1*> \brief <b> CHEEVR computes the eigenvalues and, optionally, the left and/or right eigenvectors for HE matrices</b>
2*
3*  =========== DOCUMENTATION ===========
4*
5* Online html documentation available at
6*            http://www.netlib.org/lapack/explore-html/
7*
8*> \htmlonly
9*> Download CHEEVR + dependencies
10*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/cheevr.f">
11*> [TGZ]</a>
12*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/cheevr.f">
13*> [ZIP]</a>
14*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/cheevr.f">
15*> [TXT]</a>
16*> \endhtmlonly
17*
18*  Definition:
19*  ===========
20*
21*       SUBROUTINE CHEEVR( JOBZ, RANGE, UPLO, N, A, LDA, VL, VU, IL, IU,
22*                          ABSTOL, M, W, Z, LDZ, ISUPPZ, WORK, LWORK,
23*                          RWORK, LRWORK, IWORK, LIWORK, INFO )
24*
25*       .. Scalar Arguments ..
26*       CHARACTER          JOBZ, RANGE, UPLO
27*       INTEGER            IL, INFO, IU, LDA, LDZ, LIWORK, LRWORK, LWORK,
28*      $                   M, N
29*       REAL               ABSTOL, VL, VU
30*       ..
31*       .. Array Arguments ..
32*       INTEGER            ISUPPZ( * ), IWORK( * )
33*       REAL               RWORK( * ), W( * )
34*       COMPLEX            A( LDA, * ), WORK( * ), Z( LDZ, * )
35*       ..
36*
37*
38*> \par Purpose:
39*  =============
40*>
41*> \verbatim
42*>
43*> CHEEVR computes selected eigenvalues and, optionally, eigenvectors
44*> of a complex Hermitian matrix A.  Eigenvalues and eigenvectors can
45*> be selected by specifying either a range of values or a range of
46*> indices for the desired eigenvalues.
47*>
48*> CHEEVR first reduces the matrix A to tridiagonal form T with a call
49*> to CHETRD.  Then, whenever possible, CHEEVR calls CSTEMR to compute
50*> the eigenspectrum using Relatively Robust Representations.  CSTEMR
51*> computes eigenvalues by the dqds algorithm, while orthogonal
52*> eigenvectors are computed from various "good" L D L^T representations
53*> (also known as Relatively Robust Representations). Gram-Schmidt
54*> orthogonalization is avoided as far as possible. More specifically,
55*> the various steps of the algorithm are as follows.
56*>
57*> For each unreduced block (submatrix) of T,
58*>    (a) Compute T - sigma I  = L D L^T, so that L and D
59*>        define all the wanted eigenvalues to high relative accuracy.
60*>        This means that small relative changes in the entries of D and L
61*>        cause only small relative changes in the eigenvalues and
62*>        eigenvectors. The standard (unfactored) representation of the
63*>        tridiagonal matrix T does not have this property in general.
64*>    (b) Compute the eigenvalues to suitable accuracy.
65*>        If the eigenvectors are desired, the algorithm attains full
66*>        accuracy of the computed eigenvalues only right before
67*>        the corresponding vectors have to be computed, see steps c) and d).
68*>    (c) For each cluster of close eigenvalues, select a new
69*>        shift close to the cluster, find a new factorization, and refine
70*>        the shifted eigenvalues to suitable accuracy.
71*>    (d) For each eigenvalue with a large enough relative separation compute
72*>        the corresponding eigenvector by forming a rank revealing twisted
73*>        factorization. Go back to (c) for any clusters that remain.
74*>
75*> The desired accuracy of the output can be specified by the input
76*> parameter ABSTOL.
77*>
78*> For more details, see DSTEMR's documentation and:
79*> - Inderjit S. Dhillon and Beresford N. Parlett: "Multiple representations
80*>   to compute orthogonal eigenvectors of symmetric tridiagonal matrices,"
81*>   Linear Algebra and its Applications, 387(1), pp. 1-28, August 2004.
82*> - Inderjit Dhillon and Beresford Parlett: "Orthogonal Eigenvectors and
83*>   Relative Gaps," SIAM Journal on Matrix Analysis and Applications, Vol. 25,
84*>   2004.  Also LAPACK Working Note 154.
85*> - Inderjit Dhillon: "A new O(n^2) algorithm for the symmetric
86*>   tridiagonal eigenvalue/eigenvector problem",
87*>   Computer Science Division Technical Report No. UCB/CSD-97-971,
88*>   UC Berkeley, May 1997.
89*>
90*>
91*> Note 1 : CHEEVR calls CSTEMR when the full spectrum is requested
92*> on machines which conform to the ieee-754 floating point standard.
93*> CHEEVR calls SSTEBZ and CSTEIN on non-ieee machines and
94*> when partial spectrum requests are made.
95*>
96*> Normal execution of CSTEMR may create NaNs and infinities and
97*> hence may abort due to a floating point exception in environments
98*> which do not handle NaNs and infinities in the ieee standard default
99*> manner.
100*> \endverbatim
101*
102*  Arguments:
103*  ==========
104*
105*> \param[in] JOBZ
106*> \verbatim
107*>          JOBZ is CHARACTER*1
108*>          = 'N':  Compute eigenvalues only;
109*>          = 'V':  Compute eigenvalues and eigenvectors.
110*> \endverbatim
111*>
112*> \param[in] RANGE
113*> \verbatim
114*>          RANGE is CHARACTER*1
115*>          = 'A': all eigenvalues will be found.
116*>          = 'V': all eigenvalues in the half-open interval (VL,VU]
117*>                 will be found.
118*>          = 'I': the IL-th through IU-th eigenvalues will be found.
119*>          For RANGE = 'V' or 'I' and IU - IL < N - 1, SSTEBZ and
120*>          CSTEIN are called
121*> \endverbatim
122*>
123*> \param[in] UPLO
124*> \verbatim
125*>          UPLO is CHARACTER*1
126*>          = 'U':  Upper triangle of A is stored;
127*>          = 'L':  Lower triangle of A is stored.
128*> \endverbatim
129*>
130*> \param[in] N
131*> \verbatim
132*>          N is INTEGER
133*>          The order of the matrix A.  N >= 0.
134*> \endverbatim
135*>
136*> \param[in,out] A
137*> \verbatim
138*>          A is COMPLEX array, dimension (LDA, N)
139*>          On entry, the Hermitian matrix A.  If UPLO = 'U', the
140*>          leading N-by-N upper triangular part of A contains the
141*>          upper triangular part of the matrix A.  If UPLO = 'L',
142*>          the leading N-by-N lower triangular part of A contains
143*>          the lower triangular part of the matrix A.
144*>          On exit, the lower triangle (if UPLO='L') or the upper
145*>          triangle (if UPLO='U') of A, including the diagonal, is
146*>          destroyed.
147*> \endverbatim
148*>
149*> \param[in] LDA
150*> \verbatim
151*>          LDA is INTEGER
152*>          The leading dimension of the array A.  LDA >= max(1,N).
153*> \endverbatim
154*>
155*> \param[in] VL
156*> \verbatim
157*>          VL is REAL
158*> \endverbatim
159*>
160*> \param[in] VU
161*> \verbatim
162*>          VU is REAL
163*>          If RANGE='V', the lower and upper bounds of the interval to
164*>          be searched for eigenvalues. VL < VU.
165*>          Not referenced if RANGE = 'A' or 'I'.
166*> \endverbatim
167*>
168*> \param[in] IL
169*> \verbatim
170*>          IL is INTEGER
171*> \endverbatim
172*>
173*> \param[in] IU
174*> \verbatim
175*>          IU is INTEGER
176*>          If RANGE='I', the indices (in ascending order) of the
177*>          smallest and largest eigenvalues to be returned.
178*>          1 <= IL <= IU <= N, if N > 0; IL = 1 and IU = 0 if N = 0.
179*>          Not referenced if RANGE = 'A' or 'V'.
180*> \endverbatim
181*>
182*> \param[in] ABSTOL
183*> \verbatim
184*>          ABSTOL is REAL
185*>          The absolute error tolerance for the eigenvalues.
186*>          An approximate eigenvalue is accepted as converged
187*>          when it is determined to lie in an interval [a,b]
188*>          of width less than or equal to
189*>
190*>                  ABSTOL + EPS *   max( |a|,|b| ) ,
191*>
192*>          where EPS is the machine precision.  If ABSTOL is less than
193*>          or equal to zero, then  EPS*|T|  will be used in its place,
194*>          where |T| is the 1-norm of the tridiagonal matrix obtained
195*>          by reducing A to tridiagonal form.
196*>
197*>          See "Computing Small Singular Values of Bidiagonal Matrices
198*>          with Guaranteed High Relative Accuracy," by Demmel and
199*>          Kahan, LAPACK Working Note #3.
200*>
201*>          If high relative accuracy is important, set ABSTOL to
202*>          SLAMCH( 'Safe minimum' ).  Doing so will guarantee that
203*>          eigenvalues are computed to high relative accuracy when
204*>          possible in future releases.  The current code does not
205*>          make any guarantees about high relative accuracy, but
206*>          furutre releases will. See J. Barlow and J. Demmel,
207*>          "Computing Accurate Eigensystems of Scaled Diagonally
208*>          Dominant Matrices", LAPACK Working Note #7, for a discussion
209*>          of which matrices define their eigenvalues to high relative
210*>          accuracy.
211*> \endverbatim
212*>
213*> \param[out] M
214*> \verbatim
215*>          M is INTEGER
216*>          The total number of eigenvalues found.  0 <= M <= N.
217*>          If RANGE = 'A', M = N, and if RANGE = 'I', M = IU-IL+1.
218*> \endverbatim
219*>
220*> \param[out] W
221*> \verbatim
222*>          W is REAL array, dimension (N)
223*>          The first M elements contain the selected eigenvalues in
224*>          ascending order.
225*> \endverbatim
226*>
227*> \param[out] Z
228*> \verbatim
229*>          Z is COMPLEX array, dimension (LDZ, max(1,M))
230*>          If JOBZ = 'V', then if INFO = 0, the first M columns of Z
231*>          contain the orthonormal eigenvectors of the matrix A
232*>          corresponding to the selected eigenvalues, with the i-th
233*>          column of Z holding the eigenvector associated with W(i).
234*>          If JOBZ = 'N', then Z is not referenced.
235*>          Note: the user must ensure that at least max(1,M) columns are
236*>          supplied in the array Z; if RANGE = 'V', the exact value of M
237*>          is not known in advance and an upper bound must be used.
238*> \endverbatim
239*>
240*> \param[in] LDZ
241*> \verbatim
242*>          LDZ is INTEGER
243*>          The leading dimension of the array Z.  LDZ >= 1, and if
244*>          JOBZ = 'V', LDZ >= max(1,N).
245*> \endverbatim
246*>
247*> \param[out] ISUPPZ
248*> \verbatim
249*>          ISUPPZ is INTEGER array, dimension ( 2*max(1,M) )
250*>          The support of the eigenvectors in Z, i.e., the indices
251*>          indicating the nonzero elements in Z. The i-th eigenvector
252*>          is nonzero only in elements ISUPPZ( 2*i-1 ) through
253*>          ISUPPZ( 2*i ).
254*>          Implemented only for RANGE = 'A' or 'I' and IU - IL = N - 1
255*> \endverbatim
256*>
257*> \param[out] WORK
258*> \verbatim
259*>          WORK is COMPLEX array, dimension (MAX(1,LWORK))
260*>          On exit, if INFO = 0, WORK(1) returns the optimal LWORK.
261*> \endverbatim
262*>
263*> \param[in] LWORK
264*> \verbatim
265*>          LWORK is INTEGER
266*>          The length of the array WORK.  LWORK >= max(1,2*N).
267*>          For optimal efficiency, LWORK >= (NB+1)*N,
268*>          where NB is the max of the blocksize for CHETRD and for
269*>          CUNMTR as returned by ILAENV.
270*>
271*>          If LWORK = -1, then a workspace query is assumed; the routine
272*>          only calculates the optimal sizes of the WORK, RWORK and
273*>          IWORK arrays, returns these values as the first entries of
274*>          the WORK, RWORK and IWORK arrays, and no error message
275*>          related to LWORK or LRWORK or LIWORK is issued by XERBLA.
276*> \endverbatim
277*>
278*> \param[out] RWORK
279*> \verbatim
280*>          RWORK is REAL array, dimension (MAX(1,LRWORK))
281*>          On exit, if INFO = 0, RWORK(1) returns the optimal
282*>          (and minimal) LRWORK.
283*> \endverbatim
284*>
285*> \param[in] LRWORK
286*> \verbatim
287*>          LRWORK is INTEGER
288*>          The length of the array RWORK.  LRWORK >= max(1,24*N).
289*>
290*>          If LRWORK = -1, then a workspace query is assumed; the
291*>          routine only calculates the optimal sizes of the WORK, RWORK
292*>          and IWORK arrays, returns these values as the first entries
293*>          of the WORK, RWORK and IWORK arrays, and no error message
294*>          related to LWORK or LRWORK or LIWORK is issued by XERBLA.
295*> \endverbatim
296*>
297*> \param[out] IWORK
298*> \verbatim
299*>          IWORK is INTEGER array, dimension (MAX(1,LIWORK))
300*>          On exit, if INFO = 0, IWORK(1) returns the optimal
301*>          (and minimal) LIWORK.
302*> \endverbatim
303*>
304*> \param[in] LIWORK
305*> \verbatim
306*>          LIWORK is INTEGER
307*>          The dimension of the array IWORK.  LIWORK >= max(1,10*N).
308*>
309*>          If LIWORK = -1, then a workspace query is assumed; the
310*>          routine only calculates the optimal sizes of the WORK, RWORK
311*>          and IWORK arrays, returns these values as the first entries
312*>          of the WORK, RWORK and IWORK arrays, and no error message
313*>          related to LWORK or LRWORK or LIWORK is issued by XERBLA.
314*> \endverbatim
315*>
316*> \param[out] INFO
317*> \verbatim
318*>          INFO is INTEGER
319*>          = 0:  successful exit
320*>          < 0:  if INFO = -i, the i-th argument had an illegal value
321*>          > 0:  Internal error
322*> \endverbatim
323*
324*  Authors:
325*  ========
326*
327*> \author Univ. of Tennessee
328*> \author Univ. of California Berkeley
329*> \author Univ. of Colorado Denver
330*> \author NAG Ltd.
331*
332*> \date September 2012
333*
334*> \ingroup complexHEeigen
335*
336*> \par Contributors:
337*  ==================
338*>
339*>     Inderjit Dhillon, IBM Almaden, USA \n
340*>     Osni Marques, LBNL/NERSC, USA \n
341*>     Ken Stanley, Computer Science Division, University of
342*>       California at Berkeley, USA \n
343*>     Jason Riedy, Computer Science Division, University of
344*>       California at Berkeley, USA \n
345*>
346*  =====================================================================
347      SUBROUTINE CHEEVR( JOBZ, RANGE, UPLO, N, A, LDA, VL, VU, IL, IU,
348     $                   ABSTOL, M, W, Z, LDZ, ISUPPZ, WORK, LWORK,
349     $                   RWORK, LRWORK, IWORK, LIWORK, INFO )
350*
351*  -- LAPACK driver routine (version 3.4.2) --
352*  -- LAPACK is a software package provided by Univ. of Tennessee,    --
353*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
354*     September 2012
355*
356*     .. Scalar Arguments ..
357      CHARACTER          JOBZ, RANGE, UPLO
358      INTEGER            IL, INFO, IU, LDA, LDZ, LIWORK, LRWORK, LWORK,
359     $                   M, N
360      REAL               ABSTOL, VL, VU
361*     ..
362*     .. Array Arguments ..
363      INTEGER            ISUPPZ( * ), IWORK( * )
364      REAL               RWORK( * ), W( * )
365      COMPLEX            A( LDA, * ), WORK( * ), Z( LDZ, * )
366*     ..
367*
368* =====================================================================
369*
370*     .. Parameters ..
371      REAL               ZERO, ONE, TWO
372      PARAMETER          ( ZERO = 0.0E+0, ONE = 1.0E+0, TWO = 2.0E+0 )
373*     ..
374*     .. Local Scalars ..
375      LOGICAL            ALLEIG, INDEIG, LOWER, LQUERY, TEST, VALEIG,
376     $                   WANTZ, TRYRAC
377      CHARACTER          ORDER
378      INTEGER            I, IEEEOK, IINFO, IMAX, INDIBL, INDIFL, INDISP,
379     $                   INDIWO, INDRD, INDRDD, INDRE, INDREE, INDRWK,
380     $                   INDTAU, INDWK, INDWKN, ISCALE, ITMP1, J, JJ,
381     $                   LIWMIN, LLWORK, LLRWORK, LLWRKN, LRWMIN,
382     $                   LWKOPT, LWMIN, NB, NSPLIT
383      REAL               ABSTLL, ANRM, BIGNUM, EPS, RMAX, RMIN, SAFMIN,
384     $                   SIGMA, SMLNUM, TMP1, VLL, VUU
385*     ..
386*     .. External Functions ..
387      LOGICAL            LSAME
388      INTEGER            ILAENV
389      REAL               CLANSY, SLAMCH
390      EXTERNAL           LSAME, ILAENV, CLANSY, SLAMCH
391*     ..
392*     .. External Subroutines ..
393      EXTERNAL           CHETRD, CSSCAL, CSTEMR, CSTEIN, CSWAP, CUNMTR,
394     $                   SCOPY, SSCAL, SSTEBZ, SSTERF, XERBLA
395*     ..
396*     .. Intrinsic Functions ..
397      INTRINSIC          MAX, MIN, REAL, SQRT
398*     ..
399*     .. Executable Statements ..
400*
401*     Test the input parameters.
402*
403      IEEEOK = ILAENV( 10, 'CHEEVR', 'N', 1, 2, 3, 4 )
404*
405      LOWER = LSAME( UPLO, 'L' )
406      WANTZ = LSAME( JOBZ, 'V' )
407      ALLEIG = LSAME( RANGE, 'A' )
408      VALEIG = LSAME( RANGE, 'V' )
409      INDEIG = LSAME( RANGE, 'I' )
410*
411      LQUERY = ( ( LWORK.EQ.-1 ) .OR. ( LRWORK.EQ.-1 ) .OR.
412     $         ( LIWORK.EQ.-1 ) )
413*
414      LRWMIN = MAX( 1, 24*N )
415      LIWMIN = MAX( 1, 10*N )
416      LWMIN = MAX( 1, 2*N )
417*
418      INFO = 0
419      IF( .NOT.( WANTZ .OR. LSAME( JOBZ, 'N' ) ) ) THEN
420         INFO = -1
421      ELSE IF( .NOT.( ALLEIG .OR. VALEIG .OR. INDEIG ) ) THEN
422         INFO = -2
423      ELSE IF( .NOT.( LOWER .OR. LSAME( UPLO, 'U' ) ) ) THEN
424         INFO = -3
425      ELSE IF( N.LT.0 ) THEN
426         INFO = -4
427      ELSE IF( LDA.LT.MAX( 1, N ) ) THEN
428         INFO = -6
429      ELSE
430         IF( VALEIG ) THEN
431            IF( N.GT.0 .AND. VU.LE.VL )
432     $         INFO = -8
433         ELSE IF( INDEIG ) THEN
434            IF( IL.LT.1 .OR. IL.GT.MAX( 1, N ) ) THEN
435               INFO = -9
436            ELSE IF( IU.LT.MIN( N, IL ) .OR. IU.GT.N ) THEN
437               INFO = -10
438            END IF
439         END IF
440      END IF
441      IF( INFO.EQ.0 ) THEN
442         IF( LDZ.LT.1 .OR. ( WANTZ .AND. LDZ.LT.N ) ) THEN
443            INFO = -15
444         END IF
445      END IF
446*
447      IF( INFO.EQ.0 ) THEN
448         NB = ILAENV( 1, 'CHETRD', UPLO, N, -1, -1, -1 )
449         NB = MAX( NB, ILAENV( 1, 'CUNMTR', UPLO, N, -1, -1, -1 ) )
450         LWKOPT = MAX( ( NB+1 )*N, LWMIN )
451         WORK( 1 ) = LWKOPT
452         RWORK( 1 ) = LRWMIN
453         IWORK( 1 ) = LIWMIN
454*
455         IF( LWORK.LT.LWMIN .AND. .NOT.LQUERY ) THEN
456            INFO = -18
457         ELSE IF( LRWORK.LT.LRWMIN .AND. .NOT.LQUERY ) THEN
458            INFO = -20
459         ELSE IF( LIWORK.LT.LIWMIN .AND. .NOT.LQUERY ) THEN
460            INFO = -22
461         END IF
462      END IF
463*
464      IF( INFO.NE.0 ) THEN
465         CALL XERBLA( 'CHEEVR', -INFO )
466         RETURN
467      ELSE IF( LQUERY ) THEN
468         RETURN
469      END IF
470*
471*     Quick return if possible
472*
473      M = 0
474      IF( N.EQ.0 ) THEN
475         WORK( 1 ) = 1
476         RETURN
477      END IF
478*
479      IF( N.EQ.1 ) THEN
480         WORK( 1 ) = 2
481         IF( ALLEIG .OR. INDEIG ) THEN
482            M = 1
483            W( 1 ) = REAL( A( 1, 1 ) )
484         ELSE
485            IF( VL.LT.REAL( A( 1, 1 ) ) .AND. VU.GE.REAL( A( 1, 1 ) ) )
486     $           THEN
487               M = 1
488               W( 1 ) = REAL( A( 1, 1 ) )
489            END IF
490         END IF
491         IF( WANTZ ) THEN
492            Z( 1, 1 ) = ONE
493            ISUPPZ( 1 ) = 1
494            ISUPPZ( 2 ) = 1
495         END IF
496         RETURN
497      END IF
498*
499*     Get machine constants.
500*
501      SAFMIN = SLAMCH( 'Safe minimum' )
502      EPS = SLAMCH( 'Precision' )
503      SMLNUM = SAFMIN / EPS
504      BIGNUM = ONE / SMLNUM
505      RMIN = SQRT( SMLNUM )
506      RMAX = MIN( SQRT( BIGNUM ), ONE / SQRT( SQRT( SAFMIN ) ) )
507*
508*     Scale matrix to allowable range, if necessary.
509*
510      ISCALE = 0
511      ABSTLL = ABSTOL
512      IF (VALEIG) THEN
513         VLL = VL
514         VUU = VU
515      END IF
516      ANRM = CLANSY( 'M', UPLO, N, A, LDA, RWORK )
517      IF( ANRM.GT.ZERO .AND. ANRM.LT.RMIN ) THEN
518         ISCALE = 1
519         SIGMA = RMIN / ANRM
520      ELSE IF( ANRM.GT.RMAX ) THEN
521         ISCALE = 1
522         SIGMA = RMAX / ANRM
523      END IF
524      IF( ISCALE.EQ.1 ) THEN
525         IF( LOWER ) THEN
526            DO 10 J = 1, N
527               CALL CSSCAL( N-J+1, SIGMA, A( J, J ), 1 )
528   10       CONTINUE
529         ELSE
530            DO 20 J = 1, N
531               CALL CSSCAL( J, SIGMA, A( 1, J ), 1 )
532   20       CONTINUE
533         END IF
534         IF( ABSTOL.GT.0 )
535     $      ABSTLL = ABSTOL*SIGMA
536         IF( VALEIG ) THEN
537            VLL = VL*SIGMA
538            VUU = VU*SIGMA
539         END IF
540      END IF
541
542*     Initialize indices into workspaces.  Note: The IWORK indices are
543*     used only if SSTERF or CSTEMR fail.
544
545*     WORK(INDTAU:INDTAU+N-1) stores the complex scalar factors of the
546*     elementary reflectors used in CHETRD.
547      INDTAU = 1
548*     INDWK is the starting offset of the remaining complex workspace,
549*     and LLWORK is the remaining complex workspace size.
550      INDWK = INDTAU + N
551      LLWORK = LWORK - INDWK + 1
552
553*     RWORK(INDRD:INDRD+N-1) stores the real tridiagonal's diagonal
554*     entries.
555      INDRD = 1
556*     RWORK(INDRE:INDRE+N-1) stores the off-diagonal entries of the
557*     tridiagonal matrix from CHETRD.
558      INDRE = INDRD + N
559*     RWORK(INDRDD:INDRDD+N-1) is a copy of the diagonal entries over
560*     -written by CSTEMR (the SSTERF path copies the diagonal to W).
561      INDRDD = INDRE + N
562*     RWORK(INDREE:INDREE+N-1) is a copy of the off-diagonal entries over
563*     -written while computing the eigenvalues in SSTERF and CSTEMR.
564      INDREE = INDRDD + N
565*     INDRWK is the starting offset of the left-over real workspace, and
566*     LLRWORK is the remaining workspace size.
567      INDRWK = INDREE + N
568      LLRWORK = LRWORK - INDRWK + 1
569
570*     IWORK(INDIBL:INDIBL+M-1) corresponds to IBLOCK in SSTEBZ and
571*     stores the block indices of each of the M<=N eigenvalues.
572      INDIBL = 1
573*     IWORK(INDISP:INDISP+NSPLIT-1) corresponds to ISPLIT in SSTEBZ and
574*     stores the starting and finishing indices of each block.
575      INDISP = INDIBL + N
576*     IWORK(INDIFL:INDIFL+N-1) stores the indices of eigenvectors
577*     that corresponding to eigenvectors that fail to converge in
578*     SSTEIN.  This information is discarded; if any fail, the driver
579*     returns INFO > 0.
580      INDIFL = INDISP + N
581*     INDIWO is the offset of the remaining integer workspace.
582      INDIWO = INDIFL + N
583
584*
585*     Call CHETRD to reduce Hermitian matrix to tridiagonal form.
586*
587      CALL CHETRD( UPLO, N, A, LDA, RWORK( INDRD ), RWORK( INDRE ),
588     $             WORK( INDTAU ), WORK( INDWK ), LLWORK, IINFO )
589*
590*     If all eigenvalues are desired
591*     then call SSTERF or CSTEMR and CUNMTR.
592*
593      TEST = .FALSE.
594      IF( INDEIG ) THEN
595         IF( IL.EQ.1 .AND. IU.EQ.N ) THEN
596            TEST = .TRUE.
597         END IF
598      END IF
599      IF( ( ALLEIG.OR.TEST ) .AND. ( IEEEOK.EQ.1 ) ) THEN
600         IF( .NOT.WANTZ ) THEN
601            CALL SCOPY( N, RWORK( INDRD ), 1, W, 1 )
602            CALL SCOPY( N-1, RWORK( INDRE ), 1, RWORK( INDREE ), 1 )
603            CALL SSTERF( N, W, RWORK( INDREE ), INFO )
604         ELSE
605            CALL SCOPY( N-1, RWORK( INDRE ), 1, RWORK( INDREE ), 1 )
606            CALL SCOPY( N, RWORK( INDRD ), 1, RWORK( INDRDD ), 1 )
607*
608            IF (ABSTOL .LE. TWO*N*EPS) THEN
609               TRYRAC = .TRUE.
610            ELSE
611               TRYRAC = .FALSE.
612            END IF
613            CALL CSTEMR( JOBZ, 'A', N, RWORK( INDRDD ),
614     $                   RWORK( INDREE ), VL, VU, IL, IU, M, W,
615     $                   Z, LDZ, N, ISUPPZ, TRYRAC,
616     $                   RWORK( INDRWK ), LLRWORK,
617     $                   IWORK, LIWORK, INFO )
618*
619*           Apply unitary matrix used in reduction to tridiagonal
620*           form to eigenvectors returned by CSTEIN.
621*
622            IF( WANTZ .AND. INFO.EQ.0 ) THEN
623               INDWKN = INDWK
624               LLWRKN = LWORK - INDWKN + 1
625               CALL CUNMTR( 'L', UPLO, 'N', N, M, A, LDA,
626     $                      WORK( INDTAU ), Z, LDZ, WORK( INDWKN ),
627     $                      LLWRKN, IINFO )
628            END IF
629         END IF
630*
631*
632         IF( INFO.EQ.0 ) THEN
633            M = N
634            GO TO 30
635         END IF
636         INFO = 0
637      END IF
638*
639*     Otherwise, call SSTEBZ and, if eigenvectors are desired, CSTEIN.
640*     Also call SSTEBZ and CSTEIN if CSTEMR fails.
641*
642      IF( WANTZ ) THEN
643         ORDER = 'B'
644      ELSE
645         ORDER = 'E'
646      END IF
647
648      CALL SSTEBZ( RANGE, ORDER, N, VLL, VUU, IL, IU, ABSTLL,
649     $             RWORK( INDRD ), RWORK( INDRE ), M, NSPLIT, W,
650     $             IWORK( INDIBL ), IWORK( INDISP ), RWORK( INDRWK ),
651     $             IWORK( INDIWO ), INFO )
652*
653      IF( WANTZ ) THEN
654         CALL CSTEIN( N, RWORK( INDRD ), RWORK( INDRE ), M, W,
655     $                IWORK( INDIBL ), IWORK( INDISP ), Z, LDZ,
656     $                RWORK( INDRWK ), IWORK( INDIWO ), IWORK( INDIFL ),
657     $                INFO )
658*
659*        Apply unitary matrix used in reduction to tridiagonal
660*        form to eigenvectors returned by CSTEIN.
661*
662         INDWKN = INDWK
663         LLWRKN = LWORK - INDWKN + 1
664         CALL CUNMTR( 'L', UPLO, 'N', N, M, A, LDA, WORK( INDTAU ), Z,
665     $                LDZ, WORK( INDWKN ), LLWRKN, IINFO )
666      END IF
667*
668*     If matrix was scaled, then rescale eigenvalues appropriately.
669*
670   30 CONTINUE
671      IF( ISCALE.EQ.1 ) THEN
672         IF( INFO.EQ.0 ) THEN
673            IMAX = M
674         ELSE
675            IMAX = INFO - 1
676         END IF
677         CALL SSCAL( IMAX, ONE / SIGMA, W, 1 )
678      END IF
679*
680*     If eigenvalues are not in order, then sort them, along with
681*     eigenvectors.
682*
683      IF( WANTZ ) THEN
684         DO 50 J = 1, M - 1
685            I = 0
686            TMP1 = W( J )
687            DO 40 JJ = J + 1, M
688               IF( W( JJ ).LT.TMP1 ) THEN
689                  I = JJ
690                  TMP1 = W( JJ )
691               END IF
692   40       CONTINUE
693*
694            IF( I.NE.0 ) THEN
695               ITMP1 = IWORK( INDIBL+I-1 )
696               W( I ) = W( J )
697               IWORK( INDIBL+I-1 ) = IWORK( INDIBL+J-1 )
698               W( J ) = TMP1
699               IWORK( INDIBL+J-1 ) = ITMP1
700               CALL CSWAP( N, Z( 1, I ), 1, Z( 1, J ), 1 )
701            END IF
702   50    CONTINUE
703      END IF
704*
705*     Set WORK(1) to optimal workspace size.
706*
707      WORK( 1 ) = LWKOPT
708      RWORK( 1 ) = LRWMIN
709      IWORK( 1 ) = LIWMIN
710*
711      RETURN
712*
713*     End of CHEEVR
714*
715      END
716