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 | /* Copyright (C) 2017 Free Software Foundation, Inc. |
5 | This file is part of the GNU C Library. |
6 | |
7 | The GNU C Library is free software; you can redistribute it and/or |
8 | modify it under the terms of the GNU Lesser General Public |
9 | License as published by the Free Software Foundation; either |
10 | version 2.1 of the License, or (at your option) any later version. |
11 | |
12 | The GNU C Library is distributed in the hope that it will be useful, |
13 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
15 | Lesser General Public License for more details. |
16 | |
17 | You should have received a copy of the GNU Lesser General Public |
18 | License along with the GNU C Library; if not, see |
19 | <http://www.gnu.org/licenses/>. */ |
20 | |
21 | /* |
22 | * ==================================================== |
23 | * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved. |
24 | * |
25 | * Developed at SunPro, a Sun Microsystems, Inc. business. |
26 | * Permission to use, copy, modify, and distribute this |
27 | * software is freely granted, provided that this notice |
28 | * is preserved. |
29 | * ==================================================== |
30 | */ |
31 | |
32 | #include <math.h> |
33 | #include <math_private.h> |
34 | |
35 | static const float huge = 1.0e+30, tiny = 1.0e-30; |
36 | |
37 | static const float |
38 | bp[] = {1.0, 1.5,}, |
39 | zero = 0.0, |
40 | one = 1.0, |
41 | two = 2.0, |
42 | two24 = 16777216.0, /* 0x4b800000 */ |
43 | /* poly coefs for (3/2)*(log(x)-2s-2/3*s**3 */ |
44 | L1 = 6.0000002384e-01, /* 0x3f19999a */ |
45 | L2 = 4.2857143283e-01, /* 0x3edb6db7 */ |
46 | L3 = 3.3333334327e-01, /* 0x3eaaaaab */ |
47 | L4 = 2.7272811532e-01, /* 0x3e8ba305 */ |
48 | L5 = 2.3066075146e-01, /* 0x3e6c3255 */ |
49 | L6 = 2.0697501302e-01, /* 0x3e53f142 */ |
50 | P1 = 1.6666667163e-01, /* 0x3e2aaaab */ |
51 | P2 = -2.7777778450e-03, /* 0xbb360b61 */ |
52 | P3 = 6.6137559770e-05, /* 0x388ab355 */ |
53 | P4 = -1.6533901999e-06, /* 0xb5ddea0e */ |
54 | P5 = 4.1381369442e-08, /* 0x3331bb4c */ |
55 | ovt = 4.2995665694e-08; /* -(128-log2(ovfl+.5ulp)) */ |
56 | |
57 | static const double |
58 | dp[] = { 0.0, 0x1.2b803473f7ad1p-1, }, /* log2(1.5) */ |
59 | lg2 = M_LN2, |
60 | cp = 2.0/3.0/M_LN2, |
61 | invln2 = 1.0/M_LN2; |
62 | |
63 | float |
64 | __ieee754_powf(float x, float y) |
65 | { |
66 | float z, ax, s; |
67 | double d1, d2; |
68 | int32_t i,j,k,yisint,n; |
69 | int32_t hx,hy,ix,iy; |
70 | |
71 | GET_FLOAT_WORD(hy,y); |
72 | iy = hy&0x7fffffff; |
73 | |
74 | /* y==zero: x**0 = 1 */ |
75 | if(iy==0 && !issignaling (x)) return one; |
76 | |
77 | /* x==+-1 */ |
78 | if(x == 1.0 && !issignaling (y)) return one; |
79 | if(x == -1.0 && isinf(y)) return one; |
80 | |
81 | GET_FLOAT_WORD(hx,x); |
82 | ix = hx&0x7fffffff; |
83 | |
84 | /* +-NaN return x+y */ |
85 | if(__builtin_expect(ix > 0x7f800000 || |
86 | iy > 0x7f800000, 0)) |
87 | return x+y; |
88 | |
89 | /* special value of y */ |
90 | if (__builtin_expect(iy==0x7f800000, 0)) { /* y is +-inf */ |
91 | if (ix==0x3f800000) |
92 | return y - y; /* inf**+-1 is NaN */ |
93 | else if (ix > 0x3f800000)/* (|x|>1)**+-inf = inf,0 */ |
94 | return (hy>=0)? y: zero; |
95 | else /* (|x|<1)**-,+inf = inf,0 */ |
96 | return (hy<0)?-y: zero; |
97 | } |
98 | if(iy==0x3f800000) { /* y is +-1 */ |
99 | if(hy<0) return one/x; else return x; |
100 | } |
101 | if(hy==0x40000000) return x*x; /* y is 2 */ |
102 | if(hy==0x3f000000) { /* y is 0.5 */ |
103 | if(__builtin_expect(hx>=0, 1)) /* x >= +0 */ |
104 | return __ieee754_sqrtf(x); |
105 | } |
106 | |
107 | /* determine if y is an odd int when x < 0 |
108 | * yisint = 0 ... y is not an integer |
109 | * yisint = 1 ... y is an odd int |
110 | * yisint = 2 ... y is an even int |
111 | */ |
112 | yisint = 0; |
113 | if(hx<0) { |
114 | if(iy>=0x4b800000) yisint = 2; /* even integer y */ |
115 | else if(iy>=0x3f800000) { |
116 | k = (iy>>23)-0x7f; /* exponent */ |
117 | j = iy>>(23-k); |
118 | if((j<<(23-k))==iy) yisint = 2-(j&1); |
119 | } |
120 | } |
121 | |
122 | ax = fabsf(x); |
123 | /* special value of x */ |
124 | if(__builtin_expect(ix==0x7f800000||ix==0||ix==0x3f800000, 0)){ |
125 | z = ax; /*x is +-0,+-inf,+-1*/ |
126 | if(hy<0) z = one/z; /* z = (1/|x|) */ |
127 | if(hx<0) { |
128 | if(((ix-0x3f800000)|yisint)==0) { |
129 | z = (z-z)/(z-z); /* (-1)**non-int is NaN */ |
130 | } else if(yisint==1) |
131 | z = -z; /* (x<0)**odd = -(|x|**odd) */ |
132 | } |
133 | return z; |
134 | } |
135 | |
136 | /* (x<0)**(non-int) is NaN */ |
137 | if(__builtin_expect(((((u_int32_t)hx>>31)-1)|yisint)==0, 0)) |
138 | return (x-x)/(x-x); |
139 | |
140 | /* |y| is huge */ |
141 | if(__builtin_expect(iy>0x4d000000, 0)) { /* if |y| > 2**27 */ |
142 | /* over/underflow if x is not close to one */ |
143 | if(ix<0x3f7ffff8) return (hy<0)? huge*huge:tiny*tiny; |
144 | if(ix>0x3f800007) return (hy>0)? huge*huge:tiny*tiny; |
145 | /* now |1-x| is tiny <= 2**-20, suffice to compute |
146 | log(x) by x-x^2/2+x^3/3-x^4/4 */ |
147 | d2 = ax-1; /* d2 has 20 trailing zeros. */ |
148 | d2 = d2 * invln2 - |
149 | (d2 * d2) * (0.5 - d2 * (0.333333333333 - d2 * 0.25)) * invln2; |
150 | } else { |
151 | /* Avoid internal underflow for tiny y. The exact value |
152 | of y does not matter if |y| <= 2**-32. */ |
153 | if (iy < 0x2f800000) |
154 | SET_FLOAT_WORD (y, (hy & 0x80000000) | 0x2f800000); |
155 | n = 0; |
156 | /* take care subnormal number */ |
157 | if(ix<0x00800000) |
158 | {ax *= two24; n -= 24; GET_FLOAT_WORD(ix,ax); } |
159 | n += ((ix)>>23)-0x7f; |
160 | j = ix&0x007fffff; |
161 | /* determine interval */ |
162 | ix = j|0x3f800000; /* normalize ix */ |
163 | if(j<=0x1cc471) k=0; /* |x|<sqrt(3/2) */ |
164 | else if(j<0x5db3d7) k=1; /* |x|<sqrt(3) */ |
165 | else {k=0;n+=1;ix -= 0x00800000;} |
166 | SET_FLOAT_WORD(ax,ix); |
167 | |
168 | /* compute d1 = (x-1)/(x+1) or (x-1.5)/(x+1.5) */ |
169 | d1 = (ax-(double)bp[k])/(ax+(double)bp[k]); |
170 | /* compute d2 = log(ax) */ |
171 | d2 = d1 * d1; |
172 | d2 = 3.0 + d2 + d2*d2*(L1+d2*(L2+d2*(L3+d2*(L4+d2*(L5+d2*L6))))); |
173 | /* 2/(3log2)*(d2+...) */ |
174 | d2 = d1*d2*cp; |
175 | /* log2(ax) = (d2+..)*2/(3*log2) */ |
176 | d2 = d2+dp[k]+(double)n; |
177 | } |
178 | |
179 | s = one; /* s (sign of result -ve**odd) = -1 else = 1 */ |
180 | if(((((u_int32_t)hx>>31)-1)|(yisint-1))==0) |
181 | s = -one; /* (-ve)**(odd int) */ |
182 | |
183 | /* compute y * d2 */ |
184 | d1 = y * d2; |
185 | z = d1; |
186 | GET_FLOAT_WORD(j,z); |
187 | if (__builtin_expect(j>0x43000000, 0)) /* if z > 128 */ |
188 | return s*huge*huge; /* overflow */ |
189 | else if (__builtin_expect(j==0x43000000, 0)) { /* if z == 128 */ |
190 | if(ovt>(z-d1)) return s*huge*huge; /* overflow */ |
191 | } |
192 | else if (__builtin_expect((j&0x7fffffff)>0x43160000, 0))/* z <= -150 */ |
193 | return s*tiny*tiny; /* underflow */ |
194 | else if (__builtin_expect((u_int32_t) j==0xc3160000, 0)){/* z == -150*/ |
195 | if(0.0<=(z-d1)) return s*tiny*tiny; /* underflow */ |
196 | } |
197 | /* |
198 | * compute 2**d1 |
199 | */ |
200 | i = j&0x7fffffff; |
201 | k = (i>>23)-0x7f; |
202 | n = 0; |
203 | if(i>0x3f000000) { /* if |z| > 0.5, set n = [z+0.5] */ |
204 | n = j+(0x00800000>>(k+1)); |
205 | k = ((n&0x7fffffff)>>23)-0x7f; /* new k for n */ |
206 | SET_FLOAT_WORD(z,n&~(0x007fffff>>k)); |
207 | n = ((n&0x007fffff)|0x00800000)>>(23-k); |
208 | if(j<0) n = -n; |
209 | d1 -= z; |
210 | } |
211 | d1 = d1 * lg2; |
212 | d2 = d1*d1; |
213 | d2 = d1 - d2*(P1+d2*(P2+d2*(P3+d2*(P4+d2*P5)))); |
214 | d2 = (d1*d2)/(d2-two); |
215 | z = one - (d2-d1); |
216 | GET_FLOAT_WORD(j,z); |
217 | j += (n<<23); |
218 | if((j>>23)<=0) /* subnormal output */ |
219 | { |
220 | z = __scalbnf (z, n); |
221 | float force_underflow = z * z; |
222 | math_force_eval (force_underflow); |
223 | } |
224 | else SET_FLOAT_WORD(z,j); |
225 | return s*z; |
226 | } |
227 | strong_alias (__ieee754_powf, __powf_finite) |
228 | |