1 |
|
---|
2 | /* @(#)e_pow.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_pow(x,y) return x**y
|
---|
15 | *
|
---|
16 | * n
|
---|
17 | * Method: Let x = 2 * (1+f)
|
---|
18 | * 1. Compute and return log2(x) in two pieces:
|
---|
19 | * log2(x) = w1 + w2,
|
---|
20 | * where w1 has 53-24 = 29 bit trailing zeros.
|
---|
21 | * 2. Perform y*log2(x) = n+y' by simulating muti-precision
|
---|
22 | * arithmetic, where |y'|<=0.5.
|
---|
23 | * 3. Return x**y = 2**n*exp(y'*log2)
|
---|
24 | *
|
---|
25 | * Special cases:
|
---|
26 | * 1. (anything) ** 0 is 1
|
---|
27 | * 2. (anything) ** 1 is itself
|
---|
28 | * 3. (anything) ** NAN is NAN
|
---|
29 | * 4. NAN ** (anything except 0) is NAN
|
---|
30 | * 5. +-(|x| > 1) ** +INF is +INF
|
---|
31 | * 6. +-(|x| > 1) ** -INF is +0
|
---|
32 | * 7. +-(|x| < 1) ** +INF is +0
|
---|
33 | * 8. +-(|x| < 1) ** -INF is +INF
|
---|
34 | * 9. +-1 ** +-INF is NAN
|
---|
35 | * 10. +0 ** (+anything except 0, NAN) is +0
|
---|
36 | * 11. -0 ** (+anything except 0, NAN, odd integer) is +0
|
---|
37 | * 12. +0 ** (-anything except 0, NAN) is +INF
|
---|
38 | * 13. -0 ** (-anything except 0, NAN, odd integer) is +INF
|
---|
39 | * 14. -0 ** (odd integer) = -( +0 ** (odd integer) )
|
---|
40 | * 15. +INF ** (+anything except 0,NAN) is +INF
|
---|
41 | * 16. +INF ** (-anything except 0,NAN) is +0
|
---|
42 | * 17. -INF ** (anything) = -0 ** (-anything)
|
---|
43 | * 18. (-anything) ** (integer) is (-1)**(integer)*(+anything**integer)
|
---|
44 | * 19. (-anything except 0 and inf) ** (non-integer) is NAN
|
---|
45 | *
|
---|
46 | * Accuracy:
|
---|
47 | * pow(x,y) returns x**y nearly rounded. In particular
|
---|
48 | * pow(integer,integer)
|
---|
49 | * always returns the correct integer provided it is
|
---|
50 | * representable.
|
---|
51 | *
|
---|
52 | * Constants :
|
---|
53 | * The hexadecimal values are the intended ones for the following
|
---|
54 | * constants. The decimal values may be used, provided that the
|
---|
55 | * compiler will convert from decimal to binary accurately enough
|
---|
56 | * to produce the hexadecimal values shown.
|
---|
57 | */
|
---|
58 |
|
---|
59 | #include "fdlibm.h"
|
---|
60 |
|
---|
61 | #ifndef _DOUBLE_IS_32BITS
|
---|
62 |
|
---|
63 | #ifdef __STDC__
|
---|
64 | static const double
|
---|
65 | #else
|
---|
66 | static double
|
---|
67 | #endif
|
---|
68 | bp[] = {1.0, 1.5,},
|
---|
69 | dp_h[] = { 0.0, 5.84962487220764160156e-01,}, /* 0x3FE2B803, 0x40000000 */
|
---|
70 | dp_l[] = { 0.0, 1.35003920212974897128e-08,}, /* 0x3E4CFDEB, 0x43CFD006 */
|
---|
71 | zero = 0.0,
|
---|
72 | one = 1.0,
|
---|
73 | two = 2.0,
|
---|
74 | two53 = 9007199254740992.0, /* 0x43400000, 0x00000000 */
|
---|
75 | huge = 1.0e300,
|
---|
76 | tiny = 1.0e-300,
|
---|
77 | /* poly coefs for (3/2)*(log(x)-2s-2/3*s**3 */
|
---|
78 | L1 = 5.99999999999994648725e-01, /* 0x3FE33333, 0x33333303 */
|
---|
79 | L2 = 4.28571428578550184252e-01, /* 0x3FDB6DB6, 0xDB6FABFF */
|
---|
80 | L3 = 3.33333329818377432918e-01, /* 0x3FD55555, 0x518F264D */
|
---|
81 | L4 = 2.72728123808534006489e-01, /* 0x3FD17460, 0xA91D4101 */
|
---|
82 | L5 = 2.30660745775561754067e-01, /* 0x3FCD864A, 0x93C9DB65 */
|
---|
83 | L6 = 2.06975017800338417784e-01, /* 0x3FCA7E28, 0x4A454EEF */
|
---|
84 | P1 = 1.66666666666666019037e-01, /* 0x3FC55555, 0x5555553E */
|
---|
85 | P2 = -2.77777777770155933842e-03, /* 0xBF66C16C, 0x16BEBD93 */
|
---|
86 | P3 = 6.61375632143793436117e-05, /* 0x3F11566A, 0xAF25DE2C */
|
---|
87 | P4 = -1.65339022054652515390e-06, /* 0xBEBBBD41, 0xC5D26BF1 */
|
---|
88 | P5 = 4.13813679705723846039e-08, /* 0x3E663769, 0x72BEA4D0 */
|
---|
89 | lg2 = 6.93147180559945286227e-01, /* 0x3FE62E42, 0xFEFA39EF */
|
---|
90 | lg2_h = 6.93147182464599609375e-01, /* 0x3FE62E43, 0x00000000 */
|
---|
91 | lg2_l = -1.90465429995776804525e-09, /* 0xBE205C61, 0x0CA86C39 */
|
---|
92 | ovt = 8.0085662595372944372e-0017, /* -(1024-log2(ovfl+.5ulp)) */
|
---|
93 | cp = 9.61796693925975554329e-01, /* 0x3FEEC709, 0xDC3A03FD =2/(3ln2) */
|
---|
94 | cp_h = 9.61796700954437255859e-01, /* 0x3FEEC709, 0xE0000000 =(float)cp */
|
---|
95 | cp_l = -7.02846165095275826516e-09, /* 0xBE3E2FE0, 0x145B01F5 =tail of cp_h*/
|
---|
96 | ivln2 = 1.44269504088896338700e+00, /* 0x3FF71547, 0x652B82FE =1/ln2 */
|
---|
97 | ivln2_h = 1.44269502162933349609e+00, /* 0x3FF71547, 0x60000000 =24b 1/ln2*/
|
---|
98 | ivln2_l = 1.92596299112661746887e-08; /* 0x3E54AE0B, 0xF85DDF44 =1/ln2 tail*/
|
---|
99 |
|
---|
100 | #ifdef __STDC__
|
---|
101 | double __ieee754_pow(double x, double y)
|
---|
102 | #else
|
---|
103 | double __ieee754_pow(x,y)
|
---|
104 | double x, y;
|
---|
105 | #endif
|
---|
106 | {
|
---|
107 | double z,ax,z_h,z_l,p_h,p_l;
|
---|
108 | double y1,t1,t2,r,s,t,u,v,w;
|
---|
109 | int32_t i,j,k,yisint,n;
|
---|
110 | int32_t hx,hy,ix,iy;
|
---|
111 | uint32_t lx,ly;
|
---|
112 |
|
---|
113 | EXTRACT_WORDS(hx,lx,x);
|
---|
114 | EXTRACT_WORDS(hy,ly,y);
|
---|
115 | ix = hx&0x7fffffff; iy = hy&0x7fffffff;
|
---|
116 |
|
---|
117 | /* y==zero: x**0 = 1 */
|
---|
118 | if((iy|ly)==0) return one;
|
---|
119 |
|
---|
120 | /* +-NaN return x+y */
|
---|
121 | if(ix > 0x7ff00000 || ((ix==0x7ff00000)&&(lx!=0)) ||
|
---|
122 | iy > 0x7ff00000 || ((iy==0x7ff00000)&&(ly!=0)))
|
---|
123 | return x+y;
|
---|
124 |
|
---|
125 | /* determine if y is an odd int when x < 0
|
---|
126 | * yisint = 0 ... y is not an integer
|
---|
127 | * yisint = 1 ... y is an odd int
|
---|
128 | * yisint = 2 ... y is an even int
|
---|
129 | */
|
---|
130 | yisint = 0;
|
---|
131 | if(hx<0) {
|
---|
132 | if(iy>=0x43400000) yisint = 2; /* even integer y */
|
---|
133 | else if(iy>=0x3ff00000) {
|
---|
134 | k = (iy>>20)-0x3ff; /* exponent */
|
---|
135 | if(k>20) {
|
---|
136 | j = ly>>(52-k);
|
---|
137 | if((uint32_t)(j<<(52-k))==ly) yisint = 2-(j&1);
|
---|
138 | } else if(ly==0) {
|
---|
139 | j = iy>>(20-k);
|
---|
140 | if((j<<(20-k))==iy) yisint = 2-(j&1);
|
---|
141 | }
|
---|
142 | }
|
---|
143 | }
|
---|
144 |
|
---|
145 | /* special value of y */
|
---|
146 | if(ly==0) {
|
---|
147 | if (iy==0x7ff00000) { /* y is +-inf */
|
---|
148 | if(((ix-0x3ff00000)|lx)==0)
|
---|
149 | return y - y; /* inf**+-1 is NaN */
|
---|
150 | else if (ix >= 0x3ff00000)/* (|x|>1)**+-inf = inf,0 */
|
---|
151 | return (hy>=0)? y: zero;
|
---|
152 | else /* (|x|<1)**-,+inf = inf,0 */
|
---|
153 | return (hy<0)?-y: zero;
|
---|
154 | }
|
---|
155 | if(iy==0x3ff00000) { /* y is +-1 */
|
---|
156 | if(hy<0) return one/x; else return x;
|
---|
157 | }
|
---|
158 | if(hy==0x40000000) return x*x; /* y is 2 */
|
---|
159 | if(hy==0x3fe00000) { /* y is 0.5 */
|
---|
160 | if(hx>=0) /* x >= +0 */
|
---|
161 | return __ieee754_sqrt(x);
|
---|
162 | }
|
---|
163 | }
|
---|
164 |
|
---|
165 | ax = fabs(x);
|
---|
166 | /* special value of x */
|
---|
167 | if(lx==0) {
|
---|
168 | if(ix==0x7ff00000||ix==0||ix==0x3ff00000){
|
---|
169 | z = ax; /*x is +-0,+-inf,+-1*/
|
---|
170 | if(hy<0) z = one/z; /* z = (1/|x|) */
|
---|
171 | if(hx<0) {
|
---|
172 | if(((ix-0x3ff00000)|yisint)==0) {
|
---|
173 | z = (z-z)/(z-z); /* (-1)**non-int is NaN */
|
---|
174 | } else if(yisint==1)
|
---|
175 | z = -z; /* (x<0)**odd = -(|x|**odd) */
|
---|
176 | }
|
---|
177 | return z;
|
---|
178 | }
|
---|
179 | }
|
---|
180 |
|
---|
181 | /* (x<0)**(non-int) is NaN */
|
---|
182 | /* CYGNUS LOCAL: This used to be
|
---|
183 | if((((hx>>31)+1)|yisint)==0) return (x-x)/(x-x);
|
---|
184 | but ANSI C says a right shift of a signed negative quantity is
|
---|
185 | implementation defined. */
|
---|
186 | if(((((uint32_t)hx>>31)-1)|yisint)==0) return (x-x)/(x-x);
|
---|
187 |
|
---|
188 | /* |y| is huge */
|
---|
189 | if(iy>0x41e00000) { /* if |y| > 2**31 */
|
---|
190 | if(iy>0x43f00000){ /* if |y| > 2**64, must o/uflow */
|
---|
191 | if(ix<=0x3fefffff) return (hy<0)? huge*huge:tiny*tiny;
|
---|
192 | if(ix>=0x3ff00000) return (hy>0)? huge*huge:tiny*tiny;
|
---|
193 | }
|
---|
194 | /* over/underflow if x is not close to one */
|
---|
195 | if(ix<0x3fefffff) return (hy<0)? huge*huge:tiny*tiny;
|
---|
196 | if(ix>0x3ff00000) return (hy>0)? huge*huge:tiny*tiny;
|
---|
197 | /* now |1-x| is tiny <= 2**-20, suffice to compute
|
---|
198 | log(x) by x-x^2/2+x^3/3-x^4/4 */
|
---|
199 | t = x-1; /* t has 20 trailing zeros */
|
---|
200 | w = (t*t)*(0.5-t*(0.3333333333333333333333-t*0.25));
|
---|
201 | u = ivln2_h*t; /* ivln2_h has 21 sig. bits */
|
---|
202 | v = t*ivln2_l-w*ivln2;
|
---|
203 | t1 = u+v;
|
---|
204 | SET_LOW_WORD(t1,0);
|
---|
205 | t2 = v-(t1-u);
|
---|
206 | } else {
|
---|
207 | double s2,s_h,s_l,t_h,t_l;
|
---|
208 | n = 0;
|
---|
209 | /* take care subnormal number */
|
---|
210 | if(ix<0x00100000)
|
---|
211 | {ax *= two53; n -= 53; GET_HIGH_WORD(ix,ax); }
|
---|
212 | n += ((ix)>>20)-0x3ff;
|
---|
213 | j = ix&0x000fffff;
|
---|
214 | /* determine interval */
|
---|
215 | ix = j|0x3ff00000; /* normalize ix */
|
---|
216 | if(j<=0x3988E) k=0; /* |x|<sqrt(3/2) */
|
---|
217 | else if(j<0xBB67A) k=1; /* |x|<sqrt(3) */
|
---|
218 | else {k=0;n+=1;ix -= 0x00100000;}
|
---|
219 | SET_HIGH_WORD(ax,ix);
|
---|
220 |
|
---|
221 | /* compute s = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */
|
---|
222 | u = ax-bp[k]; /* bp[0]=1.0, bp[1]=1.5 */
|
---|
223 | v = one/(ax+bp[k]);
|
---|
224 | s = u*v;
|
---|
225 | s_h = s;
|
---|
226 | SET_LOW_WORD(s_h,0);
|
---|
227 | /* t_h=ax+bp[k] High */
|
---|
228 | t_h = zero;
|
---|
229 | SET_HIGH_WORD(t_h,((ix>>1)|0x20000000)+0x00080000+(k<<18));
|
---|
230 | t_l = ax - (t_h-bp[k]);
|
---|
231 | s_l = v*((u-s_h*t_h)-s_h*t_l);
|
---|
232 | /* compute log(ax) */
|
---|
233 | s2 = s*s;
|
---|
234 | r = s2*s2*(L1+s2*(L2+s2*(L3+s2*(L4+s2*(L5+s2*L6)))));
|
---|
235 | r += s_l*(s_h+s);
|
---|
236 | s2 = s_h*s_h;
|
---|
237 | t_h = 3.0+s2+r;
|
---|
238 | SET_LOW_WORD(t_h,0);
|
---|
239 | t_l = r-((t_h-3.0)-s2);
|
---|
240 | /* u+v = s*(1+...) */
|
---|
241 | u = s_h*t_h;
|
---|
242 | v = s_l*t_h+t_l*s;
|
---|
243 | /* 2/(3log2)*(s+...) */
|
---|
244 | p_h = u+v;
|
---|
245 | SET_LOW_WORD(p_h,0);
|
---|
246 | p_l = v-(p_h-u);
|
---|
247 | z_h = cp_h*p_h; /* cp_h+cp_l = 2/(3*log2) */
|
---|
248 | z_l = cp_l*p_h+p_l*cp+dp_l[k];
|
---|
249 | /* log2(ax) = (s+..)*2/(3*log2) = n + dp_h + z_h + z_l */
|
---|
250 | t = (double)n;
|
---|
251 | t1 = (((z_h+z_l)+dp_h[k])+t);
|
---|
252 | SET_LOW_WORD(t1,0);
|
---|
253 | t2 = z_l-(((t1-t)-dp_h[k])-z_h);
|
---|
254 | }
|
---|
255 |
|
---|
256 | s = one; /* s (sign of result -ve**odd) = -1 else = 1 */
|
---|
257 | if(((((uint32_t)hx>>31)-1)|(yisint-1))==0)
|
---|
258 | s = -one;/* (-ve)**(odd int) */
|
---|
259 |
|
---|
260 | /* split up y into y1+y2 and compute (y1+y2)*(t1+t2) */
|
---|
261 | y1 = y;
|
---|
262 | SET_LOW_WORD(y1,0);
|
---|
263 | p_l = (y-y1)*t1+y*t2;
|
---|
264 | p_h = y1*t1;
|
---|
265 | z = p_l+p_h;
|
---|
266 | EXTRACT_WORDS(j,i,z);
|
---|
267 | if (j>=0x40900000) { /* z >= 1024 */
|
---|
268 | if(((j-0x40900000)|i)!=0) /* if z > 1024 */
|
---|
269 | return s*huge*huge; /* overflow */
|
---|
270 | else {
|
---|
271 | if(p_l+ovt>z-p_h) return s*huge*huge; /* overflow */
|
---|
272 | }
|
---|
273 | } else if((j&0x7fffffff)>=0x4090cc00 ) { /* z <= -1075 */
|
---|
274 | if(((j-0xc090cc00)|i)!=0) /* z < -1075 */
|
---|
275 | return s*tiny*tiny; /* underflow */
|
---|
276 | else {
|
---|
277 | if(p_l<=z-p_h) return s*tiny*tiny; /* underflow */
|
---|
278 | }
|
---|
279 | }
|
---|
280 | /*
|
---|
281 | * compute 2**(p_h+p_l)
|
---|
282 | */
|
---|
283 | i = j&0x7fffffff;
|
---|
284 | k = (i>>20)-0x3ff;
|
---|
285 | n = 0;
|
---|
286 | if(i>0x3fe00000) { /* if |z| > 0.5, set n = [z+0.5] */
|
---|
287 | n = j+(0x00100000>>(k+1));
|
---|
288 | k = ((n&0x7fffffff)>>20)-0x3ff; /* new k for n */
|
---|
289 | t = zero;
|
---|
290 | SET_HIGH_WORD(t,n&~(0x000fffff>>k));
|
---|
291 | n = ((n&0x000fffff)|0x00100000)>>(20-k);
|
---|
292 | if(j<0) n = -n;
|
---|
293 | p_h -= t;
|
---|
294 | }
|
---|
295 | t = p_l+p_h;
|
---|
296 | SET_LOW_WORD(t,0);
|
---|
297 | u = t*lg2_h;
|
---|
298 | v = (p_l-(t-p_h))*lg2+t*lg2_l;
|
---|
299 | z = u+v;
|
---|
300 | w = v-(z-u);
|
---|
301 | t = z*z;
|
---|
302 | t1 = z - t*(P1+t*(P2+t*(P3+t*(P4+t*P5))));
|
---|
303 | r = (z*t1)/(t1-two)-(w+z*w);
|
---|
304 | z = one-(r-z);
|
---|
305 | GET_HIGH_WORD(j,z);
|
---|
306 | j += (n<<20);
|
---|
307 | if((j>>20)<=0) z = scalbn(z,(int)n); /* subnormal output */
|
---|
308 | else SET_HIGH_WORD(z,j);
|
---|
309 | return s*z;
|
---|
310 | }
|
---|
311 |
|
---|
312 | #endif /* defined(_DOUBLE_IS_32BITS) */
|
---|