1 /* 2 * ==================================================== 3 * Copyright (C) 2004 by Sun Microsystems, Inc. All rights reserved. 4 * 5 * Permission to use, copy, modify, and distribute this 6 * software is freely granted, provided that this notice 7 * is preserved. 8 * ==================================================== 9 */ 10 11 #include <sys/cdefs.h> 12 /* pow(x,y) return x**y 13 * 14 * n 15 * Method: Let x = 2 * (1+f) 16 * 1. Compute and return log2(x) in two pieces: 17 * log2(x) = w1 + w2, 18 * where w1 has 53-24 = 29 bit trailing zeros. 19 * 2. Perform y*log2(x) = n+y' by simulating multi-precision 20 * arithmetic, where |y'|<=0.5. 21 * 3. Return x**y = 2**n*exp(y'*log2) 22 * 23 * Special cases: 24 * 1. (anything) ** 0 is 1 25 * 2. (anything) ** 1 is itself 26 * 3. (anything) ** NAN is NAN except 1 ** NAN = 1 27 * 4. NAN ** (anything except 0) is NAN 28 * 5. +-(|x| > 1) ** +INF is +INF 29 * 6. +-(|x| > 1) ** -INF is +0 30 * 7. +-(|x| < 1) ** +INF is +0 31 * 8. +-(|x| < 1) ** -INF is +INF 32 * 9. +-1 ** +-INF is 1 33 * 10. +0 ** (+anything except 0, NAN) is +0 34 * 11. -0 ** (+anything except 0, NAN, odd integer) is +0 35 * 12. +0 ** (-anything except 0, NAN) is +INF 36 * 13. -0 ** (-anything except 0, NAN, odd integer) is +INF 37 * 14. -0 ** (odd integer) = -( +0 ** (odd integer) ) 38 * 15. +INF ** (+anything except 0,NAN) is +INF 39 * 16. +INF ** (-anything except 0,NAN) is +0 40 * 17. -INF ** (anything) = -0 ** (-anything) 41 * 18. (-anything) ** (integer) is (-1)**(integer)*(+anything**integer) 42 * 19. (-anything except 0 and inf) ** (non-integer) is NAN 43 * 44 * Accuracy: 45 * pow(x,y) returns x**y nearly rounded. In particular 46 * pow(integer,integer) 47 * always returns the correct integer provided it is 48 * representable. 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 <float.h> 58 #include "math.h" 59 #include "math_private.h" 60 61 static const double 62 bp[] = {1.0, 1.5,}, 63 dp_h[] = { 0.0, 5.84962487220764160156e-01,}, /* 0x3FE2B803, 0x40000000 */ 64 dp_l[] = { 0.0, 1.35003920212974897128e-08,}, /* 0x3E4CFDEB, 0x43CFD006 */ 65 zero = 0.0, 66 half = 0.5, 67 qrtr = 0.25, 68 thrd = 3.3333333333333331e-01, /* 0x3fd55555, 0x55555555 */ 69 one = 1.0, 70 two = 2.0, 71 two53 = 9007199254740992.0, /* 0x43400000, 0x00000000 */ 72 huge = 1.0e300, 73 tiny = 1.0e-300, 74 /* poly coefs for (3/2)*(log(x)-2s-2/3*s**3 */ 75 L1 = 5.99999999999994648725e-01, /* 0x3FE33333, 0x33333303 */ 76 L2 = 4.28571428578550184252e-01, /* 0x3FDB6DB6, 0xDB6FABFF */ 77 L3 = 3.33333329818377432918e-01, /* 0x3FD55555, 0x518F264D */ 78 L4 = 2.72728123808534006489e-01, /* 0x3FD17460, 0xA91D4101 */ 79 L5 = 2.30660745775561754067e-01, /* 0x3FCD864A, 0x93C9DB65 */ 80 L6 = 2.06975017800338417784e-01, /* 0x3FCA7E28, 0x4A454EEF */ 81 P1 = 1.66666666666666019037e-01, /* 0x3FC55555, 0x5555553E */ 82 P2 = -2.77777777770155933842e-03, /* 0xBF66C16C, 0x16BEBD93 */ 83 P3 = 6.61375632143793436117e-05, /* 0x3F11566A, 0xAF25DE2C */ 84 P4 = -1.65339022054652515390e-06, /* 0xBEBBBD41, 0xC5D26BF1 */ 85 P5 = 4.13813679705723846039e-08, /* 0x3E663769, 0x72BEA4D0 */ 86 lg2 = 6.93147180559945286227e-01, /* 0x3FE62E42, 0xFEFA39EF */ 87 lg2_h = 6.93147182464599609375e-01, /* 0x3FE62E43, 0x00000000 */ 88 lg2_l = -1.90465429995776804525e-09, /* 0xBE205C61, 0x0CA86C39 */ 89 ovt = 8.0085662595372944372e-0017, /* -(1024-log2(ovfl+.5ulp)) */ 90 cp = 9.61796693925975554329e-01, /* 0x3FEEC709, 0xDC3A03FD =2/(3ln2) */ 91 cp_h = 9.61796700954437255859e-01, /* 0x3FEEC709, 0xE0000000 =(float)cp */ 92 cp_l = -7.02846165095275826516e-09, /* 0xBE3E2FE0, 0x145B01F5 =tail of cp_h*/ 93 ivln2 = 1.44269504088896338700e+00, /* 0x3FF71547, 0x652B82FE =1/ln2 */ 94 ivln2_h = 1.44269502162933349609e+00, /* 0x3FF71547, 0x60000000 =24b 1/ln2*/ 95 ivln2_l = 1.92596299112661746887e-08; /* 0x3E54AE0B, 0xF85DDF44 =1/ln2 tail*/ 96 97 double 98 pow(double x, double y) 99 { 100 double z,ax,z_h,z_l,p_h,p_l; 101 double y1,t1,t2,r,s,t,u,v,w; 102 int32_t i,j,k,yisint,n; 103 int32_t hx,hy,ix,iy; 104 u_int32_t lx,ly; 105 106 EXTRACT_WORDS(hx,lx,x); 107 EXTRACT_WORDS(hy,ly,y); 108 ix = hx&0x7fffffff; iy = hy&0x7fffffff; 109 110 /* y==zero: x**0 = 1 */ 111 if((iy|ly)==0) return one; 112 113 /* x==1: 1**y = 1, even if y is NaN */ 114 if (hx==0x3ff00000 && lx == 0) return one; 115 116 /* y!=zero: result is NaN if either arg is NaN */ 117 if(ix > 0x7ff00000 || ((ix==0x7ff00000)&&(lx!=0)) || 118 iy > 0x7ff00000 || ((iy==0x7ff00000)&&(ly!=0))) 119 return nan_mix(x, y); 120 121 /* determine if y is an odd int when x < 0 122 * yisint = 0 ... y is not an integer 123 * yisint = 1 ... y is an odd int 124 * yisint = 2 ... y is an even int 125 */ 126 yisint = 0; 127 if(hx<0) { 128 if(iy>=0x43400000) yisint = 2; /* even integer y */ 129 else if(iy>=0x3ff00000) { 130 k = (iy>>20)-0x3ff; /* exponent */ 131 if(k>20) { 132 j = ly>>(52-k); 133 if(((u_int32_t)j<<(52-k))==ly) yisint = 2-(j&1); 134 } else if(ly==0) { 135 j = iy>>(20-k); 136 if((j<<(20-k))==iy) yisint = 2-(j&1); 137 } 138 } 139 } 140 141 /* special value of y */ 142 if(ly==0) { 143 if (iy==0x7ff00000) { /* y is +-inf */ 144 if(((ix-0x3ff00000)|lx)==0) 145 return one; /* (-1)**+-inf is 1 */ 146 else if (ix >= 0x3ff00000)/* (|x|>1)**+-inf = inf,0 */ 147 return (hy>=0)? y: zero; 148 else /* (|x|<1)**-,+inf = inf,0 */ 149 return (hy<0)?-y: zero; 150 } 151 if(iy==0x3ff00000) { /* y is +-1 */ 152 if(hy<0) return one/x; else return x; 153 } 154 if(hy==0x40000000) return x*x; /* y is 2 */ 155 if(hy==0x3fe00000) { /* y is 0.5 */ 156 if(hx>=0) /* x >= +0 */ 157 return sqrt(x); 158 } 159 } 160 161 ax = fabs(x); 162 /* special value of x */ 163 if(lx==0) { 164 if(ix==0x7ff00000||ix==0||ix==0x3ff00000){ 165 z = ax; /*x is +-0,+-inf,+-1*/ 166 if(hy<0) z = one/z; /* z = (1/|x|) */ 167 if(hx<0) { 168 if(((ix-0x3ff00000)|yisint)==0) { 169 z = (z-z)/(z-z); /* (-1)**non-int is NaN */ 170 } else if(yisint==1) 171 z = -z; /* (x<0)**odd = -(|x|**odd) */ 172 } 173 return z; 174 } 175 } 176 177 /* CYGNUS LOCAL + fdlibm-5.3 fix: This used to be 178 n = (hx>>31)+1; 179 but ANSI C says a right shift of a signed negative quantity is 180 implementation defined. */ 181 n = ((u_int32_t)hx>>31)-1; 182 183 /* (x<0)**(non-int) is NaN */ 184 if((n|yisint)==0) return (x-x)/(x-x); 185 186 s = one; /* s (sign of result -ve**odd) = -1 else = 1 */ 187 if((n|(yisint-1))==0) s = -one;/* (-ve)**(odd int) */ 188 189 /* |y| is huge */ 190 if(iy>0x41e00000) { /* if |y| > 2**31 */ 191 if(iy>0x43f00000){ /* if |y| > 2**64, must o/uflow */ 192 if(ix<=0x3fefffff) return (hy<0)? huge*huge:tiny*tiny; 193 if(ix>=0x3ff00000) return (hy>0)? huge*huge:tiny*tiny; 194 } 195 /* over/underflow if x is not close to one */ 196 if(ix<0x3fefffff) return (hy<0)? s*huge*huge:s*tiny*tiny; 197 if(ix>0x3ff00000) return (hy>0)? s*huge*huge:s*tiny*tiny; 198 /* now |1-x| is tiny <= 2**-20, suffice to compute 199 log(x) by x-x^2/2+x^3/3-x^4/4 */ 200 t = ax-one; /* t has 20 trailing zeros */ 201 w = (t*t)*(half-t*(thrd-t*qrtr)); 202 u = ivln2_h*t; /* ivln2_h has 21 sig. bits */ 203 v = t*ivln2_l-w*ivln2; 204 t1 = u+v; 205 SET_LOW_WORD(t1,0); 206 t2 = v-(t1-u); 207 } else { 208 double ss,s2,s_h,s_l,t_h,t_l; 209 n = 0; 210 /* take care subnormal number */ 211 if(ix<0x00100000) 212 {ax *= two53; n -= 53; GET_HIGH_WORD(ix,ax); } 213 n += ((ix)>>20)-0x3ff; 214 j = ix&0x000fffff; 215 /* determine interval */ 216 ix = j|0x3ff00000; /* normalize ix */ 217 if(j<=0x3988E) k=0; /* |x|<sqrt(3/2) */ 218 else if(j<0xBB67A) k=1; /* |x|<sqrt(3) */ 219 else {k=0;n+=1;ix -= 0x00100000;} 220 SET_HIGH_WORD(ax,ix); 221 222 /* compute ss = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */ 223 u = ax-bp[k]; /* bp[0]=1.0, bp[1]=1.5 */ 224 v = one/(ax+bp[k]); 225 ss = u*v; 226 s_h = ss; 227 SET_LOW_WORD(s_h,0); 228 /* t_h=ax+bp[k] High */ 229 t_h = zero; 230 SET_HIGH_WORD(t_h,((ix>>1)|0x20000000)+0x00080000+(k<<18)); 231 t_l = ax - (t_h-bp[k]); 232 s_l = v*((u-s_h*t_h)-s_h*t_l); 233 /* compute log(ax) */ 234 s2 = ss*ss; 235 r = s2*s2*(L1+s2*(L2+s2*(L3+s2*(L4+s2*(L5+s2*L6))))); 236 r += s_l*(s_h+ss); 237 s2 = s_h*s_h; 238 t_h = 3+s2+r; 239 SET_LOW_WORD(t_h,0); 240 t_l = r-((t_h-3)-s2); 241 /* u+v = ss*(1+...) */ 242 u = s_h*t_h; 243 v = s_l*t_h+t_l*ss; 244 /* 2/(3log2)*(ss+...) */ 245 p_h = u+v; 246 SET_LOW_WORD(p_h,0); 247 p_l = v-(p_h-u); 248 z_h = cp_h*p_h; /* cp_h+cp_l = 2/(3*log2) */ 249 z_l = cp_l*p_h+p_l*cp+dp_l[k]; 250 /* log2(ax) = (ss+..)*2/(3*log2) = n + dp_h + z_h + z_l */ 251 t = n; 252 t1 = (((z_h+z_l)+dp_h[k])+t); 253 SET_LOW_WORD(t1,0); 254 t2 = z_l-(((t1-t)-dp_h[k])-z_h); 255 } 256 257 /* split up y into y1+y2 and compute (y1+y2)*(t1+t2) */ 258 y1 = y; 259 SET_LOW_WORD(y1,0); 260 p_l = (y-y1)*t1+y*t2; 261 p_h = y1*t1; 262 z = p_l+p_h; 263 EXTRACT_WORDS(j,i,z); 264 if (j>=0x40900000) { /* z >= 1024 */ 265 if(((j-0x40900000)|i)!=0) /* if z > 1024 */ 266 return s*huge*huge; /* overflow */ 267 else { 268 if(p_l+ovt>z-p_h) return s*huge*huge; /* overflow */ 269 } 270 } else if((j&0x7fffffff)>=0x4090cc00 ) { /* z <= -1075 */ 271 if(((j-0xc090cc00)|i)!=0) /* z < -1075 */ 272 return s*tiny*tiny; /* underflow */ 273 else { 274 if(p_l<=z-p_h) return s*tiny*tiny; /* underflow */ 275 } 276 } 277 /* 278 * compute 2**(p_h+p_l) 279 */ 280 i = j&0x7fffffff; 281 k = (i>>20)-0x3ff; 282 n = 0; 283 if(i>0x3fe00000) { /* if |z| > 0.5, set n = [z+0.5] */ 284 n = j+(0x00100000>>(k+1)); 285 k = ((n&0x7fffffff)>>20)-0x3ff; /* new k for n */ 286 t = zero; 287 SET_HIGH_WORD(t,n&~(0x000fffff>>k)); 288 n = ((n&0x000fffff)|0x00100000)>>(20-k); 289 if(j<0) n = -n; 290 p_h -= t; 291 } 292 t = p_l+p_h; 293 SET_LOW_WORD(t,0); 294 u = t*lg2_h; 295 v = (p_l-(t-p_h))*lg2+t*lg2_l; 296 z = u+v; 297 w = v-(z-u); 298 t = z*z; 299 t1 = z - t*(P1+t*(P2+t*(P3+t*(P4+t*P5)))); 300 r = (z*t1)/(t1-two)-(w+z*w); 301 z = one-(r-z); 302 GET_HIGH_WORD(j,z); 303 j += (n<<20); 304 if((j>>20)<=0) z = scalbn(z,n); /* subnormal output */ 305 else SET_HIGH_WORD(z,j); 306 return s*z; 307 } 308 309 #if (LDBL_MANT_DIG == 53) 310 __weak_reference(pow, powl); 311 #endif 312