1 2 /* @(#)e_exp.c 5.1 93/09/24 */ 3 /* 4 * ==================================================== 5 * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved. 6 * 7 * Developed at SunPro, a Sun Microsystems, Inc. business. 8 * Permission to use, copy, modify, and distribute this 9 * software is freely granted, provided that this notice 10 * is preserved. 11 * ==================================================== 12 */ 13 14 /* __ieee754_exp(x) 15 * Returns the exponential of x. 16 * 17 * Method 18 * 1. Argument reduction: 19 * Reduce x to an r so that |r| <= 0.5*ln2 ~ 0.34658. 20 * Given x, find r and integer k such that 21 * 22 * x = k*ln2 + r, |r| <= 0.5*ln2. 23 * 24 * Here r will be represented as r = hi-lo for better 25 * accuracy. 26 * 27 * 2. Approximation of exp(r) by a special rational function on 28 * the interval [0,0.34658]: 29 * Write 30 * R(r**2) = r*(exp(r)+1)/(exp(r)-1) = 2 + r*r/6 - r**4/360 + ... 31 * We use a special Reme algorithm on [0,0.34658] to generate 32 * a polynomial of degree 5 to approximate R. The maximum error 33 * of this polynomial approximation is bounded by 2**-59. In 34 * other words, 35 * R(z) ~ 2.0 + P1*z + P2*z**2 + P3*z**3 + P4*z**4 + P5*z**5 36 * (where z=r*r, and the values of P1 to P5 are listed below) 37 * and 38 * | 5 | -59 39 * | 2.0+P1*z+...+P5*z - R(z) | <= 2 40 * | | 41 * The computation of exp(r) thus becomes 42 * 2*r 43 * exp(r) = 1 + ------- 44 * R - r 45 * r*R1(r) 46 * = 1 + r + ----------- (for better accuracy) 47 * 2 - R1(r) 48 * where 49 * 2 4 10 50 * R1(r) = r - (P1*r + P2*r + ... + P5*r ). 51 * 52 * 3. Scale back to obtain exp(x): 53 * From step 1, we have 54 * exp(x) = 2^k * exp(r) 55 * 56 * Special cases: 57 * exp(INF) is INF, exp(NaN) is NaN; 58 * exp(-INF) is 0, and 59 * for finite argument, only exp(0)=1 is exact. 60 * 61 * Accuracy: 62 * according to an error analysis, the error is always less than 63 * 1 ulp (unit in the last place). 64 * 65 * Misc. info. 66 * For IEEE double 67 * if x > 7.09782712893383973096e+02 then exp(x) overflow 68 * if x < -7.45133219101941108420e+02 then exp(x) underflow 69 * 70 * Constants: 71 * The hexadecimal values are the intended ones for the following 72 * constants. The decimal values may be used, provided that the 73 * compiler will convert from decimal to binary accurately enough 74 * to produce the hexadecimal values shown. 75 */ 76 77 #include "fdlibm.h" 78 79 #ifndef _DOUBLE_IS_32BITS 80 81 #ifdef __STDC__ 82 static const double 83 #else 84 static double 85 #endif 86 one = 1.0, 87 halF[2] = {0.5,-0.5,}, 88 huge = 1.0e+300, 89 twom1000= 9.33263618503218878990e-302, /* 2**-1000=0x01700000,0*/ 90 o_threshold= 7.09782712893383973096e+02, /* 0x40862E42, 0xFEFA39EF */ 91 u_threshold= -7.45133219101941108420e+02, /* 0xc0874910, 0xD52D3051 */ 92 ln2HI[2] ={ 6.93147180369123816490e-01, /* 0x3fe62e42, 0xfee00000 */ 93 -6.93147180369123816490e-01,},/* 0xbfe62e42, 0xfee00000 */ 94 ln2LO[2] ={ 1.90821492927058770002e-10, /* 0x3dea39ef, 0x35793c76 */ 95 -1.90821492927058770002e-10,},/* 0xbdea39ef, 0x35793c76 */ 96 invln2 = 1.44269504088896338700e+00, /* 0x3ff71547, 0x652b82fe */ 97 P1 = 1.66666666666666019037e-01, /* 0x3FC55555, 0x5555553E */ 98 P2 = -2.77777777770155933842e-03, /* 0xBF66C16C, 0x16BEBD93 */ 99 P3 = 6.61375632143793436117e-05, /* 0x3F11566A, 0xAF25DE2C */ 100 P4 = -1.65339022054652515390e-06, /* 0xBEBBBD41, 0xC5D26BF1 */ 101 P5 = 4.13813679705723846039e-08; /* 0x3E663769, 0x72BEA4D0 */ 102 103 104 #ifdef __STDC__ __ieee754_exp(double x)105 double __ieee754_exp(double x) /* default IEEE double exp */ 106 #else 107 double __ieee754_exp(x) /* default IEEE double exp */ 108 double x; 109 #endif 110 { 111 double y,hi,lo,c,t; 112 __int32_t k = 0,xsb; 113 __uint32_t hx; 114 115 GET_HIGH_WORD(hx,x); 116 xsb = (hx>>31)&1; /* sign bit of x */ 117 hx &= 0x7fffffff; /* high word of |x| */ 118 119 /* filter out non-finite argument */ 120 if(hx >= 0x40862E42) { /* if |x|>=709.78... */ 121 if(hx>=0x7ff00000) { 122 __uint32_t lx; 123 GET_LOW_WORD(lx,x); 124 if(((hx&0xfffff)|lx)!=0) 125 return x+x; /* NaN */ 126 else return (xsb==0)? x:0.0; /* exp(+-inf)={inf,0} */ 127 } 128 if(x > o_threshold) return huge*huge; /* overflow */ 129 if(x < u_threshold) return twom1000*twom1000; /* underflow */ 130 } 131 132 /* argument reduction */ 133 if(hx > 0x3fd62e42) { /* if |x| > 0.5 ln2 */ 134 if(hx < 0x3FF0A2B2) { /* and |x| < 1.5 ln2 */ 135 hi = x-ln2HI[xsb]; lo=ln2LO[xsb]; k = 1-xsb-xsb; 136 } else { 137 k = invln2*x+halF[xsb]; 138 t = k; 139 hi = x - t*ln2HI[0]; /* t*ln2HI is exact here */ 140 lo = t*ln2LO[0]; 141 } 142 x = hi - lo; 143 } 144 else if(hx < 0x3e300000) { /* when |x|<2**-28 */ 145 if(huge+x>one) return one+x;/* trigger inexact */ 146 } 147 148 /* x is now in primary range */ 149 t = x*x; 150 c = x - t*(P1+t*(P2+t*(P3+t*(P4+t*P5)))); 151 if(k==0) return one-((x*c)/(c-2.0)-x); 152 else y = one-((lo-(x*c)/(2.0-c))-hi); 153 if(k >= -1021) { 154 __uint32_t hy; 155 GET_HIGH_WORD(hy,y); 156 SET_HIGH_WORD(y,hy+(k<<20)); /* add k to y's exponent */ 157 return y; 158 } else { 159 __uint32_t hy; 160 GET_HIGH_WORD(hy,y); 161 SET_HIGH_WORD(y,hy+((k+1000)<<20)); /* add k to y's exponent */ 162 return y*twom1000; 163 } 164 } 165 166 #endif /* defined(_DOUBLE_IS_32BITS) */ 167