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.

cgbrfs.c 28 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971
  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. #if defined(_WIN64)
  18. typedef long long BLASLONG;
  19. typedef unsigned long long BLASULONG;
  20. #else
  21. typedef long BLASLONG;
  22. typedef unsigned long BLASULONG;
  23. #endif
  24. #ifdef LAPACK_ILP64
  25. typedef BLASLONG blasint;
  26. #if defined(_WIN64)
  27. #define blasabs(x) llabs(x)
  28. #else
  29. #define blasabs(x) labs(x)
  30. #endif
  31. #else
  32. typedef int blasint;
  33. #define blasabs(x) abs(x)
  34. #endif
  35. typedef blasint integer;
  36. typedef unsigned int uinteger;
  37. typedef char *address;
  38. typedef short int shortint;
  39. typedef float real;
  40. typedef double doublereal;
  41. typedef struct { real r, i; } complex;
  42. typedef struct { doublereal r, i; } doublecomplex;
  43. static inline _Complex float Cf(complex *z) {return z->r + z->i*_Complex_I;}
  44. static inline _Complex double Cd(doublecomplex *z) {return z->r + z->i*_Complex_I;}
  45. static inline _Complex float * _pCf(complex *z) {return (_Complex float*)z;}
  46. static inline _Complex double * _pCd(doublecomplex *z) {return (_Complex double*)z;}
  47. #define pCf(z) (*_pCf(z))
  48. #define pCd(z) (*_pCd(z))
  49. typedef int logical;
  50. typedef short int shortlogical;
  51. typedef char logical1;
  52. typedef char integer1;
  53. #define TRUE_ (1)
  54. #define FALSE_ (0)
  55. /* Extern is for use with -E */
  56. #ifndef Extern
  57. #define Extern extern
  58. #endif
  59. /* I/O stuff */
  60. typedef int flag;
  61. typedef int ftnlen;
  62. typedef int ftnint;
  63. /*external read, write*/
  64. typedef struct
  65. { flag cierr;
  66. ftnint ciunit;
  67. flag ciend;
  68. char *cifmt;
  69. ftnint cirec;
  70. } cilist;
  71. /*internal read, write*/
  72. typedef struct
  73. { flag icierr;
  74. char *iciunit;
  75. flag iciend;
  76. char *icifmt;
  77. ftnint icirlen;
  78. ftnint icirnum;
  79. } icilist;
  80. /*open*/
  81. typedef struct
  82. { flag oerr;
  83. ftnint ounit;
  84. char *ofnm;
  85. ftnlen ofnmlen;
  86. char *osta;
  87. char *oacc;
  88. char *ofm;
  89. ftnint orl;
  90. char *oblnk;
  91. } olist;
  92. /*close*/
  93. typedef struct
  94. { flag cerr;
  95. ftnint cunit;
  96. char *csta;
  97. } cllist;
  98. /*rewind, backspace, endfile*/
  99. typedef struct
  100. { flag aerr;
  101. ftnint aunit;
  102. } alist;
  103. /* inquire */
  104. typedef struct
  105. { flag inerr;
  106. ftnint inunit;
  107. char *infile;
  108. ftnlen infilen;
  109. ftnint *inex; /*parameters in standard's order*/
  110. ftnint *inopen;
  111. ftnint *innum;
  112. ftnint *innamed;
  113. char *inname;
  114. ftnlen innamlen;
  115. char *inacc;
  116. ftnlen inacclen;
  117. char *inseq;
  118. ftnlen inseqlen;
  119. char *indir;
  120. ftnlen indirlen;
  121. char *infmt;
  122. ftnlen infmtlen;
  123. char *inform;
  124. ftnint informlen;
  125. char *inunf;
  126. ftnlen inunflen;
  127. ftnint *inrecl;
  128. ftnint *innrec;
  129. char *inblank;
  130. ftnlen inblanklen;
  131. } inlist;
  132. #define VOID void
  133. union Multitype { /* for multiple entry points */
  134. integer1 g;
  135. shortint h;
  136. integer i;
  137. /* longint j; */
  138. real r;
  139. doublereal d;
  140. complex c;
  141. doublecomplex z;
  142. };
  143. typedef union Multitype Multitype;
  144. struct Vardesc { /* for Namelist */
  145. char *name;
  146. char *addr;
  147. ftnlen *dims;
  148. int type;
  149. };
  150. typedef struct Vardesc Vardesc;
  151. struct Namelist {
  152. char *name;
  153. Vardesc **vars;
  154. int nvars;
  155. };
  156. typedef struct Namelist Namelist;
  157. #define abs(x) ((x) >= 0 ? (x) : -(x))
  158. #define dabs(x) (fabs(x))
  159. #define f2cmin(a,b) ((a) <= (b) ? (a) : (b))
  160. #define f2cmax(a,b) ((a) >= (b) ? (a) : (b))
  161. #define dmin(a,b) (f2cmin(a,b))
  162. #define dmax(a,b) (f2cmax(a,b))
  163. #define bit_test(a,b) ((a) >> (b) & 1)
  164. #define bit_clear(a,b) ((a) & ~((uinteger)1 << (b)))
  165. #define bit_set(a,b) ((a) | ((uinteger)1 << (b)))
  166. #define abort_() { sig_die("Fortran abort routine called", 1); }
  167. #define c_abs(z) (cabsf(Cf(z)))
  168. #define c_cos(R,Z) { pCf(R)=ccos(Cf(Z)); }
  169. #define c_div(c, a, b) {pCf(c) = Cf(a)/Cf(b);}
  170. #define z_div(c, a, b) {pCd(c) = Cd(a)/Cd(b);}
  171. #define c_exp(R, Z) {pCf(R) = cexpf(Cf(Z));}
  172. #define c_log(R, Z) {pCf(R) = clogf(Cf(Z));}
  173. #define c_sin(R, Z) {pCf(R) = csinf(Cf(Z));}
  174. //#define c_sqrt(R, Z) {*(R) = csqrtf(Cf(Z));}
  175. #define c_sqrt(R, Z) {pCf(R) = csqrtf(Cf(Z));}
  176. #define d_abs(x) (fabs(*(x)))
  177. #define d_acos(x) (acos(*(x)))
  178. #define d_asin(x) (asin(*(x)))
  179. #define d_atan(x) (atan(*(x)))
  180. #define d_atn2(x, y) (atan2(*(x),*(y)))
  181. #define d_cnjg(R, Z) { pCd(R) = conj(Cd(Z)); }
  182. #define r_cnjg(R, Z) { pCf(R) = conj(Cf(Z)); }
  183. #define d_cos(x) (cos(*(x)))
  184. #define d_cosh(x) (cosh(*(x)))
  185. #define d_dim(__a, __b) ( *(__a) > *(__b) ? *(__a) - *(__b) : 0.0 )
  186. #define d_exp(x) (exp(*(x)))
  187. #define d_imag(z) (cimag(Cd(z)))
  188. #define r_imag(z) (cimag(Cf(z)))
  189. #define d_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
  190. #define r_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
  191. #define d_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
  192. #define r_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
  193. #define d_log(x) (log(*(x)))
  194. #define d_mod(x, y) (fmod(*(x), *(y)))
  195. #define u_nint(__x) ((__x)>=0 ? floor((__x) + .5) : -floor(.5 - (__x)))
  196. #define d_nint(x) u_nint(*(x))
  197. #define u_sign(__a,__b) ((__b) >= 0 ? ((__a) >= 0 ? (__a) : -(__a)) : -((__a) >= 0 ? (__a) : -(__a)))
  198. #define d_sign(a,b) u_sign(*(a),*(b))
  199. #define r_sign(a,b) u_sign(*(a),*(b))
  200. #define d_sin(x) (sin(*(x)))
  201. #define d_sinh(x) (sinh(*(x)))
  202. #define d_sqrt(x) (sqrt(*(x)))
  203. #define d_tan(x) (tan(*(x)))
  204. #define d_tanh(x) (tanh(*(x)))
  205. #define i_abs(x) abs(*(x))
  206. #define i_dnnt(x) ((integer)u_nint(*(x)))
  207. #define i_len(s, n) (n)
  208. #define i_nint(x) ((integer)u_nint(*(x)))
  209. #define i_sign(a,b) ((integer)u_sign((integer)*(a),(integer)*(b)))
  210. #define pow_dd(ap, bp) ( pow(*(ap), *(bp)))
  211. #define pow_si(B,E) spow_ui(*(B),*(E))
  212. #define pow_ri(B,E) spow_ui(*(B),*(E))
  213. #define pow_di(B,E) dpow_ui(*(B),*(E))
  214. #define pow_zi(p, a, b) {pCd(p) = zpow_ui(Cd(a), *(b));}
  215. #define pow_ci(p, a, b) {pCf(p) = cpow_ui(Cf(a), *(b));}
  216. #define pow_zz(R,A,B) {pCd(R) = cpow(Cd(A),*(B));}
  217. #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++ = ' '; }
  218. #define s_cmp(a,b,c,d) ((integer)strncmp((a),(b),f2cmin((c),(d))))
  219. #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]; }
  220. #define sig_die(s, kill) { exit(1); }
  221. #define s_stop(s, n) {exit(0);}
  222. static char junk[] = "\n@(#)LIBF77 VERSION 19990503\n";
  223. #define z_abs(z) (cabs(Cd(z)))
  224. #define z_exp(R, Z) {pCd(R) = cexp(Cd(Z));}
  225. #define z_sqrt(R, Z) {pCd(R) = csqrt(Cd(Z));}
  226. #define myexit_() break;
  227. #define mycycle() continue;
  228. #define myceiling(w) {ceil(w)}
  229. #define myhuge(w) {HUGE_VAL}
  230. //#define mymaxloc_(w,s,e,n) {if (sizeof(*(w)) == sizeof(double)) dmaxloc_((w),*(s),*(e),n); else dmaxloc_((w),*(s),*(e),n);}
  231. #define mymaxloc(w,s,e,n) {dmaxloc_(w,*(s),*(e),n)}
  232. /* procedure parameter types for -A and -C++ */
  233. #define F2C_proc_par_types 1
  234. #ifdef __cplusplus
  235. typedef logical (*L_fp)(...);
  236. #else
  237. typedef logical (*L_fp)();
  238. #endif
  239. static float spow_ui(float x, integer n) {
  240. float pow=1.0; unsigned long int u;
  241. if(n != 0) {
  242. if(n < 0) n = -n, x = 1/x;
  243. for(u = n; ; ) {
  244. if(u & 01) pow *= x;
  245. if(u >>= 1) x *= x;
  246. else break;
  247. }
  248. }
  249. return pow;
  250. }
  251. static double dpow_ui(double x, integer n) {
  252. double pow=1.0; unsigned long int u;
  253. if(n != 0) {
  254. if(n < 0) n = -n, x = 1/x;
  255. for(u = n; ; ) {
  256. if(u & 01) pow *= x;
  257. if(u >>= 1) x *= x;
  258. else break;
  259. }
  260. }
  261. return pow;
  262. }
  263. static _Complex float cpow_ui(_Complex float x, integer n) {
  264. _Complex float pow=1.0; unsigned long int u;
  265. if(n != 0) {
  266. if(n < 0) n = -n, x = 1/x;
  267. for(u = n; ; ) {
  268. if(u & 01) pow *= x;
  269. if(u >>= 1) x *= x;
  270. else break;
  271. }
  272. }
  273. return pow;
  274. }
  275. static _Complex double zpow_ui(_Complex double x, integer n) {
  276. _Complex double pow=1.0; unsigned long int u;
  277. if(n != 0) {
  278. if(n < 0) n = -n, x = 1/x;
  279. for(u = n; ; ) {
  280. if(u & 01) pow *= x;
  281. if(u >>= 1) x *= x;
  282. else break;
  283. }
  284. }
  285. return pow;
  286. }
  287. static integer pow_ii(integer x, integer n) {
  288. integer pow; unsigned long int u;
  289. if (n <= 0) {
  290. if (n == 0 || x == 1) pow = 1;
  291. else if (x != -1) pow = x == 0 ? 1/x : 0;
  292. else n = -n;
  293. }
  294. if ((n > 0) || !(n == 0 || x == 1 || x != -1)) {
  295. u = n;
  296. for(pow = 1; ; ) {
  297. if(u & 01) pow *= x;
  298. if(u >>= 1) x *= x;
  299. else break;
  300. }
  301. }
  302. return pow;
  303. }
  304. static integer dmaxloc_(double *w, integer s, integer e, integer *n)
  305. {
  306. double m; integer i, mi;
  307. for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
  308. if (w[i-1]>m) mi=i ,m=w[i-1];
  309. return mi-s+1;
  310. }
  311. static integer smaxloc_(float *w, integer s, integer e, integer *n)
  312. {
  313. float m; integer i, mi;
  314. for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
  315. if (w[i-1]>m) mi=i ,m=w[i-1];
  316. return mi-s+1;
  317. }
  318. static inline void cdotc_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
  319. integer n = *n_, incx = *incx_, incy = *incy_, i;
  320. _Complex float zdotc = 0.0;
  321. if (incx == 1 && incy == 1) {
  322. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  323. zdotc += conjf(Cf(&x[i])) * Cf(&y[i]);
  324. }
  325. } else {
  326. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  327. zdotc += conjf(Cf(&x[i*incx])) * Cf(&y[i*incy]);
  328. }
  329. }
  330. pCf(z) = zdotc;
  331. }
  332. static inline void zdotc_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
  333. integer n = *n_, incx = *incx_, incy = *incy_, i;
  334. _Complex double zdotc = 0.0;
  335. if (incx == 1 && incy == 1) {
  336. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  337. zdotc += conj(Cd(&x[i])) * Cd(&y[i]);
  338. }
  339. } else {
  340. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  341. zdotc += conj(Cd(&x[i*incx])) * Cd(&y[i*incy]);
  342. }
  343. }
  344. pCd(z) = zdotc;
  345. }
  346. static inline void cdotu_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
  347. integer n = *n_, incx = *incx_, incy = *incy_, i;
  348. _Complex float zdotc = 0.0;
  349. if (incx == 1 && incy == 1) {
  350. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  351. zdotc += Cf(&x[i]) * Cf(&y[i]);
  352. }
  353. } else {
  354. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  355. zdotc += Cf(&x[i*incx]) * Cf(&y[i*incy]);
  356. }
  357. }
  358. pCf(z) = zdotc;
  359. }
  360. static inline void zdotu_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
  361. integer n = *n_, incx = *incx_, incy = *incy_, i;
  362. _Complex double zdotc = 0.0;
  363. if (incx == 1 && incy == 1) {
  364. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  365. zdotc += Cd(&x[i]) * Cd(&y[i]);
  366. }
  367. } else {
  368. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  369. zdotc += Cd(&x[i*incx]) * Cd(&y[i*incy]);
  370. }
  371. }
  372. pCd(z) = zdotc;
  373. }
  374. #endif
  375. /* -- translated by f2c (version 20000121).
  376. You must link the resulting object file with the libraries:
  377. -lf2c -lm (in that order)
  378. */
  379. /* Table of constant values */
  380. static complex c_b1 = {1.f,0.f};
  381. static integer c__1 = 1;
  382. /* > \brief \b CGBRFS */
  383. /* =========== DOCUMENTATION =========== */
  384. /* Online html documentation available at */
  385. /* http://www.netlib.org/lapack/explore-html/ */
  386. /* > \htmlonly */
  387. /* > Download CGBRFS + dependencies */
  388. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/cgbrfs.
  389. f"> */
  390. /* > [TGZ]</a> */
  391. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/cgbrfs.
  392. f"> */
  393. /* > [ZIP]</a> */
  394. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/cgbrfs.
  395. f"> */
  396. /* > [TXT]</a> */
  397. /* > \endhtmlonly */
  398. /* Definition: */
  399. /* =========== */
  400. /* SUBROUTINE CGBRFS( TRANS, N, KL, KU, NRHS, AB, LDAB, AFB, LDAFB, */
  401. /* IPIV, B, LDB, X, LDX, FERR, BERR, WORK, RWORK, */
  402. /* INFO ) */
  403. /* CHARACTER TRANS */
  404. /* INTEGER INFO, KL, KU, LDAB, LDAFB, LDB, LDX, N, NRHS */
  405. /* INTEGER IPIV( * ) */
  406. /* REAL BERR( * ), FERR( * ), RWORK( * ) */
  407. /* COMPLEX AB( LDAB, * ), AFB( LDAFB, * ), B( LDB, * ), */
  408. /* $ WORK( * ), X( LDX, * ) */
  409. /* > \par Purpose: */
  410. /* ============= */
  411. /* > */
  412. /* > \verbatim */
  413. /* > */
  414. /* > CGBRFS improves the computed solution to a system of linear */
  415. /* > equations when the coefficient matrix is banded, and provides */
  416. /* > error bounds and backward error estimates for the solution. */
  417. /* > \endverbatim */
  418. /* Arguments: */
  419. /* ========== */
  420. /* > \param[in] TRANS */
  421. /* > \verbatim */
  422. /* > TRANS is CHARACTER*1 */
  423. /* > Specifies the form of the system of equations: */
  424. /* > = 'N': A * X = B (No transpose) */
  425. /* > = 'T': A**T * X = B (Transpose) */
  426. /* > = 'C': A**H * X = B (Conjugate transpose) */
  427. /* > \endverbatim */
  428. /* > */
  429. /* > \param[in] N */
  430. /* > \verbatim */
  431. /* > N is INTEGER */
  432. /* > The order of the matrix A. N >= 0. */
  433. /* > \endverbatim */
  434. /* > */
  435. /* > \param[in] KL */
  436. /* > \verbatim */
  437. /* > KL is INTEGER */
  438. /* > The number of subdiagonals within the band of A. KL >= 0. */
  439. /* > \endverbatim */
  440. /* > */
  441. /* > \param[in] KU */
  442. /* > \verbatim */
  443. /* > KU is INTEGER */
  444. /* > The number of superdiagonals within the band of A. KU >= 0. */
  445. /* > \endverbatim */
  446. /* > */
  447. /* > \param[in] NRHS */
  448. /* > \verbatim */
  449. /* > NRHS is INTEGER */
  450. /* > The number of right hand sides, i.e., the number of columns */
  451. /* > of the matrices B and X. NRHS >= 0. */
  452. /* > \endverbatim */
  453. /* > */
  454. /* > \param[in] AB */
  455. /* > \verbatim */
  456. /* > AB is COMPLEX array, dimension (LDAB,N) */
  457. /* > The original band matrix A, stored in rows 1 to KL+KU+1. */
  458. /* > The j-th column of A is stored in the j-th column of the */
  459. /* > array AB as follows: */
  460. /* > AB(ku+1+i-j,j) = A(i,j) for f2cmax(1,j-ku)<=i<=f2cmin(n,j+kl). */
  461. /* > \endverbatim */
  462. /* > */
  463. /* > \param[in] LDAB */
  464. /* > \verbatim */
  465. /* > LDAB is INTEGER */
  466. /* > The leading dimension of the array AB. LDAB >= KL+KU+1. */
  467. /* > \endverbatim */
  468. /* > */
  469. /* > \param[in] AFB */
  470. /* > \verbatim */
  471. /* > AFB is COMPLEX array, dimension (LDAFB,N) */
  472. /* > Details of the LU factorization of the band matrix A, as */
  473. /* > computed by CGBTRF. U is stored as an upper triangular band */
  474. /* > matrix with KL+KU superdiagonals in rows 1 to KL+KU+1, and */
  475. /* > the multipliers used during the factorization are stored in */
  476. /* > rows KL+KU+2 to 2*KL+KU+1. */
  477. /* > \endverbatim */
  478. /* > */
  479. /* > \param[in] LDAFB */
  480. /* > \verbatim */
  481. /* > LDAFB is INTEGER */
  482. /* > The leading dimension of the array AFB. LDAFB >= 2*KL*KU+1. */
  483. /* > \endverbatim */
  484. /* > */
  485. /* > \param[in] IPIV */
  486. /* > \verbatim */
  487. /* > IPIV is INTEGER array, dimension (N) */
  488. /* > The pivot indices from CGBTRF; for 1<=i<=N, row i of the */
  489. /* > matrix was interchanged with row IPIV(i). */
  490. /* > \endverbatim */
  491. /* > */
  492. /* > \param[in] B */
  493. /* > \verbatim */
  494. /* > B is COMPLEX array, dimension (LDB,NRHS) */
  495. /* > The right hand side matrix B. */
  496. /* > \endverbatim */
  497. /* > */
  498. /* > \param[in] LDB */
  499. /* > \verbatim */
  500. /* > LDB is INTEGER */
  501. /* > The leading dimension of the array B. LDB >= f2cmax(1,N). */
  502. /* > \endverbatim */
  503. /* > */
  504. /* > \param[in,out] X */
  505. /* > \verbatim */
  506. /* > X is COMPLEX array, dimension (LDX,NRHS) */
  507. /* > On entry, the solution matrix X, as computed by CGBTRS. */
  508. /* > On exit, the improved solution matrix X. */
  509. /* > \endverbatim */
  510. /* > */
  511. /* > \param[in] LDX */
  512. /* > \verbatim */
  513. /* > LDX is INTEGER */
  514. /* > The leading dimension of the array X. LDX >= f2cmax(1,N). */
  515. /* > \endverbatim */
  516. /* > */
  517. /* > \param[out] FERR */
  518. /* > \verbatim */
  519. /* > FERR is REAL array, dimension (NRHS) */
  520. /* > The estimated forward error bound for each solution vector */
  521. /* > X(j) (the j-th column of the solution matrix X). */
  522. /* > If XTRUE is the true solution corresponding to X(j), FERR(j) */
  523. /* > is an estimated upper bound for the magnitude of the largest */
  524. /* > element in (X(j) - XTRUE) divided by the magnitude of the */
  525. /* > largest element in X(j). The estimate is as reliable as */
  526. /* > the estimate for RCOND, and is almost always a slight */
  527. /* > overestimate of the true error. */
  528. /* > \endverbatim */
  529. /* > */
  530. /* > \param[out] BERR */
  531. /* > \verbatim */
  532. /* > BERR is REAL array, dimension (NRHS) */
  533. /* > The componentwise relative backward error of each solution */
  534. /* > vector X(j) (i.e., the smallest relative change in */
  535. /* > any element of A or B that makes X(j) an exact solution). */
  536. /* > \endverbatim */
  537. /* > */
  538. /* > \param[out] WORK */
  539. /* > \verbatim */
  540. /* > WORK is COMPLEX array, dimension (2*N) */
  541. /* > \endverbatim */
  542. /* > */
  543. /* > \param[out] RWORK */
  544. /* > \verbatim */
  545. /* > RWORK is REAL array, dimension (N) */
  546. /* > \endverbatim */
  547. /* > */
  548. /* > \param[out] INFO */
  549. /* > \verbatim */
  550. /* > INFO is INTEGER */
  551. /* > = 0: successful exit */
  552. /* > < 0: if INFO = -i, the i-th argument had an illegal value */
  553. /* > \endverbatim */
  554. /* > \par Internal Parameters: */
  555. /* ========================= */
  556. /* > */
  557. /* > \verbatim */
  558. /* > ITMAX is the maximum number of steps of iterative refinement. */
  559. /* > \endverbatim */
  560. /* Authors: */
  561. /* ======== */
  562. /* > \author Univ. of Tennessee */
  563. /* > \author Univ. of California Berkeley */
  564. /* > \author Univ. of Colorado Denver */
  565. /* > \author NAG Ltd. */
  566. /* > \date December 2016 */
  567. /* > \ingroup complexGBcomputational */
  568. /* ===================================================================== */
  569. /* Subroutine */ int cgbrfs_(char *trans, integer *n, integer *kl, integer *
  570. ku, integer *nrhs, complex *ab, integer *ldab, complex *afb, integer *
  571. ldafb, integer *ipiv, complex *b, integer *ldb, complex *x, integer *
  572. ldx, real *ferr, real *berr, complex *work, real *rwork, integer *
  573. info)
  574. {
  575. /* System generated locals */
  576. integer ab_dim1, ab_offset, afb_dim1, afb_offset, b_dim1, b_offset,
  577. x_dim1, x_offset, i__1, i__2, i__3, i__4, i__5, i__6, i__7;
  578. real r__1, r__2, r__3, r__4;
  579. complex q__1;
  580. /* Local variables */
  581. integer kase;
  582. real safe1, safe2;
  583. integer i__, j, k;
  584. real s;
  585. extern /* Subroutine */ int cgbmv_(char *, integer *, integer *, integer *
  586. , integer *, complex *, complex *, integer *, complex *, integer *
  587. , complex *, complex *, integer *);
  588. extern logical lsame_(char *, char *);
  589. integer isave[3];
  590. extern /* Subroutine */ int ccopy_(integer *, complex *, integer *,
  591. complex *, integer *), caxpy_(integer *, complex *, complex *,
  592. integer *, complex *, integer *);
  593. integer count;
  594. extern /* Subroutine */ int clacn2_(integer *, complex *, complex *, real
  595. *, integer *, integer *);
  596. integer kk;
  597. real xk;
  598. extern real slamch_(char *);
  599. integer nz;
  600. real safmin;
  601. extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen), cgbtrs_(
  602. char *, integer *, integer *, integer *, integer *, complex *,
  603. integer *, integer *, complex *, integer *, integer *);
  604. logical notran;
  605. char transn[1], transt[1];
  606. real lstres, eps;
  607. /* -- LAPACK computational routine (version 3.7.0) -- */
  608. /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
  609. /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
  610. /* December 2016 */
  611. /* ===================================================================== */
  612. /* Test the input parameters. */
  613. /* Parameter adjustments */
  614. ab_dim1 = *ldab;
  615. ab_offset = 1 + ab_dim1 * 1;
  616. ab -= ab_offset;
  617. afb_dim1 = *ldafb;
  618. afb_offset = 1 + afb_dim1 * 1;
  619. afb -= afb_offset;
  620. --ipiv;
  621. b_dim1 = *ldb;
  622. b_offset = 1 + b_dim1 * 1;
  623. b -= b_offset;
  624. x_dim1 = *ldx;
  625. x_offset = 1 + x_dim1 * 1;
  626. x -= x_offset;
  627. --ferr;
  628. --berr;
  629. --work;
  630. --rwork;
  631. /* Function Body */
  632. *info = 0;
  633. notran = lsame_(trans, "N");
  634. if (! notran && ! lsame_(trans, "T") && ! lsame_(
  635. trans, "C")) {
  636. *info = -1;
  637. } else if (*n < 0) {
  638. *info = -2;
  639. } else if (*kl < 0) {
  640. *info = -3;
  641. } else if (*ku < 0) {
  642. *info = -4;
  643. } else if (*nrhs < 0) {
  644. *info = -5;
  645. } else if (*ldab < *kl + *ku + 1) {
  646. *info = -7;
  647. } else if (*ldafb < (*kl << 1) + *ku + 1) {
  648. *info = -9;
  649. } else if (*ldb < f2cmax(1,*n)) {
  650. *info = -12;
  651. } else if (*ldx < f2cmax(1,*n)) {
  652. *info = -14;
  653. }
  654. if (*info != 0) {
  655. i__1 = -(*info);
  656. xerbla_("CGBRFS", &i__1, (ftnlen)6);
  657. return 0;
  658. }
  659. /* Quick return if possible */
  660. if (*n == 0 || *nrhs == 0) {
  661. i__1 = *nrhs;
  662. for (j = 1; j <= i__1; ++j) {
  663. ferr[j] = 0.f;
  664. berr[j] = 0.f;
  665. /* L10: */
  666. }
  667. return 0;
  668. }
  669. if (notran) {
  670. *(unsigned char *)transn = 'N';
  671. *(unsigned char *)transt = 'C';
  672. } else {
  673. *(unsigned char *)transn = 'C';
  674. *(unsigned char *)transt = 'N';
  675. }
  676. /* NZ = maximum number of nonzero elements in each row of A, plus 1 */
  677. /* Computing MIN */
  678. i__1 = *kl + *ku + 2, i__2 = *n + 1;
  679. nz = f2cmin(i__1,i__2);
  680. eps = slamch_("Epsilon");
  681. safmin = slamch_("Safe minimum");
  682. safe1 = nz * safmin;
  683. safe2 = safe1 / eps;
  684. /* Do for each right hand side */
  685. i__1 = *nrhs;
  686. for (j = 1; j <= i__1; ++j) {
  687. count = 1;
  688. lstres = 3.f;
  689. L20:
  690. /* Loop until stopping criterion is satisfied. */
  691. /* Compute residual R = B - op(A) * X, */
  692. /* where op(A) = A, A**T, or A**H, depending on TRANS. */
  693. ccopy_(n, &b[j * b_dim1 + 1], &c__1, &work[1], &c__1);
  694. q__1.r = -1.f, q__1.i = 0.f;
  695. cgbmv_(trans, n, n, kl, ku, &q__1, &ab[ab_offset], ldab, &x[j *
  696. x_dim1 + 1], &c__1, &c_b1, &work[1], &c__1);
  697. /* Compute componentwise relative backward error from formula */
  698. /* f2cmax(i) ( abs(R(i)) / ( abs(op(A))*abs(X) + abs(B) )(i) ) */
  699. /* where abs(Z) is the componentwise absolute value of the matrix */
  700. /* or vector Z. If the i-th component of the denominator is less */
  701. /* than SAFE2, then SAFE1 is added to the i-th components of the */
  702. /* numerator and denominator before dividing. */
  703. i__2 = *n;
  704. for (i__ = 1; i__ <= i__2; ++i__) {
  705. i__3 = i__ + j * b_dim1;
  706. rwork[i__] = (r__1 = b[i__3].r, abs(r__1)) + (r__2 = r_imag(&b[
  707. i__ + j * b_dim1]), abs(r__2));
  708. /* L30: */
  709. }
  710. /* Compute abs(op(A))*abs(X) + abs(B). */
  711. if (notran) {
  712. i__2 = *n;
  713. for (k = 1; k <= i__2; ++k) {
  714. kk = *ku + 1 - k;
  715. i__3 = k + j * x_dim1;
  716. xk = (r__1 = x[i__3].r, abs(r__1)) + (r__2 = r_imag(&x[k + j *
  717. x_dim1]), abs(r__2));
  718. /* Computing MAX */
  719. i__3 = 1, i__4 = k - *ku;
  720. /* Computing MIN */
  721. i__6 = *n, i__7 = k + *kl;
  722. i__5 = f2cmin(i__6,i__7);
  723. for (i__ = f2cmax(i__3,i__4); i__ <= i__5; ++i__) {
  724. i__3 = kk + i__ + k * ab_dim1;
  725. rwork[i__] += ((r__1 = ab[i__3].r, abs(r__1)) + (r__2 =
  726. r_imag(&ab[kk + i__ + k * ab_dim1]), abs(r__2))) *
  727. xk;
  728. /* L40: */
  729. }
  730. /* L50: */
  731. }
  732. } else {
  733. i__2 = *n;
  734. for (k = 1; k <= i__2; ++k) {
  735. s = 0.f;
  736. kk = *ku + 1 - k;
  737. /* Computing MAX */
  738. i__5 = 1, i__3 = k - *ku;
  739. /* Computing MIN */
  740. i__6 = *n, i__7 = k + *kl;
  741. i__4 = f2cmin(i__6,i__7);
  742. for (i__ = f2cmax(i__5,i__3); i__ <= i__4; ++i__) {
  743. i__5 = kk + i__ + k * ab_dim1;
  744. i__3 = i__ + j * x_dim1;
  745. s += ((r__1 = ab[i__5].r, abs(r__1)) + (r__2 = r_imag(&ab[
  746. kk + i__ + k * ab_dim1]), abs(r__2))) * ((r__3 =
  747. x[i__3].r, abs(r__3)) + (r__4 = r_imag(&x[i__ + j
  748. * x_dim1]), abs(r__4)));
  749. /* L60: */
  750. }
  751. rwork[k] += s;
  752. /* L70: */
  753. }
  754. }
  755. s = 0.f;
  756. i__2 = *n;
  757. for (i__ = 1; i__ <= i__2; ++i__) {
  758. if (rwork[i__] > safe2) {
  759. /* Computing MAX */
  760. i__4 = i__;
  761. r__3 = s, r__4 = ((r__1 = work[i__4].r, abs(r__1)) + (r__2 =
  762. r_imag(&work[i__]), abs(r__2))) / rwork[i__];
  763. s = f2cmax(r__3,r__4);
  764. } else {
  765. /* Computing MAX */
  766. i__4 = i__;
  767. r__3 = s, r__4 = ((r__1 = work[i__4].r, abs(r__1)) + (r__2 =
  768. r_imag(&work[i__]), abs(r__2)) + safe1) / (rwork[i__]
  769. + safe1);
  770. s = f2cmax(r__3,r__4);
  771. }
  772. /* L80: */
  773. }
  774. berr[j] = s;
  775. /* Test stopping criterion. Continue iterating if */
  776. /* 1) The residual BERR(J) is larger than machine epsilon, and */
  777. /* 2) BERR(J) decreased by at least a factor of 2 during the */
  778. /* last iteration, and */
  779. /* 3) At most ITMAX iterations tried. */
  780. if (berr[j] > eps && berr[j] * 2.f <= lstres && count <= 5) {
  781. /* Update solution and try again. */
  782. cgbtrs_(trans, n, kl, ku, &c__1, &afb[afb_offset], ldafb, &ipiv[1]
  783. , &work[1], n, info);
  784. caxpy_(n, &c_b1, &work[1], &c__1, &x[j * x_dim1 + 1], &c__1);
  785. lstres = berr[j];
  786. ++count;
  787. goto L20;
  788. }
  789. /* Bound error from formula */
  790. /* norm(X - XTRUE) / norm(X) .le. FERR = */
  791. /* norm( abs(inv(op(A)))* */
  792. /* ( abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) / norm(X) */
  793. /* where */
  794. /* norm(Z) is the magnitude of the largest component of Z */
  795. /* inv(op(A)) is the inverse of op(A) */
  796. /* abs(Z) is the componentwise absolute value of the matrix or */
  797. /* vector Z */
  798. /* NZ is the maximum number of nonzeros in any row of A, plus 1 */
  799. /* EPS is machine epsilon */
  800. /* The i-th component of abs(R)+NZ*EPS*(abs(op(A))*abs(X)+abs(B)) */
  801. /* is incremented by SAFE1 if the i-th component of */
  802. /* abs(op(A))*abs(X) + abs(B) is less than SAFE2. */
  803. /* Use CLACN2 to estimate the infinity-norm of the matrix */
  804. /* inv(op(A)) * diag(W), */
  805. /* where W = abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) */
  806. i__2 = *n;
  807. for (i__ = 1; i__ <= i__2; ++i__) {
  808. if (rwork[i__] > safe2) {
  809. i__4 = i__;
  810. rwork[i__] = (r__1 = work[i__4].r, abs(r__1)) + (r__2 =
  811. r_imag(&work[i__]), abs(r__2)) + nz * eps * rwork[i__]
  812. ;
  813. } else {
  814. i__4 = i__;
  815. rwork[i__] = (r__1 = work[i__4].r, abs(r__1)) + (r__2 =
  816. r_imag(&work[i__]), abs(r__2)) + nz * eps * rwork[i__]
  817. + safe1;
  818. }
  819. /* L90: */
  820. }
  821. kase = 0;
  822. L100:
  823. clacn2_(n, &work[*n + 1], &work[1], &ferr[j], &kase, isave);
  824. if (kase != 0) {
  825. if (kase == 1) {
  826. /* Multiply by diag(W)*inv(op(A)**H). */
  827. cgbtrs_(transt, n, kl, ku, &c__1, &afb[afb_offset], ldafb, &
  828. ipiv[1], &work[1], n, info);
  829. i__2 = *n;
  830. for (i__ = 1; i__ <= i__2; ++i__) {
  831. i__4 = i__;
  832. i__5 = i__;
  833. i__3 = i__;
  834. q__1.r = rwork[i__5] * work[i__3].r, q__1.i = rwork[i__5]
  835. * work[i__3].i;
  836. work[i__4].r = q__1.r, work[i__4].i = q__1.i;
  837. /* L110: */
  838. }
  839. } else {
  840. /* Multiply by inv(op(A))*diag(W). */
  841. i__2 = *n;
  842. for (i__ = 1; i__ <= i__2; ++i__) {
  843. i__4 = i__;
  844. i__5 = i__;
  845. i__3 = i__;
  846. q__1.r = rwork[i__5] * work[i__3].r, q__1.i = rwork[i__5]
  847. * work[i__3].i;
  848. work[i__4].r = q__1.r, work[i__4].i = q__1.i;
  849. /* L120: */
  850. }
  851. cgbtrs_(transn, n, kl, ku, &c__1, &afb[afb_offset], ldafb, &
  852. ipiv[1], &work[1], n, info);
  853. }
  854. goto L100;
  855. }
  856. /* Normalize error. */
  857. lstres = 0.f;
  858. i__2 = *n;
  859. for (i__ = 1; i__ <= i__2; ++i__) {
  860. /* Computing MAX */
  861. i__4 = i__ + j * x_dim1;
  862. r__3 = lstres, r__4 = (r__1 = x[i__4].r, abs(r__1)) + (r__2 =
  863. r_imag(&x[i__ + j * x_dim1]), abs(r__2));
  864. lstres = f2cmax(r__3,r__4);
  865. /* L130: */
  866. }
  867. if (lstres != 0.f) {
  868. ferr[j] /= lstres;
  869. }
  870. /* L140: */
  871. }
  872. return 0;
  873. /* End of CGBRFS */
  874. } /* cgbrfs_ */