lh | 9ed821d | 2023-04-07 01:36:19 -0700 | [diff] [blame] | 1 | /* Adapted for log2 by Ulrich Drepper <drepper@cygnus.com>. */ |
| 2 | /* |
| 3 | * ==================================================== |
| 4 | * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved. |
| 5 | * |
| 6 | * Developed at SunPro, 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 | /* __ieee754_log2(x) |
| 14 | * Return the logarithm to base 2 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 + log(1+f). |
| 43 | * = k+(f-(hfsq-(s*(hfsq+R)))) |
| 44 | * |
| 45 | * Special cases: |
| 46 | * log2(x) is NaN with signal if x < 0 (including -INF) ; |
| 47 | * log2(+INF) is +INF; log(0) is -INF with signal; |
| 48 | * log2(NaN) is that NaN with no signal. |
| 49 | * |
| 50 | * Constants: |
| 51 | * The hexadecimal values are the intended ones for the following |
| 52 | * constants. The decimal values may be used, provided that the |
| 53 | * compiler will convert from decimal to binary accurately enough |
| 54 | * to produce the hexadecimal values shown. |
| 55 | */ |
| 56 | |
| 57 | #include "math.h" |
| 58 | #include "math_private.h" |
| 59 | |
| 60 | static const double |
| 61 | ln2 = 0.69314718055994530942, |
| 62 | two54 = 1.80143985094819840000e+16, /* 43500000 00000000 */ |
| 63 | Lg1 = 6.666666666666735130e-01, /* 3FE55555 55555593 */ |
| 64 | Lg2 = 3.999999999940941908e-01, /* 3FD99999 9997FA04 */ |
| 65 | Lg3 = 2.857142874366239149e-01, /* 3FD24924 94229359 */ |
| 66 | Lg4 = 2.222219843214978396e-01, /* 3FCC71C5 1D8E78AF */ |
| 67 | Lg5 = 1.818357216161805012e-01, /* 3FC74664 96CB03DE */ |
| 68 | Lg6 = 1.531383769920937332e-01, /* 3FC39A09 D078C69F */ |
| 69 | Lg7 = 1.479819860511658591e-01; /* 3FC2F112 DF3E5244 */ |
| 70 | |
| 71 | static const double zero = 0.0; |
| 72 | |
| 73 | double __ieee754_log2(double x) |
| 74 | { |
| 75 | double hfsq,f,s,z,R,w,t1,t2,dk; |
| 76 | int32_t k,hx,i,j; |
| 77 | u_int32_t lx; |
| 78 | |
| 79 | EXTRACT_WORDS(hx,lx,x); |
| 80 | |
| 81 | k=0; |
| 82 | if (hx < 0x00100000) { /* x < 2**-1022 */ |
| 83 | if (((hx&0x7fffffff)|lx)==0) |
| 84 | return -two54/(x-x); /* log(+-0)=-inf */ |
| 85 | if (hx<0) return (x-x)/(x-x); /* log(-#) = NaN */ |
| 86 | k -= 54; x *= two54; /* subnormal number, scale up x */ |
| 87 | GET_HIGH_WORD(hx,x); |
| 88 | } |
| 89 | if (hx >= 0x7ff00000) return x+x; |
| 90 | k += (hx>>20)-1023; |
| 91 | hx &= 0x000fffff; |
| 92 | i = (hx+0x95f64)&0x100000; |
| 93 | SET_HIGH_WORD(x,hx|(i^0x3ff00000)); /* normalize x or x/2 */ |
| 94 | k += (i>>20); |
| 95 | dk = (double) k; |
| 96 | f = x-1.0; |
| 97 | if((0x000fffff&(2+hx))<3) { /* |f| < 2**-20 */ |
| 98 | if(f==zero) return dk; |
| 99 | R = f*f*(0.5-0.33333333333333333*f); |
| 100 | return dk-(R-f)/ln2; |
| 101 | } |
| 102 | s = f/(2.0+f); |
| 103 | z = s*s; |
| 104 | i = hx-0x6147a; |
| 105 | w = z*z; |
| 106 | j = 0x6b851-hx; |
| 107 | t1= w*(Lg2+w*(Lg4+w*Lg6)); |
| 108 | t2= z*(Lg1+w*(Lg3+w*(Lg5+w*Lg7))); |
| 109 | i |= j; |
| 110 | R = t2+t1; |
| 111 | if(i>0) { |
| 112 | hfsq=0.5*f*f; |
| 113 | return dk-((hfsq-(s*(hfsq+R)))-f)/ln2; |
| 114 | } else { |
| 115 | return dk-((s*(f-R))-f)/ln2; |
| 116 | } |
| 117 | } |
| 118 | strong_alias(__ieee754_log2,log2) |