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_b21 = 1.;
520 static doublereal c_b29 = 0.;
521 static integer c__1 = 1;
523 /* > \brief <b> DSBEVD_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 DSBEVD_2STAGE + dependencies */
535 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dsbevd_
538 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dsbevd_
541 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dsbevd_
549 /* SUBROUTINE DSBEVD_2STAGE( JOBZ, UPLO, N, KD, AB, LDAB, W, Z, LDZ, */
550 /* WORK, LWORK, IWORK, LIWORK, INFO ) */
554 /* CHARACTER JOBZ, UPLO */
555 /* INTEGER INFO, KD, LDAB, LDZ, LIWORK, LWORK, N */
556 /* INTEGER IWORK( * ) */
557 /* DOUBLE PRECISION AB( LDAB, * ), W( * ), WORK( * ), Z( LDZ, * ) */
560 /* > \par Purpose: */
565 /* > DSBEVD_2STAGE computes all the eigenvalues and, optionally, eigenvectors of */
566 /* > a real symmetric band matrix A using the 2stage technique for */
567 /* > the reduction to tridiagonal. If eigenvectors are desired, it uses */
568 /* > a divide and conquer algorithm. */
570 /* > The divide and conquer algorithm makes very mild assumptions about */
571 /* > floating point arithmetic. It will work on machines with a guard */
572 /* > digit in add/subtract, or on those binary machines without guard */
573 /* > digits which subtract like the Cray X-MP, Cray Y-MP, Cray C-90, or */
574 /* > Cray-2. It could conceivably fail on hexadecimal or decimal machines */
575 /* > without guard digits, but we know of none. */
581 /* > \param[in] JOBZ */
583 /* > JOBZ is CHARACTER*1 */
584 /* > = 'N': Compute eigenvalues only; */
585 /* > = 'V': Compute eigenvalues and eigenvectors. */
586 /* > Not available in this release. */
589 /* > \param[in] UPLO */
591 /* > UPLO is CHARACTER*1 */
592 /* > = 'U': Upper triangle of A is stored; */
593 /* > = 'L': Lower triangle of A is stored. */
599 /* > The order of the matrix A. N >= 0. */
602 /* > \param[in] KD */
604 /* > KD is INTEGER */
605 /* > The number of superdiagonals of the matrix A if UPLO = 'U', */
606 /* > or the number of subdiagonals if UPLO = 'L'. KD >= 0. */
609 /* > \param[in,out] AB */
611 /* > AB is DOUBLE PRECISION array, dimension (LDAB, N) */
612 /* > On entry, the upper or lower triangle of the symmetric band */
613 /* > matrix A, stored in the first KD+1 rows of the array. The */
614 /* > j-th column of A is stored in the j-th column of the array AB */
616 /* > if UPLO = 'U', AB(kd+1+i-j,j) = A(i,j) for f2cmax(1,j-kd)<=i<=j; */
617 /* > if UPLO = 'L', AB(1+i-j,j) = A(i,j) for j<=i<=f2cmin(n,j+kd). */
619 /* > On exit, AB is overwritten by values generated during the */
620 /* > reduction to tridiagonal form. If UPLO = 'U', the first */
621 /* > superdiagonal and the diagonal of the tridiagonal matrix T */
622 /* > are returned in rows KD and KD+1 of AB, and if UPLO = 'L', */
623 /* > the diagonal and first subdiagonal of T are returned in the */
624 /* > first two rows of AB. */
627 /* > \param[in] LDAB */
629 /* > LDAB is INTEGER */
630 /* > The leading dimension of the array AB. LDAB >= KD + 1. */
633 /* > \param[out] W */
635 /* > W is DOUBLE PRECISION array, dimension (N) */
636 /* > If INFO = 0, the eigenvalues in ascending order. */
639 /* > \param[out] Z */
641 /* > Z is DOUBLE PRECISION array, dimension (LDZ, N) */
642 /* > If JOBZ = 'V', then if INFO = 0, Z contains the orthonormal */
643 /* > eigenvectors of the matrix A, with the i-th column of Z */
644 /* > holding the eigenvector associated with W(i). */
645 /* > If JOBZ = 'N', then Z is not referenced. */
648 /* > \param[in] LDZ */
650 /* > LDZ is INTEGER */
651 /* > The leading dimension of the array Z. LDZ >= 1, and if */
652 /* > JOBZ = 'V', LDZ >= f2cmax(1,N). */
655 /* > \param[out] WORK */
657 /* > WORK is DOUBLE PRECISION array, dimension LWORK */
658 /* > On exit, if INFO = 0, WORK(1) returns the optimal LWORK. */
661 /* > \param[in] LWORK */
663 /* > LWORK is INTEGER */
664 /* > The length of the array WORK. LWORK >= 1, when N <= 1; */
666 /* > If JOBZ = 'N' and N > 1, LWORK must be queried. */
667 /* > LWORK = MAX(1, dimension) where */
668 /* > dimension = (2KD+1)*N + KD*NTHREADS + N */
669 /* > where KD is the size of the band. */
670 /* > NTHREADS is the number of threads used when */
671 /* > openMP compilation is enabled, otherwise =1. */
672 /* > If JOBZ = 'V' and N > 1, LWORK must be queried. Not yet available. */
674 /* > If LWORK = -1, then a workspace query is assumed; the routine */
675 /* > only calculates the optimal sizes of the WORK and IWORK */
676 /* > arrays, returns these values as the first entries of the WORK */
677 /* > and IWORK arrays, and no error message related to LWORK or */
678 /* > LIWORK is issued by XERBLA. */
681 /* > \param[out] IWORK */
683 /* > IWORK is INTEGER array, dimension (MAX(1,LIWORK)) */
684 /* > On exit, if INFO = 0, IWORK(1) returns the optimal LIWORK. */
687 /* > \param[in] LIWORK */
689 /* > LIWORK is INTEGER */
690 /* > The dimension of the array IWORK. */
691 /* > If JOBZ = 'N' or N <= 1, LIWORK must be at least 1. */
692 /* > If JOBZ = 'V' and N > 2, LIWORK must be at least 3 + 5*N. */
694 /* > If LIWORK = -1, then a workspace query is assumed; the */
695 /* > routine only calculates the optimal sizes of the WORK and */
696 /* > IWORK arrays, returns these values as the first entries of */
697 /* > the WORK and IWORK arrays, and no error message related to */
698 /* > LWORK or LIWORK is issued by XERBLA. */
701 /* > \param[out] INFO */
703 /* > INFO is INTEGER */
704 /* > = 0: successful exit */
705 /* > < 0: if INFO = -i, the i-th argument had an illegal value */
706 /* > > 0: if INFO = i, the algorithm failed to converge; i */
707 /* > off-diagonal elements of an intermediate tridiagonal */
708 /* > form did not converge to zero. */
714 /* > \author Univ. of Tennessee */
715 /* > \author Univ. of California Berkeley */
716 /* > \author Univ. of Colorado Denver */
717 /* > \author NAG Ltd. */
719 /* > \date November 2017 */
721 /* > \ingroup doubleOTHEReigen */
723 /* > \par Further Details: */
724 /* ===================== */
728 /* > All details about the 2stage techniques are available in: */
730 /* > Azzam Haidar, Hatem Ltaief, and Jack Dongarra. */
731 /* > Parallel reduction to condensed forms for symmetric eigenvalue problems */
732 /* > using aggregated fine-grained and memory-aware kernels. In Proceedings */
733 /* > of 2011 International Conference for High Performance Computing, */
734 /* > Networking, Storage and Analysis (SC '11), New York, NY, USA, */
735 /* > Article 8 , 11 pages. */
736 /* > http://doi.acm.org/10.1145/2063384.2063394 */
738 /* > A. Haidar, J. Kurzak, P. Luszczek, 2013. */
739 /* > An improved parallel singular value algorithm and its implementation */
740 /* > for multicore hardware, In Proceedings of 2013 International Conference */
741 /* > for High Performance Computing, Networking, Storage and Analysis (SC '13). */
742 /* > Denver, Colorado, USA, 2013. */
743 /* > Article 90, 12 pages. */
744 /* > http://doi.acm.org/10.1145/2503210.2503292 */
746 /* > A. Haidar, R. Solca, S. Tomov, T. Schulthess and J. Dongarra. */
747 /* > A novel hybrid CPU-GPU generalized eigensolver for electronic structure */
748 /* > calculations based on fine-grained memory aware tasks. */
749 /* > International Journal of High Performance Computing Applications. */
750 /* > Volume 28 Issue 2, Pages 196-209, May 2014. */
751 /* > http://hpc.sagepub.com/content/28/2/196 */
755 /* ===================================================================== */
756 /* Subroutine */ int dsbevd_2stage_(char *jobz, char *uplo, integer *n,
757 integer *kd, doublereal *ab, integer *ldab, doublereal *w, doublereal
758 *z__, integer *ldz, doublereal *work, integer *lwork, integer *iwork,
759 integer *liwork, integer *info)
761 /* System generated locals */
762 integer ab_dim1, ab_offset, z_dim1, z_offset, i__1, i__2;
765 /* Local variables */
767 extern integer ilaenv2stage_(integer *, char *, char *, integer *,
768 integer *, integer *, integer *);
769 doublereal anrm, rmin, rmax;
770 extern /* Subroutine */ int dsytrd_sb2st_(char *, char *, char *,
771 integer *, integer *, doublereal *, integer *, doublereal *,
772 doublereal *, doublereal *, integer *, doublereal *, integer *,
773 integer *), dscal_(integer *, doublereal *
774 , doublereal *, integer *), dgemm_(char *, char *, integer *,
775 integer *, integer *, doublereal *, doublereal *, integer *,
776 doublereal *, integer *, doublereal *, doublereal *, integer *);
778 extern logical lsame_(char *, char *);
779 integer iinfo, lhtrd, lwmin;
783 integer indwk2, ib, llwrk2;
784 extern doublereal dlamch_(char *);
786 extern /* Subroutine */ int dlascl_(char *, integer *, integer *,
787 doublereal *, doublereal *, integer *, integer *, doublereal *,
788 integer *, integer *);
789 extern doublereal dlansb_(char *, char *, integer *, integer *,
790 doublereal *, integer *, doublereal *);
791 extern /* Subroutine */ int dstedc_(char *, integer *, doublereal *,
792 doublereal *, doublereal *, integer *, doublereal *, integer *,
793 integer *, integer *, integer *), dlacpy_(char *, integer
794 *, integer *, doublereal *, integer *, doublereal *, integer *);
796 extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
798 extern /* Subroutine */ int dsterf_(integer *, doublereal *, doublereal *,
800 integer indwrk, liwmin, llwork;
808 /* -- LAPACK driver routine (version 3.8.0) -- */
809 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
810 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
814 /* ===================================================================== */
817 /* Test the input parameters. */
819 /* Parameter adjustments */
821 ab_offset = 1 + ab_dim1 * 1;
825 z_offset = 1 + z_dim1 * 1;
831 wantz = lsame_(jobz, "V");
832 lower = lsame_(uplo, "L");
833 lquery = *lwork == -1 || *liwork == -1;
840 ib = ilaenv2stage_(&c__2, "DSYTRD_SB2ST", jobz, n, kd, &c_n1, &c_n1);
841 lhtrd = ilaenv2stage_(&c__3, "DSYTRD_SB2ST", jobz, n, kd, &ib, &c_n1);
842 lwtrd = ilaenv2stage_(&c__4, "DSYTRD_SB2ST", jobz, n, kd, &ib, &c_n1);
845 /* Computing 2nd power */
847 lwmin = *n * 5 + 1 + (i__1 * i__1 << 1);
851 i__1 = *n << 1, i__2 = *n + lhtrd + lwtrd;
852 lwmin = f2cmax(i__1,i__2);
855 if (! lsame_(jobz, "N")) {
857 } else if (! (lower || lsame_(uplo, "U"))) {
861 } else if (*kd < 0) {
863 } else if (*ldab < *kd + 1) {
865 } else if (*ldz < 1 || wantz && *ldz < *n) {
870 work[1] = (doublereal) lwmin;
873 if (*lwork < lwmin && ! lquery) {
875 } else if (*liwork < liwmin && ! lquery) {
882 xerbla_("DSBEVD_2STAGE", &i__1, (ftnlen)13);
888 /* Quick return if possible */
895 w[1] = ab[ab_dim1 + 1];
897 z__[z_dim1 + 1] = 1.;
902 /* Get machine constants. */
904 safmin = dlamch_("Safe minimum");
905 eps = dlamch_("Precision");
906 smlnum = safmin / eps;
907 bignum = 1. / smlnum;
911 /* Scale matrix to allowable range, if necessary. */
913 anrm = dlansb_("M", uplo, n, kd, &ab[ab_offset], ldab, &work[1]);
915 if (anrm > 0. && anrm < rmin) {
918 } else if (anrm > rmax) {
924 dlascl_("B", kd, kd, &c_b21, &sigma, n, n, &ab[ab_offset], ldab,
927 dlascl_("Q", kd, kd, &c_b21, &sigma, n, n, &ab[ab_offset], ldab,
932 /* Call DSYTRD_SB2ST to reduce band symmetric matrix to tridiagonal form. */
936 indwrk = indhous + lhtrd;
937 llwork = *lwork - indwrk + 1;
938 indwk2 = indwrk + *n * *n;
939 llwrk2 = *lwork - indwk2 + 1;
941 dsytrd_sb2st_("N", jobz, uplo, n, kd, &ab[ab_offset], ldab, &w[1], &work[
942 inde], &work[indhous], &lhtrd, &work[indwrk], &llwork, &iinfo);
944 /* For eigenvalues only, call DSTERF. For eigenvectors, call SSTEDC. */
947 dsterf_(n, &w[1], &work[inde], info);
949 dstedc_("I", n, &w[1], &work[inde], &work[indwrk], n, &work[indwk2], &
950 llwrk2, &iwork[1], liwork, info);
951 dgemm_("N", "N", n, n, n, &c_b21, &z__[z_offset], ldz, &work[indwrk],
952 n, &c_b29, &work[indwk2], n);
953 dlacpy_("A", n, n, &work[indwk2], n, &z__[z_offset], ldz);
956 /* If matrix was scaled, then rescale eigenvalues appropriately. */
960 dscal_(n, &d__1, &w[1], &c__1);
963 work[1] = (doublereal) lwmin;
967 /* End of DSBEVD_2STAGE */
969 } /* dsbevd_2stage__ */