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;
521 /* > \brief \b CHETRD_2STAGE */
523 /* @generated from zhetrd_2stage.f, fortran z -> c, Sun Nov 6 19:34:06 2016 */
525 /* =========== DOCUMENTATION =========== */
527 /* Online html documentation available at */
528 /* http://www.netlib.org/lapack/explore-html/ */
531 /* > Download CHETRD_2STAGE + dependencies */
532 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/chetrd_
535 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/chetrd_
538 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/chetrd_
546 /* SUBROUTINE CHETRD_2STAGE( VECT, UPLO, N, A, LDA, D, E, TAU, */
547 /* HOUS2, LHOUS2, WORK, LWORK, INFO ) */
551 /* CHARACTER VECT, UPLO */
552 /* INTEGER N, LDA, LWORK, LHOUS2, INFO */
553 /* REAL D( * ), E( * ) */
554 /* COMPLEX A( LDA, * ), TAU( * ), */
555 /* HOUS2( * ), WORK( * ) */
558 /* > \par Purpose: */
563 /* > CHETRD_2STAGE reduces a complex Hermitian matrix A to real symmetric */
564 /* > tridiagonal form T by a unitary similarity transformation: */
565 /* > Q1**H Q2**H* A * Q2 * Q1 = T. */
571 /* > \param[in] VECT */
573 /* > VECT is CHARACTER*1 */
574 /* > = 'N': No need for the Housholder representation, */
575 /* > in particular for the second stage (Band to */
576 /* > tridiagonal) and thus LHOUS2 is of size f2cmax(1, 4*N); */
577 /* > = 'V': the Householder representation is needed to */
578 /* > either generate Q1 Q2 or to apply Q1 Q2, */
579 /* > then LHOUS2 is to be queried and computed. */
580 /* > (NOT AVAILABLE IN THIS RELEASE). */
583 /* > \param[in] UPLO */
585 /* > UPLO is CHARACTER*1 */
586 /* > = 'U': Upper triangle of A is stored; */
587 /* > = 'L': Lower triangle of A is stored. */
593 /* > The order of the matrix A. N >= 0. */
596 /* > \param[in,out] A */
598 /* > A is COMPLEX array, dimension (LDA,N) */
599 /* > On entry, the Hermitian matrix A. If UPLO = 'U', the leading */
600 /* > N-by-N upper triangular part of A contains the upper */
601 /* > triangular part of the matrix A, and the strictly lower */
602 /* > triangular part of A is not referenced. If UPLO = 'L', the */
603 /* > leading N-by-N lower triangular part of A contains the lower */
604 /* > triangular part of the matrix A, and the strictly upper */
605 /* > triangular part of A is not referenced. */
606 /* > On exit, if UPLO = 'U', the band superdiagonal */
607 /* > of A are overwritten by the corresponding elements of the */
608 /* > internal band-diagonal matrix AB, and the elements above */
609 /* > the KD superdiagonal, with the array TAU, represent the unitary */
610 /* > matrix Q1 as a product of elementary reflectors; if UPLO */
611 /* > = 'L', the diagonal and band subdiagonal of A are over- */
612 /* > written by the corresponding elements of the internal band-diagonal */
613 /* > matrix AB, and the elements below the KD subdiagonal, with */
614 /* > the array TAU, represent the unitary matrix Q1 as a product */
615 /* > of elementary reflectors. See Further Details. */
618 /* > \param[in] LDA */
620 /* > LDA is INTEGER */
621 /* > The leading dimension of the array A. LDA >= f2cmax(1,N). */
624 /* > \param[out] D */
626 /* > D is REAL array, dimension (N) */
627 /* > The diagonal elements of the tridiagonal matrix T. */
630 /* > \param[out] E */
632 /* > E is REAL array, dimension (N-1) */
633 /* > The off-diagonal elements of the tridiagonal matrix T. */
636 /* > \param[out] TAU */
638 /* > TAU is COMPLEX array, dimension (N-KD) */
639 /* > The scalar factors of the elementary reflectors of */
640 /* > the first stage (see Further Details). */
643 /* > \param[out] HOUS2 */
645 /* > HOUS2 is COMPLEX array, dimension (LHOUS2) */
646 /* > Stores the Householder representation of the stage2 */
647 /* > band to tridiagonal. */
650 /* > \param[in] LHOUS2 */
652 /* > LHOUS2 is INTEGER */
653 /* > The dimension of the array HOUS2. */
654 /* > If LWORK = -1, or LHOUS2=-1, */
655 /* > then a query is assumed; the routine */
656 /* > only calculates the optimal size of the HOUS2 array, returns */
657 /* > this value as the first entry of the HOUS2 array, and no error */
658 /* > message related to LHOUS2 is issued by XERBLA. */
659 /* > If VECT='N', LHOUS2 = f2cmax(1, 4*n); */
660 /* > if VECT='V', option not yet available. */
663 /* > \param[out] WORK */
665 /* > WORK is COMPLEX array, dimension (LWORK) */
668 /* > \param[in] LWORK */
670 /* > LWORK is INTEGER */
671 /* > The dimension of the array WORK. LWORK = MAX(1, dimension) */
672 /* > If LWORK = -1, or LHOUS2 = -1, */
673 /* > then a workspace query is assumed; the routine */
674 /* > only calculates the optimal size of the WORK array, returns */
675 /* > this value as the first entry of the WORK array, and no error */
676 /* > message related to LWORK is issued by XERBLA. */
677 /* > LWORK = MAX(1, dimension) where */
678 /* > dimension = f2cmax(stage1,stage2) + (KD+1)*N */
679 /* > = N*KD + N*f2cmax(KD+1,FACTOPTNB) */
680 /* > + f2cmax(2*KD*KD, KD*NTHREADS) */
682 /* > where KD is the blocking size of the reduction, */
683 /* > FACTOPTNB is the blocking used by the QR or LQ */
684 /* > algorithm, usually FACTOPTNB=128 is a good choice */
685 /* > NTHREADS is the number of threads used when */
686 /* > openMP compilation is enabled, otherwise =1. */
689 /* > \param[out] INFO */
691 /* > INFO is INTEGER */
692 /* > = 0: successful exit */
693 /* > < 0: if INFO = -i, the i-th argument had an illegal value */
699 /* > \author Univ. of Tennessee */
700 /* > \author Univ. of California Berkeley */
701 /* > \author Univ. of Colorado Denver */
702 /* > \author NAG Ltd. */
704 /* > \date November 2017 */
706 /* > \ingroup complexHEcomputational */
708 /* > \par Further Details: */
709 /* ===================== */
713 /* > Implemented by Azzam Haidar. */
715 /* > All details are available on technical report, SC11, SC13 papers. */
717 /* > Azzam Haidar, Hatem Ltaief, and Jack Dongarra. */
718 /* > Parallel reduction to condensed forms for symmetric eigenvalue problems */
719 /* > using aggregated fine-grained and memory-aware kernels. In Proceedings */
720 /* > of 2011 International Conference for High Performance Computing, */
721 /* > Networking, Storage and Analysis (SC '11), New York, NY, USA, */
722 /* > Article 8 , 11 pages. */
723 /* > http://doi.acm.org/10.1145/2063384.2063394 */
725 /* > A. Haidar, J. Kurzak, P. Luszczek, 2013. */
726 /* > An improved parallel singular value algorithm and its implementation */
727 /* > for multicore hardware, In Proceedings of 2013 International Conference */
728 /* > for High Performance Computing, Networking, Storage and Analysis (SC '13). */
729 /* > Denver, Colorado, USA, 2013. */
730 /* > Article 90, 12 pages. */
731 /* > http://doi.acm.org/10.1145/2503210.2503292 */
733 /* > A. Haidar, R. Solca, S. Tomov, T. Schulthess and J. Dongarra. */
734 /* > A novel hybrid CPU-GPU generalized eigensolver for electronic structure */
735 /* > calculations based on fine-grained memory aware tasks. */
736 /* > International Journal of High Performance Computing Applications. */
737 /* > Volume 28 Issue 2, Pages 196-209, May 2014. */
738 /* > http://hpc.sagepub.com/content/28/2/196 */
742 /* ===================================================================== */
743 /* Subroutine */ int chetrd_2stage_(char *vect, char *uplo, integer *n,
744 complex *a, integer *lda, real *d__, real *e, complex *tau, complex *
745 hous2, integer *lhous2, complex *work, integer *lwork, integer *info)
747 /* System generated locals */
748 integer a_dim1, a_offset, i__1;
750 /* Local variables */
752 extern /* Subroutine */ int chetrd_hb2st_(char *, char *, char *,
753 integer *, integer *, complex *, integer *, real *, real *,
754 complex *, integer *, complex *, integer *, integer *);
755 extern integer ilaenv2stage_(integer *, char *, char *, integer *,
756 integer *, integer *, integer *);
758 extern logical lsame_(char *, char *);
759 integer abpos, lhmin, lwmin;
760 logical wantq, upper;
762 extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
764 extern /* Subroutine */ int chetrd_he2hb_(char *, integer *, integer *,
765 complex *, integer *, complex *, integer *, complex *, complex *,
766 integer *, integer *);
770 /* -- LAPACK computational routine (version 3.8.0) -- */
771 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
772 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
776 /* ===================================================================== */
778 /* Test the input parameters */
780 /* Parameter adjustments */
782 a_offset = 1 + a_dim1 * 1;
792 wantq = lsame_(vect, "V");
793 upper = lsame_(uplo, "U");
794 lquery = *lwork == -1 || *lhous2 == -1;
796 /* Determine the block size, the workspace size and the hous size. */
798 kd = ilaenv2stage_(&c__1, "CHETRD_2STAGE", vect, n, &c_n1, &c_n1, &c_n1);
799 ib = ilaenv2stage_(&c__2, "CHETRD_2STAGE", vect, n, &kd, &c_n1, &c_n1);
800 lhmin = ilaenv2stage_(&c__3, "CHETRD_2STAGE", vect, n, &kd, &ib, &c_n1);
801 lwmin = ilaenv2stage_(&c__4, "CHETRD_2STAGE", vect, n, &kd, &ib, &c_n1);
802 /* WRITE(*,*),'CHETRD_2STAGE N KD UPLO LHMIN LWMIN ',N, KD, UPLO, */
805 if (! lsame_(vect, "N")) {
807 } else if (! upper && ! lsame_(uplo, "L")) {
811 } else if (*lda < f2cmax(1,*n)) {
813 } else if (*lhous2 < lhmin && ! lquery) {
815 } else if (*lwork < lwmin && ! lquery) {
820 hous2[1].r = (real) lhmin, hous2[1].i = 0.f;
821 work[1].r = (real) lwmin, work[1].i = 0.f;
826 xerbla_("CHETRD_2STAGE", &i__1, (ftnlen)13);
832 /* Quick return if possible */
835 work[1].r = 1.f, work[1].i = 0.f;
839 /* Determine pointer position */
842 lwrk = *lwork - ldab * *n;
844 wpos = abpos + ldab * *n;
845 chetrd_he2hb_(uplo, n, &kd, &a[a_offset], lda, &work[abpos], &ldab, &tau[
846 1], &work[wpos], &lwrk, info);
849 xerbla_("CHETRD_HE2HB", &i__1, (ftnlen)12);
852 chetrd_hb2st_("Y", vect, uplo, n, &kd, &work[abpos], &ldab, &d__[1], &e[
853 1], &hous2[1], lhous2, &work[wpos], &lwrk, info);
856 xerbla_("CHETRD_HB2ST", &i__1, (ftnlen)12);
861 hous2[1].r = (real) lhmin, hous2[1].i = 0.f;
862 work[1].r = (real) lwmin, work[1].i = 0.f;
865 /* End of CHETRD_2STAGE */
867 } /* chetrd_2stage__ */