1 /* @(#)e_pow.c 5.1 93/09/24 */
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
10 * ====================================================
13 #include <sys/cdefs.h>
14 #if defined(LIBM_SCCS) && !defined(lint)
15 __RCSID("$NetBSD: e_pow.c,v 1.16 2010/04/23 19:17:07 drochner Exp $");
18 /* __ieee754_pow(x,y) return x**y
21 * Method: Let x = 2 * (1+f)
22 * 1. Compute and return log2(x) in two pieces:
24 * where w1 has 53-24 = 29 bit trailing zeros.
25 * 2. Perform y*log2(x) = n+y' by simulating multi-precision
26 * arithmetic, where |y'|<=0.5.
27 * 3. Return x**y = 2**n*exp(y'*log2)
30 * 1. (anything) ** 0 is 1
31 * 2. (anything) ** 1 is itself
32 * 3. (anything) ** NAN is NAN
33 * 4. NAN ** (anything except 0) is NAN
34 * 5. +-(|x| > 1) ** +INF is +INF
35 * 6. +-(|x| > 1) ** -INF is +0
36 * 7. +-(|x| < 1) ** +INF is +0
37 * 8. +-(|x| < 1) ** -INF is +INF
38 * 9. +-1 ** +-INF is NAN
39 * 10. +0 ** (+anything except 0, NAN) is +0
40 * 11. -0 ** (+anything except 0, NAN, odd integer) is +0
41 * 12. +0 ** (-anything except 0, NAN) is +INF
42 * 13. -0 ** (-anything except 0, NAN, odd integer) is +INF
43 * 14. -0 ** (odd integer) = -( +0 ** (odd integer) )
44 * 15. +INF ** (+anything except 0,NAN) is +INF
45 * 16. +INF ** (-anything except 0,NAN) is +0
46 * 17. -INF ** (anything) = -0 ** (-anything)
47 * 18. (-anything) ** (integer) is (-1)**(integer)*(+anything**integer)
48 * 19. (-anything except 0 and inf) ** (non-integer) is NAN
51 * pow(x,y) returns x**y nearly rounded. In particular
52 * pow(integer,integer)
53 * always returns the correct integer provided it is
57 * The hexadecimal values are the intended ones for the following
58 * constants. The decimal values may be used, provided that the
59 * compiler will convert from decimal to binary accurately enough
60 * to produce the hexadecimal values shown.
63 #include "namespace.h"
65 #include "math_private.h"
69 dp_h
[] = { 0.0, 5.84962487220764160156e-01,}, /* 0x3FE2B803, 0x40000000 */
70 dp_l
[] = { 0.0, 1.35003920212974897128e-08,}, /* 0x3E4CFDEB, 0x43CFD006 */
74 two53
= 9007199254740992.0, /* 0x43400000, 0x00000000 */
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*/
101 __ieee754_pow(double x
, double y
)
103 double z
,ax
,z_h
,z_l
,p_h
,p_l
;
104 double yy1
,t1
,t2
,r
,s
,t
,u
,v
,w
;
105 int32_t i
,j
,k
,yisint
,n
;
109 EXTRACT_WORDS(hx
,lx
,x
);
110 EXTRACT_WORDS(hy
,ly
,y
);
111 ix
= hx
&0x7fffffff; iy
= hy
&0x7fffffff;
113 /* y==zero: x**0 = 1 */
114 if((iy
|ly
)==0) return one
;
116 /* +-NaN return x+y */
117 if(ix
> 0x7ff00000 || ((ix
==0x7ff00000)&&(lx
!=0)) ||
118 iy
> 0x7ff00000 || ((iy
==0x7ff00000)&&(ly
!=0)))
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
128 if(iy
>=0x43400000) yisint
= 2; /* even integer y */
129 else if(iy
>=0x3ff00000) {
130 k
= (iy
>>20)-0x3ff; /* exponent */
133 if((uint32_t)(j
<<(52-k
))==ly
) yisint
= 2-(j
&1);
136 if((j
<<(20-k
))==iy
) yisint
= 2-(j
&1);
141 /* special value of y */
143 if (iy
==0x7ff00000) { /* y is +-inf */
144 if(((ix
-0x3ff00000)|lx
)==0)
145 return y
- y
; /* inf**+-1 is NaN */
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
;
151 if(iy
==0x3ff00000) { /* y is +-1 */
152 if(hy
<0) return one
/x
; else return x
;
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 __ieee754_sqrt(x
);
162 /* special value of x */
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|) */
168 if(((ix
-0x3ff00000)|yisint
)==0) {
169 z
= (z
-z
)/(z
-z
); /* (-1)**non-int is NaN */
171 z
= -z
; /* (x<0)**odd = -(|x|**odd) */
179 /* (x<0)**(non-int) is NaN */
180 if((n
|yisint
)==0) return (x
-x
)/(x
-x
);
182 s
= one
; /* s (sign of result -ve**odd) = -1 else = 1 */
183 if((n
|(yisint
-1))==0) s
= -one
;/* (-ve)**(odd int) */
186 if(iy
>0x41e00000) { /* if |y| > 2**31 */
187 if(iy
>0x43f00000){ /* if |y| > 2**64, must o/uflow */
188 if(ix
<=0x3fefffff) return (hy
<0)? huge
*huge
:tiny
*tiny
;
189 if(ix
>=0x3ff00000) return (hy
>0)? huge
*huge
:tiny
*tiny
;
191 /* over/underflow if x is not close to one */
192 if(ix
<0x3fefffff) return (hy
<0)? s
*huge
*huge
:s
*tiny
*tiny
;
193 if(ix
>0x3ff00000) return (hy
>0)? s
*huge
*huge
:s
*tiny
*tiny
;
194 /* now |1-x| is tiny <= 2**-20, suffice to compute
195 log(x) by x-x^2/2+x^3/3-x^4/4 */
196 t
= ax
-one
; /* t has 20 trailing zeros */
197 w
= (t
*t
)*(0.5-t
*(0.3333333333333333333333-t
*0.25));
198 u
= ivln2_h
*t
; /* ivln2_h has 21 sig. bits */
199 v
= t
*ivln2_l
-w
*ivln2
;
204 double ss
,s2
,s_h
,s_l
,t_h
,t_l
;
206 /* take care subnormal number */
208 {ax
*= two53
; n
-= 53; GET_HIGH_WORD(ix
,ax
); }
209 n
+= ((ix
)>>20)-0x3ff;
211 /* determine interval */
212 ix
= j
|0x3ff00000; /* normalize ix */
213 if(j
<=0x3988E) k
=0; /* |x|<sqrt(3/2) */
214 else if(j
<0xBB67A) k
=1; /* |x|<sqrt(3) */
215 else {k
=0;n
+=1;ix
-= 0x00100000;}
216 SET_HIGH_WORD(ax
,ix
);
218 /* compute ss = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */
219 u
= ax
-bp
[k
]; /* bp[0]=1.0, bp[1]=1.5 */
224 /* t_h=ax+bp[k] High */
226 SET_HIGH_WORD(t_h
,((ix
>>1)|0x20000000)+0x00080000+(k
<<18));
227 t_l
= ax
- (t_h
-bp
[k
]);
228 s_l
= v
*((u
-s_h
*t_h
)-s_h
*t_l
);
229 /* compute log(ax) */
231 r
= s2
*s2
*(L1
+s2
*(L2
+s2
*(L3
+s2
*(L4
+s2
*(L5
+s2
*L6
)))));
236 t_l
= r
-((t_h
-3.0)-s2
);
237 /* u+v = ss*(1+...) */
240 /* 2/(3log2)*(ss+...) */
244 z_h
= cp_h
*p_h
; /* cp_h+cp_l = 2/(3*log2) */
245 z_l
= cp_l
*p_h
+p_l
*cp
+dp_l
[k
];
246 /* log2(ax) = (ss+..)*2/(3*log2) = n + dp_h + z_h + z_l */
248 t1
= (((z_h
+z_l
)+dp_h
[k
])+t
);
250 t2
= z_l
-(((t1
-t
)-dp_h
[k
])-z_h
);
253 /* split up y into yy1+y2 and compute (yy1+y2)*(t1+t2) */
256 p_l
= (y
-yy1
)*t1
+y
*t2
;
259 EXTRACT_WORDS(j
,i
,z
);
260 if (j
>=0x40900000) { /* z >= 1024 */
261 if(((j
-0x40900000)|i
)!=0) /* if z > 1024 */
262 return s
*huge
*huge
; /* overflow */
264 if(p_l
+ovt
>z
-p_h
) return s
*huge
*huge
; /* overflow */
266 } else if((j
&0x7fffffff)>=0x4090cc00 ) { /* z <= -1075 */
267 if(((j
-0xc090cc00)|i
)!=0) /* z < -1075 */
268 return s
*tiny
*tiny
; /* underflow */
270 if(p_l
<=z
-p_h
) return s
*tiny
*tiny
; /* underflow */
274 * compute 2**(p_h+p_l)
279 if(i
>0x3fe00000) { /* if |z| > 0.5, set n = [z+0.5] */
280 n
= j
+(0x00100000>>(k
+1));
281 k
= ((n
&0x7fffffff)>>20)-0x3ff; /* new k for n */
283 SET_HIGH_WORD(t
,n
&~(0x000fffff>>k
));
284 n
= ((n
&0x000fffff)|0x00100000)>>(20-k
);
291 v
= (p_l
-(t
-p_h
))*lg2
+t
*lg2_l
;
295 t1
= z
- t
*(P1
+t
*(P2
+t
*(P3
+t
*(P4
+t
*P5
))));
296 r
= (z
*t1
)/(t1
-two
)-(w
+z
*w
);
300 if((j
>>20)<=0) z
= scalbn(z
,n
); /* subnormal output */
301 else SET_HIGH_WORD(z
,j
);