e_lgammal_r.c 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358
  1. /* @(#)e_lgamma_r.c 1.3 95/01/18 */
  2. /*
  3. * ====================================================
  4. * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
  5. *
  6. * Developed at SunSoft, a Sun Microsystems, Inc. business.
  7. * Permission to use, copy, modify, and distribute this
  8. * software is freely granted, provided that this notice
  9. * is preserved.
  10. * ====================================================
  11. */
  12. #include <openlibm_compat.h>
  13. __FBSDID("$FreeBSD$");
  14. /*
  15. * See e_lgamma_r.c for complete comments.
  16. *
  17. * Converted to long double by Steven G. Kargl.
  18. */
  19. #ifdef __i386__
  20. #include <ieeefp.h>
  21. #endif
  22. #include "fpmath.h"
  23. #include <openlibm_math.h>
  24. #include "math_private.h"
  25. static const volatile double vzero = 0;
  26. static const double
  27. zero= 0,
  28. half= 0.5,
  29. one = 1;
  30. static const union IEEEl2bits
  31. piu = LD80C(0xc90fdaa22168c235, 1, 3.14159265358979323851e+00L);
  32. #define pi (piu.e)
  33. /*
  34. * Domain y in [0x1p-70, 0.27], range ~[-4.5264e-22, 4.5264e-22]:
  35. * |(lgamma(2 - y) + y / 2) / y - a(y)| < 2**-70.9
  36. */
  37. static const union IEEEl2bits
  38. a0u = LD80C(0x9e233f1bed863d26, -4, 7.72156649015328606028e-02L),
  39. a1u = LD80C(0xa51a6625307d3249, -2, 3.22467033424113218889e-01L),
  40. a2u = LD80C(0x89f000d2abafda8c, -4, 6.73523010531979398946e-02L),
  41. a3u = LD80C(0xa8991563eca75f26, -6, 2.05808084277991211934e-02L),
  42. a4u = LD80C(0xf2027e10634ce6b6, -8, 7.38555102796070454026e-03L),
  43. a5u = LD80C(0xbd6eb76dd22187f4, -9, 2.89051035162703932972e-03L),
  44. a6u = LD80C(0x9c562ab05e0458ed, -10, 1.19275351624639999297e-03L),
  45. a7u = LD80C(0x859baed93ee48e46, -11, 5.09674593842117925320e-04L),
  46. a8u = LD80C(0xe9f28a4432949af2, -13, 2.23109648015769155122e-04L),
  47. a9u = LD80C(0xd12ad0d9b93c6bb0, -14, 9.97387167479808509830e-05L),
  48. a10u= LD80C(0xb7522643c78a219b, -15, 4.37071076331030136818e-05L),
  49. a11u= LD80C(0xca024dcdece2cb79, -16, 2.40813493372040143061e-05L),
  50. a12u= LD80C(0xbb90fb6968ebdbf9, -19, 2.79495621083634031729e-06L),
  51. a13u= LD80C(0xba1c9ffeeae07b37, -17, 1.10931287015513924136e-05L);
  52. #define a0 (a0u.e)
  53. #define a1 (a1u.e)
  54. #define a2 (a2u.e)
  55. #define a3 (a3u.e)
  56. #define a4 (a4u.e)
  57. #define a5 (a5u.e)
  58. #define a6 (a6u.e)
  59. #define a7 (a7u.e)
  60. #define a8 (a8u.e)
  61. #define a9 (a9u.e)
  62. #define a10 (a10u.e)
  63. #define a11 (a11u.e)
  64. #define a12 (a12u.e)
  65. #define a13 (a13u.e)
  66. /*
  67. * Domain x in [tc-0.24, tc+0.28], range ~[-6.1205e-22, 6.1205e-22]:
  68. * |(lgamma(x) - tf) - t(x - tc)| < 2**-70.5
  69. */
  70. static const union IEEEl2bits
  71. tcu = LD80C(0xbb16c31ab5f1fb71, 0, 1.46163214496836234128e+00L),
  72. tfu = LD80C(0xf8cdcde61c520e0f, -4, -1.21486290535849608093e-01L),
  73. ttu = LD80C(0xd46ee54b27d4de99, -69, -2.81152980996018785880e-21L),
  74. t0u = LD80C(0x80b9406556a62a6b, -68, 3.40728634996055147231e-21L),
  75. t1u = LD80C(0xc7e9c6f6df3f8c39, -67, -1.05833162742737073665e-20L),
  76. t2u = LD80C(0xf7b95e4771c55d51, -2, 4.83836122723810583532e-01L),
  77. t3u = LD80C(0x97213c6e35e119ff, -3, -1.47587722994530691476e-01L),
  78. t4u = LD80C(0x845a14a6a81dc94b, -4, 6.46249402389135358063e-02L),
  79. t5u = LD80C(0x864d46fa89997796, -5, -3.27885410884846056084e-02L),
  80. t6u = LD80C(0x93373cbd00297438, -6, 1.79706751150707171293e-02L),
  81. t7u = LD80C(0xa8fcfca7eddc8d1d, -7, -1.03142230361450732547e-02L),
  82. t8u = LD80C(0xc7e7015ff4bc45af, -8, 6.10053603296546099193e-03L),
  83. t9u = LD80C(0xf178d2247adc5093, -9, -3.68456964904901200152e-03L),
  84. t10u = LD80C(0x94188d58f12e5e9f, -9, 2.25976420273774583089e-03L),
  85. t11u = LD80C(0xb7cbaef14e1406f1, -10, -1.40224943666225639823e-03L),
  86. t12u = LD80C(0xe63a671e6704ea4d, -11, 8.78250640744776944887e-04L),
  87. t13u = LD80C(0x914b6c9cae61783e, -11, -5.54255012657716808811e-04L),
  88. t14u = LD80C(0xb858f5bdb79276fe, -12, 3.51614951536825927370e-04L),
  89. t15u = LD80C(0xea73e744c34b9591, -13, -2.23591563824520112236e-04L),
  90. t16u = LD80C(0x99aeabb0d67ba835, -13, 1.46562869351659194136e-04L),
  91. t17u = LD80C(0xd7c6938325db2024, -14, -1.02889866046435680588e-04L),
  92. t18u = LD80C(0xe24cb1e3b0474775, -15, 5.39540265505221957652e-05L);
  93. #define tc (tcu.e)
  94. #define tf (tfu.e)
  95. #define tt (ttu.e)
  96. #define t0 (t0u.e)
  97. #define t1 (t1u.e)
  98. #define t2 (t2u.e)
  99. #define t3 (t3u.e)
  100. #define t4 (t4u.e)
  101. #define t5 (t5u.e)
  102. #define t6 (t6u.e)
  103. #define t7 (t7u.e)
  104. #define t8 (t8u.e)
  105. #define t9 (t9u.e)
  106. #define t10 (t10u.e)
  107. #define t11 (t11u.e)
  108. #define t12 (t12u.e)
  109. #define t13 (t13u.e)
  110. #define t14 (t14u.e)
  111. #define t15 (t15u.e)
  112. #define t16 (t16u.e)
  113. #define t17 (t17u.e)
  114. #define t18 (t18u.e)
  115. /*
  116. * Domain y in [-0.1, 0.232], range ~[-8.1938e-22, 8.3815e-22]:
  117. * |(lgamma(1 + y) + 0.5 * y) / y - u(y) / v(y)| < 2**-71.2
  118. */
  119. static const union IEEEl2bits
  120. u0u = LD80C(0x9e233f1bed863d27, -4, -7.72156649015328606095e-02L),
  121. u1u = LD80C(0x98280ee45e4ddd3d, -1, 5.94361239198682739769e-01L),
  122. u2u = LD80C(0xe330c8ead4130733, 0, 1.77492629495841234275e+00L),
  123. u3u = LD80C(0xd4a213f1a002ec52, 0, 1.66119622514818078064e+00L),
  124. u4u = LD80C(0xa5a9ca6f5bc62163, -1, 6.47122051417476492989e-01L),
  125. u5u = LD80C(0xc980e49cd5b019e6, -4, 9.83903751718671509455e-02L),
  126. u6u = LD80C(0xff636a8bdce7025b, -9, 3.89691687802305743450e-03L),
  127. v1u = LD80C(0xbd109c533a19fbf5, 1, 2.95413883330948556544e+00L),
  128. v2u = LD80C(0xd295cbf96f31f099, 1, 3.29039286955665403176e+00L),
  129. v3u = LD80C(0xdab8bcfee40496cb, 0, 1.70876276441416471410e+00L),
  130. v4u = LD80C(0xd2f2dc3638567e9f, -2, 4.12009126299534668571e-01L),
  131. v5u = LD80C(0xa07d9b0851070f41, -5, 3.91822868305682491442e-02L),
  132. v6u = LD80C(0xe3cd8318f7adb2c4, -11, 8.68998648222144351114e-04L);
  133. #define u0 (u0u.e)
  134. #define u1 (u1u.e)
  135. #define u2 (u2u.e)
  136. #define u3 (u3u.e)
  137. #define u4 (u4u.e)
  138. #define u5 (u5u.e)
  139. #define u6 (u6u.e)
  140. #define v1 (v1u.e)
  141. #define v2 (v2u.e)
  142. #define v3 (v3u.e)
  143. #define v4 (v4u.e)
  144. #define v5 (v5u.e)
  145. #define v6 (v6u.e)
  146. /*
  147. * Domain x in (2, 3], range ~[-3.3648e-22, 3.4416e-22]:
  148. * |(lgamma(y+2) - 0.5 * y) / y - s(y)/r(y)| < 2**-72.3
  149. * with y = x - 2.
  150. */
  151. static const union IEEEl2bits
  152. s0u = LD80C(0x9e233f1bed863d27, -4, -7.72156649015328606095e-02L),
  153. s1u = LD80C(0xd3ff0dcc7fa91f94, -3, 2.07027640921219389860e-01L),
  154. s2u = LD80C(0xb2bb62782478ef31, -2, 3.49085881391362090549e-01L),
  155. s3u = LD80C(0xb49f7438c4611a74, -3, 1.76389518704213357954e-01L),
  156. s4u = LD80C(0x9a957008fa27ecf9, -5, 3.77401710862930008071e-02L),
  157. s5u = LD80C(0xda9b389a6ca7a7ac, -9, 3.33566791452943399399e-03L),
  158. s6u = LD80C(0xbc7a2263faf59c14, -14, 8.98728786745638844395e-05L),
  159. r1u = LD80C(0xbf5cff5b11477d4d, 0, 1.49502555796294337722e+00L),
  160. r2u = LD80C(0xd9aec89de08e3da6, -1, 8.50323236984473285866e-01L),
  161. r3u = LD80C(0xeab7ae5057c443f9, -3, 2.29216312078225806131e-01L),
  162. r4u = LD80C(0xf29707d9bd2b1e37, -6, 2.96130326586640089145e-02L),
  163. r5u = LD80C(0xd376c2f09736c5a3, -10, 1.61334161411590662495e-03L),
  164. r6u = LD80C(0xc985983d0cd34e3d, -16, 2.40232770710953450636e-05L),
  165. r7u = LD80C(0xe5c7a4f7fc2ef13d, -25, -5.34997929289167573510e-08L);
  166. #define s0 (s0u.e)
  167. #define s1 (s1u.e)
  168. #define s2 (s2u.e)
  169. #define s3 (s3u.e)
  170. #define s4 (s4u.e)
  171. #define s5 (s5u.e)
  172. #define s6 (s6u.e)
  173. #define r1 (r1u.e)
  174. #define r2 (r2u.e)
  175. #define r3 (r3u.e)
  176. #define r4 (r4u.e)
  177. #define r5 (r5u.e)
  178. #define r6 (r6u.e)
  179. #define r7 (r7u.e)
  180. /*
  181. * Domain z in [8, 0x1p70], range ~[-3.0235e-22, 3.0563e-22]:
  182. * |lgamma(x) - (x - 0.5) * (log(x) - 1) - w(1/x)| < 2**-71.7
  183. */
  184. static const union IEEEl2bits
  185. w0u = LD80C(0xd67f1c864beb4a69, -2, 4.18938533204672741776e-01L),
  186. w1u = LD80C(0xaaaaaaaaaaaaaaa1, -4, 8.33333333333333332678e-02L),
  187. w2u = LD80C(0xb60b60b60b5491c9, -9, -2.77777777777760927870e-03L),
  188. w3u = LD80C(0xd00d00cf58aede4c, -11, 7.93650793490637233668e-04L),
  189. w4u = LD80C(0x9c09bf626783d4a5, -11, -5.95238023926039051268e-04L),
  190. w5u = LD80C(0xdca7cadc5baa517b, -11, 8.41733700408000822962e-04L),
  191. w6u = LD80C(0xfb060e361e1ffd07, -10, -1.91515849570245136604e-03L),
  192. w7u = LD80C(0xcbd5101bb58d1f2b, -8, 6.22046743903262649294e-03L),
  193. w8u = LD80C(0xad27a668d32c821b, -6, -2.11370706734662081843e-02L);
  194. #define w0 (w0u.e)
  195. #define w1 (w1u.e)
  196. #define w2 (w2u.e)
  197. #define w3 (w3u.e)
  198. #define w4 (w4u.e)
  199. #define w5 (w5u.e)
  200. #define w6 (w6u.e)
  201. #define w7 (w7u.e)
  202. #define w8 (w8u.e)
  203. static long double
  204. sin_pil(long double x)
  205. {
  206. volatile long double vz;
  207. long double y,z;
  208. uint64_t n;
  209. uint16_t hx;
  210. y = -x;
  211. vz = y+0x1p63;
  212. z = vz-0x1p63;
  213. if (z == y)
  214. return zero;
  215. vz = y+0x1p61;
  216. EXTRACT_LDBL80_WORDS(hx,n,vz);
  217. z = vz-0x1p61;
  218. if (z > y) {
  219. z -= 0.25; /* adjust to round down */
  220. n--;
  221. }
  222. n &= 7; /* octant of y mod 2 */
  223. y = y - z + n * 0.25; /* y mod 2 */
  224. switch (n) {
  225. case 0: y = __kernel_sinl(pi*y,zero,0); break;
  226. case 1:
  227. case 2: y = __kernel_cosl(pi*(0.5-y),zero); break;
  228. case 3:
  229. case 4: y = __kernel_sinl(pi*(one-y),zero,0); break;
  230. case 5:
  231. case 6: y = -__kernel_cosl(pi*(y-1.5),zero); break;
  232. default: y = __kernel_sinl(pi*(y-2.0),zero,0); break;
  233. }
  234. return -y;
  235. }
  236. long double
  237. lgammal_r(long double x, int *signgamp)
  238. {
  239. long double nadj,p,p1,p2,p3,q,r,t,w,y,z;
  240. uint64_t lx;
  241. int i;
  242. uint16_t hx,ix;
  243. EXTRACT_LDBL80_WORDS(hx,lx,x);
  244. /* purge +-Inf and NaNs */
  245. *signgamp = 1;
  246. ix = hx&0x7fff;
  247. if(ix==0x7fff) return x*x;
  248. ENTERI();
  249. /* purge +-0 and tiny arguments */
  250. *signgamp = 1-2*(hx>>15);
  251. if(ix<0x3fff-67) { /* |x|<2**-(p+3), return -log(|x|) */
  252. if((ix|lx)==0)
  253. RETURNI(one/vzero);
  254. RETURNI(-logl(fabsl(x)));
  255. }
  256. /* purge negative integers and start evaluation for other x < 0 */
  257. if(hx&0x8000) {
  258. *signgamp = 1;
  259. if(ix>=0x3fff+63) /* |x|>=2**(p-1), must be -integer */
  260. RETURNI(one/vzero);
  261. t = sin_pil(x);
  262. if(t==zero) RETURNI(one/vzero); /* -integer */
  263. nadj = logl(pi/fabsl(t*x));
  264. if(t<zero) *signgamp = -1;
  265. x = -x;
  266. }
  267. /* purge 1 and 2 */
  268. if((ix==0x3fff || ix==0x4000) && lx==0x8000000000000000ULL) r = 0;
  269. /* for x < 2.0 */
  270. else if(ix<0x4000) {
  271. /*
  272. * XXX Supposedly, one can use the following information to replace the
  273. * XXX FP rational expressions. A similar approach is appropriate
  274. * XXX for ld128, but one (may need?) needs to consider llx, too.
  275. *
  276. * 8.9999961853027344e-01 3ffe e666600000000000
  277. * 7.3159980773925781e-01 3ffe bb4a200000000000
  278. * 2.3163998126983643e-01 3ffc ed33080000000000
  279. * 1.7316312789916992e+00 3fff dda6180000000000
  280. * 1.2316322326660156e+00 3fff 9da6200000000000
  281. */
  282. if(x<8.9999961853027344e-01) {
  283. r = -logl(x);
  284. if(x>=7.3159980773925781e-01) {y = 1-x; i= 0;}
  285. else if(x>=2.3163998126983643e-01) {y= x-(tc-1); i=1;}
  286. else {y = x; i=2;}
  287. } else {
  288. r = 0;
  289. if(x>=1.7316312789916992e+00) {y=2-x;i=0;}
  290. else if(x>=1.2316322326660156e+00) {y=x-tc;i=1;}
  291. else {y=x-1;i=2;}
  292. }
  293. switch(i) {
  294. case 0:
  295. z = y*y;
  296. p1 = a0+z*(a2+z*(a4+z*(a6+z*(a8+z*(a10+z*a12)))));
  297. p2 = z*(a1+z*(a3+z*(a5+z*(a7+z*(a9+z*(a11+z*a13))))));
  298. p = y*p1+p2;
  299. r += p-y/2; break;
  300. case 1:
  301. p = t0+y*t1+tt+y*y*(t2+y*(t3+y*(t4+y*(t5+y*(t6+y*(t7+y*(t8+
  302. y*(t9+y*(t10+y*(t11+y*(t12+y*(t13+y*(t14+y*(t15+y*(t16+
  303. y*(t17+y*t18))))))))))))))));
  304. r += tf + p; break;
  305. case 2:
  306. p1 = y*(u0+y*(u1+y*(u2+y*(u3+y*(u4+y*(u5+y*u6))))));
  307. p2 = 1+y*(v1+y*(v2+y*(v3+y*(v4+y*(v5+y*v6)))));
  308. r += p1/p2-y/2;
  309. }
  310. }
  311. /* x < 8.0 */
  312. else if(ix<0x4002) {
  313. i = x;
  314. y = x-i;
  315. p = y*(s0+y*(s1+y*(s2+y*(s3+y*(s4+y*(s5+y*s6))))));
  316. q = 1+y*(r1+y*(r2+y*(r3+y*(r4+y*(r5+y*(r6+y*r7))))));
  317. r = y/2+p/q;
  318. z = 1; /* lgamma(1+s) = log(s) + lgamma(s) */
  319. switch(i) {
  320. case 7: z *= (y+6); /* FALLTHRU */
  321. case 6: z *= (y+5); /* FALLTHRU */
  322. case 5: z *= (y+4); /* FALLTHRU */
  323. case 4: z *= (y+3); /* FALLTHRU */
  324. case 3: z *= (y+2); /* FALLTHRU */
  325. r += logl(z); break;
  326. }
  327. /* 8.0 <= x < 2**(p+3) */
  328. } else if (ix<0x3fff+67) {
  329. t = logl(x);
  330. z = one/x;
  331. y = z*z;
  332. w = w0+z*(w1+y*(w2+y*(w3+y*(w4+y*(w5+y*(w6+y*(w7+y*w8)))))));
  333. r = (x-half)*(t-one)+w;
  334. /* 2**(p+3) <= x <= inf */
  335. } else
  336. r = x*(logl(x)-1);
  337. if(hx&0x8000) r = nadj - r;
  338. RETURNI(r);
  339. }