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 blasint logical;
  52. typedef char logical1;
  53. typedef char integer1;
  54. #define TRUE_ (1)
  55. #define FALSE_ (0)
  56. /* Extern is for use with -E */
  57. #ifndef Extern
  58. #define Extern extern
  59. #endif
  60. /* I/O stuff */
  61. typedef int flag;
  62. typedef int ftnlen;
  63. typedef int ftnint;
  64. /*external read, write*/
  65. typedef struct
  66. { flag cierr;
  67. ftnint ciunit;
  68. flag ciend;
  69. char *cifmt;
  70. ftnint cirec;
  71. } cilist;
  72. /*internal read, write*/
  73. typedef struct
  74. { flag icierr;
  75. char *iciunit;
  76. flag iciend;
  77. char *icifmt;
  78. ftnint icirlen;
  79. ftnint icirnum;
  80. } icilist;
  81. /*open*/
  82. typedef struct
  83. { flag oerr;
  84. ftnint ounit;
  85. char *ofnm;
  86. ftnlen ofnmlen;
  87. char *osta;
  88. char *oacc;
  89. char *ofm;
  90. ftnint orl;
  91. char *oblnk;
  92. } olist;
  93. /*close*/
  94. typedef struct
  95. { flag cerr;
  96. ftnint cunit;
  97. char *csta;
  98. } cllist;
  99. /*rewind, backspace, endfile*/
  100. typedef struct
  101. { flag aerr;
  102. ftnint aunit;
  103. } alist;
  104. /* inquire */
  105. typedef struct
  106. { flag inerr;
  107. ftnint inunit;
  108. char *infile;
  109. ftnlen infilen;
  110. ftnint *inex; /*parameters in standard's order*/
  111. ftnint *inopen;
  112. ftnint *innum;
  113. ftnint *innamed;
  114. char *inname;
  115. ftnlen innamlen;
  116. char *inacc;
  117. ftnlen inacclen;
  118. char *inseq;
  119. ftnlen inseqlen;
  120. char *indir;
  121. ftnlen indirlen;
  122. char *infmt;
  123. ftnlen infmtlen;
  124. char *inform;
  125. ftnint informlen;
  126. char *inunf;
  127. ftnlen inunflen;
  128. ftnint *inrecl;
  129. ftnint *innrec;
  130. char *inblank;
  131. ftnlen inblanklen;
  132. } inlist;
  133. #define VOID void
  134. union Multitype { /* for multiple entry points */
  135. integer1 g;
  136. shortint h;
  137. integer i;
  138. /* longint j; */
  139. real r;
  140. doublereal d;
  141. complex c;
  142. doublecomplex z;
  143. };
  144. typedef union Multitype Multitype;
  145. struct Vardesc { /* for Namelist */
  146. char *name;
  147. char *addr;
  148. ftnlen *dims;
  149. int type;
  150. };
  151. typedef struct Vardesc Vardesc;
  152. struct Namelist {
  153. char *name;
  154. Vardesc **vars;
  155. int nvars;
  156. };
  157. typedef struct Namelist Namelist;
  158. #define abs(x) ((x) >= 0 ? (x) : -(x))
  159. #define dabs(x) (fabs(x))
  160. #define f2cmin(a,b) ((a) <= (b) ? (a) : (b))
  161. #define f2cmax(a,b) ((a) >= (b) ? (a) : (b))
  162. #define dmin(a,b) (f2cmin(a,b))
  163. #define dmax(a,b) (f2cmax(a,b))
  164. #define bit_test(a,b) ((a) >> (b) & 1)
  165. #define bit_clear(a,b) ((a) & ~((uinteger)1 << (b)))
  166. #define bit_set(a,b) ((a) | ((uinteger)1 << (b)))
  167. #define abort_() { sig_die("Fortran abort routine called", 1); }
  168. #define c_abs(z) (cabsf(Cf(z)))
  169. #define c_cos(R,Z) { pCf(R)=ccos(Cf(Z)); }
  170. #ifdef _MSC_VER
  171. #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]);}
  172. #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]);}
  173. #else
  174. #define c_div(c, a, b) {pCf(c) = Cf(a)/Cf(b);}
  175. #define z_div(c, a, b) {pCd(c) = Cd(a)/Cd(b);}
  176. #endif
  177. #define c_exp(R, Z) {pCf(R) = cexpf(Cf(Z));}
  178. #define c_log(R, Z) {pCf(R) = clogf(Cf(Z));}
  179. #define c_sin(R, Z) {pCf(R) = csinf(Cf(Z));}
  180. //#define c_sqrt(R, Z) {*(R) = csqrtf(Cf(Z));}
  181. #define c_sqrt(R, Z) {pCf(R) = csqrtf(Cf(Z));}
  182. #define d_abs(x) (fabs(*(x)))
  183. #define d_acos(x) (acos(*(x)))
  184. #define d_asin(x) (asin(*(x)))
  185. #define d_atan(x) (atan(*(x)))
  186. #define d_atn2(x, y) (atan2(*(x),*(y)))
  187. #define d_cnjg(R, Z) { pCd(R) = conj(Cd(Z)); }
  188. #define r_cnjg(R, Z) { pCf(R) = conjf(Cf(Z)); }
  189. #define d_cos(x) (cos(*(x)))
  190. #define d_cosh(x) (cosh(*(x)))
  191. #define d_dim(__a, __b) ( *(__a) > *(__b) ? *(__a) - *(__b) : 0.0 )
  192. #define d_exp(x) (exp(*(x)))
  193. #define d_imag(z) (cimag(Cd(z)))
  194. #define r_imag(z) (cimagf(Cf(z)))
  195. #define d_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
  196. #define r_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
  197. #define d_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
  198. #define r_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
  199. #define d_log(x) (log(*(x)))
  200. #define d_mod(x, y) (fmod(*(x), *(y)))
  201. #define u_nint(__x) ((__x)>=0 ? floor((__x) + .5) : -floor(.5 - (__x)))
  202. #define d_nint(x) u_nint(*(x))
  203. #define u_sign(__a,__b) ((__b) >= 0 ? ((__a) >= 0 ? (__a) : -(__a)) : -((__a) >= 0 ? (__a) : -(__a)))
  204. #define d_sign(a,b) u_sign(*(a),*(b))
  205. #define r_sign(a,b) u_sign(*(a),*(b))
  206. #define d_sin(x) (sin(*(x)))
  207. #define d_sinh(x) (sinh(*(x)))
  208. #define d_sqrt(x) (sqrt(*(x)))
  209. #define d_tan(x) (tan(*(x)))
  210. #define d_tanh(x) (tanh(*(x)))
  211. #define i_abs(x) abs(*(x))
  212. #define i_dnnt(x) ((integer)u_nint(*(x)))
  213. #define i_len(s, n) (n)
  214. #define i_nint(x) ((integer)u_nint(*(x)))
  215. #define i_sign(a,b) ((integer)u_sign((integer)*(a),(integer)*(b)))
  216. #define pow_dd(ap, bp) ( pow(*(ap), *(bp)))
  217. #define pow_si(B,E) spow_ui(*(B),*(E))
  218. #define pow_ri(B,E) spow_ui(*(B),*(E))
  219. #define pow_di(B,E) dpow_ui(*(B),*(E))
  220. #define pow_zi(p, a, b) {pCd(p) = zpow_ui(Cd(a), *(b));}
  221. #define pow_ci(p, a, b) {pCf(p) = cpow_ui(Cf(a), *(b));}
  222. #define pow_zz(R,A,B) {pCd(R) = cpow(Cd(A),*(B));}
  223. #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++ = ' '; }
  224. #define s_cmp(a,b,c,d) ((integer)strncmp((a),(b),f2cmin((c),(d))))
  225. #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]; }
  226. #define sig_die(s, kill) { exit(1); }
  227. #define s_stop(s, n) {exit(0);}
  228. static char junk[] = "\n@(#)LIBF77 VERSION 19990503\n";
  229. #define z_abs(z) (cabs(Cd(z)))
  230. #define z_exp(R, Z) {pCd(R) = cexp(Cd(Z));}
  231. #define z_sqrt(R, Z) {pCd(R) = csqrt(Cd(Z));}
  232. #define myexit_() break;
  233. #define mycycle() continue;
  234. #define myceiling(w) {ceil(w)}
  235. #define myhuge(w) {HUGE_VAL}
  236. //#define mymaxloc_(w,s,e,n) {if (sizeof(*(w)) == sizeof(double)) dmaxloc_((w),*(s),*(e),n); else dmaxloc_((w),*(s),*(e),n);}
  237. #define mymaxloc(w,s,e,n) {dmaxloc_(w,*(s),*(e),n)}
  238. /* procedure parameter types for -A and -C++ */
  239. #ifdef __cplusplus
  240. typedef logical (*L_fp)(...);
  241. #else
  242. typedef logical (*L_fp)();
  243. #endif
  244. static float spow_ui(float x, integer n) {
  245. float pow=1.0; unsigned long int u;
  246. if(n != 0) {
  247. if(n < 0) n = -n, x = 1/x;
  248. for(u = n; ; ) {
  249. if(u & 01) pow *= x;
  250. if(u >>= 1) x *= x;
  251. else break;
  252. }
  253. }
  254. return pow;
  255. }
  256. static double dpow_ui(double x, integer n) {
  257. double pow=1.0; unsigned long int u;
  258. if(n != 0) {
  259. if(n < 0) n = -n, x = 1/x;
  260. for(u = n; ; ) {
  261. if(u & 01) pow *= x;
  262. if(u >>= 1) x *= x;
  263. else break;
  264. }
  265. }
  266. return pow;
  267. }
  268. #ifdef _MSC_VER
  269. static _Fcomplex cpow_ui(complex x, integer n) {
  270. complex pow={1.0,0.0}; unsigned long int u;
  271. if(n != 0) {
  272. if(n < 0) n = -n, x.r = 1/x.r, x.i=1/x.i;
  273. for(u = n; ; ) {
  274. if(u & 01) pow.r *= x.r, pow.i *= x.i;
  275. if(u >>= 1) x.r *= x.r, x.i *= x.i;
  276. else break;
  277. }
  278. }
  279. _Fcomplex p={pow.r, pow.i};
  280. return p;
  281. }
  282. #else
  283. static _Complex float cpow_ui(_Complex float x, integer n) {
  284. _Complex float pow=1.0; unsigned long int u;
  285. if(n != 0) {
  286. if(n < 0) n = -n, x = 1/x;
  287. for(u = n; ; ) {
  288. if(u & 01) pow *= x;
  289. if(u >>= 1) x *= x;
  290. else break;
  291. }
  292. }
  293. return pow;
  294. }
  295. #endif
  296. #ifdef _MSC_VER
  297. static _Dcomplex zpow_ui(_Dcomplex x, integer n) {
  298. _Dcomplex pow={1.0,0.0}; unsigned long int u;
  299. if(n != 0) {
  300. if(n < 0) n = -n, x._Val[0] = 1/x._Val[0], x._Val[1] =1/x._Val[1];
  301. for(u = n; ; ) {
  302. if(u & 01) pow._Val[0] *= x._Val[0], pow._Val[1] *= x._Val[1];
  303. if(u >>= 1) x._Val[0] *= x._Val[0], x._Val[1] *= x._Val[1];
  304. else break;
  305. }
  306. }
  307. _Dcomplex p = {pow._Val[0], pow._Val[1]};
  308. return p;
  309. }
  310. #else
  311. static _Complex double zpow_ui(_Complex double x, integer n) {
  312. _Complex double pow=1.0; unsigned long int u;
  313. if(n != 0) {
  314. if(n < 0) n = -n, x = 1/x;
  315. for(u = n; ; ) {
  316. if(u & 01) pow *= x;
  317. if(u >>= 1) x *= x;
  318. else break;
  319. }
  320. }
  321. return pow;
  322. }
  323. #endif
  324. static integer pow_ii(integer x, integer n) {
  325. integer pow; unsigned long int u;
  326. if (n <= 0) {
  327. if (n == 0 || x == 1) pow = 1;
  328. else if (x != -1) pow = x == 0 ? 1/x : 0;
  329. else n = -n;
  330. }
  331. if ((n > 0) || !(n == 0 || x == 1 || x != -1)) {
  332. u = n;
  333. for(pow = 1; ; ) {
  334. if(u & 01) pow *= x;
  335. if(u >>= 1) x *= x;
  336. else break;
  337. }
  338. }
  339. return pow;
  340. }
  341. static integer dmaxloc_(double *w, integer s, integer e, integer *n)
  342. {
  343. double m; integer i, mi;
  344. for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
  345. if (w[i-1]>m) mi=i ,m=w[i-1];
  346. return mi-s+1;
  347. }
  348. static integer smaxloc_(float *w, integer s, integer e, integer *n)
  349. {
  350. float m; integer i, mi;
  351. for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
  352. if (w[i-1]>m) mi=i ,m=w[i-1];
  353. return mi-s+1;
  354. }
  355. static inline void cdotc_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
  356. integer n = *n_, incx = *incx_, incy = *incy_, i;
  357. #ifdef _MSC_VER
  358. _Fcomplex zdotc = {0.0, 0.0};
  359. if (incx == 1 && incy == 1) {
  360. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  361. zdotc._Val[0] += conjf(Cf(&x[i]))._Val[0] * Cf(&y[i])._Val[0];
  362. zdotc._Val[1] += conjf(Cf(&x[i]))._Val[1] * Cf(&y[i])._Val[1];
  363. }
  364. } else {
  365. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  366. zdotc._Val[0] += conjf(Cf(&x[i*incx]))._Val[0] * Cf(&y[i*incy])._Val[0];
  367. zdotc._Val[1] += conjf(Cf(&x[i*incx]))._Val[1] * Cf(&y[i*incy])._Val[1];
  368. }
  369. }
  370. pCf(z) = zdotc;
  371. }
  372. #else
  373. _Complex float zdotc = 0.0;
  374. if (incx == 1 && incy == 1) {
  375. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  376. zdotc += conjf(Cf(&x[i])) * Cf(&y[i]);
  377. }
  378. } else {
  379. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  380. zdotc += conjf(Cf(&x[i*incx])) * Cf(&y[i*incy]);
  381. }
  382. }
  383. pCf(z) = zdotc;
  384. }
  385. #endif
  386. static inline void zdotc_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
  387. integer n = *n_, incx = *incx_, incy = *incy_, i;
  388. #ifdef _MSC_VER
  389. _Dcomplex zdotc = {0.0, 0.0};
  390. if (incx == 1 && incy == 1) {
  391. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  392. zdotc._Val[0] += conj(Cd(&x[i]))._Val[0] * Cd(&y[i])._Val[0];
  393. zdotc._Val[1] += conj(Cd(&x[i]))._Val[1] * Cd(&y[i])._Val[1];
  394. }
  395. } else {
  396. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  397. zdotc._Val[0] += conj(Cd(&x[i*incx]))._Val[0] * Cd(&y[i*incy])._Val[0];
  398. zdotc._Val[1] += conj(Cd(&x[i*incx]))._Val[1] * Cd(&y[i*incy])._Val[1];
  399. }
  400. }
  401. pCd(z) = zdotc;
  402. }
  403. #else
  404. _Complex double zdotc = 0.0;
  405. if (incx == 1 && incy == 1) {
  406. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  407. zdotc += conj(Cd(&x[i])) * Cd(&y[i]);
  408. }
  409. } else {
  410. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  411. zdotc += conj(Cd(&x[i*incx])) * Cd(&y[i*incy]);
  412. }
  413. }
  414. pCd(z) = zdotc;
  415. }
  416. #endif
  417. static inline void cdotu_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
  418. integer n = *n_, incx = *incx_, incy = *incy_, i;
  419. #ifdef _MSC_VER
  420. _Fcomplex zdotc = {0.0, 0.0};
  421. if (incx == 1 && incy == 1) {
  422. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  423. zdotc._Val[0] += Cf(&x[i])._Val[0] * Cf(&y[i])._Val[0];
  424. zdotc._Val[1] += Cf(&x[i])._Val[1] * Cf(&y[i])._Val[1];
  425. }
  426. } else {
  427. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  428. zdotc._Val[0] += Cf(&x[i*incx])._Val[0] * Cf(&y[i*incy])._Val[0];
  429. zdotc._Val[1] += Cf(&x[i*incx])._Val[1] * Cf(&y[i*incy])._Val[1];
  430. }
  431. }
  432. pCf(z) = zdotc;
  433. }
  434. #else
  435. _Complex float zdotc = 0.0;
  436. if (incx == 1 && incy == 1) {
  437. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  438. zdotc += Cf(&x[i]) * Cf(&y[i]);
  439. }
  440. } else {
  441. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  442. zdotc += Cf(&x[i*incx]) * Cf(&y[i*incy]);
  443. }
  444. }
  445. pCf(z) = zdotc;
  446. }
  447. #endif
  448. static inline void zdotu_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
  449. integer n = *n_, incx = *incx_, incy = *incy_, i;
  450. #ifdef _MSC_VER
  451. _Dcomplex zdotc = {0.0, 0.0};
  452. if (incx == 1 && incy == 1) {
  453. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  454. zdotc._Val[0] += Cd(&x[i])._Val[0] * Cd(&y[i])._Val[0];
  455. zdotc._Val[1] += Cd(&x[i])._Val[1] * Cd(&y[i])._Val[1];
  456. }
  457. } else {
  458. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  459. zdotc._Val[0] += Cd(&x[i*incx])._Val[0] * Cd(&y[i*incy])._Val[0];
  460. zdotc._Val[1] += Cd(&x[i*incx])._Val[1] * Cd(&y[i*incy])._Val[1];
  461. }
  462. }
  463. pCd(z) = zdotc;
  464. }
  465. #else
  466. _Complex double zdotc = 0.0;
  467. if (incx == 1 && incy == 1) {
  468. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  469. zdotc += Cd(&x[i]) * Cd(&y[i]);
  470. }
  471. } else {
  472. for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
  473. zdotc += Cd(&x[i*incx]) * Cd(&y[i*incy]);
  474. }
  475. }
  476. pCd(z) = zdotc;
  477. }
  478. #endif
  479. /* -- translated by f2c (version 20000121).
  480. You must link the resulting object file with the libraries:
  481. -lf2c -lm (in that order)
  482. */
  483. /* Table of constant values */
  484. static integer c__1 = 1;
  485. static doublereal c_b36 = .5;
  486. /* > \brief \b DLATRS solves a triangular system of equations with the scale factor set to prevent overflow.
  487. */
  488. /* =========== DOCUMENTATION =========== */
  489. /* Online html documentation available at */
  490. /* http://www.netlib.org/lapack/explore-html/ */
  491. /* > \htmlonly */
  492. /* > Download DLATRS + dependencies */
  493. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlatrs.
  494. f"> */
  495. /* > [TGZ]</a> */
  496. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlatrs.
  497. f"> */
  498. /* > [ZIP]</a> */
  499. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlatrs.
  500. f"> */
  501. /* > [TXT]</a> */
  502. /* > \endhtmlonly */
  503. /* Definition: */
  504. /* =========== */
  505. /* SUBROUTINE DLATRS( UPLO, TRANS, DIAG, NORMIN, N, A, LDA, X, SCALE, */
  506. /* CNORM, INFO ) */
  507. /* CHARACTER DIAG, NORMIN, TRANS, UPLO */
  508. /* INTEGER INFO, LDA, N */
  509. /* DOUBLE PRECISION SCALE */
  510. /* DOUBLE PRECISION A( LDA, * ), CNORM( * ), X( * ) */
  511. /* > \par Purpose: */
  512. /* ============= */
  513. /* > */
  514. /* > \verbatim */
  515. /* > */
  516. /* > DLATRS solves one of the triangular systems */
  517. /* > */
  518. /* > A *x = s*b or A**T *x = s*b */
  519. /* > */
  520. /* > with scaling to prevent overflow. Here A is an upper or lower */
  521. /* > triangular matrix, A**T denotes the transpose of A, x and b are */
  522. /* > n-element vectors, and s is a scaling factor, usually less than */
  523. /* > or equal to 1, chosen so that the components of x will be less than */
  524. /* > the overflow threshold. If the unscaled problem will not cause */
  525. /* > overflow, the Level 2 BLAS routine DTRSV is called. If the matrix A */
  526. /* > is singular (A(j,j) = 0 for some j), then s is set to 0 and a */
  527. /* > non-trivial solution to A*x = 0 is returned. */
  528. /* > \endverbatim */
  529. /* Arguments: */
  530. /* ========== */
  531. /* > \param[in] UPLO */
  532. /* > \verbatim */
  533. /* > UPLO is CHARACTER*1 */
  534. /* > Specifies whether the matrix A is upper or lower triangular. */
  535. /* > = 'U': Upper triangular */
  536. /* > = 'L': Lower triangular */
  537. /* > \endverbatim */
  538. /* > */
  539. /* > \param[in] TRANS */
  540. /* > \verbatim */
  541. /* > TRANS is CHARACTER*1 */
  542. /* > Specifies the operation applied to A. */
  543. /* > = 'N': Solve A * x = s*b (No transpose) */
  544. /* > = 'T': Solve A**T* x = s*b (Transpose) */
  545. /* > = 'C': Solve A**T* x = s*b (Conjugate transpose = Transpose) */
  546. /* > \endverbatim */
  547. /* > */
  548. /* > \param[in] DIAG */
  549. /* > \verbatim */
  550. /* > DIAG is CHARACTER*1 */
  551. /* > Specifies whether or not the matrix A is unit triangular. */
  552. /* > = 'N': Non-unit triangular */
  553. /* > = 'U': Unit triangular */
  554. /* > \endverbatim */
  555. /* > */
  556. /* > \param[in] NORMIN */
  557. /* > \verbatim */
  558. /* > NORMIN is CHARACTER*1 */
  559. /* > Specifies whether CNORM has been set or not. */
  560. /* > = 'Y': CNORM contains the column norms on entry */
  561. /* > = 'N': CNORM is not set on entry. On exit, the norms will */
  562. /* > be computed and stored in CNORM. */
  563. /* > \endverbatim */
  564. /* > */
  565. /* > \param[in] N */
  566. /* > \verbatim */
  567. /* > N is INTEGER */
  568. /* > The order of the matrix A. N >= 0. */
  569. /* > \endverbatim */
  570. /* > */
  571. /* > \param[in] A */
  572. /* > \verbatim */
  573. /* > A is DOUBLE PRECISION array, dimension (LDA,N) */
  574. /* > The triangular matrix A. If UPLO = 'U', the leading n by n */
  575. /* > upper triangular part of the array A contains the upper */
  576. /* > triangular matrix, and the strictly lower triangular part of */
  577. /* > A is not referenced. If UPLO = 'L', the leading n by n lower */
  578. /* > triangular part of the array A contains the lower triangular */
  579. /* > matrix, and the strictly upper triangular part of A is not */
  580. /* > referenced. If DIAG = 'U', the diagonal elements of A are */
  581. /* > also not referenced and are assumed to be 1. */
  582. /* > \endverbatim */
  583. /* > */
  584. /* > \param[in] LDA */
  585. /* > \verbatim */
  586. /* > LDA is INTEGER */
  587. /* > The leading dimension of the array A. LDA >= f2cmax (1,N). */
  588. /* > \endverbatim */
  589. /* > */
  590. /* > \param[in,out] X */
  591. /* > \verbatim */
  592. /* > X is DOUBLE PRECISION array, dimension (N) */
  593. /* > On entry, the right hand side b of the triangular system. */
  594. /* > On exit, X is overwritten by the solution vector x. */
  595. /* > \endverbatim */
  596. /* > */
  597. /* > \param[out] SCALE */
  598. /* > \verbatim */
  599. /* > SCALE is DOUBLE PRECISION */
  600. /* > The scaling factor s for the triangular system */
  601. /* > A * x = s*b or A**T* x = s*b. */
  602. /* > If SCALE = 0, the matrix A is singular or badly scaled, and */
  603. /* > the vector x is an exact or approximate solution to A*x = 0. */
  604. /* > \endverbatim */
  605. /* > */
  606. /* > \param[in,out] CNORM */
  607. /* > \verbatim */
  608. /* > CNORM is DOUBLE PRECISION array, dimension (N) */
  609. /* > */
  610. /* > If NORMIN = 'Y', CNORM is an input argument and CNORM(j) */
  611. /* > contains the norm of the off-diagonal part of the j-th column */
  612. /* > of A. If TRANS = 'N', CNORM(j) must be greater than or equal */
  613. /* > to the infinity-norm, and if TRANS = 'T' or 'C', CNORM(j) */
  614. /* > must be greater than or equal to the 1-norm. */
  615. /* > */
  616. /* > If NORMIN = 'N', CNORM is an output argument and CNORM(j) */
  617. /* > returns the 1-norm of the offdiagonal part of the j-th column */
  618. /* > of A. */
  619. /* > \endverbatim */
  620. /* > */
  621. /* > \param[out] INFO */
  622. /* > \verbatim */
  623. /* > INFO is INTEGER */
  624. /* > = 0: successful exit */
  625. /* > < 0: if INFO = -k, the k-th argument had an illegal value */
  626. /* > \endverbatim */
  627. /* Authors: */
  628. /* ======== */
  629. /* > \author Univ. of Tennessee */
  630. /* > \author Univ. of California Berkeley */
  631. /* > \author Univ. of Colorado Denver */
  632. /* > \author NAG Ltd. */
  633. /* > \date December 2016 */
  634. /* > \ingroup doubleOTHERauxiliary */
  635. /* > \par Further Details: */
  636. /* ===================== */
  637. /* > */
  638. /* > \verbatim */
  639. /* > */
  640. /* > A rough bound on x is computed; if that is less than overflow, DTRSV */
  641. /* > is called, otherwise, specific code is used which checks for possible */
  642. /* > overflow or divide-by-zero at every operation. */
  643. /* > */
  644. /* > A columnwise scheme is used for solving A*x = b. The basic algorithm */
  645. /* > if A is lower triangular is */
  646. /* > */
  647. /* > x[1:n] := b[1:n] */
  648. /* > for j = 1, ..., n */
  649. /* > x(j) := x(j) / A(j,j) */
  650. /* > x[j+1:n] := x[j+1:n] - x(j) * A[j+1:n,j] */
  651. /* > end */
  652. /* > */
  653. /* > Define bounds on the components of x after j iterations of the loop: */
  654. /* > M(j) = bound on x[1:j] */
  655. /* > G(j) = bound on x[j+1:n] */
  656. /* > Initially, let M(0) = 0 and G(0) = f2cmax{x(i), i=1,...,n}. */
  657. /* > */
  658. /* > Then for iteration j+1 we have */
  659. /* > M(j+1) <= G(j) / | A(j+1,j+1) | */
  660. /* > G(j+1) <= G(j) + M(j+1) * | A[j+2:n,j+1] | */
  661. /* > <= G(j) ( 1 + CNORM(j+1) / | A(j+1,j+1) | ) */
  662. /* > */
  663. /* > where CNORM(j+1) is greater than or equal to the infinity-norm of */
  664. /* > column j+1 of A, not counting the diagonal. Hence */
  665. /* > */
  666. /* > G(j) <= G(0) product ( 1 + CNORM(i) / | A(i,i) | ) */
  667. /* > 1<=i<=j */
  668. /* > and */
  669. /* > */
  670. /* > |x(j)| <= ( G(0) / |A(j,j)| ) product ( 1 + CNORM(i) / |A(i,i)| ) */
  671. /* > 1<=i< j */
  672. /* > */
  673. /* > Since |x(j)| <= M(j), we use the Level 2 BLAS routine DTRSV if the */
  674. /* > reciprocal of the largest M(j), j=1,..,n, is larger than */
  675. /* > f2cmax(underflow, 1/overflow). */
  676. /* > */
  677. /* > The bound on x(j) is also used to determine when a step in the */
  678. /* > columnwise method can be performed without fear of overflow. If */
  679. /* > the computed bound is greater than a large constant, x is scaled to */
  680. /* > prevent overflow, but if the bound overflows, x is set to 0, x(j) to */
  681. /* > 1, and scale to 0, and a non-trivial solution to A*x = 0 is found. */
  682. /* > */
  683. /* > Similarly, a row-wise scheme is used to solve A**T*x = b. The basic */
  684. /* > algorithm for A upper triangular is */
  685. /* > */
  686. /* > for j = 1, ..., n */
  687. /* > x(j) := ( b(j) - A[1:j-1,j]**T * x[1:j-1] ) / A(j,j) */
  688. /* > end */
  689. /* > */
  690. /* > We simultaneously compute two bounds */
  691. /* > G(j) = bound on ( b(i) - A[1:i-1,i]**T * x[1:i-1] ), 1<=i<=j */
  692. /* > M(j) = bound on x(i), 1<=i<=j */
  693. /* > */
  694. /* > The initial values are G(0) = 0, M(0) = f2cmax{b(i), i=1,..,n}, and we */
  695. /* > add the constraint G(j) >= G(j-1) and M(j) >= M(j-1) for j >= 1. */
  696. /* > Then the bound on x(j) is */
  697. /* > */
  698. /* > M(j) <= M(j-1) * ( 1 + CNORM(j) ) / | A(j,j) | */
  699. /* > */
  700. /* > <= M(0) * product ( ( 1 + CNORM(i) ) / |A(i,i)| ) */
  701. /* > 1<=i<=j */
  702. /* > */
  703. /* > and we can safely call DTRSV if 1/M(n) and 1/G(n) are both greater */
  704. /* > than f2cmax(underflow, 1/overflow). */
  705. /* > \endverbatim */
  706. /* > */
  707. /* ===================================================================== */
  708. /* Subroutine */ void dlatrs_(char *uplo, char *trans, char *diag, char *
  709. normin, integer *n, doublereal *a, integer *lda, doublereal *x,
  710. doublereal *scale, doublereal *cnorm, integer *info)
  711. {
  712. /* System generated locals */
  713. integer a_dim1, a_offset, i__1, i__2, i__3;
  714. doublereal d__1, d__2, d__3;
  715. /* Local variables */
  716. integer jinc;
  717. extern doublereal ddot_(integer *, doublereal *, integer *, doublereal *,
  718. integer *);
  719. doublereal xbnd;
  720. integer imax;
  721. doublereal tmax, tjjs, xmax, grow, sumj;
  722. integer i__, j;
  723. extern /* Subroutine */ void dscal_(integer *, doublereal *, doublereal *,
  724. integer *);
  725. extern logical lsame_(char *, char *);
  726. doublereal tscal, uscal;
  727. extern doublereal dasum_(integer *, doublereal *, integer *);
  728. integer jlast;
  729. extern /* Subroutine */ void daxpy_(integer *, doublereal *, doublereal *,
  730. integer *, doublereal *, integer *);
  731. logical upper;
  732. extern /* Subroutine */ void dtrsv_(char *, char *, char *, integer *,
  733. doublereal *, integer *, doublereal *, integer *);
  734. extern doublereal dlamch_(char *);
  735. doublereal xj;
  736. extern integer idamax_(integer *, doublereal *, integer *);
  737. extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
  738. doublereal bignum;
  739. logical notran;
  740. integer jfirst;
  741. doublereal smlnum;
  742. logical nounit;
  743. doublereal rec, tjj;
  744. /* -- LAPACK auxiliary routine (version 3.7.0) -- */
  745. /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
  746. /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
  747. /* December 2016 */
  748. /* ===================================================================== */
  749. /* Parameter adjustments */
  750. a_dim1 = *lda;
  751. a_offset = 1 + a_dim1 * 1;
  752. a -= a_offset;
  753. --x;
  754. --cnorm;
  755. /* Function Body */
  756. *info = 0;
  757. upper = lsame_(uplo, "U");
  758. notran = lsame_(trans, "N");
  759. nounit = lsame_(diag, "N");
  760. /* Test the input parameters. */
  761. if (! upper && ! lsame_(uplo, "L")) {
  762. *info = -1;
  763. } else if (! notran && ! lsame_(trans, "T") && !
  764. lsame_(trans, "C")) {
  765. *info = -2;
  766. } else if (! nounit && ! lsame_(diag, "U")) {
  767. *info = -3;
  768. } else if (! lsame_(normin, "Y") && ! lsame_(normin,
  769. "N")) {
  770. *info = -4;
  771. } else if (*n < 0) {
  772. *info = -5;
  773. } else if (*lda < f2cmax(1,*n)) {
  774. *info = -7;
  775. }
  776. if (*info != 0) {
  777. i__1 = -(*info);
  778. xerbla_("DLATRS", &i__1, (ftnlen)6);
  779. return;
  780. }
  781. /* Quick return if possible */
  782. if (*n == 0) {
  783. return;
  784. }
  785. /* Determine machine dependent parameters to control overflow. */
  786. smlnum = dlamch_("Safe minimum") / dlamch_("Precision");
  787. bignum = 1. / smlnum;
  788. *scale = 1.;
  789. if (lsame_(normin, "N")) {
  790. /* Compute the 1-norm of each column, not including the diagonal. */
  791. if (upper) {
  792. /* A is upper triangular. */
  793. i__1 = *n;
  794. for (j = 1; j <= i__1; ++j) {
  795. i__2 = j - 1;
  796. cnorm[j] = dasum_(&i__2, &a[j * a_dim1 + 1], &c__1);
  797. /* L10: */
  798. }
  799. } else {
  800. /* A is lower triangular. */
  801. i__1 = *n - 1;
  802. for (j = 1; j <= i__1; ++j) {
  803. i__2 = *n - j;
  804. cnorm[j] = dasum_(&i__2, &a[j + 1 + j * a_dim1], &c__1);
  805. /* L20: */
  806. }
  807. cnorm[*n] = 0.;
  808. }
  809. }
  810. /* Scale the column norms by TSCAL if the maximum element in CNORM is */
  811. /* greater than BIGNUM. */
  812. imax = idamax_(n, &cnorm[1], &c__1);
  813. tmax = cnorm[imax];
  814. if (tmax <= bignum) {
  815. tscal = 1.;
  816. } else {
  817. tscal = 1. / (smlnum * tmax);
  818. dscal_(n, &tscal, &cnorm[1], &c__1);
  819. }
  820. /* Compute a bound on the computed solution vector to see if the */
  821. /* Level 2 BLAS routine DTRSV can be used. */
  822. j = idamax_(n, &x[1], &c__1);
  823. xmax = (d__1 = x[j], abs(d__1));
  824. xbnd = xmax;
  825. if (notran) {
  826. /* Compute the growth in A * x = b. */
  827. if (upper) {
  828. jfirst = *n;
  829. jlast = 1;
  830. jinc = -1;
  831. } else {
  832. jfirst = 1;
  833. jlast = *n;
  834. jinc = 1;
  835. }
  836. if (tscal != 1.) {
  837. grow = 0.;
  838. goto L50;
  839. }
  840. if (nounit) {
  841. /* A is non-unit triangular. */
  842. /* Compute GROW = 1/G(j) and XBND = 1/M(j). */
  843. /* Initially, G(0) = f2cmax{x(i), i=1,...,n}. */
  844. grow = 1. / f2cmax(xbnd,smlnum);
  845. xbnd = grow;
  846. i__1 = jlast;
  847. i__2 = jinc;
  848. for (j = jfirst; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
  849. /* Exit the loop if the growth factor is too small. */
  850. if (grow <= smlnum) {
  851. goto L50;
  852. }
  853. /* M(j) = G(j-1) / abs(A(j,j)) */
  854. tjj = (d__1 = a[j + j * a_dim1], abs(d__1));
  855. /* Computing MIN */
  856. d__1 = xbnd, d__2 = f2cmin(1.,tjj) * grow;
  857. xbnd = f2cmin(d__1,d__2);
  858. if (tjj + cnorm[j] >= smlnum) {
  859. /* G(j) = G(j-1)*( 1 + CNORM(j) / abs(A(j,j)) ) */
  860. grow *= tjj / (tjj + cnorm[j]);
  861. } else {
  862. /* G(j) could overflow, set GROW to 0. */
  863. grow = 0.;
  864. }
  865. /* L30: */
  866. }
  867. grow = xbnd;
  868. } else {
  869. /* A is unit triangular. */
  870. /* Compute GROW = 1/G(j), where G(0) = f2cmax{x(i), i=1,...,n}. */
  871. /* Computing MIN */
  872. d__1 = 1., d__2 = 1. / f2cmax(xbnd,smlnum);
  873. grow = f2cmin(d__1,d__2);
  874. i__2 = jlast;
  875. i__1 = jinc;
  876. for (j = jfirst; i__1 < 0 ? j >= i__2 : j <= i__2; j += i__1) {
  877. /* Exit the loop if the growth factor is too small. */
  878. if (grow <= smlnum) {
  879. goto L50;
  880. }
  881. /* G(j) = G(j-1)*( 1 + CNORM(j) ) */
  882. grow *= 1. / (cnorm[j] + 1.);
  883. /* L40: */
  884. }
  885. }
  886. L50:
  887. ;
  888. } else {
  889. /* Compute the growth in A**T * x = b. */
  890. if (upper) {
  891. jfirst = 1;
  892. jlast = *n;
  893. jinc = 1;
  894. } else {
  895. jfirst = *n;
  896. jlast = 1;
  897. jinc = -1;
  898. }
  899. if (tscal != 1.) {
  900. grow = 0.;
  901. goto L80;
  902. }
  903. if (nounit) {
  904. /* A is non-unit triangular. */
  905. /* Compute GROW = 1/G(j) and XBND = 1/M(j). */
  906. /* Initially, M(0) = f2cmax{x(i), i=1,...,n}. */
  907. grow = 1. / f2cmax(xbnd,smlnum);
  908. xbnd = grow;
  909. i__1 = jlast;
  910. i__2 = jinc;
  911. for (j = jfirst; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
  912. /* Exit the loop if the growth factor is too small. */
  913. if (grow <= smlnum) {
  914. goto L80;
  915. }
  916. /* G(j) = f2cmax( G(j-1), M(j-1)*( 1 + CNORM(j) ) ) */
  917. xj = cnorm[j] + 1.;
  918. /* Computing MIN */
  919. d__1 = grow, d__2 = xbnd / xj;
  920. grow = f2cmin(d__1,d__2);
  921. /* M(j) = M(j-1)*( 1 + CNORM(j) ) / abs(A(j,j)) */
  922. tjj = (d__1 = a[j + j * a_dim1], abs(d__1));
  923. if (xj > tjj) {
  924. xbnd *= tjj / xj;
  925. }
  926. /* L60: */
  927. }
  928. grow = f2cmin(grow,xbnd);
  929. } else {
  930. /* A is unit triangular. */
  931. /* Compute GROW = 1/G(j), where G(0) = f2cmax{x(i), i=1,...,n}. */
  932. /* Computing MIN */
  933. d__1 = 1., d__2 = 1. / f2cmax(xbnd,smlnum);
  934. grow = f2cmin(d__1,d__2);
  935. i__2 = jlast;
  936. i__1 = jinc;
  937. for (j = jfirst; i__1 < 0 ? j >= i__2 : j <= i__2; j += i__1) {
  938. /* Exit the loop if the growth factor is too small. */
  939. if (grow <= smlnum) {
  940. goto L80;
  941. }
  942. /* G(j) = ( 1 + CNORM(j) )*G(j-1) */
  943. xj = cnorm[j] + 1.;
  944. grow /= xj;
  945. /* L70: */
  946. }
  947. }
  948. L80:
  949. ;
  950. }
  951. if (grow * tscal > smlnum) {
  952. /* Use the Level 2 BLAS solve if the reciprocal of the bound on */
  953. /* elements of X is not too small. */
  954. dtrsv_(uplo, trans, diag, n, &a[a_offset], lda, &x[1], &c__1);
  955. } else {
  956. /* Use a Level 1 BLAS solve, scaling intermediate results. */
  957. if (xmax > bignum) {
  958. /* Scale X so that its components are less than or equal to */
  959. /* BIGNUM in absolute value. */
  960. *scale = bignum / xmax;
  961. dscal_(n, scale, &x[1], &c__1);
  962. xmax = bignum;
  963. }
  964. if (notran) {
  965. /* Solve A * x = b */
  966. i__1 = jlast;
  967. i__2 = jinc;
  968. for (j = jfirst; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
  969. /* Compute x(j) = b(j) / A(j,j), scaling x if necessary. */
  970. xj = (d__1 = x[j], abs(d__1));
  971. if (nounit) {
  972. tjjs = a[j + j * a_dim1] * tscal;
  973. } else {
  974. tjjs = tscal;
  975. if (tscal == 1.) {
  976. goto L100;
  977. }
  978. }
  979. tjj = abs(tjjs);
  980. if (tjj > smlnum) {
  981. /* abs(A(j,j)) > SMLNUM: */
  982. if (tjj < 1.) {
  983. if (xj > tjj * bignum) {
  984. /* Scale x by 1/b(j). */
  985. rec = 1. / xj;
  986. dscal_(n, &rec, &x[1], &c__1);
  987. *scale *= rec;
  988. xmax *= rec;
  989. }
  990. }
  991. x[j] /= tjjs;
  992. xj = (d__1 = x[j], abs(d__1));
  993. } else if (tjj > 0.) {
  994. /* 0 < abs(A(j,j)) <= SMLNUM: */
  995. if (xj > tjj * bignum) {
  996. /* Scale x by (1/abs(x(j)))*abs(A(j,j))*BIGNUM */
  997. /* to avoid overflow when dividing by A(j,j). */
  998. rec = tjj * bignum / xj;
  999. if (cnorm[j] > 1.) {
  1000. /* Scale by 1/CNORM(j) to avoid overflow when */
  1001. /* multiplying x(j) times column j. */
  1002. rec /= cnorm[j];
  1003. }
  1004. dscal_(n, &rec, &x[1], &c__1);
  1005. *scale *= rec;
  1006. xmax *= rec;
  1007. }
  1008. x[j] /= tjjs;
  1009. xj = (d__1 = x[j], abs(d__1));
  1010. } else {
  1011. /* A(j,j) = 0: Set x(1:n) = 0, x(j) = 1, and */
  1012. /* scale = 0, and compute a solution to A*x = 0. */
  1013. i__3 = *n;
  1014. for (i__ = 1; i__ <= i__3; ++i__) {
  1015. x[i__] = 0.;
  1016. /* L90: */
  1017. }
  1018. x[j] = 1.;
  1019. xj = 1.;
  1020. *scale = 0.;
  1021. xmax = 0.;
  1022. }
  1023. L100:
  1024. /* Scale x if necessary to avoid overflow when adding a */
  1025. /* multiple of column j of A. */
  1026. if (xj > 1.) {
  1027. rec = 1. / xj;
  1028. if (cnorm[j] > (bignum - xmax) * rec) {
  1029. /* Scale x by 1/(2*abs(x(j))). */
  1030. rec *= .5;
  1031. dscal_(n, &rec, &x[1], &c__1);
  1032. *scale *= rec;
  1033. }
  1034. } else if (xj * cnorm[j] > bignum - xmax) {
  1035. /* Scale x by 1/2. */
  1036. dscal_(n, &c_b36, &x[1], &c__1);
  1037. *scale *= .5;
  1038. }
  1039. if (upper) {
  1040. if (j > 1) {
  1041. /* Compute the update */
  1042. /* x(1:j-1) := x(1:j-1) - x(j) * A(1:j-1,j) */
  1043. i__3 = j - 1;
  1044. d__1 = -x[j] * tscal;
  1045. daxpy_(&i__3, &d__1, &a[j * a_dim1 + 1], &c__1, &x[1],
  1046. &c__1);
  1047. i__3 = j - 1;
  1048. i__ = idamax_(&i__3, &x[1], &c__1);
  1049. xmax = (d__1 = x[i__], abs(d__1));
  1050. }
  1051. } else {
  1052. if (j < *n) {
  1053. /* Compute the update */
  1054. /* x(j+1:n) := x(j+1:n) - x(j) * A(j+1:n,j) */
  1055. i__3 = *n - j;
  1056. d__1 = -x[j] * tscal;
  1057. daxpy_(&i__3, &d__1, &a[j + 1 + j * a_dim1], &c__1, &
  1058. x[j + 1], &c__1);
  1059. i__3 = *n - j;
  1060. i__ = j + idamax_(&i__3, &x[j + 1], &c__1);
  1061. xmax = (d__1 = x[i__], abs(d__1));
  1062. }
  1063. }
  1064. /* L110: */
  1065. }
  1066. } else {
  1067. /* Solve A**T * x = b */
  1068. i__2 = jlast;
  1069. i__1 = jinc;
  1070. for (j = jfirst; i__1 < 0 ? j >= i__2 : j <= i__2; j += i__1) {
  1071. /* Compute x(j) = b(j) - sum A(k,j)*x(k). */
  1072. /* k<>j */
  1073. xj = (d__1 = x[j], abs(d__1));
  1074. uscal = tscal;
  1075. rec = 1. / f2cmax(xmax,1.);
  1076. if (cnorm[j] > (bignum - xj) * rec) {
  1077. /* If x(j) could overflow, scale x by 1/(2*XMAX). */
  1078. rec *= .5;
  1079. if (nounit) {
  1080. tjjs = a[j + j * a_dim1] * tscal;
  1081. } else {
  1082. tjjs = tscal;
  1083. }
  1084. tjj = abs(tjjs);
  1085. if (tjj > 1.) {
  1086. /* Divide by A(j,j) when scaling x if A(j,j) > 1. */
  1087. /* Computing MIN */
  1088. d__1 = 1., d__2 = rec * tjj;
  1089. rec = f2cmin(d__1,d__2);
  1090. uscal /= tjjs;
  1091. }
  1092. if (rec < 1.) {
  1093. dscal_(n, &rec, &x[1], &c__1);
  1094. *scale *= rec;
  1095. xmax *= rec;
  1096. }
  1097. }
  1098. sumj = 0.;
  1099. if (uscal == 1.) {
  1100. /* If the scaling needed for A in the dot product is 1, */
  1101. /* call DDOT to perform the dot product. */
  1102. if (upper) {
  1103. i__3 = j - 1;
  1104. sumj = ddot_(&i__3, &a[j * a_dim1 + 1], &c__1, &x[1],
  1105. &c__1);
  1106. } else if (j < *n) {
  1107. i__3 = *n - j;
  1108. sumj = ddot_(&i__3, &a[j + 1 + j * a_dim1], &c__1, &x[
  1109. j + 1], &c__1);
  1110. }
  1111. } else {
  1112. /* Otherwise, use in-line code for the dot product. */
  1113. if (upper) {
  1114. i__3 = j - 1;
  1115. for (i__ = 1; i__ <= i__3; ++i__) {
  1116. sumj += a[i__ + j * a_dim1] * uscal * x[i__];
  1117. /* L120: */
  1118. }
  1119. } else if (j < *n) {
  1120. i__3 = *n;
  1121. for (i__ = j + 1; i__ <= i__3; ++i__) {
  1122. sumj += a[i__ + j * a_dim1] * uscal * x[i__];
  1123. /* L130: */
  1124. }
  1125. }
  1126. }
  1127. if (uscal == tscal) {
  1128. /* Compute x(j) := ( x(j) - sumj ) / A(j,j) if 1/A(j,j) */
  1129. /* was not used to scale the dotproduct. */
  1130. x[j] -= sumj;
  1131. xj = (d__1 = x[j], abs(d__1));
  1132. if (nounit) {
  1133. tjjs = a[j + j * a_dim1] * tscal;
  1134. } else {
  1135. tjjs = tscal;
  1136. if (tscal == 1.) {
  1137. goto L150;
  1138. }
  1139. }
  1140. /* Compute x(j) = x(j) / A(j,j), scaling if necessary. */
  1141. tjj = abs(tjjs);
  1142. if (tjj > smlnum) {
  1143. /* abs(A(j,j)) > SMLNUM: */
  1144. if (tjj < 1.) {
  1145. if (xj > tjj * bignum) {
  1146. /* Scale X by 1/abs(x(j)). */
  1147. rec = 1. / xj;
  1148. dscal_(n, &rec, &x[1], &c__1);
  1149. *scale *= rec;
  1150. xmax *= rec;
  1151. }
  1152. }
  1153. x[j] /= tjjs;
  1154. } else if (tjj > 0.) {
  1155. /* 0 < abs(A(j,j)) <= SMLNUM: */
  1156. if (xj > tjj * bignum) {
  1157. /* Scale x by (1/abs(x(j)))*abs(A(j,j))*BIGNUM. */
  1158. rec = tjj * bignum / xj;
  1159. dscal_(n, &rec, &x[1], &c__1);
  1160. *scale *= rec;
  1161. xmax *= rec;
  1162. }
  1163. x[j] /= tjjs;
  1164. } else {
  1165. /* A(j,j) = 0: Set x(1:n) = 0, x(j) = 1, and */
  1166. /* scale = 0, and compute a solution to A**T*x = 0. */
  1167. i__3 = *n;
  1168. for (i__ = 1; i__ <= i__3; ++i__) {
  1169. x[i__] = 0.;
  1170. /* L140: */
  1171. }
  1172. x[j] = 1.;
  1173. *scale = 0.;
  1174. xmax = 0.;
  1175. }
  1176. L150:
  1177. ;
  1178. } else {
  1179. /* Compute x(j) := x(j) / A(j,j) - sumj if the dot */
  1180. /* product has already been divided by 1/A(j,j). */
  1181. x[j] = x[j] / tjjs - sumj;
  1182. }
  1183. /* Computing MAX */
  1184. d__2 = xmax, d__3 = (d__1 = x[j], abs(d__1));
  1185. xmax = f2cmax(d__2,d__3);
  1186. /* L160: */
  1187. }
  1188. }
  1189. *scale /= tscal;
  1190. }
  1191. /* Scale the column norms by 1/TSCAL for return. */
  1192. if (tscal != 1.) {
  1193. d__1 = 1. / tscal;
  1194. dscal_(n, &d__1, &cnorm[1], &c__1);
  1195. }
  1196. return;
  1197. /* End of DLATRS */
  1198. } /* dlatrs_ */