14 typedef long long BLASLONG;
15 typedef unsigned long long BLASULONG;
17 typedef long BLASLONG;
18 typedef unsigned long BLASULONG;
22 typedef BLASLONG blasint;
24 #define blasabs(x) llabs(x)
26 #define blasabs(x) labs(x)
30 #define blasabs(x) abs(x)
33 typedef blasint integer;
35 typedef unsigned int uinteger;
36 typedef char *address;
37 typedef short int shortint;
39 typedef double doublereal;
40 typedef struct { real r, i; } complex;
41 typedef struct { doublereal r, i; } doublecomplex;
43 static inline _Fcomplex Cf(complex *z) {_Fcomplex zz={z->r , z->i}; return zz;}
44 static inline _Dcomplex Cd(doublecomplex *z) {_Dcomplex zz={z->r , z->i};return zz;}
45 static inline _Fcomplex * _pCf(complex *z) {return (_Fcomplex*)z;}
46 static inline _Dcomplex * _pCd(doublecomplex *z) {return (_Dcomplex*)z;}
48 static inline _Complex float Cf(complex *z) {return z->r + z->i*_Complex_I;}
49 static inline _Complex double Cd(doublecomplex *z) {return z->r + z->i*_Complex_I;}
50 static inline _Complex float * _pCf(complex *z) {return (_Complex float*)z;}
51 static inline _Complex double * _pCd(doublecomplex *z) {return (_Complex double*)z;}
53 #define pCf(z) (*_pCf(z))
54 #define pCd(z) (*_pCd(z))
56 typedef short int shortlogical;
57 typedef char logical1;
58 typedef char integer1;
63 /* Extern is for use with -E */
74 /*external read, write*/
83 /*internal read, write*/
113 /*rewind, backspace, endfile*/
125 ftnint *inex; /*parameters in standard's order*/
151 union Multitype { /* for multiple entry points */
162 typedef union Multitype Multitype;
164 struct Vardesc { /* for Namelist */
170 typedef struct Vardesc Vardesc;
177 typedef struct Namelist Namelist;
179 #define abs(x) ((x) >= 0 ? (x) : -(x))
180 #define dabs(x) (fabs(x))
181 #define f2cmin(a,b) ((a) <= (b) ? (a) : (b))
182 #define f2cmax(a,b) ((a) >= (b) ? (a) : (b))
183 #define dmin(a,b) (f2cmin(a,b))
184 #define dmax(a,b) (f2cmax(a,b))
185 #define bit_test(a,b) ((a) >> (b) & 1)
186 #define bit_clear(a,b) ((a) & ~((uinteger)1 << (b)))
187 #define bit_set(a,b) ((a) | ((uinteger)1 << (b)))
189 #define abort_() { sig_die("Fortran abort routine called", 1); }
190 #define c_abs(z) (cabsf(Cf(z)))
191 #define c_cos(R,Z) { pCf(R)=ccos(Cf(Z)); }
193 #define c_div(c, a, b) {Cf(c)._Val[0] = (Cf(a)._Val[0]/Cf(b)._Val[0]); Cf(c)._Val[1]=(Cf(a)._Val[1]/Cf(b)._Val[1]);}
194 #define z_div(c, a, b) {Cd(c)._Val[0] = (Cd(a)._Val[0]/Cd(b)._Val[0]); Cd(c)._Val[1]=(Cd(a)._Val[1]/df(b)._Val[1]);}
196 #define c_div(c, a, b) {pCf(c) = Cf(a)/Cf(b);}
197 #define z_div(c, a, b) {pCd(c) = Cd(a)/Cd(b);}
199 #define c_exp(R, Z) {pCf(R) = cexpf(Cf(Z));}
200 #define c_log(R, Z) {pCf(R) = clogf(Cf(Z));}
201 #define c_sin(R, Z) {pCf(R) = csinf(Cf(Z));}
202 //#define c_sqrt(R, Z) {*(R) = csqrtf(Cf(Z));}
203 #define c_sqrt(R, Z) {pCf(R) = csqrtf(Cf(Z));}
204 #define d_abs(x) (fabs(*(x)))
205 #define d_acos(x) (acos(*(x)))
206 #define d_asin(x) (asin(*(x)))
207 #define d_atan(x) (atan(*(x)))
208 #define d_atn2(x, y) (atan2(*(x),*(y)))
209 #define d_cnjg(R, Z) { pCd(R) = conj(Cd(Z)); }
210 #define r_cnjg(R, Z) { pCf(R) = conjf(Cf(Z)); }
211 #define d_cos(x) (cos(*(x)))
212 #define d_cosh(x) (cosh(*(x)))
213 #define d_dim(__a, __b) ( *(__a) > *(__b) ? *(__a) - *(__b) : 0.0 )
214 #define d_exp(x) (exp(*(x)))
215 #define d_imag(z) (cimag(Cd(z)))
216 #define r_imag(z) (cimagf(Cf(z)))
217 #define d_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
218 #define r_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
219 #define d_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
220 #define r_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
221 #define d_log(x) (log(*(x)))
222 #define d_mod(x, y) (fmod(*(x), *(y)))
223 #define u_nint(__x) ((__x)>=0 ? floor((__x) + .5) : -floor(.5 - (__x)))
224 #define d_nint(x) u_nint(*(x))
225 #define u_sign(__a,__b) ((__b) >= 0 ? ((__a) >= 0 ? (__a) : -(__a)) : -((__a) >= 0 ? (__a) : -(__a)))
226 #define d_sign(a,b) u_sign(*(a),*(b))
227 #define r_sign(a,b) u_sign(*(a),*(b))
228 #define d_sin(x) (sin(*(x)))
229 #define d_sinh(x) (sinh(*(x)))
230 #define d_sqrt(x) (sqrt(*(x)))
231 #define d_tan(x) (tan(*(x)))
232 #define d_tanh(x) (tanh(*(x)))
233 #define i_abs(x) abs(*(x))
234 #define i_dnnt(x) ((integer)u_nint(*(x)))
235 #define i_len(s, n) (n)
236 #define i_nint(x) ((integer)u_nint(*(x)))
237 #define i_sign(a,b) ((integer)u_sign((integer)*(a),(integer)*(b)))
238 #define pow_dd(ap, bp) ( pow(*(ap), *(bp)))
239 #define pow_si(B,E) spow_ui(*(B),*(E))
240 #define pow_ri(B,E) spow_ui(*(B),*(E))
241 #define pow_di(B,E) dpow_ui(*(B),*(E))
242 #define pow_zi(p, a, b) {pCd(p) = zpow_ui(Cd(a), *(b));}
243 #define pow_ci(p, a, b) {pCf(p) = cpow_ui(Cf(a), *(b));}
244 #define pow_zz(R,A,B) {pCd(R) = cpow(Cd(A),*(B));}
245 #define s_cat(lpp, rpp, rnp, np, llp) { ftnlen i, nc, ll; char *f__rp, *lp; ll = (llp); lp = (lpp); for(i=0; i < (int)*(np); ++i) { nc = ll; if((rnp)[i] < nc) nc = (rnp)[i]; ll -= nc; f__rp = (rpp)[i]; while(--nc >= 0) *lp++ = *(f__rp)++; } while(--ll >= 0) *lp++ = ' '; }
246 #define s_cmp(a,b,c,d) ((integer)strncmp((a),(b),f2cmin((c),(d))))
247 #define s_copy(A,B,C,D) { int __i,__m; for (__i=0, __m=f2cmin((C),(D)); __i<__m && (B)[__i] != 0; ++__i) (A)[__i] = (B)[__i]; }
248 #define sig_die(s, kill) { exit(1); }
249 #define s_stop(s, n) {exit(0);}
250 static char junk[] = "\n@(#)LIBF77 VERSION 19990503\n";
251 #define z_abs(z) (cabs(Cd(z)))
252 #define z_exp(R, Z) {pCd(R) = cexp(Cd(Z));}
253 #define z_sqrt(R, Z) {pCd(R) = csqrt(Cd(Z));}
254 #define myexit_() break;
255 #define mycycle() continue;
256 #define myceiling(w) {ceil(w)}
257 #define myhuge(w) {HUGE_VAL}
258 //#define mymaxloc_(w,s,e,n) {if (sizeof(*(w)) == sizeof(double)) dmaxloc_((w),*(s),*(e),n); else dmaxloc_((w),*(s),*(e),n);}
259 #define mymaxloc(w,s,e,n) {dmaxloc_(w,*(s),*(e),n)}
261 /* procedure parameter types for -A and -C++ */
263 #define F2C_proc_par_types 1
265 typedef logical (*L_fp)(...);
267 typedef logical (*L_fp)();
270 static float spow_ui(float x, integer n) {
271 float pow=1.0; unsigned long int u;
273 if(n < 0) n = -n, x = 1/x;
282 static double dpow_ui(double x, integer n) {
283 double pow=1.0; unsigned long int u;
285 if(n < 0) n = -n, x = 1/x;
295 static _Fcomplex cpow_ui(complex x, integer n) {
296 complex pow={1.0,0.0}; unsigned long int u;
298 if(n < 0) n = -n, x.r = 1/x.r, x.i=1/x.i;
300 if(u & 01) pow.r *= x.r, pow.i *= x.i;
301 if(u >>= 1) x.r *= x.r, x.i *= x.i;
305 _Fcomplex p={pow.r, pow.i};
309 static _Complex float cpow_ui(_Complex float x, integer n) {
310 _Complex float pow=1.0; unsigned long int u;
312 if(n < 0) n = -n, x = 1/x;
323 static _Dcomplex zpow_ui(_Dcomplex x, integer n) {
324 _Dcomplex pow={1.0,0.0}; unsigned long int u;
326 if(n < 0) n = -n, x._Val[0] = 1/x._Val[0], x._Val[1] =1/x._Val[1];
328 if(u & 01) pow._Val[0] *= x._Val[0], pow._Val[1] *= x._Val[1];
329 if(u >>= 1) x._Val[0] *= x._Val[0], x._Val[1] *= x._Val[1];
333 _Dcomplex p = {pow._Val[0], pow._Val[1]};
337 static _Complex double zpow_ui(_Complex double x, integer n) {
338 _Complex double pow=1.0; unsigned long int u;
340 if(n < 0) n = -n, x = 1/x;
350 static integer pow_ii(integer x, integer n) {
351 integer pow; unsigned long int u;
353 if (n == 0 || x == 1) pow = 1;
354 else if (x != -1) pow = x == 0 ? 1/x : 0;
357 if ((n > 0) || !(n == 0 || x == 1 || x != -1)) {
367 static integer dmaxloc_(double *w, integer s, integer e, integer *n)
369 double m; integer i, mi;
370 for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
371 if (w[i-1]>m) mi=i ,m=w[i-1];
374 static integer smaxloc_(float *w, integer s, integer e, integer *n)
376 float m; integer i, mi;
377 for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
378 if (w[i-1]>m) mi=i ,m=w[i-1];
381 static inline void cdotc_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
382 integer n = *n_, incx = *incx_, incy = *incy_, i;
384 _Fcomplex zdotc = {0.0, 0.0};
385 if (incx == 1 && incy == 1) {
386 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
387 zdotc._Val[0] += conjf(Cf(&x[i]))._Val[0] * Cf(&y[i])._Val[0];
388 zdotc._Val[1] += conjf(Cf(&x[i]))._Val[1] * Cf(&y[i])._Val[1];
391 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
392 zdotc._Val[0] += conjf(Cf(&x[i*incx]))._Val[0] * Cf(&y[i*incy])._Val[0];
393 zdotc._Val[1] += conjf(Cf(&x[i*incx]))._Val[1] * Cf(&y[i*incy])._Val[1];
399 _Complex float zdotc = 0.0;
400 if (incx == 1 && incy == 1) {
401 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
402 zdotc += conjf(Cf(&x[i])) * Cf(&y[i]);
405 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
406 zdotc += conjf(Cf(&x[i*incx])) * Cf(&y[i*incy]);
412 static inline void zdotc_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
413 integer n = *n_, incx = *incx_, incy = *incy_, i;
415 _Dcomplex zdotc = {0.0, 0.0};
416 if (incx == 1 && incy == 1) {
417 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
418 zdotc._Val[0] += conj(Cd(&x[i]))._Val[0] * Cd(&y[i])._Val[0];
419 zdotc._Val[1] += conj(Cd(&x[i]))._Val[1] * Cd(&y[i])._Val[1];
422 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
423 zdotc._Val[0] += conj(Cd(&x[i*incx]))._Val[0] * Cd(&y[i*incy])._Val[0];
424 zdotc._Val[1] += conj(Cd(&x[i*incx]))._Val[1] * Cd(&y[i*incy])._Val[1];
430 _Complex double zdotc = 0.0;
431 if (incx == 1 && incy == 1) {
432 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
433 zdotc += conj(Cd(&x[i])) * Cd(&y[i]);
436 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
437 zdotc += conj(Cd(&x[i*incx])) * Cd(&y[i*incy]);
443 static inline void cdotu_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
444 integer n = *n_, incx = *incx_, incy = *incy_, i;
446 _Fcomplex zdotc = {0.0, 0.0};
447 if (incx == 1 && incy == 1) {
448 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
449 zdotc._Val[0] += Cf(&x[i])._Val[0] * Cf(&y[i])._Val[0];
450 zdotc._Val[1] += Cf(&x[i])._Val[1] * Cf(&y[i])._Val[1];
453 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
454 zdotc._Val[0] += Cf(&x[i*incx])._Val[0] * Cf(&y[i*incy])._Val[0];
455 zdotc._Val[1] += Cf(&x[i*incx])._Val[1] * Cf(&y[i*incy])._Val[1];
461 _Complex float zdotc = 0.0;
462 if (incx == 1 && incy == 1) {
463 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
464 zdotc += Cf(&x[i]) * Cf(&y[i]);
467 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
468 zdotc += Cf(&x[i*incx]) * Cf(&y[i*incy]);
474 static inline void zdotu_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
475 integer n = *n_, incx = *incx_, incy = *incy_, i;
477 _Dcomplex zdotc = {0.0, 0.0};
478 if (incx == 1 && incy == 1) {
479 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
480 zdotc._Val[0] += Cd(&x[i])._Val[0] * Cd(&y[i])._Val[0];
481 zdotc._Val[1] += Cd(&x[i])._Val[1] * Cd(&y[i])._Val[1];
484 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
485 zdotc._Val[0] += Cd(&x[i*incx])._Val[0] * Cd(&y[i*incy])._Val[0];
486 zdotc._Val[1] += Cd(&x[i*incx])._Val[1] * Cd(&y[i*incy])._Val[1];
492 _Complex double zdotc = 0.0;
493 if (incx == 1 && incy == 1) {
494 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
495 zdotc += Cd(&x[i]) * Cd(&y[i]);
498 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
499 zdotc += Cd(&x[i*incx]) * Cd(&y[i*incy]);
505 /* -- translated by f2c (version 20000121).
506 You must link the resulting object file with the libraries:
507 -lf2c -lm (in that order)
513 /* Table of constant values */
515 static integer c__1 = 1;
516 static integer c_n1 = -1;
517 static integer c__2 = 2;
518 static integer c__3 = 3;
519 static integer c__4 = 4;
520 static real c_b26 = 1.f;
522 /* > \brief \b SSYGV_2STAGE */
524 /* @generated from dsygv_2stage.f, fortran d -> s, Sun Nov 6 12:54:29 2016 */
526 /* =========== DOCUMENTATION =========== */
528 /* Online html documentation available at */
529 /* http://www.netlib.org/lapack/explore-html/ */
532 /* > Download SSYGV_2STAGE + dependencies */
533 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/ssygv_2
536 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/ssygv_2
539 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/ssygv_2
547 /* SUBROUTINE SSYGV_2STAGE( ITYPE, JOBZ, UPLO, N, A, LDA, B, LDB, W, */
548 /* WORK, LWORK, INFO ) */
552 /* CHARACTER JOBZ, UPLO */
553 /* INTEGER INFO, ITYPE, LDA, LDB, LWORK, N */
554 /* REAL A( LDA, * ), B( LDB, * ), W( * ), WORK( * ) */
557 /* > \par Purpose: */
562 /* > SSYGV_2STAGE computes all the eigenvalues, and optionally, the eigenvectors */
563 /* > of a real generalized symmetric-definite eigenproblem, of the form */
564 /* > A*x=(lambda)*B*x, A*Bx=(lambda)*x, or B*A*x=(lambda)*x. */
565 /* > Here A and B are assumed to be symmetric and B is also */
566 /* > positive definite. */
567 /* > This routine use the 2stage technique for the reduction to tridiagonal */
568 /* > which showed higher performance on recent architecture and for large */
569 /* > sizes N>2000. */
575 /* > \param[in] ITYPE */
577 /* > ITYPE is INTEGER */
578 /* > Specifies the problem type to be solved: */
579 /* > = 1: A*x = (lambda)*B*x */
580 /* > = 2: A*B*x = (lambda)*x */
581 /* > = 3: B*A*x = (lambda)*x */
584 /* > \param[in] JOBZ */
586 /* > JOBZ is CHARACTER*1 */
587 /* > = 'N': Compute eigenvalues only; */
588 /* > = 'V': Compute eigenvalues and eigenvectors. */
589 /* > Not available in this release. */
592 /* > \param[in] UPLO */
594 /* > UPLO is CHARACTER*1 */
595 /* > = 'U': Upper triangles of A and B are stored; */
596 /* > = 'L': Lower triangles of A and B are stored. */
602 /* > The order of the matrices A and B. N >= 0. */
605 /* > \param[in,out] A */
607 /* > A is REAL array, dimension (LDA, N) */
608 /* > On entry, the symmetric matrix A. If UPLO = 'U', the */
609 /* > leading N-by-N upper triangular part of A contains the */
610 /* > upper triangular part of the matrix A. If UPLO = 'L', */
611 /* > the leading N-by-N lower triangular part of A contains */
612 /* > the lower triangular part of the matrix A. */
614 /* > On exit, if JOBZ = 'V', then if INFO = 0, A contains the */
615 /* > matrix Z of eigenvectors. The eigenvectors are normalized */
617 /* > if ITYPE = 1 or 2, Z**T*B*Z = I; */
618 /* > if ITYPE = 3, Z**T*inv(B)*Z = I. */
619 /* > If JOBZ = 'N', then on exit the upper triangle (if UPLO='U') */
620 /* > or the lower triangle (if UPLO='L') of A, including the */
621 /* > diagonal, is destroyed. */
624 /* > \param[in] LDA */
626 /* > LDA is INTEGER */
627 /* > The leading dimension of the array A. LDA >= f2cmax(1,N). */
630 /* > \param[in,out] B */
632 /* > B is REAL array, dimension (LDB, N) */
633 /* > On entry, the symmetric positive definite matrix B. */
634 /* > If UPLO = 'U', the leading N-by-N upper triangular part of B */
635 /* > contains the upper triangular part of the matrix B. */
636 /* > If UPLO = 'L', the leading N-by-N lower triangular part of B */
637 /* > contains the lower triangular part of the matrix B. */
639 /* > On exit, if INFO <= N, the part of B containing the matrix is */
640 /* > overwritten by the triangular factor U or L from the Cholesky */
641 /* > factorization B = U**T*U or B = L*L**T. */
644 /* > \param[in] LDB */
646 /* > LDB is INTEGER */
647 /* > The leading dimension of the array B. LDB >= f2cmax(1,N). */
650 /* > \param[out] W */
652 /* > W is REAL array, dimension (N) */
653 /* > If INFO = 0, the eigenvalues in ascending order. */
656 /* > \param[out] WORK */
658 /* > WORK is REAL array, dimension (MAX(1,LWORK)) */
659 /* > On exit, if INFO = 0, WORK(1) returns the optimal LWORK. */
662 /* > \param[in] LWORK */
664 /* > LWORK is INTEGER */
665 /* > The length of the array WORK. LWORK >= 1, when N <= 1; */
667 /* > If JOBZ = 'N' and N > 1, LWORK must be queried. */
668 /* > LWORK = MAX(1, dimension) where */
669 /* > dimension = f2cmax(stage1,stage2) + (KD+1)*N + 2*N */
670 /* > = N*KD + N*f2cmax(KD+1,FACTOPTNB) */
671 /* > + f2cmax(2*KD*KD, KD*NTHREADS) */
672 /* > + (KD+1)*N + 2*N */
673 /* > where KD is the blocking size of the reduction, */
674 /* > FACTOPTNB is the blocking used by the QR or LQ */
675 /* > algorithm, usually FACTOPTNB=128 is a good choice */
676 /* > NTHREADS is the number of threads used when */
677 /* > openMP compilation is enabled, otherwise =1. */
678 /* > If JOBZ = 'V' and N > 1, LWORK must be queried. Not yet available */
680 /* > If LWORK = -1, then a workspace query is assumed; the routine */
681 /* > only calculates the optimal size of the WORK array, returns */
682 /* > this value as the first entry of the WORK array, and no error */
683 /* > message related to LWORK is issued by XERBLA. */
686 /* > \param[out] INFO */
688 /* > INFO is INTEGER */
689 /* > = 0: successful exit */
690 /* > < 0: if INFO = -i, the i-th argument had an illegal value */
691 /* > > 0: SPOTRF or SSYEV returned an error code: */
692 /* > <= N: if INFO = i, SSYEV failed to converge; */
693 /* > i off-diagonal elements of an intermediate */
694 /* > tridiagonal form did not converge to zero; */
695 /* > > N: if INFO = N + i, for 1 <= i <= N, then the leading */
696 /* > minor of order i of B is not positive definite. */
697 /* > The factorization of B could not be completed and */
698 /* > no eigenvalues or eigenvectors were computed. */
704 /* > \author Univ. of Tennessee */
705 /* > \author Univ. of California Berkeley */
706 /* > \author Univ. of Colorado Denver */
707 /* > \author NAG Ltd. */
709 /* > \date November 2017 */
711 /* > \ingroup realSYeigen */
713 /* > \par Further Details: */
714 /* ===================== */
718 /* > All details about the 2stage techniques are available in: */
720 /* > Azzam Haidar, Hatem Ltaief, and Jack Dongarra. */
721 /* > Parallel reduction to condensed forms for symmetric eigenvalue problems */
722 /* > using aggregated fine-grained and memory-aware kernels. In Proceedings */
723 /* > of 2011 International Conference for High Performance Computing, */
724 /* > Networking, Storage and Analysis (SC '11), New York, NY, USA, */
725 /* > Article 8 , 11 pages. */
726 /* > http://doi.acm.org/10.1145/2063384.2063394 */
728 /* > A. Haidar, J. Kurzak, P. Luszczek, 2013. */
729 /* > An improved parallel singular value algorithm and its implementation */
730 /* > for multicore hardware, In Proceedings of 2013 International Conference */
731 /* > for High Performance Computing, Networking, Storage and Analysis (SC '13). */
732 /* > Denver, Colorado, USA, 2013. */
733 /* > Article 90, 12 pages. */
734 /* > http://doi.acm.org/10.1145/2503210.2503292 */
736 /* > A. Haidar, R. Solca, S. Tomov, T. Schulthess and J. Dongarra. */
737 /* > A novel hybrid CPU-GPU generalized eigensolver for electronic structure */
738 /* > calculations based on fine-grained memory aware tasks. */
739 /* > International Journal of High Performance Computing Applications. */
740 /* > Volume 28 Issue 2, Pages 196-209, May 2014. */
741 /* > http://hpc.sagepub.com/content/28/2/196 */
745 /* ===================================================================== */
746 /* Subroutine */ int ssygv_2stage_(integer *itype, char *jobz, char *uplo,
747 integer *n, real *a, integer *lda, real *b, integer *ldb, real *w,
748 real *work, integer *lwork, integer *info)
750 /* System generated locals */
751 integer a_dim1, a_offset, b_dim1, b_offset, i__1;
753 /* Local variables */
755 extern integer ilaenv2stage_(integer *, char *, char *, integer *,
756 integer *, integer *, integer *);
757 extern /* Subroutine */ int ssyev_2stage_(char *, char *, integer *,
758 real *, integer *, real *, real *, integer *, integer *);
759 extern logical lsame_(char *, char *);
760 integer lhtrd, lwmin;
764 extern /* Subroutine */ int strmm_(char *, char *, char *, char *,
765 integer *, integer *, real *, real *, integer *, real *, integer *
768 extern /* Subroutine */ int strsm_(char *, char *, char *, char *,
769 integer *, integer *, real *, real *, integer *, real *, integer *
772 extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen), spotrf_(
773 char *, integer *, real *, integer *, integer *);
775 extern /* Subroutine */ int ssygst_(integer *, char *, integer *, real *,
776 integer *, real *, integer *, integer *);
780 /* -- LAPACK driver routine (version 3.8.0) -- */
781 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
782 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
786 /* ===================================================================== */
789 /* Test the input parameters. */
791 /* Parameter adjustments */
793 a_offset = 1 + a_dim1 * 1;
796 b_offset = 1 + b_dim1 * 1;
802 wantz = lsame_(jobz, "V");
803 upper = lsame_(uplo, "U");
804 lquery = *lwork == -1;
807 if (*itype < 1 || *itype > 3) {
809 } else if (! lsame_(jobz, "N")) {
811 } else if (! (upper || lsame_(uplo, "L"))) {
815 } else if (*lda < f2cmax(1,*n)) {
817 } else if (*ldb < f2cmax(1,*n)) {
822 kd = ilaenv2stage_(&c__1, "SSYTRD_2STAGE", jobz, n, &c_n1, &c_n1, &
824 ib = ilaenv2stage_(&c__2, "SSYTRD_2STAGE", jobz, n, &kd, &c_n1, &c_n1);
825 lhtrd = ilaenv2stage_(&c__3, "SSYTRD_2STAGE", jobz, n, &kd, &ib, &
827 lwtrd = ilaenv2stage_(&c__4, "SSYTRD_2STAGE", jobz, n, &kd, &ib, &
829 lwmin = (*n << 1) + lhtrd + lwtrd;
830 work[1] = (real) lwmin;
832 if (*lwork < lwmin && ! lquery) {
839 xerbla_("SSYGV_2STAGE ", &i__1, (ftnlen)13);
845 /* Quick return if possible */
851 /* Form a Cholesky factorization of B. */
853 spotrf_(uplo, n, &b[b_offset], ldb, info);
859 /* Transform problem to standard eigenvalue problem and solve. */
861 ssygst_(itype, uplo, n, &a[a_offset], lda, &b[b_offset], ldb, info);
862 ssyev_2stage_(jobz, uplo, n, &a[a_offset], lda, &w[1], &work[1], lwork,
867 /* Backtransform eigenvectors to the original problem. */
873 if (*itype == 1 || *itype == 2) {
875 /* For A*x=(lambda)*B*x and A*B*x=(lambda)*x; */
876 /* backtransform eigenvectors: x = inv(L)**T*y or inv(U)*y */
879 *(unsigned char *)trans = 'N';
881 *(unsigned char *)trans = 'T';
884 strsm_("Left", uplo, trans, "Non-unit", n, &neig, &c_b26, &b[
885 b_offset], ldb, &a[a_offset], lda);
887 } else if (*itype == 3) {
889 /* For B*A*x=(lambda)*x; */
890 /* backtransform eigenvectors: x = L*y or U**T*y */
893 *(unsigned char *)trans = 'T';
895 *(unsigned char *)trans = 'N';
898 strmm_("Left", uplo, trans, "Non-unit", n, &neig, &c_b26, &b[
899 b_offset], ldb, &a[a_offset], lda);
903 work[1] = (real) lwmin;
906 /* End of SSYGV_2STAGE */
908 } /* ssygv_2stage__ */