aboutsummaryrefslogtreecommitdiff
path: root/sysdeps/libm-ieee754/e_expf.c
blob: 08103aa2716eb79afef0f92b685973252c7a94ab (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
/* Single-precision floating point e^x.
   Copyright (C) 1997, 1998 Free Software Foundation, Inc.
   This file is part of the GNU C Library.
   Contributed by Geoffrey Keating <geoffk@ozemail.com.au>

   The GNU C Library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Library General Public License as
   published by the Free Software Foundation; either version 2 of the
   License, or (at your option) any later version.

   The GNU C Library is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   Library General Public License for more details.

   You should have received a copy of the GNU Library General Public
   License along with the GNU C Library; see the file COPYING.LIB.  If not,
   write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
   Boston, MA 02111-1307, USA.  */

/* How this works:

   The input value, x, is written as

   x = n * ln(2) + t/512 + delta[t] + x;

   where:
   - n is an integer, 127 >= n >= -150;
   - t is an integer, 177 >= t >= -177
   - delta is based on a table entry, delta[t] < 2^-28
   - x is whatever is left, |x| < 2^-10

   Then e^x is approximated as

   e^x = 2^n ( e^(t/512 + delta[t])
               + ( e^(t/512 + delta[t])
                   * ( p(x + delta[t] + n * ln(2)) - delta ) ) )

   where
   - p(x) is a polynomial approximating e(x)-1;
   - e^(t/512 + delta[t]) is obtained from a table.

   The table used is the same one as for the double precision version;
   since we have the table, we might as well use it.

   It turns out to be faster to do calculations in double precision than
   to perform an 'accurate table method' expf, because of the range reduction
   overhead (compare exp2f).
   */
#ifndef _GNU_SOURCE
#define _GNU_SOURCE
#endif
#include <float.h>
#include <ieee754.h>
#include <math.h>
#include <fenv.h>
#include <inttypes.h>
#include <math_private.h>

extern const float __exp_deltatable[178];
extern const double __exp_atable[355] /* __attribute__((mode(DF))) */;

static const volatile float TWOM100 = 7.88860905e-31;
static const volatile float TWO127 = 1.7014118346e+38;

float
__ieee754_expf (float x)
{
  static const uint32_t a_minf = 0xff800000;
  static const float himark = 88.72283935546875;
  static const float lomark = -103.972084045410;
  /* Check for usual case.  */
  if (isless (x, himark) && isgreater (x, lomark))
    {
      static const float TWO43 = 8796093022208.0;
      static const float TWO23 = 8388608.0;
      /* 1/ln(2).  */
#undef M_1_LN2
      static const float M_1_LN2 = 1.44269502163f;
      /* ln(2) */
#undef M_LN2
      static const double M_LN2 = .6931471805599452862;

      int tval;
      double x22, t, result, dx;
      float n, delta;
      union ieee754_double ex2_u;
      fenv_t oldenv;

      feholdexcept (&oldenv);
      fesetround (FE_TONEAREST);

      /* Calculate n.  */
      if (x >= 0)
	{
	  n = x * M_1_LN2 + TWO23;
	  n -= TWO23;
	}
      else
	{
	  n = x * M_1_LN2 - TWO23;
	  n += TWO23;
	}
      dx = x - n*M_LN2;
      if (dx >= 0)
	{
	  /* Calculate t/512.  */
	  t = dx + TWO43;
	  t -= TWO43;
	  dx -= t;

	  /* Compute tval = t.  */
	  tval = (int) (t * 512.0);

	  delta = - __exp_deltatable[tval];
	}
      else
	{
	  /* As above, but x is negative.  */
	  t = dx - TWO43;
	  t += TWO43;
	  dx -= t;

	  tval = (int) (t * 512.0);

	  delta = __exp_deltatable[-tval];
	}

      /* Compute ex2 = 2^n e^(t/512+delta[t]).  */
      ex2_u.d = __exp_atable[tval+177];
      ex2_u.ieee.exponent += (int) n;

      /* Approximate e^(dx+delta) - 1, using a second-degree polynomial,
	 with maximum error in [-2^-10-2^-28,2^-10+2^-28]
	 less than 5e-11.  */
      x22 = (0.5000000496709180453 * dx + 1.0000001192102037084) * dx + delta;

      /* Return result.  */
      fesetenv (&oldenv);

      result = x22 * ex2_u.d + ex2_u.d;
      return (float) result;
    }
  /* Exceptional cases:  */
  else if (isless (x, himark))
    {
      if (x == *(const float *) &a_minf)
	/* e^-inf == 0, with no error.  */
	return 0;
      else
	/* Underflow */
	return TWOM100 * TWOM100;
    }
  else
    /* Return x, if x is a NaN or Inf; or overflow, otherwise.  */
    return TWO127*x;
}