14 typedef long long BLASLONG;
15 typedef unsigned long long BLASULONG;
17 typedef long BLASLONG;
18 typedef unsigned long BLASULONG;
22 typedef BLASLONG blasint;
24 #define blasabs(x) llabs(x)
26 #define blasabs(x) labs(x)
30 #define blasabs(x) abs(x)
33 typedef blasint integer;
35 typedef unsigned int uinteger;
36 typedef char *address;
37 typedef short int shortint;
39 typedef double doublereal;
40 typedef struct { real r, i; } complex;
41 typedef struct { doublereal r, i; } doublecomplex;
43 static inline _Fcomplex Cf(complex *z) {_Fcomplex zz={z->r , z->i}; return zz;}
44 static inline _Dcomplex Cd(doublecomplex *z) {_Dcomplex zz={z->r , z->i};return zz;}
45 static inline _Fcomplex * _pCf(complex *z) {return (_Fcomplex*)z;}
46 static inline _Dcomplex * _pCd(doublecomplex *z) {return (_Dcomplex*)z;}
48 static inline _Complex float Cf(complex *z) {return z->r + z->i*_Complex_I;}
49 static inline _Complex double Cd(doublecomplex *z) {return z->r + z->i*_Complex_I;}
50 static inline _Complex float * _pCf(complex *z) {return (_Complex float*)z;}
51 static inline _Complex double * _pCd(doublecomplex *z) {return (_Complex double*)z;}
53 #define pCf(z) (*_pCf(z))
54 #define pCd(z) (*_pCd(z))
56 typedef short int shortlogical;
57 typedef char logical1;
58 typedef char integer1;
63 /* Extern is for use with -E */
74 /*external read, write*/
83 /*internal read, write*/
113 /*rewind, backspace, endfile*/
125 ftnint *inex; /*parameters in standard's order*/
151 union Multitype { /* for multiple entry points */
162 typedef union Multitype Multitype;
164 struct Vardesc { /* for Namelist */
170 typedef struct Vardesc Vardesc;
177 typedef struct Namelist Namelist;
179 #define abs(x) ((x) >= 0 ? (x) : -(x))
180 #define dabs(x) (fabs(x))
181 #define f2cmin(a,b) ((a) <= (b) ? (a) : (b))
182 #define f2cmax(a,b) ((a) >= (b) ? (a) : (b))
183 #define dmin(a,b) (f2cmin(a,b))
184 #define dmax(a,b) (f2cmax(a,b))
185 #define bit_test(a,b) ((a) >> (b) & 1)
186 #define bit_clear(a,b) ((a) & ~((uinteger)1 << (b)))
187 #define bit_set(a,b) ((a) | ((uinteger)1 << (b)))
189 #define abort_() { sig_die("Fortran abort routine called", 1); }
190 #define c_abs(z) (cabsf(Cf(z)))
191 #define c_cos(R,Z) { pCf(R)=ccos(Cf(Z)); }
193 #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]);}
194 #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]);}
196 #define c_div(c, a, b) {pCf(c) = Cf(a)/Cf(b);}
197 #define z_div(c, a, b) {pCd(c) = Cd(a)/Cd(b);}
199 #define c_exp(R, Z) {pCf(R) = cexpf(Cf(Z));}
200 #define c_log(R, Z) {pCf(R) = clogf(Cf(Z));}
201 #define c_sin(R, Z) {pCf(R) = csinf(Cf(Z));}
202 //#define c_sqrt(R, Z) {*(R) = csqrtf(Cf(Z));}
203 #define c_sqrt(R, Z) {pCf(R) = csqrtf(Cf(Z));}
204 #define d_abs(x) (fabs(*(x)))
205 #define d_acos(x) (acos(*(x)))
206 #define d_asin(x) (asin(*(x)))
207 #define d_atan(x) (atan(*(x)))
208 #define d_atn2(x, y) (atan2(*(x),*(y)))
209 #define d_cnjg(R, Z) { pCd(R) = conj(Cd(Z)); }
210 #define r_cnjg(R, Z) { pCf(R) = conjf(Cf(Z)); }
211 #define d_cos(x) (cos(*(x)))
212 #define d_cosh(x) (cosh(*(x)))
213 #define d_dim(__a, __b) ( *(__a) > *(__b) ? *(__a) - *(__b) : 0.0 )
214 #define d_exp(x) (exp(*(x)))
215 #define d_imag(z) (cimag(Cd(z)))
216 #define r_imag(z) (cimagf(Cf(z)))
217 #define d_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
218 #define r_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
219 #define d_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
220 #define r_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
221 #define d_log(x) (log(*(x)))
222 #define d_mod(x, y) (fmod(*(x), *(y)))
223 #define u_nint(__x) ((__x)>=0 ? floor((__x) + .5) : -floor(.5 - (__x)))
224 #define d_nint(x) u_nint(*(x))
225 #define u_sign(__a,__b) ((__b) >= 0 ? ((__a) >= 0 ? (__a) : -(__a)) : -((__a) >= 0 ? (__a) : -(__a)))
226 #define d_sign(a,b) u_sign(*(a),*(b))
227 #define r_sign(a,b) u_sign(*(a),*(b))
228 #define d_sin(x) (sin(*(x)))
229 #define d_sinh(x) (sinh(*(x)))
230 #define d_sqrt(x) (sqrt(*(x)))
231 #define d_tan(x) (tan(*(x)))
232 #define d_tanh(x) (tanh(*(x)))
233 #define i_abs(x) abs(*(x))
234 #define i_dnnt(x) ((integer)u_nint(*(x)))
235 #define i_len(s, n) (n)
236 #define i_nint(x) ((integer)u_nint(*(x)))
237 #define i_sign(a,b) ((integer)u_sign((integer)*(a),(integer)*(b)))
238 #define pow_dd(ap, bp) ( pow(*(ap), *(bp)))
239 #define pow_si(B,E) spow_ui(*(B),*(E))
240 #define pow_ri(B,E) spow_ui(*(B),*(E))
241 #define pow_di(B,E) dpow_ui(*(B),*(E))
242 #define pow_zi(p, a, b) {pCd(p) = zpow_ui(Cd(a), *(b));}
243 #define pow_ci(p, a, b) {pCf(p) = cpow_ui(Cf(a), *(b));}
244 #define pow_zz(R,A,B) {pCd(R) = cpow(Cd(A),*(B));}
245 #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++ = ' '; }
246 #define s_cmp(a,b,c,d) ((integer)strncmp((a),(b),f2cmin((c),(d))))
247 #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]; }
248 #define sig_die(s, kill) { exit(1); }
249 #define s_stop(s, n) {exit(0);}
250 static char junk[] = "\n@(#)LIBF77 VERSION 19990503\n";
251 #define z_abs(z) (cabs(Cd(z)))
252 #define z_exp(R, Z) {pCd(R) = cexp(Cd(Z));}
253 #define z_sqrt(R, Z) {pCd(R) = csqrt(Cd(Z));}
254 #define myexit_() break;
255 #define mycycle() continue;
256 #define myceiling(w) {ceil(w)}
257 #define myhuge(w) {HUGE_VAL}
258 //#define mymaxloc_(w,s,e,n) {if (sizeof(*(w)) == sizeof(double)) dmaxloc_((w),*(s),*(e),n); else dmaxloc_((w),*(s),*(e),n);}
259 #define mymaxloc(w,s,e,n) {dmaxloc_(w,*(s),*(e),n)}
261 /* procedure parameter types for -A and -C++ */
263 #define F2C_proc_par_types 1
265 typedef logical (*L_fp)(...);
267 typedef logical (*L_fp)();
270 static float spow_ui(float x, integer n) {
271 float pow=1.0; unsigned long int u;
273 if(n < 0) n = -n, x = 1/x;
282 static double dpow_ui(double x, integer n) {
283 double pow=1.0; unsigned long int u;
285 if(n < 0) n = -n, x = 1/x;
295 static _Fcomplex cpow_ui(complex x, integer n) {
296 complex pow={1.0,0.0}; unsigned long int u;
298 if(n < 0) n = -n, x.r = 1/x.r, x.i=1/x.i;
300 if(u & 01) pow.r *= x.r, pow.i *= x.i;
301 if(u >>= 1) x.r *= x.r, x.i *= x.i;
305 _Fcomplex p={pow.r, pow.i};
309 static _Complex float cpow_ui(_Complex float x, integer n) {
310 _Complex float pow=1.0; unsigned long int u;
312 if(n < 0) n = -n, x = 1/x;
323 static _Dcomplex zpow_ui(_Dcomplex x, integer n) {
324 _Dcomplex pow={1.0,0.0}; unsigned long int u;
326 if(n < 0) n = -n, x._Val[0] = 1/x._Val[0], x._Val[1] =1/x._Val[1];
328 if(u & 01) pow._Val[0] *= x._Val[0], pow._Val[1] *= x._Val[1];
329 if(u >>= 1) x._Val[0] *= x._Val[0], x._Val[1] *= x._Val[1];
333 _Dcomplex p = {pow._Val[0], pow._Val[1]};
337 static _Complex double zpow_ui(_Complex double x, integer n) {
338 _Complex double pow=1.0; unsigned long int u;
340 if(n < 0) n = -n, x = 1/x;
350 static integer pow_ii(integer x, integer n) {
351 integer pow; unsigned long int u;
353 if (n == 0 || x == 1) pow = 1;
354 else if (x != -1) pow = x == 0 ? 1/x : 0;
357 if ((n > 0) || !(n == 0 || x == 1 || x != -1)) {
367 static integer dmaxloc_(double *w, integer s, integer e, integer *n)
369 double m; integer i, mi;
370 for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
371 if (w[i-1]>m) mi=i ,m=w[i-1];
374 static integer smaxloc_(float *w, integer s, integer e, integer *n)
376 float m; integer i, mi;
377 for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
378 if (w[i-1]>m) mi=i ,m=w[i-1];
381 static inline void cdotc_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
382 integer n = *n_, incx = *incx_, incy = *incy_, i;
384 _Fcomplex zdotc = {0.0, 0.0};
385 if (incx == 1 && incy == 1) {
386 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
387 zdotc._Val[0] += conjf(Cf(&x[i]))._Val[0] * Cf(&y[i])._Val[0];
388 zdotc._Val[1] += conjf(Cf(&x[i]))._Val[1] * Cf(&y[i])._Val[1];
391 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
392 zdotc._Val[0] += conjf(Cf(&x[i*incx]))._Val[0] * Cf(&y[i*incy])._Val[0];
393 zdotc._Val[1] += conjf(Cf(&x[i*incx]))._Val[1] * Cf(&y[i*incy])._Val[1];
399 _Complex float zdotc = 0.0;
400 if (incx == 1 && incy == 1) {
401 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
402 zdotc += conjf(Cf(&x[i])) * Cf(&y[i]);
405 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
406 zdotc += conjf(Cf(&x[i*incx])) * Cf(&y[i*incy]);
412 static inline void zdotc_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
413 integer n = *n_, incx = *incx_, incy = *incy_, i;
415 _Dcomplex zdotc = {0.0, 0.0};
416 if (incx == 1 && incy == 1) {
417 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
418 zdotc._Val[0] += conj(Cd(&x[i]))._Val[0] * Cd(&y[i])._Val[0];
419 zdotc._Val[1] += conj(Cd(&x[i]))._Val[1] * Cd(&y[i])._Val[1];
422 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
423 zdotc._Val[0] += conj(Cd(&x[i*incx]))._Val[0] * Cd(&y[i*incy])._Val[0];
424 zdotc._Val[1] += conj(Cd(&x[i*incx]))._Val[1] * Cd(&y[i*incy])._Val[1];
430 _Complex double zdotc = 0.0;
431 if (incx == 1 && incy == 1) {
432 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
433 zdotc += conj(Cd(&x[i])) * Cd(&y[i]);
436 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
437 zdotc += conj(Cd(&x[i*incx])) * Cd(&y[i*incy]);
443 static inline void cdotu_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
444 integer n = *n_, incx = *incx_, incy = *incy_, i;
446 _Fcomplex zdotc = {0.0, 0.0};
447 if (incx == 1 && incy == 1) {
448 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
449 zdotc._Val[0] += Cf(&x[i])._Val[0] * Cf(&y[i])._Val[0];
450 zdotc._Val[1] += Cf(&x[i])._Val[1] * Cf(&y[i])._Val[1];
453 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
454 zdotc._Val[0] += Cf(&x[i*incx])._Val[0] * Cf(&y[i*incy])._Val[0];
455 zdotc._Val[1] += Cf(&x[i*incx])._Val[1] * Cf(&y[i*incy])._Val[1];
461 _Complex float zdotc = 0.0;
462 if (incx == 1 && incy == 1) {
463 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
464 zdotc += Cf(&x[i]) * Cf(&y[i]);
467 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
468 zdotc += Cf(&x[i*incx]) * Cf(&y[i*incy]);
474 static inline void zdotu_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
475 integer n = *n_, incx = *incx_, incy = *incy_, i;
477 _Dcomplex zdotc = {0.0, 0.0};
478 if (incx == 1 && incy == 1) {
479 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
480 zdotc._Val[0] += Cd(&x[i])._Val[0] * Cd(&y[i])._Val[0];
481 zdotc._Val[1] += Cd(&x[i])._Val[1] * Cd(&y[i])._Val[1];
484 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
485 zdotc._Val[0] += Cd(&x[i*incx])._Val[0] * Cd(&y[i*incy])._Val[0];
486 zdotc._Val[1] += Cd(&x[i*incx])._Val[1] * Cd(&y[i*incy])._Val[1];
492 _Complex double zdotc = 0.0;
493 if (incx == 1 && incy == 1) {
494 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
495 zdotc += Cd(&x[i]) * Cd(&y[i]);
498 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
499 zdotc += Cd(&x[i*incx]) * Cd(&y[i*incy]);
505 /* -- translated by f2c (version 20000121).
506 You must link the resulting object file with the libraries:
507 -lf2c -lm (in that order)
513 /* Table of constant values */
515 static integer c__1 = 1;
516 static real c_b9 = -1.f;
517 static real c_b10 = 1.f;
519 /* > \brief \b SLASYF_ROOK computes a partial factorization of a real symmetric matrix using the bounded Bunch
520 -Kaufman ("rook") diagonal pivoting method. */
522 /* =========== DOCUMENTATION =========== */
524 /* Online html documentation available at */
525 /* http://www.netlib.org/lapack/explore-html/ */
528 /* > Download SLASYF_ROOK + dependencies */
529 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/slasyf_
532 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/slasyf_
535 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/slasyf_
543 /* SUBROUTINE SLASYF_ROOK( UPLO, N, NB, KB, A, LDA, IPIV, W, LDW, INFO ) */
546 /* INTEGER INFO, KB, LDA, LDW, N, NB */
547 /* INTEGER IPIV( * ) */
548 /* REAL A( LDA, * ), W( LDW, * ) */
551 /* > \par Purpose: */
556 /* > SLASYF_ROOK computes a partial factorization of a real symmetric */
557 /* > matrix A using the bounded Bunch-Kaufman ("rook") diagonal */
558 /* > pivoting method. The partial factorization has the form: */
560 /* > A = ( I U12 ) ( A11 0 ) ( I 0 ) if UPLO = 'U', or: */
561 /* > ( 0 U22 ) ( 0 D ) ( U12**T U22**T ) */
563 /* > A = ( L11 0 ) ( D 0 ) ( L11**T L21**T ) if UPLO = 'L' */
564 /* > ( L21 I ) ( 0 A22 ) ( 0 I ) */
566 /* > where the order of D is at most NB. The actual order is returned in */
567 /* > the argument KB, and is either NB or NB-1, or N if N <= NB. */
569 /* > SLASYF_ROOK is an auxiliary routine called by SSYTRF_ROOK. It uses */
570 /* > blocked code (calling Level 3 BLAS) to update the submatrix */
571 /* > A11 (if UPLO = 'U') or A22 (if UPLO = 'L'). */
577 /* > \param[in] UPLO */
579 /* > UPLO is CHARACTER*1 */
580 /* > Specifies whether the upper or lower triangular part of the */
581 /* > symmetric matrix A is stored: */
582 /* > = 'U': Upper triangular */
583 /* > = 'L': Lower triangular */
589 /* > The order of the matrix A. N >= 0. */
592 /* > \param[in] NB */
594 /* > NB is INTEGER */
595 /* > The maximum number of columns of the matrix A that should be */
596 /* > factored. NB should be at least 2 to allow for 2-by-2 pivot */
600 /* > \param[out] KB */
602 /* > KB is INTEGER */
603 /* > The number of columns of A that were actually factored. */
604 /* > KB is either NB-1 or NB, or N if N <= NB. */
607 /* > \param[in,out] A */
609 /* > A is REAL array, dimension (LDA,N) */
610 /* > On entry, the symmetric matrix A. If UPLO = 'U', the leading */
611 /* > n-by-n upper triangular part of A contains the upper */
612 /* > triangular part of the matrix A, and the strictly lower */
613 /* > triangular part of A is not referenced. If UPLO = 'L', the */
614 /* > leading n-by-n lower triangular part of A contains the lower */
615 /* > triangular part of the matrix A, and the strictly upper */
616 /* > triangular part of A is not referenced. */
617 /* > On exit, A contains details of the partial factorization. */
620 /* > \param[in] LDA */
622 /* > LDA is INTEGER */
623 /* > The leading dimension of the array A. LDA >= f2cmax(1,N). */
626 /* > \param[out] IPIV */
628 /* > IPIV is INTEGER array, dimension (N) */
629 /* > Details of the interchanges and the block structure of D. */
631 /* > If UPLO = 'U': */
632 /* > Only the last KB elements of IPIV are set. */
634 /* > If IPIV(k) > 0, then rows and columns k and IPIV(k) were */
635 /* > interchanged and D(k,k) is a 1-by-1 diagonal block. */
637 /* > If IPIV(k) < 0 and IPIV(k-1) < 0, then rows and */
638 /* > columns k and -IPIV(k) were interchanged and rows and */
639 /* > columns k-1 and -IPIV(k-1) were inerchaged, */
640 /* > D(k-1:k,k-1:k) is a 2-by-2 diagonal block. */
642 /* > If UPLO = 'L': */
643 /* > Only the first KB elements of IPIV are set. */
645 /* > If IPIV(k) > 0, then rows and columns k and IPIV(k) */
646 /* > were interchanged and D(k,k) is a 1-by-1 diagonal block. */
648 /* > If IPIV(k) < 0 and IPIV(k+1) < 0, then rows and */
649 /* > columns k and -IPIV(k) were interchanged and rows and */
650 /* > columns k+1 and -IPIV(k+1) were inerchaged, */
651 /* > D(k:k+1,k:k+1) is a 2-by-2 diagonal block. */
654 /* > \param[out] W */
656 /* > W is REAL array, dimension (LDW,NB) */
659 /* > \param[in] LDW */
661 /* > LDW is INTEGER */
662 /* > The leading dimension of the array W. LDW >= f2cmax(1,N). */
665 /* > \param[out] INFO */
667 /* > INFO is INTEGER */
668 /* > = 0: successful exit */
669 /* > > 0: if INFO = k, D(k,k) is exactly zero. The factorization */
670 /* > has been completed, but the block diagonal matrix D is */
671 /* > exactly singular. */
677 /* > \author Univ. of Tennessee */
678 /* > \author Univ. of California Berkeley */
679 /* > \author Univ. of Colorado Denver */
680 /* > \author NAG Ltd. */
682 /* > \date November 2013 */
684 /* > \ingroup realSYcomputational */
686 /* > \par Contributors: */
687 /* ================== */
691 /* > November 2013, Igor Kozachenko, */
692 /* > Computer Science Division, */
693 /* > University of California, Berkeley */
695 /* > September 2007, Sven Hammarling, Nicholas J. Higham, Craig Lucas, */
696 /* > School of Mathematics, */
697 /* > University of Manchester */
701 /* ===================================================================== */
702 /* Subroutine */ int slasyf_rook_(char *uplo, integer *n, integer *nb,
703 integer *kb, real *a, integer *lda, integer *ipiv, real *w, integer *
706 /* System generated locals */
707 integer a_dim1, a_offset, w_dim1, w_offset, i__1, i__2, i__3, i__4, i__5;
710 /* Local variables */
712 integer imax, jmax, j, k, p;
714 extern logical lsame_(char *, char *);
715 extern /* Subroutine */ int sscal_(integer *, real *, real *, integer *),
716 sgemm_(char *, char *, integer *, integer *, integer *, real *,
717 real *, integer *, real *, integer *, real *, real *, integer *);
720 extern /* Subroutine */ int sgemv_(char *, integer *, integer *, real *,
721 real *, integer *, real *, integer *, real *, real *, integer *);
724 extern /* Subroutine */ int scopy_(integer *, real *, integer *, real *,
725 integer *), sswap_(integer *, real *, integer *, real *, integer *
727 real r1, d11, d12, d21, d22;
728 integer jb, ii, jj, kk, kp;
731 extern real slamch_(char *);
732 extern integer isamax_(integer *, real *, integer *);
739 /* -- LAPACK computational routine (version 3.5.0) -- */
740 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
741 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
745 /* ===================================================================== */
748 /* Parameter adjustments */
750 a_offset = 1 + a_dim1 * 1;
754 w_offset = 1 + w_dim1 * 1;
760 /* Initialize ALPHA for use in choosing pivot block size. */
762 alpha = (sqrt(17.f) + 1.f) / 8.f;
764 /* Compute machine safe minimum */
766 sfmin = slamch_("S");
768 if (lsame_(uplo, "U")) {
770 /* Factorize the trailing columns of A using the upper triangle */
771 /* of A and working backwards, and compute the matrix W = U12*D */
772 /* for use in updating A11 */
774 /* K is the main loop index, decreasing from N in steps of 1 or 2 */
779 /* KW is the column of W which corresponds to column K of A */
785 if (k <= *n - *nb + 1 && *nb < *n || k < 1) {
792 /* Copy column K of A to column KW of W and update it */
794 scopy_(&k, &a[k * a_dim1 + 1], &c__1, &w[kw * w_dim1 + 1], &c__1);
797 sgemv_("No transpose", &k, &i__1, &c_b9, &a[(k + 1) * a_dim1 + 1],
798 lda, &w[k + (kw + 1) * w_dim1], ldw, &c_b10, &w[kw *
802 /* Determine rows and columns to be interchanged and whether */
803 /* a 1-by-1 or 2-by-2 pivot block will be used */
805 absakk = (r__1 = w[k + kw * w_dim1], abs(r__1));
807 /* IMAX is the row-index of the largest off-diagonal element in */
808 /* column K, and COLMAX is its absolute value. */
809 /* Determine both COLMAX and IMAX. */
813 imax = isamax_(&i__1, &w[kw * w_dim1 + 1], &c__1);
814 colmax = (r__1 = w[imax + kw * w_dim1], abs(r__1));
819 if (f2cmax(absakk,colmax) == 0.f) {
821 /* Column K is zero or underflow: set INFO and continue */
827 scopy_(&k, &w[kw * w_dim1 + 1], &c__1, &a[k * a_dim1 + 1], &c__1);
830 /* ============================================================ */
832 /* Test for interchange */
834 /* Equivalent to testing for ABSAKK.GE.ALPHA*COLMAX */
835 /* (used to handle NaN and Inf) */
837 if (! (absakk < alpha * colmax)) {
839 /* no interchange, use 1-by-1 pivot block */
847 /* Loop until pivot found */
851 /* Begin pivot search loop body */
854 /* Copy column IMAX to column KW-1 of W and update it */
856 scopy_(&imax, &a[imax * a_dim1 + 1], &c__1, &w[(kw - 1) *
859 scopy_(&i__1, &a[imax + (imax + 1) * a_dim1], lda, &w[imax +
860 1 + (kw - 1) * w_dim1], &c__1);
864 sgemv_("No transpose", &k, &i__1, &c_b9, &a[(k + 1) *
865 a_dim1 + 1], lda, &w[imax + (kw + 1) * w_dim1],
866 ldw, &c_b10, &w[(kw - 1) * w_dim1 + 1], &c__1);
869 /* JMAX is the column-index of the largest off-diagonal */
870 /* element in row IMAX, and ROWMAX is its absolute value. */
871 /* Determine both ROWMAX and JMAX. */
875 jmax = imax + isamax_(&i__1, &w[imax + 1 + (kw - 1) *
877 rowmax = (r__1 = w[jmax + (kw - 1) * w_dim1], abs(r__1));
884 itemp = isamax_(&i__1, &w[(kw - 1) * w_dim1 + 1], &c__1);
885 stemp = (r__1 = w[itemp + (kw - 1) * w_dim1], abs(r__1));
886 if (stemp > rowmax) {
892 /* Equivalent to testing for */
893 /* ABS( W( IMAX, KW-1 ) ).GE.ALPHA*ROWMAX */
894 /* (used to handle NaN and Inf) */
896 if (! ((r__1 = w[imax + (kw - 1) * w_dim1], abs(r__1)) <
899 /* interchange rows and columns K and IMAX, */
900 /* use 1-by-1 pivot block */
904 /* copy column KW-1 of W to column KW of W */
906 scopy_(&k, &w[(kw - 1) * w_dim1 + 1], &c__1, &w[kw *
911 /* Equivalent to testing for ROWMAX.EQ.COLMAX, */
912 /* (used to handle NaN and Inf) */
914 } else if (p == jmax || rowmax <= colmax) {
916 /* interchange rows and columns K-1 and IMAX, */
917 /* use 2-by-2 pivot block */
924 /* Pivot not found: set params and repeat */
930 /* Copy updated JMAXth (next IMAXth) column to Kth of W */
932 scopy_(&k, &w[(kw - 1) * w_dim1 + 1], &c__1, &w[kw *
937 /* End pivot search loop body */
945 /* ============================================================ */
949 /* KKW is the column of W which corresponds to column KK of A */
953 if (kstep == 2 && p != k) {
955 /* Copy non-updated column K to column P */
958 scopy_(&i__1, &a[p + 1 + k * a_dim1], &c__1, &a[p + (p + 1) *
960 scopy_(&p, &a[k * a_dim1 + 1], &c__1, &a[p * a_dim1 + 1], &
963 /* Interchange rows K and P in last N-K+1 columns of A */
964 /* and last N-K+2 columns of W */
967 sswap_(&i__1, &a[k + k * a_dim1], lda, &a[p + k * a_dim1],
970 sswap_(&i__1, &w[k + kkw * w_dim1], ldw, &w[p + kkw * w_dim1],
974 /* Updated column KP is already stored in column KKW of W */
978 /* Copy non-updated column KK to column KP */
980 a[kp + k * a_dim1] = a[kk + k * a_dim1];
982 scopy_(&i__1, &a[kp + 1 + kk * a_dim1], &c__1, &a[kp + (kp +
984 scopy_(&kp, &a[kk * a_dim1 + 1], &c__1, &a[kp * a_dim1 + 1], &
987 /* Interchange rows KK and KP in last N-KK+1 columns */
991 sswap_(&i__1, &a[kk + kk * a_dim1], lda, &a[kp + kk * a_dim1],
994 sswap_(&i__1, &w[kk + kkw * w_dim1], ldw, &w[kp + kkw *
1000 /* 1-by-1 pivot block D(k): column KW of W now holds */
1002 /* W(k) = U(k)*D(k) */
1004 /* where U(k) is the k-th column of U */
1006 /* Store U(k) in column k of A */
1008 scopy_(&k, &w[kw * w_dim1 + 1], &c__1, &a[k * a_dim1 + 1], &
1011 if ((r__1 = a[k + k * a_dim1], abs(r__1)) >= sfmin) {
1012 r1 = 1.f / a[k + k * a_dim1];
1014 sscal_(&i__1, &r1, &a[k * a_dim1 + 1], &c__1);
1015 } else if (a[k + k * a_dim1] != 0.f) {
1017 for (ii = 1; ii <= i__1; ++ii) {
1018 a[ii + k * a_dim1] /= a[k + k * a_dim1];
1026 /* 2-by-2 pivot block D(k): columns KW and KW-1 of W now */
1029 /* ( W(k-1) W(k) ) = ( U(k-1) U(k) )*D(k) */
1031 /* where U(k) and U(k-1) are the k-th and (k-1)-th columns */
1036 /* Store U(k) and U(k-1) in columns k and k-1 of A */
1038 d12 = w[k - 1 + kw * w_dim1];
1039 d11 = w[k + kw * w_dim1] / d12;
1040 d22 = w[k - 1 + (kw - 1) * w_dim1] / d12;
1041 t = 1.f / (d11 * d22 - 1.f);
1043 for (j = 1; j <= i__1; ++j) {
1044 a[j + (k - 1) * a_dim1] = t * ((d11 * w[j + (kw - 1) *
1045 w_dim1] - w[j + kw * w_dim1]) / d12);
1046 a[j + k * a_dim1] = t * ((d22 * w[j + kw * w_dim1] -
1047 w[j + (kw - 1) * w_dim1]) / d12);
1052 /* Copy D(k) to A */
1054 a[k - 1 + (k - 1) * a_dim1] = w[k - 1 + (kw - 1) * w_dim1];
1055 a[k - 1 + k * a_dim1] = w[k - 1 + kw * w_dim1];
1056 a[k + k * a_dim1] = w[k + kw * w_dim1];
1060 /* Store details of the interchanges in IPIV */
1069 /* Decrease K and return to the start of the main loop */
1076 /* Update the upper triangle of A11 (= A(1:k,1:k)) as */
1078 /* A11 := A11 - U12*D*U12**T = A11 - U12*W**T */
1080 /* computing blocks of NB columns at a time */
1083 for (j = (k - 1) / *nb * *nb + 1; i__1 < 0 ? j >= 1 : j <= 1; j +=
1086 i__2 = *nb, i__3 = k - j + 1;
1087 jb = f2cmin(i__2,i__3);
1089 /* Update the upper triangle of the diagonal block */
1092 for (jj = j; jj <= i__2; ++jj) {
1095 sgemv_("No transpose", &i__3, &i__4, &c_b9, &a[j + (k + 1) *
1096 a_dim1], lda, &w[jj + (kw + 1) * w_dim1], ldw, &c_b10,
1097 &a[j + jj * a_dim1], &c__1);
1101 /* Update the rectangular superdiagonal block */
1106 sgemm_("No transpose", "Transpose", &i__2, &jb, &i__3, &c_b9,
1107 &a[(k + 1) * a_dim1 + 1], lda, &w[j + (kw + 1) *
1108 w_dim1], ldw, &c_b10, &a[j * a_dim1 + 1], lda);
1113 /* Put U12 in standard form by partially undoing the interchanges */
1114 /* in columns k+1:n */
1131 if (jp2 != jj && j <= *n) {
1133 sswap_(&i__1, &a[jp2 + j * a_dim1], lda, &a[jj + j * a_dim1], lda)
1137 if (jp1 != jj && kstep == 2) {
1139 sswap_(&i__1, &a[jp1 + j * a_dim1], lda, &a[jj + j * a_dim1], lda)
1146 /* Set KB to the number of columns factorized */
1152 /* Factorize the leading columns of A using the lower triangle */
1153 /* of A and working forwards, and compute the matrix W = L21*D */
1154 /* for use in updating A22 */
1156 /* K is the main loop index, increasing from 1 in steps of 1 or 2 */
1161 /* Exit from loop */
1163 if (k >= *nb && *nb < *n || k > *n) {
1170 /* Copy column K of A to column K of W and update it */
1173 scopy_(&i__1, &a[k + k * a_dim1], &c__1, &w[k + k * w_dim1], &c__1);
1177 sgemv_("No transpose", &i__1, &i__2, &c_b9, &a[k + a_dim1], lda, &
1178 w[k + w_dim1], ldw, &c_b10, &w[k + k * w_dim1], &c__1);
1181 /* Determine rows and columns to be interchanged and whether */
1182 /* a 1-by-1 or 2-by-2 pivot block will be used */
1184 absakk = (r__1 = w[k + k * w_dim1], abs(r__1));
1186 /* IMAX is the row-index of the largest off-diagonal element in */
1187 /* column K, and COLMAX is its absolute value. */
1188 /* Determine both COLMAX and IMAX. */
1192 imax = k + isamax_(&i__1, &w[k + 1 + k * w_dim1], &c__1);
1193 colmax = (r__1 = w[imax + k * w_dim1], abs(r__1));
1198 if (f2cmax(absakk,colmax) == 0.f) {
1200 /* Column K is zero or underflow: set INFO and continue */
1207 scopy_(&i__1, &w[k + k * w_dim1], &c__1, &a[k + k * a_dim1], &
1211 /* ============================================================ */
1213 /* Test for interchange */
1215 /* Equivalent to testing for ABSAKK.GE.ALPHA*COLMAX */
1216 /* (used to handle NaN and Inf) */
1218 if (! (absakk < alpha * colmax)) {
1220 /* no interchange, use 1-by-1 pivot block */
1228 /* Loop until pivot found */
1232 /* Begin pivot search loop body */
1235 /* Copy column IMAX to column K+1 of W and update it */
1238 scopy_(&i__1, &a[imax + k * a_dim1], lda, &w[k + (k + 1) *
1240 i__1 = *n - imax + 1;
1241 scopy_(&i__1, &a[imax + imax * a_dim1], &c__1, &w[imax + (k +
1242 1) * w_dim1], &c__1);
1246 sgemv_("No transpose", &i__1, &i__2, &c_b9, &a[k + a_dim1]
1247 , lda, &w[imax + w_dim1], ldw, &c_b10, &w[k + (k
1248 + 1) * w_dim1], &c__1);
1251 /* JMAX is the column-index of the largest off-diagonal */
1252 /* element in row IMAX, and ROWMAX is its absolute value. */
1253 /* Determine both ROWMAX and JMAX. */
1257 jmax = k - 1 + isamax_(&i__1, &w[k + (k + 1) * w_dim1], &
1259 rowmax = (r__1 = w[jmax + (k + 1) * w_dim1], abs(r__1));
1266 itemp = imax + isamax_(&i__1, &w[imax + 1 + (k + 1) *
1268 stemp = (r__1 = w[itemp + (k + 1) * w_dim1], abs(r__1));
1269 if (stemp > rowmax) {
1275 /* Equivalent to testing for */
1276 /* ABS( W( IMAX, K+1 ) ).GE.ALPHA*ROWMAX */
1277 /* (used to handle NaN and Inf) */
1279 if (! ((r__1 = w[imax + (k + 1) * w_dim1], abs(r__1)) < alpha
1282 /* interchange rows and columns K and IMAX, */
1283 /* use 1-by-1 pivot block */
1287 /* copy column K+1 of W to column K of W */
1290 scopy_(&i__1, &w[k + (k + 1) * w_dim1], &c__1, &w[k + k *
1295 /* Equivalent to testing for ROWMAX.EQ.COLMAX, */
1296 /* (used to handle NaN and Inf) */
1298 } else if (p == jmax || rowmax <= colmax) {
1300 /* interchange rows and columns K+1 and IMAX, */
1301 /* use 2-by-2 pivot block */
1308 /* Pivot not found: set params and repeat */
1314 /* Copy updated JMAXth (next IMAXth) column to Kth of W */
1317 scopy_(&i__1, &w[k + (k + 1) * w_dim1], &c__1, &w[k + k *
1322 /* End pivot search loop body */
1330 /* ============================================================ */
1334 if (kstep == 2 && p != k) {
1336 /* Copy non-updated column K to column P */
1339 scopy_(&i__1, &a[k + k * a_dim1], &c__1, &a[p + k * a_dim1],
1342 scopy_(&i__1, &a[p + k * a_dim1], &c__1, &a[p + p * a_dim1], &
1345 /* Interchange rows K and P in first K columns of A */
1346 /* and first K+1 columns of W */
1348 sswap_(&k, &a[k + a_dim1], lda, &a[p + a_dim1], lda);
1349 sswap_(&kk, &w[k + w_dim1], ldw, &w[p + w_dim1], ldw);
1352 /* Updated column KP is already stored in column KK of W */
1356 /* Copy non-updated column KK to column KP */
1358 a[kp + k * a_dim1] = a[kk + k * a_dim1];
1360 scopy_(&i__1, &a[k + 1 + kk * a_dim1], &c__1, &a[kp + (k + 1)
1363 scopy_(&i__1, &a[kp + kk * a_dim1], &c__1, &a[kp + kp *
1366 /* Interchange rows KK and KP in first KK columns of A and W */
1368 sswap_(&kk, &a[kk + a_dim1], lda, &a[kp + a_dim1], lda);
1369 sswap_(&kk, &w[kk + w_dim1], ldw, &w[kp + w_dim1], ldw);
1374 /* 1-by-1 pivot block D(k): column k of W now holds */
1376 /* W(k) = L(k)*D(k) */
1378 /* where L(k) is the k-th column of L */
1380 /* Store L(k) in column k of A */
1383 scopy_(&i__1, &w[k + k * w_dim1], &c__1, &a[k + k * a_dim1], &
1386 if ((r__1 = a[k + k * a_dim1], abs(r__1)) >= sfmin) {
1387 r1 = 1.f / a[k + k * a_dim1];
1389 sscal_(&i__1, &r1, &a[k + 1 + k * a_dim1], &c__1);
1390 } else if (a[k + k * a_dim1] != 0.f) {
1392 for (ii = k + 1; ii <= i__1; ++ii) {
1393 a[ii + k * a_dim1] /= a[k + k * a_dim1];
1401 /* 2-by-2 pivot block D(k): columns k and k+1 of W now hold */
1403 /* ( W(k) W(k+1) ) = ( L(k) L(k+1) )*D(k) */
1405 /* where L(k) and L(k+1) are the k-th and (k+1)-th columns */
1410 /* Store L(k) and L(k+1) in columns k and k+1 of A */
1412 d21 = w[k + 1 + k * w_dim1];
1413 d11 = w[k + 1 + (k + 1) * w_dim1] / d21;
1414 d22 = w[k + k * w_dim1] / d21;
1415 t = 1.f / (d11 * d22 - 1.f);
1417 for (j = k + 2; j <= i__1; ++j) {
1418 a[j + k * a_dim1] = t * ((d11 * w[j + k * w_dim1] - w[
1419 j + (k + 1) * w_dim1]) / d21);
1420 a[j + (k + 1) * a_dim1] = t * ((d22 * w[j + (k + 1) *
1421 w_dim1] - w[j + k * w_dim1]) / d21);
1426 /* Copy D(k) to A */
1428 a[k + k * a_dim1] = w[k + k * w_dim1];
1429 a[k + 1 + k * a_dim1] = w[k + 1 + k * w_dim1];
1430 a[k + 1 + (k + 1) * a_dim1] = w[k + 1 + (k + 1) * w_dim1];
1434 /* Store details of the interchanges in IPIV */
1443 /* Increase K and return to the start of the main loop */
1450 /* Update the lower triangle of A22 (= A(k:n,k:n)) as */
1452 /* A22 := A22 - L21*D*L21**T = A22 - L21*W**T */
1454 /* computing blocks of NB columns at a time */
1458 for (j = k; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
1460 i__3 = *nb, i__4 = *n - j + 1;
1461 jb = f2cmin(i__3,i__4);
1463 /* Update the lower triangle of the diagonal block */
1466 for (jj = j; jj <= i__3; ++jj) {
1469 sgemv_("No transpose", &i__4, &i__5, &c_b9, &a[jj + a_dim1],
1470 lda, &w[jj + w_dim1], ldw, &c_b10, &a[jj + jj *
1475 /* Update the rectangular subdiagonal block */
1478 i__3 = *n - j - jb + 1;
1480 sgemm_("No transpose", "Transpose", &i__3, &jb, &i__4, &c_b9,
1481 &a[j + jb + a_dim1], lda, &w[j + w_dim1], ldw, &c_b10,
1482 &a[j + jb + j * a_dim1], lda);
1487 /* Put L21 in standard form by partially undoing the interchanges */
1488 /* in columns 1:k-1 */
1505 if (jp2 != jj && j >= 1) {
1506 sswap_(&j, &a[jp2 + a_dim1], lda, &a[jj + a_dim1], lda);
1509 if (jp1 != jj && kstep == 2) {
1510 sswap_(&j, &a[jp1 + a_dim1], lda, &a[jj + a_dim1], lda);
1516 /* Set KB to the number of columns factorized */
1523 /* End of SLASYF_ROOK */
1525 } /* slasyf_rook__ */