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 doublereal c_b3 = -1.;
516 static integer c__1 = 1;
518 /* > \brief \b DLAED2 used by sstedc. Merges eigenvalues and deflates secular equation. Used when the original
519 matrix is tridiagonal. */
521 /* =========== DOCUMENTATION =========== */
523 /* Online html documentation available at */
524 /* http://www.netlib.org/lapack/explore-html/ */
527 /* > Download DLAED2 + dependencies */
528 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlaed2.
531 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlaed2.
534 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlaed2.
542 /* SUBROUTINE DLAED2( K, N, N1, D, Q, LDQ, INDXQ, RHO, Z, DLAMDA, W, */
543 /* Q2, INDX, INDXC, INDXP, COLTYP, INFO ) */
545 /* INTEGER INFO, K, LDQ, N, N1 */
546 /* DOUBLE PRECISION RHO */
547 /* INTEGER COLTYP( * ), INDX( * ), INDXC( * ), INDXP( * ), */
549 /* DOUBLE PRECISION D( * ), DLAMDA( * ), Q( LDQ, * ), Q2( * ), */
550 /* $ W( * ), Z( * ) */
553 /* > \par Purpose: */
558 /* > DLAED2 merges the two sets of eigenvalues together into a single */
559 /* > sorted set. Then it tries to deflate the size of the problem. */
560 /* > There are two ways in which deflation can occur: when two or more */
561 /* > eigenvalues are close together or if there is a tiny entry in the */
562 /* > Z vector. For each such occurrence the order of the related secular */
563 /* > equation problem is reduced by one. */
569 /* > \param[out] K */
572 /* > The number of non-deflated eigenvalues, and the order of the */
573 /* > related secular equation. 0 <= K <=N. */
579 /* > The dimension of the symmetric tridiagonal matrix. N >= 0. */
582 /* > \param[in] N1 */
584 /* > N1 is INTEGER */
585 /* > The location of the last eigenvalue in the leading sub-matrix. */
586 /* > f2cmin(1,N) <= N1 <= N/2. */
589 /* > \param[in,out] D */
591 /* > D is DOUBLE PRECISION array, dimension (N) */
592 /* > On entry, D contains the eigenvalues of the two submatrices to */
594 /* > On exit, D contains the trailing (N-K) updated eigenvalues */
595 /* > (those which were deflated) sorted into increasing order. */
598 /* > \param[in,out] Q */
600 /* > Q is DOUBLE PRECISION array, dimension (LDQ, N) */
601 /* > On entry, Q contains the eigenvectors of two submatrices in */
602 /* > the two square blocks with corners at (1,1), (N1,N1) */
603 /* > and (N1+1, N1+1), (N,N). */
604 /* > On exit, Q contains the trailing (N-K) updated eigenvectors */
605 /* > (those which were deflated) in its last N-K columns. */
608 /* > \param[in] LDQ */
610 /* > LDQ is INTEGER */
611 /* > The leading dimension of the array Q. LDQ >= f2cmax(1,N). */
614 /* > \param[in,out] INDXQ */
616 /* > INDXQ is INTEGER array, dimension (N) */
617 /* > The permutation which separately sorts the two sub-problems */
618 /* > in D into ascending order. Note that elements in the second */
619 /* > half of this permutation must first have N1 added to their */
620 /* > values. Destroyed on exit. */
623 /* > \param[in,out] RHO */
625 /* > RHO is DOUBLE PRECISION */
626 /* > On entry, the off-diagonal element associated with the rank-1 */
627 /* > cut which originally split the two submatrices which are now */
628 /* > being recombined. */
629 /* > On exit, RHO has been modified to the value required by */
635 /* > Z is DOUBLE PRECISION array, dimension (N) */
636 /* > On entry, Z contains the updating vector (the last */
637 /* > row of the first sub-eigenvector matrix and the first row of */
638 /* > the second sub-eigenvector matrix). */
639 /* > On exit, the contents of Z have been destroyed by the updating */
643 /* > \param[out] DLAMDA */
645 /* > DLAMDA is DOUBLE PRECISION array, dimension (N) */
646 /* > A copy of the first K eigenvalues which will be used by */
647 /* > DLAED3 to form the secular equation. */
650 /* > \param[out] W */
652 /* > W is DOUBLE PRECISION array, dimension (N) */
653 /* > The first k values of the final deflation-altered z-vector */
654 /* > which will be passed to DLAED3. */
657 /* > \param[out] Q2 */
659 /* > Q2 is DOUBLE PRECISION array, dimension (N1**2+(N-N1)**2) */
660 /* > A copy of the first K eigenvectors which will be used by */
661 /* > DLAED3 in a matrix multiply (DGEMM) to solve for the new */
662 /* > eigenvectors. */
665 /* > \param[out] INDX */
667 /* > INDX is INTEGER array, dimension (N) */
668 /* > The permutation used to sort the contents of DLAMDA into */
669 /* > ascending order. */
672 /* > \param[out] INDXC */
674 /* > INDXC is INTEGER array, dimension (N) */
675 /* > The permutation used to arrange the columns of the deflated */
676 /* > Q matrix into three groups: the first group contains non-zero */
677 /* > elements only at and above N1, the second contains */
678 /* > non-zero elements only below N1, and the third is dense. */
681 /* > \param[out] INDXP */
683 /* > INDXP is INTEGER array, dimension (N) */
684 /* > The permutation used to place deflated values of D at the end */
685 /* > of the array. INDXP(1:K) points to the nondeflated D-values */
686 /* > and INDXP(K+1:N) points to the deflated eigenvalues. */
689 /* > \param[out] COLTYP */
691 /* > COLTYP is INTEGER array, dimension (N) */
692 /* > During execution, a label which will indicate which of the */
693 /* > following types a column in the Q2 matrix is: */
694 /* > 1 : non-zero in the upper half only; */
696 /* > 3 : non-zero in the lower half only; */
697 /* > 4 : deflated. */
698 /* > On exit, COLTYP(i) is the number of columns of type i, */
699 /* > for i=1 to 4 only. */
702 /* > \param[out] INFO */
704 /* > INFO is INTEGER */
705 /* > = 0: successful exit. */
706 /* > < 0: if INFO = -i, the i-th argument had an illegal value. */
712 /* > \author Univ. of Tennessee */
713 /* > \author Univ. of California Berkeley */
714 /* > \author Univ. of Colorado Denver */
715 /* > \author NAG Ltd. */
717 /* > \date December 2016 */
719 /* > \ingroup auxOTHERcomputational */
721 /* > \par Contributors: */
722 /* ================== */
724 /* > Jeff Rutter, Computer Science Division, University of California */
725 /* > at Berkeley, USA \n */
726 /* > Modified by Francoise Tisseur, University of Tennessee */
728 /* ===================================================================== */
729 /* Subroutine */ int dlaed2_(integer *k, integer *n, integer *n1, doublereal *
730 d__, doublereal *q, integer *ldq, integer *indxq, doublereal *rho,
731 doublereal *z__, doublereal *dlamda, doublereal *w, doublereal *q2,
732 integer *indx, integer *indxc, integer *indxp, integer *coltyp,
735 /* System generated locals */
736 integer q_dim1, q_offset, i__1, i__2;
737 doublereal d__1, d__2, d__3, d__4;
739 /* Local variables */
741 extern /* Subroutine */ int drot_(integer *, doublereal *, integer *,
742 doublereal *, integer *, doublereal *, doublereal *);
747 extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
748 integer *), dcopy_(integer *, doublereal *, integer *, doublereal
751 extern doublereal dlapy2_(doublereal *, doublereal *);
753 extern doublereal dlamch_(char *);
755 extern integer idamax_(integer *, doublereal *, integer *);
756 extern /* Subroutine */ int dlamrg_(integer *, integer *, doublereal *,
757 integer *, integer *, integer *), dlacpy_(char *, integer *,
758 integer *, doublereal *, integer *, doublereal *, integer *), xerbla_(char *, integer *, ftnlen);
759 integer iq1, iq2, n1p1;
760 doublereal eps, tau, tol;
764 /* -- LAPACK computational routine (version 3.7.0) -- */
765 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
766 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
770 /* ===================================================================== */
773 /* Test the input parameters. */
775 /* Parameter adjustments */
778 q_offset = 1 + q_dim1 * 1;
795 } else if (*ldq < f2cmax(1,*n)) {
797 } else /* if(complicated condition) */ {
799 i__1 = 1, i__2 = *n / 2;
800 if (f2cmin(i__1,i__2) > *n1 || *n / 2 < *n1) {
806 xerbla_("DLAED2", &i__1, (ftnlen)6);
810 /* Quick return if possible */
820 dscal_(&n2, &c_b3, &z__[n1p1], &c__1);
823 /* Normalize z so that norm(z) = 1. Since z is the concatenation of */
824 /* two normalized vectors, norm2(z) = sqrt(2). */
827 dscal_(n, &t, &z__[1], &c__1);
829 /* RHO = ABS( norm(z)**2 * RHO ) */
831 *rho = (d__1 = *rho * 2., abs(d__1));
833 /* Sort the eigenvalues into increasing order */
836 for (i__ = n1p1; i__ <= i__1; ++i__) {
841 /* re-integrate the deflated parts from the last pass */
844 for (i__ = 1; i__ <= i__1; ++i__) {
845 dlamda[i__] = d__[indxq[i__]];
848 dlamrg_(n1, &n2, &dlamda[1], &c__1, &c__1, &indxc[1]);
850 for (i__ = 1; i__ <= i__1; ++i__) {
851 indx[i__] = indxq[indxc[i__]];
855 /* Calculate the allowable deflation tolerance */
857 imax = idamax_(n, &z__[1], &c__1);
858 jmax = idamax_(n, &d__[1], &c__1);
859 eps = dlamch_("Epsilon");
861 d__3 = (d__1 = d__[jmax], abs(d__1)), d__4 = (d__2 = z__[imax], abs(d__2))
863 tol = eps * 8. * f2cmax(d__3,d__4);
865 /* If the rank-1 modifier is small enough, no more needs to be done */
866 /* except to reorganize Q so that its columns correspond with the */
869 if (*rho * (d__1 = z__[imax], abs(d__1)) <= tol) {
873 for (j = 1; j <= i__1; ++j) {
875 dcopy_(n, &q[i__ * q_dim1 + 1], &c__1, &q2[iq2], &c__1);
876 dlamda[j] = d__[i__];
880 dlacpy_("A", n, n, &q2[1], n, &q[q_offset], ldq);
881 dcopy_(n, &dlamda[1], &c__1, &d__[1], &c__1);
885 /* If there are multiple eigenvalues then the problem deflates. Here */
886 /* the number of equal eigenvalues are found. As each equal */
887 /* eigenvalue is found, an elementary reflector is computed to rotate */
888 /* the corresponding eigensubspace so that the corresponding */
889 /* components of Z are zero in this new basis. */
892 for (i__ = 1; i__ <= i__1; ++i__) {
897 for (i__ = n1p1; i__ <= i__1; ++i__) {
906 for (j = 1; j <= i__1; ++j) {
908 if (*rho * (d__1 = z__[nj], abs(d__1)) <= tol) {
910 /* Deflate due to small z component. */
930 if (*rho * (d__1 = z__[nj], abs(d__1)) <= tol) {
932 /* Deflate due to small z component. */
939 /* Check if eigenvalues are close enough to allow deflation. */
944 /* Find sqrt(a**2+b**2) without overflow or */
945 /* destructive underflow. */
947 tau = dlapy2_(&c__, &s);
948 t = d__[nj] - d__[pj];
951 if ((d__1 = t * c__ * s, abs(d__1)) <= tol) {
953 /* Deflation is possible. */
957 if (coltyp[nj] != coltyp[pj]) {
961 drot_(n, &q[pj * q_dim1 + 1], &c__1, &q[nj * q_dim1 + 1], &c__1, &
963 /* Computing 2nd power */
965 /* Computing 2nd power */
967 t = d__[pj] * (d__1 * d__1) + d__[nj] * (d__2 * d__2);
968 /* Computing 2nd power */
970 /* Computing 2nd power */
972 d__[nj] = d__[pj] * (d__1 * d__1) + d__[nj] * (d__2 * d__2);
977 if (k2 + i__ <= *n) {
978 if (d__[pj] < d__[indxp[k2 + i__]]) {
979 indxp[k2 + i__ - 1] = indxp[k2 + i__];
980 indxp[k2 + i__] = pj;
984 indxp[k2 + i__ - 1] = pj;
987 indxp[k2 + i__ - 1] = pj;
992 dlamda[*k] = d__[pj];
1001 /* Record the last eigenvalue. */
1004 dlamda[*k] = d__[pj];
1008 /* Count up the total number of the various types of columns, then */
1009 /* form a permutation which positions the four column types into */
1010 /* four uniform groups (although one or more of these groups may be */
1013 for (j = 1; j <= 4; ++j) {
1018 for (j = 1; j <= i__1; ++j) {
1024 /* PSM(*) = Position in SubMatrix (of types 1 through 4) */
1027 psm[1] = ctot[0] + 1;
1028 psm[2] = psm[1] + ctot[1];
1029 psm[3] = psm[2] + ctot[2];
1032 /* Fill out the INDXC array so that the permutation which it induces */
1033 /* will place all type-1 columns first, all type-2 columns next, */
1034 /* then all type-3's, and finally all type-4's. */
1037 for (j = 1; j <= i__1; ++j) {
1040 indx[psm[ct - 1]] = js;
1041 indxc[psm[ct - 1]] = j;
1046 /* Sort the eigenvalues and corresponding eigenvectors into DLAMDA */
1047 /* and Q2 respectively. The eigenvalues/vectors which were not */
1048 /* deflated go into the first K slots of DLAMDA and Q2 respectively, */
1049 /* while those which were deflated go into the last N - K slots. */
1053 iq2 = (ctot[0] + ctot[1]) * *n1 + 1;
1055 for (j = 1; j <= i__1; ++j) {
1057 dcopy_(n1, &q[js * q_dim1 + 1], &c__1, &q2[iq1], &c__1);
1065 for (j = 1; j <= i__1; ++j) {
1067 dcopy_(n1, &q[js * q_dim1 + 1], &c__1, &q2[iq1], &c__1);
1068 dcopy_(&n2, &q[*n1 + 1 + js * q_dim1], &c__1, &q2[iq2], &c__1);
1077 for (j = 1; j <= i__1; ++j) {
1079 dcopy_(&n2, &q[*n1 + 1 + js * q_dim1], &c__1, &q2[iq2], &c__1);
1088 for (j = 1; j <= i__1; ++j) {
1090 dcopy_(n, &q[js * q_dim1 + 1], &c__1, &q2[iq2], &c__1);
1097 /* The deflated eigenvalues and their corresponding vectors go back */
1098 /* into the last N - K slots of D and Q respectively. */
1101 dlacpy_("A", n, &ctot[3], &q2[iq1], n, &q[(*k + 1) * q_dim1 + 1], ldq);
1103 dcopy_(&i__1, &z__[*k + 1], &c__1, &d__[*k + 1], &c__1);
1106 /* Copy CTOT into COLTYP for referencing in DLAED3. */
1108 for (j = 1; j <= 4; ++j) {
1109 coltyp[j] = ctot[j - 1];