1       SUBROUTINE ZLAHEF( UPLO, N, NB, KB, A, LDA, IPIV, W, LDW, INFO )
  2 *
  3 *  -- LAPACK routine (version 3.3.1) --
  4 *  -- LAPACK is a software package provided by Univ. of Tennessee,    --
  5 *  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
  6 *  -- April 2011                                                      --
  7 *
  8 *     .. Scalar Arguments ..
  9       CHARACTER          UPLO
 10       INTEGER            INFO, KB, LDA, LDW, N, NB
 11 *     ..
 12 *     .. Array Arguments ..
 13       INTEGER            IPIV( * )
 14       COMPLEX*16         A( LDA, * ), W( LDW, * )
 15 *     ..
 16 *
 17 *  Purpose
 18 *  =======
 19 *
 20 *  ZLAHEF computes a partial factorization of a complex Hermitian
 21 *  matrix A using the Bunch-Kaufman diagonal pivoting method. The
 22 *  partial factorization has the form:
 23 *
 24 *  A  =  ( I  U12 ) ( A11  0  ) (  I      0     )  if UPLO = 'U', or:
 25 *        ( 0  U22 ) (  0   D  ) ( U12**H U22**H )
 26 *
 27 *  A  =  ( L11  0 ) (  D   0  ) ( L11**H L21**H )  if UPLO = 'L'
 28 *        ( L21  I ) (  0  A22 ) (  0      I     )
 29 *
 30 *  where the order of D is at most NB. The actual order is returned in
 31 *  the argument KB, and is either NB or NB-1, or N if N <= NB.
 32 *  Note that U**H denotes the conjugate transpose of U.
 33 *
 34 *  ZLAHEF is an auxiliary routine called by ZHETRF. It uses blocked code
 35 *  (calling Level 3 BLAS) to update the submatrix A11 (if UPLO = 'U') or
 36 *  A22 (if UPLO = 'L').
 37 *
 38 *  Arguments
 39 *  =========
 40 *
 41 *  UPLO    (input) CHARACTER*1
 42 *          Specifies whether the upper or lower triangular part of the
 43 *          Hermitian matrix A is stored:
 44 *          = 'U':  Upper triangular
 45 *          = 'L':  Lower triangular
 46 *
 47 *  N       (input) INTEGER
 48 *          The order of the matrix A.  N >= 0.
 49 *
 50 *  NB      (input) INTEGER
 51 *          The maximum number of columns of the matrix A that should be
 52 *          factored.  NB should be at least 2 to allow for 2-by-2 pivot
 53 *          blocks.
 54 *
 55 *  KB      (output) INTEGER
 56 *          The number of columns of A that were actually factored.
 57 *          KB is either NB-1 or NB, or N if N <= NB.
 58 *
 59 *  A       (input/output) COMPLEX*16 array, dimension (LDA,N)
 60 *          On entry, the Hermitian matrix A.  If UPLO = 'U', the leading
 61 *          n-by-n upper triangular part of A contains the upper
 62 *          triangular part of the matrix A, and the strictly lower
 63 *          triangular part of A is not referenced.  If UPLO = 'L', the
 64 *          leading n-by-n lower triangular part of A contains the lower
 65 *          triangular part of the matrix A, and the strictly upper
 66 *          triangular part of A is not referenced.
 67 *          On exit, A contains details of the partial factorization.
 68 *
 69 *  LDA     (input) INTEGER
 70 *          The leading dimension of the array A.  LDA >= max(1,N).
 71 *
 72 *  IPIV    (output) INTEGER array, dimension (N)
 73 *          Details of the interchanges and the block structure of D.
 74 *          If UPLO = 'U', only the last KB elements of IPIV are set;
 75 *          if UPLO = 'L', only the first KB elements are set.
 76 *
 77 *          If IPIV(k) > 0, then rows and columns k and IPIV(k) were
 78 *          interchanged and D(k,k) is a 1-by-1 diagonal block.
 79 *          If UPLO = 'U' and IPIV(k) = IPIV(k-1) < 0, then rows and
 80 *          columns k-1 and -IPIV(k) were interchanged and D(k-1:k,k-1:k)
 81 *          is a 2-by-2 diagonal block.  If UPLO = 'L' and IPIV(k) =
 82 *          IPIV(k+1) < 0, then rows and columns k+1 and -IPIV(k) were
 83 *          interchanged and D(k:k+1,k:k+1) is a 2-by-2 diagonal block.
 84 *
 85 *  W       (workspace) COMPLEX*16 array, dimension (LDW,NB)
 86 *
 87 *  LDW     (input) INTEGER
 88 *          The leading dimension of the array W.  LDW >= max(1,N).
 89 *
 90 *  INFO    (output) INTEGER
 91 *          = 0: successful exit
 92 *          > 0: if INFO = k, D(k,k) is exactly zero.  The factorization
 93 *               has been completed, but the block diagonal matrix D is
 94 *               exactly singular.
 95 *
 96 *  =====================================================================
 97 *
 98 *     .. Parameters ..
 99       DOUBLE PRECISION   ZERO, ONE
100       PARAMETER          ( ZERO = 0.0D+0, ONE = 1.0D+0 )
101       COMPLEX*16         CONE
102       PARAMETER          ( CONE = ( 1.0D+00.0D+0 ) )
103       DOUBLE PRECISION   EIGHT, SEVTEN
104       PARAMETER          ( EIGHT = 8.0D+0, SEVTEN = 17.0D+0 )
105 *     ..
106 *     .. Local Scalars ..
107       INTEGER            IMAX, J, JB, JJ, JMAX, JP, K, KK, KKW, KP,
108      $                   KSTEP, KW
109       DOUBLE PRECISION   ABSAKK, ALPHA, COLMAX, R1, ROWMAX, T
110       COMPLEX*16         D11, D21, D22, Z
111 *     ..
112 *     .. External Functions ..
113       LOGICAL            LSAME
114       INTEGER            IZAMAX
115       EXTERNAL           LSAME, IZAMAX
116 *     ..
117 *     .. External Subroutines ..
118       EXTERNAL           ZCOPY, ZDSCAL, ZGEMM, ZGEMV, ZLACGV, ZSWAP
119 *     ..
120 *     .. Intrinsic Functions ..
121       INTRINSIC          ABSDBLEDCONJGDIMAGMAXMINSQRT
122 *     ..
123 *     .. Statement Functions ..
124       DOUBLE PRECISION   CABS1
125 *     ..
126 *     .. Statement Function definitions ..
127       CABS1( Z ) = ABSDBLE( Z ) ) + ABSDIMAG( Z ) )
128 *     ..
129 *     .. Executable Statements ..
130 *
131       INFO = 0
132 *
133 *     Initialize ALPHA for use in choosing pivot block size.
134 *
135       ALPHA = ( ONE+SQRT( SEVTEN ) ) / EIGHT
136 *
137       IF( LSAME( UPLO, 'U' ) ) THEN
138 *
139 *        Factorize the trailing columns of A using the upper triangle
140 *        of A and working backwards, and compute the matrix W = U12*D
141 *        for use in updating A11 (note that conjg(W) is actually stored)
142 *
143 *        K is the main loop index, decreasing from N in steps of 1 or 2
144 *
145 *        KW is the column of W which corresponds to column K of A
146 *
147          K = N
148    10    CONTINUE
149          KW = NB + K - N
150 *
151 *        Exit from loop
152 *
153          IF( ( K.LE.N-NB+1 .AND. NB.LT.N ) .OR. K.LT.1 )
154      $      GO TO 30
155 *
156 *        Copy column K of A to column KW of W and update it
157 *
158          CALL ZCOPY( K-1, A( 1, K ), 1, W( 1, KW ), 1 )
159          W( K, KW ) = DBLE( A( K, K ) )
160          IF( K.LT.N ) THEN
161             CALL ZGEMV( 'No transpose', K, N-K, -CONE, A( 1, K+1 ), LDA,
162      $                  W( K, KW+1 ), LDW, CONE, W( 1, KW ), 1 )
163             W( K, KW ) = DBLE( W( K, KW ) )
164          END IF
165 *
166          KSTEP = 1
167 *
168 *        Determine rows and columns to be interchanged and whether
169 *        a 1-by-1 or 2-by-2 pivot block will be used
170 *
171          ABSAKK = ABSDBLE( W( K, KW ) ) )
172 *
173 *        IMAX is the row-index of the largest off-diagonal element in
174 *        column K, and COLMAX is its absolute value
175 *
176          IF( K.GT.1 ) THEN
177             IMAX = IZAMAX( K-1, W( 1, KW ), 1 )
178             COLMAX = CABS1( W( IMAX, KW ) )
179          ELSE
180             COLMAX = ZERO
181          END IF
182 *
183          IFMAX( ABSAKK, COLMAX ).EQ.ZERO ) THEN
184 *
185 *           Column K is zero: set INFO and continue
186 *
187             IF( INFO.EQ.0 )
188      $         INFO = K
189             KP = K
190             A( K, K ) = DBLE( A( K, K ) )
191          ELSE
192             IF( ABSAKK.GE.ALPHA*COLMAX ) THEN
193 *
194 *              no interchange, use 1-by-1 pivot block
195 *
196                KP = K
197             ELSE
198 *
199 *              Copy column IMAX to column KW-1 of W and update it
200 *
201                CALL ZCOPY( IMAX-1, A( 1, IMAX ), 1, W( 1, KW-1 ), 1 )
202                W( IMAX, KW-1 ) = DBLE( A( IMAX, IMAX ) )
203                CALL ZCOPY( K-IMAX, A( IMAX, IMAX+1 ), LDA,
204      $                     W( IMAX+1, KW-1 ), 1 )
205                CALL ZLACGV( K-IMAX, W( IMAX+1, KW-1 ), 1 )
206                IF( K.LT.N ) THEN
207                   CALL ZGEMV( 'No transpose', K, N-K, -CONE,
208      $                        A( 1, K+1 ), LDA, W( IMAX, KW+1 ), LDW,
209      $                        CONE, W( 1, KW-1 ), 1 )
210                   W( IMAX, KW-1 ) = DBLE( W( IMAX, KW-1 ) )
211                END IF
212 *
213 *              JMAX is the column-index of the largest off-diagonal
214 *              element in row IMAX, and ROWMAX is its absolute value
215 *
216                JMAX = IMAX + IZAMAX( K-IMAX, W( IMAX+1, KW-1 ), 1 )
217                ROWMAX = CABS1( W( JMAX, KW-1 ) )
218                IF( IMAX.GT.1 ) THEN
219                   JMAX = IZAMAX( IMAX-1, W( 1, KW-1 ), 1 )
220                   ROWMAX = MAX( ROWMAX, CABS1( W( JMAX, KW-1 ) ) )
221                END IF
222 *
223                IF( ABSAKK.GE.ALPHA*COLMAX*( COLMAX / ROWMAX ) ) THEN
224 *
225 *                 no interchange, use 1-by-1 pivot block
226 *
227                   KP = K
228                ELSE IFABSDBLE( W( IMAX, KW-1 ) ) ).GE.ALPHA*ROWMAX )
229      $                   THEN
230 *
231 *                 interchange rows and columns K and IMAX, use 1-by-1
232 *                 pivot block
233 *
234                   KP = IMAX
235 *
236 *                 copy column KW-1 of W to column KW
237 *
238                   CALL ZCOPY( K, W( 1, KW-1 ), 1, W( 1, KW ), 1 )
239                ELSE
240 *
241 *                 interchange rows and columns K-1 and IMAX, use 2-by-2
242 *                 pivot block
243 *
244                   KP = IMAX
245                   KSTEP = 2
246                END IF
247             END IF
248 *
249             KK = K - KSTEP + 1
250             KKW = NB + KK - N
251 *
252 *           Updated column KP is already stored in column KKW of W
253 *
254             IF( KP.NE.KK ) THEN
255 *
256 *              Copy non-updated column KK to column KP
257 *
258                A( KP, KP ) = DBLE( A( KK, KK ) )
259                CALL ZCOPY( KK-1-KP, A( KP+1, KK ), 1, A( KP, KP+1 ),
260      $                     LDA )
261                CALL ZLACGV( KK-1-KP, A( KP, KP+1 ), LDA )
262                CALL ZCOPY( KP-1, A( 1, KK ), 1, A( 1, KP ), 1 )
263 *
264 *              Interchange rows KK and KP in last KK columns of A and W
265 *
266                IF( KK.LT.N )
267      $            CALL ZSWAP( N-KK, A( KK, KK+1 ), LDA, A( KP, KK+1 ),
268      $                        LDA )
269                CALL ZSWAP( N-KK+1, W( KK, KKW ), LDW, W( KP, KKW ),
270      $                     LDW )
271             END IF
272 *
273             IF( KSTEP.EQ.1 ) THEN
274 *
275 *              1-by-1 pivot block D(k): column KW of W now holds
276 *
277 *              W(k) = U(k)*D(k)
278 *
279 *              where U(k) is the k-th column of U
280 *
281 *              Store U(k) in column k of A
282 *
283                CALL ZCOPY( K, W( 1, KW ), 1, A( 1, K ), 1 )
284                R1 = ONE / DBLE( A( K, K ) )
285                CALL ZDSCAL( K-1, R1, A( 1, K ), 1 )
286 *
287 *              Conjugate W(k)
288 *
289                CALL ZLACGV( K-1, W( 1, KW ), 1 )
290             ELSE
291 *
292 *              2-by-2 pivot block D(k): columns KW and KW-1 of W now
293 *              hold
294 *
295 *              ( W(k-1) W(k) ) = ( U(k-1) U(k) )*D(k)
296 *
297 *              where U(k) and U(k-1) are the k-th and (k-1)-th columns
298 *              of U
299 *
300                IF( K.GT.2 ) THEN
301 *
302 *                 Store U(k) and U(k-1) in columns k and k-1 of A
303 *
304                   D21 = W( K-1, KW )
305                   D11 = W( K, KW ) / DCONJG( D21 )
306                   D22 = W( K-1, KW-1 ) / D21
307                   T = ONE / ( DBLE( D11*D22 )-ONE )
308                   D21 = T / D21
309                   DO 20 J = 1, K - 2
310                      A( J, K-1 ) = D21*( D11*W( J, KW-1 )-W( J, KW ) )
311                      A( J, K ) = DCONJG( D21 )*
312      $                           ( D22*W( J, KW )-W( J, KW-1 ) )
313    20             CONTINUE
314                END IF
315 *
316 *              Copy D(k) to A
317 *
318                A( K-1, K-1 ) = W( K-1, KW-1 )
319                A( K-1, K ) = W( K-1, KW )
320                A( K, K ) = W( K, KW )
321 *
322 *              Conjugate W(k) and W(k-1)
323 *
324                CALL ZLACGV( K-1, W( 1, KW ), 1 )
325                CALL ZLACGV( K-2, W( 1, KW-1 ), 1 )
326             END IF
327          END IF
328 *
329 *        Store details of the interchanges in IPIV
330 *
331          IF( KSTEP.EQ.1 ) THEN
332             IPIV( K ) = KP
333          ELSE
334             IPIV( K ) = -KP
335             IPIV( K-1 ) = -KP
336          END IF
337 *
338 *        Decrease K and return to the start of the main loop
339 *
340          K = K - KSTEP
341          GO TO 10
342 *
343    30    CONTINUE
344 *
345 *        Update the upper triangle of A11 (= A(1:k,1:k)) as
346 *
347 *        A11 := A11 - U12*D*U12**H = A11 - U12*W**H
348 *
349 *        computing blocks of NB columns at a time (note that conjg(W) is
350 *        actually stored)
351 *
352          DO 50 J = ( ( K-1 ) / NB )*NB + 11-NB
353             JB = MIN( NB, K-J+1 )
354 *
355 *           Update the upper triangle of the diagonal block
356 *
357             DO 40 JJ = J, J + JB - 1
358                A( JJ, JJ ) = DBLE( A( JJ, JJ ) )
359                CALL ZGEMV( 'No transpose', JJ-J+1, N-K, -CONE,
360      $                     A( J, K+1 ), LDA, W( JJ, KW+1 ), LDW, CONE,
361      $                     A( J, JJ ), 1 )
362                A( JJ, JJ ) = DBLE( A( JJ, JJ ) )
363    40       CONTINUE
364 *
365 *           Update the rectangular superdiagonal block
366 *
367             CALL ZGEMM( 'No transpose''Transpose', J-1, JB, N-K,
368      $                  -CONE, A( 1, K+1 ), LDA, W( J, KW+1 ), LDW,
369      $                  CONE, A( 1, J ), LDA )
370    50    CONTINUE
371 *
372 *        Put U12 in standard form by partially undoing the interchanges
373 *        in columns k+1:n
374 *
375          J = K + 1
376    60    CONTINUE
377          JJ = J
378          JP = IPIV( J )
379          IF( JP.LT.0 ) THEN
380             JP = -JP
381             J = J + 1
382          END IF
383          J = J + 1
384          IF( JP.NE.JJ .AND. J.LE.N )
385      $      CALL ZSWAP( N-J+1, A( JP, J ), LDA, A( JJ, J ), LDA )
386          IF( J.LE.N )
387      $      GO TO 60
388 *
389 *        Set KB to the number of columns factorized
390 *
391          KB = N - K
392 *
393       ELSE
394 *
395 *        Factorize the leading columns of A using the lower triangle
396 *        of A and working forwards, and compute the matrix W = L21*D
397 *        for use in updating A22 (note that conjg(W) is actually stored)
398 *
399 *        K is the main loop index, increasing from 1 in steps of 1 or 2
400 *
401          K = 1
402    70    CONTINUE
403 *
404 *        Exit from loop
405 *
406          IF( ( K.GE.NB .AND. NB.LT.N ) .OR. K.GT.N )
407      $      GO TO 90
408 *
409 *        Copy column K of A to column K of W and update it
410 *
411          W( K, K ) = DBLE( A( K, K ) )
412          IF( K.LT.N )
413      $      CALL ZCOPY( N-K, A( K+1, K ), 1, W( K+1, K ), 1 )
414          CALL ZGEMV( 'No transpose', N-K+1, K-1-CONE, A( K, 1 ), LDA,
415      $               W( K, 1 ), LDW, CONE, W( K, K ), 1 )
416          W( K, K ) = DBLE( W( K, K ) )
417 *
418          KSTEP = 1
419 *
420 *        Determine rows and columns to be interchanged and whether
421 *        a 1-by-1 or 2-by-2 pivot block will be used
422 *
423          ABSAKK = ABSDBLE( W( K, K ) ) )
424 *
425 *        IMAX is the row-index of the largest off-diagonal element in
426 *        column K, and COLMAX is its absolute value
427 *
428          IF( K.LT.N ) THEN
429             IMAX = K + IZAMAX( N-K, W( K+1, K ), 1 )
430             COLMAX = CABS1( W( IMAX, K ) )
431          ELSE
432             COLMAX = ZERO
433          END IF
434 *
435          IFMAX( ABSAKK, COLMAX ).EQ.ZERO ) THEN
436 *
437 *           Column K is zero: set INFO and continue
438 *
439             IF( INFO.EQ.0 )
440      $         INFO = K
441             KP = K
442             A( K, K ) = DBLE( A( K, K ) )
443          ELSE
444             IF( ABSAKK.GE.ALPHA*COLMAX ) THEN
445 *
446 *              no interchange, use 1-by-1 pivot block
447 *
448                KP = K
449             ELSE
450 *
451 *              Copy column IMAX to column K+1 of W and update it
452 *
453                CALL ZCOPY( IMAX-K, A( IMAX, K ), LDA, W( K, K+1 ), 1 )
454                CALL ZLACGV( IMAX-K, W( K, K+1 ), 1 )
455                W( IMAX, K+1 ) = DBLE( A( IMAX, IMAX ) )
456                IF( IMAX.LT.N )
457      $            CALL ZCOPY( N-IMAX, A( IMAX+1, IMAX ), 1,
458      $                        W( IMAX+1, K+1 ), 1 )
459                CALL ZGEMV( 'No transpose', N-K+1, K-1-CONE, A( K, 1 ),
460      $                     LDA, W( IMAX, 1 ), LDW, CONE, W( K, K+1 ),
461      $                     1 )
462                W( IMAX, K+1 ) = DBLE( W( IMAX, K+1 ) )
463 *
464 *              JMAX is the column-index of the largest off-diagonal
465 *              element in row IMAX, and ROWMAX is its absolute value
466 *
467                JMAX = K - 1 + IZAMAX( IMAX-K, W( K, K+1 ), 1 )
468                ROWMAX = CABS1( W( JMAX, K+1 ) )
469                IF( IMAX.LT.N ) THEN
470                   JMAX = IMAX + IZAMAX( N-IMAX, W( IMAX+1, K+1 ), 1 )
471                   ROWMAX = MAX( ROWMAX, CABS1( W( JMAX, K+1 ) ) )
472                END IF
473 *
474                IF( ABSAKK.GE.ALPHA*COLMAX*( COLMAX / ROWMAX ) ) THEN
475 *
476 *                 no interchange, use 1-by-1 pivot block
477 *
478                   KP = K
479                ELSE IFABSDBLE( W( IMAX, K+1 ) ) ).GE.ALPHA*ROWMAX )
480      $                   THEN
481 *
482 *                 interchange rows and columns K and IMAX, use 1-by-1
483 *                 pivot block
484 *
485                   KP = IMAX
486 *
487 *                 copy column K+1 of W to column K
488 *
489                   CALL ZCOPY( N-K+1, W( K, K+1 ), 1, W( K, K ), 1 )
490                ELSE
491 *
492 *                 interchange rows and columns K+1 and IMAX, use 2-by-2
493 *                 pivot block
494 *
495                   KP = IMAX
496                   KSTEP = 2
497                END IF
498             END IF
499 *
500             KK = K + KSTEP - 1
501 *
502 *           Updated column KP is already stored in column KK of W
503 *
504             IF( KP.NE.KK ) THEN
505 *
506 *              Copy non-updated column KK to column KP
507 *
508                A( KP, KP ) = DBLE( A( KK, KK ) )
509                CALL ZCOPY( KP-KK-1, A( KK+1, KK ), 1, A( KP, KK+1 ),
510      $                     LDA )
511                CALL ZLACGV( KP-KK-1, A( KP, KK+1 ), LDA )
512                IF( KP.LT.N )
513      $            CALL ZCOPY( N-KP, A( KP+1, KK ), 1, A( KP+1, KP ), 1 )
514 *
515 *              Interchange rows KK and KP in first KK columns of A and W
516 *
517                CALL ZSWAP( KK-1, A( KK, 1 ), LDA, A( KP, 1 ), LDA )
518                CALL ZSWAP( KK, W( KK, 1 ), LDW, W( KP, 1 ), LDW )
519             END IF
520 *
521             IF( KSTEP.EQ.1 ) THEN
522 *
523 *              1-by-1 pivot block D(k): column k of W now holds
524 *
525 *              W(k) = L(k)*D(k)
526 *
527 *              where L(k) is the k-th column of L
528 *
529 *              Store L(k) in column k of A
530 *
531                CALL ZCOPY( N-K+1, W( K, K ), 1, A( K, K ), 1 )
532                IF( K.LT.N ) THEN
533                   R1 = ONE / DBLE( A( K, K ) )
534                   CALL ZDSCAL( N-K, R1, A( K+1, K ), 1 )
535 *
536 *                 Conjugate W(k)
537 *
538                   CALL ZLACGV( N-K, W( K+1, K ), 1 )
539                END IF
540             ELSE
541 *
542 *              2-by-2 pivot block D(k): columns k and k+1 of W now hold
543 *
544 *              ( W(k) W(k+1) ) = ( L(k) L(k+1) )*D(k)
545 *
546 *              where L(k) and L(k+1) are the k-th and (k+1)-th columns
547 *              of L
548 *
549                IF( K.LT.N-1 ) THEN
550 *
551 *                 Store L(k) and L(k+1) in columns k and k+1 of A
552 *
553                   D21 = W( K+1, K )
554                   D11 = W( K+1, K+1 ) / D21
555                   D22 = W( K, K ) / DCONJG( D21 )
556                   T = ONE / ( DBLE( D11*D22 )-ONE )
557                   D21 = T / D21
558                   DO 80 J = K + 2, N
559                      A( J, K ) = DCONJG( D21 )*
560      $                           ( D11*W( J, K )-W( J, K+1 ) )
561                      A( J, K+1 ) = D21*( D22*W( J, K+1 )-W( J, K ) )
562    80             CONTINUE
563                END IF
564 *
565 *              Copy D(k) to A
566 *
567                A( K, K ) = W( K, K )
568                A( K+1, K ) = W( K+1, K )
569                A( K+1, K+1 ) = W( K+1, K+1 )
570 *
571 *              Conjugate W(k) and W(k+1)
572 *
573                CALL ZLACGV( N-K, W( K+1, K ), 1 )
574                CALL ZLACGV( N-K-1, W( K+2, K+1 ), 1 )
575             END IF
576          END IF
577 *
578 *        Store details of the interchanges in IPIV
579 *
580          IF( KSTEP.EQ.1 ) THEN
581             IPIV( K ) = KP
582          ELSE
583             IPIV( K ) = -KP
584             IPIV( K+1 ) = -KP
585          END IF
586 *
587 *        Increase K and return to the start of the main loop
588 *
589          K = K + KSTEP
590          GO TO 70
591 *
592    90    CONTINUE
593 *
594 *        Update the lower triangle of A22 (= A(k:n,k:n)) as
595 *
596 *        A22 := A22 - L21*D*L21**H = A22 - L21*W**H
597 *
598 *        computing blocks of NB columns at a time (note that conjg(W) is
599 *        actually stored)
600 *
601          DO 110 J = K, N, NB
602             JB = MIN( NB, N-J+1 )
603 *
604 *           Update the lower triangle of the diagonal block
605 *
606             DO 100 JJ = J, J + JB - 1
607                A( JJ, JJ ) = DBLE( A( JJ, JJ ) )
608                CALL ZGEMV( 'No transpose', J+JB-JJ, K-1-CONE,
609      $                     A( JJ, 1 ), LDA, W( JJ, 1 ), LDW, CONE,
610      $                     A( JJ, JJ ), 1 )
611                A( JJ, JJ ) = DBLE( A( JJ, JJ ) )
612   100       CONTINUE
613 *
614 *           Update the rectangular subdiagonal block
615 *
616             IF( J+JB.LE.N )
617      $         CALL ZGEMM( 'No transpose''Transpose', N-J-JB+1, JB,
618      $                     K-1-CONE, A( J+JB, 1 ), LDA, W( J, 1 ),
619      $                     LDW, CONE, A( J+JB, J ), LDA )
620   110    CONTINUE
621 *
622 *        Put L21 in standard form by partially undoing the interchanges
623 *        in columns 1:k-1
624 *
625          J = K - 1
626   120    CONTINUE
627          JJ = J
628          JP = IPIV( J )
629          IF( JP.LT.0 ) THEN
630             JP = -JP
631             J = J - 1
632          END IF
633          J = J - 1
634          IF( JP.NE.JJ .AND. J.GE.1 )
635      $      CALL ZSWAP( J, A( JP, 1 ), LDA, A( JJ, 1 ), LDA )
636          IF( J.GE.1 )
637      $      GO TO 120
638 *
639 *        Set KB to the number of columns factorized
640 *
641          KB = K - 1
642 *
643       END IF
644       RETURN
645 *
646 *     End of ZLAHEF
647 *
648       END