1 | /* e_powf.c -- float version of e_pow.c.
|
---|
2 | * Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
|
---|
3 | */
|
---|
4 |
|
---|
5 | /*
|
---|
6 | * ====================================================
|
---|
7 | * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
|
---|
8 | *
|
---|
9 | * Developed at SunPro, a Sun Microsystems, Inc. business.
|
---|
10 | * Permission to use, copy, modify, and distribute this
|
---|
11 | * software is freely granted, provided that this notice
|
---|
12 | * is preserved.
|
---|
13 | * ====================================================
|
---|
14 | */
|
---|
15 |
|
---|
16 | #ifndef lint
|
---|
17 | static char rcsid[] = "$FreeBSD: src/lib/msun/src/e_powf.c,v 1.12 2004/06/01 19:33:30 bde Exp $";
|
---|
18 | #endif
|
---|
19 |
|
---|
20 | #include "math.h"
|
---|
21 | #include "math_private.h"
|
---|
22 |
|
---|
23 | static const float
|
---|
24 | bp[] = {1.0, 1.5,},
|
---|
25 | dp_h[] = { 0.0, 5.84960938e-01,}, /* 0x3f15c000 */
|
---|
26 | dp_l[] = { 0.0, 1.56322085e-06,}, /* 0x35d1cfdc */
|
---|
27 | zero = 0.0,
|
---|
28 | one = 1.0,
|
---|
29 | two = 2.0,
|
---|
30 | two24 = 16777216.0, /* 0x4b800000 */
|
---|
31 | huge = 1.0e30,
|
---|
32 | tiny = 1.0e-30,
|
---|
33 | /* poly coefs for (3/2)*(log(x)-2s-2/3*s**3 */
|
---|
34 | L1 = 6.0000002384e-01, /* 0x3f19999a */
|
---|
35 | L2 = 4.2857143283e-01, /* 0x3edb6db7 */
|
---|
36 | L3 = 3.3333334327e-01, /* 0x3eaaaaab */
|
---|
37 | L4 = 2.7272811532e-01, /* 0x3e8ba305 */
|
---|
38 | L5 = 2.3066075146e-01, /* 0x3e6c3255 */
|
---|
39 | L6 = 2.0697501302e-01, /* 0x3e53f142 */
|
---|
40 | P1 = 1.6666667163e-01, /* 0x3e2aaaab */
|
---|
41 | P2 = -2.7777778450e-03, /* 0xbb360b61 */
|
---|
42 | P3 = 6.6137559770e-05, /* 0x388ab355 */
|
---|
43 | P4 = -1.6533901999e-06, /* 0xb5ddea0e */
|
---|
44 | P5 = 4.1381369442e-08, /* 0x3331bb4c */
|
---|
45 | lg2 = 6.9314718246e-01, /* 0x3f317218 */
|
---|
46 | lg2_h = 6.93145752e-01, /* 0x3f317200 */
|
---|
47 | lg2_l = 1.42860654e-06, /* 0x35bfbe8c */
|
---|
48 | ovt = 4.2995665694e-08, /* -(128-log2(ovfl+.5ulp)) */
|
---|
49 | cp = 9.6179670095e-01, /* 0x3f76384f =2/(3ln2) */
|
---|
50 | cp_h = 9.6179199219e-01, /* 0x3f763800 =head of cp */
|
---|
51 | cp_l = 4.7017383622e-06, /* 0x369dc3a0 =tail of cp_h */
|
---|
52 | ivln2 = 1.4426950216e+00, /* 0x3fb8aa3b =1/ln2 */
|
---|
53 | ivln2_h = 1.4426879883e+00, /* 0x3fb8aa00 =16b 1/ln2*/
|
---|
54 | ivln2_l = 7.0526075433e-06; /* 0x36eca570 =1/ln2 tail*/
|
---|
55 |
|
---|
56 | float
|
---|
57 | __ieee754_powf(float x, float y)
|
---|
58 | {
|
---|
59 | float z,ax,z_h,z_l,p_h,p_l;
|
---|
60 | float y1,t1,t2,r,s,sn,t,u,v,w;
|
---|
61 | int32_t i,j,k,yisint,n;
|
---|
62 | int32_t hx,hy,ix,iy,is;
|
---|
63 |
|
---|
64 | GET_FLOAT_WORD(hx,x);
|
---|
65 | GET_FLOAT_WORD(hy,y);
|
---|
66 | ix = hx&0x7fffffff; iy = hy&0x7fffffff;
|
---|
67 |
|
---|
68 | /* y==zero: x**0 = 1 */
|
---|
69 | if(iy==0) return one;
|
---|
70 |
|
---|
71 | /* +-NaN return x+y */
|
---|
72 | if(ix > 0x7f800000 ||
|
---|
73 | iy > 0x7f800000)
|
---|
74 | return x+y;
|
---|
75 |
|
---|
76 | /* determine if y is an odd int when x < 0
|
---|
77 | * yisint = 0 ... y is not an integer
|
---|
78 | * yisint = 1 ... y is an odd int
|
---|
79 | * yisint = 2 ... y is an even int
|
---|
80 | */
|
---|
81 | yisint = 0;
|
---|
82 | if(hx<0) {
|
---|
83 | if(iy>=0x4b800000) yisint = 2; /* even integer y */
|
---|
84 | else if(iy>=0x3f800000) {
|
---|
85 | k = (iy>>23)-0x7f; /* exponent */
|
---|
86 | j = iy>>(23-k);
|
---|
87 | if((j<<(23-k))==iy) yisint = 2-(j&1);
|
---|
88 | }
|
---|
89 | }
|
---|
90 |
|
---|
91 | /* special value of y */
|
---|
92 | if (iy==0x7f800000) { /* y is +-inf */
|
---|
93 | if (ix==0x3f800000)
|
---|
94 | return y - y; /* inf**+-1 is NaN */
|
---|
95 | else if (ix > 0x3f800000)/* (|x|>1)**+-inf = inf,0 */
|
---|
96 | return (hy>=0)? y: zero;
|
---|
97 | else /* (|x|<1)**-,+inf = inf,0 */
|
---|
98 | return (hy<0)?-y: zero;
|
---|
99 | }
|
---|
100 | if(iy==0x3f800000) { /* y is +-1 */
|
---|
101 | if(hy<0) return one/x; else return x;
|
---|
102 | }
|
---|
103 | if(hy==0x40000000) return x*x; /* y is 2 */
|
---|
104 | if(hy==0x3f000000) { /* y is 0.5 */
|
---|
105 | if(hx>=0) /* x >= +0 */
|
---|
106 | return __ieee754_sqrtf(x);
|
---|
107 | }
|
---|
108 |
|
---|
109 | ax = fabsf(x);
|
---|
110 | /* special value of x */
|
---|
111 | if(ix==0x7f800000||ix==0||ix==0x3f800000){
|
---|
112 | z = ax; /*x is +-0,+-inf,+-1*/
|
---|
113 | if(hy<0) z = one/z; /* z = (1/|x|) */
|
---|
114 | if(hx<0) {
|
---|
115 | if(((ix-0x3f800000)|yisint)==0) {
|
---|
116 | z = (z-z)/(z-z); /* (-1)**non-int is NaN */
|
---|
117 | } else if(yisint==1)
|
---|
118 | z = -z; /* (x<0)**odd = -(|x|**odd) */
|
---|
119 | }
|
---|
120 | return z;
|
---|
121 | }
|
---|
122 |
|
---|
123 | n = ((u_int32_t)hx>>31)-1;
|
---|
124 |
|
---|
125 | /* (x<0)**(non-int) is NaN */
|
---|
126 | if((n|yisint)==0) return (x-x)/(x-x);
|
---|
127 |
|
---|
128 | sn = one; /* s (sign of result -ve**odd) = -1 else = 1 */
|
---|
129 | if((n|(yisint-1))==0) sn = -one;/* (-ve)**(odd int) */
|
---|
130 |
|
---|
131 | /* |y| is huge */
|
---|
132 | if(iy>0x4d000000) { /* if |y| > 2**27 */
|
---|
133 | /* over/underflow if x is not close to one */
|
---|
134 | if(ix<0x3f7ffff8) return (hy<0)? sn*huge*huge:sn*tiny*tiny;
|
---|
135 | if(ix>0x3f800007) return (hy>0)? sn*huge*huge:sn*tiny*tiny;
|
---|
136 | /* now |1-x| is tiny <= 2**-20, suffice to compute
|
---|
137 | log(x) by x-x^2/2+x^3/3-x^4/4 */
|
---|
138 | t = ax-1; /* t has 20 trailing zeros */
|
---|
139 | w = (t*t)*((float)0.5-t*((float)0.333333333333-t*(float)0.25));
|
---|
140 | u = ivln2_h*t; /* ivln2_h has 16 sig. bits */
|
---|
141 | v = t*ivln2_l-w*ivln2;
|
---|
142 | t1 = u+v;
|
---|
143 | GET_FLOAT_WORD(is,t1);
|
---|
144 | SET_FLOAT_WORD(t1,is&0xfffff000);
|
---|
145 | t2 = v-(t1-u);
|
---|
146 | } else {
|
---|
147 | float s2,s_h,s_l,t_h,t_l;
|
---|
148 | n = 0;
|
---|
149 | /* take care subnormal number */
|
---|
150 | if(ix<0x00800000)
|
---|
151 | {ax *= two24; n -= 24; GET_FLOAT_WORD(ix,ax); }
|
---|
152 | n += ((ix)>>23)-0x7f;
|
---|
153 | j = ix&0x007fffff;
|
---|
154 | /* determine interval */
|
---|
155 | ix = j|0x3f800000; /* normalize ix */
|
---|
156 | if(j<=0x1cc471) k=0; /* |x|<sqrt(3/2) */
|
---|
157 | else if(j<0x5db3d7) k=1; /* |x|<sqrt(3) */
|
---|
158 | else {k=0;n+=1;ix -= 0x00800000;}
|
---|
159 | SET_FLOAT_WORD(ax,ix);
|
---|
160 |
|
---|
161 | /* compute s = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */
|
---|
162 | u = ax-bp[k]; /* bp[0]=1.0, bp[1]=1.5 */
|
---|
163 | v = one/(ax+bp[k]);
|
---|
164 | s = u*v;
|
---|
165 | s_h = s;
|
---|
166 | GET_FLOAT_WORD(is,s_h);
|
---|
167 | SET_FLOAT_WORD(s_h,is&0xfffff000);
|
---|
168 | /* t_h=ax+bp[k] High */
|
---|
169 | is = ((ix>>1)&0xfffff000)|0x20000000;
|
---|
170 | SET_FLOAT_WORD(t_h,is+0x00400000+(k<<21));
|
---|
171 | t_l = ax - (t_h-bp[k]);
|
---|
172 | s_l = v*((u-s_h*t_h)-s_h*t_l);
|
---|
173 | /* compute log(ax) */
|
---|
174 | s2 = s*s;
|
---|
175 | r = s2*s2*(L1+s2*(L2+s2*(L3+s2*(L4+s2*(L5+s2*L6)))));
|
---|
176 | r += s_l*(s_h+s);
|
---|
177 | s2 = s_h*s_h;
|
---|
178 | t_h = (float)3.0+s2+r;
|
---|
179 | GET_FLOAT_WORD(is,t_h);
|
---|
180 | SET_FLOAT_WORD(t_h,is&0xfffff000);
|
---|
181 | t_l = r-((t_h-(float)3.0)-s2);
|
---|
182 | /* u+v = s*(1+...) */
|
---|
183 | u = s_h*t_h;
|
---|
184 | v = s_l*t_h+t_l*s;
|
---|
185 | /* 2/(3log2)*(s+...) */
|
---|
186 | p_h = u+v;
|
---|
187 | GET_FLOAT_WORD(is,p_h);
|
---|
188 | SET_FLOAT_WORD(p_h,is&0xfffff000);
|
---|
189 | p_l = v-(p_h-u);
|
---|
190 | z_h = cp_h*p_h; /* cp_h+cp_l = 2/(3*log2) */
|
---|
191 | z_l = cp_l*p_h+p_l*cp+dp_l[k];
|
---|
192 | /* log2(ax) = (s+..)*2/(3*log2) = n + dp_h + z_h + z_l */
|
---|
193 | t = (float)n;
|
---|
194 | t1 = (((z_h+z_l)+dp_h[k])+t);
|
---|
195 | GET_FLOAT_WORD(is,t1);
|
---|
196 | SET_FLOAT_WORD(t1,is&0xfffff000);
|
---|
197 | t2 = z_l-(((t1-t)-dp_h[k])-z_h);
|
---|
198 | }
|
---|
199 |
|
---|
200 | /* split up y into y1+y2 and compute (y1+y2)*(t1+t2) */
|
---|
201 | GET_FLOAT_WORD(is,y);
|
---|
202 | SET_FLOAT_WORD(y1,is&0xfffff000);
|
---|
203 | p_l = (y-y1)*t1+y*t2;
|
---|
204 | p_h = y1*t1;
|
---|
205 | z = p_l+p_h;
|
---|
206 | GET_FLOAT_WORD(j,z);
|
---|
207 | if (j>0x43000000) /* if z > 128 */
|
---|
208 | return sn*huge*huge; /* overflow */
|
---|
209 | else if (j==0x43000000) { /* if z == 128 */
|
---|
210 | if(p_l+ovt>z-p_h) return sn*huge*huge; /* overflow */
|
---|
211 | }
|
---|
212 | else if ((j&0x7fffffff)>0x43160000) /* z <= -150 */
|
---|
213 | return sn*tiny*tiny; /* underflow */
|
---|
214 | else if (j==0xc3160000){ /* z == -150 */
|
---|
215 | if(p_l<=z-p_h) return sn*tiny*tiny; /* underflow */
|
---|
216 | }
|
---|
217 | /*
|
---|
218 | * compute 2**(p_h+p_l)
|
---|
219 | */
|
---|
220 | i = j&0x7fffffff;
|
---|
221 | k = (i>>23)-0x7f;
|
---|
222 | n = 0;
|
---|
223 | if(i>0x3f000000) { /* if |z| > 0.5, set n = [z+0.5] */
|
---|
224 | n = j+(0x00800000>>(k+1));
|
---|
225 | k = ((n&0x7fffffff)>>23)-0x7f; /* new k for n */
|
---|
226 | SET_FLOAT_WORD(t,n&~(0x007fffff>>k));
|
---|
227 | n = ((n&0x007fffff)|0x00800000)>>(23-k);
|
---|
228 | if(j<0) n = -n;
|
---|
229 | p_h -= t;
|
---|
230 | }
|
---|
231 | t = p_l+p_h;
|
---|
232 | GET_FLOAT_WORD(is,t);
|
---|
233 | SET_FLOAT_WORD(t,is&0xffff8000);
|
---|
234 | u = t*lg2_h;
|
---|
235 | v = (p_l-(t-p_h))*lg2+t*lg2_l;
|
---|
236 | z = u+v;
|
---|
237 | w = v-(z-u);
|
---|
238 | t = z*z;
|
---|
239 | t1 = z - t*(P1+t*(P2+t*(P3+t*(P4+t*P5))));
|
---|
240 | r = (z*t1)/(t1-two)-(w+z*w);
|
---|
241 | z = one-(r-z);
|
---|
242 | GET_FLOAT_WORD(j,z);
|
---|
243 | j += (n<<23);
|
---|
244 | if((j>>23)<=0) z = scalbnf(z,n); /* subnormal output */
|
---|
245 | else SET_FLOAT_WORD(z,j);
|
---|
246 | return sn*z;
|
---|
247 | }
|
---|