1 | /* |
2 | * Copyright (c) 1998, 2003, Oracle and/or its affiliates. All rights reserved. |
3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
4 | * |
5 | * This code is free software; you can redistribute it and/or modify it |
6 | * under the terms of the GNU General Public License version 2 only, as |
7 | * published by the Free Software Foundation. Oracle designates this |
8 | * particular file as subject to the "Classpath" exception as provided |
9 | * by Oracle in the LICENSE file that accompanied this code. |
10 | * |
11 | * This code is distributed in the hope that it will be useful, but WITHOUT |
12 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
13 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
14 | * version 2 for more details (a copy is included in the LICENSE file that |
15 | * accompanied this code). |
16 | * |
17 | * You should have received a copy of the GNU General Public License version |
18 | * 2 along with this work; if not, write to the Free Software Foundation, |
19 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
20 | * |
21 | * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
22 | * or visit www.oracle.com if you need additional information or have any |
23 | * questions. |
24 | */ |
25 | |
26 | /* double log1p(double x) |
27 | * |
28 | * Method : |
29 | * 1. Argument Reduction: find k and f such that |
30 | * 1+x = 2^k * (1+f), |
31 | * where sqrt(2)/2 < 1+f < sqrt(2) . |
32 | * |
33 | * Note. If k=0, then f=x is exact. However, if k!=0, then f |
34 | * may not be representable exactly. In that case, a correction |
35 | * term is need. Let u=1+x rounded. Let c = (1+x)-u, then |
36 | * log(1+x) - log(u) ~ c/u. Thus, we proceed to compute log(u), |
37 | * and add back the correction term c/u. |
38 | * (Note: when x > 2**53, one can simply return log(x)) |
39 | * |
40 | * 2. Approximation of log1p(f). |
41 | * Let s = f/(2+f) ; based on log(1+f) = log(1+s) - log(1-s) |
42 | * = 2s + 2/3 s**3 + 2/5 s**5 + ....., |
43 | * = 2s + s*R |
44 | * We use a special Reme algorithm on [0,0.1716] to generate |
45 | * a polynomial of degree 14 to approximate R The maximum error |
46 | * of this polynomial approximation is bounded by 2**-58.45. In |
47 | * other words, |
48 | * 2 4 6 8 10 12 14 |
49 | * R(z) ~ Lp1*s +Lp2*s +Lp3*s +Lp4*s +Lp5*s +Lp6*s +Lp7*s |
50 | * (the values of Lp1 to Lp7 are listed in the program) |
51 | * and |
52 | * | 2 14 | -58.45 |
53 | * | Lp1*s +...+Lp7*s - R(z) | <= 2 |
54 | * | | |
55 | * Note that 2s = f - s*f = f - hfsq + s*hfsq, where hfsq = f*f/2. |
56 | * In order to guarantee error in log below 1ulp, we compute log |
57 | * by |
58 | * log1p(f) = f - (hfsq - s*(hfsq+R)). |
59 | * |
60 | * 3. Finally, log1p(x) = k*ln2 + log1p(f). |
61 | * = k*ln2_hi+(f-(hfsq-(s*(hfsq+R)+k*ln2_lo))) |
62 | * Here ln2 is split into two floating point number: |
63 | * ln2_hi + ln2_lo, |
64 | * where n*ln2_hi is always exact for |n| < 2000. |
65 | * |
66 | * Special cases: |
67 | * log1p(x) is NaN with signal if x < -1 (including -INF) ; |
68 | * log1p(+INF) is +INF; log1p(-1) is -INF with signal; |
69 | * log1p(NaN) is that NaN with no signal. |
70 | * |
71 | * Accuracy: |
72 | * according to an error analysis, the error is always less than |
73 | * 1 ulp (unit in the last place). |
74 | * |
75 | * Constants: |
76 | * The hexadecimal values are the intended ones for the following |
77 | * constants. The decimal values may be used, provided that the |
78 | * compiler will convert from decimal to binary accurately enough |
79 | * to produce the hexadecimal values shown. |
80 | * |
81 | * Note: Assuming log() return accurate answer, the following |
82 | * algorithm can be used to compute log1p(x) to within a few ULP: |
83 | * |
84 | * u = 1+x; |
85 | * if(u==1.0) return x ; else |
86 | * return log(u)*(x/(u-1.0)); |
87 | * |
88 | * See HP-15C Advanced Functions Handbook, p.193. |
89 | */ |
90 | |
91 | #include "fdlibm.h" |
92 | |
93 | #ifdef __STDC__ |
94 | static const double |
95 | #else |
96 | static double |
97 | #endif |
98 | ln2_hi = 6.93147180369123816490e-01, /* 3fe62e42 fee00000 */ |
99 | ln2_lo = 1.90821492927058770002e-10, /* 3dea39ef 35793c76 */ |
100 | two54 = 1.80143985094819840000e+16, /* 43500000 00000000 */ |
101 | Lp1 = 6.666666666666735130e-01, /* 3FE55555 55555593 */ |
102 | Lp2 = 3.999999999940941908e-01, /* 3FD99999 9997FA04 */ |
103 | Lp3 = 2.857142874366239149e-01, /* 3FD24924 94229359 */ |
104 | Lp4 = 2.222219843214978396e-01, /* 3FCC71C5 1D8E78AF */ |
105 | Lp5 = 1.818357216161805012e-01, /* 3FC74664 96CB03DE */ |
106 | Lp6 = 1.531383769920937332e-01, /* 3FC39A09 D078C69F */ |
107 | Lp7 = 1.479819860511658591e-01; /* 3FC2F112 DF3E5244 */ |
108 | |
109 | static double zero = 0.0; |
110 | |
111 | #ifdef __STDC__ |
112 | double log1p(double x) |
113 | #else |
114 | double log1p(x) |
115 | double x; |
116 | #endif |
117 | { |
118 | double hfsq,f=0,c=0,s,z,R,u; |
119 | int k,hx,hu=0,ax; |
120 | |
121 | hx = __HI(x); /* high word of x */ |
122 | ax = hx&0x7fffffff; |
123 | |
124 | k = 1; |
125 | if (hx < 0x3FDA827A) { /* x < 0.41422 */ |
126 | if(ax>=0x3ff00000) { /* x <= -1.0 */ |
127 | /* |
128 | * Added redundant test against hx to work around VC++ |
129 | * code generation problem. |
130 | */ |
131 | if(x==-1.0 && (hx==0xbff00000)) /* log1p(-1)=-inf */ |
132 | return -two54/zero; |
133 | else |
134 | return (x-x)/(x-x); /* log1p(x<-1)=NaN */ |
135 | } |
136 | if(ax<0x3e200000) { /* |x| < 2**-29 */ |
137 | if(two54+x>zero /* raise inexact */ |
138 | &&ax<0x3c900000) /* |x| < 2**-54 */ |
139 | return x; |
140 | else |
141 | return x - x*x*0.5; |
142 | } |
143 | if(hx>0||hx<=((int)0xbfd2bec3)) { |
144 | k=0;f=x;hu=1;} /* -0.2929<x<0.41422 */ |
145 | } |
146 | if (hx >= 0x7ff00000) return x+x; |
147 | if(k!=0) { |
148 | if(hx<0x43400000) { |
149 | u = 1.0+x; |
150 | hu = __HI(u); /* high word of u */ |
151 | k = (hu>>20)-1023; |
152 | c = (k>0)? 1.0-(u-x):x-(u-1.0);/* correction term */ |
153 | c /= u; |
154 | } else { |
155 | u = x; |
156 | hu = __HI(u); /* high word of u */ |
157 | k = (hu>>20)-1023; |
158 | c = 0; |
159 | } |
160 | hu &= 0x000fffff; |
161 | if(hu<0x6a09e) { |
162 | __HI(u) = hu|0x3ff00000; /* normalize u */ |
163 | } else { |
164 | k += 1; |
165 | __HI(u) = hu|0x3fe00000; /* normalize u/2 */ |
166 | hu = (0x00100000-hu)>>2; |
167 | } |
168 | f = u-1.0; |
169 | } |
170 | hfsq=0.5*f*f; |
171 | if(hu==0) { /* |f| < 2**-20 */ |
172 | if(f==zero) { if(k==0) return zero; |
173 | else {c += k*ln2_lo; return k*ln2_hi+c;}} |
174 | R = hfsq*(1.0-0.66666666666666666*f); |
175 | if(k==0) return f-R; else |
176 | return k*ln2_hi-((R-(k*ln2_lo+c))-f); |
177 | } |
178 | s = f/(2.0+f); |
179 | z = s*s; |
180 | R = z*(Lp1+z*(Lp2+z*(Lp3+z*(Lp4+z*(Lp5+z*(Lp6+z*Lp7)))))); |
181 | if(k==0) return f-(hfsq-s*(hfsq+R)); else |
182 | return k*ln2_hi-((hfsq-(s*(hfsq+R)+(k*ln2_lo+c)))-f); |
183 | } |
184 | |