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]/Cd(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 = {1.,0.};
516 static integer c__1 = 1;
518 /* > \brief \b ZLAHEF computes a partial factorization of a complex Hermitian indefinite matrix using the Bunc
519 h-Kaufman diagonal pivoting method (blocked algorithm, calling Level 3 BLAS). */
521 /* =========== DOCUMENTATION =========== */
523 /* Online html documentation available at */
524 /* http://www.netlib.org/lapack/explore-html/ */
527 /* > Download ZLAHEF + dependencies */
528 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/zlahef.
531 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/zlahef.
534 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/zlahef.
542 /* SUBROUTINE ZLAHEF( UPLO, N, NB, KB, A, LDA, IPIV, W, LDW, INFO ) */
545 /* INTEGER INFO, KB, LDA, LDW, N, NB */
546 /* INTEGER IPIV( * ) */
547 /* COMPLEX*16 A( LDA, * ), W( LDW, * ) */
550 /* > \par Purpose: */
555 /* > ZLAHEF computes a partial factorization of a complex Hermitian */
556 /* > matrix A using the Bunch-Kaufman diagonal pivoting method. The */
557 /* > partial factorization has the form: */
559 /* > A = ( I U12 ) ( A11 0 ) ( I 0 ) if UPLO = 'U', or: */
560 /* > ( 0 U22 ) ( 0 D ) ( U12**H U22**H ) */
562 /* > A = ( L11 0 ) ( D 0 ) ( L11**H L21**H ) if UPLO = 'L' */
563 /* > ( L21 I ) ( 0 A22 ) ( 0 I ) */
565 /* > where the order of D is at most NB. The actual order is returned in */
566 /* > the argument KB, and is either NB or NB-1, or N if N <= NB. */
567 /* > Note that U**H denotes the conjugate transpose of U. */
569 /* > ZLAHEF is an auxiliary routine called by ZHETRF. It uses blocked code */
570 /* > (calling Level 3 BLAS) to update the submatrix A11 (if UPLO = 'U') or */
571 /* > A22 (if UPLO = 'L'). */
577 /* > \param[in] UPLO */
579 /* > UPLO is CHARACTER*1 */
580 /* > Specifies whether the upper or lower triangular part of the */
581 /* > Hermitian matrix A is stored: */
582 /* > = 'U': Upper triangular */
583 /* > = 'L': Lower triangular */
589 /* > The order of the matrix A. N >= 0. */
592 /* > \param[in] NB */
594 /* > NB is INTEGER */
595 /* > The maximum number of columns of the matrix A that should be */
596 /* > factored. NB should be at least 2 to allow for 2-by-2 pivot */
600 /* > \param[out] KB */
602 /* > KB is INTEGER */
603 /* > The number of columns of A that were actually factored. */
604 /* > KB is either NB-1 or NB, or N if N <= NB. */
607 /* > \param[in,out] A */
609 /* > A is COMPLEX*16 array, dimension (LDA,N) */
610 /* > On entry, the Hermitian matrix A. If UPLO = 'U', the leading */
611 /* > n-by-n upper triangular part of A contains the upper */
612 /* > triangular part of the matrix A, and the strictly lower */
613 /* > triangular part of A is not referenced. If UPLO = 'L', the */
614 /* > leading n-by-n lower triangular part of A contains the lower */
615 /* > triangular part of the matrix A, and the strictly upper */
616 /* > triangular part of A is not referenced. */
617 /* > On exit, A contains details of the partial factorization. */
620 /* > \param[in] LDA */
622 /* > LDA is INTEGER */
623 /* > The leading dimension of the array A. LDA >= f2cmax(1,N). */
626 /* > \param[out] IPIV */
628 /* > IPIV is INTEGER array, dimension (N) */
629 /* > Details of the interchanges and the block structure of D. */
631 /* > If UPLO = 'U': */
632 /* > Only the last KB elements of IPIV are set. */
634 /* > If IPIV(k) > 0, then rows and columns k and IPIV(k) were */
635 /* > interchanged and D(k,k) is a 1-by-1 diagonal block. */
637 /* > If IPIV(k) = IPIV(k-1) < 0, then rows and columns */
638 /* > k-1 and -IPIV(k) were interchanged and D(k-1:k,k-1:k) */
639 /* > is a 2-by-2 diagonal block. */
641 /* > If UPLO = 'L': */
642 /* > Only the first KB elements of IPIV are set. */
644 /* > If IPIV(k) > 0, then rows and columns k and IPIV(k) were */
645 /* > interchanged and D(k,k) is a 1-by-1 diagonal block. */
647 /* > If IPIV(k) = IPIV(k+1) < 0, then rows and columns */
648 /* > k+1 and -IPIV(k) were interchanged and D(k:k+1,k:k+1) */
649 /* > is a 2-by-2 diagonal block. */
652 /* > \param[out] W */
654 /* > W is COMPLEX*16 array, dimension (LDW,NB) */
657 /* > \param[in] LDW */
659 /* > LDW is INTEGER */
660 /* > The leading dimension of the array W. LDW >= f2cmax(1,N). */
663 /* > \param[out] INFO */
665 /* > INFO is INTEGER */
666 /* > = 0: successful exit */
667 /* > > 0: if INFO = k, D(k,k) is exactly zero. The factorization */
668 /* > has been completed, but the block diagonal matrix D is */
669 /* > exactly singular. */
675 /* > \author Univ. of Tennessee */
676 /* > \author Univ. of California Berkeley */
677 /* > \author Univ. of Colorado Denver */
678 /* > \author NAG Ltd. */
680 /* > \date December 2016 */
682 /* > \ingroup complex16HEcomputational */
684 /* > \par Contributors: */
685 /* ================== */
689 /* > December 2016, Igor Kozachenko, */
690 /* > Computer Science Division, */
691 /* > University of California, Berkeley */
694 /* ===================================================================== */
695 /* Subroutine */ int zlahef_(char *uplo, integer *n, integer *nb, integer *kb,
696 doublecomplex *a, integer *lda, integer *ipiv, doublecomplex *w,
697 integer *ldw, integer *info)
699 /* System generated locals */
700 integer a_dim1, a_offset, w_dim1, w_offset, i__1, i__2, i__3, i__4, i__5;
701 doublereal d__1, d__2, d__3, d__4;
702 doublecomplex z__1, z__2, z__3, z__4;
704 /* Local variables */
705 integer imax, jmax, j, k;
707 extern logical lsame_(char *, char *);
708 extern /* Subroutine */ int zgemm_(char *, char *, integer *, integer *,
709 integer *, doublecomplex *, doublecomplex *, integer *,
710 doublecomplex *, integer *, doublecomplex *, doublecomplex *,
713 extern /* Subroutine */ int zgemv_(char *, integer *, integer *,
714 doublecomplex *, doublecomplex *, integer *, doublecomplex *,
715 integer *, doublecomplex *, doublecomplex *, integer *);
717 extern /* Subroutine */ int zcopy_(integer *, doublecomplex *, integer *,
718 doublecomplex *, integer *), zswap_(integer *, doublecomplex *,
719 integer *, doublecomplex *, integer *);
720 doublecomplex d11, d21, d22;
721 integer jb, jj, kk, jp, kp;
724 extern /* Subroutine */ int zdscal_(integer *, doublereal *,
725 doublecomplex *, integer *);
727 extern /* Subroutine */ int zlacgv_(integer *, doublecomplex *, integer *)
729 extern integer izamax_(integer *, doublecomplex *, integer *);
734 /* -- LAPACK computational routine (version 3.7.0) -- */
735 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
736 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
740 /* ===================================================================== */
743 /* Parameter adjustments */
745 a_offset = 1 + a_dim1 * 1;
749 w_offset = 1 + w_dim1 * 1;
755 /* Initialize ALPHA for use in choosing pivot block size. */
757 alpha = (sqrt(17.) + 1.) / 8.;
759 if (lsame_(uplo, "U")) {
761 /* Factorize the trailing columns of A using the upper triangle */
762 /* of A and working backwards, and compute the matrix W = U12*D */
763 /* for use in updating A11 (note that conjg(W) is actually stored) */
765 /* K is the main loop index, decreasing from N in steps of 1 or 2 */
767 /* KW is the column of W which corresponds to column K of A */
775 if (k <= *n - *nb + 1 && *nb < *n || k < 1) {
781 /* Copy column K of A to column KW of W and update it */
784 zcopy_(&i__1, &a[k * a_dim1 + 1], &c__1, &w[kw * w_dim1 + 1], &c__1);
785 i__1 = k + kw * w_dim1;
786 i__2 = k + k * a_dim1;
788 w[i__1].r = d__1, w[i__1].i = 0.;
791 z__1.r = -1., z__1.i = 0.;
792 zgemv_("No transpose", &k, &i__1, &z__1, &a[(k + 1) * a_dim1 + 1],
793 lda, &w[k + (kw + 1) * w_dim1], ldw, &c_b1, &w[kw *
795 i__1 = k + kw * w_dim1;
796 i__2 = k + kw * w_dim1;
798 w[i__1].r = d__1, w[i__1].i = 0.;
801 /* Determine rows and columns to be interchanged and whether */
802 /* a 1-by-1 or 2-by-2 pivot block will be used */
804 i__1 = k + kw * w_dim1;
805 absakk = (d__1 = w[i__1].r, abs(d__1));
807 /* IMAX is the row-index of the largest off-diagonal element in */
808 /* column K, and COLMAX is its absolute value. */
809 /* Determine both COLMAX and IMAX. */
813 imax = izamax_(&i__1, &w[kw * w_dim1 + 1], &c__1);
814 i__1 = imax + kw * w_dim1;
815 colmax = (d__1 = w[i__1].r, abs(d__1)) + (d__2 = d_imag(&w[imax +
816 kw * w_dim1]), abs(d__2));
821 if (f2cmax(absakk,colmax) == 0.) {
823 /* Column K is zero or underflow: set INFO and continue */
829 i__1 = k + k * a_dim1;
830 i__2 = k + k * a_dim1;
832 a[i__1].r = d__1, a[i__1].i = 0.;
835 /* ============================================================ */
837 /* BEGIN pivot search */
840 if (absakk >= alpha * colmax) {
842 /* no interchange, use 1-by-1 pivot block */
847 /* BEGIN pivot search along IMAX row */
850 /* Copy column IMAX to column KW-1 of W and update it */
853 zcopy_(&i__1, &a[imax * a_dim1 + 1], &c__1, &w[(kw - 1) *
855 i__1 = imax + (kw - 1) * w_dim1;
856 i__2 = imax + imax * a_dim1;
858 w[i__1].r = d__1, w[i__1].i = 0.;
860 zcopy_(&i__1, &a[imax + (imax + 1) * a_dim1], lda, &w[imax +
861 1 + (kw - 1) * w_dim1], &c__1);
863 zlacgv_(&i__1, &w[imax + 1 + (kw - 1) * w_dim1], &c__1);
866 z__1.r = -1., z__1.i = 0.;
867 zgemv_("No transpose", &k, &i__1, &z__1, &a[(k + 1) *
868 a_dim1 + 1], lda, &w[imax + (kw + 1) * w_dim1],
869 ldw, &c_b1, &w[(kw - 1) * w_dim1 + 1], &c__1);
870 i__1 = imax + (kw - 1) * w_dim1;
871 i__2 = imax + (kw - 1) * w_dim1;
873 w[i__1].r = d__1, w[i__1].i = 0.;
876 /* JMAX is the column-index of the largest off-diagonal */
877 /* element in row IMAX, and ROWMAX is its absolute value. */
878 /* Determine only ROWMAX. */
881 jmax = imax + izamax_(&i__1, &w[imax + 1 + (kw - 1) * w_dim1],
883 i__1 = jmax + (kw - 1) * w_dim1;
884 rowmax = (d__1 = w[i__1].r, abs(d__1)) + (d__2 = d_imag(&w[
885 jmax + (kw - 1) * w_dim1]), abs(d__2));
888 jmax = izamax_(&i__1, &w[(kw - 1) * w_dim1 + 1], &c__1);
890 i__1 = jmax + (kw - 1) * w_dim1;
891 d__3 = rowmax, d__4 = (d__1 = w[i__1].r, abs(d__1)) + (
892 d__2 = d_imag(&w[jmax + (kw - 1) * w_dim1]), abs(
894 rowmax = f2cmax(d__3,d__4);
898 if (absakk >= alpha * colmax * (colmax / rowmax)) {
900 /* no interchange, use 1-by-1 pivot block */
905 } else /* if(complicated condition) */ {
906 i__1 = imax + (kw - 1) * w_dim1;
907 if ((d__1 = w[i__1].r, abs(d__1)) >= alpha * rowmax) {
909 /* interchange rows and columns K and IMAX, use 1-by-1 */
914 /* copy column KW-1 of W to column KW of W */
916 zcopy_(&k, &w[(kw - 1) * w_dim1 + 1], &c__1, &w[kw *
922 /* interchange rows and columns K-1 and IMAX, use 2-by-2 */
931 /* END pivot search along IMAX row */
935 /* END pivot search */
937 /* ============================================================ */
939 /* KK is the column of A where pivoting step stopped */
943 /* KKW is the column of W which corresponds to column KK of A */
947 /* Interchange rows and columns KP and KK. */
948 /* Updated column KP is already stored in column KKW of W. */
952 /* Copy non-updated column KK to column KP of submatrix A */
953 /* at step K. No need to copy element into column K */
954 /* (or K and K-1 for 2-by-2 pivot) of A, since these columns */
955 /* will be later overwritten. */
957 i__1 = kp + kp * a_dim1;
958 i__2 = kk + kk * a_dim1;
960 a[i__1].r = d__1, a[i__1].i = 0.;
962 zcopy_(&i__1, &a[kp + 1 + kk * a_dim1], &c__1, &a[kp + (kp +
965 zlacgv_(&i__1, &a[kp + (kp + 1) * a_dim1], lda);
968 zcopy_(&i__1, &a[kk * a_dim1 + 1], &c__1, &a[kp * a_dim1
972 /* Interchange rows KK and KP in last K+1 to N columns of A */
973 /* (columns K (or K and K-1 for 2-by-2 pivot) of A will be */
974 /* later overwritten). Interchange rows KK and KP */
975 /* in last KKW to NB columns of W. */
979 zswap_(&i__1, &a[kk + (k + 1) * a_dim1], lda, &a[kp + (k
980 + 1) * a_dim1], lda);
983 zswap_(&i__1, &w[kk + kkw * w_dim1], ldw, &w[kp + kkw *
989 /* 1-by-1 pivot block D(k): column kw of W now holds */
991 /* W(kw) = U(k)*D(k), */
993 /* where U(k) is the k-th column of U */
995 /* (1) Store subdiag. elements of column U(k) */
996 /* and 1-by-1 block D(k) in column k of A. */
997 /* (NOTE: Diagonal element U(k,k) is a UNIT element */
998 /* and not stored) */
999 /* A(k,k) := D(k,k) = W(k,kw) */
1000 /* A(1:k-1,k) := U(1:k-1,k) = W(1:k-1,kw)/D(k,k) */
1002 /* (NOTE: No need to use for Hermitian matrix */
1003 /* A( K, K ) = DBLE( W( K, K) ) to separately copy diagonal */
1004 /* element D(k,k) from W (potentially saves only one load)) */
1005 zcopy_(&k, &w[kw * w_dim1 + 1], &c__1, &a[k * a_dim1 + 1], &
1009 /* (NOTE: No need to check if A(k,k) is NOT ZERO, */
1010 /* since that was ensured earlier in pivot search: */
1011 /* case A(k,k) = 0 falls into 2x2 pivot case(4)) */
1013 i__1 = k + k * a_dim1;
1014 r1 = 1. / a[i__1].r;
1016 zdscal_(&i__1, &r1, &a[k * a_dim1 + 1], &c__1);
1018 /* (2) Conjugate column W(kw) */
1021 zlacgv_(&i__1, &w[kw * w_dim1 + 1], &c__1);
1026 /* 2-by-2 pivot block D(k): columns kw and kw-1 of W now hold */
1028 /* ( W(kw-1) W(kw) ) = ( U(k-1) U(k) )*D(k) */
1030 /* where U(k) and U(k-1) are the k-th and (k-1)-th columns */
1033 /* (1) Store U(1:k-2,k-1) and U(1:k-2,k) and 2-by-2 */
1034 /* block D(k-1:k,k-1:k) in columns k-1 and k of A. */
1035 /* (NOTE: 2-by-2 diagonal block U(k-1:k,k-1:k) is a UNIT */
1036 /* block and not stored) */
1037 /* A(k-1:k,k-1:k) := D(k-1:k,k-1:k) = W(k-1:k,kw-1:kw) */
1038 /* A(1:k-2,k-1:k) := U(1:k-2,k:k-1:k) = */
1039 /* = W(1:k-2,kw-1:kw) * ( D(k-1:k,k-1:k)**(-1) ) */
1043 /* Factor out the columns of the inverse of 2-by-2 pivot */
1044 /* block D, so that each column contains 1, to reduce the */
1045 /* number of FLOPS when we multiply panel */
1046 /* ( W(kw-1) W(kw) ) by this inverse, i.e. by D**(-1). */
1048 /* D**(-1) = ( d11 cj(d21) )**(-1) = */
1051 /* = 1/(d11*d22-|d21|**2) * ( ( d22) (-cj(d21) ) ) = */
1052 /* ( (-d21) ( d11 ) ) */
1054 /* = 1/(|d21|**2) * 1/((d11/cj(d21))*(d22/d21)-1) * */
1056 /* * ( d21*( d22/d21 ) conj(d21)*( - 1 ) ) = */
1057 /* ( ( -1 ) ( d11/conj(d21) ) ) */
1059 /* = 1/(|d21|**2) * 1/(D22*D11-1) * */
1061 /* * ( d21*( D11 ) conj(d21)*( -1 ) ) = */
1062 /* ( ( -1 ) ( D22 ) ) */
1064 /* = (1/|d21|**2) * T * ( d21*( D11 ) conj(d21)*( -1 ) ) = */
1065 /* ( ( -1 ) ( D22 ) ) */
1067 /* = ( (T/conj(d21))*( D11 ) (T/d21)*( -1 ) ) = */
1068 /* ( ( -1 ) ( D22 ) ) */
1070 /* = ( conj(D21)*( D11 ) D21*( -1 ) ) */
1071 /* ( ( -1 ) ( D22 ) ), */
1073 /* where D11 = d22/d21, */
1074 /* D22 = d11/conj(d21), */
1076 /* T = 1/(D22*D11-1). */
1078 /* (NOTE: No need to check for division by ZERO, */
1079 /* since that was ensured earlier in pivot search: */
1080 /* (a) d21 != 0, since in 2x2 pivot case(4) */
1081 /* |d21| should be larger than |d11| and |d22|; */
1082 /* (b) (D22*D11 - 1) != 0, since from (a), */
1083 /* both |D11| < 1, |D22| < 1, hence |D22*D11| << 1.) */
1085 i__1 = k - 1 + kw * w_dim1;
1086 d21.r = w[i__1].r, d21.i = w[i__1].i;
1087 d_cnjg(&z__2, &d21);
1088 z_div(&z__1, &w[k + kw * w_dim1], &z__2);
1089 d11.r = z__1.r, d11.i = z__1.i;
1090 z_div(&z__1, &w[k - 1 + (kw - 1) * w_dim1], &d21);
1091 d22.r = z__1.r, d22.i = z__1.i;
1092 z__1.r = d11.r * d22.r - d11.i * d22.i, z__1.i = d11.r *
1093 d22.i + d11.i * d22.r;
1094 t = 1. / (z__1.r - 1.);
1095 z__2.r = t, z__2.i = 0.;
1096 z_div(&z__1, &z__2, &d21);
1097 d21.r = z__1.r, d21.i = z__1.i;
1099 /* Update elements in columns A(k-1) and A(k) as */
1100 /* dot products of rows of ( W(kw-1) W(kw) ) and columns */
1104 for (j = 1; j <= i__1; ++j) {
1105 i__2 = j + (k - 1) * a_dim1;
1106 i__3 = j + (kw - 1) * w_dim1;
1107 z__3.r = d11.r * w[i__3].r - d11.i * w[i__3].i,
1108 z__3.i = d11.r * w[i__3].i + d11.i * w[i__3]
1110 i__4 = j + kw * w_dim1;
1111 z__2.r = z__3.r - w[i__4].r, z__2.i = z__3.i - w[i__4]
1113 z__1.r = d21.r * z__2.r - d21.i * z__2.i, z__1.i =
1114 d21.r * z__2.i + d21.i * z__2.r;
1115 a[i__2].r = z__1.r, a[i__2].i = z__1.i;
1116 i__2 = j + k * a_dim1;
1117 d_cnjg(&z__2, &d21);
1118 i__3 = j + kw * w_dim1;
1119 z__4.r = d22.r * w[i__3].r - d22.i * w[i__3].i,
1120 z__4.i = d22.r * w[i__3].i + d22.i * w[i__3]
1122 i__4 = j + (kw - 1) * w_dim1;
1123 z__3.r = z__4.r - w[i__4].r, z__3.i = z__4.i - w[i__4]
1125 z__1.r = z__2.r * z__3.r - z__2.i * z__3.i, z__1.i =
1126 z__2.r * z__3.i + z__2.i * z__3.r;
1127 a[i__2].r = z__1.r, a[i__2].i = z__1.i;
1132 /* Copy D(k) to A */
1134 i__1 = k - 1 + (k - 1) * a_dim1;
1135 i__2 = k - 1 + (kw - 1) * w_dim1;
1136 a[i__1].r = w[i__2].r, a[i__1].i = w[i__2].i;
1137 i__1 = k - 1 + k * a_dim1;
1138 i__2 = k - 1 + kw * w_dim1;
1139 a[i__1].r = w[i__2].r, a[i__1].i = w[i__2].i;
1140 i__1 = k + k * a_dim1;
1141 i__2 = k + kw * w_dim1;
1142 a[i__1].r = w[i__2].r, a[i__1].i = w[i__2].i;
1144 /* (2) Conjugate columns W(kw) and W(kw-1) */
1147 zlacgv_(&i__1, &w[kw * w_dim1 + 1], &c__1);
1149 zlacgv_(&i__1, &w[(kw - 1) * w_dim1 + 1], &c__1);
1155 /* Store details of the interchanges in IPIV */
1164 /* Decrease K and return to the start of the main loop */
1171 /* Update the upper triangle of A11 (= A(1:k,1:k)) as */
1173 /* A11 := A11 - U12*D*U12**H = A11 - U12*W**H */
1175 /* computing blocks of NB columns at a time (note that conjg(W) is */
1176 /* actually stored) */
1179 for (j = (k - 1) / *nb * *nb + 1; i__1 < 0 ? j >= 1 : j <= 1; j +=
1182 i__2 = *nb, i__3 = k - j + 1;
1183 jb = f2cmin(i__2,i__3);
1185 /* Update the upper triangle of the diagonal block */
1188 for (jj = j; jj <= i__2; ++jj) {
1189 i__3 = jj + jj * a_dim1;
1190 i__4 = jj + jj * a_dim1;
1192 a[i__3].r = d__1, a[i__3].i = 0.;
1195 z__1.r = -1., z__1.i = 0.;
1196 zgemv_("No transpose", &i__3, &i__4, &z__1, &a[j + (k + 1) *
1197 a_dim1], lda, &w[jj + (kw + 1) * w_dim1], ldw, &c_b1,
1198 &a[j + jj * a_dim1], &c__1);
1199 i__3 = jj + jj * a_dim1;
1200 i__4 = jj + jj * a_dim1;
1202 a[i__3].r = d__1, a[i__3].i = 0.;
1206 /* Update the rectangular superdiagonal block */
1210 z__1.r = -1., z__1.i = 0.;
1211 zgemm_("No transpose", "Transpose", &i__2, &jb, &i__3, &z__1, &a[(
1212 k + 1) * a_dim1 + 1], lda, &w[j + (kw + 1) * w_dim1], ldw,
1213 &c_b1, &a[j * a_dim1 + 1], lda);
1217 /* Put U12 in standard form by partially undoing the interchanges */
1218 /* in columns k+1:n looping backwards from k+1 to n */
1223 /* Undo the interchanges (if any) of rows JJ and JP at each */
1226 /* (Here, J is a diagonal index) */
1231 /* (Here, J is a diagonal index) */
1234 /* (NOTE: Here, J is used to determine row length. Length N-J+1 */
1235 /* of the rows to swap back doesn't include diagonal element) */
1237 if (jp != jj && j <= *n) {
1239 zswap_(&i__1, &a[jp + j * a_dim1], lda, &a[jj + j * a_dim1], lda);
1245 /* Set KB to the number of columns factorized */
1251 /* Factorize the leading columns of A using the lower triangle */
1252 /* of A and working forwards, and compute the matrix W = L21*D */
1253 /* for use in updating A22 (note that conjg(W) is actually stored) */
1255 /* K is the main loop index, increasing from 1 in steps of 1 or 2 */
1260 /* Exit from loop */
1262 if (k >= *nb && *nb < *n || k > *n) {
1268 /* Copy column K of A to column K of W and update it */
1270 i__1 = k + k * w_dim1;
1271 i__2 = k + k * a_dim1;
1273 w[i__1].r = d__1, w[i__1].i = 0.;
1276 zcopy_(&i__1, &a[k + 1 + k * a_dim1], &c__1, &w[k + 1 + k *
1281 z__1.r = -1., z__1.i = 0.;
1282 zgemv_("No transpose", &i__1, &i__2, &z__1, &a[k + a_dim1], lda, &w[k
1283 + w_dim1], ldw, &c_b1, &w[k + k * w_dim1], &c__1);
1284 i__1 = k + k * w_dim1;
1285 i__2 = k + k * w_dim1;
1287 w[i__1].r = d__1, w[i__1].i = 0.;
1289 /* Determine rows and columns to be interchanged and whether */
1290 /* a 1-by-1 or 2-by-2 pivot block will be used */
1292 i__1 = k + k * w_dim1;
1293 absakk = (d__1 = w[i__1].r, abs(d__1));
1295 /* IMAX is the row-index of the largest off-diagonal element in */
1296 /* column K, and COLMAX is its absolute value. */
1297 /* Determine both COLMAX and IMAX. */
1301 imax = k + izamax_(&i__1, &w[k + 1 + k * w_dim1], &c__1);
1302 i__1 = imax + k * w_dim1;
1303 colmax = (d__1 = w[i__1].r, abs(d__1)) + (d__2 = d_imag(&w[imax +
1304 k * w_dim1]), abs(d__2));
1309 if (f2cmax(absakk,colmax) == 0.) {
1311 /* Column K is zero or underflow: set INFO and continue */
1317 i__1 = k + k * a_dim1;
1318 i__2 = k + k * a_dim1;
1320 a[i__1].r = d__1, a[i__1].i = 0.;
1323 /* ============================================================ */
1325 /* BEGIN pivot search */
1328 if (absakk >= alpha * colmax) {
1330 /* no interchange, use 1-by-1 pivot block */
1335 /* BEGIN pivot search along IMAX row */
1338 /* Copy column IMAX to column K+1 of W and update it */
1341 zcopy_(&i__1, &a[imax + k * a_dim1], lda, &w[k + (k + 1) *
1344 zlacgv_(&i__1, &w[k + (k + 1) * w_dim1], &c__1);
1345 i__1 = imax + (k + 1) * w_dim1;
1346 i__2 = imax + imax * a_dim1;
1348 w[i__1].r = d__1, w[i__1].i = 0.;
1351 zcopy_(&i__1, &a[imax + 1 + imax * a_dim1], &c__1, &w[
1352 imax + 1 + (k + 1) * w_dim1], &c__1);
1356 z__1.r = -1., z__1.i = 0.;
1357 zgemv_("No transpose", &i__1, &i__2, &z__1, &a[k + a_dim1],
1358 lda, &w[imax + w_dim1], ldw, &c_b1, &w[k + (k + 1) *
1360 i__1 = imax + (k + 1) * w_dim1;
1361 i__2 = imax + (k + 1) * w_dim1;
1363 w[i__1].r = d__1, w[i__1].i = 0.;
1365 /* JMAX is the column-index of the largest off-diagonal */
1366 /* element in row IMAX, and ROWMAX is its absolute value. */
1367 /* Determine only ROWMAX. */
1370 jmax = k - 1 + izamax_(&i__1, &w[k + (k + 1) * w_dim1], &c__1)
1372 i__1 = jmax + (k + 1) * w_dim1;
1373 rowmax = (d__1 = w[i__1].r, abs(d__1)) + (d__2 = d_imag(&w[
1374 jmax + (k + 1) * w_dim1]), abs(d__2));
1377 jmax = imax + izamax_(&i__1, &w[imax + 1 + (k + 1) *
1380 i__1 = jmax + (k + 1) * w_dim1;
1381 d__3 = rowmax, d__4 = (d__1 = w[i__1].r, abs(d__1)) + (
1382 d__2 = d_imag(&w[jmax + (k + 1) * w_dim1]), abs(
1384 rowmax = f2cmax(d__3,d__4);
1388 if (absakk >= alpha * colmax * (colmax / rowmax)) {
1390 /* no interchange, use 1-by-1 pivot block */
1395 } else /* if(complicated condition) */ {
1396 i__1 = imax + (k + 1) * w_dim1;
1397 if ((d__1 = w[i__1].r, abs(d__1)) >= alpha * rowmax) {
1399 /* interchange rows and columns K and IMAX, use 1-by-1 */
1404 /* copy column K+1 of W to column K of W */
1407 zcopy_(&i__1, &w[k + (k + 1) * w_dim1], &c__1, &w[k +
1408 k * w_dim1], &c__1);
1413 /* interchange rows and columns K+1 and IMAX, use 2-by-2 */
1422 /* END pivot search along IMAX row */
1426 /* END pivot search */
1428 /* ============================================================ */
1430 /* KK is the column of A where pivoting step stopped */
1434 /* Interchange rows and columns KP and KK. */
1435 /* Updated column KP is already stored in column KK of W. */
1439 /* Copy non-updated column KK to column KP of submatrix A */
1440 /* at step K. No need to copy element into column K */
1441 /* (or K and K+1 for 2-by-2 pivot) of A, since these columns */
1442 /* will be later overwritten. */
1444 i__1 = kp + kp * a_dim1;
1445 i__2 = kk + kk * a_dim1;
1447 a[i__1].r = d__1, a[i__1].i = 0.;
1449 zcopy_(&i__1, &a[kk + 1 + kk * a_dim1], &c__1, &a[kp + (kk +
1452 zlacgv_(&i__1, &a[kp + (kk + 1) * a_dim1], lda);
1455 zcopy_(&i__1, &a[kp + 1 + kk * a_dim1], &c__1, &a[kp + 1
1456 + kp * a_dim1], &c__1);
1459 /* Interchange rows KK and KP in first K-1 columns of A */
1460 /* (columns K (or K and K+1 for 2-by-2 pivot) of A will be */
1461 /* later overwritten). Interchange rows KK and KP */
1462 /* in first KK columns of W. */
1466 zswap_(&i__1, &a[kk + a_dim1], lda, &a[kp + a_dim1], lda);
1468 zswap_(&kk, &w[kk + w_dim1], ldw, &w[kp + w_dim1], ldw);
1473 /* 1-by-1 pivot block D(k): column k of W now holds */
1475 /* W(k) = L(k)*D(k), */
1477 /* where L(k) is the k-th column of L */
1479 /* (1) Store subdiag. elements of column L(k) */
1480 /* and 1-by-1 block D(k) in column k of A. */
1481 /* (NOTE: Diagonal element L(k,k) is a UNIT element */
1482 /* and not stored) */
1483 /* A(k,k) := D(k,k) = W(k,k) */
1484 /* A(k+1:N,k) := L(k+1:N,k) = W(k+1:N,k)/D(k,k) */
1486 /* (NOTE: No need to use for Hermitian matrix */
1487 /* A( K, K ) = DBLE( W( K, K) ) to separately copy diagonal */
1488 /* element D(k,k) from W (potentially saves only one load)) */
1490 zcopy_(&i__1, &w[k + k * w_dim1], &c__1, &a[k + k * a_dim1], &
1494 /* (NOTE: No need to check if A(k,k) is NOT ZERO, */
1495 /* since that was ensured earlier in pivot search: */
1496 /* case A(k,k) = 0 falls into 2x2 pivot case(4)) */
1498 i__1 = k + k * a_dim1;
1499 r1 = 1. / a[i__1].r;
1501 zdscal_(&i__1, &r1, &a[k + 1 + k * a_dim1], &c__1);
1503 /* (2) Conjugate column W(k) */
1506 zlacgv_(&i__1, &w[k + 1 + k * w_dim1], &c__1);
1511 /* 2-by-2 pivot block D(k): columns k and k+1 of W now hold */
1513 /* ( W(k) W(k+1) ) = ( L(k) L(k+1) )*D(k) */
1515 /* where L(k) and L(k+1) are the k-th and (k+1)-th columns */
1518 /* (1) Store L(k+2:N,k) and L(k+2:N,k+1) and 2-by-2 */
1519 /* block D(k:k+1,k:k+1) in columns k and k+1 of A. */
1520 /* (NOTE: 2-by-2 diagonal block L(k:k+1,k:k+1) is a UNIT */
1521 /* block and not stored) */
1522 /* A(k:k+1,k:k+1) := D(k:k+1,k:k+1) = W(k:k+1,k:k+1) */
1523 /* A(k+2:N,k:k+1) := L(k+2:N,k:k+1) = */
1524 /* = W(k+2:N,k:k+1) * ( D(k:k+1,k:k+1)**(-1) ) */
1528 /* Factor out the columns of the inverse of 2-by-2 pivot */
1529 /* block D, so that each column contains 1, to reduce the */
1530 /* number of FLOPS when we multiply panel */
1531 /* ( W(kw-1) W(kw) ) by this inverse, i.e. by D**(-1). */
1533 /* D**(-1) = ( d11 cj(d21) )**(-1) = */
1536 /* = 1/(d11*d22-|d21|**2) * ( ( d22) (-cj(d21) ) ) = */
1537 /* ( (-d21) ( d11 ) ) */
1539 /* = 1/(|d21|**2) * 1/((d11/cj(d21))*(d22/d21)-1) * */
1541 /* * ( d21*( d22/d21 ) conj(d21)*( - 1 ) ) = */
1542 /* ( ( -1 ) ( d11/conj(d21) ) ) */
1544 /* = 1/(|d21|**2) * 1/(D22*D11-1) * */
1546 /* * ( d21*( D11 ) conj(d21)*( -1 ) ) = */
1547 /* ( ( -1 ) ( D22 ) ) */
1549 /* = (1/|d21|**2) * T * ( d21*( D11 ) conj(d21)*( -1 ) ) = */
1550 /* ( ( -1 ) ( D22 ) ) */
1552 /* = ( (T/conj(d21))*( D11 ) (T/d21)*( -1 ) ) = */
1553 /* ( ( -1 ) ( D22 ) ) */
1555 /* = ( conj(D21)*( D11 ) D21*( -1 ) ) */
1556 /* ( ( -1 ) ( D22 ) ), */
1558 /* where D11 = d22/d21, */
1559 /* D22 = d11/conj(d21), */
1561 /* T = 1/(D22*D11-1). */
1563 /* (NOTE: No need to check for division by ZERO, */
1564 /* since that was ensured earlier in pivot search: */
1565 /* (a) d21 != 0, since in 2x2 pivot case(4) */
1566 /* |d21| should be larger than |d11| and |d22|; */
1567 /* (b) (D22*D11 - 1) != 0, since from (a), */
1568 /* both |D11| < 1, |D22| < 1, hence |D22*D11| << 1.) */
1570 i__1 = k + 1 + k * w_dim1;
1571 d21.r = w[i__1].r, d21.i = w[i__1].i;
1572 z_div(&z__1, &w[k + 1 + (k + 1) * w_dim1], &d21);
1573 d11.r = z__1.r, d11.i = z__1.i;
1574 d_cnjg(&z__2, &d21);
1575 z_div(&z__1, &w[k + k * w_dim1], &z__2);
1576 d22.r = z__1.r, d22.i = z__1.i;
1577 z__1.r = d11.r * d22.r - d11.i * d22.i, z__1.i = d11.r *
1578 d22.i + d11.i * d22.r;
1579 t = 1. / (z__1.r - 1.);
1580 z__2.r = t, z__2.i = 0.;
1581 z_div(&z__1, &z__2, &d21);
1582 d21.r = z__1.r, d21.i = z__1.i;
1584 /* Update elements in columns A(k) and A(k+1) as */
1585 /* dot products of rows of ( W(k) W(k+1) ) and columns */
1589 for (j = k + 2; j <= i__1; ++j) {
1590 i__2 = j + k * a_dim1;
1591 d_cnjg(&z__2, &d21);
1592 i__3 = j + k * w_dim1;
1593 z__4.r = d11.r * w[i__3].r - d11.i * w[i__3].i,
1594 z__4.i = d11.r * w[i__3].i + d11.i * w[i__3]
1596 i__4 = j + (k + 1) * w_dim1;
1597 z__3.r = z__4.r - w[i__4].r, z__3.i = z__4.i - w[i__4]
1599 z__1.r = z__2.r * z__3.r - z__2.i * z__3.i, z__1.i =
1600 z__2.r * z__3.i + z__2.i * z__3.r;
1601 a[i__2].r = z__1.r, a[i__2].i = z__1.i;
1602 i__2 = j + (k + 1) * a_dim1;
1603 i__3 = j + (k + 1) * w_dim1;
1604 z__3.r = d22.r * w[i__3].r - d22.i * w[i__3].i,
1605 z__3.i = d22.r * w[i__3].i + d22.i * w[i__3]
1607 i__4 = j + k * w_dim1;
1608 z__2.r = z__3.r - w[i__4].r, z__2.i = z__3.i - w[i__4]
1610 z__1.r = d21.r * z__2.r - d21.i * z__2.i, z__1.i =
1611 d21.r * z__2.i + d21.i * z__2.r;
1612 a[i__2].r = z__1.r, a[i__2].i = z__1.i;
1617 /* Copy D(k) to A */
1619 i__1 = k + k * a_dim1;
1620 i__2 = k + k * w_dim1;
1621 a[i__1].r = w[i__2].r, a[i__1].i = w[i__2].i;
1622 i__1 = k + 1 + k * a_dim1;
1623 i__2 = k + 1 + k * w_dim1;
1624 a[i__1].r = w[i__2].r, a[i__1].i = w[i__2].i;
1625 i__1 = k + 1 + (k + 1) * a_dim1;
1626 i__2 = k + 1 + (k + 1) * w_dim1;
1627 a[i__1].r = w[i__2].r, a[i__1].i = w[i__2].i;
1629 /* (2) Conjugate columns W(k) and W(k+1) */
1632 zlacgv_(&i__1, &w[k + 1 + k * w_dim1], &c__1);
1634 zlacgv_(&i__1, &w[k + 2 + (k + 1) * w_dim1], &c__1);
1640 /* Store details of the interchanges in IPIV */
1649 /* Increase K and return to the start of the main loop */
1656 /* Update the lower triangle of A22 (= A(k:n,k:n)) as */
1658 /* A22 := A22 - L21*D*L21**H = A22 - L21*W**H */
1660 /* computing blocks of NB columns at a time (note that conjg(W) is */
1661 /* actually stored) */
1665 for (j = k; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
1667 i__3 = *nb, i__4 = *n - j + 1;
1668 jb = f2cmin(i__3,i__4);
1670 /* Update the lower triangle of the diagonal block */
1673 for (jj = j; jj <= i__3; ++jj) {
1674 i__4 = jj + jj * a_dim1;
1675 i__5 = jj + jj * a_dim1;
1677 a[i__4].r = d__1, a[i__4].i = 0.;
1680 z__1.r = -1., z__1.i = 0.;
1681 zgemv_("No transpose", &i__4, &i__5, &z__1, &a[jj + a_dim1],
1682 lda, &w[jj + w_dim1], ldw, &c_b1, &a[jj + jj * a_dim1]
1684 i__4 = jj + jj * a_dim1;
1685 i__5 = jj + jj * a_dim1;
1687 a[i__4].r = d__1, a[i__4].i = 0.;
1691 /* Update the rectangular subdiagonal block */
1694 i__3 = *n - j - jb + 1;
1696 z__1.r = -1., z__1.i = 0.;
1697 zgemm_("No transpose", "Transpose", &i__3, &jb, &i__4, &z__1,
1698 &a[j + jb + a_dim1], lda, &w[j + w_dim1], ldw, &c_b1,
1699 &a[j + jb + j * a_dim1], lda);
1704 /* Put L21 in standard form by partially undoing the interchanges */
1705 /* of rows in columns 1:k-1 looping backwards from k-1 to 1 */
1710 /* Undo the interchanges (if any) of rows JJ and JP at each */
1713 /* (Here, J is a diagonal index) */
1718 /* (Here, J is a diagonal index) */
1721 /* (NOTE: Here, J is used to determine row length. Length J */
1722 /* of the rows to swap back doesn't include diagonal element) */
1724 if (jp != jj && j >= 1) {
1725 zswap_(&j, &a[jp + a_dim1], lda, &a[jj + a_dim1], lda);
1731 /* Set KB to the number of columns factorized */