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__2 = 2;
516 static integer c_n1 = -1;
517 static integer c__3 = 3;
518 static integer c__4 = 4;
519 static doublereal c_b24 = 1.;
520 static integer c__1 = 1;
521 static doublereal c_b45 = 0.;
523 /* > \brief <b> DSBEVX_2STAGE computes the eigenvalues and, optionally, the left and/or right eigenvectors for
524 OTHER matrices</b> */
526 /* @precisions fortran d -> s */
528 /* =========== DOCUMENTATION =========== */
530 /* Online html documentation available at */
531 /* http://www.netlib.org/lapack/explore-html/ */
534 /* > Download DSBEVX_2STAGE + dependencies */
535 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dsbevx_
538 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dsbevx_
541 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dsbevx_
549 /* SUBROUTINE DSBEVX_2STAGE( JOBZ, RANGE, UPLO, N, KD, AB, LDAB, Q, */
550 /* LDQ, VL, VU, IL, IU, ABSTOL, M, W, Z, */
551 /* LDZ, WORK, LWORK, IWORK, IFAIL, INFO ) */
555 /* CHARACTER JOBZ, RANGE, UPLO */
556 /* INTEGER IL, INFO, IU, KD, LDAB, LDQ, LDZ, M, N, LWORK */
557 /* DOUBLE PRECISION ABSTOL, VL, VU */
558 /* INTEGER IFAIL( * ), IWORK( * ) */
559 /* DOUBLE PRECISION AB( LDAB, * ), Q( LDQ, * ), W( * ), WORK( * ), */
563 /* > \par Purpose: */
568 /* > DSBEVX_2STAGE computes selected eigenvalues and, optionally, eigenvectors */
569 /* > of a real symmetric band matrix A using the 2stage technique for */
570 /* > the reduction to tridiagonal. Eigenvalues and eigenvectors can */
571 /* > be selected by specifying either a range of values or a range of */
572 /* > indices for the desired eigenvalues. */
578 /* > \param[in] JOBZ */
580 /* > JOBZ is CHARACTER*1 */
581 /* > = 'N': Compute eigenvalues only; */
582 /* > = 'V': Compute eigenvalues and eigenvectors. */
583 /* > Not available in this release. */
586 /* > \param[in] RANGE */
588 /* > RANGE is CHARACTER*1 */
589 /* > = 'A': all eigenvalues will be found; */
590 /* > = 'V': all eigenvalues in the half-open interval (VL,VU] */
591 /* > will be found; */
592 /* > = 'I': the IL-th through IU-th eigenvalues will be found. */
595 /* > \param[in] UPLO */
597 /* > UPLO is CHARACTER*1 */
598 /* > = 'U': Upper triangle of A is stored; */
599 /* > = 'L': Lower triangle of A is stored. */
605 /* > The order of the matrix A. N >= 0. */
608 /* > \param[in] KD */
610 /* > KD is INTEGER */
611 /* > The number of superdiagonals of the matrix A if UPLO = 'U', */
612 /* > or the number of subdiagonals if UPLO = 'L'. KD >= 0. */
615 /* > \param[in,out] AB */
617 /* > AB is DOUBLE PRECISION array, dimension (LDAB, N) */
618 /* > On entry, the upper or lower triangle of the symmetric band */
619 /* > matrix A, stored in the first KD+1 rows of the array. The */
620 /* > j-th column of A is stored in the j-th column of the array AB */
622 /* > if UPLO = 'U', AB(kd+1+i-j,j) = A(i,j) for f2cmax(1,j-kd)<=i<=j; */
623 /* > if UPLO = 'L', AB(1+i-j,j) = A(i,j) for j<=i<=f2cmin(n,j+kd). */
625 /* > On exit, AB is overwritten by values generated during the */
626 /* > reduction to tridiagonal form. If UPLO = 'U', the first */
627 /* > superdiagonal and the diagonal of the tridiagonal matrix T */
628 /* > are returned in rows KD and KD+1 of AB, and if UPLO = 'L', */
629 /* > the diagonal and first subdiagonal of T are returned in the */
630 /* > first two rows of AB. */
633 /* > \param[in] LDAB */
635 /* > LDAB is INTEGER */
636 /* > The leading dimension of the array AB. LDAB >= KD + 1. */
639 /* > \param[out] Q */
641 /* > Q is DOUBLE PRECISION array, dimension (LDQ, N) */
642 /* > If JOBZ = 'V', the N-by-N orthogonal matrix used in the */
643 /* > reduction to tridiagonal form. */
644 /* > If JOBZ = 'N', the array Q is not referenced. */
647 /* > \param[in] LDQ */
649 /* > LDQ is INTEGER */
650 /* > The leading dimension of the array Q. If JOBZ = 'V', then */
651 /* > LDQ >= f2cmax(1,N). */
654 /* > \param[in] VL */
656 /* > VL is DOUBLE PRECISION */
657 /* > If RANGE='V', the lower bound of the interval to */
658 /* > be searched for eigenvalues. VL < VU. */
659 /* > Not referenced if RANGE = 'A' or 'I'. */
662 /* > \param[in] VU */
664 /* > VU is DOUBLE PRECISION */
665 /* > If RANGE='V', the upper bound of the interval to */
666 /* > be searched for eigenvalues. VL < VU. */
667 /* > Not referenced if RANGE = 'A' or 'I'. */
670 /* > \param[in] IL */
672 /* > IL is INTEGER */
673 /* > If RANGE='I', the index of the */
674 /* > smallest eigenvalue to be returned. */
675 /* > 1 <= IL <= IU <= N, if N > 0; IL = 1 and IU = 0 if N = 0. */
676 /* > Not referenced if RANGE = 'A' or 'V'. */
679 /* > \param[in] IU */
681 /* > IU is INTEGER */
682 /* > If RANGE='I', the index of the */
683 /* > largest eigenvalue to be returned. */
684 /* > 1 <= IL <= IU <= N, if N > 0; IL = 1 and IU = 0 if N = 0. */
685 /* > Not referenced if RANGE = 'A' or 'V'. */
688 /* > \param[in] ABSTOL */
690 /* > ABSTOL is DOUBLE PRECISION */
691 /* > The absolute error tolerance for the eigenvalues. */
692 /* > An approximate eigenvalue is accepted as converged */
693 /* > when it is determined to lie in an interval [a,b] */
694 /* > of width less than or equal to */
696 /* > ABSTOL + EPS * f2cmax( |a|,|b| ) , */
698 /* > where EPS is the machine precision. If ABSTOL is less than */
699 /* > or equal to zero, then EPS*|T| will be used in its place, */
700 /* > where |T| is the 1-norm of the tridiagonal matrix obtained */
701 /* > by reducing AB to tridiagonal form. */
703 /* > Eigenvalues will be computed most accurately when ABSTOL is */
704 /* > set to twice the underflow threshold 2*DLAMCH('S'), not zero. */
705 /* > If this routine returns with INFO>0, indicating that some */
706 /* > eigenvectors did not converge, try setting ABSTOL to */
707 /* > 2*DLAMCH('S'). */
709 /* > See "Computing Small Singular Values of Bidiagonal Matrices */
710 /* > with Guaranteed High Relative Accuracy," by Demmel and */
711 /* > Kahan, LAPACK Working Note #3. */
714 /* > \param[out] M */
717 /* > The total number of eigenvalues found. 0 <= M <= N. */
718 /* > If RANGE = 'A', M = N, and if RANGE = 'I', M = IU-IL+1. */
721 /* > \param[out] W */
723 /* > W is DOUBLE PRECISION array, dimension (N) */
724 /* > The first M elements contain the selected eigenvalues in */
725 /* > ascending order. */
728 /* > \param[out] Z */
730 /* > Z is DOUBLE PRECISION array, dimension (LDZ, f2cmax(1,M)) */
731 /* > If JOBZ = 'V', then if INFO = 0, the first M columns of Z */
732 /* > contain the orthonormal eigenvectors of the matrix A */
733 /* > corresponding to the selected eigenvalues, with the i-th */
734 /* > column of Z holding the eigenvector associated with W(i). */
735 /* > If an eigenvector fails to converge, then that column of Z */
736 /* > contains the latest approximation to the eigenvector, and the */
737 /* > index of the eigenvector is returned in IFAIL. */
738 /* > If JOBZ = 'N', then Z is not referenced. */
739 /* > Note: the user must ensure that at least f2cmax(1,M) columns are */
740 /* > supplied in the array Z; if RANGE = 'V', the exact value of M */
741 /* > is not known in advance and an upper bound must be used. */
744 /* > \param[in] LDZ */
746 /* > LDZ is INTEGER */
747 /* > The leading dimension of the array Z. LDZ >= 1, and if */
748 /* > JOBZ = 'V', LDZ >= f2cmax(1,N). */
751 /* > \param[out] WORK */
753 /* > WORK is DOUBLE PRECISION array, dimension (LWORK) */
756 /* > \param[in] LWORK */
758 /* > LWORK is INTEGER */
759 /* > The length of the array WORK. LWORK >= 1, when N <= 1; */
761 /* > If JOBZ = 'N' and N > 1, LWORK must be queried. */
762 /* > LWORK = MAX(1, 7*N, dimension) where */
763 /* > dimension = (2KD+1)*N + KD*NTHREADS + 2*N */
764 /* > where KD is the size of the band. */
765 /* > NTHREADS is the number of threads used when */
766 /* > openMP compilation is enabled, otherwise =1. */
767 /* > If JOBZ = 'V' and N > 1, LWORK must be queried. Not yet available */
769 /* > If LWORK = -1, then a workspace query is assumed; the routine */
770 /* > only calculates the optimal size of the WORK array, returns */
771 /* > this value as the first entry of the WORK array, and no error */
772 /* > message related to LWORK is issued by XERBLA. */
775 /* > \param[out] IWORK */
777 /* > IWORK is INTEGER array, dimension (5*N) */
780 /* > \param[out] IFAIL */
782 /* > IFAIL is INTEGER array, dimension (N) */
783 /* > If JOBZ = 'V', then if INFO = 0, the first M elements of */
784 /* > IFAIL are zero. If INFO > 0, then IFAIL contains the */
785 /* > indices of the eigenvectors that failed to converge. */
786 /* > If JOBZ = 'N', then IFAIL is not referenced. */
789 /* > \param[out] INFO */
791 /* > INFO is INTEGER */
792 /* > = 0: successful exit. */
793 /* > < 0: if INFO = -i, the i-th argument had an illegal value. */
794 /* > > 0: if INFO = i, then i eigenvectors failed to converge. */
795 /* > Their indices are stored in array IFAIL. */
801 /* > \author Univ. of Tennessee */
802 /* > \author Univ. of California Berkeley */
803 /* > \author Univ. of Colorado Denver */
804 /* > \author NAG Ltd. */
806 /* > \date June 2016 */
808 /* > \ingroup doubleOTHEReigen */
810 /* > \par Further Details: */
811 /* ===================== */
815 /* > All details about the 2stage techniques are available in: */
817 /* > Azzam Haidar, Hatem Ltaief, and Jack Dongarra. */
818 /* > Parallel reduction to condensed forms for symmetric eigenvalue problems */
819 /* > using aggregated fine-grained and memory-aware kernels. In Proceedings */
820 /* > of 2011 International Conference for High Performance Computing, */
821 /* > Networking, Storage and Analysis (SC '11), New York, NY, USA, */
822 /* > Article 8 , 11 pages. */
823 /* > http://doi.acm.org/10.1145/2063384.2063394 */
825 /* > A. Haidar, J. Kurzak, P. Luszczek, 2013. */
826 /* > An improved parallel singular value algorithm and its implementation */
827 /* > for multicore hardware, In Proceedings of 2013 International Conference */
828 /* > for High Performance Computing, Networking, Storage and Analysis (SC '13). */
829 /* > Denver, Colorado, USA, 2013. */
830 /* > Article 90, 12 pages. */
831 /* > http://doi.acm.org/10.1145/2503210.2503292 */
833 /* > A. Haidar, R. Solca, S. Tomov, T. Schulthess and J. Dongarra. */
834 /* > A novel hybrid CPU-GPU generalized eigensolver for electronic structure */
835 /* > calculations based on fine-grained memory aware tasks. */
836 /* > International Journal of High Performance Computing Applications. */
837 /* > Volume 28 Issue 2, Pages 196-209, May 2014. */
838 /* > http://hpc.sagepub.com/content/28/2/196 */
842 /* ===================================================================== */
843 /* Subroutine */ int dsbevx_2stage_(char *jobz, char *range, char *uplo,
844 integer *n, integer *kd, doublereal *ab, integer *ldab, doublereal *q,
845 integer *ldq, doublereal *vl, doublereal *vu, integer *il, integer *
846 iu, doublereal *abstol, integer *m, doublereal *w, doublereal *z__,
847 integer *ldz, doublereal *work, integer *lwork, integer *iwork,
848 integer *ifail, integer *info)
850 /* System generated locals */
851 integer ab_dim1, ab_offset, q_dim1, q_offset, z_dim1, z_offset, i__1,
853 doublereal d__1, d__2;
855 /* Local variables */
857 extern integer ilaenv2stage_(integer *, char *, char *, integer *,
858 integer *, integer *, integer *);
861 doublereal rmin, rmax;
863 extern /* Subroutine */ int dsytrd_sb2st_(char *, char *, char *,
864 integer *, integer *, doublereal *, integer *, doublereal *,
865 doublereal *, doublereal *, integer *, doublereal *, integer *,
867 integer itmp1, i__, j, indee;
868 extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
871 extern logical lsame_(char *, char *);
872 extern /* Subroutine */ int dgemv_(char *, integer *, integer *,
873 doublereal *, doublereal *, integer *, doublereal *, integer *,
874 doublereal *, doublereal *, integer *);
878 extern /* Subroutine */ int dcopy_(integer *, doublereal *, integer *,
879 doublereal *, integer *), dswap_(integer *, doublereal *, integer
880 *, doublereal *, integer *);
886 extern doublereal dlamch_(char *);
887 logical alleig, indeig;
888 integer iscale, indibl;
889 extern doublereal dlansb_(char *, char *, integer *, integer *,
890 doublereal *, integer *, doublereal *);
891 extern /* Subroutine */ int dlascl_(char *, integer *, integer *,
892 doublereal *, doublereal *, integer *, integer *, doublereal *,
893 integer *, integer *);
895 extern /* Subroutine */ int dlacpy_(char *, integer *, integer *,
896 doublereal *, integer *, doublereal *, integer *);
898 extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
899 doublereal abstll, bignum;
901 extern /* Subroutine */ int dstein_(integer *, doublereal *, doublereal *,
902 integer *, doublereal *, integer *, integer *, doublereal *,
903 integer *, doublereal *, integer *, integer *, integer *),
904 dsterf_(integer *, doublereal *, doublereal *, integer *);
906 extern /* Subroutine */ int dstebz_(char *, char *, integer *, doublereal
907 *, doublereal *, integer *, integer *, doublereal *, doublereal *,
908 doublereal *, integer *, integer *, doublereal *, integer *,
909 integer *, doublereal *, integer *, integer *);
911 extern /* Subroutine */ int dsteqr_(char *, integer *, doublereal *,
912 doublereal *, doublereal *, integer *, doublereal *, integer *);
913 integer nsplit, llwork;
916 doublereal eps, vll, vuu;
922 /* -- LAPACK driver routine (version 3.8.0) -- */
923 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
924 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
928 /* ===================================================================== */
931 /* Test the input parameters. */
933 /* Parameter adjustments */
935 ab_offset = 1 + ab_dim1 * 1;
938 q_offset = 1 + q_dim1 * 1;
942 z_offset = 1 + z_dim1 * 1;
949 wantz = lsame_(jobz, "V");
950 alleig = lsame_(range, "A");
951 valeig = lsame_(range, "V");
952 indeig = lsame_(range, "I");
953 lower = lsame_(uplo, "L");
954 lquery = *lwork == -1;
957 if (! lsame_(jobz, "N")) {
959 } else if (! (alleig || valeig || indeig)) {
961 } else if (! (lower || lsame_(uplo, "U"))) {
965 } else if (*kd < 0) {
967 } else if (*ldab < *kd + 1) {
969 } else if (wantz && *ldq < f2cmax(1,*n)) {
973 if (*n > 0 && *vu <= *vl) {
977 if (*il < 1 || *il > f2cmax(1,*n)) {
979 } else if (*iu < f2cmin(*n,*il) || *iu > *n) {
985 if (*ldz < 1 || wantz && *ldz < *n) {
993 work[1] = (doublereal) lwmin;
995 ib = ilaenv2stage_(&c__2, "DSYTRD_SB2ST", jobz, n, kd, &c_n1, &
997 lhtrd = ilaenv2stage_(&c__3, "DSYTRD_SB2ST", jobz, n, kd, &ib, &
999 lwtrd = ilaenv2stage_(&c__4, "DSYTRD_SB2ST", jobz, n, kd, &ib, &
1001 lwmin = (*n << 1) + lhtrd + lwtrd;
1002 work[1] = (doublereal) lwmin;
1005 if (*lwork < lwmin && ! lquery) {
1012 xerbla_("DSBEVX_2STAGE ", &i__1, (ftnlen)13);
1014 } else if (lquery) {
1018 /* Quick return if possible */
1028 tmp1 = ab[ab_dim1 + 1];
1030 tmp1 = ab[*kd + 1 + ab_dim1];
1033 if (! (*vl < tmp1 && *vu >= tmp1)) {
1040 z__[z_dim1 + 1] = 1.;
1046 /* Get machine constants. */
1048 safmin = dlamch_("Safe minimum");
1049 eps = dlamch_("Precision");
1050 smlnum = safmin / eps;
1051 bignum = 1. / smlnum;
1052 rmin = sqrt(smlnum);
1054 d__1 = sqrt(bignum), d__2 = 1. / sqrt(sqrt(safmin));
1055 rmax = f2cmin(d__1,d__2);
1057 /* Scale matrix to allowable range, if necessary. */
1068 anrm = dlansb_("M", uplo, n, kd, &ab[ab_offset], ldab, &work[1]);
1069 if (anrm > 0. && anrm < rmin) {
1071 sigma = rmin / anrm;
1072 } else if (anrm > rmax) {
1074 sigma = rmax / anrm;
1078 dlascl_("B", kd, kd, &c_b24, &sigma, n, n, &ab[ab_offset], ldab,
1081 dlascl_("Q", kd, kd, &c_b24, &sigma, n, n, &ab[ab_offset], ldab,
1085 abstll = *abstol * sigma;
1093 /* Call DSYTRD_SB2ST to reduce symmetric band matrix to tridiagonal form. */
1097 indhous = inde + *n;
1098 indwrk = indhous + lhtrd;
1099 llwork = *lwork - indwrk + 1;
1101 dsytrd_sb2st_("N", jobz, uplo, n, kd, &ab[ab_offset], ldab, &work[indd],
1102 &work[inde], &work[indhous], &lhtrd, &work[indwrk], &llwork, &
1105 /* If all eigenvalues are desired and ABSTOL is less than or equal */
1106 /* to zero, then call DSTERF or SSTEQR. If this fails for some */
1107 /* eigenvalue, then try DSTEBZ. */
1111 if (*il == 1 && *iu == *n) {
1115 if ((alleig || test) && *abstol <= 0.) {
1116 dcopy_(n, &work[indd], &c__1, &w[1], &c__1);
1117 indee = indwrk + (*n << 1);
1120 dcopy_(&i__1, &work[inde], &c__1, &work[indee], &c__1);
1121 dsterf_(n, &w[1], &work[indee], info);
1123 dlacpy_("A", n, n, &q[q_offset], ldq, &z__[z_offset], ldz);
1125 dcopy_(&i__1, &work[inde], &c__1, &work[indee], &c__1);
1126 dsteqr_(jobz, n, &w[1], &work[indee], &z__[z_offset], ldz, &work[
1130 for (i__ = 1; i__ <= i__1; ++i__) {
1143 /* Otherwise, call DSTEBZ and, if eigenvectors are desired, SSTEIN. */
1146 *(unsigned char *)order = 'B';
1148 *(unsigned char *)order = 'E';
1151 indisp = indibl + *n;
1152 indiwo = indisp + *n;
1153 dstebz_(range, order, n, &vll, &vuu, il, iu, &abstll, &work[indd], &work[
1154 inde], m, &nsplit, &w[1], &iwork[indibl], &iwork[indisp], &work[
1155 indwrk], &iwork[indiwo], info);
1158 dstein_(n, &work[indd], &work[inde], m, &w[1], &iwork[indibl], &iwork[
1159 indisp], &z__[z_offset], ldz, &work[indwrk], &iwork[indiwo], &
1162 /* Apply orthogonal matrix used in reduction to tridiagonal */
1163 /* form to eigenvectors returned by DSTEIN. */
1166 for (j = 1; j <= i__1; ++j) {
1167 dcopy_(n, &z__[j * z_dim1 + 1], &c__1, &work[1], &c__1);
1168 dgemv_("N", n, n, &c_b24, &q[q_offset], ldq, &work[1], &c__1, &
1169 c_b45, &z__[j * z_dim1 + 1], &c__1);
1174 /* If matrix was scaled, then rescale eigenvalues appropriately. */
1184 dscal_(&imax, &d__1, &w[1], &c__1);
1187 /* If eigenvalues are not in order, then sort them, along with */
1192 for (j = 1; j <= i__1; ++j) {
1196 for (jj = j + 1; jj <= i__2; ++jj) {
1205 itmp1 = iwork[indibl + i__ - 1];
1207 iwork[indibl + i__ - 1] = iwork[indibl + j - 1];
1209 iwork[indibl + j - 1] = itmp1;
1210 dswap_(n, &z__[i__ * z_dim1 + 1], &c__1, &z__[j * z_dim1 + 1],
1214 ifail[i__] = ifail[j];
1222 /* Set WORK(1) to optimal workspace size. */
1224 work[1] = (doublereal) lwmin;
1228 /* End of DSBEVX_2STAGE */
1230 } /* dsbevx_2stage__ */