xref: /third_party/jerryscript/jerry-libm/pow.c (revision 425bb815)
1/* Copyright JS Foundation and other contributors, http://js.foundation
2 *
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at
6 *
7 *     http://www.apache.org/licenses/LICENSE-2.0
8 *
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
14 *
15 * This file is based on work under the following copyright and permission
16 * notice:
17 *
18 *     Copyright (C) 2004 by Sun Microsystems, Inc. All rights reserved.
19 *
20 *     Permission to use, copy, modify, and distribute this
21 *     software is freely granted, provided that this notice
22 *     is preserved.
23 *
24 *     @(#)e_pow.c 1.5 04/04/22
25 */
26
27#include "jerry-libm-internal.h"
28
29/* pow(x,y) return x**y
30 *
31 *                    n
32 * Method:  Let x =  2   * (1+f)
33 *      1. Compute and return log2(x) in two pieces:
34 *              log2(x) = w1 + w2,
35 *         where w1 has 53-24 = 29 bit trailing zeros.
36 *      2. Perform y*log2(x) = n+y' by simulating muti-precision
37 *         arithmetic, where |y'|<=0.5.
38 *      3. Return x**y = 2**n*exp(y'*log2)
39 *
40 * Special cases:
41 *      0.  +1 ** (anything) is 1
42 *      1.  (anything) ** 0  is 1
43 *      2.  (anything) ** 1  is itself
44 *      3.  (anything) ** NAN is NAN
45 *      4.  NAN ** (anything except 0) is NAN
46 *      5.  +-(|x| > 1) **  +INF is +INF
47 *      6.  +-(|x| > 1) **  -INF is +0
48 *      7.  +-(|x| < 1) **  +INF is +0
49 *      8.  +-(|x| < 1) **  -INF is +INF
50 *      9.  -1          ** +-INF is 1
51 *      10. +0 ** (+anything except 0, NAN)               is +0
52 *      11. -0 ** (+anything except 0, NAN, odd integer)  is +0
53 *      12. +0 ** (-anything except 0, NAN)               is +INF
54 *      13. -0 ** (-anything except 0, NAN, odd integer)  is +INF
55 *      14. -0 ** (odd integer) = -( +0 ** (odd integer) )
56 *      15. +INF ** (+anything except 0,NAN) is +INF
57 *      16. +INF ** (-anything except 0,NAN) is +0
58 *      17. -INF ** (anything)  = -0 ** (-anything)
59 *      18. (-anything) ** (integer) is (-1)**(integer)*(+anything**integer)
60 *      19. (-anything except 0 and inf) ** (non-integer) is NAN
61 *
62 * Accuracy:
63 *      pow(x,y) returns x**y nearly rounded. In particular
64 *                      pow(integer,integer)
65 *      always returns the correct integer provided it is
66 *      representable.
67 *
68 * Constants:
69 * The hexadecimal values are the intended ones for the following
70 * constants. The decimal values may be used, provided that the
71 * compiler will convert from decimal to binary accurately enough
72 * to produce the hexadecimal values shown.
73 */
74
75static const double bp[] =
76{
77  1.0,
78  1.5,
79};
80static const double dp_h[] =
81{
82  0.0,
83  5.84962487220764160156e-01, /* 0x3FE2B803, 0x40000000 */
84};
85static const double dp_l[] =
86{
87  0.0,
88  1.35003920212974897128e-08, /* 0x3E4CFDEB, 0x43CFD006 */
89};
90
91#define zero     0.0
92#define one      1.0
93#define two      2.0
94#define two53    9007199254740992.0 /* 0x43400000, 0x00000000 */
95#define huge     1.0e300
96#define tiny     1.0e-300
97/* poly coefs for (3/2) * (log(x) - 2s - 2/3 * s**3 */
98#define L1       5.99999999999994648725e-01 /* 0x3FE33333, 0x33333303 */
99#define L2       4.28571428578550184252e-01 /* 0x3FDB6DB6, 0xDB6FABFF */
100#define L3       3.33333329818377432918e-01 /* 0x3FD55555, 0x518F264D */
101#define L4       2.72728123808534006489e-01 /* 0x3FD17460, 0xA91D4101 */
102#define L5       2.30660745775561754067e-01 /* 0x3FCD864A, 0x93C9DB65 */
103#define L6       2.06975017800338417784e-01 /* 0x3FCA7E28, 0x4A454EEF */
104#define P1       1.66666666666666019037e-01 /* 0x3FC55555, 0x5555553E */
105#define P2      -2.77777777770155933842e-03 /* 0xBF66C16C, 0x16BEBD93 */
106#define P3       6.61375632143793436117e-05 /* 0x3F11566A, 0xAF25DE2C */
107#define P4      -1.65339022054652515390e-06 /* 0xBEBBBD41, 0xC5D26BF1 */
108#define P5       4.13813679705723846039e-08 /* 0x3E663769, 0x72BEA4D0 */
109#define lg2      6.93147180559945286227e-01 /* 0x3FE62E42, 0xFEFA39EF */
110#define lg2_h    6.93147182464599609375e-01 /* 0x3FE62E43, 0x00000000 */
111#define lg2_l   -1.90465429995776804525e-09 /* 0xBE205C61, 0x0CA86C39 */
112#define ovt      8.0085662595372944372e-0017 /* -(1024-log2(ovfl+.5ulp)) */
113#define cp       9.61796693925975554329e-01 /* 0x3FEEC709, 0xDC3A03FD = 2 / (3 ln2) */
114#define cp_h     9.61796700954437255859e-01 /* 0x3FEEC709, 0xE0000000 = (float) cp */
115#define cp_l    -7.02846165095275826516e-09 /* 0xBE3E2FE0, 0x145B01F5 = tail of cp_h */
116#define ivln2    1.44269504088896338700e+00 /* 0x3FF71547, 0x652B82FE = 1 / ln2 */
117#define ivln2_h  1.44269502162933349609e+00 /* 0x3FF71547, 0x60000000 = 24b 1 / ln2 */
118#define ivln2_l  1.92596299112661746887e-08 /* 0x3E54AE0B, 0xF85DDF44 = 1 / ln2 tail */
119
120double
121pow (double x, double y)
122{
123  double_accessor t1, ax, p_h, y1, t, z;
124  double z_h, z_l, p_l;
125  double t2, r, s, u, v, w;
126  int i, j, k, yisint, n;
127  int hx, hy, ix, iy;
128  unsigned lx, ly;
129
130  hx = __HI (x);
131  lx = __LO (x);
132  hy = __HI (y);
133  ly = __LO (y);
134  ix = hx & 0x7fffffff;
135  iy = hy & 0x7fffffff;
136
137  /* x == one: 1**y = 1 */
138  if (((hx - 0x3ff00000) | lx) == 0)
139  {
140    return one;
141  }
142
143  /* y == zero: x**0 = 1 */
144  if ((iy | ly) == 0)
145  {
146    return one;
147  }
148
149  /* +-NaN return x + y */
150  if (ix > 0x7ff00000 || ((ix == 0x7ff00000) && (lx != 0)) || iy > 0x7ff00000 || ((iy == 0x7ff00000) && (ly != 0)))
151  {
152    return x + y;
153  }
154
155  /* determine if y is an odd int when x < 0
156   * yisint = 0 ... y is not an integer
157   * yisint = 1 ... y is an odd int
158   * yisint = 2 ... y is an even int
159   */
160  yisint = 0;
161  if (hx < 0)
162  {
163    if (iy >= 0x43400000) /* even integer y */
164    {
165      yisint = 2;
166    }
167    else if (iy >= 0x3ff00000)
168    {
169      k = (iy >> 20) - 0x3ff; /* exponent */
170      if (k > 20)
171      {
172        j = ly >> (52 - k);
173        if ((j << (52 - k)) == ly)
174        {
175          yisint = 2 - (j & 1);
176        }
177      }
178      else if (ly == 0)
179      {
180        j = iy >> (20 - k);
181        if ((j << (20 - k)) == iy)
182        {
183          yisint = 2 - (j & 1);
184        }
185      }
186    }
187  }
188
189  /* special value of y */
190  if (ly == 0)
191  {
192    if (iy == 0x7ff00000) /* y is +-inf */
193    {
194      if (((ix - 0x3ff00000) | lx) == 0) /* +-1**+-inf is 1 */
195      {
196        return one;
197      }
198      else if (ix >= 0x3ff00000) /* (|x|>1)**+-inf = inf,0 */
199      {
200        return (hy >= 0) ? y : zero;
201      }
202      else /* (|x|<1)**-,+inf = inf,0 */
203      {
204        return (hy < 0) ? -y : zero;
205      }
206    }
207    if (iy == 0x3ff00000) /* y is +-1 */
208    {
209      if (hy < 0)
210      {
211        return one / x;
212      }
213      else
214      {
215        return x;
216      }
217    }
218    if (hy == 0x40000000) /* y is 2 */
219    {
220      return x * x;
221    }
222    if (hy == 0x3fe00000) /* y is 0.5 */
223    {
224      if (hx >= 0) /* x >= +0 */
225      {
226        return sqrt (x);
227      }
228    }
229  }
230
231  ax.dbl = fabs (x);
232  /* special value of x */
233  if (lx == 0)
234  {
235    if (ix == 0x7ff00000 || ix == 0 || ix == 0x3ff00000)
236    {
237      z.dbl = ax.dbl; /* x is +-0,+-inf,+-1 */
238      if (hy < 0)
239      {
240        z.dbl = one / z.dbl; /* z = (1 / |x|) */
241      }
242      if (hx < 0)
243      {
244        if (((ix - 0x3ff00000) | yisint) == 0)
245        {
246          z.dbl = NAN; /* (-1)**non-int is NaN */
247        }
248        else if (yisint == 1)
249        {
250          z.dbl = -z.dbl; /* (x<0)**odd = -(|x|**odd) */
251        }
252      }
253      return z.dbl;
254    }
255  }
256
257  n = (hx < 0) ? 0 : 1;
258
259  /* (x<0)**(non-int) is NaN */
260  if ((n | yisint) == 0)
261  {
262    return NAN;
263  }
264
265  s = one; /* s (sign of result -ve**odd) = -1 else = 1 */
266  if ((n | (yisint - 1)) == 0)
267  {
268    s = -one; /* (-ve)**(odd int) */
269  }
270
271  /* |y| is huge */
272  if (iy > 0x41e00000) /* if |y| > 2**31 */
273  {
274    if (iy > 0x43f00000) /* if |y| > 2**64, must o/uflow */
275    {
276      if (ix <= 0x3fefffff)
277      {
278        return (hy < 0) ? huge * huge : tiny * tiny;
279      }
280      if (ix >= 0x3ff00000)
281      {
282        return (hy > 0) ? huge * huge : tiny * tiny;
283      }
284    }
285    /* over/underflow if x is not close to one */
286    if (ix < 0x3fefffff)
287    {
288      return (hy < 0) ? s * huge * huge : s * tiny * tiny;
289    }
290    if (ix > 0x3ff00000)
291    {
292      return (hy > 0) ? s * huge * huge : s * tiny * tiny;
293    }
294    /* now |1 - x| is tiny <= 2**-20, suffice to compute
295       log(x) by x - x^2 / 2 + x^3 / 3 - x^4 / 4 */
296    t.dbl = ax.dbl - one; /* t has 20 trailing zeros */
297    w = (t.dbl * t.dbl) * (0.5 - t.dbl * (0.3333333333333333333333 - t.dbl * 0.25));
298    u = ivln2_h * t.dbl; /* ivln2_h has 21 sig. bits */
299    v = t.dbl * ivln2_l - w * ivln2;
300    t1.dbl = u + v;
301    t1.as_int.lo = 0;
302    t2 = v - (t1.dbl - u);
303  }
304  else
305  {
306    double_accessor s_h, t_h;
307    double ss, s2, s_l, t_l;
308
309    n = 0;
310    /* take care subnormal number */
311    if (ix < 0x00100000)
312    {
313      ax.dbl *= two53;
314      n -= 53;
315      ix = ax.as_int.hi;
316    }
317    n += ((ix) >> 20) - 0x3ff;
318    j = ix & 0x000fffff;
319    /* determine interval */
320    ix = j | 0x3ff00000; /* normalize ix */
321    if (j <= 0x3988E) /* |x| < sqrt(3/2) */
322    {
323      k = 0;
324    }
325    else if (j < 0xBB67A) /* |x| < sqrt(3) */
326    {
327      k = 1;
328    }
329    else
330    {
331      k = 0;
332      n += 1;
333      ix -= 0x00100000;
334    }
335    ax.as_int.hi = ix;
336
337    /* compute ss = s_h + s_l = (x - 1) / (x + 1) or (x - 1.5) / (x + 1.5) */
338    u = ax.dbl - bp[k]; /* bp[0] = 1.0, bp[1] = 1.5 */
339    v = one / (ax.dbl + bp[k]);
340    ss = u * v;
341    s_h.dbl = ss;
342    s_h.as_int.lo = 0;
343    /* t_h = ax + bp[k] High */
344    t_h.dbl = zero;
345    t_h.as_int.hi = ((ix >> 1) | 0x20000000) + 0x00080000 + (k << 18);
346    t_l = ax.dbl - (t_h.dbl - bp[k]);
347    s_l = v * ((u - s_h.dbl * t_h.dbl) - s_h.dbl * t_l);
348    /* compute log(ax) */
349    s2 = ss * ss;
350    r = s2 * s2 * (L1 + s2 * (L2 + s2 * (L3 + s2 * (L4 + s2 * (L5 + s2 * L6)))));
351    r += s_l * (s_h.dbl + ss);
352    s2 = s_h.dbl * s_h.dbl;
353    t_h.dbl = 3.0 + s2 + r;
354    t_h.as_int.lo = 0;
355    t_l = r - ((t_h.dbl - 3.0) - s2);
356    /* u + v = ss * (1 + ...) */
357    u = s_h.dbl * t_h.dbl;
358    v = s_l * t_h.dbl + t_l * ss;
359    /* 2 / (3 * log2) * (ss + ...) */
360    p_h.dbl = u + v;
361    p_h.as_int.lo = 0;
362    p_l = v - (p_h.dbl - u);
363    z_h = cp_h * p_h.dbl; /* cp_h + cp_l = 2 / (3 * log2) */
364    z_l = cp_l * p_h.dbl + p_l * cp + dp_l[k];
365    /* log2(ax) = (ss + ...) * 2 / (3 * log2) = n + dp_h + z_h + z_l */
366    t.dbl = (double) n;
367    t1.dbl = (((z_h + z_l) + dp_h[k]) + t.dbl);
368    t1.as_int.lo = 0;
369    t2 = z_l - (((t1.dbl - t.dbl) - dp_h[k]) - z_h);
370  }
371
372  /* split up y into y1 + y2 and compute (y1 + y2) * (t1 + t2) */
373  y1.dbl = y;
374  y1.as_int.lo = 0;
375  p_l = (y - y1.dbl) * t1.dbl + y * t2;
376  p_h.dbl = y1.dbl * t1.dbl;
377  z.dbl = p_l + p_h.dbl;
378  j = z.as_int.hi;
379  i = z.as_int.lo;
380  if (j >= 0x40900000) /* z >= 1024 */
381  {
382    if (((j - 0x40900000) | i) != 0) /* if z > 1024 */
383    {
384      return s * huge * huge; /* overflow */
385    }
386    else
387    {
388      if (p_l + ovt > z.dbl - p_h.dbl)
389      {
390        return s * huge * huge; /* overflow */
391      }
392    }
393  }
394  else if ((j & 0x7fffffff) >= 0x4090cc00) /* z <= -1075 */
395  {
396    if (((j - 0xc090cc00) | i) != 0) /* z < -1075 */
397    {
398      return s * tiny * tiny; /* underflow */
399    }
400    else
401    {
402      if (p_l <= z.dbl - p_h.dbl)
403      {
404        return s * tiny * tiny; /* underflow */
405      }
406    }
407  }
408  /*
409   * compute 2**(p_h + p_l)
410   */
411  i = j & 0x7fffffff;
412  k = (i >> 20) - 0x3ff;
413  n = 0;
414  if (i > 0x3fe00000) /* if |z| > 0.5, set n = [z + 0.5] */
415  {
416    n = j + (0x00100000 >> (k + 1));
417    k = ((n & 0x7fffffff) >> 20) - 0x3ff; /* new k for n */
418    t.dbl = zero;
419    t.as_int.hi = (n & ~(0x000fffff >> k));
420    n = ((n & 0x000fffff) | 0x00100000) >> (20 - k);
421    if (j < 0)
422    {
423      n = -n;
424    }
425    p_h.dbl -= t.dbl;
426  }
427  t.dbl = p_l + p_h.dbl;
428  t.as_int.lo = 0;
429  u = t.dbl * lg2_h;
430  v = (p_l - (t.dbl - p_h.dbl)) * lg2 + t.dbl * lg2_l;
431  z.dbl = u + v;
432  w = v - (z.dbl - u);
433  t.dbl = z.dbl * z.dbl;
434  t1.dbl = z.dbl - t.dbl * (P1 + t.dbl * (P2 + t.dbl * (P3 + t.dbl * (P4 + t.dbl * P5))));
435  r = (z.dbl * t1.dbl) / (t1.dbl - two) - (w + z.dbl * w);
436  z.dbl = one - (r - z.dbl);
437  j = z.as_int.hi;
438  j += (n << 20);
439  if ((j >> 20) <= 0) /* subnormal output */
440  {
441    z.dbl = scalbn (z.dbl, n);
442  }
443  else
444  {
445    z.as_int.hi += (n << 20);
446  }
447  return s * z.dbl;
448} /* pow */
449
450#undef zero
451#undef one
452#undef two
453#undef two53
454#undef huge
455#undef tiny
456#undef L1
457#undef L2
458#undef L3
459#undef L4
460#undef L5
461#undef L6
462#undef P1
463#undef P2
464#undef P3
465#undef P4
466#undef P5
467#undef lg2
468#undef lg2_h
469#undef lg2_l
470#undef ovt
471#undef cp
472#undef cp_h
473#undef cp_l
474#undef ivln2
475#undef ivln2_h
476#undef ivln2_l
477