1*> \brief \b ZSTEMR
2*
3*  =========== DOCUMENTATION ===========
4*
5* Online html documentation available at
6*            http://www.netlib.org/lapack/explore-html/
7*
8*> \htmlonly
9*> Download ZSTEMR + dependencies
10*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/zstemr.f">
11*> [TGZ]</a>
12*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/zstemr.f">
13*> [ZIP]</a>
14*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/zstemr.f">
15*> [TXT]</a>
16*> \endhtmlonly
17*
18*  Definition:
19*  ===========
20*
21*       SUBROUTINE ZSTEMR( JOBZ, RANGE, N, D, E, VL, VU, IL, IU,
22*                          M, W, Z, LDZ, NZC, ISUPPZ, TRYRAC, WORK, LWORK,
23*                          IWORK, LIWORK, INFO )
24*
25*       .. Scalar Arguments ..
26*       CHARACTER          JOBZ, RANGE
27*       LOGICAL            TRYRAC
28*       INTEGER            IL, INFO, IU, LDZ, NZC, LIWORK, LWORK, M, N
29*       DOUBLE PRECISION VL, VU
30*       ..
31*       .. Array Arguments ..
32*       INTEGER            ISUPPZ( * ), IWORK( * )
33*       DOUBLE PRECISION   D( * ), E( * ), W( * ), WORK( * )
34*       COMPLEX*16         Z( LDZ, * )
35*       ..
36*
37*
38*> \par Purpose:
39*  =============
40*>
41*> \verbatim
42*>
43*> ZSTEMR computes selected eigenvalues and, optionally, eigenvectors
44*> of a real symmetric tridiagonal matrix T. Any such unreduced matrix has
45*> a well defined set of pairwise different real eigenvalues, the corresponding
46*> real eigenvectors are pairwise orthogonal.
47*>
48*> The spectrum may be computed either completely or partially by specifying
49*> either an interval (VL,VU] or a range of indices IL:IU for the desired
50*> eigenvalues.
51*>
52*> Depending on the number of desired eigenvalues, these are computed either
53*> by bisection or the dqds algorithm. Numerically orthogonal eigenvectors are
54*> computed by the use of various suitable L D L^T factorizations near clusters
55*> of close eigenvalues (referred to as RRRs, Relatively Robust
56*> Representations). An informal sketch of the algorithm follows.
57*>
58*> For each unreduced block (submatrix) of T,
59*>    (a) Compute T - sigma I  = L D L^T, so that L and D
60*>        define all the wanted eigenvalues to high relative accuracy.
61*>        This means that small relative changes in the entries of D and L
62*>        cause only small relative changes in the eigenvalues and
63*>        eigenvectors. The standard (unfactored) representation of the
64*>        tridiagonal matrix T does not have this property in general.
65*>    (b) Compute the eigenvalues to suitable accuracy.
66*>        If the eigenvectors are desired, the algorithm attains full
67*>        accuracy of the computed eigenvalues only right before
68*>        the corresponding vectors have to be computed, see steps c) and d).
69*>    (c) For each cluster of close eigenvalues, select a new
70*>        shift close to the cluster, find a new factorization, and refine
71*>        the shifted eigenvalues to suitable accuracy.
72*>    (d) For each eigenvalue with a large enough relative separation compute
73*>        the corresponding eigenvector by forming a rank revealing twisted
74*>        factorization. Go back to (c) for any clusters that remain.
75*>
76*> For more details, see:
77*> - Inderjit S. Dhillon and Beresford N. Parlett: "Multiple representations
78*>   to compute orthogonal eigenvectors of symmetric tridiagonal matrices,"
79*>   Linear Algebra and its Applications, 387(1), pp. 1-28, August 2004.
80*> - Inderjit Dhillon and Beresford Parlett: "Orthogonal Eigenvectors and
81*>   Relative Gaps," SIAM Journal on Matrix Analysis and Applications, Vol. 25,
82*>   2004.  Also LAPACK Working Note 154.
83*> - Inderjit Dhillon: "A new O(n^2) algorithm for the symmetric
84*>   tridiagonal eigenvalue/eigenvector problem",
85*>   Computer Science Division Technical Report No. UCB/CSD-97-971,
86*>   UC Berkeley, May 1997.
87*>
88*> Further Details
89*> 1.ZSTEMR works only on machines which follow IEEE-754
90*> floating-point standard in their handling of infinities and NaNs.
91*> This permits the use of efficient inner loops avoiding a check for
92*> zero divisors.
93*>
94*> 2. LAPACK routines can be used to reduce a complex Hermitean matrix to
95*> real symmetric tridiagonal form.
96*>
97*> (Any complex Hermitean tridiagonal matrix has real values on its diagonal
98*> and potentially complex numbers on its off-diagonals. By applying a
99*> similarity transform with an appropriate diagonal matrix
100*> diag(1,e^{i \phy_1}, ... , e^{i \phy_{n-1}}), the complex Hermitean
101*> matrix can be transformed into a real symmetric matrix and complex
102*> arithmetic can be entirely avoided.)
103*>
104*> While the eigenvectors of the real symmetric tridiagonal matrix are real,
105*> the eigenvectors of original complex Hermitean matrix have complex entries
106*> in general.
107*> Since LAPACK drivers overwrite the matrix data with the eigenvectors,
108*> ZSTEMR accepts complex workspace to facilitate interoperability
109*> with ZUNMTR or ZUPMTR.
110*> \endverbatim
111*
112*  Arguments:
113*  ==========
114*
115*> \param[in] JOBZ
116*> \verbatim
117*>          JOBZ is CHARACTER*1
118*>          = 'N':  Compute eigenvalues only;
119*>          = 'V':  Compute eigenvalues and eigenvectors.
120*> \endverbatim
121*>
122*> \param[in] RANGE
123*> \verbatim
124*>          RANGE is CHARACTER*1
125*>          = 'A': all eigenvalues will be found.
126*>          = 'V': all eigenvalues in the half-open interval (VL,VU]
127*>                 will be found.
128*>          = 'I': the IL-th through IU-th eigenvalues will be found.
129*> \endverbatim
130*>
131*> \param[in] N
132*> \verbatim
133*>          N is INTEGER
134*>          The order of the matrix.  N >= 0.
135*> \endverbatim
136*>
137*> \param[in,out] D
138*> \verbatim
139*>          D is DOUBLE PRECISION array, dimension (N)
140*>          On entry, the N diagonal elements of the tridiagonal matrix
141*>          T. On exit, D is overwritten.
142*> \endverbatim
143*>
144*> \param[in,out] E
145*> \verbatim
146*>          E is DOUBLE PRECISION array, dimension (N)
147*>          On entry, the (N-1) subdiagonal elements of the tridiagonal
148*>          matrix T in elements 1 to N-1 of E. E(N) need not be set on
149*>          input, but is used internally as workspace.
150*>          On exit, E is overwritten.
151*> \endverbatim
152*>
153*> \param[in] VL
154*> \verbatim
155*>          VL is DOUBLE PRECISION
156*> \endverbatim
157*>
158*> \param[in] VU
159*> \verbatim
160*>          VU is DOUBLE PRECISION
161*>
162*>          If RANGE='V', the lower and upper bounds of the interval to
163*>          be searched for eigenvalues. VL < VU.
164*>          Not referenced if RANGE = 'A' or 'I'.
165*> \endverbatim
166*>
167*> \param[in] IL
168*> \verbatim
169*>          IL is INTEGER
170*> \endverbatim
171*>
172*> \param[in] IU
173*> \verbatim
174*>          IU is INTEGER
175*>
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.
179*>          Not referenced if RANGE = 'A' or 'V'.
180*> \endverbatim
181*>
182*> \param[out] M
183*> \verbatim
184*>          M is INTEGER
185*>          The total number of eigenvalues found.  0 <= M <= N.
186*>          If RANGE = 'A', M = N, and if RANGE = 'I', M = IU-IL+1.
187*> \endverbatim
188*>
189*> \param[out] W
190*> \verbatim
191*>          W is DOUBLE PRECISION array, dimension (N)
192*>          The first M elements contain the selected eigenvalues in
193*>          ascending order.
194*> \endverbatim
195*>
196*> \param[out] Z
197*> \verbatim
198*>          Z is COMPLEX*16 array, dimension (LDZ, max(1,M) )
199*>          If JOBZ = 'V', and if INFO = 0, then the first M columns of Z
200*>          contain the orthonormal eigenvectors of the matrix T
201*>          corresponding to the selected eigenvalues, with the i-th
202*>          column of Z holding the eigenvector associated with W(i).
203*>          If JOBZ = 'N', then Z is not referenced.
204*>          Note: the user must ensure that at least max(1,M) columns are
205*>          supplied in the array Z; if RANGE = 'V', the exact value of M
206*>          is not known in advance and can be computed with a workspace
207*>          query by setting NZC = -1, see below.
208*> \endverbatim
209*>
210*> \param[in] LDZ
211*> \verbatim
212*>          LDZ is INTEGER
213*>          The leading dimension of the array Z.  LDZ >= 1, and if
214*>          JOBZ = 'V', then LDZ >= max(1,N).
215*> \endverbatim
216*>
217*> \param[in] NZC
218*> \verbatim
219*>          NZC is INTEGER
220*>          The number of eigenvectors to be held in the array Z.
221*>          If RANGE = 'A', then NZC >= max(1,N).
222*>          If RANGE = 'V', then NZC >= the number of eigenvalues in (VL,VU].
223*>          If RANGE = 'I', then NZC >= IU-IL+1.
224*>          If NZC = -1, then a workspace query is assumed; the
225*>          routine calculates the number of columns of the array Z that
226*>          are needed to hold the eigenvectors.
227*>          This value is returned as the first entry of the Z array, and
228*>          no error message related to NZC is issued by XERBLA.
229*> \endverbatim
230*>
231*> \param[out] ISUPPZ
232*> \verbatim
233*>          ISUPPZ is INTEGER ARRAY, dimension ( 2*max(1,M) )
234*>          The support of the eigenvectors in Z, i.e., the indices
235*>          indicating the nonzero elements in Z. The i-th computed eigenvector
236*>          is nonzero only in elements ISUPPZ( 2*i-1 ) through
237*>          ISUPPZ( 2*i ). This is relevant in the case when the matrix
238*>          is split. ISUPPZ is only accessed when JOBZ is 'V' and N > 0.
239*> \endverbatim
240*>
241*> \param[in,out] TRYRAC
242*> \verbatim
243*>          TRYRAC is LOGICAL
244*>          If TRYRAC.EQ..TRUE., indicates that the code should check whether
245*>          the tridiagonal matrix defines its eigenvalues to high relative
246*>          accuracy.  If so, the code uses relative-accuracy preserving
247*>          algorithms that might be (a bit) slower depending on the matrix.
248*>          If the matrix does not define its eigenvalues to high relative
249*>          accuracy, the code can uses possibly faster algorithms.
250*>          If TRYRAC.EQ..FALSE., the code is not required to guarantee
251*>          relatively accurate eigenvalues and can use the fastest possible
252*>          techniques.
253*>          On exit, a .TRUE. TRYRAC will be set to .FALSE. if the matrix
254*>          does not define its eigenvalues to high relative accuracy.
255*> \endverbatim
256*>
257*> \param[out] WORK
258*> \verbatim
259*>          WORK is DOUBLE PRECISION array, dimension (LWORK)
260*>          On exit, if INFO = 0, WORK(1) returns the optimal
261*>          (and minimal) LWORK.
262*> \endverbatim
263*>
264*> \param[in] LWORK
265*> \verbatim
266*>          LWORK is INTEGER
267*>          The dimension of the array WORK. LWORK >= max(1,18*N)
268*>          if JOBZ = 'V', and LWORK >= max(1,12*N) if JOBZ = 'N'.
269*>          If LWORK = -1, then a workspace query is assumed; the routine
270*>          only calculates the optimal size of the WORK array, returns
271*>          this value as the first entry of the WORK array, and no error
272*>          message related to LWORK is issued by XERBLA.
273*> \endverbatim
274*>
275*> \param[out] IWORK
276*> \verbatim
277*>          IWORK is INTEGER array, dimension (LIWORK)
278*>          On exit, if INFO = 0, IWORK(1) returns the optimal LIWORK.
279*> \endverbatim
280*>
281*> \param[in] LIWORK
282*> \verbatim
283*>          LIWORK is INTEGER
284*>          The dimension of the array IWORK.  LIWORK >= max(1,10*N)
285*>          if the eigenvectors are desired, and LIWORK >= max(1,8*N)
286*>          if only the eigenvalues are to be computed.
287*>          If LIWORK = -1, then a workspace query is assumed; the
288*>          routine only calculates the optimal size of the IWORK array,
289*>          returns this value as the first entry of the IWORK array, and
290*>          no error message related to LIWORK is issued by XERBLA.
291*> \endverbatim
292*>
293*> \param[out] INFO
294*> \verbatim
295*>          INFO is INTEGER
296*>          On exit, INFO
297*>          = 0:  successful exit
298*>          < 0:  if INFO = -i, the i-th argument had an illegal value
299*>          > 0:  if INFO = 1X, internal error in DLARRE,
300*>                if INFO = 2X, internal error in ZLARRV.
301*>                Here, the digit X = ABS( IINFO ) < 10, where IINFO is
302*>                the nonzero error code returned by DLARRE or
303*>                ZLARRV, respectively.
304*> \endverbatim
305*
306*  Authors:
307*  ========
308*
309*> \author Univ. of Tennessee
310*> \author Univ. of California Berkeley
311*> \author Univ. of Colorado Denver
312*> \author NAG Ltd.
313*
314*> \date November 2015
315*
316*> \ingroup complex16OTHERcomputational
317*
318*> \par Contributors:
319*  ==================
320*>
321*> Beresford Parlett, University of California, Berkeley, USA \n
322*> Jim Demmel, University of California, Berkeley, USA \n
323*> Inderjit Dhillon, University of Texas, Austin, USA \n
324*> Osni Marques, LBNL/NERSC, USA \n
325*> Christof Voemel, University of California, Berkeley, USA \n
326*
327*  =====================================================================
328      SUBROUTINE ZSTEMR( JOBZ, RANGE, N, D, E, VL, VU, IL, IU,
329     $                   M, W, Z, LDZ, NZC, ISUPPZ, TRYRAC, WORK, LWORK,
330     $                   IWORK, LIWORK, INFO )
331*
332*  -- LAPACK computational routine (version 3.6.0) --
333*  -- LAPACK is a software package provided by Univ. of Tennessee,    --
334*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
335*     November 2015
336*
337*     .. Scalar Arguments ..
338      CHARACTER          JOBZ, RANGE
339      LOGICAL            TRYRAC
340      INTEGER            IL, INFO, IU, LDZ, NZC, LIWORK, LWORK, M, N
341      DOUBLE PRECISION VL, VU
342*     ..
343*     .. Array Arguments ..
344      INTEGER            ISUPPZ( * ), IWORK( * )
345      DOUBLE PRECISION   D( * ), E( * ), W( * ), WORK( * )
346      COMPLEX*16         Z( LDZ, * )
347*     ..
348*
349*  =====================================================================
350*
351*     .. Parameters ..
352      DOUBLE PRECISION   ZERO, ONE, FOUR, MINRGP
353      PARAMETER          ( ZERO = 0.0D0, ONE = 1.0D0,
354     $                     FOUR = 4.0D0,
355     $                     MINRGP = 1.0D-3 )
356*     ..
357*     .. Local Scalars ..
358      LOGICAL            ALLEIG, INDEIG, LQUERY, VALEIG, WANTZ, ZQUERY
359      INTEGER            I, IBEGIN, IEND, IFIRST, IIL, IINDBL, IINDW,
360     $                   IINDWK, IINFO, IINSPL, IIU, ILAST, IN, INDD,
361     $                   INDE2, INDERR, INDGP, INDGRS, INDWRK, ITMP,
362     $                   ITMP2, J, JBLK, JJ, LIWMIN, LWMIN, NSPLIT,
363     $                   NZCMIN, OFFSET, WBEGIN, WEND
364      DOUBLE PRECISION   BIGNUM, CS, EPS, PIVMIN, R1, R2, RMAX, RMIN,
365     $                   RTOL1, RTOL2, SAFMIN, SCALE, SMLNUM, SN,
366     $                   THRESH, TMP, TNRM, WL, WU
367*     ..
368*     ..
369*     .. External Functions ..
370      LOGICAL            LSAME
371      DOUBLE PRECISION   DLAMCH, DLANST
372      EXTERNAL           LSAME, DLAMCH, DLANST
373*     ..
374*     .. External Subroutines ..
375      EXTERNAL           DCOPY, DLAE2, DLAEV2, DLARRC, DLARRE, DLARRJ,
376     $                   DLARRR, DLASRT, DSCAL, XERBLA, ZLARRV, ZSWAP
377*     ..
378*     .. Intrinsic Functions ..
379      INTRINSIC          MAX, MIN, SQRT
380
381
382*     ..
383*     .. Executable Statements ..
384*
385*     Test the input parameters.
386*
387      WANTZ = LSAME( JOBZ, 'V' )
388      ALLEIG = LSAME( RANGE, 'A' )
389      VALEIG = LSAME( RANGE, 'V' )
390      INDEIG = LSAME( RANGE, 'I' )
391*
392      LQUERY = ( ( LWORK.EQ.-1 ).OR.( LIWORK.EQ.-1 ) )
393      ZQUERY = ( NZC.EQ.-1 )
394
395*     DSTEMR needs WORK of size 6*N, IWORK of size 3*N.
396*     In addition, DLARRE needs WORK of size 6*N, IWORK of size 5*N.
397*     Furthermore, ZLARRV needs WORK of size 12*N, IWORK of size 7*N.
398      IF( WANTZ ) THEN
399         LWMIN = 18*N
400         LIWMIN = 10*N
401      ELSE
402*        need less workspace if only the eigenvalues are wanted
403         LWMIN = 12*N
404         LIWMIN = 8*N
405      ENDIF
406
407      WL = ZERO
408      WU = ZERO
409      IIL = 0
410      IIU = 0
411      NSPLIT = 0
412
413      IF( VALEIG ) THEN
414*        We do not reference VL, VU in the cases RANGE = 'I','A'
415*        The interval (WL, WU] contains all the wanted eigenvalues.
416*        It is either given by the user or computed in DLARRE.
417         WL = VL
418         WU = VU
419      ELSEIF( INDEIG ) THEN
420*        We do not reference IL, IU in the cases RANGE = 'V','A'
421         IIL = IL
422         IIU = IU
423      ENDIF
424*
425      INFO = 0
426      IF( .NOT.( WANTZ .OR. LSAME( JOBZ, 'N' ) ) ) THEN
427         INFO = -1
428      ELSE IF( .NOT.( ALLEIG .OR. VALEIG .OR. INDEIG ) ) THEN
429         INFO = -2
430      ELSE IF( N.LT.0 ) THEN
431         INFO = -3
432      ELSE IF( VALEIG .AND. N.GT.0 .AND. WU.LE.WL ) THEN
433         INFO = -7
434      ELSE IF( INDEIG .AND. ( IIL.LT.1 .OR. IIL.GT.N ) ) THEN
435         INFO = -8
436      ELSE IF( INDEIG .AND. ( IIU.LT.IIL .OR. IIU.GT.N ) ) THEN
437         INFO = -9
438      ELSE IF( LDZ.LT.1 .OR. ( WANTZ .AND. LDZ.LT.N ) ) THEN
439         INFO = -13
440      ELSE IF( LWORK.LT.LWMIN .AND. .NOT.LQUERY ) THEN
441         INFO = -17
442      ELSE IF( LIWORK.LT.LIWMIN .AND. .NOT.LQUERY ) THEN
443         INFO = -19
444      END IF
445*
446*     Get machine constants.
447*
448      SAFMIN = DLAMCH( 'Safe minimum' )
449      EPS = DLAMCH( 'Precision' )
450      SMLNUM = SAFMIN / EPS
451      BIGNUM = ONE / SMLNUM
452      RMIN = SQRT( SMLNUM )
453      RMAX = MIN( SQRT( BIGNUM ), ONE / SQRT( SQRT( SAFMIN ) ) )
454*
455      IF( INFO.EQ.0 ) THEN
456         WORK( 1 ) = LWMIN
457         IWORK( 1 ) = LIWMIN
458*
459         IF( WANTZ .AND. ALLEIG ) THEN
460            NZCMIN = N
461         ELSE IF( WANTZ .AND. VALEIG ) THEN
462            CALL DLARRC( 'T', N, VL, VU, D, E, SAFMIN,
463     $                            NZCMIN, ITMP, ITMP2, INFO )
464         ELSE IF( WANTZ .AND. INDEIG ) THEN
465            NZCMIN = IIU-IIL+1
466         ELSE
467*           WANTZ .EQ. FALSE.
468            NZCMIN = 0
469         ENDIF
470         IF( ZQUERY .AND. INFO.EQ.0 ) THEN
471            Z( 1,1 ) = NZCMIN
472         ELSE IF( NZC.LT.NZCMIN .AND. .NOT.ZQUERY ) THEN
473            INFO = -14
474         END IF
475      END IF
476
477      IF( INFO.NE.0 ) THEN
478*
479         CALL XERBLA( 'ZSTEMR', -INFO )
480*
481         RETURN
482      ELSE IF( LQUERY .OR. ZQUERY ) THEN
483         RETURN
484      END IF
485*
486*     Handle N = 0, 1, and 2 cases immediately
487*
488      M = 0
489      IF( N.EQ.0 )
490     $   RETURN
491*
492      IF( N.EQ.1 ) THEN
493         IF( ALLEIG .OR. INDEIG ) THEN
494            M = 1
495            W( 1 ) = D( 1 )
496         ELSE
497            IF( WL.LT.D( 1 ) .AND. WU.GE.D( 1 ) ) THEN
498               M = 1
499               W( 1 ) = D( 1 )
500            END IF
501         END IF
502         IF( WANTZ.AND.(.NOT.ZQUERY) ) THEN
503            Z( 1, 1 ) = ONE
504            ISUPPZ(1) = 1
505            ISUPPZ(2) = 1
506         END IF
507         RETURN
508      END IF
509*
510      IF( N.EQ.2 ) THEN
511         IF( .NOT.WANTZ ) THEN
512            CALL DLAE2( D(1), E(1), D(2), R1, R2 )
513         ELSE IF( WANTZ.AND.(.NOT.ZQUERY) ) THEN
514            CALL DLAEV2( D(1), E(1), D(2), R1, R2, CS, SN )
515         END IF
516         IF( ALLEIG.OR.
517     $      (VALEIG.AND.(R2.GT.WL).AND.
518     $                  (R2.LE.WU)).OR.
519     $      (INDEIG.AND.(IIL.EQ.1)) ) THEN
520            M = M+1
521            W( M ) = R2
522            IF( WANTZ.AND.(.NOT.ZQUERY) ) THEN
523               Z( 1, M ) = -SN
524               Z( 2, M ) = CS
525*              Note: At most one of SN and CS can be zero.
526               IF (SN.NE.ZERO) THEN
527                  IF (CS.NE.ZERO) THEN
528                     ISUPPZ(2*M-1) = 1
529                     ISUPPZ(2*M) = 2
530                  ELSE
531                     ISUPPZ(2*M-1) = 1
532                     ISUPPZ(2*M) = 1
533                  END IF
534               ELSE
535                  ISUPPZ(2*M-1) = 2
536                  ISUPPZ(2*M) = 2
537               END IF
538            ENDIF
539         ENDIF
540         IF( ALLEIG.OR.
541     $      (VALEIG.AND.(R1.GT.WL).AND.
542     $                  (R1.LE.WU)).OR.
543     $      (INDEIG.AND.(IIU.EQ.2)) ) THEN
544            M = M+1
545            W( M ) = R1
546            IF( WANTZ.AND.(.NOT.ZQUERY) ) THEN
547               Z( 1, M ) = CS
548               Z( 2, M ) = SN
549*              Note: At most one of SN and CS can be zero.
550               IF (SN.NE.ZERO) THEN
551                  IF (CS.NE.ZERO) THEN
552                     ISUPPZ(2*M-1) = 1
553                     ISUPPZ(2*M) = 2
554                  ELSE
555                     ISUPPZ(2*M-1) = 1
556                     ISUPPZ(2*M) = 1
557                  END IF
558               ELSE
559                  ISUPPZ(2*M-1) = 2
560                  ISUPPZ(2*M) = 2
561               END IF
562            ENDIF
563         ENDIF
564      ELSE
565
566*        Continue with general N
567
568         INDGRS = 1
569         INDERR = 2*N + 1
570         INDGP = 3*N + 1
571         INDD = 4*N + 1
572         INDE2 = 5*N + 1
573         INDWRK = 6*N + 1
574*
575         IINSPL = 1
576         IINDBL = N + 1
577         IINDW = 2*N + 1
578         IINDWK = 3*N + 1
579*
580*        Scale matrix to allowable range, if necessary.
581*        The allowable range is related to the PIVMIN parameter; see the
582*        comments in DLARRD.  The preference for scaling small values
583*        up is heuristic; we expect users' matrices not to be close to the
584*        RMAX threshold.
585*
586         SCALE = ONE
587         TNRM = DLANST( 'M', N, D, E )
588         IF( TNRM.GT.ZERO .AND. TNRM.LT.RMIN ) THEN
589            SCALE = RMIN / TNRM
590         ELSE IF( TNRM.GT.RMAX ) THEN
591            SCALE = RMAX / TNRM
592         END IF
593         IF( SCALE.NE.ONE ) THEN
594            CALL DSCAL( N, SCALE, D, 1 )
595            CALL DSCAL( N-1, SCALE, E, 1 )
596            TNRM = TNRM*SCALE
597            IF( VALEIG ) THEN
598*              If eigenvalues in interval have to be found,
599*              scale (WL, WU] accordingly
600               WL = WL*SCALE
601               WU = WU*SCALE
602            ENDIF
603         END IF
604*
605*        Compute the desired eigenvalues of the tridiagonal after splitting
606*        into smaller subblocks if the corresponding off-diagonal elements
607*        are small
608*        THRESH is the splitting parameter for DLARRE
609*        A negative THRESH forces the old splitting criterion based on the
610*        size of the off-diagonal. A positive THRESH switches to splitting
611*        which preserves relative accuracy.
612*
613         IF( TRYRAC ) THEN
614*           Test whether the matrix warrants the more expensive relative approach.
615            CALL DLARRR( N, D, E, IINFO )
616         ELSE
617*           The user does not care about relative accurately eigenvalues
618            IINFO = -1
619         ENDIF
620*        Set the splitting criterion
621         IF (IINFO.EQ.0) THEN
622            THRESH = EPS
623         ELSE
624            THRESH = -EPS
625*           relative accuracy is desired but T does not guarantee it
626            TRYRAC = .FALSE.
627         ENDIF
628*
629         IF( TRYRAC ) THEN
630*           Copy original diagonal, needed to guarantee relative accuracy
631            CALL DCOPY(N,D,1,WORK(INDD),1)
632         ENDIF
633*        Store the squares of the offdiagonal values of T
634         DO 5 J = 1, N-1
635            WORK( INDE2+J-1 ) = E(J)**2
636 5    CONTINUE
637
638*        Set the tolerance parameters for bisection
639         IF( .NOT.WANTZ ) THEN
640*           DLARRE computes the eigenvalues to full precision.
641            RTOL1 = FOUR * EPS
642            RTOL2 = FOUR * EPS
643         ELSE
644*           DLARRE computes the eigenvalues to less than full precision.
645*           ZLARRV will refine the eigenvalue approximations, and we only
646*           need less accurate initial bisection in DLARRE.
647*           Note: these settings do only affect the subset case and DLARRE
648            RTOL1 = SQRT(EPS)
649            RTOL2 = MAX( SQRT(EPS)*5.0D-3, FOUR * EPS )
650         ENDIF
651         CALL DLARRE( RANGE, N, WL, WU, IIL, IIU, D, E,
652     $             WORK(INDE2), RTOL1, RTOL2, THRESH, NSPLIT,
653     $             IWORK( IINSPL ), M, W, WORK( INDERR ),
654     $             WORK( INDGP ), IWORK( IINDBL ),
655     $             IWORK( IINDW ), WORK( INDGRS ), PIVMIN,
656     $             WORK( INDWRK ), IWORK( IINDWK ), IINFO )
657         IF( IINFO.NE.0 ) THEN
658            INFO = 10 + ABS( IINFO )
659            RETURN
660         END IF
661*        Note that if RANGE .NE. 'V', DLARRE computes bounds on the desired
662*        part of the spectrum. All desired eigenvalues are contained in
663*        (WL,WU]
664
665
666         IF( WANTZ ) THEN
667*
668*           Compute the desired eigenvectors corresponding to the computed
669*           eigenvalues
670*
671            CALL ZLARRV( N, WL, WU, D, E,
672     $                PIVMIN, IWORK( IINSPL ), M,
673     $                1, M, MINRGP, RTOL1, RTOL2,
674     $                W, WORK( INDERR ), WORK( INDGP ), IWORK( IINDBL ),
675     $                IWORK( IINDW ), WORK( INDGRS ), Z, LDZ,
676     $                ISUPPZ, WORK( INDWRK ), IWORK( IINDWK ), IINFO )
677            IF( IINFO.NE.0 ) THEN
678               INFO = 20 + ABS( IINFO )
679               RETURN
680            END IF
681         ELSE
682*           DLARRE computes eigenvalues of the (shifted) root representation
683*           ZLARRV returns the eigenvalues of the unshifted matrix.
684*           However, if the eigenvectors are not desired by the user, we need
685*           to apply the corresponding shifts from DLARRE to obtain the
686*           eigenvalues of the original matrix.
687            DO 20 J = 1, M
688               ITMP = IWORK( IINDBL+J-1 )
689               W( J ) = W( J ) + E( IWORK( IINSPL+ITMP-1 ) )
690 20      CONTINUE
691         END IF
692*
693
694         IF ( TRYRAC ) THEN
695*           Refine computed eigenvalues so that they are relatively accurate
696*           with respect to the original matrix T.
697            IBEGIN = 1
698            WBEGIN = 1
699            DO 39  JBLK = 1, IWORK( IINDBL+M-1 )
700               IEND = IWORK( IINSPL+JBLK-1 )
701               IN = IEND - IBEGIN + 1
702               WEND = WBEGIN - 1
703*              check if any eigenvalues have to be refined in this block
704 36         CONTINUE
705               IF( WEND.LT.M ) THEN
706                  IF( IWORK( IINDBL+WEND ).EQ.JBLK ) THEN
707                     WEND = WEND + 1
708                     GO TO 36
709                  END IF
710               END IF
711               IF( WEND.LT.WBEGIN ) THEN
712                  IBEGIN = IEND + 1
713                  GO TO 39
714               END IF
715
716               OFFSET = IWORK(IINDW+WBEGIN-1)-1
717               IFIRST = IWORK(IINDW+WBEGIN-1)
718               ILAST = IWORK(IINDW+WEND-1)
719               RTOL2 = FOUR * EPS
720               CALL DLARRJ( IN,
721     $                   WORK(INDD+IBEGIN-1), WORK(INDE2+IBEGIN-1),
722     $                   IFIRST, ILAST, RTOL2, OFFSET, W(WBEGIN),
723     $                   WORK( INDERR+WBEGIN-1 ),
724     $                   WORK( INDWRK ), IWORK( IINDWK ), PIVMIN,
725     $                   TNRM, IINFO )
726               IBEGIN = IEND + 1
727               WBEGIN = WEND + 1
728 39      CONTINUE
729         ENDIF
730*
731*        If matrix was scaled, then rescale eigenvalues appropriately.
732*
733         IF( SCALE.NE.ONE ) THEN
734            CALL DSCAL( M, ONE / SCALE, W, 1 )
735         END IF
736      END IF
737*
738*     If eigenvalues are not in increasing order, then sort them,
739*     possibly along with eigenvectors.
740*
741      IF( NSPLIT.GT.1 .OR. N.EQ.2 ) THEN
742         IF( .NOT. WANTZ ) THEN
743            CALL DLASRT( 'I', M, W, IINFO )
744            IF( IINFO.NE.0 ) THEN
745               INFO = 3
746               RETURN
747            END IF
748         ELSE
749            DO 60 J = 1, M - 1
750               I = 0
751               TMP = W( J )
752               DO 50 JJ = J + 1, M
753                  IF( W( JJ ).LT.TMP ) THEN
754                     I = JJ
755                     TMP = W( JJ )
756                  END IF
757 50            CONTINUE
758               IF( I.NE.0 ) THEN
759                  W( I ) = W( J )
760                  W( J ) = TMP
761                  IF( WANTZ ) THEN
762                     CALL ZSWAP( N, Z( 1, I ), 1, Z( 1, J ), 1 )
763                     ITMP = ISUPPZ( 2*I-1 )
764                     ISUPPZ( 2*I-1 ) = ISUPPZ( 2*J-1 )
765                     ISUPPZ( 2*J-1 ) = ITMP
766                     ITMP = ISUPPZ( 2*I )
767                     ISUPPZ( 2*I ) = ISUPPZ( 2*J )
768                     ISUPPZ( 2*J ) = ITMP
769                  END IF
770               END IF
771 60         CONTINUE
772         END IF
773      ENDIF
774*
775*
776      WORK( 1 ) = LWMIN
777      IWORK( 1 ) = LIWMIN
778      RETURN
779*
780*     End of ZSTEMR
781*
782      END
783