compiler/mlib/s_erff.c

   1 /* s_erff.c -- float version of s_erf.c.
   2  * Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
   3  */
   4
   5 /*
   6  * ====================================================
   7  * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
   8  *
   9  * Developed at SunPro, a Sun Microsystems, Inc. business.
  10  * Permission to use, copy, modify, and distribute this
  11  * software is freely granted, provided that this notice
  12  * is preserved.
  13  * ====================================================
  14  */
  15
  16 #ifndef lint
  17 static char rcsid[] = "$FreeBSD: src/lib/msun/src/s_erff.c,v 1.5 1999/08/28 00:06:46 peter Exp $";
  18 #endif
  19
  20 #include "math.h"
  21 #include "math_private.h"
  22
  23 #ifdef __STDC__
  24 static const float
  25 #else
  26 static float
  27 #endif
  28 tiny        = 1e-30,
  29 half=  5.0000000000e-01, /* 0x3F000000 */
  30 one =  1.0000000000e+00, /* 0x3F800000 */
  31 two =  2.0000000000e+00, /* 0x40000000 */
  32         /* c = (subfloat)0.84506291151 */
  33 erx =  8.4506291151e-01, /* 0x3f58560b */
  34 /*
  35  * Coefficients for approximation to  erf on [0,0.84375]
  36  */
  37 efx =  1.2837916613e-01, /* 0x3e0375d4 */
  38 efx8=  1.0270333290e+00, /* 0x3f8375d4 */
  39 pp0  =  1.2837916613e-01, /* 0x3e0375d4 */
  40 pp1  = -3.2504209876e-01, /* 0xbea66beb */
  41 pp2  = -2.8481749818e-02, /* 0xbce9528f */
  42 pp3  = -5.7702702470e-03, /* 0xbbbd1489 */
  43 pp4  = -2.3763017452e-05, /* 0xb7c756b1 */
  44 qq1  =  3.9791721106e-01, /* 0x3ecbbbce */
  45 qq2  =  6.5022252500e-02, /* 0x3d852a63 */
  46 qq3  =  5.0813062117e-03, /* 0x3ba68116 */
  47 qq4  =  1.3249473704e-04, /* 0x390aee49 */
  48 qq5  = -3.9602282413e-06, /* 0xb684e21a */
  49 /*
  50  * Coefficients for approximation to  erf  in [0.84375,1.25]
  51  */
  52 pa0  = -2.3621185683e-03, /* 0xbb1acdc6 */
  53 pa1  =  4.1485610604e-01, /* 0x3ed46805 */
  54 pa2  = -3.7220788002e-01, /* 0xbebe9208 */
  55 pa3  =  3.1834661961e-01, /* 0x3ea2fe54 */
  56 pa4  = -1.1089469492e-01, /* 0xbde31cc2 */
  57 pa5  =  3.5478305072e-02, /* 0x3d1151b3 */
  58 pa6  = -2.1663755178e-03, /* 0xbb0df9c0 */
  59 qa1  =  1.0642088205e-01, /* 0x3dd9f331 */
  60 qa2  =  5.4039794207e-01, /* 0x3f0a5785 */
  61 qa3  =  7.1828655899e-02, /* 0x3d931ae7 */
  62 qa4  =  1.2617121637e-01, /* 0x3e013307 */
  63 qa5  =  1.3637083583e-02, /* 0x3c5f6e13 */
  64 qa6  =  1.1984500103e-02, /* 0x3c445aa3 */
  65 /*
  66  * Coefficients for approximation to  erfc in [1.25,1/0.35]
  67  */
  68 ra0  = -9.8649440333e-03, /* 0xbc21a093 */
  69 ra1  = -6.9385856390e-01, /* 0xbf31a0b7 */
  70 ra2  = -1.0558626175e+01, /* 0xc128f022 */
  71 ra3  = -6.2375331879e+01, /* 0xc2798057 */
  72 ra4  = -1.6239666748e+02, /* 0xc322658c */
  73 ra5  = -1.8460508728e+02, /* 0xc3389ae7 */
  74 ra6  = -8.1287437439e+01, /* 0xc2a2932b */
  75 ra7  = -9.8143291473e+00, /* 0xc11d077e */
  76 sa1  =  1.9651271820e+01, /* 0x419d35ce */
  77 sa2  =  1.3765776062e+02, /* 0x4309a863 */
  78 sa3  =  4.3456588745e+02, /* 0x43d9486f */
  79 sa4  =  6.4538726807e+02, /* 0x442158c9 */
  80 sa5  =  4.2900814819e+02, /* 0x43d6810b */
  81 sa6  =  1.0863500214e+02, /* 0x42d9451f */
  82 sa7  =  6.5702495575e+00, /* 0x40d23f7c */
  83 sa8  = -6.0424413532e-02, /* 0xbd777f97 */
  84 /*
  85  * Coefficients for approximation to  erfc in [1/.35,28]
  86  */
  87 rb0  = -9.8649431020e-03, /* 0xbc21a092 */
  88 rb1  = -7.9928326607e-01, /* 0xbf4c9dd4 */
  89 rb2  = -1.7757955551e+01, /* 0xc18e104b */
  90 rb3  = -1.6063638306e+02, /* 0xc320a2ea */
  91 rb4  = -6.3756646729e+02, /* 0xc41f6441 */
  92 rb5  = -1.0250950928e+03, /* 0xc480230b */
  93 rb6  = -4.8351919556e+02, /* 0xc3f1c275 */
  94 sb1  =  3.0338060379e+01, /* 0x41f2b459 */
  95 sb2  =  3.2579251099e+02, /* 0x43a2e571 */
  96 sb3  =  1.5367296143e+03, /* 0x44c01759 */
  97 sb4  =  3.1998581543e+03, /* 0x4547fdbb */
  98 sb5  =  2.5530502930e+03, /* 0x451f90ce */
  99 sb6  =  4.7452853394e+02, /* 0x43ed43a7 */
 100 sb7  = -2.2440952301e+01; /* 0xc1b38712 */
 101
 102 #ifdef __STDC__
 103         float erff(float x)
 104 #else
 105         float erff(x)
 106         float x;
 107 #endif
 108 {
 109         int32_t hx,ix,i;
 110         float R,S,P,Q,s,y,z,r;
 111         GET_FLOAT_WORD(hx,x);
 112         ix = hx&0x7fffffff;
 113         if(ix>=0x7f800000) {            /* erf(nan)=nan */
 114             i = ((uint32_t)hx>>31)<<1;
 115             return (float)(1-i)+one/x;  /* erf(+-inf)=+-1 */
 116         }
 117
 118         if(ix < 0x3f580000) {           /* |x|<0.84375 */
 119             if(ix < 0x31800000) {       /* |x|<2**-28 */
 120                 if (ix < 0x04000000)
 121                     /*avoid underflow */
 122                     return (float)0.125*((float)8.0*x+efx8*x);
 123                 return x + efx*x;
 124             }
 125             z = x*x;
 126             r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4)));
 127             s = one+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5))));
 128             y = r/s;
 129             return x + x*y;
 130         }
 131         if(ix < 0x3fa00000) {           /* 0.84375 <= |x| < 1.25 */
 132             s = fabsf(x)-one;
 133             P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6)))));
 134             Q = one+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6)))));
 135             if(hx>=0) return erx + P/Q; else return -erx - P/Q;
 136         }
 137         if (ix >= 0x40c00000) {         /* inf>|x|>=6 */
 138             if(hx>=0) return one-tiny; else return tiny-one;
 139         }
 140         x = fabsf(x);
 141         s = one/(x*x);
 142         if(ix< 0x4036DB6E) {    /* |x| < 1/0.35 */
 143             R=ra0+s*(ra1+s*(ra2+s*(ra3+s*(ra4+s*(
 144                                 ra5+s*(ra6+s*ra7))))));
 145             S=one+s*(sa1+s*(sa2+s*(sa3+s*(sa4+s*(
 146                                 sa5+s*(sa6+s*(sa7+s*sa8)))))));
 147         } else {        /* |x| >= 1/0.35 */
 148             R=rb0+s*(rb1+s*(rb2+s*(rb3+s*(rb4+s*(
 149                                 rb5+s*rb6)))));
 150             S=one+s*(sb1+s*(sb2+s*(sb3+s*(sb4+s*(
 151                                 sb5+s*(sb6+s*sb7))))));
 152         }
 153         GET_FLOAT_WORD(ix,x);
 154         SET_FLOAT_WORD(z,ix&0xfffff000);
 155         r  =  __ieee754_expf(-z*z-(float)0.5625)*__ieee754_expf((z-x)*(z+x)+R/S);
 156         if(hx>=0) return one-r/x; else return  r/x-one;
 157 }
 158
 159 #ifdef __STDC__
 160         float erfcf(float x)
 161 #else
 162         float erfcf(x)
 163         float x;
 164 #endif
 165 {
 166         int32_t hx,ix;
 167         float R,S,P,Q,s,y,z,r;
 168         GET_FLOAT_WORD(hx,x);
 169         ix = hx&0x7fffffff;
 170         if(ix>=0x7f800000) {                    /* erfc(nan)=nan */
 171                                                 /* erfc(+-inf)=0,2 */
 172             return (float)(((uint32_t)hx>>31)<<1)+one/x;
 173         }
 174
 175         if(ix < 0x3f580000) {           /* |x|<0.84375 */
 176             if(ix < 0x23800000)         /* |x|<2**-56 */
 177                 return one-x;
 178             z = x*x;
 179             r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4)));
 180             s = one+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5))));
 181             y = r/s;
 182             if(hx < 0x3e800000) {       /* x<1/4 */
 183                 return one-(x+x*y);
 184             } else {
 185                 r = x*y;
 186                 r += (x-half);
 187                 return half - r ;
 188             }
 189         }
 190         if(ix < 0x3fa00000) {           /* 0.84375 <= |x| < 1.25 */
 191             s = fabsf(x)-one;
 192             P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6)))));
 193             Q = one+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6)))));
 194             if(hx>=0) {
 195                 z  = one-erx; return z - P/Q;
 196             } else {
 197                 z = erx+P/Q; return one+z;
 198             }
 199         }
 200         if (ix < 0x41e00000) {          /* |x|<28 */
 201             x = fabsf(x);
 202             s = one/(x*x);
 203             if(ix< 0x4036DB6D) {        /* |x| < 1/.35 ~ 2.857143*/
 204                 R=ra0+s*(ra1+s*(ra2+s*(ra3+s*(ra4+s*(
 205                                 ra5+s*(ra6+s*ra7))))));
 206                 S=one+s*(sa1+s*(sa2+s*(sa3+s*(sa4+s*(
 207                                 sa5+s*(sa6+s*(sa7+s*sa8)))))));
 208             } else {                    /* |x| >= 1/.35 ~ 2.857143 */
 209                 if(hx<0&&ix>=0x40c00000) return two-tiny;/* x < -6 */
 210                 R=rb0+s*(rb1+s*(rb2+s*(rb3+s*(rb4+s*(
 211                                 rb5+s*rb6)))));
 212                 S=one+s*(sb1+s*(sb2+s*(sb3+s*(sb4+s*(
 213                                 sb5+s*(sb6+s*sb7))))));
 214             }
 215             GET_FLOAT_WORD(ix,x);
 216             SET_FLOAT_WORD(z,ix&0xfffff000);
 217             r  =  __ieee754_expf(-z*z-(float)0.5625)*
 218                         __ieee754_expf((z-x)*(z+x)+R/S);
 219             if(hx>0) return r/x; else return two-r/x;
 220         } else {
 221             if(hx>0) return tiny*tiny; else return two-tiny;
 222         }
 223 }