1 | /* |
---|
2 | * ==================================================== |
---|
3 | * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved. |
---|
4 | * |
---|
5 | * Developed at SunPro, a Sun Microsystems, Inc. business. |
---|
6 | * Permission to use, copy, modify, and distribute this |
---|
7 | * software is freely granted, provided that this notice |
---|
8 | * is preserved. |
---|
9 | * ==================================================== |
---|
10 | */ |
---|
11 | |
---|
12 | /* |
---|
13 | * Modified for ALMOS-MKH OS at UPMC, France, August 2018. (Alain Greiner) |
---|
14 | */ |
---|
15 | |
---|
16 | /* __ieee754_pow(x,y) return x**y |
---|
17 | * |
---|
18 | * n |
---|
19 | * Method: Let x = 2 * (1+f) |
---|
20 | * 1. Compute and return log2(x) in two pieces: |
---|
21 | * log2(x) = w1 + w2, |
---|
22 | * where w1 has 53-24 = 29 bit trailing zeros. |
---|
23 | * 2. Perform y*log2(x) = n+y' by simulating muti-precision |
---|
24 | * arithmetic, where |y'|<=0.5. |
---|
25 | * 3. Return x**y = 2**n*exp(y'*log2) |
---|
26 | * |
---|
27 | * Special cases: |
---|
28 | * 1. +-1 ** anything is 1.0 |
---|
29 | * 2. +-1 ** +-INF is 1.0 |
---|
30 | * 3. (anything) ** 0 is 1 |
---|
31 | * 4. (anything) ** 1 is itself |
---|
32 | * 5. (anything) ** NAN is NAN |
---|
33 | * 6. NAN ** (anything except 0) is NAN |
---|
34 | * 7. +-(|x| > 1) ** +INF is +INF |
---|
35 | * 8. +-(|x| > 1) ** -INF is +0 |
---|
36 | * 9. +-(|x| < 1) ** +INF is +0 |
---|
37 | * 10 +-(|x| < 1) ** -INF is +INF |
---|
38 | * 11. +0 ** (+anything except 0, NAN) is +0 |
---|
39 | * 12. -0 ** (+anything except 0, NAN, odd integer) is +0 |
---|
40 | * 13. +0 ** (-anything except 0, NAN) is +INF |
---|
41 | * 14. -0 ** (-anything except 0, NAN, odd integer) is +INF |
---|
42 | * 15. -0 ** (odd integer) = -( +0 ** (odd integer) ) |
---|
43 | * 16. +INF ** (+anything except 0,NAN) is +INF |
---|
44 | * 17. +INF ** (-anything except 0,NAN) is +0 |
---|
45 | * 18. -INF ** (anything) = -0 ** (-anything) |
---|
46 | * 19. (-anything) ** (integer) is (-1)**(integer)*(+anything**integer) |
---|
47 | * 20. (-anything except 0 and inf) ** (non-integer) is NAN |
---|
48 | * |
---|
49 | * Accuracy: |
---|
50 | * pow(x,y) returns x**y nearly rounded. In particular |
---|
51 | * pow(integer,integer) |
---|
52 | * always returns the correct integer provided it is |
---|
53 | * representable. |
---|
54 | * |
---|
55 | * Constants : |
---|
56 | * The hexadecimal values are the intended ones for the following |
---|
57 | * constants. The decimal values may be used, provided that the |
---|
58 | * compiler will convert from decimal to binary accurately enough |
---|
59 | * to produce the hexadecimal values shown. |
---|
60 | */ |
---|
61 | |
---|
62 | #include "math.h" |
---|
63 | #include "math_private.h" |
---|
64 | |
---|
65 | static const double |
---|
66 | bp[] = {1.0, 1.5,}, |
---|
67 | dp_h[] = { 0.0, 5.84962487220764160156e-01,}, /* 0x3FE2B803, 0x40000000 */ |
---|
68 | dp_l[] = { 0.0, 1.35003920212974897128e-08,}, /* 0x3E4CFDEB, 0x43CFD006 */ |
---|
69 | zero = 0.0, |
---|
70 | one = 1.0, |
---|
71 | two = 2.0, |
---|
72 | two53 = 9007199254740992.0, /* 0x43400000, 0x00000000 */ |
---|
73 | huge = 1.0e300, |
---|
74 | tiny = 1.0e-300, |
---|
75 | /* poly coefs for (3/2)*(log(x)-2s-2/3*s**3 */ |
---|
76 | L1 = 5.99999999999994648725e-01, /* 0x3FE33333, 0x33333303 */ |
---|
77 | L2 = 4.28571428578550184252e-01, /* 0x3FDB6DB6, 0xDB6FABFF */ |
---|
78 | L3 = 3.33333329818377432918e-01, /* 0x3FD55555, 0x518F264D */ |
---|
79 | L4 = 2.72728123808534006489e-01, /* 0x3FD17460, 0xA91D4101 */ |
---|
80 | L5 = 2.30660745775561754067e-01, /* 0x3FCD864A, 0x93C9DB65 */ |
---|
81 | L6 = 2.06975017800338417784e-01, /* 0x3FCA7E28, 0x4A454EEF */ |
---|
82 | P1 = 1.66666666666666019037e-01, /* 0x3FC55555, 0x5555553E */ |
---|
83 | P2 = -2.77777777770155933842e-03, /* 0xBF66C16C, 0x16BEBD93 */ |
---|
84 | P3 = 6.61375632143793436117e-05, /* 0x3F11566A, 0xAF25DE2C */ |
---|
85 | P4 = -1.65339022054652515390e-06, /* 0xBEBBBD41, 0xC5D26BF1 */ |
---|
86 | P5 = 4.13813679705723846039e-08, /* 0x3E663769, 0x72BEA4D0 */ |
---|
87 | lg2 = 6.93147180559945286227e-01, /* 0x3FE62E42, 0xFEFA39EF */ |
---|
88 | lg2_h = 6.93147182464599609375e-01, /* 0x3FE62E43, 0x00000000 */ |
---|
89 | lg2_l = -1.90465429995776804525e-09, /* 0xBE205C61, 0x0CA86C39 */ |
---|
90 | ovt = 8.0085662595372944372e-0017, /* -(1024-log2(ovfl+.5ulp)) */ |
---|
91 | cp = 9.61796693925975554329e-01, /* 0x3FEEC709, 0xDC3A03FD =2/(3ln2) */ |
---|
92 | cp_h = 9.61796700954437255859e-01, /* 0x3FEEC709, 0xE0000000 =(float)cp */ |
---|
93 | cp_l = -7.02846165095275826516e-09, /* 0xBE3E2FE0, 0x145B01F5 =tail of cp_h*/ |
---|
94 | ivln2 = 1.44269504088896338700e+00, /* 0x3FF71547, 0x652B82FE =1/ln2 */ |
---|
95 | ivln2_h = 1.44269502162933349609e+00, /* 0x3FF71547, 0x60000000 =24b 1/ln2*/ |
---|
96 | ivln2_l = 1.92596299112661746887e-08; /* 0x3E54AE0B, 0xF85DDF44 =1/ln2 tail*/ |
---|
97 | |
---|
98 | double __ieee754_pow(double x, double y) |
---|
99 | { |
---|
100 | double z,ax,z_h,z_l,p_h,p_l; |
---|
101 | double y1,t1,t2,r,s,t,u,v,w; |
---|
102 | int32_t i,j,k,yisint,n; |
---|
103 | int32_t hx,hy,ix,iy; |
---|
104 | uint32_t lx,ly; |
---|
105 | |
---|
106 | EXTRACT_WORDS(hx,lx,x); |
---|
107 | /* x==1: 1**y = 1 (even if y is NaN) */ |
---|
108 | if (hx==0x3ff00000 && lx==0) { |
---|
109 | return x; |
---|
110 | } |
---|
111 | ix = hx&0x7fffffff; |
---|
112 | |
---|
113 | EXTRACT_WORDS(hy,ly,y); |
---|
114 | iy = hy&0x7fffffff; |
---|
115 | |
---|
116 | /* y==zero: x**0 = 1 */ |
---|
117 | if((iy|ly)==0) return one; |
---|
118 | |
---|
119 | /* +-NaN return x+y */ |
---|
120 | if(ix > 0x7ff00000 || ((ix==0x7ff00000)&&(lx!=0)) || |
---|
121 | iy > 0x7ff00000 || ((iy==0x7ff00000)&&(ly!=0))) |
---|
122 | return x+y; |
---|
123 | |
---|
124 | /* determine if y is an odd int when x < 0 |
---|
125 | * yisint = 0 ... y is not an integer |
---|
126 | * yisint = 1 ... y is an odd int |
---|
127 | * yisint = 2 ... y is an even int |
---|
128 | */ |
---|
129 | yisint = 0; |
---|
130 | if(hx<0) { |
---|
131 | if(iy>=0x43400000) yisint = 2; /* even integer y */ |
---|
132 | else if(iy>=0x3ff00000) { |
---|
133 | k = (iy>>20)-0x3ff; /* exponent */ |
---|
134 | if(k>20) { |
---|
135 | j = ly>>(52-k); |
---|
136 | if((j<<(52-k))==(int32_t)ly) yisint = 2-(j&1); |
---|
137 | } else if(ly==0) { |
---|
138 | j = iy>>(20-k); |
---|
139 | if((j<<(20-k))==iy) yisint = 2-(j&1); |
---|
140 | } |
---|
141 | } |
---|
142 | } |
---|
143 | |
---|
144 | /* special value of y */ |
---|
145 | if(ly==0) { |
---|
146 | if (iy==0x7ff00000) { /* y is +-inf */ |
---|
147 | if (((ix-0x3ff00000)|lx)==0) |
---|
148 | return one; /* +-1**+-inf is 1 (yes, weird rule) */ |
---|
149 | if (ix >= 0x3ff00000) /* (|x|>1)**+-inf = inf,0 */ |
---|
150 | return (hy>=0) ? y : zero; |
---|
151 | /* (|x|<1)**-,+inf = inf,0 */ |
---|
152 | return (hy<0) ? -y : zero; |
---|
153 | } |
---|
154 | if(iy==0x3ff00000) { /* y is +-1 */ |
---|
155 | if(hy<0) return one/x; else return x; |
---|
156 | } |
---|
157 | if(hy==0x40000000) return x*x; /* y is 2 */ |
---|
158 | if(hy==0x3fe00000) { /* y is 0.5 */ |
---|
159 | if(hx>=0) /* x >= +0 */ |
---|
160 | return sqrt(x); |
---|
161 | } |
---|
162 | } |
---|
163 | |
---|
164 | ax = fabs(x); |
---|
165 | /* special value of x */ |
---|
166 | if(lx==0) { |
---|
167 | if(ix==0x7ff00000||ix==0||ix==0x3ff00000){ |
---|
168 | z = ax; /*x is +-0,+-inf,+-1*/ |
---|
169 | if(hy<0) z = one/z; /* z = (1/|x|) */ |
---|
170 | if(hx<0) { |
---|
171 | if(((ix-0x3ff00000)|yisint)==0) { |
---|
172 | z = (z-z)/(z-z); /* (-1)**non-int is NaN */ |
---|
173 | } else if(yisint==1) |
---|
174 | z = -z; /* (x<0)**odd = -(|x|**odd) */ |
---|
175 | } |
---|
176 | return z; |
---|
177 | } |
---|
178 | } |
---|
179 | |
---|
180 | /* (x<0)**(non-int) is NaN */ |
---|
181 | if(((((uint32_t)hx>>31)-1)|yisint)==0) return (x-x)/(x-x); |
---|
182 | |
---|
183 | /* |y| is huge */ |
---|
184 | if(iy>0x41e00000) { /* if |y| > 2**31 */ |
---|
185 | if(iy>0x43f00000){ /* if |y| > 2**64, must o/uflow */ |
---|
186 | if(ix<=0x3fefffff) return (hy<0)? huge*huge:tiny*tiny; |
---|
187 | if(ix>=0x3ff00000) return (hy>0)? huge*huge:tiny*tiny; |
---|
188 | } |
---|
189 | /* over/underflow if x is not close to one */ |
---|
190 | if(ix<0x3fefffff) return (hy<0)? huge*huge:tiny*tiny; |
---|
191 | if(ix>0x3ff00000) return (hy>0)? huge*huge:tiny*tiny; |
---|
192 | /* now |1-x| is tiny <= 2**-20, suffice to compute |
---|
193 | log(x) by x-x^2/2+x^3/3-x^4/4 */ |
---|
194 | t = x-1; /* t has 20 trailing zeros */ |
---|
195 | w = (t*t)*(0.5-t*(0.3333333333333333333333-t*0.25)); |
---|
196 | u = ivln2_h*t; /* ivln2_h has 21 sig. bits */ |
---|
197 | v = t*ivln2_l-w*ivln2; |
---|
198 | t1 = u+v; |
---|
199 | SET_LOW_WORD(t1,0); |
---|
200 | t2 = v-(t1-u); |
---|
201 | } else { |
---|
202 | double s2,s_h,s_l,t_h,t_l; |
---|
203 | n = 0; |
---|
204 | /* take care subnormal number */ |
---|
205 | if(ix<0x00100000) |
---|
206 | {ax *= two53; n -= 53; GET_HIGH_WORD(ix,ax); } |
---|
207 | n += ((ix)>>20)-0x3ff; |
---|
208 | j = ix&0x000fffff; |
---|
209 | /* determine interval */ |
---|
210 | ix = j|0x3ff00000; /* normalize ix */ |
---|
211 | if(j<=0x3988E) k=0; /* |x|<sqrt(3/2) */ |
---|
212 | else if(j<0xBB67A) k=1; /* |x|<sqrt(3) */ |
---|
213 | else {k=0;n+=1;ix -= 0x00100000;} |
---|
214 | SET_HIGH_WORD(ax,ix); |
---|
215 | |
---|
216 | /* compute s = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */ |
---|
217 | u = ax-bp[k]; /* bp[0]=1.0, bp[1]=1.5 */ |
---|
218 | v = one/(ax+bp[k]); |
---|
219 | s = u*v; |
---|
220 | s_h = s; |
---|
221 | SET_LOW_WORD(s_h,0); |
---|
222 | /* t_h=ax+bp[k] High */ |
---|
223 | t_h = zero; |
---|
224 | SET_HIGH_WORD(t_h,((ix>>1)|0x20000000)+0x00080000+(k<<18)); |
---|
225 | t_l = ax - (t_h-bp[k]); |
---|
226 | s_l = v*((u-s_h*t_h)-s_h*t_l); |
---|
227 | /* compute log(ax) */ |
---|
228 | s2 = s*s; |
---|
229 | r = s2*s2*(L1+s2*(L2+s2*(L3+s2*(L4+s2*(L5+s2*L6))))); |
---|
230 | r += s_l*(s_h+s); |
---|
231 | s2 = s_h*s_h; |
---|
232 | t_h = 3.0+s2+r; |
---|
233 | SET_LOW_WORD(t_h,0); |
---|
234 | t_l = r-((t_h-3.0)-s2); |
---|
235 | /* u+v = s*(1+...) */ |
---|
236 | u = s_h*t_h; |
---|
237 | v = s_l*t_h+t_l*s; |
---|
238 | /* 2/(3log2)*(s+...) */ |
---|
239 | p_h = u+v; |
---|
240 | SET_LOW_WORD(p_h,0); |
---|
241 | p_l = v-(p_h-u); |
---|
242 | z_h = cp_h*p_h; /* cp_h+cp_l = 2/(3*log2) */ |
---|
243 | z_l = cp_l*p_h+p_l*cp+dp_l[k]; |
---|
244 | /* log2(ax) = (s+..)*2/(3*log2) = n + dp_h + z_h + z_l */ |
---|
245 | t = (double)n; |
---|
246 | t1 = (((z_h+z_l)+dp_h[k])+t); |
---|
247 | SET_LOW_WORD(t1,0); |
---|
248 | t2 = z_l-(((t1-t)-dp_h[k])-z_h); |
---|
249 | } |
---|
250 | |
---|
251 | s = one; /* s (sign of result -ve**odd) = -1 else = 1 */ |
---|
252 | if(((((uint32_t)hx>>31)-1)|(yisint-1))==0) |
---|
253 | s = -one;/* (-ve)**(odd int) */ |
---|
254 | |
---|
255 | /* split up y into y1+y2 and compute (y1+y2)*(t1+t2) */ |
---|
256 | y1 = y; |
---|
257 | SET_LOW_WORD(y1,0); |
---|
258 | p_l = (y-y1)*t1+y*t2; |
---|
259 | p_h = y1*t1; |
---|
260 | z = p_l+p_h; |
---|
261 | EXTRACT_WORDS(j,i,z); |
---|
262 | if (j>=0x40900000) { /* z >= 1024 */ |
---|
263 | if(((j-0x40900000)|i)!=0) /* if z > 1024 */ |
---|
264 | return s*huge*huge; /* overflow */ |
---|
265 | else { |
---|
266 | if(p_l+ovt>z-p_h) return s*huge*huge; /* overflow */ |
---|
267 | } |
---|
268 | } else if((j&0x7fffffff)>=0x4090cc00 ) { /* z <= -1075 */ |
---|
269 | if(((j-0xc090cc00)|i)!=0) /* z < -1075 */ |
---|
270 | return s*tiny*tiny; /* underflow */ |
---|
271 | else { |
---|
272 | if(p_l<=z-p_h) return s*tiny*tiny; /* underflow */ |
---|
273 | } |
---|
274 | } |
---|
275 | /* |
---|
276 | * compute 2**(p_h+p_l) |
---|
277 | */ |
---|
278 | i = j&0x7fffffff; |
---|
279 | k = (i>>20)-0x3ff; |
---|
280 | n = 0; |
---|
281 | if(i>0x3fe00000) { /* if |z| > 0.5, set n = [z+0.5] */ |
---|
282 | n = j+(0x00100000>>(k+1)); |
---|
283 | k = ((n&0x7fffffff)>>20)-0x3ff; /* new k for n */ |
---|
284 | t = zero; |
---|
285 | SET_HIGH_WORD(t,n&~(0x000fffff>>k)); |
---|
286 | n = ((n&0x000fffff)|0x00100000)>>(20-k); |
---|
287 | if(j<0) n = -n; |
---|
288 | p_h -= t; |
---|
289 | } |
---|
290 | t = p_l+p_h; |
---|
291 | SET_LOW_WORD(t,0); |
---|
292 | u = t*lg2_h; |
---|
293 | v = (p_l-(t-p_h))*lg2+t*lg2_l; |
---|
294 | z = u+v; |
---|
295 | w = v-(z-u); |
---|
296 | t = z*z; |
---|
297 | t1 = z - t*(P1+t*(P2+t*(P3+t*(P4+t*P5)))); |
---|
298 | r = (z*t1)/(t1-two)-(w+z*w); |
---|
299 | z = one-(r-z); |
---|
300 | GET_HIGH_WORD(j,z); |
---|
301 | j += (n<<20); |
---|
302 | if((j>>20)<=0) z = scalbn(z,n); /* subnormal output */ |
---|
303 | else SET_HIGH_WORD(z,j); |
---|
304 | return s*z; |
---|
305 | } |
---|
306 | |
---|
307 | /* |
---|
308 | * wrapper pow(x,y) return x**y |
---|
309 | */ |
---|
310 | double pow(double x, double y) |
---|
311 | { |
---|
312 | double z = __ieee754_pow(x, y); |
---|
313 | if (isnan(y)) { |
---|
314 | return z; |
---|
315 | } |
---|
316 | if (isnan(x)) { |
---|
317 | if (y == 0.0) { |
---|
318 | return x; /* pow(NaN,0.0) */ |
---|
319 | } |
---|
320 | return z; |
---|
321 | } |
---|
322 | if (x == 0.0) { |
---|
323 | if (y == 0.0) { |
---|
324 | return 0; /* pow(0.0,0.0) */ |
---|
325 | } |
---|
326 | if (isfinite(y) && y < 0.0) { |
---|
327 | return -huge; /* pow(0.0,negative) */ |
---|
328 | } |
---|
329 | return z; |
---|
330 | } |
---|
331 | if (!isfinite(z)) { |
---|
332 | if (isfinite(x) && isfinite(y)) { |
---|
333 | if (isnan(z)) { |
---|
334 | return 0.0 / 0.0; /* pow neg**non-int */ |
---|
335 | } |
---|
336 | { |
---|
337 | double ret = huge; |
---|
338 | y *= 0.5; |
---|
339 | if (x < zero && rint(y) !=y ) { |
---|
340 | ret = -huge; |
---|
341 | } |
---|
342 | return ret; |
---|
343 | } |
---|
344 | } |
---|
345 | } |
---|
346 | if (z == 0.0 && isfinite(x) && isfinite(y)) { |
---|
347 | return 0.0; |
---|
348 | } |
---|
349 | return z; |
---|
350 | } |
---|
351 | |
---|
352 | /* |
---|
353 | * Pseudo exp function (QM) |
---|
354 | * There probably is a better implementation |
---|
355 | */ |
---|
356 | double exp(double x) |
---|
357 | { |
---|
358 | const double e = 2.71828182846; |
---|
359 | return pow(e, x); |
---|
360 | } |
---|
361 | |
---|