e_pow.c 9.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313
  1. /*
  2. * ====================================================
  3. * Copyright (C) 2004 by Sun Microsystems, Inc. All rights reserved.
  4. *
  5. * Permission to use, copy, modify, and distribute this
  6. * software is freely granted, provided that this notice
  7. * is preserved.
  8. * ====================================================
  9. */
  10. /* __ieee754_pow(x,y) return x**y
  11. *
  12. * n
  13. * Method: Let x = 2 * (1+f)
  14. * 1. Compute and return log2(x) in two pieces:
  15. * log2(x) = w1 + w2,
  16. * where w1 has 53-24 = 29 bit trailing zeros.
  17. * 2. Perform y*log2(x) = n+y' by simulating muti-precision
  18. * arithmetic, where |y'|<=0.5.
  19. * 3. Return x**y = 2**n*exp(y'*log2)
  20. *
  21. * Special cases:
  22. * 1. (anything) ** 0 is 1
  23. * 2. (anything) ** 1 is itself
  24. * 3. (anything) ** NAN is NAN
  25. * 4. NAN ** (anything except 0) is NAN
  26. * 5. +-(|x| > 1) ** +INF is +INF
  27. * 6. +-(|x| > 1) ** -INF is +0
  28. * 7. +-(|x| < 1) ** +INF is +0
  29. * 8. +-(|x| < 1) ** -INF is +INF
  30. * 9. +-1 ** +-INF is NAN
  31. * 10. +0 ** (+anything except 0, NAN) is +0
  32. * 11. -0 ** (+anything except 0, NAN, odd integer) is +0
  33. * 12. +0 ** (-anything except 0, NAN) is +INF
  34. * 13. -0 ** (-anything except 0, NAN, odd integer) is +INF
  35. * 14. -0 ** (odd integer) = -( +0 ** (odd integer) )
  36. * 15. +INF ** (+anything except 0,NAN) is +INF
  37. * 16. +INF ** (-anything except 0,NAN) is +0
  38. * 17. -INF ** (anything) = -0 ** (-anything)
  39. * 18. (-anything) ** (integer) is (-1)**(integer)*(+anything**integer)
  40. * 19. (-anything except 0 and inf) ** (non-integer) is NAN
  41. *
  42. * Accuracy:
  43. * pow(x,y) returns x**y nearly rounded. In particular
  44. * pow(integer,integer)
  45. * always returns the correct integer provided it is
  46. * representable.
  47. *
  48. * Constants :
  49. * The hexadecimal values are the intended ones for the following
  50. * constants. The decimal values may be used, provided that the
  51. * compiler will convert from decimal to binary accurately enough
  52. * to produce the hexadecimal values shown.
  53. */
  54. #include "fdlibm.h"
  55. #ifndef _DOUBLE_IS_32BITS
  56. #ifdef __STDC__
  57. static const double
  58. #else
  59. static double
  60. #endif
  61. bp[] = {1.0, 1.5,},
  62. dp_h[] = { 0.0, 5.84962487220764160156e-01,}, /* 0x3FE2B803, 0x40000000 */
  63. dp_l[] = { 0.0, 1.35003920212974897128e-08,}, /* 0x3E4CFDEB, 0x43CFD006 */
  64. zero = 0.0,
  65. one = 1.0,
  66. two = 2.0,
  67. two53 = 9007199254740992.0, /* 0x43400000, 0x00000000 */
  68. huge = 1.0e300,
  69. tiny = 1.0e-300,
  70. /* poly coefs for (3/2)*(log(x)-2s-2/3*s**3 */
  71. L1 = 5.99999999999994648725e-01, /* 0x3FE33333, 0x33333303 */
  72. L2 = 4.28571428578550184252e-01, /* 0x3FDB6DB6, 0xDB6FABFF */
  73. L3 = 3.33333329818377432918e-01, /* 0x3FD55555, 0x518F264D */
  74. L4 = 2.72728123808534006489e-01, /* 0x3FD17460, 0xA91D4101 */
  75. L5 = 2.30660745775561754067e-01, /* 0x3FCD864A, 0x93C9DB65 */
  76. L6 = 2.06975017800338417784e-01, /* 0x3FCA7E28, 0x4A454EEF */
  77. P1 = 1.66666666666666019037e-01, /* 0x3FC55555, 0x5555553E */
  78. P2 = -2.77777777770155933842e-03, /* 0xBF66C16C, 0x16BEBD93 */
  79. P3 = 6.61375632143793436117e-05, /* 0x3F11566A, 0xAF25DE2C */
  80. P4 = -1.65339022054652515390e-06, /* 0xBEBBBD41, 0xC5D26BF1 */
  81. P5 = 4.13813679705723846039e-08, /* 0x3E663769, 0x72BEA4D0 */
  82. lg2 = 6.93147180559945286227e-01, /* 0x3FE62E42, 0xFEFA39EF */
  83. lg2_h = 6.93147182464599609375e-01, /* 0x3FE62E43, 0x00000000 */
  84. lg2_l = -1.90465429995776804525e-09, /* 0xBE205C61, 0x0CA86C39 */
  85. ovt = 8.0085662595372944372e-0017, /* -(1024-log2(ovfl+.5ulp)) */
  86. cp = 9.61796693925975554329e-01, /* 0x3FEEC709, 0xDC3A03FD =2/(3ln2) */
  87. cp_h = 9.61796700954437255859e-01, /* 0x3FEEC709, 0xE0000000 =(float)cp */
  88. cp_l = -7.02846165095275826516e-09, /* 0xBE3E2FE0, 0x145B01F5 =tail of cp_h*/
  89. ivln2 = 1.44269504088896338700e+00, /* 0x3FF71547, 0x652B82FE =1/ln2 */
  90. ivln2_h = 1.44269502162933349609e+00, /* 0x3FF71547, 0x60000000 =24b 1/ln2*/
  91. ivln2_l = 1.92596299112661746887e-08; /* 0x3E54AE0B, 0xF85DDF44 =1/ln2 tail*/
  92. #ifdef __STDC__
  93. double __ieee754_pow(double x, double y)
  94. #else
  95. double __ieee754_pow(x,y)
  96. double x, y;
  97. #endif
  98. {
  99. double z,ax,z_h,z_l,p_h,p_l;
  100. double y1,t1,t2,r,s,t,u,v,w;
  101. int32_t i0,i1,i,j,k,yisint,n;
  102. int32_t hx,hy,ix,iy;
  103. uint32_t lx,ly;
  104. i0 = ((*(int*)&one)>>29)^1; i1=1-i0;
  105. EXTRACT_WORDS(hx,lx,x);
  106. EXTRACT_WORDS(hy,ly,y);
  107. ix = hx&0x7fffffff; iy = hy&0x7fffffff;
  108. /* y==zero: x**0 = 1 */
  109. if((iy|ly)==0) return one;
  110. /* +-NaN return x+y */
  111. if(ix > 0x7ff00000 || ((ix==0x7ff00000)&&(lx!=0)) ||
  112. iy > 0x7ff00000 || ((iy==0x7ff00000)&&(ly!=0)))
  113. return x+y;
  114. /* determine if y is an odd int when x < 0
  115. * yisint = 0 ... y is not an integer
  116. * yisint = 1 ... y is an odd int
  117. * yisint = 2 ... y is an even int
  118. */
  119. yisint = 0;
  120. if(hx<0) {
  121. if(iy>=0x43400000) yisint = 2; /* even integer y */
  122. else if(iy>=0x3ff00000) {
  123. k = (iy>>20)-0x3ff; /* exponent */
  124. if(k>20) {
  125. j = ly>>(52-k);
  126. if((uint32_t)(j<<(52-k))==ly) yisint = 2-(j&1);
  127. } else if(ly==0) {
  128. j = iy>>(20-k);
  129. if((j<<(20-k))==iy) yisint = 2-(j&1);
  130. }
  131. }
  132. }
  133. /* special value of y */
  134. if(ly==0) {
  135. if (iy==0x7ff00000) { /* y is +-inf */
  136. if(((ix-0x3ff00000)|lx)==0)
  137. return y - y; /* inf**+-1 is NaN */
  138. else if (ix >= 0x3ff00000)/* (|x|>1)**+-inf = inf,0 */
  139. return (hy>=0)? y: zero;
  140. else /* (|x|<1)**-,+inf = inf,0 */
  141. return (hy<0)?-y: zero;
  142. }
  143. if(iy==0x3ff00000) { /* y is +-1 */
  144. if(hy<0) return one/x; else return x;
  145. }
  146. if(hy==0x40000000) return x*x; /* y is 2 */
  147. if(hy==0x3fe00000) { /* y is 0.5 */
  148. if(hx>=0) /* x >= +0 */
  149. return __ieee754_sqrt(x);
  150. }
  151. }
  152. ax = fabs(x);
  153. /* special value of x */
  154. if(lx==0) {
  155. if(ix==0x7ff00000||ix==0||ix==0x3ff00000){
  156. z = ax; /*x is +-0,+-inf,+-1*/
  157. if(hy<0) z = one/z; /* z = (1/|x|) */
  158. if(hx<0) {
  159. if(((ix-0x3ff00000)|yisint)==0) {
  160. z = (z-z)/(z-z); /* (-1)**non-int is NaN */
  161. } else if(yisint==1)
  162. z = -z; /* (x<0)**odd = -(|x|**odd) */
  163. }
  164. return z;
  165. }
  166. }
  167. n = (hx>>31)+1;
  168. /* (x<0)**(non-int) is NaN */
  169. if((n|yisint)==0) return (x-x)/(x-x);
  170. s = one; /* s (sign of result -ve**odd) = -1 else = 1 */
  171. if((n|(yisint-1))==0) s = -one;/* (-ve)**(odd int) */
  172. /* |y| is huge */
  173. if(iy>0x41e00000) { /* if |y| > 2**31 */
  174. if(iy>0x43f00000){ /* if |y| > 2**64, must o/uflow */
  175. if(ix<=0x3fefffff) return (hy<0)? huge*huge:tiny*tiny;
  176. if(ix>=0x3ff00000) return (hy>0)? huge*huge:tiny*tiny;
  177. }
  178. /* over/underflow if x is not close to one */
  179. if(ix<0x3fefffff) return (hy<0)? s*huge*huge:s*tiny*tiny;
  180. if(ix>0x3ff00000) return (hy>0)? s*huge*huge:s*tiny*tiny;
  181. /* now |1-x| is tiny <= 2**-20, suffice to compute
  182. log(x) by x-x^2/2+x^3/3-x^4/4 */
  183. t = ax-one; /* t has 20 trailing zeros */
  184. w = (t*t)*(0.5-t*(0.3333333333333333333333-t*0.25));
  185. u = ivln2_h*t; /* ivln2_h has 21 sig. bits */
  186. v = t*ivln2_l-w*ivln2;
  187. t1 = u+v;
  188. SET_LOW_WORD(t1,0);
  189. t2 = v-(t1-u);
  190. } else {
  191. double ss,s2,s_h,s_l,t_h,t_l;
  192. n = 0;
  193. /* take care subnormal number */
  194. if(ix<0x00100000)
  195. {ax *= two53; n -= 53; GET_HIGH_WORD(ix,ax); }
  196. n += ((ix)>>20)-0x3ff;
  197. j = ix&0x000fffff;
  198. /* determine interval */
  199. ix = j|0x3ff00000; /* normalize ix */
  200. if(j<=0x3988E) k=0; /* |x|<sqrt(3/2) */
  201. else if(j<0xBB67A) k=1; /* |x|<sqrt(3) */
  202. else {k=0;n+=1;ix -= 0x00100000;}
  203. SET_HIGH_WORD(ax,ix);
  204. /* compute ss = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */
  205. u = ax-bp[k]; /* bp[0]=1.0, bp[1]=1.5 */
  206. v = one/(ax+bp[k]);
  207. ss = u*v;
  208. s_h = ss;
  209. SET_LOW_WORD(s_h,0);
  210. /* t_h=ax+bp[k] High */
  211. t_h = zero;
  212. SET_HIGH_WORD(t_h,((ix>>1)|0x20000000)+0x00080000+(k<<18));
  213. t_l = ax - (t_h-bp[k]);
  214. s_l = v*((u-s_h*t_h)-s_h*t_l);
  215. /* compute log(ax) */
  216. s2 = ss*ss;
  217. r = s2*s2*(L1+s2*(L2+s2*(L3+s2*(L4+s2*(L5+s2*L6)))));
  218. r += s_l*(s_h+ss);
  219. s2 = s_h*s_h;
  220. t_h = 3.0+s2+r;
  221. SET_LOW_WORD(t_h,0);
  222. t_l = r-((t_h-3.0)-s2);
  223. /* u+v = ss*(1+...) */
  224. u = s_h*t_h;
  225. v = s_l*t_h+t_l*ss;
  226. /* 2/(3log2)*(ss+...) */
  227. p_h = u+v;
  228. SET_LOW_WORD(p_h,0);
  229. p_l = v-(p_h-u);
  230. z_h = cp_h*p_h; /* cp_h+cp_l = 2/(3*log2) */
  231. z_l = cp_l*p_h+p_l*cp+dp_l[k];
  232. /* log2(ax) = (ss+..)*2/(3*log2) = n + dp_h + z_h + z_l */
  233. t = (double)n;
  234. t1 = (((z_h+z_l)+dp_h[k])+t);
  235. SET_LOW_WORD(t1,0);
  236. t2 = z_l-(((t1-t)-dp_h[k])-z_h);
  237. }
  238. /* split up y into y1+y2 and compute (y1+y2)*(t1+t2) */
  239. y1 = y;
  240. SET_LOW_WORD(y1,0);
  241. p_l = (y-y1)*t1+y*t2;
  242. p_h = y1*t1;
  243. z = p_l+p_h;
  244. EXTRACT_WORDS(j,i,z);
  245. if (j>=0x40900000) { /* z >= 1024 */
  246. if(((j-0x40900000)|i)!=0) /* if z > 1024 */
  247. return s*huge*huge; /* overflow */
  248. else {
  249. if(p_l+ovt>z-p_h) return s*huge*huge; /* overflow */
  250. }
  251. } else if((j&0x7fffffff)>=0x4090cc00 ) { /* z <= -1075 */
  252. if(((j-0xc090cc00)|i)!=0) /* z < -1075 */
  253. return s*tiny*tiny; /* underflow */
  254. else {
  255. if(p_l<=z-p_h) return s*tiny*tiny; /* underflow */
  256. }
  257. }
  258. /*
  259. * compute 2**(p_h+p_l)
  260. */
  261. i = j&0x7fffffff;
  262. k = (i>>20)-0x3ff;
  263. n = 0;
  264. if(i>0x3fe00000) { /* if |z| > 0.5, set n = [z+0.5] */
  265. n = j+(0x00100000>>(k+1));
  266. k = ((n&0x7fffffff)>>20)-0x3ff; /* new k for n */
  267. t = zero;
  268. SET_HIGH_WORD(t,(n&~(0x000fffff>>k)));
  269. n = ((n&0x000fffff)|0x00100000)>>(20-k);
  270. if(j<0) n = -n;
  271. p_h -= t;
  272. }
  273. t = p_l+p_h;
  274. SET_LOW_WORD(t,0);
  275. u = t*lg2_h;
  276. v = (p_l-(t-p_h))*lg2+t*lg2_l;
  277. z = u+v;
  278. w = v-(z-u);
  279. t = z*z;
  280. t1 = z - t*(P1+t*(P2+t*(P3+t*(P4+t*P5))));
  281. r = (z*t1)/(t1-two)-(w+z*w);
  282. z = one-(r-z);
  283. GET_HIGH_WORD(j,z);
  284. j += (n<<20);
  285. if((j>>20)<=0) z = scalbn(z,n); /* subnormal output */
  286. else
  287. {
  288. uint32_t hz;
  289. GET_HIGH_WORD(hz,z);
  290. SET_HIGH_WORD(z,hz + (n<<20));
  291. }
  292. return s*z;
  293. }
  294. #endif /* defined(_DOUBLE_IS_32BITS) */