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)
514 /* Table of constant values */
516 static doublecomplex c_b1 = {1.,0.};
517 static doublecomplex c_b2 = {0.,0.};
519 /* > \brief \b ZHETRI2X */
521 /* =========== DOCUMENTATION =========== */
523 /* Online html documentation available at */
524 /* http://www.netlib.org/lapack/explore-html/ */
527 /* > Download ZHETRI2X + dependencies */
528 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/zhetri2
531 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/zhetri2
534 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/zhetri2
542 /* SUBROUTINE ZHETRI2X( UPLO, N, A, LDA, IPIV, WORK, NB, INFO ) */
545 /* INTEGER INFO, LDA, N, NB */
546 /* INTEGER IPIV( * ) */
547 /* COMPLEX*16 A( LDA, * ), WORK( N+NB+1,* ) */
550 /* > \par Purpose: */
555 /* > ZHETRI2X computes the inverse of a COMPLEX*16 Hermitian indefinite matrix */
556 /* > A using the factorization A = U*D*U**H or A = L*D*L**H computed by */
563 /* > \param[in] UPLO */
565 /* > UPLO is CHARACTER*1 */
566 /* > Specifies whether the details of the factorization are stored */
567 /* > as an upper or lower triangular matrix. */
568 /* > = 'U': Upper triangular, form is A = U*D*U**H; */
569 /* > = 'L': Lower triangular, form is A = L*D*L**H. */
575 /* > The order of the matrix A. N >= 0. */
578 /* > \param[in,out] A */
580 /* > A is COMPLEX*16 array, dimension (LDA,N) */
581 /* > On entry, the NNB diagonal matrix D and the multipliers */
582 /* > used to obtain the factor U or L as computed by ZHETRF. */
584 /* > On exit, if INFO = 0, the (symmetric) inverse of the original */
585 /* > matrix. If UPLO = 'U', the upper triangular part of the */
586 /* > inverse is formed and the part of A below the diagonal is not */
587 /* > referenced; if UPLO = 'L' the lower triangular part of the */
588 /* > inverse is formed and the part of A above the diagonal is */
589 /* > not referenced. */
592 /* > \param[in] LDA */
594 /* > LDA is INTEGER */
595 /* > The leading dimension of the array A. LDA >= f2cmax(1,N). */
598 /* > \param[in] IPIV */
600 /* > IPIV is INTEGER array, dimension (N) */
601 /* > Details of the interchanges and the NNB structure of D */
602 /* > as determined by ZHETRF. */
605 /* > \param[out] WORK */
607 /* > WORK is COMPLEX*16 array, dimension (N+NB+1,NB+3) */
610 /* > \param[in] NB */
612 /* > NB is INTEGER */
616 /* > \param[out] INFO */
618 /* > INFO is INTEGER */
619 /* > = 0: successful exit */
620 /* > < 0: if INFO = -i, the i-th argument had an illegal value */
621 /* > > 0: if INFO = i, D(i,i) = 0; the matrix is singular and its */
622 /* > inverse could not be computed. */
628 /* > \author Univ. of Tennessee */
629 /* > \author Univ. of California Berkeley */
630 /* > \author Univ. of Colorado Denver */
631 /* > \author NAG Ltd. */
633 /* > \date June 2017 */
635 /* > \ingroup complex16HEcomputational */
637 /* ===================================================================== */
638 /* Subroutine */ int zhetri2x_(char *uplo, integer *n, doublecomplex *a,
639 integer *lda, integer *ipiv, doublecomplex *work, integer *nb,
642 /* System generated locals */
643 integer a_dim1, a_offset, work_dim1, work_offset, i__1, i__2, i__3, i__4,
646 doublecomplex z__1, z__2, z__3;
648 /* Local variables */
651 extern /* Subroutine */ int zheswapr_(char *, integer *, doublecomplex *,
652 integer *, integer *, integer *);
656 extern logical lsame_(char *, char *);
658 extern /* Subroutine */ int zgemm_(char *, char *, integer *, integer *,
659 integer *, doublecomplex *, doublecomplex *, integer *,
660 doublecomplex *, integer *, doublecomplex *, doublecomplex *,
664 extern /* Subroutine */ int ztrmm_(char *, char *, char *, char *,
665 integer *, integer *, doublecomplex *, doublecomplex *, integer *,
666 doublecomplex *, integer *);
667 doublecomplex ak, u01_i_j__;
669 doublecomplex u11_i_j__;
671 extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen), ztrtri_(
672 char *, char *, integer *, doublecomplex *, integer *, integer *);
674 doublecomplex akp1, u01_ip1_j__, u11_ip1_j__;
675 extern /* Subroutine */ int zsyconv_(char *, char *, integer *,
676 doublecomplex *, integer *, integer *, doublecomplex *, integer *);
679 /* -- LAPACK computational routine (version 3.7.1) -- */
680 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
681 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
685 /* ===================================================================== */
688 /* Test the input parameters. */
690 /* Parameter adjustments */
692 a_offset = 1 + a_dim1 * 1;
695 work_dim1 = *n + *nb + 1;
696 work_offset = 1 + work_dim1 * 1;
701 upper = lsame_(uplo, "U");
702 if (! upper && ! lsame_(uplo, "L")) {
706 } else if (*lda < f2cmax(1,*n)) {
710 /* Quick return if possible */
715 xerbla_("ZHETRI2X", &i__1, (ftnlen)8);
723 /* Workspace got Non-diag elements of D */
725 zsyconv_(uplo, "C", n, &a[a_offset], lda, &ipiv[1], &work[work_offset], &
728 /* Check that the diagonal matrix D is nonsingular. */
732 /* Upper triangular storage: examine D from bottom to top */
734 for (*info = *n; *info >= 1; --(*info)) {
735 i__1 = *info + *info * a_dim1;
736 if (ipiv[*info] > 0 && (a[i__1].r == 0. && a[i__1].i == 0.)) {
742 /* Lower triangular storage: examine D from top to bottom. */
745 for (*info = 1; *info <= i__1; ++(*info)) {
746 i__2 = *info + *info * a_dim1;
747 if (ipiv[*info] > 0 && (a[i__2].r == 0. && a[i__2].i == 0.)) {
754 /* Splitting Workspace */
755 /* U01 is a block (N,NB+1) */
756 /* The first element of U01 is in WORK(1,1) */
757 /* U11 is a block (NB+1,NB+1) */
758 /* The first element of U11 is in WORK(N+1,1) */
760 /* INVD is a block (N,2) */
761 /* The first element of INVD is in WORK(1,INVD) */
765 /* invA = P * inv(U**H)*inv(D)*inv(U)*P**H. */
767 ztrtri_(uplo, "U", n, &a[a_offset], lda, info);
769 /* inv(D) and inv(D)*inv(U) */
774 /* 1 x 1 diagonal NNB */
775 i__1 = k + invd * work_dim1;
776 i__2 = k + k * a_dim1;
777 d__1 = 1. / a[i__2].r;
778 work[i__1].r = d__1, work[i__1].i = 0.;
779 i__1 = k + (invd + 1) * work_dim1;
780 work[i__1].r = 0., work[i__1].i = 0.;
783 /* 2 x 2 diagonal NNB */
784 d__1 = z_abs(&work[k + 1 + work_dim1]);
785 t.r = d__1, t.i = 0.;
786 i__1 = k + k * a_dim1;
788 z__2.r = d__1, z__2.i = 0.;
789 z_div(&z__1, &z__2, &t);
790 ak.r = z__1.r, ak.i = z__1.i;
791 i__1 = k + 1 + (k + 1) * a_dim1;
793 z__2.r = d__1, z__2.i = 0.;
794 z_div(&z__1, &z__2, &t);
795 akp1.r = z__1.r, akp1.i = z__1.i;
796 z_div(&z__1, &work[k + 1 + work_dim1], &t);
797 akkp1.r = z__1.r, akkp1.i = z__1.i;
798 z__3.r = ak.r * akp1.r - ak.i * akp1.i, z__3.i = ak.r *
799 akp1.i + ak.i * akp1.r;
800 z__2.r = z__3.r - 1., z__2.i = z__3.i;
801 z__1.r = t.r * z__2.r - t.i * z__2.i, z__1.i = t.r * z__2.i +
803 d__.r = z__1.r, d__.i = z__1.i;
804 i__1 = k + invd * work_dim1;
805 z_div(&z__1, &akp1, &d__);
806 work[i__1].r = z__1.r, work[i__1].i = z__1.i;
807 i__1 = k + 1 + (invd + 1) * work_dim1;
808 z_div(&z__1, &ak, &d__);
809 work[i__1].r = z__1.r, work[i__1].i = z__1.i;
810 i__1 = k + (invd + 1) * work_dim1;
811 z__2.r = -akkp1.r, z__2.i = -akkp1.i;
812 z_div(&z__1, &z__2, &d__);
813 work[i__1].r = z__1.r, work[i__1].i = z__1.i;
814 i__1 = k + 1 + invd * work_dim1;
815 d_cnjg(&z__1, &work[k + (invd + 1) * work_dim1]);
816 work[i__1].r = z__1.r, work[i__1].i = z__1.i;
821 /* inv(U**H) = (inv(U))**H */
823 /* inv(U**H)*inv(D)*inv(U) */
832 /* count negative elements, */
834 for (i__ = cut + 1 - nnb; i__ <= i__1; ++i__) {
839 /* need a even number for a clear cut */
840 if (count % 2 == 1) {
849 for (i__ = 1; i__ <= i__1; ++i__) {
851 for (j = 1; j <= i__2; ++j) {
852 i__3 = i__ + j * work_dim1;
853 i__4 = i__ + (cut + j) * a_dim1;
854 work[i__3].r = a[i__4].r, work[i__3].i = a[i__4].i;
861 for (i__ = 1; i__ <= i__1; ++i__) {
862 i__2 = u11 + i__ + i__ * work_dim1;
863 work[i__2].r = 1., work[i__2].i = 0.;
865 for (j = 1; j <= i__2; ++j) {
866 i__3 = u11 + i__ + j * work_dim1;
867 work[i__3].r = 0., work[i__3].i = 0.;
870 for (j = i__ + 1; j <= i__2; ++j) {
871 i__3 = u11 + i__ + j * work_dim1;
872 i__4 = cut + i__ + (cut + j) * a_dim1;
873 work[i__3].r = a[i__4].r, work[i__3].i = a[i__4].i;
883 for (j = 1; j <= i__1; ++j) {
884 i__2 = i__ + j * work_dim1;
885 i__3 = i__ + invd * work_dim1;
886 i__4 = i__ + j * work_dim1;
887 z__1.r = work[i__3].r * work[i__4].r - work[i__3].i *
888 work[i__4].i, z__1.i = work[i__3].r * work[
889 i__4].i + work[i__3].i * work[i__4].r;
890 work[i__2].r = z__1.r, work[i__2].i = z__1.i;
895 for (j = 1; j <= i__1; ++j) {
896 i__2 = i__ + j * work_dim1;
897 u01_i_j__.r = work[i__2].r, u01_i_j__.i = work[i__2]
899 i__2 = i__ + 1 + j * work_dim1;
900 u01_ip1_j__.r = work[i__2].r, u01_ip1_j__.i = work[
902 i__2 = i__ + j * work_dim1;
903 i__3 = i__ + invd * work_dim1;
904 z__2.r = work[i__3].r * u01_i_j__.r - work[i__3].i *
905 u01_i_j__.i, z__2.i = work[i__3].r *
906 u01_i_j__.i + work[i__3].i * u01_i_j__.r;
907 i__4 = i__ + (invd + 1) * work_dim1;
908 z__3.r = work[i__4].r * u01_ip1_j__.r - work[i__4].i *
909 u01_ip1_j__.i, z__3.i = work[i__4].r *
910 u01_ip1_j__.i + work[i__4].i * u01_ip1_j__.r;
911 z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
912 work[i__2].r = z__1.r, work[i__2].i = z__1.i;
913 i__2 = i__ + 1 + j * work_dim1;
914 i__3 = i__ + 1 + invd * work_dim1;
915 z__2.r = work[i__3].r * u01_i_j__.r - work[i__3].i *
916 u01_i_j__.i, z__2.i = work[i__3].r *
917 u01_i_j__.i + work[i__3].i * u01_i_j__.r;
918 i__4 = i__ + 1 + (invd + 1) * work_dim1;
919 z__3.r = work[i__4].r * u01_ip1_j__.r - work[i__4].i *
920 u01_ip1_j__.i, z__3.i = work[i__4].r *
921 u01_ip1_j__.i + work[i__4].i * u01_ip1_j__.r;
922 z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
923 work[i__2].r = z__1.r, work[i__2].i = z__1.i;
933 if (ipiv[cut + i__] > 0) {
935 for (j = i__; j <= i__1; ++j) {
936 i__2 = u11 + i__ + j * work_dim1;
937 i__3 = cut + i__ + invd * work_dim1;
938 i__4 = u11 + i__ + j * work_dim1;
939 z__1.r = work[i__3].r * work[i__4].r - work[i__3].i *
940 work[i__4].i, z__1.i = work[i__3].r * work[
941 i__4].i + work[i__3].i * work[i__4].r;
942 work[i__2].r = z__1.r, work[i__2].i = z__1.i;
947 for (j = i__; j <= i__1; ++j) {
948 i__2 = u11 + i__ + j * work_dim1;
949 u11_i_j__.r = work[i__2].r, u11_i_j__.i = work[i__2]
951 i__2 = u11 + i__ + 1 + j * work_dim1;
952 u11_ip1_j__.r = work[i__2].r, u11_ip1_j__.i = work[
954 i__2 = u11 + i__ + j * work_dim1;
955 i__3 = cut + i__ + invd * work_dim1;
956 i__4 = u11 + i__ + j * work_dim1;
957 z__2.r = work[i__3].r * work[i__4].r - work[i__3].i *
958 work[i__4].i, z__2.i = work[i__3].r * work[
959 i__4].i + work[i__3].i * work[i__4].r;
960 i__5 = cut + i__ + (invd + 1) * work_dim1;
961 i__6 = u11 + i__ + 1 + j * work_dim1;
962 z__3.r = work[i__5].r * work[i__6].r - work[i__5].i *
963 work[i__6].i, z__3.i = work[i__5].r * work[
964 i__6].i + work[i__5].i * work[i__6].r;
965 z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
966 work[i__2].r = z__1.r, work[i__2].i = z__1.i;
967 i__2 = u11 + i__ + 1 + j * work_dim1;
968 i__3 = cut + i__ + 1 + invd * work_dim1;
969 z__2.r = work[i__3].r * u11_i_j__.r - work[i__3].i *
970 u11_i_j__.i, z__2.i = work[i__3].r *
971 u11_i_j__.i + work[i__3].i * u11_i_j__.r;
972 i__4 = cut + i__ + 1 + (invd + 1) * work_dim1;
973 z__3.r = work[i__4].r * u11_ip1_j__.r - work[i__4].i *
974 u11_ip1_j__.i, z__3.i = work[i__4].r *
975 u11_ip1_j__.i + work[i__4].i * u11_ip1_j__.r;
976 z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
977 work[i__2].r = z__1.r, work[i__2].i = z__1.i;
983 /* U11**H*invD1*U11->U11 */
986 ztrmm_("L", "U", "C", "U", &nnb, &nnb, &c_b1, &a[cut + 1 + (cut +
987 1) * a_dim1], lda, &work[u11 + 1 + work_dim1], &i__1);
990 for (i__ = 1; i__ <= i__1; ++i__) {
992 for (j = i__; j <= i__2; ++j) {
993 i__3 = cut + i__ + (cut + j) * a_dim1;
994 i__4 = u11 + i__ + j * work_dim1;
995 a[i__3].r = work[i__4].r, a[i__3].i = work[i__4].i;
999 /* U01**H*invD*U01->A(CUT+I,CUT+J) */
1001 i__1 = *n + *nb + 1;
1002 i__2 = *n + *nb + 1;
1003 zgemm_("C", "N", &nnb, &nnb, &cut, &c_b1, &a[(cut + 1) * a_dim1 +
1004 1], lda, &work[work_offset], &i__1, &c_b2, &work[u11 + 1
1005 + work_dim1], &i__2);
1007 /* U11 = U11**H*invD1*U11 + U01**H*invD*U01 */
1010 for (i__ = 1; i__ <= i__1; ++i__) {
1012 for (j = i__; j <= i__2; ++j) {
1013 i__3 = cut + i__ + (cut + j) * a_dim1;
1014 i__4 = cut + i__ + (cut + j) * a_dim1;
1015 i__5 = u11 + i__ + j * work_dim1;
1016 z__1.r = a[i__4].r + work[i__5].r, z__1.i = a[i__4].i +
1018 a[i__3].r = z__1.r, a[i__3].i = z__1.i;
1022 /* U01 = U00**H*invD0*U01 */
1024 i__1 = *n + *nb + 1;
1025 ztrmm_("L", uplo, "C", "U", &cut, &nnb, &c_b1, &a[a_offset], lda,
1026 &work[work_offset], &i__1);
1031 for (i__ = 1; i__ <= i__1; ++i__) {
1033 for (j = 1; j <= i__2; ++j) {
1034 i__3 = i__ + (cut + j) * a_dim1;
1035 i__4 = i__ + j * work_dim1;
1036 a[i__3].r = work[i__4].r, a[i__3].i = work[i__4].i;
1044 /* Apply PERMUTATIONS P and P**H: P * inv(U**H)*inv(D)*inv(U) *P**H */
1048 if (ipiv[i__] > 0) {
1051 zheswapr_(uplo, n, &a[a_offset], lda, &i__, &ip);
1054 zheswapr_(uplo, n, &a[a_offset], lda, &ip, &i__);
1061 zheswapr_(uplo, n, &a[a_offset], lda, &i__1, &ip);
1065 zheswapr_(uplo, n, &a[a_offset], lda, &ip, &i__1);
1074 /* invA = P * inv(U**H)*inv(D)*inv(U)*P**H. */
1076 ztrtri_(uplo, "U", n, &a[a_offset], lda, info);
1078 /* inv(D) and inv(D)*inv(U) */
1083 /* 1 x 1 diagonal NNB */
1084 i__1 = k + invd * work_dim1;
1085 i__2 = k + k * a_dim1;
1086 d__1 = 1. / a[i__2].r;
1087 work[i__1].r = d__1, work[i__1].i = 0.;
1088 i__1 = k + (invd + 1) * work_dim1;
1089 work[i__1].r = 0., work[i__1].i = 0.;
1092 /* 2 x 2 diagonal NNB */
1093 d__1 = z_abs(&work[k - 1 + work_dim1]);
1094 t.r = d__1, t.i = 0.;
1095 i__1 = k - 1 + (k - 1) * a_dim1;
1097 z__2.r = d__1, z__2.i = 0.;
1098 z_div(&z__1, &z__2, &t);
1099 ak.r = z__1.r, ak.i = z__1.i;
1100 i__1 = k + k * a_dim1;
1102 z__2.r = d__1, z__2.i = 0.;
1103 z_div(&z__1, &z__2, &t);
1104 akp1.r = z__1.r, akp1.i = z__1.i;
1105 z_div(&z__1, &work[k - 1 + work_dim1], &t);
1106 akkp1.r = z__1.r, akkp1.i = z__1.i;
1107 z__3.r = ak.r * akp1.r - ak.i * akp1.i, z__3.i = ak.r *
1108 akp1.i + ak.i * akp1.r;
1109 z__2.r = z__3.r - 1., z__2.i = z__3.i;
1110 z__1.r = t.r * z__2.r - t.i * z__2.i, z__1.i = t.r * z__2.i +
1112 d__.r = z__1.r, d__.i = z__1.i;
1113 i__1 = k - 1 + invd * work_dim1;
1114 z_div(&z__1, &akp1, &d__);
1115 work[i__1].r = z__1.r, work[i__1].i = z__1.i;
1116 i__1 = k + invd * work_dim1;
1117 z_div(&z__1, &ak, &d__);
1118 work[i__1].r = z__1.r, work[i__1].i = z__1.i;
1119 i__1 = k + (invd + 1) * work_dim1;
1120 z__2.r = -akkp1.r, z__2.i = -akkp1.i;
1121 z_div(&z__1, &z__2, &d__);
1122 work[i__1].r = z__1.r, work[i__1].i = z__1.i;
1123 i__1 = k - 1 + (invd + 1) * work_dim1;
1124 d_cnjg(&z__1, &work[k + (invd + 1) * work_dim1]);
1125 work[i__1].r = z__1.r, work[i__1].i = z__1.i;
1130 /* inv(U**H) = (inv(U))**H */
1132 /* inv(U**H)*inv(D)*inv(U) */
1137 if (cut + nnb >= *n) {
1141 /* count negative elements, */
1143 for (i__ = cut + 1; i__ <= i__1; ++i__) {
1144 if (ipiv[i__] < 0) {
1148 /* need a even number for a clear cut */
1149 if (count % 2 == 1) {
1154 i__1 = *n - cut - nnb;
1155 for (i__ = 1; i__ <= i__1; ++i__) {
1157 for (j = 1; j <= i__2; ++j) {
1158 i__3 = i__ + j * work_dim1;
1159 i__4 = cut + nnb + i__ + (cut + j) * a_dim1;
1160 work[i__3].r = a[i__4].r, work[i__3].i = a[i__4].i;
1165 for (i__ = 1; i__ <= i__1; ++i__) {
1166 i__2 = u11 + i__ + i__ * work_dim1;
1167 work[i__2].r = 1., work[i__2].i = 0.;
1169 for (j = i__ + 1; j <= i__2; ++j) {
1170 i__3 = u11 + i__ + j * work_dim1;
1171 work[i__3].r = 0., work[i__3].i = 0.;
1174 for (j = 1; j <= i__2; ++j) {
1175 i__3 = u11 + i__ + j * work_dim1;
1176 i__4 = cut + i__ + (cut + j) * a_dim1;
1177 work[i__3].r = a[i__4].r, work[i__3].i = a[i__4].i;
1183 i__ = *n - cut - nnb;
1185 if (ipiv[cut + nnb + i__] > 0) {
1187 for (j = 1; j <= i__1; ++j) {
1188 i__2 = i__ + j * work_dim1;
1189 i__3 = cut + nnb + i__ + invd * work_dim1;
1190 i__4 = i__ + j * work_dim1;
1191 z__1.r = work[i__3].r * work[i__4].r - work[i__3].i *
1192 work[i__4].i, z__1.i = work[i__3].r * work[
1193 i__4].i + work[i__3].i * work[i__4].r;
1194 work[i__2].r = z__1.r, work[i__2].i = z__1.i;
1199 for (j = 1; j <= i__1; ++j) {
1200 i__2 = i__ + j * work_dim1;
1201 u01_i_j__.r = work[i__2].r, u01_i_j__.i = work[i__2]
1203 i__2 = i__ - 1 + j * work_dim1;
1204 u01_ip1_j__.r = work[i__2].r, u01_ip1_j__.i = work[
1206 i__2 = i__ + j * work_dim1;
1207 i__3 = cut + nnb + i__ + invd * work_dim1;
1208 z__2.r = work[i__3].r * u01_i_j__.r - work[i__3].i *
1209 u01_i_j__.i, z__2.i = work[i__3].r *
1210 u01_i_j__.i + work[i__3].i * u01_i_j__.r;
1211 i__4 = cut + nnb + i__ + (invd + 1) * work_dim1;
1212 z__3.r = work[i__4].r * u01_ip1_j__.r - work[i__4].i *
1213 u01_ip1_j__.i, z__3.i = work[i__4].r *
1214 u01_ip1_j__.i + work[i__4].i * u01_ip1_j__.r;
1215 z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
1216 work[i__2].r = z__1.r, work[i__2].i = z__1.i;
1217 i__2 = i__ - 1 + j * work_dim1;
1218 i__3 = cut + nnb + i__ - 1 + (invd + 1) * work_dim1;
1219 z__2.r = work[i__3].r * u01_i_j__.r - work[i__3].i *
1220 u01_i_j__.i, z__2.i = work[i__3].r *
1221 u01_i_j__.i + work[i__3].i * u01_i_j__.r;
1222 i__4 = cut + nnb + i__ - 1 + invd * work_dim1;
1223 z__3.r = work[i__4].r * u01_ip1_j__.r - work[i__4].i *
1224 u01_ip1_j__.i, z__3.i = work[i__4].r *
1225 u01_ip1_j__.i + work[i__4].i * u01_ip1_j__.r;
1226 z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
1227 work[i__2].r = z__1.r, work[i__2].i = z__1.i;
1237 if (ipiv[cut + i__] > 0) {
1239 for (j = 1; j <= i__1; ++j) {
1240 i__2 = u11 + i__ + j * work_dim1;
1241 i__3 = cut + i__ + invd * work_dim1;
1242 i__4 = u11 + i__ + j * work_dim1;
1243 z__1.r = work[i__3].r * work[i__4].r - work[i__3].i *
1244 work[i__4].i, z__1.i = work[i__3].r * work[
1245 i__4].i + work[i__3].i * work[i__4].r;
1246 work[i__2].r = z__1.r, work[i__2].i = z__1.i;
1251 for (j = 1; j <= i__1; ++j) {
1252 i__2 = u11 + i__ + j * work_dim1;
1253 u11_i_j__.r = work[i__2].r, u11_i_j__.i = work[i__2]
1255 i__2 = u11 + i__ - 1 + j * work_dim1;
1256 u11_ip1_j__.r = work[i__2].r, u11_ip1_j__.i = work[
1258 i__2 = u11 + i__ + j * work_dim1;
1259 i__3 = cut + i__ + invd * work_dim1;
1260 i__4 = u11 + i__ + j * work_dim1;
1261 z__2.r = work[i__3].r * work[i__4].r - work[i__3].i *
1262 work[i__4].i, z__2.i = work[i__3].r * work[
1263 i__4].i + work[i__3].i * work[i__4].r;
1264 i__5 = cut + i__ + (invd + 1) * work_dim1;
1265 z__3.r = work[i__5].r * u11_ip1_j__.r - work[i__5].i *
1266 u11_ip1_j__.i, z__3.i = work[i__5].r *
1267 u11_ip1_j__.i + work[i__5].i * u11_ip1_j__.r;
1268 z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
1269 work[i__2].r = z__1.r, work[i__2].i = z__1.i;
1270 i__2 = u11 + i__ - 1 + j * work_dim1;
1271 i__3 = cut + i__ - 1 + (invd + 1) * work_dim1;
1272 z__2.r = work[i__3].r * u11_i_j__.r - work[i__3].i *
1273 u11_i_j__.i, z__2.i = work[i__3].r *
1274 u11_i_j__.i + work[i__3].i * u11_i_j__.r;
1275 i__4 = cut + i__ - 1 + invd * work_dim1;
1276 z__3.r = work[i__4].r * u11_ip1_j__.r - work[i__4].i *
1277 u11_ip1_j__.i, z__3.i = work[i__4].r *
1278 u11_ip1_j__.i + work[i__4].i * u11_ip1_j__.r;
1279 z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
1280 work[i__2].r = z__1.r, work[i__2].i = z__1.i;
1286 /* L11**H*invD1*L11->L11 */
1288 i__1 = *n + *nb + 1;
1289 ztrmm_("L", uplo, "C", "U", &nnb, &nnb, &c_b1, &a[cut + 1 + (cut
1290 + 1) * a_dim1], lda, &work[u11 + 1 + work_dim1], &i__1);
1293 for (i__ = 1; i__ <= i__1; ++i__) {
1295 for (j = 1; j <= i__2; ++j) {
1296 i__3 = cut + i__ + (cut + j) * a_dim1;
1297 i__4 = u11 + i__ + j * work_dim1;
1298 a[i__3].r = work[i__4].r, a[i__3].i = work[i__4].i;
1302 if (cut + nnb < *n) {
1304 /* L21**H*invD2*L21->A(CUT+I,CUT+J) */
1306 i__1 = *n - nnb - cut;
1307 i__2 = *n + *nb + 1;
1308 i__3 = *n + *nb + 1;
1309 zgemm_("C", "N", &nnb, &nnb, &i__1, &c_b1, &a[cut + nnb + 1 +
1310 (cut + 1) * a_dim1], lda, &work[work_offset], &i__2, &
1311 c_b2, &work[u11 + 1 + work_dim1], &i__3);
1313 /* L11 = L11**H*invD1*L11 + U01**H*invD*U01 */
1316 for (i__ = 1; i__ <= i__1; ++i__) {
1318 for (j = 1; j <= i__2; ++j) {
1319 i__3 = cut + i__ + (cut + j) * a_dim1;
1320 i__4 = cut + i__ + (cut + j) * a_dim1;
1321 i__5 = u11 + i__ + j * work_dim1;
1322 z__1.r = a[i__4].r + work[i__5].r, z__1.i = a[i__4].i
1324 a[i__3].r = z__1.r, a[i__3].i = z__1.i;
1328 /* L01 = L22**H*invD2*L21 */
1330 i__1 = *n - nnb - cut;
1331 i__2 = *n + *nb + 1;
1332 ztrmm_("L", uplo, "C", "U", &i__1, &nnb, &c_b1, &a[cut + nnb
1333 + 1 + (cut + nnb + 1) * a_dim1], lda, &work[
1334 work_offset], &i__2);
1336 i__1 = *n - cut - nnb;
1337 for (i__ = 1; i__ <= i__1; ++i__) {
1339 for (j = 1; j <= i__2; ++j) {
1340 i__3 = cut + nnb + i__ + (cut + j) * a_dim1;
1341 i__4 = i__ + j * work_dim1;
1342 a[i__3].r = work[i__4].r, a[i__3].i = work[i__4].i;
1347 /* L11 = L11**H*invD1*L11 */
1350 for (i__ = 1; i__ <= i__1; ++i__) {
1352 for (j = 1; j <= i__2; ++j) {
1353 i__3 = cut + i__ + (cut + j) * a_dim1;
1354 i__4 = u11 + i__ + j * work_dim1;
1355 a[i__3].r = work[i__4].r, a[i__3].i = work[i__4].i;
1365 /* Apply PERMUTATIONS P and P**H: P * inv(U**H)*inv(D)*inv(U) *P**H */
1369 if (ipiv[i__] > 0) {
1372 zheswapr_(uplo, n, &a[a_offset], lda, &i__, &ip);
1375 zheswapr_(uplo, n, &a[a_offset], lda, &ip, &i__);
1380 zheswapr_(uplo, n, &a[a_offset], lda, &i__, &ip);
1383 zheswapr_(uplo, n, &a[a_offset], lda, &ip, &i__);
1393 /* End of ZHETRI2X */