1 *> \brief <b> SSYSVX computes the solution to system of linear equations A * X = B for SY matrices</b>
3 * =========== DOCUMENTATION ===========
5 * Online html documentation available at
6 * http://www.netlib.org/lapack/explore-html/
9 *> Download SSYSVX + dependencies
10 *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/ssysvx.f">
12 *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/ssysvx.f">
14 *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/ssysvx.f">
21 * SUBROUTINE SSYSVX( FACT, UPLO, N, NRHS, A, LDA, AF, LDAF, IPIV, B,
22 * LDB, X, LDX, RCOND, FERR, BERR, WORK, LWORK,
25 * .. Scalar Arguments ..
26 * CHARACTER FACT, UPLO
27 * INTEGER INFO, LDA, LDAF, LDB, LDX, LWORK, N, NRHS
30 * .. Array Arguments ..
31 * INTEGER IPIV( * ), IWORK( * )
32 * REAL A( LDA, * ), AF( LDAF, * ), B( LDB, * ),
33 * $ BERR( * ), FERR( * ), WORK( * ), X( LDX, * )
42 *> SSYSVX uses the diagonal pivoting factorization to compute the
43 *> solution to a real system of linear equations A * X = B,
44 *> where A is an N-by-N symmetric matrix and X and B are N-by-NRHS
47 *> Error bounds on the solution and a condition estimate are also
56 *> The following steps are performed:
58 *> 1. If FACT = 'N', the diagonal pivoting method is used to factor A.
59 *> The form of the factorization is
60 *> A = U * D * U**T, if UPLO = 'U', or
61 *> A = L * D * L**T, if UPLO = 'L',
62 *> where U (or L) is a product of permutation and unit upper (lower)
63 *> triangular matrices, and D is symmetric and block diagonal with
64 *> 1-by-1 and 2-by-2 diagonal blocks.
66 *> 2. If some D(i,i)=0, so that D is exactly singular, then the routine
67 *> returns with INFO = i. Otherwise, the factored form of A is used
68 *> to estimate the condition number of the matrix A. If the
69 *> reciprocal of the condition number is less than machine precision,
70 *> INFO = N+1 is returned as a warning, but the routine still goes on
71 *> to solve for X and compute error bounds as described below.
73 *> 3. The system of equations is solved for X using the factored form
76 *> 4. Iterative refinement is applied to improve the computed solution
77 *> matrix and calculate error bounds and backward error estimates
86 *> FACT is CHARACTER*1
87 *> Specifies whether or not the factored form of A has been
89 *> = 'F': On entry, AF and IPIV contain the factored form of
90 *> A. AF and IPIV will not be modified.
91 *> = 'N': The matrix A will be copied to AF and factored.
96 *> UPLO is CHARACTER*1
97 *> = 'U': Upper triangle of A is stored;
98 *> = 'L': Lower triangle of A is stored.
104 *> The number of linear equations, i.e., the order of the
111 *> The number of right hand sides, i.e., the number of columns
112 *> of the matrices B and X. NRHS >= 0.
117 *> A is REAL array, dimension (LDA,N)
118 *> The symmetric matrix A. If UPLO = 'U', the leading N-by-N
119 *> upper triangular part of A contains the upper triangular part
120 *> of the matrix A, and the strictly lower triangular part of A
121 *> is not referenced. If UPLO = 'L', the leading N-by-N lower
122 *> triangular part of A contains the lower triangular part of
123 *> the matrix A, and the strictly upper triangular part of A is
130 *> The leading dimension of the array A. LDA >= max(1,N).
135 *> AF is REAL array, dimension (LDAF,N)
136 *> If FACT = 'F', then AF is an input argument and on entry
137 *> contains the block diagonal matrix D and the multipliers used
138 *> to obtain the factor U or L from the factorization
139 *> A = U*D*U**T or A = L*D*L**T as computed by SSYTRF.
141 *> If FACT = 'N', then AF is an output argument and on exit
142 *> returns the block diagonal matrix D and the multipliers used
143 *> to obtain the factor U or L from the factorization
144 *> A = U*D*U**T or A = L*D*L**T.
150 *> The leading dimension of the array AF. LDAF >= max(1,N).
153 *> \param[in,out] IPIV
155 *> IPIV is INTEGER array, dimension (N)
156 *> If FACT = 'F', then IPIV is an input argument and on entry
157 *> contains details of the interchanges and the block structure
158 *> of D, as determined by SSYTRF.
159 *> If IPIV(k) > 0, then rows and columns k and IPIV(k) were
160 *> interchanged and D(k,k) is a 1-by-1 diagonal block.
161 *> If UPLO = 'U' and IPIV(k) = IPIV(k-1) < 0, then rows and
162 *> columns k-1 and -IPIV(k) were interchanged and D(k-1:k,k-1:k)
163 *> is a 2-by-2 diagonal block. If UPLO = 'L' and IPIV(k) =
164 *> IPIV(k+1) < 0, then rows and columns k+1 and -IPIV(k) were
165 *> interchanged and D(k:k+1,k:k+1) is a 2-by-2 diagonal block.
167 *> If FACT = 'N', then IPIV is an output argument and on exit
168 *> contains details of the interchanges and the block structure
169 *> of D, as determined by SSYTRF.
174 *> B is REAL array, dimension (LDB,NRHS)
175 *> The N-by-NRHS right hand side matrix B.
181 *> The leading dimension of the array B. LDB >= max(1,N).
186 *> X is REAL array, dimension (LDX,NRHS)
187 *> If INFO = 0 or INFO = N+1, the N-by-NRHS solution matrix X.
193 *> The leading dimension of the array X. LDX >= max(1,N).
199 *> The estimate of the reciprocal condition number of the matrix
200 *> A. If RCOND is less than the machine precision (in
201 *> particular, if RCOND = 0), the matrix is singular to working
202 *> precision. This condition is indicated by a return code of
208 *> FERR is REAL array, dimension (NRHS)
209 *> The estimated forward error bound for each solution vector
210 *> X(j) (the j-th column of the solution matrix X).
211 *> If XTRUE is the true solution corresponding to X(j), FERR(j)
212 *> is an estimated upper bound for the magnitude of the largest
213 *> element in (X(j) - XTRUE) divided by the magnitude of the
214 *> largest element in X(j). The estimate is as reliable as
215 *> the estimate for RCOND, and is almost always a slight
216 *> overestimate of the true error.
221 *> BERR is REAL array, dimension (NRHS)
222 *> The componentwise relative backward error of each solution
223 *> vector X(j) (i.e., the smallest relative change in
224 *> any element of A or B that makes X(j) an exact solution).
229 *> WORK is REAL array, dimension (MAX(1,LWORK))
230 *> On exit, if INFO = 0, WORK(1) returns the optimal LWORK.
236 *> The length of WORK. LWORK >= max(1,3*N), and for best
237 *> performance, when FACT = 'N', LWORK >= max(1,3*N,N*NB), where
238 *> NB is the optimal blocksize for SSYTRF.
240 *> If LWORK = -1, then a workspace query is assumed; the routine
241 *> only calculates the optimal size of the WORK array, returns
242 *> this value as the first entry of the WORK array, and no error
243 *> message related to LWORK is issued by XERBLA.
248 *> IWORK is INTEGER array, dimension (N)
254 *> = 0: successful exit
255 *> < 0: if INFO = -i, the i-th argument had an illegal value
256 *> > 0: if INFO = i, and i is
257 *> <= N: D(i,i) is exactly zero. The factorization
258 *> has been completed but the factor D is exactly
259 *> singular, so the solution and error bounds could
260 *> not be computed. RCOND = 0 is returned.
261 *> = N+1: D is nonsingular, but RCOND is less than machine
262 *> precision, meaning that the matrix is singular
263 *> to working precision. Nevertheless, the
264 *> solution and error bounds are computed because
265 *> there are a number of situations where the
266 *> computed solution can be more accurate than the
267 *> value of RCOND would suggest.
273 *> \author Univ. of Tennessee
274 *> \author Univ. of California Berkeley
275 *> \author Univ. of Colorado Denver
280 *> \ingroup realSYsolve
282 * =====================================================================
283 SUBROUTINE SSYSVX( FACT, UPLO, N, NRHS, A, LDA, AF, LDAF, IPIV, B,
284 $ LDB, X, LDX, RCOND, FERR, BERR, WORK, LWORK,
287 * -- LAPACK driver routine (version 3.4.1) --
288 * -- LAPACK is a software package provided by Univ. of Tennessee, --
289 * -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
292 * .. Scalar Arguments ..
294 INTEGER INFO, LDA, LDAF, LDB, LDX, LWORK, N, NRHS
297 * .. Array Arguments ..
298 INTEGER IPIV( * ), IWORK( * )
299 REAL A( LDA, * ), AF( LDAF, * ), B( LDB, * ),
300 $ BERR( * ), FERR( * ), WORK( * ), X( LDX, * )
303 * =====================================================================
307 PARAMETER ( ZERO = 0.0E+0 )
309 * .. Local Scalars ..
310 LOGICAL LQUERY, NOFACT
314 * .. External Functions ..
318 EXTERNAL ILAENV, LSAME, SLAMCH, SLANSY
320 * .. External Subroutines ..
321 EXTERNAL SLACPY, SSYCON, SSYRFS, SSYTRF, SSYTRS, XERBLA
323 * .. Intrinsic Functions ..
326 * .. Executable Statements ..
328 * Test the input parameters.
331 NOFACT = LSAME( FACT, 'N' )
332 LQUERY = ( LWORK.EQ.-1 )
333 IF( .NOT.NOFACT .AND. .NOT.LSAME( FACT, 'F' ) ) THEN
335 ELSE IF( .NOT.LSAME( UPLO, 'U' ) .AND. .NOT.LSAME( UPLO, 'L' ) )
338 ELSE IF( N.LT.0 ) THEN
340 ELSE IF( NRHS.LT.0 ) THEN
342 ELSE IF( LDA.LT.MAX( 1, N ) ) THEN
344 ELSE IF( LDAF.LT.MAX( 1, N ) ) THEN
346 ELSE IF( LDB.LT.MAX( 1, N ) ) THEN
348 ELSE IF( LDX.LT.MAX( 1, N ) ) THEN
350 ELSE IF( LWORK.LT.MAX( 1, 3*N ) .AND. .NOT.LQUERY ) THEN
355 LWKOPT = MAX( 1, 3*N )
357 NB = ILAENV( 1, 'SSYTRF', UPLO, N, -1, -1, -1 )
358 LWKOPT = MAX( LWKOPT, N*NB )
364 CALL XERBLA( 'SSYSVX', -INFO )
366 ELSE IF( LQUERY ) THEN
372 * Compute the factorization A = U*D*U**T or A = L*D*L**T.
374 CALL SLACPY( UPLO, N, N, A, LDA, AF, LDAF )
375 CALL SSYTRF( UPLO, N, AF, LDAF, IPIV, WORK, LWORK, INFO )
377 * Return if INFO is non-zero.
385 * Compute the norm of the matrix A.
387 ANORM = SLANSY( 'I', UPLO, N, A, LDA, WORK )
389 * Compute the reciprocal of the condition number of A.
391 CALL SSYCON( UPLO, N, AF, LDAF, IPIV, ANORM, RCOND, WORK, IWORK,
394 * Compute the solution vectors X.
396 CALL SLACPY( 'Full', N, NRHS, B, LDB, X, LDX )
397 CALL SSYTRS( UPLO, N, NRHS, AF, LDAF, IPIV, X, LDX, INFO )
399 * Use iterative refinement to improve the computed solutions and
400 * compute error bounds and backward error estimates for them.
402 CALL SSYRFS( UPLO, N, NRHS, A, LDA, AF, LDAF, IPIV, B, LDB, X,
403 $ LDX, FERR, BERR, WORK, IWORK, INFO )
405 * Set INFO = N+1 if the matrix is singular to working precision.
407 IF( RCOND.LT.SLAMCH( 'Epsilon' ) )