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 complex c_b1 = {1.f,0.f};
516 static complex c_b2 = {0.f,0.f};
517 static integer c__1 = 1;
518 static integer c_n1 = -1;
519 static integer c__2 = 2;
520 static integer c__3 = 3;
521 static integer c__16 = 16;
523 /* > \brief \b CGGHD3 */
525 /* =========== DOCUMENTATION =========== */
527 /* Online html documentation available at */
528 /* http://www.netlib.org/lapack/explore-html/ */
531 /* > Download CGGHD3 + dependencies */
532 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/cgghd3.
535 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/cgghd3.
538 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/cgghd3.
546 /* SUBROUTINE CGGHD3( COMPQ, COMPZ, N, ILO, IHI, A, LDA, B, LDB, Q, */
547 /* $ LDQ, Z, LDZ, WORK, LWORK, INFO ) */
549 /* CHARACTER COMPQ, COMPZ */
550 /* INTEGER IHI, ILO, INFO, LDA, LDB, LDQ, LDZ, N, LWORK */
551 /* COMPLEX A( LDA, * ), B( LDB, * ), Q( LDQ, * ), */
552 /* $ Z( LDZ, * ), WORK( * ) */
555 /* > \par Purpose: */
561 /* > CGGHD3 reduces a pair of complex matrices (A,B) to generalized upper */
562 /* > Hessenberg form using unitary transformations, where A is a */
563 /* > general matrix and B is upper triangular. The form of the */
564 /* > generalized eigenvalue problem is */
565 /* > A*x = lambda*B*x, */
566 /* > and B is typically made upper triangular by computing its QR */
567 /* > factorization and moving the unitary matrix Q to the left side */
568 /* > of the equation. */
570 /* > This subroutine simultaneously reduces A to a Hessenberg matrix H: */
572 /* > and transforms B to another upper triangular matrix T: */
574 /* > in order to reduce the problem to its standard form */
575 /* > H*y = lambda*T*y */
576 /* > where y = Z**H*x. */
578 /* > The unitary matrices Q and Z are determined as products of Givens */
579 /* > rotations. They may either be formed explicitly, or they may be */
580 /* > postmultiplied into input matrices Q1 and Z1, so that */
582 /* > Q1 * A * Z1**H = (Q1*Q) * H * (Z1*Z)**H */
584 /* > Q1 * B * Z1**H = (Q1*Q) * T * (Z1*Z)**H */
586 /* > If Q1 is the unitary matrix from the QR factorization of B in the */
587 /* > original equation A*x = lambda*B*x, then CGGHD3 reduces the original */
588 /* > problem to generalized Hessenberg form. */
590 /* > This is a blocked variant of CGGHRD, using matrix-matrix */
591 /* > multiplications for parts of the computation to enhance performance. */
597 /* > \param[in] COMPQ */
599 /* > COMPQ is CHARACTER*1 */
600 /* > = 'N': do not compute Q; */
601 /* > = 'I': Q is initialized to the unit matrix, and the */
602 /* > unitary matrix Q is returned; */
603 /* > = 'V': Q must contain a unitary matrix Q1 on entry, */
604 /* > and the product Q1*Q is returned. */
607 /* > \param[in] COMPZ */
609 /* > COMPZ is CHARACTER*1 */
610 /* > = 'N': do not compute Z; */
611 /* > = 'I': Z is initialized to the unit matrix, and the */
612 /* > unitary matrix Z is returned; */
613 /* > = 'V': Z must contain a unitary matrix Z1 on entry, */
614 /* > and the product Z1*Z is returned. */
620 /* > The order of the matrices A and B. N >= 0. */
623 /* > \param[in] ILO */
625 /* > ILO is INTEGER */
628 /* > \param[in] IHI */
630 /* > IHI is INTEGER */
632 /* > ILO and IHI mark the rows and columns of A which are to be */
633 /* > reduced. It is assumed that A is already upper triangular */
634 /* > in rows and columns 1:ILO-1 and IHI+1:N. ILO and IHI are */
635 /* > normally set by a previous call to CGGBAL; otherwise they */
636 /* > should be set to 1 and N respectively. */
637 /* > 1 <= ILO <= IHI <= N, if N > 0; ILO=1 and IHI=0, if N=0. */
640 /* > \param[in,out] A */
642 /* > A is COMPLEX array, dimension (LDA, N) */
643 /* > On entry, the N-by-N general matrix to be reduced. */
644 /* > On exit, the upper triangle and the first subdiagonal of A */
645 /* > are overwritten with the upper Hessenberg matrix H, and the */
646 /* > rest is set to zero. */
649 /* > \param[in] LDA */
651 /* > LDA is INTEGER */
652 /* > The leading dimension of the array A. LDA >= f2cmax(1,N). */
655 /* > \param[in,out] B */
657 /* > B is COMPLEX array, dimension (LDB, N) */
658 /* > On entry, the N-by-N upper triangular matrix B. */
659 /* > On exit, the upper triangular matrix T = Q**H B Z. The */
660 /* > elements below the diagonal are set to zero. */
663 /* > \param[in] LDB */
665 /* > LDB is INTEGER */
666 /* > The leading dimension of the array B. LDB >= f2cmax(1,N). */
669 /* > \param[in,out] Q */
671 /* > Q is COMPLEX array, dimension (LDQ, N) */
672 /* > On entry, if COMPQ = 'V', the unitary matrix Q1, typically */
673 /* > from the QR factorization of B. */
674 /* > On exit, if COMPQ='I', the unitary matrix Q, and if */
675 /* > COMPQ = 'V', the product Q1*Q. */
676 /* > Not referenced if COMPQ='N'. */
679 /* > \param[in] LDQ */
681 /* > LDQ is INTEGER */
682 /* > The leading dimension of the array Q. */
683 /* > LDQ >= N if COMPQ='V' or 'I'; LDQ >= 1 otherwise. */
686 /* > \param[in,out] Z */
688 /* > Z is COMPLEX array, dimension (LDZ, N) */
689 /* > On entry, if COMPZ = 'V', the unitary matrix Z1. */
690 /* > On exit, if COMPZ='I', the unitary matrix Z, and if */
691 /* > COMPZ = 'V', the product Z1*Z. */
692 /* > Not referenced if COMPZ='N'. */
695 /* > \param[in] LDZ */
697 /* > LDZ is INTEGER */
698 /* > The leading dimension of the array Z. */
699 /* > LDZ >= N if COMPZ='V' or 'I'; LDZ >= 1 otherwise. */
702 /* > \param[out] WORK */
704 /* > WORK is COMPLEX array, dimension (LWORK) */
705 /* > On exit, if INFO = 0, WORK(1) returns the optimal LWORK. */
708 /* > \param[in] LWORK */
710 /* > LWORK is INTEGER */
711 /* > The length of the array WORK. LWORK >= 1. */
712 /* > For optimum performance LWORK >= 6*N*NB, where NB is the */
713 /* > optimal blocksize. */
715 /* > If LWORK = -1, then a workspace query is assumed; the routine */
716 /* > only calculates the optimal size of the WORK array, returns */
717 /* > this value as the first entry of the WORK array, and no error */
718 /* > message related to LWORK is issued by XERBLA. */
721 /* > \param[out] INFO */
723 /* > INFO is INTEGER */
724 /* > = 0: successful exit. */
725 /* > < 0: if INFO = -i, the i-th argument had an illegal value. */
731 /* > \author Univ. of Tennessee */
732 /* > \author Univ. of California Berkeley */
733 /* > \author Univ. of Colorado Denver */
734 /* > \author NAG Ltd. */
736 /* > \date January 2015 */
738 /* > \ingroup complexOTHERcomputational */
740 /* > \par Further Details: */
741 /* ===================== */
745 /* > This routine reduces A to Hessenberg form and maintains B in */
746 /* > using a blocked variant of Moler and Stewart's original algorithm, */
747 /* > as described by Kagstrom, Kressner, Quintana-Orti, and Quintana-Orti */
751 /* ===================================================================== */
752 /* Subroutine */ int cgghd3_(char *compq, char *compz, integer *n, integer *
753 ilo, integer *ihi, complex *a, integer *lda, complex *b, integer *ldb,
754 complex *q, integer *ldq, complex *z__, integer *ldz, complex *work,
755 integer *lwork, integer *info)
757 /* System generated locals */
758 integer a_dim1, a_offset, b_dim1, b_offset, q_dim1, q_offset, z_dim1,
759 z_offset, i__1, i__2, i__3, i__4, i__5, i__6, i__7, i__8, i__9;
760 complex q__1, q__2, q__3, q__4;
762 /* Local variables */
764 integer cola, jcol, ierr;
766 extern /* Subroutine */ int crot_(integer *, complex *, integer *,
767 complex *, integer *, real *, complex *);
768 integer jrow, topq, ppwo;
769 complex temp1, temp2, temp3;
771 integer kacc22, i__, j, k;
773 extern /* Subroutine */ int cgemm_(char *, char *, integer *, integer *,
774 integer *, complex *, complex *, integer *, complex *, integer *,
775 complex *, complex *, integer *);
776 extern logical lsame_(char *, char *);
777 extern /* Subroutine */ int cgemv_(char *, integer *, integer *, complex *
778 , complex *, integer *, complex *, integer *, complex *, complex *
781 extern /* Subroutine */ int cunm22_(char *, char *, integer *, integer *,
782 integer *, integer *, complex *, integer *, complex *, integer *,
783 complex *, integer *, integer *);
790 extern /* Subroutine */ int ctrmv_(char *, char *, char *, integer *,
791 complex *, integer *, complex *, integer *);
792 logical initz, wantz;
794 char compq2[1], compz2[1];
796 extern /* Subroutine */ int cgghrd_(char *, char *, integer *, integer *,
797 integer *, complex *, integer *, complex *, integer *, complex *,
798 integer *, complex *, integer *, integer *);
800 extern /* Subroutine */ int claset_(char *, integer *, integer *, complex
801 *, complex *, complex *, integer *), clartg_(complex *,
802 complex *, real *, complex *, complex *), clacpy_(char *, integer
803 *, integer *, complex *, integer *, complex *, integer *);
804 extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
805 integer *, integer *, ftnlen, ftnlen);
806 extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
809 integer nnb, len, top, ppw, n2nb;
812 /* -- LAPACK computational routine (version 3.8.0) -- */
813 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
814 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
820 /* ===================================================================== */
823 /* Decode and test the input parameters. */
825 /* Parameter adjustments */
827 a_offset = 1 + a_dim1 * 1;
830 b_offset = 1 + b_dim1 * 1;
833 q_offset = 1 + q_dim1 * 1;
836 z_offset = 1 + z_dim1 * 1;
842 nb = ilaenv_(&c__1, "CGGHD3", " ", n, ilo, ihi, &c_n1, (ftnlen)6, (ftnlen)
846 lwkopt = f2cmax(i__1,1);
847 q__1.r = (real) lwkopt, q__1.i = 0.f;
848 work[1].r = q__1.r, work[1].i = q__1.i;
849 initq = lsame_(compq, "I");
850 wantq = initq || lsame_(compq, "V");
851 initz = lsame_(compz, "I");
852 wantz = initz || lsame_(compz, "V");
853 lquery = *lwork == -1;
855 if (! lsame_(compq, "N") && ! wantq) {
857 } else if (! lsame_(compz, "N") && ! wantz) {
861 } else if (*ilo < 1) {
863 } else if (*ihi > *n || *ihi < *ilo - 1) {
865 } else if (*lda < f2cmax(1,*n)) {
867 } else if (*ldb < f2cmax(1,*n)) {
869 } else if (wantq && *ldq < *n || *ldq < 1) {
871 } else if (wantz && *ldz < *n || *ldz < 1) {
873 } else if (*lwork < 1 && ! lquery) {
878 xerbla_("CGGHD3", &i__1, (ftnlen)6);
884 /* Initialize Q and Z if desired. */
887 claset_("All", n, n, &c_b2, &c_b1, &q[q_offset], ldq);
890 claset_("All", n, n, &c_b2, &c_b1, &z__[z_offset], ldz);
893 /* Zero out lower triangle of B. */
898 claset_("Lower", &i__1, &i__2, &c_b2, &c_b2, &b[b_dim1 + 2], ldb);
901 /* Quick return if possible */
903 nh = *ihi - *ilo + 1;
905 work[1].r = 1.f, work[1].i = 0.f;
909 /* Determine the blocksize. */
911 nbmin = ilaenv_(&c__2, "CGGHD3", " ", n, ilo, ihi, &c_n1, (ftnlen)6, (
913 if (nb > 1 && nb < nh) {
915 /* Determine when to use unblocked instead of blocked code. */
918 i__1 = nb, i__2 = ilaenv_(&c__3, "CGGHD3", " ", n, ilo, ihi, &c_n1, (
919 ftnlen)6, (ftnlen)1);
920 nx = f2cmax(i__1,i__2);
923 /* Determine if workspace is large enough for blocked code. */
925 if (*lwork < lwkopt) {
927 /* Not enough workspace to use optimal NB: determine the */
928 /* minimum value of NB, and reduce NB or force use of */
929 /* unblocked code. */
932 i__1 = 2, i__2 = ilaenv_(&c__2, "CGGHD3", " ", n, ilo, ihi, &
933 c_n1, (ftnlen)6, (ftnlen)1);
934 nbmin = f2cmax(i__1,i__2);
935 if (*lwork >= *n * 6 * nbmin) {
936 nb = *lwork / (*n * 6);
944 if (nb < nbmin || nb >= nh) {
946 /* Use unblocked code below */
952 /* Use blocked code */
954 kacc22 = ilaenv_(&c__16, "CGGHD3", " ", n, ilo, ihi, &c_n1, (ftnlen)6,
959 for (jcol = *ilo; i__2 < 0 ? jcol >= i__1 : jcol <= i__1; jcol +=
962 i__3 = nb, i__4 = *ihi - jcol - 1;
963 nnb = f2cmin(i__3,i__4);
965 /* Initialize small unitary factors that will hold the */
966 /* accumulated Givens rotations in workspace. */
967 /* N2NB denotes the number of 2*NNB-by-2*NNB factors */
968 /* NBLST denotes the (possibly smaller) order of the last */
971 n2nb = (*ihi - jcol - 1) / nnb - 1;
972 nblst = *ihi - jcol - n2nb * nnb;
973 claset_("All", &nblst, &nblst, &c_b2, &c_b1, &work[1], &nblst);
974 pw = nblst * nblst + 1;
976 for (i__ = 1; i__ <= i__3; ++i__) {
980 claset_("All", &i__4, &i__5, &c_b2, &c_b1, &work[pw], &i__6);
981 pw += (nnb << 2) * nnb;
984 /* Reduce columns JCOL:JCOL+NNB-1 of A to Hessenberg form. */
986 i__3 = jcol + nnb - 1;
987 for (j = jcol; j <= i__3; ++j) {
989 /* Reduce Jth column of A. Store cosines and sines in Jth */
990 /* column of A and B, respectively. */
993 for (i__ = *ihi; i__ >= i__4; --i__) {
994 i__5 = i__ - 1 + j * a_dim1;
995 temp.r = a[i__5].r, temp.i = a[i__5].i;
996 clartg_(&temp, &a[i__ + j * a_dim1], &c__, &s, &a[i__ - 1
998 i__5 = i__ + j * a_dim1;
999 q__1.r = c__, q__1.i = 0.f;
1000 a[i__5].r = q__1.r, a[i__5].i = q__1.i;
1001 i__5 = i__ + j * b_dim1;
1002 b[i__5].r = s.r, b[i__5].i = s.i;
1005 /* Accumulate Givens rotations into workspace array. */
1007 ppw = (nblst + 1) * (nblst - 2) - j + jcol + 1;
1009 jrow = j + n2nb * nnb + 2;
1011 for (i__ = *ihi; i__ >= i__4; --i__) {
1012 i__5 = i__ + j * a_dim1;
1013 ctemp.r = a[i__5].r, ctemp.i = a[i__5].i;
1014 i__5 = i__ + j * b_dim1;
1015 s.r = b[i__5].r, s.i = b[i__5].i;
1016 i__5 = ppw + len - 1;
1017 for (jj = ppw; jj <= i__5; ++jj) {
1019 temp.r = work[i__6].r, temp.i = work[i__6].i;
1021 q__2.r = ctemp.r * temp.r - ctemp.i * temp.i, q__2.i =
1022 ctemp.r * temp.i + ctemp.i * temp.r;
1024 q__3.r = s.r * work[i__7].r - s.i * work[i__7].i,
1025 q__3.i = s.r * work[i__7].i + s.i * work[i__7]
1027 q__1.r = q__2.r - q__3.r, q__1.i = q__2.i - q__3.i;
1028 work[i__6].r = q__1.r, work[i__6].i = q__1.i;
1031 q__2.r = q__3.r * temp.r - q__3.i * temp.i, q__2.i =
1032 q__3.r * temp.i + q__3.i * temp.r;
1034 q__4.r = ctemp.r * work[i__7].r - ctemp.i * work[i__7]
1035 .i, q__4.i = ctemp.r * work[i__7].i + ctemp.i
1037 q__1.r = q__2.r + q__4.r, q__1.i = q__2.i + q__4.i;
1038 work[i__6].r = q__1.r, work[i__6].i = q__1.i;
1041 ppw = ppw - nblst - 1;
1044 ppwo = nblst * nblst + (nnb + j - jcol - 1 << 1) * nnb + nnb;
1048 for (jrow = j0; i__5 < 0 ? jrow >= i__4 : jrow <= i__4; jrow
1053 for (i__ = jrow + nnb - 1; i__ >= i__6; --i__) {
1054 i__7 = i__ + j * a_dim1;
1055 ctemp.r = a[i__7].r, ctemp.i = a[i__7].i;
1056 i__7 = i__ + j * b_dim1;
1057 s.r = b[i__7].r, s.i = b[i__7].i;
1058 i__7 = ppw + len - 1;
1059 for (jj = ppw; jj <= i__7; ++jj) {
1060 i__8 = jj + (nnb << 1);
1061 temp.r = work[i__8].r, temp.i = work[i__8].i;
1062 i__8 = jj + (nnb << 1);
1063 q__2.r = ctemp.r * temp.r - ctemp.i * temp.i,
1064 q__2.i = ctemp.r * temp.i + ctemp.i *
1067 q__3.r = s.r * work[i__9].r - s.i * work[i__9].i,
1068 q__3.i = s.r * work[i__9].i + s.i * work[
1070 q__1.r = q__2.r - q__3.r, q__1.i = q__2.i -
1072 work[i__8].r = q__1.r, work[i__8].i = q__1.i;
1075 q__2.r = q__3.r * temp.r - q__3.i * temp.i,
1076 q__2.i = q__3.r * temp.i + q__3.i *
1079 q__4.r = ctemp.r * work[i__9].r - ctemp.i * work[
1080 i__9].i, q__4.i = ctemp.r * work[i__9].i
1081 + ctemp.i * work[i__9].r;
1082 q__1.r = q__2.r + q__4.r, q__1.i = q__2.i +
1084 work[i__8].r = q__1.r, work[i__8].i = q__1.i;
1087 ppw = ppw - (nnb << 1) - 1;
1089 ppwo += (nnb << 2) * nnb;
1092 /* TOP denotes the number of top rows in A and B that will */
1093 /* not be updated during the next steps. */
1101 /* Propagate transformations through B and replace stored */
1102 /* left sines/cosines by right sines/cosines. */
1105 for (jj = *n; jj >= i__5; --jj) {
1107 /* Update JJth column of B. */
1112 for (i__ = f2cmin(i__4,*ihi); i__ >= i__6; --i__) {
1113 i__4 = i__ + j * a_dim1;
1114 ctemp.r = a[i__4].r, ctemp.i = a[i__4].i;
1115 i__4 = i__ + j * b_dim1;
1116 s.r = b[i__4].r, s.i = b[i__4].i;
1117 i__4 = i__ + jj * b_dim1;
1118 temp.r = b[i__4].r, temp.i = b[i__4].i;
1119 i__4 = i__ + jj * b_dim1;
1120 q__2.r = ctemp.r * temp.r - ctemp.i * temp.i, q__2.i =
1121 ctemp.r * temp.i + ctemp.i * temp.r;
1123 i__7 = i__ - 1 + jj * b_dim1;
1124 q__3.r = q__4.r * b[i__7].r - q__4.i * b[i__7].i,
1125 q__3.i = q__4.r * b[i__7].i + q__4.i * b[i__7]
1127 q__1.r = q__2.r - q__3.r, q__1.i = q__2.i - q__3.i;
1128 b[i__4].r = q__1.r, b[i__4].i = q__1.i;
1129 i__4 = i__ - 1 + jj * b_dim1;
1130 q__2.r = s.r * temp.r - s.i * temp.i, q__2.i = s.r *
1131 temp.i + s.i * temp.r;
1132 i__7 = i__ - 1 + jj * b_dim1;
1133 q__3.r = ctemp.r * b[i__7].r - ctemp.i * b[i__7].i,
1134 q__3.i = ctemp.r * b[i__7].i + ctemp.i * b[
1136 q__1.r = q__2.r + q__3.r, q__1.i = q__2.i + q__3.i;
1137 b[i__4].r = q__1.r, b[i__4].i = q__1.i;
1140 /* Annihilate B( JJ+1, JJ ). */
1143 i__6 = jj + 1 + (jj + 1) * b_dim1;
1144 temp.r = b[i__6].r, temp.i = b[i__6].i;
1145 clartg_(&temp, &b[jj + 1 + jj * b_dim1], &c__, &s, &b[
1146 jj + 1 + (jj + 1) * b_dim1]);
1147 i__6 = jj + 1 + jj * b_dim1;
1148 b[i__6].r = 0.f, b[i__6].i = 0.f;
1150 crot_(&i__6, &b[top + 1 + (jj + 1) * b_dim1], &c__1, &
1151 b[top + 1 + jj * b_dim1], &c__1, &c__, &s);
1152 i__6 = jj + 1 + j * a_dim1;
1153 q__1.r = c__, q__1.i = 0.f;
1154 a[i__6].r = q__1.r, a[i__6].i = q__1.i;
1155 i__6 = jj + 1 + j * b_dim1;
1157 q__1.r = -q__2.r, q__1.i = -q__2.i;
1158 b[i__6].r = q__1.r, b[i__6].i = q__1.i;
1162 /* Update A by transformations from right. */
1164 jj = (*ihi - j - 1) % 3;
1166 for (i__ = *ihi - j - 3; i__ >= i__5; i__ += -3) {
1167 i__6 = j + 1 + i__ + j * a_dim1;
1168 ctemp.r = a[i__6].r, ctemp.i = a[i__6].i;
1169 i__6 = j + 1 + i__ + j * b_dim1;
1170 q__1.r = -b[i__6].r, q__1.i = -b[i__6].i;
1171 s.r = q__1.r, s.i = q__1.i;
1172 i__6 = j + 2 + i__ + j * a_dim1;
1173 c1.r = a[i__6].r, c1.i = a[i__6].i;
1174 i__6 = j + 2 + i__ + j * b_dim1;
1175 q__1.r = -b[i__6].r, q__1.i = -b[i__6].i;
1176 s1.r = q__1.r, s1.i = q__1.i;
1177 i__6 = j + 3 + i__ + j * a_dim1;
1178 c2.r = a[i__6].r, c2.i = a[i__6].i;
1179 i__6 = j + 3 + i__ + j * b_dim1;
1180 q__1.r = -b[i__6].r, q__1.i = -b[i__6].i;
1181 s2.r = q__1.r, s2.i = q__1.i;
1184 for (k = top + 1; k <= i__6; ++k) {
1185 i__4 = k + (j + i__) * a_dim1;
1186 temp.r = a[i__4].r, temp.i = a[i__4].i;
1187 i__4 = k + (j + i__ + 1) * a_dim1;
1188 temp1.r = a[i__4].r, temp1.i = a[i__4].i;
1189 i__4 = k + (j + i__ + 2) * a_dim1;
1190 temp2.r = a[i__4].r, temp2.i = a[i__4].i;
1191 i__4 = k + (j + i__ + 3) * a_dim1;
1192 temp3.r = a[i__4].r, temp3.i = a[i__4].i;
1193 i__4 = k + (j + i__ + 3) * a_dim1;
1194 q__2.r = c2.r * temp3.r - c2.i * temp3.i, q__2.i =
1195 c2.r * temp3.i + c2.i * temp3.r;
1197 q__3.r = q__4.r * temp2.r - q__4.i * temp2.i, q__3.i =
1198 q__4.r * temp2.i + q__4.i * temp2.r;
1199 q__1.r = q__2.r + q__3.r, q__1.i = q__2.i + q__3.i;
1200 a[i__4].r = q__1.r, a[i__4].i = q__1.i;
1201 q__3.r = -s2.r, q__3.i = -s2.i;
1202 q__2.r = q__3.r * temp3.r - q__3.i * temp3.i, q__2.i =
1203 q__3.r * temp3.i + q__3.i * temp3.r;
1204 q__4.r = c2.r * temp2.r - c2.i * temp2.i, q__4.i =
1205 c2.r * temp2.i + c2.i * temp2.r;
1206 q__1.r = q__2.r + q__4.r, q__1.i = q__2.i + q__4.i;
1207 temp2.r = q__1.r, temp2.i = q__1.i;
1208 i__4 = k + (j + i__ + 2) * a_dim1;
1209 q__2.r = c1.r * temp2.r - c1.i * temp2.i, q__2.i =
1210 c1.r * temp2.i + c1.i * temp2.r;
1212 q__3.r = q__4.r * temp1.r - q__4.i * temp1.i, q__3.i =
1213 q__4.r * temp1.i + q__4.i * temp1.r;
1214 q__1.r = q__2.r + q__3.r, q__1.i = q__2.i + q__3.i;
1215 a[i__4].r = q__1.r, a[i__4].i = q__1.i;
1216 q__3.r = -s1.r, q__3.i = -s1.i;
1217 q__2.r = q__3.r * temp2.r - q__3.i * temp2.i, q__2.i =
1218 q__3.r * temp2.i + q__3.i * temp2.r;
1219 q__4.r = c1.r * temp1.r - c1.i * temp1.i, q__4.i =
1220 c1.r * temp1.i + c1.i * temp1.r;
1221 q__1.r = q__2.r + q__4.r, q__1.i = q__2.i + q__4.i;
1222 temp1.r = q__1.r, temp1.i = q__1.i;
1223 i__4 = k + (j + i__ + 1) * a_dim1;
1224 q__2.r = ctemp.r * temp1.r - ctemp.i * temp1.i,
1225 q__2.i = ctemp.r * temp1.i + ctemp.i *
1228 q__3.r = q__4.r * temp.r - q__4.i * temp.i, q__3.i =
1229 q__4.r * temp.i + q__4.i * temp.r;
1230 q__1.r = q__2.r + q__3.r, q__1.i = q__2.i + q__3.i;
1231 a[i__4].r = q__1.r, a[i__4].i = q__1.i;
1232 i__4 = k + (j + i__) * a_dim1;
1233 q__3.r = -s.r, q__3.i = -s.i;
1234 q__2.r = q__3.r * temp1.r - q__3.i * temp1.i, q__2.i =
1235 q__3.r * temp1.i + q__3.i * temp1.r;
1236 q__4.r = ctemp.r * temp.r - ctemp.i * temp.i, q__4.i =
1237 ctemp.r * temp.i + ctemp.i * temp.r;
1238 q__1.r = q__2.r + q__4.r, q__1.i = q__2.i + q__4.i;
1239 a[i__4].r = q__1.r, a[i__4].i = q__1.i;
1244 for (i__ = jj; i__ >= 1; --i__) {
1245 i__5 = j + 1 + i__ + j * a_dim1;
1246 c__ = (doublereal) a[i__5].r;
1248 r_cnjg(&q__2, &b[j + 1 + i__ + j * b_dim1]);
1249 q__1.r = -q__2.r, q__1.i = -q__2.i;
1250 crot_(&i__5, &a[top + 1 + (j + i__ + 1) * a_dim1], &
1251 c__1, &a[top + 1 + (j + i__) * a_dim1], &c__1,
1256 /* Update (J+1)th column of A by transformations from left. */
1258 if (j < jcol + nnb - 1) {
1261 /* Multiply with the trailing accumulated unitary */
1262 /* matrix, which takes the form */
1268 /* where U21 is a LEN-by-LEN matrix and U12 is lower */
1271 jrow = *ihi - nblst + 1;
1272 cgemv_("Conjugate", &nblst, &len, &c_b1, &work[1], &nblst,
1273 &a[jrow + (j + 1) * a_dim1], &c__1, &c_b2, &work[
1276 i__5 = jrow + nblst - len - 1;
1277 for (i__ = jrow; i__ <= i__5; ++i__) {
1279 i__4 = i__ + (j + 1) * a_dim1;
1280 work[i__6].r = a[i__4].r, work[i__6].i = a[i__4].i;
1284 ctrmv_("Lower", "Conjugate", "Non-unit", &i__5, &work[len
1285 * nblst + 1], &nblst, &work[pw + len], &c__1);
1287 cgemv_("Conjugate", &len, &i__5, &c_b1, &work[(len + 1) *
1288 nblst - len + 1], &nblst, &a[jrow + nblst - len +
1289 (j + 1) * a_dim1], &c__1, &c_b1, &work[pw + len],
1292 i__5 = jrow + nblst - 1;
1293 for (i__ = jrow; i__ <= i__5; ++i__) {
1294 i__6 = i__ + (j + 1) * a_dim1;
1296 a[i__6].r = work[i__4].r, a[i__6].i = work[i__4].i;
1300 /* Multiply with the other accumulated unitary */
1301 /* matrices, which take the form */
1305 /* U = [ U21 U22 0 ], */
1309 /* where I denotes the (NNB-LEN)-by-(NNB-LEN) identity */
1310 /* matrix, U21 is a LEN-by-LEN upper triangular matrix */
1311 /* and U12 is an NNB-by-NNB lower triangular matrix. */
1313 ppwo = nblst * nblst + 1;
1317 for (jrow = j0; i__6 < 0 ? jrow >= i__5 : jrow <= i__5;
1320 i__4 = jrow + nnb - 1;
1321 for (i__ = jrow; i__ <= i__4; ++i__) {
1323 i__8 = i__ + (j + 1) * a_dim1;
1324 work[i__7].r = a[i__8].r, work[i__7].i = a[i__8]
1329 i__4 = jrow + nnb + len - 1;
1330 for (i__ = jrow + nnb; i__ <= i__4; ++i__) {
1332 i__8 = i__ + (j + 1) * a_dim1;
1333 work[i__7].r = a[i__8].r, work[i__7].i = a[i__8]
1338 ctrmv_("Upper", "Conjugate", "Non-unit", &len, &work[
1339 ppwo + nnb], &i__4, &work[pw], &c__1);
1341 ctrmv_("Lower", "Conjugate", "Non-unit", &nnb, &work[
1342 ppwo + (len << 1) * nnb], &i__4, &work[pw +
1345 cgemv_("Conjugate", &nnb, &len, &c_b1, &work[ppwo], &
1346 i__4, &a[jrow + (j + 1) * a_dim1], &c__1, &
1347 c_b1, &work[pw], &c__1);
1349 cgemv_("Conjugate", &len, &nnb, &c_b1, &work[ppwo + (
1350 len << 1) * nnb + nnb], &i__4, &a[jrow + nnb
1351 + (j + 1) * a_dim1], &c__1, &c_b1, &work[pw +
1354 i__4 = jrow + len + nnb - 1;
1355 for (i__ = jrow; i__ <= i__4; ++i__) {
1356 i__7 = i__ + (j + 1) * a_dim1;
1358 a[i__7].r = work[i__8].r, a[i__7].i = work[i__8]
1362 ppwo += (nnb << 2) * nnb;
1367 /* Apply accumulated unitary matrices to A. */
1369 cola = *n - jcol - nnb + 1;
1370 j = *ihi - nblst + 1;
1371 cgemm_("Conjugate", "No Transpose", &nblst, &cola, &nblst, &c_b1,
1372 &work[1], &nblst, &a[j + (jcol + nnb) * a_dim1], lda, &
1373 c_b2, &work[pw], &nblst);
1374 clacpy_("All", &nblst, &cola, &work[pw], &nblst, &a[j + (jcol +
1375 nnb) * a_dim1], lda);
1376 ppwo = nblst * nblst + 1;
1380 for (j = j0; i__6 < 0 ? j >= i__3 : j <= i__3; j += i__6) {
1383 /* Exploit the structure of */
1389 /* where all blocks are NNB-by-NNB, U21 is upper */
1390 /* triangular and U12 is lower triangular. */
1394 i__7 = *lwork - pw + 1;
1395 cunm22_("Left", "Conjugate", &i__5, &cola, &nnb, &nnb, &
1396 work[ppwo], &i__4, &a[j + (jcol + nnb) * a_dim1],
1397 lda, &work[pw], &i__7, &ierr);
1400 /* Ignore the structure of U. */
1406 cgemm_("Conjugate", "No Transpose", &i__5, &cola, &i__4, &
1407 c_b1, &work[ppwo], &i__7, &a[j + (jcol + nnb) *
1408 a_dim1], lda, &c_b2, &work[pw], &i__8);
1411 clacpy_("All", &i__5, &cola, &work[pw], &i__4, &a[j + (
1412 jcol + nnb) * a_dim1], lda);
1414 ppwo += (nnb << 2) * nnb;
1417 /* Apply accumulated unitary matrices to Q. */
1420 j = *ihi - nblst + 1;
1423 i__6 = 2, i__3 = j - jcol + 1;
1424 topq = f2cmax(i__6,i__3);
1425 nh = *ihi - topq + 1;
1430 cgemm_("No Transpose", "No Transpose", &nh, &nblst, &nblst, &
1431 c_b1, &q[topq + j * q_dim1], ldq, &work[1], &nblst, &
1432 c_b2, &work[pw], &nh);
1433 clacpy_("All", &nh, &nblst, &work[pw], &nh, &q[topq + j *
1435 ppwo = nblst * nblst + 1;
1439 for (j = j0; i__3 < 0 ? j >= i__6 : j <= i__6; j += i__3) {
1442 i__5 = 2, i__4 = j - jcol + 1;
1443 topq = f2cmax(i__5,i__4);
1444 nh = *ihi - topq + 1;
1448 /* Exploit the structure of U. */
1452 i__7 = *lwork - pw + 1;
1453 cunm22_("Right", "No Transpose", &nh, &i__5, &nnb, &
1454 nnb, &work[ppwo], &i__4, &q[topq + j * q_dim1]
1455 , ldq, &work[pw], &i__7, &ierr);
1458 /* Ignore the structure of U. */
1463 cgemm_("No Transpose", "No Transpose", &nh, &i__5, &
1464 i__4, &c_b1, &q[topq + j * q_dim1], ldq, &
1465 work[ppwo], &i__7, &c_b2, &work[pw], &nh);
1467 clacpy_("All", &nh, &i__5, &work[pw], &nh, &q[topq +
1470 ppwo += (nnb << 2) * nnb;
1474 /* Accumulate right Givens rotations if required. */
1476 if (wantz || top > 0) {
1478 /* Initialize small unitary factors that will hold the */
1479 /* accumulated Givens rotations in workspace. */
1481 claset_("All", &nblst, &nblst, &c_b2, &c_b1, &work[1], &nblst);
1482 pw = nblst * nblst + 1;
1484 for (i__ = 1; i__ <= i__3; ++i__) {
1488 claset_("All", &i__6, &i__5, &c_b2, &c_b1, &work[pw], &
1490 pw += (nnb << 2) * nnb;
1493 /* Accumulate Givens rotations into workspace array. */
1495 i__3 = jcol + nnb - 1;
1496 for (j = jcol; j <= i__3; ++j) {
1497 ppw = (nblst + 1) * (nblst - 2) - j + jcol + 1;
1499 jrow = j + n2nb * nnb + 2;
1501 for (i__ = *ihi; i__ >= i__6; --i__) {
1502 i__5 = i__ + j * a_dim1;
1503 ctemp.r = a[i__5].r, ctemp.i = a[i__5].i;
1504 i__5 = i__ + j * a_dim1;
1505 a[i__5].r = 0.f, a[i__5].i = 0.f;
1506 i__5 = i__ + j * b_dim1;
1507 s.r = b[i__5].r, s.i = b[i__5].i;
1508 i__5 = i__ + j * b_dim1;
1509 b[i__5].r = 0.f, b[i__5].i = 0.f;
1510 i__5 = ppw + len - 1;
1511 for (jj = ppw; jj <= i__5; ++jj) {
1513 temp.r = work[i__4].r, temp.i = work[i__4].i;
1515 q__2.r = ctemp.r * temp.r - ctemp.i * temp.i,
1516 q__2.i = ctemp.r * temp.i + ctemp.i *
1520 q__3.r = q__4.r * work[i__7].r - q__4.i * work[
1521 i__7].i, q__3.i = q__4.r * work[i__7].i +
1522 q__4.i * work[i__7].r;
1523 q__1.r = q__2.r - q__3.r, q__1.i = q__2.i -
1525 work[i__4].r = q__1.r, work[i__4].i = q__1.i;
1527 q__2.r = s.r * temp.r - s.i * temp.i, q__2.i =
1528 s.r * temp.i + s.i * temp.r;
1530 q__3.r = ctemp.r * work[i__7].r - ctemp.i * work[
1531 i__7].i, q__3.i = ctemp.r * work[i__7].i
1532 + ctemp.i * work[i__7].r;
1533 q__1.r = q__2.r + q__3.r, q__1.i = q__2.i +
1535 work[i__4].r = q__1.r, work[i__4].i = q__1.i;
1538 ppw = ppw - nblst - 1;
1541 ppwo = nblst * nblst + (nnb + j - jcol - 1 << 1) * nnb +
1546 for (jrow = j0; i__5 < 0 ? jrow >= i__6 : jrow <= i__6;
1551 for (i__ = jrow + nnb - 1; i__ >= i__4; --i__) {
1552 i__7 = i__ + j * a_dim1;
1553 ctemp.r = a[i__7].r, ctemp.i = a[i__7].i;
1554 i__7 = i__ + j * a_dim1;
1555 a[i__7].r = 0.f, a[i__7].i = 0.f;
1556 i__7 = i__ + j * b_dim1;
1557 s.r = b[i__7].r, s.i = b[i__7].i;
1558 i__7 = i__ + j * b_dim1;
1559 b[i__7].r = 0.f, b[i__7].i = 0.f;
1560 i__7 = ppw + len - 1;
1561 for (jj = ppw; jj <= i__7; ++jj) {
1562 i__8 = jj + (nnb << 1);
1563 temp.r = work[i__8].r, temp.i = work[i__8].i;
1564 i__8 = jj + (nnb << 1);
1565 q__2.r = ctemp.r * temp.r - ctemp.i * temp.i,
1566 q__2.i = ctemp.r * temp.i + ctemp.i *
1570 q__3.r = q__4.r * work[i__9].r - q__4.i *
1571 work[i__9].i, q__3.i = q__4.r * work[
1572 i__9].i + q__4.i * work[i__9].r;
1573 q__1.r = q__2.r - q__3.r, q__1.i = q__2.i -
1575 work[i__8].r = q__1.r, work[i__8].i = q__1.i;
1577 q__2.r = s.r * temp.r - s.i * temp.i, q__2.i =
1578 s.r * temp.i + s.i * temp.r;
1580 q__3.r = ctemp.r * work[i__9].r - ctemp.i *
1581 work[i__9].i, q__3.i = ctemp.r * work[
1582 i__9].i + ctemp.i * work[i__9].r;
1583 q__1.r = q__2.r + q__3.r, q__1.i = q__2.i +
1585 work[i__8].r = q__1.r, work[i__8].i = q__1.i;
1588 ppw = ppw - (nnb << 1) - 1;
1590 ppwo += (nnb << 2) * nnb;
1595 i__3 = *ihi - jcol - 1;
1596 claset_("Lower", &i__3, &nnb, &c_b2, &c_b2, &a[jcol + 2 +
1597 jcol * a_dim1], lda);
1598 i__3 = *ihi - jcol - 1;
1599 claset_("Lower", &i__3, &nnb, &c_b2, &c_b2, &b[jcol + 2 +
1600 jcol * b_dim1], ldb);
1603 /* Apply accumulated unitary matrices to A and B. */
1606 j = *ihi - nblst + 1;
1607 cgemm_("No Transpose", "No Transpose", &top, &nblst, &nblst, &
1608 c_b1, &a[j * a_dim1 + 1], lda, &work[1], &nblst, &
1609 c_b2, &work[pw], &top);
1610 clacpy_("All", &top, &nblst, &work[pw], &top, &a[j * a_dim1 +
1612 ppwo = nblst * nblst + 1;
1616 for (j = j0; i__5 < 0 ? j >= i__3 : j <= i__3; j += i__5) {
1619 /* Exploit the structure of U. */
1623 i__7 = *lwork - pw + 1;
1624 cunm22_("Right", "No Transpose", &top, &i__6, &nnb, &
1625 nnb, &work[ppwo], &i__4, &a[j * a_dim1 + 1],
1626 lda, &work[pw], &i__7, &ierr);
1629 /* Ignore the structure of U. */
1634 cgemm_("No Transpose", "No Transpose", &top, &i__6, &
1635 i__4, &c_b1, &a[j * a_dim1 + 1], lda, &work[
1636 ppwo], &i__7, &c_b2, &work[pw], &top);
1638 clacpy_("All", &top, &i__6, &work[pw], &top, &a[j *
1641 ppwo += (nnb << 2) * nnb;
1644 j = *ihi - nblst + 1;
1645 cgemm_("No Transpose", "No Transpose", &top, &nblst, &nblst, &
1646 c_b1, &b[j * b_dim1 + 1], ldb, &work[1], &nblst, &
1647 c_b2, &work[pw], &top);
1648 clacpy_("All", &top, &nblst, &work[pw], &top, &b[j * b_dim1 +
1650 ppwo = nblst * nblst + 1;
1654 for (j = j0; i__3 < 0 ? j >= i__5 : j <= i__5; j += i__3) {
1657 /* Exploit the structure of U. */
1661 i__7 = *lwork - pw + 1;
1662 cunm22_("Right", "No Transpose", &top, &i__6, &nnb, &
1663 nnb, &work[ppwo], &i__4, &b[j * b_dim1 + 1],
1664 ldb, &work[pw], &i__7, &ierr);
1667 /* Ignore the structure of U. */
1672 cgemm_("No Transpose", "No Transpose", &top, &i__6, &
1673 i__4, &c_b1, &b[j * b_dim1 + 1], ldb, &work[
1674 ppwo], &i__7, &c_b2, &work[pw], &top);
1676 clacpy_("All", &top, &i__6, &work[pw], &top, &b[j *
1679 ppwo += (nnb << 2) * nnb;
1683 /* Apply accumulated unitary matrices to Z. */
1686 j = *ihi - nblst + 1;
1689 i__3 = 2, i__5 = j - jcol + 1;
1690 topq = f2cmax(i__3,i__5);
1691 nh = *ihi - topq + 1;
1696 cgemm_("No Transpose", "No Transpose", &nh, &nblst, &nblst, &
1697 c_b1, &z__[topq + j * z_dim1], ldz, &work[1], &nblst,
1698 &c_b2, &work[pw], &nh);
1699 clacpy_("All", &nh, &nblst, &work[pw], &nh, &z__[topq + j *
1701 ppwo = nblst * nblst + 1;
1705 for (j = j0; i__5 < 0 ? j >= i__3 : j <= i__3; j += i__5) {
1708 i__6 = 2, i__4 = j - jcol + 1;
1709 topq = f2cmax(i__6,i__4);
1710 nh = *ihi - topq + 1;
1714 /* Exploit the structure of U. */
1718 i__7 = *lwork - pw + 1;
1719 cunm22_("Right", "No Transpose", &nh, &i__6, &nnb, &
1720 nnb, &work[ppwo], &i__4, &z__[topq + j *
1721 z_dim1], ldz, &work[pw], &i__7, &ierr);
1724 /* Ignore the structure of U. */
1729 cgemm_("No Transpose", "No Transpose", &nh, &i__6, &
1730 i__4, &c_b1, &z__[topq + j * z_dim1], ldz, &
1731 work[ppwo], &i__7, &c_b2, &work[pw], &nh);
1733 clacpy_("All", &nh, &i__6, &work[pw], &nh, &z__[topq
1734 + j * z_dim1], ldz);
1736 ppwo += (nnb << 2) * nnb;
1742 /* Use unblocked code to reduce the rest of the matrix */
1743 /* Avoid re-initialization of modified Q and Z. */
1745 *(unsigned char *)compq2 = *(unsigned char *)compq;
1746 *(unsigned char *)compz2 = *(unsigned char *)compz;
1749 *(unsigned char *)compq2 = 'V';
1752 *(unsigned char *)compz2 = 'V';
1757 cgghrd_(compq2, compz2, n, &jcol, ihi, &a[a_offset], lda, &b[b_offset]
1758 , ldb, &q[q_offset], ldq, &z__[z_offset], ldz, &ierr);
1760 q__1.r = (real) lwkopt, q__1.i = 0.f;
1761 work[1].r = q__1.r, work[1].i = q__1.i;