14 typedef long long BLASLONG;
15 typedef unsigned long long BLASULONG;
17 typedef long BLASLONG;
18 typedef unsigned long BLASULONG;
22 typedef BLASLONG blasint;
24 #define blasabs(x) llabs(x)
26 #define blasabs(x) labs(x)
30 #define blasabs(x) abs(x)
33 typedef blasint integer;
35 typedef unsigned int uinteger;
36 typedef char *address;
37 typedef short int shortint;
39 typedef double doublereal;
40 typedef struct { real r, i; } complex;
41 typedef struct { doublereal r, i; } doublecomplex;
43 static inline _Fcomplex Cf(complex *z) {_Fcomplex zz={z->r , z->i}; return zz;}
44 static inline _Dcomplex Cd(doublecomplex *z) {_Dcomplex zz={z->r , z->i};return zz;}
45 static inline _Fcomplex * _pCf(complex *z) {return (_Fcomplex*)z;}
46 static inline _Dcomplex * _pCd(doublecomplex *z) {return (_Dcomplex*)z;}
48 static inline _Complex float Cf(complex *z) {return z->r + z->i*_Complex_I;}
49 static inline _Complex double Cd(doublecomplex *z) {return z->r + z->i*_Complex_I;}
50 static inline _Complex float * _pCf(complex *z) {return (_Complex float*)z;}
51 static inline _Complex double * _pCd(doublecomplex *z) {return (_Complex double*)z;}
53 #define pCf(z) (*_pCf(z))
54 #define pCd(z) (*_pCd(z))
56 typedef short int shortlogical;
57 typedef char logical1;
58 typedef char integer1;
63 /* Extern is for use with -E */
74 /*external read, write*/
83 /*internal read, write*/
113 /*rewind, backspace, endfile*/
125 ftnint *inex; /*parameters in standard's order*/
151 union Multitype { /* for multiple entry points */
162 typedef union Multitype Multitype;
164 struct Vardesc { /* for Namelist */
170 typedef struct Vardesc Vardesc;
177 typedef struct Namelist Namelist;
179 #define abs(x) ((x) >= 0 ? (x) : -(x))
180 #define dabs(x) (fabs(x))
181 #define f2cmin(a,b) ((a) <= (b) ? (a) : (b))
182 #define f2cmax(a,b) ((a) >= (b) ? (a) : (b))
183 #define dmin(a,b) (f2cmin(a,b))
184 #define dmax(a,b) (f2cmax(a,b))
185 #define bit_test(a,b) ((a) >> (b) & 1)
186 #define bit_clear(a,b) ((a) & ~((uinteger)1 << (b)))
187 #define bit_set(a,b) ((a) | ((uinteger)1 << (b)))
189 #define abort_() { sig_die("Fortran abort routine called", 1); }
190 #define c_abs(z) (cabsf(Cf(z)))
191 #define c_cos(R,Z) { pCf(R)=ccos(Cf(Z)); }
193 #define c_div(c, a, b) {Cf(c)._Val[0] = (Cf(a)._Val[0]/Cf(b)._Val[0]); Cf(c)._Val[1]=(Cf(a)._Val[1]/Cf(b)._Val[1]);}
194 #define z_div(c, a, b) {Cd(c)._Val[0] = (Cd(a)._Val[0]/Cd(b)._Val[0]); Cd(c)._Val[1]=(Cd(a)._Val[1]/df(b)._Val[1]);}
196 #define c_div(c, a, b) {pCf(c) = Cf(a)/Cf(b);}
197 #define z_div(c, a, b) {pCd(c) = Cd(a)/Cd(b);}
199 #define c_exp(R, Z) {pCf(R) = cexpf(Cf(Z));}
200 #define c_log(R, Z) {pCf(R) = clogf(Cf(Z));}
201 #define c_sin(R, Z) {pCf(R) = csinf(Cf(Z));}
202 //#define c_sqrt(R, Z) {*(R) = csqrtf(Cf(Z));}
203 #define c_sqrt(R, Z) {pCf(R) = csqrtf(Cf(Z));}
204 #define d_abs(x) (fabs(*(x)))
205 #define d_acos(x) (acos(*(x)))
206 #define d_asin(x) (asin(*(x)))
207 #define d_atan(x) (atan(*(x)))
208 #define d_atn2(x, y) (atan2(*(x),*(y)))
209 #define d_cnjg(R, Z) { pCd(R) = conj(Cd(Z)); }
210 #define r_cnjg(R, Z) { pCf(R) = conjf(Cf(Z)); }
211 #define d_cos(x) (cos(*(x)))
212 #define d_cosh(x) (cosh(*(x)))
213 #define d_dim(__a, __b) ( *(__a) > *(__b) ? *(__a) - *(__b) : 0.0 )
214 #define d_exp(x) (exp(*(x)))
215 #define d_imag(z) (cimag(Cd(z)))
216 #define r_imag(z) (cimagf(Cf(z)))
217 #define d_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
218 #define r_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
219 #define d_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
220 #define r_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
221 #define d_log(x) (log(*(x)))
222 #define d_mod(x, y) (fmod(*(x), *(y)))
223 #define u_nint(__x) ((__x)>=0 ? floor((__x) + .5) : -floor(.5 - (__x)))
224 #define d_nint(x) u_nint(*(x))
225 #define u_sign(__a,__b) ((__b) >= 0 ? ((__a) >= 0 ? (__a) : -(__a)) : -((__a) >= 0 ? (__a) : -(__a)))
226 #define d_sign(a,b) u_sign(*(a),*(b))
227 #define r_sign(a,b) u_sign(*(a),*(b))
228 #define d_sin(x) (sin(*(x)))
229 #define d_sinh(x) (sinh(*(x)))
230 #define d_sqrt(x) (sqrt(*(x)))
231 #define d_tan(x) (tan(*(x)))
232 #define d_tanh(x) (tanh(*(x)))
233 #define i_abs(x) abs(*(x))
234 #define i_dnnt(x) ((integer)u_nint(*(x)))
235 #define i_len(s, n) (n)
236 #define i_nint(x) ((integer)u_nint(*(x)))
237 #define i_sign(a,b) ((integer)u_sign((integer)*(a),(integer)*(b)))
238 #define pow_dd(ap, bp) ( pow(*(ap), *(bp)))
239 #define pow_si(B,E) spow_ui(*(B),*(E))
240 #define pow_ri(B,E) spow_ui(*(B),*(E))
241 #define pow_di(B,E) dpow_ui(*(B),*(E))
242 #define pow_zi(p, a, b) {pCd(p) = zpow_ui(Cd(a), *(b));}
243 #define pow_ci(p, a, b) {pCf(p) = cpow_ui(Cf(a), *(b));}
244 #define pow_zz(R,A,B) {pCd(R) = cpow(Cd(A),*(B));}
245 #define s_cat(lpp, rpp, rnp, np, llp) { ftnlen i, nc, ll; char *f__rp, *lp; ll = (llp); lp = (lpp); for(i=0; i < (int)*(np); ++i) { nc = ll; if((rnp)[i] < nc) nc = (rnp)[i]; ll -= nc; f__rp = (rpp)[i]; while(--nc >= 0) *lp++ = *(f__rp)++; } while(--ll >= 0) *lp++ = ' '; }
246 #define s_cmp(a,b,c,d) ((integer)strncmp((a),(b),f2cmin((c),(d))))
247 #define s_copy(A,B,C,D) { int __i,__m; for (__i=0, __m=f2cmin((C),(D)); __i<__m && (B)[__i] != 0; ++__i) (A)[__i] = (B)[__i]; }
248 #define sig_die(s, kill) { exit(1); }
249 #define s_stop(s, n) {exit(0);}
250 static char junk[] = "\n@(#)LIBF77 VERSION 19990503\n";
251 #define z_abs(z) (cabs(Cd(z)))
252 #define z_exp(R, Z) {pCd(R) = cexp(Cd(Z));}
253 #define z_sqrt(R, Z) {pCd(R) = csqrt(Cd(Z));}
254 #define myexit_() break;
255 #define mycycle() continue;
256 #define myceiling(w) {ceil(w)}
257 #define myhuge(w) {HUGE_VAL}
258 //#define mymaxloc_(w,s,e,n) {if (sizeof(*(w)) == sizeof(double)) dmaxloc_((w),*(s),*(e),n); else dmaxloc_((w),*(s),*(e),n);}
259 #define mymaxloc(w,s,e,n) {dmaxloc_(w,*(s),*(e),n)}
261 /* procedure parameter types for -A and -C++ */
263 #define F2C_proc_par_types 1
265 typedef logical (*L_fp)(...);
267 typedef logical (*L_fp)();
270 static float spow_ui(float x, integer n) {
271 float pow=1.0; unsigned long int u;
273 if(n < 0) n = -n, x = 1/x;
282 static double dpow_ui(double x, integer n) {
283 double pow=1.0; unsigned long int u;
285 if(n < 0) n = -n, x = 1/x;
295 static _Fcomplex cpow_ui(complex x, integer n) {
296 complex pow={1.0,0.0}; unsigned long int u;
298 if(n < 0) n = -n, x.r = 1/x.r, x.i=1/x.i;
300 if(u & 01) pow.r *= x.r, pow.i *= x.i;
301 if(u >>= 1) x.r *= x.r, x.i *= x.i;
305 _Fcomplex p={pow.r, pow.i};
309 static _Complex float cpow_ui(_Complex float x, integer n) {
310 _Complex float pow=1.0; unsigned long int u;
312 if(n < 0) n = -n, x = 1/x;
323 static _Dcomplex zpow_ui(_Dcomplex x, integer n) {
324 _Dcomplex pow={1.0,0.0}; unsigned long int u;
326 if(n < 0) n = -n, x._Val[0] = 1/x._Val[0], x._Val[1] =1/x._Val[1];
328 if(u & 01) pow._Val[0] *= x._Val[0], pow._Val[1] *= x._Val[1];
329 if(u >>= 1) x._Val[0] *= x._Val[0], x._Val[1] *= x._Val[1];
333 _Dcomplex p = {pow._Val[0], pow._Val[1]};
337 static _Complex double zpow_ui(_Complex double x, integer n) {
338 _Complex double pow=1.0; unsigned long int u;
340 if(n < 0) n = -n, x = 1/x;
350 static integer pow_ii(integer x, integer n) {
351 integer pow; unsigned long int u;
353 if (n == 0 || x == 1) pow = 1;
354 else if (x != -1) pow = x == 0 ? 1/x : 0;
357 if ((n > 0) || !(n == 0 || x == 1 || x != -1)) {
367 static integer dmaxloc_(double *w, integer s, integer e, integer *n)
369 double m; integer i, mi;
370 for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
371 if (w[i-1]>m) mi=i ,m=w[i-1];
374 static integer smaxloc_(float *w, integer s, integer e, integer *n)
376 float m; integer i, mi;
377 for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
378 if (w[i-1]>m) mi=i ,m=w[i-1];
381 static inline void cdotc_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
382 integer n = *n_, incx = *incx_, incy = *incy_, i;
384 _Fcomplex zdotc = {0.0, 0.0};
385 if (incx == 1 && incy == 1) {
386 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
387 zdotc._Val[0] += conjf(Cf(&x[i]))._Val[0] * Cf(&y[i])._Val[0];
388 zdotc._Val[1] += conjf(Cf(&x[i]))._Val[1] * Cf(&y[i])._Val[1];
391 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
392 zdotc._Val[0] += conjf(Cf(&x[i*incx]))._Val[0] * Cf(&y[i*incy])._Val[0];
393 zdotc._Val[1] += conjf(Cf(&x[i*incx]))._Val[1] * Cf(&y[i*incy])._Val[1];
399 _Complex float zdotc = 0.0;
400 if (incx == 1 && incy == 1) {
401 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
402 zdotc += conjf(Cf(&x[i])) * Cf(&y[i]);
405 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
406 zdotc += conjf(Cf(&x[i*incx])) * Cf(&y[i*incy]);
412 static inline void zdotc_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
413 integer n = *n_, incx = *incx_, incy = *incy_, i;
415 _Dcomplex zdotc = {0.0, 0.0};
416 if (incx == 1 && incy == 1) {
417 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
418 zdotc._Val[0] += conj(Cd(&x[i]))._Val[0] * Cd(&y[i])._Val[0];
419 zdotc._Val[1] += conj(Cd(&x[i]))._Val[1] * Cd(&y[i])._Val[1];
422 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
423 zdotc._Val[0] += conj(Cd(&x[i*incx]))._Val[0] * Cd(&y[i*incy])._Val[0];
424 zdotc._Val[1] += conj(Cd(&x[i*incx]))._Val[1] * Cd(&y[i*incy])._Val[1];
430 _Complex double zdotc = 0.0;
431 if (incx == 1 && incy == 1) {
432 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
433 zdotc += conj(Cd(&x[i])) * Cd(&y[i]);
436 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
437 zdotc += conj(Cd(&x[i*incx])) * Cd(&y[i*incy]);
443 static inline void cdotu_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
444 integer n = *n_, incx = *incx_, incy = *incy_, i;
446 _Fcomplex zdotc = {0.0, 0.0};
447 if (incx == 1 && incy == 1) {
448 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
449 zdotc._Val[0] += Cf(&x[i])._Val[0] * Cf(&y[i])._Val[0];
450 zdotc._Val[1] += Cf(&x[i])._Val[1] * Cf(&y[i])._Val[1];
453 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
454 zdotc._Val[0] += Cf(&x[i*incx])._Val[0] * Cf(&y[i*incy])._Val[0];
455 zdotc._Val[1] += Cf(&x[i*incx])._Val[1] * Cf(&y[i*incy])._Val[1];
461 _Complex float zdotc = 0.0;
462 if (incx == 1 && incy == 1) {
463 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
464 zdotc += Cf(&x[i]) * Cf(&y[i]);
467 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
468 zdotc += Cf(&x[i*incx]) * Cf(&y[i*incy]);
474 static inline void zdotu_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
475 integer n = *n_, incx = *incx_, incy = *incy_, i;
477 _Dcomplex zdotc = {0.0, 0.0};
478 if (incx == 1 && incy == 1) {
479 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
480 zdotc._Val[0] += Cd(&x[i])._Val[0] * Cd(&y[i])._Val[0];
481 zdotc._Val[1] += Cd(&x[i])._Val[1] * Cd(&y[i])._Val[1];
484 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
485 zdotc._Val[0] += Cd(&x[i*incx])._Val[0] * Cd(&y[i*incy])._Val[0];
486 zdotc._Val[1] += Cd(&x[i*incx])._Val[1] * Cd(&y[i*incy])._Val[1];
492 _Complex double zdotc = 0.0;
493 if (incx == 1 && incy == 1) {
494 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
495 zdotc += Cd(&x[i]) * Cd(&y[i]);
498 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
499 zdotc += Cd(&x[i*incx]) * Cd(&y[i*incy]);
505 /* -- translated by f2c (version 20000121).
506 You must link the resulting object file with the libraries:
507 -lf2c -lm (in that order)
513 /* Table of constant values */
515 static integer c__1 = 1;
516 static integer c_n1 = -1;
517 static real c_b14 = 0.f;
518 static real c_b15 = 1.f;
519 static integer c__2 = 2;
520 static integer c__3 = 3;
521 static integer c__16 = 16;
523 /* > \brief \b SGGHD3 */
525 /* =========== DOCUMENTATION =========== */
527 /* Online html documentation available at */
528 /* http://www.netlib.org/lapack/explore-html/ */
531 /* > Download SGGHRD + dependencies */
532 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/sgghd3.
535 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/sgghd3.
538 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/sgghd3.
546 /* SUBROUTINE SGGHD3( 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 /* REAL A( LDA, * ), B( LDB, * ), Q( LDQ, * ), */
552 /* $ Z( LDZ, * ), WORK( * ) */
555 /* > \par Purpose: */
560 /* > SGGHD3 reduces a pair of real matrices (A,B) to generalized upper */
561 /* > Hessenberg form using orthogonal transformations, where A is a */
562 /* > general matrix and B is upper triangular. The form of the */
563 /* > generalized eigenvalue problem is */
564 /* > A*x = lambda*B*x, */
565 /* > and B is typically made upper triangular by computing its QR */
566 /* > factorization and moving the orthogonal matrix Q to the left side */
567 /* > of the equation. */
569 /* > This subroutine simultaneously reduces A to a Hessenberg matrix H: */
571 /* > and transforms B to another upper triangular matrix T: */
573 /* > in order to reduce the problem to its standard form */
574 /* > H*y = lambda*T*y */
575 /* > where y = Z**T*x. */
577 /* > The orthogonal matrices Q and Z are determined as products of Givens */
578 /* > rotations. They may either be formed explicitly, or they may be */
579 /* > postmultiplied into input matrices Q1 and Z1, so that */
581 /* > Q1 * A * Z1**T = (Q1*Q) * H * (Z1*Z)**T */
583 /* > Q1 * B * Z1**T = (Q1*Q) * T * (Z1*Z)**T */
585 /* > If Q1 is the orthogonal matrix from the QR factorization of B in the */
586 /* > original equation A*x = lambda*B*x, then SGGHD3 reduces the original */
587 /* > problem to generalized Hessenberg form. */
589 /* > This is a blocked variant of SGGHRD, using matrix-matrix */
590 /* > multiplications for parts of the computation to enhance performance. */
596 /* > \param[in] COMPQ */
598 /* > COMPQ is CHARACTER*1 */
599 /* > = 'N': do not compute Q; */
600 /* > = 'I': Q is initialized to the unit matrix, and the */
601 /* > orthogonal matrix Q is returned; */
602 /* > = 'V': Q must contain an orthogonal matrix Q1 on entry, */
603 /* > and the product Q1*Q is returned. */
606 /* > \param[in] COMPZ */
608 /* > COMPZ is CHARACTER*1 */
609 /* > = 'N': do not compute Z; */
610 /* > = 'I': Z is initialized to the unit matrix, and the */
611 /* > orthogonal matrix Z is returned; */
612 /* > = 'V': Z must contain an orthogonal matrix Z1 on entry, */
613 /* > and the product Z1*Z is returned. */
619 /* > The order of the matrices A and B. N >= 0. */
622 /* > \param[in] ILO */
624 /* > ILO is INTEGER */
627 /* > \param[in] IHI */
629 /* > IHI is INTEGER */
631 /* > ILO and IHI mark the rows and columns of A which are to be */
632 /* > reduced. It is assumed that A is already upper triangular */
633 /* > in rows and columns 1:ILO-1 and IHI+1:N. ILO and IHI are */
634 /* > normally set by a previous call to SGGBAL; otherwise they */
635 /* > should be set to 1 and N respectively. */
636 /* > 1 <= ILO <= IHI <= N, if N > 0; ILO=1 and IHI=0, if N=0. */
639 /* > \param[in,out] A */
641 /* > A is REAL array, dimension (LDA, N) */
642 /* > On entry, the N-by-N general matrix to be reduced. */
643 /* > On exit, the upper triangle and the first subdiagonal of A */
644 /* > are overwritten with the upper Hessenberg matrix H, and the */
645 /* > rest is set to zero. */
648 /* > \param[in] LDA */
650 /* > LDA is INTEGER */
651 /* > The leading dimension of the array A. LDA >= f2cmax(1,N). */
654 /* > \param[in,out] B */
656 /* > B is REAL array, dimension (LDB, N) */
657 /* > On entry, the N-by-N upper triangular matrix B. */
658 /* > On exit, the upper triangular matrix T = Q**T B Z. The */
659 /* > elements below the diagonal are set to zero. */
662 /* > \param[in] LDB */
664 /* > LDB is INTEGER */
665 /* > The leading dimension of the array B. LDB >= f2cmax(1,N). */
668 /* > \param[in,out] Q */
670 /* > Q is REAL array, dimension (LDQ, N) */
671 /* > On entry, if COMPQ = 'V', the orthogonal matrix Q1, */
672 /* > typically from the QR factorization of B. */
673 /* > On exit, if COMPQ='I', the orthogonal matrix Q, and if */
674 /* > COMPQ = 'V', the product Q1*Q. */
675 /* > Not referenced if COMPQ='N'. */
678 /* > \param[in] LDQ */
680 /* > LDQ is INTEGER */
681 /* > The leading dimension of the array Q. */
682 /* > LDQ >= N if COMPQ='V' or 'I'; LDQ >= 1 otherwise. */
685 /* > \param[in,out] Z */
687 /* > Z is REAL array, dimension (LDZ, N) */
688 /* > On entry, if COMPZ = 'V', the orthogonal matrix Z1. */
689 /* > On exit, if COMPZ='I', the orthogonal matrix Z, and if */
690 /* > COMPZ = 'V', the product Z1*Z. */
691 /* > Not referenced if COMPZ='N'. */
694 /* > \param[in] LDZ */
696 /* > LDZ is INTEGER */
697 /* > The leading dimension of the array Z. */
698 /* > LDZ >= N if COMPZ='V' or 'I'; LDZ >= 1 otherwise. */
701 /* > \param[out] WORK */
703 /* > WORK is REAL array, dimension (LWORK) */
704 /* > On exit, if INFO = 0, WORK(1) returns the optimal LWORK. */
707 /* > \param[in] LWORK */
709 /* > LWORK is INTEGER */
710 /* > The length of the array WORK. LWORK >= 1. */
711 /* > For optimum performance LWORK >= 6*N*NB, where NB is the */
712 /* > optimal blocksize. */
714 /* > If LWORK = -1, then a workspace query is assumed; the routine */
715 /* > only calculates the optimal size of the WORK array, returns */
716 /* > this value as the first entry of the WORK array, and no error */
717 /* > message related to LWORK is issued by XERBLA. */
720 /* > \param[out] INFO */
722 /* > INFO is INTEGER */
723 /* > = 0: successful exit. */
724 /* > < 0: if INFO = -i, the i-th argument had an illegal value. */
730 /* > \author Univ. of Tennessee */
731 /* > \author Univ. of California Berkeley */
732 /* > \author Univ. of Colorado Denver */
733 /* > \author NAG Ltd. */
735 /* > \date January 2015 */
737 /* > \ingroup realOTHERcomputational */
739 /* > \par Further Details: */
740 /* ===================== */
744 /* > This routine reduces A to Hessenberg form and maintains B in */
745 /* > using a blocked variant of Moler and Stewart's original algorithm, */
746 /* > as described by Kagstrom, Kressner, Quintana-Orti, and Quintana-Orti */
750 /* ===================================================================== */
751 /* Subroutine */ int sgghd3_(char *compq, char *compz, integer *n, integer *
752 ilo, integer *ihi, real *a, integer *lda, real *b, integer *ldb, real
753 *q, integer *ldq, real *z__, integer *ldz, real *work, integer *lwork,
756 /* System generated locals */
757 integer a_dim1, a_offset, b_dim1, b_offset, q_dim1, q_offset, z_dim1,
758 z_offset, i__1, i__2, i__3, i__4, i__5, i__6, i__7, i__8;
761 /* Local variables */
763 integer cola, jcol, ierr;
765 integer jrow, topq, ppwo;
766 extern /* Subroutine */ int srot_(integer *, real *, integer *, real *,
767 integer *, real *, real *);
768 real temp1, temp2, temp3, c__;
769 integer kacc22, i__, j, k;
771 extern logical lsame_(char *, char *);
773 extern /* Subroutine */ int sgemm_(char *, char *, integer *, integer *,
774 integer *, real *, real *, integer *, real *, integer *, real *,
775 real *, integer *), sgemv_(char *, integer *,
776 integer *, real *, real *, integer *, real *, integer *, real *,
781 extern /* Subroutine */ int sorm22_(char *, char *, integer *, integer *,
782 integer *, integer *, real *, integer *, real *, integer *, real *
783 , integer *, integer *);
786 logical initz, wantz;
788 extern /* Subroutine */ int strmv_(char *, char *, char *, integer *,
789 real *, integer *, real *, integer *);
790 char compq2[1], compz2[1];
791 integer nb, jj, nh, nx, pw;
792 extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
793 extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
794 integer *, integer *, ftnlen, ftnlen);
795 extern /* Subroutine */ int sgghrd_(char *, char *, integer *, integer *,
796 integer *, real *, integer *, real *, integer *, real *, integer *
797 , real *, integer *, integer *), slaset_(char *,
798 integer *, integer *, real *, real *, real *, integer *),
799 slartg_(real *, real *, real *, real *, real *), slacpy_(char *,
800 integer *, integer *, real *, integer *, real *, integer *);
803 integer nnb, len, top, ppw, n2nb;
806 /* -- LAPACK computational routine (version 3.8.0) -- */
807 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
808 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
813 /* ===================================================================== */
816 /* Decode and test the input parameters. */
818 /* Parameter adjustments */
820 a_offset = 1 + a_dim1 * 1;
823 b_offset = 1 + b_dim1 * 1;
826 q_offset = 1 + q_dim1 * 1;
829 z_offset = 1 + z_dim1 * 1;
835 nb = ilaenv_(&c__1, "SGGHD3", " ", n, ilo, ihi, &c_n1, (ftnlen)6, (ftnlen)
839 lwkopt = f2cmax(i__1,1);
840 work[1] = (real) lwkopt;
841 initq = lsame_(compq, "I");
842 wantq = initq || lsame_(compq, "V");
843 initz = lsame_(compz, "I");
844 wantz = initz || lsame_(compz, "V");
845 lquery = *lwork == -1;
847 if (! lsame_(compq, "N") && ! wantq) {
849 } else if (! lsame_(compz, "N") && ! wantz) {
853 } else if (*ilo < 1) {
855 } else if (*ihi > *n || *ihi < *ilo - 1) {
857 } else if (*lda < f2cmax(1,*n)) {
859 } else if (*ldb < f2cmax(1,*n)) {
861 } else if (wantq && *ldq < *n || *ldq < 1) {
863 } else if (wantz && *ldz < *n || *ldz < 1) {
865 } else if (*lwork < 1 && ! lquery) {
870 xerbla_("SGGHD3", &i__1, (ftnlen)6);
876 /* Initialize Q and Z if desired. */
879 slaset_("All", n, n, &c_b14, &c_b15, &q[q_offset], ldq);
882 slaset_("All", n, n, &c_b14, &c_b15, &z__[z_offset], ldz);
885 /* Zero out lower triangle of B. */
890 slaset_("Lower", &i__1, &i__2, &c_b14, &c_b14, &b[b_dim1 + 2], ldb);
893 /* Quick return if possible */
895 nh = *ihi - *ilo + 1;
901 /* Determine the blocksize. */
903 nbmin = ilaenv_(&c__2, "SGGHD3", " ", n, ilo, ihi, &c_n1, (ftnlen)6, (
905 if (nb > 1 && nb < nh) {
907 /* Determine when to use unblocked instead of blocked code. */
910 i__1 = nb, i__2 = ilaenv_(&c__3, "SGGHD3", " ", n, ilo, ihi, &c_n1, (
911 ftnlen)6, (ftnlen)1);
912 nx = f2cmax(i__1,i__2);
915 /* Determine if workspace is large enough for blocked code. */
917 if (*lwork < lwkopt) {
919 /* Not enough workspace to use optimal NB: determine the */
920 /* minimum value of NB, and reduce NB or force use of */
921 /* unblocked code. */
924 i__1 = 2, i__2 = ilaenv_(&c__2, "SGGHD3", " ", n, ilo, ihi, &
925 c_n1, (ftnlen)6, (ftnlen)1);
926 nbmin = f2cmax(i__1,i__2);
927 if (*lwork >= *n * 6 * nbmin) {
928 nb = *lwork / (*n * 6);
936 if (nb < nbmin || nb >= nh) {
938 /* Use unblocked code below */
944 /* Use blocked code */
946 kacc22 = ilaenv_(&c__16, "SGGHD3", " ", n, ilo, ihi, &c_n1, (ftnlen)6,
951 for (jcol = *ilo; i__2 < 0 ? jcol >= i__1 : jcol <= i__1; jcol +=
954 i__3 = nb, i__4 = *ihi - jcol - 1;
955 nnb = f2cmin(i__3,i__4);
957 /* Initialize small orthogonal factors that will hold the */
958 /* accumulated Givens rotations in workspace. */
959 /* N2NB denotes the number of 2*NNB-by-2*NNB factors */
960 /* NBLST denotes the (possibly smaller) order of the last */
963 n2nb = (*ihi - jcol - 1) / nnb - 1;
964 nblst = *ihi - jcol - n2nb * nnb;
965 slaset_("All", &nblst, &nblst, &c_b14, &c_b15, &work[1], &nblst);
966 pw = nblst * nblst + 1;
968 for (i__ = 1; i__ <= i__3; ++i__) {
972 slaset_("All", &i__4, &i__5, &c_b14, &c_b15, &work[pw], &i__6);
973 pw += (nnb << 2) * nnb;
976 /* Reduce columns JCOL:JCOL+NNB-1 of A to Hessenberg form. */
978 i__3 = jcol + nnb - 1;
979 for (j = jcol; j <= i__3; ++j) {
981 /* Reduce Jth column of A. Store cosines and sines in Jth */
982 /* column of A and B, respectively. */
985 for (i__ = *ihi; i__ >= i__4; --i__) {
986 temp = a[i__ - 1 + j * a_dim1];
987 slartg_(&temp, &a[i__ + j * a_dim1], &c__, &s, &a[i__ - 1
989 a[i__ + j * a_dim1] = c__;
990 b[i__ + j * b_dim1] = s;
993 /* Accumulate Givens rotations into workspace array. */
995 ppw = (nblst + 1) * (nblst - 2) - j + jcol + 1;
997 jrow = j + n2nb * nnb + 2;
999 for (i__ = *ihi; i__ >= i__4; --i__) {
1000 c__ = a[i__ + j * a_dim1];
1001 s = b[i__ + j * b_dim1];
1002 i__5 = ppw + len - 1;
1003 for (jj = ppw; jj <= i__5; ++jj) {
1004 temp = work[jj + nblst];
1005 work[jj + nblst] = c__ * temp - s * work[jj];
1006 work[jj] = s * temp + c__ * work[jj];
1009 ppw = ppw - nblst - 1;
1012 ppwo = nblst * nblst + (nnb + j - jcol - 1 << 1) * nnb + nnb;
1016 for (jrow = j0; i__5 < 0 ? jrow >= i__4 : jrow <= i__4; jrow
1021 for (i__ = jrow + nnb - 1; i__ >= i__6; --i__) {
1022 c__ = a[i__ + j * a_dim1];
1023 s = b[i__ + j * b_dim1];
1024 i__7 = ppw + len - 1;
1025 for (jj = ppw; jj <= i__7; ++jj) {
1026 temp = work[jj + (nnb << 1)];
1027 work[jj + (nnb << 1)] = c__ * temp - s * work[jj];
1028 work[jj] = s * temp + c__ * work[jj];
1031 ppw = ppw - (nnb << 1) - 1;
1033 ppwo += (nnb << 2) * nnb;
1036 /* TOP denotes the number of top rows in A and B that will */
1037 /* not be updated during the next steps. */
1045 /* Propagate transformations through B and replace stored */
1046 /* left sines/cosines by right sines/cosines. */
1049 for (jj = *n; jj >= i__5; --jj) {
1051 /* Update JJth column of B. */
1056 for (i__ = f2cmin(i__4,*ihi); i__ >= i__6; --i__) {
1057 c__ = a[i__ + j * a_dim1];
1058 s = b[i__ + j * b_dim1];
1059 temp = b[i__ + jj * b_dim1];
1060 b[i__ + jj * b_dim1] = c__ * temp - s * b[i__ - 1 +
1062 b[i__ - 1 + jj * b_dim1] = s * temp + c__ * b[i__ - 1
1066 /* Annihilate B( JJ+1, JJ ). */
1069 temp = b[jj + 1 + (jj + 1) * b_dim1];
1070 slartg_(&temp, &b[jj + 1 + jj * b_dim1], &c__, &s, &b[
1071 jj + 1 + (jj + 1) * b_dim1]);
1072 b[jj + 1 + jj * b_dim1] = 0.f;
1074 srot_(&i__6, &b[top + 1 + (jj + 1) * b_dim1], &c__1, &
1075 b[top + 1 + jj * b_dim1], &c__1, &c__, &s);
1076 a[jj + 1 + j * a_dim1] = c__;
1077 b[jj + 1 + j * b_dim1] = -s;
1081 /* Update A by transformations from right. */
1082 /* Explicit loop unrolling provides better performance */
1083 /* compared to SLASR. */
1084 /* CALL SLASR( 'Right', 'Variable', 'Backward', IHI-TOP, */
1085 /* $ IHI-J, A( J+2, J ), B( J+2, J ), */
1086 /* $ A( TOP+1, J+1 ), LDA ) */
1088 jj = (*ihi - j - 1) % 3;
1090 for (i__ = *ihi - j - 3; i__ >= i__5; i__ += -3) {
1091 c__ = a[j + 1 + i__ + j * a_dim1];
1092 s = -b[j + 1 + i__ + j * b_dim1];
1093 c1 = a[j + 2 + i__ + j * a_dim1];
1094 s1 = -b[j + 2 + i__ + j * b_dim1];
1095 c2 = a[j + 3 + i__ + j * a_dim1];
1096 s2 = -b[j + 3 + i__ + j * b_dim1];
1099 for (k = top + 1; k <= i__6; ++k) {
1100 temp = a[k + (j + i__) * a_dim1];
1101 temp1 = a[k + (j + i__ + 1) * a_dim1];
1102 temp2 = a[k + (j + i__ + 2) * a_dim1];
1103 temp3 = a[k + (j + i__ + 3) * a_dim1];
1104 a[k + (j + i__ + 3) * a_dim1] = c2 * temp3 + s2 *
1106 temp2 = -s2 * temp3 + c2 * temp2;
1107 a[k + (j + i__ + 2) * a_dim1] = c1 * temp2 + s1 *
1109 temp1 = -s1 * temp2 + c1 * temp1;
1110 a[k + (j + i__ + 1) * a_dim1] = c__ * temp1 + s *
1112 a[k + (j + i__) * a_dim1] = -s * temp1 + c__ * temp;
1117 for (i__ = jj; i__ >= 1; --i__) {
1119 r__1 = -b[j + 1 + i__ + j * b_dim1];
1120 srot_(&i__5, &a[top + 1 + (j + i__ + 1) * a_dim1], &
1121 c__1, &a[top + 1 + (j + i__) * a_dim1], &c__1,
1122 &a[j + 1 + i__ + j * a_dim1], &r__1);
1126 /* Update (J+1)th column of A by transformations from left. */
1128 if (j < jcol + nnb - 1) {
1131 /* Multiply with the trailing accumulated orthogonal */
1132 /* matrix, which takes the form */
1138 /* where U21 is a LEN-by-LEN matrix and U12 is lower */
1141 jrow = *ihi - nblst + 1;
1142 sgemv_("Transpose", &nblst, &len, &c_b15, &work[1], &
1143 nblst, &a[jrow + (j + 1) * a_dim1], &c__1, &c_b14,
1146 i__5 = jrow + nblst - len - 1;
1147 for (i__ = jrow; i__ <= i__5; ++i__) {
1148 work[ppw] = a[i__ + (j + 1) * a_dim1];
1152 strmv_("Lower", "Transpose", "Non-unit", &i__5, &work[len
1153 * nblst + 1], &nblst, &work[pw + len], &c__1);
1155 sgemv_("Transpose", &len, &i__5, &c_b15, &work[(len + 1) *
1156 nblst - len + 1], &nblst, &a[jrow + nblst - len
1157 + (j + 1) * a_dim1], &c__1, &c_b15, &work[pw +
1160 i__5 = jrow + nblst - 1;
1161 for (i__ = jrow; i__ <= i__5; ++i__) {
1162 a[i__ + (j + 1) * a_dim1] = work[ppw];
1166 /* Multiply with the other accumulated orthogonal */
1167 /* matrices, which take the form */
1171 /* U = [ U21 U22 0 ], */
1175 /* where I denotes the (NNB-LEN)-by-(NNB-LEN) identity */
1176 /* matrix, U21 is a LEN-by-LEN upper triangular matrix */
1177 /* and U12 is an NNB-by-NNB lower triangular matrix. */
1179 ppwo = nblst * nblst + 1;
1183 for (jrow = j0; i__6 < 0 ? jrow >= i__5 : jrow <= i__5;
1186 i__4 = jrow + nnb - 1;
1187 for (i__ = jrow; i__ <= i__4; ++i__) {
1188 work[ppw] = a[i__ + (j + 1) * a_dim1];
1192 i__4 = jrow + nnb + len - 1;
1193 for (i__ = jrow + nnb; i__ <= i__4; ++i__) {
1194 work[ppw] = a[i__ + (j + 1) * a_dim1];
1198 strmv_("Upper", "Transpose", "Non-unit", &len, &work[
1199 ppwo + nnb], &i__4, &work[pw], &c__1);
1201 strmv_("Lower", "Transpose", "Non-unit", &nnb, &work[
1202 ppwo + (len << 1) * nnb], &i__4, &work[pw +
1205 sgemv_("Transpose", &nnb, &len, &c_b15, &work[ppwo], &
1206 i__4, &a[jrow + (j + 1) * a_dim1], &c__1, &
1207 c_b15, &work[pw], &c__1);
1209 sgemv_("Transpose", &len, &nnb, &c_b15, &work[ppwo + (
1210 len << 1) * nnb + nnb], &i__4, &a[jrow + nnb
1211 + (j + 1) * a_dim1], &c__1, &c_b15, &work[pw
1214 i__4 = jrow + len + nnb - 1;
1215 for (i__ = jrow; i__ <= i__4; ++i__) {
1216 a[i__ + (j + 1) * a_dim1] = work[ppw];
1219 ppwo += (nnb << 2) * nnb;
1224 /* Apply accumulated orthogonal matrices to A. */
1226 cola = *n - jcol - nnb + 1;
1227 j = *ihi - nblst + 1;
1228 sgemm_("Transpose", "No Transpose", &nblst, &cola, &nblst, &c_b15,
1229 &work[1], &nblst, &a[j + (jcol + nnb) * a_dim1], lda, &
1230 c_b14, &work[pw], &nblst);
1231 slacpy_("All", &nblst, &cola, &work[pw], &nblst, &a[j + (jcol +
1232 nnb) * a_dim1], lda);
1233 ppwo = nblst * nblst + 1;
1237 for (j = j0; i__6 < 0 ? j >= i__3 : j <= i__3; j += i__6) {
1240 /* Exploit the structure of */
1246 /* where all blocks are NNB-by-NNB, U21 is upper */
1247 /* triangular and U12 is lower triangular. */
1251 i__7 = *lwork - pw + 1;
1252 sorm22_("Left", "Transpose", &i__5, &cola, &nnb, &nnb, &
1253 work[ppwo], &i__4, &a[j + (jcol + nnb) * a_dim1],
1254 lda, &work[pw], &i__7, &ierr);
1257 /* Ignore the structure of U. */
1263 sgemm_("Transpose", "No Transpose", &i__5, &cola, &i__4, &
1264 c_b15, &work[ppwo], &i__7, &a[j + (jcol + nnb) *
1265 a_dim1], lda, &c_b14, &work[pw], &i__8);
1268 slacpy_("All", &i__5, &cola, &work[pw], &i__4, &a[j + (
1269 jcol + nnb) * a_dim1], lda);
1271 ppwo += (nnb << 2) * nnb;
1274 /* Apply accumulated orthogonal matrices to Q. */
1277 j = *ihi - nblst + 1;
1280 i__6 = 2, i__3 = j - jcol + 1;
1281 topq = f2cmax(i__6,i__3);
1282 nh = *ihi - topq + 1;
1287 sgemm_("No Transpose", "No Transpose", &nh, &nblst, &nblst, &
1288 c_b15, &q[topq + j * q_dim1], ldq, &work[1], &nblst, &
1289 c_b14, &work[pw], &nh);
1290 slacpy_("All", &nh, &nblst, &work[pw], &nh, &q[topq + j *
1292 ppwo = nblst * nblst + 1;
1296 for (j = j0; i__3 < 0 ? j >= i__6 : j <= i__6; j += i__3) {
1299 i__5 = 2, i__4 = j - jcol + 1;
1300 topq = f2cmax(i__5,i__4);
1301 nh = *ihi - topq + 1;
1305 /* Exploit the structure of U. */
1309 i__7 = *lwork - pw + 1;
1310 sorm22_("Right", "No Transpose", &nh, &i__5, &nnb, &
1311 nnb, &work[ppwo], &i__4, &q[topq + j * q_dim1]
1312 , ldq, &work[pw], &i__7, &ierr);
1315 /* Ignore the structure of U. */
1320 sgemm_("No Transpose", "No Transpose", &nh, &i__5, &
1321 i__4, &c_b15, &q[topq + j * q_dim1], ldq, &
1322 work[ppwo], &i__7, &c_b14, &work[pw], &nh);
1324 slacpy_("All", &nh, &i__5, &work[pw], &nh, &q[topq +
1327 ppwo += (nnb << 2) * nnb;
1331 /* Accumulate right Givens rotations if required. */
1333 if (wantz || top > 0) {
1335 /* Initialize small orthogonal factors that will hold the */
1336 /* accumulated Givens rotations in workspace. */
1338 slaset_("All", &nblst, &nblst, &c_b14, &c_b15, &work[1], &
1340 pw = nblst * nblst + 1;
1342 for (i__ = 1; i__ <= i__3; ++i__) {
1346 slaset_("All", &i__6, &i__5, &c_b14, &c_b15, &work[pw], &
1348 pw += (nnb << 2) * nnb;
1351 /* Accumulate Givens rotations into workspace array. */
1353 i__3 = jcol + nnb - 1;
1354 for (j = jcol; j <= i__3; ++j) {
1355 ppw = (nblst + 1) * (nblst - 2) - j + jcol + 1;
1357 jrow = j + n2nb * nnb + 2;
1359 for (i__ = *ihi; i__ >= i__6; --i__) {
1360 c__ = a[i__ + j * a_dim1];
1361 a[i__ + j * a_dim1] = 0.f;
1362 s = b[i__ + j * b_dim1];
1363 b[i__ + j * b_dim1] = 0.f;
1364 i__5 = ppw + len - 1;
1365 for (jj = ppw; jj <= i__5; ++jj) {
1366 temp = work[jj + nblst];
1367 work[jj + nblst] = c__ * temp - s * work[jj];
1368 work[jj] = s * temp + c__ * work[jj];
1371 ppw = ppw - nblst - 1;
1374 ppwo = nblst * nblst + (nnb + j - jcol - 1 << 1) * nnb +
1379 for (jrow = j0; i__5 < 0 ? jrow >= i__6 : jrow <= i__6;
1384 for (i__ = jrow + nnb - 1; i__ >= i__4; --i__) {
1385 c__ = a[i__ + j * a_dim1];
1386 a[i__ + j * a_dim1] = 0.f;
1387 s = b[i__ + j * b_dim1];
1388 b[i__ + j * b_dim1] = 0.f;
1389 i__7 = ppw + len - 1;
1390 for (jj = ppw; jj <= i__7; ++jj) {
1391 temp = work[jj + (nnb << 1)];
1392 work[jj + (nnb << 1)] = c__ * temp - s * work[
1394 work[jj] = s * temp + c__ * work[jj];
1397 ppw = ppw - (nnb << 1) - 1;
1399 ppwo += (nnb << 2) * nnb;
1404 i__3 = *ihi - jcol - 1;
1405 slaset_("Lower", &i__3, &nnb, &c_b14, &c_b14, &a[jcol + 2 +
1406 jcol * a_dim1], lda);
1407 i__3 = *ihi - jcol - 1;
1408 slaset_("Lower", &i__3, &nnb, &c_b14, &c_b14, &b[jcol + 2 +
1409 jcol * b_dim1], ldb);
1412 /* Apply accumulated orthogonal matrices to A and B. */
1415 j = *ihi - nblst + 1;
1416 sgemm_("No Transpose", "No Transpose", &top, &nblst, &nblst, &
1417 c_b15, &a[j * a_dim1 + 1], lda, &work[1], &nblst, &
1418 c_b14, &work[pw], &top);
1419 slacpy_("All", &top, &nblst, &work[pw], &top, &a[j * a_dim1 +
1421 ppwo = nblst * nblst + 1;
1425 for (j = j0; i__5 < 0 ? j >= i__3 : j <= i__3; j += i__5) {
1428 /* Exploit the structure of U. */
1432 i__7 = *lwork - pw + 1;
1433 sorm22_("Right", "No Transpose", &top, &i__6, &nnb, &
1434 nnb, &work[ppwo], &i__4, &a[j * a_dim1 + 1],
1435 lda, &work[pw], &i__7, &ierr);
1438 /* Ignore the structure of U. */
1443 sgemm_("No Transpose", "No Transpose", &top, &i__6, &
1444 i__4, &c_b15, &a[j * a_dim1 + 1], lda, &work[
1445 ppwo], &i__7, &c_b14, &work[pw], &top);
1447 slacpy_("All", &top, &i__6, &work[pw], &top, &a[j *
1450 ppwo += (nnb << 2) * nnb;
1453 j = *ihi - nblst + 1;
1454 sgemm_("No Transpose", "No Transpose", &top, &nblst, &nblst, &
1455 c_b15, &b[j * b_dim1 + 1], ldb, &work[1], &nblst, &
1456 c_b14, &work[pw], &top);
1457 slacpy_("All", &top, &nblst, &work[pw], &top, &b[j * b_dim1 +
1459 ppwo = nblst * nblst + 1;
1463 for (j = j0; i__3 < 0 ? j >= i__5 : j <= i__5; j += i__3) {
1466 /* Exploit the structure of U. */
1470 i__7 = *lwork - pw + 1;
1471 sorm22_("Right", "No Transpose", &top, &i__6, &nnb, &
1472 nnb, &work[ppwo], &i__4, &b[j * b_dim1 + 1],
1473 ldb, &work[pw], &i__7, &ierr);
1476 /* Ignore the structure of U. */
1481 sgemm_("No Transpose", "No Transpose", &top, &i__6, &
1482 i__4, &c_b15, &b[j * b_dim1 + 1], ldb, &work[
1483 ppwo], &i__7, &c_b14, &work[pw], &top);
1485 slacpy_("All", &top, &i__6, &work[pw], &top, &b[j *
1488 ppwo += (nnb << 2) * nnb;
1492 /* Apply accumulated orthogonal matrices to Z. */
1495 j = *ihi - nblst + 1;
1498 i__3 = 2, i__5 = j - jcol + 1;
1499 topq = f2cmax(i__3,i__5);
1500 nh = *ihi - topq + 1;
1505 sgemm_("No Transpose", "No Transpose", &nh, &nblst, &nblst, &
1506 c_b15, &z__[topq + j * z_dim1], ldz, &work[1], &nblst,
1507 &c_b14, &work[pw], &nh);
1508 slacpy_("All", &nh, &nblst, &work[pw], &nh, &z__[topq + j *
1510 ppwo = nblst * nblst + 1;
1514 for (j = j0; i__5 < 0 ? j >= i__3 : j <= i__3; j += i__5) {
1517 i__6 = 2, i__4 = j - jcol + 1;
1518 topq = f2cmax(i__6,i__4);
1519 nh = *ihi - topq + 1;
1523 /* Exploit the structure of U. */
1527 i__7 = *lwork - pw + 1;
1528 sorm22_("Right", "No Transpose", &nh, &i__6, &nnb, &
1529 nnb, &work[ppwo], &i__4, &z__[topq + j *
1530 z_dim1], ldz, &work[pw], &i__7, &ierr);
1533 /* Ignore the structure of U. */
1538 sgemm_("No Transpose", "No Transpose", &nh, &i__6, &
1539 i__4, &c_b15, &z__[topq + j * z_dim1], ldz, &
1540 work[ppwo], &i__7, &c_b14, &work[pw], &nh);
1542 slacpy_("All", &nh, &i__6, &work[pw], &nh, &z__[topq
1543 + j * z_dim1], ldz);
1545 ppwo += (nnb << 2) * nnb;
1551 /* Use unblocked code to reduce the rest of the matrix */
1552 /* Avoid re-initialization of modified Q and Z. */
1554 *(unsigned char *)compq2 = *(unsigned char *)compq;
1555 *(unsigned char *)compz2 = *(unsigned char *)compz;
1558 *(unsigned char *)compq2 = 'V';
1561 *(unsigned char *)compz2 = 'V';
1566 sgghrd_(compq2, compz2, n, &jcol, ihi, &a[a_offset], lda, &b[b_offset]
1567 , ldb, &q[q_offset], ldq, &z__[z_offset], ldz, &ierr);
1569 work[1] = (real) lwkopt;