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.

sgeqp3rk.c 23 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796
  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]/Cd(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 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++ = ' '; }
  217. #define s_cmp(a,b,c,d) ((integer)strncmp((a),(b),f2cmin((c),(d))))
  218. #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]; }
  219. #define sig_die(s, kill) { exit(1); }
  220. #define s_stop(s, n) {exit(0);}
  221. #define z_abs(z) (cabs(Cd(z)))
  222. #define z_exp(R, Z) {pCd(R) = cexp(Cd(Z));}
  223. #define z_sqrt(R, Z) {pCd(R) = csqrt(Cd(Z));}
  224. #define myexit_() break;
  225. #define mycycle_() continue;
  226. #define myceiling_(w) {ceil(w)}
  227. #define myhuge_(w) {HUGE_VAL}
  228. //#define mymaxloc_(w,s,e,n) {if (sizeof(*(w)) == sizeof(double)) dmaxloc_((w),*(s),*(e),n); else dmaxloc_((w),*(s),*(e),n);}
  229. #define mymaxloc_(w,s,e,n) dmaxloc_(w,*(s),*(e),n)
  230. /* -- translated by f2c (version 20000121).
  231. You must link the resulting object file with the libraries:
  232. -lf2c -lm (in that order)
  233. */
  234. /* Table of constant values */
  235. static integer c__1 = 1;
  236. static integer c_n1 = -1;
  237. static integer c__3 = 3;
  238. static integer c__2 = 2;
  239. /* Subroutine */ int sgeqp3rk_(integer *m, integer *n, integer *nrhs, integer
  240. *kmax, real *abstol, real *reltol, real *a, integer *lda, integer *k,
  241. real *maxc2nrmk, real *relmaxc2nrmk, integer *jpiv, real *tau, real *
  242. work, integer *lwork, integer *iwork, integer *info)
  243. {
  244. /* System generated locals */
  245. integer a_dim1, a_offset, i__1, i__2;
  246. real r__1, r__2;
  247. /* Local variables */
  248. real maxc2nrm;
  249. extern /* Subroutine */ int slaqp2rk_(integer *, integer *, integer *,
  250. integer *, integer *, real *, real *, integer *, real *, real *,
  251. integer *, integer *, real *, real *, integer *, real *, real *,
  252. real *, real *, integer *), slaqp3rk_(integer *, integer *,
  253. integer *, integer *, integer *, real *, real *, integer *, real *
  254. , real *, integer *, logical *, integer *, real *, real *,
  255. integer *, real *, real *, real *, real *, real *, integer *,
  256. integer *, integer *);
  257. logical done;
  258. integer jmax;
  259. extern real snrm2_(integer *, real *, integer *);
  260. integer j, jmaxc2nrm, jmaxb, nbmin, iinfo, n_sub__, minmn;
  261. real myhugeval;
  262. integer jb, nb, kf, nx;
  263. extern real slamch_(char *);
  264. real safmin;
  265. extern /* Subroutine */ int xerbla_(char *, integer *);
  266. extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
  267. integer *, integer *, ftnlen, ftnlen), isamax_(integer *, real *,
  268. integer *);
  269. extern logical sisnan_(real *);
  270. integer kp1, lwkopt;
  271. logical lquery;
  272. integer jbf;
  273. real eps;
  274. integer iws, ioffset;
  275. /* -- LAPACK computational routine -- */
  276. /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
  277. /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
  278. /* ===================================================================== */
  279. /* Test input arguments */
  280. /* ==================== */
  281. /* Parameter adjustments */
  282. a_dim1 = *lda;
  283. a_offset = 1 + a_dim1 * 1;
  284. a -= a_offset;
  285. --jpiv;
  286. --tau;
  287. --work;
  288. --iwork;
  289. /* Function Body */
  290. *info = 0;
  291. lquery = *lwork == -1;
  292. if (*m < 0) {
  293. *info = -1;
  294. } else if (*n < 0) {
  295. *info = -2;
  296. } else if (*nrhs < 0) {
  297. *info = -3;
  298. } else if (*kmax < 0) {
  299. *info = -4;
  300. } else if (sisnan_(abstol)) {
  301. *info = -5;
  302. } else if (sisnan_(reltol)) {
  303. *info = -6;
  304. } else if (*lda < f2cmax(1,*m)) {
  305. *info = -8;
  306. }
  307. /* If the input parameters M, N, NRHS, KMAX, LDA are valid: */
  308. /* a) Test the input workspace size LWORK for the minimum */
  309. /* size requirement IWS. */
  310. /* b) Determine the optimal block size NB and optimal */
  311. /* workspace size LWKOPT to be returned in WORK(1) */
  312. /* in case of (1) LWORK < IWS, (2) LQUERY = .TRUE., */
  313. /* (3) when routine exits. */
  314. /* Here, IWS is the miminum workspace required for unblocked */
  315. /* code. */
  316. if (*info == 0) {
  317. minmn = f2cmin(*m,*n);
  318. if (minmn == 0) {
  319. iws = 1;
  320. lwkopt = 1;
  321. } else {
  322. /* Minimal workspace size in case of using only unblocked */
  323. /* BLAS 2 code in SLAQP2RK. */
  324. /* 1) SGEQP3RK and SLAQP2RK: 2*N to store full and partial */
  325. /* column 2-norms. */
  326. /* 2) SLAQP2RK: N+NRHS-1 to use in WORK array that is used */
  327. /* in SLARF subroutine inside SLAQP2RK to apply an */
  328. /* elementary reflector from the left. */
  329. /* TOTAL_WORK_SIZE = 3*N + NRHS - 1 */
  330. iws = *n * 3 + *nrhs - 1;
  331. /* Assign to NB optimal block size. */
  332. nb = ilaenv_(&c__1, "SGEQP3RK", " ", m, n, &c_n1, &c_n1, (ftnlen)
  333. 8, (ftnlen)1);
  334. /* A formula for the optimal workspace size in case of using */
  335. /* both unblocked BLAS 2 in SLAQP2RK and blocked BLAS 3 code */
  336. /* in SLAQP3RK. */
  337. /* 1) SGEQP3RK, SLAQP2RK, SLAQP3RK: 2*N to store full and */
  338. /* partial column 2-norms. */
  339. /* 2) SLAQP2RK: N+NRHS-1 to use in WORK array that is used */
  340. /* in SLARF subroutine to apply an elementary reflector */
  341. /* from the left. */
  342. /* 3) SLAQP3RK: NB*(N+NRHS) to use in the work array F that */
  343. /* is used to apply a block reflector from */
  344. /* the left. */
  345. /* 4) SLAQP3RK: NB to use in the auxilixary array AUX. */
  346. /* Sizes (2) and ((3) + (4)) should intersect, therefore */
  347. /* TOTAL_WORK_SIZE = 2*N + NB*( N+NRHS+1 ), given NBMIN=2. */
  348. lwkopt = (*n << 1) + nb * (*n + *nrhs + 1);
  349. }
  350. work[1] = (real) lwkopt;
  351. if (*lwork < iws && ! lquery) {
  352. *info = -15;
  353. }
  354. }
  355. /* NOTE: The optimal workspace size is returned in WORK(1), if */
  356. /* the input parameters M, N, NRHS, KMAX, LDA are valid. */
  357. if (*info != 0) {
  358. i__1 = -(*info);
  359. xerbla_("SGEQP3RK", &i__1);
  360. return 0;
  361. } else if (lquery) {
  362. return 0;
  363. }
  364. /* Quick return if possible for M=0 or N=0. */
  365. if (minmn == 0) {
  366. *k = 0;
  367. *maxc2nrmk = 0.f;
  368. *relmaxc2nrmk = 0.f;
  369. work[1] = (real) lwkopt;
  370. return 0;
  371. }
  372. /* ================================================================== */
  373. /* Initialize column pivot array JPIV. */
  374. i__1 = *n;
  375. for (j = 1; j <= i__1; ++j) {
  376. jpiv[j] = j;
  377. }
  378. /* ================================================================== */
  379. /* Initialize storage for partial and exact column 2-norms. */
  380. /* a) The elements WORK(1:N) are used to store partial column */
  381. /* 2-norms of the matrix A, and may decrease in each computation */
  382. /* step; initialize to the values of complete columns 2-norms. */
  383. /* b) The elements WORK(N+1:2*N) are used to store complete column */
  384. /* 2-norms of the matrix A, they are not changed during the */
  385. /* computation; initialize the values of complete columns 2-norms. */
  386. i__1 = *n;
  387. for (j = 1; j <= i__1; ++j) {
  388. work[j] = snrm2_(m, &a[j * a_dim1 + 1], &c__1);
  389. work[*n + j] = work[j];
  390. }
  391. /* ================================================================== */
  392. /* Compute the pivot column index and the maximum column 2-norm */
  393. /* for the whole original matrix stored in A(1:M,1:N). */
  394. kp1 = isamax_(n, &work[1], &c__1);
  395. maxc2nrm = work[kp1];
  396. /* ==================================================================. */
  397. if (sisnan_(&maxc2nrm)) {
  398. /* Check if the matrix A contains NaN, set INFO parameter */
  399. /* to the column number where the first NaN is found and return */
  400. /* from the routine. */
  401. *k = 0;
  402. *info = kp1;
  403. /* Set MAXC2NRMK and RELMAXC2NRMK to NaN. */
  404. *maxc2nrmk = maxc2nrm;
  405. *relmaxc2nrmk = maxc2nrm;
  406. /* Array TAU is not set and contains undefined elements. */
  407. work[1] = (real) lwkopt;
  408. return 0;
  409. }
  410. /* =================================================================== */
  411. if (maxc2nrm == 0.f) {
  412. /* Check is the matrix A is a zero matrix, set array TAU and */
  413. /* return from the routine. */
  414. *k = 0;
  415. *maxc2nrmk = 0.f;
  416. *relmaxc2nrmk = 0.f;
  417. i__1 = minmn;
  418. for (j = 1; j <= i__1; ++j) {
  419. tau[j] = 0.f;
  420. }
  421. work[1] = (real) lwkopt;
  422. return 0;
  423. }
  424. /* =================================================================== */
  425. myhugeval = slamch_("Overflow");
  426. if (maxc2nrm > myhugeval) {
  427. /* Check if the matrix A contains +Inf or -Inf, set INFO parameter */
  428. /* to the column number, where the first +/-Inf is found plus N, */
  429. /* and continue the computation. */
  430. *info = *n + kp1;
  431. }
  432. /* ================================================================== */
  433. /* Quick return if possible for the case when the first */
  434. /* stopping criterion is satisfied, i.e. KMAX = 0. */
  435. if (*kmax == 0) {
  436. *k = 0;
  437. *maxc2nrmk = maxc2nrm;
  438. *relmaxc2nrmk = 1.f;
  439. i__1 = minmn;
  440. for (j = 1; j <= i__1; ++j) {
  441. tau[j] = 0.f;
  442. }
  443. work[1] = (real) lwkopt;
  444. return 0;
  445. }
  446. /* ================================================================== */
  447. eps = slamch_("Epsilon");
  448. /* Adjust ABSTOL */
  449. if (*abstol >= 0.f) {
  450. safmin = slamch_("Safe minimum");
  451. /* Computing MAX */
  452. r__1 = *abstol, r__2 = safmin * 2.f;
  453. *abstol = f2cmax(r__1,r__2);
  454. }
  455. /* Adjust RELTOL */
  456. if (*reltol >= 0.f) {
  457. *reltol = f2cmax(*reltol,eps);
  458. }
  459. /* =================================================================== */
  460. /* JMAX is the maximum index of the column to be factorized, */
  461. /* which is also limited by the first stopping criterion KMAX. */
  462. jmax = f2cmin(*kmax,minmn);
  463. /* =================================================================== */
  464. /* Quick return if possible for the case when the second or third */
  465. /* stopping criterion for the whole original matrix is satified, */
  466. /* i.e. MAXC2NRM <= ABSTOL or RELMAXC2NRM <= RELTOL */
  467. /* (which is ONE <= RELTOL). */
  468. if (maxc2nrm <= *abstol || 1.f <= *reltol) {
  469. *k = 0;
  470. *maxc2nrmk = maxc2nrm;
  471. *relmaxc2nrmk = 1.f;
  472. i__1 = minmn;
  473. for (j = 1; j <= i__1; ++j) {
  474. tau[j] = 0.f;
  475. }
  476. work[1] = (real) lwkopt;
  477. return 0;
  478. }
  479. /* ================================================================== */
  480. /* Factorize columns */
  481. /* ================================================================== */
  482. /* Determine the block size. */
  483. nbmin = 2;
  484. nx = 0;
  485. if (nb > 1 && nb < minmn) {
  486. /* Determine when to cross over from blocked to unblocked code. */
  487. /* (for N less than NX, unblocked code should be used). */
  488. /* Computing MAX */
  489. i__1 = 0, i__2 = ilaenv_(&c__3, "SGEQP3RK", " ", m, n, &c_n1, &c_n1, (
  490. ftnlen)8, (ftnlen)1);
  491. nx = f2cmax(i__1,i__2);
  492. if (nx < minmn) {
  493. /* Determine if workspace is large enough for blocked code. */
  494. if (*lwork < lwkopt) {
  495. /* Not enough workspace to use optimal block size that */
  496. /* is currently stored in NB. */
  497. /* Reduce NB and determine the minimum value of NB. */
  498. nb = (*lwork - (*n << 1)) / (*n + 1);
  499. /* Computing MAX */
  500. i__1 = 2, i__2 = ilaenv_(&c__2, "SGEQP3RK", " ", m, n, &c_n1,
  501. &c_n1, (ftnlen)8, (ftnlen)1);
  502. nbmin = f2cmax(i__1,i__2);
  503. }
  504. }
  505. }
  506. /* ================================================================== */
  507. /* DONE is the boolean flag to rerpresent the case when the */
  508. /* factorization completed in the block factorization routine, */
  509. /* before the end of the block. */
  510. done = FALSE_;
  511. /* J is the column index. */
  512. j = 1;
  513. /* (1) Use blocked code initially. */
  514. /* JMAXB is the maximum column index of the block, when the */
  515. /* blocked code is used, is also limited by the first stopping */
  516. /* criterion KMAX. */
  517. /* Computing MIN */
  518. i__1 = *kmax, i__2 = minmn - nx;
  519. jmaxb = f2cmin(i__1,i__2);
  520. if (nb >= nbmin && nb < jmax && jmaxb > 0) {
  521. /* Loop over the column blocks of the matrix A(1:M,1:JMAXB). Here: */
  522. /* J is the column index of a column block; */
  523. /* JB is the column block size to pass to block factorization */
  524. /* routine in a loop step; */
  525. /* JBF is the number of columns that were actually factorized */
  526. /* that was returned by the block factorization routine */
  527. /* in a loop step, JBF <= JB; */
  528. /* N_SUB is the number of columns in the submatrix; */
  529. /* IOFFSET is the number of rows that should not be factorized. */
  530. while(j <= jmaxb) {
  531. /* Computing MIN */
  532. i__1 = nb, i__2 = jmaxb - j + 1;
  533. jb = f2cmin(i__1,i__2);
  534. n_sub__ = *n - j + 1;
  535. ioffset = j - 1;
  536. /* Factorize JB columns among the columns A(J:N). */
  537. i__1 = *n + *nrhs - j + 1;
  538. slaqp3rk_(m, &n_sub__, nrhs, &ioffset, &jb, abstol, reltol, &kp1,
  539. &maxc2nrm, &a[j * a_dim1 + 1], lda, &done, &jbf,
  540. maxc2nrmk, relmaxc2nrmk, &jpiv[j], &tau[j], &work[j], &
  541. work[*n + j], &work[(*n << 1) + 1], &work[(*n << 1) + jb
  542. + 1], &i__1, &iwork[1], &iinfo);
  543. /* Set INFO on the first occurence of Inf. */
  544. if (iinfo > n_sub__ && *info == 0) {
  545. *info = (ioffset << 1) + iinfo;
  546. }
  547. if (done) {
  548. /* Either the submatrix is zero before the end of the */
  549. /* column block, or ABSTOL or RELTOL criterion is */
  550. /* satisfied before the end of the column block, we can */
  551. /* return from the routine. Perform the following before */
  552. /* returning: */
  553. /* a) Set the number of factorized columns K, */
  554. /* K = IOFFSET + JBF from the last call of blocked */
  555. /* routine. */
  556. /* NOTE: 1) MAXC2NRMK and RELMAXC2NRMK are returned */
  557. /* by the block factorization routine; */
  558. /* 2) The remaining TAUs are set to ZERO by the */
  559. /* block factorization routine. */
  560. *k = ioffset + jbf;
  561. /* Set INFO on the first occurrence of NaN, NaN takes */
  562. /* prcedence over Inf. */
  563. if (iinfo <= n_sub__ && iinfo > 0) {
  564. *info = ioffset + iinfo;
  565. }
  566. /* Return from the routine. */
  567. work[1] = (real) lwkopt;
  568. return 0;
  569. }
  570. j += jbf;
  571. }
  572. }
  573. /* Use unblocked code to factor the last or only block. */
  574. /* J = JMAX+1 means we factorized the maximum possible number of */
  575. /* columns, that is in ELSE clause we need to compute */
  576. /* the MAXC2NORM and RELMAXC2NORM to return after we processed */
  577. /* the blocks. */
  578. if (j <= jmax) {
  579. /* N_SUB is the number of columns in the submatrix; */
  580. /* IOFFSET is the number of rows that should not be factorized. */
  581. n_sub__ = *n - j + 1;
  582. ioffset = j - 1;
  583. i__1 = jmax - j + 1;
  584. slaqp2rk_(m, &n_sub__, nrhs, &ioffset, &i__1, abstol, reltol, &kp1, &
  585. maxc2nrm, &a[j * a_dim1 + 1], lda, &kf, maxc2nrmk,
  586. relmaxc2nrmk, &jpiv[j], &tau[j], &work[j], &work[*n + j], &
  587. work[(*n << 1) + 1], &iinfo);
  588. /* ABSTOL or RELTOL criterion is satisfied when the number of */
  589. /* the factorized columns KF is smaller then the number */
  590. /* of columns JMAX-J+1 supplied to be factorized by the */
  591. /* unblocked routine, we can return from */
  592. /* the routine. Perform the following before returning: */
  593. /* a) Set the number of factorized columns K, */
  594. /* b) MAXC2NRMK and RELMAXC2NRMK are returned by the */
  595. /* unblocked factorization routine above. */
  596. *k = j - 1 + kf;
  597. /* Set INFO on the first exception occurence. */
  598. /* Set INFO on the first exception occurence of Inf or NaN, */
  599. /* (NaN takes precedence over Inf). */
  600. if (iinfo > n_sub__ && *info == 0) {
  601. *info = (ioffset << 1) + iinfo;
  602. } else if (iinfo <= n_sub__ && iinfo > 0) {
  603. *info = ioffset + iinfo;
  604. }
  605. } else {
  606. /* Compute the return values for blocked code. */
  607. /* Set the number of factorized columns if the unblocked routine */
  608. /* was not called. */
  609. *k = jmax;
  610. /* If there exits a residual matrix after the blocked code: */
  611. /* 1) compute the values of MAXC2NRMK, RELMAXC2NRMK of the */
  612. /* residual matrix, otherwise set them to ZERO; */
  613. /* 2) Set TAU(K+1:MINMN) to ZERO. */
  614. if (*k < minmn) {
  615. i__1 = *n - *k;
  616. jmaxc2nrm = *k + isamax_(&i__1, &work[*k + 1], &c__1);
  617. *maxc2nrmk = work[jmaxc2nrm];
  618. if (*k == 0) {
  619. *relmaxc2nrmk = 1.f;
  620. } else {
  621. *relmaxc2nrmk = *maxc2nrmk / maxc2nrm;
  622. }
  623. i__1 = minmn;
  624. for (j = *k + 1; j <= i__1; ++j) {
  625. tau[j] = 0.f;
  626. }
  627. }
  628. /* END IF( J.LE.JMAX ) THEN */
  629. }
  630. work[1] = (real) lwkopt;
  631. return 0;
  632. /* End of SGEQP3RK */
  633. } /* sgeqp3rk_ */