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.

sgbrfs.c 23 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790
  1. #include <math.h>
  2. #include <stdlib.h>
  3. #include <string.h>
  4. #include <stdio.h>
  5. #include <complex.h>
  6. #ifdef complex
  7. #undef complex
  8. #endif
  9. #ifdef I
  10. #undef I
  11. #endif
  12. #if defined(_WIN64)
  13. typedef long long BLASLONG;
  14. typedef unsigned long long BLASULONG;
  15. #else
  16. typedef long BLASLONG;
  17. typedef unsigned long BLASULONG;
  18. #endif
  19. #ifdef LAPACK_ILP64
  20. typedef BLASLONG blasint;
  21. #if defined(_WIN64)
  22. #define blasabs(x) llabs(x)
  23. #else
  24. #define blasabs(x) labs(x)
  25. #endif
  26. #else
  27. typedef int blasint;
  28. #define blasabs(x) abs(x)
  29. #endif
  30. typedef blasint integer;
  31. typedef unsigned int uinteger;
  32. typedef char *address;
  33. typedef short int shortint;
  34. typedef float real;
  35. typedef double doublereal;
  36. typedef struct { real r, i; } complex;
  37. typedef struct { doublereal r, i; } doublecomplex;
  38. #ifdef _MSC_VER
  39. static inline _Fcomplex Cf(complex *z) {_Fcomplex zz={z->r , z->i}; return zz;}
  40. static inline _Dcomplex Cd(doublecomplex *z) {_Dcomplex zz={z->r , z->i};return zz;}
  41. static inline _Fcomplex * _pCf(complex *z) {return (_Fcomplex*)z;}
  42. static inline _Dcomplex * _pCd(doublecomplex *z) {return (_Dcomplex*)z;}
  43. #else
  44. static inline _Complex float Cf(complex *z) {return z->r + z->i*_Complex_I;}
  45. static inline _Complex double Cd(doublecomplex *z) {return z->r + z->i*_Complex_I;}
  46. static inline _Complex float * _pCf(complex *z) {return (_Complex float*)z;}
  47. static inline _Complex double * _pCd(doublecomplex *z) {return (_Complex double*)z;}
  48. #endif
  49. #define pCf(z) (*_pCf(z))
  50. #define pCd(z) (*_pCd(z))
  51. typedef blasint logical;
  52. typedef char logical1;
  53. typedef char integer1;
  54. #define TRUE_ (1)
  55. #define FALSE_ (0)
  56. /* Extern is for use with -E */
  57. #ifndef Extern
  58. #define Extern extern
  59. #endif
  60. /* I/O stuff */
  61. typedef int flag;
  62. typedef int ftnlen;
  63. typedef int ftnint;
  64. /*external read, write*/
  65. typedef struct
  66. { flag cierr;
  67. ftnint ciunit;
  68. flag ciend;
  69. char *cifmt;
  70. ftnint cirec;
  71. } cilist;
  72. /*internal read, write*/
  73. typedef struct
  74. { flag icierr;
  75. char *iciunit;
  76. flag iciend;
  77. char *icifmt;
  78. ftnint icirlen;
  79. ftnint icirnum;
  80. } icilist;
  81. /*open*/
  82. typedef struct
  83. { flag oerr;
  84. ftnint ounit;
  85. char *ofnm;
  86. ftnlen ofnmlen;
  87. char *osta;
  88. char *oacc;
  89. char *ofm;
  90. ftnint orl;
  91. char *oblnk;
  92. } olist;
  93. /*close*/
  94. typedef struct
  95. { flag cerr;
  96. ftnint cunit;
  97. char *csta;
  98. } cllist;
  99. /*rewind, backspace, endfile*/
  100. typedef struct
  101. { flag aerr;
  102. ftnint aunit;
  103. } alist;
  104. /* inquire */
  105. typedef struct
  106. { flag inerr;
  107. ftnint inunit;
  108. char *infile;
  109. ftnlen infilen;
  110. ftnint *inex; /*parameters in standard's order*/
  111. ftnint *inopen;
  112. ftnint *innum;
  113. ftnint *innamed;
  114. char *inname;
  115. ftnlen innamlen;
  116. char *inacc;
  117. ftnlen inacclen;
  118. char *inseq;
  119. ftnlen inseqlen;
  120. char *indir;
  121. ftnlen indirlen;
  122. char *infmt;
  123. ftnlen infmtlen;
  124. char *inform;
  125. ftnint informlen;
  126. char *inunf;
  127. ftnlen inunflen;
  128. ftnint *inrecl;
  129. ftnint *innrec;
  130. char *inblank;
  131. ftnlen inblanklen;
  132. } inlist;
  133. #define VOID void
  134. union Multitype { /* for multiple entry points */
  135. integer1 g;
  136. shortint h;
  137. integer i;
  138. /* longint j; */
  139. real r;
  140. doublereal d;
  141. complex c;
  142. doublecomplex z;
  143. };
  144. typedef union Multitype Multitype;
  145. struct Vardesc { /* for Namelist */
  146. char *name;
  147. char *addr;
  148. ftnlen *dims;
  149. int type;
  150. };
  151. typedef struct Vardesc Vardesc;
  152. struct Namelist {
  153. char *name;
  154. Vardesc **vars;
  155. int nvars;
  156. };
  157. typedef struct Namelist Namelist;
  158. #define abs(x) ((x) >= 0 ? (x) : -(x))
  159. #define dabs(x) (fabs(x))
  160. #define f2cmin(a,b) ((a) <= (b) ? (a) : (b))
  161. #define f2cmax(a,b) ((a) >= (b) ? (a) : (b))
  162. #define dmin(a,b) (f2cmin(a,b))
  163. #define dmax(a,b) (f2cmax(a,b))
  164. #define bit_test(a,b) ((a) >> (b) & 1)
  165. #define bit_clear(a,b) ((a) & ~((uinteger)1 << (b)))
  166. #define bit_set(a,b) ((a) | ((uinteger)1 << (b)))
  167. #define abort_() { sig_die("Fortran abort routine called", 1); }
  168. #define c_abs(z) (cabsf(Cf(z)))
  169. #define c_cos(R,Z) { pCf(R)=ccos(Cf(Z)); }
  170. #ifdef _MSC_VER
  171. #define c_div(c, a, b) {Cf(c)._Val[0] = (Cf(a)._Val[0]/Cf(b)._Val[0]); Cf(c)._Val[1]=(Cf(a)._Val[1]/Cf(b)._Val[1]);}
  172. #define z_div(c, a, b) {Cd(c)._Val[0] = (Cd(a)._Val[0]/Cd(b)._Val[0]); Cd(c)._Val[1]=(Cd(a)._Val[1]/df(b)._Val[1]);}
  173. #else
  174. #define c_div(c, a, b) {pCf(c) = Cf(a)/Cf(b);}
  175. #define z_div(c, a, b) {pCd(c) = Cd(a)/Cd(b);}
  176. #endif
  177. #define c_exp(R, Z) {pCf(R) = cexpf(Cf(Z));}
  178. #define c_log(R, Z) {pCf(R) = clogf(Cf(Z));}
  179. #define c_sin(R, Z) {pCf(R) = csinf(Cf(Z));}
  180. //#define c_sqrt(R, Z) {*(R) = csqrtf(Cf(Z));}
  181. #define c_sqrt(R, Z) {pCf(R) = csqrtf(Cf(Z));}
  182. #define d_abs(x) (fabs(*(x)))
  183. #define d_acos(x) (acos(*(x)))
  184. #define d_asin(x) (asin(*(x)))
  185. #define d_atan(x) (atan(*(x)))
  186. #define d_atn2(x, y) (atan2(*(x),*(y)))
  187. #define d_cnjg(R, Z) { pCd(R) = conj(Cd(Z)); }
  188. #define r_cnjg(R, Z) { pCf(R) = conjf(Cf(Z)); }
  189. #define d_cos(x) (cos(*(x)))
  190. #define d_cosh(x) (cosh(*(x)))
  191. #define d_dim(__a, __b) ( *(__a) > *(__b) ? *(__a) - *(__b) : 0.0 )
  192. #define d_exp(x) (exp(*(x)))
  193. #define d_imag(z) (cimag(Cd(z)))
  194. #define r_imag(z) (cimagf(Cf(z)))
  195. #define d_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
  196. #define r_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
  197. #define d_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
  198. #define r_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
  199. #define d_log(x) (log(*(x)))
  200. #define d_mod(x, y) (fmod(*(x), *(y)))
  201. #define u_nint(__x) ((__x)>=0 ? floor((__x) + .5) : -floor(.5 - (__x)))
  202. #define d_nint(x) u_nint(*(x))
  203. #define u_sign(__a,__b) ((__b) >= 0 ? ((__a) >= 0 ? (__a) : -(__a)) : -((__a) >= 0 ? (__a) : -(__a)))
  204. #define d_sign(a,b) u_sign(*(a),*(b))
  205. #define r_sign(a,b) u_sign(*(a),*(b))
  206. #define d_sin(x) (sin(*(x)))
  207. #define d_sinh(x) (sinh(*(x)))
  208. #define d_sqrt(x) (sqrt(*(x)))
  209. #define d_tan(x) (tan(*(x)))
  210. #define d_tanh(x) (tanh(*(x)))
  211. #define i_abs(x) abs(*(x))
  212. #define i_dnnt(x) ((integer)u_nint(*(x)))
  213. #define i_len(s, n) (n)
  214. #define i_nint(x) ((integer)u_nint(*(x)))
  215. #define i_sign(a,b) ((integer)u_sign((integer)*(a),(integer)*(b)))
  216. #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++ = ' '; }
  217. #define s_cmp(a,b,c,d) ((integer)strncmp((a),(b),f2cmin((c),(d))))
  218. #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]; }
  219. #define sig_die(s, kill) { exit(1); }
  220. #define s_stop(s, n) {exit(0);}
  221. #define z_abs(z) (cabs(Cd(z)))
  222. #define z_exp(R, Z) {pCd(R) = cexp(Cd(Z));}
  223. #define z_sqrt(R, Z) {pCd(R) = csqrt(Cd(Z));}
  224. #define myexit_() break;
  225. #define mycycle() continue;
  226. #define myceiling(w) {ceil(w)}
  227. #define myhuge(w) {HUGE_VAL}
  228. //#define mymaxloc_(w,s,e,n) {if (sizeof(*(w)) == sizeof(double)) dmaxloc_((w),*(s),*(e),n); else dmaxloc_((w),*(s),*(e),n);}
  229. #define mymaxloc(w,s,e,n) {dmaxloc_(w,*(s),*(e),n)}
  230. /* -- translated by f2c (version 20000121).
  231. You must link the resulting object file with the libraries:
  232. -lf2c -lm (in that order)
  233. */
  234. /* Table of constant values */
  235. static integer c__1 = 1;
  236. static real c_b15 = -1.f;
  237. static real c_b17 = 1.f;
  238. /* > \brief \b SGBRFS */
  239. /* =========== DOCUMENTATION =========== */
  240. /* Online html documentation available at */
  241. /* http://www.netlib.org/lapack/explore-html/ */
  242. /* > \htmlonly */
  243. /* > Download SGBRFS + dependencies */
  244. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/sgbrfs.
  245. f"> */
  246. /* > [TGZ]</a> */
  247. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/sgbrfs.
  248. f"> */
  249. /* > [ZIP]</a> */
  250. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/sgbrfs.
  251. f"> */
  252. /* > [TXT]</a> */
  253. /* > \endhtmlonly */
  254. /* Definition: */
  255. /* =========== */
  256. /* SUBROUTINE SGBRFS( TRANS, N, KL, KU, NRHS, AB, LDAB, AFB, LDAFB, */
  257. /* IPIV, B, LDB, X, LDX, FERR, BERR, WORK, IWORK, */
  258. /* INFO ) */
  259. /* CHARACTER TRANS */
  260. /* INTEGER INFO, KL, KU, LDAB, LDAFB, LDB, LDX, N, NRHS */
  261. /* INTEGER IPIV( * ), IWORK( * ) */
  262. /* REAL AB( LDAB, * ), AFB( LDAFB, * ), B( LDB, * ), */
  263. /* $ BERR( * ), FERR( * ), WORK( * ), X( LDX, * ) */
  264. /* > \par Purpose: */
  265. /* ============= */
  266. /* > */
  267. /* > \verbatim */
  268. /* > */
  269. /* > SGBRFS improves the computed solution to a system of linear */
  270. /* > equations when the coefficient matrix is banded, and provides */
  271. /* > error bounds and backward error estimates for the solution. */
  272. /* > \endverbatim */
  273. /* Arguments: */
  274. /* ========== */
  275. /* > \param[in] TRANS */
  276. /* > \verbatim */
  277. /* > TRANS is CHARACTER*1 */
  278. /* > Specifies the form of the system of equations: */
  279. /* > = 'N': A * X = B (No transpose) */
  280. /* > = 'T': A**T * X = B (Transpose) */
  281. /* > = 'C': A**H * X = B (Conjugate transpose = Transpose) */
  282. /* > \endverbatim */
  283. /* > */
  284. /* > \param[in] N */
  285. /* > \verbatim */
  286. /* > N is INTEGER */
  287. /* > The order of the matrix A. N >= 0. */
  288. /* > \endverbatim */
  289. /* > */
  290. /* > \param[in] KL */
  291. /* > \verbatim */
  292. /* > KL is INTEGER */
  293. /* > The number of subdiagonals within the band of A. KL >= 0. */
  294. /* > \endverbatim */
  295. /* > */
  296. /* > \param[in] KU */
  297. /* > \verbatim */
  298. /* > KU is INTEGER */
  299. /* > The number of superdiagonals within the band of A. KU >= 0. */
  300. /* > \endverbatim */
  301. /* > */
  302. /* > \param[in] NRHS */
  303. /* > \verbatim */
  304. /* > NRHS is INTEGER */
  305. /* > The number of right hand sides, i.e., the number of columns */
  306. /* > of the matrices B and X. NRHS >= 0. */
  307. /* > \endverbatim */
  308. /* > */
  309. /* > \param[in] AB */
  310. /* > \verbatim */
  311. /* > AB is REAL array, dimension (LDAB,N) */
  312. /* > The original band matrix A, stored in rows 1 to KL+KU+1. */
  313. /* > The j-th column of A is stored in the j-th column of the */
  314. /* > array AB as follows: */
  315. /* > AB(ku+1+i-j,j) = A(i,j) for f2cmax(1,j-ku)<=i<=f2cmin(n,j+kl). */
  316. /* > \endverbatim */
  317. /* > */
  318. /* > \param[in] LDAB */
  319. /* > \verbatim */
  320. /* > LDAB is INTEGER */
  321. /* > The leading dimension of the array AB. LDAB >= KL+KU+1. */
  322. /* > \endverbatim */
  323. /* > */
  324. /* > \param[in] AFB */
  325. /* > \verbatim */
  326. /* > AFB is REAL array, dimension (LDAFB,N) */
  327. /* > Details of the LU factorization of the band matrix A, as */
  328. /* > computed by SGBTRF. U is stored as an upper triangular band */
  329. /* > matrix with KL+KU superdiagonals in rows 1 to KL+KU+1, and */
  330. /* > the multipliers used during the factorization are stored in */
  331. /* > rows KL+KU+2 to 2*KL+KU+1. */
  332. /* > \endverbatim */
  333. /* > */
  334. /* > \param[in] LDAFB */
  335. /* > \verbatim */
  336. /* > LDAFB is INTEGER */
  337. /* > The leading dimension of the array AFB. LDAFB >= 2*KL*KU+1. */
  338. /* > \endverbatim */
  339. /* > */
  340. /* > \param[in] IPIV */
  341. /* > \verbatim */
  342. /* > IPIV is INTEGER array, dimension (N) */
  343. /* > The pivot indices from SGBTRF; for 1<=i<=N, row i of the */
  344. /* > matrix was interchanged with row IPIV(i). */
  345. /* > \endverbatim */
  346. /* > */
  347. /* > \param[in] B */
  348. /* > \verbatim */
  349. /* > B is REAL array, dimension (LDB,NRHS) */
  350. /* > The right hand side matrix B. */
  351. /* > \endverbatim */
  352. /* > */
  353. /* > \param[in] LDB */
  354. /* > \verbatim */
  355. /* > LDB is INTEGER */
  356. /* > The leading dimension of the array B. LDB >= f2cmax(1,N). */
  357. /* > \endverbatim */
  358. /* > */
  359. /* > \param[in,out] X */
  360. /* > \verbatim */
  361. /* > X is REAL array, dimension (LDX,NRHS) */
  362. /* > On entry, the solution matrix X, as computed by SGBTRS. */
  363. /* > On exit, the improved solution matrix X. */
  364. /* > \endverbatim */
  365. /* > */
  366. /* > \param[in] LDX */
  367. /* > \verbatim */
  368. /* > LDX is INTEGER */
  369. /* > The leading dimension of the array X. LDX >= f2cmax(1,N). */
  370. /* > \endverbatim */
  371. /* > */
  372. /* > \param[out] FERR */
  373. /* > \verbatim */
  374. /* > FERR is REAL array, dimension (NRHS) */
  375. /* > The estimated forward error bound for each solution vector */
  376. /* > X(j) (the j-th column of the solution matrix X). */
  377. /* > If XTRUE is the true solution corresponding to X(j), FERR(j) */
  378. /* > is an estimated upper bound for the magnitude of the largest */
  379. /* > element in (X(j) - XTRUE) divided by the magnitude of the */
  380. /* > largest element in X(j). The estimate is as reliable as */
  381. /* > the estimate for RCOND, and is almost always a slight */
  382. /* > overestimate of the true error. */
  383. /* > \endverbatim */
  384. /* > */
  385. /* > \param[out] BERR */
  386. /* > \verbatim */
  387. /* > BERR is REAL array, dimension (NRHS) */
  388. /* > The componentwise relative backward error of each solution */
  389. /* > vector X(j) (i.e., the smallest relative change in */
  390. /* > any element of A or B that makes X(j) an exact solution). */
  391. /* > \endverbatim */
  392. /* > */
  393. /* > \param[out] WORK */
  394. /* > \verbatim */
  395. /* > WORK is REAL array, dimension (3*N) */
  396. /* > \endverbatim */
  397. /* > */
  398. /* > \param[out] IWORK */
  399. /* > \verbatim */
  400. /* > IWORK is INTEGER array, dimension (N) */
  401. /* > \endverbatim */
  402. /* > */
  403. /* > \param[out] INFO */
  404. /* > \verbatim */
  405. /* > INFO is INTEGER */
  406. /* > = 0: successful exit */
  407. /* > < 0: if INFO = -i, the i-th argument had an illegal value */
  408. /* > \endverbatim */
  409. /* > \par Internal Parameters: */
  410. /* ========================= */
  411. /* > */
  412. /* > \verbatim */
  413. /* > ITMAX is the maximum number of steps of iterative refinement. */
  414. /* > \endverbatim */
  415. /* Authors: */
  416. /* ======== */
  417. /* > \author Univ. of Tennessee */
  418. /* > \author Univ. of California Berkeley */
  419. /* > \author Univ. of Colorado Denver */
  420. /* > \author NAG Ltd. */
  421. /* > \date December 2016 */
  422. /* > \ingroup realGBcomputational */
  423. /* ===================================================================== */
  424. /* Subroutine */ void sgbrfs_(char *trans, integer *n, integer *kl, integer *
  425. ku, integer *nrhs, real *ab, integer *ldab, real *afb, integer *ldafb,
  426. integer *ipiv, real *b, integer *ldb, real *x, integer *ldx, real *
  427. ferr, real *berr, real *work, integer *iwork, integer *info)
  428. {
  429. /* System generated locals */
  430. integer ab_dim1, ab_offset, afb_dim1, afb_offset, b_dim1, b_offset,
  431. x_dim1, x_offset, i__1, i__2, i__3, i__4, i__5, i__6, i__7;
  432. real r__1, r__2, r__3;
  433. /* Local variables */
  434. integer kase;
  435. real safe1, safe2;
  436. integer i__, j, k;
  437. real s;
  438. extern logical lsame_(char *, char *);
  439. integer isave[3];
  440. extern /* Subroutine */ void sgbmv_(char *, integer *, integer *, integer *
  441. , integer *, real *, real *, integer *, real *, integer *, real *,
  442. real *, integer *);
  443. integer count;
  444. extern /* Subroutine */ void scopy_(integer *, real *, integer *, real *,
  445. integer *), saxpy_(integer *, real *, real *, integer *, real *,
  446. integer *), slacn2_(integer *, real *, real *, integer *, real *,
  447. integer *, integer *);
  448. integer kk;
  449. real xk;
  450. extern real slamch_(char *);
  451. integer nz;
  452. real safmin;
  453. extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
  454. logical notran;
  455. extern /* Subroutine */ void sgbtrs_(char *, integer *, integer *, integer
  456. *, integer *, real *, integer *, integer *, real *, integer *,
  457. integer *);
  458. char transt[1];
  459. real lstres, eps;
  460. /* -- LAPACK computational routine (version 3.7.0) -- */
  461. /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
  462. /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
  463. /* December 2016 */
  464. /* ===================================================================== */
  465. /* Test the input parameters. */
  466. /* Parameter adjustments */
  467. ab_dim1 = *ldab;
  468. ab_offset = 1 + ab_dim1 * 1;
  469. ab -= ab_offset;
  470. afb_dim1 = *ldafb;
  471. afb_offset = 1 + afb_dim1 * 1;
  472. afb -= afb_offset;
  473. --ipiv;
  474. b_dim1 = *ldb;
  475. b_offset = 1 + b_dim1 * 1;
  476. b -= b_offset;
  477. x_dim1 = *ldx;
  478. x_offset = 1 + x_dim1 * 1;
  479. x -= x_offset;
  480. --ferr;
  481. --berr;
  482. --work;
  483. --iwork;
  484. /* Function Body */
  485. *info = 0;
  486. notran = lsame_(trans, "N");
  487. if (! notran && ! lsame_(trans, "T") && ! lsame_(
  488. trans, "C")) {
  489. *info = -1;
  490. } else if (*n < 0) {
  491. *info = -2;
  492. } else if (*kl < 0) {
  493. *info = -3;
  494. } else if (*ku < 0) {
  495. *info = -4;
  496. } else if (*nrhs < 0) {
  497. *info = -5;
  498. } else if (*ldab < *kl + *ku + 1) {
  499. *info = -7;
  500. } else if (*ldafb < (*kl << 1) + *ku + 1) {
  501. *info = -9;
  502. } else if (*ldb < f2cmax(1,*n)) {
  503. *info = -12;
  504. } else if (*ldx < f2cmax(1,*n)) {
  505. *info = -14;
  506. }
  507. if (*info != 0) {
  508. i__1 = -(*info);
  509. xerbla_("SGBRFS", &i__1, (ftnlen)6);
  510. return;
  511. }
  512. /* Quick return if possible */
  513. if (*n == 0 || *nrhs == 0) {
  514. i__1 = *nrhs;
  515. for (j = 1; j <= i__1; ++j) {
  516. ferr[j] = 0.f;
  517. berr[j] = 0.f;
  518. /* L10: */
  519. }
  520. return;
  521. }
  522. if (notran) {
  523. *(unsigned char *)transt = 'T';
  524. } else {
  525. *(unsigned char *)transt = 'N';
  526. }
  527. /* NZ = maximum number of nonzero elements in each row of A, plus 1 */
  528. /* Computing MIN */
  529. i__1 = *kl + *ku + 2, i__2 = *n + 1;
  530. nz = f2cmin(i__1,i__2);
  531. eps = slamch_("Epsilon");
  532. safmin = slamch_("Safe minimum");
  533. safe1 = nz * safmin;
  534. safe2 = safe1 / eps;
  535. /* Do for each right hand side */
  536. i__1 = *nrhs;
  537. for (j = 1; j <= i__1; ++j) {
  538. count = 1;
  539. lstres = 3.f;
  540. L20:
  541. /* Loop until stopping criterion is satisfied. */
  542. /* Compute residual R = B - op(A) * X, */
  543. /* where op(A) = A, A**T, or A**H, depending on TRANS. */
  544. scopy_(n, &b[j * b_dim1 + 1], &c__1, &work[*n + 1], &c__1);
  545. sgbmv_(trans, n, n, kl, ku, &c_b15, &ab[ab_offset], ldab, &x[j *
  546. x_dim1 + 1], &c__1, &c_b17, &work[*n + 1], &c__1);
  547. /* Compute componentwise relative backward error from formula */
  548. /* f2cmax(i) ( abs(R(i)) / ( abs(op(A))*abs(X) + abs(B) )(i) ) */
  549. /* where abs(Z) is the componentwise absolute value of the matrix */
  550. /* or vector Z. If the i-th component of the denominator is less */
  551. /* than SAFE2, then SAFE1 is added to the i-th components of the */
  552. /* numerator and denominator before dividing. */
  553. i__2 = *n;
  554. for (i__ = 1; i__ <= i__2; ++i__) {
  555. work[i__] = (r__1 = b[i__ + j * b_dim1], abs(r__1));
  556. /* L30: */
  557. }
  558. /* Compute abs(op(A))*abs(X) + abs(B). */
  559. if (notran) {
  560. i__2 = *n;
  561. for (k = 1; k <= i__2; ++k) {
  562. kk = *ku + 1 - k;
  563. xk = (r__1 = x[k + j * x_dim1], abs(r__1));
  564. /* Computing MAX */
  565. i__3 = 1, i__4 = k - *ku;
  566. /* Computing MIN */
  567. i__6 = *n, i__7 = k + *kl;
  568. i__5 = f2cmin(i__6,i__7);
  569. for (i__ = f2cmax(i__3,i__4); i__ <= i__5; ++i__) {
  570. work[i__] += (r__1 = ab[kk + i__ + k * ab_dim1], abs(r__1)
  571. ) * xk;
  572. /* L40: */
  573. }
  574. /* L50: */
  575. }
  576. } else {
  577. i__2 = *n;
  578. for (k = 1; k <= i__2; ++k) {
  579. s = 0.f;
  580. kk = *ku + 1 - k;
  581. /* Computing MAX */
  582. i__5 = 1, i__3 = k - *ku;
  583. /* Computing MIN */
  584. i__6 = *n, i__7 = k + *kl;
  585. i__4 = f2cmin(i__6,i__7);
  586. for (i__ = f2cmax(i__5,i__3); i__ <= i__4; ++i__) {
  587. s += (r__1 = ab[kk + i__ + k * ab_dim1], abs(r__1)) * (
  588. r__2 = x[i__ + j * x_dim1], abs(r__2));
  589. /* L60: */
  590. }
  591. work[k] += s;
  592. /* L70: */
  593. }
  594. }
  595. s = 0.f;
  596. i__2 = *n;
  597. for (i__ = 1; i__ <= i__2; ++i__) {
  598. if (work[i__] > safe2) {
  599. /* Computing MAX */
  600. r__2 = s, r__3 = (r__1 = work[*n + i__], abs(r__1)) / work[
  601. i__];
  602. s = f2cmax(r__2,r__3);
  603. } else {
  604. /* Computing MAX */
  605. r__2 = s, r__3 = ((r__1 = work[*n + i__], abs(r__1)) + safe1)
  606. / (work[i__] + safe1);
  607. s = f2cmax(r__2,r__3);
  608. }
  609. /* L80: */
  610. }
  611. berr[j] = s;
  612. /* Test stopping criterion. Continue iterating if */
  613. /* 1) The residual BERR(J) is larger than machine epsilon, and */
  614. /* 2) BERR(J) decreased by at least a factor of 2 during the */
  615. /* last iteration, and */
  616. /* 3) At most ITMAX iterations tried. */
  617. if (berr[j] > eps && berr[j] * 2.f <= lstres && count <= 5) {
  618. /* Update solution and try again. */
  619. sgbtrs_(trans, n, kl, ku, &c__1, &afb[afb_offset], ldafb, &ipiv[1]
  620. , &work[*n + 1], n, info);
  621. saxpy_(n, &c_b17, &work[*n + 1], &c__1, &x[j * x_dim1 + 1], &c__1)
  622. ;
  623. lstres = berr[j];
  624. ++count;
  625. goto L20;
  626. }
  627. /* Bound error from formula */
  628. /* norm(X - XTRUE) / norm(X) .le. FERR = */
  629. /* norm( abs(inv(op(A)))* */
  630. /* ( abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) / norm(X) */
  631. /* where */
  632. /* norm(Z) is the magnitude of the largest component of Z */
  633. /* inv(op(A)) is the inverse of op(A) */
  634. /* abs(Z) is the componentwise absolute value of the matrix or */
  635. /* vector Z */
  636. /* NZ is the maximum number of nonzeros in any row of A, plus 1 */
  637. /* EPS is machine epsilon */
  638. /* The i-th component of abs(R)+NZ*EPS*(abs(op(A))*abs(X)+abs(B)) */
  639. /* is incremented by SAFE1 if the i-th component of */
  640. /* abs(op(A))*abs(X) + abs(B) is less than SAFE2. */
  641. /* Use SLACN2 to estimate the infinity-norm of the matrix */
  642. /* inv(op(A)) * diag(W), */
  643. /* where W = abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) */
  644. i__2 = *n;
  645. for (i__ = 1; i__ <= i__2; ++i__) {
  646. if (work[i__] > safe2) {
  647. work[i__] = (r__1 = work[*n + i__], abs(r__1)) + nz * eps *
  648. work[i__];
  649. } else {
  650. work[i__] = (r__1 = work[*n + i__], abs(r__1)) + nz * eps *
  651. work[i__] + safe1;
  652. }
  653. /* L90: */
  654. }
  655. kase = 0;
  656. L100:
  657. slacn2_(n, &work[(*n << 1) + 1], &work[*n + 1], &iwork[1], &ferr[j], &
  658. kase, isave);
  659. if (kase != 0) {
  660. if (kase == 1) {
  661. /* Multiply by diag(W)*inv(op(A)**T). */
  662. sgbtrs_(transt, n, kl, ku, &c__1, &afb[afb_offset], ldafb, &
  663. ipiv[1], &work[*n + 1], n, info);
  664. i__2 = *n;
  665. for (i__ = 1; i__ <= i__2; ++i__) {
  666. work[*n + i__] *= work[i__];
  667. /* L110: */
  668. }
  669. } else {
  670. /* Multiply by inv(op(A))*diag(W). */
  671. i__2 = *n;
  672. for (i__ = 1; i__ <= i__2; ++i__) {
  673. work[*n + i__] *= work[i__];
  674. /* L120: */
  675. }
  676. sgbtrs_(trans, n, kl, ku, &c__1, &afb[afb_offset], ldafb, &
  677. ipiv[1], &work[*n + 1], n, info);
  678. }
  679. goto L100;
  680. }
  681. /* Normalize error. */
  682. lstres = 0.f;
  683. i__2 = *n;
  684. for (i__ = 1; i__ <= i__2; ++i__) {
  685. /* Computing MAX */
  686. r__2 = lstres, r__3 = (r__1 = x[i__ + j * x_dim1], abs(r__1));
  687. lstres = f2cmax(r__2,r__3);
  688. /* L130: */
  689. }
  690. if (lstres != 0.f) {
  691. ferr[j] /= lstres;
  692. }
  693. /* L140: */
  694. }
  695. return;
  696. /* End of SGBRFS */
  697. } /* sgbrfs_ */