LAPACK 3.12.0
LAPACK: Linear Algebra PACKage
Loading...
Searching...
No Matches
sgesvxx.f
Go to the documentation of this file.
1*> \brief <b> SGESVXX computes the solution to system of linear equations A * X = B for GE matrices</b>
2*
3* =========== DOCUMENTATION ===========
4*
5* Online html documentation available at
6* http://www.netlib.org/lapack/explore-html/
7*
8*> \htmlonly
9*> Download SGESVXX + dependencies
10*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/sgesvxx.f">
11*> [TGZ]</a>
12*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/sgesvxx.f">
13*> [ZIP]</a>
14*> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/sgesvxx.f">
15*> [TXT]</a>
16*> \endhtmlonly
17*
18* Definition:
19* ===========
20*
21* SUBROUTINE SGESVXX( FACT, TRANS, N, NRHS, A, LDA, AF, LDAF, IPIV,
22* EQUED, R, C, B, LDB, X, LDX, RCOND, RPVGRW,
23* BERR, N_ERR_BNDS, ERR_BNDS_NORM,
24* ERR_BNDS_COMP, NPARAMS, PARAMS, WORK, IWORK,
25* INFO )
26*
27* .. Scalar Arguments ..
28* CHARACTER EQUED, FACT, TRANS
29* INTEGER INFO, LDA, LDAF, LDB, LDX, N, NRHS, NPARAMS,
30* $ N_ERR_BNDS
31* REAL RCOND, RPVGRW
32* ..
33* .. Array Arguments ..
34* INTEGER IPIV( * ), IWORK( * )
35* REAL A( LDA, * ), AF( LDAF, * ), B( LDB, * ),
36* $ X( LDX , * ),WORK( * )
37* REAL R( * ), C( * ), PARAMS( * ), BERR( * ),
38* $ ERR_BNDS_NORM( NRHS, * ),
39* $ ERR_BNDS_COMP( NRHS, * )
40* ..
41*
42*
43*> \par Purpose:
44* =============
45*>
46*> \verbatim
47*>
48*> SGESVXX uses the LU factorization to compute the solution to a
49*> real system of linear equations A * X = B, where A is an
50*> N-by-N matrix and X and B are N-by-NRHS matrices.
51*>
52*> If requested, both normwise and maximum componentwise error bounds
53*> are returned. SGESVXX will return a solution with a tiny
54*> guaranteed error (O(eps) where eps is the working machine
55*> precision) unless the matrix is very ill-conditioned, in which
56*> case a warning is returned. Relevant condition numbers also are
57*> calculated and returned.
58*>
59*> SGESVXX accepts user-provided factorizations and equilibration
60*> factors; see the definitions of the FACT and EQUED options.
61*> Solving with refinement and using a factorization from a previous
62*> SGESVXX call will also produce a solution with either O(eps)
63*> errors or warnings, but we cannot make that claim for general
64*> user-provided factorizations and equilibration factors if they
65*> differ from what SGESVXX would itself produce.
66*> \endverbatim
67*
68*> \par Description:
69* =================
70*>
71*> \verbatim
72*>
73*> The following steps are performed:
74*>
75*> 1. If FACT = 'E', real scaling factors are computed to equilibrate
76*> the system:
77*>
78*> TRANS = 'N': diag(R)*A*diag(C) *inv(diag(C))*X = diag(R)*B
79*> TRANS = 'T': (diag(R)*A*diag(C))**T *inv(diag(R))*X = diag(C)*B
80*> TRANS = 'C': (diag(R)*A*diag(C))**H *inv(diag(R))*X = diag(C)*B
81*>
82*> Whether or not the system will be equilibrated depends on the
83*> scaling of the matrix A, but if equilibration is used, A is
84*> overwritten by diag(R)*A*diag(C) and B by diag(R)*B (if TRANS='N')
85*> or diag(C)*B (if TRANS = 'T' or 'C').
86*>
87*> 2. If FACT = 'N' or 'E', the LU decomposition is used to factor
88*> the matrix A (after equilibration if FACT = 'E') as
89*>
90*> A = P * L * U,
91*>
92*> where P is a permutation matrix, L is a unit lower triangular
93*> matrix, and U is upper triangular.
94*>
95*> 3. If some U(i,i)=0, so that U is exactly singular, then the
96*> routine returns with INFO = i. Otherwise, the factored form of A
97*> is used to estimate the condition number of the matrix A (see
98*> argument RCOND). If the reciprocal of the condition number is less
99*> than machine precision, the routine still goes on to solve for X
100*> and compute error bounds as described below.
101*>
102*> 4. The system of equations is solved for X using the factored form
103*> of A.
104*>
105*> 5. By default (unless PARAMS(LA_LINRX_ITREF_I) is set to zero),
106*> the routine will use iterative refinement to try to get a small
107*> error and error bounds. Refinement calculates the residual to at
108*> least twice the working precision.
109*>
110*> 6. If equilibration was used, the matrix X is premultiplied by
111*> diag(C) (if TRANS = 'N') or diag(R) (if TRANS = 'T' or 'C') so
112*> that it solves the original system before equilibration.
113*> \endverbatim
114*
115* Arguments:
116* ==========
117*
118*> \verbatim
119*> Some optional parameters are bundled in the PARAMS array. These
120*> settings determine how refinement is performed, but often the
121*> defaults are acceptable. If the defaults are acceptable, users
122*> can pass NPARAMS = 0 which prevents the source code from accessing
123*> the PARAMS argument.
124*> \endverbatim
125*>
126*> \param[in] FACT
127*> \verbatim
128*> FACT is CHARACTER*1
129*> Specifies whether or not the factored form of the matrix A is
130*> supplied on entry, and if not, whether the matrix A should be
131*> equilibrated before it is factored.
132*> = 'F': On entry, AF and IPIV contain the factored form of A.
133*> If EQUED is not 'N', the matrix A has been
134*> equilibrated with scaling factors given by R and C.
135*> A, AF, and IPIV are not modified.
136*> = 'N': The matrix A will be copied to AF and factored.
137*> = 'E': The matrix A will be equilibrated if necessary, then
138*> copied to AF and factored.
139*> \endverbatim
140*>
141*> \param[in] TRANS
142*> \verbatim
143*> TRANS is CHARACTER*1
144*> Specifies the form of the system of equations:
145*> = 'N': A * X = B (No transpose)
146*> = 'T': A**T * X = B (Transpose)
147*> = 'C': A**H * X = B (Conjugate Transpose = Transpose)
148*> \endverbatim
149*>
150*> \param[in] N
151*> \verbatim
152*> N is INTEGER
153*> The number of linear equations, i.e., the order of the
154*> matrix A. N >= 0.
155*> \endverbatim
156*>
157*> \param[in] NRHS
158*> \verbatim
159*> NRHS is INTEGER
160*> The number of right hand sides, i.e., the number of columns
161*> of the matrices B and X. NRHS >= 0.
162*> \endverbatim
163*>
164*> \param[in,out] A
165*> \verbatim
166*> A is REAL array, dimension (LDA,N)
167*> On entry, the N-by-N matrix A. If FACT = 'F' and EQUED is
168*> not 'N', then A must have been equilibrated by the scaling
169*> factors in R and/or C. A is not modified if FACT = 'F' or
170*> 'N', or if FACT = 'E' and EQUED = 'N' on exit.
171*>
172*> On exit, if EQUED .ne. 'N', A is scaled as follows:
173*> EQUED = 'R': A := diag(R) * A
174*> EQUED = 'C': A := A * diag(C)
175*> EQUED = 'B': A := diag(R) * A * diag(C).
176*> \endverbatim
177*>
178*> \param[in] LDA
179*> \verbatim
180*> LDA is INTEGER
181*> The leading dimension of the array A. LDA >= max(1,N).
182*> \endverbatim
183*>
184*> \param[in,out] AF
185*> \verbatim
186*> AF is REAL array, dimension (LDAF,N)
187*> If FACT = 'F', then AF is an input argument and on entry
188*> contains the factors L and U from the factorization
189*> A = P*L*U as computed by SGETRF. If EQUED .ne. 'N', then
190*> AF is the factored form of the equilibrated matrix A.
191*>
192*> If FACT = 'N', then AF is an output argument and on exit
193*> returns the factors L and U from the factorization A = P*L*U
194*> of the original matrix A.
195*>
196*> If FACT = 'E', then AF is an output argument and on exit
197*> returns the factors L and U from the factorization A = P*L*U
198*> of the equilibrated matrix A (see the description of A for
199*> the form of the equilibrated matrix).
200*> \endverbatim
201*>
202*> \param[in] LDAF
203*> \verbatim
204*> LDAF is INTEGER
205*> The leading dimension of the array AF. LDAF >= max(1,N).
206*> \endverbatim
207*>
208*> \param[in,out] IPIV
209*> \verbatim
210*> IPIV is INTEGER array, dimension (N)
211*> If FACT = 'F', then IPIV is an input argument and on entry
212*> contains the pivot indices from the factorization A = P*L*U
213*> as computed by SGETRF; row i of the matrix was interchanged
214*> with row IPIV(i).
215*>
216*> If FACT = 'N', then IPIV is an output argument and on exit
217*> contains the pivot indices from the factorization A = P*L*U
218*> of the original matrix A.
219*>
220*> If FACT = 'E', then IPIV is an output argument and on exit
221*> contains the pivot indices from the factorization A = P*L*U
222*> of the equilibrated matrix A.
223*> \endverbatim
224*>
225*> \param[in,out] EQUED
226*> \verbatim
227*> EQUED is CHARACTER*1
228*> Specifies the form of equilibration that was done.
229*> = 'N': No equilibration (always true if FACT = 'N').
230*> = 'R': Row equilibration, i.e., A has been premultiplied by
231*> diag(R).
232*> = 'C': Column equilibration, i.e., A has been postmultiplied
233*> by diag(C).
234*> = 'B': Both row and column equilibration, i.e., A has been
235*> replaced by diag(R) * A * diag(C).
236*> EQUED is an input argument if FACT = 'F'; otherwise, it is an
237*> output argument.
238*> \endverbatim
239*>
240*> \param[in,out] R
241*> \verbatim
242*> R is REAL array, dimension (N)
243*> The row scale factors for A. If EQUED = 'R' or 'B', A is
244*> multiplied on the left by diag(R); if EQUED = 'N' or 'C', R
245*> is not accessed. R is an input argument if FACT = 'F';
246*> otherwise, R is an output argument. If FACT = 'F' and
247*> EQUED = 'R' or 'B', each element of R must be positive.
248*> If R is output, each element of R is a power of the radix.
249*> If R is input, each element of R should be a power of the radix
250*> to ensure a reliable solution and error estimates. Scaling by
251*> powers of the radix does not cause rounding errors unless the
252*> result underflows or overflows. Rounding errors during scaling
253*> lead to refining with a matrix that is not equivalent to the
254*> input matrix, producing error estimates that may not be
255*> reliable.
256*> \endverbatim
257*>
258*> \param[in,out] C
259*> \verbatim
260*> C is REAL array, dimension (N)
261*> The column scale factors for A. If EQUED = 'C' or 'B', A is
262*> multiplied on the right by diag(C); if EQUED = 'N' or 'R', C
263*> is not accessed. C is an input argument if FACT = 'F';
264*> otherwise, C is an output argument. If FACT = 'F' and
265*> EQUED = 'C' or 'B', each element of C must be positive.
266*> If C is output, each element of C is a power of the radix.
267*> If C is input, each element of C should be a power of the radix
268*> to ensure a reliable solution and error estimates. Scaling by
269*> powers of the radix does not cause rounding errors unless the
270*> result underflows or overflows. Rounding errors during scaling
271*> lead to refining with a matrix that is not equivalent to the
272*> input matrix, producing error estimates that may not be
273*> reliable.
274*> \endverbatim
275*>
276*> \param[in,out] B
277*> \verbatim
278*> B is REAL array, dimension (LDB,NRHS)
279*> On entry, the N-by-NRHS right hand side matrix B.
280*> On exit,
281*> if EQUED = 'N', B is not modified;
282*> if TRANS = 'N' and EQUED = 'R' or 'B', B is overwritten by
283*> diag(R)*B;
284*> if TRANS = 'T' or 'C' and EQUED = 'C' or 'B', B is
285*> overwritten by diag(C)*B.
286*> \endverbatim
287*>
288*> \param[in] LDB
289*> \verbatim
290*> LDB is INTEGER
291*> The leading dimension of the array B. LDB >= max(1,N).
292*> \endverbatim
293*>
294*> \param[out] X
295*> \verbatim
296*> X is REAL array, dimension (LDX,NRHS)
297*> If INFO = 0, the N-by-NRHS solution matrix X to the original
298*> system of equations. Note that A and B are modified on exit
299*> if EQUED .ne. 'N', and the solution to the equilibrated system is
300*> inv(diag(C))*X if TRANS = 'N' and EQUED = 'C' or 'B', or
301*> inv(diag(R))*X if TRANS = 'T' or 'C' and EQUED = 'R' or 'B'.
302*> \endverbatim
303*>
304*> \param[in] LDX
305*> \verbatim
306*> LDX is INTEGER
307*> The leading dimension of the array X. LDX >= max(1,N).
308*> \endverbatim
309*>
310*> \param[out] RCOND
311*> \verbatim
312*> RCOND is REAL
313*> Reciprocal scaled condition number. This is an estimate of the
314*> reciprocal Skeel condition number of the matrix A after
315*> equilibration (if done). If this is less than the machine
316*> precision (in particular, if it is zero), the matrix is singular
317*> to working precision. Note that the error may still be small even
318*> if this number is very small and the matrix appears ill-
319*> conditioned.
320*> \endverbatim
321*>
322*> \param[out] RPVGRW
323*> \verbatim
324*> RPVGRW is REAL
325*> Reciprocal pivot growth. On exit, this contains the reciprocal
326*> pivot growth factor norm(A)/norm(U). The "max absolute element"
327*> norm is used. If this is much less than 1, then the stability of
328*> the LU factorization of the (equilibrated) matrix A could be poor.
329*> This also means that the solution X, estimated condition numbers,
330*> and error bounds could be unreliable. If factorization fails with
331*> 0<INFO<=N, then this contains the reciprocal pivot growth factor
332*> for the leading INFO columns of A. In SGESVX, this quantity is
333*> returned in WORK(1).
334*> \endverbatim
335*>
336*> \param[out] BERR
337*> \verbatim
338*> BERR is REAL array, dimension (NRHS)
339*> Componentwise relative backward error. This is the
340*> componentwise relative backward error of each solution vector X(j)
341*> (i.e., the smallest relative change in any element of A or B that
342*> makes X(j) an exact solution).
343*> \endverbatim
344*>
345*> \param[in] N_ERR_BNDS
346*> \verbatim
347*> N_ERR_BNDS is INTEGER
348*> Number of error bounds to return for each right hand side
349*> and each type (normwise or componentwise). See ERR_BNDS_NORM and
350*> ERR_BNDS_COMP below.
351*> \endverbatim
352*>
353*> \param[out] ERR_BNDS_NORM
354*> \verbatim
355*> ERR_BNDS_NORM is REAL array, dimension (NRHS, N_ERR_BNDS)
356*> For each right-hand side, this array contains information about
357*> various error bounds and condition numbers corresponding to the
358*> normwise relative error, which is defined as follows:
359*>
360*> Normwise relative error in the ith solution vector:
361*> max_j (abs(XTRUE(j,i) - X(j,i)))
362*> ------------------------------
363*> max_j abs(X(j,i))
364*>
365*> The array is indexed by the type of error information as described
366*> below. There currently are up to three pieces of information
367*> returned.
368*>
369*> The first index in ERR_BNDS_NORM(i,:) corresponds to the ith
370*> right-hand side.
371*>
372*> The second index in ERR_BNDS_NORM(:,err) contains the following
373*> three fields:
374*> err = 1 "Trust/don't trust" boolean. Trust the answer if the
375*> reciprocal condition number is less than the threshold
376*> sqrt(n) * slamch('Epsilon').
377*>
378*> err = 2 "Guaranteed" error bound: The estimated forward error,
379*> almost certainly within a factor of 10 of the true error
380*> so long as the next entry is greater than the threshold
381*> sqrt(n) * slamch('Epsilon'). This error bound should only
382*> be trusted if the previous boolean is true.
383*>
384*> err = 3 Reciprocal condition number: Estimated normwise
385*> reciprocal condition number. Compared with the threshold
386*> sqrt(n) * slamch('Epsilon') to determine if the error
387*> estimate is "guaranteed". These reciprocal condition
388*> numbers are 1 / (norm(Z^{-1},inf) * norm(Z,inf)) for some
389*> appropriately scaled matrix Z.
390*> Let Z = S*A, where S scales each row by a power of the
391*> radix so all absolute row sums of Z are approximately 1.
392*>
393*> See Lapack Working Note 165 for further details and extra
394*> cautions.
395*> \endverbatim
396*>
397*> \param[out] ERR_BNDS_COMP
398*> \verbatim
399*> ERR_BNDS_COMP is REAL array, dimension (NRHS, N_ERR_BNDS)
400*> For each right-hand side, this array contains information about
401*> various error bounds and condition numbers corresponding to the
402*> componentwise relative error, which is defined as follows:
403*>
404*> Componentwise relative error in the ith solution vector:
405*> abs(XTRUE(j,i) - X(j,i))
406*> max_j ----------------------
407*> abs(X(j,i))
408*>
409*> The array is indexed by the right-hand side i (on which the
410*> componentwise relative error depends), and the type of error
411*> information as described below. There currently are up to three
412*> pieces of information returned for each right-hand side. If
413*> componentwise accuracy is not requested (PARAMS(3) = 0.0), then
414*> ERR_BNDS_COMP is not accessed. If N_ERR_BNDS < 3, then at most
415*> the first (:,N_ERR_BNDS) entries are returned.
416*>
417*> The first index in ERR_BNDS_COMP(i,:) corresponds to the ith
418*> right-hand side.
419*>
420*> The second index in ERR_BNDS_COMP(:,err) contains the following
421*> three fields:
422*> err = 1 "Trust/don't trust" boolean. Trust the answer if the
423*> reciprocal condition number is less than the threshold
424*> sqrt(n) * slamch('Epsilon').
425*>
426*> err = 2 "Guaranteed" error bound: The estimated forward error,
427*> almost certainly within a factor of 10 of the true error
428*> so long as the next entry is greater than the threshold
429*> sqrt(n) * slamch('Epsilon'). This error bound should only
430*> be trusted if the previous boolean is true.
431*>
432*> err = 3 Reciprocal condition number: Estimated componentwise
433*> reciprocal condition number. Compared with the threshold
434*> sqrt(n) * slamch('Epsilon') to determine if the error
435*> estimate is "guaranteed". These reciprocal condition
436*> numbers are 1 / (norm(Z^{-1},inf) * norm(Z,inf)) for some
437*> appropriately scaled matrix Z.
438*> Let Z = S*(A*diag(x)), where x is the solution for the
439*> current right-hand side and S scales each row of
440*> A*diag(x) by a power of the radix so all absolute row
441*> sums of Z are approximately 1.
442*>
443*> See Lapack Working Note 165 for further details and extra
444*> cautions.
445*> \endverbatim
446*>
447*> \param[in] NPARAMS
448*> \verbatim
449*> NPARAMS is INTEGER
450*> Specifies the number of parameters set in PARAMS. If <= 0, the
451*> PARAMS array is never referenced and default values are used.
452*> \endverbatim
453*>
454*> \param[in,out] PARAMS
455*> \verbatim
456*> PARAMS is REAL array, dimension NPARAMS
457*> Specifies algorithm parameters. If an entry is < 0.0, then
458*> that entry will be filled with default value used for that
459*> parameter. Only positions up to NPARAMS are accessed; defaults
460*> are used for higher-numbered parameters.
461*>
462*> PARAMS(LA_LINRX_ITREF_I = 1) : Whether to perform iterative
463*> refinement or not.
464*> Default: 1.0
465*> = 0.0: No refinement is performed, and no error bounds are
466*> computed.
467*> = 1.0: Use the double-precision refinement algorithm,
468*> possibly with doubled-single computations if the
469*> compilation environment does not support DOUBLE
470*> PRECISION.
471*> (other values are reserved for future use)
472*>
473*> PARAMS(LA_LINRX_ITHRESH_I = 2) : Maximum number of residual
474*> computations allowed for refinement.
475*> Default: 10
476*> Aggressive: Set to 100 to permit convergence using approximate
477*> factorizations or factorizations other than LU. If
478*> the factorization uses a technique other than
479*> Gaussian elimination, the guarantees in
480*> err_bnds_norm and err_bnds_comp may no longer be
481*> trustworthy.
482*>
483*> PARAMS(LA_LINRX_CWISE_I = 3) : Flag determining if the code
484*> will attempt to find a solution with small componentwise
485*> relative error in the double-precision algorithm. Positive
486*> is true, 0.0 is false.
487*> Default: 1.0 (attempt componentwise convergence)
488*> \endverbatim
489*>
490*> \param[out] WORK
491*> \verbatim
492*> WORK is REAL array, dimension (4*N)
493*> \endverbatim
494*>
495*> \param[out] IWORK
496*> \verbatim
497*> IWORK is INTEGER array, dimension (N)
498*> \endverbatim
499*>
500*> \param[out] INFO
501*> \verbatim
502*> INFO is INTEGER
503*> = 0: Successful exit. The solution to every right-hand side is
504*> guaranteed.
505*> < 0: If INFO = -i, the i-th argument had an illegal value
506*> > 0 and <= N: U(INFO,INFO) is exactly zero. The factorization
507*> has been completed, but the factor U is exactly singular, so
508*> the solution and error bounds could not be computed. RCOND = 0
509*> is returned.
510*> = N+J: The solution corresponding to the Jth right-hand side is
511*> not guaranteed. The solutions corresponding to other right-
512*> hand sides K with K > J may not be guaranteed as well, but
513*> only the first such right-hand side is reported. If a small
514*> componentwise error is not requested (PARAMS(3) = 0.0) then
515*> the Jth right-hand side is the first with a normwise error
516*> bound that is not guaranteed (the smallest J such
517*> that ERR_BNDS_NORM(J,1) = 0.0). By default (PARAMS(3) = 1.0)
518*> the Jth right-hand side is the first with either a normwise or
519*> componentwise error bound that is not guaranteed (the smallest
520*> J such that either ERR_BNDS_NORM(J,1) = 0.0 or
521*> ERR_BNDS_COMP(J,1) = 0.0). See the definition of
522*> ERR_BNDS_NORM(:,1) and ERR_BNDS_COMP(:,1). To get information
523*> about all of the right-hand sides check ERR_BNDS_NORM or
524*> ERR_BNDS_COMP.
525*> \endverbatim
526*
527* Authors:
528* ========
529*
530*> \author Univ. of Tennessee
531*> \author Univ. of California Berkeley
532*> \author Univ. of Colorado Denver
533*> \author NAG Ltd.
534*
535*> \ingroup gesvxx
536*
537* =====================================================================
538 SUBROUTINE sgesvxx( FACT, TRANS, N, NRHS, A, LDA, AF, LDAF, IPIV,
539 $ EQUED, R, C, B, LDB, X, LDX, RCOND, RPVGRW,
540 $ BERR, N_ERR_BNDS, ERR_BNDS_NORM,
541 $ ERR_BNDS_COMP, NPARAMS, PARAMS, WORK, IWORK,
542 $ INFO )
543*
544* -- LAPACK driver routine --
545* -- LAPACK is a software package provided by Univ. of Tennessee, --
546* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
547*
548* .. Scalar Arguments ..
549 CHARACTER EQUED, FACT, TRANS
550 INTEGER INFO, LDA, LDAF, LDB, LDX, N, NRHS, NPARAMS,
551 $ N_ERR_BNDS
552 REAL RCOND, RPVGRW
553* ..
554* .. Array Arguments ..
555 INTEGER IPIV( * ), IWORK( * )
556 REAL A( LDA, * ), AF( LDAF, * ), B( LDB, * ),
557 $ X( LDX , * ),WORK( * )
558 REAL R( * ), C( * ), PARAMS( * ), BERR( * ),
559 $ err_bnds_norm( nrhs, * ),
560 $ err_bnds_comp( nrhs, * )
561* ..
562*
563* ==================================================================
564*
565* .. Parameters ..
566 REAL ZERO, ONE
567 PARAMETER ( ZERO = 0.0e+0, one = 1.0e+0 )
568 INTEGER FINAL_NRM_ERR_I, FINAL_CMP_ERR_I, BERR_I
569 INTEGER RCOND_I, NRM_RCOND_I, NRM_ERR_I, CMP_RCOND_I
570 INTEGER CMP_ERR_I, PIV_GROWTH_I
571 parameter( final_nrm_err_i = 1, final_cmp_err_i = 2,
572 $ berr_i = 3 )
573 parameter( rcond_i = 4, nrm_rcond_i = 5, nrm_err_i = 6 )
574 parameter( cmp_rcond_i = 7, cmp_err_i = 8,
575 $ piv_growth_i = 9 )
576* ..
577* .. Local Scalars ..
578 LOGICAL COLEQU, EQUIL, NOFACT, NOTRAN, ROWEQU
579 INTEGER INFEQU, J
580 REAL AMAX, BIGNUM, COLCND, RCMAX, RCMIN, ROWCND,
581 $ SMLNUM
582* ..
583* .. External Functions ..
584 EXTERNAL lsame, slamch, sla_gerpvgrw
585 LOGICAL LSAME
586 REAL SLAMCH, SLA_GERPVGRW
587* ..
588* .. External Subroutines ..
589 EXTERNAL sgeequb, sgetrf, sgetrs, slacpy, slaqge,
591* ..
592* .. Intrinsic Functions ..
593 INTRINSIC max, min
594* ..
595* .. Executable Statements ..
596*
597 info = 0
598 nofact = lsame( fact, 'N' )
599 equil = lsame( fact, 'E' )
600 notran = lsame( trans, 'N' )
601 smlnum = slamch( 'Safe minimum' )
602 bignum = one / smlnum
603 IF( nofact .OR. equil ) THEN
604 equed = 'N'
605 rowequ = .false.
606 colequ = .false.
607 ELSE
608 rowequ = lsame( equed, 'R' ) .OR. lsame( equed, 'B' )
609 colequ = lsame( equed, 'C' ) .OR. lsame( equed, 'B' )
610 END IF
611*
612* Default is failure. If an input parameter is wrong or
613* factorization fails, make everything look horrible. Only the
614* pivot growth is set here, the rest is initialized in SGERFSX.
615*
616 rpvgrw = zero
617*
618* Test the input parameters. PARAMS is not tested until SGERFSX.
619*
620 IF( .NOT.nofact .AND. .NOT.equil .AND. .NOT.
621 $ lsame( fact, 'F' ) ) THEN
622 info = -1
623 ELSE IF( .NOT.notran .AND. .NOT.lsame( trans, 'T' ) .AND. .NOT.
624 $ lsame( trans, 'C' ) ) THEN
625 info = -2
626 ELSE IF( n.LT.0 ) THEN
627 info = -3
628 ELSE IF( nrhs.LT.0 ) THEN
629 info = -4
630 ELSE IF( lda.LT.max( 1, n ) ) THEN
631 info = -6
632 ELSE IF( ldaf.LT.max( 1, n ) ) THEN
633 info = -8
634 ELSE IF( lsame( fact, 'F' ) .AND. .NOT.
635 $ ( rowequ .OR. colequ .OR. lsame( equed, 'N' ) ) ) THEN
636 info = -10
637 ELSE
638 IF( rowequ ) THEN
639 rcmin = bignum
640 rcmax = zero
641 DO 10 j = 1, n
642 rcmin = min( rcmin, r( j ) )
643 rcmax = max( rcmax, r( j ) )
644 10 CONTINUE
645 IF( rcmin.LE.zero ) THEN
646 info = -11
647 ELSE IF( n.GT.0 ) THEN
648 rowcnd = max( rcmin, smlnum ) / min( rcmax, bignum )
649 ELSE
650 rowcnd = one
651 END IF
652 END IF
653 IF( colequ .AND. info.EQ.0 ) THEN
654 rcmin = bignum
655 rcmax = zero
656 DO 20 j = 1, n
657 rcmin = min( rcmin, c( j ) )
658 rcmax = max( rcmax, c( j ) )
659 20 CONTINUE
660 IF( rcmin.LE.zero ) THEN
661 info = -12
662 ELSE IF( n.GT.0 ) THEN
663 colcnd = max( rcmin, smlnum ) / min( rcmax, bignum )
664 ELSE
665 colcnd = one
666 END IF
667 END IF
668 IF( info.EQ.0 ) THEN
669 IF( ldb.LT.max( 1, n ) ) THEN
670 info = -14
671 ELSE IF( ldx.LT.max( 1, n ) ) THEN
672 info = -16
673 END IF
674 END IF
675 END IF
676*
677 IF( info.NE.0 ) THEN
678 CALL xerbla( 'SGESVXX', -info )
679 RETURN
680 END IF
681*
682 IF( equil ) THEN
683*
684* Compute row and column scalings to equilibrate the matrix A.
685*
686 CALL sgeequb( n, n, a, lda, r, c, rowcnd, colcnd, amax,
687 $ infequ )
688 IF( infequ.EQ.0 ) THEN
689*
690* Equilibrate the matrix.
691*
692 CALL slaqge( n, n, a, lda, r, c, rowcnd, colcnd, amax,
693 $ equed )
694 rowequ = lsame( equed, 'R' ) .OR. lsame( equed, 'B' )
695 colequ = lsame( equed, 'C' ) .OR. lsame( equed, 'B' )
696 END IF
697*
698* If the scaling factors are not applied, set them to 1.0.
699*
700 IF ( .NOT.rowequ ) THEN
701 DO j = 1, n
702 r( j ) = 1.0
703 END DO
704 END IF
705 IF ( .NOT.colequ ) THEN
706 DO j = 1, n
707 c( j ) = 1.0
708 END DO
709 END IF
710 END IF
711*
712* Scale the right-hand side.
713*
714 IF( notran ) THEN
715 IF( rowequ ) CALL slascl2( n, nrhs, r, b, ldb )
716 ELSE
717 IF( colequ ) CALL slascl2( n, nrhs, c, b, ldb )
718 END IF
719*
720 IF( nofact .OR. equil ) THEN
721*
722* Compute the LU factorization of A.
723*
724 CALL slacpy( 'Full', n, n, a, lda, af, ldaf )
725 CALL sgetrf( n, n, af, ldaf, ipiv, info )
726*
727* Return if INFO is non-zero.
728*
729 IF( info.GT.0 ) THEN
730*
731* Pivot in column INFO is exactly 0
732* Compute the reciprocal pivot growth factor of the
733* leading rank-deficient INFO columns of A.
734*
735 rpvgrw = sla_gerpvgrw( n, info, a, lda, af, ldaf )
736 RETURN
737 END IF
738 END IF
739*
740* Compute the reciprocal pivot growth factor RPVGRW.
741*
742 rpvgrw = sla_gerpvgrw( n, n, a, lda, af, ldaf )
743*
744* Compute the solution matrix X.
745*
746 CALL slacpy( 'Full', n, nrhs, b, ldb, x, ldx )
747 CALL sgetrs( trans, n, nrhs, af, ldaf, ipiv, x, ldx, info )
748*
749* Use iterative refinement to improve the computed solution and
750* compute error bounds and backward error estimates for it.
751*
752 CALL sgerfsx( trans, equed, n, nrhs, a, lda, af, ldaf,
753 $ ipiv, r, c, b, ldb, x, ldx, rcond, berr,
754 $ n_err_bnds, err_bnds_norm, err_bnds_comp, nparams, params,
755 $ work, iwork, info )
756*
757* Scale solutions.
758*
759 IF ( colequ .AND. notran ) THEN
760 CALL slascl2 ( n, nrhs, c, x, ldx )
761 ELSE IF ( rowequ .AND. .NOT.notran ) THEN
762 CALL slascl2 ( n, nrhs, r, x, ldx )
763 END IF
764*
765 RETURN
766*
767* End of SGESVXX
768
769 END
subroutine xerbla(srname, info)
Definition cblat2.f:3285
subroutine sgeequb(m, n, a, lda, r, c, rowcnd, colcnd, amax, info)
SGEEQUB
Definition sgeequb.f:146
subroutine sgerfsx(trans, equed, n, nrhs, a, lda, af, ldaf, ipiv, r, c, b, ldb, x, ldx, rcond, berr, n_err_bnds, err_bnds_norm, err_bnds_comp, nparams, params, work, iwork, info)
SGERFSX
Definition sgerfsx.f:414
subroutine sgesvxx(fact, trans, n, nrhs, a, lda, af, ldaf, ipiv, equed, r, c, b, ldb, x, ldx, rcond, rpvgrw, berr, n_err_bnds, err_bnds_norm, err_bnds_comp, nparams, params, work, iwork, info)
SGESVXX computes the solution to system of linear equations A * X = B for GE matrices
Definition sgesvxx.f:543
subroutine sgetrf(m, n, a, lda, ipiv, info)
SGETRF
Definition sgetrf.f:108
subroutine sgetrs(trans, n, nrhs, a, lda, ipiv, b, ldb, info)
SGETRS
Definition sgetrs.f:121
real function sla_gerpvgrw(n, ncols, a, lda, af, ldaf)
SLA_GERPVGRW
subroutine slacpy(uplo, m, n, a, lda, b, ldb)
SLACPY copies all or part of one two-dimensional array to another.
Definition slacpy.f:103
real function slamch(cmach)
SLAMCH
Definition slamch.f:68
subroutine slaqge(m, n, a, lda, r, c, rowcnd, colcnd, amax, equed)
SLAQGE scales a general rectangular matrix, using row and column scaling factors computed by sgeequ.
Definition slaqge.f:142
subroutine slascl2(m, n, d, x, ldx)
SLASCL2 performs diagonal scaling on a matrix.
Definition slascl2.f:90
logical function lsame(ca, cb)
LSAME
Definition lsame.f:48