OpenCores
URL https://opencores.org/ocsvn/openrisc/openrisc/trunk

Subversion Repositories openrisc

[/] [openrisc/] [trunk/] [gnu-dev/] [or1k-gcc/] [libstdc++-v3/] [include/] [tr1/] [gamma.tcc] - Blame information for rev 742

Details | Compare with Previous | View Log

Line No. Rev Author Line
1 742 jeremybenn
// Special functions -*- C++ -*-
2
 
3
// Copyright (C) 2006, 2007, 2008, 2009, 2010
4
// Free Software Foundation, Inc.
5
//
6
// This file is part of the GNU ISO C++ Library.  This library is free
7
// software; you can redistribute it and/or modify it under the
8
// terms of the GNU General Public License as published by the
9
// Free Software Foundation; either version 3, or (at your option)
10
// any later version.
11
//
12
// This library is distributed in the hope that it will be useful,
13
// but WITHOUT ANY WARRANTY; without even the implied warranty of
14
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15
// GNU General Public License for more details.
16
//
17
// Under Section 7 of GPL version 3, you are granted additional
18
// permissions described in the GCC Runtime Library Exception, version
19
// 3.1, as published by the Free Software Foundation.
20
 
21
// You should have received a copy of the GNU General Public License and
22
// a copy of the GCC Runtime Library Exception along with this program;
23
// see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
24
// .
25
 
26
/** @file tr1/gamma.tcc
27
 *  This is an internal header file, included by other library headers.
28
 *  Do not attempt to use it directly. @headername{tr1/cmath}
29
 */
30
 
31
//
32
// ISO C++ 14882 TR1: 5.2  Special functions
33
//
34
 
35
// Written by Edward Smith-Rowland based on:
36
//   (1) Handbook of Mathematical Functions,
37
//       ed. Milton Abramowitz and Irene A. Stegun,
38
//       Dover Publications,
39
//       Section 6, pp. 253-266
40
//   (2) The Gnu Scientific Library, http://www.gnu.org/software/gsl
41
//   (3) Numerical Recipes in C, by W. H. Press, S. A. Teukolsky,
42
//       W. T. Vetterling, B. P. Flannery, Cambridge University Press (1992),
43
//       2nd ed, pp. 213-216
44
//   (4) Gamma, Exploring Euler's Constant, Julian Havil,
45
//       Princeton, 2003.
46
 
47
#ifndef _GLIBCXX_TR1_GAMMA_TCC
48
#define _GLIBCXX_TR1_GAMMA_TCC 1
49
 
50
#include "special_function_util.h"
51
 
52
namespace std _GLIBCXX_VISIBILITY(default)
53
{
54
namespace tr1
55
{
56
  // Implementation-space details.
57
  namespace __detail
58
  {
59
  _GLIBCXX_BEGIN_NAMESPACE_VERSION
60
 
61
    /**
62
     *   @brief This returns Bernoulli numbers from a table or by summation
63
     *          for larger values.
64
     *
65
     *   Recursion is unstable.
66
     *
67
     *   @param __n the order n of the Bernoulli number.
68
     *   @return  The Bernoulli number of order n.
69
     */
70
    template 
71
    _Tp __bernoulli_series(unsigned int __n)
72
    {
73
 
74
      static const _Tp __num[28] = {
75
        _Tp(1UL),                        -_Tp(1UL) / _Tp(2UL),
76
        _Tp(1UL) / _Tp(6UL),             _Tp(0UL),
77
        -_Tp(1UL) / _Tp(30UL),           _Tp(0UL),
78
        _Tp(1UL) / _Tp(42UL),            _Tp(0UL),
79
        -_Tp(1UL) / _Tp(30UL),           _Tp(0UL),
80
        _Tp(5UL) / _Tp(66UL),            _Tp(0UL),
81
        -_Tp(691UL) / _Tp(2730UL),       _Tp(0UL),
82
        _Tp(7UL) / _Tp(6UL),             _Tp(0UL),
83
        -_Tp(3617UL) / _Tp(510UL),       _Tp(0UL),
84
        _Tp(43867UL) / _Tp(798UL),       _Tp(0UL),
85
        -_Tp(174611) / _Tp(330UL),       _Tp(0UL),
86
        _Tp(854513UL) / _Tp(138UL),      _Tp(0UL),
87
        -_Tp(236364091UL) / _Tp(2730UL), _Tp(0UL),
88
        _Tp(8553103UL) / _Tp(6UL),       _Tp(0UL)
89
      };
90
 
91
      if (__n == 0)
92
        return _Tp(1);
93
 
94
      if (__n == 1)
95
        return -_Tp(1) / _Tp(2);
96
 
97
      //  Take care of the rest of the odd ones.
98
      if (__n % 2 == 1)
99
        return _Tp(0);
100
 
101
      //  Take care of some small evens that are painful for the series.
102
      if (__n < 28)
103
        return __num[__n];
104
 
105
 
106
      _Tp __fact = _Tp(1);
107
      if ((__n / 2) % 2 == 0)
108
        __fact *= _Tp(-1);
109
      for (unsigned int __k = 1; __k <= __n; ++__k)
110
        __fact *= __k / (_Tp(2) * __numeric_constants<_Tp>::__pi());
111
      __fact *= _Tp(2);
112
 
113
      _Tp __sum = _Tp(0);
114
      for (unsigned int __i = 1; __i < 1000; ++__i)
115
        {
116
          _Tp __term = std::pow(_Tp(__i), -_Tp(__n));
117
          if (__term < std::numeric_limits<_Tp>::epsilon())
118
            break;
119
          __sum += __term;
120
        }
121
 
122
      return __fact * __sum;
123
    }
124
 
125
 
126
    /**
127
     *   @brief This returns Bernoulli number \f$B_n\f$.
128
     *
129
     *   @param __n the order n of the Bernoulli number.
130
     *   @return  The Bernoulli number of order n.
131
     */
132
    template
133
    inline _Tp
134
    __bernoulli(const int __n)
135
    {
136
      return __bernoulli_series<_Tp>(__n);
137
    }
138
 
139
 
140
    /**
141
     *   @brief Return \f$log(\Gamma(x))\f$ by asymptotic expansion
142
     *          with Bernoulli number coefficients.  This is like
143
     *          Sterling's approximation.
144
     *
145
     *   @param __x The argument of the log of the gamma function.
146
     *   @return  The logarithm of the gamma function.
147
     */
148
    template
149
    _Tp
150
    __log_gamma_bernoulli(const _Tp __x)
151
    {
152
      _Tp __lg = (__x - _Tp(0.5L)) * std::log(__x) - __x
153
               + _Tp(0.5L) * std::log(_Tp(2)
154
               * __numeric_constants<_Tp>::__pi());
155
 
156
      const _Tp __xx = __x * __x;
157
      _Tp __help = _Tp(1) / __x;
158
      for ( unsigned int __i = 1; __i < 20; ++__i )
159
        {
160
          const _Tp __2i = _Tp(2 * __i);
161
          __help /= __2i * (__2i - _Tp(1)) * __xx;
162
          __lg += __bernoulli<_Tp>(2 * __i) * __help;
163
        }
164
 
165
      return __lg;
166
    }
167
 
168
 
169
    /**
170
     *   @brief Return \f$log(\Gamma(x))\f$ by the Lanczos method.
171
     *          This method dominates all others on the positive axis I think.
172
     *
173
     *   @param __x The argument of the log of the gamma function.
174
     *   @return  The logarithm of the gamma function.
175
     */
176
    template
177
    _Tp
178
    __log_gamma_lanczos(const _Tp __x)
179
    {
180
      const _Tp __xm1 = __x - _Tp(1);
181
 
182
      static const _Tp __lanczos_cheb_7[9] = {
183
       _Tp( 0.99999999999980993227684700473478L),
184
       _Tp( 676.520368121885098567009190444019L),
185
       _Tp(-1259.13921672240287047156078755283L),
186
       _Tp( 771.3234287776530788486528258894L),
187
       _Tp(-176.61502916214059906584551354L),
188
       _Tp( 12.507343278686904814458936853L),
189
       _Tp(-0.13857109526572011689554707L),
190
       _Tp( 9.984369578019570859563e-6L),
191
       _Tp( 1.50563273514931155834e-7L)
192
      };
193
 
194
      static const _Tp __LOGROOT2PI
195
          = _Tp(0.9189385332046727417803297364056176L);
196
 
197
      _Tp __sum = __lanczos_cheb_7[0];
198
      for(unsigned int __k = 1; __k < 9; ++__k)
199
        __sum += __lanczos_cheb_7[__k] / (__xm1 + __k);
200
 
201
      const _Tp __term1 = (__xm1 + _Tp(0.5L))
202
                        * std::log((__xm1 + _Tp(7.5L))
203
                       / __numeric_constants<_Tp>::__euler());
204
      const _Tp __term2 = __LOGROOT2PI + std::log(__sum);
205
      const _Tp __result = __term1 + (__term2 - _Tp(7));
206
 
207
      return __result;
208
    }
209
 
210
 
211
    /**
212
     *   @brief Return \f$ log(|\Gamma(x)|) \f$.
213
     *          This will return values even for \f$ x < 0 \f$.
214
     *          To recover the sign of \f$ \Gamma(x) \f$ for
215
     *          any argument use @a __log_gamma_sign.
216
     *
217
     *   @param __x The argument of the log of the gamma function.
218
     *   @return  The logarithm of the gamma function.
219
     */
220
    template
221
    _Tp
222
    __log_gamma(const _Tp __x)
223
    {
224
      if (__x > _Tp(0.5L))
225
        return __log_gamma_lanczos(__x);
226
      else
227
        {
228
          const _Tp __sin_fact
229
                 = std::abs(std::sin(__numeric_constants<_Tp>::__pi() * __x));
230
          if (__sin_fact == _Tp(0))
231
            std::__throw_domain_error(__N("Argument is nonpositive integer "
232
                                          "in __log_gamma"));
233
          return __numeric_constants<_Tp>::__lnpi()
234
                     - std::log(__sin_fact)
235
                     - __log_gamma_lanczos(_Tp(1) - __x);
236
        }
237
    }
238
 
239
 
240
    /**
241
     *   @brief Return the sign of \f$ \Gamma(x) \f$.
242
     *          At nonpositive integers zero is returned.
243
     *
244
     *   @param __x The argument of the gamma function.
245
     *   @return  The sign of the gamma function.
246
     */
247
    template
248
    _Tp
249
    __log_gamma_sign(const _Tp __x)
250
    {
251
      if (__x > _Tp(0))
252
        return _Tp(1);
253
      else
254
        {
255
          const _Tp __sin_fact
256
                  = std::sin(__numeric_constants<_Tp>::__pi() * __x);
257
          if (__sin_fact > _Tp(0))
258
            return (1);
259
          else if (__sin_fact < _Tp(0))
260
            return -_Tp(1);
261
          else
262
            return _Tp(0);
263
        }
264
    }
265
 
266
 
267
    /**
268
     *   @brief Return the logarithm of the binomial coefficient.
269
     *   The binomial coefficient is given by:
270
     *   @f[
271
     *   \left(  \right) = \frac{n!}{(n-k)! k!}
272
     *   @f]
273
     *
274
     *   @param __n The first argument of the binomial coefficient.
275
     *   @param __k The second argument of the binomial coefficient.
276
     *   @return  The binomial coefficient.
277
     */
278
    template
279
    _Tp
280
    __log_bincoef(const unsigned int __n, const unsigned int __k)
281
    {
282
      //  Max e exponent before overflow.
283
      static const _Tp __max_bincoeff
284
                      = std::numeric_limits<_Tp>::max_exponent10
285
                      * std::log(_Tp(10)) - _Tp(1);
286
#if _GLIBCXX_USE_C99_MATH_TR1
287
      _Tp __coeff =  std::tr1::lgamma(_Tp(1 + __n))
288
                  - std::tr1::lgamma(_Tp(1 + __k))
289
                  - std::tr1::lgamma(_Tp(1 + __n - __k));
290
#else
291
      _Tp __coeff =  __log_gamma(_Tp(1 + __n))
292
                  - __log_gamma(_Tp(1 + __k))
293
                  - __log_gamma(_Tp(1 + __n - __k));
294
#endif
295
    }
296
 
297
 
298
    /**
299
     *   @brief Return the binomial coefficient.
300
     *   The binomial coefficient is given by:
301
     *   @f[
302
     *   \left(  \right) = \frac{n!}{(n-k)! k!}
303
     *   @f]
304
     *
305
     *   @param __n The first argument of the binomial coefficient.
306
     *   @param __k The second argument of the binomial coefficient.
307
     *   @return  The binomial coefficient.
308
     */
309
    template
310
    _Tp
311
    __bincoef(const unsigned int __n, const unsigned int __k)
312
    {
313
      //  Max e exponent before overflow.
314
      static const _Tp __max_bincoeff
315
                      = std::numeric_limits<_Tp>::max_exponent10
316
                      * std::log(_Tp(10)) - _Tp(1);
317
 
318
      const _Tp __log_coeff = __log_bincoef<_Tp>(__n, __k);
319
      if (__log_coeff > __max_bincoeff)
320
        return std::numeric_limits<_Tp>::quiet_NaN();
321
      else
322
        return std::exp(__log_coeff);
323
    }
324
 
325
 
326
    /**
327
     *   @brief Return \f$ \Gamma(x) \f$.
328
     *
329
     *   @param __x The argument of the gamma function.
330
     *   @return  The gamma function.
331
     */
332
    template
333
    inline _Tp
334
    __gamma(const _Tp __x)
335
    {
336
      return std::exp(__log_gamma(__x));
337
    }
338
 
339
 
340
    /**
341
     *   @brief  Return the digamma function by series expansion.
342
     *   The digamma or @f$ \psi(x) @f$ function is defined by
343
     *   @f[
344
     *     \psi(x) = \frac{\Gamma'(x)}{\Gamma(x)}
345
     *   @f]
346
     *
347
     *   The series is given by:
348
     *   @f[
349
     *     \psi(x) = -\gamma_E - \frac{1}{x}
350
     *              \sum_{k=1}^{\infty} \frac{x}{k(x + k)}
351
     *   @f]
352
     */
353
    template
354
    _Tp
355
    __psi_series(const _Tp __x)
356
    {
357
      _Tp __sum = -__numeric_constants<_Tp>::__gamma_e() - _Tp(1) / __x;
358
      const unsigned int __max_iter = 100000;
359
      for (unsigned int __k = 1; __k < __max_iter; ++__k)
360
        {
361
          const _Tp __term = __x / (__k * (__k + __x));
362
          __sum += __term;
363
          if (std::abs(__term / __sum) < std::numeric_limits<_Tp>::epsilon())
364
            break;
365
        }
366
      return __sum;
367
    }
368
 
369
 
370
    /**
371
     *   @brief  Return the digamma function for large argument.
372
     *   The digamma or @f$ \psi(x) @f$ function is defined by
373
     *   @f[
374
     *     \psi(x) = \frac{\Gamma'(x)}{\Gamma(x)}
375
     *   @f]
376
     *
377
     *   The asymptotic series is given by:
378
     *   @f[
379
     *     \psi(x) = \ln(x) - \frac{1}{2x}
380
     *             - \sum_{n=1}^{\infty} \frac{B_{2n}}{2 n x^{2n}}
381
     *   @f]
382
     */
383
    template
384
    _Tp
385
    __psi_asymp(const _Tp __x)
386
    {
387
      _Tp __sum = std::log(__x) - _Tp(0.5L) / __x;
388
      const _Tp __xx = __x * __x;
389
      _Tp __xp = __xx;
390
      const unsigned int __max_iter = 100;
391
      for (unsigned int __k = 1; __k < __max_iter; ++__k)
392
        {
393
          const _Tp __term = __bernoulli<_Tp>(2 * __k) / (2 * __k * __xp);
394
          __sum -= __term;
395
          if (std::abs(__term / __sum) < std::numeric_limits<_Tp>::epsilon())
396
            break;
397
          __xp *= __xx;
398
        }
399
      return __sum;
400
    }
401
 
402
 
403
    /**
404
     *   @brief  Return the digamma function.
405
     *   The digamma or @f$ \psi(x) @f$ function is defined by
406
     *   @f[
407
     *     \psi(x) = \frac{\Gamma'(x)}{\Gamma(x)}
408
     *   @f]
409
     *   For negative argument the reflection formula is used:
410
     *   @f[
411
     *     \psi(x) = \psi(1-x) - \pi \cot(\pi x)
412
     *   @f]
413
     */
414
    template
415
    _Tp
416
    __psi(const _Tp __x)
417
    {
418
      const int __n = static_cast(__x + 0.5L);
419
      const _Tp __eps = _Tp(4) * std::numeric_limits<_Tp>::epsilon();
420
      if (__n <= 0 && std::abs(__x - _Tp(__n)) < __eps)
421
        return std::numeric_limits<_Tp>::quiet_NaN();
422
      else if (__x < _Tp(0))
423
        {
424
          const _Tp __pi = __numeric_constants<_Tp>::__pi();
425
          return __psi(_Tp(1) - __x)
426
               - __pi * std::cos(__pi * __x) / std::sin(__pi * __x);
427
        }
428
      else if (__x > _Tp(100))
429
        return __psi_asymp(__x);
430
      else
431
        return __psi_series(__x);
432
    }
433
 
434
 
435
    /**
436
     *   @brief  Return the polygamma function @f$ \psi^{(n)}(x) @f$.
437
     *
438
     *   The polygamma function is related to the Hurwitz zeta function:
439
     *   @f[
440
     *     \psi^{(n)}(x) = (-1)^{n+1} m! \zeta(m+1,x)
441
     *   @f]
442
     */
443
    template
444
    _Tp
445
    __psi(const unsigned int __n, const _Tp __x)
446
    {
447
      if (__x <= _Tp(0))
448
        std::__throw_domain_error(__N("Argument out of range "
449
                                      "in __psi"));
450
      else if (__n == 0)
451
        return __psi(__x);
452
      else
453
        {
454
          const _Tp __hzeta = __hurwitz_zeta(_Tp(__n + 1), __x);
455
#if _GLIBCXX_USE_C99_MATH_TR1
456
          const _Tp __ln_nfact = std::tr1::lgamma(_Tp(__n + 1));
457
#else
458
          const _Tp __ln_nfact = __log_gamma(_Tp(__n + 1));
459
#endif
460
          _Tp __result = std::exp(__ln_nfact) * __hzeta;
461
          if (__n % 2 == 1)
462
            __result = -__result;
463
          return __result;
464
        }
465
    }
466
 
467
  _GLIBCXX_END_NAMESPACE_VERSION
468
  } // namespace std::tr1::__detail
469
}
470
}
471
 
472
#endif // _GLIBCXX_TR1_GAMMA_TCC
473
 

powered by: WebSVN 2.1.0

© copyright 1999-2024 OpenCores.org, equivalent to Oliscience, all rights reserved. OpenCores®, registered trademark.