sync
[bitrig.git] / lib / libm / src / e_pow.c
blobafd10469227e06e259be532edb86b17bbd7f89be
1 /* @(#)e_pow.c 5.1 93/09/24 */
2 /*
3 * ====================================================
4 * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
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 * ====================================================
13 /* pow(x,y) return x**y
15 * n
16 * Method: Let x = 2 * (1+f)
17 * 1. Compute and return log2(x) in two pieces:
18 * log2(x) = w1 + w2,
19 * where w1 has 53-24 = 29 bit trailing zeros.
20 * 2. Perform y*log2(x) = n+y' by simulating multi-precision
21 * arithmetic, where |y'|<=0.5.
22 * 3. Return x**y = 2**n*exp(y'*log2)
24 * Special cases:
25 * 1. (anything) ** 0 is 1
26 * 2. (anything) ** 1 is itself
27 * 3. (anything) ** NAN is NAN
28 * 4. NAN ** (anything except 0) is NAN
29 * 5. +-(|x| > 1) ** +INF is +INF
30 * 6. +-(|x| > 1) ** -INF is +0
31 * 7. +-(|x| < 1) ** +INF is +0
32 * 8. +-(|x| < 1) ** -INF is +INF
33 * 9. +-1 ** +-INF is NAN
34 * 10. +0 ** (+anything except 0, NAN) is +0
35 * 11. -0 ** (+anything except 0, NAN, odd integer) is +0
36 * 12. +0 ** (-anything except 0, NAN) is +INF
37 * 13. -0 ** (-anything except 0, NAN, odd integer) is +INF
38 * 14. -0 ** (odd integer) = -( +0 ** (odd integer) )
39 * 15. +INF ** (+anything except 0,NAN) is +INF
40 * 16. +INF ** (-anything except 0,NAN) is +0
41 * 17. -INF ** (anything) = -0 ** (-anything)
42 * 18. (-anything) ** (integer) is (-1)**(integer)*(+anything**integer)
43 * 19. (-anything except 0 and inf) ** (non-integer) is NAN
45 * Accuracy:
46 * pow(x,y) returns x**y nearly rounded. In particular
47 * pow(integer,integer)
48 * always returns the correct integer provided it is
49 * representable.
51 * Constants :
52 * The hexadecimal values are the intended ones for the following
53 * constants. The decimal values may be used, provided that the
54 * compiler will convert from decimal to binary accurately enough
55 * to produce the hexadecimal values shown.
58 #include <float.h>
59 #include <math.h>
61 #include "math_private.h"
63 static const double
64 bp[] = {1.0, 1.5,},
65 dp_h[] = { 0.0, 5.84962487220764160156e-01,}, /* 0x3FE2B803, 0x40000000 */
66 dp_l[] = { 0.0, 1.35003920212974897128e-08,}, /* 0x3E4CFDEB, 0x43CFD006 */
67 zero = 0.0,
68 one = 1.0,
69 two = 2.0,
70 two53 = 9007199254740992.0, /* 0x43400000, 0x00000000 */
71 huge = 1.0e300,
72 tiny = 1.0e-300,
73 /* poly coefs for (3/2)*(log(x)-2s-2/3*s**3 */
74 L1 = 5.99999999999994648725e-01, /* 0x3FE33333, 0x33333303 */
75 L2 = 4.28571428578550184252e-01, /* 0x3FDB6DB6, 0xDB6FABFF */
76 L3 = 3.33333329818377432918e-01, /* 0x3FD55555, 0x518F264D */
77 L4 = 2.72728123808534006489e-01, /* 0x3FD17460, 0xA91D4101 */
78 L5 = 2.30660745775561754067e-01, /* 0x3FCD864A, 0x93C9DB65 */
79 L6 = 2.06975017800338417784e-01, /* 0x3FCA7E28, 0x4A454EEF */
80 P1 = 1.66666666666666019037e-01, /* 0x3FC55555, 0x5555553E */
81 P2 = -2.77777777770155933842e-03, /* 0xBF66C16C, 0x16BEBD93 */
82 P3 = 6.61375632143793436117e-05, /* 0x3F11566A, 0xAF25DE2C */
83 P4 = -1.65339022054652515390e-06, /* 0xBEBBBD41, 0xC5D26BF1 */
84 P5 = 4.13813679705723846039e-08, /* 0x3E663769, 0x72BEA4D0 */
85 lg2 = 6.93147180559945286227e-01, /* 0x3FE62E42, 0xFEFA39EF */
86 lg2_h = 6.93147182464599609375e-01, /* 0x3FE62E43, 0x00000000 */
87 lg2_l = -1.90465429995776804525e-09, /* 0xBE205C61, 0x0CA86C39 */
88 ovt = 8.0085662595372944372e-0017, /* -(1024-log2(ovfl+.5ulp)) */
89 cp = 9.61796693925975554329e-01, /* 0x3FEEC709, 0xDC3A03FD =2/(3ln2) */
90 cp_h = 9.61796700954437255859e-01, /* 0x3FEEC709, 0xE0000000 =(float)cp */
91 cp_l = -7.02846165095275826516e-09, /* 0xBE3E2FE0, 0x145B01F5 =tail of cp_h*/
92 ivln2 = 1.44269504088896338700e+00, /* 0x3FF71547, 0x652B82FE =1/ln2 */
93 ivln2_h = 1.44269502162933349609e+00, /* 0x3FF71547, 0x60000000 =24b 1/ln2*/
94 ivln2_l = 1.92596299112661746887e-08; /* 0x3E54AE0B, 0xF85DDF44 =1/ln2 tail*/
96 double
97 pow(double x, double y)
99 double z,ax,z_h,z_l,p_h,p_l;
100 double yy1,t1,t2,r,s,t,u,v,w;
101 int32_t i,j,k,yisint,n;
102 int32_t hx,hy,ix,iy;
103 u_int32_t lx,ly;
105 EXTRACT_WORDS(hx,lx,x);
106 EXTRACT_WORDS(hy,ly,y);
107 ix = hx&0x7fffffff; iy = hy&0x7fffffff;
109 /* y==zero: x**0 = 1 */
110 if((iy|ly)==0) return one;
112 /* +-NaN return x+y */
113 if(ix > 0x7ff00000 || ((ix==0x7ff00000)&&(lx!=0)) ||
114 iy > 0x7ff00000 || ((iy==0x7ff00000)&&(ly!=0)))
115 return x+y;
117 /* determine if y is an odd int when x < 0
118 * yisint = 0 ... y is not an integer
119 * yisint = 1 ... y is an odd int
120 * yisint = 2 ... y is an even int
122 yisint = 0;
123 if(hx<0) {
124 if(iy>=0x43400000) yisint = 2; /* even integer y */
125 else if(iy>=0x3ff00000) {
126 k = (iy>>20)-0x3ff; /* exponent */
127 if(k>20) {
128 j = ly>>(52-k);
129 if((j<<(52-k))==ly) yisint = 2-(j&1);
130 } else if(ly==0) {
131 j = iy>>(20-k);
132 if((j<<(20-k))==iy) yisint = 2-(j&1);
137 /* special value of y */
138 if(ly==0) {
139 if (iy==0x7ff00000) { /* y is +-inf */
140 if(((ix-0x3ff00000)|lx)==0)
141 return y - y; /* inf**+-1 is NaN */
142 else if (ix >= 0x3ff00000)/* (|x|>1)**+-inf = inf,0 */
143 return (hy>=0)? y: zero;
144 else /* (|x|<1)**-,+inf = inf,0 */
145 return (hy<0)?-y: zero;
147 if(iy==0x3ff00000) { /* y is +-1 */
148 if(hy<0) return one/x; else return x;
150 if(hy==0x40000000) return x*x; /* y is 2 */
151 if(hy==0x3fe00000) { /* y is 0.5 */
152 if(hx>=0) /* x >= +0 */
153 return sqrt(x);
157 ax = fabs(x);
158 /* special value of x */
159 if(lx==0) {
160 if(ix==0x7ff00000||ix==0||ix==0x3ff00000){
161 z = ax; /*x is +-0,+-inf,+-1*/
162 if(hy<0) z = one/z; /* z = (1/|x|) */
163 if(hx<0) {
164 if(((ix-0x3ff00000)|yisint)==0) {
165 z = (z-z)/(z-z); /* (-1)**non-int is NaN */
166 } else if(yisint==1)
167 z = -z; /* (x<0)**odd = -(|x|**odd) */
169 return z;
173 n = (hx>>31)+1;
175 /* (x<0)**(non-int) is NaN */
176 if((n|yisint)==0) return (x-x)/(x-x);
178 s = one; /* s (sign of result -ve**odd) = -1 else = 1 */
179 if((n|(yisint-1))==0) s = -one;/* (-ve)**(odd int) */
181 /* |y| is huge */
182 if(iy>0x41e00000) { /* if |y| > 2**31 */
183 if(iy>0x43f00000){ /* if |y| > 2**64, must o/uflow */
184 if(ix<=0x3fefffff) return (hy<0)? huge*huge:tiny*tiny;
185 if(ix>=0x3ff00000) return (hy>0)? huge*huge:tiny*tiny;
187 /* over/underflow if x is not close to one */
188 if(ix<0x3fefffff) return (hy<0)? s*huge*huge:s*tiny*tiny;
189 if(ix>0x3ff00000) return (hy>0)? s*huge*huge:s*tiny*tiny;
190 /* now |1-x| is tiny <= 2**-20, suffice to compute
191 log(x) by x-x^2/2+x^3/3-x^4/4 */
192 t = ax-one; /* t has 20 trailing zeros */
193 w = (t*t)*(0.5-t*(0.3333333333333333333333-t*0.25));
194 u = ivln2_h*t; /* ivln2_h has 21 sig. bits */
195 v = t*ivln2_l-w*ivln2;
196 t1 = u+v;
197 SET_LOW_WORD(t1,0);
198 t2 = v-(t1-u);
199 } else {
200 double ss,s2,s_h,s_l,t_h,t_l;
201 n = 0;
202 /* take care subnormal number */
203 if(ix<0x00100000)
204 {ax *= two53; n -= 53; GET_HIGH_WORD(ix,ax); }
205 n += ((ix)>>20)-0x3ff;
206 j = ix&0x000fffff;
207 /* determine interval */
208 ix = j|0x3ff00000; /* normalize ix */
209 if(j<=0x3988E) k=0; /* |x|<sqrt(3/2) */
210 else if(j<0xBB67A) k=1; /* |x|<sqrt(3) */
211 else {k=0;n+=1;ix -= 0x00100000;}
212 SET_HIGH_WORD(ax,ix);
214 /* compute ss = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */
215 u = ax-bp[k]; /* bp[0]=1.0, bp[1]=1.5 */
216 v = one/(ax+bp[k]);
217 ss = u*v;
218 s_h = ss;
219 SET_LOW_WORD(s_h,0);
220 /* t_h=ax+bp[k] High */
221 t_h = zero;
222 SET_HIGH_WORD(t_h,((ix>>1)|0x20000000)+0x00080000+(k<<18));
223 t_l = ax - (t_h-bp[k]);
224 s_l = v*((u-s_h*t_h)-s_h*t_l);
225 /* compute log(ax) */
226 s2 = ss*ss;
227 r = s2*s2*(L1+s2*(L2+s2*(L3+s2*(L4+s2*(L5+s2*L6)))));
228 r += s_l*(s_h+ss);
229 s2 = s_h*s_h;
230 t_h = 3.0+s2+r;
231 SET_LOW_WORD(t_h,0);
232 t_l = r-((t_h-3.0)-s2);
233 /* u+v = ss*(1+...) */
234 u = s_h*t_h;
235 v = s_l*t_h+t_l*ss;
236 /* 2/(3log2)*(ss+...) */
237 p_h = u+v;
238 SET_LOW_WORD(p_h,0);
239 p_l = v-(p_h-u);
240 z_h = cp_h*p_h; /* cp_h+cp_l = 2/(3*log2) */
241 z_l = cp_l*p_h+p_l*cp+dp_l[k];
242 /* log2(ax) = (ss+..)*2/(3*log2) = n + dp_h + z_h + z_l */
243 t = (double)n;
244 t1 = (((z_h+z_l)+dp_h[k])+t);
245 SET_LOW_WORD(t1,0);
246 t2 = z_l-(((t1-t)-dp_h[k])-z_h);
249 /* split up y into yy1+y2 and compute (yy1+y2)*(t1+t2) */
250 yy1 = y;
251 SET_LOW_WORD(yy1,0);
252 p_l = (y-yy1)*t1+y*t2;
253 p_h = yy1*t1;
254 z = p_l+p_h;
255 EXTRACT_WORDS(j,i,z);
256 if (j>=0x40900000) { /* z >= 1024 */
257 if(((j-0x40900000)|i)!=0) /* if z > 1024 */
258 return s*huge*huge; /* overflow */
259 else {
260 if(p_l+ovt>z-p_h) return s*huge*huge; /* overflow */
262 } else if((j&0x7fffffff)>=0x4090cc00 ) { /* z <= -1075 */
263 if(((j-0xc090cc00)|i)!=0) /* z < -1075 */
264 return s*tiny*tiny; /* underflow */
265 else {
266 if(p_l<=z-p_h) return s*tiny*tiny; /* underflow */
270 * compute 2**(p_h+p_l)
272 i = j&0x7fffffff;
273 k = (i>>20)-0x3ff;
274 n = 0;
275 if(i>0x3fe00000) { /* if |z| > 0.5, set n = [z+0.5] */
276 n = j+(0x00100000>>(k+1));
277 k = ((n&0x7fffffff)>>20)-0x3ff; /* new k for n */
278 t = zero;
279 SET_HIGH_WORD(t,n&~(0x000fffff>>k));
280 n = ((n&0x000fffff)|0x00100000)>>(20-k);
281 if(j<0) n = -n;
282 p_h -= t;
284 t = p_l+p_h;
285 SET_LOW_WORD(t,0);
286 u = t*lg2_h;
287 v = (p_l-(t-p_h))*lg2+t*lg2_l;
288 z = u+v;
289 w = v-(z-u);
290 t = z*z;
291 t1 = z - t*(P1+t*(P2+t*(P3+t*(P4+t*P5))));
292 r = (z*t1)/(t1-two)-(w+z*w);
293 z = one-(r-z);
294 GET_HIGH_WORD(j,z);
295 j += (n<<20);
296 if((j>>20)<=0) z = scalbn(z,n); /* subnormal output */
297 else SET_HIGH_WORD(z,j);
298 return s*z;
301 #if LDBL_MANT_DIG == 53
302 __strong_alias(powl, pow);
303 #endif /* LDBL_MANT_DIG == 53 */