1*> \brief \b DLAQR0 computes the eigenvalues of a Hessenberg matrix, and optionally the matrices from the Schur decomposition.
2*
3*  =========== DOCUMENTATION ===========
4*
5* Online html documentation available at
6*            http://www.netlib.org/lapack/explore-html/
7*
8*> \htmlonly
9*> Download DLAQR0 + dependencies
10*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlaqr0.f">
11*> [TGZ]</a>
12*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlaqr0.f">
13*> [ZIP]</a>
14*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlaqr0.f">
15*> [TXT]</a>
16*> \endhtmlonly
17*
18*  Definition:
19*  ===========
20*
21*       SUBROUTINE DLAQR0( WANTT, WANTZ, N, ILO, IHI, H, LDH, WR, WI,
22*                          ILOZ, IHIZ, Z, LDZ, WORK, LWORK, INFO )
23*
24*       .. Scalar Arguments ..
25*       INTEGER            IHI, IHIZ, ILO, ILOZ, INFO, LDH, LDZ, LWORK, N
26*       LOGICAL            WANTT, WANTZ
27*       ..
28*       .. Array Arguments ..
29*       DOUBLE PRECISION   H( LDH, * ), WI( * ), WORK( * ), WR( * ),
30*      $                   Z( LDZ, * )
31*       ..
32*
33*
34*> \par Purpose:
35*  =============
36*>
37*> \verbatim
38*>
39*>    DLAQR0 computes the eigenvalues of a Hessenberg matrix H
40*>    and, optionally, the matrices T and Z from the Schur decomposition
41*>    H = Z T Z**T, where T is an upper quasi-triangular matrix (the
42*>    Schur form), and Z is the orthogonal matrix of Schur vectors.
43*>
44*>    Optionally Z may be postmultiplied into an input orthogonal
45*>    matrix Q so that this routine can give the Schur factorization
46*>    of a matrix A which has been reduced to the Hessenberg form H
47*>    by the orthogonal matrix Q:  A = Q*H*Q**T = (QZ)*T*(QZ)**T.
48*> \endverbatim
49*
50*  Arguments:
51*  ==========
52*
53*> \param[in] WANTT
54*> \verbatim
55*>          WANTT is LOGICAL
56*>          = .TRUE. : the full Schur form T is required;
57*>          = .FALSE.: only eigenvalues are required.
58*> \endverbatim
59*>
60*> \param[in] WANTZ
61*> \verbatim
62*>          WANTZ is LOGICAL
63*>          = .TRUE. : the matrix of Schur vectors Z is required;
64*>          = .FALSE.: Schur vectors are not required.
65*> \endverbatim
66*>
67*> \param[in] N
68*> \verbatim
69*>          N is INTEGER
70*>           The order of the matrix H.  N >= 0.
71*> \endverbatim
72*>
73*> \param[in] ILO
74*> \verbatim
75*>          ILO is INTEGER
76*> \endverbatim
77*>
78*> \param[in] IHI
79*> \verbatim
80*>          IHI is INTEGER
81*>           It is assumed that H is already upper triangular in rows
82*>           and columns 1:ILO-1 and IHI+1:N and, if ILO > 1,
83*>           H(ILO,ILO-1) is zero. ILO and IHI are normally set by a
84*>           previous call to DGEBAL, and then passed to DGEHRD when the
85*>           matrix output by DGEBAL is reduced to Hessenberg form.
86*>           Otherwise, ILO and IHI should be set to 1 and N,
87*>           respectively.  If N > 0, then 1 <= ILO <= IHI <= N.
88*>           If N = 0, then ILO = 1 and IHI = 0.
89*> \endverbatim
90*>
91*> \param[in,out] H
92*> \verbatim
93*>          H is DOUBLE PRECISION array, dimension (LDH,N)
94*>           On entry, the upper Hessenberg matrix H.
95*>           On exit, if INFO = 0 and WANTT is .TRUE., then H contains
96*>           the upper quasi-triangular matrix T from the Schur
97*>           decomposition (the Schur form); 2-by-2 diagonal blocks
98*>           (corresponding to complex conjugate pairs of eigenvalues)
99*>           are returned in standard form, with H(i,i) = H(i+1,i+1)
100*>           and H(i+1,i)*H(i,i+1) < 0. If INFO = 0 and WANTT is
101*>           .FALSE., then the contents of H are unspecified on exit.
102*>           (The output value of H when INFO > 0 is given under the
103*>           description of INFO below.)
104*>
105*>           This subroutine may explicitly set H(i,j) = 0 for i > j and
106*>           j = 1, 2, ... ILO-1 or j = IHI+1, IHI+2, ... N.
107*> \endverbatim
108*>
109*> \param[in] LDH
110*> \verbatim
111*>          LDH is INTEGER
112*>           The leading dimension of the array H. LDH >= max(1,N).
113*> \endverbatim
114*>
115*> \param[out] WR
116*> \verbatim
117*>          WR is DOUBLE PRECISION array, dimension (IHI)
118*> \endverbatim
119*>
120*> \param[out] WI
121*> \verbatim
122*>          WI is DOUBLE PRECISION array, dimension (IHI)
123*>           The real and imaginary parts, respectively, of the computed
124*>           eigenvalues of H(ILO:IHI,ILO:IHI) are stored in WR(ILO:IHI)
125*>           and WI(ILO:IHI). If two eigenvalues are computed as a
126*>           complex conjugate pair, they are stored in consecutive
127*>           elements of WR and WI, say the i-th and (i+1)th, with
128*>           WI(i) > 0 and WI(i+1) < 0. If WANTT is .TRUE., then
129*>           the eigenvalues are stored in the same order as on the
130*>           diagonal of the Schur form returned in H, with
131*>           WR(i) = H(i,i) and, if H(i:i+1,i:i+1) is a 2-by-2 diagonal
132*>           block, WI(i) = sqrt(-H(i+1,i)*H(i,i+1)) and
133*>           WI(i+1) = -WI(i).
134*> \endverbatim
135*>
136*> \param[in] ILOZ
137*> \verbatim
138*>          ILOZ is INTEGER
139*> \endverbatim
140*>
141*> \param[in] IHIZ
142*> \verbatim
143*>          IHIZ is INTEGER
144*>           Specify the rows of Z to which transformations must be
145*>           applied if WANTZ is .TRUE..
146*>           1 <= ILOZ <= ILO; IHI <= IHIZ <= N.
147*> \endverbatim
148*>
149*> \param[in,out] Z
150*> \verbatim
151*>          Z is DOUBLE PRECISION array, dimension (LDZ,IHI)
152*>           If WANTZ is .FALSE., then Z is not referenced.
153*>           If WANTZ is .TRUE., then Z(ILO:IHI,ILOZ:IHIZ) is
154*>           replaced by Z(ILO:IHI,ILOZ:IHIZ)*U where U is the
155*>           orthogonal Schur factor of H(ILO:IHI,ILO:IHI).
156*>           (The output value of Z when INFO > 0 is given under
157*>           the description of INFO below.)
158*> \endverbatim
159*>
160*> \param[in] LDZ
161*> \verbatim
162*>          LDZ is INTEGER
163*>           The leading dimension of the array Z.  if WANTZ is .TRUE.
164*>           then LDZ >= MAX(1,IHIZ).  Otherwise, LDZ >= 1.
165*> \endverbatim
166*>
167*> \param[out] WORK
168*> \verbatim
169*>          WORK is DOUBLE PRECISION array, dimension LWORK
170*>           On exit, if LWORK = -1, WORK(1) returns an estimate of
171*>           the optimal value for LWORK.
172*> \endverbatim
173*>
174*> \param[in] LWORK
175*> \verbatim
176*>          LWORK is INTEGER
177*>           The dimension of the array WORK.  LWORK >= max(1,N)
178*>           is sufficient, but LWORK typically as large as 6*N may
179*>           be required for optimal performance.  A workspace query
180*>           to determine the optimal workspace size is recommended.
181*>
182*>           If LWORK = -1, then DLAQR0 does a workspace query.
183*>           In this case, DLAQR0 checks the input parameters and
184*>           estimates the optimal workspace size for the given
185*>           values of N, ILO and IHI.  The estimate is returned
186*>           in WORK(1).  No error message related to LWORK is
187*>           issued by XERBLA.  Neither H nor Z are accessed.
188*> \endverbatim
189*>
190*> \param[out] INFO
191*> \verbatim
192*>          INFO is INTEGER
193*>             = 0:  successful exit
194*>             > 0:  if INFO = i, DLAQR0 failed to compute all of
195*>                the eigenvalues.  Elements 1:ilo-1 and i+1:n of WR
196*>                and WI contain those eigenvalues which have been
197*>                successfully computed.  (Failures are rare.)
198*>
199*>                If INFO > 0 and WANT is .FALSE., then on exit,
200*>                the remaining unconverged eigenvalues are the eigen-
201*>                values of the upper Hessenberg matrix rows and
202*>                columns ILO through INFO of the final, output
203*>                value of H.
204*>
205*>                If INFO > 0 and WANTT is .TRUE., then on exit
206*>
207*>           (*)  (initial value of H)*U  = U*(final value of H)
208*>
209*>                where U is an orthogonal matrix.  The final
210*>                value of H is upper Hessenberg and quasi-triangular
211*>                in rows and columns INFO+1 through IHI.
212*>
213*>                If INFO > 0 and WANTZ is .TRUE., then on exit
214*>
215*>                  (final value of Z(ILO:IHI,ILOZ:IHIZ)
216*>                   =  (initial value of Z(ILO:IHI,ILOZ:IHIZ)*U
217*>
218*>                where U is the orthogonal matrix in (*) (regard-
219*>                less of the value of WANTT.)
220*>
221*>                If INFO > 0 and WANTZ is .FALSE., then Z is not
222*>                accessed.
223*> \endverbatim
224*
225*> \par Contributors:
226*  ==================
227*>
228*>       Karen Braman and Ralph Byers, Department of Mathematics,
229*>       University of Kansas, USA
230*
231*> \par References:
232*  ================
233*>
234*>       K. Braman, R. Byers and R. Mathias, The Multi-Shift QR
235*>       Algorithm Part I: Maintaining Well Focused Shifts, and Level 3
236*>       Performance, SIAM Journal of Matrix Analysis, volume 23, pages
237*>       929--947, 2002.
238*> \n
239*>       K. Braman, R. Byers and R. Mathias, The Multi-Shift QR
240*>       Algorithm Part II: Aggressive Early Deflation, SIAM Journal
241*>       of Matrix Analysis, volume 23, pages 948--973, 2002.
242*
243*  Authors:
244*  ========
245*
246*> \author Univ. of Tennessee
247*> \author Univ. of California Berkeley
248*> \author Univ. of Colorado Denver
249*> \author NAG Ltd.
250*
251*> \ingroup doubleOTHERauxiliary
252*
253*  =====================================================================
254      SUBROUTINE DLAQR0( WANTT, WANTZ, N, ILO, IHI, H, LDH, WR, WI,
255     $                   ILOZ, IHIZ, Z, LDZ, WORK, LWORK, INFO )
256*
257*  -- LAPACK auxiliary routine --
258*  -- LAPACK is a software package provided by Univ. of Tennessee,    --
259*  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
260*
261*     .. Scalar Arguments ..
262      INTEGER            IHI, IHIZ, ILO, ILOZ, INFO, LDH, LDZ, LWORK, N
263      LOGICAL            WANTT, WANTZ
264*     ..
265*     .. Array Arguments ..
266      DOUBLE PRECISION   H( LDH, * ), WI( * ), WORK( * ), WR( * ),
267     $                   Z( LDZ, * )
268*     ..
269*
270*  ================================================================
271*
272*     .. Parameters ..
273*
274*     ==== Matrices of order NTINY or smaller must be processed by
275*     .    DLAHQR because of insufficient subdiagonal scratch space.
276*     .    (This is a hard limit.) ====
277      INTEGER            NTINY
278      PARAMETER          ( NTINY = 15 )
279*
280*     ==== Exceptional deflation windows:  try to cure rare
281*     .    slow convergence by varying the size of the
282*     .    deflation window after KEXNW iterations. ====
283      INTEGER            KEXNW
284      PARAMETER          ( KEXNW = 5 )
285*
286*     ==== Exceptional shifts: try to cure rare slow convergence
287*     .    with ad-hoc exceptional shifts every KEXSH iterations.
288*     .    ====
289      INTEGER            KEXSH
290      PARAMETER          ( KEXSH = 6 )
291*
292*     ==== The constants WILK1 and WILK2 are used to form the
293*     .    exceptional shifts. ====
294      DOUBLE PRECISION   WILK1, WILK2
295      PARAMETER          ( WILK1 = 0.75d0, WILK2 = -0.4375d0 )
296      DOUBLE PRECISION   ZERO, ONE
297      PARAMETER          ( ZERO = 0.0d0, ONE = 1.0d0 )
298*     ..
299*     .. Local Scalars ..
300      DOUBLE PRECISION   AA, BB, CC, CS, DD, SN, SS, SWAP
301      INTEGER            I, INF, IT, ITMAX, K, KACC22, KBOT, KDU, KS,
302     $                   KT, KTOP, KU, KV, KWH, KWTOP, KWV, LD, LS,
303     $                   LWKOPT, NDEC, NDFL, NH, NHO, NIBBLE, NMIN, NS,
304     $                   NSMAX, NSR, NVE, NW, NWMAX, NWR, NWUPBD
305      LOGICAL            SORTED
306      CHARACTER          JBCMPZ*2
307*     ..
308*     .. External Functions ..
309      INTEGER            ILAENV
310      EXTERNAL           ILAENV
311*     ..
312*     .. Local Arrays ..
313      DOUBLE PRECISION   ZDUM( 1, 1 )
314*     ..
315*     .. External Subroutines ..
316      EXTERNAL           DLACPY, DLAHQR, DLANV2, DLAQR3, DLAQR4, DLAQR5
317*     ..
318*     .. Intrinsic Functions ..
319      INTRINSIC          ABS, DBLE, INT, MAX, MIN, MOD
320*     ..
321*     .. Executable Statements ..
322      INFO = 0
323*
324*     ==== Quick return for N = 0: nothing to do. ====
325*
326      IF( N.EQ.0 ) THEN
327         WORK( 1 ) = ONE
328         RETURN
329      END IF
330*
331      IF( N.LE.NTINY ) THEN
332*
333*        ==== Tiny matrices must use DLAHQR. ====
334*
335         LWKOPT = 1
336         IF( LWORK.NE.-1 )
337     $      CALL DLAHQR( WANTT, WANTZ, N, ILO, IHI, H, LDH, WR, WI,
338     $                   ILOZ, IHIZ, Z, LDZ, INFO )
339      ELSE
340*
341*        ==== Use small bulge multi-shift QR with aggressive early
342*        .    deflation on larger-than-tiny matrices. ====
343*
344*        ==== Hope for the best. ====
345*
346         INFO = 0
347*
348*        ==== Set up job flags for ILAENV. ====
349*
350         IF( WANTT ) THEN
351            JBCMPZ( 1: 1 ) = 'S'
352         ELSE
353            JBCMPZ( 1: 1 ) = 'E'
354         END IF
355         IF( WANTZ ) THEN
356            JBCMPZ( 2: 2 ) = 'V'
357         ELSE
358            JBCMPZ( 2: 2 ) = 'N'
359         END IF
360*
361*        ==== NWR = recommended deflation window size.  At this
362*        .    point,  N .GT. NTINY = 15, so there is enough
363*        .    subdiagonal workspace for NWR.GE.2 as required.
364*        .    (In fact, there is enough subdiagonal space for
365*        .    NWR.GE.4.) ====
366*
367         NWR = ILAENV( 13, 'DLAQR0', JBCMPZ, N, ILO, IHI, LWORK )
368         NWR = MAX( 2, NWR )
369         NWR = MIN( IHI-ILO+1, ( N-1 ) / 3, NWR )
370*
371*        ==== NSR = recommended number of simultaneous shifts.
372*        .    At this point N .GT. NTINY = 15, so there is at
373*        .    enough subdiagonal workspace for NSR to be even
374*        .    and greater than or equal to two as required. ====
375*
376         NSR = ILAENV( 15, 'DLAQR0', JBCMPZ, N, ILO, IHI, LWORK )
377         NSR = MIN( NSR, ( N-3 ) / 6, IHI-ILO )
378         NSR = MAX( 2, NSR-MOD( NSR, 2 ) )
379*
380*        ==== Estimate optimal workspace ====
381*
382*        ==== Workspace query call to DLAQR3 ====
383*
384         CALL DLAQR3( WANTT, WANTZ, N, ILO, IHI, NWR+1, H, LDH, ILOZ,
385     $                IHIZ, Z, LDZ, LS, LD, WR, WI, H, LDH, N, H, LDH,
386     $                N, H, LDH, WORK, -1 )
387*
388*        ==== Optimal workspace = MAX(DLAQR5, DLAQR3) ====
389*
390         LWKOPT = MAX( 3*NSR / 2, INT( WORK( 1 ) ) )
391*
392*        ==== Quick return in case of workspace query. ====
393*
394         IF( LWORK.EQ.-1 ) THEN
395            WORK( 1 ) = DBLE( LWKOPT )
396            RETURN
397         END IF
398*
399*        ==== DLAHQR/DLAQR0 crossover point ====
400*
401         NMIN = ILAENV( 12, 'DLAQR0', JBCMPZ, N, ILO, IHI, LWORK )
402         NMIN = MAX( NTINY, NMIN )
403*
404*        ==== Nibble crossover point ====
405*
406         NIBBLE = ILAENV( 14, 'DLAQR0', JBCMPZ, N, ILO, IHI, LWORK )
407         NIBBLE = MAX( 0, NIBBLE )
408*
409*        ==== Accumulate reflections during ttswp?  Use block
410*        .    2-by-2 structure during matrix-matrix multiply? ====
411*
412         KACC22 = ILAENV( 16, 'DLAQR0', JBCMPZ, N, ILO, IHI, LWORK )
413         KACC22 = MAX( 0, KACC22 )
414         KACC22 = MIN( 2, KACC22 )
415*
416*        ==== NWMAX = the largest possible deflation window for
417*        .    which there is sufficient workspace. ====
418*
419         NWMAX = MIN( ( N-1 ) / 3, LWORK / 2 )
420         NW = NWMAX
421*
422*        ==== NSMAX = the Largest number of simultaneous shifts
423*        .    for which there is sufficient workspace. ====
424*
425         NSMAX = MIN( ( N-3 ) / 6, 2*LWORK / 3 )
426         NSMAX = NSMAX - MOD( NSMAX, 2 )
427*
428*        ==== NDFL: an iteration count restarted at deflation. ====
429*
430         NDFL = 1
431*
432*        ==== ITMAX = iteration limit ====
433*
434         ITMAX = MAX( 30, 2*KEXSH )*MAX( 10, ( IHI-ILO+1 ) )
435*
436*        ==== Last row and column in the active block ====
437*
438         KBOT = IHI
439*
440*        ==== Main Loop ====
441*
442         DO 80 IT = 1, ITMAX
443*
444*           ==== Done when KBOT falls below ILO ====
445*
446            IF( KBOT.LT.ILO )
447     $         GO TO 90
448*
449*           ==== Locate active block ====
450*
451            DO 10 K = KBOT, ILO + 1, -1
452               IF( H( K, K-1 ).EQ.ZERO )
453     $            GO TO 20
454   10       CONTINUE
455            K = ILO
456   20       CONTINUE
457            KTOP = K
458*
459*           ==== Select deflation window size:
460*           .    Typical Case:
461*           .      If possible and advisable, nibble the entire
462*           .      active block.  If not, use size MIN(NWR,NWMAX)
463*           .      or MIN(NWR+1,NWMAX) depending upon which has
464*           .      the smaller corresponding subdiagonal entry
465*           .      (a heuristic).
466*           .
467*           .    Exceptional Case:
468*           .      If there have been no deflations in KEXNW or
469*           .      more iterations, then vary the deflation window
470*           .      size.   At first, because, larger windows are,
471*           .      in general, more powerful than smaller ones,
472*           .      rapidly increase the window to the maximum possible.
473*           .      Then, gradually reduce the window size. ====
474*
475            NH = KBOT - KTOP + 1
476            NWUPBD = MIN( NH, NWMAX )
477            IF( NDFL.LT.KEXNW ) THEN
478               NW = MIN( NWUPBD, NWR )
479            ELSE
480               NW = MIN( NWUPBD, 2*NW )
481            END IF
482            IF( NW.LT.NWMAX ) THEN
483               IF( NW.GE.NH-1 ) THEN
484                  NW = NH
485               ELSE
486                  KWTOP = KBOT - NW + 1
487                  IF( ABS( H( KWTOP, KWTOP-1 ) ).GT.
488     $                ABS( H( KWTOP-1, KWTOP-2 ) ) )NW = NW + 1
489               END IF
490            END IF
491            IF( NDFL.LT.KEXNW ) THEN
492               NDEC = -1
493            ELSE IF( NDEC.GE.0 .OR. NW.GE.NWUPBD ) THEN
494               NDEC = NDEC + 1
495               IF( NW-NDEC.LT.2 )
496     $            NDEC = 0
497               NW = NW - NDEC
498            END IF
499*
500*           ==== Aggressive early deflation:
501*           .    split workspace under the subdiagonal into
502*           .      - an nw-by-nw work array V in the lower
503*           .        left-hand-corner,
504*           .      - an NW-by-at-least-NW-but-more-is-better
505*           .        (NW-by-NHO) horizontal work array along
506*           .        the bottom edge,
507*           .      - an at-least-NW-but-more-is-better (NHV-by-NW)
508*           .        vertical work array along the left-hand-edge.
509*           .        ====
510*
511            KV = N - NW + 1
512            KT = NW + 1
513            NHO = ( N-NW-1 ) - KT + 1
514            KWV = NW + 2
515            NVE = ( N-NW ) - KWV + 1
516*
517*           ==== Aggressive early deflation ====
518*
519            CALL DLAQR3( WANTT, WANTZ, N, KTOP, KBOT, NW, H, LDH, ILOZ,
520     $                   IHIZ, Z, LDZ, LS, LD, WR, WI, H( KV, 1 ), LDH,
521     $                   NHO, H( KV, KT ), LDH, NVE, H( KWV, 1 ), LDH,
522     $                   WORK, LWORK )
523*
524*           ==== Adjust KBOT accounting for new deflations. ====
525*
526            KBOT = KBOT - LD
527*
528*           ==== KS points to the shifts. ====
529*
530            KS = KBOT - LS + 1
531*
532*           ==== Skip an expensive QR sweep if there is a (partly
533*           .    heuristic) reason to expect that many eigenvalues
534*           .    will deflate without it.  Here, the QR sweep is
535*           .    skipped if many eigenvalues have just been deflated
536*           .    or if the remaining active block is small.
537*
538            IF( ( LD.EQ.0 ) .OR. ( ( 100*LD.LE.NW*NIBBLE ) .AND. ( KBOT-
539     $          KTOP+1.GT.MIN( NMIN, NWMAX ) ) ) ) THEN
540*
541*              ==== NS = nominal number of simultaneous shifts.
542*              .    This may be lowered (slightly) if DLAQR3
543*              .    did not provide that many shifts. ====
544*
545               NS = MIN( NSMAX, NSR, MAX( 2, KBOT-KTOP ) )
546               NS = NS - MOD( NS, 2 )
547*
548*              ==== If there have been no deflations
549*              .    in a multiple of KEXSH iterations,
550*              .    then try exceptional shifts.
551*              .    Otherwise use shifts provided by
552*              .    DLAQR3 above or from the eigenvalues
553*              .    of a trailing principal submatrix. ====
554*
555               IF( MOD( NDFL, KEXSH ).EQ.0 ) THEN
556                  KS = KBOT - NS + 1
557                  DO 30 I = KBOT, MAX( KS+1, KTOP+2 ), -2
558                     SS = ABS( H( I, I-1 ) ) + ABS( H( I-1, I-2 ) )
559                     AA = WILK1*SS + H( I, I )
560                     BB = SS
561                     CC = WILK2*SS
562                     DD = AA
563                     CALL DLANV2( AA, BB, CC, DD, WR( I-1 ), WI( I-1 ),
564     $                            WR( I ), WI( I ), CS, SN )
565   30             CONTINUE
566                  IF( KS.EQ.KTOP ) THEN
567                     WR( KS+1 ) = H( KS+1, KS+1 )
568                     WI( KS+1 ) = ZERO
569                     WR( KS ) = WR( KS+1 )
570                     WI( KS ) = WI( KS+1 )
571                  END IF
572               ELSE
573*
574*                 ==== Got NS/2 or fewer shifts? Use DLAQR4 or
575*                 .    DLAHQR on a trailing principal submatrix to
576*                 .    get more. (Since NS.LE.NSMAX.LE.(N-3)/6,
577*                 .    there is enough space below the subdiagonal
578*                 .    to fit an NS-by-NS scratch array.) ====
579*
580                  IF( KBOT-KS+1.LE.NS / 2 ) THEN
581                     KS = KBOT - NS + 1
582                     KT = N - NS + 1
583                     CALL DLACPY( 'A', NS, NS, H( KS, KS ), LDH,
584     $                            H( KT, 1 ), LDH )
585                     IF( NS.GT.NMIN ) THEN
586                        CALL DLAQR4( .false., .false., NS, 1, NS,
587     $                               H( KT, 1 ), LDH, WR( KS ),
588     $                               WI( KS ), 1, 1, ZDUM, 1, WORK,
589     $                               LWORK, INF )
590                     ELSE
591                        CALL DLAHQR( .false., .false., NS, 1, NS,
592     $                               H( KT, 1 ), LDH, WR( KS ),
593     $                               WI( KS ), 1, 1, ZDUM, 1, INF )
594                     END IF
595                     KS = KS + INF
596*
597*                    ==== In case of a rare QR failure use
598*                    .    eigenvalues of the trailing 2-by-2
599*                    .    principal submatrix.  ====
600*
601                     IF( KS.GE.KBOT ) THEN
602                        AA = H( KBOT-1, KBOT-1 )
603                        CC = H( KBOT, KBOT-1 )
604                        BB = H( KBOT-1, KBOT )
605                        DD = H( KBOT, KBOT )
606                        CALL DLANV2( AA, BB, CC, DD, WR( KBOT-1 ),
607     $                               WI( KBOT-1 ), WR( KBOT ),
608     $                               WI( KBOT ), CS, SN )
609                        KS = KBOT - 1
610                     END IF
611                  END IF
612*
613                  IF( KBOT-KS+1.GT.NS ) THEN
614*
615*                    ==== Sort the shifts (Helps a little)
616*                    .    Bubble sort keeps complex conjugate
617*                    .    pairs together. ====
618*
619                     SORTED = .false.
620                     DO 50 K = KBOT, KS + 1, -1
621                        IF( SORTED )
622     $                     GO TO 60
623                        SORTED = .true.
624                        DO 40 I = KS, K - 1
625                           IF( ABS( WR( I ) )+ABS( WI( I ) ).LT.
626     $                         ABS( WR( I+1 ) )+ABS( WI( I+1 ) ) ) THEN
627                              SORTED = .false.
628*
629                              SWAP = WR( I )
630                              WR( I ) = WR( I+1 )
631                              WR( I+1 ) = SWAP
632*
633                              SWAP = WI( I )
634                              WI( I ) = WI( I+1 )
635                              WI( I+1 ) = SWAP
636                           END IF
637   40                   CONTINUE
638   50                CONTINUE
639   60                CONTINUE
640                  END IF
641*
642*                 ==== Shuffle shifts into pairs of real shifts
643*                 .    and pairs of complex conjugate shifts
644*                 .    assuming complex conjugate shifts are
645*                 .    already adjacent to one another. (Yes,
646*                 .    they are.)  ====
647*
648                  DO 70 I = KBOT, KS + 2, -2
649                     IF( WI( I ).NE.-WI( I-1 ) ) THEN
650*
651                        SWAP = WR( I )
652                        WR( I ) = WR( I-1 )
653                        WR( I-1 ) = WR( I-2 )
654                        WR( I-2 ) = SWAP
655*
656                        SWAP = WI( I )
657                        WI( I ) = WI( I-1 )
658                        WI( I-1 ) = WI( I-2 )
659                        WI( I-2 ) = SWAP
660                     END IF
661   70             CONTINUE
662               END IF
663*
664*              ==== If there are only two shifts and both are
665*              .    real, then use only one.  ====
666*
667               IF( KBOT-KS+1.EQ.2 ) THEN
668                  IF( WI( KBOT ).EQ.ZERO ) THEN
669                     IF( ABS( WR( KBOT )-H( KBOT, KBOT ) ).LT.
670     $                   ABS( WR( KBOT-1 )-H( KBOT, KBOT ) ) ) THEN
671                        WR( KBOT-1 ) = WR( KBOT )
672                     ELSE
673                        WR( KBOT ) = WR( KBOT-1 )
674                     END IF
675                  END IF
676               END IF
677*
678*              ==== Use up to NS of the the smallest magnitude
679*              .    shifts.  If there aren't NS shifts available,
680*              .    then use them all, possibly dropping one to
681*              .    make the number of shifts even. ====
682*
683               NS = MIN( NS, KBOT-KS+1 )
684               NS = NS - MOD( NS, 2 )
685               KS = KBOT - NS + 1
686*
687*              ==== Small-bulge multi-shift QR sweep:
688*              .    split workspace under the subdiagonal into
689*              .    - a KDU-by-KDU work array U in the lower
690*              .      left-hand-corner,
691*              .    - a KDU-by-at-least-KDU-but-more-is-better
692*              .      (KDU-by-NHo) horizontal work array WH along
693*              .      the bottom edge,
694*              .    - and an at-least-KDU-but-more-is-better-by-KDU
695*              .      (NVE-by-KDU) vertical work WV arrow along
696*              .      the left-hand-edge. ====
697*
698               KDU = 2*NS
699               KU = N - KDU + 1
700               KWH = KDU + 1
701               NHO = ( N-KDU+1-4 ) - ( KDU+1 ) + 1
702               KWV = KDU + 4
703               NVE = N - KDU - KWV + 1
704*
705*              ==== Small-bulge multi-shift QR sweep ====
706*
707               CALL DLAQR5( WANTT, WANTZ, KACC22, N, KTOP, KBOT, NS,
708     $                      WR( KS ), WI( KS ), H, LDH, ILOZ, IHIZ, Z,
709     $                      LDZ, WORK, 3, H( KU, 1 ), LDH, NVE,
710     $                      H( KWV, 1 ), LDH, NHO, H( KU, KWH ), LDH )
711            END IF
712*
713*           ==== Note progress (or the lack of it). ====
714*
715            IF( LD.GT.0 ) THEN
716               NDFL = 1
717            ELSE
718               NDFL = NDFL + 1
719            END IF
720*
721*           ==== End of main loop ====
722   80    CONTINUE
723*
724*        ==== Iteration limit exceeded.  Set INFO to show where
725*        .    the problem occurred and exit. ====
726*
727         INFO = KBOT
728   90    CONTINUE
729      END IF
730*
731*     ==== Return the optimal value of LWORK. ====
732*
733      WORK( 1 ) = DBLE( LWKOPT )
734*
735*     ==== End of DLAQR0 ====
736*
737      END
738