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 doublecomplex c_b1 = {0.,0.};
516 static doublecomplex c_b2 = {1.,0.};
517 static integer c__1 = 1;
519 /* > \brief \b ZHBGST */
521 /* =========== DOCUMENTATION =========== */
523 /* Online html documentation available at */
524 /* http://www.netlib.org/lapack/explore-html/ */
527 /* > Download ZHBGST + dependencies */
528 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/zhbgst.
531 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/zhbgst.
534 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/zhbgst.
542 /* SUBROUTINE ZHBGST( VECT, UPLO, N, KA, KB, AB, LDAB, BB, LDBB, X, */
543 /* LDX, WORK, RWORK, INFO ) */
545 /* CHARACTER UPLO, VECT */
546 /* INTEGER INFO, KA, KB, LDAB, LDBB, LDX, N */
547 /* DOUBLE PRECISION RWORK( * ) */
548 /* COMPLEX*16 AB( LDAB, * ), BB( LDBB, * ), WORK( * ), */
552 /* > \par Purpose: */
557 /* > ZHBGST reduces a complex Hermitian-definite banded generalized */
558 /* > eigenproblem A*x = lambda*B*x to standard form C*y = lambda*y, */
559 /* > such that C has the same bandwidth as A. */
561 /* > B must have been previously factorized as S**H*S by ZPBSTF, using a */
562 /* > split Cholesky factorization. A is overwritten by C = X**H*A*X, where */
563 /* > X = S**(-1)*Q and Q is a unitary matrix chosen to preserve the */
564 /* > bandwidth of A. */
570 /* > \param[in] VECT */
572 /* > VECT is CHARACTER*1 */
573 /* > = 'N': do not form the transformation matrix X; */
574 /* > = 'V': form X. */
577 /* > \param[in] UPLO */
579 /* > UPLO is CHARACTER*1 */
580 /* > = 'U': Upper triangle of A is stored; */
581 /* > = 'L': Lower triangle of A is stored. */
587 /* > The order of the matrices A and B. N >= 0. */
590 /* > \param[in] KA */
592 /* > KA is INTEGER */
593 /* > The number of superdiagonals of the matrix A if UPLO = 'U', */
594 /* > or the number of subdiagonals if UPLO = 'L'. KA >= 0. */
597 /* > \param[in] KB */
599 /* > KB is INTEGER */
600 /* > The number of superdiagonals of the matrix B if UPLO = 'U', */
601 /* > or the number of subdiagonals if UPLO = 'L'. KA >= KB >= 0. */
604 /* > \param[in,out] AB */
606 /* > AB is COMPLEX*16 array, dimension (LDAB,N) */
607 /* > On entry, the upper or lower triangle of the Hermitian band */
608 /* > matrix A, stored in the first ka+1 rows of the array. The */
609 /* > j-th column of A is stored in the j-th column of the array AB */
611 /* > if UPLO = 'U', AB(ka+1+i-j,j) = A(i,j) for f2cmax(1,j-ka)<=i<=j; */
612 /* > if UPLO = 'L', AB(1+i-j,j) = A(i,j) for j<=i<=f2cmin(n,j+ka). */
614 /* > On exit, the transformed matrix X**H*A*X, stored in the same */
618 /* > \param[in] LDAB */
620 /* > LDAB is INTEGER */
621 /* > The leading dimension of the array AB. LDAB >= KA+1. */
624 /* > \param[in] BB */
626 /* > BB is COMPLEX*16 array, dimension (LDBB,N) */
627 /* > The banded factor S from the split Cholesky factorization of */
628 /* > B, as returned by ZPBSTF, stored in the first kb+1 rows of */
632 /* > \param[in] LDBB */
634 /* > LDBB is INTEGER */
635 /* > The leading dimension of the array BB. LDBB >= KB+1. */
638 /* > \param[out] X */
640 /* > X is COMPLEX*16 array, dimension (LDX,N) */
641 /* > If VECT = 'V', the n-by-n matrix X. */
642 /* > If VECT = 'N', the array X is not referenced. */
645 /* > \param[in] LDX */
647 /* > LDX is INTEGER */
648 /* > The leading dimension of the array X. */
649 /* > LDX >= f2cmax(1,N) if VECT = 'V'; LDX >= 1 otherwise. */
652 /* > \param[out] WORK */
654 /* > WORK is COMPLEX*16 array, dimension (N) */
657 /* > \param[out] RWORK */
659 /* > RWORK is DOUBLE PRECISION array, dimension (N) */
662 /* > \param[out] INFO */
664 /* > INFO is INTEGER */
665 /* > = 0: successful exit */
666 /* > < 0: if INFO = -i, the i-th argument had an illegal value. */
672 /* > \author Univ. of Tennessee */
673 /* > \author Univ. of California Berkeley */
674 /* > \author Univ. of Colorado Denver */
675 /* > \author NAG Ltd. */
677 /* > \date December 2016 */
679 /* > \ingroup complex16OTHERcomputational */
681 /* ===================================================================== */
682 /* Subroutine */ int zhbgst_(char *vect, char *uplo, integer *n, integer *ka,
683 integer *kb, doublecomplex *ab, integer *ldab, doublecomplex *bb,
684 integer *ldbb, doublecomplex *x, integer *ldx, doublecomplex *work,
685 doublereal *rwork, integer *info)
687 /* System generated locals */
688 integer ab_dim1, ab_offset, bb_dim1, bb_offset, x_dim1, x_offset, i__1,
689 i__2, i__3, i__4, i__5, i__6, i__7, i__8;
691 doublecomplex z__1, z__2, z__3, z__4, z__5, z__6, z__7, z__8, z__9, z__10;
693 /* Local variables */
695 extern /* Subroutine */ int zrot_(integer *, doublecomplex *, integer *,
696 doublecomplex *, integer *, doublereal *, doublecomplex *);
697 integer i__, j, k, l, m;
699 extern logical lsame_(char *, char *);
700 extern /* Subroutine */ int zgerc_(integer *, integer *, doublecomplex *,
701 doublecomplex *, integer *, doublecomplex *, integer *,
702 doublecomplex *, integer *);
706 extern /* Subroutine */ int zgeru_(integer *, integer *, doublecomplex *,
707 doublecomplex *, integer *, doublecomplex *, integer *,
708 doublecomplex *, integer *);
710 extern /* Subroutine */ int zlar2v_(integer *, doublecomplex *,
711 doublecomplex *, doublecomplex *, integer *, doublereal *,
712 doublecomplex *, integer *);
715 extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen), zdscal_(
716 integer *, doublereal *, doublecomplex *, integer *);
718 extern /* Subroutine */ int zlacgv_(integer *, doublecomplex *, integer *)
721 extern /* Subroutine */ int zlaset_(char *, integer *, integer *,
722 doublecomplex *, doublecomplex *, doublecomplex *, integer *), zlartg_(doublecomplex *, doublecomplex *, doublereal *,
723 doublecomplex *, doublecomplex *);
725 extern /* Subroutine */ int zlargv_(integer *, doublecomplex *, integer *,
726 doublecomplex *, integer *, doublereal *, integer *);
728 extern /* Subroutine */ int zlartv_(integer *, doublecomplex *, integer *,
729 doublecomplex *, integer *, doublereal *, doublecomplex *,
735 /* -- LAPACK computational 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..-- */
741 /* ===================================================================== */
744 /* Test the input parameters */
746 /* Parameter adjustments */
748 ab_offset = 1 + ab_dim1 * 1;
751 bb_offset = 1 + bb_dim1 * 1;
754 x_offset = 1 + x_dim1 * 1;
760 wantx = lsame_(vect, "V");
761 upper = lsame_(uplo, "U");
765 if (! wantx && ! lsame_(vect, "N")) {
767 } else if (! upper && ! lsame_(uplo, "L")) {
771 } else if (*ka < 0) {
773 } else if (*kb < 0 || *kb > *ka) {
775 } else if (*ldab < *ka + 1) {
777 } else if (*ldbb < *kb + 1) {
779 } else if (*ldx < 1 || wantx && *ldx < f2cmax(1,*n)) {
784 xerbla_("ZHBGST", &i__1, (ftnlen)6);
788 /* Quick return if possible */
796 /* Initialize X to the unit matrix, if needed */
799 zlaset_("Full", n, n, &c_b1, &c_b2, &x[x_offset], ldx);
802 /* Set M to the splitting point m. It must be the same value as is */
803 /* used in ZPBSTF. The chosen value allows the arrays WORK and RWORK */
804 /* to be of dimension (N). */
808 /* The routine works in two phases, corresponding to the two halves */
809 /* of the split Cholesky factorization of B as S**H*S where */
814 /* with U upper triangular of order m, and L lower triangular of */
815 /* order n-m. S has the same bandwidth as B. */
817 /* S is treated as a product of elementary matrices: */
819 /* S = S(m)*S(m-1)*...*S(2)*S(1)*S(m+1)*S(m+2)*...*S(n-1)*S(n) */
821 /* where S(i) is determined by the i-th row of S. */
823 /* In phase 1, the index i takes the values n, n-1, ... , m+1; */
824 /* in phase 2, it takes the values 1, 2, ... , m. */
826 /* For each value of i, the current matrix A is updated by forming */
827 /* inv(S(i))**H*A*inv(S(i)). This creates a triangular bulge outside */
828 /* the band of A. The bulge is then pushed down toward the bottom of */
829 /* A in phase 1, and up toward the top of A in phase 2, by applying */
830 /* plane rotations. */
832 /* There are kb*(kb+1)/2 elements in the bulge, but at most 2*kb-1 */
833 /* of them are linearly independent, so annihilating a bulge requires */
834 /* only 2*kb-1 plane rotations. The rotations are divided into a 1st */
835 /* set of kb-1 rotations, and a 2nd set of kb rotations. */
837 /* Wherever possible, rotations are generated and applied in vector */
838 /* operations of length NR between the indices J1 and J2 (sometimes */
839 /* replaced by modified values NRT, J1T or J2T). */
841 /* The real cosines and complex sines of the rotations are stored in */
842 /* the arrays RWORK and WORK, those of the 1st set in elements */
843 /* 2:m-kb-1, and those of the 2nd set in elements m-kb+1:n. */
845 /* The bulges are not formed explicitly; nonzero elements outside the */
846 /* band are created only when they are required for generating new */
847 /* rotations; they are stored in the array WORK, in positions where */
848 /* they are later overwritten by the sines of the rotations which */
849 /* annihilate them. */
851 /* **************************** Phase 1 ***************************** */
853 /* The logical structure of this phase is: */
855 /* UPDATE = .TRUE. */
856 /* DO I = N, M + 1, -1 */
857 /* use S(i) to update A and create a new bulge */
858 /* apply rotations to push all bulges KA positions downward */
860 /* UPDATE = .FALSE. */
861 /* DO I = M + KA + 1, N - 1 */
862 /* apply rotations to push all bulges KA positions downward */
865 /* To avoid duplicating code, the two loops are merged. */
873 i__1 = *kb, i__2 = i__ - 1;
874 kbt = f2cmin(i__1,i__2);
877 i__1 = *n, i__2 = i__ + *ka;
878 i1 = f2cmin(i__1,i__2);
879 i2 = i__ - kbt + ka1;
898 /* Transform A, working with the upper triangle */
902 /* Form inv(S(i))**H * A * inv(S(i)) */
904 i__1 = kb1 + i__ * bb_dim1;
906 i__1 = ka1 + i__ * ab_dim1;
907 i__2 = ka1 + i__ * ab_dim1;
908 d__1 = ab[i__2].r / bii / bii;
909 ab[i__1].r = d__1, ab[i__1].i = 0.;
911 for (j = i__ + 1; j <= i__1; ++j) {
912 i__2 = i__ - j + ka1 + j * ab_dim1;
913 i__3 = i__ - j + ka1 + j * ab_dim1;
914 z__1.r = ab[i__3].r / bii, z__1.i = ab[i__3].i / bii;
915 ab[i__2].r = z__1.r, ab[i__2].i = z__1.i;
919 i__1 = 1, i__2 = i__ - *ka;
921 for (j = f2cmax(i__1,i__2); j <= i__3; ++j) {
922 i__1 = j - i__ + ka1 + i__ * ab_dim1;
923 i__2 = j - i__ + ka1 + i__ * ab_dim1;
924 z__1.r = ab[i__2].r / bii, z__1.i = ab[i__2].i / bii;
925 ab[i__1].r = z__1.r, ab[i__1].i = z__1.i;
929 for (k = i__ - kbt; k <= i__3; ++k) {
931 for (j = i__ - kbt; j <= i__1; ++j) {
932 i__2 = j - k + ka1 + k * ab_dim1;
933 i__4 = j - k + ka1 + k * ab_dim1;
934 i__5 = j - i__ + kb1 + i__ * bb_dim1;
935 d_cnjg(&z__5, &ab[k - i__ + ka1 + i__ * ab_dim1]);
936 z__4.r = bb[i__5].r * z__5.r - bb[i__5].i * z__5.i,
937 z__4.i = bb[i__5].r * z__5.i + bb[i__5].i *
939 z__3.r = ab[i__4].r - z__4.r, z__3.i = ab[i__4].i -
941 d_cnjg(&z__7, &bb[k - i__ + kb1 + i__ * bb_dim1]);
942 i__6 = j - i__ + ka1 + i__ * ab_dim1;
943 z__6.r = z__7.r * ab[i__6].r - z__7.i * ab[i__6].i,
944 z__6.i = z__7.r * ab[i__6].i + z__7.i * ab[i__6]
946 z__2.r = z__3.r - z__6.r, z__2.i = z__3.i - z__6.i;
947 i__7 = ka1 + i__ * ab_dim1;
949 i__8 = j - i__ + kb1 + i__ * bb_dim1;
950 z__9.r = d__1 * bb[i__8].r, z__9.i = d__1 * bb[i__8].i;
951 d_cnjg(&z__10, &bb[k - i__ + kb1 + i__ * bb_dim1]);
952 z__8.r = z__9.r * z__10.r - z__9.i * z__10.i, z__8.i =
953 z__9.r * z__10.i + z__9.i * z__10.r;
954 z__1.r = z__2.r + z__8.r, z__1.i = z__2.i + z__8.i;
955 ab[i__2].r = z__1.r, ab[i__2].i = z__1.i;
959 i__1 = 1, i__2 = i__ - *ka;
960 i__4 = i__ - kbt - 1;
961 for (j = f2cmax(i__1,i__2); j <= i__4; ++j) {
962 i__1 = j - k + ka1 + k * ab_dim1;
963 i__2 = j - k + ka1 + k * ab_dim1;
964 d_cnjg(&z__3, &bb[k - i__ + kb1 + i__ * bb_dim1]);
965 i__5 = j - i__ + ka1 + i__ * ab_dim1;
966 z__2.r = z__3.r * ab[i__5].r - z__3.i * ab[i__5].i,
967 z__2.i = z__3.r * ab[i__5].i + z__3.i * ab[i__5]
969 z__1.r = ab[i__2].r - z__2.r, z__1.i = ab[i__2].i -
971 ab[i__1].r = z__1.r, ab[i__1].i = z__1.i;
977 for (j = i__; j <= i__3; ++j) {
979 i__4 = j - *ka, i__1 = i__ - kbt;
981 for (k = f2cmax(i__4,i__1); k <= i__2; ++k) {
982 i__4 = k - j + ka1 + j * ab_dim1;
983 i__1 = k - j + ka1 + j * ab_dim1;
984 i__5 = k - i__ + kb1 + i__ * bb_dim1;
985 i__6 = i__ - j + ka1 + j * ab_dim1;
986 z__2.r = bb[i__5].r * ab[i__6].r - bb[i__5].i * ab[i__6]
987 .i, z__2.i = bb[i__5].r * ab[i__6].i + bb[i__5].i
989 z__1.r = ab[i__1].r - z__2.r, z__1.i = ab[i__1].i -
991 ab[i__4].r = z__1.r, ab[i__4].i = z__1.i;
999 /* post-multiply X by inv(S(i)) */
1003 zdscal_(&i__3, &d__1, &x[m + 1 + i__ * x_dim1], &c__1);
1006 z__1.r = -1., z__1.i = 0.;
1007 zgerc_(&i__3, &kbt, &z__1, &x[m + 1 + i__ * x_dim1], &
1008 c__1, &bb[kb1 - kbt + i__ * bb_dim1], &c__1, &x[m
1009 + 1 + (i__ - kbt) * x_dim1], ldx);
1013 /* store a(i,i1) in RA1 for use in next loop over K */
1015 i__3 = i__ - i1 + ka1 + i1 * ab_dim1;
1016 ra1.r = ab[i__3].r, ra1.i = ab[i__3].i;
1019 /* Generate and apply vectors of rotations to chase all the */
1020 /* existing bulges KA positions down toward the bottom of the */
1024 for (k = 1; k <= i__3; ++k) {
1027 /* Determine the rotations which would annihilate the bulge */
1028 /* which has in theory just been created */
1030 if (i__ - k + *ka < *n && i__ - k > 1) {
1032 /* generate rotation to annihilate a(i,i-k+ka+1) */
1034 zlartg_(&ab[k + 1 + (i__ - k + *ka) * ab_dim1], &ra1, &
1035 rwork[i__ - k + *ka - m], &work[i__ - k + *ka - m]
1038 /* create nonzero element a(i-k,i-k+ka+1) outside the */
1039 /* band and store it in WORK(i-k) */
1041 i__2 = kb1 - k + i__ * bb_dim1;
1042 z__2.r = -bb[i__2].r, z__2.i = -bb[i__2].i;
1043 z__1.r = z__2.r * ra1.r - z__2.i * ra1.i, z__1.i = z__2.r
1044 * ra1.i + z__2.i * ra1.r;
1045 t.r = z__1.r, t.i = z__1.i;
1047 i__4 = i__ - k + *ka - m;
1048 z__2.r = rwork[i__4] * t.r, z__2.i = rwork[i__4] * t.i;
1049 d_cnjg(&z__4, &work[i__ - k + *ka - m]);
1050 i__1 = (i__ - k + *ka) * ab_dim1 + 1;
1051 z__3.r = z__4.r * ab[i__1].r - z__4.i * ab[i__1].i,
1052 z__3.i = z__4.r * ab[i__1].i + z__4.i * ab[i__1]
1054 z__1.r = z__2.r - z__3.r, z__1.i = z__2.i - z__3.i;
1055 work[i__2].r = z__1.r, work[i__2].i = z__1.i;
1056 i__2 = (i__ - k + *ka) * ab_dim1 + 1;
1057 i__4 = i__ - k + *ka - m;
1058 z__2.r = work[i__4].r * t.r - work[i__4].i * t.i, z__2.i =
1059 work[i__4].r * t.i + work[i__4].i * t.r;
1060 i__1 = i__ - k + *ka - m;
1061 i__5 = (i__ - k + *ka) * ab_dim1 + 1;
1062 z__3.r = rwork[i__1] * ab[i__5].r, z__3.i = rwork[i__1] *
1064 z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
1065 ab[i__2].r = z__1.r, ab[i__2].i = z__1.i;
1066 ra1.r = ra.r, ra1.i = ra.i;
1070 i__2 = 1, i__4 = k - i0 + 2;
1071 j2 = i__ - k - 1 + f2cmax(i__2,i__4) * ka1;
1072 nr = (*n - j2 + *ka) / ka1;
1073 j1 = j2 + (nr - 1) * ka1;
1076 i__2 = j2, i__4 = i__ + (*ka << 1) - k + 1;
1077 j2t = f2cmax(i__2,i__4);
1081 nrt = (*n - j2t + *ka) / ka1;
1084 for (j = j2t; i__4 < 0 ? j >= i__2 : j <= i__2; j += i__4) {
1086 /* create nonzero element a(j-ka,j+1) outside the band */
1087 /* and store it in WORK(j-m) */
1091 i__6 = (j + 1) * ab_dim1 + 1;
1092 z__1.r = work[i__5].r * ab[i__6].r - work[i__5].i * ab[i__6]
1093 .i, z__1.i = work[i__5].r * ab[i__6].i + work[i__5].i
1095 work[i__1].r = z__1.r, work[i__1].i = z__1.i;
1096 i__1 = (j + 1) * ab_dim1 + 1;
1098 i__6 = (j + 1) * ab_dim1 + 1;
1099 z__1.r = rwork[i__5] * ab[i__6].r, z__1.i = rwork[i__5] * ab[
1101 ab[i__1].r = z__1.r, ab[i__1].i = z__1.i;
1105 /* generate rotations in 1st set to annihilate elements which */
1106 /* have been created outside the band */
1109 zlargv_(&nrt, &ab[j2t * ab_dim1 + 1], &inca, &work[j2t - m], &
1110 ka1, &rwork[j2t - m], &ka1);
1114 /* apply rotations in 1st set from the right */
1117 for (l = 1; l <= i__4; ++l) {
1118 zlartv_(&nr, &ab[ka1 - l + j2 * ab_dim1], &inca, &ab[*ka
1119 - l + (j2 + 1) * ab_dim1], &inca, &rwork[j2 - m],
1120 &work[j2 - m], &ka1);
1124 /* apply rotations in 1st set from both sides to diagonal */
1127 zlar2v_(&nr, &ab[ka1 + j2 * ab_dim1], &ab[ka1 + (j2 + 1) *
1128 ab_dim1], &ab[*ka + (j2 + 1) * ab_dim1], &inca, &
1129 rwork[j2 - m], &work[j2 - m], &ka1);
1131 zlacgv_(&nr, &work[j2 - m], &ka1);
1134 /* start applying rotations in 1st set from the left */
1137 for (l = *ka - 1; l >= i__4; --l) {
1138 nrt = (*n - j2 + l) / ka1;
1140 zlartv_(&nrt, &ab[l + (j2 + ka1 - l) * ab_dim1], &inca, &
1141 ab[l + 1 + (j2 + ka1 - l) * ab_dim1], &inca, &
1142 rwork[j2 - m], &work[j2 - m], &ka1);
1149 /* post-multiply X by product of rotations in 1st set */
1153 for (j = j2; i__2 < 0 ? j >= i__4 : j <= i__4; j += i__2) {
1155 d_cnjg(&z__1, &work[j - m]);
1156 zrot_(&i__1, &x[m + 1 + j * x_dim1], &c__1, &x[m + 1 + (j
1157 + 1) * x_dim1], &c__1, &rwork[j - m], &z__1);
1165 if (i2 <= *n && kbt > 0) {
1167 /* create nonzero element a(i-kbt,i-kbt+ka+1) outside the */
1168 /* band and store it in WORK(i-kbt) */
1171 i__2 = kb1 - kbt + i__ * bb_dim1;
1172 z__2.r = -bb[i__2].r, z__2.i = -bb[i__2].i;
1173 z__1.r = z__2.r * ra1.r - z__2.i * ra1.i, z__1.i = z__2.r *
1174 ra1.i + z__2.i * ra1.r;
1175 work[i__3].r = z__1.r, work[i__3].i = z__1.i;
1179 for (k = *kb; k >= 1; --k) {
1182 i__3 = 2, i__2 = k - i0 + 1;
1183 j2 = i__ - k - 1 + f2cmax(i__3,i__2) * ka1;
1186 i__3 = 1, i__2 = k - i0 + 1;
1187 j2 = i__ - k - 1 + f2cmax(i__3,i__2) * ka1;
1190 /* finish applying rotations in 2nd set from the left */
1192 for (l = *kb - k; l >= 1; --l) {
1193 nrt = (*n - j2 + *ka + l) / ka1;
1195 zlartv_(&nrt, &ab[l + (j2 - l + 1) * ab_dim1], &inca, &ab[
1196 l + 1 + (j2 - l + 1) * ab_dim1], &inca, &rwork[j2
1197 - *ka], &work[j2 - *ka], &ka1);
1201 nr = (*n - j2 + *ka) / ka1;
1202 j1 = j2 + (nr - 1) * ka1;
1205 for (j = j1; i__2 < 0 ? j >= i__3 : j <= i__3; j += i__2) {
1208 work[i__4].r = work[i__1].r, work[i__4].i = work[i__1].i;
1209 rwork[j] = rwork[j - *ka];
1214 for (j = j2; i__3 < 0 ? j >= i__2 : j <= i__2; j += i__3) {
1216 /* create nonzero element a(j-ka,j+1) outside the band */
1217 /* and store it in WORK(j) */
1221 i__5 = (j + 1) * ab_dim1 + 1;
1222 z__1.r = work[i__1].r * ab[i__5].r - work[i__1].i * ab[i__5]
1223 .i, z__1.i = work[i__1].r * ab[i__5].i + work[i__1].i
1225 work[i__4].r = z__1.r, work[i__4].i = z__1.i;
1226 i__4 = (j + 1) * ab_dim1 + 1;
1228 i__5 = (j + 1) * ab_dim1 + 1;
1229 z__1.r = rwork[i__1] * ab[i__5].r, z__1.i = rwork[i__1] * ab[
1231 ab[i__4].r = z__1.r, ab[i__4].i = z__1.i;
1235 if (i__ - k < *n - *ka && k <= kbt) {
1236 i__3 = i__ - k + *ka;
1238 work[i__3].r = work[i__2].r, work[i__3].i = work[i__2].i;
1244 for (k = *kb; k >= 1; --k) {
1246 i__3 = 1, i__2 = k - i0 + 1;
1247 j2 = i__ - k - 1 + f2cmax(i__3,i__2) * ka1;
1248 nr = (*n - j2 + *ka) / ka1;
1249 j1 = j2 + (nr - 1) * ka1;
1252 /* generate rotations in 2nd set to annihilate elements */
1253 /* which have been created outside the band */
1255 zlargv_(&nr, &ab[j2 * ab_dim1 + 1], &inca, &work[j2], &ka1, &
1258 /* apply rotations in 2nd set from the right */
1261 for (l = 1; l <= i__3; ++l) {
1262 zlartv_(&nr, &ab[ka1 - l + j2 * ab_dim1], &inca, &ab[*ka
1263 - l + (j2 + 1) * ab_dim1], &inca, &rwork[j2], &
1268 /* apply rotations in 2nd set from both sides to diagonal */
1271 zlar2v_(&nr, &ab[ka1 + j2 * ab_dim1], &ab[ka1 + (j2 + 1) *
1272 ab_dim1], &ab[*ka + (j2 + 1) * ab_dim1], &inca, &
1273 rwork[j2], &work[j2], &ka1);
1275 zlacgv_(&nr, &work[j2], &ka1);
1278 /* start applying rotations in 2nd set from the left */
1281 for (l = *ka - 1; l >= i__3; --l) {
1282 nrt = (*n - j2 + l) / ka1;
1284 zlartv_(&nrt, &ab[l + (j2 + ka1 - l) * ab_dim1], &inca, &
1285 ab[l + 1 + (j2 + ka1 - l) * ab_dim1], &inca, &
1286 rwork[j2], &work[j2], &ka1);
1293 /* post-multiply X by product of rotations in 2nd set */
1297 for (j = j2; i__2 < 0 ? j >= i__3 : j <= i__3; j += i__2) {
1299 d_cnjg(&z__1, &work[j]);
1300 zrot_(&i__4, &x[m + 1 + j * x_dim1], &c__1, &x[m + 1 + (j
1301 + 1) * x_dim1], &c__1, &rwork[j], &z__1);
1309 for (k = 1; k <= i__2; ++k) {
1311 i__3 = 1, i__4 = k - i0 + 2;
1312 j2 = i__ - k - 1 + f2cmax(i__3,i__4) * ka1;
1314 /* finish applying rotations in 1st set from the left */
1316 for (l = *kb - k; l >= 1; --l) {
1317 nrt = (*n - j2 + l) / ka1;
1319 zlartv_(&nrt, &ab[l + (j2 + ka1 - l) * ab_dim1], &inca, &
1320 ab[l + 1 + (j2 + ka1 - l) * ab_dim1], &inca, &
1321 rwork[j2 - m], &work[j2 - m], &ka1);
1330 for (j = *n - 1; j >= i__2; --j) {
1331 rwork[j - m] = rwork[j - *ka - m];
1334 work[i__3].r = work[i__4].r, work[i__3].i = work[i__4].i;
1341 /* Transform A, working with the lower triangle */
1345 /* Form inv(S(i))**H * A * inv(S(i)) */
1347 i__2 = i__ * bb_dim1 + 1;
1349 i__2 = i__ * ab_dim1 + 1;
1350 i__3 = i__ * ab_dim1 + 1;
1351 d__1 = ab[i__3].r / bii / bii;
1352 ab[i__2].r = d__1, ab[i__2].i = 0.;
1354 for (j = i__ + 1; j <= i__2; ++j) {
1355 i__3 = j - i__ + 1 + i__ * ab_dim1;
1356 i__4 = j - i__ + 1 + i__ * ab_dim1;
1357 z__1.r = ab[i__4].r / bii, z__1.i = ab[i__4].i / bii;
1358 ab[i__3].r = z__1.r, ab[i__3].i = z__1.i;
1362 i__2 = 1, i__3 = i__ - *ka;
1364 for (j = f2cmax(i__2,i__3); j <= i__4; ++j) {
1365 i__2 = i__ - j + 1 + j * ab_dim1;
1366 i__3 = i__ - j + 1 + j * ab_dim1;
1367 z__1.r = ab[i__3].r / bii, z__1.i = ab[i__3].i / bii;
1368 ab[i__2].r = z__1.r, ab[i__2].i = z__1.i;
1372 for (k = i__ - kbt; k <= i__4; ++k) {
1374 for (j = i__ - kbt; j <= i__2; ++j) {
1375 i__3 = k - j + 1 + j * ab_dim1;
1376 i__1 = k - j + 1 + j * ab_dim1;
1377 i__5 = i__ - j + 1 + j * bb_dim1;
1378 d_cnjg(&z__5, &ab[i__ - k + 1 + k * ab_dim1]);
1379 z__4.r = bb[i__5].r * z__5.r - bb[i__5].i * z__5.i,
1380 z__4.i = bb[i__5].r * z__5.i + bb[i__5].i *
1382 z__3.r = ab[i__1].r - z__4.r, z__3.i = ab[i__1].i -
1384 d_cnjg(&z__7, &bb[i__ - k + 1 + k * bb_dim1]);
1385 i__6 = i__ - j + 1 + j * ab_dim1;
1386 z__6.r = z__7.r * ab[i__6].r - z__7.i * ab[i__6].i,
1387 z__6.i = z__7.r * ab[i__6].i + z__7.i * ab[i__6]
1389 z__2.r = z__3.r - z__6.r, z__2.i = z__3.i - z__6.i;
1390 i__7 = i__ * ab_dim1 + 1;
1392 i__8 = i__ - j + 1 + j * bb_dim1;
1393 z__9.r = d__1 * bb[i__8].r, z__9.i = d__1 * bb[i__8].i;
1394 d_cnjg(&z__10, &bb[i__ - k + 1 + k * bb_dim1]);
1395 z__8.r = z__9.r * z__10.r - z__9.i * z__10.i, z__8.i =
1396 z__9.r * z__10.i + z__9.i * z__10.r;
1397 z__1.r = z__2.r + z__8.r, z__1.i = z__2.i + z__8.i;
1398 ab[i__3].r = z__1.r, ab[i__3].i = z__1.i;
1402 i__2 = 1, i__3 = i__ - *ka;
1403 i__1 = i__ - kbt - 1;
1404 for (j = f2cmax(i__2,i__3); j <= i__1; ++j) {
1405 i__2 = k - j + 1 + j * ab_dim1;
1406 i__3 = k - j + 1 + j * ab_dim1;
1407 d_cnjg(&z__3, &bb[i__ - k + 1 + k * bb_dim1]);
1408 i__5 = i__ - j + 1 + j * ab_dim1;
1409 z__2.r = z__3.r * ab[i__5].r - z__3.i * ab[i__5].i,
1410 z__2.i = z__3.r * ab[i__5].i + z__3.i * ab[i__5]
1412 z__1.r = ab[i__3].r - z__2.r, z__1.i = ab[i__3].i -
1414 ab[i__2].r = z__1.r, ab[i__2].i = z__1.i;
1420 for (j = i__; j <= i__4; ++j) {
1422 i__1 = j - *ka, i__2 = i__ - kbt;
1424 for (k = f2cmax(i__1,i__2); k <= i__3; ++k) {
1425 i__1 = j - k + 1 + k * ab_dim1;
1426 i__2 = j - k + 1 + k * ab_dim1;
1427 i__5 = i__ - k + 1 + k * bb_dim1;
1428 i__6 = j - i__ + 1 + i__ * ab_dim1;
1429 z__2.r = bb[i__5].r * ab[i__6].r - bb[i__5].i * ab[i__6]
1430 .i, z__2.i = bb[i__5].r * ab[i__6].i + bb[i__5].i
1432 z__1.r = ab[i__2].r - z__2.r, z__1.i = ab[i__2].i -
1434 ab[i__1].r = z__1.r, ab[i__1].i = z__1.i;
1442 /* post-multiply X by inv(S(i)) */
1446 zdscal_(&i__4, &d__1, &x[m + 1 + i__ * x_dim1], &c__1);
1449 z__1.r = -1., z__1.i = 0.;
1451 zgeru_(&i__4, &kbt, &z__1, &x[m + 1 + i__ * x_dim1], &
1452 c__1, &bb[kbt + 1 + (i__ - kbt) * bb_dim1], &i__3,
1453 &x[m + 1 + (i__ - kbt) * x_dim1], ldx);
1457 /* store a(i1,i) in RA1 for use in next loop over K */
1459 i__4 = i1 - i__ + 1 + i__ * ab_dim1;
1460 ra1.r = ab[i__4].r, ra1.i = ab[i__4].i;
1463 /* Generate and apply vectors of rotations to chase all the */
1464 /* existing bulges KA positions down toward the bottom of the */
1468 for (k = 1; k <= i__4; ++k) {
1471 /* Determine the rotations which would annihilate the bulge */
1472 /* which has in theory just been created */
1474 if (i__ - k + *ka < *n && i__ - k > 1) {
1476 /* generate rotation to annihilate a(i-k+ka+1,i) */
1478 zlartg_(&ab[ka1 - k + i__ * ab_dim1], &ra1, &rwork[i__ -
1479 k + *ka - m], &work[i__ - k + *ka - m], &ra);
1481 /* create nonzero element a(i-k+ka+1,i-k) outside the */
1482 /* band and store it in WORK(i-k) */
1484 i__3 = k + 1 + (i__ - k) * bb_dim1;
1485 z__2.r = -bb[i__3].r, z__2.i = -bb[i__3].i;
1486 z__1.r = z__2.r * ra1.r - z__2.i * ra1.i, z__1.i = z__2.r
1487 * ra1.i + z__2.i * ra1.r;
1488 t.r = z__1.r, t.i = z__1.i;
1490 i__1 = i__ - k + *ka - m;
1491 z__2.r = rwork[i__1] * t.r, z__2.i = rwork[i__1] * t.i;
1492 d_cnjg(&z__4, &work[i__ - k + *ka - m]);
1493 i__2 = ka1 + (i__ - k) * ab_dim1;
1494 z__3.r = z__4.r * ab[i__2].r - z__4.i * ab[i__2].i,
1495 z__3.i = z__4.r * ab[i__2].i + z__4.i * ab[i__2]
1497 z__1.r = z__2.r - z__3.r, z__1.i = z__2.i - z__3.i;
1498 work[i__3].r = z__1.r, work[i__3].i = z__1.i;
1499 i__3 = ka1 + (i__ - k) * ab_dim1;
1500 i__1 = i__ - k + *ka - m;
1501 z__2.r = work[i__1].r * t.r - work[i__1].i * t.i, z__2.i =
1502 work[i__1].r * t.i + work[i__1].i * t.r;
1503 i__2 = i__ - k + *ka - m;
1504 i__5 = ka1 + (i__ - k) * ab_dim1;
1505 z__3.r = rwork[i__2] * ab[i__5].r, z__3.i = rwork[i__2] *
1507 z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
1508 ab[i__3].r = z__1.r, ab[i__3].i = z__1.i;
1509 ra1.r = ra.r, ra1.i = ra.i;
1513 i__3 = 1, i__1 = k - i0 + 2;
1514 j2 = i__ - k - 1 + f2cmax(i__3,i__1) * ka1;
1515 nr = (*n - j2 + *ka) / ka1;
1516 j1 = j2 + (nr - 1) * ka1;
1519 i__3 = j2, i__1 = i__ + (*ka << 1) - k + 1;
1520 j2t = f2cmax(i__3,i__1);
1524 nrt = (*n - j2t + *ka) / ka1;
1527 for (j = j2t; i__1 < 0 ? j >= i__3 : j <= i__3; j += i__1) {
1529 /* create nonzero element a(j+1,j-ka) outside the band */
1530 /* and store it in WORK(j-m) */
1534 i__6 = ka1 + (j - *ka + 1) * ab_dim1;
1535 z__1.r = work[i__5].r * ab[i__6].r - work[i__5].i * ab[i__6]
1536 .i, z__1.i = work[i__5].r * ab[i__6].i + work[i__5].i
1538 work[i__2].r = z__1.r, work[i__2].i = z__1.i;
1539 i__2 = ka1 + (j - *ka + 1) * ab_dim1;
1541 i__6 = ka1 + (j - *ka + 1) * ab_dim1;
1542 z__1.r = rwork[i__5] * ab[i__6].r, z__1.i = rwork[i__5] * ab[
1544 ab[i__2].r = z__1.r, ab[i__2].i = z__1.i;
1548 /* generate rotations in 1st set to annihilate elements which */
1549 /* have been created outside the band */
1552 zlargv_(&nrt, &ab[ka1 + (j2t - *ka) * ab_dim1], &inca, &work[
1553 j2t - m], &ka1, &rwork[j2t - m], &ka1);
1557 /* apply rotations in 1st set from the left */
1560 for (l = 1; l <= i__1; ++l) {
1561 zlartv_(&nr, &ab[l + 1 + (j2 - l) * ab_dim1], &inca, &ab[
1562 l + 2 + (j2 - l) * ab_dim1], &inca, &rwork[j2 - m]
1563 , &work[j2 - m], &ka1);
1567 /* apply rotations in 1st set from both sides to diagonal */
1570 zlar2v_(&nr, &ab[j2 * ab_dim1 + 1], &ab[(j2 + 1) * ab_dim1 +
1571 1], &ab[j2 * ab_dim1 + 2], &inca, &rwork[j2 - m], &
1572 work[j2 - m], &ka1);
1574 zlacgv_(&nr, &work[j2 - m], &ka1);
1577 /* start applying rotations in 1st set from the right */
1580 for (l = *ka - 1; l >= i__1; --l) {
1581 nrt = (*n - j2 + l) / ka1;
1583 zlartv_(&nrt, &ab[ka1 - l + 1 + j2 * ab_dim1], &inca, &ab[
1584 ka1 - l + (j2 + 1) * ab_dim1], &inca, &rwork[j2 -
1585 m], &work[j2 - m], &ka1);
1592 /* post-multiply X by product of rotations in 1st set */
1596 for (j = j2; i__3 < 0 ? j >= i__1 : j <= i__1; j += i__3) {
1598 zrot_(&i__2, &x[m + 1 + j * x_dim1], &c__1, &x[m + 1 + (j
1599 + 1) * x_dim1], &c__1, &rwork[j - m], &work[j - m]
1608 if (i2 <= *n && kbt > 0) {
1610 /* create nonzero element a(i-kbt+ka+1,i-kbt) outside the */
1611 /* band and store it in WORK(i-kbt) */
1614 i__3 = kbt + 1 + (i__ - kbt) * bb_dim1;
1615 z__2.r = -bb[i__3].r, z__2.i = -bb[i__3].i;
1616 z__1.r = z__2.r * ra1.r - z__2.i * ra1.i, z__1.i = z__2.r *
1617 ra1.i + z__2.i * ra1.r;
1618 work[i__4].r = z__1.r, work[i__4].i = z__1.i;
1622 for (k = *kb; k >= 1; --k) {
1625 i__4 = 2, i__3 = k - i0 + 1;
1626 j2 = i__ - k - 1 + f2cmax(i__4,i__3) * ka1;
1629 i__4 = 1, i__3 = k - i0 + 1;
1630 j2 = i__ - k - 1 + f2cmax(i__4,i__3) * ka1;
1633 /* finish applying rotations in 2nd set from the right */
1635 for (l = *kb - k; l >= 1; --l) {
1636 nrt = (*n - j2 + *ka + l) / ka1;
1638 zlartv_(&nrt, &ab[ka1 - l + 1 + (j2 - *ka) * ab_dim1], &
1639 inca, &ab[ka1 - l + (j2 - *ka + 1) * ab_dim1], &
1640 inca, &rwork[j2 - *ka], &work[j2 - *ka], &ka1);
1644 nr = (*n - j2 + *ka) / ka1;
1645 j1 = j2 + (nr - 1) * ka1;
1648 for (j = j1; i__3 < 0 ? j >= i__4 : j <= i__4; j += i__3) {
1651 work[i__1].r = work[i__2].r, work[i__1].i = work[i__2].i;
1652 rwork[j] = rwork[j - *ka];
1657 for (j = j2; i__4 < 0 ? j >= i__3 : j <= i__3; j += i__4) {
1659 /* create nonzero element a(j+1,j-ka) outside the band */
1660 /* and store it in WORK(j) */
1664 i__5 = ka1 + (j - *ka + 1) * ab_dim1;
1665 z__1.r = work[i__2].r * ab[i__5].r - work[i__2].i * ab[i__5]
1666 .i, z__1.i = work[i__2].r * ab[i__5].i + work[i__2].i
1668 work[i__1].r = z__1.r, work[i__1].i = z__1.i;
1669 i__1 = ka1 + (j - *ka + 1) * ab_dim1;
1671 i__5 = ka1 + (j - *ka + 1) * ab_dim1;
1672 z__1.r = rwork[i__2] * ab[i__5].r, z__1.i = rwork[i__2] * ab[
1674 ab[i__1].r = z__1.r, ab[i__1].i = z__1.i;
1678 if (i__ - k < *n - *ka && k <= kbt) {
1679 i__4 = i__ - k + *ka;
1681 work[i__4].r = work[i__3].r, work[i__4].i = work[i__3].i;
1687 for (k = *kb; k >= 1; --k) {
1689 i__4 = 1, i__3 = k - i0 + 1;
1690 j2 = i__ - k - 1 + f2cmax(i__4,i__3) * ka1;
1691 nr = (*n - j2 + *ka) / ka1;
1692 j1 = j2 + (nr - 1) * ka1;
1695 /* generate rotations in 2nd set to annihilate elements */
1696 /* which have been created outside the band */
1698 zlargv_(&nr, &ab[ka1 + (j2 - *ka) * ab_dim1], &inca, &work[j2]
1699 , &ka1, &rwork[j2], &ka1);
1701 /* apply rotations in 2nd set from the left */
1704 for (l = 1; l <= i__4; ++l) {
1705 zlartv_(&nr, &ab[l + 1 + (j2 - l) * ab_dim1], &inca, &ab[
1706 l + 2 + (j2 - l) * ab_dim1], &inca, &rwork[j2], &
1711 /* apply rotations in 2nd set from both sides to diagonal */
1714 zlar2v_(&nr, &ab[j2 * ab_dim1 + 1], &ab[(j2 + 1) * ab_dim1 +
1715 1], &ab[j2 * ab_dim1 + 2], &inca, &rwork[j2], &work[
1718 zlacgv_(&nr, &work[j2], &ka1);
1721 /* start applying rotations in 2nd set from the right */
1724 for (l = *ka - 1; l >= i__4; --l) {
1725 nrt = (*n - j2 + l) / ka1;
1727 zlartv_(&nrt, &ab[ka1 - l + 1 + j2 * ab_dim1], &inca, &ab[
1728 ka1 - l + (j2 + 1) * ab_dim1], &inca, &rwork[j2],
1736 /* post-multiply X by product of rotations in 2nd set */
1740 for (j = j2; i__3 < 0 ? j >= i__4 : j <= i__4; j += i__3) {
1742 zrot_(&i__1, &x[m + 1 + j * x_dim1], &c__1, &x[m + 1 + (j
1743 + 1) * x_dim1], &c__1, &rwork[j], &work[j]);
1751 for (k = 1; k <= i__3; ++k) {
1753 i__4 = 1, i__1 = k - i0 + 2;
1754 j2 = i__ - k - 1 + f2cmax(i__4,i__1) * ka1;
1756 /* finish applying rotations in 1st set from the right */
1758 for (l = *kb - k; l >= 1; --l) {
1759 nrt = (*n - j2 + l) / ka1;
1761 zlartv_(&nrt, &ab[ka1 - l + 1 + j2 * ab_dim1], &inca, &ab[
1762 ka1 - l + (j2 + 1) * ab_dim1], &inca, &rwork[j2 -
1763 m], &work[j2 - m], &ka1);
1772 for (j = *n - 1; j >= i__3; --j) {
1773 rwork[j - m] = rwork[j - *ka - m];
1776 work[i__4].r = work[i__1].r, work[i__4].i = work[i__1].i;
1787 /* **************************** Phase 2 ***************************** */
1789 /* The logical structure of this phase is: */
1791 /* UPDATE = .TRUE. */
1793 /* use S(i) to update A and create a new bulge */
1794 /* apply rotations to push all bulges KA positions upward */
1796 /* UPDATE = .FALSE. */
1797 /* DO I = M - KA - 1, 2, -1 */
1798 /* apply rotations to push all bulges KA positions upward */
1801 /* To avoid duplicating code, the two loops are merged. */
1809 i__3 = *kb, i__4 = m - i__;
1810 kbt = f2cmin(i__3,i__4);
1813 i__3 = 1, i__4 = i__ - *ka;
1814 i1 = f2cmax(i__3,i__4);
1815 i2 = i__ + kbt - ka1;
1832 if (i__ < m - kbt) {
1840 /* Transform A, working with the upper triangle */
1844 /* Form inv(S(i))**H * A * inv(S(i)) */
1846 i__3 = kb1 + i__ * bb_dim1;
1848 i__3 = ka1 + i__ * ab_dim1;
1849 i__4 = ka1 + i__ * ab_dim1;
1850 d__1 = ab[i__4].r / bii / bii;
1851 ab[i__3].r = d__1, ab[i__3].i = 0.;
1853 for (j = i1; j <= i__3; ++j) {
1854 i__4 = j - i__ + ka1 + i__ * ab_dim1;
1855 i__1 = j - i__ + ka1 + i__ * ab_dim1;
1856 z__1.r = ab[i__1].r / bii, z__1.i = ab[i__1].i / bii;
1857 ab[i__4].r = z__1.r, ab[i__4].i = z__1.i;
1861 i__4 = *n, i__1 = i__ + *ka;
1862 i__3 = f2cmin(i__4,i__1);
1863 for (j = i__ + 1; j <= i__3; ++j) {
1864 i__4 = i__ - j + ka1 + j * ab_dim1;
1865 i__1 = i__ - j + ka1 + j * ab_dim1;
1866 z__1.r = ab[i__1].r / bii, z__1.i = ab[i__1].i / bii;
1867 ab[i__4].r = z__1.r, ab[i__4].i = z__1.i;
1871 for (k = i__ + 1; k <= i__3; ++k) {
1873 for (j = k; j <= i__4; ++j) {
1874 i__1 = k - j + ka1 + j * ab_dim1;
1875 i__2 = k - j + ka1 + j * ab_dim1;
1876 i__5 = i__ - j + kb1 + j * bb_dim1;
1877 d_cnjg(&z__5, &ab[i__ - k + ka1 + k * ab_dim1]);
1878 z__4.r = bb[i__5].r * z__5.r - bb[i__5].i * z__5.i,
1879 z__4.i = bb[i__5].r * z__5.i + bb[i__5].i *
1881 z__3.r = ab[i__2].r - z__4.r, z__3.i = ab[i__2].i -
1883 d_cnjg(&z__7, &bb[i__ - k + kb1 + k * bb_dim1]);
1884 i__6 = i__ - j + ka1 + j * ab_dim1;
1885 z__6.r = z__7.r * ab[i__6].r - z__7.i * ab[i__6].i,
1886 z__6.i = z__7.r * ab[i__6].i + z__7.i * ab[i__6]
1888 z__2.r = z__3.r - z__6.r, z__2.i = z__3.i - z__6.i;
1889 i__7 = ka1 + i__ * ab_dim1;
1891 i__8 = i__ - j + kb1 + j * bb_dim1;
1892 z__9.r = d__1 * bb[i__8].r, z__9.i = d__1 * bb[i__8].i;
1893 d_cnjg(&z__10, &bb[i__ - k + kb1 + k * bb_dim1]);
1894 z__8.r = z__9.r * z__10.r - z__9.i * z__10.i, z__8.i =
1895 z__9.r * z__10.i + z__9.i * z__10.r;
1896 z__1.r = z__2.r + z__8.r, z__1.i = z__2.i + z__8.i;
1897 ab[i__1].r = z__1.r, ab[i__1].i = z__1.i;
1901 i__1 = *n, i__2 = i__ + *ka;
1902 i__4 = f2cmin(i__1,i__2);
1903 for (j = i__ + kbt + 1; j <= i__4; ++j) {
1904 i__1 = k - j + ka1 + j * ab_dim1;
1905 i__2 = k - j + ka1 + j * ab_dim1;
1906 d_cnjg(&z__3, &bb[i__ - k + kb1 + k * bb_dim1]);
1907 i__5 = i__ - j + ka1 + j * ab_dim1;
1908 z__2.r = z__3.r * ab[i__5].r - z__3.i * ab[i__5].i,
1909 z__2.i = z__3.r * ab[i__5].i + z__3.i * ab[i__5]
1911 z__1.r = ab[i__2].r - z__2.r, z__1.i = ab[i__2].i -
1913 ab[i__1].r = z__1.r, ab[i__1].i = z__1.i;
1919 for (j = i1; j <= i__3; ++j) {
1921 i__1 = j + *ka, i__2 = i__ + kbt;
1922 i__4 = f2cmin(i__1,i__2);
1923 for (k = i__ + 1; k <= i__4; ++k) {
1924 i__1 = j - k + ka1 + k * ab_dim1;
1925 i__2 = j - k + ka1 + k * ab_dim1;
1926 i__5 = i__ - k + kb1 + k * bb_dim1;
1927 i__6 = j - i__ + ka1 + i__ * ab_dim1;
1928 z__2.r = bb[i__5].r * ab[i__6].r - bb[i__5].i * ab[i__6]
1929 .i, z__2.i = bb[i__5].r * ab[i__6].i + bb[i__5].i
1931 z__1.r = ab[i__2].r - z__2.r, z__1.i = ab[i__2].i -
1933 ab[i__1].r = z__1.r, ab[i__1].i = z__1.i;
1941 /* post-multiply X by inv(S(i)) */
1944 zdscal_(&nx, &d__1, &x[i__ * x_dim1 + 1], &c__1);
1946 z__1.r = -1., z__1.i = 0.;
1948 zgeru_(&nx, &kbt, &z__1, &x[i__ * x_dim1 + 1], &c__1, &bb[
1949 *kb + (i__ + 1) * bb_dim1], &i__3, &x[(i__ + 1) *
1954 /* store a(i1,i) in RA1 for use in next loop over K */
1956 i__3 = i1 - i__ + ka1 + i__ * ab_dim1;
1957 ra1.r = ab[i__3].r, ra1.i = ab[i__3].i;
1960 /* Generate and apply vectors of rotations to chase all the */
1961 /* existing bulges KA positions up toward the top of the band */
1964 for (k = 1; k <= i__3; ++k) {
1967 /* Determine the rotations which would annihilate the bulge */
1968 /* which has in theory just been created */
1970 if (i__ + k - ka1 > 0 && i__ + k < m) {
1972 /* generate rotation to annihilate a(i+k-ka-1,i) */
1974 zlartg_(&ab[k + 1 + i__ * ab_dim1], &ra1, &rwork[i__ + k
1975 - *ka], &work[i__ + k - *ka], &ra);
1977 /* create nonzero element a(i+k-ka-1,i+k) outside the */
1978 /* band and store it in WORK(m-kb+i+k) */
1980 i__4 = kb1 - k + (i__ + k) * bb_dim1;
1981 z__2.r = -bb[i__4].r, z__2.i = -bb[i__4].i;
1982 z__1.r = z__2.r * ra1.r - z__2.i * ra1.i, z__1.i = z__2.r
1983 * ra1.i + z__2.i * ra1.r;
1984 t.r = z__1.r, t.i = z__1.i;
1985 i__4 = m - *kb + i__ + k;
1986 i__1 = i__ + k - *ka;
1987 z__2.r = rwork[i__1] * t.r, z__2.i = rwork[i__1] * t.i;
1988 d_cnjg(&z__4, &work[i__ + k - *ka]);
1989 i__2 = (i__ + k) * ab_dim1 + 1;
1990 z__3.r = z__4.r * ab[i__2].r - z__4.i * ab[i__2].i,
1991 z__3.i = z__4.r * ab[i__2].i + z__4.i * ab[i__2]
1993 z__1.r = z__2.r - z__3.r, z__1.i = z__2.i - z__3.i;
1994 work[i__4].r = z__1.r, work[i__4].i = z__1.i;
1995 i__4 = (i__ + k) * ab_dim1 + 1;
1996 i__1 = i__ + k - *ka;
1997 z__2.r = work[i__1].r * t.r - work[i__1].i * t.i, z__2.i =
1998 work[i__1].r * t.i + work[i__1].i * t.r;
1999 i__2 = i__ + k - *ka;
2000 i__5 = (i__ + k) * ab_dim1 + 1;
2001 z__3.r = rwork[i__2] * ab[i__5].r, z__3.i = rwork[i__2] *
2003 z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
2004 ab[i__4].r = z__1.r, ab[i__4].i = z__1.i;
2005 ra1.r = ra.r, ra1.i = ra.i;
2009 i__4 = 1, i__1 = k + i0 - m + 1;
2010 j2 = i__ + k + 1 - f2cmax(i__4,i__1) * ka1;
2011 nr = (j2 + *ka - 1) / ka1;
2012 j1 = j2 - (nr - 1) * ka1;
2015 i__4 = j2, i__1 = i__ - (*ka << 1) + k - 1;
2016 j2t = f2cmin(i__4,i__1);
2020 nrt = (j2t + *ka - 1) / ka1;
2023 for (j = j1; i__1 < 0 ? j >= i__4 : j <= i__4; j += i__1) {
2025 /* create nonzero element a(j-1,j+ka) outside the band */
2026 /* and store it in WORK(j) */
2030 i__6 = (j + *ka - 1) * ab_dim1 + 1;
2031 z__1.r = work[i__5].r * ab[i__6].r - work[i__5].i * ab[i__6]
2032 .i, z__1.i = work[i__5].r * ab[i__6].i + work[i__5].i
2034 work[i__2].r = z__1.r, work[i__2].i = z__1.i;
2035 i__2 = (j + *ka - 1) * ab_dim1 + 1;
2037 i__6 = (j + *ka - 1) * ab_dim1 + 1;
2038 z__1.r = rwork[i__5] * ab[i__6].r, z__1.i = rwork[i__5] * ab[
2040 ab[i__2].r = z__1.r, ab[i__2].i = z__1.i;
2044 /* generate rotations in 1st set to annihilate elements which */
2045 /* have been created outside the band */
2048 zlargv_(&nrt, &ab[(j1 + *ka) * ab_dim1 + 1], &inca, &work[j1],
2049 &ka1, &rwork[j1], &ka1);
2053 /* apply rotations in 1st set from the left */
2056 for (l = 1; l <= i__1; ++l) {
2057 zlartv_(&nr, &ab[ka1 - l + (j1 + l) * ab_dim1], &inca, &
2058 ab[*ka - l + (j1 + l) * ab_dim1], &inca, &rwork[
2059 j1], &work[j1], &ka1);
2063 /* apply rotations in 1st set from both sides to diagonal */
2066 zlar2v_(&nr, &ab[ka1 + j1 * ab_dim1], &ab[ka1 + (j1 - 1) *
2067 ab_dim1], &ab[*ka + j1 * ab_dim1], &inca, &rwork[j1],
2070 zlacgv_(&nr, &work[j1], &ka1);
2073 /* start applying rotations in 1st set from the right */
2076 for (l = *ka - 1; l >= i__1; --l) {
2077 nrt = (j2 + l - 1) / ka1;
2078 j1t = j2 - (nrt - 1) * ka1;
2080 zlartv_(&nrt, &ab[l + j1t * ab_dim1], &inca, &ab[l + 1 + (
2081 j1t - 1) * ab_dim1], &inca, &rwork[j1t], &work[
2089 /* post-multiply X by product of rotations in 1st set */
2093 for (j = j1; i__4 < 0 ? j >= i__1 : j <= i__1; j += i__4) {
2094 zrot_(&nx, &x[j * x_dim1 + 1], &c__1, &x[(j - 1) * x_dim1
2095 + 1], &c__1, &rwork[j], &work[j]);
2103 if (i2 > 0 && kbt > 0) {
2105 /* create nonzero element a(i+kbt-ka-1,i+kbt) outside the */
2106 /* band and store it in WORK(m-kb+i+kbt) */
2108 i__3 = m - *kb + i__ + kbt;
2109 i__4 = kb1 - kbt + (i__ + kbt) * bb_dim1;
2110 z__2.r = -bb[i__4].r, z__2.i = -bb[i__4].i;
2111 z__1.r = z__2.r * ra1.r - z__2.i * ra1.i, z__1.i = z__2.r *
2112 ra1.i + z__2.i * ra1.r;
2113 work[i__3].r = z__1.r, work[i__3].i = z__1.i;
2117 for (k = *kb; k >= 1; --k) {
2120 i__3 = 2, i__4 = k + i0 - m;
2121 j2 = i__ + k + 1 - f2cmax(i__3,i__4) * ka1;
2124 i__3 = 1, i__4 = k + i0 - m;
2125 j2 = i__ + k + 1 - f2cmax(i__3,i__4) * ka1;
2128 /* finish applying rotations in 2nd set from the right */
2130 for (l = *kb - k; l >= 1; --l) {
2131 nrt = (j2 + *ka + l - 1) / ka1;
2132 j1t = j2 - (nrt - 1) * ka1;
2134 zlartv_(&nrt, &ab[l + (j1t + *ka) * ab_dim1], &inca, &ab[
2135 l + 1 + (j1t + *ka - 1) * ab_dim1], &inca, &rwork[
2136 m - *kb + j1t + *ka], &work[m - *kb + j1t + *ka],
2141 nr = (j2 + *ka - 1) / ka1;
2142 j1 = j2 - (nr - 1) * ka1;
2145 for (j = j1; i__4 < 0 ? j >= i__3 : j <= i__3; j += i__4) {
2147 i__2 = m - *kb + j + *ka;
2148 work[i__1].r = work[i__2].r, work[i__1].i = work[i__2].i;
2149 rwork[m - *kb + j] = rwork[m - *kb + j + *ka];
2154 for (j = j1; i__3 < 0 ? j >= i__4 : j <= i__4; j += i__3) {
2156 /* create nonzero element a(j-1,j+ka) outside the band */
2157 /* and store it in WORK(m-kb+j) */
2161 i__5 = (j + *ka - 1) * ab_dim1 + 1;
2162 z__1.r = work[i__2].r * ab[i__5].r - work[i__2].i * ab[i__5]
2163 .i, z__1.i = work[i__2].r * ab[i__5].i + work[i__2].i
2165 work[i__1].r = z__1.r, work[i__1].i = z__1.i;
2166 i__1 = (j + *ka - 1) * ab_dim1 + 1;
2168 i__5 = (j + *ka - 1) * ab_dim1 + 1;
2169 z__1.r = rwork[i__2] * ab[i__5].r, z__1.i = rwork[i__2] * ab[
2171 ab[i__1].r = z__1.r, ab[i__1].i = z__1.i;
2175 if (i__ + k > ka1 && k <= kbt) {
2176 i__3 = m - *kb + i__ + k - *ka;
2177 i__4 = m - *kb + i__ + k;
2178 work[i__3].r = work[i__4].r, work[i__3].i = work[i__4].i;
2184 for (k = *kb; k >= 1; --k) {
2186 i__3 = 1, i__4 = k + i0 - m;
2187 j2 = i__ + k + 1 - f2cmax(i__3,i__4) * ka1;
2188 nr = (j2 + *ka - 1) / ka1;
2189 j1 = j2 - (nr - 1) * ka1;
2192 /* generate rotations in 2nd set to annihilate elements */
2193 /* which have been created outside the band */
2195 zlargv_(&nr, &ab[(j1 + *ka) * ab_dim1 + 1], &inca, &work[m - *
2196 kb + j1], &ka1, &rwork[m - *kb + j1], &ka1);
2198 /* apply rotations in 2nd set from the left */
2201 for (l = 1; l <= i__3; ++l) {
2202 zlartv_(&nr, &ab[ka1 - l + (j1 + l) * ab_dim1], &inca, &
2203 ab[*ka - l + (j1 + l) * ab_dim1], &inca, &rwork[m
2204 - *kb + j1], &work[m - *kb + j1], &ka1);
2208 /* apply rotations in 2nd set from both sides to diagonal */
2211 zlar2v_(&nr, &ab[ka1 + j1 * ab_dim1], &ab[ka1 + (j1 - 1) *
2212 ab_dim1], &ab[*ka + j1 * ab_dim1], &inca, &rwork[m - *
2213 kb + j1], &work[m - *kb + j1], &ka1);
2215 zlacgv_(&nr, &work[m - *kb + j1], &ka1);
2218 /* start applying rotations in 2nd set from the right */
2221 for (l = *ka - 1; l >= i__3; --l) {
2222 nrt = (j2 + l - 1) / ka1;
2223 j1t = j2 - (nrt - 1) * ka1;
2225 zlartv_(&nrt, &ab[l + j1t * ab_dim1], &inca, &ab[l + 1 + (
2226 j1t - 1) * ab_dim1], &inca, &rwork[m - *kb + j1t],
2227 &work[m - *kb + j1t], &ka1);
2234 /* post-multiply X by product of rotations in 2nd set */
2238 for (j = j1; i__4 < 0 ? j >= i__3 : j <= i__3; j += i__4) {
2239 zrot_(&nx, &x[j * x_dim1 + 1], &c__1, &x[(j - 1) * x_dim1
2240 + 1], &c__1, &rwork[m - *kb + j], &work[m - *kb +
2249 for (k = 1; k <= i__4; ++k) {
2251 i__3 = 1, i__1 = k + i0 - m + 1;
2252 j2 = i__ + k + 1 - f2cmax(i__3,i__1) * ka1;
2254 /* finish applying rotations in 1st set from the right */
2256 for (l = *kb - k; l >= 1; --l) {
2257 nrt = (j2 + l - 1) / ka1;
2258 j1t = j2 - (nrt - 1) * ka1;
2260 zlartv_(&nrt, &ab[l + j1t * ab_dim1], &inca, &ab[l + 1 + (
2261 j1t - 1) * ab_dim1], &inca, &rwork[j1t], &work[
2271 for (j = 2; j <= i__4; ++j) {
2272 rwork[j] = rwork[j + *ka];
2275 work[i__3].r = work[i__1].r, work[i__3].i = work[i__1].i;
2282 /* Transform A, working with the lower triangle */
2286 /* Form inv(S(i))**H * A * inv(S(i)) */
2288 i__4 = i__ * bb_dim1 + 1;
2290 i__4 = i__ * ab_dim1 + 1;
2291 i__3 = i__ * ab_dim1 + 1;
2292 d__1 = ab[i__3].r / bii / bii;
2293 ab[i__4].r = d__1, ab[i__4].i = 0.;
2295 for (j = i1; j <= i__4; ++j) {
2296 i__3 = i__ - j + 1 + j * ab_dim1;
2297 i__1 = i__ - j + 1 + j * ab_dim1;
2298 z__1.r = ab[i__1].r / bii, z__1.i = ab[i__1].i / bii;
2299 ab[i__3].r = z__1.r, ab[i__3].i = z__1.i;
2303 i__3 = *n, i__1 = i__ + *ka;
2304 i__4 = f2cmin(i__3,i__1);
2305 for (j = i__ + 1; j <= i__4; ++j) {
2306 i__3 = j - i__ + 1 + i__ * ab_dim1;
2307 i__1 = j - i__ + 1 + i__ * ab_dim1;
2308 z__1.r = ab[i__1].r / bii, z__1.i = ab[i__1].i / bii;
2309 ab[i__3].r = z__1.r, ab[i__3].i = z__1.i;
2313 for (k = i__ + 1; k <= i__4; ++k) {
2315 for (j = k; j <= i__3; ++j) {
2316 i__1 = j - k + 1 + k * ab_dim1;
2317 i__2 = j - k + 1 + k * ab_dim1;
2318 i__5 = j - i__ + 1 + i__ * bb_dim1;
2319 d_cnjg(&z__5, &ab[k - i__ + 1 + i__ * ab_dim1]);
2320 z__4.r = bb[i__5].r * z__5.r - bb[i__5].i * z__5.i,
2321 z__4.i = bb[i__5].r * z__5.i + bb[i__5].i *
2323 z__3.r = ab[i__2].r - z__4.r, z__3.i = ab[i__2].i -
2325 d_cnjg(&z__7, &bb[k - i__ + 1 + i__ * bb_dim1]);
2326 i__6 = j - i__ + 1 + i__ * ab_dim1;
2327 z__6.r = z__7.r * ab[i__6].r - z__7.i * ab[i__6].i,
2328 z__6.i = z__7.r * ab[i__6].i + z__7.i * ab[i__6]
2330 z__2.r = z__3.r - z__6.r, z__2.i = z__3.i - z__6.i;
2331 i__7 = i__ * ab_dim1 + 1;
2333 i__8 = j - i__ + 1 + i__ * bb_dim1;
2334 z__9.r = d__1 * bb[i__8].r, z__9.i = d__1 * bb[i__8].i;
2335 d_cnjg(&z__10, &bb[k - i__ + 1 + i__ * bb_dim1]);
2336 z__8.r = z__9.r * z__10.r - z__9.i * z__10.i, z__8.i =
2337 z__9.r * z__10.i + z__9.i * z__10.r;
2338 z__1.r = z__2.r + z__8.r, z__1.i = z__2.i + z__8.i;
2339 ab[i__1].r = z__1.r, ab[i__1].i = z__1.i;
2343 i__1 = *n, i__2 = i__ + *ka;
2344 i__3 = f2cmin(i__1,i__2);
2345 for (j = i__ + kbt + 1; j <= i__3; ++j) {
2346 i__1 = j - k + 1 + k * ab_dim1;
2347 i__2 = j - k + 1 + k * ab_dim1;
2348 d_cnjg(&z__3, &bb[k - i__ + 1 + i__ * bb_dim1]);
2349 i__5 = j - i__ + 1 + i__ * ab_dim1;
2350 z__2.r = z__3.r * ab[i__5].r - z__3.i * ab[i__5].i,
2351 z__2.i = z__3.r * ab[i__5].i + z__3.i * ab[i__5]
2353 z__1.r = ab[i__2].r - z__2.r, z__1.i = ab[i__2].i -
2355 ab[i__1].r = z__1.r, ab[i__1].i = z__1.i;
2361 for (j = i1; j <= i__4; ++j) {
2363 i__1 = j + *ka, i__2 = i__ + kbt;
2364 i__3 = f2cmin(i__1,i__2);
2365 for (k = i__ + 1; k <= i__3; ++k) {
2366 i__1 = k - j + 1 + j * ab_dim1;
2367 i__2 = k - j + 1 + j * ab_dim1;
2368 i__5 = k - i__ + 1 + i__ * bb_dim1;
2369 i__6 = i__ - j + 1 + j * ab_dim1;
2370 z__2.r = bb[i__5].r * ab[i__6].r - bb[i__5].i * ab[i__6]
2371 .i, z__2.i = bb[i__5].r * ab[i__6].i + bb[i__5].i
2373 z__1.r = ab[i__2].r - z__2.r, z__1.i = ab[i__2].i -
2375 ab[i__1].r = z__1.r, ab[i__1].i = z__1.i;
2383 /* post-multiply X by inv(S(i)) */
2386 zdscal_(&nx, &d__1, &x[i__ * x_dim1 + 1], &c__1);
2388 z__1.r = -1., z__1.i = 0.;
2389 zgerc_(&nx, &kbt, &z__1, &x[i__ * x_dim1 + 1], &c__1, &bb[
2390 i__ * bb_dim1 + 2], &c__1, &x[(i__ + 1) * x_dim1
2395 /* store a(i,i1) in RA1 for use in next loop over K */
2397 i__4 = i__ - i1 + 1 + i1 * ab_dim1;
2398 ra1.r = ab[i__4].r, ra1.i = ab[i__4].i;
2401 /* Generate and apply vectors of rotations to chase all the */
2402 /* existing bulges KA positions up toward the top of the band */
2405 for (k = 1; k <= i__4; ++k) {
2408 /* Determine the rotations which would annihilate the bulge */
2409 /* which has in theory just been created */
2411 if (i__ + k - ka1 > 0 && i__ + k < m) {
2413 /* generate rotation to annihilate a(i,i+k-ka-1) */
2415 zlartg_(&ab[ka1 - k + (i__ + k - *ka) * ab_dim1], &ra1, &
2416 rwork[i__ + k - *ka], &work[i__ + k - *ka], &ra);
2418 /* create nonzero element a(i+k,i+k-ka-1) outside the */
2419 /* band and store it in WORK(m-kb+i+k) */
2421 i__3 = k + 1 + i__ * bb_dim1;
2422 z__2.r = -bb[i__3].r, z__2.i = -bb[i__3].i;
2423 z__1.r = z__2.r * ra1.r - z__2.i * ra1.i, z__1.i = z__2.r
2424 * ra1.i + z__2.i * ra1.r;
2425 t.r = z__1.r, t.i = z__1.i;
2426 i__3 = m - *kb + i__ + k;
2427 i__1 = i__ + k - *ka;
2428 z__2.r = rwork[i__1] * t.r, z__2.i = rwork[i__1] * t.i;
2429 d_cnjg(&z__4, &work[i__ + k - *ka]);
2430 i__2 = ka1 + (i__ + k - *ka) * ab_dim1;
2431 z__3.r = z__4.r * ab[i__2].r - z__4.i * ab[i__2].i,
2432 z__3.i = z__4.r * ab[i__2].i + z__4.i * ab[i__2]
2434 z__1.r = z__2.r - z__3.r, z__1.i = z__2.i - z__3.i;
2435 work[i__3].r = z__1.r, work[i__3].i = z__1.i;
2436 i__3 = ka1 + (i__ + k - *ka) * ab_dim1;
2437 i__1 = i__ + k - *ka;
2438 z__2.r = work[i__1].r * t.r - work[i__1].i * t.i, z__2.i =
2439 work[i__1].r * t.i + work[i__1].i * t.r;
2440 i__2 = i__ + k - *ka;
2441 i__5 = ka1 + (i__ + k - *ka) * ab_dim1;
2442 z__3.r = rwork[i__2] * ab[i__5].r, z__3.i = rwork[i__2] *
2444 z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
2445 ab[i__3].r = z__1.r, ab[i__3].i = z__1.i;
2446 ra1.r = ra.r, ra1.i = ra.i;
2450 i__3 = 1, i__1 = k + i0 - m + 1;
2451 j2 = i__ + k + 1 - f2cmax(i__3,i__1) * ka1;
2452 nr = (j2 + *ka - 1) / ka1;
2453 j1 = j2 - (nr - 1) * ka1;
2456 i__3 = j2, i__1 = i__ - (*ka << 1) + k - 1;
2457 j2t = f2cmin(i__3,i__1);
2461 nrt = (j2t + *ka - 1) / ka1;
2464 for (j = j1; i__1 < 0 ? j >= i__3 : j <= i__3; j += i__1) {
2466 /* create nonzero element a(j+ka,j-1) outside the band */
2467 /* and store it in WORK(j) */
2471 i__6 = ka1 + (j - 1) * ab_dim1;
2472 z__1.r = work[i__5].r * ab[i__6].r - work[i__5].i * ab[i__6]
2473 .i, z__1.i = work[i__5].r * ab[i__6].i + work[i__5].i
2475 work[i__2].r = z__1.r, work[i__2].i = z__1.i;
2476 i__2 = ka1 + (j - 1) * ab_dim1;
2478 i__6 = ka1 + (j - 1) * ab_dim1;
2479 z__1.r = rwork[i__5] * ab[i__6].r, z__1.i = rwork[i__5] * ab[
2481 ab[i__2].r = z__1.r, ab[i__2].i = z__1.i;
2485 /* generate rotations in 1st set to annihilate elements which */
2486 /* have been created outside the band */
2489 zlargv_(&nrt, &ab[ka1 + j1 * ab_dim1], &inca, &work[j1], &ka1,
2494 /* apply rotations in 1st set from the right */
2497 for (l = 1; l <= i__1; ++l) {
2498 zlartv_(&nr, &ab[l + 1 + j1 * ab_dim1], &inca, &ab[l + 2
2499 + (j1 - 1) * ab_dim1], &inca, &rwork[j1], &work[
2504 /* apply rotations in 1st set from both sides to diagonal */
2507 zlar2v_(&nr, &ab[j1 * ab_dim1 + 1], &ab[(j1 - 1) * ab_dim1 +
2508 1], &ab[(j1 - 1) * ab_dim1 + 2], &inca, &rwork[j1], &
2511 zlacgv_(&nr, &work[j1], &ka1);
2514 /* start applying rotations in 1st set from the left */
2517 for (l = *ka - 1; l >= i__1; --l) {
2518 nrt = (j2 + l - 1) / ka1;
2519 j1t = j2 - (nrt - 1) * ka1;
2521 zlartv_(&nrt, &ab[ka1 - l + 1 + (j1t - ka1 + l) * ab_dim1]
2522 , &inca, &ab[ka1 - l + (j1t - ka1 + l) * ab_dim1],
2523 &inca, &rwork[j1t], &work[j1t], &ka1);
2530 /* post-multiply X by product of rotations in 1st set */
2534 for (j = j1; i__3 < 0 ? j >= i__1 : j <= i__1; j += i__3) {
2535 d_cnjg(&z__1, &work[j]);
2536 zrot_(&nx, &x[j * x_dim1 + 1], &c__1, &x[(j - 1) * x_dim1
2537 + 1], &c__1, &rwork[j], &z__1);
2545 if (i2 > 0 && kbt > 0) {
2547 /* create nonzero element a(i+kbt,i+kbt-ka-1) outside the */
2548 /* band and store it in WORK(m-kb+i+kbt) */
2550 i__4 = m - *kb + i__ + kbt;
2551 i__3 = kbt + 1 + i__ * bb_dim1;
2552 z__2.r = -bb[i__3].r, z__2.i = -bb[i__3].i;
2553 z__1.r = z__2.r * ra1.r - z__2.i * ra1.i, z__1.i = z__2.r *
2554 ra1.i + z__2.i * ra1.r;
2555 work[i__4].r = z__1.r, work[i__4].i = z__1.i;
2559 for (k = *kb; k >= 1; --k) {
2562 i__4 = 2, i__3 = k + i0 - m;
2563 j2 = i__ + k + 1 - f2cmax(i__4,i__3) * ka1;
2566 i__4 = 1, i__3 = k + i0 - m;
2567 j2 = i__ + k + 1 - f2cmax(i__4,i__3) * ka1;
2570 /* finish applying rotations in 2nd set from the left */
2572 for (l = *kb - k; l >= 1; --l) {
2573 nrt = (j2 + *ka + l - 1) / ka1;
2574 j1t = j2 - (nrt - 1) * ka1;
2576 zlartv_(&nrt, &ab[ka1 - l + 1 + (j1t + l - 1) * ab_dim1],
2577 &inca, &ab[ka1 - l + (j1t + l - 1) * ab_dim1], &
2578 inca, &rwork[m - *kb + j1t + *ka], &work[m - *kb
2579 + j1t + *ka], &ka1);
2583 nr = (j2 + *ka - 1) / ka1;
2584 j1 = j2 - (nr - 1) * ka1;
2587 for (j = j1; i__3 < 0 ? j >= i__4 : j <= i__4; j += i__3) {
2589 i__2 = m - *kb + j + *ka;
2590 work[i__1].r = work[i__2].r, work[i__1].i = work[i__2].i;
2591 rwork[m - *kb + j] = rwork[m - *kb + j + *ka];
2596 for (j = j1; i__4 < 0 ? j >= i__3 : j <= i__3; j += i__4) {
2598 /* create nonzero element a(j+ka,j-1) outside the band */
2599 /* and store it in WORK(m-kb+j) */
2603 i__5 = ka1 + (j - 1) * ab_dim1;
2604 z__1.r = work[i__2].r * ab[i__5].r - work[i__2].i * ab[i__5]
2605 .i, z__1.i = work[i__2].r * ab[i__5].i + work[i__2].i
2607 work[i__1].r = z__1.r, work[i__1].i = z__1.i;
2608 i__1 = ka1 + (j - 1) * ab_dim1;
2610 i__5 = ka1 + (j - 1) * ab_dim1;
2611 z__1.r = rwork[i__2] * ab[i__5].r, z__1.i = rwork[i__2] * ab[
2613 ab[i__1].r = z__1.r, ab[i__1].i = z__1.i;
2617 if (i__ + k > ka1 && k <= kbt) {
2618 i__4 = m - *kb + i__ + k - *ka;
2619 i__3 = m - *kb + i__ + k;
2620 work[i__4].r = work[i__3].r, work[i__4].i = work[i__3].i;
2626 for (k = *kb; k >= 1; --k) {
2628 i__4 = 1, i__3 = k + i0 - m;
2629 j2 = i__ + k + 1 - f2cmax(i__4,i__3) * ka1;
2630 nr = (j2 + *ka - 1) / ka1;
2631 j1 = j2 - (nr - 1) * ka1;
2634 /* generate rotations in 2nd set to annihilate elements */
2635 /* which have been created outside the band */
2637 zlargv_(&nr, &ab[ka1 + j1 * ab_dim1], &inca, &work[m - *kb +
2638 j1], &ka1, &rwork[m - *kb + j1], &ka1);
2640 /* apply rotations in 2nd set from the right */
2643 for (l = 1; l <= i__4; ++l) {
2644 zlartv_(&nr, &ab[l + 1 + j1 * ab_dim1], &inca, &ab[l + 2
2645 + (j1 - 1) * ab_dim1], &inca, &rwork[m - *kb + j1]
2646 , &work[m - *kb + j1], &ka1);
2650 /* apply rotations in 2nd set from both sides to diagonal */
2653 zlar2v_(&nr, &ab[j1 * ab_dim1 + 1], &ab[(j1 - 1) * ab_dim1 +
2654 1], &ab[(j1 - 1) * ab_dim1 + 2], &inca, &rwork[m - *
2655 kb + j1], &work[m - *kb + j1], &ka1);
2657 zlacgv_(&nr, &work[m - *kb + j1], &ka1);
2660 /* start applying rotations in 2nd set from the left */
2663 for (l = *ka - 1; l >= i__4; --l) {
2664 nrt = (j2 + l - 1) / ka1;
2665 j1t = j2 - (nrt - 1) * ka1;
2667 zlartv_(&nrt, &ab[ka1 - l + 1 + (j1t - ka1 + l) * ab_dim1]
2668 , &inca, &ab[ka1 - l + (j1t - ka1 + l) * ab_dim1],
2669 &inca, &rwork[m - *kb + j1t], &work[m - *kb +
2677 /* post-multiply X by product of rotations in 2nd set */
2681 for (j = j1; i__3 < 0 ? j >= i__4 : j <= i__4; j += i__3) {
2682 d_cnjg(&z__1, &work[m - *kb + j]);
2683 zrot_(&nx, &x[j * x_dim1 + 1], &c__1, &x[(j - 1) * x_dim1
2684 + 1], &c__1, &rwork[m - *kb + j], &z__1);
2692 for (k = 1; k <= i__3; ++k) {
2694 i__4 = 1, i__1 = k + i0 - m + 1;
2695 j2 = i__ + k + 1 - f2cmax(i__4,i__1) * ka1;
2697 /* finish applying rotations in 1st set from the left */
2699 for (l = *kb - k; l >= 1; --l) {
2700 nrt = (j2 + l - 1) / ka1;
2701 j1t = j2 - (nrt - 1) * ka1;
2703 zlartv_(&nrt, &ab[ka1 - l + 1 + (j1t - ka1 + l) * ab_dim1]
2704 , &inca, &ab[ka1 - l + (j1t - ka1 + l) * ab_dim1],
2705 &inca, &rwork[j1t], &work[j1t], &ka1);
2714 for (j = 2; j <= i__3; ++j) {
2715 rwork[j] = rwork[j + *ka];
2718 work[i__4].r = work[i__1].r, work[i__4].i = work[i__1].i;