1: *> \brief \b DLA_PORFSX_EXTENDED improves the computed solution to a system of linear equations for symmetric or Hermitian positive-definite matrices by performing extra-precise iterative refinement and provides error bounds and backward error estimates for the solution.
2: *
3: * =========== DOCUMENTATION ===========
4: *
5: * Online html documentation available at
6: * http://www.netlib.org/lapack/explore-html/
7: *
8: *> \htmlonly
9: *> Download DLA_PORFSX_EXTENDED + dependencies
10: *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dla_porfsx_extended.f">
11: *> [TGZ]</a>
12: *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dla_porfsx_extended.f">
13: *> [ZIP]</a>
14: *> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dla_porfsx_extended.f">
15: *> [TXT]</a>
16: *> \endhtmlonly
17: *
18: * Definition:
19: * ===========
20: *
21: * SUBROUTINE DLA_PORFSX_EXTENDED( PREC_TYPE, UPLO, N, NRHS, A, LDA,
22: * AF, LDAF, COLEQU, C, B, LDB, Y,
23: * LDY, BERR_OUT, N_NORMS,
24: * ERR_BNDS_NORM, ERR_BNDS_COMP, RES,
25: * AYB, DY, Y_TAIL, RCOND, ITHRESH,
26: * RTHRESH, DZ_UB, IGNORE_CWISE,
27: * INFO )
28: *
29: * .. Scalar Arguments ..
30: * INTEGER INFO, LDA, LDAF, LDB, LDY, N, NRHS, PREC_TYPE,
31: * $ N_NORMS, ITHRESH
32: * CHARACTER UPLO
33: * LOGICAL COLEQU, IGNORE_CWISE
34: * DOUBLE PRECISION RTHRESH, DZ_UB
35: * ..
36: * .. Array Arguments ..
37: * DOUBLE PRECISION A( LDA, * ), AF( LDAF, * ), B( LDB, * ),
38: * $ Y( LDY, * ), RES( * ), DY( * ), Y_TAIL( * )
39: * DOUBLE PRECISION C( * ), AYB(*), RCOND, BERR_OUT( * ),
40: * $ ERR_BNDS_NORM( NRHS, * ),
41: * $ ERR_BNDS_COMP( NRHS, * )
42: * ..
43: *
44: *
45: *> \par Purpose:
46: * =============
47: *>
48: *> \verbatim
49: *>
50: *> DLA_PORFSX_EXTENDED improves the computed solution to a system of
51: *> linear equations by performing extra-precise iterative refinement
52: *> and provides error bounds and backward error estimates for the solution.
53: *> This subroutine is called by DPORFSX to perform iterative refinement.
54: *> In addition to normwise error bound, the code provides maximum
55: *> componentwise error bound if possible. See comments for ERR_BNDS_NORM
56: *> and ERR_BNDS_COMP for details of the error bounds. Note that this
57: *> subroutine is only responsible for setting the second fields of
58: *> ERR_BNDS_NORM and ERR_BNDS_COMP.
59: *> \endverbatim
60: *
61: * Arguments:
62: * ==========
63: *
64: *> \param[in] PREC_TYPE
65: *> \verbatim
66: *> PREC_TYPE is INTEGER
67: *> Specifies the intermediate precision to be used in refinement.
68: *> The value is defined by ILAPREC(P) where P is a CHARACTER and P
69: *> = 'S': Single
70: *> = 'D': Double
71: *> = 'I': Indigenous
72: *> = 'X' or 'E': Extra
73: *> \endverbatim
74: *>
75: *> \param[in] UPLO
76: *> \verbatim
77: *> UPLO is CHARACTER*1
78: *> = 'U': Upper triangle of A is stored;
79: *> = 'L': Lower triangle of A is stored.
80: *> \endverbatim
81: *>
82: *> \param[in] N
83: *> \verbatim
84: *> N is INTEGER
85: *> The number of linear equations, i.e., the order of the
86: *> matrix A. N >= 0.
87: *> \endverbatim
88: *>
89: *> \param[in] NRHS
90: *> \verbatim
91: *> NRHS is INTEGER
92: *> The number of right-hand-sides, i.e., the number of columns of the
93: *> matrix B.
94: *> \endverbatim
95: *>
96: *> \param[in] A
97: *> \verbatim
98: *> A is DOUBLE PRECISION array, dimension (LDA,N)
99: *> On entry, the N-by-N matrix A.
100: *> \endverbatim
101: *>
102: *> \param[in] LDA
103: *> \verbatim
104: *> LDA is INTEGER
105: *> The leading dimension of the array A. LDA >= max(1,N).
106: *> \endverbatim
107: *>
108: *> \param[in] AF
109: *> \verbatim
110: *> AF is DOUBLE PRECISION array, dimension (LDAF,N)
111: *> The triangular factor U or L from the Cholesky factorization
112: *> A = U**T*U or A = L*L**T, as computed by DPOTRF.
113: *> \endverbatim
114: *>
115: *> \param[in] LDAF
116: *> \verbatim
117: *> LDAF is INTEGER
118: *> The leading dimension of the array AF. LDAF >= max(1,N).
119: *> \endverbatim
120: *>
121: *> \param[in] COLEQU
122: *> \verbatim
123: *> COLEQU is LOGICAL
124: *> If .TRUE. then column equilibration was done to A before calling
125: *> this routine. This is needed to compute the solution and error
126: *> bounds correctly.
127: *> \endverbatim
128: *>
129: *> \param[in] C
130: *> \verbatim
131: *> C is DOUBLE PRECISION array, dimension (N)
132: *> The column scale factors for A. If COLEQU = .FALSE., C
133: *> is not accessed. If C is input, each element of C should be a power
134: *> of the radix to ensure a reliable solution and error estimates.
135: *> Scaling by powers of the radix does not cause rounding errors unless
136: *> the result underflows or overflows. Rounding errors during scaling
137: *> lead to refining with a matrix that is not equivalent to the
138: *> input matrix, producing error estimates that may not be
139: *> reliable.
140: *> \endverbatim
141: *>
142: *> \param[in] B
143: *> \verbatim
144: *> B is DOUBLE PRECISION array, dimension (LDB,NRHS)
145: *> The right-hand-side matrix B.
146: *> \endverbatim
147: *>
148: *> \param[in] LDB
149: *> \verbatim
150: *> LDB is INTEGER
151: *> The leading dimension of the array B. LDB >= max(1,N).
152: *> \endverbatim
153: *>
154: *> \param[in,out] Y
155: *> \verbatim
156: *> Y is DOUBLE PRECISION array, dimension (LDY,NRHS)
157: *> On entry, the solution matrix X, as computed by DPOTRS.
158: *> On exit, the improved solution matrix Y.
159: *> \endverbatim
160: *>
161: *> \param[in] LDY
162: *> \verbatim
163: *> LDY is INTEGER
164: *> The leading dimension of the array Y. LDY >= max(1,N).
165: *> \endverbatim
166: *>
167: *> \param[out] BERR_OUT
168: *> \verbatim
169: *> BERR_OUT is DOUBLE PRECISION array, dimension (NRHS)
170: *> On exit, BERR_OUT(j) contains the componentwise relative backward
171: *> error for right-hand-side j from the formula
172: *> max(i) ( abs(RES(i)) / ( abs(op(A_s))*abs(Y) + abs(B_s) )(i) )
173: *> where abs(Z) is the componentwise absolute value of the matrix
174: *> or vector Z. This is computed by DLA_LIN_BERR.
175: *> \endverbatim
176: *>
177: *> \param[in] N_NORMS
178: *> \verbatim
179: *> N_NORMS is INTEGER
180: *> Determines which error bounds to return (see ERR_BNDS_NORM
181: *> and ERR_BNDS_COMP).
182: *> If N_NORMS >= 1 return normwise error bounds.
183: *> If N_NORMS >= 2 return componentwise error bounds.
184: *> \endverbatim
185: *>
186: *> \param[in,out] ERR_BNDS_NORM
187: *> \verbatim
188: *> ERR_BNDS_NORM is DOUBLE PRECISION array, dimension (NRHS, N_ERR_BNDS)
189: *> For each right-hand side, this array contains information about
190: *> various error bounds and condition numbers corresponding to the
191: *> normwise relative error, which is defined as follows:
192: *>
193: *> Normwise relative error in the ith solution vector:
194: *> max_j (abs(XTRUE(j,i) - X(j,i)))
195: *> ------------------------------
196: *> max_j abs(X(j,i))
197: *>
198: *> The array is indexed by the type of error information as described
199: *> below. There currently are up to three pieces of information
200: *> returned.
201: *>
202: *> The first index in ERR_BNDS_NORM(i,:) corresponds to the ith
203: *> right-hand side.
204: *>
205: *> The second index in ERR_BNDS_NORM(:,err) contains the following
206: *> three fields:
207: *> err = 1 "Trust/don't trust" boolean. Trust the answer if the
208: *> reciprocal condition number is less than the threshold
209: *> sqrt(n) * slamch('Epsilon').
210: *>
211: *> err = 2 "Guaranteed" error bound: The estimated forward error,
212: *> almost certainly within a factor of 10 of the true error
213: *> so long as the next entry is greater than the threshold
214: *> sqrt(n) * slamch('Epsilon'). This error bound should only
215: *> be trusted if the previous boolean is true.
216: *>
217: *> err = 3 Reciprocal condition number: Estimated normwise
218: *> reciprocal condition number. Compared with the threshold
219: *> sqrt(n) * slamch('Epsilon') to determine if the error
220: *> estimate is "guaranteed". These reciprocal condition
221: *> numbers are 1 / (norm(Z^{-1},inf) * norm(Z,inf)) for some
222: *> appropriately scaled matrix Z.
223: *> Let Z = S*A, where S scales each row by a power of the
224: *> radix so all absolute row sums of Z are approximately 1.
225: *>
226: *> This subroutine is only responsible for setting the second field
227: *> above.
228: *> See Lapack Working Note 165 for further details and extra
229: *> cautions.
230: *> \endverbatim
231: *>
232: *> \param[in,out] ERR_BNDS_COMP
233: *> \verbatim
234: *> ERR_BNDS_COMP is DOUBLE PRECISION array, dimension (NRHS, N_ERR_BNDS)
235: *> For each right-hand side, this array contains information about
236: *> various error bounds and condition numbers corresponding to the
237: *> componentwise relative error, which is defined as follows:
238: *>
239: *> Componentwise relative error in the ith solution vector:
240: *> abs(XTRUE(j,i) - X(j,i))
241: *> max_j ----------------------
242: *> abs(X(j,i))
243: *>
244: *> The array is indexed by the right-hand side i (on which the
245: *> componentwise relative error depends), and the type of error
246: *> information as described below. There currently are up to three
247: *> pieces of information returned for each right-hand side. If
248: *> componentwise accuracy is not requested (PARAMS(3) = 0.0), then
249: *> ERR_BNDS_COMP is not accessed. If N_ERR_BNDS < 3, then at most
250: *> the first (:,N_ERR_BNDS) entries are returned.
251: *>
252: *> The first index in ERR_BNDS_COMP(i,:) corresponds to the ith
253: *> right-hand side.
254: *>
255: *> The second index in ERR_BNDS_COMP(:,err) contains the following
256: *> three fields:
257: *> err = 1 "Trust/don't trust" boolean. Trust the answer if the
258: *> reciprocal condition number is less than the threshold
259: *> sqrt(n) * slamch('Epsilon').
260: *>
261: *> err = 2 "Guaranteed" error bound: The estimated forward error,
262: *> almost certainly within a factor of 10 of the true error
263: *> so long as the next entry is greater than the threshold
264: *> sqrt(n) * slamch('Epsilon'). This error bound should only
265: *> be trusted if the previous boolean is true.
266: *>
267: *> err = 3 Reciprocal condition number: Estimated componentwise
268: *> reciprocal condition number. Compared with the threshold
269: *> sqrt(n) * slamch('Epsilon') to determine if the error
270: *> estimate is "guaranteed". These reciprocal condition
271: *> numbers are 1 / (norm(Z^{-1},inf) * norm(Z,inf)) for some
272: *> appropriately scaled matrix Z.
273: *> Let Z = S*(A*diag(x)), where x is the solution for the
274: *> current right-hand side and S scales each row of
275: *> A*diag(x) by a power of the radix so all absolute row
276: *> sums of Z are approximately 1.
277: *>
278: *> This subroutine is only responsible for setting the second field
279: *> above.
280: *> See Lapack Working Note 165 for further details and extra
281: *> cautions.
282: *> \endverbatim
283: *>
284: *> \param[in] RES
285: *> \verbatim
286: *> RES is DOUBLE PRECISION array, dimension (N)
287: *> Workspace to hold the intermediate residual.
288: *> \endverbatim
289: *>
290: *> \param[in] AYB
291: *> \verbatim
292: *> AYB is DOUBLE PRECISION array, dimension (N)
293: *> Workspace. This can be the same workspace passed for Y_TAIL.
294: *> \endverbatim
295: *>
296: *> \param[in] DY
297: *> \verbatim
298: *> DY is DOUBLE PRECISION array, dimension (N)
299: *> Workspace to hold the intermediate solution.
300: *> \endverbatim
301: *>
302: *> \param[in] Y_TAIL
303: *> \verbatim
304: *> Y_TAIL is DOUBLE PRECISION array, dimension (N)
305: *> Workspace to hold the trailing bits of the intermediate solution.
306: *> \endverbatim
307: *>
308: *> \param[in] RCOND
309: *> \verbatim
310: *> RCOND is DOUBLE PRECISION
311: *> Reciprocal scaled condition number. This is an estimate of the
312: *> reciprocal Skeel condition number of the matrix A after
313: *> equilibration (if done). If this is less than the machine
314: *> precision (in particular, if it is zero), the matrix is singular
315: *> to working precision. Note that the error may still be small even
316: *> if this number is very small and the matrix appears ill-
317: *> conditioned.
318: *> \endverbatim
319: *>
320: *> \param[in] ITHRESH
321: *> \verbatim
322: *> ITHRESH is INTEGER
323: *> The maximum number of residual computations allowed for
324: *> refinement. The default is 10. For 'aggressive' set to 100 to
325: *> permit convergence using approximate factorizations or
326: *> factorizations other than LU. If the factorization uses a
327: *> technique other than Gaussian elimination, the guarantees in
328: *> ERR_BNDS_NORM and ERR_BNDS_COMP may no longer be trustworthy.
329: *> \endverbatim
330: *>
331: *> \param[in] RTHRESH
332: *> \verbatim
333: *> RTHRESH is DOUBLE PRECISION
334: *> Determines when to stop refinement if the error estimate stops
335: *> decreasing. Refinement will stop when the next solution no longer
336: *> satisfies norm(dx_{i+1}) < RTHRESH * norm(dx_i) where norm(Z) is
337: *> the infinity norm of Z. RTHRESH satisfies 0 < RTHRESH <= 1. The
338: *> default value is 0.5. For 'aggressive' set to 0.9 to permit
339: *> convergence on extremely ill-conditioned matrices. See LAWN 165
340: *> for more details.
341: *> \endverbatim
342: *>
343: *> \param[in] DZ_UB
344: *> \verbatim
345: *> DZ_UB is DOUBLE PRECISION
346: *> Determines when to start considering componentwise convergence.
347: *> Componentwise convergence is only considered after each component
348: *> of the solution Y is stable, which we define as the relative
349: *> change in each component being less than DZ_UB. The default value
350: *> is 0.25, requiring the first bit to be stable. See LAWN 165 for
351: *> more details.
352: *> \endverbatim
353: *>
354: *> \param[in] IGNORE_CWISE
355: *> \verbatim
356: *> IGNORE_CWISE is LOGICAL
357: *> If .TRUE. then ignore componentwise convergence. Default value
358: *> is .FALSE..
359: *> \endverbatim
360: *>
361: *> \param[out] INFO
362: *> \verbatim
363: *> INFO is INTEGER
364: *> = 0: Successful exit.
365: *> < 0: if INFO = -i, the ith argument to DPOTRS had an illegal
366: *> value
367: *> \endverbatim
368: *
369: * Authors:
370: * ========
371: *
372: *> \author Univ. of Tennessee
373: *> \author Univ. of California Berkeley
374: *> \author Univ. of Colorado Denver
375: *> \author NAG Ltd.
376: *
377: *> \ingroup doublePOcomputational
378: *
379: * =====================================================================
380: SUBROUTINE DLA_PORFSX_EXTENDED( PREC_TYPE, UPLO, N, NRHS, A, LDA,
381: $ AF, LDAF, COLEQU, C, B, LDB, Y,
382: $ LDY, BERR_OUT, N_NORMS,
383: $ ERR_BNDS_NORM, ERR_BNDS_COMP, RES,
384: $ AYB, DY, Y_TAIL, RCOND, ITHRESH,
385: $ RTHRESH, DZ_UB, IGNORE_CWISE,
386: $ INFO )
387: *
388: * -- LAPACK computational routine --
389: * -- LAPACK is a software package provided by Univ. of Tennessee, --
390: * -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
391: *
392: * .. Scalar Arguments ..
393: INTEGER INFO, LDA, LDAF, LDB, LDY, N, NRHS, PREC_TYPE,
394: $ N_NORMS, ITHRESH
395: CHARACTER UPLO
396: LOGICAL COLEQU, IGNORE_CWISE
397: DOUBLE PRECISION RTHRESH, DZ_UB
398: * ..
399: * .. Array Arguments ..
400: DOUBLE PRECISION A( LDA, * ), AF( LDAF, * ), B( LDB, * ),
401: $ Y( LDY, * ), RES( * ), DY( * ), Y_TAIL( * )
402: DOUBLE PRECISION C( * ), AYB(*), RCOND, BERR_OUT( * ),
403: $ ERR_BNDS_NORM( NRHS, * ),
404: $ ERR_BNDS_COMP( NRHS, * )
405: * ..
406: *
407: * =====================================================================
408: *
409: * .. Local Scalars ..
410: INTEGER UPLO2, CNT, I, J, X_STATE, Z_STATE
411: DOUBLE PRECISION YK, DYK, YMIN, NORMY, NORMX, NORMDX, DXRAT,
412: $ DZRAT, PREVNORMDX, PREV_DZ_Z, DXRATMAX,
413: $ DZRATMAX, DX_X, DZ_Z, FINAL_DX_X, FINAL_DZ_Z,
414: $ EPS, HUGEVAL, INCR_THRESH
415: LOGICAL INCR_PREC
416: * ..
417: * .. Parameters ..
418: INTEGER UNSTABLE_STATE, WORKING_STATE, CONV_STATE,
419: $ NOPROG_STATE, Y_PREC_STATE, BASE_RESIDUAL,
420: $ EXTRA_RESIDUAL, EXTRA_Y
421: PARAMETER ( UNSTABLE_STATE = 0, WORKING_STATE = 1,
422: $ CONV_STATE = 2, NOPROG_STATE = 3 )
423: PARAMETER ( BASE_RESIDUAL = 0, EXTRA_RESIDUAL = 1,
424: $ EXTRA_Y = 2 )
425: INTEGER FINAL_NRM_ERR_I, FINAL_CMP_ERR_I, BERR_I
426: INTEGER RCOND_I, NRM_RCOND_I, NRM_ERR_I, CMP_RCOND_I
427: INTEGER CMP_ERR_I, PIV_GROWTH_I
428: PARAMETER ( FINAL_NRM_ERR_I = 1, FINAL_CMP_ERR_I = 2,
429: $ BERR_I = 3 )
430: PARAMETER ( RCOND_I = 4, NRM_RCOND_I = 5, NRM_ERR_I = 6 )
431: PARAMETER ( CMP_RCOND_I = 7, CMP_ERR_I = 8,
432: $ PIV_GROWTH_I = 9 )
433: INTEGER LA_LINRX_ITREF_I, LA_LINRX_ITHRESH_I,
434: $ LA_LINRX_CWISE_I
435: PARAMETER ( LA_LINRX_ITREF_I = 1,
436: $ LA_LINRX_ITHRESH_I = 2 )
437: PARAMETER ( LA_LINRX_CWISE_I = 3 )
438: INTEGER LA_LINRX_TRUST_I, LA_LINRX_ERR_I,
439: $ LA_LINRX_RCOND_I
440: PARAMETER ( LA_LINRX_TRUST_I = 1, LA_LINRX_ERR_I = 2 )
441: PARAMETER ( LA_LINRX_RCOND_I = 3 )
442: * ..
443: * .. External Functions ..
444: LOGICAL LSAME
445: EXTERNAL ILAUPLO
446: INTEGER ILAUPLO
447: * ..
448: * .. External Subroutines ..
449: EXTERNAL DAXPY, DCOPY, DPOTRS, DSYMV, BLAS_DSYMV_X,
450: $ BLAS_DSYMV2_X, DLA_SYAMV, DLA_WWADDW,
451: $ DLA_LIN_BERR
452: DOUBLE PRECISION DLAMCH
453: * ..
454: * .. Intrinsic Functions ..
455: INTRINSIC ABS, MAX, MIN
456: * ..
457: * .. Executable Statements ..
458: *
459: IF (INFO.NE.0) RETURN
460: EPS = DLAMCH( 'Epsilon' )
461: HUGEVAL = DLAMCH( 'Overflow' )
462: * Force HUGEVAL to Inf
463: HUGEVAL = HUGEVAL * HUGEVAL
464: * Using HUGEVAL may lead to spurious underflows.
465: INCR_THRESH = DBLE( N ) * EPS
466:
467: IF ( LSAME ( UPLO, 'L' ) ) THEN
468: UPLO2 = ILAUPLO( 'L' )
469: ELSE
470: UPLO2 = ILAUPLO( 'U' )
471: ENDIF
472:
473: DO J = 1, NRHS
474: Y_PREC_STATE = EXTRA_RESIDUAL
475: IF ( Y_PREC_STATE .EQ. EXTRA_Y ) THEN
476: DO I = 1, N
477: Y_TAIL( I ) = 0.0D+0
478: END DO
479: END IF
480:
481: DXRAT = 0.0D+0
482: DXRATMAX = 0.0D+0
483: DZRAT = 0.0D+0
484: DZRATMAX = 0.0D+0
485: FINAL_DX_X = HUGEVAL
486: FINAL_DZ_Z = HUGEVAL
487: PREVNORMDX = HUGEVAL
488: PREV_DZ_Z = HUGEVAL
489: DZ_Z = HUGEVAL
490: DX_X = HUGEVAL
491:
492: X_STATE = WORKING_STATE
493: Z_STATE = UNSTABLE_STATE
494: INCR_PREC = .FALSE.
495:
496: DO CNT = 1, ITHRESH
497: *
498: * Compute residual RES = B_s - op(A_s) * Y,
499: * op(A) = A, A**T, or A**H depending on TRANS (and type).
500: *
501: CALL DCOPY( N, B( 1, J ), 1, RES, 1 )
502: IF ( Y_PREC_STATE .EQ. BASE_RESIDUAL ) THEN
503: CALL DSYMV( UPLO, N, -1.0D+0, A, LDA, Y(1,J), 1,
504: $ 1.0D+0, RES, 1 )
505: ELSE IF ( Y_PREC_STATE .EQ. EXTRA_RESIDUAL ) THEN
506: CALL BLAS_DSYMV_X( UPLO2, N, -1.0D+0, A, LDA,
507: $ Y( 1, J ), 1, 1.0D+0, RES, 1, PREC_TYPE )
508: ELSE
509: CALL BLAS_DSYMV2_X(UPLO2, N, -1.0D+0, A, LDA,
510: $ Y(1, J), Y_TAIL, 1, 1.0D+0, RES, 1, PREC_TYPE)
511: END IF
512:
513: ! XXX: RES is no longer needed.
514: CALL DCOPY( N, RES, 1, DY, 1 )
515: CALL DPOTRS( UPLO, N, 1, AF, LDAF, DY, N, INFO )
516: *
517: * Calculate relative changes DX_X, DZ_Z and ratios DXRAT, DZRAT.
518: *
519: NORMX = 0.0D+0
520: NORMY = 0.0D+0
521: NORMDX = 0.0D+0
522: DZ_Z = 0.0D+0
523: YMIN = HUGEVAL
524:
525: DO I = 1, N
526: YK = ABS( Y( I, J ) )
527: DYK = ABS( DY( I ) )
528:
529: IF ( YK .NE. 0.0D+0 ) THEN
530: DZ_Z = MAX( DZ_Z, DYK / YK )
531: ELSE IF ( DYK .NE. 0.0D+0 ) THEN
532: DZ_Z = HUGEVAL
533: END IF
534:
535: YMIN = MIN( YMIN, YK )
536:
537: NORMY = MAX( NORMY, YK )
538:
539: IF ( COLEQU ) THEN
540: NORMX = MAX( NORMX, YK * C( I ) )
541: NORMDX = MAX( NORMDX, DYK * C( I ) )
542: ELSE
543: NORMX = NORMY
544: NORMDX = MAX( NORMDX, DYK )
545: END IF
546: END DO
547:
548: IF ( NORMX .NE. 0.0D+0 ) THEN
549: DX_X = NORMDX / NORMX
550: ELSE IF ( NORMDX .EQ. 0.0D+0 ) THEN
551: DX_X = 0.0D+0
552: ELSE
553: DX_X = HUGEVAL
554: END IF
555:
556: DXRAT = NORMDX / PREVNORMDX
557: DZRAT = DZ_Z / PREV_DZ_Z
558: *
559: * Check termination criteria.
560: *
561: IF ( YMIN*RCOND .LT. INCR_THRESH*NORMY
562: $ .AND. Y_PREC_STATE .LT. EXTRA_Y )
563: $ INCR_PREC = .TRUE.
564:
565: IF ( X_STATE .EQ. NOPROG_STATE .AND. DXRAT .LE. RTHRESH )
566: $ X_STATE = WORKING_STATE
567: IF ( X_STATE .EQ. WORKING_STATE ) THEN
568: IF ( DX_X .LE. EPS ) THEN
569: X_STATE = CONV_STATE
570: ELSE IF ( DXRAT .GT. RTHRESH ) THEN
571: IF ( Y_PREC_STATE .NE. EXTRA_Y ) THEN
572: INCR_PREC = .TRUE.
573: ELSE
574: X_STATE = NOPROG_STATE
575: END IF
576: ELSE
577: IF ( DXRAT .GT. DXRATMAX ) DXRATMAX = DXRAT
578: END IF
579: IF ( X_STATE .GT. WORKING_STATE ) FINAL_DX_X = DX_X
580: END IF
581:
582: IF ( Z_STATE .EQ. UNSTABLE_STATE .AND. DZ_Z .LE. DZ_UB )
583: $ Z_STATE = WORKING_STATE
584: IF ( Z_STATE .EQ. NOPROG_STATE .AND. DZRAT .LE. RTHRESH )
585: $ Z_STATE = WORKING_STATE
586: IF ( Z_STATE .EQ. WORKING_STATE ) THEN
587: IF ( DZ_Z .LE. EPS ) THEN
588: Z_STATE = CONV_STATE
589: ELSE IF ( DZ_Z .GT. DZ_UB ) THEN
590: Z_STATE = UNSTABLE_STATE
591: DZRATMAX = 0.0D+0
592: FINAL_DZ_Z = HUGEVAL
593: ELSE IF ( DZRAT .GT. RTHRESH ) THEN
594: IF ( Y_PREC_STATE .NE. EXTRA_Y ) THEN
595: INCR_PREC = .TRUE.
596: ELSE
597: Z_STATE = NOPROG_STATE
598: END IF
599: ELSE
600: IF ( DZRAT .GT. DZRATMAX ) DZRATMAX = DZRAT
601: END IF
602: IF ( Z_STATE .GT. WORKING_STATE ) FINAL_DZ_Z = DZ_Z
603: END IF
604:
605: IF ( X_STATE.NE.WORKING_STATE.AND.
606: $ ( IGNORE_CWISE.OR.Z_STATE.NE.WORKING_STATE ) )
607: $ GOTO 666
608:
609: IF ( INCR_PREC ) THEN
610: INCR_PREC = .FALSE.
611: Y_PREC_STATE = Y_PREC_STATE + 1
612: DO I = 1, N
613: Y_TAIL( I ) = 0.0D+0
614: END DO
615: END IF
616:
617: PREVNORMDX = NORMDX
618: PREV_DZ_Z = DZ_Z
619: *
620: * Update soluton.
621: *
622: IF (Y_PREC_STATE .LT. EXTRA_Y) THEN
623: CALL DAXPY( N, 1.0D+0, DY, 1, Y(1,J), 1 )
624: ELSE
625: CALL DLA_WWADDW( N, Y( 1, J ), Y_TAIL, DY )
626: END IF
627:
628: END DO
629: * Target of "IF (Z_STOP .AND. X_STOP)". Sun's f77 won't EXIT.
630: 666 CONTINUE
631: *
632: * Set final_* when cnt hits ithresh.
633: *
634: IF ( X_STATE .EQ. WORKING_STATE ) FINAL_DX_X = DX_X
635: IF ( Z_STATE .EQ. WORKING_STATE ) FINAL_DZ_Z = DZ_Z
636: *
637: * Compute error bounds.
638: *
639: IF ( N_NORMS .GE. 1 ) THEN
640: ERR_BNDS_NORM( J, LA_LINRX_ERR_I ) =
641: $ FINAL_DX_X / (1 - DXRATMAX)
642: END IF
643: IF ( N_NORMS .GE. 2 ) THEN
644: ERR_BNDS_COMP( J, LA_LINRX_ERR_I ) =
645: $ FINAL_DZ_Z / (1 - DZRATMAX)
646: END IF
647: *
648: * Compute componentwise relative backward error from formula
649: * max(i) ( abs(R(i)) / ( abs(op(A_s))*abs(Y) + abs(B_s) )(i) )
650: * where abs(Z) is the componentwise absolute value of the matrix
651: * or vector Z.
652: *
653: * Compute residual RES = B_s - op(A_s) * Y,
654: * op(A) = A, A**T, or A**H depending on TRANS (and type).
655: *
656: CALL DCOPY( N, B( 1, J ), 1, RES, 1 )
657: CALL DSYMV( UPLO, N, -1.0D+0, A, LDA, Y(1,J), 1, 1.0D+0, RES,
658: $ 1 )
659:
660: DO I = 1, N
661: AYB( I ) = ABS( B( I, J ) )
662: END DO
663: *
664: * Compute abs(op(A_s))*abs(Y) + abs(B_s).
665: *
666: CALL DLA_SYAMV( UPLO2, N, 1.0D+0,
667: $ A, LDA, Y(1, J), 1, 1.0D+0, AYB, 1 )
668:
669: CALL DLA_LIN_BERR( N, N, 1, RES, AYB, BERR_OUT( J ) )
670: *
671: * End of loop for each RHS.
672: *
673: END DO
674: *
675: RETURN
676: *
677: * End of DLA_PORFSX_EXTENDED
678: *
679: END
CVSweb interface <joel.bertrand@systella.fr>