
| 194 | typedef const T* const_iterator; |
| 195 | fixed_vector() : m_used(0) |
| 196 | { |
| 197 | std::size_t overflow_limit = 5 + b2n_overflow_limit<T, policies::policy<> >(); |
| 198 | m_capacity = static_cast<unsigned>((std::min)(a: overflow_limit, b: static_cast<std::size_t>(100000u))); |
| 199 | m_data = this->allocate(m_capacity); |
| 200 | } |
| 201 | ~fixed_vector() |
| 202 | { |
| 203 | #ifdef BOOST_NO_CXX11_ALLOCATOR |
| 204 | for(unsigned i = 0; i < m_used; ++i) |
| 205 | this->destroy(&m_data[i]); |
| 206 | this->deallocate(m_data, m_capacity); |
| 207 | #else |
| 208 | typedef std::allocator<T> allocator_type; |
| 209 | typedef std::allocator_traits<allocator_type> allocator_traits; |
| 210 | allocator_type& alloc = *this; |
| 211 | for(unsigned i = 0; i < m_used; ++i) |
| 212 | allocator_traits::destroy(alloc, &m_data[i]); |
| 213 | allocator_traits::deallocate(alloc, m_data, m_capacity); |
| 214 | #endif |
| 215 | } |
| 216 | T& operator[](unsigned n) { BOOST_ASSERT(n < m_used); return m_data[n]; } |
| 217 | const T& operator[](unsigned n)const { BOOST_ASSERT(n < m_used); return m_data[n]; } |
| 218 | unsigned size()const { return m_used; } |
| 219 | unsigned size() { return m_used; } |
| 220 | void resize(unsigned n, const T& val) |
| 221 | { |
| 222 | if(n > m_capacity) |
| 223 | { |
| 224 | BOOST_THROW_EXCEPTION(std::runtime_error("Exhausted storage for Bernoulli numbers." )); |
| 225 | } |
| 226 | for(unsigned i = m_used; i < n; ++i) |
| 227 | new (m_data + i) T(val); |
| 228 | m_used = n; |
| 229 | } |
| 230 | void resize(unsigned n) { resize(n, T()); } |
| 231 | T* begin() { return m_data; } |
| 232 | T* end() { return m_data + m_used; } |
| 233 | T* begin()const { return m_data; } |
| 234 | T* end()const { return m_data + m_used; } |
| 235 | unsigned capacity()const { return m_capacity; } |
| 236 | void clear() { m_used = 0; } |
| 237 | private: |
| 238 | T* m_data; |
| 239 | unsigned m_used, m_capacity; |
| 240 | }; |
| 241 | |
| 242 | template <class T, class Policy> |
| 243 | class bernoulli_numbers_cache |
| 244 | { |
| 245 | public: |
| 246 | bernoulli_numbers_cache() : m_overflow_limit((std::numeric_limits<std::size_t>::max)()) |
| 247 | #if defined(BOOST_HAS_THREADS) && !defined(BOOST_MATH_NO_ATOMIC_INT) |
| 248 | , m_counter(0) |
| 249 | #endif |
| 250 | , m_current_precision(boost::math::tools::digits<T>()) |
| 251 | {} |
| 252 | |
| 253 | typedef fixed_vector<T> container_type; |
| 254 | |
| 255 | void tangent(std::size_t m) |
| 256 | { |
| 257 | static const std::size_t min_overflow_index = b2n_overflow_limit<T, Policy>() - 1; |
| 258 | tn.resize(static_cast<typename container_type::size_type>(m), T(0U)); |
| 259 | |
| 260 | BOOST_MATH_INSTRUMENT_VARIABLE(min_overflow_index); |
| 261 | |
| 262 | std::size_t prev_size = m_intermediates.size(); |
| 263 | m_intermediates.resize(m, T(0U)); |
| 264 | |
| 265 | if(prev_size == 0) |
| 266 | { |
| 267 | m_intermediates[1] = tangent_scale_factor<T>() /*T(1U)*/; |
| 268 | tn[0U] = T(0U); |
| 269 | tn[1U] = tangent_scale_factor<T>()/* T(1U)*/; |
| 270 | BOOST_MATH_INSTRUMENT_VARIABLE(tn[0]); |
| 271 | BOOST_MATH_INSTRUMENT_VARIABLE(tn[1]); |
| 272 | } |
| 273 | |
| 274 | for(std::size_t i = std::max<size_t>(a: 2, b: prev_size); i < m; i++) |
| 275 | { |
| 276 | bool overflow_check = false; |
| 277 | if(i >= min_overflow_index && (boost::math::tools::max_value<T>() / (i-1) < m_intermediates[1]) ) |
| 278 | { |
| 279 | std::fill(tn.begin() + i, tn.end(), boost::math::tools::max_value<T>()); |
| 280 | break; |
| 281 | } |
| 282 | m_intermediates[1] = m_intermediates[1] * (i-1); |
| 283 | for(std::size_t j = 2; j <= i; j++) |
| 284 | { |
| 285 | overflow_check = |
| 286 | (i >= min_overflow_index) && ( |
| 287 | (boost::math::tools::max_value<T>() / (i - j) < m_intermediates[j]) |
| 288 | || (boost::math::tools::max_value<T>() / (i - j + 2) < m_intermediates[j-1]) |
| 289 | || (boost::math::tools::max_value<T>() - m_intermediates[j] * (i - j) < m_intermediates[j-1] * (i - j + 2)) |
| 290 | || ((boost::math::isinf)(m_intermediates[j])) |
| 291 | ); |
| 292 | |
| 293 | if(overflow_check) |
| 294 | { |
| 295 | std::fill(tn.begin() + i, tn.end(), boost::math::tools::max_value<T>()); |
| 296 | break; |
| 297 | } |
| 298 | m_intermediates[j] = m_intermediates[j] * (i - j) + m_intermediates[j-1] * (i - j + 2); |
| 299 | } |
| 300 | if(overflow_check) |
| 301 | break; // already filled the tn... |
| 302 | tn[static_cast<typename container_type::size_type>(i)] = m_intermediates[i]; |
| 303 | BOOST_MATH_INSTRUMENT_VARIABLE(i); |
| 304 | BOOST_MATH_INSTRUMENT_VARIABLE(tn[static_cast<typename container_type::size_type>(i)]); |
| 305 | } |
| 306 | } |
| 307 | |
| 308 | void tangent_numbers_series(const std::size_t m) |
| 309 | { |
| 310 | BOOST_MATH_STD_USING |
| 311 | static const std::size_t min_overflow_index = b2n_overflow_limit<T, Policy>() - 1; |
| 312 | |
| 313 | typename container_type::size_type old_size = bn.size(); |
| 314 | |
| 315 | tangent(m); |
| 316 | bn.resize(static_cast<typename container_type::size_type>(m)); |
| 317 | |
| 318 | if(!old_size) |
| 319 | { |
| 320 | bn[0] = 1; |
| 321 | old_size = 1; |
| 322 | } |
| 323 | |
| 324 | T power_two(ldexp(T(1), static_cast<int>(2 * old_size))); |
| 325 | |
| 326 | for(std::size_t i = old_size; i < m; i++) |
| 327 | { |
| 328 | T b(static_cast<T>(i * 2)); |
| 329 | // |
| 330 | // Not only do we need to take care to avoid spurious over/under flow in |
| 331 | // the calculation, but we also need to avoid overflow altogether in case |
| 332 | // we're calculating with a type where "bad things" happen in that case: |
| 333 | // |
| 334 | b = b / (power_two * tangent_scale_factor<T>()); |
| 335 | b /= (power_two - 1); |
| 336 | bool overflow_check = (i >= min_overflow_index) && (tools::max_value<T>() / tn[static_cast<typename container_type::size_type>(i)] < b); |
| 337 | if(overflow_check) |
| 338 | { |
| 339 | m_overflow_limit = i; |
| 340 | while(i < m) |
| 341 | { |
| 342 | b = std::numeric_limits<T>::has_infinity ? std::numeric_limits<T>::infinity() : tools::max_value<T>(); |
| 343 | bn[static_cast<typename container_type::size_type>(i)] = ((i % 2U) ? b : T(-b)); |
| 344 | ++i; |
| 345 | } |
| 346 | break; |
| 347 | } |
| 348 | else |
| 349 | { |
| 350 | b *= tn[static_cast<typename container_type::size_type>(i)]; |
| 351 | } |
| 352 | |
| 353 | power_two = ldexp(power_two, 2); |
| 354 | |
| 355 | const bool b_neg = i % 2 == 0; |
| 356 | |
| 357 | bn[static_cast<typename container_type::size_type>(i)] = ((!b_neg) ? b : T(-b)); |
| 358 | } |
| 359 | } |
| 360 | |
| 361 | template <class OutputIterator> |
| 362 | OutputIterator copy_bernoulli_numbers(OutputIterator out, std::size_t start, std::size_t n, const Policy& pol) |
| 363 | { |
| 364 | // |
| 365 | // There are basically 3 thread safety options: |
| 366 | // |
| 367 | // 1) There are no threads (BOOST_HAS_THREADS is not defined). |
| 368 | // 2) There are threads, but we do not have a true atomic integer type, |
| 369 | // in this case we just use a mutex to guard against race conditions. |
| 370 | // 3) There are threads, and we have an atomic integer: in this case we can |
| 371 | // use the double-checked locking pattern to avoid thread synchronisation |
| 372 | // when accessing values already in the cache. |
| 373 | // |
| 374 | // First off handle the common case for overflow and/or asymptotic expansion: |
| 375 | // |
| 376 | if(start + n > bn.capacity()) |
| 377 | { |
| 378 | if(start < bn.capacity()) |
| 379 | { |
| 380 | out = copy_bernoulli_numbers(out, start, bn.capacity() - start, pol); |
| 381 | n -= bn.capacity() - start; |
| 382 | start = static_cast<std::size_t>(bn.capacity()); |
| 383 | } |
| 384 | if(start < b2n_overflow_limit<T, Policy>() + 2u) |
| 385 | { |
| 386 | for(; n; ++start, --n) |
| 387 | { |
| 388 | *out = b2n_asymptotic<T, Policy>(static_cast<typename container_type::size_type>(start * 2U)); |
| 389 | ++out; |
| 390 | } |
| 391 | } |
| 392 | for(; n; ++start, --n) |
| 393 | { |
| 394 | *out = policies::raise_overflow_error<T>("boost::math::bernoulli_b2n<%1%>(std::size_t)" , 0, T(start), pol); |
| 395 | ++out; |
| 396 | } |
| 397 | return out; |
| 398 | } |
| 399 | #if !defined(BOOST_HAS_THREADS) |
| 400 | // |
| 401 | // Single threaded code, very simple: |
| 402 | // |
| 403 | if(m_current_precision < boost::math::tools::digits<T>()) |
| 404 | { |
| 405 | bn.clear(); |
| 406 | tn.clear(); |
| 407 | m_intermediates.clear(); |
| 408 | m_current_precision = boost::math::tools::digits<T>(); |
| 409 | } |
| 410 | if(start + n >= bn.size()) |
| 411 | { |
| 412 | std::size_t new_size = (std::min)((std::max)((std::max)(std::size_t(start + n), std::size_t(bn.size() + 20)), std::size_t(50)), std::size_t(bn.capacity())); |
| 413 | tangent_numbers_series(new_size); |
| 414 | } |
| 415 | |
| 416 | for(std::size_t i = (std::max)(std::size_t(max_bernoulli_b2n<T>::value + 1), start); i < start + n; ++i) |
| 417 | { |
| 418 | *out = (i >= m_overflow_limit) ? policies::raise_overflow_error<T>("boost::math::bernoulli_b2n<%1%>(std::size_t)" , 0, T(i), pol) : bn[i]; |
| 419 | ++out; |
| 420 | } |
| 421 | #elif defined(BOOST_MATH_NO_ATOMIC_INT) |
| 422 | // |
| 423 | // We need to grab a mutex every time we get here, for both readers and writers: |
| 424 | // |
| 425 | boost::detail::lightweight_mutex::scoped_lock l(m_mutex); |
| 426 | if(m_current_precision < boost::math::tools::digits<T>()) |
| 427 | { |
| 428 | bn.clear(); |
| 429 | tn.clear(); |
| 430 | m_intermediates.clear(); |
| 431 | m_current_precision = boost::math::tools::digits<T>(); |
| 432 | } |
| 433 | if(start + n >= bn.size()) |
| 434 | { |
| 435 | std::size_t new_size = (std::min)((std::max)((std::max)(std::size_t(start + n), std::size_t(bn.size() + 20)), std::size_t(50)), std::size_t(bn.capacity())); |
| 436 | tangent_numbers_series(new_size); |
| 437 | } |
| 438 | |
| 439 | for(std::size_t i = (std::max)(std::size_t(max_bernoulli_b2n<T>::value + 1), start); i < start + n; ++i) |
| 440 | { |
| 441 | *out = (i >= m_overflow_limit) ? policies::raise_overflow_error<T>("boost::math::bernoulli_b2n<%1%>(std::size_t)" , 0, T(i), pol) : bn[i]; |
| 442 | ++out; |
| 443 | } |
| 444 | |
| 445 | #else |
| 446 | // |
| 447 | // Double-checked locking pattern, lets us access cached already cached values |
| 448 | // without locking: |
| 449 | // |
| 450 | // Get the counter and see if we need to calculate more constants: |
| 451 | // |
| 452 | if((static_cast<std::size_t>(m_counter.load(BOOST_MATH_ATOMIC_NS::memory_order_consume)) < start + n) |
| 453 | || (static_cast<int>(m_current_precision.load(BOOST_MATH_ATOMIC_NS::memory_order_consume)) < boost::math::tools::digits<T>())) |
| 454 | { |
| 455 | boost::detail::lightweight_mutex::scoped_lock l(m_mutex); |
| 456 | |
| 457 | if((static_cast<std::size_t>(m_counter.load(BOOST_MATH_ATOMIC_NS::memory_order_consume)) < start + n) |
| 458 | || (static_cast<int>(m_current_precision.load(BOOST_MATH_ATOMIC_NS::memory_order_consume)) < boost::math::tools::digits<T>())) |
| 459 | { |
| 460 | if(static_cast<int>(m_current_precision.load(BOOST_MATH_ATOMIC_NS::memory_order_consume)) < boost::math::tools::digits<T>()) |
| 461 | { |
| 462 | bn.clear(); |
| 463 | tn.clear(); |
| 464 | m_intermediates.clear(); |
| 465 | m_counter.store(i: 0, BOOST_MATH_ATOMIC_NS::memory_order_release); |
| 466 | m_current_precision = boost::math::tools::digits<T>(); |
| 467 | } |
| 468 | if(start + n >= bn.size()) |
| 469 | { |
| 470 | std::size_t new_size = (std::min)(a: (std::max)(a: (std::max)(a: std::size_t(start + n), b: std::size_t(bn.size() + 20)), b: std::size_t(50)), b: std::size_t(bn.capacity())); |
| 471 | tangent_numbers_series(m: new_size); |
| 472 | } |
| 473 | m_counter.store(i: static_cast<atomic_integer_type>(bn.size()), BOOST_MATH_ATOMIC_NS::memory_order_release); |
| 474 | } |
| 475 | } |
| 476 | |
| 477 | for(std::size_t i = (std::max)(a: static_cast<std::size_t>(max_bernoulli_b2n<T>::value + 1), b: start); i < start + n; ++i) |
| 478 | { |
| 479 | *out = (i >= m_overflow_limit) ? policies::raise_overflow_error<T>("boost::math::bernoulli_b2n<%1%>(std::size_t)" , 0, T(i), pol) : bn[static_cast<typename container_type::size_type>(i)]; |
| 480 | ++out; |
| 481 | } |
| 482 | |
| 483 | #endif |
| 484 | return out; |
| 485 | } |
| 486 | |
| 487 | template <class OutputIterator> |
| 488 | OutputIterator copy_tangent_numbers(OutputIterator out, std::size_t start, std::size_t n, const Policy& pol) |
| 489 | { |
| 490 | // |
| 491 | // There are basically 3 thread safety options: |
| 492 | // |
| 493 | // 1) There are no threads (BOOST_HAS_THREADS is not defined). |
| 494 | // 2) There are threads, but we do not have a true atomic integer type, |
| 495 | // in this case we just use a mutex to guard against race conditions. |
| 496 | // 3) There are threads, and we have an atomic integer: in this case we can |
| 497 | // use the double-checked locking pattern to avoid thread synchronisation |
| 498 | // when accessing values already in the cache. |
| 499 | // |
| 500 | // |
| 501 | // First off handle the common case for overflow and/or asymptotic expansion: |
| 502 | // |
| 503 | if(start + n > bn.capacity()) |
| 504 | { |
| 505 | if(start < bn.capacity()) |
| 506 | { |
| 507 | out = copy_tangent_numbers(out, start, bn.capacity() - start, pol); |
| 508 | n -= bn.capacity() - start; |
| 509 | start = static_cast<std::size_t>(bn.capacity()); |
| 510 | } |
| 511 | if(start < b2n_overflow_limit<T, Policy>() + 2u) |
| 512 | { |
| 513 | for(; n; ++start, --n) |
| 514 | { |
| 515 | *out = t2n_asymptotic<T, Policy>(static_cast<typename container_type::size_type>(start)); |
| 516 | ++out; |
| 517 | } |
| 518 | } |
| 519 | for(; n; ++start, --n) |
| 520 | { |
| 521 | *out = policies::raise_overflow_error<T>("boost::math::bernoulli_b2n<%1%>(std::size_t)" , 0, T(start), pol); |
| 522 | ++out; |
| 523 | } |
| 524 | return out; |
| 525 | } |
| 526 | #if !defined(BOOST_HAS_THREADS) |
| 527 | // |
| 528 | // Single threaded code, very simple: |
| 529 | // |
| 530 | if(m_current_precision < boost::math::tools::digits<T>()) |
| 531 | { |
| 532 | bn.clear(); |
| 533 | tn.clear(); |
| 534 | m_intermediates.clear(); |
| 535 | m_current_precision = boost::math::tools::digits<T>(); |
| 536 | } |
| 537 | if(start + n >= bn.size()) |
| 538 | { |
| 539 | std::size_t new_size = (std::min)((std::max)((std::max)(start + n, std::size_t(bn.size() + 20)), std::size_t(50)), std::size_t(bn.capacity())); |
| 540 | tangent_numbers_series(new_size); |
| 541 | } |
| 542 | |
| 543 | for(std::size_t i = start; i < start + n; ++i) |
| 544 | { |
| 545 | if(i >= m_overflow_limit) |
| 546 | *out = policies::raise_overflow_error<T>("boost::math::bernoulli_b2n<%1%>(std::size_t)" , 0, T(i), pol); |
| 547 | else |
| 548 | { |
| 549 | if(tools::max_value<T>() * tangent_scale_factor<T>() < tn[static_cast<typename container_type::size_type>(i)]) |
| 550 | *out = policies::raise_overflow_error<T>("boost::math::bernoulli_b2n<%1%>(std::size_t)" , 0, T(i), pol); |
| 551 | else |
| 552 | *out = tn[static_cast<typename container_type::size_type>(i)] / tangent_scale_factor<T>(); |
| 553 | } |
| 554 | ++out; |
| 555 | } |
| 556 | #elif defined(BOOST_MATH_NO_ATOMIC_INT) |
| 557 | // |
| 558 | // We need to grab a mutex every time we get here, for both readers and writers: |
| 559 | // |
| 560 | boost::detail::lightweight_mutex::scoped_lock l(m_mutex); |
| 561 | if(m_current_precision < boost::math::tools::digits<T>()) |
| 562 | { |
| 563 | bn.clear(); |
| 564 | tn.clear(); |
| 565 | m_intermediates.clear(); |
| 566 | m_current_precision = boost::math::tools::digits<T>(); |
| 567 | } |
| 568 | if(start + n >= bn.size()) |
| 569 | { |
| 570 | std::size_t new_size = (std::min)((std::max)((std::max)(start + n, std::size_t(bn.size() + 20)), std::size_t(50)), std::size_t(bn.capacity())); |
| 571 | tangent_numbers_series(new_size); |
| 572 | } |
| 573 | |
| 574 | for(std::size_t i = start; i < start + n; ++i) |
| 575 | { |
| 576 | if(i >= m_overflow_limit) |
| 577 | *out = policies::raise_overflow_error<T>("boost::math::bernoulli_b2n<%1%>(std::size_t)" , 0, T(i), pol); |
| 578 | else |
| 579 | { |
| 580 | if(tools::max_value<T>() * tangent_scale_factor<T>() < tn[static_cast<typename container_type::size_type>(i)]) |
| 581 | *out = policies::raise_overflow_error<T>("boost::math::bernoulli_b2n<%1%>(std::size_t)" , 0, T(i), pol); |
| 582 | else |
| 583 | *out = tn[static_cast<typename container_type::size_type>(i)] / tangent_scale_factor<T>(); |
| 584 | } |
| 585 | ++out; |
| 586 | } |
| 587 | |
| 588 | #else |
| 589 | // |
| 590 | // Double-checked locking pattern, lets us access cached already cached values |
| 591 | // without locking: |
| 592 | // |
| 593 | // Get the counter and see if we need to calculate more constants: |
| 594 | // |
| 595 | if((static_cast<std::size_t>(m_counter.load(BOOST_MATH_ATOMIC_NS::memory_order_consume)) < start + n) |
| 596 | || (static_cast<int>(m_current_precision.load(BOOST_MATH_ATOMIC_NS::memory_order_consume)) < boost::math::tools::digits<T>())) |
| 597 | { |
| 598 | boost::detail::lightweight_mutex::scoped_lock l(m_mutex); |
| 599 | |
| 600 | if((static_cast<std::size_t>(m_counter.load(BOOST_MATH_ATOMIC_NS::memory_order_consume)) < start + n) |
| 601 | || (static_cast<int>(m_current_precision.load(BOOST_MATH_ATOMIC_NS::memory_order_consume)) < boost::math::tools::digits<T>())) |
| 602 | { |
| 603 | if(static_cast<int>(m_current_precision.load(BOOST_MATH_ATOMIC_NS::memory_order_consume)) < boost::math::tools::digits<T>()) |
| 604 | { |
| 605 | bn.clear(); |
| 606 | tn.clear(); |
| 607 | m_intermediates.clear(); |
| 608 | m_counter.store(i: 0, BOOST_MATH_ATOMIC_NS::memory_order_release); |
| 609 | m_current_precision = boost::math::tools::digits<T>(); |
| 610 | } |
| 611 | if(start + n >= bn.size()) |
| 612 | { |
| 613 | std::size_t new_size = (std::min)(a: (std::max)(a: (std::max)(a: start + n, b: std::size_t(bn.size() + 20)), b: std::size_t(50)), b: std::size_t(bn.capacity())); |
| 614 | tangent_numbers_series(m: new_size); |
| 615 | } |
| 616 | m_counter.store(i: static_cast<atomic_integer_type>(bn.size()), BOOST_MATH_ATOMIC_NS::memory_order_release); |
| 617 | } |
| 618 | } |
| 619 | |
| 620 | for(std::size_t i = start; i < start + n; ++i) |
| 621 | { |
| 622 | if(i >= m_overflow_limit) |
| 623 | *out = policies::raise_overflow_error<T>("boost::math::bernoulli_b2n<%1%>(std::size_t)" , 0, T(i), pol); |
| 624 | else |
| 625 | { |
| 626 | if(tools::max_value<T>() * tangent_scale_factor<T>() < tn[static_cast<typename container_type::size_type>(i)]) |
| 627 | *out = policies::raise_overflow_error<T>("boost::math::bernoulli_b2n<%1%>(std::size_t)" , 0, T(i), pol); |
| 628 | else |
| 629 | *out = tn[static_cast<typename container_type::size_type>(i)] / tangent_scale_factor<T>(); |
| 630 | } |
| 631 | ++out; |
| 632 | } |
| 633 | |
| 634 | #endif |
| 635 | return out; |
| 636 | } |
| 637 | |
| 638 | private: |
| 639 | // |
| 640 | // The caches for Bernoulli and tangent numbers, once allocated, |
| 641 | // these must NEVER EVER reallocate as it breaks our thread |
| 642 | // safety guarantees: |
| 643 | // |
| 644 | fixed_vector<T> bn, tn; |
| 645 | std::vector<T> m_intermediates; |
| 646 | // The value at which we know overflow has already occurred for the Bn: |
| 647 | std::size_t m_overflow_limit; |
| 648 | #if !defined(BOOST_HAS_THREADS) |
| 649 | int m_current_precision; |
| 650 | #elif defined(BOOST_MATH_NO_ATOMIC_INT) |
| 651 | boost::detail::lightweight_mutex m_mutex; |
| 652 | int m_current_precision; |
| 653 | #else |
| 654 | boost::detail::lightweight_mutex m_mutex; |
| 655 | atomic_counter_type m_counter, m_current_precision; |
| 656 | #endif |
| 657 | }; |
| 658 | |
| 659 | template <class T, class Policy> |
| 660 | inline bernoulli_numbers_cache<T, Policy>& get_bernoulli_numbers_cache() |
| 661 | { |
| 662 | // |
| 663 | // Force this function to be called at program startup so all the static variables |
| 664 | // get initialized then (thread safety). |
| 665 | // |
| 666 | bernoulli_initializer<T, Policy>::force_instantiate(); |
| 667 | static bernoulli_numbers_cache<T, Policy> data; |
| 668 | return data; |
| 669 | } |
| 670 | |
| 671 | }}} |
| 672 | |
| 673 | #endif // BOOST_MATH_BERNOULLI_DETAIL_HPP |
| 674 | |