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.

dlatrs.c 38 kB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385
  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 DLATRS solves a triangular system of equations with the scale factor set to prevent overflow.
  489. */
  490. /* =========== DOCUMENTATION =========== */
  491. /* Online html documentation available at */
  492. /* http://www.netlib.org/lapack/explore-html/ */
  493. /* > \htmlonly */
  494. /* > Download DLATRS + dependencies */
  495. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlatrs.
  496. f"> */
  497. /* > [TGZ]</a> */
  498. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlatrs.
  499. f"> */
  500. /* > [ZIP]</a> */
  501. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlatrs.
  502. f"> */
  503. /* > [TXT]</a> */
  504. /* > \endhtmlonly */
  505. /* Definition: */
  506. /* =========== */
  507. /* SUBROUTINE DLATRS( UPLO, TRANS, DIAG, NORMIN, N, A, LDA, X, SCALE, */
  508. /* CNORM, INFO ) */
  509. /* CHARACTER DIAG, NORMIN, TRANS, UPLO */
  510. /* INTEGER INFO, LDA, N */
  511. /* DOUBLE PRECISION SCALE */
  512. /* DOUBLE PRECISION A( LDA, * ), CNORM( * ), X( * ) */
  513. /* > \par Purpose: */
  514. /* ============= */
  515. /* > */
  516. /* > \verbatim */
  517. /* > */
  518. /* > DLATRS solves one of the triangular systems */
  519. /* > */
  520. /* > A *x = s*b or A**T *x = s*b */
  521. /* > */
  522. /* > with scaling to prevent overflow. Here A is an upper or lower */
  523. /* > triangular matrix, A**T denotes the transpose of A, x and b are */
  524. /* > n-element vectors, and s is a scaling factor, usually less than */
  525. /* > or equal to 1, chosen so that the components of x will be less than */
  526. /* > the overflow threshold. If the unscaled problem will not cause */
  527. /* > overflow, the Level 2 BLAS routine DTRSV is called. If the matrix A */
  528. /* > is singular (A(j,j) = 0 for some j), then s is set to 0 and a */
  529. /* > non-trivial solution to A*x = 0 is returned. */
  530. /* > \endverbatim */
  531. /* Arguments: */
  532. /* ========== */
  533. /* > \param[in] UPLO */
  534. /* > \verbatim */
  535. /* > UPLO is CHARACTER*1 */
  536. /* > Specifies whether the matrix A is upper or lower triangular. */
  537. /* > = 'U': Upper triangular */
  538. /* > = 'L': Lower triangular */
  539. /* > \endverbatim */
  540. /* > */
  541. /* > \param[in] TRANS */
  542. /* > \verbatim */
  543. /* > TRANS is CHARACTER*1 */
  544. /* > Specifies the operation applied to A. */
  545. /* > = 'N': Solve A * x = s*b (No transpose) */
  546. /* > = 'T': Solve A**T* x = s*b (Transpose) */
  547. /* > = 'C': Solve A**T* x = s*b (Conjugate transpose = Transpose) */
  548. /* > \endverbatim */
  549. /* > */
  550. /* > \param[in] DIAG */
  551. /* > \verbatim */
  552. /* > DIAG is CHARACTER*1 */
  553. /* > Specifies whether or not the matrix A is unit triangular. */
  554. /* > = 'N': Non-unit triangular */
  555. /* > = 'U': Unit triangular */
  556. /* > \endverbatim */
  557. /* > */
  558. /* > \param[in] NORMIN */
  559. /* > \verbatim */
  560. /* > NORMIN is CHARACTER*1 */
  561. /* > Specifies whether CNORM has been set or not. */
  562. /* > = 'Y': CNORM contains the column norms on entry */
  563. /* > = 'N': CNORM is not set on entry. On exit, the norms will */
  564. /* > be computed and stored in CNORM. */
  565. /* > \endverbatim */
  566. /* > */
  567. /* > \param[in] N */
  568. /* > \verbatim */
  569. /* > N is INTEGER */
  570. /* > The order of the matrix A. N >= 0. */
  571. /* > \endverbatim */
  572. /* > */
  573. /* > \param[in] A */
  574. /* > \verbatim */
  575. /* > A is DOUBLE PRECISION array, dimension (LDA,N) */
  576. /* > The triangular matrix A. If UPLO = 'U', the leading n by n */
  577. /* > upper triangular part of the array A contains the upper */
  578. /* > triangular matrix, and the strictly lower triangular part of */
  579. /* > A is not referenced. If UPLO = 'L', the leading n by n lower */
  580. /* > triangular part of the array A contains the lower triangular */
  581. /* > matrix, and the strictly upper triangular part of A is not */
  582. /* > referenced. If DIAG = 'U', the diagonal elements of A are */
  583. /* > also not referenced and are assumed to be 1. */
  584. /* > \endverbatim */
  585. /* > */
  586. /* > \param[in] LDA */
  587. /* > \verbatim */
  588. /* > LDA is INTEGER */
  589. /* > The leading dimension of the array A. LDA >= f2cmax (1,N). */
  590. /* > \endverbatim */
  591. /* > */
  592. /* > \param[in,out] X */
  593. /* > \verbatim */
  594. /* > X is DOUBLE PRECISION array, dimension (N) */
  595. /* > On entry, the right hand side b of the triangular system. */
  596. /* > On exit, X is overwritten by the solution vector x. */
  597. /* > \endverbatim */
  598. /* > */
  599. /* > \param[out] SCALE */
  600. /* > \verbatim */
  601. /* > SCALE is DOUBLE PRECISION */
  602. /* > The scaling factor s for the triangular system */
  603. /* > A * x = s*b or A**T* x = s*b. */
  604. /* > If SCALE = 0, the matrix A is singular or badly scaled, and */
  605. /* > the vector x is an exact or approximate solution to A*x = 0. */
  606. /* > \endverbatim */
  607. /* > */
  608. /* > \param[in,out] CNORM */
  609. /* > \verbatim */
  610. /* > CNORM is DOUBLE PRECISION array, dimension (N) */
  611. /* > */
  612. /* > If NORMIN = 'Y', CNORM is an input argument and CNORM(j) */
  613. /* > contains the norm of the off-diagonal part of the j-th column */
  614. /* > of A. If TRANS = 'N', CNORM(j) must be greater than or equal */
  615. /* > to the infinity-norm, and if TRANS = 'T' or 'C', CNORM(j) */
  616. /* > must be greater than or equal to the 1-norm. */
  617. /* > */
  618. /* > If NORMIN = 'N', CNORM is an output argument and CNORM(j) */
  619. /* > returns the 1-norm of the offdiagonal part of the j-th column */
  620. /* > of A. */
  621. /* > \endverbatim */
  622. /* > */
  623. /* > \param[out] INFO */
  624. /* > \verbatim */
  625. /* > INFO is INTEGER */
  626. /* > = 0: successful exit */
  627. /* > < 0: if INFO = -k, the k-th argument had an illegal value */
  628. /* > \endverbatim */
  629. /* Authors: */
  630. /* ======== */
  631. /* > \author Univ. of Tennessee */
  632. /* > \author Univ. of California Berkeley */
  633. /* > \author Univ. of Colorado Denver */
  634. /* > \author NAG Ltd. */
  635. /* > \date December 2016 */
  636. /* > \ingroup doubleOTHERauxiliary */
  637. /* > \par Further Details: */
  638. /* ===================== */
  639. /* > */
  640. /* > \verbatim */
  641. /* > */
  642. /* > A rough bound on x is computed; if that is less than overflow, DTRSV */
  643. /* > is called, otherwise, specific code is used which checks for possible */
  644. /* > overflow or divide-by-zero at every operation. */
  645. /* > */
  646. /* > A columnwise scheme is used for solving A*x = b. The basic algorithm */
  647. /* > if A is lower triangular is */
  648. /* > */
  649. /* > x[1:n] := b[1:n] */
  650. /* > for j = 1, ..., n */
  651. /* > x(j) := x(j) / A(j,j) */
  652. /* > x[j+1:n] := x[j+1:n] - x(j) * A[j+1:n,j] */
  653. /* > end */
  654. /* > */
  655. /* > Define bounds on the components of x after j iterations of the loop: */
  656. /* > M(j) = bound on x[1:j] */
  657. /* > G(j) = bound on x[j+1:n] */
  658. /* > Initially, let M(0) = 0 and G(0) = f2cmax{x(i), i=1,...,n}. */
  659. /* > */
  660. /* > Then for iteration j+1 we have */
  661. /* > M(j+1) <= G(j) / | A(j+1,j+1) | */
  662. /* > G(j+1) <= G(j) + M(j+1) * | A[j+2:n,j+1] | */
  663. /* > <= G(j) ( 1 + CNORM(j+1) / | A(j+1,j+1) | ) */
  664. /* > */
  665. /* > where CNORM(j+1) is greater than or equal to the infinity-norm of */
  666. /* > column j+1 of A, not counting the diagonal. Hence */
  667. /* > */
  668. /* > G(j) <= G(0) product ( 1 + CNORM(i) / | A(i,i) | ) */
  669. /* > 1<=i<=j */
  670. /* > and */
  671. /* > */
  672. /* > |x(j)| <= ( G(0) / |A(j,j)| ) product ( 1 + CNORM(i) / |A(i,i)| ) */
  673. /* > 1<=i< j */
  674. /* > */
  675. /* > Since |x(j)| <= M(j), we use the Level 2 BLAS routine DTRSV if the */
  676. /* > reciprocal of the largest M(j), j=1,..,n, is larger than */
  677. /* > f2cmax(underflow, 1/overflow). */
  678. /* > */
  679. /* > The bound on x(j) is also used to determine when a step in the */
  680. /* > columnwise method can be performed without fear of overflow. If */
  681. /* > the computed bound is greater than a large constant, x is scaled to */
  682. /* > prevent overflow, but if the bound overflows, x is set to 0, x(j) to */
  683. /* > 1, and scale to 0, and a non-trivial solution to A*x = 0 is found. */
  684. /* > */
  685. /* > Similarly, a row-wise scheme is used to solve A**T*x = b. The basic */
  686. /* > algorithm for A upper triangular is */
  687. /* > */
  688. /* > for j = 1, ..., n */
  689. /* > x(j) := ( b(j) - A[1:j-1,j]**T * x[1:j-1] ) / A(j,j) */
  690. /* > end */
  691. /* > */
  692. /* > We simultaneously compute two bounds */
  693. /* > G(j) = bound on ( b(i) - A[1:i-1,i]**T * x[1:i-1] ), 1<=i<=j */
  694. /* > M(j) = bound on x(i), 1<=i<=j */
  695. /* > */
  696. /* > The initial values are G(0) = 0, M(0) = f2cmax{b(i), i=1,..,n}, and we */
  697. /* > add the constraint G(j) >= G(j-1) and M(j) >= M(j-1) for j >= 1. */
  698. /* > Then the bound on x(j) is */
  699. /* > */
  700. /* > M(j) <= M(j-1) * ( 1 + CNORM(j) ) / | A(j,j) | */
  701. /* > */
  702. /* > <= M(0) * product ( ( 1 + CNORM(i) ) / |A(i,i)| ) */
  703. /* > 1<=i<=j */
  704. /* > */
  705. /* > and we can safely call DTRSV if 1/M(n) and 1/G(n) are both greater */
  706. /* > than f2cmax(underflow, 1/overflow). */
  707. /* > \endverbatim */
  708. /* > */
  709. /* ===================================================================== */
  710. /* Subroutine */ int dlatrs_(char *uplo, char *trans, char *diag, char *
  711. normin, integer *n, doublereal *a, integer *lda, doublereal *x,
  712. doublereal *scale, doublereal *cnorm, integer *info)
  713. {
  714. /* System generated locals */
  715. integer a_dim1, a_offset, i__1, i__2, i__3;
  716. doublereal d__1, d__2, d__3;
  717. /* Local variables */
  718. integer jinc;
  719. extern doublereal ddot_(integer *, doublereal *, integer *, doublereal *,
  720. integer *);
  721. doublereal xbnd;
  722. integer imax;
  723. doublereal tmax, tjjs, xmax, grow, sumj;
  724. integer i__, j;
  725. extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
  726. integer *);
  727. extern logical lsame_(char *, char *);
  728. doublereal tscal, uscal;
  729. extern doublereal dasum_(integer *, doublereal *, integer *);
  730. integer jlast;
  731. extern /* Subroutine */ int daxpy_(integer *, doublereal *, doublereal *,
  732. integer *, doublereal *, integer *);
  733. logical upper;
  734. extern /* Subroutine */ int dtrsv_(char *, char *, char *, integer *,
  735. doublereal *, integer *, doublereal *, integer *);
  736. extern doublereal dlamch_(char *);
  737. doublereal xj;
  738. extern integer idamax_(integer *, doublereal *, integer *);
  739. extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
  740. doublereal bignum;
  741. logical notran;
  742. integer jfirst;
  743. doublereal smlnum;
  744. logical nounit;
  745. doublereal rec, tjj;
  746. /* -- LAPACK auxiliary routine (version 3.7.0) -- */
  747. /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
  748. /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
  749. /* December 2016 */
  750. /* ===================================================================== */
  751. /* Parameter adjustments */
  752. a_dim1 = *lda;
  753. a_offset = 1 + a_dim1 * 1;
  754. a -= a_offset;
  755. --x;
  756. --cnorm;
  757. /* Function Body */
  758. *info = 0;
  759. upper = lsame_(uplo, "U");
  760. notran = lsame_(trans, "N");
  761. nounit = lsame_(diag, "N");
  762. /* Test the input parameters. */
  763. if (! upper && ! lsame_(uplo, "L")) {
  764. *info = -1;
  765. } else if (! notran && ! lsame_(trans, "T") && !
  766. lsame_(trans, "C")) {
  767. *info = -2;
  768. } else if (! nounit && ! lsame_(diag, "U")) {
  769. *info = -3;
  770. } else if (! lsame_(normin, "Y") && ! lsame_(normin,
  771. "N")) {
  772. *info = -4;
  773. } else if (*n < 0) {
  774. *info = -5;
  775. } else if (*lda < f2cmax(1,*n)) {
  776. *info = -7;
  777. }
  778. if (*info != 0) {
  779. i__1 = -(*info);
  780. xerbla_("DLATRS", &i__1, (ftnlen)6);
  781. return 0;
  782. }
  783. /* Quick return if possible */
  784. if (*n == 0) {
  785. return 0;
  786. }
  787. /* Determine machine dependent parameters to control overflow. */
  788. smlnum = dlamch_("Safe minimum") / dlamch_("Precision");
  789. bignum = 1. / smlnum;
  790. *scale = 1.;
  791. if (lsame_(normin, "N")) {
  792. /* Compute the 1-norm of each column, not including the diagonal. */
  793. if (upper) {
  794. /* A is upper triangular. */
  795. i__1 = *n;
  796. for (j = 1; j <= i__1; ++j) {
  797. i__2 = j - 1;
  798. cnorm[j] = dasum_(&i__2, &a[j * a_dim1 + 1], &c__1);
  799. /* L10: */
  800. }
  801. } else {
  802. /* A is lower triangular. */
  803. i__1 = *n - 1;
  804. for (j = 1; j <= i__1; ++j) {
  805. i__2 = *n - j;
  806. cnorm[j] = dasum_(&i__2, &a[j + 1 + j * a_dim1], &c__1);
  807. /* L20: */
  808. }
  809. cnorm[*n] = 0.;
  810. }
  811. }
  812. /* Scale the column norms by TSCAL if the maximum element in CNORM is */
  813. /* greater than BIGNUM. */
  814. imax = idamax_(n, &cnorm[1], &c__1);
  815. tmax = cnorm[imax];
  816. if (tmax <= bignum) {
  817. tscal = 1.;
  818. } else {
  819. tscal = 1. / (smlnum * tmax);
  820. dscal_(n, &tscal, &cnorm[1], &c__1);
  821. }
  822. /* Compute a bound on the computed solution vector to see if the */
  823. /* Level 2 BLAS routine DTRSV can be used. */
  824. j = idamax_(n, &x[1], &c__1);
  825. xmax = (d__1 = x[j], abs(d__1));
  826. xbnd = xmax;
  827. if (notran) {
  828. /* Compute the growth in A * x = b. */
  829. if (upper) {
  830. jfirst = *n;
  831. jlast = 1;
  832. jinc = -1;
  833. } else {
  834. jfirst = 1;
  835. jlast = *n;
  836. jinc = 1;
  837. }
  838. if (tscal != 1.) {
  839. grow = 0.;
  840. goto L50;
  841. }
  842. if (nounit) {
  843. /* A is non-unit triangular. */
  844. /* Compute GROW = 1/G(j) and XBND = 1/M(j). */
  845. /* Initially, G(0) = f2cmax{x(i), i=1,...,n}. */
  846. grow = 1. / f2cmax(xbnd,smlnum);
  847. xbnd = grow;
  848. i__1 = jlast;
  849. i__2 = jinc;
  850. for (j = jfirst; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
  851. /* Exit the loop if the growth factor is too small. */
  852. if (grow <= smlnum) {
  853. goto L50;
  854. }
  855. /* M(j) = G(j-1) / abs(A(j,j)) */
  856. tjj = (d__1 = a[j + j * a_dim1], abs(d__1));
  857. /* Computing MIN */
  858. d__1 = xbnd, d__2 = f2cmin(1.,tjj) * grow;
  859. xbnd = f2cmin(d__1,d__2);
  860. if (tjj + cnorm[j] >= smlnum) {
  861. /* G(j) = G(j-1)*( 1 + CNORM(j) / abs(A(j,j)) ) */
  862. grow *= tjj / (tjj + cnorm[j]);
  863. } else {
  864. /* G(j) could overflow, set GROW to 0. */
  865. grow = 0.;
  866. }
  867. /* L30: */
  868. }
  869. grow = xbnd;
  870. } else {
  871. /* A is unit triangular. */
  872. /* Compute GROW = 1/G(j), where G(0) = f2cmax{x(i), i=1,...,n}. */
  873. /* Computing MIN */
  874. d__1 = 1., d__2 = 1. / f2cmax(xbnd,smlnum);
  875. grow = f2cmin(d__1,d__2);
  876. i__2 = jlast;
  877. i__1 = jinc;
  878. for (j = jfirst; i__1 < 0 ? j >= i__2 : j <= i__2; j += i__1) {
  879. /* Exit the loop if the growth factor is too small. */
  880. if (grow <= smlnum) {
  881. goto L50;
  882. }
  883. /* G(j) = G(j-1)*( 1 + CNORM(j) ) */
  884. grow *= 1. / (cnorm[j] + 1.);
  885. /* L40: */
  886. }
  887. }
  888. L50:
  889. ;
  890. } else {
  891. /* Compute the growth in A**T * x = b. */
  892. if (upper) {
  893. jfirst = 1;
  894. jlast = *n;
  895. jinc = 1;
  896. } else {
  897. jfirst = *n;
  898. jlast = 1;
  899. jinc = -1;
  900. }
  901. if (tscal != 1.) {
  902. grow = 0.;
  903. goto L80;
  904. }
  905. if (nounit) {
  906. /* A is non-unit triangular. */
  907. /* Compute GROW = 1/G(j) and XBND = 1/M(j). */
  908. /* Initially, M(0) = f2cmax{x(i), i=1,...,n}. */
  909. grow = 1. / f2cmax(xbnd,smlnum);
  910. xbnd = grow;
  911. i__1 = jlast;
  912. i__2 = jinc;
  913. for (j = jfirst; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
  914. /* Exit the loop if the growth factor is too small. */
  915. if (grow <= smlnum) {
  916. goto L80;
  917. }
  918. /* G(j) = f2cmax( G(j-1), M(j-1)*( 1 + CNORM(j) ) ) */
  919. xj = cnorm[j] + 1.;
  920. /* Computing MIN */
  921. d__1 = grow, d__2 = xbnd / xj;
  922. grow = f2cmin(d__1,d__2);
  923. /* M(j) = M(j-1)*( 1 + CNORM(j) ) / abs(A(j,j)) */
  924. tjj = (d__1 = a[j + j * a_dim1], abs(d__1));
  925. if (xj > tjj) {
  926. xbnd *= tjj / xj;
  927. }
  928. /* L60: */
  929. }
  930. grow = f2cmin(grow,xbnd);
  931. } else {
  932. /* A is unit triangular. */
  933. /* Compute GROW = 1/G(j), where G(0) = f2cmax{x(i), i=1,...,n}. */
  934. /* Computing MIN */
  935. d__1 = 1., d__2 = 1. / f2cmax(xbnd,smlnum);
  936. grow = f2cmin(d__1,d__2);
  937. i__2 = jlast;
  938. i__1 = jinc;
  939. for (j = jfirst; i__1 < 0 ? j >= i__2 : j <= i__2; j += i__1) {
  940. /* Exit the loop if the growth factor is too small. */
  941. if (grow <= smlnum) {
  942. goto L80;
  943. }
  944. /* G(j) = ( 1 + CNORM(j) )*G(j-1) */
  945. xj = cnorm[j] + 1.;
  946. grow /= xj;
  947. /* L70: */
  948. }
  949. }
  950. L80:
  951. ;
  952. }
  953. if (grow * tscal > smlnum) {
  954. /* Use the Level 2 BLAS solve if the reciprocal of the bound on */
  955. /* elements of X is not too small. */
  956. dtrsv_(uplo, trans, diag, n, &a[a_offset], lda, &x[1], &c__1);
  957. } else {
  958. /* Use a Level 1 BLAS solve, scaling intermediate results. */
  959. if (xmax > bignum) {
  960. /* Scale X so that its components are less than or equal to */
  961. /* BIGNUM in absolute value. */
  962. *scale = bignum / xmax;
  963. dscal_(n, scale, &x[1], &c__1);
  964. xmax = bignum;
  965. }
  966. if (notran) {
  967. /* Solve A * x = b */
  968. i__1 = jlast;
  969. i__2 = jinc;
  970. for (j = jfirst; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
  971. /* Compute x(j) = b(j) / A(j,j), scaling x if necessary. */
  972. xj = (d__1 = x[j], abs(d__1));
  973. if (nounit) {
  974. tjjs = a[j + j * a_dim1] * tscal;
  975. } else {
  976. tjjs = tscal;
  977. if (tscal == 1.) {
  978. goto L100;
  979. }
  980. }
  981. tjj = abs(tjjs);
  982. if (tjj > smlnum) {
  983. /* abs(A(j,j)) > SMLNUM: */
  984. if (tjj < 1.) {
  985. if (xj > tjj * bignum) {
  986. /* Scale x by 1/b(j). */
  987. rec = 1. / xj;
  988. dscal_(n, &rec, &x[1], &c__1);
  989. *scale *= rec;
  990. xmax *= rec;
  991. }
  992. }
  993. x[j] /= tjjs;
  994. xj = (d__1 = x[j], abs(d__1));
  995. } else if (tjj > 0.) {
  996. /* 0 < abs(A(j,j)) <= SMLNUM: */
  997. if (xj > tjj * bignum) {
  998. /* Scale x by (1/abs(x(j)))*abs(A(j,j))*BIGNUM */
  999. /* to avoid overflow when dividing by A(j,j). */
  1000. rec = tjj * bignum / xj;
  1001. if (cnorm[j] > 1.) {
  1002. /* Scale by 1/CNORM(j) to avoid overflow when */
  1003. /* multiplying x(j) times column j. */
  1004. rec /= cnorm[j];
  1005. }
  1006. dscal_(n, &rec, &x[1], &c__1);
  1007. *scale *= rec;
  1008. xmax *= rec;
  1009. }
  1010. x[j] /= tjjs;
  1011. xj = (d__1 = x[j], abs(d__1));
  1012. } else {
  1013. /* A(j,j) = 0: Set x(1:n) = 0, x(j) = 1, and */
  1014. /* scale = 0, and compute a solution to A*x = 0. */
  1015. i__3 = *n;
  1016. for (i__ = 1; i__ <= i__3; ++i__) {
  1017. x[i__] = 0.;
  1018. /* L90: */
  1019. }
  1020. x[j] = 1.;
  1021. xj = 1.;
  1022. *scale = 0.;
  1023. xmax = 0.;
  1024. }
  1025. L100:
  1026. /* Scale x if necessary to avoid overflow when adding a */
  1027. /* multiple of column j of A. */
  1028. if (xj > 1.) {
  1029. rec = 1. / xj;
  1030. if (cnorm[j] > (bignum - xmax) * rec) {
  1031. /* Scale x by 1/(2*abs(x(j))). */
  1032. rec *= .5;
  1033. dscal_(n, &rec, &x[1], &c__1);
  1034. *scale *= rec;
  1035. }
  1036. } else if (xj * cnorm[j] > bignum - xmax) {
  1037. /* Scale x by 1/2. */
  1038. dscal_(n, &c_b36, &x[1], &c__1);
  1039. *scale *= .5;
  1040. }
  1041. if (upper) {
  1042. if (j > 1) {
  1043. /* Compute the update */
  1044. /* x(1:j-1) := x(1:j-1) - x(j) * A(1:j-1,j) */
  1045. i__3 = j - 1;
  1046. d__1 = -x[j] * tscal;
  1047. daxpy_(&i__3, &d__1, &a[j * a_dim1 + 1], &c__1, &x[1],
  1048. &c__1);
  1049. i__3 = j - 1;
  1050. i__ = idamax_(&i__3, &x[1], &c__1);
  1051. xmax = (d__1 = x[i__], abs(d__1));
  1052. }
  1053. } else {
  1054. if (j < *n) {
  1055. /* Compute the update */
  1056. /* x(j+1:n) := x(j+1:n) - x(j) * A(j+1:n,j) */
  1057. i__3 = *n - j;
  1058. d__1 = -x[j] * tscal;
  1059. daxpy_(&i__3, &d__1, &a[j + 1 + j * a_dim1], &c__1, &
  1060. x[j + 1], &c__1);
  1061. i__3 = *n - j;
  1062. i__ = j + idamax_(&i__3, &x[j + 1], &c__1);
  1063. xmax = (d__1 = x[i__], abs(d__1));
  1064. }
  1065. }
  1066. /* L110: */
  1067. }
  1068. } else {
  1069. /* Solve A**T * x = b */
  1070. i__2 = jlast;
  1071. i__1 = jinc;
  1072. for (j = jfirst; i__1 < 0 ? j >= i__2 : j <= i__2; j += i__1) {
  1073. /* Compute x(j) = b(j) - sum A(k,j)*x(k). */
  1074. /* k<>j */
  1075. xj = (d__1 = x[j], abs(d__1));
  1076. uscal = tscal;
  1077. rec = 1. / f2cmax(xmax,1.);
  1078. if (cnorm[j] > (bignum - xj) * rec) {
  1079. /* If x(j) could overflow, scale x by 1/(2*XMAX). */
  1080. rec *= .5;
  1081. if (nounit) {
  1082. tjjs = a[j + j * a_dim1] * tscal;
  1083. } else {
  1084. tjjs = tscal;
  1085. }
  1086. tjj = abs(tjjs);
  1087. if (tjj > 1.) {
  1088. /* Divide by A(j,j) when scaling x if A(j,j) > 1. */
  1089. /* Computing MIN */
  1090. d__1 = 1., d__2 = rec * tjj;
  1091. rec = f2cmin(d__1,d__2);
  1092. uscal /= tjjs;
  1093. }
  1094. if (rec < 1.) {
  1095. dscal_(n, &rec, &x[1], &c__1);
  1096. *scale *= rec;
  1097. xmax *= rec;
  1098. }
  1099. }
  1100. sumj = 0.;
  1101. if (uscal == 1.) {
  1102. /* If the scaling needed for A in the dot product is 1, */
  1103. /* call DDOT to perform the dot product. */
  1104. if (upper) {
  1105. i__3 = j - 1;
  1106. sumj = ddot_(&i__3, &a[j * a_dim1 + 1], &c__1, &x[1],
  1107. &c__1);
  1108. } else if (j < *n) {
  1109. i__3 = *n - j;
  1110. sumj = ddot_(&i__3, &a[j + 1 + j * a_dim1], &c__1, &x[
  1111. j + 1], &c__1);
  1112. }
  1113. } else {
  1114. /* Otherwise, use in-line code for the dot product. */
  1115. if (upper) {
  1116. i__3 = j - 1;
  1117. for (i__ = 1; i__ <= i__3; ++i__) {
  1118. sumj += a[i__ + j * a_dim1] * uscal * x[i__];
  1119. /* L120: */
  1120. }
  1121. } else if (j < *n) {
  1122. i__3 = *n;
  1123. for (i__ = j + 1; i__ <= i__3; ++i__) {
  1124. sumj += a[i__ + j * a_dim1] * uscal * x[i__];
  1125. /* L130: */
  1126. }
  1127. }
  1128. }
  1129. if (uscal == tscal) {
  1130. /* Compute x(j) := ( x(j) - sumj ) / A(j,j) if 1/A(j,j) */
  1131. /* was not used to scale the dotproduct. */
  1132. x[j] -= sumj;
  1133. xj = (d__1 = x[j], abs(d__1));
  1134. if (nounit) {
  1135. tjjs = a[j + j * a_dim1] * tscal;
  1136. } else {
  1137. tjjs = tscal;
  1138. if (tscal == 1.) {
  1139. goto L150;
  1140. }
  1141. }
  1142. /* Compute x(j) = x(j) / A(j,j), scaling if necessary. */
  1143. tjj = abs(tjjs);
  1144. if (tjj > smlnum) {
  1145. /* abs(A(j,j)) > SMLNUM: */
  1146. if (tjj < 1.) {
  1147. if (xj > tjj * bignum) {
  1148. /* Scale X by 1/abs(x(j)). */
  1149. rec = 1. / xj;
  1150. dscal_(n, &rec, &x[1], &c__1);
  1151. *scale *= rec;
  1152. xmax *= rec;
  1153. }
  1154. }
  1155. x[j] /= tjjs;
  1156. } else if (tjj > 0.) {
  1157. /* 0 < abs(A(j,j)) <= SMLNUM: */
  1158. if (xj > tjj * bignum) {
  1159. /* Scale x by (1/abs(x(j)))*abs(A(j,j))*BIGNUM. */
  1160. rec = tjj * bignum / xj;
  1161. dscal_(n, &rec, &x[1], &c__1);
  1162. *scale *= rec;
  1163. xmax *= rec;
  1164. }
  1165. x[j] /= tjjs;
  1166. } else {
  1167. /* A(j,j) = 0: Set x(1:n) = 0, x(j) = 1, and */
  1168. /* scale = 0, and compute a solution to A**T*x = 0. */
  1169. i__3 = *n;
  1170. for (i__ = 1; i__ <= i__3; ++i__) {
  1171. x[i__] = 0.;
  1172. /* L140: */
  1173. }
  1174. x[j] = 1.;
  1175. *scale = 0.;
  1176. xmax = 0.;
  1177. }
  1178. L150:
  1179. ;
  1180. } else {
  1181. /* Compute x(j) := x(j) / A(j,j) - sumj if the dot */
  1182. /* product has already been divided by 1/A(j,j). */
  1183. x[j] = x[j] / tjjs - sumj;
  1184. }
  1185. /* Computing MAX */
  1186. d__2 = xmax, d__3 = (d__1 = x[j], abs(d__1));
  1187. xmax = f2cmax(d__2,d__3);
  1188. /* L160: */
  1189. }
  1190. }
  1191. *scale /= tscal;
  1192. }
  1193. /* Scale the column norms by 1/TSCAL for return. */
  1194. if (tscal != 1.) {
  1195. d__1 = 1. / tscal;
  1196. dscal_(n, &d__1, &cnorm[1], &c__1);
  1197. }
  1198. return 0;
  1199. /* End of DLATRS */
  1200. } /* dlatrs_ */