1 |
|
2 /* |
|
3 * Copyright (c) 1998, 2004, Oracle and/or its affiliates. All rights reserved. |
|
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
|
5 * |
|
6 * This code is free software; you can redistribute it and/or modify it |
|
7 * under the terms of the GNU General Public License version 2 only, as |
|
8 * published by the Free Software Foundation. Oracle designates this |
|
9 * particular file as subject to the "Classpath" exception as provided |
|
10 * by Oracle in the LICENSE file that accompanied this code. |
|
11 * |
|
12 * This code is distributed in the hope that it will be useful, but WITHOUT |
|
13 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
|
14 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
|
15 * version 2 for more details (a copy is included in the LICENSE file that |
|
16 * accompanied this code). |
|
17 * |
|
18 * You should have received a copy of the GNU General Public License version |
|
19 * 2 along with this work; if not, write to the Free Software Foundation, |
|
20 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
|
21 * |
|
22 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
|
23 * or visit www.oracle.com if you need additional information or have any |
|
24 * questions. |
|
25 */ |
|
26 |
|
27 /* __ieee754_pow(x,y) return x**y |
|
28 * |
|
29 * n |
|
30 * Method: Let x = 2 * (1+f) |
|
31 * 1. Compute and return log2(x) in two pieces: |
|
32 * log2(x) = w1 + w2, |
|
33 * where w1 has 53-24 = 29 bit trailing zeros. |
|
34 * 2. Perform y*log2(x) = n+y' by simulating muti-precision |
|
35 * arithmetic, where |y'|<=0.5. |
|
36 * 3. Return x**y = 2**n*exp(y'*log2) |
|
37 * |
|
38 * Special cases: |
|
39 * 1. (anything) ** 0 is 1 |
|
40 * 2. (anything) ** 1 is itself |
|
41 * 3. (anything) ** NAN is NAN |
|
42 * 4. NAN ** (anything except 0) is NAN |
|
43 * 5. +-(|x| > 1) ** +INF is +INF |
|
44 * 6. +-(|x| > 1) ** -INF is +0 |
|
45 * 7. +-(|x| < 1) ** +INF is +0 |
|
46 * 8. +-(|x| < 1) ** -INF is +INF |
|
47 * 9. +-1 ** +-INF is NAN |
|
48 * 10. +0 ** (+anything except 0, NAN) is +0 |
|
49 * 11. -0 ** (+anything except 0, NAN, odd integer) is +0 |
|
50 * 12. +0 ** (-anything except 0, NAN) is +INF |
|
51 * 13. -0 ** (-anything except 0, NAN, odd integer) is +INF |
|
52 * 14. -0 ** (odd integer) = -( +0 ** (odd integer) ) |
|
53 * 15. +INF ** (+anything except 0,NAN) is +INF |
|
54 * 16. +INF ** (-anything except 0,NAN) is +0 |
|
55 * 17. -INF ** (anything) = -0 ** (-anything) |
|
56 * 18. (-anything) ** (integer) is (-1)**(integer)*(+anything**integer) |
|
57 * 19. (-anything except 0 and inf) ** (non-integer) is NAN |
|
58 * |
|
59 * Accuracy: |
|
60 * pow(x,y) returns x**y nearly rounded. In particular |
|
61 * pow(integer,integer) |
|
62 * always returns the correct integer provided it is |
|
63 * representable. |
|
64 * |
|
65 * Constants : |
|
66 * The hexadecimal values are the intended ones for the following |
|
67 * constants. The decimal values may be used, provided that the |
|
68 * compiler will convert from decimal to binary accurately enough |
|
69 * to produce the hexadecimal values shown. |
|
70 */ |
|
71 |
|
72 #include "fdlibm.h" |
|
73 |
|
74 #ifdef __STDC__ |
|
75 static const double |
|
76 #else |
|
77 static double |
|
78 #endif |
|
79 bp[] = {1.0, 1.5,}, |
|
80 dp_h[] = { 0.0, 5.84962487220764160156e-01,}, /* 0x3FE2B803, 0x40000000 */ |
|
81 dp_l[] = { 0.0, 1.35003920212974897128e-08,}, /* 0x3E4CFDEB, 0x43CFD006 */ |
|
82 zero = 0.0, |
|
83 one = 1.0, |
|
84 two = 2.0, |
|
85 two53 = 9007199254740992.0, /* 0x43400000, 0x00000000 */ |
|
86 huge = 1.0e300, |
|
87 tiny = 1.0e-300, |
|
88 /* poly coefs for (3/2)*(log(x)-2s-2/3*s**3 */ |
|
89 L1 = 5.99999999999994648725e-01, /* 0x3FE33333, 0x33333303 */ |
|
90 L2 = 4.28571428578550184252e-01, /* 0x3FDB6DB6, 0xDB6FABFF */ |
|
91 L3 = 3.33333329818377432918e-01, /* 0x3FD55555, 0x518F264D */ |
|
92 L4 = 2.72728123808534006489e-01, /* 0x3FD17460, 0xA91D4101 */ |
|
93 L5 = 2.30660745775561754067e-01, /* 0x3FCD864A, 0x93C9DB65 */ |
|
94 L6 = 2.06975017800338417784e-01, /* 0x3FCA7E28, 0x4A454EEF */ |
|
95 P1 = 1.66666666666666019037e-01, /* 0x3FC55555, 0x5555553E */ |
|
96 P2 = -2.77777777770155933842e-03, /* 0xBF66C16C, 0x16BEBD93 */ |
|
97 P3 = 6.61375632143793436117e-05, /* 0x3F11566A, 0xAF25DE2C */ |
|
98 P4 = -1.65339022054652515390e-06, /* 0xBEBBBD41, 0xC5D26BF1 */ |
|
99 P5 = 4.13813679705723846039e-08, /* 0x3E663769, 0x72BEA4D0 */ |
|
100 lg2 = 6.93147180559945286227e-01, /* 0x3FE62E42, 0xFEFA39EF */ |
|
101 lg2_h = 6.93147182464599609375e-01, /* 0x3FE62E43, 0x00000000 */ |
|
102 lg2_l = -1.90465429995776804525e-09, /* 0xBE205C61, 0x0CA86C39 */ |
|
103 ovt = 8.0085662595372944372e-0017, /* -(1024-log2(ovfl+.5ulp)) */ |
|
104 cp = 9.61796693925975554329e-01, /* 0x3FEEC709, 0xDC3A03FD =2/(3ln2) */ |
|
105 cp_h = 9.61796700954437255859e-01, /* 0x3FEEC709, 0xE0000000 =(float)cp */ |
|
106 cp_l = -7.02846165095275826516e-09, /* 0xBE3E2FE0, 0x145B01F5 =tail of cp_h*/ |
|
107 ivln2 = 1.44269504088896338700e+00, /* 0x3FF71547, 0x652B82FE =1/ln2 */ |
|
108 ivln2_h = 1.44269502162933349609e+00, /* 0x3FF71547, 0x60000000 =24b 1/ln2*/ |
|
109 ivln2_l = 1.92596299112661746887e-08; /* 0x3E54AE0B, 0xF85DDF44 =1/ln2 tail*/ |
|
110 |
|
111 #ifdef __STDC__ |
|
112 double __ieee754_pow(double x, double y) |
|
113 #else |
|
114 double __ieee754_pow(x,y) |
|
115 double x, y; |
|
116 #endif |
|
117 { |
|
118 double z,ax,z_h,z_l,p_h,p_l; |
|
119 double y1,t1,t2,r,s,t,u,v,w; |
|
120 int i0,i1,i,j,k,yisint,n; |
|
121 int hx,hy,ix,iy; |
|
122 unsigned lx,ly; |
|
123 |
|
124 i0 = ((*(int*)&one)>>29)^1; i1=1-i0; |
|
125 hx = __HI(x); lx = __LO(x); |
|
126 hy = __HI(y); ly = __LO(y); |
|
127 ix = hx&0x7fffffff; iy = hy&0x7fffffff; |
|
128 |
|
129 /* y==zero: x**0 = 1 */ |
|
130 if((iy|ly)==0) return one; |
|
131 |
|
132 /* +-NaN return x+y */ |
|
133 if(ix > 0x7ff00000 || ((ix==0x7ff00000)&&(lx!=0)) || |
|
134 iy > 0x7ff00000 || ((iy==0x7ff00000)&&(ly!=0))) |
|
135 return x+y; |
|
136 |
|
137 /* determine if y is an odd int when x < 0 |
|
138 * yisint = 0 ... y is not an integer |
|
139 * yisint = 1 ... y is an odd int |
|
140 * yisint = 2 ... y is an even int |
|
141 */ |
|
142 yisint = 0; |
|
143 if(hx<0) { |
|
144 if(iy>=0x43400000) yisint = 2; /* even integer y */ |
|
145 else if(iy>=0x3ff00000) { |
|
146 k = (iy>>20)-0x3ff; /* exponent */ |
|
147 if(k>20) { |
|
148 j = ly>>(52-k); |
|
149 if((j<<(52-k))==ly) yisint = 2-(j&1); |
|
150 } else if(ly==0) { |
|
151 j = iy>>(20-k); |
|
152 if((j<<(20-k))==iy) yisint = 2-(j&1); |
|
153 } |
|
154 } |
|
155 } |
|
156 |
|
157 /* special value of y */ |
|
158 if(ly==0) { |
|
159 if (iy==0x7ff00000) { /* y is +-inf */ |
|
160 if(((ix-0x3ff00000)|lx)==0) |
|
161 return y - y; /* inf**+-1 is NaN */ |
|
162 else if (ix >= 0x3ff00000)/* (|x|>1)**+-inf = inf,0 */ |
|
163 return (hy>=0)? y: zero; |
|
164 else /* (|x|<1)**-,+inf = inf,0 */ |
|
165 return (hy<0)?-y: zero; |
|
166 } |
|
167 if(iy==0x3ff00000) { /* y is +-1 */ |
|
168 if(hy<0) return one/x; else return x; |
|
169 } |
|
170 if(hy==0x40000000) return x*x; /* y is 2 */ |
|
171 if(hy==0x3fe00000) { /* y is 0.5 */ |
|
172 if(hx>=0) /* x >= +0 */ |
|
173 return sqrt(x); |
|
174 } |
|
175 } |
|
176 |
|
177 ax = fabs(x); |
|
178 /* special value of x */ |
|
179 if(lx==0) { |
|
180 if(ix==0x7ff00000||ix==0||ix==0x3ff00000){ |
|
181 z = ax; /*x is +-0,+-inf,+-1*/ |
|
182 if(hy<0) z = one/z; /* z = (1/|x|) */ |
|
183 if(hx<0) { |
|
184 if(((ix-0x3ff00000)|yisint)==0) { |
|
185 z = (z-z)/(z-z); /* (-1)**non-int is NaN */ |
|
186 } else if(yisint==1) |
|
187 z = -1.0*z; /* (x<0)**odd = -(|x|**odd) */ |
|
188 } |
|
189 return z; |
|
190 } |
|
191 } |
|
192 |
|
193 n = (hx>>31)+1; |
|
194 |
|
195 /* (x<0)**(non-int) is NaN */ |
|
196 if((n|yisint)==0) return (x-x)/(x-x); |
|
197 |
|
198 s = one; /* s (sign of result -ve**odd) = -1 else = 1 */ |
|
199 if((n|(yisint-1))==0) s = -one;/* (-ve)**(odd int) */ |
|
200 |
|
201 /* |y| is huge */ |
|
202 if(iy>0x41e00000) { /* if |y| > 2**31 */ |
|
203 if(iy>0x43f00000){ /* if |y| > 2**64, must o/uflow */ |
|
204 if(ix<=0x3fefffff) return (hy<0)? huge*huge:tiny*tiny; |
|
205 if(ix>=0x3ff00000) return (hy>0)? huge*huge:tiny*tiny; |
|
206 } |
|
207 /* over/underflow if x is not close to one */ |
|
208 if(ix<0x3fefffff) return (hy<0)? s*huge*huge:s*tiny*tiny; |
|
209 if(ix>0x3ff00000) return (hy>0)? s*huge*huge:s*tiny*tiny; |
|
210 /* now |1-x| is tiny <= 2**-20, suffice to compute |
|
211 log(x) by x-x^2/2+x^3/3-x^4/4 */ |
|
212 t = ax-one; /* t has 20 trailing zeros */ |
|
213 w = (t*t)*(0.5-t*(0.3333333333333333333333-t*0.25)); |
|
214 u = ivln2_h*t; /* ivln2_h has 21 sig. bits */ |
|
215 v = t*ivln2_l-w*ivln2; |
|
216 t1 = u+v; |
|
217 __LO(t1) = 0; |
|
218 t2 = v-(t1-u); |
|
219 } else { |
|
220 double ss,s2,s_h,s_l,t_h,t_l; |
|
221 n = 0; |
|
222 /* take care subnormal number */ |
|
223 if(ix<0x00100000) |
|
224 {ax *= two53; n -= 53; ix = __HI(ax); } |
|
225 n += ((ix)>>20)-0x3ff; |
|
226 j = ix&0x000fffff; |
|
227 /* determine interval */ |
|
228 ix = j|0x3ff00000; /* normalize ix */ |
|
229 if(j<=0x3988E) k=0; /* |x|<sqrt(3/2) */ |
|
230 else if(j<0xBB67A) k=1; /* |x|<sqrt(3) */ |
|
231 else {k=0;n+=1;ix -= 0x00100000;} |
|
232 __HI(ax) = ix; |
|
233 |
|
234 /* compute ss = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */ |
|
235 u = ax-bp[k]; /* bp[0]=1.0, bp[1]=1.5 */ |
|
236 v = one/(ax+bp[k]); |
|
237 ss = u*v; |
|
238 s_h = ss; |
|
239 __LO(s_h) = 0; |
|
240 /* t_h=ax+bp[k] High */ |
|
241 t_h = zero; |
|
242 __HI(t_h)=((ix>>1)|0x20000000)+0x00080000+(k<<18); |
|
243 t_l = ax - (t_h-bp[k]); |
|
244 s_l = v*((u-s_h*t_h)-s_h*t_l); |
|
245 /* compute log(ax) */ |
|
246 s2 = ss*ss; |
|
247 r = s2*s2*(L1+s2*(L2+s2*(L3+s2*(L4+s2*(L5+s2*L6))))); |
|
248 r += s_l*(s_h+ss); |
|
249 s2 = s_h*s_h; |
|
250 t_h = 3.0+s2+r; |
|
251 __LO(t_h) = 0; |
|
252 t_l = r-((t_h-3.0)-s2); |
|
253 /* u+v = ss*(1+...) */ |
|
254 u = s_h*t_h; |
|
255 v = s_l*t_h+t_l*ss; |
|
256 /* 2/(3log2)*(ss+...) */ |
|
257 p_h = u+v; |
|
258 __LO(p_h) = 0; |
|
259 p_l = v-(p_h-u); |
|
260 z_h = cp_h*p_h; /* cp_h+cp_l = 2/(3*log2) */ |
|
261 z_l = cp_l*p_h+p_l*cp+dp_l[k]; |
|
262 /* log2(ax) = (ss+..)*2/(3*log2) = n + dp_h + z_h + z_l */ |
|
263 t = (double)n; |
|
264 t1 = (((z_h+z_l)+dp_h[k])+t); |
|
265 __LO(t1) = 0; |
|
266 t2 = z_l-(((t1-t)-dp_h[k])-z_h); |
|
267 } |
|
268 |
|
269 /* split up y into y1+y2 and compute (y1+y2)*(t1+t2) */ |
|
270 y1 = y; |
|
271 __LO(y1) = 0; |
|
272 p_l = (y-y1)*t1+y*t2; |
|
273 p_h = y1*t1; |
|
274 z = p_l+p_h; |
|
275 j = __HI(z); |
|
276 i = __LO(z); |
|
277 if (j>=0x40900000) { /* z >= 1024 */ |
|
278 if(((j-0x40900000)|i)!=0) /* if z > 1024 */ |
|
279 return s*huge*huge; /* overflow */ |
|
280 else { |
|
281 if(p_l+ovt>z-p_h) return s*huge*huge; /* overflow */ |
|
282 } |
|
283 } else if((j&0x7fffffff)>=0x4090cc00 ) { /* z <= -1075 */ |
|
284 if(((j-0xc090cc00)|i)!=0) /* z < -1075 */ |
|
285 return s*tiny*tiny; /* underflow */ |
|
286 else { |
|
287 if(p_l<=z-p_h) return s*tiny*tiny; /* underflow */ |
|
288 } |
|
289 } |
|
290 /* |
|
291 * compute 2**(p_h+p_l) |
|
292 */ |
|
293 i = j&0x7fffffff; |
|
294 k = (i>>20)-0x3ff; |
|
295 n = 0; |
|
296 if(i>0x3fe00000) { /* if |z| > 0.5, set n = [z+0.5] */ |
|
297 n = j+(0x00100000>>(k+1)); |
|
298 k = ((n&0x7fffffff)>>20)-0x3ff; /* new k for n */ |
|
299 t = zero; |
|
300 __HI(t) = (n&~(0x000fffff>>k)); |
|
301 n = ((n&0x000fffff)|0x00100000)>>(20-k); |
|
302 if(j<0) n = -n; |
|
303 p_h -= t; |
|
304 } |
|
305 t = p_l+p_h; |
|
306 __LO(t) = 0; |
|
307 u = t*lg2_h; |
|
308 v = (p_l-(t-p_h))*lg2+t*lg2_l; |
|
309 z = u+v; |
|
310 w = v-(z-u); |
|
311 t = z*z; |
|
312 t1 = z - t*(P1+t*(P2+t*(P3+t*(P4+t*P5)))); |
|
313 r = (z*t1)/(t1-two)-(w+z*w); |
|
314 z = one-(r-z); |
|
315 j = __HI(z); |
|
316 j += (n<<20); |
|
317 if((j>>20)<=0) z = scalbn(z,n); /* subnormal output */ |
|
318 else __HI(z) += (n<<20); |
|
319 return s*z; |
|
320 } |
|