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.

slarre.c 47 kB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492
  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 int logical;
  52. typedef short int shortlogical;
  53. typedef char logical1;
  54. typedef char integer1;
  55. #define TRUE_ (1)
  56. #define FALSE_ (0)
  57. /* Extern is for use with -E */
  58. #ifndef Extern
  59. #define Extern extern
  60. #endif
  61. /* I/O stuff */
  62. typedef int flag;
  63. typedef int ftnlen;
  64. typedef int ftnint;
  65. /*external read, write*/
  66. typedef struct
  67. { flag cierr;
  68. ftnint ciunit;
  69. flag ciend;
  70. char *cifmt;
  71. ftnint cirec;
  72. } cilist;
  73. /*internal read, write*/
  74. typedef struct
  75. { flag icierr;
  76. char *iciunit;
  77. flag iciend;
  78. char *icifmt;
  79. ftnint icirlen;
  80. ftnint icirnum;
  81. } icilist;
  82. /*open*/
  83. typedef struct
  84. { flag oerr;
  85. ftnint ounit;
  86. char *ofnm;
  87. ftnlen ofnmlen;
  88. char *osta;
  89. char *oacc;
  90. char *ofm;
  91. ftnint orl;
  92. char *oblnk;
  93. } olist;
  94. /*close*/
  95. typedef struct
  96. { flag cerr;
  97. ftnint cunit;
  98. char *csta;
  99. } cllist;
  100. /*rewind, backspace, endfile*/
  101. typedef struct
  102. { flag aerr;
  103. ftnint aunit;
  104. } alist;
  105. /* inquire */
  106. typedef struct
  107. { flag inerr;
  108. ftnint inunit;
  109. char *infile;
  110. ftnlen infilen;
  111. ftnint *inex; /*parameters in standard's order*/
  112. ftnint *inopen;
  113. ftnint *innum;
  114. ftnint *innamed;
  115. char *inname;
  116. ftnlen innamlen;
  117. char *inacc;
  118. ftnlen inacclen;
  119. char *inseq;
  120. ftnlen inseqlen;
  121. char *indir;
  122. ftnlen indirlen;
  123. char *infmt;
  124. ftnlen infmtlen;
  125. char *inform;
  126. ftnint informlen;
  127. char *inunf;
  128. ftnlen inunflen;
  129. ftnint *inrecl;
  130. ftnint *innrec;
  131. char *inblank;
  132. ftnlen inblanklen;
  133. } inlist;
  134. #define VOID void
  135. union Multitype { /* for multiple entry points */
  136. integer1 g;
  137. shortint h;
  138. integer i;
  139. /* longint j; */
  140. real r;
  141. doublereal d;
  142. complex c;
  143. doublecomplex z;
  144. };
  145. typedef union Multitype Multitype;
  146. struct Vardesc { /* for Namelist */
  147. char *name;
  148. char *addr;
  149. ftnlen *dims;
  150. int type;
  151. };
  152. typedef struct Vardesc Vardesc;
  153. struct Namelist {
  154. char *name;
  155. Vardesc **vars;
  156. int nvars;
  157. };
  158. typedef struct Namelist Namelist;
  159. #define abs(x) ((x) >= 0 ? (x) : -(x))
  160. #define dabs(x) (fabs(x))
  161. #define f2cmin(a,b) ((a) <= (b) ? (a) : (b))
  162. #define f2cmax(a,b) ((a) >= (b) ? (a) : (b))
  163. #define dmin(a,b) (f2cmin(a,b))
  164. #define dmax(a,b) (f2cmax(a,b))
  165. #define bit_test(a,b) ((a) >> (b) & 1)
  166. #define bit_clear(a,b) ((a) & ~((uinteger)1 << (b)))
  167. #define bit_set(a,b) ((a) | ((uinteger)1 << (b)))
  168. #define abort_() { sig_die("Fortran abort routine called", 1); }
  169. #define c_abs(z) (cabsf(Cf(z)))
  170. #define c_cos(R,Z) { pCf(R)=ccos(Cf(Z)); }
  171. #ifdef _MSC_VER
  172. #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]);}
  173. #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]);}
  174. #else
  175. #define c_div(c, a, b) {pCf(c) = Cf(a)/Cf(b);}
  176. #define z_div(c, a, b) {pCd(c) = Cd(a)/Cd(b);}
  177. #endif
  178. #define c_exp(R, Z) {pCf(R) = cexpf(Cf(Z));}
  179. #define c_log(R, Z) {pCf(R) = clogf(Cf(Z));}
  180. #define c_sin(R, Z) {pCf(R) = csinf(Cf(Z));}
  181. //#define c_sqrt(R, Z) {*(R) = csqrtf(Cf(Z));}
  182. #define c_sqrt(R, Z) {pCf(R) = csqrtf(Cf(Z));}
  183. #define d_abs(x) (fabs(*(x)))
  184. #define d_acos(x) (acos(*(x)))
  185. #define d_asin(x) (asin(*(x)))
  186. #define d_atan(x) (atan(*(x)))
  187. #define d_atn2(x, y) (atan2(*(x),*(y)))
  188. #define d_cnjg(R, Z) { pCd(R) = conj(Cd(Z)); }
  189. #define r_cnjg(R, Z) { pCf(R) = conjf(Cf(Z)); }
  190. #define d_cos(x) (cos(*(x)))
  191. #define d_cosh(x) (cosh(*(x)))
  192. #define d_dim(__a, __b) ( *(__a) > *(__b) ? *(__a) - *(__b) : 0.0 )
  193. #define d_exp(x) (exp(*(x)))
  194. #define d_imag(z) (cimag(Cd(z)))
  195. #define r_imag(z) (cimagf(Cf(z)))
  196. #define d_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
  197. #define r_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
  198. #define d_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
  199. #define r_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
  200. #define d_log(x) (log(*(x)))
  201. #define d_mod(x, y) (fmod(*(x), *(y)))
  202. #define u_nint(__x) ((__x)>=0 ? floor((__x) + .5) : -floor(.5 - (__x)))
  203. #define d_nint(x) u_nint(*(x))
  204. #define u_sign(__a,__b) ((__b) >= 0 ? ((__a) >= 0 ? (__a) : -(__a)) : -((__a) >= 0 ? (__a) : -(__a)))
  205. #define d_sign(a,b) u_sign(*(a),*(b))
  206. #define r_sign(a,b) u_sign(*(a),*(b))
  207. #define d_sin(x) (sin(*(x)))
  208. #define d_sinh(x) (sinh(*(x)))
  209. #define d_sqrt(x) (sqrt(*(x)))
  210. #define d_tan(x) (tan(*(x)))
  211. #define d_tanh(x) (tanh(*(x)))
  212. #define i_abs(x) abs(*(x))
  213. #define i_dnnt(x) ((integer)u_nint(*(x)))
  214. #define i_len(s, n) (n)
  215. #define i_nint(x) ((integer)u_nint(*(x)))
  216. #define i_sign(a,b) ((integer)u_sign((integer)*(a),(integer)*(b)))
  217. #define pow_dd(ap, bp) ( pow(*(ap), *(bp)))
  218. #define pow_si(B,E) spow_ui(*(B),*(E))
  219. #define pow_ri(B,E) spow_ui(*(B),*(E))
  220. #define pow_di(B,E) dpow_ui(*(B),*(E))
  221. #define pow_zi(p, a, b) {pCd(p) = zpow_ui(Cd(a), *(b));}
  222. #define pow_ci(p, a, b) {pCf(p) = cpow_ui(Cf(a), *(b));}
  223. #define pow_zz(R,A,B) {pCd(R) = cpow(Cd(A),*(B));}
  224. #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++ = ' '; }
  225. #define s_cmp(a,b,c,d) ((integer)strncmp((a),(b),f2cmin((c),(d))))
  226. #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]; }
  227. #define sig_die(s, kill) { exit(1); }
  228. #define s_stop(s, n) {exit(0);}
  229. static char junk[] = "\n@(#)LIBF77 VERSION 19990503\n";
  230. #define z_abs(z) (cabs(Cd(z)))
  231. #define z_exp(R, Z) {pCd(R) = cexp(Cd(Z));}
  232. #define z_sqrt(R, Z) {pCd(R) = csqrt(Cd(Z));}
  233. #define myexit_() break;
  234. #define mycycle() continue;
  235. #define myceiling(w) {ceil(w)}
  236. #define myhuge(w) {HUGE_VAL}
  237. //#define mymaxloc_(w,s,e,n) {if (sizeof(*(w)) == sizeof(double)) dmaxloc_((w),*(s),*(e),n); else dmaxloc_((w),*(s),*(e),n);}
  238. #define mymaxloc(w,s,e,n) {dmaxloc_(w,*(s),*(e),n)}
  239. /* procedure parameter types for -A and -C++ */
  240. #define F2C_proc_par_types 1
  241. #ifdef __cplusplus
  242. typedef logical (*L_fp)(...);
  243. #else
  244. typedef logical (*L_fp)();
  245. #endif
  246. static float spow_ui(float x, integer n) {
  247. float pow=1.0; unsigned long int u;
  248. if(n != 0) {
  249. if(n < 0) n = -n, x = 1/x;
  250. for(u = n; ; ) {
  251. if(u & 01) pow *= x;
  252. if(u >>= 1) x *= x;
  253. else break;
  254. }
  255. }
  256. return pow;
  257. }
  258. static double dpow_ui(double x, integer n) {
  259. double pow=1.0; unsigned long int u;
  260. if(n != 0) {
  261. if(n < 0) n = -n, x = 1/x;
  262. for(u = n; ; ) {
  263. if(u & 01) pow *= x;
  264. if(u >>= 1) x *= x;
  265. else break;
  266. }
  267. }
  268. return pow;
  269. }
  270. #ifdef _MSC_VER
  271. static _Fcomplex cpow_ui(complex x, integer n) {
  272. complex pow={1.0,0.0}; unsigned long int u;
  273. if(n != 0) {
  274. if(n < 0) n = -n, x.r = 1/x.r, x.i=1/x.i;
  275. for(u = n; ; ) {
  276. if(u & 01) pow.r *= x.r, pow.i *= x.i;
  277. if(u >>= 1) x.r *= x.r, x.i *= x.i;
  278. else break;
  279. }
  280. }
  281. _Fcomplex p={pow.r, pow.i};
  282. return p;
  283. }
  284. #else
  285. static _Complex float cpow_ui(_Complex float x, integer n) {
  286. _Complex float pow=1.0; unsigned long int u;
  287. if(n != 0) {
  288. if(n < 0) n = -n, x = 1/x;
  289. for(u = n; ; ) {
  290. if(u & 01) pow *= x;
  291. if(u >>= 1) x *= x;
  292. else break;
  293. }
  294. }
  295. return pow;
  296. }
  297. #endif
  298. #ifdef _MSC_VER
  299. static _Dcomplex zpow_ui(_Dcomplex x, integer n) {
  300. _Dcomplex pow={1.0,0.0}; unsigned long int u;
  301. if(n != 0) {
  302. if(n < 0) n = -n, x._Val[0] = 1/x._Val[0], x._Val[1] =1/x._Val[1];
  303. for(u = n; ; ) {
  304. if(u & 01) pow._Val[0] *= x._Val[0], pow._Val[1] *= x._Val[1];
  305. if(u >>= 1) x._Val[0] *= x._Val[0], x._Val[1] *= x._Val[1];
  306. else break;
  307. }
  308. }
  309. _Dcomplex p = {pow._Val[0], pow._Val[1]};
  310. return p;
  311. }
  312. #else
  313. static _Complex double zpow_ui(_Complex double x, integer n) {
  314. _Complex double pow=1.0; unsigned long int u;
  315. if(n != 0) {
  316. if(n < 0) n = -n, x = 1/x;
  317. for(u = n; ; ) {
  318. if(u & 01) pow *= x;
  319. if(u >>= 1) x *= x;
  320. else break;
  321. }
  322. }
  323. return pow;
  324. }
  325. #endif
  326. static integer pow_ii(integer x, integer n) {
  327. integer pow; unsigned long int u;
  328. if (n <= 0) {
  329. if (n == 0 || x == 1) pow = 1;
  330. else if (x != -1) pow = x == 0 ? 1/x : 0;
  331. else n = -n;
  332. }
  333. if ((n > 0) || !(n == 0 || x == 1 || x != -1)) {
  334. u = n;
  335. for(pow = 1; ; ) {
  336. if(u & 01) pow *= x;
  337. if(u >>= 1) x *= x;
  338. else break;
  339. }
  340. }
  341. return pow;
  342. }
  343. static integer dmaxloc_(double *w, integer s, integer e, integer *n)
  344. {
  345. double m; integer i, mi;
  346. for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
  347. if (w[i-1]>m) mi=i ,m=w[i-1];
  348. return mi-s+1;
  349. }
  350. static integer smaxloc_(float *w, integer s, integer e, integer *n)
  351. {
  352. float m; integer i, mi;
  353. for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
  354. if (w[i-1]>m) mi=i ,m=w[i-1];
  355. return mi-s+1;
  356. }
  357. static inline void cdotc_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
  358. integer n = *n_, incx = *incx_, incy = *incy_, i;
  359. #ifdef _MSC_VER
  360. _Fcomplex zdotc = {0.0, 0.0};
  361. if (incx == 1 && incy == 1) {
  362. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  363. zdotc._Val[0] += conjf(Cf(&x[i]))._Val[0] * Cf(&y[i])._Val[0];
  364. zdotc._Val[1] += conjf(Cf(&x[i]))._Val[1] * Cf(&y[i])._Val[1];
  365. }
  366. } else {
  367. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  368. zdotc._Val[0] += conjf(Cf(&x[i*incx]))._Val[0] * Cf(&y[i*incy])._Val[0];
  369. zdotc._Val[1] += conjf(Cf(&x[i*incx]))._Val[1] * Cf(&y[i*incy])._Val[1];
  370. }
  371. }
  372. pCf(z) = zdotc;
  373. }
  374. #else
  375. _Complex float zdotc = 0.0;
  376. if (incx == 1 && incy == 1) {
  377. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  378. zdotc += conjf(Cf(&x[i])) * Cf(&y[i]);
  379. }
  380. } else {
  381. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  382. zdotc += conjf(Cf(&x[i*incx])) * Cf(&y[i*incy]);
  383. }
  384. }
  385. pCf(z) = zdotc;
  386. }
  387. #endif
  388. static inline void zdotc_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
  389. integer n = *n_, incx = *incx_, incy = *incy_, i;
  390. #ifdef _MSC_VER
  391. _Dcomplex zdotc = {0.0, 0.0};
  392. if (incx == 1 && incy == 1) {
  393. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  394. zdotc._Val[0] += conj(Cd(&x[i]))._Val[0] * Cd(&y[i])._Val[0];
  395. zdotc._Val[1] += conj(Cd(&x[i]))._Val[1] * Cd(&y[i])._Val[1];
  396. }
  397. } else {
  398. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  399. zdotc._Val[0] += conj(Cd(&x[i*incx]))._Val[0] * Cd(&y[i*incy])._Val[0];
  400. zdotc._Val[1] += conj(Cd(&x[i*incx]))._Val[1] * Cd(&y[i*incy])._Val[1];
  401. }
  402. }
  403. pCd(z) = zdotc;
  404. }
  405. #else
  406. _Complex double zdotc = 0.0;
  407. if (incx == 1 && incy == 1) {
  408. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  409. zdotc += conj(Cd(&x[i])) * Cd(&y[i]);
  410. }
  411. } else {
  412. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  413. zdotc += conj(Cd(&x[i*incx])) * Cd(&y[i*incy]);
  414. }
  415. }
  416. pCd(z) = zdotc;
  417. }
  418. #endif
  419. static inline void cdotu_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
  420. integer n = *n_, incx = *incx_, incy = *incy_, i;
  421. #ifdef _MSC_VER
  422. _Fcomplex zdotc = {0.0, 0.0};
  423. if (incx == 1 && incy == 1) {
  424. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  425. zdotc._Val[0] += Cf(&x[i])._Val[0] * Cf(&y[i])._Val[0];
  426. zdotc._Val[1] += Cf(&x[i])._Val[1] * Cf(&y[i])._Val[1];
  427. }
  428. } else {
  429. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  430. zdotc._Val[0] += Cf(&x[i*incx])._Val[0] * Cf(&y[i*incy])._Val[0];
  431. zdotc._Val[1] += Cf(&x[i*incx])._Val[1] * Cf(&y[i*incy])._Val[1];
  432. }
  433. }
  434. pCf(z) = zdotc;
  435. }
  436. #else
  437. _Complex float zdotc = 0.0;
  438. if (incx == 1 && incy == 1) {
  439. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  440. zdotc += Cf(&x[i]) * Cf(&y[i]);
  441. }
  442. } else {
  443. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  444. zdotc += Cf(&x[i*incx]) * Cf(&y[i*incy]);
  445. }
  446. }
  447. pCf(z) = zdotc;
  448. }
  449. #endif
  450. static inline void zdotu_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
  451. integer n = *n_, incx = *incx_, incy = *incy_, i;
  452. #ifdef _MSC_VER
  453. _Dcomplex zdotc = {0.0, 0.0};
  454. if (incx == 1 && incy == 1) {
  455. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  456. zdotc._Val[0] += Cd(&x[i])._Val[0] * Cd(&y[i])._Val[0];
  457. zdotc._Val[1] += Cd(&x[i])._Val[1] * Cd(&y[i])._Val[1];
  458. }
  459. } else {
  460. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  461. zdotc._Val[0] += Cd(&x[i*incx])._Val[0] * Cd(&y[i*incy])._Val[0];
  462. zdotc._Val[1] += Cd(&x[i*incx])._Val[1] * Cd(&y[i*incy])._Val[1];
  463. }
  464. }
  465. pCd(z) = zdotc;
  466. }
  467. #else
  468. _Complex double zdotc = 0.0;
  469. if (incx == 1 && incy == 1) {
  470. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  471. zdotc += Cd(&x[i]) * Cd(&y[i]);
  472. }
  473. } else {
  474. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  475. zdotc += Cd(&x[i*incx]) * Cd(&y[i*incy]);
  476. }
  477. }
  478. pCd(z) = zdotc;
  479. }
  480. #endif
  481. /* -- translated by f2c (version 20000121).
  482. You must link the resulting object file with the libraries:
  483. -lf2c -lm (in that order)
  484. */
  485. /* Table of constant values */
  486. static integer c__1 = 1;
  487. static integer c__2 = 2;
  488. /* > \brief \b SLARRE given the tridiagonal matrix T, sets small off-diagonal elements to zero and for each un
  489. reduced block Ti, finds base representations and eigenvalues. */
  490. /* =========== DOCUMENTATION =========== */
  491. /* Online html documentation available at */
  492. /* http://www.netlib.org/lapack/explore-html/ */
  493. /* > \htmlonly */
  494. /* > Download SLARRE + dependencies */
  495. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/slarre.
  496. f"> */
  497. /* > [TGZ]</a> */
  498. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/slarre.
  499. f"> */
  500. /* > [ZIP]</a> */
  501. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/slarre.
  502. f"> */
  503. /* > [TXT]</a> */
  504. /* > \endhtmlonly */
  505. /* Definition: */
  506. /* =========== */
  507. /* SUBROUTINE SLARRE( RANGE, N, VL, VU, IL, IU, D, E, E2, */
  508. /* RTOL1, RTOL2, SPLTOL, NSPLIT, ISPLIT, M, */
  509. /* W, WERR, WGAP, IBLOCK, INDEXW, GERS, PIVMIN, */
  510. /* WORK, IWORK, INFO ) */
  511. /* CHARACTER RANGE */
  512. /* INTEGER IL, INFO, IU, M, N, NSPLIT */
  513. /* REAL PIVMIN, RTOL1, RTOL2, SPLTOL, VL, VU */
  514. /* INTEGER IBLOCK( * ), ISPLIT( * ), IWORK( * ), */
  515. /* $ INDEXW( * ) */
  516. /* REAL D( * ), E( * ), E2( * ), GERS( * ), */
  517. /* $ W( * ),WERR( * ), WGAP( * ), WORK( * ) */
  518. /* > \par Purpose: */
  519. /* ============= */
  520. /* > */
  521. /* > \verbatim */
  522. /* > */
  523. /* > To find the desired eigenvalues of a given real symmetric */
  524. /* > tridiagonal matrix T, SLARRE sets any "small" off-diagonal */
  525. /* > elements to zero, and for each unreduced block T_i, it finds */
  526. /* > (a) a suitable shift at one end of the block's spectrum, */
  527. /* > (b) the base representation, T_i - sigma_i I = L_i D_i L_i^T, and */
  528. /* > (c) eigenvalues of each L_i D_i L_i^T. */
  529. /* > The representations and eigenvalues found are then used by */
  530. /* > SSTEMR to compute the eigenvectors of T. */
  531. /* > The accuracy varies depending on whether bisection is used to */
  532. /* > find a few eigenvalues or the dqds algorithm (subroutine SLASQ2) to */
  533. /* > conpute all and then discard any unwanted one. */
  534. /* > As an added benefit, SLARRE also outputs the n */
  535. /* > Gerschgorin intervals for the matrices L_i D_i L_i^T. */
  536. /* > \endverbatim */
  537. /* Arguments: */
  538. /* ========== */
  539. /* > \param[in] RANGE */
  540. /* > \verbatim */
  541. /* > RANGE is CHARACTER*1 */
  542. /* > = 'A': ("All") all eigenvalues will be found. */
  543. /* > = 'V': ("Value") all eigenvalues in the half-open interval */
  544. /* > (VL, VU] will be found. */
  545. /* > = 'I': ("Index") the IL-th through IU-th eigenvalues (of the */
  546. /* > entire matrix) will be found. */
  547. /* > \endverbatim */
  548. /* > */
  549. /* > \param[in] N */
  550. /* > \verbatim */
  551. /* > N is INTEGER */
  552. /* > The order of the matrix. N > 0. */
  553. /* > \endverbatim */
  554. /* > */
  555. /* > \param[in,out] VL */
  556. /* > \verbatim */
  557. /* > VL is REAL */
  558. /* > If RANGE='V', the lower bound for the eigenvalues. */
  559. /* > Eigenvalues less than or equal to VL, or greater than VU, */
  560. /* > will not be returned. VL < VU. */
  561. /* > If RANGE='I' or ='A', SLARRE computes bounds on the desired */
  562. /* > part of the spectrum. */
  563. /* > \endverbatim */
  564. /* > */
  565. /* > \param[in,out] VU */
  566. /* > \verbatim */
  567. /* > VU is REAL */
  568. /* > If RANGE='V', the upper bound for the eigenvalues. */
  569. /* > Eigenvalues less than or equal to VL, or greater than VU, */
  570. /* > will not be returned. VL < VU. */
  571. /* > If RANGE='I' or ='A', SLARRE computes bounds on the desired */
  572. /* > part of the spectrum. */
  573. /* > \endverbatim */
  574. /* > */
  575. /* > \param[in] IL */
  576. /* > \verbatim */
  577. /* > IL is INTEGER */
  578. /* > If RANGE='I', the index of the */
  579. /* > smallest eigenvalue to be returned. */
  580. /* > 1 <= IL <= IU <= N. */
  581. /* > \endverbatim */
  582. /* > */
  583. /* > \param[in] IU */
  584. /* > \verbatim */
  585. /* > IU is INTEGER */
  586. /* > If RANGE='I', the index of the */
  587. /* > largest eigenvalue to be returned. */
  588. /* > 1 <= IL <= IU <= N. */
  589. /* > \endverbatim */
  590. /* > */
  591. /* > \param[in,out] D */
  592. /* > \verbatim */
  593. /* > D is REAL array, dimension (N) */
  594. /* > On entry, the N diagonal elements of the tridiagonal */
  595. /* > matrix T. */
  596. /* > On exit, the N diagonal elements of the diagonal */
  597. /* > matrices D_i. */
  598. /* > \endverbatim */
  599. /* > */
  600. /* > \param[in,out] E */
  601. /* > \verbatim */
  602. /* > E is REAL array, dimension (N) */
  603. /* > On entry, the first (N-1) entries contain the subdiagonal */
  604. /* > elements of the tridiagonal matrix T; E(N) need not be set. */
  605. /* > On exit, E contains the subdiagonal elements of the unit */
  606. /* > bidiagonal matrices L_i. The entries E( ISPLIT( I ) ), */
  607. /* > 1 <= I <= NSPLIT, contain the base points sigma_i on output. */
  608. /* > \endverbatim */
  609. /* > */
  610. /* > \param[in,out] E2 */
  611. /* > \verbatim */
  612. /* > E2 is REAL array, dimension (N) */
  613. /* > On entry, the first (N-1) entries contain the SQUARES of the */
  614. /* > subdiagonal elements of the tridiagonal matrix T; */
  615. /* > E2(N) need not be set. */
  616. /* > On exit, the entries E2( ISPLIT( I ) ), */
  617. /* > 1 <= I <= NSPLIT, have been set to zero */
  618. /* > \endverbatim */
  619. /* > */
  620. /* > \param[in] RTOL1 */
  621. /* > \verbatim */
  622. /* > RTOL1 is REAL */
  623. /* > \endverbatim */
  624. /* > */
  625. /* > \param[in] RTOL2 */
  626. /* > \verbatim */
  627. /* > RTOL2 is REAL */
  628. /* > Parameters for bisection. */
  629. /* > An interval [LEFT,RIGHT] has converged if */
  630. /* > RIGHT-LEFT < MAX( RTOL1*GAP, RTOL2*MAX(|LEFT|,|RIGHT|) ) */
  631. /* > \endverbatim */
  632. /* > */
  633. /* > \param[in] SPLTOL */
  634. /* > \verbatim */
  635. /* > SPLTOL is REAL */
  636. /* > The threshold for splitting. */
  637. /* > \endverbatim */
  638. /* > */
  639. /* > \param[out] NSPLIT */
  640. /* > \verbatim */
  641. /* > NSPLIT is INTEGER */
  642. /* > The number of blocks T splits into. 1 <= NSPLIT <= N. */
  643. /* > \endverbatim */
  644. /* > */
  645. /* > \param[out] ISPLIT */
  646. /* > \verbatim */
  647. /* > ISPLIT is INTEGER array, dimension (N) */
  648. /* > The splitting points, at which T breaks up into blocks. */
  649. /* > The first block consists of rows/columns 1 to ISPLIT(1), */
  650. /* > the second of rows/columns ISPLIT(1)+1 through ISPLIT(2), */
  651. /* > etc., and the NSPLIT-th consists of rows/columns */
  652. /* > ISPLIT(NSPLIT-1)+1 through ISPLIT(NSPLIT)=N. */
  653. /* > \endverbatim */
  654. /* > */
  655. /* > \param[out] M */
  656. /* > \verbatim */
  657. /* > M is INTEGER */
  658. /* > The total number of eigenvalues (of all L_i D_i L_i^T) */
  659. /* > found. */
  660. /* > \endverbatim */
  661. /* > */
  662. /* > \param[out] W */
  663. /* > \verbatim */
  664. /* > W is REAL array, dimension (N) */
  665. /* > The first M elements contain the eigenvalues. The */
  666. /* > eigenvalues of each of the blocks, L_i D_i L_i^T, are */
  667. /* > sorted in ascending order ( SLARRE may use the */
  668. /* > remaining N-M elements as workspace). */
  669. /* > \endverbatim */
  670. /* > */
  671. /* > \param[out] WERR */
  672. /* > \verbatim */
  673. /* > WERR is REAL array, dimension (N) */
  674. /* > The error bound on the corresponding eigenvalue in W. */
  675. /* > \endverbatim */
  676. /* > */
  677. /* > \param[out] WGAP */
  678. /* > \verbatim */
  679. /* > WGAP is REAL array, dimension (N) */
  680. /* > The separation from the right neighbor eigenvalue in W. */
  681. /* > The gap is only with respect to the eigenvalues of the same block */
  682. /* > as each block has its own representation tree. */
  683. /* > Exception: at the right end of a block we store the left gap */
  684. /* > \endverbatim */
  685. /* > */
  686. /* > \param[out] IBLOCK */
  687. /* > \verbatim */
  688. /* > IBLOCK is INTEGER array, dimension (N) */
  689. /* > The indices of the blocks (submatrices) associated with the */
  690. /* > corresponding eigenvalues in W; IBLOCK(i)=1 if eigenvalue */
  691. /* > W(i) belongs to the first block from the top, =2 if W(i) */
  692. /* > belongs to the second block, etc. */
  693. /* > \endverbatim */
  694. /* > */
  695. /* > \param[out] INDEXW */
  696. /* > \verbatim */
  697. /* > INDEXW is INTEGER array, dimension (N) */
  698. /* > The indices of the eigenvalues within each block (submatrix); */
  699. /* > for example, INDEXW(i)= 10 and IBLOCK(i)=2 imply that the */
  700. /* > i-th eigenvalue W(i) is the 10-th eigenvalue in block 2 */
  701. /* > \endverbatim */
  702. /* > */
  703. /* > \param[out] GERS */
  704. /* > \verbatim */
  705. /* > GERS is REAL array, dimension (2*N) */
  706. /* > The N Gerschgorin intervals (the i-th Gerschgorin interval */
  707. /* > is (GERS(2*i-1), GERS(2*i)). */
  708. /* > \endverbatim */
  709. /* > */
  710. /* > \param[out] PIVMIN */
  711. /* > \verbatim */
  712. /* > PIVMIN is REAL */
  713. /* > The minimum pivot in the Sturm sequence for T. */
  714. /* > \endverbatim */
  715. /* > */
  716. /* > \param[out] WORK */
  717. /* > \verbatim */
  718. /* > WORK is REAL array, dimension (6*N) */
  719. /* > Workspace. */
  720. /* > \endverbatim */
  721. /* > */
  722. /* > \param[out] IWORK */
  723. /* > \verbatim */
  724. /* > IWORK is INTEGER array, dimension (5*N) */
  725. /* > Workspace. */
  726. /* > \endverbatim */
  727. /* > */
  728. /* > \param[out] INFO */
  729. /* > \verbatim */
  730. /* > INFO is INTEGER */
  731. /* > = 0: successful exit */
  732. /* > > 0: A problem occurred in SLARRE. */
  733. /* > < 0: One of the called subroutines signaled an internal problem. */
  734. /* > Needs inspection of the corresponding parameter IINFO */
  735. /* > for further information. */
  736. /* > */
  737. /* > =-1: Problem in SLARRD. */
  738. /* > = 2: No base representation could be found in MAXTRY iterations. */
  739. /* > Increasing MAXTRY and recompilation might be a remedy. */
  740. /* > =-3: Problem in SLARRB when computing the refined root */
  741. /* > representation for SLASQ2. */
  742. /* > =-4: Problem in SLARRB when preforming bisection on the */
  743. /* > desired part of the spectrum. */
  744. /* > =-5: Problem in SLASQ2. */
  745. /* > =-6: Problem in SLASQ2. */
  746. /* > \endverbatim */
  747. /* Authors: */
  748. /* ======== */
  749. /* > \author Univ. of Tennessee */
  750. /* > \author Univ. of California Berkeley */
  751. /* > \author Univ. of Colorado Denver */
  752. /* > \author NAG Ltd. */
  753. /* > \date June 2016 */
  754. /* > \ingroup OTHERauxiliary */
  755. /* > \par Further Details: */
  756. /* ===================== */
  757. /* > */
  758. /* > \verbatim */
  759. /* > */
  760. /* > The base representations are required to suffer very little */
  761. /* > element growth and consequently define all their eigenvalues to */
  762. /* > high relative accuracy. */
  763. /* > \endverbatim */
  764. /* > \par Contributors: */
  765. /* ================== */
  766. /* > */
  767. /* > Beresford Parlett, University of California, Berkeley, USA \n */
  768. /* > Jim Demmel, University of California, Berkeley, USA \n */
  769. /* > Inderjit Dhillon, University of Texas, Austin, USA \n */
  770. /* > Osni Marques, LBNL/NERSC, USA \n */
  771. /* > Christof Voemel, University of California, Berkeley, USA \n */
  772. /* > */
  773. /* ===================================================================== */
  774. /* Subroutine */ void slarre_(char *range, integer *n, real *vl, real *vu,
  775. integer *il, integer *iu, real *d__, real *e, real *e2, real *rtol1,
  776. real *rtol2, real *spltol, integer *nsplit, integer *isplit, integer *
  777. m, real *w, real *werr, real *wgap, integer *iblock, integer *indexw,
  778. real *gers, real *pivmin, real *work, integer *iwork, integer *info)
  779. {
  780. /* System generated locals */
  781. integer i__1, i__2;
  782. real r__1, r__2, r__3;
  783. /* Local variables */
  784. real eabs;
  785. integer iend, jblk;
  786. real eold;
  787. integer indl;
  788. real dmax__, emax;
  789. integer wend, idum, indu;
  790. real rtol;
  791. integer i__, j, iseed[4];
  792. real avgap, sigma;
  793. extern logical lsame_(char *, char *);
  794. integer iinfo;
  795. logical norep;
  796. extern /* Subroutine */ void scopy_(integer *, real *, integer *, real *,
  797. integer *);
  798. real s1, s2;
  799. extern /* Subroutine */ void slasq2_(integer *, real *, integer *);
  800. integer mb;
  801. real gl;
  802. integer in, mm;
  803. real gu;
  804. integer ibegin;
  805. logical forceb;
  806. integer irange;
  807. real sgndef;
  808. extern real slamch_(char *);
  809. integer wbegin;
  810. real safmin, spdiam;
  811. extern /* Subroutine */ void slarra_(integer *, real *, real *, real *,
  812. real *, real *, integer *, integer *, integer *);
  813. logical usedqd;
  814. real clwdth, isleft;
  815. extern /* Subroutine */ void slarrb_(integer *, real *, real *, integer *,
  816. integer *, real *, real *, integer *, real *, real *, real *,
  817. real *, integer *, real *, real *, integer *, integer *), slarrc_(
  818. char *, integer *, real *, real *, real *, real *, real *,
  819. integer *, integer *, integer *, integer *), slarrd_(char
  820. *, char *, integer *, real *, real *, integer *, integer *, real *
  821. , real *, real *, real *, real *, real *, integer *, integer *,
  822. integer *, real *, real *, real *, real *, integer *, integer *,
  823. real *, integer *, integer *), slarrk_(integer *,
  824. integer *, real *, real *, real *, real *, real *, real *, real *,
  825. real *, integer *);
  826. real isrght, bsrtol, dpivot;
  827. extern /* Subroutine */ void slarnv_(integer *, integer *, integer *, real
  828. *);
  829. integer cnt;
  830. real eps, tau, tmp, rtl;
  831. integer cnt1, cnt2;
  832. real tmp1;
  833. /* -- LAPACK auxiliary routine (version 3.8.0) -- */
  834. /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
  835. /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
  836. /* June 2016 */
  837. /* ===================================================================== */
  838. /* Parameter adjustments */
  839. --iwork;
  840. --work;
  841. --gers;
  842. --indexw;
  843. --iblock;
  844. --wgap;
  845. --werr;
  846. --w;
  847. --isplit;
  848. --e2;
  849. --e;
  850. --d__;
  851. /* Function Body */
  852. *info = 0;
  853. /* Quick return if possible */
  854. if (*n <= 0) {
  855. return;
  856. }
  857. /* Decode RANGE */
  858. if (lsame_(range, "A")) {
  859. irange = 1;
  860. } else if (lsame_(range, "V")) {
  861. irange = 3;
  862. } else if (lsame_(range, "I")) {
  863. irange = 2;
  864. }
  865. *m = 0;
  866. /* Get machine constants */
  867. safmin = slamch_("S");
  868. eps = slamch_("P");
  869. /* Set parameters */
  870. rtl = eps * 100.f;
  871. /* If one were ever to ask for less initial precision in BSRTOL, */
  872. /* one should keep in mind that for the subset case, the extremal */
  873. /* eigenvalues must be at least as accurate as the current setting */
  874. /* (eigenvalues in the middle need not as much accuracy) */
  875. bsrtol = sqrt(eps) * 5e-4f;
  876. /* Treat case of 1x1 matrix for quick return */
  877. if (*n == 1) {
  878. if (irange == 1 || irange == 3 && d__[1] > *vl && d__[1] <= *vu ||
  879. irange == 2 && *il == 1 && *iu == 1) {
  880. *m = 1;
  881. w[1] = d__[1];
  882. /* The computation error of the eigenvalue is zero */
  883. werr[1] = 0.f;
  884. wgap[1] = 0.f;
  885. iblock[1] = 1;
  886. indexw[1] = 1;
  887. gers[1] = d__[1];
  888. gers[2] = d__[1];
  889. }
  890. /* store the shift for the initial RRR, which is zero in this case */
  891. e[1] = 0.f;
  892. return;
  893. }
  894. /* General case: tridiagonal matrix of order > 1 */
  895. /* Init WERR, WGAP. Compute Gerschgorin intervals and spectral diameter. */
  896. /* Compute maximum off-diagonal entry and pivmin. */
  897. gl = d__[1];
  898. gu = d__[1];
  899. eold = 0.f;
  900. emax = 0.f;
  901. e[*n] = 0.f;
  902. i__1 = *n;
  903. for (i__ = 1; i__ <= i__1; ++i__) {
  904. werr[i__] = 0.f;
  905. wgap[i__] = 0.f;
  906. eabs = (r__1 = e[i__], abs(r__1));
  907. if (eabs >= emax) {
  908. emax = eabs;
  909. }
  910. tmp1 = eabs + eold;
  911. gers[(i__ << 1) - 1] = d__[i__] - tmp1;
  912. /* Computing MIN */
  913. r__1 = gl, r__2 = gers[(i__ << 1) - 1];
  914. gl = f2cmin(r__1,r__2);
  915. gers[i__ * 2] = d__[i__] + tmp1;
  916. /* Computing MAX */
  917. r__1 = gu, r__2 = gers[i__ * 2];
  918. gu = f2cmax(r__1,r__2);
  919. eold = eabs;
  920. /* L5: */
  921. }
  922. /* The minimum pivot allowed in the Sturm sequence for T */
  923. /* Computing MAX */
  924. /* Computing 2nd power */
  925. r__3 = emax;
  926. r__1 = 1.f, r__2 = r__3 * r__3;
  927. *pivmin = safmin * f2cmax(r__1,r__2);
  928. /* Compute spectral diameter. The Gerschgorin bounds give an */
  929. /* estimate that is wrong by at most a factor of SQRT(2) */
  930. spdiam = gu - gl;
  931. /* Compute splitting points */
  932. slarra_(n, &d__[1], &e[1], &e2[1], spltol, &spdiam, nsplit, &isplit[1], &
  933. iinfo);
  934. /* Can force use of bisection instead of faster DQDS. */
  935. /* Option left in the code for future multisection work. */
  936. forceb = FALSE_;
  937. /* Initialize USEDQD, DQDS should be used for ALLRNG unless someone */
  938. /* explicitly wants bisection. */
  939. usedqd = irange == 1 && ! forceb;
  940. if (irange == 1 && ! forceb) {
  941. /* Set interval [VL,VU] that contains all eigenvalues */
  942. *vl = gl;
  943. *vu = gu;
  944. } else {
  945. /* We call SLARRD to find crude approximations to the eigenvalues */
  946. /* in the desired range. In case IRANGE = INDRNG, we also obtain the */
  947. /* interval (VL,VU] that contains all the wanted eigenvalues. */
  948. /* An interval [LEFT,RIGHT] has converged if */
  949. /* RIGHT-LEFT.LT.RTOL*MAX(ABS(LEFT),ABS(RIGHT)) */
  950. /* SLARRD needs a WORK of size 4*N, IWORK of size 3*N */
  951. slarrd_(range, "B", n, vl, vu, il, iu, &gers[1], &bsrtol, &d__[1], &e[
  952. 1], &e2[1], pivmin, nsplit, &isplit[1], &mm, &w[1], &werr[1],
  953. vl, vu, &iblock[1], &indexw[1], &work[1], &iwork[1], &iinfo);
  954. if (iinfo != 0) {
  955. *info = -1;
  956. return;
  957. }
  958. /* Make sure that the entries M+1 to N in W, WERR, IBLOCK, INDEXW are 0 */
  959. i__1 = *n;
  960. for (i__ = mm + 1; i__ <= i__1; ++i__) {
  961. w[i__] = 0.f;
  962. werr[i__] = 0.f;
  963. iblock[i__] = 0;
  964. indexw[i__] = 0;
  965. /* L14: */
  966. }
  967. }
  968. /* ** */
  969. /* Loop over unreduced blocks */
  970. ibegin = 1;
  971. wbegin = 1;
  972. i__1 = *nsplit;
  973. for (jblk = 1; jblk <= i__1; ++jblk) {
  974. iend = isplit[jblk];
  975. in = iend - ibegin + 1;
  976. /* 1 X 1 block */
  977. if (in == 1) {
  978. if (irange == 1 || irange == 3 && d__[ibegin] > *vl && d__[ibegin]
  979. <= *vu || irange == 2 && iblock[wbegin] == jblk) {
  980. ++(*m);
  981. w[*m] = d__[ibegin];
  982. werr[*m] = 0.f;
  983. /* The gap for a single block doesn't matter for the later */
  984. /* algorithm and is assigned an arbitrary large value */
  985. wgap[*m] = 0.f;
  986. iblock[*m] = jblk;
  987. indexw[*m] = 1;
  988. ++wbegin;
  989. }
  990. /* E( IEND ) holds the shift for the initial RRR */
  991. e[iend] = 0.f;
  992. ibegin = iend + 1;
  993. goto L170;
  994. }
  995. /* Blocks of size larger than 1x1 */
  996. /* E( IEND ) will hold the shift for the initial RRR, for now set it =0 */
  997. e[iend] = 0.f;
  998. /* Find local outer bounds GL,GU for the block */
  999. gl = d__[ibegin];
  1000. gu = d__[ibegin];
  1001. i__2 = iend;
  1002. for (i__ = ibegin; i__ <= i__2; ++i__) {
  1003. /* Computing MIN */
  1004. r__1 = gers[(i__ << 1) - 1];
  1005. gl = f2cmin(r__1,gl);
  1006. /* Computing MAX */
  1007. r__1 = gers[i__ * 2];
  1008. gu = f2cmax(r__1,gu);
  1009. /* L15: */
  1010. }
  1011. spdiam = gu - gl;
  1012. if (! (irange == 1 && ! forceb)) {
  1013. /* Count the number of eigenvalues in the current block. */
  1014. mb = 0;
  1015. i__2 = mm;
  1016. for (i__ = wbegin; i__ <= i__2; ++i__) {
  1017. if (iblock[i__] == jblk) {
  1018. ++mb;
  1019. } else {
  1020. goto L21;
  1021. }
  1022. /* L20: */
  1023. }
  1024. L21:
  1025. if (mb == 0) {
  1026. /* No eigenvalue in the current block lies in the desired range */
  1027. /* E( IEND ) holds the shift for the initial RRR */
  1028. e[iend] = 0.f;
  1029. ibegin = iend + 1;
  1030. goto L170;
  1031. } else {
  1032. /* Decide whether dqds or bisection is more efficient */
  1033. usedqd = (real) mb > in * .5f && ! forceb;
  1034. wend = wbegin + mb - 1;
  1035. /* Calculate gaps for the current block */
  1036. /* In later stages, when representations for individual */
  1037. /* eigenvalues are different, we use SIGMA = E( IEND ). */
  1038. sigma = 0.f;
  1039. i__2 = wend - 1;
  1040. for (i__ = wbegin; i__ <= i__2; ++i__) {
  1041. /* Computing MAX */
  1042. r__1 = 0.f, r__2 = w[i__ + 1] - werr[i__ + 1] - (w[i__] +
  1043. werr[i__]);
  1044. wgap[i__] = f2cmax(r__1,r__2);
  1045. /* L30: */
  1046. }
  1047. /* Computing MAX */
  1048. r__1 = 0.f, r__2 = *vu - sigma - (w[wend] + werr[wend]);
  1049. wgap[wend] = f2cmax(r__1,r__2);
  1050. /* Find local index of the first and last desired evalue. */
  1051. indl = indexw[wbegin];
  1052. indu = indexw[wend];
  1053. }
  1054. }
  1055. if (irange == 1 && ! forceb || usedqd) {
  1056. /* Case of DQDS */
  1057. /* Find approximations to the extremal eigenvalues of the block */
  1058. slarrk_(&in, &c__1, &gl, &gu, &d__[ibegin], &e2[ibegin], pivmin, &
  1059. rtl, &tmp, &tmp1, &iinfo);
  1060. if (iinfo != 0) {
  1061. *info = -1;
  1062. return;
  1063. }
  1064. /* Computing MAX */
  1065. r__2 = gl, r__3 = tmp - tmp1 - eps * 100.f * (r__1 = tmp - tmp1,
  1066. abs(r__1));
  1067. isleft = f2cmax(r__2,r__3);
  1068. slarrk_(&in, &in, &gl, &gu, &d__[ibegin], &e2[ibegin], pivmin, &
  1069. rtl, &tmp, &tmp1, &iinfo);
  1070. if (iinfo != 0) {
  1071. *info = -1;
  1072. return;
  1073. }
  1074. /* Computing MIN */
  1075. r__2 = gu, r__3 = tmp + tmp1 + eps * 100.f * (r__1 = tmp + tmp1,
  1076. abs(r__1));
  1077. isrght = f2cmin(r__2,r__3);
  1078. /* Improve the estimate of the spectral diameter */
  1079. spdiam = isrght - isleft;
  1080. } else {
  1081. /* Case of bisection */
  1082. /* Find approximations to the wanted extremal eigenvalues */
  1083. /* Computing MAX */
  1084. r__2 = gl, r__3 = w[wbegin] - werr[wbegin] - eps * 100.f * (r__1 =
  1085. w[wbegin] - werr[wbegin], abs(r__1));
  1086. isleft = f2cmax(r__2,r__3);
  1087. /* Computing MIN */
  1088. r__2 = gu, r__3 = w[wend] + werr[wend] + eps * 100.f * (r__1 = w[
  1089. wend] + werr[wend], abs(r__1));
  1090. isrght = f2cmin(r__2,r__3);
  1091. }
  1092. /* Decide whether the base representation for the current block */
  1093. /* L_JBLK D_JBLK L_JBLK^T = T_JBLK - sigma_JBLK I */
  1094. /* should be on the left or the right end of the current block. */
  1095. /* The strategy is to shift to the end which is "more populated" */
  1096. /* Furthermore, decide whether to use DQDS for the computation of */
  1097. /* the eigenvalue approximations at the end of SLARRE or bisection. */
  1098. /* dqds is chosen if all eigenvalues are desired or the number of */
  1099. /* eigenvalues to be computed is large compared to the blocksize. */
  1100. if (irange == 1 && ! forceb) {
  1101. /* If all the eigenvalues have to be computed, we use dqd */
  1102. usedqd = TRUE_;
  1103. /* INDL is the local index of the first eigenvalue to compute */
  1104. indl = 1;
  1105. indu = in;
  1106. /* MB = number of eigenvalues to compute */
  1107. mb = in;
  1108. wend = wbegin + mb - 1;
  1109. /* Define 1/4 and 3/4 points of the spectrum */
  1110. s1 = isleft + spdiam * .25f;
  1111. s2 = isrght - spdiam * .25f;
  1112. } else {
  1113. /* SLARRD has computed IBLOCK and INDEXW for each eigenvalue */
  1114. /* approximation. */
  1115. /* choose sigma */
  1116. if (usedqd) {
  1117. s1 = isleft + spdiam * .25f;
  1118. s2 = isrght - spdiam * .25f;
  1119. } else {
  1120. tmp = f2cmin(isrght,*vu) - f2cmax(isleft,*vl);
  1121. s1 = f2cmax(isleft,*vl) + tmp * .25f;
  1122. s2 = f2cmin(isrght,*vu) - tmp * .25f;
  1123. }
  1124. }
  1125. /* Compute the negcount at the 1/4 and 3/4 points */
  1126. if (mb > 1) {
  1127. slarrc_("T", &in, &s1, &s2, &d__[ibegin], &e[ibegin], pivmin, &
  1128. cnt, &cnt1, &cnt2, &iinfo);
  1129. }
  1130. if (mb == 1) {
  1131. sigma = gl;
  1132. sgndef = 1.f;
  1133. } else if (cnt1 - indl >= indu - cnt2) {
  1134. if (irange == 1 && ! forceb) {
  1135. sigma = f2cmax(isleft,gl);
  1136. } else if (usedqd) {
  1137. /* use Gerschgorin bound as shift to get pos def matrix */
  1138. /* for dqds */
  1139. sigma = isleft;
  1140. } else {
  1141. /* use approximation of the first desired eigenvalue of the */
  1142. /* block as shift */
  1143. sigma = f2cmax(isleft,*vl);
  1144. }
  1145. sgndef = 1.f;
  1146. } else {
  1147. if (irange == 1 && ! forceb) {
  1148. sigma = f2cmin(isrght,gu);
  1149. } else if (usedqd) {
  1150. /* use Gerschgorin bound as shift to get neg def matrix */
  1151. /* for dqds */
  1152. sigma = isrght;
  1153. } else {
  1154. /* use approximation of the first desired eigenvalue of the */
  1155. /* block as shift */
  1156. sigma = f2cmin(isrght,*vu);
  1157. }
  1158. sgndef = -1.f;
  1159. }
  1160. /* An initial SIGMA has been chosen that will be used for computing */
  1161. /* T - SIGMA I = L D L^T */
  1162. /* Define the increment TAU of the shift in case the initial shift */
  1163. /* needs to be refined to obtain a factorization with not too much */
  1164. /* element growth. */
  1165. if (usedqd) {
  1166. /* The initial SIGMA was to the outer end of the spectrum */
  1167. /* the matrix is definite and we need not retreat. */
  1168. tau = spdiam * eps * *n + *pivmin * 2.f;
  1169. /* Computing MAX */
  1170. r__1 = tau, r__2 = eps * 2.f * abs(sigma);
  1171. tau = f2cmax(r__1,r__2);
  1172. } else {
  1173. if (mb > 1) {
  1174. clwdth = w[wend] + werr[wend] - w[wbegin] - werr[wbegin];
  1175. avgap = (r__1 = clwdth / (real) (wend - wbegin), abs(r__1));
  1176. if (sgndef == 1.f) {
  1177. /* Computing MAX */
  1178. r__1 = wgap[wbegin];
  1179. tau = f2cmax(r__1,avgap) * .5f;
  1180. /* Computing MAX */
  1181. r__1 = tau, r__2 = werr[wbegin];
  1182. tau = f2cmax(r__1,r__2);
  1183. } else {
  1184. /* Computing MAX */
  1185. r__1 = wgap[wend - 1];
  1186. tau = f2cmax(r__1,avgap) * .5f;
  1187. /* Computing MAX */
  1188. r__1 = tau, r__2 = werr[wend];
  1189. tau = f2cmax(r__1,r__2);
  1190. }
  1191. } else {
  1192. tau = werr[wbegin];
  1193. }
  1194. }
  1195. for (idum = 1; idum <= 6; ++idum) {
  1196. /* Compute L D L^T factorization of tridiagonal matrix T - sigma I. */
  1197. /* Store D in WORK(1:IN), L in WORK(IN+1:2*IN), and reciprocals of */
  1198. /* pivots in WORK(2*IN+1:3*IN) */
  1199. dpivot = d__[ibegin] - sigma;
  1200. work[1] = dpivot;
  1201. dmax__ = abs(work[1]);
  1202. j = ibegin;
  1203. i__2 = in - 1;
  1204. for (i__ = 1; i__ <= i__2; ++i__) {
  1205. work[(in << 1) + i__] = 1.f / work[i__];
  1206. tmp = e[j] * work[(in << 1) + i__];
  1207. work[in + i__] = tmp;
  1208. dpivot = d__[j + 1] - sigma - tmp * e[j];
  1209. work[i__ + 1] = dpivot;
  1210. /* Computing MAX */
  1211. r__1 = dmax__, r__2 = abs(dpivot);
  1212. dmax__ = f2cmax(r__1,r__2);
  1213. ++j;
  1214. /* L70: */
  1215. }
  1216. /* check for element growth */
  1217. if (dmax__ > spdiam * 64.f) {
  1218. norep = TRUE_;
  1219. } else {
  1220. norep = FALSE_;
  1221. }
  1222. if (usedqd && ! norep) {
  1223. /* Ensure the definiteness of the representation */
  1224. /* All entries of D (of L D L^T) must have the same sign */
  1225. i__2 = in;
  1226. for (i__ = 1; i__ <= i__2; ++i__) {
  1227. tmp = sgndef * work[i__];
  1228. if (tmp < 0.f) {
  1229. norep = TRUE_;
  1230. }
  1231. /* L71: */
  1232. }
  1233. }
  1234. if (norep) {
  1235. /* Note that in the case of IRANGE=ALLRNG, we use the Gerschgorin */
  1236. /* shift which makes the matrix definite. So we should end up */
  1237. /* here really only in the case of IRANGE = VALRNG or INDRNG. */
  1238. if (idum == 5) {
  1239. if (sgndef == 1.f) {
  1240. /* The fudged Gerschgorin shift should succeed */
  1241. sigma = gl - spdiam * 2.f * eps * *n - *pivmin * 4.f;
  1242. } else {
  1243. sigma = gu + spdiam * 2.f * eps * *n + *pivmin * 4.f;
  1244. }
  1245. } else {
  1246. sigma -= sgndef * tau;
  1247. tau *= 2.f;
  1248. }
  1249. } else {
  1250. /* an initial RRR is found */
  1251. goto L83;
  1252. }
  1253. /* L80: */
  1254. }
  1255. /* if the program reaches this point, no base representation could be */
  1256. /* found in MAXTRY iterations. */
  1257. *info = 2;
  1258. return;
  1259. L83:
  1260. /* At this point, we have found an initial base representation */
  1261. /* T - SIGMA I = L D L^T with not too much element growth. */
  1262. /* Store the shift. */
  1263. e[iend] = sigma;
  1264. /* Store D and L. */
  1265. scopy_(&in, &work[1], &c__1, &d__[ibegin], &c__1);
  1266. i__2 = in - 1;
  1267. scopy_(&i__2, &work[in + 1], &c__1, &e[ibegin], &c__1);
  1268. if (mb > 1) {
  1269. /* Perturb each entry of the base representation by a small */
  1270. /* (but random) relative amount to overcome difficulties with */
  1271. /* glued matrices. */
  1272. for (i__ = 1; i__ <= 4; ++i__) {
  1273. iseed[i__ - 1] = 1;
  1274. /* L122: */
  1275. }
  1276. i__2 = (in << 1) - 1;
  1277. slarnv_(&c__2, iseed, &i__2, &work[1]);
  1278. i__2 = in - 1;
  1279. for (i__ = 1; i__ <= i__2; ++i__) {
  1280. d__[ibegin + i__ - 1] *= eps * 4.f * work[i__] + 1.f;
  1281. e[ibegin + i__ - 1] *= eps * 4.f * work[in + i__] + 1.f;
  1282. /* L125: */
  1283. }
  1284. d__[iend] *= eps * 4.f * work[in] + 1.f;
  1285. }
  1286. /* Don't update the Gerschgorin intervals because keeping track */
  1287. /* of the updates would be too much work in SLARRV. */
  1288. /* We update W instead and use it to locate the proper Gerschgorin */
  1289. /* intervals. */
  1290. /* Compute the required eigenvalues of L D L' by bisection or dqds */
  1291. if (! usedqd) {
  1292. /* If SLARRD has been used, shift the eigenvalue approximations */
  1293. /* according to their representation. This is necessary for */
  1294. /* a uniform SLARRV since dqds computes eigenvalues of the */
  1295. /* shifted representation. In SLARRV, W will always hold the */
  1296. /* UNshifted eigenvalue approximation. */
  1297. i__2 = wend;
  1298. for (j = wbegin; j <= i__2; ++j) {
  1299. w[j] -= sigma;
  1300. werr[j] += (r__1 = w[j], abs(r__1)) * eps;
  1301. /* L134: */
  1302. }
  1303. /* call SLARRB to reduce eigenvalue error of the approximations */
  1304. /* from SLARRD */
  1305. i__2 = iend - 1;
  1306. for (i__ = ibegin; i__ <= i__2; ++i__) {
  1307. /* Computing 2nd power */
  1308. r__1 = e[i__];
  1309. work[i__] = d__[i__] * (r__1 * r__1);
  1310. /* L135: */
  1311. }
  1312. /* use bisection to find EV from INDL to INDU */
  1313. i__2 = indl - 1;
  1314. slarrb_(&in, &d__[ibegin], &work[ibegin], &indl, &indu, rtol1,
  1315. rtol2, &i__2, &w[wbegin], &wgap[wbegin], &werr[wbegin], &
  1316. work[(*n << 1) + 1], &iwork[1], pivmin, &spdiam, &in, &
  1317. iinfo);
  1318. if (iinfo != 0) {
  1319. *info = -4;
  1320. return;
  1321. }
  1322. /* SLARRB computes all gaps correctly except for the last one */
  1323. /* Record distance to VU/GU */
  1324. /* Computing MAX */
  1325. r__1 = 0.f, r__2 = *vu - sigma - (w[wend] + werr[wend]);
  1326. wgap[wend] = f2cmax(r__1,r__2);
  1327. i__2 = indu;
  1328. for (i__ = indl; i__ <= i__2; ++i__) {
  1329. ++(*m);
  1330. iblock[*m] = jblk;
  1331. indexw[*m] = i__;
  1332. /* L138: */
  1333. }
  1334. } else {
  1335. /* Call dqds to get all eigs (and then possibly delete unwanted */
  1336. /* eigenvalues). */
  1337. /* Note that dqds finds the eigenvalues of the L D L^T representation */
  1338. /* of T to high relative accuracy. High relative accuracy */
  1339. /* might be lost when the shift of the RRR is subtracted to obtain */
  1340. /* the eigenvalues of T. However, T is not guaranteed to define its */
  1341. /* eigenvalues to high relative accuracy anyway. */
  1342. /* Set RTOL to the order of the tolerance used in SLASQ2 */
  1343. /* This is an ESTIMATED error, the worst case bound is 4*N*EPS */
  1344. /* which is usually too large and requires unnecessary work to be */
  1345. /* done by bisection when computing the eigenvectors */
  1346. rtol = log((real) in) * 4.f * eps;
  1347. j = ibegin;
  1348. i__2 = in - 1;
  1349. for (i__ = 1; i__ <= i__2; ++i__) {
  1350. work[(i__ << 1) - 1] = (r__1 = d__[j], abs(r__1));
  1351. work[i__ * 2] = e[j] * e[j] * work[(i__ << 1) - 1];
  1352. ++j;
  1353. /* L140: */
  1354. }
  1355. work[(in << 1) - 1] = (r__1 = d__[iend], abs(r__1));
  1356. work[in * 2] = 0.f;
  1357. slasq2_(&in, &work[1], &iinfo);
  1358. if (iinfo != 0) {
  1359. /* If IINFO = -5 then an index is part of a tight cluster */
  1360. /* and should be changed. The index is in IWORK(1) and the */
  1361. /* gap is in WORK(N+1) */
  1362. *info = -5;
  1363. return;
  1364. } else {
  1365. /* Test that all eigenvalues are positive as expected */
  1366. i__2 = in;
  1367. for (i__ = 1; i__ <= i__2; ++i__) {
  1368. if (work[i__] < 0.f) {
  1369. *info = -6;
  1370. return;
  1371. }
  1372. /* L149: */
  1373. }
  1374. }
  1375. if (sgndef > 0.f) {
  1376. i__2 = indu;
  1377. for (i__ = indl; i__ <= i__2; ++i__) {
  1378. ++(*m);
  1379. w[*m] = work[in - i__ + 1];
  1380. iblock[*m] = jblk;
  1381. indexw[*m] = i__;
  1382. /* L150: */
  1383. }
  1384. } else {
  1385. i__2 = indu;
  1386. for (i__ = indl; i__ <= i__2; ++i__) {
  1387. ++(*m);
  1388. w[*m] = -work[i__];
  1389. iblock[*m] = jblk;
  1390. indexw[*m] = i__;
  1391. /* L160: */
  1392. }
  1393. }
  1394. i__2 = *m;
  1395. for (i__ = *m - mb + 1; i__ <= i__2; ++i__) {
  1396. /* the value of RTOL below should be the tolerance in SLASQ2 */
  1397. werr[i__] = rtol * (r__1 = w[i__], abs(r__1));
  1398. /* L165: */
  1399. }
  1400. i__2 = *m - 1;
  1401. for (i__ = *m - mb + 1; i__ <= i__2; ++i__) {
  1402. /* compute the right gap between the intervals */
  1403. /* Computing MAX */
  1404. r__1 = 0.f, r__2 = w[i__ + 1] - werr[i__ + 1] - (w[i__] +
  1405. werr[i__]);
  1406. wgap[i__] = f2cmax(r__1,r__2);
  1407. /* L166: */
  1408. }
  1409. /* Computing MAX */
  1410. r__1 = 0.f, r__2 = *vu - sigma - (w[*m] + werr[*m]);
  1411. wgap[*m] = f2cmax(r__1,r__2);
  1412. }
  1413. /* proceed with next block */
  1414. ibegin = iend + 1;
  1415. wbegin = wend + 1;
  1416. L170:
  1417. ;
  1418. }
  1419. return;
  1420. /* end of SLARRE */
  1421. } /* slarre_ */