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.

slarrf.c 30 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029
  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. /* > \brief \b SLARRF finds a new relatively robust representation such that at least one of the eigenvalues i
  488. s relatively isolated. */
  489. /* =========== DOCUMENTATION =========== */
  490. /* Online html documentation available at */
  491. /* http://www.netlib.org/lapack/explore-html/ */
  492. /* > \htmlonly */
  493. /* > Download SLARRF + dependencies */
  494. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/slarrf.
  495. f"> */
  496. /* > [TGZ]</a> */
  497. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/slarrf.
  498. f"> */
  499. /* > [ZIP]</a> */
  500. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/slarrf.
  501. f"> */
  502. /* > [TXT]</a> */
  503. /* > \endhtmlonly */
  504. /* Definition: */
  505. /* =========== */
  506. /* SUBROUTINE SLARRF( N, D, L, LD, CLSTRT, CLEND, */
  507. /* W, WGAP, WERR, */
  508. /* SPDIAM, CLGAPL, CLGAPR, PIVMIN, SIGMA, */
  509. /* DPLUS, LPLUS, WORK, INFO ) */
  510. /* INTEGER CLSTRT, CLEND, INFO, N */
  511. /* REAL CLGAPL, CLGAPR, PIVMIN, SIGMA, SPDIAM */
  512. /* REAL D( * ), DPLUS( * ), L( * ), LD( * ), */
  513. /* $ LPLUS( * ), W( * ), WGAP( * ), WERR( * ), WORK( * ) */
  514. /* > \par Purpose: */
  515. /* ============= */
  516. /* > */
  517. /* > \verbatim */
  518. /* > */
  519. /* > Given the initial representation L D L^T and its cluster of close */
  520. /* > eigenvalues (in a relative measure), W( CLSTRT ), W( CLSTRT+1 ), ... */
  521. /* > W( CLEND ), SLARRF finds a new relatively robust representation */
  522. /* > L D L^T - SIGMA I = L(+) D(+) L(+)^T such that at least one of the */
  523. /* > eigenvalues of L(+) D(+) L(+)^T is relatively isolated. */
  524. /* > \endverbatim */
  525. /* Arguments: */
  526. /* ========== */
  527. /* > \param[in] N */
  528. /* > \verbatim */
  529. /* > N is INTEGER */
  530. /* > The order of the matrix (subblock, if the matrix split). */
  531. /* > \endverbatim */
  532. /* > */
  533. /* > \param[in] D */
  534. /* > \verbatim */
  535. /* > D is REAL array, dimension (N) */
  536. /* > The N diagonal elements of the diagonal matrix D. */
  537. /* > \endverbatim */
  538. /* > */
  539. /* > \param[in] L */
  540. /* > \verbatim */
  541. /* > L is REAL array, dimension (N-1) */
  542. /* > The (N-1) subdiagonal elements of the unit bidiagonal */
  543. /* > matrix L. */
  544. /* > \endverbatim */
  545. /* > */
  546. /* > \param[in] LD */
  547. /* > \verbatim */
  548. /* > LD is REAL array, dimension (N-1) */
  549. /* > The (N-1) elements L(i)*D(i). */
  550. /* > \endverbatim */
  551. /* > */
  552. /* > \param[in] CLSTRT */
  553. /* > \verbatim */
  554. /* > CLSTRT is INTEGER */
  555. /* > The index of the first eigenvalue in the cluster. */
  556. /* > \endverbatim */
  557. /* > */
  558. /* > \param[in] CLEND */
  559. /* > \verbatim */
  560. /* > CLEND is INTEGER */
  561. /* > The index of the last eigenvalue in the cluster. */
  562. /* > \endverbatim */
  563. /* > */
  564. /* > \param[in] W */
  565. /* > \verbatim */
  566. /* > W is REAL array, dimension */
  567. /* > dimension is >= (CLEND-CLSTRT+1) */
  568. /* > The eigenvalue APPROXIMATIONS of L D L^T in ascending order. */
  569. /* > W( CLSTRT ) through W( CLEND ) form the cluster of relatively */
  570. /* > close eigenalues. */
  571. /* > \endverbatim */
  572. /* > */
  573. /* > \param[in,out] WGAP */
  574. /* > \verbatim */
  575. /* > WGAP is REAL array, dimension */
  576. /* > dimension is >= (CLEND-CLSTRT+1) */
  577. /* > The separation from the right neighbor eigenvalue in W. */
  578. /* > \endverbatim */
  579. /* > */
  580. /* > \param[in] WERR */
  581. /* > \verbatim */
  582. /* > WERR is REAL array, dimension */
  583. /* > dimension is >= (CLEND-CLSTRT+1) */
  584. /* > WERR contain the semiwidth of the uncertainty */
  585. /* > interval of the corresponding eigenvalue APPROXIMATION in W */
  586. /* > \endverbatim */
  587. /* > */
  588. /* > \param[in] SPDIAM */
  589. /* > \verbatim */
  590. /* > SPDIAM is REAL */
  591. /* > estimate of the spectral diameter obtained from the */
  592. /* > Gerschgorin intervals */
  593. /* > \endverbatim */
  594. /* > */
  595. /* > \param[in] CLGAPL */
  596. /* > \verbatim */
  597. /* > CLGAPL is REAL */
  598. /* > \endverbatim */
  599. /* > */
  600. /* > \param[in] CLGAPR */
  601. /* > \verbatim */
  602. /* > CLGAPR is REAL */
  603. /* > absolute gap on each end of the cluster. */
  604. /* > Set by the calling routine to protect against shifts too close */
  605. /* > to eigenvalues outside the cluster. */
  606. /* > \endverbatim */
  607. /* > */
  608. /* > \param[in] PIVMIN */
  609. /* > \verbatim */
  610. /* > PIVMIN is REAL */
  611. /* > The minimum pivot allowed in the Sturm sequence. */
  612. /* > \endverbatim */
  613. /* > */
  614. /* > \param[out] SIGMA */
  615. /* > \verbatim */
  616. /* > SIGMA is REAL */
  617. /* > The shift used to form L(+) D(+) L(+)^T. */
  618. /* > \endverbatim */
  619. /* > */
  620. /* > \param[out] DPLUS */
  621. /* > \verbatim */
  622. /* > DPLUS is REAL array, dimension (N) */
  623. /* > The N diagonal elements of the diagonal matrix D(+). */
  624. /* > \endverbatim */
  625. /* > */
  626. /* > \param[out] LPLUS */
  627. /* > \verbatim */
  628. /* > LPLUS is REAL array, dimension (N-1) */
  629. /* > The first (N-1) elements of LPLUS contain the subdiagonal */
  630. /* > elements of the unit bidiagonal matrix L(+). */
  631. /* > \endverbatim */
  632. /* > */
  633. /* > \param[out] WORK */
  634. /* > \verbatim */
  635. /* > WORK is REAL array, dimension (2*N) */
  636. /* > Workspace. */
  637. /* > \endverbatim */
  638. /* > */
  639. /* > \param[out] INFO */
  640. /* > \verbatim */
  641. /* > INFO is INTEGER */
  642. /* > Signals processing OK (=0) or failure (=1) */
  643. /* > \endverbatim */
  644. /* Authors: */
  645. /* ======== */
  646. /* > \author Univ. of Tennessee */
  647. /* > \author Univ. of California Berkeley */
  648. /* > \author Univ. of Colorado Denver */
  649. /* > \author NAG Ltd. */
  650. /* > \date June 2016 */
  651. /* > \ingroup OTHERauxiliary */
  652. /* > \par Contributors: */
  653. /* ================== */
  654. /* > */
  655. /* > Beresford Parlett, University of California, Berkeley, USA \n */
  656. /* > Jim Demmel, University of California, Berkeley, USA \n */
  657. /* > Inderjit Dhillon, University of Texas, Austin, USA \n */
  658. /* > Osni Marques, LBNL/NERSC, USA \n */
  659. /* > Christof Voemel, University of California, Berkeley, USA */
  660. /* ===================================================================== */
  661. /* Subroutine */ void slarrf_(integer *n, real *d__, real *l, real *ld,
  662. integer *clstrt, integer *clend, real *w, real *wgap, real *werr,
  663. real *spdiam, real *clgapl, real *clgapr, real *pivmin, real *sigma,
  664. real *dplus, real *lplus, real *work, integer *info)
  665. {
  666. /* System generated locals */
  667. integer i__1;
  668. real r__1, r__2, r__3;
  669. /* Local variables */
  670. real growthbound, fail, fact, oldp;
  671. integer indx;
  672. real prod;
  673. integer ktry;
  674. real fail2;
  675. integer i__;
  676. real s, avgap, ldmax, rdmax;
  677. integer shift;
  678. extern /* Subroutine */ void scopy_(integer *, real *, integer *, real *,
  679. integer *);
  680. real bestshift, smlgrowth;
  681. logical dorrr1;
  682. real ldelta;
  683. extern real slamch_(char *);
  684. logical nofail;
  685. real mingap, lsigma, rdelta;
  686. logical forcer;
  687. real rsigma, clwdth;
  688. extern logical sisnan_(real *);
  689. logical sawnan1, sawnan2;
  690. real eps, tmp;
  691. logical tryrrr1;
  692. real max1, max2, rrr1, rrr2, znm2;
  693. /* -- LAPACK auxiliary routine (version 3.7.1) -- */
  694. /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
  695. /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
  696. /* June 2016 */
  697. /* ===================================================================== */
  698. /* Parameter adjustments */
  699. --work;
  700. --lplus;
  701. --dplus;
  702. --werr;
  703. --wgap;
  704. --w;
  705. --ld;
  706. --l;
  707. --d__;
  708. /* Function Body */
  709. *info = 0;
  710. /* Quick return if possible */
  711. if (*n <= 0) {
  712. return;
  713. }
  714. fact = 2.f;
  715. eps = slamch_("Precision");
  716. shift = 0;
  717. forcer = FALSE_;
  718. /* Note that we cannot guarantee that for any of the shifts tried, */
  719. /* the factorization has a small or even moderate element growth. */
  720. /* There could be Ritz values at both ends of the cluster and despite */
  721. /* backing off, there are examples where all factorizations tried */
  722. /* (in IEEE mode, allowing zero pivots & infinities) have INFINITE */
  723. /* element growth. */
  724. /* For this reason, we should use PIVMIN in this subroutine so that at */
  725. /* least the L D L^T factorization exists. It can be checked afterwards */
  726. /* whether the element growth caused bad residuals/orthogonality. */
  727. /* Decide whether the code should accept the best among all */
  728. /* representations despite large element growth or signal INFO=1 */
  729. /* Setting NOFAIL to .FALSE. for quick fix for bug 113 */
  730. nofail = FALSE_;
  731. /* Compute the average gap length of the cluster */
  732. clwdth = (r__1 = w[*clend] - w[*clstrt], abs(r__1)) + werr[*clend] + werr[
  733. *clstrt];
  734. avgap = clwdth / (real) (*clend - *clstrt);
  735. mingap = f2cmin(*clgapl,*clgapr);
  736. /* Initial values for shifts to both ends of cluster */
  737. /* Computing MIN */
  738. r__1 = w[*clstrt], r__2 = w[*clend];
  739. lsigma = f2cmin(r__1,r__2) - werr[*clstrt];
  740. /* Computing MAX */
  741. r__1 = w[*clstrt], r__2 = w[*clend];
  742. rsigma = f2cmax(r__1,r__2) + werr[*clend];
  743. /* Use a small fudge to make sure that we really shift to the outside */
  744. lsigma -= abs(lsigma) * 2.f * eps;
  745. rsigma += abs(rsigma) * 2.f * eps;
  746. /* Compute upper bounds for how much to back off the initial shifts */
  747. ldmax = mingap * .25f + *pivmin * 2.f;
  748. rdmax = mingap * .25f + *pivmin * 2.f;
  749. /* Computing MAX */
  750. r__1 = avgap, r__2 = wgap[*clstrt];
  751. ldelta = f2cmax(r__1,r__2) / fact;
  752. /* Computing MAX */
  753. r__1 = avgap, r__2 = wgap[*clend - 1];
  754. rdelta = f2cmax(r__1,r__2) / fact;
  755. /* Initialize the record of the best representation found */
  756. s = slamch_("S");
  757. smlgrowth = 1.f / s;
  758. fail = (real) (*n - 1) * mingap / (*spdiam * eps);
  759. fail2 = (real) (*n - 1) * mingap / (*spdiam * sqrt(eps));
  760. bestshift = lsigma;
  761. /* while (KTRY <= KTRYMAX) */
  762. ktry = 0;
  763. growthbound = *spdiam * 8.f;
  764. L5:
  765. sawnan1 = FALSE_;
  766. sawnan2 = FALSE_;
  767. /* Ensure that we do not back off too much of the initial shifts */
  768. ldelta = f2cmin(ldmax,ldelta);
  769. rdelta = f2cmin(rdmax,rdelta);
  770. /* Compute the element growth when shifting to both ends of the cluster */
  771. /* accept the shift if there is no element growth at one of the two ends */
  772. /* Left end */
  773. s = -lsigma;
  774. dplus[1] = d__[1] + s;
  775. if (abs(dplus[1]) < *pivmin) {
  776. dplus[1] = -(*pivmin);
  777. /* Need to set SAWNAN1 because refined RRR test should not be used */
  778. /* in this case */
  779. sawnan1 = TRUE_;
  780. }
  781. max1 = abs(dplus[1]);
  782. i__1 = *n - 1;
  783. for (i__ = 1; i__ <= i__1; ++i__) {
  784. lplus[i__] = ld[i__] / dplus[i__];
  785. s = s * lplus[i__] * l[i__] - lsigma;
  786. dplus[i__ + 1] = d__[i__ + 1] + s;
  787. if ((r__1 = dplus[i__ + 1], abs(r__1)) < *pivmin) {
  788. dplus[i__ + 1] = -(*pivmin);
  789. /* Need to set SAWNAN1 because refined RRR test should not be used */
  790. /* in this case */
  791. sawnan1 = TRUE_;
  792. }
  793. /* Computing MAX */
  794. r__2 = max1, r__3 = (r__1 = dplus[i__ + 1], abs(r__1));
  795. max1 = f2cmax(r__2,r__3);
  796. /* L6: */
  797. }
  798. sawnan1 = sawnan1 || sisnan_(&max1);
  799. if (forcer || max1 <= growthbound && ! sawnan1) {
  800. *sigma = lsigma;
  801. shift = 1;
  802. goto L100;
  803. }
  804. /* Right end */
  805. s = -rsigma;
  806. work[1] = d__[1] + s;
  807. if (abs(work[1]) < *pivmin) {
  808. work[1] = -(*pivmin);
  809. /* Need to set SAWNAN2 because refined RRR test should not be used */
  810. /* in this case */
  811. sawnan2 = TRUE_;
  812. }
  813. max2 = abs(work[1]);
  814. i__1 = *n - 1;
  815. for (i__ = 1; i__ <= i__1; ++i__) {
  816. work[*n + i__] = ld[i__] / work[i__];
  817. s = s * work[*n + i__] * l[i__] - rsigma;
  818. work[i__ + 1] = d__[i__ + 1] + s;
  819. if ((r__1 = work[i__ + 1], abs(r__1)) < *pivmin) {
  820. work[i__ + 1] = -(*pivmin);
  821. /* Need to set SAWNAN2 because refined RRR test should not be used */
  822. /* in this case */
  823. sawnan2 = TRUE_;
  824. }
  825. /* Computing MAX */
  826. r__2 = max2, r__3 = (r__1 = work[i__ + 1], abs(r__1));
  827. max2 = f2cmax(r__2,r__3);
  828. /* L7: */
  829. }
  830. sawnan2 = sawnan2 || sisnan_(&max2);
  831. if (forcer || max2 <= growthbound && ! sawnan2) {
  832. *sigma = rsigma;
  833. shift = 2;
  834. goto L100;
  835. }
  836. /* If we are at this point, both shifts led to too much element growth */
  837. /* Record the better of the two shifts (provided it didn't lead to NaN) */
  838. if (sawnan1 && sawnan2) {
  839. /* both MAX1 and MAX2 are NaN */
  840. goto L50;
  841. } else {
  842. if (! sawnan1) {
  843. indx = 1;
  844. if (max1 <= smlgrowth) {
  845. smlgrowth = max1;
  846. bestshift = lsigma;
  847. }
  848. }
  849. if (! sawnan2) {
  850. if (sawnan1 || max2 <= max1) {
  851. indx = 2;
  852. }
  853. if (max2 <= smlgrowth) {
  854. smlgrowth = max2;
  855. bestshift = rsigma;
  856. }
  857. }
  858. }
  859. /* If we are here, both the left and the right shift led to */
  860. /* element growth. If the element growth is moderate, then */
  861. /* we may still accept the representation, if it passes a */
  862. /* refined test for RRR. This test supposes that no NaN occurred. */
  863. /* Moreover, we use the refined RRR test only for isolated clusters. */
  864. if (clwdth < mingap / 128.f && f2cmin(max1,max2) < fail2 && ! sawnan1 && !
  865. sawnan2) {
  866. dorrr1 = TRUE_;
  867. } else {
  868. dorrr1 = FALSE_;
  869. }
  870. tryrrr1 = TRUE_;
  871. if (tryrrr1 && dorrr1) {
  872. if (indx == 1) {
  873. tmp = (r__1 = dplus[*n], abs(r__1));
  874. znm2 = 1.f;
  875. prod = 1.f;
  876. oldp = 1.f;
  877. for (i__ = *n - 1; i__ >= 1; --i__) {
  878. if (prod <= eps) {
  879. prod = dplus[i__ + 1] * work[*n + i__ + 1] / (dplus[i__] *
  880. work[*n + i__]) * oldp;
  881. } else {
  882. prod *= (r__1 = work[*n + i__], abs(r__1));
  883. }
  884. oldp = prod;
  885. /* Computing 2nd power */
  886. r__1 = prod;
  887. znm2 += r__1 * r__1;
  888. /* Computing MAX */
  889. r__2 = tmp, r__3 = (r__1 = dplus[i__] * prod, abs(r__1));
  890. tmp = f2cmax(r__2,r__3);
  891. /* L15: */
  892. }
  893. rrr1 = tmp / (*spdiam * sqrt(znm2));
  894. if (rrr1 <= 8.f) {
  895. *sigma = lsigma;
  896. shift = 1;
  897. goto L100;
  898. }
  899. } else if (indx == 2) {
  900. tmp = (r__1 = work[*n], abs(r__1));
  901. znm2 = 1.f;
  902. prod = 1.f;
  903. oldp = 1.f;
  904. for (i__ = *n - 1; i__ >= 1; --i__) {
  905. if (prod <= eps) {
  906. prod = work[i__ + 1] * lplus[i__ + 1] / (work[i__] *
  907. lplus[i__]) * oldp;
  908. } else {
  909. prod *= (r__1 = lplus[i__], abs(r__1));
  910. }
  911. oldp = prod;
  912. /* Computing 2nd power */
  913. r__1 = prod;
  914. znm2 += r__1 * r__1;
  915. /* Computing MAX */
  916. r__2 = tmp, r__3 = (r__1 = work[i__] * prod, abs(r__1));
  917. tmp = f2cmax(r__2,r__3);
  918. /* L16: */
  919. }
  920. rrr2 = tmp / (*spdiam * sqrt(znm2));
  921. if (rrr2 <= 8.f) {
  922. *sigma = rsigma;
  923. shift = 2;
  924. goto L100;
  925. }
  926. }
  927. }
  928. L50:
  929. if (ktry < 1) {
  930. /* If we are here, both shifts failed also the RRR test. */
  931. /* Back off to the outside */
  932. /* Computing MAX */
  933. r__1 = lsigma - ldelta, r__2 = lsigma - ldmax;
  934. lsigma = f2cmax(r__1,r__2);
  935. /* Computing MIN */
  936. r__1 = rsigma + rdelta, r__2 = rsigma + rdmax;
  937. rsigma = f2cmin(r__1,r__2);
  938. ldelta *= 2.f;
  939. rdelta *= 2.f;
  940. ++ktry;
  941. goto L5;
  942. } else {
  943. /* None of the representations investigated satisfied our */
  944. /* criteria. Take the best one we found. */
  945. if (smlgrowth < fail || nofail) {
  946. lsigma = bestshift;
  947. rsigma = bestshift;
  948. forcer = TRUE_;
  949. goto L5;
  950. } else {
  951. *info = 1;
  952. return;
  953. }
  954. }
  955. L100:
  956. if (shift == 1) {
  957. } else if (shift == 2) {
  958. /* store new L and D back into DPLUS, LPLUS */
  959. scopy_(n, &work[1], &c__1, &dplus[1], &c__1);
  960. i__1 = *n - 1;
  961. scopy_(&i__1, &work[*n + 1], &c__1, &lplus[1], &c__1);
  962. }
  963. return;
  964. /* End of SLARRF */
  965. } /* slarrf_ */