corrected typo
[uclibc-ng.git] / libm / e_pow.c
blob137f600c32daa60ba45feaaa18d0af0814ffb830
1 /*
2 * ====================================================
3 * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
5 * Developed at SunPro, a Sun Microsystems, Inc. business.
6 * Permission to use, copy, modify, and distribute this
7 * software is freely granted, provided that this notice
8 * is preserved.
9 * ====================================================
12 /* __ieee754_pow(x,y) return x**y
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 muti-precision
20 * arithmetic, where |y'|<=0.5.
21 * 3. Return x**y = 2**n*exp(y'*log2)
23 * Special cases:
24 * 1. (anything) ** 0 is 1
25 * 2. (anything) ** 1 is itself
26 * 3. (anything) ** NAN is NAN
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 NAN
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
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.
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.
57 #include "math.h"
58 #include "math_private.h"
60 static const double
61 bp[] = {1.0, 1.5,},
62 dp_h[] = { 0.0, 5.84962487220764160156e-01,}, /* 0x3FE2B803, 0x40000000 */
63 dp_l[] = { 0.0, 1.35003920212974897128e-08,}, /* 0x3E4CFDEB, 0x43CFD006 */
64 zero = 0.0,
65 one = 1.0,
66 two = 2.0,
67 two53 = 9007199254740992.0, /* 0x43400000, 0x00000000 */
68 huge = 1.0e300,
69 tiny = 1.0e-300,
70 /* poly coefs for (3/2)*(log(x)-2s-2/3*s**3 */
71 L1 = 5.99999999999994648725e-01, /* 0x3FE33333, 0x33333303 */
72 L2 = 4.28571428578550184252e-01, /* 0x3FDB6DB6, 0xDB6FABFF */
73 L3 = 3.33333329818377432918e-01, /* 0x3FD55555, 0x518F264D */
74 L4 = 2.72728123808534006489e-01, /* 0x3FD17460, 0xA91D4101 */
75 L5 = 2.30660745775561754067e-01, /* 0x3FCD864A, 0x93C9DB65 */
76 L6 = 2.06975017800338417784e-01, /* 0x3FCA7E28, 0x4A454EEF */
77 P1 = 1.66666666666666019037e-01, /* 0x3FC55555, 0x5555553E */
78 P2 = -2.77777777770155933842e-03, /* 0xBF66C16C, 0x16BEBD93 */
79 P3 = 6.61375632143793436117e-05, /* 0x3F11566A, 0xAF25DE2C */
80 P4 = -1.65339022054652515390e-06, /* 0xBEBBBD41, 0xC5D26BF1 */
81 P5 = 4.13813679705723846039e-08, /* 0x3E663769, 0x72BEA4D0 */
82 lg2 = 6.93147180559945286227e-01, /* 0x3FE62E42, 0xFEFA39EF */
83 lg2_h = 6.93147182464599609375e-01, /* 0x3FE62E43, 0x00000000 */
84 lg2_l = -1.90465429995776804525e-09, /* 0xBE205C61, 0x0CA86C39 */
85 ovt = 8.0085662595372944372e-0017, /* -(1024-log2(ovfl+.5ulp)) */
86 cp = 9.61796693925975554329e-01, /* 0x3FEEC709, 0xDC3A03FD =2/(3ln2) */
87 cp_h = 9.61796700954437255859e-01, /* 0x3FEEC709, 0xE0000000 =(float)cp */
88 cp_l = -7.02846165095275826516e-09, /* 0xBE3E2FE0, 0x145B01F5 =tail of cp_h*/
89 ivln2 = 1.44269504088896338700e+00, /* 0x3FF71547, 0x652B82FE =1/ln2 */
90 ivln2_h = 1.44269502162933349609e+00, /* 0x3FF71547, 0x60000000 =24b 1/ln2*/
91 ivln2_l = 1.92596299112661746887e-08; /* 0x3E54AE0B, 0xF85DDF44 =1/ln2 tail*/
93 double attribute_hidden __ieee754_pow(double x, double y)
95 double z,ax,z_h,z_l,p_h,p_l;
96 double y1,t1,t2,r,s,t,u,v,w;
97 int32_t i,j,k,yisint,n;
98 int32_t hx,hy,ix,iy;
99 u_int32_t lx,ly;
101 EXTRACT_WORDS(hx,lx,x);
102 EXTRACT_WORDS(hy,ly,y);
103 ix = hx&0x7fffffff; iy = hy&0x7fffffff;
105 /* y==zero: x**0 = 1 */
106 if((iy|ly)==0) return one;
108 /* +-NaN return x+y */
109 if(ix > 0x7ff00000 || ((ix==0x7ff00000)&&(lx!=0)) ||
110 iy > 0x7ff00000 || ((iy==0x7ff00000)&&(ly!=0)))
111 return x+y;
113 /* determine if y is an odd int when x < 0
114 * yisint = 0 ... y is not an integer
115 * yisint = 1 ... y is an odd int
116 * yisint = 2 ... y is an even int
118 yisint = 0;
119 if(hx<0) {
120 if(iy>=0x43400000) yisint = 2; /* even integer y */
121 else if(iy>=0x3ff00000) {
122 k = (iy>>20)-0x3ff; /* exponent */
123 if(k>20) {
124 j = ly>>(52-k);
125 if((j<<(52-k))==ly) yisint = 2-(j&1);
126 } else if(ly==0) {
127 j = iy>>(20-k);
128 if((j<<(20-k))==iy) yisint = 2-(j&1);
133 /* special value of y */
134 if(ly==0) {
135 if (iy==0x7ff00000) { /* y is +-inf */
136 if(((ix-0x3ff00000)|lx)==0)
137 return y - y; /* inf**+-1 is NaN */
138 else if (ix >= 0x3ff00000)/* (|x|>1)**+-inf = inf,0 */
139 return (hy>=0)? y: zero;
140 else /* (|x|<1)**-,+inf = inf,0 */
141 return (hy<0)?-y: zero;
143 if(iy==0x3ff00000) { /* y is +-1 */
144 if(hy<0) return one/x; else return x;
146 if(hy==0x40000000) return x*x; /* y is 2 */
147 if(hy==0x3fe00000) { /* y is 0.5 */
148 if(hx>=0) /* x >= +0 */
149 return __ieee754_sqrt(x);
153 ax = fabs(x);
154 /* special value of x */
155 if(lx==0) {
156 if(ix==0x7ff00000||ix==0||ix==0x3ff00000){
157 z = ax; /*x is +-0,+-inf,+-1*/
158 if(hy<0) z = one/z; /* z = (1/|x|) */
159 if(hx<0) {
160 if(((ix-0x3ff00000)|yisint)==0) {
161 z = (z-z)/(z-z); /* (-1)**non-int is NaN */
162 } else if(yisint==1)
163 z = -z; /* (x<0)**odd = -(|x|**odd) */
165 return z;
169 /* (x<0)**(non-int) is NaN */
170 if(((((u_int32_t)hx>>31)-1)|yisint)==0) return (x-x)/(x-x);
172 /* |y| is huge */
173 if(iy>0x41e00000) { /* if |y| > 2**31 */
174 if(iy>0x43f00000){ /* if |y| > 2**64, must o/uflow */
175 if(ix<=0x3fefffff) return (hy<0)? huge*huge:tiny*tiny;
176 if(ix>=0x3ff00000) return (hy>0)? huge*huge:tiny*tiny;
178 /* over/underflow if x is not close to one */
179 if(ix<0x3fefffff) return (hy<0)? huge*huge:tiny*tiny;
180 if(ix>0x3ff00000) return (hy>0)? huge*huge:tiny*tiny;
181 /* now |1-x| is tiny <= 2**-20, suffice to compute
182 log(x) by x-x^2/2+x^3/3-x^4/4 */
183 t = x-1; /* t has 20 trailing zeros */
184 w = (t*t)*(0.5-t*(0.3333333333333333333333-t*0.25));
185 u = ivln2_h*t; /* ivln2_h has 21 sig. bits */
186 v = t*ivln2_l-w*ivln2;
187 t1 = u+v;
188 SET_LOW_WORD(t1,0);
189 t2 = v-(t1-u);
190 } else {
191 double s2,s_h,s_l,t_h,t_l;
192 n = 0;
193 /* take care subnormal number */
194 if(ix<0x00100000)
195 {ax *= two53; n -= 53; GET_HIGH_WORD(ix,ax); }
196 n += ((ix)>>20)-0x3ff;
197 j = ix&0x000fffff;
198 /* determine interval */
199 ix = j|0x3ff00000; /* normalize ix */
200 if(j<=0x3988E) k=0; /* |x|<sqrt(3/2) */
201 else if(j<0xBB67A) k=1; /* |x|<sqrt(3) */
202 else {k=0;n+=1;ix -= 0x00100000;}
203 SET_HIGH_WORD(ax,ix);
205 /* compute s = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */
206 u = ax-bp[k]; /* bp[0]=1.0, bp[1]=1.5 */
207 v = one/(ax+bp[k]);
208 s = u*v;
209 s_h = s;
210 SET_LOW_WORD(s_h,0);
211 /* t_h=ax+bp[k] High */
212 t_h = zero;
213 SET_HIGH_WORD(t_h,((ix>>1)|0x20000000)+0x00080000+(k<<18));
214 t_l = ax - (t_h-bp[k]);
215 s_l = v*((u-s_h*t_h)-s_h*t_l);
216 /* compute log(ax) */
217 s2 = s*s;
218 r = s2*s2*(L1+s2*(L2+s2*(L3+s2*(L4+s2*(L5+s2*L6)))));
219 r += s_l*(s_h+s);
220 s2 = s_h*s_h;
221 t_h = 3.0+s2+r;
222 SET_LOW_WORD(t_h,0);
223 t_l = r-((t_h-3.0)-s2);
224 /* u+v = s*(1+...) */
225 u = s_h*t_h;
226 v = s_l*t_h+t_l*s;
227 /* 2/(3log2)*(s+...) */
228 p_h = u+v;
229 SET_LOW_WORD(p_h,0);
230 p_l = v-(p_h-u);
231 z_h = cp_h*p_h; /* cp_h+cp_l = 2/(3*log2) */
232 z_l = cp_l*p_h+p_l*cp+dp_l[k];
233 /* log2(ax) = (s+..)*2/(3*log2) = n + dp_h + z_h + z_l */
234 t = (double)n;
235 t1 = (((z_h+z_l)+dp_h[k])+t);
236 SET_LOW_WORD(t1,0);
237 t2 = z_l-(((t1-t)-dp_h[k])-z_h);
240 s = one; /* s (sign of result -ve**odd) = -1 else = 1 */
241 if(((((u_int32_t)hx>>31)-1)|(yisint-1))==0)
242 s = -one;/* (-ve)**(odd int) */
244 /* split up y into y1+y2 and compute (y1+y2)*(t1+t2) */
245 y1 = y;
246 SET_LOW_WORD(y1,0);
247 p_l = (y-y1)*t1+y*t2;
248 p_h = y1*t1;
249 z = p_l+p_h;
250 EXTRACT_WORDS(j,i,z);
251 if (j>=0x40900000) { /* z >= 1024 */
252 if(((j-0x40900000)|i)!=0) /* if z > 1024 */
253 return s*huge*huge; /* overflow */
254 else {
255 if(p_l+ovt>z-p_h) return s*huge*huge; /* overflow */
257 } else if((j&0x7fffffff)>=0x4090cc00 ) { /* z <= -1075 */
258 if(((j-0xc090cc00)|i)!=0) /* z < -1075 */
259 return s*tiny*tiny; /* underflow */
260 else {
261 if(p_l<=z-p_h) return s*tiny*tiny; /* underflow */
265 * compute 2**(p_h+p_l)
267 i = j&0x7fffffff;
268 k = (i>>20)-0x3ff;
269 n = 0;
270 if(i>0x3fe00000) { /* if |z| > 0.5, set n = [z+0.5] */
271 n = j+(0x00100000>>(k+1));
272 k = ((n&0x7fffffff)>>20)-0x3ff; /* new k for n */
273 t = zero;
274 SET_HIGH_WORD(t,n&~(0x000fffff>>k));
275 n = ((n&0x000fffff)|0x00100000)>>(20-k);
276 if(j<0) n = -n;
277 p_h -= t;
279 t = p_l+p_h;
280 SET_LOW_WORD(t,0);
281 u = t*lg2_h;
282 v = (p_l-(t-p_h))*lg2+t*lg2_l;
283 z = u+v;
284 w = v-(z-u);
285 t = z*z;
286 t1 = z - t*(P1+t*(P2+t*(P3+t*(P4+t*P5))));
287 r = (z*t1)/(t1-two)-(w+z*w);
288 z = one-(r-z);
289 GET_HIGH_WORD(j,z);
290 j += (n<<20);
291 if((j>>20)<=0) z = scalbn(z,n); /* subnormal output */
292 else SET_HIGH_WORD(z,j);
293 return s*z;
297 * wrapper pow(x,y) return x**y
299 #ifndef _IEEE_LIBM
300 double pow(double x, double y)
302 double z = __ieee754_pow(x, y);
303 if (_LIB_VERSION == _IEEE_|| isnan(y))
304 return z;
305 if (isnan(x)) {
306 if (y == 0.0)
307 return __kernel_standard(x, y, 42); /* pow(NaN,0.0) */
308 return z;
310 if (x == 0.0) {
311 if (y == 0.0)
312 return __kernel_standard(x, y, 20); /* pow(0.0,0.0) */
313 if (isfinite(y) && y < 0.0)
314 return __kernel_standard(x,y,23); /* pow(0.0,negative) */
315 return z;
317 if (!isfinite(z)) {
318 if (isfinite(x) && isfinite(y)) {
319 if (isnan(z))
320 return __kernel_standard(x, y, 24); /* pow neg**non-int */
321 return __kernel_standard(x, y, 21); /* pow overflow */
324 if (z == 0.0 && isfinite(x) && isfinite(y))
325 return __kernel_standard(x, y, 22); /* pow underflow */
326 return z;
328 #else
329 strong_alias(__ieee754_pow, pow)
330 #endif
331 libm_hidden_def(pow)