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.

dlaed3.c 23 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786
  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_b22 = 1.;
  364. static doublereal c_b23 = 0.;
  365. /* > \brief \b DLAED3 used by sstedc. Finds the roots of the secular equation and updates the eigenvectors. Us
  366. ed when the original matrix is tridiagonal. */
  367. /* =========== DOCUMENTATION =========== */
  368. /* Online html documentation available at */
  369. /* http://www.netlib.org/lapack/explore-html/ */
  370. /* > \htmlonly */
  371. /* > Download DLAED3 + dependencies */
  372. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlaed3.
  373. f"> */
  374. /* > [TGZ]</a> */
  375. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlaed3.
  376. f"> */
  377. /* > [ZIP]</a> */
  378. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlaed3.
  379. f"> */
  380. /* > [TXT]</a> */
  381. /* > \endhtmlonly */
  382. /* Definition: */
  383. /* =========== */
  384. /* SUBROUTINE DLAED3( K, N, N1, D, Q, LDQ, RHO, DLAMDA, Q2, INDX, */
  385. /* CTOT, W, S, INFO ) */
  386. /* INTEGER INFO, K, LDQ, N, N1 */
  387. /* DOUBLE PRECISION RHO */
  388. /* INTEGER CTOT( * ), INDX( * ) */
  389. /* DOUBLE PRECISION D( * ), DLAMDA( * ), Q( LDQ, * ), Q2( * ), */
  390. /* $ S( * ), W( * ) */
  391. /* > \par Purpose: */
  392. /* ============= */
  393. /* > */
  394. /* > \verbatim */
  395. /* > */
  396. /* > DLAED3 finds the roots of the secular equation, as defined by the */
  397. /* > values in D, W, and RHO, between 1 and K. It makes the */
  398. /* > appropriate calls to DLAED4 and then updates the eigenvectors by */
  399. /* > multiplying the matrix of eigenvectors of the pair of eigensystems */
  400. /* > being combined by the matrix of eigenvectors of the K-by-K system */
  401. /* > which is solved here. */
  402. /* > */
  403. /* > This code makes very mild assumptions about floating point */
  404. /* > arithmetic. It will work on machines with a guard digit in */
  405. /* > add/subtract, or on those binary machines without guard digits */
  406. /* > which subtract like the Cray X-MP, Cray Y-MP, Cray C-90, or Cray-2. */
  407. /* > It could conceivably fail on hexadecimal or decimal machines */
  408. /* > without guard digits, but we know of none. */
  409. /* > \endverbatim */
  410. /* Arguments: */
  411. /* ========== */
  412. /* > \param[in] K */
  413. /* > \verbatim */
  414. /* > K is INTEGER */
  415. /* > The number of terms in the rational function to be solved by */
  416. /* > DLAED4. K >= 0. */
  417. /* > \endverbatim */
  418. /* > */
  419. /* > \param[in] N */
  420. /* > \verbatim */
  421. /* > N is INTEGER */
  422. /* > The number of rows and columns in the Q matrix. */
  423. /* > N >= K (deflation may result in N>K). */
  424. /* > \endverbatim */
  425. /* > */
  426. /* > \param[in] N1 */
  427. /* > \verbatim */
  428. /* > N1 is INTEGER */
  429. /* > The location of the last eigenvalue in the leading submatrix. */
  430. /* > f2cmin(1,N) <= N1 <= N/2. */
  431. /* > \endverbatim */
  432. /* > */
  433. /* > \param[out] D */
  434. /* > \verbatim */
  435. /* > D is DOUBLE PRECISION array, dimension (N) */
  436. /* > D(I) contains the updated eigenvalues for */
  437. /* > 1 <= I <= K. */
  438. /* > \endverbatim */
  439. /* > */
  440. /* > \param[out] Q */
  441. /* > \verbatim */
  442. /* > Q is DOUBLE PRECISION array, dimension (LDQ,N) */
  443. /* > Initially the first K columns are used as workspace. */
  444. /* > On output the columns 1 to K contain */
  445. /* > the updated eigenvectors. */
  446. /* > \endverbatim */
  447. /* > */
  448. /* > \param[in] LDQ */
  449. /* > \verbatim */
  450. /* > LDQ is INTEGER */
  451. /* > The leading dimension of the array Q. LDQ >= f2cmax(1,N). */
  452. /* > \endverbatim */
  453. /* > */
  454. /* > \param[in] RHO */
  455. /* > \verbatim */
  456. /* > RHO is DOUBLE PRECISION */
  457. /* > The value of the parameter in the rank one update equation. */
  458. /* > RHO >= 0 required. */
  459. /* > \endverbatim */
  460. /* > */
  461. /* > \param[in,out] DLAMDA */
  462. /* > \verbatim */
  463. /* > DLAMDA is DOUBLE PRECISION array, dimension (K) */
  464. /* > The first K elements of this array contain the old roots */
  465. /* > of the deflated updating problem. These are the poles */
  466. /* > of the secular equation. May be changed on output by */
  467. /* > having lowest order bit set to zero on Cray X-MP, Cray Y-MP, */
  468. /* > Cray-2, or Cray C-90, as described above. */
  469. /* > \endverbatim */
  470. /* > */
  471. /* > \param[in] Q2 */
  472. /* > \verbatim */
  473. /* > Q2 is DOUBLE PRECISION array, dimension (LDQ2*N) */
  474. /* > The first K columns of this matrix contain the non-deflated */
  475. /* > eigenvectors for the split problem. */
  476. /* > \endverbatim */
  477. /* > */
  478. /* > \param[in] INDX */
  479. /* > \verbatim */
  480. /* > INDX is INTEGER array, dimension (N) */
  481. /* > The permutation used to arrange the columns of the deflated */
  482. /* > Q matrix into three groups (see DLAED2). */
  483. /* > The rows of the eigenvectors found by DLAED4 must be likewise */
  484. /* > permuted before the matrix multiply can take place. */
  485. /* > \endverbatim */
  486. /* > */
  487. /* > \param[in] CTOT */
  488. /* > \verbatim */
  489. /* > CTOT is INTEGER array, dimension (4) */
  490. /* > A count of the total number of the various types of columns */
  491. /* > in Q, as described in INDX. The fourth column type is any */
  492. /* > column which has been deflated. */
  493. /* > \endverbatim */
  494. /* > */
  495. /* > \param[in,out] W */
  496. /* > \verbatim */
  497. /* > W is DOUBLE PRECISION array, dimension (K) */
  498. /* > The first K elements of this array contain the components */
  499. /* > of the deflation-adjusted updating vector. Destroyed on */
  500. /* > output. */
  501. /* > \endverbatim */
  502. /* > */
  503. /* > \param[out] S */
  504. /* > \verbatim */
  505. /* > S is DOUBLE PRECISION array, dimension (N1 + 1)*K */
  506. /* > Will contain the eigenvectors of the repaired matrix which */
  507. /* > will be multiplied by the previously accumulated eigenvectors */
  508. /* > to update the system. */
  509. /* > \endverbatim */
  510. /* > */
  511. /* > \param[out] INFO */
  512. /* > \verbatim */
  513. /* > INFO is INTEGER */
  514. /* > = 0: successful exit. */
  515. /* > < 0: if INFO = -i, the i-th argument had an illegal value. */
  516. /* > > 0: if INFO = 1, an eigenvalue did not converge */
  517. /* > \endverbatim */
  518. /* Authors: */
  519. /* ======== */
  520. /* > \author Univ. of Tennessee */
  521. /* > \author Univ. of California Berkeley */
  522. /* > \author Univ. of Colorado Denver */
  523. /* > \author NAG Ltd. */
  524. /* > \date June 2017 */
  525. /* > \ingroup auxOTHERcomputational */
  526. /* > \par Contributors: */
  527. /* ================== */
  528. /* > */
  529. /* > Jeff Rutter, Computer Science Division, University of California */
  530. /* > at Berkeley, USA \n */
  531. /* > Modified by Francoise Tisseur, University of Tennessee */
  532. /* > */
  533. /* ===================================================================== */
  534. /* Subroutine */ int dlaed3_(integer *k, integer *n, integer *n1, doublereal *
  535. d__, doublereal *q, integer *ldq, doublereal *rho, doublereal *dlamda,
  536. doublereal *q2, integer *indx, integer *ctot, doublereal *w,
  537. doublereal *s, integer *info)
  538. {
  539. /* System generated locals */
  540. integer q_dim1, q_offset, i__1, i__2;
  541. doublereal d__1;
  542. /* Local variables */
  543. doublereal temp;
  544. extern doublereal dnrm2_(integer *, doublereal *, integer *);
  545. integer i__, j;
  546. extern /* Subroutine */ int dgemm_(char *, char *, integer *, integer *,
  547. integer *, doublereal *, doublereal *, integer *, doublereal *,
  548. integer *, doublereal *, doublereal *, integer *),
  549. dcopy_(integer *, doublereal *, integer *, doublereal *, integer
  550. *), dlaed4_(integer *, integer *, doublereal *, doublereal *,
  551. doublereal *, doublereal *, doublereal *, integer *);
  552. integer n2;
  553. extern doublereal dlamc3_(doublereal *, doublereal *);
  554. integer n12, ii, n23;
  555. extern /* Subroutine */ int dlacpy_(char *, integer *, integer *,
  556. doublereal *, integer *, doublereal *, integer *),
  557. dlaset_(char *, integer *, integer *, doublereal *, doublereal *,
  558. doublereal *, integer *), xerbla_(char *, integer *, ftnlen);
  559. integer iq2;
  560. /* -- LAPACK computational routine (version 3.7.1) -- */
  561. /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
  562. /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
  563. /* June 2017 */
  564. /* ===================================================================== */
  565. /* Test the input parameters. */
  566. /* Parameter adjustments */
  567. --d__;
  568. q_dim1 = *ldq;
  569. q_offset = 1 + q_dim1 * 1;
  570. q -= q_offset;
  571. --dlamda;
  572. --q2;
  573. --indx;
  574. --ctot;
  575. --w;
  576. --s;
  577. /* Function Body */
  578. *info = 0;
  579. if (*k < 0) {
  580. *info = -1;
  581. } else if (*n < *k) {
  582. *info = -2;
  583. } else if (*ldq < f2cmax(1,*n)) {
  584. *info = -6;
  585. }
  586. if (*info != 0) {
  587. i__1 = -(*info);
  588. xerbla_("DLAED3", &i__1, (ftnlen)6);
  589. return 0;
  590. }
  591. /* Quick return if possible */
  592. if (*k == 0) {
  593. return 0;
  594. }
  595. /* Modify values DLAMDA(i) to make sure all DLAMDA(i)-DLAMDA(j) can */
  596. /* be computed with high relative accuracy (barring over/underflow). */
  597. /* This is a problem on machines without a guard digit in */
  598. /* add/subtract (Cray XMP, Cray YMP, Cray C 90 and Cray 2). */
  599. /* The following code replaces DLAMDA(I) by 2*DLAMDA(I)-DLAMDA(I), */
  600. /* which on any of these machines zeros out the bottommost */
  601. /* bit of DLAMDA(I) if it is 1; this makes the subsequent */
  602. /* subtractions DLAMDA(I)-DLAMDA(J) unproblematic when cancellation */
  603. /* occurs. On binary machines with a guard digit (almost all */
  604. /* machines) it does not change DLAMDA(I) at all. On hexadecimal */
  605. /* and decimal machines with a guard digit, it slightly */
  606. /* changes the bottommost bits of DLAMDA(I). It does not account */
  607. /* for hexadecimal or decimal machines without guard digits */
  608. /* (we know of none). We use a subroutine call to compute */
  609. /* 2*DLAMBDA(I) to prevent optimizing compilers from eliminating */
  610. /* this code. */
  611. i__1 = *k;
  612. for (i__ = 1; i__ <= i__1; ++i__) {
  613. dlamda[i__] = dlamc3_(&dlamda[i__], &dlamda[i__]) - dlamda[i__];
  614. /* L10: */
  615. }
  616. i__1 = *k;
  617. for (j = 1; j <= i__1; ++j) {
  618. dlaed4_(k, &j, &dlamda[1], &w[1], &q[j * q_dim1 + 1], rho, &d__[j],
  619. info);
  620. /* If the zero finder fails, the computation is terminated. */
  621. if (*info != 0) {
  622. goto L120;
  623. }
  624. /* L20: */
  625. }
  626. if (*k == 1) {
  627. goto L110;
  628. }
  629. if (*k == 2) {
  630. i__1 = *k;
  631. for (j = 1; j <= i__1; ++j) {
  632. w[1] = q[j * q_dim1 + 1];
  633. w[2] = q[j * q_dim1 + 2];
  634. ii = indx[1];
  635. q[j * q_dim1 + 1] = w[ii];
  636. ii = indx[2];
  637. q[j * q_dim1 + 2] = w[ii];
  638. /* L30: */
  639. }
  640. goto L110;
  641. }
  642. /* Compute updated W. */
  643. dcopy_(k, &w[1], &c__1, &s[1], &c__1);
  644. /* Initialize W(I) = Q(I,I) */
  645. i__1 = *ldq + 1;
  646. dcopy_(k, &q[q_offset], &i__1, &w[1], &c__1);
  647. i__1 = *k;
  648. for (j = 1; j <= i__1; ++j) {
  649. i__2 = j - 1;
  650. for (i__ = 1; i__ <= i__2; ++i__) {
  651. w[i__] *= q[i__ + j * q_dim1] / (dlamda[i__] - dlamda[j]);
  652. /* L40: */
  653. }
  654. i__2 = *k;
  655. for (i__ = j + 1; i__ <= i__2; ++i__) {
  656. w[i__] *= q[i__ + j * q_dim1] / (dlamda[i__] - dlamda[j]);
  657. /* L50: */
  658. }
  659. /* L60: */
  660. }
  661. i__1 = *k;
  662. for (i__ = 1; i__ <= i__1; ++i__) {
  663. d__1 = sqrt(-w[i__]);
  664. w[i__] = d_sign(&d__1, &s[i__]);
  665. /* L70: */
  666. }
  667. /* Compute eigenvectors of the modified rank-1 modification. */
  668. i__1 = *k;
  669. for (j = 1; j <= i__1; ++j) {
  670. i__2 = *k;
  671. for (i__ = 1; i__ <= i__2; ++i__) {
  672. s[i__] = w[i__] / q[i__ + j * q_dim1];
  673. /* L80: */
  674. }
  675. temp = dnrm2_(k, &s[1], &c__1);
  676. i__2 = *k;
  677. for (i__ = 1; i__ <= i__2; ++i__) {
  678. ii = indx[i__];
  679. q[i__ + j * q_dim1] = s[ii] / temp;
  680. /* L90: */
  681. }
  682. /* L100: */
  683. }
  684. /* Compute the updated eigenvectors. */
  685. L110:
  686. n2 = *n - *n1;
  687. n12 = ctot[1] + ctot[2];
  688. n23 = ctot[2] + ctot[3];
  689. dlacpy_("A", &n23, k, &q[ctot[1] + 1 + q_dim1], ldq, &s[1], &n23);
  690. iq2 = *n1 * n12 + 1;
  691. if (n23 != 0) {
  692. dgemm_("N", "N", &n2, k, &n23, &c_b22, &q2[iq2], &n2, &s[1], &n23, &
  693. c_b23, &q[*n1 + 1 + q_dim1], ldq);
  694. } else {
  695. dlaset_("A", &n2, k, &c_b23, &c_b23, &q[*n1 + 1 + q_dim1], ldq);
  696. }
  697. dlacpy_("A", &n12, k, &q[q_offset], ldq, &s[1], &n12);
  698. if (n12 != 0) {
  699. dgemm_("N", "N", n1, k, &n12, &c_b22, &q2[1], n1, &s[1], &n12, &c_b23,
  700. &q[q_offset], ldq);
  701. } else {
  702. dlaset_("A", n1, k, &c_b23, &c_b23, &q[q_dim1 + 1], ldq);
  703. }
  704. L120:
  705. return 0;
  706. /* End of DLAED3 */
  707. } /* dlaed3_ */