1       SUBROUTINE DGBEQU( M, N, KL, KU, AB, LDAB, R, C, ROWCND, COLCND,
  2      $                   AMAX, INFO )
  3 *
  4 *  -- LAPACK routine (version 3.2) --
  5 *  -- LAPACK is a software package provided by Univ. of Tennessee,    --
  6 *  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
  7 *     November 2006
  8 *
  9 *     .. Scalar Arguments ..
 10       INTEGER            INFO, KL, KU, LDAB, M, N
 11       DOUBLE PRECISION   AMAX, COLCND, ROWCND
 12 *     ..
 13 *     .. Array Arguments ..
 14       DOUBLE PRECISION   AB( LDAB, * ), C( * ), R( * )
 15 *     ..
 16 *
 17 *  Purpose
 18 *  =======
 19 *
 20 *  DGBEQU computes row and column scalings intended to equilibrate an
 21 *  M-by-N band matrix A and reduce its condition number.  R returns the
 22 *  row scale factors and C the column scale factors, chosen to try to
 23 *  make the largest element in each row and column of the matrix B with
 24 *  elements B(i,j)=R(i)*A(i,j)*C(j) have absolute value 1.
 25 *
 26 *  R(i) and C(j) are restricted to be between SMLNUM = smallest safe
 27 *  number and BIGNUM = largest safe number.  Use of these scaling
 28 *  factors is not guaranteed to reduce the condition number of A but
 29 *  works well in practice.
 30 *
 31 *  Arguments
 32 *  =========
 33 *
 34 *  M       (input) INTEGER
 35 *          The number of rows of the matrix A.  M >= 0.
 36 *
 37 *  N       (input) INTEGER
 38 *          The number of columns of the matrix A.  N >= 0.
 39 *
 40 *  KL      (input) INTEGER
 41 *          The number of subdiagonals within the band of A.  KL >= 0.
 42 *
 43 *  KU      (input) INTEGER
 44 *          The number of superdiagonals within the band of A.  KU >= 0.
 45 *
 46 *  AB      (input) DOUBLE PRECISION array, dimension (LDAB,N)
 47 *          The band matrix A, stored in rows 1 to KL+KU+1.  The j-th
 48 *          column of A is stored in the j-th column of the array AB as
 49 *          follows:
 50 *          AB(ku+1+i-j,j) = A(i,j) for max(1,j-ku)<=i<=min(m,j+kl).
 51 *
 52 *  LDAB    (input) INTEGER
 53 *          The leading dimension of the array AB.  LDAB >= KL+KU+1.
 54 *
 55 *  R       (output) DOUBLE PRECISION array, dimension (M)
 56 *          If INFO = 0, or INFO > M, R contains the row scale factors
 57 *          for A.
 58 *
 59 *  C       (output) DOUBLE PRECISION array, dimension (N)
 60 *          If INFO = 0, C contains the column scale factors for A.
 61 *
 62 *  ROWCND  (output) DOUBLE PRECISION
 63 *          If INFO = 0 or INFO > M, ROWCND contains the ratio of the
 64 *          smallest R(i) to the largest R(i).  If ROWCND >= 0.1 and
 65 *          AMAX is neither too large nor too small, it is not worth
 66 *          scaling by R.
 67 *
 68 *  COLCND  (output) DOUBLE PRECISION
 69 *          If INFO = 0, COLCND contains the ratio of the smallest
 70 *          C(i) to the largest C(i).  If COLCND >= 0.1, it is not
 71 *          worth scaling by C.
 72 *
 73 *  AMAX    (output) DOUBLE PRECISION
 74 *          Absolute value of largest matrix element.  If AMAX is very
 75 *          close to overflow or very close to underflow, the matrix
 76 *          should be scaled.
 77 *
 78 *  INFO    (output) INTEGER
 79 *          = 0:  successful exit
 80 *          < 0:  if INFO = -i, the i-th argument had an illegal value
 81 *          > 0:  if INFO = i, and i is
 82 *                <= M:  the i-th row of A is exactly zero
 83 *                >  M:  the (i-M)-th column of A is exactly zero
 84 *
 85 *  =====================================================================
 86 *
 87 *     .. Parameters ..
 88       DOUBLE PRECISION   ONE, ZERO
 89       PARAMETER          ( ONE = 1.0D+0, ZERO = 0.0D+0 )
 90 *     ..
 91 *     .. Local Scalars ..
 92       INTEGER            I, J, KD
 93       DOUBLE PRECISION   BIGNUM, RCMAX, RCMIN, SMLNUM
 94 *     ..
 95 *     .. External Functions ..
 96       DOUBLE PRECISION   DLAMCH
 97       EXTERNAL           DLAMCH
 98 *     ..
 99 *     .. External Subroutines ..
100       EXTERNAL           XERBLA
101 *     ..
102 *     .. Intrinsic Functions ..
103       INTRINSIC          ABSMAXMIN
104 *     ..
105 *     .. Executable Statements ..
106 *
107 *     Test the input parameters
108 *
109       INFO = 0
110       IF( M.LT.0 ) THEN
111          INFO = -1
112       ELSE IF( N.LT.0 ) THEN
113          INFO = -2
114       ELSE IF( KL.LT.0 ) THEN
115          INFO = -3
116       ELSE IF( KU.LT.0 ) THEN
117          INFO = -4
118       ELSE IF( LDAB.LT.KL+KU+1 ) THEN
119          INFO = -6
120       END IF
121       IF( INFO.NE.0 ) THEN
122          CALL XERBLA( 'DGBEQU'-INFO )
123          RETURN
124       END IF
125 *
126 *     Quick return if possible
127 *
128       IF( M.EQ.0 .OR. N.EQ.0 ) THEN
129          ROWCND = ONE
130          COLCND = ONE
131          AMAX = ZERO
132          RETURN
133       END IF
134 *
135 *     Get machine constants.
136 *
137       SMLNUM = DLAMCH( 'S' )
138       BIGNUM = ONE / SMLNUM
139 *
140 *     Compute row scale factors.
141 *
142       DO 10 I = 1, M
143          R( I ) = ZERO
144    10 CONTINUE
145 *
146 *     Find the maximum element in each row.
147 *
148       KD = KU + 1
149       DO 30 J = 1, N
150          DO 20 I = MAX( J-KU, 1 ), MIN( J+KL, M )
151             R( I ) = MAX( R( I ), ABS( AB( KD+I-J, J ) ) )
152    20    CONTINUE
153    30 CONTINUE
154 *
155 *     Find the maximum and minimum scale factors.
156 *
157       RCMIN = BIGNUM
158       RCMAX = ZERO
159       DO 40 I = 1, M
160          RCMAX = MAX( RCMAX, R( I ) )
161          RCMIN = MIN( RCMIN, R( I ) )
162    40 CONTINUE
163       AMAX = RCMAX
164 *
165       IF( RCMIN.EQ.ZERO ) THEN
166 *
167 *        Find the first zero scale factor and return an error code.
168 *
169          DO 50 I = 1, M
170             IF( R( I ).EQ.ZERO ) THEN
171                INFO = I
172                RETURN
173             END IF
174    50    CONTINUE
175       ELSE
176 *
177 *        Invert the scale factors.
178 *
179          DO 60 I = 1, M
180             R( I ) = ONE / MINMAX( R( I ), SMLNUM ), BIGNUM )
181    60    CONTINUE
182 *
183 *        Compute ROWCND = min(R(I)) / max(R(I))
184 *
185          ROWCND = MAX( RCMIN, SMLNUM ) / MIN( RCMAX, BIGNUM )
186       END IF
187 *
188 *     Compute column scale factors
189 *
190       DO 70 J = 1, N
191          C( J ) = ZERO
192    70 CONTINUE
193 *
194 *     Find the maximum element in each column,
195 *     assuming the row scaling computed above.
196 *
197       KD = KU + 1
198       DO 90 J = 1, N
199          DO 80 I = MAX( J-KU, 1 ), MIN( J+KL, M )
200             C( J ) = MAX( C( J ), ABS( AB( KD+I-J, J ) )*R( I ) )
201    80    CONTINUE
202    90 CONTINUE
203 *
204 *     Find the maximum and minimum scale factors.
205 *
206       RCMIN = BIGNUM
207       RCMAX = ZERO
208       DO 100 J = 1, N
209          RCMIN = MIN( RCMIN, C( J ) )
210          RCMAX = MAX( RCMAX, C( J ) )
211   100 CONTINUE
212 *
213       IF( RCMIN.EQ.ZERO ) THEN
214 *
215 *        Find the first zero scale factor and return an error code.
216 *
217          DO 110 J = 1, N
218             IF( C( J ).EQ.ZERO ) THEN
219                INFO = M + J
220                RETURN
221             END IF
222   110    CONTINUE
223       ELSE
224 *
225 *        Invert the scale factors.
226 *
227          DO 120 J = 1, N
228             C( J ) = ONE / MINMAX( C( J ), SMLNUM ), BIGNUM )
229   120    CONTINUE
230 *
231 *        Compute COLCND = min(C(J)) / max(C(J))
232 *
233          COLCND = MAX( RCMIN, SMLNUM ) / MIN( RCMAX, BIGNUM )
234       END IF
235 *
236       RETURN
237 *
238 *     End of DGBEQU
239 *
240       END