Actual source code: itfunc.c
1: /*
2: Interface KSP routines that the user calls.
3: */
5: #include <petsc/private/kspimpl.h>
6: #include <petsc/private/matimpl.h>
7: #include <petscdm.h>
9: /* number of nested levels of KSPSetUp/Solve(). This is used to determine if KSP_DIVERGED_ITS should be fatal. */
10: static PetscInt level = 0;
12: static inline PetscErrorCode ObjectView(PetscObject obj, PetscViewer viewer, PetscViewerFormat format)
13: {
15: PetscViewerPushFormat(viewer, format);
16: PetscObjectView(obj, viewer);
17: PetscViewerPopFormat(viewer);
18: return(0);
19: }
21: /*@
22: KSPComputeExtremeSingularValues - Computes the extreme singular values
23: for the preconditioned operator. Called after or during KSPSolve().
25: Not Collective
27: Input Parameter:
28: . ksp - iterative context obtained from KSPCreate()
30: Output Parameters:
31: . emin, emax - extreme singular values
33: Options Database Keys:
34: . -ksp_view_singularvalues - compute extreme singular values and print when KSPSolve() completes.
36: Notes:
37: One must call KSPSetComputeSingularValues() before calling KSPSetUp()
38: (or use the option -ksp_view_eigenvalues) in order for this routine to work correctly.
40: Many users may just want to use the monitoring routine
41: KSPMonitorSingularValue() (which can be set with option -ksp_monitor_singular_value)
42: to print the extreme singular values at each iteration of the linear solve.
44: Estimates of the smallest singular value may be very inaccurate, especially if the Krylov method has not converged.
45: The largest singular value is usually accurate to within a few percent if the method has converged, but is still not
46: intended for eigenanalysis.
48: Disable restarts if using KSPGMRES, otherwise this estimate will only be using those iterations after the last
49: restart. See KSPGMRESSetRestart() for more details.
51: Level: advanced
53: .seealso: KSPSetComputeSingularValues(), KSPMonitorSingularValue(), KSPComputeEigenvalues(), KSP
54: @*/
55: PetscErrorCode KSPComputeExtremeSingularValues(KSP ksp,PetscReal *emax,PetscReal *emin)
56: {
62: if (ksp->ops->computeextremesingularvalues) {
63: (*ksp->ops->computeextremesingularvalues)(ksp,emax,emin);
64: } else {
65: *emin = -1.0;
66: *emax = -1.0;
67: }
68: return 0;
69: }
71: /*@
72: KSPComputeEigenvalues - Computes the extreme eigenvalues for the
73: preconditioned operator. Called after or during KSPSolve().
75: Not Collective
77: Input Parameters:
78: + ksp - iterative context obtained from KSPCreate()
79: - n - size of arrays r and c. The number of eigenvalues computed (neig) will, in
80: general, be less than this.
82: Output Parameters:
83: + r - real part of computed eigenvalues, provided by user with a dimension of at least n
84: . c - complex part of computed eigenvalues, provided by user with a dimension of at least n
85: - neig - actual number of eigenvalues computed (will be less than or equal to n)
87: Options Database Keys:
88: . -ksp_view_eigenvalues - Prints eigenvalues to stdout
90: Notes:
91: The number of eigenvalues estimated depends on the size of the Krylov space
92: generated during the KSPSolve() ; for example, with
93: CG it corresponds to the number of CG iterations, for GMRES it is the number
94: of GMRES iterations SINCE the last restart. Any extra space in r[] and c[]
95: will be ignored.
97: KSPComputeEigenvalues() does not usually provide accurate estimates; it is
98: intended only for assistance in understanding the convergence of iterative
99: methods, not for eigenanalysis. For accurate computation of eigenvalues we recommend using
100: the excellent package SLEPc.
102: One must call KSPSetComputeEigenvalues() before calling KSPSetUp()
103: in order for this routine to work correctly.
105: Many users may just want to use the monitoring routine
106: KSPMonitorSingularValue() (which can be set with option -ksp_monitor_singular_value)
107: to print the singular values at each iteration of the linear solve.
109: Level: advanced
111: .seealso: KSPSetComputeSingularValues(), KSPMonitorSingularValue(), KSPComputeExtremeSingularValues(), KSP
112: @*/
113: PetscErrorCode KSPComputeEigenvalues(KSP ksp,PetscInt n,PetscReal r[],PetscReal c[],PetscInt *neig)
114: {
122: if (n && ksp->ops->computeeigenvalues) {
123: (*ksp->ops->computeeigenvalues)(ksp,n,r,c,neig);
124: } else {
125: *neig = 0;
126: }
127: return 0;
128: }
130: /*@
131: KSPComputeRitz - Computes the Ritz or harmonic Ritz pairs associated to the
132: smallest or largest in modulus, for the preconditioned operator.
133: Called after KSPSolve().
135: Not Collective
137: Input Parameters:
138: + ksp - iterative context obtained from KSPCreate()
139: . ritz - PETSC_TRUE or PETSC_FALSE for Ritz pairs or harmonic Ritz pairs, respectively
140: - small - PETSC_TRUE or PETSC_FALSE for smallest or largest (harmonic) Ritz values, respectively
142: Output Parameters:
143: + nrit - On input number of (harmonic) Ritz pairs to compute; on output, actual number of computed (harmonic) Ritz pairs
144: . S - an array of the Ritz vectors
145: . tetar - real part of the Ritz values
146: - tetai - imaginary part of the Ritz values
148: Notes:
149: -For GMRES, the (harmonic) Ritz pairs are computed from the Hessenberg matrix obtained during
150: the last complete cycle, or obtained at the end of the solution if the method is stopped before
151: a restart. Then, the number of actual (harmonic) Ritz pairs computed is less or equal to the restart
152: parameter for GMRES if a complete cycle has been performed or less or equal to the number of GMRES
153: iterations.
154: -Moreover, for real matrices, the (harmonic) Ritz pairs are possibly complex-valued. In such a case,
155: the routine selects the complex (harmonic) Ritz value and its conjugate, and two successive entries of S
156: are equal to the real and the imaginary parts of the associated vectors.
157: -the (harmonic) Ritz pairs are given in order of increasing (harmonic) Ritz values in modulus
158: -this is currently not implemented when PETSc is built with complex numbers
160: One must call KSPSetComputeRitz() before calling KSPSetUp()
161: in order for this routine to work correctly.
163: Level: advanced
165: .seealso: KSPSetComputeRitz(), KSP
166: @*/
167: PetscErrorCode KSPComputeRitz(KSP ksp,PetscBool ritz,PetscBool small,PetscInt *nrit,Vec S[],PetscReal tetar[],PetscReal tetai[])
168: {
171: if (ksp->ops->computeritz) (*ksp->ops->computeritz)(ksp,ritz,small,nrit,S,tetar,tetai);
172: return 0;
173: }
174: /*@
175: KSPSetUpOnBlocks - Sets up the preconditioner for each block in
176: the block Jacobi, block Gauss-Seidel, and overlapping Schwarz
177: methods.
179: Collective on ksp
181: Input Parameter:
182: . ksp - the KSP context
184: Notes:
185: KSPSetUpOnBlocks() is a routine that the user can optinally call for
186: more precise profiling (via -log_view) of the setup phase for these
187: block preconditioners. If the user does not call KSPSetUpOnBlocks(),
188: it will automatically be called from within KSPSolve().
190: Calling KSPSetUpOnBlocks() is the same as calling PCSetUpOnBlocks()
191: on the PC context within the KSP context.
193: Level: advanced
195: .seealso: PCSetUpOnBlocks(), KSPSetUp(), PCSetUp(), KSP
196: @*/
197: PetscErrorCode KSPSetUpOnBlocks(KSP ksp)
198: {
199: PC pc;
200: PCFailedReason pcreason;
203: level++;
204: KSPGetPC(ksp,&pc);
205: PCSetUpOnBlocks(pc);
206: PCGetFailedReasonRank(pc,&pcreason);
207: level--;
208: /*
209: This is tricky since only a subset of MPI ranks may set this; each KSPSolve_*() is responsible for checking
210: this flag and initializing an appropriate vector with VecSetInf() so that the first norm computation can
211: produce a result at KSPCheckNorm() thus communicating the known problem to all MPI ranks so they may
212: terminate the Krylov solve. For many KSP implementations this is handled within KSPInitialResidual()
213: */
214: if (pcreason) {
215: ksp->reason = KSP_DIVERGED_PC_FAILED;
216: }
217: return 0;
218: }
220: /*@
221: KSPSetReusePreconditioner - reuse the current preconditioner, do not construct a new one even if the operator changes
223: Collective on ksp
225: Input Parameters:
226: + ksp - iterative context obtained from KSPCreate()
227: - flag - PETSC_TRUE to reuse the current preconditioner
229: Level: intermediate
231: .seealso: KSPCreate(), KSPSolve(), KSPDestroy(), PCSetReusePreconditioner(), KSP
232: @*/
233: PetscErrorCode KSPSetReusePreconditioner(KSP ksp,PetscBool flag)
234: {
235: PC pc;
238: KSPGetPC(ksp,&pc);
239: PCSetReusePreconditioner(pc,flag);
240: return 0;
241: }
243: /*@
244: KSPGetReusePreconditioner - Determines if the KSP reuses the current preconditioner even if the operator in the preconditioner has changed.
246: Collective on ksp
248: Input Parameters:
249: . ksp - iterative context obtained from KSPCreate()
251: Output Parameters:
252: . flag - the boolean flag
254: Level: intermediate
256: .seealso: KSPCreate(), KSPSolve(), KSPDestroy(), KSPSetReusePreconditioner(), KSP
257: @*/
258: PetscErrorCode KSPGetReusePreconditioner(KSP ksp,PetscBool *flag)
259: {
262: *flag = PETSC_FALSE;
263: if (ksp->pc) {
264: PCGetReusePreconditioner(ksp->pc,flag);
265: }
266: return 0;
267: }
269: /*@
270: KSPSetSkipPCSetFromOptions - prevents KSPSetFromOptions() from call PCSetFromOptions(). This is used if the same PC is shared by more than one KSP so its options are not resetable for each KSP
272: Collective on ksp
274: Input Parameters:
275: + ksp - iterative context obtained from KSPCreate()
276: - flag - PETSC_TRUE to skip calling the PCSetFromOptions()
278: Level: intermediate
280: .seealso: KSPCreate(), KSPSolve(), KSPDestroy(), PCSetReusePreconditioner(), KSP
281: @*/
282: PetscErrorCode KSPSetSkipPCSetFromOptions(KSP ksp,PetscBool flag)
283: {
285: ksp->skippcsetfromoptions = flag;
286: return 0;
287: }
289: /*@
290: KSPSetUp - Sets up the internal data structures for the
291: later use of an iterative solver.
293: Collective on ksp
295: Input Parameter:
296: . ksp - iterative context obtained from KSPCreate()
298: Level: developer
300: .seealso: KSPCreate(), KSPSolve(), KSPDestroy(), KSP
301: @*/
302: PetscErrorCode KSPSetUp(KSP ksp)
303: {
304: Mat A,B;
305: Mat mat,pmat;
306: MatNullSpace nullsp;
307: PCFailedReason pcreason;
310: level++;
312: /* reset the convergence flag from the previous solves */
313: ksp->reason = KSP_CONVERGED_ITERATING;
315: if (!((PetscObject)ksp)->type_name) {
316: KSPSetType(ksp,KSPGMRES);
317: }
318: KSPSetUpNorms_Private(ksp,PETSC_TRUE,&ksp->normtype,&ksp->pc_side);
320: if (ksp->dmActive && !ksp->setupstage) {
321: /* first time in so build matrix and vector data structures using DM */
322: if (!ksp->vec_rhs) DMCreateGlobalVector(ksp->dm,&ksp->vec_rhs);
323: if (!ksp->vec_sol) DMCreateGlobalVector(ksp->dm,&ksp->vec_sol);
324: DMCreateMatrix(ksp->dm,&A);
325: KSPSetOperators(ksp,A,A);
326: PetscObjectDereference((PetscObject)A);
327: }
329: if (ksp->dmActive) {
330: DMKSP kdm;
331: DMGetDMKSP(ksp->dm,&kdm);
333: if (kdm->ops->computeinitialguess && ksp->setupstage != KSP_SETUP_NEWRHS) {
334: /* only computes initial guess the first time through */
335: (*kdm->ops->computeinitialguess)(ksp,ksp->vec_sol,kdm->initialguessctx);
336: KSPSetInitialGuessNonzero(ksp,PETSC_TRUE);
337: }
338: if (kdm->ops->computerhs) {
339: (*kdm->ops->computerhs)(ksp,ksp->vec_rhs,kdm->rhsctx);
340: }
342: if (ksp->setupstage != KSP_SETUP_NEWRHS) {
343: if (kdm->ops->computeoperators) {
344: KSPGetOperators(ksp,&A,&B);
345: (*kdm->ops->computeoperators)(ksp,A,B,kdm->operatorsctx);
346: } else SETERRQ(PetscObjectComm((PetscObject)ksp),PETSC_ERR_ARG_WRONGSTATE,"You called KSPSetDM() but did not use DMKSPSetComputeOperators() or KSPSetDMActive(ksp,PETSC_FALSE);");
347: }
348: }
350: if (ksp->setupstage == KSP_SETUP_NEWRHS) {
351: level--;
352: return 0;
353: }
354: PetscLogEventBegin(KSP_SetUp,ksp,ksp->vec_rhs,ksp->vec_sol,0);
356: switch (ksp->setupstage) {
357: case KSP_SETUP_NEW:
358: (*ksp->ops->setup)(ksp);
359: break;
360: case KSP_SETUP_NEWMATRIX: { /* This should be replaced with a more general mechanism */
361: if (ksp->setupnewmatrix) {
362: (*ksp->ops->setup)(ksp);
363: }
364: } break;
365: default: break;
366: }
368: if (!ksp->pc) KSPGetPC(ksp,&ksp->pc);
369: PCGetOperators(ksp->pc,&mat,&pmat);
370: /* scale the matrix if requested */
371: if (ksp->dscale) {
372: PetscScalar *xx;
373: PetscInt i,n;
374: PetscBool zeroflag = PETSC_FALSE;
375: if (!ksp->pc) KSPGetPC(ksp,&ksp->pc);
376: if (!ksp->diagonal) { /* allocate vector to hold diagonal */
377: MatCreateVecs(pmat,&ksp->diagonal,NULL);
378: }
379: MatGetDiagonal(pmat,ksp->diagonal);
380: VecGetLocalSize(ksp->diagonal,&n);
381: VecGetArray(ksp->diagonal,&xx);
382: for (i=0; i<n; i++) {
383: if (xx[i] != 0.0) xx[i] = 1.0/PetscSqrtReal(PetscAbsScalar(xx[i]));
384: else {
385: xx[i] = 1.0;
386: zeroflag = PETSC_TRUE;
387: }
388: }
389: VecRestoreArray(ksp->diagonal,&xx);
390: if (zeroflag) {
391: PetscInfo(ksp,"Zero detected in diagonal of matrix, using 1 at those locations\n");
392: }
393: MatDiagonalScale(pmat,ksp->diagonal,ksp->diagonal);
394: if (mat != pmat) MatDiagonalScale(mat,ksp->diagonal,ksp->diagonal);
395: ksp->dscalefix2 = PETSC_FALSE;
396: }
397: PetscLogEventEnd(KSP_SetUp,ksp,ksp->vec_rhs,ksp->vec_sol,0);
398: PCSetErrorIfFailure(ksp->pc,ksp->errorifnotconverged);
399: PCSetUp(ksp->pc);
400: PCGetFailedReasonRank(ksp->pc,&pcreason);
401: /* TODO: this code was wrong and is still wrong, there is no way to propagate the failure to all processes; their is no code to handle a ksp->reason on only some ranks */
402: if (pcreason) {
403: ksp->reason = KSP_DIVERGED_PC_FAILED;
404: }
406: MatGetNullSpace(mat,&nullsp);
407: if (nullsp) {
408: PetscBool test = PETSC_FALSE;
409: PetscOptionsGetBool(((PetscObject)ksp)->options,((PetscObject)ksp)->prefix,"-ksp_test_null_space",&test,NULL);
410: if (test) {
411: MatNullSpaceTest(nullsp,mat,NULL);
412: }
413: }
414: ksp->setupstage = KSP_SETUP_NEWRHS;
415: level--;
416: return 0;
417: }
419: /*@C
420: KSPConvergedReasonView - Displays the reason a KSP solve converged or diverged to a viewer
422: Collective on ksp
424: Parameter:
425: + ksp - iterative context obtained from KSPCreate()
426: - viewer - the viewer to display the reason
428: Options Database Keys:
429: + -ksp_converged_reason - print reason for converged or diverged, also prints number of iterations
430: - -ksp_converged_reason ::failed - only print reason and number of iterations when diverged
432: Notes:
433: To change the format of the output call PetscViewerPushFormat(viewer,format) before this call. Use PETSC_VIEWER_DEFAULT for the default,
434: use PETSC_VIEWER_FAILED to only display a reason if it fails.
436: Level: beginner
438: .seealso: KSPCreate(), KSPSetUp(), KSPDestroy(), KSPSetTolerances(), KSPConvergedDefault(),
439: KSPSolveTranspose(), KSPGetIterationNumber(), KSP, KSPGetConvergedReason(), PetscViewerPushFormat(), PetscViewerPopFormat()
440: @*/
441: PetscErrorCode KSPConvergedReasonView(KSP ksp, PetscViewer viewer)
442: {
443: PetscBool isAscii;
444: PetscViewerFormat format;
446: if (!viewer) viewer = PETSC_VIEWER_STDOUT_(PetscObjectComm((PetscObject)ksp));
447: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&isAscii);
448: if (isAscii) {
449: PetscViewerGetFormat(viewer, &format);
450: PetscViewerASCIIAddTab(viewer,((PetscObject)ksp)->tablevel);
451: if (ksp->reason > 0 && format != PETSC_VIEWER_FAILED) {
452: if (((PetscObject) ksp)->prefix) {
453: PetscViewerASCIIPrintf(viewer,"Linear %s solve converged due to %s iterations %D\n",((PetscObject) ksp)->prefix,KSPConvergedReasons[ksp->reason],ksp->its);
454: } else {
455: PetscViewerASCIIPrintf(viewer,"Linear solve converged due to %s iterations %D\n",KSPConvergedReasons[ksp->reason],ksp->its);
456: }
457: } else if (ksp->reason <= 0) {
458: if (((PetscObject) ksp)->prefix) {
459: PetscViewerASCIIPrintf(viewer,"Linear %s solve did not converge due to %s iterations %D\n",((PetscObject) ksp)->prefix,KSPConvergedReasons[ksp->reason],ksp->its);
460: } else {
461: PetscViewerASCIIPrintf(viewer,"Linear solve did not converge due to %s iterations %D\n",KSPConvergedReasons[ksp->reason],ksp->its);
462: }
463: if (ksp->reason == KSP_DIVERGED_PC_FAILED) {
464: PCFailedReason reason;
465: PCGetFailedReason(ksp->pc,&reason);
466: PetscViewerASCIIPrintf(viewer," PC failed due to %s \n",PCFailedReasons[reason]);
467: }
468: }
469: PetscViewerASCIISubtractTab(viewer,((PetscObject)ksp)->tablevel);
470: }
471: return 0;
472: }
474: /*@C
475: KSPConvergedReasonViewSet - Sets an ADDITIONAL function that is to be used at the
476: end of the linear solver to display the convergence reason of the linear solver.
478: Logically Collective on KSP
480: Input Parameters:
481: + ksp - the KSP context
482: . f - the ksp converged reason view function
483: . vctx - [optional] user-defined context for private data for the
484: ksp converged reason view routine (use NULL if no context is desired)
485: - reasonviewdestroy - [optional] routine that frees reasonview context
486: (may be NULL)
488: Options Database Keys:
489: + -ksp_converged_reason - sets a default KSPConvergedReasonView()
490: - -ksp_converged_reason_view_cancel - cancels all converged reason viewers that have
491: been hardwired into a code by
492: calls to KSPConvergedReasonViewSet(), but
493: does not cancel those set via
494: the options database.
496: Notes:
497: Several different converged reason view routines may be set by calling
498: KSPConvergedReasonViewSet() multiple times; all will be called in the
499: order in which they were set.
501: Level: intermediate
503: .seealso: KSPConvergedReasonView(), KSPConvergedReasonViewCancel()
504: @*/
505: PetscErrorCode KSPConvergedReasonViewSet(KSP ksp,PetscErrorCode (*f)(KSP,void*),void *vctx,PetscErrorCode (*reasonviewdestroy)(void**))
506: {
507: PetscInt i;
508: PetscBool identical;
511: for (i=0; i<ksp->numberreasonviews;i++) {
512: PetscMonitorCompare((PetscErrorCode (*)(void))f,vctx,reasonviewdestroy,(PetscErrorCode (*)(void))ksp->reasonview[i],ksp->reasonviewcontext[i],ksp->reasonviewdestroy[i],&identical);
513: if (identical) return 0;
514: }
516: ksp->reasonview[ksp->numberreasonviews] = f;
517: ksp->reasonviewdestroy[ksp->numberreasonviews] = reasonviewdestroy;
518: ksp->reasonviewcontext[ksp->numberreasonviews++] = (void*)vctx;
519: return 0;
520: }
522: /*@
523: KSPConvergedReasonViewCancel - Clears all the reasonview functions for a KSP object.
525: Collective on KSP
527: Input Parameter:
528: . ksp - iterative context obtained from KSPCreate()
530: Level: intermediate
532: .seealso: KSPCreate(), KSPDestroy(), KSPReset()
533: @*/
534: PetscErrorCode KSPConvergedReasonViewCancel(KSP ksp)
535: {
536: PetscInt i;
539: for (i=0; i<ksp->numberreasonviews; i++) {
540: if (ksp->reasonviewdestroy[i]) {
541: (*ksp->reasonviewdestroy[i])(&ksp->reasonviewcontext[i]);
542: }
543: }
544: ksp->numberreasonviews = 0;
545: return 0;
546: }
548: /*@
549: KSPConvergedReasonViewFromOptions - Processes command line options to determine if/how a KSPReason is to be viewed.
551: Collective on ksp
553: Input Parameters:
554: . ksp - the KSP object
556: Level: intermediate
558: .seealso: KSPConvergedReasonView()
559: @*/
560: PetscErrorCode KSPConvergedReasonViewFromOptions(KSP ksp)
561: {
562: PetscViewer viewer;
563: PetscBool flg;
564: PetscViewerFormat format;
565: PetscInt i;
568: /* Call all user-provided reason review routines */
569: for (i=0; i<ksp->numberreasonviews; i++) {
570: (*ksp->reasonview[i])(ksp,ksp->reasonviewcontext[i]);
571: }
573: /* Call the default PETSc routine */
574: PetscOptionsGetViewer(PetscObjectComm((PetscObject)ksp),((PetscObject)ksp)->options,((PetscObject)ksp)->prefix,"-ksp_converged_reason",&viewer,&format,&flg);
575: if (flg) {
576: PetscViewerPushFormat(viewer,format);
577: KSPConvergedReasonView(ksp, viewer);
578: PetscViewerPopFormat(viewer);
579: PetscViewerDestroy(&viewer);
580: }
581: return 0;
582: }
584: /*@C
585: KSPConvergedRateView - Displays the reason a KSP solve converged or diverged to a viewer
587: Collective on ksp
589: Input Parameters:
590: + ksp - iterative context obtained from KSPCreate()
591: - viewer - the viewer to display the reason
593: Options Database Keys:
594: . -ksp_converged_rate - print reason for convergence or divergence and the convergence rate (or 0.0 for divergence)
596: Notes:
597: To change the format of the output, call PetscViewerPushFormat(viewer,format) before this call.
599: Suppose that the residual is reduced linearly, $r_k = c^k r_0$, which means $log r_k = log r_0 + k log c$. After linear regression,
600: the slope is $\log c$. The coefficient of determination is given by $1 - \frac{\sum_i (y_i - f(x_i))^2}{\sum_i (y_i - \bar y)}$,
601: see also https://en.wikipedia.org/wiki/Coefficient_of_determination
603: Level: intermediate
605: .seealso: KSPConvergedReasonView(), KSPGetConvergedRate(), KSPSetTolerances(), KSPConvergedDefault()
606: @*/
607: PetscErrorCode KSPConvergedRateView(KSP ksp, PetscViewer viewer)
608: {
609: PetscViewerFormat format;
610: PetscBool isAscii;
611: PetscReal rrate, rRsq, erate = 0.0, eRsq = 0.0;
612: PetscInt its;
613: const char *prefix, *reason = KSPConvergedReasons[ksp->reason];
615: KSPGetOptionsPrefix(ksp, &prefix);
616: KSPGetIterationNumber(ksp, &its);
617: KSPComputeConvergenceRate(ksp, &rrate, &rRsq, &erate, &eRsq);
618: if (!viewer) viewer = PETSC_VIEWER_STDOUT_(PetscObjectComm((PetscObject)ksp));
619: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&isAscii);
620: if (isAscii) {
621: PetscViewerGetFormat(viewer, &format);
622: PetscViewerASCIIAddTab(viewer,((PetscObject)ksp)->tablevel);
623: if (ksp->reason > 0) {
624: if (prefix) PetscViewerASCIIPrintf(viewer, "Linear %s solve converged due to %s iterations %D", prefix, reason, its);
625: else PetscViewerASCIIPrintf(viewer, "Linear solve converged due to %s iterations %D", reason, its);
626: PetscViewerASCIIUseTabs(viewer, PETSC_FALSE);
627: if (rRsq >= 0.0) PetscViewerASCIIPrintf(viewer, " res rate %g R^2 %g", rrate, rRsq);
628: if (eRsq >= 0.0) PetscViewerASCIIPrintf(viewer, " error rate %g R^2 %g", erate, eRsq);
629: PetscViewerASCIIPrintf(viewer, "\n");
630: PetscViewerASCIIUseTabs(viewer, PETSC_TRUE);
631: } else if (ksp->reason <= 0) {
632: if (prefix) PetscViewerASCIIPrintf(viewer, "Linear %s solve did not converge due to %s iterations %D", prefix, reason, its);
633: else PetscViewerASCIIPrintf(viewer, "Linear solve did not converge due to %s iterations %D", reason, its);
634: PetscViewerASCIIUseTabs(viewer, PETSC_FALSE);
635: if (rRsq >= 0.0) PetscViewerASCIIPrintf(viewer, " res rate %g R^2 %g", rrate, rRsq);
636: if (eRsq >= 0.0) PetscViewerASCIIPrintf(viewer, " error rate %g R^2 %g", erate, eRsq);
637: PetscViewerASCIIPrintf(viewer, "\n");
638: PetscViewerASCIIUseTabs(viewer, PETSC_TRUE);
639: if (ksp->reason == KSP_DIVERGED_PC_FAILED) {
640: PCFailedReason reason;
641: PCGetFailedReason(ksp->pc,&reason);
642: PetscViewerASCIIPrintf(viewer," PC failed due to %s \n",PCFailedReasons[reason]);
643: }
644: }
645: PetscViewerASCIISubtractTab(viewer,((PetscObject)ksp)->tablevel);
646: }
647: return 0;
648: }
650: #include <petscdraw.h>
652: static PetscErrorCode KSPViewEigenvalues_Internal(KSP ksp, PetscBool isExplicit, PetscViewer viewer, PetscViewerFormat format)
653: {
654: PetscReal *r, *c;
655: PetscInt n, i, neig;
656: PetscBool isascii, isdraw;
657: PetscMPIInt rank;
659: MPI_Comm_rank(PetscObjectComm((PetscObject) ksp), &rank);
660: PetscObjectTypeCompare((PetscObject) viewer, PETSCVIEWERASCII, &isascii);
661: PetscObjectTypeCompare((PetscObject) viewer, PETSCVIEWERDRAW, &isdraw);
662: if (isExplicit) {
663: VecGetSize(ksp->vec_sol,&n);
664: PetscMalloc2(n, &r, n, &c);
665: KSPComputeEigenvaluesExplicitly(ksp, n, r, c);
666: neig = n;
667: } else {
668: PetscInt nits;
670: KSPGetIterationNumber(ksp, &nits);
671: n = nits+2;
672: if (!nits) {PetscViewerASCIIPrintf(viewer, "Zero iterations in solver, cannot approximate any eigenvalues\n"));PetscFunctionReturn(0;}
673: PetscMalloc2(n, &r, n, &c);
674: KSPComputeEigenvalues(ksp, n, r, c, &neig);
675: }
676: if (isascii) {
677: PetscViewerASCIIPrintf(viewer, "%s computed eigenvalues\n", isExplicit ? "Explicitly" : "Iteratively");
678: for (i = 0; i < neig; ++i) {
679: if (c[i] >= 0.0) PetscViewerASCIIPrintf(viewer, "%g + %gi\n", (double) r[i], (double) c[i]);
680: else PetscViewerASCIIPrintf(viewer, "%g - %gi\n", (double) r[i], -(double) c[i]);
681: }
682: } else if (isdraw && rank == 0) {
683: PetscDraw draw;
684: PetscDrawSP drawsp;
686: if (format == PETSC_VIEWER_DRAW_CONTOUR) {
687: KSPPlotEigenContours_Private(ksp,neig,r,c);
688: } else {
689: if (!ksp->eigviewer) PetscViewerDrawOpen(PETSC_COMM_SELF,NULL,isExplicit ? "Explicitly Computed Eigenvalues" : "Iteratively Computed Eigenvalues",PETSC_DECIDE,PETSC_DECIDE,400,400,&ksp->eigviewer);
690: PetscViewerDrawGetDraw(ksp->eigviewer,0,&draw);
691: PetscDrawSPCreate(draw,1,&drawsp);
692: PetscDrawSPReset(drawsp);
693: for (i = 0; i < neig; ++i) PetscDrawSPAddPoint(drawsp,r+i,c+i);
694: PetscDrawSPDraw(drawsp,PETSC_TRUE);
695: PetscDrawSPSave(drawsp);
696: PetscDrawSPDestroy(&drawsp);
697: }
698: }
699: PetscFree2(r, c);
700: return 0;
701: }
703: static PetscErrorCode KSPViewSingularvalues_Internal(KSP ksp, PetscViewer viewer, PetscViewerFormat format)
704: {
705: PetscReal smax, smin;
706: PetscInt nits;
707: PetscBool isascii;
709: PetscObjectTypeCompare((PetscObject) viewer, PETSCVIEWERASCII, &isascii);
710: KSPGetIterationNumber(ksp, &nits);
711: if (!nits) {PetscViewerASCIIPrintf(viewer, "Zero iterations in solver, cannot approximate any singular values\n"));PetscFunctionReturn(0;}
712: KSPComputeExtremeSingularValues(ksp, &smax, &smin);
713: if (isascii) PetscViewerASCIIPrintf(viewer, "Iteratively computed extreme singular values: max %g min %g max/min %g\n",(double)smax,(double)smin,(double)(smax/smin));
714: return 0;
715: }
717: static PetscErrorCode KSPViewFinalResidual_Internal(KSP ksp, PetscViewer viewer, PetscViewerFormat format)
718: {
719: PetscBool isascii;
721: PetscObjectTypeCompare((PetscObject) viewer, PETSCVIEWERASCII, &isascii);
723: if (isascii) {
724: Mat A;
725: Vec t;
726: PetscReal norm;
728: PCGetOperators(ksp->pc, &A, NULL);
729: VecDuplicate(ksp->vec_rhs, &t);
730: KSP_MatMult(ksp, A, ksp->vec_sol, t);
731: VecAYPX(t, -1.0, ksp->vec_rhs);
732: VecNorm(t, NORM_2, &norm);
733: VecDestroy(&t);
734: PetscViewerASCIIPrintf(viewer, "KSP final norm of residual %g\n", (double) norm);
735: }
736: return 0;
737: }
739: static PetscErrorCode KSPMonitorPauseFinal_Internal(KSP ksp)
740: {
741: PetscInt i;
743: if (!ksp->pauseFinal) return 0;
744: for (i = 0; i < ksp->numbermonitors; ++i) {
745: PetscViewerAndFormat *vf = (PetscViewerAndFormat *) ksp->monitorcontext[i];
746: PetscDraw draw;
747: PetscReal lpause;
749: if (!vf) continue;
750: if (vf->lg) {
752: if (((PetscObject) vf->lg)->classid != PETSC_DRAWLG_CLASSID) continue;
753: PetscDrawLGGetDraw(vf->lg, &draw);
754: PetscDrawGetPause(draw, &lpause);
755: PetscDrawSetPause(draw, -1.0);
756: PetscDrawPause(draw);
757: PetscDrawSetPause(draw, lpause);
758: } else {
759: PetscBool isdraw;
762: if (((PetscObject) vf->viewer)->classid != PETSC_VIEWER_CLASSID) continue;
763: PetscObjectTypeCompare((PetscObject) vf->viewer, PETSCVIEWERDRAW, &isdraw);
764: if (!isdraw) continue;
765: PetscViewerDrawGetDraw(vf->viewer, 0, &draw);
766: PetscDrawGetPause(draw, &lpause);
767: PetscDrawSetPause(draw, -1.0);
768: PetscDrawPause(draw);
769: PetscDrawSetPause(draw, lpause);
770: }
771: }
772: return 0;
773: }
775: static PetscErrorCode KSPSolve_Private(KSP ksp,Vec b,Vec x)
776: {
777: PetscBool flg = PETSC_FALSE,inXisinB = PETSC_FALSE,guess_zero;
778: Mat mat,pmat;
779: MPI_Comm comm;
780: MatNullSpace nullsp;
781: Vec btmp,vec_rhs = NULL;
783: level++;
784: comm = PetscObjectComm((PetscObject)ksp);
785: if (x && x == b) {
787: VecDuplicate(b,&x);
788: inXisinB = PETSC_TRUE;
789: }
790: if (b) {
791: PetscObjectReference((PetscObject)b);
792: VecDestroy(&ksp->vec_rhs);
793: ksp->vec_rhs = b;
794: }
795: if (x) {
796: PetscObjectReference((PetscObject)x);
797: VecDestroy(&ksp->vec_sol);
798: ksp->vec_sol = x;
799: }
801: if (ksp->viewPre) ObjectView((PetscObject) ksp, ksp->viewerPre, ksp->formatPre);
803: if (ksp->presolve) (*ksp->presolve)(ksp,ksp->vec_rhs,ksp->vec_sol,ksp->prectx);
805: /* reset the residual history list if requested */
806: if (ksp->res_hist_reset) ksp->res_hist_len = 0;
807: if (ksp->err_hist_reset) ksp->err_hist_len = 0;
809: if (ksp->guess) {
810: PetscObjectState ostate,state;
812: KSPGuessSetUp(ksp->guess);
813: PetscObjectStateGet((PetscObject)ksp->vec_sol,&ostate);
814: KSPGuessFormGuess(ksp->guess,ksp->vec_rhs,ksp->vec_sol);
815: PetscObjectStateGet((PetscObject)ksp->vec_sol,&state);
816: if (state != ostate) {
817: ksp->guess_zero = PETSC_FALSE;
818: } else {
819: PetscInfo(ksp,"Using zero initial guess since the KSPGuess object did not change the vector\n");
820: ksp->guess_zero = PETSC_TRUE;
821: }
822: }
824: /* KSPSetUp() scales the matrix if needed */
825: KSPSetUp(ksp);
826: KSPSetUpOnBlocks(ksp);
828: VecSetErrorIfLocked(ksp->vec_sol,3);
830: PetscLogEventBegin(KSP_Solve,ksp,ksp->vec_rhs,ksp->vec_sol,0);
831: PCGetOperators(ksp->pc,&mat,&pmat);
832: /* diagonal scale RHS if called for */
833: if (ksp->dscale) {
834: VecPointwiseMult(ksp->vec_rhs,ksp->vec_rhs,ksp->diagonal);
835: /* second time in, but matrix was scaled back to original */
836: if (ksp->dscalefix && ksp->dscalefix2) {
837: Mat mat,pmat;
839: PCGetOperators(ksp->pc,&mat,&pmat);
840: MatDiagonalScale(pmat,ksp->diagonal,ksp->diagonal);
841: if (mat != pmat) MatDiagonalScale(mat,ksp->diagonal,ksp->diagonal);
842: }
844: /* scale initial guess */
845: if (!ksp->guess_zero) {
846: if (!ksp->truediagonal) {
847: VecDuplicate(ksp->diagonal,&ksp->truediagonal);
848: VecCopy(ksp->diagonal,ksp->truediagonal);
849: VecReciprocal(ksp->truediagonal);
850: }
851: VecPointwiseMult(ksp->vec_sol,ksp->vec_sol,ksp->truediagonal);
852: }
853: }
854: PCPreSolve(ksp->pc,ksp);
856: if (ksp->guess_zero) VecSet(ksp->vec_sol,0.0);
857: if (ksp->guess_knoll) { /* The Knoll trick is independent on the KSPGuess specified */
858: PCApply(ksp->pc,ksp->vec_rhs,ksp->vec_sol);
859: KSP_RemoveNullSpace(ksp,ksp->vec_sol);
860: ksp->guess_zero = PETSC_FALSE;
861: }
863: /* can we mark the initial guess as zero for this solve? */
864: guess_zero = ksp->guess_zero;
865: if (!ksp->guess_zero) {
866: PetscReal norm;
868: VecNormAvailable(ksp->vec_sol,NORM_2,&flg,&norm);
869: if (flg && !norm) ksp->guess_zero = PETSC_TRUE;
870: }
871: if (ksp->transpose_solve) {
872: MatGetNullSpace(pmat,&nullsp);
873: } else {
874: MatGetTransposeNullSpace(pmat,&nullsp);
875: }
876: if (nullsp) {
877: VecDuplicate(ksp->vec_rhs,&btmp);
878: VecCopy(ksp->vec_rhs,btmp);
879: MatNullSpaceRemove(nullsp,btmp);
880: vec_rhs = ksp->vec_rhs;
881: ksp->vec_rhs = btmp;
882: }
883: VecLockReadPush(ksp->vec_rhs);
884: (*ksp->ops->solve)(ksp);
885: KSPMonitorPauseFinal_Internal(ksp);
887: VecLockReadPop(ksp->vec_rhs);
888: if (nullsp) {
889: ksp->vec_rhs = vec_rhs;
890: VecDestroy(&btmp);
891: }
893: ksp->guess_zero = guess_zero;
896: ksp->totalits += ksp->its;
898: KSPConvergedReasonViewFromOptions(ksp);
900: if (ksp->viewRate) {
901: PetscViewerPushFormat(ksp->viewerRate,ksp->formatRate);
902: KSPConvergedRateView(ksp, ksp->viewerRate);
903: PetscViewerPopFormat(ksp->viewerRate);
904: }
905: PCPostSolve(ksp->pc,ksp);
907: /* diagonal scale solution if called for */
908: if (ksp->dscale) {
909: VecPointwiseMult(ksp->vec_sol,ksp->vec_sol,ksp->diagonal);
910: /* unscale right hand side and matrix */
911: if (ksp->dscalefix) {
912: Mat mat,pmat;
914: VecReciprocal(ksp->diagonal);
915: VecPointwiseMult(ksp->vec_rhs,ksp->vec_rhs,ksp->diagonal);
916: PCGetOperators(ksp->pc,&mat,&pmat);
917: MatDiagonalScale(pmat,ksp->diagonal,ksp->diagonal);
918: if (mat != pmat) MatDiagonalScale(mat,ksp->diagonal,ksp->diagonal);
919: VecReciprocal(ksp->diagonal);
920: ksp->dscalefix2 = PETSC_TRUE;
921: }
922: }
923: PetscLogEventEnd(KSP_Solve,ksp,ksp->vec_rhs,ksp->vec_sol,0);
924: if (ksp->guess) {
925: KSPGuessUpdate(ksp->guess,ksp->vec_rhs,ksp->vec_sol);
926: }
927: if (ksp->postsolve) {
928: (*ksp->postsolve)(ksp,ksp->vec_rhs,ksp->vec_sol,ksp->postctx);
929: }
931: PCGetOperators(ksp->pc,&mat,&pmat);
932: if (ksp->viewEV) KSPViewEigenvalues_Internal(ksp, PETSC_FALSE, ksp->viewerEV, ksp->formatEV);
933: if (ksp->viewEVExp) KSPViewEigenvalues_Internal(ksp, PETSC_TRUE, ksp->viewerEVExp, ksp->formatEVExp);
934: if (ksp->viewSV) KSPViewSingularvalues_Internal(ksp, ksp->viewerSV, ksp->formatSV);
935: if (ksp->viewFinalRes) KSPViewFinalResidual_Internal(ksp, ksp->viewerFinalRes, ksp->formatFinalRes);
936: if (ksp->viewMat) ObjectView((PetscObject) mat, ksp->viewerMat, ksp->formatMat);
937: if (ksp->viewPMat) ObjectView((PetscObject) pmat, ksp->viewerPMat, ksp->formatPMat);
938: if (ksp->viewRhs) ObjectView((PetscObject) ksp->vec_rhs, ksp->viewerRhs, ksp->formatRhs);
939: if (ksp->viewSol) ObjectView((PetscObject) ksp->vec_sol, ksp->viewerSol, ksp->formatSol);
940: if (ksp->view) ObjectView((PetscObject) ksp, ksp->viewer, ksp->format);
941: if (ksp->viewDScale) ObjectView((PetscObject) ksp->diagonal, ksp->viewerDScale, ksp->formatDScale);
942: if (ksp->viewMatExp) {
943: Mat A, B;
945: PCGetOperators(ksp->pc, &A, NULL);
946: if (ksp->transpose_solve) {
947: Mat AT;
949: MatCreateTranspose(A, &AT);
950: MatComputeOperator(AT, MATAIJ, &B);
951: MatDestroy(&AT);
952: } else {
953: MatComputeOperator(A, MATAIJ, &B);
954: }
955: ObjectView((PetscObject) B, ksp->viewerMatExp, ksp->formatMatExp);
956: MatDestroy(&B);
957: }
958: if (ksp->viewPOpExp) {
959: Mat B;
961: KSPComputeOperator(ksp, MATAIJ, &B);
962: ObjectView((PetscObject) B, ksp->viewerPOpExp, ksp->formatPOpExp);
963: MatDestroy(&B);
964: }
966: if (inXisinB) {
967: VecCopy(x,b);
968: VecDestroy(&x);
969: }
970: PetscObjectSAWsBlock((PetscObject)ksp);
971: if (ksp->errorifnotconverged && ksp->reason < 0 && ((level == 1) || (ksp->reason != KSP_DIVERGED_ITS))) {
972: if (ksp->reason == KSP_DIVERGED_PC_FAILED) {
973: PCFailedReason reason;
974: PCGetFailedReason(ksp->pc,&reason);
975: SETERRQ(comm,PETSC_ERR_NOT_CONVERGED,"KSPSolve has not converged, reason %s PC failed due to %s",KSPConvergedReasons[ksp->reason],PCFailedReasons[reason]);
976: } else SETERRQ(comm,PETSC_ERR_NOT_CONVERGED,"KSPSolve has not converged, reason %s",KSPConvergedReasons[ksp->reason]);
977: }
978: level--;
979: return 0;
980: }
982: /*@
983: KSPSolve - Solves linear system.
985: Collective on ksp
987: Parameters:
988: + ksp - iterative context obtained from KSPCreate()
989: . b - the right hand side vector
990: - x - the solution (this may be the same vector as b, then b will be overwritten with answer)
992: Options Database Keys:
993: + -ksp_view_eigenvalues - compute preconditioned operators eigenvalues
994: . -ksp_view_eigenvalues_explicit - compute the eigenvalues by forming the dense operator and using LAPACK
995: . -ksp_view_mat binary - save matrix to the default binary viewer
996: . -ksp_view_pmat binary - save matrix used to build preconditioner to the default binary viewer
997: . -ksp_view_rhs binary - save right hand side vector to the default binary viewer
998: . -ksp_view_solution binary - save computed solution vector to the default binary viewer
999: (can be read later with src/ksp/tutorials/ex10.c for testing solvers)
1000: . -ksp_view_mat_explicit - for matrix-free operators, computes the matrix entries and views them
1001: . -ksp_view_preconditioned_operator_explicit - computes the product of the preconditioner and matrix as an explicit matrix and views it
1002: . -ksp_converged_reason - print reason for converged or diverged, also prints number of iterations
1003: . -ksp_view_final_residual - print 2-norm of true linear system residual at the end of the solution process
1004: . -ksp_error_if_not_converged - stop the program as soon as an error is detected in a KSPSolve()
1005: - -ksp_view - print the ksp data structure at the end of the system solution
1007: Notes:
1009: If one uses KSPSetDM() then x or b need not be passed. Use KSPGetSolution() to access the solution in this case.
1011: The operator is specified with KSPSetOperators().
1013: KSPSolve() will normally return without generating an error regardless of whether the linear system was solved or if constructing the preconditioner failed.
1014: Call KSPGetConvergedReason() to determine if the solver converged or failed and why. The option -ksp_error_if_not_converged or function KSPSetErrorIfNotConverged()
1015: will cause KSPSolve() to error as soon as an error occurs in the linear solver. In inner KSPSolves() KSP_DIVERGED_ITS is not treated as an error because when using nested solvers
1016: it may be fine that inner solvers in the preconditioner do not converge during the solution process.
1018: The number of iterations can be obtained from KSPGetIterationNumber().
1020: If you provide a matrix that has a MatSetNullSpace() and MatSetTransposeNullSpace() this will use that information to solve singular systems
1021: in the least squares sense with a norm minimizing solution.
1022: $
1023: $ A x = b where b = b_p + b_t where b_t is not in the range of A (and hence by the fundamental theorem of linear algebra is in the nullspace(A') see MatSetNullSpace()
1024: $
1025: $ KSP first removes b_t producing the linear system A x = b_p (which has multiple solutions) and solves this to find the ||x|| minimizing solution (and hence
1026: $ it finds the solution x orthogonal to the nullspace(A). The algorithm is simply in each iteration of the Krylov method we remove the nullspace(A) from the search
1027: $ direction thus the solution which is a linear combination of the search directions has no component in the nullspace(A).
1028: $
1029: $ We recommend always using GMRES for such singular systems.
1030: $ If nullspace(A) = nullspace(A') (note symmetric matrices always satisfy this property) then both left and right preconditioning will work
1031: $ If nullspace(A) != nullspace(A') then left preconditioning will work but right preconditioning may not work (or it may).
1033: Developer Note: The reason we cannot always solve nullspace(A) != nullspace(A') systems with right preconditioning is because we need to remove at each iteration
1034: the nullspace(AB) from the search direction. While we know the nullspace(A) the nullspace(AB) equals B^-1 times the nullspace(A) but except for trivial preconditioners
1035: such as diagonal scaling we cannot apply the inverse of the preconditioner to a vector and thus cannot compute the nullspace(AB).
1037: If using a direct method (e.g., via the KSP solver
1038: KSPPREONLY and a preconditioner such as PCLU/PCILU),
1039: then its=1. See KSPSetTolerances() and KSPConvergedDefault()
1040: for more details.
1042: Understanding Convergence:
1043: The routines KSPMonitorSet(), KSPComputeEigenvalues(), and
1044: KSPComputeEigenvaluesExplicitly() provide information on additional
1045: options to monitor convergence and print eigenvalue information.
1047: Level: beginner
1049: .seealso: KSPCreate(), KSPSetUp(), KSPDestroy(), KSPSetTolerances(), KSPConvergedDefault(),
1050: KSPSolveTranspose(), KSPGetIterationNumber(), MatNullSpaceCreate(), MatSetNullSpace(), MatSetTransposeNullSpace(), KSP,
1051: KSPConvergedReasonView(), KSPCheckSolve(), KSPSetErrorIfNotConverged()
1052: @*/
1053: PetscErrorCode KSPSolve(KSP ksp,Vec b,Vec x)
1054: {
1058: ksp->transpose_solve = PETSC_FALSE;
1059: KSPSolve_Private(ksp,b,x);
1060: return 0;
1061: }
1063: /*@
1064: KSPSolveTranspose - Solves the transpose of a linear system.
1066: Collective on ksp
1068: Input Parameters:
1069: + ksp - iterative context obtained from KSPCreate()
1070: . b - right hand side vector
1071: - x - solution vector
1073: Notes:
1074: For complex numbers this solve the non-Hermitian transpose system.
1076: Developer Notes:
1077: We need to implement a KSPSolveHermitianTranspose()
1079: Level: developer
1081: .seealso: KSPCreate(), KSPSetUp(), KSPDestroy(), KSPSetTolerances(), KSPConvergedDefault(),
1082: KSPSolve(), KSP
1083: @*/
1084: PetscErrorCode KSPSolveTranspose(KSP ksp,Vec b,Vec x)
1085: {
1089: if (ksp->transpose.use_explicittranspose) {
1090: Mat J,Jpre;
1091: KSPGetOperators(ksp,&J,&Jpre);
1092: if (!ksp->transpose.reuse_transpose) {
1093: MatTranspose(J,MAT_INITIAL_MATRIX,&ksp->transpose.AT);
1094: if (J != Jpre) {
1095: MatTranspose(Jpre,MAT_INITIAL_MATRIX,&ksp->transpose.BT);
1096: }
1097: ksp->transpose.reuse_transpose = PETSC_TRUE;
1098: } else {
1099: MatTranspose(J,MAT_REUSE_MATRIX,&ksp->transpose.AT);
1100: if (J != Jpre) {
1101: MatTranspose(Jpre,MAT_REUSE_MATRIX,&ksp->transpose.BT);
1102: }
1103: }
1104: if (J == Jpre && ksp->transpose.BT != ksp->transpose.AT) {
1105: PetscObjectReference((PetscObject)ksp->transpose.AT);
1106: ksp->transpose.BT = ksp->transpose.AT;
1107: }
1108: KSPSetOperators(ksp,ksp->transpose.AT,ksp->transpose.BT);
1109: } else {
1110: ksp->transpose_solve = PETSC_TRUE;
1111: }
1112: KSPSolve_Private(ksp,b,x);
1113: return 0;
1114: }
1116: static PetscErrorCode KSPViewFinalMatResidual_Internal(KSP ksp, Mat B, Mat X, PetscViewer viewer, PetscViewerFormat format, PetscInt shift)
1117: {
1118: Mat A, R;
1119: PetscReal *norms;
1120: PetscInt i, N;
1121: PetscBool flg;
1123: PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &flg);
1124: if (flg) {
1125: PCGetOperators(ksp->pc, &A, NULL);
1126: MatMatMult(A, X, MAT_INITIAL_MATRIX, PETSC_DEFAULT, &R);
1127: MatAYPX(R, -1.0, B, SAME_NONZERO_PATTERN);
1128: MatGetSize(R, NULL, &N);
1129: PetscMalloc1(N, &norms);
1130: MatGetColumnNorms(R, NORM_2, norms);
1131: MatDestroy(&R);
1132: for (i = 0; i < N; ++i) {
1133: PetscViewerASCIIPrintf(viewer, "%s #%D %g\n", i == 0 ? "KSP final norm of residual" : " ", shift + i, (double)norms[i]);
1134: }
1135: PetscFree(norms);
1136: }
1137: return 0;
1138: }
1140: /*@
1141: KSPMatSolve - Solves a linear system with multiple right-hand sides stored as a MATDENSE. Unlike KSPSolve(), B and X must be different matrices.
1143: Input Parameters:
1144: + ksp - iterative context
1145: - B - block of right-hand sides
1147: Output Parameter:
1148: . X - block of solutions
1150: Notes:
1151: This is a stripped-down version of KSPSolve(), which only handles -ksp_view, -ksp_converged_reason, and -ksp_view_final_residual.
1153: Level: intermediate
1155: .seealso: KSPSolve(), MatMatSolve(), MATDENSE, KSPHPDDM, PCBJACOBI, PCASM
1156: @*/
1157: PetscErrorCode KSPMatSolve(KSP ksp, Mat B, Mat X)
1158: {
1159: Mat A, P, vB, vX;
1160: Vec cb, cx;
1161: PetscInt n1, N1, n2, N2, Bbn = PETSC_DECIDE;
1162: PetscBool match;
1170: MatCheckPreallocated(X, 3);
1171: if (!X->assembled) {
1172: MatSetOption(X, MAT_NO_OFF_PROC_ENTRIES, PETSC_TRUE);
1173: MatAssemblyBegin(X, MAT_FINAL_ASSEMBLY);
1174: MatAssemblyEnd(X, MAT_FINAL_ASSEMBLY);
1175: }
1177: KSPGetOperators(ksp, &A, &P);
1178: MatGetLocalSize(B, NULL, &n2);
1179: MatGetLocalSize(X, NULL, &n1);
1180: MatGetSize(B, NULL, &N2);
1181: MatGetSize(X, NULL, &N1);
1183: PetscObjectBaseTypeCompareAny((PetscObject)B, &match, MATSEQDENSE, MATMPIDENSE, "");
1185: PetscObjectBaseTypeCompareAny((PetscObject)X, &match, MATSEQDENSE, MATMPIDENSE, "");
1187: KSPSetUp(ksp);
1188: KSPSetUpOnBlocks(ksp);
1189: if (ksp->ops->matsolve) {
1190: if (ksp->guess_zero) {
1191: MatZeroEntries(X);
1192: }
1193: PetscLogEventBegin(KSP_MatSolve, ksp, B, X, 0);
1194: KSPGetMatSolveBatchSize(ksp, &Bbn);
1195: /* by default, do a single solve with all columns */
1196: if (Bbn == PETSC_DECIDE) Bbn = N2;
1198: PetscInfo(ksp, "KSP type %s solving using batches of width at most %D\n", ((PetscObject)ksp)->type_name, Bbn);
1199: /* if -ksp_matsolve_batch_size is greater than the actual number of columns, do a single solve with all columns */
1200: if (Bbn >= N2) {
1201: (*ksp->ops->matsolve)(ksp, B, X);
1202: if (ksp->viewFinalRes) {
1203: KSPViewFinalMatResidual_Internal(ksp, B, X, ksp->viewerFinalRes, ksp->formatFinalRes, 0);
1204: }
1206: KSPConvergedReasonViewFromOptions(ksp);
1208: if (ksp->viewRate) {
1209: PetscViewerPushFormat(ksp->viewerRate,PETSC_VIEWER_DEFAULT);
1210: KSPConvergedRateView(ksp, ksp->viewerRate);
1211: PetscViewerPopFormat(ksp->viewerRate);
1212: }
1213: } else {
1214: for (n2 = 0; n2 < N2; n2 += Bbn) {
1215: MatDenseGetSubMatrix(B, n2, PetscMin(n2+Bbn, N2), &vB);
1216: MatDenseGetSubMatrix(X, n2, PetscMin(n2+Bbn, N2), &vX);
1217: (*ksp->ops->matsolve)(ksp, vB, vX);
1218: if (ksp->viewFinalRes) {
1219: KSPViewFinalMatResidual_Internal(ksp, vB, vX, ksp->viewerFinalRes, ksp->formatFinalRes, n2);
1220: }
1222: KSPConvergedReasonViewFromOptions(ksp);
1224: if (ksp->viewRate) {
1225: PetscViewerPushFormat(ksp->viewerRate,PETSC_VIEWER_DEFAULT);
1226: KSPConvergedRateView(ksp, ksp->viewerRate);
1227: PetscViewerPopFormat(ksp->viewerRate);
1228: }
1229: MatDenseRestoreSubMatrix(B, &vB);
1230: MatDenseRestoreSubMatrix(X, &vX);
1231: }
1232: }
1233: if (ksp->viewMat) ObjectView((PetscObject) A, ksp->viewerMat, ksp->formatMat);
1234: if (ksp->viewPMat) ObjectView((PetscObject) P, ksp->viewerPMat,ksp->formatPMat);
1235: if (ksp->viewRhs) ObjectView((PetscObject) B, ksp->viewerRhs, ksp->formatRhs);
1236: if (ksp->viewSol) ObjectView((PetscObject) X, ksp->viewerSol, ksp->formatSol);
1237: if (ksp->view) {
1238: KSPView(ksp, ksp->viewer);
1239: }
1240: PetscLogEventEnd(KSP_MatSolve, ksp, B, X, 0);
1241: } else {
1242: PetscInfo(ksp, "KSP type %s solving column by column\n", ((PetscObject)ksp)->type_name);
1243: for (n2 = 0; n2 < N2; ++n2) {
1244: MatDenseGetColumnVecRead(B, n2, &cb);
1245: MatDenseGetColumnVecWrite(X, n2, &cx);
1246: KSPSolve(ksp, cb, cx);
1247: MatDenseRestoreColumnVecWrite(X, n2, &cx);
1248: MatDenseRestoreColumnVecRead(B, n2, &cb);
1249: }
1250: }
1251: return 0;
1252: }
1254: /*@
1255: KSPSetMatSolveBatchSize - Sets the maximum number of columns treated simultaneously in KSPMatSolve().
1257: Logically collective
1259: Input Parameters:
1260: + ksp - iterative context
1261: - bs - batch size
1263: Level: advanced
1265: .seealso: KSPMatSolve(), KSPGetMatSolveBatchSize(), -mat_mumps_icntl_27, -matmatmult_Bbn
1266: @*/
1267: PetscErrorCode KSPSetMatSolveBatchSize(KSP ksp, PetscInt bs)
1268: {
1271: ksp->nmax = bs;
1272: return 0;
1273: }
1275: /*@
1276: KSPGetMatSolveBatchSize - Gets the maximum number of columns treated simultaneously in KSPMatSolve().
1278: Input Parameter:
1279: . ksp - iterative context
1281: Output Parameter:
1282: . bs - batch size
1284: Level: advanced
1286: .seealso: KSPMatSolve(), KSPSetMatSolveBatchSize(), -mat_mumps_icntl_27, -matmatmult_Bbn
1287: @*/
1288: PetscErrorCode KSPGetMatSolveBatchSize(KSP ksp, PetscInt *bs)
1289: {
1292: *bs = ksp->nmax;
1293: return 0;
1294: }
1296: /*@
1297: KSPResetViewers - Resets all the viewers set from the options database during KSPSetFromOptions()
1299: Collective on ksp
1301: Input Parameter:
1302: . ksp - iterative context obtained from KSPCreate()
1304: Level: beginner
1306: .seealso: KSPCreate(), KSPSetUp(), KSPSolve(), KSPSetFromOptions(), KSP
1307: @*/
1308: PetscErrorCode KSPResetViewers(KSP ksp)
1309: {
1311: if (!ksp) return 0;
1312: PetscViewerDestroy(&ksp->viewer);
1313: PetscViewerDestroy(&ksp->viewerPre);
1314: PetscViewerDestroy(&ksp->viewerRate);
1315: PetscViewerDestroy(&ksp->viewerMat);
1316: PetscViewerDestroy(&ksp->viewerPMat);
1317: PetscViewerDestroy(&ksp->viewerRhs);
1318: PetscViewerDestroy(&ksp->viewerSol);
1319: PetscViewerDestroy(&ksp->viewerMatExp);
1320: PetscViewerDestroy(&ksp->viewerEV);
1321: PetscViewerDestroy(&ksp->viewerSV);
1322: PetscViewerDestroy(&ksp->viewerEVExp);
1323: PetscViewerDestroy(&ksp->viewerFinalRes);
1324: PetscViewerDestroy(&ksp->viewerPOpExp);
1325: PetscViewerDestroy(&ksp->viewerDScale);
1326: ksp->view = PETSC_FALSE;
1327: ksp->viewPre = PETSC_FALSE;
1328: ksp->viewMat = PETSC_FALSE;
1329: ksp->viewPMat = PETSC_FALSE;
1330: ksp->viewRhs = PETSC_FALSE;
1331: ksp->viewSol = PETSC_FALSE;
1332: ksp->viewMatExp = PETSC_FALSE;
1333: ksp->viewEV = PETSC_FALSE;
1334: ksp->viewSV = PETSC_FALSE;
1335: ksp->viewEVExp = PETSC_FALSE;
1336: ksp->viewFinalRes = PETSC_FALSE;
1337: ksp->viewPOpExp = PETSC_FALSE;
1338: ksp->viewDScale = PETSC_FALSE;
1339: return 0;
1340: }
1342: /*@
1343: KSPReset - Resets a KSP context to the kspsetupcalled = 0 state and removes any allocated Vecs and Mats
1345: Collective on ksp
1347: Input Parameter:
1348: . ksp - iterative context obtained from KSPCreate()
1350: Level: beginner
1352: .seealso: KSPCreate(), KSPSetUp(), KSPSolve(), KSP
1353: @*/
1354: PetscErrorCode KSPReset(KSP ksp)
1355: {
1357: if (!ksp) return 0;
1358: if (ksp->ops->reset) {
1359: (*ksp->ops->reset)(ksp);
1360: }
1361: if (ksp->pc) PCReset(ksp->pc);
1362: if (ksp->guess) {
1363: KSPGuess guess = ksp->guess;
1364: if (guess->ops->reset) (*guess->ops->reset)(guess);
1365: }
1366: VecDestroyVecs(ksp->nwork,&ksp->work);
1367: VecDestroy(&ksp->vec_rhs);
1368: VecDestroy(&ksp->vec_sol);
1369: VecDestroy(&ksp->diagonal);
1370: VecDestroy(&ksp->truediagonal);
1372: KSPResetViewers(ksp);
1374: ksp->setupstage = KSP_SETUP_NEW;
1375: ksp->nmax = PETSC_DECIDE;
1376: return 0;
1377: }
1379: /*@C
1380: KSPDestroy - Destroys KSP context.
1382: Collective on ksp
1384: Input Parameter:
1385: . ksp - iterative context obtained from KSPCreate()
1387: Level: beginner
1389: .seealso: KSPCreate(), KSPSetUp(), KSPSolve(), KSP
1390: @*/
1391: PetscErrorCode KSPDestroy(KSP *ksp)
1392: {
1393: PC pc;
1395: if (!*ksp) return 0;
1397: if (--((PetscObject)(*ksp))->refct > 0) {*ksp = NULL; return 0;}
1399: PetscObjectSAWsViewOff((PetscObject)*ksp);
1401: /*
1402: Avoid a cascading call to PCReset(ksp->pc) from the following call:
1403: PCReset() shouldn't be called from KSPDestroy() as it is unprotected by pc's
1404: refcount (and may be shared, e.g., by other ksps).
1405: */
1406: pc = (*ksp)->pc;
1407: (*ksp)->pc = NULL;
1408: KSPReset((*ksp));
1409: (*ksp)->pc = pc;
1410: if ((*ksp)->ops->destroy) (*(*ksp)->ops->destroy)(*ksp);
1412: if ((*ksp)->transpose.use_explicittranspose) {
1413: MatDestroy(&(*ksp)->transpose.AT);
1414: MatDestroy(&(*ksp)->transpose.BT);
1415: (*ksp)->transpose.reuse_transpose = PETSC_FALSE;
1416: }
1418: KSPGuessDestroy(&(*ksp)->guess);
1419: DMDestroy(&(*ksp)->dm);
1420: PCDestroy(&(*ksp)->pc);
1421: PetscFree((*ksp)->res_hist_alloc);
1422: PetscFree((*ksp)->err_hist_alloc);
1423: if ((*ksp)->convergeddestroy) {
1424: (*(*ksp)->convergeddestroy)((*ksp)->cnvP);
1425: }
1426: KSPMonitorCancel((*ksp));
1427: KSPConvergedReasonViewCancel((*ksp));
1428: PetscViewerDestroy(&(*ksp)->eigviewer);
1429: PetscHeaderDestroy(ksp);
1430: return 0;
1431: }
1433: /*@
1434: KSPSetPCSide - Sets the preconditioning side.
1436: Logically Collective on ksp
1438: Input Parameter:
1439: . ksp - iterative context obtained from KSPCreate()
1441: Output Parameter:
1442: . side - the preconditioning side, where side is one of
1443: .vb
1444: PC_LEFT - left preconditioning (default)
1445: PC_RIGHT - right preconditioning
1446: PC_SYMMETRIC - symmetric preconditioning
1447: .ve
1449: Options Database Keys:
1450: . -ksp_pc_side <right,left,symmetric> - KSP preconditioner side
1452: Notes:
1453: Left preconditioning is used by default for most Krylov methods except KSPFGMRES which only supports right preconditioning.
1455: For methods changing the side of the preconditioner changes the norm type that is used, see KSPSetNormType().
1457: Symmetric preconditioning is currently available only for the KSPQCG method. Note, however, that
1458: symmetric preconditioning can be emulated by using either right or left
1459: preconditioning and a pre or post processing step.
1461: Setting the PC side often affects the default norm type. See KSPSetNormType() for details.
1463: Level: intermediate
1465: .seealso: KSPGetPCSide(), KSPSetNormType(), KSPGetNormType(), KSP
1466: @*/
1467: PetscErrorCode KSPSetPCSide(KSP ksp,PCSide side)
1468: {
1471: ksp->pc_side = ksp->pc_side_set = side;
1472: return 0;
1473: }
1475: /*@
1476: KSPGetPCSide - Gets the preconditioning side.
1478: Not Collective
1480: Input Parameter:
1481: . ksp - iterative context obtained from KSPCreate()
1483: Output Parameter:
1484: . side - the preconditioning side, where side is one of
1485: .vb
1486: PC_LEFT - left preconditioning (default)
1487: PC_RIGHT - right preconditioning
1488: PC_SYMMETRIC - symmetric preconditioning
1489: .ve
1491: Level: intermediate
1493: .seealso: KSPSetPCSide(), KSP
1494: @*/
1495: PetscErrorCode KSPGetPCSide(KSP ksp,PCSide *side)
1496: {
1499: KSPSetUpNorms_Private(ksp,PETSC_TRUE,&ksp->normtype,&ksp->pc_side);
1500: *side = ksp->pc_side;
1501: return 0;
1502: }
1504: /*@
1505: KSPGetTolerances - Gets the relative, absolute, divergence, and maximum
1506: iteration tolerances used by the default KSP convergence tests.
1508: Not Collective
1510: Input Parameter:
1511: . ksp - the Krylov subspace context
1513: Output Parameters:
1514: + rtol - the relative convergence tolerance
1515: . abstol - the absolute convergence tolerance
1516: . dtol - the divergence tolerance
1517: - maxits - maximum number of iterations
1519: Notes:
1520: The user can specify NULL for any parameter that is not needed.
1522: Level: intermediate
1524: maximum, iterations
1526: .seealso: KSPSetTolerances(), KSP
1527: @*/
1528: PetscErrorCode KSPGetTolerances(KSP ksp,PetscReal *rtol,PetscReal *abstol,PetscReal *dtol,PetscInt *maxits)
1529: {
1531: if (abstol) *abstol = ksp->abstol;
1532: if (rtol) *rtol = ksp->rtol;
1533: if (dtol) *dtol = ksp->divtol;
1534: if (maxits) *maxits = ksp->max_it;
1535: return 0;
1536: }
1538: /*@
1539: KSPSetTolerances - Sets the relative, absolute, divergence, and maximum
1540: iteration tolerances used by the default KSP convergence testers.
1542: Logically Collective on ksp
1544: Input Parameters:
1545: + ksp - the Krylov subspace context
1546: . rtol - the relative convergence tolerance, relative decrease in the (possibly preconditioned) residual norm
1547: . abstol - the absolute convergence tolerance absolute size of the (possibly preconditioned) residual norm
1548: . dtol - the divergence tolerance, amount (possibly preconditioned) residual norm can increase before KSPConvergedDefault() concludes that the method is diverging
1549: - maxits - maximum number of iterations to use
1551: Options Database Keys:
1552: + -ksp_atol <abstol> - Sets abstol
1553: . -ksp_rtol <rtol> - Sets rtol
1554: . -ksp_divtol <dtol> - Sets dtol
1555: - -ksp_max_it <maxits> - Sets maxits
1557: Notes:
1558: Use PETSC_DEFAULT to retain the default value of any of the tolerances.
1560: See KSPConvergedDefault() for details how these parameters are used in the default convergence test. See also KSPSetConvergenceTest()
1561: for setting user-defined stopping criteria.
1563: Level: intermediate
1565: convergence, maximum, iterations
1567: .seealso: KSPGetTolerances(), KSPConvergedDefault(), KSPSetConvergenceTest(), KSP
1568: @*/
1569: PetscErrorCode KSPSetTolerances(KSP ksp,PetscReal rtol,PetscReal abstol,PetscReal dtol,PetscInt maxits)
1570: {
1577: if (rtol != PETSC_DEFAULT) {
1579: ksp->rtol = rtol;
1580: }
1581: if (abstol != PETSC_DEFAULT) {
1583: ksp->abstol = abstol;
1584: }
1585: if (dtol != PETSC_DEFAULT) {
1587: ksp->divtol = dtol;
1588: }
1589: if (maxits != PETSC_DEFAULT) {
1591: ksp->max_it = maxits;
1592: }
1593: return 0;
1594: }
1596: /*@
1597: KSPSetInitialGuessNonzero - Tells the iterative solver that the
1598: initial guess is nonzero; otherwise KSP assumes the initial guess
1599: is to be zero (and thus zeros it out before solving).
1601: Logically Collective on ksp
1603: Input Parameters:
1604: + ksp - iterative context obtained from KSPCreate()
1605: - flg - PETSC_TRUE indicates the guess is non-zero, PETSC_FALSE indicates the guess is zero
1607: Options database keys:
1608: . -ksp_initial_guess_nonzero <true,false> - use nonzero initial guess
1610: Level: beginner
1612: Notes:
1613: If this is not called the X vector is zeroed in the call to KSPSolve().
1615: .seealso: KSPGetInitialGuessNonzero(), KSPSetGuessType(), KSPGuessType, KSP
1616: @*/
1617: PetscErrorCode KSPSetInitialGuessNonzero(KSP ksp,PetscBool flg)
1618: {
1621: ksp->guess_zero = (PetscBool) !(int)flg;
1622: return 0;
1623: }
1625: /*@
1626: KSPGetInitialGuessNonzero - Determines whether the KSP solver is using
1627: a zero initial guess.
1629: Not Collective
1631: Input Parameter:
1632: . ksp - iterative context obtained from KSPCreate()
1634: Output Parameter:
1635: . flag - PETSC_TRUE if guess is nonzero, else PETSC_FALSE
1637: Level: intermediate
1639: .seealso: KSPSetInitialGuessNonzero(), KSP
1640: @*/
1641: PetscErrorCode KSPGetInitialGuessNonzero(KSP ksp,PetscBool *flag)
1642: {
1645: if (ksp->guess_zero) *flag = PETSC_FALSE;
1646: else *flag = PETSC_TRUE;
1647: return 0;
1648: }
1650: /*@
1651: KSPSetErrorIfNotConverged - Causes KSPSolve() to generate an error if the solver has not converged.
1653: Logically Collective on ksp
1655: Input Parameters:
1656: + ksp - iterative context obtained from KSPCreate()
1657: - flg - PETSC_TRUE indicates you want the error generated
1659: Options database keys:
1660: . -ksp_error_if_not_converged <true,false> - generate an error and stop the program
1662: Level: intermediate
1664: Notes:
1665: Normally PETSc continues if a linear solver fails to converge, you can call KSPGetConvergedReason() after a KSPSolve()
1666: to determine if it has converged.
1668: A KSP_DIVERGED_ITS will not generate an error in a KSPSolve() inside a nested linear solver
1670: .seealso: KSPGetErrorIfNotConverged(), KSP
1671: @*/
1672: PetscErrorCode KSPSetErrorIfNotConverged(KSP ksp,PetscBool flg)
1673: {
1676: ksp->errorifnotconverged = flg;
1677: return 0;
1678: }
1680: /*@
1681: KSPGetErrorIfNotConverged - Will KSPSolve() generate an error if the solver does not converge?
1683: Not Collective
1685: Input Parameter:
1686: . ksp - iterative context obtained from KSPCreate()
1688: Output Parameter:
1689: . flag - PETSC_TRUE if it will generate an error, else PETSC_FALSE
1691: Level: intermediate
1693: .seealso: KSPSetErrorIfNotConverged(), KSP
1694: @*/
1695: PetscErrorCode KSPGetErrorIfNotConverged(KSP ksp,PetscBool *flag)
1696: {
1699: *flag = ksp->errorifnotconverged;
1700: return 0;
1701: }
1703: /*@
1704: KSPSetInitialGuessKnoll - Tells the iterative solver to use PCApply(pc,b,..) to compute the initial guess (The Knoll trick)
1706: Logically Collective on ksp
1708: Input Parameters:
1709: + ksp - iterative context obtained from KSPCreate()
1710: - flg - PETSC_TRUE or PETSC_FALSE
1712: Level: advanced
1714: Developer Note: the Knoll trick is not currently implemented using the KSPGuess class
1716: .seealso: KSPGetInitialGuessKnoll(), KSPSetInitialGuessNonzero(), KSPGetInitialGuessNonzero(), KSP
1717: @*/
1718: PetscErrorCode KSPSetInitialGuessKnoll(KSP ksp,PetscBool flg)
1719: {
1722: ksp->guess_knoll = flg;
1723: return 0;
1724: }
1726: /*@
1727: KSPGetInitialGuessKnoll - Determines whether the KSP solver is using the Knoll trick (using PCApply(pc,b,...) to compute
1728: the initial guess
1730: Not Collective
1732: Input Parameter:
1733: . ksp - iterative context obtained from KSPCreate()
1735: Output Parameter:
1736: . flag - PETSC_TRUE if using Knoll trick, else PETSC_FALSE
1738: Level: advanced
1740: .seealso: KSPSetInitialGuessKnoll(), KSPSetInitialGuessNonzero(), KSPGetInitialGuessNonzero(), KSP
1741: @*/
1742: PetscErrorCode KSPGetInitialGuessKnoll(KSP ksp,PetscBool *flag)
1743: {
1746: *flag = ksp->guess_knoll;
1747: return 0;
1748: }
1750: /*@
1751: KSPGetComputeSingularValues - Gets the flag indicating whether the extreme singular
1752: values will be calculated via a Lanczos or Arnoldi process as the linear
1753: system is solved.
1755: Not Collective
1757: Input Parameter:
1758: . ksp - iterative context obtained from KSPCreate()
1760: Output Parameter:
1761: . flg - PETSC_TRUE or PETSC_FALSE
1763: Options Database Key:
1764: . -ksp_monitor_singular_value - Activates KSPSetComputeSingularValues()
1766: Notes:
1767: Currently this option is not valid for all iterative methods.
1769: Many users may just want to use the monitoring routine
1770: KSPMonitorSingularValue() (which can be set with option -ksp_monitor_singular_value)
1771: to print the singular values at each iteration of the linear solve.
1773: Level: advanced
1775: .seealso: KSPComputeExtremeSingularValues(), KSPMonitorSingularValue(), KSP
1776: @*/
1777: PetscErrorCode KSPGetComputeSingularValues(KSP ksp,PetscBool *flg)
1778: {
1781: *flg = ksp->calc_sings;
1782: return 0;
1783: }
1785: /*@
1786: KSPSetComputeSingularValues - Sets a flag so that the extreme singular
1787: values will be calculated via a Lanczos or Arnoldi process as the linear
1788: system is solved.
1790: Logically Collective on ksp
1792: Input Parameters:
1793: + ksp - iterative context obtained from KSPCreate()
1794: - flg - PETSC_TRUE or PETSC_FALSE
1796: Options Database Key:
1797: . -ksp_monitor_singular_value - Activates KSPSetComputeSingularValues()
1799: Notes:
1800: Currently this option is not valid for all iterative methods.
1802: Many users may just want to use the monitoring routine
1803: KSPMonitorSingularValue() (which can be set with option -ksp_monitor_singular_value)
1804: to print the singular values at each iteration of the linear solve.
1806: Level: advanced
1808: .seealso: KSPComputeExtremeSingularValues(), KSPMonitorSingularValue(), KSP
1809: @*/
1810: PetscErrorCode KSPSetComputeSingularValues(KSP ksp,PetscBool flg)
1811: {
1814: ksp->calc_sings = flg;
1815: return 0;
1816: }
1818: /*@
1819: KSPGetComputeEigenvalues - Gets the flag indicating that the extreme eigenvalues
1820: values will be calculated via a Lanczos or Arnoldi process as the linear
1821: system is solved.
1823: Not Collective
1825: Input Parameter:
1826: . ksp - iterative context obtained from KSPCreate()
1828: Output Parameter:
1829: . flg - PETSC_TRUE or PETSC_FALSE
1831: Notes:
1832: Currently this option is not valid for all iterative methods.
1834: Level: advanced
1836: .seealso: KSPComputeEigenvalues(), KSPComputeEigenvaluesExplicitly(), KSP
1837: @*/
1838: PetscErrorCode KSPGetComputeEigenvalues(KSP ksp,PetscBool *flg)
1839: {
1842: *flg = ksp->calc_sings;
1843: return 0;
1844: }
1846: /*@
1847: KSPSetComputeEigenvalues - Sets a flag so that the extreme eigenvalues
1848: values will be calculated via a Lanczos or Arnoldi process as the linear
1849: system is solved.
1851: Logically Collective on ksp
1853: Input Parameters:
1854: + ksp - iterative context obtained from KSPCreate()
1855: - flg - PETSC_TRUE or PETSC_FALSE
1857: Notes:
1858: Currently this option is not valid for all iterative methods.
1860: Level: advanced
1862: .seealso: KSPComputeEigenvalues(), KSPComputeEigenvaluesExplicitly(), KSP
1863: @*/
1864: PetscErrorCode KSPSetComputeEigenvalues(KSP ksp,PetscBool flg)
1865: {
1868: ksp->calc_sings = flg;
1869: return 0;
1870: }
1872: /*@
1873: KSPSetComputeRitz - Sets a flag so that the Ritz or harmonic Ritz pairs
1874: will be calculated via a Lanczos or Arnoldi process as the linear
1875: system is solved.
1877: Logically Collective on ksp
1879: Input Parameters:
1880: + ksp - iterative context obtained from KSPCreate()
1881: - flg - PETSC_TRUE or PETSC_FALSE
1883: Notes:
1884: Currently this option is only valid for the GMRES method.
1886: Level: advanced
1888: .seealso: KSPComputeRitz(), KSP
1889: @*/
1890: PetscErrorCode KSPSetComputeRitz(KSP ksp, PetscBool flg)
1891: {
1894: ksp->calc_ritz = flg;
1895: return 0;
1896: }
1898: /*@
1899: KSPGetRhs - Gets the right-hand-side vector for the linear system to
1900: be solved.
1902: Not Collective
1904: Input Parameter:
1905: . ksp - iterative context obtained from KSPCreate()
1907: Output Parameter:
1908: . r - right-hand-side vector
1910: Level: developer
1912: .seealso: KSPGetSolution(), KSPSolve(), KSP
1913: @*/
1914: PetscErrorCode KSPGetRhs(KSP ksp,Vec *r)
1915: {
1918: *r = ksp->vec_rhs;
1919: return 0;
1920: }
1922: /*@
1923: KSPGetSolution - Gets the location of the solution for the
1924: linear system to be solved. Note that this may not be where the solution
1925: is stored during the iterative process; see KSPBuildSolution().
1927: Not Collective
1929: Input Parameters:
1930: . ksp - iterative context obtained from KSPCreate()
1932: Output Parameters:
1933: . v - solution vector
1935: Level: developer
1937: .seealso: KSPGetRhs(), KSPBuildSolution(), KSPSolve(), KSP
1938: @*/
1939: PetscErrorCode KSPGetSolution(KSP ksp,Vec *v)
1940: {
1943: *v = ksp->vec_sol;
1944: return 0;
1945: }
1947: /*@
1948: KSPSetPC - Sets the preconditioner to be used to calculate the
1949: application of the preconditioner on a vector.
1951: Collective on ksp
1953: Input Parameters:
1954: + ksp - iterative context obtained from KSPCreate()
1955: - pc - the preconditioner object (can be NULL)
1957: Notes:
1958: Use KSPGetPC() to retrieve the preconditioner context.
1960: Level: developer
1962: .seealso: KSPGetPC(), KSP
1963: @*/
1964: PetscErrorCode KSPSetPC(KSP ksp,PC pc)
1965: {
1967: if (pc) {
1970: }
1971: PetscObjectReference((PetscObject)pc);
1972: PCDestroy(&ksp->pc);
1973: ksp->pc = pc;
1974: PetscLogObjectParent((PetscObject)ksp,(PetscObject)ksp->pc);
1975: return 0;
1976: }
1978: /*@
1979: KSPGetPC - Returns a pointer to the preconditioner context
1980: set with KSPSetPC().
1982: Not Collective
1984: Input Parameters:
1985: . ksp - iterative context obtained from KSPCreate()
1987: Output Parameter:
1988: . pc - preconditioner context
1990: Level: developer
1992: .seealso: KSPSetPC(), KSP
1993: @*/
1994: PetscErrorCode KSPGetPC(KSP ksp,PC *pc)
1995: {
1998: if (!ksp->pc) {
1999: PCCreate(PetscObjectComm((PetscObject)ksp),&ksp->pc);
2000: PetscObjectIncrementTabLevel((PetscObject)ksp->pc,(PetscObject)ksp,0);
2001: PetscLogObjectParent((PetscObject)ksp,(PetscObject)ksp->pc);
2002: PetscObjectSetOptions((PetscObject)ksp->pc,((PetscObject)ksp)->options);
2003: }
2004: *pc = ksp->pc;
2005: return 0;
2006: }
2008: /*@
2009: KSPMonitor - runs the user provided monitor routines, if they exist
2011: Collective on ksp
2013: Input Parameters:
2014: + ksp - iterative context obtained from KSPCreate()
2015: . it - iteration number
2016: - rnorm - relative norm of the residual
2018: Notes:
2019: This routine is called by the KSP implementations.
2020: It does not typically need to be called by the user.
2022: Level: developer
2024: .seealso: KSPMonitorSet()
2025: @*/
2026: PetscErrorCode KSPMonitor(KSP ksp,PetscInt it,PetscReal rnorm)
2027: {
2028: PetscInt i, n = ksp->numbermonitors;
2030: for (i=0; i<n; i++) {
2031: (*ksp->monitor[i])(ksp,it,rnorm,ksp->monitorcontext[i]);
2032: }
2033: return 0;
2034: }
2036: /*@C
2037: KSPMonitorSet - Sets an ADDITIONAL function to be called at every iteration to monitor
2038: the residual/error etc.
2040: Logically Collective on ksp
2042: Input Parameters:
2043: + ksp - iterative context obtained from KSPCreate()
2044: . monitor - pointer to function (if this is NULL, it turns off monitoring
2045: . mctx - [optional] context for private data for the
2046: monitor routine (use NULL if no context is desired)
2047: - monitordestroy - [optional] routine that frees monitor context
2048: (may be NULL)
2050: Calling Sequence of monitor:
2051: $ monitor (KSP ksp, PetscInt it, PetscReal rnorm, void *mctx)
2053: + ksp - iterative context obtained from KSPCreate()
2054: . it - iteration number
2055: . rnorm - (estimated) 2-norm of (preconditioned) residual
2056: - mctx - optional monitoring context, as set by KSPMonitorSet()
2058: Options Database Keys:
2059: + -ksp_monitor - sets KSPMonitorResidual()
2060: . -ksp_monitor draw - sets KSPMonitorResidualDraw() and plots residual
2061: . -ksp_monitor draw::draw_lg - sets KSPMonitorResidualDrawLG() and plots residual
2062: . -ksp_monitor_pause_final - Pauses any graphics when the solve finishes (only works for internal monitors)
2063: . -ksp_monitor_true_residual - sets KSPMonitorTrueResidual()
2064: . -ksp_monitor_true_residual draw::draw_lg - sets KSPMonitorTrueResidualDrawLG() and plots residual
2065: . -ksp_monitor_max - sets KSPMonitorTrueResidualMax()
2066: . -ksp_monitor_singular_value - sets KSPMonitorSingularValue()
2067: - -ksp_monitor_cancel - cancels all monitors that have
2068: been hardwired into a code by
2069: calls to KSPMonitorSet(), but
2070: does not cancel those set via
2071: the options database.
2073: Notes:
2074: The default is to do nothing. To print the residual, or preconditioned
2075: residual if KSPSetNormType(ksp,KSP_NORM_PRECONDITIONED) was called, use
2076: KSPMonitorResidual() as the monitoring routine, with a ASCII viewer as the
2077: context.
2079: Several different monitoring routines may be set by calling
2080: KSPMonitorSet() multiple times; all will be called in the
2081: order in which they were set.
2083: Fortran Notes:
2084: Only a single monitor function can be set for each KSP object
2086: Level: beginner
2088: .seealso: KSPMonitorResidual(), KSPMonitorCancel(), KSP
2089: @*/
2090: PetscErrorCode KSPMonitorSet(KSP ksp,PetscErrorCode (*monitor)(KSP,PetscInt,PetscReal,void*),void *mctx,PetscErrorCode (*monitordestroy)(void**))
2091: {
2092: PetscInt i;
2093: PetscBool identical;
2096: for (i=0; i<ksp->numbermonitors;i++) {
2097: PetscMonitorCompare((PetscErrorCode (*)(void))monitor,mctx,monitordestroy,(PetscErrorCode (*)(void))ksp->monitor[i],ksp->monitorcontext[i],ksp->monitordestroy[i],&identical);
2098: if (identical) return 0;
2099: }
2101: ksp->monitor[ksp->numbermonitors] = monitor;
2102: ksp->monitordestroy[ksp->numbermonitors] = monitordestroy;
2103: ksp->monitorcontext[ksp->numbermonitors++] = (void*)mctx;
2104: return 0;
2105: }
2107: /*@
2108: KSPMonitorCancel - Clears all monitors for a KSP object.
2110: Logically Collective on ksp
2112: Input Parameters:
2113: . ksp - iterative context obtained from KSPCreate()
2115: Options Database Key:
2116: . -ksp_monitor_cancel - Cancels all monitors that have
2117: been hardwired into a code by calls to KSPMonitorSet(),
2118: but does not cancel those set via the options database.
2120: Level: intermediate
2122: .seealso: KSPMonitorResidual(), KSPMonitorSet(), KSP
2123: @*/
2124: PetscErrorCode KSPMonitorCancel(KSP ksp)
2125: {
2126: PetscInt i;
2129: for (i=0; i<ksp->numbermonitors; i++) {
2130: if (ksp->monitordestroy[i]) {
2131: (*ksp->monitordestroy[i])(&ksp->monitorcontext[i]);
2132: }
2133: }
2134: ksp->numbermonitors = 0;
2135: return 0;
2136: }
2138: /*@C
2139: KSPGetMonitorContext - Gets the monitoring context, as set by
2140: KSPMonitorSet() for the FIRST monitor only.
2142: Not Collective
2144: Input Parameter:
2145: . ksp - iterative context obtained from KSPCreate()
2147: Output Parameter:
2148: . ctx - monitoring context
2150: Level: intermediate
2152: .seealso: KSPMonitorResidual(), KSP
2153: @*/
2154: PetscErrorCode KSPGetMonitorContext(KSP ksp,void *ctx)
2155: {
2157: *(void**)ctx = ksp->monitorcontext[0];
2158: return 0;
2159: }
2161: /*@
2162: KSPSetResidualHistory - Sets the array used to hold the residual history.
2163: If set, this array will contain the residual norms computed at each
2164: iteration of the solver.
2166: Not Collective
2168: Input Parameters:
2169: + ksp - iterative context obtained from KSPCreate()
2170: . a - array to hold history
2171: . na - size of a
2172: - reset - PETSC_TRUE indicates the history counter is reset to zero
2173: for each new linear solve
2175: Level: advanced
2177: Notes:
2178: If provided, he array is NOT freed by PETSc so the user needs to keep track of it and destroy once the KSP object is destroyed.
2179: If 'a' is NULL then space is allocated for the history. If 'na' PETSC_DECIDE or PETSC_DEFAULT then a
2180: default array of length 10000 is allocated.
2182: .seealso: KSPGetResidualHistory(), KSP
2184: @*/
2185: PetscErrorCode KSPSetResidualHistory(KSP ksp,PetscReal a[],PetscInt na,PetscBool reset)
2186: {
2189: PetscFree(ksp->res_hist_alloc);
2190: if (na != PETSC_DECIDE && na != PETSC_DEFAULT && a) {
2191: ksp->res_hist = a;
2192: ksp->res_hist_max = (size_t) na;
2193: } else {
2194: if (na != PETSC_DECIDE && na != PETSC_DEFAULT) ksp->res_hist_max = (size_t) na;
2195: else ksp->res_hist_max = 10000; /* like default ksp->max_it */
2196: PetscCalloc1(ksp->res_hist_max,&ksp->res_hist_alloc);
2198: ksp->res_hist = ksp->res_hist_alloc;
2199: }
2200: ksp->res_hist_len = 0;
2201: ksp->res_hist_reset = reset;
2202: return 0;
2203: }
2205: /*@C
2206: KSPGetResidualHistory - Gets the array used to hold the residual history
2207: and the number of residuals it contains.
2209: Not Collective
2211: Input Parameter:
2212: . ksp - iterative context obtained from KSPCreate()
2214: Output Parameters:
2215: + a - pointer to array to hold history (or NULL)
2216: - na - number of used entries in a (or NULL)
2218: Level: advanced
2220: Notes:
2221: This array is borrowed and should not be freed by the caller.
2222: Can only be called after a KSPSetResidualHistory() otherwise a and na are set to zero
2224: The Fortran version of this routine has a calling sequence
2225: $ call KSPGetResidualHistory(KSP ksp, integer na, integer ierr)
2226: note that you have passed a Fortran array into KSPSetResidualHistory() and you need
2227: to access the residual values from this Fortran array you provided. Only the na (number of
2228: residual norms currently held) is set.
2230: .seealso: KSPSetResidualHistory(), KSP
2232: @*/
2233: PetscErrorCode KSPGetResidualHistory(KSP ksp, const PetscReal *a[],PetscInt *na)
2234: {
2236: if (a) *a = ksp->res_hist;
2237: if (na) *na = (PetscInt) ksp->res_hist_len;
2238: return 0;
2239: }
2241: /*@
2242: KSPSetErrorHistory - Sets the array used to hold the error history. If set, this array will contain the error norms computed at each iteration of the solver.
2244: Not Collective
2246: Input Parameters:
2247: + ksp - iterative context obtained from KSPCreate()
2248: . a - array to hold history
2249: . na - size of a
2250: - reset - PETSC_TRUE indicates the history counter is reset to zero for each new linear solve
2252: Level: advanced
2254: Notes:
2255: If provided, the array is NOT freed by PETSc so the user needs to keep track of it and destroy once the KSP object is destroyed.
2256: If 'a' is NULL then space is allocated for the history. If 'na' PETSC_DECIDE or PETSC_DEFAULT then a default array of length 10000 is allocated.
2258: .seealso: KSPGetErrorHistory(), KSPSetResidualHistory(), KSP
2259: @*/
2260: PetscErrorCode KSPSetErrorHistory(KSP ksp, PetscReal a[], PetscInt na, PetscBool reset)
2261: {
2264: PetscFree(ksp->err_hist_alloc);
2265: if (na != PETSC_DECIDE && na != PETSC_DEFAULT && a) {
2266: ksp->err_hist = a;
2267: ksp->err_hist_max = (size_t) na;
2268: } else {
2269: if (na != PETSC_DECIDE && na != PETSC_DEFAULT) ksp->err_hist_max = (size_t) na;
2270: else ksp->err_hist_max = 10000; /* like default ksp->max_it */
2271: PetscCalloc1(ksp->err_hist_max, &ksp->err_hist_alloc);
2273: ksp->err_hist = ksp->err_hist_alloc;
2274: }
2275: ksp->err_hist_len = 0;
2276: ksp->err_hist_reset = reset;
2277: return 0;
2278: }
2280: /*@C
2281: KSPGetErrorHistory - Gets the array used to hold the error history and the number of residuals it contains.
2283: Not Collective
2285: Input Parameter:
2286: . ksp - iterative context obtained from KSPCreate()
2288: Output Parameters:
2289: + a - pointer to array to hold history (or NULL)
2290: - na - number of used entries in a (or NULL)
2292: Level: advanced
2294: Notes:
2295: This array is borrowed and should not be freed by the caller.
2296: Can only be called after a KSPSetErrorHistory() otherwise a and na are set to zero
2297: The Fortran version of this routine has a calling sequence
2298: $ call KSPGetErrorHistory(KSP ksp, integer na, integer ierr)
2299: note that you have passed a Fortran array into KSPSetErrorHistory() and you need
2300: to access the residual values from this Fortran array you provided. Only the na (number of
2301: residual norms currently held) is set.
2303: .seealso: KSPSetErrorHistory(), KSPGetResidualHistory(), KSP
2304: @*/
2305: PetscErrorCode KSPGetErrorHistory(KSP ksp, const PetscReal *a[], PetscInt *na)
2306: {
2308: if (a) *a = ksp->err_hist;
2309: if (na) *na = (PetscInt) ksp->err_hist_len;
2310: return 0;
2311: }
2313: /*
2314: KSPComputeConvergenceRate - Compute the convergence rate for the iteration
2316: Not collective
2318: Input Parameter:
2319: . ksp - The KSP
2321: Output Parameters:
2322: + cr - The residual contraction rate
2323: . rRsq - The coefficient of determination, R^2, indicating the linearity of the data
2324: . ce - The error contraction rate
2325: - eRsq - The coefficient of determination, R^2, indicating the linearity of the data
2327: Note:
2328: Suppose that the residual is reduced linearly, $r_k = c^k r_0$, which means $log r_k = log r_0 + k log c$. After linear regression,
2329: the slope is $\log c$. The coefficient of determination is given by $1 - \frac{\sum_i (y_i - f(x_i))^2}{\sum_i (y_i - \bar y)}$,
2330: see also https://en.wikipedia.org/wiki/Coefficient_of_determination
2332: Level: advanced
2334: .seealso: KSPConvergedRateView()
2335: */
2336: PetscErrorCode KSPComputeConvergenceRate(KSP ksp, PetscReal *cr, PetscReal *rRsq, PetscReal *ce, PetscReal *eRsq)
2337: {
2338: PetscReal const *hist;
2339: PetscReal *x, *y, slope, intercept, mean = 0.0, var = 0.0, res = 0.0;
2340: PetscInt n, k;
2342: if (cr || rRsq) {
2343: KSPGetResidualHistory(ksp, &hist, &n);
2344: if (!n) {
2345: if (cr) *cr = 0.0;
2346: if (rRsq) *rRsq = -1.0;
2347: } else {
2348: PetscMalloc2(n, &x, n, &y);
2349: for (k = 0; k < n; ++k) {
2350: x[k] = k;
2351: y[k] = PetscLogReal(hist[k]);
2352: mean += y[k];
2353: }
2354: mean /= n;
2355: PetscLinearRegression(n, x, y, &slope, &intercept);
2356: for (k = 0; k < n; ++k) {
2357: res += PetscSqr(y[k] - (slope*x[k] + intercept));
2358: var += PetscSqr(y[k] - mean);
2359: }
2360: PetscFree2(x, y);
2361: if (cr) *cr = PetscExpReal(slope);
2362: if (rRsq) *rRsq = var < PETSC_MACHINE_EPSILON ? 0.0 : 1.0 - (res / var);
2363: }
2364: }
2365: if (ce || eRsq) {
2366: KSPGetErrorHistory(ksp, &hist, &n);
2367: if (!n) {
2368: if (ce) *ce = 0.0;
2369: if (eRsq) *eRsq = -1.0;
2370: } else {
2371: PetscMalloc2(n, &x, n, &y);
2372: for (k = 0; k < n; ++k) {
2373: x[k] = k;
2374: y[k] = PetscLogReal(hist[k]);
2375: mean += y[k];
2376: }
2377: mean /= n;
2378: PetscLinearRegression(n, x, y, &slope, &intercept);
2379: for (k = 0; k < n; ++k) {
2380: res += PetscSqr(y[k] - (slope*x[k] + intercept));
2381: var += PetscSqr(y[k] - mean);
2382: }
2383: PetscFree2(x, y);
2384: if (ce) *ce = PetscExpReal(slope);
2385: if (eRsq) *eRsq = var < PETSC_MACHINE_EPSILON ? 0.0 : 1.0 - (res / var);
2386: }
2387: }
2388: return 0;
2389: }
2391: /*@C
2392: KSPSetConvergenceTest - Sets the function to be used to determine
2393: convergence.
2395: Logically Collective on ksp
2397: Input Parameters:
2398: + ksp - iterative context obtained from KSPCreate()
2399: . converge - pointer to the function
2400: . cctx - context for private data for the convergence routine (may be null)
2401: - destroy - a routine for destroying the context (may be null)
2403: Calling sequence of converge:
2404: $ converge (KSP ksp, PetscInt it, PetscReal rnorm, KSPConvergedReason *reason,void *mctx)
2406: + ksp - iterative context obtained from KSPCreate()
2407: . it - iteration number
2408: . rnorm - (estimated) 2-norm of (preconditioned) residual
2409: . reason - the reason why it has converged or diverged
2410: - cctx - optional convergence context, as set by KSPSetConvergenceTest()
2412: Notes:
2413: Must be called after the KSP type has been set so put this after
2414: a call to KSPSetType(), or KSPSetFromOptions().
2416: The default convergence test, KSPConvergedDefault(), aborts if the
2417: residual grows to more than 10000 times the initial residual.
2419: The default is a combination of relative and absolute tolerances.
2420: The residual value that is tested may be an approximation; routines
2421: that need exact values should compute them.
2423: In the default PETSc convergence test, the precise values of reason
2424: are macros such as KSP_CONVERGED_RTOL, which are defined in petscksp.h.
2426: Level: advanced
2428: .seealso: KSPConvergedDefault(), KSPGetConvergenceContext(), KSPSetTolerances(), KSP, KSPGetConvergenceTest(), KSPGetAndClearConvergenceTest()
2429: @*/
2430: PetscErrorCode KSPSetConvergenceTest(KSP ksp,PetscErrorCode (*converge)(KSP,PetscInt,PetscReal,KSPConvergedReason*,void*),void *cctx,PetscErrorCode (*destroy)(void*))
2431: {
2433: if (ksp->convergeddestroy) {
2434: (*ksp->convergeddestroy)(ksp->cnvP);
2435: }
2436: ksp->converged = converge;
2437: ksp->convergeddestroy = destroy;
2438: ksp->cnvP = (void*)cctx;
2439: return 0;
2440: }
2442: /*@C
2443: KSPGetConvergenceTest - Gets the function to be used to determine
2444: convergence.
2446: Logically Collective on ksp
2448: Input Parameter:
2449: . ksp - iterative context obtained from KSPCreate()
2451: Output Parameters:
2452: + converge - pointer to convergence test function
2453: . cctx - context for private data for the convergence routine (may be null)
2454: - destroy - a routine for destroying the context (may be null)
2456: Calling sequence of converge:
2457: $ converge (KSP ksp, PetscInt it, PetscReal rnorm, KSPConvergedReason *reason,void *mctx)
2459: + ksp - iterative context obtained from KSPCreate()
2460: . it - iteration number
2461: . rnorm - (estimated) 2-norm of (preconditioned) residual
2462: . reason - the reason why it has converged or diverged
2463: - cctx - optional convergence context, as set by KSPSetConvergenceTest()
2465: Level: advanced
2467: .seealso: KSPConvergedDefault(), KSPGetConvergenceContext(), KSPSetTolerances(), KSP, KSPSetConvergenceTest(), KSPGetAndClearConvergenceTest()
2468: @*/
2469: PetscErrorCode KSPGetConvergenceTest(KSP ksp,PetscErrorCode (**converge)(KSP,PetscInt,PetscReal,KSPConvergedReason*,void*),void **cctx,PetscErrorCode (**destroy)(void*))
2470: {
2472: if (converge) *converge = ksp->converged;
2473: if (destroy) *destroy = ksp->convergeddestroy;
2474: if (cctx) *cctx = ksp->cnvP;
2475: return 0;
2476: }
2478: /*@C
2479: KSPGetAndClearConvergenceTest - Gets the function to be used to determine convergence. Removes the current test without calling destroy on the test context
2481: Logically Collective on ksp
2483: Input Parameter:
2484: . ksp - iterative context obtained from KSPCreate()
2486: Output Parameters:
2487: + converge - pointer to convergence test function
2488: . cctx - context for private data for the convergence routine
2489: - destroy - a routine for destroying the context
2491: Calling sequence of converge:
2492: $ converge (KSP ksp, PetscInt it, PetscReal rnorm, KSPConvergedReason *reason,void *mctx)
2494: + ksp - iterative context obtained from KSPCreate()
2495: . it - iteration number
2496: . rnorm - (estimated) 2-norm of (preconditioned) residual
2497: . reason - the reason why it has converged or diverged
2498: - cctx - optional convergence context, as set by KSPSetConvergenceTest()
2500: Level: advanced
2502: Notes: This is intended to be used to allow transferring the convergence test (and its context) to another testing object (for example another KSP) and then calling
2503: KSPSetConvergenceTest() on this original KSP. If you just called KSPGetConvergenceTest() followed by KSPSetConvergenceTest() the original context information
2504: would be destroyed and hence the transferred context would be invalid and trigger a crash on use
2506: .seealso: KSPConvergedDefault(), KSPGetConvergenceContext(), KSPSetTolerances(), KSP, KSPSetConvergenceTest(), KSPGetConvergenceTest()
2507: @*/
2508: PetscErrorCode KSPGetAndClearConvergenceTest(KSP ksp,PetscErrorCode (**converge)(KSP,PetscInt,PetscReal,KSPConvergedReason*,void*),void **cctx,PetscErrorCode (**destroy)(void*))
2509: {
2511: *converge = ksp->converged;
2512: *destroy = ksp->convergeddestroy;
2513: *cctx = ksp->cnvP;
2514: ksp->converged = NULL;
2515: ksp->cnvP = NULL;
2516: ksp->convergeddestroy = NULL;
2517: return 0;
2518: }
2520: /*@C
2521: KSPGetConvergenceContext - Gets the convergence context set with
2522: KSPSetConvergenceTest().
2524: Not Collective
2526: Input Parameter:
2527: . ksp - iterative context obtained from KSPCreate()
2529: Output Parameter:
2530: . ctx - monitoring context
2532: Level: advanced
2534: .seealso: KSPConvergedDefault(), KSPSetConvergenceTest(), KSP
2535: @*/
2536: PetscErrorCode KSPGetConvergenceContext(KSP ksp,void *ctx)
2537: {
2539: *(void**)ctx = ksp->cnvP;
2540: return 0;
2541: }
2543: /*@C
2544: KSPBuildSolution - Builds the approximate solution in a vector provided.
2545: This routine is NOT commonly needed (see KSPSolve()).
2547: Collective on ksp
2549: Input Parameter:
2550: . ctx - iterative context obtained from KSPCreate()
2552: Output Parameter:
2553: Provide exactly one of
2554: + v - location to stash solution.
2555: - V - the solution is returned in this location. This vector is created
2556: internally. This vector should NOT be destroyed by the user with
2557: VecDestroy().
2559: Notes:
2560: This routine can be used in one of two ways
2561: .vb
2562: KSPBuildSolution(ksp,NULL,&V);
2563: or
2564: KSPBuildSolution(ksp,v,NULL); or KSPBuildSolution(ksp,v,&v);
2565: .ve
2566: In the first case an internal vector is allocated to store the solution
2567: (the user cannot destroy this vector). In the second case the solution
2568: is generated in the vector that the user provides. Note that for certain
2569: methods, such as KSPCG, the second case requires a copy of the solution,
2570: while in the first case the call is essentially free since it simply
2571: returns the vector where the solution already is stored. For some methods
2572: like GMRES this is a reasonably expensive operation and should only be
2573: used in truly needed.
2575: Level: advanced
2577: .seealso: KSPGetSolution(), KSPBuildResidual(), KSP
2578: @*/
2579: PetscErrorCode KSPBuildSolution(KSP ksp,Vec v,Vec *V)
2580: {
2583: if (!V) V = &v;
2584: (*ksp->ops->buildsolution)(ksp,v,V);
2585: return 0;
2586: }
2588: /*@C
2589: KSPBuildResidual - Builds the residual in a vector provided.
2591: Collective on ksp
2593: Input Parameter:
2594: . ksp - iterative context obtained from KSPCreate()
2596: Output Parameters:
2597: + v - optional location to stash residual. If v is not provided,
2598: then a location is generated.
2599: . t - work vector. If not provided then one is generated.
2600: - V - the residual
2602: Notes:
2603: Regardless of whether or not v is provided, the residual is
2604: returned in V.
2606: Level: advanced
2608: .seealso: KSPBuildSolution()
2609: @*/
2610: PetscErrorCode KSPBuildResidual(KSP ksp,Vec t,Vec v,Vec *V)
2611: {
2612: PetscBool flag = PETSC_FALSE;
2613: Vec w = v,tt = t;
2616: if (!w) {
2617: VecDuplicate(ksp->vec_rhs,&w);
2618: PetscLogObjectParent((PetscObject)ksp,(PetscObject)w);
2619: }
2620: if (!tt) {
2621: VecDuplicate(ksp->vec_sol,&tt); flag = PETSC_TRUE;
2622: PetscLogObjectParent((PetscObject)ksp,(PetscObject)tt);
2623: }
2624: (*ksp->ops->buildresidual)(ksp,tt,w,V);
2625: if (flag) VecDestroy(&tt);
2626: return 0;
2627: }
2629: /*@
2630: KSPSetDiagonalScale - Tells KSP to symmetrically diagonally scale the system
2631: before solving. This actually CHANGES the matrix (and right hand side).
2633: Logically Collective on ksp
2635: Input Parameters:
2636: + ksp - the KSP context
2637: - scale - PETSC_TRUE or PETSC_FALSE
2639: Options Database Key:
2640: + -ksp_diagonal_scale -
2641: - -ksp_diagonal_scale_fix - scale the matrix back AFTER the solve
2643: Notes:
2644: Scales the matrix by D^(-1/2) A D^(-1/2) [D^(1/2) x ] = D^(-1/2) b
2645: where D_{ii} is 1/abs(A_{ii}) unless A_{ii} is zero and then it is 1.
2647: BE CAREFUL with this routine: it actually scales the matrix and right
2648: hand side that define the system. After the system is solved the matrix
2649: and right hand side remain scaled unless you use KSPSetDiagonalScaleFix()
2651: This should NOT be used within the SNES solves if you are using a line
2652: search.
2654: If you use this with the PCType Eisenstat preconditioner than you can
2655: use the PCEisenstatSetNoDiagonalScaling() option, or -pc_eisenstat_no_diagonal_scaling
2656: to save some unneeded, redundant flops.
2658: Level: intermediate
2660: .seealso: KSPGetDiagonalScale(), KSPSetDiagonalScaleFix(), KSP
2661: @*/
2662: PetscErrorCode KSPSetDiagonalScale(KSP ksp,PetscBool scale)
2663: {
2666: ksp->dscale = scale;
2667: return 0;
2668: }
2670: /*@
2671: KSPGetDiagonalScale - Checks if KSP solver scales the matrix and
2672: right hand side
2674: Not Collective
2676: Input Parameter:
2677: . ksp - the KSP context
2679: Output Parameter:
2680: . scale - PETSC_TRUE or PETSC_FALSE
2682: Notes:
2683: BE CAREFUL with this routine: it actually scales the matrix and right
2684: hand side that define the system. After the system is solved the matrix
2685: and right hand side remain scaled unless you use KSPSetDiagonalScaleFix()
2687: Level: intermediate
2689: .seealso: KSPSetDiagonalScale(), KSPSetDiagonalScaleFix(), KSP
2690: @*/
2691: PetscErrorCode KSPGetDiagonalScale(KSP ksp,PetscBool *scale)
2692: {
2695: *scale = ksp->dscale;
2696: return 0;
2697: }
2699: /*@
2700: KSPSetDiagonalScaleFix - Tells KSP to diagonally scale the system
2701: back after solving.
2703: Logically Collective on ksp
2705: Input Parameters:
2706: + ksp - the KSP context
2707: - fix - PETSC_TRUE to scale back after the system solve, PETSC_FALSE to not
2708: rescale (default)
2710: Notes:
2711: Must be called after KSPSetDiagonalScale()
2713: Using this will slow things down, because it rescales the matrix before and
2714: after each linear solve. This is intended mainly for testing to allow one
2715: to easily get back the original system to make sure the solution computed is
2716: accurate enough.
2718: Level: intermediate
2720: .seealso: KSPGetDiagonalScale(), KSPSetDiagonalScale(), KSPGetDiagonalScaleFix(), KSP
2721: @*/
2722: PetscErrorCode KSPSetDiagonalScaleFix(KSP ksp,PetscBool fix)
2723: {
2726: ksp->dscalefix = fix;
2727: return 0;
2728: }
2730: /*@
2731: KSPGetDiagonalScaleFix - Determines if KSP diagonally scales the system
2732: back after solving.
2734: Not Collective
2736: Input Parameter:
2737: . ksp - the KSP context
2739: Output Parameter:
2740: . fix - PETSC_TRUE to scale back after the system solve, PETSC_FALSE to not
2741: rescale (default)
2743: Notes:
2744: Must be called after KSPSetDiagonalScale()
2746: If PETSC_TRUE will slow things down, because it rescales the matrix before and
2747: after each linear solve. This is intended mainly for testing to allow one
2748: to easily get back the original system to make sure the solution computed is
2749: accurate enough.
2751: Level: intermediate
2753: .seealso: KSPGetDiagonalScale(), KSPSetDiagonalScale(), KSPSetDiagonalScaleFix(), KSP
2754: @*/
2755: PetscErrorCode KSPGetDiagonalScaleFix(KSP ksp,PetscBool *fix)
2756: {
2759: *fix = ksp->dscalefix;
2760: return 0;
2761: }
2763: /*@C
2764: KSPSetComputeOperators - set routine to compute the linear operators
2766: Logically Collective
2768: Input Parameters:
2769: + ksp - the KSP context
2770: . func - function to compute the operators
2771: - ctx - optional context
2773: Calling sequence of func:
2774: $ func(KSP ksp,Mat A,Mat B,void *ctx)
2776: + ksp - the KSP context
2777: . A - the linear operator
2778: . B - preconditioning matrix
2779: - ctx - optional user-provided context
2781: Notes:
2782: The user provided func() will be called automatically at the very next call to KSPSolve(). It will not be called at future KSPSolve() calls
2783: unless either KSPSetComputeOperators() or KSPSetOperators() is called before that KSPSolve() is called.
2785: To reuse the same preconditioner for the next KSPSolve() and not compute a new one based on the most recently computed matrix call KSPSetReusePreconditioner()
2787: Level: beginner
2789: .seealso: KSPSetOperators(), KSPSetComputeRHS(), DMKSPSetComputeOperators(), KSPSetComputeInitialGuess()
2790: @*/
2791: PetscErrorCode KSPSetComputeOperators(KSP ksp,PetscErrorCode (*func)(KSP,Mat,Mat,void*),void *ctx)
2792: {
2793: DM dm;
2796: KSPGetDM(ksp,&dm);
2797: DMKSPSetComputeOperators(dm,func,ctx);
2798: if (ksp->setupstage == KSP_SETUP_NEWRHS) ksp->setupstage = KSP_SETUP_NEWMATRIX;
2799: return 0;
2800: }
2802: /*@C
2803: KSPSetComputeRHS - set routine to compute the right hand side of the linear system
2805: Logically Collective
2807: Input Parameters:
2808: + ksp - the KSP context
2809: . func - function to compute the right hand side
2810: - ctx - optional context
2812: Calling sequence of func:
2813: $ func(KSP ksp,Vec b,void *ctx)
2815: + ksp - the KSP context
2816: . b - right hand side of linear system
2817: - ctx - optional user-provided context
2819: Notes:
2820: The routine you provide will be called EACH you call KSPSolve() to prepare the new right hand side for that solve
2822: Level: beginner
2824: .seealso: KSPSolve(), DMKSPSetComputeRHS(), KSPSetComputeOperators()
2825: @*/
2826: PetscErrorCode KSPSetComputeRHS(KSP ksp,PetscErrorCode (*func)(KSP,Vec,void*),void *ctx)
2827: {
2828: DM dm;
2831: KSPGetDM(ksp,&dm);
2832: DMKSPSetComputeRHS(dm,func,ctx);
2833: return 0;
2834: }
2836: /*@C
2837: KSPSetComputeInitialGuess - set routine to compute the initial guess of the linear system
2839: Logically Collective
2841: Input Parameters:
2842: + ksp - the KSP context
2843: . func - function to compute the initial guess
2844: - ctx - optional context
2846: Calling sequence of func:
2847: $ func(KSP ksp,Vec x,void *ctx)
2849: + ksp - the KSP context
2850: . x - solution vector
2851: - ctx - optional user-provided context
2853: Notes: This should only be used in conjunction with KSPSetComputeRHS(), KSPSetComputeOperators(), otherwise
2854: call KSPSetInitialGuessNonzero() and set the initial guess values in the solution vector passed to KSPSolve().
2856: Level: beginner
2858: .seealso: KSPSolve(), KSPSetComputeRHS(), KSPSetComputeOperators(), DMKSPSetComputeInitialGuess()
2859: @*/
2860: PetscErrorCode KSPSetComputeInitialGuess(KSP ksp,PetscErrorCode (*func)(KSP,Vec,void*),void *ctx)
2861: {
2862: DM dm;
2865: KSPGetDM(ksp,&dm);
2866: DMKSPSetComputeInitialGuess(dm,func,ctx);
2867: return 0;
2868: }
2870: /*@
2871: KSPSetUseExplicitTranspose - Determines if transpose the system explicitly
2872: in KSPSolveTranspose.
2874: Logically Collective on ksp
2876: Input Parameter:
2877: . ksp - the KSP context
2879: Output Parameter:
2880: . flg - PETSC_TRUE to transpose the system in KSPSolveTranspose, PETSC_FALSE to not
2881: transpose (default)
2883: Level: advanced
2885: .seealso: KSPSolveTranspose(), KSP
2886: @*/
2887: PetscErrorCode KSPSetUseExplicitTranspose(KSP ksp,PetscBool flg)
2888: {
2891: ksp->transpose.use_explicittranspose = flg;
2892: return 0;
2893: }