sf_erf.c 7.3 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257
  1. /*
  2. * This file is part of the MicroPython project, http://micropython.org/
  3. *
  4. * These math functions are taken from newlib-nano-2, the newlib/libm/math
  5. * directory, available from https://github.com/32bitmicro/newlib-nano-2.
  6. *
  7. * Appropriate copyright headers are reproduced below.
  8. */
  9. /* sf_erf.c -- float version of s_erf.c.
  10. * Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
  11. */
  12. /*
  13. * ====================================================
  14. * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
  15. *
  16. * Developed at SunPro, a Sun Microsystems, Inc. business.
  17. * Permission to use, copy, modify, and distribute this
  18. * software is freely granted, provided that this notice
  19. * is preserved.
  20. * ====================================================
  21. */
  22. #include "fdlibm.h"
  23. #define __ieee754_expf expf
  24. #ifdef __v810__
  25. #define const
  26. #endif
  27. #ifdef __STDC__
  28. static const float
  29. #else
  30. static float
  31. #endif
  32. tiny = 1e-30,
  33. half= 5.0000000000e-01, /* 0x3F000000 */
  34. one = 1.0000000000e+00, /* 0x3F800000 */
  35. two = 2.0000000000e+00, /* 0x40000000 */
  36. /* c = (subfloat)0.84506291151 */
  37. erx = 8.4506291151e-01, /* 0x3f58560b */
  38. /*
  39. * Coefficients for approximation to erf on [0,0.84375]
  40. */
  41. efx = 1.2837916613e-01, /* 0x3e0375d4 */
  42. efx8= 1.0270333290e+00, /* 0x3f8375d4 */
  43. pp0 = 1.2837916613e-01, /* 0x3e0375d4 */
  44. pp1 = -3.2504209876e-01, /* 0xbea66beb */
  45. pp2 = -2.8481749818e-02, /* 0xbce9528f */
  46. pp3 = -5.7702702470e-03, /* 0xbbbd1489 */
  47. pp4 = -2.3763017452e-05, /* 0xb7c756b1 */
  48. qq1 = 3.9791721106e-01, /* 0x3ecbbbce */
  49. qq2 = 6.5022252500e-02, /* 0x3d852a63 */
  50. qq3 = 5.0813062117e-03, /* 0x3ba68116 */
  51. qq4 = 1.3249473704e-04, /* 0x390aee49 */
  52. qq5 = -3.9602282413e-06, /* 0xb684e21a */
  53. /*
  54. * Coefficients for approximation to erf in [0.84375,1.25]
  55. */
  56. pa0 = -2.3621185683e-03, /* 0xbb1acdc6 */
  57. pa1 = 4.1485610604e-01, /* 0x3ed46805 */
  58. pa2 = -3.7220788002e-01, /* 0xbebe9208 */
  59. pa3 = 3.1834661961e-01, /* 0x3ea2fe54 */
  60. pa4 = -1.1089469492e-01, /* 0xbde31cc2 */
  61. pa5 = 3.5478305072e-02, /* 0x3d1151b3 */
  62. pa6 = -2.1663755178e-03, /* 0xbb0df9c0 */
  63. qa1 = 1.0642088205e-01, /* 0x3dd9f331 */
  64. qa2 = 5.4039794207e-01, /* 0x3f0a5785 */
  65. qa3 = 7.1828655899e-02, /* 0x3d931ae7 */
  66. qa4 = 1.2617121637e-01, /* 0x3e013307 */
  67. qa5 = 1.3637083583e-02, /* 0x3c5f6e13 */
  68. qa6 = 1.1984500103e-02, /* 0x3c445aa3 */
  69. /*
  70. * Coefficients for approximation to erfc in [1.25,1/0.35]
  71. */
  72. ra0 = -9.8649440333e-03, /* 0xbc21a093 */
  73. ra1 = -6.9385856390e-01, /* 0xbf31a0b7 */
  74. ra2 = -1.0558626175e+01, /* 0xc128f022 */
  75. ra3 = -6.2375331879e+01, /* 0xc2798057 */
  76. ra4 = -1.6239666748e+02, /* 0xc322658c */
  77. ra5 = -1.8460508728e+02, /* 0xc3389ae7 */
  78. ra6 = -8.1287437439e+01, /* 0xc2a2932b */
  79. ra7 = -9.8143291473e+00, /* 0xc11d077e */
  80. sa1 = 1.9651271820e+01, /* 0x419d35ce */
  81. sa2 = 1.3765776062e+02, /* 0x4309a863 */
  82. sa3 = 4.3456588745e+02, /* 0x43d9486f */
  83. sa4 = 6.4538726807e+02, /* 0x442158c9 */
  84. sa5 = 4.2900814819e+02, /* 0x43d6810b */
  85. sa6 = 1.0863500214e+02, /* 0x42d9451f */
  86. sa7 = 6.5702495575e+00, /* 0x40d23f7c */
  87. sa8 = -6.0424413532e-02, /* 0xbd777f97 */
  88. /*
  89. * Coefficients for approximation to erfc in [1/.35,28]
  90. */
  91. rb0 = -9.8649431020e-03, /* 0xbc21a092 */
  92. rb1 = -7.9928326607e-01, /* 0xbf4c9dd4 */
  93. rb2 = -1.7757955551e+01, /* 0xc18e104b */
  94. rb3 = -1.6063638306e+02, /* 0xc320a2ea */
  95. rb4 = -6.3756646729e+02, /* 0xc41f6441 */
  96. rb5 = -1.0250950928e+03, /* 0xc480230b */
  97. rb6 = -4.8351919556e+02, /* 0xc3f1c275 */
  98. sb1 = 3.0338060379e+01, /* 0x41f2b459 */
  99. sb2 = 3.2579251099e+02, /* 0x43a2e571 */
  100. sb3 = 1.5367296143e+03, /* 0x44c01759 */
  101. sb4 = 3.1998581543e+03, /* 0x4547fdbb */
  102. sb5 = 2.5530502930e+03, /* 0x451f90ce */
  103. sb6 = 4.7452853394e+02, /* 0x43ed43a7 */
  104. sb7 = -2.2440952301e+01; /* 0xc1b38712 */
  105. #ifdef __STDC__
  106. float erff(float x)
  107. #else
  108. float erff(x)
  109. float x;
  110. #endif
  111. {
  112. __int32_t hx,ix,i;
  113. float R,S,P,Q,s,y,z,r;
  114. GET_FLOAT_WORD(hx,x);
  115. ix = hx&0x7fffffff;
  116. if(!FLT_UWORD_IS_FINITE(ix)) { /* erf(nan)=nan */
  117. i = ((__uint32_t)hx>>31)<<1;
  118. return (float)(1-i)+one/x; /* erf(+-inf)=+-1 */
  119. }
  120. if(ix < 0x3f580000) { /* |x|<0.84375 */
  121. if(ix < 0x31800000) { /* |x|<2**-28 */
  122. if (ix < 0x04000000)
  123. /*avoid underflow */
  124. return (float)0.125*((float)8.0*x+efx8*x);
  125. return x + efx*x;
  126. }
  127. z = x*x;
  128. r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4)));
  129. s = one+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5))));
  130. y = r/s;
  131. return x + x*y;
  132. }
  133. if(ix < 0x3fa00000) { /* 0.84375 <= |x| < 1.25 */
  134. s = fabsf(x)-one;
  135. P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6)))));
  136. Q = one+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6)))));
  137. if(hx>=0) return erx + P/Q; else return -erx - P/Q;
  138. }
  139. if (ix >= 0x40c00000) { /* inf>|x|>=6 */
  140. if(hx>=0) return one-tiny; else return tiny-one;
  141. }
  142. x = fabsf(x);
  143. s = one/(x*x);
  144. if(ix< 0x4036DB6E) { /* |x| < 1/0.35 */
  145. R=ra0+s*(ra1+s*(ra2+s*(ra3+s*(ra4+s*(
  146. ra5+s*(ra6+s*ra7))))));
  147. S=one+s*(sa1+s*(sa2+s*(sa3+s*(sa4+s*(
  148. sa5+s*(sa6+s*(sa7+s*sa8)))))));
  149. } else { /* |x| >= 1/0.35 */
  150. R=rb0+s*(rb1+s*(rb2+s*(rb3+s*(rb4+s*(
  151. rb5+s*rb6)))));
  152. S=one+s*(sb1+s*(sb2+s*(sb3+s*(sb4+s*(
  153. sb5+s*(sb6+s*sb7))))));
  154. }
  155. GET_FLOAT_WORD(ix,x);
  156. SET_FLOAT_WORD(z,ix&0xfffff000);
  157. r = __ieee754_expf(-z*z-(float)0.5625)*__ieee754_expf((z-x)*(z+x)+R/S);
  158. if(hx>=0) return one-r/x; else return r/x-one;
  159. }
  160. #ifdef __STDC__
  161. float erfcf(float x)
  162. #else
  163. float erfcf(x)
  164. float x;
  165. #endif
  166. {
  167. __int32_t hx,ix;
  168. float R,S,P,Q,s,y,z,r;
  169. GET_FLOAT_WORD(hx,x);
  170. ix = hx&0x7fffffff;
  171. if(!FLT_UWORD_IS_FINITE(ix)) { /* erfc(nan)=nan */
  172. /* erfc(+-inf)=0,2 */
  173. return (float)(((__uint32_t)hx>>31)<<1)+one/x;
  174. }
  175. if(ix < 0x3f580000) { /* |x|<0.84375 */
  176. if(ix < 0x23800000) /* |x|<2**-56 */
  177. return one-x;
  178. z = x*x;
  179. r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4)));
  180. s = one+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5))));
  181. y = r/s;
  182. if(hx < 0x3e800000) { /* x<1/4 */
  183. return one-(x+x*y);
  184. } else {
  185. r = x*y;
  186. r += (x-half);
  187. return half - r ;
  188. }
  189. }
  190. if(ix < 0x3fa00000) { /* 0.84375 <= |x| < 1.25 */
  191. s = fabsf(x)-one;
  192. P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6)))));
  193. Q = one+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6)))));
  194. if(hx>=0) {
  195. z = one-erx; return z - P/Q;
  196. } else {
  197. z = erx+P/Q; return one+z;
  198. }
  199. }
  200. if (ix < 0x41e00000) { /* |x|<28 */
  201. x = fabsf(x);
  202. s = one/(x*x);
  203. if(ix< 0x4036DB6D) { /* |x| < 1/.35 ~ 2.857143*/
  204. R=ra0+s*(ra1+s*(ra2+s*(ra3+s*(ra4+s*(
  205. ra5+s*(ra6+s*ra7))))));
  206. S=one+s*(sa1+s*(sa2+s*(sa3+s*(sa4+s*(
  207. sa5+s*(sa6+s*(sa7+s*sa8)))))));
  208. } else { /* |x| >= 1/.35 ~ 2.857143 */
  209. if(hx<0&&ix>=0x40c00000) return two-tiny;/* x < -6 */
  210. R=rb0+s*(rb1+s*(rb2+s*(rb3+s*(rb4+s*(
  211. rb5+s*rb6)))));
  212. S=one+s*(sb1+s*(sb2+s*(sb3+s*(sb4+s*(
  213. sb5+s*(sb6+s*sb7))))));
  214. }
  215. GET_FLOAT_WORD(ix,x);
  216. SET_FLOAT_WORD(z,ix&0xfffff000);
  217. r = __ieee754_expf(-z*z-(float)0.5625)*
  218. __ieee754_expf((z-x)*(z+x)+R/S);
  219. if(hx>0) return r/x; else return two-r/x;
  220. } else {
  221. if(hx>0) return tiny*tiny; else return two-tiny;
  222. }
  223. }
  224. #ifdef _DOUBLE_IS_32BITS
  225. #ifdef __STDC__
  226. double erf(double x)
  227. #else
  228. double erf(x)
  229. double x;
  230. #endif
  231. {
  232. return (double) erff((float) x);
  233. }
  234. #ifdef __STDC__
  235. double erfc(double x)
  236. #else
  237. double erfc(x)
  238. double x;
  239. #endif
  240. {
  241. return (double) erfcf((float) x);
  242. }
  243. #endif /* defined(_DOUBLE_IS_32BITS) */