You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

dptrfs.c 23 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821
  1. /* f2c.h -- Standard Fortran to C header file */
  2. /** barf [ba:rf] 2. "He suggested using FORTRAN, and everybody barfed."
  3. - From The Shogakukan DICTIONARY OF NEW ENGLISH (Second edition) */
  4. #ifndef F2C_INCLUDE
  5. #define F2C_INCLUDE
  6. #include <math.h>
  7. #include <stdlib.h>
  8. #include <string.h>
  9. #include <stdio.h>
  10. #include <complex.h>
  11. #ifdef complex
  12. #undef complex
  13. #endif
  14. #ifdef I
  15. #undef I
  16. #endif
  17. typedef int integer;
  18. typedef unsigned int uinteger;
  19. typedef char *address;
  20. typedef short int shortint;
  21. typedef float real;
  22. typedef double doublereal;
  23. typedef struct { real r, i; } complex;
  24. typedef struct { doublereal r, i; } doublecomplex;
  25. static inline _Complex float Cf(complex *z) {return z->r + z->i*_Complex_I;}
  26. static inline _Complex double Cd(doublecomplex *z) {return z->r + z->i*_Complex_I;}
  27. static inline _Complex float * _pCf(complex *z) {return (_Complex float*)z;}
  28. static inline _Complex double * _pCd(doublecomplex *z) {return (_Complex double*)z;}
  29. #define pCf(z) (*_pCf(z))
  30. #define pCd(z) (*_pCd(z))
  31. typedef int logical;
  32. typedef short int shortlogical;
  33. typedef char logical1;
  34. typedef char integer1;
  35. #define TRUE_ (1)
  36. #define FALSE_ (0)
  37. /* Extern is for use with -E */
  38. #ifndef Extern
  39. #define Extern extern
  40. #endif
  41. /* I/O stuff */
  42. typedef int flag;
  43. typedef int ftnlen;
  44. typedef int ftnint;
  45. /*external read, write*/
  46. typedef struct
  47. { flag cierr;
  48. ftnint ciunit;
  49. flag ciend;
  50. char *cifmt;
  51. ftnint cirec;
  52. } cilist;
  53. /*internal read, write*/
  54. typedef struct
  55. { flag icierr;
  56. char *iciunit;
  57. flag iciend;
  58. char *icifmt;
  59. ftnint icirlen;
  60. ftnint icirnum;
  61. } icilist;
  62. /*open*/
  63. typedef struct
  64. { flag oerr;
  65. ftnint ounit;
  66. char *ofnm;
  67. ftnlen ofnmlen;
  68. char *osta;
  69. char *oacc;
  70. char *ofm;
  71. ftnint orl;
  72. char *oblnk;
  73. } olist;
  74. /*close*/
  75. typedef struct
  76. { flag cerr;
  77. ftnint cunit;
  78. char *csta;
  79. } cllist;
  80. /*rewind, backspace, endfile*/
  81. typedef struct
  82. { flag aerr;
  83. ftnint aunit;
  84. } alist;
  85. /* inquire */
  86. typedef struct
  87. { flag inerr;
  88. ftnint inunit;
  89. char *infile;
  90. ftnlen infilen;
  91. ftnint *inex; /*parameters in standard's order*/
  92. ftnint *inopen;
  93. ftnint *innum;
  94. ftnint *innamed;
  95. char *inname;
  96. ftnlen innamlen;
  97. char *inacc;
  98. ftnlen inacclen;
  99. char *inseq;
  100. ftnlen inseqlen;
  101. char *indir;
  102. ftnlen indirlen;
  103. char *infmt;
  104. ftnlen infmtlen;
  105. char *inform;
  106. ftnint informlen;
  107. char *inunf;
  108. ftnlen inunflen;
  109. ftnint *inrecl;
  110. ftnint *innrec;
  111. char *inblank;
  112. ftnlen inblanklen;
  113. } inlist;
  114. #define VOID void
  115. union Multitype { /* for multiple entry points */
  116. integer1 g;
  117. shortint h;
  118. integer i;
  119. /* longint j; */
  120. real r;
  121. doublereal d;
  122. complex c;
  123. doublecomplex z;
  124. };
  125. typedef union Multitype Multitype;
  126. struct Vardesc { /* for Namelist */
  127. char *name;
  128. char *addr;
  129. ftnlen *dims;
  130. int type;
  131. };
  132. typedef struct Vardesc Vardesc;
  133. struct Namelist {
  134. char *name;
  135. Vardesc **vars;
  136. int nvars;
  137. };
  138. typedef struct Namelist Namelist;
  139. #define abs(x) ((x) >= 0 ? (x) : -(x))
  140. #define dabs(x) (fabs(x))
  141. #define f2cmin(a,b) ((a) <= (b) ? (a) : (b))
  142. #define f2cmax(a,b) ((a) >= (b) ? (a) : (b))
  143. #define dmin(a,b) (f2cmin(a,b))
  144. #define dmax(a,b) (f2cmax(a,b))
  145. #define bit_test(a,b) ((a) >> (b) & 1)
  146. #define bit_clear(a,b) ((a) & ~((uinteger)1 << (b)))
  147. #define bit_set(a,b) ((a) | ((uinteger)1 << (b)))
  148. #define abort_() { sig_die("Fortran abort routine called", 1); }
  149. #define c_abs(z) (cabsf(Cf(z)))
  150. #define c_cos(R,Z) { pCf(R)=ccos(Cf(Z)); }
  151. #define c_div(c, a, b) {pCf(c) = Cf(a)/Cf(b);}
  152. #define z_div(c, a, b) {pCd(c) = Cd(a)/Cd(b);}
  153. #define c_exp(R, Z) {pCf(R) = cexpf(Cf(Z));}
  154. #define c_log(R, Z) {pCf(R) = clogf(Cf(Z));}
  155. #define c_sin(R, Z) {pCf(R) = csinf(Cf(Z));}
  156. //#define c_sqrt(R, Z) {*(R) = csqrtf(Cf(Z));}
  157. #define c_sqrt(R, Z) {pCf(R) = csqrtf(Cf(Z));}
  158. #define d_abs(x) (fabs(*(x)))
  159. #define d_acos(x) (acos(*(x)))
  160. #define d_asin(x) (asin(*(x)))
  161. #define d_atan(x) (atan(*(x)))
  162. #define d_atn2(x, y) (atan2(*(x),*(y)))
  163. #define d_cnjg(R, Z) { pCd(R) = conj(Cd(Z)); }
  164. #define r_cnjg(R, Z) { pCf(R) = conj(Cf(Z)); }
  165. #define d_cos(x) (cos(*(x)))
  166. #define d_cosh(x) (cosh(*(x)))
  167. #define d_dim(__a, __b) ( *(__a) > *(__b) ? *(__a) - *(__b) : 0.0 )
  168. #define d_exp(x) (exp(*(x)))
  169. #define d_imag(z) (cimag(Cd(z)))
  170. #define r_imag(z) (cimag(Cf(z)))
  171. #define d_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
  172. #define r_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
  173. #define d_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
  174. #define r_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
  175. #define d_log(x) (log(*(x)))
  176. #define d_mod(x, y) (fmod(*(x), *(y)))
  177. #define u_nint(__x) ((__x)>=0 ? floor((__x) + .5) : -floor(.5 - (__x)))
  178. #define d_nint(x) u_nint(*(x))
  179. #define u_sign(__a,__b) ((__b) >= 0 ? ((__a) >= 0 ? (__a) : -(__a)) : -((__a) >= 0 ? (__a) : -(__a)))
  180. #define d_sign(a,b) u_sign(*(a),*(b))
  181. #define r_sign(a,b) u_sign(*(a),*(b))
  182. #define d_sin(x) (sin(*(x)))
  183. #define d_sinh(x) (sinh(*(x)))
  184. #define d_sqrt(x) (sqrt(*(x)))
  185. #define d_tan(x) (tan(*(x)))
  186. #define d_tanh(x) (tanh(*(x)))
  187. #define i_abs(x) abs(*(x))
  188. #define i_dnnt(x) ((integer)u_nint(*(x)))
  189. #define i_len(s, n) (n)
  190. #define i_nint(x) ((integer)u_nint(*(x)))
  191. #define i_sign(a,b) ((integer)u_sign((integer)*(a),(integer)*(b)))
  192. #define pow_dd(ap, bp) ( pow(*(ap), *(bp)))
  193. #define pow_si(B,E) spow_ui(*(B),*(E))
  194. #define pow_ri(B,E) spow_ui(*(B),*(E))
  195. #define pow_di(B,E) dpow_ui(*(B),*(E))
  196. #define pow_zi(p, a, b) {pCd(p) = zpow_ui(Cd(a), *(b));}
  197. #define pow_ci(p, a, b) {pCf(p) = cpow_ui(Cf(a), *(b));}
  198. #define pow_zz(R,A,B) {pCd(R) = cpow(Cd(A),*(B));}
  199. #define s_cat(lpp, rpp, rnp, np, llp) { ftnlen i, nc, ll; char *f__rp, *lp; ll = (llp); lp = (lpp); for(i=0; i < (int)*(np); ++i) { nc = ll; if((rnp)[i] < nc) nc = (rnp)[i]; ll -= nc; f__rp = (rpp)[i]; while(--nc >= 0) *lp++ = *(f__rp)++; } while(--ll >= 0) *lp++ = ' '; }
  200. #define s_cmp(a,b,c,d) ((integer)strncmp((a),(b),f2cmin((c),(d))))
  201. #define s_copy(A,B,C,D) { int __i,__m; for (__i=0, __m=f2cmin((C),(D)); __i<__m && (B)[__i] != 0; ++__i) (A)[__i] = (B)[__i]; }
  202. #define sig_die(s, kill) { exit(1); }
  203. #define s_stop(s, n) {exit(0);}
  204. static char junk[] = "\n@(#)LIBF77 VERSION 19990503\n";
  205. #define z_abs(z) (cabs(Cd(z)))
  206. #define z_exp(R, Z) {pCd(R) = cexp(Cd(Z));}
  207. #define z_sqrt(R, Z) {pCd(R) = csqrt(Cd(Z));}
  208. #define myexit_() break;
  209. #define mycycle() continue;
  210. #define myceiling(w) {ceil(w)}
  211. #define myhuge(w) {HUGE_VAL}
  212. //#define mymaxloc_(w,s,e,n) {if (sizeof(*(w)) == sizeof(double)) dmaxloc_((w),*(s),*(e),n); else dmaxloc_((w),*(s),*(e),n);}
  213. #define mymaxloc(w,s,e,n) {dmaxloc_(w,*(s),*(e),n)}
  214. /* procedure parameter types for -A and -C++ */
  215. #define F2C_proc_par_types 1
  216. #ifdef __cplusplus
  217. typedef logical (*L_fp)(...);
  218. #else
  219. typedef logical (*L_fp)();
  220. #endif
  221. static float spow_ui(float x, integer n) {
  222. float pow=1.0; unsigned long int u;
  223. if(n != 0) {
  224. if(n < 0) n = -n, x = 1/x;
  225. for(u = n; ; ) {
  226. if(u & 01) pow *= x;
  227. if(u >>= 1) x *= x;
  228. else break;
  229. }
  230. }
  231. return pow;
  232. }
  233. static double dpow_ui(double x, integer n) {
  234. double pow=1.0; unsigned long int u;
  235. if(n != 0) {
  236. if(n < 0) n = -n, x = 1/x;
  237. for(u = n; ; ) {
  238. if(u & 01) pow *= x;
  239. if(u >>= 1) x *= x;
  240. else break;
  241. }
  242. }
  243. return pow;
  244. }
  245. static _Complex float cpow_ui(_Complex float x, integer n) {
  246. _Complex float pow=1.0; unsigned long int u;
  247. if(n != 0) {
  248. if(n < 0) n = -n, x = 1/x;
  249. for(u = n; ; ) {
  250. if(u & 01) pow *= x;
  251. if(u >>= 1) x *= x;
  252. else break;
  253. }
  254. }
  255. return pow;
  256. }
  257. static _Complex double zpow_ui(_Complex double x, integer n) {
  258. _Complex double pow=1.0; unsigned long int u;
  259. if(n != 0) {
  260. if(n < 0) n = -n, x = 1/x;
  261. for(u = n; ; ) {
  262. if(u & 01) pow *= x;
  263. if(u >>= 1) x *= x;
  264. else break;
  265. }
  266. }
  267. return pow;
  268. }
  269. static integer pow_ii(integer x, integer n) {
  270. integer pow; unsigned long int u;
  271. if (n <= 0) {
  272. if (n == 0 || x == 1) pow = 1;
  273. else if (x != -1) pow = x == 0 ? 1/x : 0;
  274. else n = -n;
  275. }
  276. if ((n > 0) || !(n == 0 || x == 1 || x != -1)) {
  277. u = n;
  278. for(pow = 1; ; ) {
  279. if(u & 01) pow *= x;
  280. if(u >>= 1) x *= x;
  281. else break;
  282. }
  283. }
  284. return pow;
  285. }
  286. static integer dmaxloc_(double *w, integer s, integer e, integer *n)
  287. {
  288. double m; integer i, mi;
  289. for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
  290. if (w[i-1]>m) mi=i ,m=w[i-1];
  291. return mi-s+1;
  292. }
  293. static integer smaxloc_(float *w, integer s, integer e, integer *n)
  294. {
  295. float m; integer i, mi;
  296. for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
  297. if (w[i-1]>m) mi=i ,m=w[i-1];
  298. return mi-s+1;
  299. }
  300. static inline void cdotc_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
  301. integer n = *n_, incx = *incx_, incy = *incy_, i;
  302. _Complex float zdotc = 0.0;
  303. if (incx == 1 && incy == 1) {
  304. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  305. zdotc += conjf(Cf(&x[i])) * Cf(&y[i]);
  306. }
  307. } else {
  308. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  309. zdotc += conjf(Cf(&x[i*incx])) * Cf(&y[i*incy]);
  310. }
  311. }
  312. pCf(z) = zdotc;
  313. }
  314. static inline void zdotc_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
  315. integer n = *n_, incx = *incx_, incy = *incy_, i;
  316. _Complex double zdotc = 0.0;
  317. if (incx == 1 && incy == 1) {
  318. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  319. zdotc += conj(Cd(&x[i])) * Cd(&y[i]);
  320. }
  321. } else {
  322. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  323. zdotc += conj(Cd(&x[i*incx])) * Cd(&y[i*incy]);
  324. }
  325. }
  326. pCd(z) = zdotc;
  327. }
  328. static inline void cdotu_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
  329. integer n = *n_, incx = *incx_, incy = *incy_, i;
  330. _Complex float zdotc = 0.0;
  331. if (incx == 1 && incy == 1) {
  332. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  333. zdotc += Cf(&x[i]) * Cf(&y[i]);
  334. }
  335. } else {
  336. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  337. zdotc += Cf(&x[i*incx]) * Cf(&y[i*incy]);
  338. }
  339. }
  340. pCf(z) = zdotc;
  341. }
  342. static inline void zdotu_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
  343. integer n = *n_, incx = *incx_, incy = *incy_, i;
  344. _Complex double zdotc = 0.0;
  345. if (incx == 1 && incy == 1) {
  346. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  347. zdotc += Cd(&x[i]) * Cd(&y[i]);
  348. }
  349. } else {
  350. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  351. zdotc += Cd(&x[i*incx]) * Cd(&y[i*incy]);
  352. }
  353. }
  354. pCd(z) = zdotc;
  355. }
  356. #endif
  357. /* -- translated by f2c (version 20000121).
  358. You must link the resulting object file with the libraries:
  359. -lf2c -lm (in that order)
  360. */
  361. /* Table of constant values */
  362. static integer c__1 = 1;
  363. static doublereal c_b11 = 1.;
  364. /* > \brief \b DPTRFS */
  365. /* =========== DOCUMENTATION =========== */
  366. /* Online html documentation available at */
  367. /* http://www.netlib.org/lapack/explore-html/ */
  368. /* > \htmlonly */
  369. /* > Download DPTRFS + dependencies */
  370. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dptrfs.
  371. f"> */
  372. /* > [TGZ]</a> */
  373. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dptrfs.
  374. f"> */
  375. /* > [ZIP]</a> */
  376. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dptrfs.
  377. f"> */
  378. /* > [TXT]</a> */
  379. /* > \endhtmlonly */
  380. /* Definition: */
  381. /* =========== */
  382. /* SUBROUTINE DPTRFS( N, NRHS, D, E, DF, EF, B, LDB, X, LDX, FERR, */
  383. /* BERR, WORK, INFO ) */
  384. /* INTEGER INFO, LDB, LDX, N, NRHS */
  385. /* DOUBLE PRECISION B( LDB, * ), BERR( * ), D( * ), DF( * ), */
  386. /* $ E( * ), EF( * ), FERR( * ), WORK( * ), */
  387. /* $ X( LDX, * ) */
  388. /* > \par Purpose: */
  389. /* ============= */
  390. /* > */
  391. /* > \verbatim */
  392. /* > */
  393. /* > DPTRFS improves the computed solution to a system of linear */
  394. /* > equations when the coefficient matrix is symmetric positive definite */
  395. /* > and tridiagonal, and provides error bounds and backward error */
  396. /* > estimates for the solution. */
  397. /* > \endverbatim */
  398. /* Arguments: */
  399. /* ========== */
  400. /* > \param[in] N */
  401. /* > \verbatim */
  402. /* > N is INTEGER */
  403. /* > The order of the matrix A. N >= 0. */
  404. /* > \endverbatim */
  405. /* > */
  406. /* > \param[in] NRHS */
  407. /* > \verbatim */
  408. /* > NRHS is INTEGER */
  409. /* > The number of right hand sides, i.e., the number of columns */
  410. /* > of the matrix B. NRHS >= 0. */
  411. /* > \endverbatim */
  412. /* > */
  413. /* > \param[in] D */
  414. /* > \verbatim */
  415. /* > D is DOUBLE PRECISION array, dimension (N) */
  416. /* > The n diagonal elements of the tridiagonal matrix A. */
  417. /* > \endverbatim */
  418. /* > */
  419. /* > \param[in] E */
  420. /* > \verbatim */
  421. /* > E is DOUBLE PRECISION array, dimension (N-1) */
  422. /* > The (n-1) subdiagonal elements of the tridiagonal matrix A. */
  423. /* > \endverbatim */
  424. /* > */
  425. /* > \param[in] DF */
  426. /* > \verbatim */
  427. /* > DF is DOUBLE PRECISION array, dimension (N) */
  428. /* > The n diagonal elements of the diagonal matrix D from the */
  429. /* > factorization computed by DPTTRF. */
  430. /* > \endverbatim */
  431. /* > */
  432. /* > \param[in] EF */
  433. /* > \verbatim */
  434. /* > EF is DOUBLE PRECISION array, dimension (N-1) */
  435. /* > The (n-1) subdiagonal elements of the unit bidiagonal factor */
  436. /* > L from the factorization computed by DPTTRF. */
  437. /* > \endverbatim */
  438. /* > */
  439. /* > \param[in] B */
  440. /* > \verbatim */
  441. /* > B is DOUBLE PRECISION array, dimension (LDB,NRHS) */
  442. /* > The right hand side matrix B. */
  443. /* > \endverbatim */
  444. /* > */
  445. /* > \param[in] LDB */
  446. /* > \verbatim */
  447. /* > LDB is INTEGER */
  448. /* > The leading dimension of the array B. LDB >= f2cmax(1,N). */
  449. /* > \endverbatim */
  450. /* > */
  451. /* > \param[in,out] X */
  452. /* > \verbatim */
  453. /* > X is DOUBLE PRECISION array, dimension (LDX,NRHS) */
  454. /* > On entry, the solution matrix X, as computed by DPTTRS. */
  455. /* > On exit, the improved solution matrix X. */
  456. /* > \endverbatim */
  457. /* > */
  458. /* > \param[in] LDX */
  459. /* > \verbatim */
  460. /* > LDX is INTEGER */
  461. /* > The leading dimension of the array X. LDX >= f2cmax(1,N). */
  462. /* > \endverbatim */
  463. /* > */
  464. /* > \param[out] FERR */
  465. /* > \verbatim */
  466. /* > FERR is DOUBLE PRECISION array, dimension (NRHS) */
  467. /* > The forward error bound for each solution vector */
  468. /* > X(j) (the j-th column of the solution matrix X). */
  469. /* > If XTRUE is the true solution corresponding to X(j), FERR(j) */
  470. /* > is an estimated upper bound for the magnitude of the largest */
  471. /* > element in (X(j) - XTRUE) divided by the magnitude of the */
  472. /* > largest element in X(j). */
  473. /* > \endverbatim */
  474. /* > */
  475. /* > \param[out] BERR */
  476. /* > \verbatim */
  477. /* > BERR is DOUBLE PRECISION array, dimension (NRHS) */
  478. /* > The componentwise relative backward error of each solution */
  479. /* > vector X(j) (i.e., the smallest relative change in */
  480. /* > any element of A or B that makes X(j) an exact solution). */
  481. /* > \endverbatim */
  482. /* > */
  483. /* > \param[out] WORK */
  484. /* > \verbatim */
  485. /* > WORK is DOUBLE PRECISION array, dimension (2*N) */
  486. /* > \endverbatim */
  487. /* > */
  488. /* > \param[out] INFO */
  489. /* > \verbatim */
  490. /* > INFO is INTEGER */
  491. /* > = 0: successful exit */
  492. /* > < 0: if INFO = -i, the i-th argument had an illegal value */
  493. /* > \endverbatim */
  494. /* > \par Internal Parameters: */
  495. /* ========================= */
  496. /* > */
  497. /* > \verbatim */
  498. /* > ITMAX is the maximum number of steps of iterative refinement. */
  499. /* > \endverbatim */
  500. /* Authors: */
  501. /* ======== */
  502. /* > \author Univ. of Tennessee */
  503. /* > \author Univ. of California Berkeley */
  504. /* > \author Univ. of Colorado Denver */
  505. /* > \author NAG Ltd. */
  506. /* > \date December 2016 */
  507. /* > \ingroup doublePTcomputational */
  508. /* ===================================================================== */
  509. /* Subroutine */ int dptrfs_(integer *n, integer *nrhs, doublereal *d__,
  510. doublereal *e, doublereal *df, doublereal *ef, doublereal *b, integer
  511. *ldb, doublereal *x, integer *ldx, doublereal *ferr, doublereal *berr,
  512. doublereal *work, integer *info)
  513. {
  514. /* System generated locals */
  515. integer b_dim1, b_offset, x_dim1, x_offset, i__1, i__2;
  516. doublereal d__1, d__2, d__3;
  517. /* Local variables */
  518. doublereal safe1, safe2;
  519. integer i__, j;
  520. doublereal s;
  521. extern /* Subroutine */ int daxpy_(integer *, doublereal *, doublereal *,
  522. integer *, doublereal *, integer *);
  523. integer count;
  524. doublereal bi;
  525. extern doublereal dlamch_(char *);
  526. doublereal cx, dx, ex;
  527. integer ix;
  528. extern integer idamax_(integer *, doublereal *, integer *);
  529. integer nz;
  530. doublereal safmin;
  531. extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
  532. doublereal lstres;
  533. extern /* Subroutine */ int dpttrs_(integer *, integer *, doublereal *,
  534. doublereal *, doublereal *, integer *, integer *);
  535. doublereal eps;
  536. /* -- LAPACK computational routine (version 3.7.0) -- */
  537. /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
  538. /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
  539. /* December 2016 */
  540. /* ===================================================================== */
  541. /* Test the input parameters. */
  542. /* Parameter adjustments */
  543. --d__;
  544. --e;
  545. --df;
  546. --ef;
  547. b_dim1 = *ldb;
  548. b_offset = 1 + b_dim1 * 1;
  549. b -= b_offset;
  550. x_dim1 = *ldx;
  551. x_offset = 1 + x_dim1 * 1;
  552. x -= x_offset;
  553. --ferr;
  554. --berr;
  555. --work;
  556. /* Function Body */
  557. *info = 0;
  558. if (*n < 0) {
  559. *info = -1;
  560. } else if (*nrhs < 0) {
  561. *info = -2;
  562. } else if (*ldb < f2cmax(1,*n)) {
  563. *info = -8;
  564. } else if (*ldx < f2cmax(1,*n)) {
  565. *info = -10;
  566. }
  567. if (*info != 0) {
  568. i__1 = -(*info);
  569. xerbla_("DPTRFS", &i__1, (ftnlen)6);
  570. return 0;
  571. }
  572. /* Quick return if possible */
  573. if (*n == 0 || *nrhs == 0) {
  574. i__1 = *nrhs;
  575. for (j = 1; j <= i__1; ++j) {
  576. ferr[j] = 0.;
  577. berr[j] = 0.;
  578. /* L10: */
  579. }
  580. return 0;
  581. }
  582. /* NZ = maximum number of nonzero elements in each row of A, plus 1 */
  583. nz = 4;
  584. eps = dlamch_("Epsilon");
  585. safmin = dlamch_("Safe minimum");
  586. safe1 = nz * safmin;
  587. safe2 = safe1 / eps;
  588. /* Do for each right hand side */
  589. i__1 = *nrhs;
  590. for (j = 1; j <= i__1; ++j) {
  591. count = 1;
  592. lstres = 3.;
  593. L20:
  594. /* Loop until stopping criterion is satisfied. */
  595. /* Compute residual R = B - A * X. Also compute */
  596. /* abs(A)*abs(x) + abs(b) for use in the backward error bound. */
  597. if (*n == 1) {
  598. bi = b[j * b_dim1 + 1];
  599. dx = d__[1] * x[j * x_dim1 + 1];
  600. work[*n + 1] = bi - dx;
  601. work[1] = abs(bi) + abs(dx);
  602. } else {
  603. bi = b[j * b_dim1 + 1];
  604. dx = d__[1] * x[j * x_dim1 + 1];
  605. ex = e[1] * x[j * x_dim1 + 2];
  606. work[*n + 1] = bi - dx - ex;
  607. work[1] = abs(bi) + abs(dx) + abs(ex);
  608. i__2 = *n - 1;
  609. for (i__ = 2; i__ <= i__2; ++i__) {
  610. bi = b[i__ + j * b_dim1];
  611. cx = e[i__ - 1] * x[i__ - 1 + j * x_dim1];
  612. dx = d__[i__] * x[i__ + j * x_dim1];
  613. ex = e[i__] * x[i__ + 1 + j * x_dim1];
  614. work[*n + i__] = bi - cx - dx - ex;
  615. work[i__] = abs(bi) + abs(cx) + abs(dx) + abs(ex);
  616. /* L30: */
  617. }
  618. bi = b[*n + j * b_dim1];
  619. cx = e[*n - 1] * x[*n - 1 + j * x_dim1];
  620. dx = d__[*n] * x[*n + j * x_dim1];
  621. work[*n + *n] = bi - cx - dx;
  622. work[*n] = abs(bi) + abs(cx) + abs(dx);
  623. }
  624. /* Compute componentwise relative backward error from formula */
  625. /* f2cmax(i) ( abs(R(i)) / ( abs(A)*abs(X) + abs(B) )(i) ) */
  626. /* where abs(Z) is the componentwise absolute value of the matrix */
  627. /* or vector Z. If the i-th component of the denominator is less */
  628. /* than SAFE2, then SAFE1 is added to the i-th components of the */
  629. /* numerator and denominator before dividing. */
  630. s = 0.;
  631. i__2 = *n;
  632. for (i__ = 1; i__ <= i__2; ++i__) {
  633. if (work[i__] > safe2) {
  634. /* Computing MAX */
  635. d__2 = s, d__3 = (d__1 = work[*n + i__], abs(d__1)) / work[
  636. i__];
  637. s = f2cmax(d__2,d__3);
  638. } else {
  639. /* Computing MAX */
  640. d__2 = s, d__3 = ((d__1 = work[*n + i__], abs(d__1)) + safe1)
  641. / (work[i__] + safe1);
  642. s = f2cmax(d__2,d__3);
  643. }
  644. /* L40: */
  645. }
  646. berr[j] = s;
  647. /* Test stopping criterion. Continue iterating if */
  648. /* 1) The residual BERR(J) is larger than machine epsilon, and */
  649. /* 2) BERR(J) decreased by at least a factor of 2 during the */
  650. /* last iteration, and */
  651. /* 3) At most ITMAX iterations tried. */
  652. if (berr[j] > eps && berr[j] * 2. <= lstres && count <= 5) {
  653. /* Update solution and try again. */
  654. dpttrs_(n, &c__1, &df[1], &ef[1], &work[*n + 1], n, info);
  655. daxpy_(n, &c_b11, &work[*n + 1], &c__1, &x[j * x_dim1 + 1], &c__1)
  656. ;
  657. lstres = berr[j];
  658. ++count;
  659. goto L20;
  660. }
  661. /* Bound error from formula */
  662. /* norm(X - XTRUE) / norm(X) .le. FERR = */
  663. /* norm( abs(inv(A))* */
  664. /* ( abs(R) + NZ*EPS*( abs(A)*abs(X)+abs(B) ))) / norm(X) */
  665. /* where */
  666. /* norm(Z) is the magnitude of the largest component of Z */
  667. /* inv(A) is the inverse of A */
  668. /* abs(Z) is the componentwise absolute value of the matrix or */
  669. /* vector Z */
  670. /* NZ is the maximum number of nonzeros in any row of A, plus 1 */
  671. /* EPS is machine epsilon */
  672. /* The i-th component of abs(R)+NZ*EPS*(abs(A)*abs(X)+abs(B)) */
  673. /* is incremented by SAFE1 if the i-th component of */
  674. /* abs(A)*abs(X) + abs(B) is less than SAFE2. */
  675. i__2 = *n;
  676. for (i__ = 1; i__ <= i__2; ++i__) {
  677. if (work[i__] > safe2) {
  678. work[i__] = (d__1 = work[*n + i__], abs(d__1)) + nz * eps *
  679. work[i__];
  680. } else {
  681. work[i__] = (d__1 = work[*n + i__], abs(d__1)) + nz * eps *
  682. work[i__] + safe1;
  683. }
  684. /* L50: */
  685. }
  686. ix = idamax_(n, &work[1], &c__1);
  687. ferr[j] = work[ix];
  688. /* Estimate the norm of inv(A). */
  689. /* Solve M(A) * x = e, where M(A) = (m(i,j)) is given by */
  690. /* m(i,j) = abs(A(i,j)), i = j, */
  691. /* m(i,j) = -abs(A(i,j)), i .ne. j, */
  692. /* and e = [ 1, 1, ..., 1 ]**T. Note M(A) = M(L)*D*M(L)**T. */
  693. /* Solve M(L) * x = e. */
  694. work[1] = 1.;
  695. i__2 = *n;
  696. for (i__ = 2; i__ <= i__2; ++i__) {
  697. work[i__] = work[i__ - 1] * (d__1 = ef[i__ - 1], abs(d__1)) + 1.;
  698. /* L60: */
  699. }
  700. /* Solve D * M(L)**T * x = b. */
  701. work[*n] /= df[*n];
  702. for (i__ = *n - 1; i__ >= 1; --i__) {
  703. work[i__] = work[i__] / df[i__] + work[i__ + 1] * (d__1 = ef[i__],
  704. abs(d__1));
  705. /* L70: */
  706. }
  707. /* Compute norm(inv(A)) = f2cmax(x(i)), 1<=i<=n. */
  708. ix = idamax_(n, &work[1], &c__1);
  709. ferr[j] *= (d__1 = work[ix], abs(d__1));
  710. /* Normalize error. */
  711. lstres = 0.;
  712. i__2 = *n;
  713. for (i__ = 1; i__ <= i__2; ++i__) {
  714. /* Computing MAX */
  715. d__2 = lstres, d__3 = (d__1 = x[i__ + j * x_dim1], abs(d__1));
  716. lstres = f2cmax(d__2,d__3);
  717. /* L80: */
  718. }
  719. if (lstres != 0.) {
  720. ferr[j] /= lstres;
  721. }
  722. /* L90: */
  723. }
  724. return 0;
  725. /* End of DPTRFS */
  726. } /* dptrfs_ */