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.

dlatbs.c 39 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422
  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 doublereal c_b36 = .5;
  488. /* > \brief \b DLATBS solves a triangular banded system of equations. */
  489. /* =========== DOCUMENTATION =========== */
  490. /* Online html documentation available at */
  491. /* http://www.netlib.org/lapack/explore-html/ */
  492. /* > \htmlonly */
  493. /* > Download DLATBS + dependencies */
  494. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlatbs.
  495. f"> */
  496. /* > [TGZ]</a> */
  497. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlatbs.
  498. f"> */
  499. /* > [ZIP]</a> */
  500. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlatbs.
  501. f"> */
  502. /* > [TXT]</a> */
  503. /* > \endhtmlonly */
  504. /* Definition: */
  505. /* =========== */
  506. /* SUBROUTINE DLATBS( UPLO, TRANS, DIAG, NORMIN, N, KD, AB, LDAB, X, */
  507. /* SCALE, CNORM, INFO ) */
  508. /* CHARACTER DIAG, NORMIN, TRANS, UPLO */
  509. /* INTEGER INFO, KD, LDAB, N */
  510. /* DOUBLE PRECISION SCALE */
  511. /* DOUBLE PRECISION AB( LDAB, * ), CNORM( * ), X( * ) */
  512. /* > \par Purpose: */
  513. /* ============= */
  514. /* > */
  515. /* > \verbatim */
  516. /* > */
  517. /* > DLATBS solves one of the triangular systems */
  518. /* > */
  519. /* > A *x = s*b or A**T*x = s*b */
  520. /* > */
  521. /* > with scaling to prevent overflow, where A is an upper or lower */
  522. /* > triangular band matrix. Here A**T denotes the transpose of A, x and b */
  523. /* > are n-element vectors, and s is a scaling factor, usually less than */
  524. /* > or equal to 1, chosen so that the components of x will be less than */
  525. /* > the overflow threshold. If the unscaled problem will not cause */
  526. /* > overflow, the Level 2 BLAS routine DTBSV is called. If the matrix A */
  527. /* > is singular (A(j,j) = 0 for some j), then s is set to 0 and a */
  528. /* > non-trivial solution to A*x = 0 is returned. */
  529. /* > \endverbatim */
  530. /* Arguments: */
  531. /* ========== */
  532. /* > \param[in] UPLO */
  533. /* > \verbatim */
  534. /* > UPLO is CHARACTER*1 */
  535. /* > Specifies whether the matrix A is upper or lower triangular. */
  536. /* > = 'U': Upper triangular */
  537. /* > = 'L': Lower triangular */
  538. /* > \endverbatim */
  539. /* > */
  540. /* > \param[in] TRANS */
  541. /* > \verbatim */
  542. /* > TRANS is CHARACTER*1 */
  543. /* > Specifies the operation applied to A. */
  544. /* > = 'N': Solve A * x = s*b (No transpose) */
  545. /* > = 'T': Solve A**T* x = s*b (Transpose) */
  546. /* > = 'C': Solve A**T* x = s*b (Conjugate transpose = Transpose) */
  547. /* > \endverbatim */
  548. /* > */
  549. /* > \param[in] DIAG */
  550. /* > \verbatim */
  551. /* > DIAG is CHARACTER*1 */
  552. /* > Specifies whether or not the matrix A is unit triangular. */
  553. /* > = 'N': Non-unit triangular */
  554. /* > = 'U': Unit triangular */
  555. /* > \endverbatim */
  556. /* > */
  557. /* > \param[in] NORMIN */
  558. /* > \verbatim */
  559. /* > NORMIN is CHARACTER*1 */
  560. /* > Specifies whether CNORM has been set or not. */
  561. /* > = 'Y': CNORM contains the column norms on entry */
  562. /* > = 'N': CNORM is not set on entry. On exit, the norms will */
  563. /* > be computed and stored in CNORM. */
  564. /* > \endverbatim */
  565. /* > */
  566. /* > \param[in] N */
  567. /* > \verbatim */
  568. /* > N is INTEGER */
  569. /* > The order of the matrix A. N >= 0. */
  570. /* > \endverbatim */
  571. /* > */
  572. /* > \param[in] KD */
  573. /* > \verbatim */
  574. /* > KD is INTEGER */
  575. /* > The number of subdiagonals or superdiagonals in the */
  576. /* > triangular matrix A. KD >= 0. */
  577. /* > \endverbatim */
  578. /* > */
  579. /* > \param[in] AB */
  580. /* > \verbatim */
  581. /* > AB is DOUBLE PRECISION array, dimension (LDAB,N) */
  582. /* > The upper or lower triangular band matrix A, stored in the */
  583. /* > first KD+1 rows of the array. The j-th column of A is stored */
  584. /* > in the j-th column of the array AB as follows: */
  585. /* > if UPLO = 'U', AB(kd+1+i-j,j) = A(i,j) for f2cmax(1,j-kd)<=i<=j; */
  586. /* > if UPLO = 'L', AB(1+i-j,j) = A(i,j) for j<=i<=f2cmin(n,j+kd). */
  587. /* > \endverbatim */
  588. /* > */
  589. /* > \param[in] LDAB */
  590. /* > \verbatim */
  591. /* > LDAB is INTEGER */
  592. /* > The leading dimension of the array AB. LDAB >= KD+1. */
  593. /* > \endverbatim */
  594. /* > */
  595. /* > \param[in,out] X */
  596. /* > \verbatim */
  597. /* > X is DOUBLE PRECISION array, dimension (N) */
  598. /* > On entry, the right hand side b of the triangular system. */
  599. /* > On exit, X is overwritten by the solution vector x. */
  600. /* > \endverbatim */
  601. /* > */
  602. /* > \param[out] SCALE */
  603. /* > \verbatim */
  604. /* > SCALE is DOUBLE PRECISION */
  605. /* > The scaling factor s for the triangular system */
  606. /* > A * x = s*b or A**T* x = s*b. */
  607. /* > If SCALE = 0, the matrix A is singular or badly scaled, and */
  608. /* > the vector x is an exact or approximate solution to A*x = 0. */
  609. /* > \endverbatim */
  610. /* > */
  611. /* > \param[in,out] CNORM */
  612. /* > \verbatim */
  613. /* > CNORM is DOUBLE PRECISION array, dimension (N) */
  614. /* > */
  615. /* > If NORMIN = 'Y', CNORM is an input argument and CNORM(j) */
  616. /* > contains the norm of the off-diagonal part of the j-th column */
  617. /* > of A. If TRANS = 'N', CNORM(j) must be greater than or equal */
  618. /* > to the infinity-norm, and if TRANS = 'T' or 'C', CNORM(j) */
  619. /* > must be greater than or equal to the 1-norm. */
  620. /* > */
  621. /* > If NORMIN = 'N', CNORM is an output argument and CNORM(j) */
  622. /* > returns the 1-norm of the offdiagonal part of the j-th column */
  623. /* > of A. */
  624. /* > \endverbatim */
  625. /* > */
  626. /* > \param[out] INFO */
  627. /* > \verbatim */
  628. /* > INFO is INTEGER */
  629. /* > = 0: successful exit */
  630. /* > < 0: if INFO = -k, the k-th argument had an illegal value */
  631. /* > \endverbatim */
  632. /* Authors: */
  633. /* ======== */
  634. /* > \author Univ. of Tennessee */
  635. /* > \author Univ. of California Berkeley */
  636. /* > \author Univ. of Colorado Denver */
  637. /* > \author NAG Ltd. */
  638. /* > \date December 2016 */
  639. /* > \ingroup doubleOTHERauxiliary */
  640. /* > \par Further Details: */
  641. /* ===================== */
  642. /* > */
  643. /* > \verbatim */
  644. /* > */
  645. /* > A rough bound on x is computed; if that is less than overflow, DTBSV */
  646. /* > is called, otherwise, specific code is used which checks for possible */
  647. /* > overflow or divide-by-zero at every operation. */
  648. /* > */
  649. /* > A columnwise scheme is used for solving A*x = b. The basic algorithm */
  650. /* > if A is lower triangular is */
  651. /* > */
  652. /* > x[1:n] := b[1:n] */
  653. /* > for j = 1, ..., n */
  654. /* > x(j) := x(j) / A(j,j) */
  655. /* > x[j+1:n] := x[j+1:n] - x(j) * A[j+1:n,j] */
  656. /* > end */
  657. /* > */
  658. /* > Define bounds on the components of x after j iterations of the loop: */
  659. /* > M(j) = bound on x[1:j] */
  660. /* > G(j) = bound on x[j+1:n] */
  661. /* > Initially, let M(0) = 0 and G(0) = f2cmax{x(i), i=1,...,n}. */
  662. /* > */
  663. /* > Then for iteration j+1 we have */
  664. /* > M(j+1) <= G(j) / | A(j+1,j+1) | */
  665. /* > G(j+1) <= G(j) + M(j+1) * | A[j+2:n,j+1] | */
  666. /* > <= G(j) ( 1 + CNORM(j+1) / | A(j+1,j+1) | ) */
  667. /* > */
  668. /* > where CNORM(j+1) is greater than or equal to the infinity-norm of */
  669. /* > column j+1 of A, not counting the diagonal. Hence */
  670. /* > */
  671. /* > G(j) <= G(0) product ( 1 + CNORM(i) / | A(i,i) | ) */
  672. /* > 1<=i<=j */
  673. /* > and */
  674. /* > */
  675. /* > |x(j)| <= ( G(0) / |A(j,j)| ) product ( 1 + CNORM(i) / |A(i,i)| ) */
  676. /* > 1<=i< j */
  677. /* > */
  678. /* > Since |x(j)| <= M(j), we use the Level 2 BLAS routine DTBSV if the */
  679. /* > reciprocal of the largest M(j), j=1,..,n, is larger than */
  680. /* > f2cmax(underflow, 1/overflow). */
  681. /* > */
  682. /* > The bound on x(j) is also used to determine when a step in the */
  683. /* > columnwise method can be performed without fear of overflow. If */
  684. /* > the computed bound is greater than a large constant, x is scaled to */
  685. /* > prevent overflow, but if the bound overflows, x is set to 0, x(j) to */
  686. /* > 1, and scale to 0, and a non-trivial solution to A*x = 0 is found. */
  687. /* > */
  688. /* > Similarly, a row-wise scheme is used to solve A**T*x = b. The basic */
  689. /* > algorithm for A upper triangular is */
  690. /* > */
  691. /* > for j = 1, ..., n */
  692. /* > x(j) := ( b(j) - A[1:j-1,j]**T * x[1:j-1] ) / A(j,j) */
  693. /* > end */
  694. /* > */
  695. /* > We simultaneously compute two bounds */
  696. /* > G(j) = bound on ( b(i) - A[1:i-1,i]**T * x[1:i-1] ), 1<=i<=j */
  697. /* > M(j) = bound on x(i), 1<=i<=j */
  698. /* > */
  699. /* > The initial values are G(0) = 0, M(0) = f2cmax{b(i), i=1,..,n}, and we */
  700. /* > add the constraint G(j) >= G(j-1) and M(j) >= M(j-1) for j >= 1. */
  701. /* > Then the bound on x(j) is */
  702. /* > */
  703. /* > M(j) <= M(j-1) * ( 1 + CNORM(j) ) / | A(j,j) | */
  704. /* > */
  705. /* > <= M(0) * product ( ( 1 + CNORM(i) ) / |A(i,i)| ) */
  706. /* > 1<=i<=j */
  707. /* > */
  708. /* > and we can safely call DTBSV if 1/M(n) and 1/G(n) are both greater */
  709. /* > than f2cmax(underflow, 1/overflow). */
  710. /* > \endverbatim */
  711. /* > */
  712. /* ===================================================================== */
  713. /* Subroutine */ int dlatbs_(char *uplo, char *trans, char *diag, char *
  714. normin, integer *n, integer *kd, doublereal *ab, integer *ldab,
  715. doublereal *x, doublereal *scale, doublereal *cnorm, integer *info)
  716. {
  717. /* System generated locals */
  718. integer ab_dim1, ab_offset, i__1, i__2, i__3, i__4;
  719. doublereal d__1, d__2, d__3;
  720. /* Local variables */
  721. integer jinc, jlen;
  722. extern doublereal ddot_(integer *, doublereal *, integer *, doublereal *,
  723. integer *);
  724. doublereal xbnd;
  725. integer imax;
  726. doublereal tmax, tjjs, xmax, grow, sumj;
  727. integer i__, j;
  728. extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
  729. integer *);
  730. integer maind;
  731. extern logical lsame_(char *, char *);
  732. doublereal tscal, uscal;
  733. extern doublereal dasum_(integer *, doublereal *, integer *);
  734. integer jlast;
  735. extern /* Subroutine */ int dtbsv_(char *, char *, char *, integer *,
  736. integer *, doublereal *, integer *, doublereal *, integer *), daxpy_(integer *, doublereal *,
  737. doublereal *, integer *, doublereal *, integer *);
  738. logical upper;
  739. extern doublereal dlamch_(char *);
  740. doublereal xj;
  741. extern integer idamax_(integer *, doublereal *, integer *);
  742. extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
  743. doublereal bignum;
  744. logical notran;
  745. integer jfirst;
  746. doublereal smlnum;
  747. logical nounit;
  748. doublereal rec, tjj;
  749. /* -- LAPACK auxiliary routine (version 3.7.0) -- */
  750. /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
  751. /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
  752. /* December 2016 */
  753. /* ===================================================================== */
  754. /* Parameter adjustments */
  755. ab_dim1 = *ldab;
  756. ab_offset = 1 + ab_dim1 * 1;
  757. ab -= ab_offset;
  758. --x;
  759. --cnorm;
  760. /* Function Body */
  761. *info = 0;
  762. upper = lsame_(uplo, "U");
  763. notran = lsame_(trans, "N");
  764. nounit = lsame_(diag, "N");
  765. /* Test the input parameters. */
  766. if (! upper && ! lsame_(uplo, "L")) {
  767. *info = -1;
  768. } else if (! notran && ! lsame_(trans, "T") && !
  769. lsame_(trans, "C")) {
  770. *info = -2;
  771. } else if (! nounit && ! lsame_(diag, "U")) {
  772. *info = -3;
  773. } else if (! lsame_(normin, "Y") && ! lsame_(normin,
  774. "N")) {
  775. *info = -4;
  776. } else if (*n < 0) {
  777. *info = -5;
  778. } else if (*kd < 0) {
  779. *info = -6;
  780. } else if (*ldab < *kd + 1) {
  781. *info = -8;
  782. }
  783. if (*info != 0) {
  784. i__1 = -(*info);
  785. xerbla_("DLATBS", &i__1, (ftnlen)6);
  786. return 0;
  787. }
  788. /* Quick return if possible */
  789. if (*n == 0) {
  790. return 0;
  791. }
  792. /* Determine machine dependent parameters to control overflow. */
  793. smlnum = dlamch_("Safe minimum") / dlamch_("Precision");
  794. bignum = 1. / smlnum;
  795. *scale = 1.;
  796. if (lsame_(normin, "N")) {
  797. /* Compute the 1-norm of each column, not including the diagonal. */
  798. if (upper) {
  799. /* A is upper triangular. */
  800. i__1 = *n;
  801. for (j = 1; j <= i__1; ++j) {
  802. /* Computing MIN */
  803. i__2 = *kd, i__3 = j - 1;
  804. jlen = f2cmin(i__2,i__3);
  805. cnorm[j] = dasum_(&jlen, &ab[*kd + 1 - jlen + j * ab_dim1], &
  806. c__1);
  807. /* L10: */
  808. }
  809. } else {
  810. /* A is lower triangular. */
  811. i__1 = *n;
  812. for (j = 1; j <= i__1; ++j) {
  813. /* Computing MIN */
  814. i__2 = *kd, i__3 = *n - j;
  815. jlen = f2cmin(i__2,i__3);
  816. if (jlen > 0) {
  817. cnorm[j] = dasum_(&jlen, &ab[j * ab_dim1 + 2], &c__1);
  818. } else {
  819. cnorm[j] = 0.;
  820. }
  821. /* L20: */
  822. }
  823. }
  824. }
  825. /* Scale the column norms by TSCAL if the maximum element in CNORM is */
  826. /* greater than BIGNUM. */
  827. imax = idamax_(n, &cnorm[1], &c__1);
  828. tmax = cnorm[imax];
  829. if (tmax <= bignum) {
  830. tscal = 1.;
  831. } else {
  832. tscal = 1. / (smlnum * tmax);
  833. dscal_(n, &tscal, &cnorm[1], &c__1);
  834. }
  835. /* Compute a bound on the computed solution vector to see if the */
  836. /* Level 2 BLAS routine DTBSV can be used. */
  837. j = idamax_(n, &x[1], &c__1);
  838. xmax = (d__1 = x[j], abs(d__1));
  839. xbnd = xmax;
  840. if (notran) {
  841. /* Compute the growth in A * x = b. */
  842. if (upper) {
  843. jfirst = *n;
  844. jlast = 1;
  845. jinc = -1;
  846. maind = *kd + 1;
  847. } else {
  848. jfirst = 1;
  849. jlast = *n;
  850. jinc = 1;
  851. maind = 1;
  852. }
  853. if (tscal != 1.) {
  854. grow = 0.;
  855. goto L50;
  856. }
  857. if (nounit) {
  858. /* A is non-unit triangular. */
  859. /* Compute GROW = 1/G(j) and XBND = 1/M(j). */
  860. /* Initially, G(0) = f2cmax{x(i), i=1,...,n}. */
  861. grow = 1. / f2cmax(xbnd,smlnum);
  862. xbnd = grow;
  863. i__1 = jlast;
  864. i__2 = jinc;
  865. for (j = jfirst; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
  866. /* Exit the loop if the growth factor is too small. */
  867. if (grow <= smlnum) {
  868. goto L50;
  869. }
  870. /* M(j) = G(j-1) / abs(A(j,j)) */
  871. tjj = (d__1 = ab[maind + j * ab_dim1], abs(d__1));
  872. /* Computing MIN */
  873. d__1 = xbnd, d__2 = f2cmin(1.,tjj) * grow;
  874. xbnd = f2cmin(d__1,d__2);
  875. if (tjj + cnorm[j] >= smlnum) {
  876. /* G(j) = G(j-1)*( 1 + CNORM(j) / abs(A(j,j)) ) */
  877. grow *= tjj / (tjj + cnorm[j]);
  878. } else {
  879. /* G(j) could overflow, set GROW to 0. */
  880. grow = 0.;
  881. }
  882. /* L30: */
  883. }
  884. grow = xbnd;
  885. } else {
  886. /* A is unit triangular. */
  887. /* Compute GROW = 1/G(j), where G(0) = f2cmax{x(i), i=1,...,n}. */
  888. /* Computing MIN */
  889. d__1 = 1., d__2 = 1. / f2cmax(xbnd,smlnum);
  890. grow = f2cmin(d__1,d__2);
  891. i__2 = jlast;
  892. i__1 = jinc;
  893. for (j = jfirst; i__1 < 0 ? j >= i__2 : j <= i__2; j += i__1) {
  894. /* Exit the loop if the growth factor is too small. */
  895. if (grow <= smlnum) {
  896. goto L50;
  897. }
  898. /* G(j) = G(j-1)*( 1 + CNORM(j) ) */
  899. grow *= 1. / (cnorm[j] + 1.);
  900. /* L40: */
  901. }
  902. }
  903. L50:
  904. ;
  905. } else {
  906. /* Compute the growth in A**T * x = b. */
  907. if (upper) {
  908. jfirst = 1;
  909. jlast = *n;
  910. jinc = 1;
  911. maind = *kd + 1;
  912. } else {
  913. jfirst = *n;
  914. jlast = 1;
  915. jinc = -1;
  916. maind = 1;
  917. }
  918. if (tscal != 1.) {
  919. grow = 0.;
  920. goto L80;
  921. }
  922. if (nounit) {
  923. /* A is non-unit triangular. */
  924. /* Compute GROW = 1/G(j) and XBND = 1/M(j). */
  925. /* Initially, M(0) = f2cmax{x(i), i=1,...,n}. */
  926. grow = 1. / f2cmax(xbnd,smlnum);
  927. xbnd = grow;
  928. i__1 = jlast;
  929. i__2 = jinc;
  930. for (j = jfirst; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
  931. /* Exit the loop if the growth factor is too small. */
  932. if (grow <= smlnum) {
  933. goto L80;
  934. }
  935. /* G(j) = f2cmax( G(j-1), M(j-1)*( 1 + CNORM(j) ) ) */
  936. xj = cnorm[j] + 1.;
  937. /* Computing MIN */
  938. d__1 = grow, d__2 = xbnd / xj;
  939. grow = f2cmin(d__1,d__2);
  940. /* M(j) = M(j-1)*( 1 + CNORM(j) ) / abs(A(j,j)) */
  941. tjj = (d__1 = ab[maind + j * ab_dim1], abs(d__1));
  942. if (xj > tjj) {
  943. xbnd *= tjj / xj;
  944. }
  945. /* L60: */
  946. }
  947. grow = f2cmin(grow,xbnd);
  948. } else {
  949. /* A is unit triangular. */
  950. /* Compute GROW = 1/G(j), where G(0) = f2cmax{x(i), i=1,...,n}. */
  951. /* Computing MIN */
  952. d__1 = 1., d__2 = 1. / f2cmax(xbnd,smlnum);
  953. grow = f2cmin(d__1,d__2);
  954. i__2 = jlast;
  955. i__1 = jinc;
  956. for (j = jfirst; i__1 < 0 ? j >= i__2 : j <= i__2; j += i__1) {
  957. /* Exit the loop if the growth factor is too small. */
  958. if (grow <= smlnum) {
  959. goto L80;
  960. }
  961. /* G(j) = ( 1 + CNORM(j) )*G(j-1) */
  962. xj = cnorm[j] + 1.;
  963. grow /= xj;
  964. /* L70: */
  965. }
  966. }
  967. L80:
  968. ;
  969. }
  970. if (grow * tscal > smlnum) {
  971. /* Use the Level 2 BLAS solve if the reciprocal of the bound on */
  972. /* elements of X is not too small. */
  973. dtbsv_(uplo, trans, diag, n, kd, &ab[ab_offset], ldab, &x[1], &c__1);
  974. } else {
  975. /* Use a Level 1 BLAS solve, scaling intermediate results. */
  976. if (xmax > bignum) {
  977. /* Scale X so that its components are less than or equal to */
  978. /* BIGNUM in absolute value. */
  979. *scale = bignum / xmax;
  980. dscal_(n, scale, &x[1], &c__1);
  981. xmax = bignum;
  982. }
  983. if (notran) {
  984. /* Solve A * x = b */
  985. i__1 = jlast;
  986. i__2 = jinc;
  987. for (j = jfirst; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
  988. /* Compute x(j) = b(j) / A(j,j), scaling x if necessary. */
  989. xj = (d__1 = x[j], abs(d__1));
  990. if (nounit) {
  991. tjjs = ab[maind + j * ab_dim1] * tscal;
  992. } else {
  993. tjjs = tscal;
  994. if (tscal == 1.) {
  995. goto L100;
  996. }
  997. }
  998. tjj = abs(tjjs);
  999. if (tjj > smlnum) {
  1000. /* abs(A(j,j)) > SMLNUM: */
  1001. if (tjj < 1.) {
  1002. if (xj > tjj * bignum) {
  1003. /* Scale x by 1/b(j). */
  1004. rec = 1. / xj;
  1005. dscal_(n, &rec, &x[1], &c__1);
  1006. *scale *= rec;
  1007. xmax *= rec;
  1008. }
  1009. }
  1010. x[j] /= tjjs;
  1011. xj = (d__1 = x[j], abs(d__1));
  1012. } else if (tjj > 0.) {
  1013. /* 0 < abs(A(j,j)) <= SMLNUM: */
  1014. if (xj > tjj * bignum) {
  1015. /* Scale x by (1/abs(x(j)))*abs(A(j,j))*BIGNUM */
  1016. /* to avoid overflow when dividing by A(j,j). */
  1017. rec = tjj * bignum / xj;
  1018. if (cnorm[j] > 1.) {
  1019. /* Scale by 1/CNORM(j) to avoid overflow when */
  1020. /* multiplying x(j) times column j. */
  1021. rec /= cnorm[j];
  1022. }
  1023. dscal_(n, &rec, &x[1], &c__1);
  1024. *scale *= rec;
  1025. xmax *= rec;
  1026. }
  1027. x[j] /= tjjs;
  1028. xj = (d__1 = x[j], abs(d__1));
  1029. } else {
  1030. /* A(j,j) = 0: Set x(1:n) = 0, x(j) = 1, and */
  1031. /* scale = 0, and compute a solution to A*x = 0. */
  1032. i__3 = *n;
  1033. for (i__ = 1; i__ <= i__3; ++i__) {
  1034. x[i__] = 0.;
  1035. /* L90: */
  1036. }
  1037. x[j] = 1.;
  1038. xj = 1.;
  1039. *scale = 0.;
  1040. xmax = 0.;
  1041. }
  1042. L100:
  1043. /* Scale x if necessary to avoid overflow when adding a */
  1044. /* multiple of column j of A. */
  1045. if (xj > 1.) {
  1046. rec = 1. / xj;
  1047. if (cnorm[j] > (bignum - xmax) * rec) {
  1048. /* Scale x by 1/(2*abs(x(j))). */
  1049. rec *= .5;
  1050. dscal_(n, &rec, &x[1], &c__1);
  1051. *scale *= rec;
  1052. }
  1053. } else if (xj * cnorm[j] > bignum - xmax) {
  1054. /* Scale x by 1/2. */
  1055. dscal_(n, &c_b36, &x[1], &c__1);
  1056. *scale *= .5;
  1057. }
  1058. if (upper) {
  1059. if (j > 1) {
  1060. /* Compute the update */
  1061. /* x(f2cmax(1,j-kd):j-1) := x(f2cmax(1,j-kd):j-1) - */
  1062. /* x(j)* A(f2cmax(1,j-kd):j-1,j) */
  1063. /* Computing MIN */
  1064. i__3 = *kd, i__4 = j - 1;
  1065. jlen = f2cmin(i__3,i__4);
  1066. d__1 = -x[j] * tscal;
  1067. daxpy_(&jlen, &d__1, &ab[*kd + 1 - jlen + j * ab_dim1]
  1068. , &c__1, &x[j - jlen], &c__1);
  1069. i__3 = j - 1;
  1070. i__ = idamax_(&i__3, &x[1], &c__1);
  1071. xmax = (d__1 = x[i__], abs(d__1));
  1072. }
  1073. } else if (j < *n) {
  1074. /* Compute the update */
  1075. /* x(j+1:f2cmin(j+kd,n)) := x(j+1:f2cmin(j+kd,n)) - */
  1076. /* x(j) * A(j+1:f2cmin(j+kd,n),j) */
  1077. /* Computing MIN */
  1078. i__3 = *kd, i__4 = *n - j;
  1079. jlen = f2cmin(i__3,i__4);
  1080. if (jlen > 0) {
  1081. d__1 = -x[j] * tscal;
  1082. daxpy_(&jlen, &d__1, &ab[j * ab_dim1 + 2], &c__1, &x[
  1083. j + 1], &c__1);
  1084. }
  1085. i__3 = *n - j;
  1086. i__ = j + idamax_(&i__3, &x[j + 1], &c__1);
  1087. xmax = (d__1 = x[i__], abs(d__1));
  1088. }
  1089. /* L110: */
  1090. }
  1091. } else {
  1092. /* Solve A**T * x = b */
  1093. i__2 = jlast;
  1094. i__1 = jinc;
  1095. for (j = jfirst; i__1 < 0 ? j >= i__2 : j <= i__2; j += i__1) {
  1096. /* Compute x(j) = b(j) - sum A(k,j)*x(k). */
  1097. /* k<>j */
  1098. xj = (d__1 = x[j], abs(d__1));
  1099. uscal = tscal;
  1100. rec = 1. / f2cmax(xmax,1.);
  1101. if (cnorm[j] > (bignum - xj) * rec) {
  1102. /* If x(j) could overflow, scale x by 1/(2*XMAX). */
  1103. rec *= .5;
  1104. if (nounit) {
  1105. tjjs = ab[maind + j * ab_dim1] * tscal;
  1106. } else {
  1107. tjjs = tscal;
  1108. }
  1109. tjj = abs(tjjs);
  1110. if (tjj > 1.) {
  1111. /* Divide by A(j,j) when scaling x if A(j,j) > 1. */
  1112. /* Computing MIN */
  1113. d__1 = 1., d__2 = rec * tjj;
  1114. rec = f2cmin(d__1,d__2);
  1115. uscal /= tjjs;
  1116. }
  1117. if (rec < 1.) {
  1118. dscal_(n, &rec, &x[1], &c__1);
  1119. *scale *= rec;
  1120. xmax *= rec;
  1121. }
  1122. }
  1123. sumj = 0.;
  1124. if (uscal == 1.) {
  1125. /* If the scaling needed for A in the dot product is 1, */
  1126. /* call DDOT to perform the dot product. */
  1127. if (upper) {
  1128. /* Computing MIN */
  1129. i__3 = *kd, i__4 = j - 1;
  1130. jlen = f2cmin(i__3,i__4);
  1131. sumj = ddot_(&jlen, &ab[*kd + 1 - jlen + j * ab_dim1],
  1132. &c__1, &x[j - jlen], &c__1);
  1133. } else {
  1134. /* Computing MIN */
  1135. i__3 = *kd, i__4 = *n - j;
  1136. jlen = f2cmin(i__3,i__4);
  1137. if (jlen > 0) {
  1138. sumj = ddot_(&jlen, &ab[j * ab_dim1 + 2], &c__1, &
  1139. x[j + 1], &c__1);
  1140. }
  1141. }
  1142. } else {
  1143. /* Otherwise, use in-line code for the dot product. */
  1144. if (upper) {
  1145. /* Computing MIN */
  1146. i__3 = *kd, i__4 = j - 1;
  1147. jlen = f2cmin(i__3,i__4);
  1148. i__3 = jlen;
  1149. for (i__ = 1; i__ <= i__3; ++i__) {
  1150. sumj += ab[*kd + i__ - jlen + j * ab_dim1] *
  1151. uscal * x[j - jlen - 1 + i__];
  1152. /* L120: */
  1153. }
  1154. } else {
  1155. /* Computing MIN */
  1156. i__3 = *kd, i__4 = *n - j;
  1157. jlen = f2cmin(i__3,i__4);
  1158. i__3 = jlen;
  1159. for (i__ = 1; i__ <= i__3; ++i__) {
  1160. sumj += ab[i__ + 1 + j * ab_dim1] * uscal * x[j +
  1161. i__];
  1162. /* L130: */
  1163. }
  1164. }
  1165. }
  1166. if (uscal == tscal) {
  1167. /* Compute x(j) := ( x(j) - sumj ) / A(j,j) if 1/A(j,j) */
  1168. /* was not used to scale the dotproduct. */
  1169. x[j] -= sumj;
  1170. xj = (d__1 = x[j], abs(d__1));
  1171. if (nounit) {
  1172. /* Compute x(j) = x(j) / A(j,j), scaling if necessary. */
  1173. tjjs = ab[maind + j * ab_dim1] * tscal;
  1174. } else {
  1175. tjjs = tscal;
  1176. if (tscal == 1.) {
  1177. goto L150;
  1178. }
  1179. }
  1180. tjj = abs(tjjs);
  1181. if (tjj > smlnum) {
  1182. /* abs(A(j,j)) > SMLNUM: */
  1183. if (tjj < 1.) {
  1184. if (xj > tjj * bignum) {
  1185. /* Scale X by 1/abs(x(j)). */
  1186. rec = 1. / xj;
  1187. dscal_(n, &rec, &x[1], &c__1);
  1188. *scale *= rec;
  1189. xmax *= rec;
  1190. }
  1191. }
  1192. x[j] /= tjjs;
  1193. } else if (tjj > 0.) {
  1194. /* 0 < abs(A(j,j)) <= SMLNUM: */
  1195. if (xj > tjj * bignum) {
  1196. /* Scale x by (1/abs(x(j)))*abs(A(j,j))*BIGNUM. */
  1197. rec = tjj * bignum / xj;
  1198. dscal_(n, &rec, &x[1], &c__1);
  1199. *scale *= rec;
  1200. xmax *= rec;
  1201. }
  1202. x[j] /= tjjs;
  1203. } else {
  1204. /* A(j,j) = 0: Set x(1:n) = 0, x(j) = 1, and */
  1205. /* scale = 0, and compute a solution to A**T*x = 0. */
  1206. i__3 = *n;
  1207. for (i__ = 1; i__ <= i__3; ++i__) {
  1208. x[i__] = 0.;
  1209. /* L140: */
  1210. }
  1211. x[j] = 1.;
  1212. *scale = 0.;
  1213. xmax = 0.;
  1214. }
  1215. L150:
  1216. ;
  1217. } else {
  1218. /* Compute x(j) := x(j) / A(j,j) - sumj if the dot */
  1219. /* product has already been divided by 1/A(j,j). */
  1220. x[j] = x[j] / tjjs - sumj;
  1221. }
  1222. /* Computing MAX */
  1223. d__2 = xmax, d__3 = (d__1 = x[j], abs(d__1));
  1224. xmax = f2cmax(d__2,d__3);
  1225. /* L160: */
  1226. }
  1227. }
  1228. *scale /= tscal;
  1229. }
  1230. /* Scale the column norms by 1/TSCAL for return. */
  1231. if (tscal != 1.) {
  1232. d__1 = 1. / tscal;
  1233. dscal_(n, &d__1, &cnorm[1], &c__1);
  1234. }
  1235. return 0;
  1236. /* End of DLATBS */
  1237. } /* dlatbs_ */