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
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
|
/* Quad-precision floating point e^x.
Copyright (C) 1999-2020 Free Software Foundation, Inc.
This file is part of the GNU C Library.
Contributed by Jakub Jelinek <jj@ultra.linux.cz>
Partly based on double-precision code
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 Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 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
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<https://www.gnu.org/licenses/>. */
/* The basic design here is from
Abraham Ziv, "Fast Evaluation of Elementary Mathematical Functions with
Correctly Rounded Last Bit", ACM Trans. Math. Soft., 17 (3), September 1991,
pp. 410-423.
We work with number pairs where the first number is the high part and
the second one is the low part. Arithmetic with the high part numbers must
be exact, without any roundoff errors.
The input value, X, is written as
X = n * ln(2)_0 + arg1[t1]_0 + arg2[t2]_0 + x
- n * ln(2)_1 + arg1[t1]_1 + arg2[t2]_1 + xl
where:
- n is an integer, 16384 >= n >= -16495;
- ln(2)_0 is the first 93 bits of ln(2), and |ln(2)_0-ln(2)-ln(2)_1| < 2^-205
- t1 is an integer, 89 >= t1 >= -89
- t2 is an integer, 65 >= t2 >= -65
- |arg1[t1]-t1/256.0| < 2^-53
- |arg2[t2]-t2/32768.0| < 2^-53
- x + xl is whatever is left, |x + xl| < 2^-16 + 2^-53
Then e^x is approximated as
e^x = 2^n_1 ( 2^n_0 e^(arg1[t1]_0 + arg1[t1]_1) e^(arg2[t2]_0 + arg2[t2]_1)
+ 2^n_0 e^(arg1[t1]_0 + arg1[t1]_1) e^(arg2[t2]_0 + arg2[t2]_1)
* p (x + xl + n * ln(2)_1))
where:
- p(x) is a polynomial approximating e(x)-1
- e^(arg1[t1]_0 + arg1[t1]_1) is obtained from a table
- e^(arg2[t2]_0 + arg2[t2]_1) likewise
- n_1 + n_0 = n, so that |n_0| < -LDBL_MIN_EXP-1.
If it happens that n_1 == 0 (this is the usual case), that multiplication
is omitted.
*/
#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>
#include <fenv_private.h>
#include "t_expl.h"
static const long double C[] = {
/* Smallest integer x for which e^x overflows. */
#define himark C[0]
709.78271289338399678773454114191496482L,
/* Largest integer x for which e^x underflows. */
#define lomark C[1]
-744.44007192138126231410729844608163411L,
/* 3x2^96 */
#define THREEp96 C[2]
59421121885698253195157962752.0L,
/* 3x2^103 */
#define THREEp103 C[3]
30423614405477505635920876929024.0L,
/* 3x2^111 */
#define THREEp111 C[4]
7788445287802241442795744493830144.0L,
/* 1/ln(2) */
#define M_1_LN2 C[5]
1.44269504088896340735992468100189204L,
/* first 93 bits of ln(2) */
#define M_LN2_0 C[6]
0.693147180559945309417232121457981864L,
/* ln2_0 - ln(2) */
#define M_LN2_1 C[7]
-1.94704509238074995158795957333327386E-31L,
/* very small number */
#define TINY C[8]
1.0e-308L,
/* 2^16383 */
#define TWO1023 C[9]
8.988465674311579538646525953945123668E+307L,
/* 256 */
#define TWO8 C[10]
256.0L,
/* 32768 */
#define TWO15 C[11]
32768.0L,
/* Chebyshev polynom coefficients for (exp(x)-1)/x */
#define P1 C[12]
#define P2 C[13]
#define P3 C[14]
#define P4 C[15]
#define P5 C[16]
#define P6 C[17]
0.5L,
1.66666666666666666666666666666666683E-01L,
4.16666666666666666666654902320001674E-02L,
8.33333333333333333333314659767198461E-03L,
1.38888888889899438565058018857254025E-03L,
1.98412698413981650382436541785404286E-04L,
};
/* Avoid local PLT entry use from (int) roundl (...) being converted
to a call to lroundl in the case of 32-bit long and roundl not
inlined. */
long int lroundl (long double) asm ("__lroundl");
long double
__ieee754_expl (long double x)
{
long double result, x22;
union ibm_extended_long_double ex2_u, scale_u;
int unsafe;
/* Check for usual case. */
if (isless (x, himark) && isgreater (x, lomark))
{
int tval1, tval2, n_i, exponent2;
long double n, xl;
SET_RESTORE_ROUND (FE_TONEAREST);
n = roundl (x*M_1_LN2);
x = x-n*M_LN2_0;
xl = n*M_LN2_1;
tval1 = roundl (x*TWO8);
x -= __expl_table[T_EXPL_ARG1+2*tval1];
xl -= __expl_table[T_EXPL_ARG1+2*tval1+1];
tval2 = roundl (x*TWO15);
x -= __expl_table[T_EXPL_ARG2+2*tval2];
xl -= __expl_table[T_EXPL_ARG2+2*tval2+1];
x = x + xl;
/* Compute ex2 = 2^n_0 e^(argtable[tval1]) e^(argtable[tval2]). */
ex2_u.ld = (__expl_table[T_EXPL_RES1 + tval1]
* __expl_table[T_EXPL_RES2 + tval2]);
n_i = (int)n;
/* 'unsafe' is 1 iff n_1 != 0. */
unsafe = fabsl(n_i) >= -LDBL_MIN_EXP - 1;
ex2_u.d[0].ieee.exponent += n_i >> unsafe;
/* Fortunately, there are no subnormal lowpart doubles in
__expl_table, only normal values and zeros.
But after scaling it can be subnormal. */
exponent2 = ex2_u.d[1].ieee.exponent + (n_i >> unsafe);
if (ex2_u.d[1].ieee.exponent == 0)
/* assert ((ex2_u.d[1].ieee.mantissa0|ex2_u.d[1].ieee.mantissa1) == 0) */;
else if (exponent2 > 0)
ex2_u.d[1].ieee.exponent = exponent2;
else if (exponent2 <= -54)
{
ex2_u.d[1].ieee.exponent = 0;
ex2_u.d[1].ieee.mantissa0 = 0;
ex2_u.d[1].ieee.mantissa1 = 0;
}
else
{
static const double
two54 = 1.80143985094819840000e+16, /* 4350000000000000 */
twom54 = 5.55111512312578270212e-17; /* 3C90000000000000 */
ex2_u.d[1].d *= two54;
ex2_u.d[1].ieee.exponent += n_i >> unsafe;
ex2_u.d[1].d *= twom54;
}
/* Compute scale = 2^n_1. */
scale_u.ld = 1.0L;
scale_u.d[0].ieee.exponent += n_i - (n_i >> unsafe);
/* Approximate e^x2 - 1, using a seventh-degree polynomial,
with maximum error in [-2^-16-2^-53,2^-16+2^-53]
less than 4.8e-39. */
x22 = x + x*x*(P1+x*(P2+x*(P3+x*(P4+x*(P5+x*P6)))));
/* Now we can test whether the result is ultimate or if we are unsure.
In the later case we should probably call a mpn based routine to give
the ultimate result.
Empirically, this routine is already ultimate in about 99.9986% of
cases, the test below for the round to nearest case will be false
in ~ 99.9963% of cases.
Without proc2 routine maximum error which has been seen is
0.5000262 ulp.
union ieee854_long_double ex3_u;
#ifdef FE_TONEAREST
fesetround (FE_TONEAREST);
#endif
ex3_u.d = (result - ex2_u.d) - x22 * ex2_u.d;
ex2_u.d = result;
ex3_u.ieee.exponent += LDBL_MANT_DIG + 15 + IEEE854_LONG_DOUBLE_BIAS
- ex2_u.ieee.exponent;
n_i = abs (ex3_u.d);
n_i = (n_i + 1) / 2;
fesetenv (&oldenv);
#ifdef FE_TONEAREST
if (fegetround () == FE_TONEAREST)
n_i -= 0x4000;
#endif
if (!n_i) {
return __ieee754_expl_proc2 (origx);
}
*/
}
/* Exceptional cases: */
else if (isless (x, himark))
{
if (isinf (x))
/* e^-inf == 0, with no error. */
return 0;
else
/* Underflow */
return TINY * TINY;
}
else
/* Return x, if x is a NaN or Inf; or overflow, otherwise. */
return TWO1023*x;
result = x22 * ex2_u.ld + ex2_u.ld;
if (!unsafe)
return result;
return result * scale_u.ld;
}
strong_alias (__ieee754_expl, __expl_finite)
|
