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.

dlatps.c 37 kB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391
  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 DLATPS solves a triangular system of equations with the matrix held in packed storage. */
  487. /* =========== DOCUMENTATION =========== */
  488. /* Online html documentation available at */
  489. /* http://www.netlib.org/lapack/explore-html/ */
  490. /* > \htmlonly */
  491. /* > Download DLATPS + dependencies */
  492. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlatps.
  493. f"> */
  494. /* > [TGZ]</a> */
  495. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlatps.
  496. f"> */
  497. /* > [ZIP]</a> */
  498. /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlatps.
  499. f"> */
  500. /* > [TXT]</a> */
  501. /* > \endhtmlonly */
  502. /* Definition: */
  503. /* =========== */
  504. /* SUBROUTINE DLATPS( UPLO, TRANS, DIAG, NORMIN, N, AP, X, SCALE, */
  505. /* CNORM, INFO ) */
  506. /* CHARACTER DIAG, NORMIN, TRANS, UPLO */
  507. /* INTEGER INFO, N */
  508. /* DOUBLE PRECISION SCALE */
  509. /* DOUBLE PRECISION AP( * ), CNORM( * ), X( * ) */
  510. /* > \par Purpose: */
  511. /* ============= */
  512. /* > */
  513. /* > \verbatim */
  514. /* > */
  515. /* > DLATPS solves one of the triangular systems */
  516. /* > */
  517. /* > A *x = s*b or A**T*x = s*b */
  518. /* > */
  519. /* > with scaling to prevent overflow, where A is an upper or lower */
  520. /* > triangular matrix stored in packed form. Here A**T denotes the */
  521. /* > transpose of A, x and b are n-element vectors, and s is a scaling */
  522. /* > factor, usually less than or equal to 1, chosen so that the */
  523. /* > components of x will be less than the overflow threshold. If the */
  524. /* > unscaled problem will not cause overflow, the Level 2 BLAS routine */
  525. /* > DTPSV is called. If the matrix A is singular (A(j,j) = 0 for some j), */
  526. /* > then s is set to 0 and a non-trivial solution to A*x = 0 is returned. */
  527. /* > \endverbatim */
  528. /* Arguments: */
  529. /* ========== */
  530. /* > \param[in] UPLO */
  531. /* > \verbatim */
  532. /* > UPLO is CHARACTER*1 */
  533. /* > Specifies whether the matrix A is upper or lower triangular. */
  534. /* > = 'U': Upper triangular */
  535. /* > = 'L': Lower triangular */
  536. /* > \endverbatim */
  537. /* > */
  538. /* > \param[in] TRANS */
  539. /* > \verbatim */
  540. /* > TRANS is CHARACTER*1 */
  541. /* > Specifies the operation applied to A. */
  542. /* > = 'N': Solve A * x = s*b (No transpose) */
  543. /* > = 'T': Solve A**T* x = s*b (Transpose) */
  544. /* > = 'C': Solve A**T* x = s*b (Conjugate transpose = Transpose) */
  545. /* > \endverbatim */
  546. /* > */
  547. /* > \param[in] DIAG */
  548. /* > \verbatim */
  549. /* > DIAG is CHARACTER*1 */
  550. /* > Specifies whether or not the matrix A is unit triangular. */
  551. /* > = 'N': Non-unit triangular */
  552. /* > = 'U': Unit triangular */
  553. /* > \endverbatim */
  554. /* > */
  555. /* > \param[in] NORMIN */
  556. /* > \verbatim */
  557. /* > NORMIN is CHARACTER*1 */
  558. /* > Specifies whether CNORM has been set or not. */
  559. /* > = 'Y': CNORM contains the column norms on entry */
  560. /* > = 'N': CNORM is not set on entry. On exit, the norms will */
  561. /* > be computed and stored in CNORM. */
  562. /* > \endverbatim */
  563. /* > */
  564. /* > \param[in] N */
  565. /* > \verbatim */
  566. /* > N is INTEGER */
  567. /* > The order of the matrix A. N >= 0. */
  568. /* > \endverbatim */
  569. /* > */
  570. /* > \param[in] AP */
  571. /* > \verbatim */
  572. /* > AP is DOUBLE PRECISION array, dimension (N*(N+1)/2) */
  573. /* > The upper or lower triangular matrix A, packed columnwise in */
  574. /* > a linear array. The j-th column of A is stored in the array */
  575. /* > AP as follows: */
  576. /* > if UPLO = 'U', AP(i + (j-1)*j/2) = A(i,j) for 1<=i<=j; */
  577. /* > if UPLO = 'L', AP(i + (j-1)*(2n-j)/2) = A(i,j) for j<=i<=n. */
  578. /* > \endverbatim */
  579. /* > */
  580. /* > \param[in,out] X */
  581. /* > \verbatim */
  582. /* > X is DOUBLE PRECISION array, dimension (N) */
  583. /* > On entry, the right hand side b of the triangular system. */
  584. /* > On exit, X is overwritten by the solution vector x. */
  585. /* > \endverbatim */
  586. /* > */
  587. /* > \param[out] SCALE */
  588. /* > \verbatim */
  589. /* > SCALE is DOUBLE PRECISION */
  590. /* > The scaling factor s for the triangular system */
  591. /* > A * x = s*b or A**T* x = s*b. */
  592. /* > If SCALE = 0, the matrix A is singular or badly scaled, and */
  593. /* > the vector x is an exact or approximate solution to A*x = 0. */
  594. /* > \endverbatim */
  595. /* > */
  596. /* > \param[in,out] CNORM */
  597. /* > \verbatim */
  598. /* > CNORM is DOUBLE PRECISION array, dimension (N) */
  599. /* > */
  600. /* > If NORMIN = 'Y', CNORM is an input argument and CNORM(j) */
  601. /* > contains the norm of the off-diagonal part of the j-th column */
  602. /* > of A. If TRANS = 'N', CNORM(j) must be greater than or equal */
  603. /* > to the infinity-norm, and if TRANS = 'T' or 'C', CNORM(j) */
  604. /* > must be greater than or equal to the 1-norm. */
  605. /* > */
  606. /* > If NORMIN = 'N', CNORM is an output argument and CNORM(j) */
  607. /* > returns the 1-norm of the offdiagonal part of the j-th column */
  608. /* > of A. */
  609. /* > \endverbatim */
  610. /* > */
  611. /* > \param[out] INFO */
  612. /* > \verbatim */
  613. /* > INFO is INTEGER */
  614. /* > = 0: successful exit */
  615. /* > < 0: if INFO = -k, the k-th argument had an illegal value */
  616. /* > \endverbatim */
  617. /* Authors: */
  618. /* ======== */
  619. /* > \author Univ. of Tennessee */
  620. /* > \author Univ. of California Berkeley */
  621. /* > \author Univ. of Colorado Denver */
  622. /* > \author NAG Ltd. */
  623. /* > \date December 2016 */
  624. /* > \ingroup doubleOTHERauxiliary */
  625. /* > \par Further Details: */
  626. /* ===================== */
  627. /* > */
  628. /* > \verbatim */
  629. /* > */
  630. /* > A rough bound on x is computed; if that is less than overflow, DTPSV */
  631. /* > is called, otherwise, specific code is used which checks for possible */
  632. /* > overflow or divide-by-zero at every operation. */
  633. /* > */
  634. /* > A columnwise scheme is used for solving A*x = b. The basic algorithm */
  635. /* > if A is lower triangular is */
  636. /* > */
  637. /* > x[1:n] := b[1:n] */
  638. /* > for j = 1, ..., n */
  639. /* > x(j) := x(j) / A(j,j) */
  640. /* > x[j+1:n] := x[j+1:n] - x(j) * A[j+1:n,j] */
  641. /* > end */
  642. /* > */
  643. /* > Define bounds on the components of x after j iterations of the loop: */
  644. /* > M(j) = bound on x[1:j] */
  645. /* > G(j) = bound on x[j+1:n] */
  646. /* > Initially, let M(0) = 0 and G(0) = f2cmax{x(i), i=1,...,n}. */
  647. /* > */
  648. /* > Then for iteration j+1 we have */
  649. /* > M(j+1) <= G(j) / | A(j+1,j+1) | */
  650. /* > G(j+1) <= G(j) + M(j+1) * | A[j+2:n,j+1] | */
  651. /* > <= G(j) ( 1 + CNORM(j+1) / | A(j+1,j+1) | ) */
  652. /* > */
  653. /* > where CNORM(j+1) is greater than or equal to the infinity-norm of */
  654. /* > column j+1 of A, not counting the diagonal. Hence */
  655. /* > */
  656. /* > G(j) <= G(0) product ( 1 + CNORM(i) / | A(i,i) | ) */
  657. /* > 1<=i<=j */
  658. /* > and */
  659. /* > */
  660. /* > |x(j)| <= ( G(0) / |A(j,j)| ) product ( 1 + CNORM(i) / |A(i,i)| ) */
  661. /* > 1<=i< j */
  662. /* > */
  663. /* > Since |x(j)| <= M(j), we use the Level 2 BLAS routine DTPSV if the */
  664. /* > reciprocal of the largest M(j), j=1,..,n, is larger than */
  665. /* > f2cmax(underflow, 1/overflow). */
  666. /* > */
  667. /* > The bound on x(j) is also used to determine when a step in the */
  668. /* > columnwise method can be performed without fear of overflow. If */
  669. /* > the computed bound is greater than a large constant, x is scaled to */
  670. /* > prevent overflow, but if the bound overflows, x is set to 0, x(j) to */
  671. /* > 1, and scale to 0, and a non-trivial solution to A*x = 0 is found. */
  672. /* > */
  673. /* > Similarly, a row-wise scheme is used to solve A**T*x = b. The basic */
  674. /* > algorithm for A upper triangular is */
  675. /* > */
  676. /* > for j = 1, ..., n */
  677. /* > x(j) := ( b(j) - A[1:j-1,j]**T * x[1:j-1] ) / A(j,j) */
  678. /* > end */
  679. /* > */
  680. /* > We simultaneously compute two bounds */
  681. /* > G(j) = bound on ( b(i) - A[1:i-1,i]**T * x[1:i-1] ), 1<=i<=j */
  682. /* > M(j) = bound on x(i), 1<=i<=j */
  683. /* > */
  684. /* > The initial values are G(0) = 0, M(0) = f2cmax{b(i), i=1,..,n}, and we */
  685. /* > add the constraint G(j) >= G(j-1) and M(j) >= M(j-1) for j >= 1. */
  686. /* > Then the bound on x(j) is */
  687. /* > */
  688. /* > M(j) <= M(j-1) * ( 1 + CNORM(j) ) / | A(j,j) | */
  689. /* > */
  690. /* > <= M(0) * product ( ( 1 + CNORM(i) ) / |A(i,i)| ) */
  691. /* > 1<=i<=j */
  692. /* > */
  693. /* > and we can safely call DTPSV if 1/M(n) and 1/G(n) are both greater */
  694. /* > than f2cmax(underflow, 1/overflow). */
  695. /* > \endverbatim */
  696. /* > */
  697. /* ===================================================================== */
  698. /* Subroutine */ void dlatps_(char *uplo, char *trans, char *diag, char *
  699. normin, integer *n, doublereal *ap, doublereal *x, doublereal *scale,
  700. doublereal *cnorm, integer *info)
  701. {
  702. /* System generated locals */
  703. integer i__1, i__2, i__3;
  704. doublereal d__1, d__2, d__3;
  705. /* Local variables */
  706. integer jinc, jlen;
  707. extern doublereal ddot_(integer *, doublereal *, integer *, doublereal *,
  708. integer *);
  709. doublereal xbnd;
  710. integer imax;
  711. doublereal tmax, tjjs, xmax, grow, sumj;
  712. integer i__, j;
  713. extern /* Subroutine */ void dscal_(integer *, doublereal *, doublereal *,
  714. integer *);
  715. extern logical lsame_(char *, char *);
  716. doublereal tscal, uscal;
  717. extern doublereal dasum_(integer *, doublereal *, integer *);
  718. integer jlast;
  719. extern /* Subroutine */ void daxpy_(integer *, doublereal *, doublereal *,
  720. integer *, doublereal *, integer *);
  721. logical upper;
  722. extern /* Subroutine */ void dtpsv_(char *, char *, char *, integer *,
  723. doublereal *, doublereal *, integer *);
  724. extern doublereal dlamch_(char *);
  725. integer ip;
  726. doublereal xj;
  727. extern integer idamax_(integer *, doublereal *, integer *);
  728. extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
  729. doublereal bignum;
  730. logical notran;
  731. integer jfirst;
  732. doublereal smlnum;
  733. logical nounit;
  734. doublereal rec, tjj;
  735. /* -- LAPACK auxiliary routine (version 3.7.0) -- */
  736. /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
  737. /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
  738. /* December 2016 */
  739. /* ===================================================================== */
  740. /* Parameter adjustments */
  741. --cnorm;
  742. --x;
  743. --ap;
  744. /* Function Body */
  745. *info = 0;
  746. upper = lsame_(uplo, "U");
  747. notran = lsame_(trans, "N");
  748. nounit = lsame_(diag, "N");
  749. /* Test the input parameters. */
  750. if (! upper && ! lsame_(uplo, "L")) {
  751. *info = -1;
  752. } else if (! notran && ! lsame_(trans, "T") && !
  753. lsame_(trans, "C")) {
  754. *info = -2;
  755. } else if (! nounit && ! lsame_(diag, "U")) {
  756. *info = -3;
  757. } else if (! lsame_(normin, "Y") && ! lsame_(normin,
  758. "N")) {
  759. *info = -4;
  760. } else if (*n < 0) {
  761. *info = -5;
  762. }
  763. if (*info != 0) {
  764. i__1 = -(*info);
  765. xerbla_("DLATPS", &i__1, (ftnlen)6);
  766. return;
  767. }
  768. /* Quick return if possible */
  769. if (*n == 0) {
  770. return;
  771. }
  772. /* Determine machine dependent parameters to control overflow. */
  773. smlnum = dlamch_("Safe minimum") / dlamch_("Precision");
  774. bignum = 1. / smlnum;
  775. *scale = 1.;
  776. if (lsame_(normin, "N")) {
  777. /* Compute the 1-norm of each column, not including the diagonal. */
  778. if (upper) {
  779. /* A is upper triangular. */
  780. ip = 1;
  781. i__1 = *n;
  782. for (j = 1; j <= i__1; ++j) {
  783. i__2 = j - 1;
  784. cnorm[j] = dasum_(&i__2, &ap[ip], &c__1);
  785. ip += j;
  786. /* L10: */
  787. }
  788. } else {
  789. /* A is lower triangular. */
  790. ip = 1;
  791. i__1 = *n - 1;
  792. for (j = 1; j <= i__1; ++j) {
  793. i__2 = *n - j;
  794. cnorm[j] = dasum_(&i__2, &ap[ip + 1], &c__1);
  795. ip = ip + *n - j + 1;
  796. /* L20: */
  797. }
  798. cnorm[*n] = 0.;
  799. }
  800. }
  801. /* Scale the column norms by TSCAL if the maximum element in CNORM is */
  802. /* greater than BIGNUM. */
  803. imax = idamax_(n, &cnorm[1], &c__1);
  804. tmax = cnorm[imax];
  805. if (tmax <= bignum) {
  806. tscal = 1.;
  807. } else {
  808. tscal = 1. / (smlnum * tmax);
  809. dscal_(n, &tscal, &cnorm[1], &c__1);
  810. }
  811. /* Compute a bound on the computed solution vector to see if the */
  812. /* Level 2 BLAS routine DTPSV can be used. */
  813. j = idamax_(n, &x[1], &c__1);
  814. xmax = (d__1 = x[j], abs(d__1));
  815. xbnd = xmax;
  816. if (notran) {
  817. /* Compute the growth in A * x = b. */
  818. if (upper) {
  819. jfirst = *n;
  820. jlast = 1;
  821. jinc = -1;
  822. } else {
  823. jfirst = 1;
  824. jlast = *n;
  825. jinc = 1;
  826. }
  827. if (tscal != 1.) {
  828. grow = 0.;
  829. goto L50;
  830. }
  831. if (nounit) {
  832. /* A is non-unit triangular. */
  833. /* Compute GROW = 1/G(j) and XBND = 1/M(j). */
  834. /* Initially, G(0) = f2cmax{x(i), i=1,...,n}. */
  835. grow = 1. / f2cmax(xbnd,smlnum);
  836. xbnd = grow;
  837. ip = jfirst * (jfirst + 1) / 2;
  838. jlen = *n;
  839. i__1 = jlast;
  840. i__2 = jinc;
  841. for (j = jfirst; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
  842. /* Exit the loop if the growth factor is too small. */
  843. if (grow <= smlnum) {
  844. goto L50;
  845. }
  846. /* M(j) = G(j-1) / abs(A(j,j)) */
  847. tjj = (d__1 = ap[ip], abs(d__1));
  848. /* Computing MIN */
  849. d__1 = xbnd, d__2 = f2cmin(1.,tjj) * grow;
  850. xbnd = f2cmin(d__1,d__2);
  851. if (tjj + cnorm[j] >= smlnum) {
  852. /* G(j) = G(j-1)*( 1 + CNORM(j) / abs(A(j,j)) ) */
  853. grow *= tjj / (tjj + cnorm[j]);
  854. } else {
  855. /* G(j) could overflow, set GROW to 0. */
  856. grow = 0.;
  857. }
  858. ip += jinc * jlen;
  859. --jlen;
  860. /* L30: */
  861. }
  862. grow = xbnd;
  863. } else {
  864. /* A is unit triangular. */
  865. /* Compute GROW = 1/G(j), where G(0) = f2cmax{x(i), i=1,...,n}. */
  866. /* Computing MIN */
  867. d__1 = 1., d__2 = 1. / f2cmax(xbnd,smlnum);
  868. grow = f2cmin(d__1,d__2);
  869. i__2 = jlast;
  870. i__1 = jinc;
  871. for (j = jfirst; i__1 < 0 ? j >= i__2 : j <= i__2; j += i__1) {
  872. /* Exit the loop if the growth factor is too small. */
  873. if (grow <= smlnum) {
  874. goto L50;
  875. }
  876. /* G(j) = G(j-1)*( 1 + CNORM(j) ) */
  877. grow *= 1. / (cnorm[j] + 1.);
  878. /* L40: */
  879. }
  880. }
  881. L50:
  882. ;
  883. } else {
  884. /* Compute the growth in A**T * x = b. */
  885. if (upper) {
  886. jfirst = 1;
  887. jlast = *n;
  888. jinc = 1;
  889. } else {
  890. jfirst = *n;
  891. jlast = 1;
  892. jinc = -1;
  893. }
  894. if (tscal != 1.) {
  895. grow = 0.;
  896. goto L80;
  897. }
  898. if (nounit) {
  899. /* A is non-unit triangular. */
  900. /* Compute GROW = 1/G(j) and XBND = 1/M(j). */
  901. /* Initially, M(0) = f2cmax{x(i), i=1,...,n}. */
  902. grow = 1. / f2cmax(xbnd,smlnum);
  903. xbnd = grow;
  904. ip = jfirst * (jfirst + 1) / 2;
  905. jlen = 1;
  906. i__1 = jlast;
  907. i__2 = jinc;
  908. for (j = jfirst; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
  909. /* Exit the loop if the growth factor is too small. */
  910. if (grow <= smlnum) {
  911. goto L80;
  912. }
  913. /* G(j) = f2cmax( G(j-1), M(j-1)*( 1 + CNORM(j) ) ) */
  914. xj = cnorm[j] + 1.;
  915. /* Computing MIN */
  916. d__1 = grow, d__2 = xbnd / xj;
  917. grow = f2cmin(d__1,d__2);
  918. /* M(j) = M(j-1)*( 1 + CNORM(j) ) / abs(A(j,j)) */
  919. tjj = (d__1 = ap[ip], abs(d__1));
  920. if (xj > tjj) {
  921. xbnd *= tjj / xj;
  922. }
  923. ++jlen;
  924. ip += jinc * jlen;
  925. /* L60: */
  926. }
  927. grow = f2cmin(grow,xbnd);
  928. } else {
  929. /* A is unit triangular. */
  930. /* Compute GROW = 1/G(j), where G(0) = f2cmax{x(i), i=1,...,n}. */
  931. /* Computing MIN */
  932. d__1 = 1., d__2 = 1. / f2cmax(xbnd,smlnum);
  933. grow = f2cmin(d__1,d__2);
  934. i__2 = jlast;
  935. i__1 = jinc;
  936. for (j = jfirst; i__1 < 0 ? j >= i__2 : j <= i__2; j += i__1) {
  937. /* Exit the loop if the growth factor is too small. */
  938. if (grow <= smlnum) {
  939. goto L80;
  940. }
  941. /* G(j) = ( 1 + CNORM(j) )*G(j-1) */
  942. xj = cnorm[j] + 1.;
  943. grow /= xj;
  944. /* L70: */
  945. }
  946. }
  947. L80:
  948. ;
  949. }
  950. if (grow * tscal > smlnum) {
  951. /* Use the Level 2 BLAS solve if the reciprocal of the bound on */
  952. /* elements of X is not too small. */
  953. dtpsv_(uplo, trans, diag, n, &ap[1], &x[1], &c__1);
  954. } else {
  955. /* Use a Level 1 BLAS solve, scaling intermediate results. */
  956. if (xmax > bignum) {
  957. /* Scale X so that its components are less than or equal to */
  958. /* BIGNUM in absolute value. */
  959. *scale = bignum / xmax;
  960. dscal_(n, scale, &x[1], &c__1);
  961. xmax = bignum;
  962. }
  963. if (notran) {
  964. /* Solve A * x = b */
  965. ip = jfirst * (jfirst + 1) / 2;
  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 = ap[ip] * 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, &ap[ip - j + 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. ip -= j;
  1052. } else {
  1053. if (j < *n) {
  1054. /* Compute the update */
  1055. /* x(j+1:n) := x(j+1:n) - x(j) * A(j+1:n,j) */
  1056. i__3 = *n - j;
  1057. d__1 = -x[j] * tscal;
  1058. daxpy_(&i__3, &d__1, &ap[ip + 1], &c__1, &x[j + 1], &
  1059. c__1);
  1060. i__3 = *n - j;
  1061. i__ = j + idamax_(&i__3, &x[j + 1], &c__1);
  1062. xmax = (d__1 = x[i__], abs(d__1));
  1063. }
  1064. ip = ip + *n - j + 1;
  1065. }
  1066. /* L110: */
  1067. }
  1068. } else {
  1069. /* Solve A**T * x = b */
  1070. ip = jfirst * (jfirst + 1) / 2;
  1071. jlen = 1;
  1072. i__2 = jlast;
  1073. i__1 = jinc;
  1074. for (j = jfirst; i__1 < 0 ? j >= i__2 : j <= i__2; j += i__1) {
  1075. /* Compute x(j) = b(j) - sum A(k,j)*x(k). */
  1076. /* k<>j */
  1077. xj = (d__1 = x[j], abs(d__1));
  1078. uscal = tscal;
  1079. rec = 1. / f2cmax(xmax,1.);
  1080. if (cnorm[j] > (bignum - xj) * rec) {
  1081. /* If x(j) could overflow, scale x by 1/(2*XMAX). */
  1082. rec *= .5;
  1083. if (nounit) {
  1084. tjjs = ap[ip] * tscal;
  1085. } else {
  1086. tjjs = tscal;
  1087. }
  1088. tjj = abs(tjjs);
  1089. if (tjj > 1.) {
  1090. /* Divide by A(j,j) when scaling x if A(j,j) > 1. */
  1091. /* Computing MIN */
  1092. d__1 = 1., d__2 = rec * tjj;
  1093. rec = f2cmin(d__1,d__2);
  1094. uscal /= tjjs;
  1095. }
  1096. if (rec < 1.) {
  1097. dscal_(n, &rec, &x[1], &c__1);
  1098. *scale *= rec;
  1099. xmax *= rec;
  1100. }
  1101. }
  1102. sumj = 0.;
  1103. if (uscal == 1.) {
  1104. /* If the scaling needed for A in the dot product is 1, */
  1105. /* call DDOT to perform the dot product. */
  1106. if (upper) {
  1107. i__3 = j - 1;
  1108. sumj = ddot_(&i__3, &ap[ip - j + 1], &c__1, &x[1], &
  1109. c__1);
  1110. } else if (j < *n) {
  1111. i__3 = *n - j;
  1112. sumj = ddot_(&i__3, &ap[ip + 1], &c__1, &x[j + 1], &
  1113. c__1);
  1114. }
  1115. } else {
  1116. /* Otherwise, use in-line code for the dot product. */
  1117. if (upper) {
  1118. i__3 = j - 1;
  1119. for (i__ = 1; i__ <= i__3; ++i__) {
  1120. sumj += ap[ip - j + i__] * uscal * x[i__];
  1121. /* L120: */
  1122. }
  1123. } else if (j < *n) {
  1124. i__3 = *n - j;
  1125. for (i__ = 1; i__ <= i__3; ++i__) {
  1126. sumj += ap[ip + i__] * uscal * x[j + i__];
  1127. /* L130: */
  1128. }
  1129. }
  1130. }
  1131. if (uscal == tscal) {
  1132. /* Compute x(j) := ( x(j) - sumj ) / A(j,j) if 1/A(j,j) */
  1133. /* was not used to scale the dotproduct. */
  1134. x[j] -= sumj;
  1135. xj = (d__1 = x[j], abs(d__1));
  1136. if (nounit) {
  1137. /* Compute x(j) = x(j) / A(j,j), scaling if necessary. */
  1138. tjjs = ap[ip] * tscal;
  1139. } else {
  1140. tjjs = tscal;
  1141. if (tscal == 1.) {
  1142. goto L150;
  1143. }
  1144. }
  1145. tjj = abs(tjjs);
  1146. if (tjj > smlnum) {
  1147. /* abs(A(j,j)) > SMLNUM: */
  1148. if (tjj < 1.) {
  1149. if (xj > tjj * bignum) {
  1150. /* Scale X by 1/abs(x(j)). */
  1151. rec = 1. / xj;
  1152. dscal_(n, &rec, &x[1], &c__1);
  1153. *scale *= rec;
  1154. xmax *= rec;
  1155. }
  1156. }
  1157. x[j] /= tjjs;
  1158. } else if (tjj > 0.) {
  1159. /* 0 < abs(A(j,j)) <= SMLNUM: */
  1160. if (xj > tjj * bignum) {
  1161. /* Scale x by (1/abs(x(j)))*abs(A(j,j))*BIGNUM. */
  1162. rec = tjj * bignum / xj;
  1163. dscal_(n, &rec, &x[1], &c__1);
  1164. *scale *= rec;
  1165. xmax *= rec;
  1166. }
  1167. x[j] /= tjjs;
  1168. } else {
  1169. /* A(j,j) = 0: Set x(1:n) = 0, x(j) = 1, and */
  1170. /* scale = 0, and compute a solution to A**T*x = 0. */
  1171. i__3 = *n;
  1172. for (i__ = 1; i__ <= i__3; ++i__) {
  1173. x[i__] = 0.;
  1174. /* L140: */
  1175. }
  1176. x[j] = 1.;
  1177. *scale = 0.;
  1178. xmax = 0.;
  1179. }
  1180. L150:
  1181. ;
  1182. } else {
  1183. /* Compute x(j) := x(j) / A(j,j) - sumj if the dot */
  1184. /* product has already been divided by 1/A(j,j). */
  1185. x[j] = x[j] / tjjs - sumj;
  1186. }
  1187. /* Computing MAX */
  1188. d__2 = xmax, d__3 = (d__1 = x[j], abs(d__1));
  1189. xmax = f2cmax(d__2,d__3);
  1190. ++jlen;
  1191. ip += jinc * jlen;
  1192. /* L160: */
  1193. }
  1194. }
  1195. *scale /= tscal;
  1196. }
  1197. /* Scale the column norms by 1/TSCAL for return. */
  1198. if (tscal != 1.) {
  1199. d__1 = 1. / tscal;
  1200. dscal_(n, &d__1, &cnorm[1], &c__1);
  1201. }
  1202. return;
  1203. /* End of DLATPS */
  1204. } /* dlatps_ */