/src/moddable/xs/tools/fdlibm/e_log.c
Line | Count | Source |
1 | | |
2 | | /* |
3 | | * ==================================================== |
4 | | * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved. |
5 | | * |
6 | | * Developed at SunSoft, a Sun Microsystems, Inc. business. |
7 | | * Permission to use, copy, modify, and distribute this |
8 | | * software is freely granted, provided that this notice |
9 | | * is preserved. |
10 | | * ==================================================== |
11 | | */ |
12 | | |
13 | | /* log(x) |
14 | | * Return the logrithm of x |
15 | | * |
16 | | * Method : |
17 | | * 1. Argument Reduction: find k and f such that |
18 | | * x = 2^k * (1+f), |
19 | | * where sqrt(2)/2 < 1+f < sqrt(2) . |
20 | | * |
21 | | * 2. Approximation of log(1+f). |
22 | | * Let s = f/(2+f) ; based on log(1+f) = log(1+s) - log(1-s) |
23 | | * = 2s + 2/3 s**3 + 2/5 s**5 + ....., |
24 | | * = 2s + s*R |
25 | | * We use a special Reme algorithm on [0,0.1716] to generate |
26 | | * a polynomial of degree 14 to approximate R The maximum error |
27 | | * of this polynomial approximation is bounded by 2**-58.45. In |
28 | | * other words, |
29 | | * 2 4 6 8 10 12 14 |
30 | | * R(z) ~ Lg1*s +Lg2*s +Lg3*s +Lg4*s +Lg5*s +Lg6*s +Lg7*s |
31 | | * (the values of Lg1 to Lg7 are listed in the program) |
32 | | * and |
33 | | * | 2 14 | -58.45 |
34 | | * | Lg1*s +...+Lg7*s - R(z) | <= 2 |
35 | | * | | |
36 | | * Note that 2s = f - s*f = f - hfsq + s*hfsq, where hfsq = f*f/2. |
37 | | * In order to guarantee error in log below 1ulp, we compute log |
38 | | * by |
39 | | * log(1+f) = f - s*(f - R) (if f is not too large) |
40 | | * log(1+f) = f - (hfsq - s*(hfsq+R)). (better accuracy) |
41 | | * |
42 | | * 3. Finally, log(x) = k*ln2 + log(1+f). |
43 | | * = k*ln2_hi+(f-(hfsq-(s*(hfsq+R)+k*ln2_lo))) |
44 | | * Here ln2 is split into two floating point number: |
45 | | * ln2_hi + ln2_lo, |
46 | | * where n*ln2_hi is always exact for |n| < 2000. |
47 | | * |
48 | | * Special cases: |
49 | | * log(x) is NaN with signal if x < 0 (including -INF) ; |
50 | | * log(+INF) is +INF; log(0) is -INF with signal; |
51 | | * log(NaN) is that NaN with no signal. |
52 | | * |
53 | | * Accuracy: |
54 | | * according to an error analysis, the error is always less than |
55 | | * 1 ulp (unit in the last place). |
56 | | * |
57 | | * Constants: |
58 | | * The hexadecimal values are the intended ones for the following |
59 | | * constants. The decimal values may be used, provided that the |
60 | | * compiler will convert from decimal to binary accurately enough |
61 | | * to produce the hexadecimal values shown. |
62 | | */ |
63 | | |
64 | | #include "math_private.h" |
65 | | |
66 | | static const double |
67 | | ln2_hi = 6.93147180369123816490e-01, /* 3fe62e42 fee00000 */ |
68 | | ln2_lo = 1.90821492927058770002e-10, /* 3dea39ef 35793c76 */ |
69 | | two54 = 1.80143985094819840000e+16, /* 43500000 00000000 */ |
70 | | Lg1 = 6.666666666666735130e-01, /* 3FE55555 55555593 */ |
71 | | Lg2 = 3.999999999940941908e-01, /* 3FD99999 9997FA04 */ |
72 | | Lg3 = 2.857142874366239149e-01, /* 3FD24924 94229359 */ |
73 | | Lg4 = 2.222219843214978396e-01, /* 3FCC71C5 1D8E78AF */ |
74 | | Lg5 = 1.818357216161805012e-01, /* 3FC74664 96CB03DE */ |
75 | | Lg6 = 1.531383769920937332e-01, /* 3FC39A09 D078C69F */ |
76 | | Lg7 = 1.479819860511658591e-01; /* 3FC2F112 DF3E5244 */ |
77 | | |
78 | | static const double zero = 0.0; |
79 | | static volatile double vzero = 0.0; |
80 | | |
81 | | double |
82 | | __ieee754_log(double x) |
83 | 853k | { |
84 | 853k | double hfsq,f,s,z,R,w,t1,t2,dk; |
85 | 853k | int32_t k,hx,i,j; |
86 | 853k | u_int32_t lx; |
87 | | |
88 | 853k | EXTRACT_WORDS(hx,lx,x); |
89 | | |
90 | 853k | k=0; |
91 | 853k | if (hx < 0x00100000) { /* x < 2**-1022 */ |
92 | 18 | if (((hx&0x7fffffff)|lx)==0) |
93 | 7 | return -two54/vzero; /* log(+-0)=-inf */ |
94 | 11 | if (hx<0) return (x-x)/zero; /* log(-#) = NaN */ |
95 | 0 | k -= 54; x *= two54; /* subnormal number, scale up x */ |
96 | 0 | GET_HIGH_WORD(hx,x); |
97 | 0 | } |
98 | 853k | if (hx >= 0x7ff00000) return x+x; |
99 | 853k | k += (hx>>20)-1023; |
100 | 853k | hx &= 0x000fffff; |
101 | 853k | i = (hx+0x95f64)&0x100000; |
102 | 853k | SET_HIGH_WORD(x,hx|(i^0x3ff00000)); /* normalize x or x/2 */ |
103 | 853k | k += (i>>20); |
104 | 853k | f = x-1.0; |
105 | 853k | if((0x000fffff&(2+hx))<3) { /* -2**-20 <= f < 2**-20 */ |
106 | 483k | if(f==zero) { |
107 | 480k | if(k==0) { |
108 | 1 | return zero; |
109 | 480k | } else { |
110 | 480k | dk=(double)k; |
111 | 480k | return dk*ln2_hi+dk*ln2_lo; |
112 | 480k | } |
113 | 480k | } |
114 | 2.27k | R = f*f*(0.5-0.33333333333333333*f); |
115 | 2.27k | if(k==0) return f-R; else {dk=(double)k; |
116 | 2.27k | return dk*ln2_hi-((R-dk*ln2_lo)-f);} |
117 | 2.27k | } |
118 | 370k | s = f/(2.0+f); |
119 | 370k | dk = (double)k; |
120 | 370k | z = s*s; |
121 | 370k | i = hx-0x6147a; |
122 | 370k | w = z*z; |
123 | 370k | j = 0x6b851-hx; |
124 | 370k | t1= w*(Lg2+w*(Lg4+w*Lg6)); |
125 | 370k | t2= z*(Lg1+w*(Lg3+w*(Lg5+w*Lg7))); |
126 | 370k | i |= j; |
127 | 370k | R = t2+t1; |
128 | 370k | if(i>0) { |
129 | 342 | hfsq=0.5*f*f; |
130 | 342 | if(k==0) return f-(hfsq-s*(hfsq+R)); else |
131 | 342 | return dk*ln2_hi-((hfsq-(s*(hfsq+R)+dk*ln2_lo))-f); |
132 | 370k | } else { |
133 | 370k | if(k==0) return f-s*(f-R); else |
134 | 370k | return dk*ln2_hi-((s*(f-R)-dk*ln2_lo)-f); |
135 | 370k | } |
136 | 370k | } |