Home | History | Annotate | Download | only in Math
      1 /** @file
      2   Compute the logrithm of x.
      3 
      4   Copyright (c) 2010 - 2011, Intel Corporation. All rights reserved.<BR>
      5   This program and the accompanying materials are licensed and made available under
      6   the terms and conditions of the BSD License that accompanies this distribution.
      7   The full text of the license may be found at
      8   http://opensource.org/licenses/bsd-license.
      9 
     10   THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS,
     11   WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED.
     12 
     13  * ====================================================
     14  * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
     15  *
     16  * Developed at SunPro, a Sun Microsystems, Inc. business.
     17  * Permission to use, copy, modify, and distribute this
     18  * software is freely granted, provided that this notice
     19  * is preserved.
     20  * ====================================================
     21 
     22   e_log.c 5.1 93/09/24
     23   NetBSD: e_log.c,v 1.12 2002/05/26 22:01:51 wiz Exp
     24 **/
     25 #include  <LibConfig.h>
     26 #include  <sys/EfiCdefs.h>
     27 
     28 #if defined(_MSC_VER)           /* Handle Microsoft VC++ compiler specifics. */
     29   // potential divide by 0 -- near line 118, (x-x)/zero is on purpose
     30   #pragma warning ( disable : 4723 )
     31 #endif
     32 
     33 /* __ieee754_log(x)
     34  * Return the logrithm of x
     35  *
     36  * Method :
     37  *   1. Argument Reduction: find k and f such that
     38  *      x = 2^k * (1+f),
     39  *     where  sqrt(2)/2 < 1+f < sqrt(2) .
     40  *
     41  *   2. Approximation of log(1+f).
     42  *  Let s = f/(2+f) ; based on log(1+f) = log(1+s) - log(1-s)
     43  *     = 2s + 2/3 s**3 + 2/5 s**5 + .....,
     44  *         = 2s + s*R
     45  *      We use a special Reme algorithm on [0,0.1716] to generate
     46  *  a polynomial of degree 14 to approximate R The maximum error
     47  *  of this polynomial approximation is bounded by 2**-58.45. In
     48  *  other words,
     49  *            2      4      6      8      10      12      14
     50  *      R(z) ~ Lg1*s +Lg2*s +Lg3*s +Lg4*s +Lg5*s  +Lg6*s  +Lg7*s
     51  *    (the values of Lg1 to Lg7 are listed in the program)
     52  *  and
     53  *      |      2          14          |     -58.45
     54  *      | Lg1*s +...+Lg7*s    -  R(z) | <= 2
     55  *      |                             |
     56  *  Note that 2s = f - s*f = f - hfsq + s*hfsq, where hfsq = f*f/2.
     57  *  In order to guarantee error in log below 1ulp, we compute log
     58  *  by
     59  *    log(1+f) = f - s*(f - R)  (if f is not too large)
     60  *    log(1+f) = f - (hfsq - s*(hfsq+R)). (better accuracy)
     61  *
     62  *  3. Finally,  log(x) = k*ln2 + log(1+f).
     63  *          = k*ln2_hi+(f-(hfsq-(s*(hfsq+R)+k*ln2_lo)))
     64  *     Here ln2 is split into two floating point number:
     65  *      ln2_hi + ln2_lo,
     66  *     where n*ln2_hi is always exact for |n| < 2000.
     67  *
     68  * Special cases:
     69  *  log(x) is NaN with signal if x < 0 (including -INF) ;
     70  *  log(+INF) is +INF; log(0) is -INF with signal;
     71  *  log(NaN) is that NaN with no signal.
     72  *
     73  * Accuracy:
     74  *  according to an error analysis, the error is always less than
     75  *  1 ulp (unit in the last place).
     76  *
     77  * Constants:
     78  * The hexadecimal values are the intended ones for the following
     79  * constants. The decimal values may be used, provided that the
     80  * compiler will convert from decimal to binary accurately enough
     81  * to produce the hexadecimal values shown.
     82  */
     83 
     84 #include "math.h"
     85 #include "math_private.h"
     86 #include  <errno.h>
     87 
     88 static const double
     89 ln2_hi  =  6.93147180369123816490e-01,  /* 3fe62e42 fee00000 */
     90 ln2_lo  =  1.90821492927058770002e-10,  /* 3dea39ef 35793c76 */
     91 two54   =  1.80143985094819840000e+16,  /* 43500000 00000000 */
     92 Lg1 = 6.666666666666735130e-01,  /* 3FE55555 55555593 */
     93 Lg2 = 3.999999999940941908e-01,  /* 3FD99999 9997FA04 */
     94 Lg3 = 2.857142874366239149e-01,  /* 3FD24924 94229359 */
     95 Lg4 = 2.222219843214978396e-01,  /* 3FCC71C5 1D8E78AF */
     96 Lg5 = 1.818357216161805012e-01,  /* 3FC74664 96CB03DE */
     97 Lg6 = 1.531383769920937332e-01,  /* 3FC39A09 D078C69F */
     98 Lg7 = 1.479819860511658591e-01;  /* 3FC2F112 DF3E5244 */
     99 
    100 static const double zero   =  0.0;
    101 
    102 double
    103 __ieee754_log(double x)
    104 {
    105   double hfsq,f,s,z,R,w,t1,t2,dk;
    106   int32_t k,hx,i,j;
    107   u_int32_t lx;
    108 
    109   EXTRACT_WORDS(hx,lx,x);
    110 
    111   k=0;
    112   if (hx < 0x00100000) {      /* x < 2**-1022  */
    113     if (((hx&0x7fffffff)|lx)==0)
    114       return -two54/zero;     /* log(+-0)=-inf */
    115     if (hx<0) {
    116       errno = EDOM;
    117       return (x-x)/zero;      /* log(-#) = NaN */
    118     }
    119       k -= 54; x *= two54;    /* subnormal number, scale up x */
    120       GET_HIGH_WORD(hx,x);
    121   }
    122   if (hx >= 0x7ff00000) return x+x;
    123   k += (hx>>20)-1023;
    124   hx &= 0x000fffff;
    125   i = (hx+0x95f64)&0x100000;
    126   SET_HIGH_WORD(x,hx|(i^0x3ff00000)); /* normalize x or x/2 */
    127   k += (i>>20);
    128   f = x-1.0;
    129   if((0x000fffff&(2+hx))<3) { /* |f| < 2**-20 */
    130     if(f==zero) { if(k==0) return zero;  else {dk=(double)k;
    131          return dk*ln2_hi+dk*ln2_lo;}
    132     }
    133     R = f*f*(0.5-0.33333333333333333*f);
    134     if(k==0) return f-R; else {dk=(double)k;
    135            return dk*ln2_hi-((R-dk*ln2_lo)-f);}
    136   }
    137   s = f/(2.0+f);
    138   dk = (double)k;
    139   z = s*s;
    140   i = hx-0x6147a;
    141   w = z*z;
    142   j = 0x6b851-hx;
    143   t1= w*(Lg2+w*(Lg4+w*Lg6));
    144   t2= z*(Lg1+w*(Lg3+w*(Lg5+w*Lg7)));
    145   i |= j;
    146   R = t2+t1;
    147   if(i>0) {
    148       hfsq=0.5*f*f;
    149       if(k==0) return f-(hfsq-s*(hfsq+R)); else
    150          return dk*ln2_hi-((hfsq-(s*(hfsq+R)+dk*ln2_lo))-f);
    151   } else {
    152       if(k==0) return f-s*(f-R); else
    153          return dk*ln2_hi-((s*(f-R)-dk*ln2_lo)-f);
    154   }
    155 }
    156