1
       2
       3
       4
       5
       6
       7
       8
       9
      10
      11
      12
      13
      14
      15
      16
      17
      18
      19
      20
      21
      22
      23
      24
      25
      26
      27
      28
      29
      30
      31
      32
      33
      34
      35
      36
      37
      38
      39
      40
      41
      42
      43
      44
      45
      46
      47
      48
      49
      50
      51
      52
      53
      54
      55
      56
      57
      58
      59
      60
      61
      62
      63
      64
      65
      66
      67
      68
      69
      70
      71
      72
      73
      74
      75
      76
      77
      78
      79
      80
      81
      82
      83
      84
      85
      86
      87
      88
      89
      90
      91
      92
      93
      94
      95
      96
      97
      98
      99
     100
     101
     102
     103
     104
     105
     106
     107
     108
     109
     110
     111
     112
     113
     114
     115
     116
     117
     118
     119
     120
     121
     122
     123
     124
     125
     126
     127
     128
     129
     130
     131
     132
     133
     134
     135
     136
     137
     138
     139
     140
     141
     142
     143
     144
     145
     146
     147
     148
     149
     150
     151
     152
     153
     154
     155
     156
     157
     158
     159
     160
     161
     162
     163
     164
     165
     166
     167
     168
     169
     170
     171
     172
     173
     174
     175
     176
     177
     178
     179
     180
     181
     182
     183
     184
     185
     186
     187
     188
     189
     190
     191
     192
     193
     194
     195
     196
     197
     198
     199
     200
     201
     202
     203
     204
     205
     206
     207
     208
     209
     210
     211
     212
     213
     214
     215
     216
     217
     218
     219
     220
     221
     222
     223
     224
     225
     226
     227
     228
     229
     230
     231
     232
     233
     234
     235
     236
     237
     238
     239
     240
     241
     242
     243
     244
     245
     246
     247
     248
     249
     250
     251
     252
     253
     254
     255
     256
     257
     258
     259
     260
     261
     262
     263
     264
     265
     266
     267
     268
     269
     270
     271
     272
     273
     274
     275
     276
     277
     278
     279
     280
     281
     282
     283
     284
     285
     286
     287
     288
     289
     290
     291
     292
     293
     294
     295
     296
     297
     298
     299
     300
     301
     302
     303
     304
     305
     306
     307
     308
     309
     310
     311
     312
     313
     314
     315
     316
     317
     318
     319
     320
     321
     322
     323
     324
     325
     326
     327
     328
     329
     330
     331
     332
     333
     334
     335
     336
     337
     338
     339
     340
     341
     342
     343
     344
     345
     346
     347
     348
     349
     350
     351
     352
     353
     354
     355
     356
     357
     358
     359
     360
     361
     362
     363
     364
     365
     366
     367
     368
     369
     370
     371
     372
     373
     374
     375
     376
     377
     378
     379
     380
     381
     382
     383
     384
     385
     386
     387
     388
     389
     390
     391
     392
     393
     394
     395
     396
     397
     398
     399
     400
     401
     402
     403
     404
     405
     406
     407
     408
     409
     410
     411
     412
     413
     414
     415
     416
     417
     418
     419
     420
     421
     422
     423
     424
     425
     426
     427
     428
     429
     430
     431
     432
     433
     434
     435
     436
     437
     438
     439
     440
     441
     442
     443
     444
     445
     446
     447
     448
     449
     450
     451
     452
     453
     454
     455
     456
     457
     458
     459
     460
     461
     462
     463
     464
     465
     466
     467
     468
     469
     470
     471
     472
     473
     474
     475
     476
     477
     478
     479
     480
     481
     482
     483
     484
     485
     486
     487
     488
     489
     490
     491
     492
     493
     494
     495
     496
     497
     498
     499
     500
     501
     502
     503
     504
     505
     506
     507
     508
     509
     510
     511
     512
     513
     514
     515
     516
     517
     518
     519
     520
     521
     522
     523
     524
     525
     526
     527
     528
     529
     530
     531
     532
     533
     534
     535
     536
     537
     538
     539
     540
     541
     542
     543
     544
     545
     546
     547
      SUBROUTINE DLA_GERFSX_EXTENDED( PREC_TYPE, TRANS_TYPE, N, NRHS, A,
     $                                LDA, AF, LDAF, IPIV, COLEQU, C, B,
     $                                LDB, Y, LDY, BERR_OUT, N_NORMS,
     $                                ERRS_N, ERRS_C, RES, AYB, DY,
     $                                Y_TAIL, RCOND, ITHRESH, RTHRESH,
     $                                DZ_UB, IGNORE_CWISE, INFO )
*
*     -- LAPACK routine (version 3.2.1)                                 --
*     -- Contributed by James Demmel, Deaglan Halligan, Yozo Hida and --
*     -- Jason Riedy of Univ. of California Berkeley.                 --
*     -- April 2009                                                   --
*
*     -- LAPACK is a software package provided by Univ. of Tennessee, --
*     -- Univ. of California Berkeley and NAG Ltd.                    --
*
      IMPLICIT NONE
*     ..
*     .. Scalar Arguments ..
      INTEGER            INFO, LDA, LDAF, LDB, LDY, N, NRHS, PREC_TYPE,
     $                   TRANS_TYPE, N_NORMS, ITHRESH
      LOGICAL            COLEQU, IGNORE_CWISE
      DOUBLE PRECISION   RTHRESH, DZ_UB
*     ..
*     .. Array Arguments ..
      INTEGER            IPIV( * )
      DOUBLE PRECISION   A( LDA, * ), AF( LDAF, * ), B( LDB, * ),
     $                   Y( LDY, * ), RES( * ), DY( * ), Y_TAIL( * )
      DOUBLE PRECISION   C( * ), AYB( * ), RCOND, BERR_OUT( * ),
     $                   ERRS_N( NRHS, * ), ERRS_C( NRHS, * )
*     ..
*
*  Purpose
*  =======

*  DLA_GERFSX_EXTENDED improves the computed solution to a system of
*  linear equations by performing extra-precise iterative refinement
*  and provides error bounds and backward error estimates for the solution.
*  This subroutine is called by DGERFSX to perform iterative refinement.
*  In addition to normwise error bound, the code provides maximum
*  componentwise error bound if possible. See comments for ERR_BNDS_NORM
*  and ERR_BNDS_COMP for details of the error bounds. Note that this
*  subroutine is only resonsible for setting the second fields of
*  ERR_BNDS_NORM and ERR_BNDS_COMP.
*
*  Arguments
*  =========
*
*     PREC_TYPE      (input) INTEGER
*     Specifies the intermediate precision to be used in refinement.
*     The value is defined by ILAPREC(P) where P is a CHARACTER and
*     P    = 'S':  Single
*          = 'D':  Double
*          = 'I':  Indigenous
*          = 'X', 'E':  Extra
*
*     TRANS_TYPE     (input) INTEGER
*     Specifies the transposition operation on A.
*     The value is defined by ILATRANS(T) where T is a CHARACTER and
*     T    = 'N':  No transpose
*          = 'T':  Transpose
*          = 'C':  Conjugate transpose
*
*     N              (input) INTEGER
*     The number of linear equations, i.e., the order of the
*     matrix A.  N >= 0.
*
*     NRHS           (input) INTEGER
*     The number of right-hand-sides, i.e., the number of columns of the
*     matrix B.
*
*     A              (input) DOUBLE PRECISION array, dimension (LDA,N)
*     On entry, the N-by-N matrix A.
*
*     LDA            (input) INTEGER
*     The leading dimension of the array A.  LDA >= max(1,N).
*
*     AF             (input) DOUBLE PRECISION array, dimension (LDAF,N)
*     The factors L and U from the factorization
*     A = P*L*U as computed by DGETRF.
*
*     LDAF           (input) INTEGER
*     The leading dimension of the array AF.  LDAF >= max(1,N).
*
*     IPIV           (input) INTEGER array, dimension (N)
*     The pivot indices from the factorization A = P*L*U
*     as computed by DGETRF; row i of the matrix was interchanged
*     with row IPIV(i).
*
*     COLEQU         (input) LOGICAL
*     If .TRUE. then column equilibration was done to A before calling
*     this routine. This is needed to compute the solution and error
*     bounds correctly.
*
*     C              (input) DOUBLE PRECISION  array, dimension (N)
*     The column scale factors for A. If COLEQU = .FALSE., C
*     is not accessed. If C is input, each element of C should be a power
*     of the radix to ensure a reliable solution and error estimates.
*     Scaling by powers of the radix does not cause rounding errors unless
*     the result underflows or overflows. Rounding errors during scaling
*     lead to refining with a matrix that is not equivalent to the
*     input matrix, producing error estimates that may not be
*     reliable.
*
*     B              (input) DOUBLE PRECISION array, dimension (LDB,NRHS)
*     The right-hand-side matrix B.
*
*     LDB            (input) INTEGER
*     The leading dimension of the array B.  LDB >= max(1,N).
*
*     Y              (input/output) DOUBLE PRECISION array, dimension
*                    (LDY,NRHS)
*     On entry, the solution matrix X, as computed by DGETRS.
*     On exit, the improved solution matrix Y.
*
*     LDY            (input) INTEGER
*     The leading dimension of the array Y.  LDY >= max(1,N).
*
*     BERR_OUT       (output) DOUBLE PRECISION array, dimension (NRHS)
*     On exit, BERR_OUT(j) contains the componentwise relative backward
*     error for right-hand-side j from the formula
*         max(i) ( abs(RES(i)) / ( abs(op(A_s))*abs(Y) + abs(B_s) )(i) )
*     where abs(Z) is the componentwise absolute value of the matrix
*     or vector Z. This is computed by DLA_LIN_BERR.
*
*     N_NORMS        (input) INTEGER
*     Determines which error bounds to return (see ERR_BNDS_NORM
*     and ERR_BNDS_COMP).
*     If N_NORMS >= 1 return normwise error bounds.
*     If N_NORMS >= 2 return componentwise error bounds.
*
*     ERR_BNDS_NORM  (input/output) DOUBLE PRECISION array, dimension
*                    (NRHS, N_ERR_BNDS)
*     For each right-hand side, this array contains information about
*     various error bounds and condition numbers corresponding to the
*     normwise relative error, which is defined as follows:
*
*     Normwise relative error in the ith solution vector:
*             max_j (abs(XTRUE(j,i) - X(j,i)))
*            ------------------------------
*                  max_j abs(X(j,i))
*
*     The array is indexed by the type of error information as described
*     below. There currently are up to three pieces of information
*     returned.
*
*     The first index in ERR_BNDS_NORM(i,:) corresponds to the ith
*     right-hand side.
*
*     The second index in ERR_BNDS_NORM(:,err) contains the following
*     three fields:
*     err = 1 "Trust/don't trust" boolean. Trust the answer if the
*              reciprocal condition number is less than the threshold
*              sqrt(n) * slamch('Epsilon').
*
*     err = 2 "Guaranteed" error bound: The estimated forward error,
*              almost certainly within a factor of 10 of the true error
*              so long as the next entry is greater than the threshold
*              sqrt(n) * slamch('Epsilon'). This error bound should only
*              be trusted if the previous boolean is true.
*
*     err = 3  Reciprocal condition number: Estimated normwise
*              reciprocal condition number.  Compared with the threshold
*              sqrt(n) * slamch('Epsilon') to determine if the error
*              estimate is "guaranteed". These reciprocal condition
*              numbers are 1 / (norm(Z^{-1},inf) * norm(Z,inf)) for some
*              appropriately scaled matrix Z.
*              Let Z = S*A, where S scales each row by a power of the
*              radix so all absolute row sums of Z are approximately 1.
*
*     This subroutine is only responsible for setting the second field
*     above.
*     See Lapack Working Note 165 for further details and extra
*     cautions.
*
*     ERR_BNDS_COMP  (input/output) DOUBLE PRECISION array, dimension
*                    (NRHS, N_ERR_BNDS)
*     For each right-hand side, this array contains information about
*     various error bounds and condition numbers corresponding to the
*     componentwise relative error, which is defined as follows:
*
*     Componentwise relative error in the ith solution vector:
*                    abs(XTRUE(j,i) - X(j,i))
*             max_j ----------------------
*                         abs(X(j,i))
*
*     The array is indexed by the right-hand side i (on which the
*     componentwise relative error depends), and the type of error
*     information as described below. There currently are up to three
*     pieces of information returned for each right-hand side. If
*     componentwise accuracy is not requested (PARAMS(3) = 0.0), then
*     ERR_BNDS_COMP is not accessed.  If N_ERR_BNDS .LT. 3, then at most
*     the first (:,N_ERR_BNDS) entries are returned.
*
*     The first index in ERR_BNDS_COMP(i,:) corresponds to the ith
*     right-hand side.
*
*     The second index in ERR_BNDS_COMP(:,err) contains the following
*     three fields:
*     err = 1 "Trust/don't trust" boolean. Trust the answer if the
*              reciprocal condition number is less than the threshold
*              sqrt(n) * slamch('Epsilon').
*
*     err = 2 "Guaranteed" error bound: The estimated forward error,
*              almost certainly within a factor of 10 of the true error
*              so long as the next entry is greater than the threshold
*              sqrt(n) * slamch('Epsilon'). This error bound should only
*              be trusted if the previous boolean is true.
*
*     err = 3  Reciprocal condition number: Estimated componentwise
*              reciprocal condition number.  Compared with the threshold
*              sqrt(n) * slamch('Epsilon') to determine if the error
*              estimate is "guaranteed". These reciprocal condition
*              numbers are 1 / (norm(Z^{-1},inf) * norm(Z,inf)) for some
*              appropriately scaled matrix Z.
*              Let Z = S*(A*diag(x)), where x is the solution for the
*              current right-hand side and S scales each row of
*              A*diag(x) by a power of the radix so all absolute row
*              sums of Z are approximately 1.
*
*     This subroutine is only responsible for setting the second field
*     above.
*     See Lapack Working Note 165 for further details and extra
*     cautions.
*
*     RES            (input) DOUBLE PRECISION array, dimension (N)
*     Workspace to hold the intermediate residual.
*
*     AYB            (input) DOUBLE PRECISION array, dimension (N)
*     Workspace. This can be the same workspace passed for Y_TAIL.
*
*     DY             (input) DOUBLE PRECISION array, dimension (N)
*     Workspace to hold the intermediate solution.
*
*     Y_TAIL         (input) DOUBLE PRECISION array, dimension (N)
*     Workspace to hold the trailing bits of the intermediate solution.
*
*     RCOND          (input) DOUBLE PRECISION
*     Reciprocal scaled condition number.  This is an estimate of the
*     reciprocal Skeel condition number of the matrix A after
*     equilibration (if done).  If this is less than the machine
*     precision (in particular, if it is zero), the matrix is singular
*     to working precision.  Note that the error may still be small even
*     if this number is very small and the matrix appears ill-
*     conditioned.
*
*     ITHRESH        (input) INTEGER
*     The maximum number of residual computations allowed for
*     refinement. The default is 10. For 'aggressive' set to 100 to
*     permit convergence using approximate factorizations or
*     factorizations other than LU. If the factorization uses a
*     technique other than Gaussian elimination, the guarantees in
*     ERR_BNDS_NORM and ERR_BNDS_COMP may no longer be trustworthy.
*
*     RTHRESH        (input) DOUBLE PRECISION
*     Determines when to stop refinement if the error estimate stops
*     decreasing. Refinement will stop when the next solution no longer
*     satisfies norm(dx_{i+1}) < RTHRESH * norm(dx_i) where norm(Z) is
*     the infinity norm of Z. RTHRESH satisfies 0 < RTHRESH <= 1. The
*     default value is 0.5. For 'aggressive' set to 0.9 to permit
*     convergence on extremely ill-conditioned matrices. See LAWN 165
*     for more details.
*
*     DZ_UB          (input) DOUBLE PRECISION
*     Determines when to start considering componentwise convergence.
*     Componentwise convergence is only considered after each component
*     of the solution Y is stable, which we definte as the relative
*     change in each component being less than DZ_UB. The default value
*     is 0.25, requiring the first bit to be stable. See LAWN 165 for
*     more details.
*
*     IGNORE_CWISE   (input) LOGICAL
*     If .TRUE. then ignore componentwise convergence. Default value
*     is .FALSE..
*
*     INFO           (output) INTEGER
*       = 0:  Successful exit.
*       < 0:  if INFO = -i, the ith argument to DGETRS had an illegal
*             value
*
*  =====================================================================
*
*     .. Local Scalars ..
      CHARACTER          TRANS
      INTEGER            CNT, I, J, X_STATE, Z_STATE, Y_PREC_STATE
      DOUBLE PRECISION   YK, DYK, YMIN, NORMY, NORMX, NORMDX, DXRAT,
     $                   DZRAT, PREVNORMDX, PREV_DZ_Z, DXRATMAX,
     $                   DZRATMAX, DX_X, DZ_Z, FINAL_DX_X, FINAL_DZ_Z,
     $                   EPS, HUGEVAL, INCR_THRESH
      LOGICAL            INCR_PREC
*     ..
*     .. Parameters ..
      INTEGER            UNSTABLE_STATE, WORKING_STATE, CONV_STATE,
     $                   NOPROG_STATE, BASE_RESIDUAL, EXTRA_RESIDUAL,
     $                   EXTRA_Y
      PARAMETER          ( UNSTABLE_STATE = 0, WORKING_STATE = 1,
     $                   CONV_STATE = 2, NOPROG_STATE = 3 )
      PARAMETER          ( BASE_RESIDUAL = 0, EXTRA_RESIDUAL = 1,
     $                   EXTRA_Y = 2 )
      INTEGER            FINAL_NRM_ERR_I, FINAL_CMP_ERR_I, BERR_I
      INTEGER            RCOND_I, NRM_RCOND_I, NRM_ERR_I, CMP_RCOND_I
      INTEGER            CMP_ERR_I, PIV_GROWTH_I
      PARAMETER          ( FINAL_NRM_ERR_I = 1, FINAL_CMP_ERR_I = 2,
     $                   BERR_I = 3 )
      PARAMETER          ( RCOND_I = 4, NRM_RCOND_I = 5, NRM_ERR_I = 6 )
      PARAMETER          ( CMP_RCOND_I = 7, CMP_ERR_I = 8,
     $                   PIV_GROWTH_I = 9 )
      INTEGER            LA_LINRX_ITREF_I, LA_LINRX_ITHRESH_I,
     $                   LA_LINRX_CWISE_I
      PARAMETER          ( LA_LINRX_ITREF_I = 1,
     $                   LA_LINRX_ITHRESH_I = 2 )
      PARAMETER          ( LA_LINRX_CWISE_I = 3 )
      INTEGER            LA_LINRX_TRUST_I, LA_LINRX_ERR_I,
     $                   LA_LINRX_RCOND_I
      PARAMETER          ( LA_LINRX_TRUST_I = 1, LA_LINRX_ERR_I = 2 )
      PARAMETER          ( LA_LINRX_RCOND_I = 3 )
*     ..
*     .. External Subroutines ..
      EXTERNAL           DAXPY, DCOPY, DGETRS, DGEMV, BLAS_DGEMV_X,
     $                   BLAS_DGEMV2_X, DLA_GEAMV, DLA_WWADDW, DLAMCH,
     $                   CHLA_TRANSTYPE, DLA_LIN_BERR
      DOUBLE PRECISION   DLAMCH
      CHARACTER          CHLA_TRANSTYPE
*     ..
*     .. Intrinsic Functions ..
      INTRINSIC          ABSMAXMIN
*     ..
*     .. Executable Statements ..
*
      IF ( INFO.NE.0 ) RETURN
      TRANS = CHLA_TRANSTYPE(TRANS_TYPE)
      EPS = DLAMCH( 'Epsilon' )
      HUGEVAL = DLAMCH( 'Overflow' )
*     Force HUGEVAL to Inf
      HUGEVAL = HUGEVAL * HUGEVAL
*     Using HUGEVAL may lead to spurious underflows.
      INCR_THRESH = DBLE( N ) * EPS
*
      DO J = 1, NRHS
         Y_PREC_STATE = EXTRA_RESIDUAL
         IF ( Y_PREC_STATE .EQ. EXTRA_Y ) THEN
            DO I = 1, N
               Y_TAIL( I ) = 0.0D+0
            END DO
         END IF

         DXRAT = 0.0D+0
         DXRATMAX = 0.0D+0
         DZRAT = 0.0D+0
         DZRATMAX = 0.0D+0
         FINAL_DX_X = HUGEVAL
         FINAL_DZ_Z = HUGEVAL
         PREVNORMDX = HUGEVAL
         PREV_DZ_Z = HUGEVAL
         DZ_Z = HUGEVAL
         DX_X = HUGEVAL

         X_STATE = WORKING_STATE
         Z_STATE = UNSTABLE_STATE
         INCR_PREC = .FALSE.

         DO CNT = 1, ITHRESH
*
*         Compute residual RES = B_s - op(A_s) * Y,
*             op(A) = A, A**T, or A**H depending on TRANS (and type).
*
            CALL DCOPY( N, B( 1, J ), 1, RES, 1 )
            IF ( Y_PREC_STATE .EQ. BASE_RESIDUAL ) THEN
               CALL DGEMV( TRANS, N, N, -1.0D+0, A, LDA, Y( 1, J ), 1,
     $              1.0D+0, RES, 1 )
            ELSE IF ( Y_PREC_STATE .EQ. EXTRA_RESIDUAL ) THEN
               CALL BLAS_DGEMV_X( TRANS_TYPE, N, N, -1.0D+0, A, LDA,
     $              Y( 1, J ), 11.0D+0, RES, 1, PREC_TYPE )
            ELSE
               CALL BLAS_DGEMV2_X( TRANS_TYPE, N, N, -1.0D+0, A, LDA,
     $              Y( 1, J ), Y_TAIL, 11.0D+0, RES, 1, PREC_TYPE )
            END IF

!        XXX: RES is no longer needed.
            CALL DCOPY( N, RES, 1, DY, 1 )
            CALL DGETRS( TRANS, N, 1, AF, LDAF, IPIV, DY, N, INFO )
*
*         Calculate relative changes DX_X, DZ_Z and ratios DXRAT, DZRAT.
*
            NORMX = 0.0D+0
            NORMY = 0.0D+0
            NORMDX = 0.0D+0
            DZ_Z = 0.0D+0
            YMIN = HUGEVAL
*
            DO I = 1, N
               YK = ABS( Y( I, J ) )
               DYK = ABS( DY( I ) )

               IF ( YK .NE. 0.0D+0 ) THEN
                  DZ_Z = MAX( DZ_Z, DYK / YK )
               ELSE IF ( DYK .NE. 0.0D+0 ) THEN
                  DZ_Z = HUGEVAL
               END IF

               YMIN = MIN( YMIN, YK )

               NORMY = MAX( NORMY, YK )

               IF ( COLEQU ) THEN
                  NORMX = MAX( NORMX, YK * C( I ) )
                  NORMDX = MAX( NORMDX, DYK * C( I ) )
               ELSE
                  NORMX = NORMY
                  NORMDX = MAX( NORMDX, DYK )
               END IF
            END DO

            IF ( NORMX .NE. 0.0D+0 ) THEN
               DX_X = NORMDX / NORMX
            ELSE IF ( NORMDX .EQ. 0.0D+0 ) THEN
               DX_X = 0.0D+0
            ELSE
               DX_X = HUGEVAL
            END IF

            DXRAT = NORMDX / PREVNORMDX
            DZRAT = DZ_Z / PREV_DZ_Z
*
*         Check termination criteria
*
            IF (.NOT.IGNORE_CWISE
     $           .AND. YMIN*RCOND .LT. INCR_THRESH*NORMY
     $           .AND. Y_PREC_STATE .LT. EXTRA_Y)
     $           INCR_PREC = .TRUE.

            IF ( X_STATE .EQ. NOPROG_STATE .AND. DXRAT .LE. RTHRESH )
     $           X_STATE = WORKING_STATE
            IF ( X_STATE .EQ. WORKING_STATE ) THEN
               IF ( DX_X .LE. EPS ) THEN
                  X_STATE = CONV_STATE
               ELSE IF ( DXRAT .GT. RTHRESH ) THEN
                  IF ( Y_PREC_STATE .NE. EXTRA_Y ) THEN
                     INCR_PREC = .TRUE.
                  ELSE
                     X_STATE = NOPROG_STATE
                  END IF
               ELSE
                  IF ( DXRAT .GT. DXRATMAX ) DXRATMAX = DXRAT
               END IF
               IF ( X_STATE .GT. WORKING_STATE ) FINAL_DX_X = DX_X
            END IF

            IF ( Z_STATE .EQ. UNSTABLE_STATE .AND. DZ_Z .LE. DZ_UB )
     $           Z_STATE = WORKING_STATE
            IF ( Z_STATE .EQ. NOPROG_STATE .AND. DZRAT .LE. RTHRESH )
     $           Z_STATE = WORKING_STATE
            IF ( Z_STATE .EQ. WORKING_STATE ) THEN
               IF ( DZ_Z .LE. EPS ) THEN
                  Z_STATE = CONV_STATE
               ELSE IF ( DZ_Z .GT. DZ_UB ) THEN
                  Z_STATE = UNSTABLE_STATE
                  DZRATMAX = 0.0D+0
                  FINAL_DZ_Z = HUGEVAL
               ELSE IF ( DZRAT .GT. RTHRESH ) THEN
                  IF ( Y_PREC_STATE .NE. EXTRA_Y ) THEN
                     INCR_PREC = .TRUE.
                  ELSE
                     Z_STATE = NOPROG_STATE
                  END IF
               ELSE
                  IF ( DZRAT .GT. DZRATMAX ) DZRATMAX = DZRAT
               END IF
               IF ( Z_STATE .GT. WORKING_STATE ) FINAL_DZ_Z = DZ_Z
            END IF
*
*           Exit if both normwise and componentwise stopped working,
*           but if componentwise is unstable, let it go at least two
*           iterations.
*
            IF ( X_STATE.NE.WORKING_STATE ) THEN
               IF ( IGNORE_CWISE) GOTO 666
               IF ( Z_STATE.EQ.NOPROG_STATE .OR. Z_STATE.EQ.CONV_STATE )
     $              GOTO 666
               IF ( Z_STATE.EQ.UNSTABLE_STATE .AND. CNT.GT.1 ) GOTO 666
            END IF

            IF ( INCR_PREC ) THEN
               INCR_PREC = .FALSE.
               Y_PREC_STATE = Y_PREC_STATE + 1
               DO I = 1, N
                  Y_TAIL( I ) = 0.0D+0
               END DO
            END IF

            PREVNORMDX = NORMDX
            PREV_DZ_Z = DZ_Z
*
*           Update soluton.
*
            IF ( Y_PREC_STATE .LT. EXTRA_Y ) THEN
               CALL DAXPY( N, 1.0D+0, DY, 1, Y( 1, J ), 1 )
            ELSE
               CALL DLA_WWADDW( N, Y( 1, J ), Y_TAIL, DY )
            END IF

         END DO
*        Target of "IF (Z_STOP .AND. X_STOP)".  Sun's f77 won't EXIT.
 666     CONTINUE
*
*     Set final_* when cnt hits ithresh.
*
         IF ( X_STATE .EQ. WORKING_STATE ) FINAL_DX_X = DX_X
         IF ( Z_STATE .EQ. WORKING_STATE ) FINAL_DZ_Z = DZ_Z
*
*     Compute error bounds
*
         IF (N_NORMS .GE. 1THEN
            ERRS_N( J, LA_LINRX_ERR_I ) = FINAL_DX_X / (1 - DXRATMAX)
         END IF
         IF ( N_NORMS .GE. 2 ) THEN
            ERRS_C( J, LA_LINRX_ERR_I ) = FINAL_DZ_Z / (1 - DZRATMAX)
         END IF
*
*     Compute componentwise relative backward error from formula
*         max(i) ( abs(R(i)) / ( abs(op(A_s))*abs(Y) + abs(B_s) )(i) )
*     where abs(Z) is the componentwise absolute value of the matrix
*     or vector Z.
*
*         Compute residual RES = B_s - op(A_s) * Y,
*             op(A) = A, A**T, or A**H depending on TRANS (and type).
*
         CALL DCOPY( N, B( 1, J ), 1, RES, 1 )
         CALL DGEMV( TRANS, N, N, -1.0D+0, A, LDA, Y(1,J), 11.0D+0
     $     RES, 1 )

         DO I = 1, N
            AYB( I ) = ABS( B( I, J ) )
         END DO
*
*     Compute abs(op(A_s))*abs(Y) + abs(B_s).
*
         CALL DLA_GEAMV ( TRANS_TYPE, N, N, 1.0D+0,
     $        A, LDA, Y(1, J), 11.0D+0, AYB, 1 )

         CALL DLA_LIN_BERR ( N, N, 1, RES, AYB, BERR_OUT( J ) )
*
*     End of loop for each RHS.
*
      END DO
*
      RETURN
      END