1       SUBROUTINE DSYSVX( FACT, UPLO, N, NRHS, A, LDA, AF, LDAF, IPIV, B,
  2      $                   LDB, X, LDX, RCOND, FERR, BERR, WORK, LWORK,
  3      $                   IWORK, INFO )
  4 *
  5 *  -- LAPACK driver routine (version 3.3.1) --
  6 *  -- LAPACK is a software package provided by Univ. of Tennessee,    --
  7 *  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
  8 *  -- April 2011                                                      --
  9 *
 10 *     .. Scalar Arguments ..
 11       CHARACTER          FACT, UPLO
 12       INTEGER            INFO, LDA, LDAF, LDB, LDX, LWORK, N, NRHS
 13       DOUBLE PRECISION   RCOND
 14 *     ..
 15 *     .. Array Arguments ..
 16       INTEGER            IPIV( * ), IWORK( * )
 17       DOUBLE PRECISION   A( LDA, * ), AF( LDAF, * ), B( LDB, * ),
 18      $                   BERR( * ), FERR( * ), WORK( * ), X( LDX, * )
 19 *     ..
 20 *
 21 *  Purpose
 22 *  =======
 23 *
 24 *  DSYSVX uses the diagonal pivoting factorization to compute the
 25 *  solution to a real system of linear equations A * X = B,
 26 *  where A is an N-by-N symmetric matrix and X and B are N-by-NRHS
 27 *  matrices.
 28 *
 29 *  Error bounds on the solution and a condition estimate are also
 30 *  provided.
 31 *
 32 *  Description
 33 *  ===========
 34 *
 35 *  The following steps are performed:
 36 *
 37 *  1. If FACT = 'N', the diagonal pivoting method is used to factor A.
 38 *     The form of the factorization is
 39 *        A = U * D * U**T,  if UPLO = 'U', or
 40 *        A = L * D * L**T,  if UPLO = 'L',
 41 *     where U (or L) is a product of permutation and unit upper (lower)
 42 *     triangular matrices, and D is symmetric and block diagonal with
 43 *     1-by-1 and 2-by-2 diagonal blocks.
 44 *
 45 *  2. If some D(i,i)=0, so that D is exactly singular, then the routine
 46 *     returns with INFO = i. Otherwise, the factored form of A is used
 47 *     to estimate the condition number of the matrix A.  If the
 48 *     reciprocal of the condition number is less than machine precision,
 49 *     INFO = N+1 is returned as a warning, but the routine still goes on
 50 *     to solve for X and compute error bounds as described below.
 51 *
 52 *  3. The system of equations is solved for X using the factored form
 53 *     of A.
 54 *
 55 *  4. Iterative refinement is applied to improve the computed solution
 56 *     matrix and calculate error bounds and backward error estimates
 57 *     for it.
 58 *
 59 *  Arguments
 60 *  =========
 61 *
 62 *  FACT    (input) CHARACTER*1
 63 *          Specifies whether or not the factored form of A has been
 64 *          supplied on entry.
 65 *          = 'F':  On entry, AF and IPIV contain the factored form of
 66 *                  A.  AF and IPIV will not be modified.
 67 *          = 'N':  The matrix A will be copied to AF and factored.
 68 *
 69 *  UPLO    (input) CHARACTER*1
 70 *          = 'U':  Upper triangle of A is stored;
 71 *          = 'L':  Lower triangle of A is stored.
 72 *
 73 *  N       (input) INTEGER
 74 *          The number of linear equations, i.e., the order of the
 75 *          matrix A.  N >= 0.
 76 *
 77 *  NRHS    (input) INTEGER
 78 *          The number of right hand sides, i.e., the number of columns
 79 *          of the matrices B and X.  NRHS >= 0.
 80 *
 81 *  A       (input) DOUBLE PRECISION array, dimension (LDA,N)
 82 *          The symmetric matrix A.  If UPLO = 'U', the leading N-by-N
 83 *          upper triangular part of A contains the upper triangular part
 84 *          of the matrix A, and the strictly lower triangular part of A
 85 *          is not referenced.  If UPLO = 'L', the leading N-by-N lower
 86 *          triangular part of A contains the lower triangular part of
 87 *          the matrix A, and the strictly upper triangular part of A is
 88 *          not referenced.
 89 *
 90 *  LDA     (input) INTEGER
 91 *          The leading dimension of the array A.  LDA >= max(1,N).
 92 *
 93 *  AF      (input or output) DOUBLE PRECISION array, dimension (LDAF,N)
 94 *          If FACT = 'F', then AF is an input argument and on entry
 95 *          contains the block diagonal matrix D and the multipliers used
 96 *          to obtain the factor U or L from the factorization
 97 *          A = U*D*U**T or A = L*D*L**T as computed by DSYTRF.
 98 *
 99 *          If FACT = 'N', then AF is an output argument and on exit
100 *          returns the block diagonal matrix D and the multipliers used
101 *          to obtain the factor U or L from the factorization
102 *          A = U*D*U**T or A = L*D*L**T.
103 *
104 *  LDAF    (input) INTEGER
105 *          The leading dimension of the array AF.  LDAF >= max(1,N).
106 *
107 *  IPIV    (input or output) INTEGER array, dimension (N)
108 *          If FACT = 'F', then IPIV is an input argument and on entry
109 *          contains details of the interchanges and the block structure
110 *          of D, as determined by DSYTRF.
111 *          If IPIV(k) > 0, then rows and columns k and IPIV(k) were
112 *          interchanged and D(k,k) is a 1-by-1 diagonal block.
113 *          If UPLO = 'U' and IPIV(k) = IPIV(k-1) < 0, then rows and
114 *          columns k-1 and -IPIV(k) were interchanged and D(k-1:k,k-1:k)
115 *          is a 2-by-2 diagonal block.  If UPLO = 'L' and IPIV(k) =
116 *          IPIV(k+1) < 0, then rows and columns k+1 and -IPIV(k) were
117 *          interchanged and D(k:k+1,k:k+1) is a 2-by-2 diagonal block.
118 *
119 *          If FACT = 'N', then IPIV is an output argument and on exit
120 *          contains details of the interchanges and the block structure
121 *          of D, as determined by DSYTRF.
122 *
123 *  B       (input) DOUBLE PRECISION array, dimension (LDB,NRHS)
124 *          The N-by-NRHS right hand side matrix B.
125 *
126 *  LDB     (input) INTEGER
127 *          The leading dimension of the array B.  LDB >= max(1,N).
128 *
129 *  X       (output) DOUBLE PRECISION array, dimension (LDX,NRHS)
130 *          If INFO = 0 or INFO = N+1, the N-by-NRHS solution matrix X.
131 *
132 *  LDX     (input) INTEGER
133 *          The leading dimension of the array X.  LDX >= max(1,N).
134 *
135 *  RCOND   (output) DOUBLE PRECISION
136 *          The estimate of the reciprocal condition number of the matrix
137 *          A.  If RCOND is less than the machine precision (in
138 *          particular, if RCOND = 0), the matrix is singular to working
139 *          precision.  This condition is indicated by a return code of
140 *          INFO > 0.
141 *
142 *  FERR    (output) DOUBLE PRECISION array, dimension (NRHS)
143 *          The estimated forward error bound for each solution vector
144 *          X(j) (the j-th column of the solution matrix X).
145 *          If XTRUE is the true solution corresponding to X(j), FERR(j)
146 *          is an estimated upper bound for the magnitude of the largest
147 *          element in (X(j) - XTRUE) divided by the magnitude of the
148 *          largest element in X(j).  The estimate is as reliable as
149 *          the estimate for RCOND, and is almost always a slight
150 *          overestimate of the true error.
151 *
152 *  BERR    (output) DOUBLE PRECISION array, dimension (NRHS)
153 *          The componentwise relative backward error of each solution
154 *          vector X(j) (i.e., the smallest relative change in
155 *          any element of A or B that makes X(j) an exact solution).
156 *
157 *  WORK    (workspace/output) DOUBLE PRECISION array, dimension (MAX(1,LWORK))
158 *          On exit, if INFO = 0, WORK(1) returns the optimal LWORK.
159 *
160 *  LWORK   (input) INTEGER
161 *          The length of WORK.  LWORK >= max(1,3*N), and for best
162 *          performance, when FACT = 'N', LWORK >= max(1,3*N,N*NB), where
163 *          NB is the optimal blocksize for DSYTRF.
164 *
165 *          If LWORK = -1, then a workspace query is assumed; the routine
166 *          only calculates the optimal size of the WORK array, returns
167 *          this value as the first entry of the WORK array, and no error
168 *          message related to LWORK is issued by XERBLA.
169 *
170 *  IWORK   (workspace) INTEGER array, dimension (N)
171 *
172 *  INFO    (output) INTEGER
173 *          = 0: successful exit
174 *          < 0: if INFO = -i, the i-th argument had an illegal value
175 *          > 0: if INFO = i, and i is
176 *                <= N:  D(i,i) is exactly zero.  The factorization
177 *                       has been completed but the factor D is exactly
178 *                       singular, so the solution and error bounds could
179 *                       not be computed. RCOND = 0 is returned.
180 *                = N+1: D is nonsingular, but RCOND is less than machine
181 *                       precision, meaning that the matrix is singular
182 *                       to working precision.  Nevertheless, the
183 *                       solution and error bounds are computed because
184 *                       there are a number of situations where the
185 *                       computed solution can be more accurate than the
186 *                       value of RCOND would suggest.
187 *
188 *  =====================================================================
189 *
190 *     .. Parameters ..
191       DOUBLE PRECISION   ZERO
192       PARAMETER          ( ZERO = 0.0D+0 )
193 *     ..
194 *     .. Local Scalars ..
195       LOGICAL            LQUERY, NOFACT
196       INTEGER            LWKOPT, NB
197       DOUBLE PRECISION   ANORM
198 *     ..
199 *     .. External Functions ..
200       LOGICAL            LSAME
201       INTEGER            ILAENV
202       DOUBLE PRECISION   DLAMCH, DLANSY
203       EXTERNAL           LSAME, ILAENV, DLAMCH, DLANSY
204 *     ..
205 *     .. External Subroutines ..
206       EXTERNAL           DLACPY, DSYCON, DSYRFS, DSYTRF, DSYTRS, XERBLA
207 *     ..
208 *     .. Intrinsic Functions ..
209       INTRINSIC          MAX
210 *     ..
211 *     .. Executable Statements ..
212 *
213 *     Test the input parameters.
214 *
215       INFO = 0
216       NOFACT = LSAME( FACT, 'N' )
217       LQUERY = ( LWORK.EQ.-1 )
218       IF.NOT.NOFACT .AND. .NOT.LSAME( FACT, 'F' ) ) THEN
219          INFO = -1
220       ELSE IF.NOT.LSAME( UPLO, 'U' ) .AND. .NOT.LSAME( UPLO, 'L' ) )
221      $          THEN
222          INFO = -2
223       ELSE IF( N.LT.0 ) THEN
224          INFO = -3
225       ELSE IF( NRHS.LT.0 ) THEN
226          INFO = -4
227       ELSE IF( LDA.LT.MAX1, N ) ) THEN
228          INFO = -6
229       ELSE IF( LDAF.LT.MAX1, N ) ) THEN
230          INFO = -8
231       ELSE IF( LDB.LT.MAX1, N ) ) THEN
232          INFO = -11
233       ELSE IF( LDX.LT.MAX1, N ) ) THEN
234          INFO = -13
235       ELSE IF( LWORK.LT.MAX13*N ) .AND. .NOT.LQUERY ) THEN
236          INFO = -18
237       END IF
238 *
239       IF( INFO.EQ.0 ) THEN
240          LWKOPT = MAX13*N )
241          IF( NOFACT ) THEN
242             NB = ILAENV( 1'DSYTRF', UPLO, N, -1-1-1 )
243             LWKOPT = MAX( LWKOPT, N*NB )
244          END IF
245          WORK( 1 ) = LWKOPT
246       END IF
247 *
248       IF( INFO.NE.0 ) THEN
249          CALL XERBLA( 'DSYSVX'-INFO )
250          RETURN
251       ELSE IF( LQUERY ) THEN
252          RETURN
253       END IF
254 *
255       IF( NOFACT ) THEN
256 *
257 *        Compute the factorization A = U*D*U**T or A = L*D*L**T.
258 *
259          CALL DLACPY( UPLO, N, N, A, LDA, AF, LDAF )
260          CALL DSYTRF( UPLO, N, AF, LDAF, IPIV, WORK, LWORK, INFO )
261 *
262 *        Return if INFO is non-zero.
263 *
264          IF( INFO.GT.0 )THEN
265             RCOND = ZERO
266             RETURN
267          END IF
268       END IF
269 *
270 *     Compute the norm of the matrix A.
271 *
272       ANORM = DLANSY( 'I', UPLO, N, A, LDA, WORK )
273 *
274 *     Compute the reciprocal of the condition number of A.
275 *
276       CALL DSYCON( UPLO, N, AF, LDAF, IPIV, ANORM, RCOND, WORK, IWORK,
277      $             INFO )
278 *
279 *     Compute the solution vectors X.
280 *
281       CALL DLACPY( 'Full', N, NRHS, B, LDB, X, LDX )
282       CALL DSYTRS( UPLO, N, NRHS, AF, LDAF, IPIV, X, LDX, INFO )
283 *
284 *     Use iterative refinement to improve the computed solutions and
285 *     compute error bounds and backward error estimates for them.
286 *
287       CALL DSYRFS( UPLO, N, NRHS, A, LDA, AF, LDAF, IPIV, B, LDB, X,
288      $             LDX, FERR, BERR, WORK, IWORK, INFO )
289 *
290 *     Set INFO = N+1 if the matrix is singular to working precision.
291 *
292       IF( RCOND.LT.DLAMCH( 'Epsilon' ) )
293      $   INFO = N + 1
294 *
295       WORK( 1 ) = LWKOPT
296 *
297       RETURN
298 *
299 *     End of DSYSVX
300 *
301       END