OpenRadioss 2025.1.11
OpenRadioss project
Loading...
Searching...
No Matches
zhbevd_2stage.f
Go to the documentation of this file.
1*> \brief <b> ZHBEVD_2STAGE computes the eigenvalues and, optionally, the left and/or right eigenvectors for OTHER matrices</b>
2*
3* @precisions fortran z -> s d c
4*
5* =========== DOCUMENTATION ===========
6*
7* Online html documentation available at
8* http://www.netlib.org/lapack/explore-html/
9*
10*> \htmlonly
11*> Download ZHBEVD_2STAGE + dependencies
12*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/zhbevd_2stage.f">
13*> [TGZ]</a>
14*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/zhbevd_2stage.f">
15*> [ZIP]</a>
16*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/zhbevd_2stage.f">
17*> [TXT]</a>
18*> \endhtmlonly
19*
20* Definition:
21* ===========
22*
23* SUBROUTINE ZHBEVD_2STAGE( JOBZ, UPLO, N, KD, AB, LDAB, W, Z, LDZ,
24* WORK, LWORK, RWORK, LRWORK, IWORK,
25* LIWORK, INFO )
26*
27* IMPLICIT NONE
28*
29* .. Scalar Arguments ..
30* CHARACTER JOBZ, UPLO
31* INTEGER INFO, KD, LDAB, LDZ, LIWORK, LRWORK, LWORK, N
32* ..
33* .. Array Arguments ..
34* INTEGER IWORK( * )
35* DOUBLE PRECISION RWORK( * ), W( * )
36* COMPLEX*16 AB( LDAB, * ), WORK( * ), Z( LDZ, * )
37* ..
38*
39*
40*> \par Purpose:
41* =============
42*>
43*> \verbatim
44*>
45*> ZHBEVD_2STAGE computes all the eigenvalues and, optionally, eigenvectors of
46*> a complex Hermitian band matrix A using the 2stage technique for
47*> the reduction to tridiagonal. If eigenvectors are desired, it
48*> uses a divide and conquer algorithm.
49*>
50*> The divide and conquer algorithm makes very mild assumptions about
51*> floating point arithmetic. It will work on machines with a guard
52*> digit in add/subtract, or on those binary machines without guard
53*> digits which subtract like the Cray X-MP, Cray Y-MP, Cray C-90, or
54*> Cray-2. It could conceivably fail on hexadecimal or decimal machines
55*> without guard digits, but we know of none.
56*> \endverbatim
57*
58* Arguments:
59* ==========
60*
61*> \param[in] JOBZ
62*> \verbatim
63*> JOBZ is CHARACTER*1
64*> = 'N': Compute eigenvalues only;
65*> = 'V': Compute eigenvalues and eigenvectors.
66*> Not available in this release.
67*> \endverbatim
68*>
69*> \param[in] UPLO
70*> \verbatim
71*> UPLO is CHARACTER*1
72*> = 'U': Upper triangle of A is stored;
73*> = 'L': Lower triangle of A is stored.
74*> \endverbatim
75*>
76*> \param[in] N
77*> \verbatim
78*> N is INTEGER
79*> The order of the matrix A. N >= 0.
80*> \endverbatim
81*>
82*> \param[in] KD
83*> \verbatim
84*> KD is INTEGER
85*> The number of superdiagonals of the matrix A if UPLO = 'U',
86*> or the number of subdiagonals if UPLO = 'L'. KD >= 0.
87*> \endverbatim
88*>
89*> \param[in,out] AB
90*> \verbatim
91*> AB is COMPLEX*16 array, dimension (LDAB, N)
92*> On entry, the upper or lower triangle of the Hermitian band
93*> matrix A, stored in the first KD+1 rows of the array. The
94*> j-th column of A is stored in the j-th column of the array AB
95*> as follows:
96*> if UPLO = 'U', AB(kd+1+i-j,j) = A(i,j) for max(1,j-kd)<=i<=j;
97*> if UPLO = 'L', AB(1+i-j,j) = A(i,j) for j<=i<=min(n,j+kd).
98*>
99*> On exit, AB is overwritten by values generated during the
100*> reduction to tridiagonal form. If UPLO = 'U', the first
101*> superdiagonal and the diagonal of the tridiagonal matrix T
102*> are returned in rows KD and KD+1 of AB, and if UPLO = 'L',
103*> the diagonal and first subdiagonal of T are returned in the
104*> first two rows of AB.
105*> \endverbatim
106*>
107*> \param[in] LDAB
108*> \verbatim
109*> LDAB is INTEGER
110*> The leading dimension of the array AB. LDAB >= KD + 1.
111*> \endverbatim
112*>
113*> \param[out] W
114*> \verbatim
115*> W is DOUBLE PRECISION array, dimension (N)
116*> If INFO = 0, the eigenvalues in ascending order.
117*> \endverbatim
118*>
119*> \param[out] Z
120*> \verbatim
121*> Z is COMPLEX*16 array, dimension (LDZ, N)
122*> If JOBZ = 'V', then if INFO = 0, Z contains the orthonormal
123*> eigenvectors of the matrix A, with the i-th column of Z
124*> holding the eigenvector associated with W(i).
125*> If JOBZ = 'N', then Z is not referenced.
126*> \endverbatim
127*>
128*> \param[in] LDZ
129*> \verbatim
130*> LDZ is INTEGER
131*> The leading dimension of the array Z. LDZ >= 1, and if
132*> JOBZ = 'V', LDZ >= max(1,N).
133*> \endverbatim
134*>
135*> \param[out] WORK
136*> \verbatim
137*> WORK is COMPLEX*16 array, dimension (MAX(1,LWORK))
138*> On exit, if INFO = 0, WORK(1) returns the optimal LWORK.
139*> \endverbatim
140*>
141*> \param[in] LWORK
142*> \verbatim
143*> LWORK is INTEGER
144*> The length of the array WORK. LWORK >= 1, when N <= 1;
145*> otherwise
146*> If JOBZ = 'N' and N > 1, LWORK must be queried.
147*> LWORK = MAX(1, dimension) where
148*> dimension = (2KD+1)*N + KD*NTHREADS
149*> where KD is the size of the band.
150*> NTHREADS is the number of threads used when
151*> openMP compilation is enabled, otherwise =1.
152*> If JOBZ = 'V' and N > 1, LWORK must be queried. Not yet available.
153*>
154*> If LWORK = -1, then a workspace query is assumed; the routine
155*> only calculates the optimal sizes of the WORK, RWORK and
156*> IWORK arrays, returns these values as the first entries of
157*> the WORK, RWORK and IWORK arrays, and no error message
158*> related to LWORK or LRWORK or LIWORK is issued by XERBLA.
159*> \endverbatim
160*>
161*> \param[out] RWORK
162*> \verbatim
163*> RWORK is DOUBLE PRECISION array,
164*> dimension (LRWORK)
165*> On exit, if INFO = 0, RWORK(1) returns the optimal LRWORK.
166*> \endverbatim
167*>
168*> \param[in] LRWORK
169*> \verbatim
170*> LRWORK is INTEGER
171*> The dimension of array RWORK.
172*> If N <= 1, LRWORK must be at least 1.
173*> If JOBZ = 'N' and N > 1, LRWORK must be at least N.
174*> If JOBZ = 'V' and N > 1, LRWORK must be at least
175*> 1 + 5*N + 2*N**2.
176*>
177*> If LRWORK = -1, then a workspace query is assumed; the
178*> routine only calculates the optimal sizes of the WORK, RWORK
179*> and IWORK arrays, returns these values as the first entries
180*> of the WORK, RWORK and IWORK arrays, and no error message
181*> related to LWORK or LRWORK or LIWORK is issued by XERBLA.
182*> \endverbatim
183*>
184*> \param[out] IWORK
185*> \verbatim
186*> IWORK is INTEGER array, dimension (MAX(1,LIWORK))
187*> On exit, if INFO = 0, IWORK(1) returns the optimal LIWORK.
188*> \endverbatim
189*>
190*> \param[in] LIWORK
191*> \verbatim
192*> LIWORK is INTEGER
193*> The dimension of array IWORK.
194*> If JOBZ = 'N' or N <= 1, LIWORK must be at least 1.
195*> If JOBZ = 'V' and N > 1, LIWORK must be at least 3 + 5*N .
196*>
197*> If LIWORK = -1, then a workspace query is assumed; the
198*> routine only calculates the optimal sizes of the WORK, RWORK
199*> and IWORK arrays, returns these values as the first entries
200*> of the WORK, RWORK and IWORK arrays, and no error message
201*> related to LWORK or LRWORK or LIWORK is issued by XERBLA.
202*> \endverbatim
203*>
204*> \param[out] INFO
205*> \verbatim
206*> INFO is INTEGER
207*> = 0: successful exit.
208*> < 0: if INFO = -i, the i-th argument had an illegal value.
209*> > 0: if INFO = i, the algorithm failed to converge; i
210*> off-diagonal elements of an intermediate tridiagonal
211*> form did not converge to zero.
212*> \endverbatim
213*
214* Authors:
215* ========
216*
217*> \author Univ. of Tennessee
218*> \author Univ. of California Berkeley
219*> \author Univ. of Colorado Denver
220*> \author NAG Ltd.
221*
222*> \ingroup complex16OTHEReigen
223*
224*> \par Further Details:
225* =====================
226*>
227*> \verbatim
228*>
229*> All details about the 2stage techniques are available in:
230*>
231*> Azzam Haidar, Hatem Ltaief, and Jack Dongarra.
232*> Parallel reduction to condensed forms for symmetric eigenvalue problems
233*> using aggregated fine-grained and memory-aware kernels. In Proceedings
234*> of 2011 International Conference for High Performance Computing,
235*> Networking, Storage and Analysis (SC '11), New York, NY, USA,
236*> Article 8 , 11 pages.
237*> http://doi.acm.org/10.1145/2063384.2063394
238*>
239*> A. Haidar, J. Kurzak, P. Luszczek, 2013.
240*> An improved parallel singular value algorithm and its implementation
241*> for multicore hardware, In Proceedings of 2013 International Conference
242*> for High Performance Computing, Networking, Storage and Analysis (SC '13).
243*> Denver, Colorado, USA, 2013.
244*> Article 90, 12 pages.
245*> http://doi.acm.org/10.1145/2503210.2503292
246*>
247*> A. Haidar, R. Solca, S. Tomov, T. Schulthess and J. Dongarra.
248*> A novel hybrid CPU-GPU generalized eigensolver for electronic structure
249*> calculations based on fine-grained memory aware tasks.
250*> International Journal of High Performance Computing Applications.
251*> Volume 28 Issue 2, Pages 196-209, May 2014.
252*> http://hpc.sagepub.com/content/28/2/196
253*>
254*> \endverbatim
255*
256* =====================================================================
257 SUBROUTINE zhbevd_2stage( JOBZ, UPLO, N, KD, AB, LDAB, W, Z, LDZ,
258 $ WORK, LWORK, RWORK, LRWORK, IWORK,
259 $ LIWORK, INFO )
260*
261 IMPLICIT NONE
262*
263* -- LAPACK driver routine --
264* -- LAPACK is a software package provided by Univ. of Tennessee, --
265* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
266*
267* .. Scalar Arguments ..
268 CHARACTER JOBZ, UPLO
269 INTEGER INFO, KD, LDAB, LDZ, LIWORK, LRWORK, LWORK, N
270* ..
271* .. Array Arguments ..
272 INTEGER IWORK( * )
273 DOUBLE PRECISION RWORK( * ), W( * )
274 COMPLEX*16 AB( LDAB, * ), WORK( * ), Z( LDZ, * )
275* ..
276*
277* =====================================================================
278*
279* .. Parameters ..
280 DOUBLE PRECISION ZERO, ONE
281 PARAMETER ( ZERO = 0.0d0, one = 1.0d0 )
282 COMPLEX*16 CZERO, CONE
283 parameter( czero = ( 0.0d0, 0.0d0 ),
284 $ cone = ( 1.0d0, 0.0d0 ) )
285* ..
286* .. Local Scalars ..
287 LOGICAL LOWER, LQUERY, WANTZ
288 INTEGER IINFO, IMAX, INDE, INDWK2, INDRWK, ISCALE,
289 $ llwork, indwk, lhtrd, lwtrd, ib, indhous,
290 $ liwmin, llrwk, llwk2, lrwmin, lwmin
291 DOUBLE PRECISION ANRM, BIGNUM, EPS, RMAX, RMIN, SAFMIN, SIGMA,
292 $ SMLNUM
293* ..
294* .. External Functions ..
295 LOGICAL LSAME
296 INTEGER ILAENV2STAGE
297 DOUBLE PRECISION DLAMCH, ZLANHB
298 EXTERNAL lsame, dlamch, zlanhb, ilaenv2stage
299* ..
300* .. External Subroutines ..
301 EXTERNAL dscal, dsterf, xerbla, zgemm, zlacpy,
303* ..
304* .. Intrinsic Functions ..
305 INTRINSIC dble, sqrt
306* ..
307* .. Executable Statements ..
308*
309* Test the input parameters.
310*
311 wantz = lsame( jobz, 'V' )
312 lower = lsame( uplo, 'l' )
313.EQ..OR..EQ..OR..EQ. LQUERY = ( LWORK-1 LIWORK-1 LRWORK-1 )
314*
315 INFO = 0
316.LE. IF( N1 ) THEN
317 LWMIN = 1
318 LRWMIN = 1
319 LIWMIN = 1
320 ELSE
321 IB = ILAENV2STAGE( 2, 'zhetrd_hb2st', JOBZ, N, KD, -1, -1 )
322 LHTRD = ILAENV2STAGE( 3, 'zhetrd_hb2st', JOBZ, N, KD, IB, -1 )
323 LWTRD = ILAENV2STAGE( 4, 'zhetrd_hb2st', JOBZ, N, KD, IB, -1 )
324 IF( WANTZ ) THEN
325 LWMIN = 2*N**2
326 LRWMIN = 1 + 5*N + 2*N**2
327 LIWMIN = 3 + 5*N
328 ELSE
329 LWMIN = MAX( N, LHTRD + LWTRD )
330 LRWMIN = N
331 LIWMIN = 1
332 END IF
333 END IF
334.NOT. IF( ( LSAME( JOBZ, 'n' ) ) ) THEN
335 INFO = -1
336.NOT..OR. ELSE IF( ( LOWER LSAME( UPLO, 'u' ) ) ) THEN
337 INFO = -2
338.LT. ELSE IF( N0 ) THEN
339 INFO = -3
340.LT. ELSE IF( KD0 ) THEN
341 INFO = -4
342.LT. ELSE IF( LDABKD+1 ) THEN
343 INFO = -6
344.LT..OR..AND..LT. ELSE IF( LDZ1 ( WANTZ LDZN ) ) THEN
345 INFO = -9
346 END IF
347*
348.EQ. IF( INFO0 ) THEN
349 WORK( 1 ) = LWMIN
350 RWORK( 1 ) = LRWMIN
351 IWORK( 1 ) = LIWMIN
352*
353.LT..AND..NOT. IF( LWORKLWMIN LQUERY ) THEN
354 INFO = -11
355.LT..AND..NOT. ELSE IF( LRWORKLRWMIN LQUERY ) THEN
356 INFO = -13
357.LT..AND..NOT. ELSE IF( LIWORKLIWMIN LQUERY ) THEN
358 INFO = -15
359 END IF
360 END IF
361*
362.NE. IF( INFO0 ) THEN
363 CALL XERBLA( 'zhbevd_2stage', -INFO )
364 RETURN
365 ELSE IF( LQUERY ) THEN
366 RETURN
367 END IF
368*
369* Quick return if possible
370*
371.EQ. IF( N0 )
372 $ RETURN
373*
374.EQ. IF( N1 ) THEN
375 W( 1 ) = DBLE( AB( 1, 1 ) )
376 IF( WANTZ )
377 $ Z( 1, 1 ) = CONE
378 RETURN
379 END IF
380*
381* Get machine constants.
382*
383 SAFMIN = DLAMCH( 'safe minimum' )
384 EPS = DLAMCH( 'precision' )
385 SMLNUM = SAFMIN / EPS
386 BIGNUM = ONE / SMLNUM
387 RMIN = SQRT( SMLNUM )
388 RMAX = SQRT( BIGNUM )
389*
390* Scale matrix to allowable range, if necessary.
391*
392 ANRM = ZLANHB( 'm', UPLO, N, KD, AB, LDAB, RWORK )
393 ISCALE = 0
394.GT..AND..LT. IF( ANRMZERO ANRMRMIN ) THEN
395 ISCALE = 1
396 SIGMA = RMIN / ANRM
397.GT. ELSE IF( ANRMRMAX ) THEN
398 ISCALE = 1
399 SIGMA = RMAX / ANRM
400 END IF
401.EQ. IF( ISCALE1 ) THEN
402 IF( LOWER ) THEN
403 CALL ZLASCL( 'b', KD, KD, ONE, SIGMA, N, N, AB, LDAB, INFO )
404 ELSE
405 CALL ZLASCL( 'q', KD, KD, ONE, SIGMA, N, N, AB, LDAB, INFO )
406 END IF
407 END IF
408*
409* Call ZHBTRD_HB2ST to reduce Hermitian band matrix to tridiagonal form.
410*
411 INDE = 1
412 INDRWK = INDE + N
413 LLRWK = LRWORK - INDRWK + 1
414 INDHOUS = 1
415 INDWK = INDHOUS + LHTRD
416 LLWORK = LWORK - INDWK + 1
417 INDWK2 = INDWK + N*N
418 LLWK2 = LWORK - INDWK2 + 1
419*
420 CALL ZHETRD_HB2ST( "N", JOBZ, UPLO, N, KD, AB, LDAB, W,
421 $ RWORK( INDE ), WORK( INDHOUS ), LHTRD,
422 $ WORK( INDWK ), LLWORK, IINFO )
423*
424* For eigenvalues only, call DSTERF. For eigenvectors, call ZSTEDC.
425*
426.NOT. IF( WANTZ ) THEN
427 CALL DSTERF( N, W, RWORK( INDE ), INFO )
428 ELSE
429 CALL ZSTEDC( 'i', N, W, RWORK( INDE ), WORK, N, WORK( INDWK2 ),
430 $ LLWK2, RWORK( INDRWK ), LLRWK, IWORK, LIWORK,
431 $ INFO )
432 CALL ZGEMM( 'n', 'n', N, N, N, CONE, Z, LDZ, WORK, N, CZERO,
433 $ WORK( INDWK2 ), N )
434 CALL ZLACPY( 'a', N, N, WORK( INDWK2 ), N, Z, LDZ )
435 END IF
436*
437* If matrix was scaled, then rescale eigenvalues appropriately.
438*
439.EQ. IF( ISCALE1 ) THEN
440.EQ. IF( INFO0 ) THEN
441 IMAX = N
442 ELSE
443 IMAX = INFO - 1
444 END IF
445 CALL DSCAL( IMAX, ONE / SIGMA, W, 1 )
446 END IF
447*
448 WORK( 1 ) = LWMIN
449 RWORK( 1 ) = LRWMIN
450 IWORK( 1 ) = LIWMIN
451 RETURN
452*
453* End of ZHBEVD_2STAGE
454*
455 END
subroutine dsterf(n, d, e, info)
DSTERF
Definition dsterf.f:86
subroutine xerbla(srname, info)
XERBLA
Definition xerbla.f:60
subroutine zlascl(type, kl, ku, cfrom, cto, m, n, a, lda, info)
ZLASCL multiplies a general rectangular matrix by a real scalar defined as cto/cfrom.
Definition zlascl.f:143
subroutine zlacpy(uplo, m, n, a, lda, b, ldb)
ZLACPY copies all or part of one two-dimensional array to another.
Definition zlacpy.f:103
subroutine zhetrd_hb2st(stage1, vect, uplo, n, kd, ab, ldab, d, e, hous, lhous, work, lwork, info)
ZHETRD_HB2ST reduces a complex Hermitian band matrix A to real symmetric tridiagonal form T
subroutine zstedc(compz, n, d, e, z, ldz, work, lwork, rwork, lrwork, iwork, liwork, info)
ZSTEDC
Definition zstedc.f:212
subroutine zhbevd_2stage(jobz, uplo, n, kd, ab, ldab, w, z, ldz, work, lwork, rwork, lrwork, iwork, liwork, info)
ZHBEVD_2STAGE computes the eigenvalues and, optionally, the left and/or right eigenvectors for OTHER ...
subroutine zgemm(transa, transb, m, n, k, alpha, a, lda, b, ldb, beta, c, ldc)
ZGEMM
Definition zgemm.f:187
subroutine dscal(n, da, dx, incx)
DSCAL
Definition dscal.f:79