Actual source code: mg.c
2: /*
3: Defines the multigrid preconditioner interface.
4: */
5: #include <petsc/private/pcmgimpl.h>
6: #include <petsc/private/kspimpl.h>
7: #include <petscdm.h>
8: PETSC_INTERN PetscErrorCode PCPreSolveChangeRHS(PC,PetscBool*);
10: /*
11: Contains the list of registered coarse space construction routines
12: */
13: PetscFunctionList PCMGCoarseList = NULL;
15: PetscErrorCode PCMGMCycle_Private(PC pc,PC_MG_Levels **mglevelsin,PetscBool transpose,PetscBool matapp,PCRichardsonConvergedReason *reason)
16: {
17: PC_MG *mg = (PC_MG*)pc->data;
18: PC_MG_Levels *mgc,*mglevels = *mglevelsin;
19: PetscInt cycles = (mglevels->level == 1) ? 1 : (PetscInt) mglevels->cycles;
21: if (mglevels->eventsmoothsolve) PetscLogEventBegin(mglevels->eventsmoothsolve,0,0,0,0);
22: if (!transpose) {
23: if (matapp) {
24: KSPMatSolve(mglevels->smoothd,mglevels->B,mglevels->X); /* pre-smooth */
25: KSPCheckSolve(mglevels->smoothd,pc,NULL);
26: } else {
27: KSPSolve(mglevels->smoothd,mglevels->b,mglevels->x); /* pre-smooth */
28: KSPCheckSolve(mglevels->smoothd,pc,mglevels->x);
29: }
30: } else {
32: KSPSolveTranspose(mglevels->smoothu,mglevels->b,mglevels->x); /* transpose of post-smooth */
33: KSPCheckSolve(mglevels->smoothu,pc,mglevels->x);
34: }
35: if (mglevels->eventsmoothsolve) PetscLogEventEnd(mglevels->eventsmoothsolve,0,0,0,0);
36: if (mglevels->level) { /* not the coarsest grid */
37: if (mglevels->eventresidual) PetscLogEventBegin(mglevels->eventresidual,0,0,0,0);
38: if (matapp && !mglevels->R) {
39: MatDuplicate(mglevels->B,MAT_DO_NOT_COPY_VALUES,&mglevels->R);
40: }
41: if (!transpose) {
42: if (matapp) (*mglevels->matresidual)(mglevels->A,mglevels->B,mglevels->X,mglevels->R);
43: else (*mglevels->residual)(mglevels->A,mglevels->b,mglevels->x,mglevels->r);
44: } else {
45: if (matapp) (*mglevels->matresidualtranspose)(mglevels->A,mglevels->B,mglevels->X,mglevels->R);
46: else (*mglevels->residualtranspose)(mglevels->A,mglevels->b,mglevels->x,mglevels->r);
47: }
48: if (mglevels->eventresidual) PetscLogEventEnd(mglevels->eventresidual,0,0,0,0);
50: /* if on finest level and have convergence criteria set */
51: if (mglevels->level == mglevels->levels-1 && mg->ttol && reason) {
52: PetscReal rnorm;
53: VecNorm(mglevels->r,NORM_2,&rnorm);
54: if (rnorm <= mg->ttol) {
55: if (rnorm < mg->abstol) {
56: *reason = PCRICHARDSON_CONVERGED_ATOL;
57: PetscInfo(pc,"Linear solver has converged. Residual norm %g is less than absolute tolerance %g\n",(double)rnorm,(double)mg->abstol);
58: } else {
59: *reason = PCRICHARDSON_CONVERGED_RTOL;
60: PetscInfo(pc,"Linear solver has converged. Residual norm %g is less than relative tolerance times initial residual norm %g\n",(double)rnorm,(double)mg->ttol);
61: }
62: return 0;
63: }
64: }
66: mgc = *(mglevelsin - 1);
67: if (mglevels->eventinterprestrict) PetscLogEventBegin(mglevels->eventinterprestrict,0,0,0,0);
68: if (!transpose) {
69: if (matapp) MatMatRestrict(mglevels->restrct,mglevels->R,&mgc->B);
70: else MatRestrict(mglevels->restrct,mglevels->r,mgc->b);
71: } else {
72: if (matapp) MatMatRestrict(mglevels->interpolate,mglevels->R,&mgc->B);
73: else MatRestrict(mglevels->interpolate,mglevels->r,mgc->b);
74: }
75: if (mglevels->eventinterprestrict) PetscLogEventEnd(mglevels->eventinterprestrict,0,0,0,0);
76: if (matapp) {
77: if (!mgc->X) {
78: MatDuplicate(mgc->B,MAT_DO_NOT_COPY_VALUES,&mgc->X);
79: } else {
80: MatZeroEntries(mgc->X);
81: }
82: } else {
83: VecZeroEntries(mgc->x);
84: }
85: while (cycles--) {
86: PCMGMCycle_Private(pc,mglevelsin-1,transpose,matapp,reason);
87: }
88: if (mglevels->eventinterprestrict) PetscLogEventBegin(mglevels->eventinterprestrict,0,0,0,0);
89: if (!transpose) {
90: if (matapp) MatMatInterpolateAdd(mglevels->interpolate,mgc->X,mglevels->X,&mglevels->X);
91: else MatInterpolateAdd(mglevels->interpolate,mgc->x,mglevels->x,mglevels->x);
92: } else {
93: MatInterpolateAdd(mglevels->restrct,mgc->x,mglevels->x,mglevels->x);
94: }
95: if (mglevels->eventinterprestrict) PetscLogEventEnd(mglevels->eventinterprestrict,0,0,0,0);
96: if (mglevels->eventsmoothsolve) PetscLogEventBegin(mglevels->eventsmoothsolve,0,0,0,0);
97: if (!transpose) {
98: if (matapp) {
99: KSPMatSolve(mglevels->smoothu,mglevels->B,mglevels->X); /* post smooth */
100: KSPCheckSolve(mglevels->smoothu,pc,NULL);
101: } else {
102: KSPSolve(mglevels->smoothu,mglevels->b,mglevels->x); /* post smooth */
103: KSPCheckSolve(mglevels->smoothu,pc,mglevels->x);
104: }
105: } else {
107: KSPSolveTranspose(mglevels->smoothd,mglevels->b,mglevels->x); /* post smooth */
108: KSPCheckSolve(mglevels->smoothd,pc,mglevels->x);
109: }
110: if (mglevels->cr) {
112: /* TODO Turn on copy and turn off noisy if we have an exact solution
113: VecCopy(mglevels->x, mglevels->crx);
114: VecCopy(mglevels->b, mglevels->crb); */
115: KSPSetNoisy_Private(mglevels->crx);
116: KSPSolve(mglevels->cr,mglevels->crb,mglevels->crx); /* compatible relaxation */
117: KSPCheckSolve(mglevels->cr,pc,mglevels->crx);
118: }
119: if (mglevels->eventsmoothsolve) PetscLogEventEnd(mglevels->eventsmoothsolve,0,0,0,0);
120: }
121: return 0;
122: }
124: static PetscErrorCode PCApplyRichardson_MG(PC pc,Vec b,Vec x,Vec w,PetscReal rtol,PetscReal abstol, PetscReal dtol,PetscInt its,PetscBool zeroguess,PetscInt *outits,PCRichardsonConvergedReason *reason)
125: {
126: PC_MG *mg = (PC_MG*)pc->data;
127: PC_MG_Levels **mglevels = mg->levels;
128: PC tpc;
129: PetscBool changeu,changed;
130: PetscInt levels = mglevels[0]->levels,i;
132: /* When the DM is supplying the matrix then it will not exist until here */
133: for (i=0; i<levels; i++) {
134: if (!mglevels[i]->A) {
135: KSPGetOperators(mglevels[i]->smoothu,&mglevels[i]->A,NULL);
136: PetscObjectReference((PetscObject)mglevels[i]->A);
137: }
138: }
140: KSPGetPC(mglevels[levels-1]->smoothd,&tpc);
141: PCPreSolveChangeRHS(tpc,&changed);
142: KSPGetPC(mglevels[levels-1]->smoothu,&tpc);
143: PCPreSolveChangeRHS(tpc,&changeu);
144: if (!changed && !changeu) {
145: VecDestroy(&mglevels[levels-1]->b);
146: mglevels[levels-1]->b = b;
147: } else { /* if the smoother changes the rhs during PreSolve, we cannot use the input vector */
148: if (!mglevels[levels-1]->b) {
149: Vec *vec;
151: KSPCreateVecs(mglevels[levels-1]->smoothd,1,&vec,0,NULL);
152: mglevels[levels-1]->b = *vec;
153: PetscFree(vec);
154: }
155: VecCopy(b,mglevels[levels-1]->b);
156: }
157: mglevels[levels-1]->x = x;
159: mg->rtol = rtol;
160: mg->abstol = abstol;
161: mg->dtol = dtol;
162: if (rtol) {
163: /* compute initial residual norm for relative convergence test */
164: PetscReal rnorm;
165: if (zeroguess) {
166: VecNorm(b,NORM_2,&rnorm);
167: } else {
168: (*mglevels[levels-1]->residual)(mglevels[levels-1]->A,b,x,w);
169: VecNorm(w,NORM_2,&rnorm);
170: }
171: mg->ttol = PetscMax(rtol*rnorm,abstol);
172: } else if (abstol) mg->ttol = abstol;
173: else mg->ttol = 0.0;
175: /* since smoother is applied to full system, not just residual we need to make sure that smoothers don't
176: stop prematurely due to small residual */
177: for (i=1; i<levels; i++) {
178: KSPSetTolerances(mglevels[i]->smoothu,0,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT);
179: if (mglevels[i]->smoothu != mglevels[i]->smoothd) {
180: /* For Richardson the initial guess is nonzero since it is solving in each cycle the original system not just applying as a preconditioner */
181: KSPSetInitialGuessNonzero(mglevels[i]->smoothd,PETSC_TRUE);
182: KSPSetTolerances(mglevels[i]->smoothd,0,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT);
183: }
184: }
186: *reason = (PCRichardsonConvergedReason)0;
187: for (i=0; i<its; i++) {
188: PCMGMCycle_Private(pc,mglevels+levels-1,PETSC_FALSE,PETSC_FALSE,reason);
189: if (*reason) break;
190: }
191: if (!*reason) *reason = PCRICHARDSON_CONVERGED_ITS;
192: *outits = i;
193: if (!changed && !changeu) mglevels[levels-1]->b = NULL;
194: return 0;
195: }
197: PetscErrorCode PCReset_MG(PC pc)
198: {
199: PC_MG *mg = (PC_MG*)pc->data;
200: PC_MG_Levels **mglevels = mg->levels;
201: PetscInt i,c,n;
203: if (mglevels) {
204: n = mglevels[0]->levels;
205: for (i=0; i<n-1; i++) {
206: VecDestroy(&mglevels[i+1]->r);
207: VecDestroy(&mglevels[i]->b);
208: VecDestroy(&mglevels[i]->x);
209: MatDestroy(&mglevels[i+1]->R);
210: MatDestroy(&mglevels[i]->B);
211: MatDestroy(&mglevels[i]->X);
212: VecDestroy(&mglevels[i]->crx);
213: VecDestroy(&mglevels[i]->crb);
214: MatDestroy(&mglevels[i+1]->restrct);
215: MatDestroy(&mglevels[i+1]->interpolate);
216: MatDestroy(&mglevels[i+1]->inject);
217: VecDestroy(&mglevels[i+1]->rscale);
218: }
219: VecDestroy(&mglevels[n-1]->crx);
220: VecDestroy(&mglevels[n-1]->crb);
221: /* this is not null only if the smoother on the finest level
222: changes the rhs during PreSolve */
223: VecDestroy(&mglevels[n-1]->b);
224: MatDestroy(&mglevels[n-1]->B);
226: for (i=0; i<n; i++) {
227: if (mglevels[i]->coarseSpace) for (c = 0; c < mg->Nc; ++c) VecDestroy(&mglevels[i]->coarseSpace[c]);
228: PetscFree(mglevels[i]->coarseSpace);
229: mglevels[i]->coarseSpace = NULL;
230: MatDestroy(&mglevels[i]->A);
231: if (mglevels[i]->smoothd != mglevels[i]->smoothu) {
232: KSPReset(mglevels[i]->smoothd);
233: }
234: KSPReset(mglevels[i]->smoothu);
235: if (mglevels[i]->cr) KSPReset(mglevels[i]->cr);
236: }
237: mg->Nc = 0;
238: }
239: return 0;
240: }
242: /* Implementing CR
244: We only want to make corrections that ``do not change'' the coarse solution. What we mean by not changing is that if I prolong my coarse solution to the fine grid and then inject that fine solution back to the coarse grid, I get the same answer. Injection is what Brannick calls R. We want the complementary projector to Inj, which we will call S, after Brannick, so that Inj S = 0. Now the orthogonal projector onto the range of Inj^T is
246: Inj^T (Inj Inj^T)^{-1} Inj
248: and if Inj is a VecScatter, as it is now in PETSc, we have
250: Inj^T Inj
252: and
254: S = I - Inj^T Inj
256: since
258: Inj S = Inj - (Inj Inj^T) Inj = 0.
260: Brannick suggests
262: A \to S^T A S \qquad\mathrm{and}\qquad M \to S^T M S
264: but I do not think his :math:`S^T S = I` is correct. Our S is an orthogonal projector, so :math:`S^T S = S^2 = S`. We will use
266: M^{-1} A \to S M^{-1} A S
268: In fact, since it is somewhat hard in PETSc to do the symmetric application, we will just apply S on the left.
270: Check: || Inj P - I ||_F < tol
271: Check: In general, Inj Inj^T = I
272: */
274: typedef struct {
275: PC mg; /* The PCMG object */
276: PetscInt l; /* The multigrid level for this solver */
277: Mat Inj; /* The injection matrix */
278: Mat S; /* I - Inj^T Inj */
279: } CRContext;
281: static PetscErrorCode CRSetup_Private(PC pc)
282: {
283: CRContext *ctx;
284: Mat It;
287: PCShellGetContext(pc, &ctx);
288: PCMGGetInjection(ctx->mg, ctx->l, &It);
290: MatCreateTranspose(It, &ctx->Inj);
291: MatCreateNormal(ctx->Inj, &ctx->S);
292: MatScale(ctx->S, -1.0);
293: MatShift(ctx->S, 1.0);
294: return 0;
295: }
297: static PetscErrorCode CRApply_Private(PC pc, Vec x, Vec y)
298: {
299: CRContext *ctx;
302: PCShellGetContext(pc, &ctx);
303: MatMult(ctx->S, x, y);
304: return 0;
305: }
307: static PetscErrorCode CRDestroy_Private(PC pc)
308: {
309: CRContext *ctx;
312: PCShellGetContext(pc, &ctx);
313: MatDestroy(&ctx->Inj);
314: MatDestroy(&ctx->S);
315: PetscFree(ctx);
316: PCShellSetContext(pc, NULL);
317: return 0;
318: }
320: static PetscErrorCode CreateCR_Private(PC pc, PetscInt l, PC *cr)
321: {
322: CRContext *ctx;
325: PCCreate(PetscObjectComm((PetscObject) pc), cr);
326: PetscObjectSetName((PetscObject) *cr, "S (complementary projector to injection)");
327: PetscCalloc1(1, &ctx);
328: ctx->mg = pc;
329: ctx->l = l;
330: PCSetType(*cr, PCSHELL);
331: PCShellSetContext(*cr, ctx);
332: PCShellSetApply(*cr, CRApply_Private);
333: PCShellSetSetUp(*cr, CRSetup_Private);
334: PCShellSetDestroy(*cr, CRDestroy_Private);
335: return 0;
336: }
338: PetscErrorCode PCMGSetLevels_MG(PC pc,PetscInt levels,MPI_Comm *comms)
339: {
340: PC_MG *mg = (PC_MG*)pc->data;
341: MPI_Comm comm;
342: PC_MG_Levels **mglevels = mg->levels;
343: PCMGType mgtype = mg->am;
344: PetscInt mgctype = (PetscInt) PC_MG_CYCLE_V;
345: PetscInt i;
346: PetscMPIInt size;
347: const char *prefix;
348: PC ipc;
349: PetscInt n;
353: if (mg->nlevels == levels) return 0;
354: PetscObjectGetComm((PetscObject)pc,&comm);
355: if (mglevels) {
356: mgctype = mglevels[0]->cycles;
357: /* changing the number of levels so free up the previous stuff */
358: PCReset_MG(pc);
359: n = mglevels[0]->levels;
360: for (i=0; i<n; i++) {
361: if (mglevels[i]->smoothd != mglevels[i]->smoothu) {
362: KSPDestroy(&mglevels[i]->smoothd);
363: }
364: KSPDestroy(&mglevels[i]->smoothu);
365: KSPDestroy(&mglevels[i]->cr);
366: PetscFree(mglevels[i]);
367: }
368: PetscFree(mg->levels);
369: }
371: mg->nlevels = levels;
373: PetscMalloc1(levels,&mglevels);
374: PetscLogObjectMemory((PetscObject)pc,levels*(sizeof(PC_MG*)));
376: PCGetOptionsPrefix(pc,&prefix);
378: mg->stageApply = 0;
379: for (i=0; i<levels; i++) {
380: PetscNewLog(pc,&mglevels[i]);
382: mglevels[i]->level = i;
383: mglevels[i]->levels = levels;
384: mglevels[i]->cycles = mgctype;
385: mg->default_smoothu = 2;
386: mg->default_smoothd = 2;
387: mglevels[i]->eventsmoothsetup = 0;
388: mglevels[i]->eventsmoothsolve = 0;
389: mglevels[i]->eventresidual = 0;
390: mglevels[i]->eventinterprestrict = 0;
392: if (comms) comm = comms[i];
393: if (comm != MPI_COMM_NULL) {
394: KSPCreate(comm,&mglevels[i]->smoothd);
395: KSPSetErrorIfNotConverged(mglevels[i]->smoothd,pc->erroriffailure);
396: PetscObjectIncrementTabLevel((PetscObject)mglevels[i]->smoothd,(PetscObject)pc,levels-i);
397: KSPSetOptionsPrefix(mglevels[i]->smoothd,prefix);
398: PetscObjectComposedDataSetInt((PetscObject) mglevels[i]->smoothd, PetscMGLevelId, mglevels[i]->level);
399: if (i || levels == 1) {
400: char tprefix[128];
402: KSPSetType(mglevels[i]->smoothd,KSPCHEBYSHEV);
403: KSPSetConvergenceTest(mglevels[i]->smoothd,KSPConvergedSkip,NULL,NULL);
404: KSPSetNormType(mglevels[i]->smoothd,KSP_NORM_NONE);
405: KSPGetPC(mglevels[i]->smoothd,&ipc);
406: PCSetType(ipc,PCSOR);
407: KSPSetTolerances(mglevels[i]->smoothd,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT, mg->default_smoothd);
409: PetscSNPrintf(tprefix,128,"mg_levels_%d_",(int)i);
410: KSPAppendOptionsPrefix(mglevels[i]->smoothd,tprefix);
411: } else {
412: KSPAppendOptionsPrefix(mglevels[0]->smoothd,"mg_coarse_");
414: /* coarse solve is (redundant) LU by default; set shifttype NONZERO to avoid annoying zero-pivot in LU preconditioner */
415: KSPSetType(mglevels[0]->smoothd,KSPPREONLY);
416: KSPGetPC(mglevels[0]->smoothd,&ipc);
417: MPI_Comm_size(comm,&size);
418: if (size > 1) {
419: PCSetType(ipc,PCREDUNDANT);
420: } else {
421: PCSetType(ipc,PCLU);
422: }
423: PCFactorSetShiftType(ipc,MAT_SHIFT_INBLOCKS);
424: }
425: PetscLogObjectParent((PetscObject)pc,(PetscObject)mglevels[i]->smoothd);
426: }
427: mglevels[i]->smoothu = mglevels[i]->smoothd;
428: mg->rtol = 0.0;
429: mg->abstol = 0.0;
430: mg->dtol = 0.0;
431: mg->ttol = 0.0;
432: mg->cyclesperpcapply = 1;
433: }
434: mg->levels = mglevels;
435: PCMGSetType(pc,mgtype);
436: return 0;
437: }
439: /*@C
440: PCMGSetLevels - Sets the number of levels to use with MG.
441: Must be called before any other MG routine.
443: Logically Collective on PC
445: Input Parameters:
446: + pc - the preconditioner context
447: . levels - the number of levels
448: - comms - optional communicators for each level; this is to allow solving the coarser problems
449: on smaller sets of processes. For processes that are not included in the computation
450: you must pass MPI_COMM_NULL. Use comms = NULL to specify that all processes
451: should participate in each level of problem.
453: Level: intermediate
455: Notes:
456: If the number of levels is one then the multigrid uses the -mg_levels prefix
457: for setting the level options rather than the -mg_coarse prefix.
459: You can free the information in comms after this routine is called.
461: The array of MPI communicators must contain MPI_COMM_NULL for those ranks that at each level
462: are not participating in the coarser solve. For example, with 2 levels and 1 and 2 ranks on
463: the two levels, rank 0 in the original communicator will pass in an array of 2 communicators
464: of size 2 and 1, while rank 1 in the original communicator will pass in array of 2 communicators
465: the first of size 2 and the second of value MPI_COMM_NULL since the rank 1 does not participate
466: in the coarse grid solve.
468: Since each coarser level may have a new MPI_Comm with fewer ranks than the previous, one
469: must take special care in providing the restriction and interpolation operation. We recommend
470: providing these as two step operations; first perform a standard restriction or interpolation on
471: the full number of ranks for that level and then use an MPI call to copy the resulting vector
472: array entries (after calls to VecGetArray()) to the smaller or larger number of ranks, note in both
473: cases the MPI calls must be made on the larger of the two communicators. Traditional MPI send and
474: recieves or MPI_AlltoAllv() could be used to do the reshuffling of the vector entries.
476: Fortran Notes:
477: Use comms = PETSC_NULL_MPI_COMM as the equivalent of NULL in the C interface. Note PETSC_NULL_MPI_COMM
478: is not MPI_COMM_NULL. It is more like PETSC_NULL_INTEGER, PETSC_NULL_REAL etc.
480: .seealso: PCMGSetType(), PCMGGetLevels()
481: @*/
482: PetscErrorCode PCMGSetLevels(PC pc,PetscInt levels,MPI_Comm *comms)
483: {
486: PetscTryMethod(pc,"PCMGSetLevels_C",(PC,PetscInt,MPI_Comm*),(pc,levels,comms));
487: return 0;
488: }
490: PetscErrorCode PCDestroy_MG(PC pc)
491: {
492: PC_MG *mg = (PC_MG*)pc->data;
493: PC_MG_Levels **mglevels = mg->levels;
494: PetscInt i,n;
496: PCReset_MG(pc);
497: if (mglevels) {
498: n = mglevels[0]->levels;
499: for (i=0; i<n; i++) {
500: if (mglevels[i]->smoothd != mglevels[i]->smoothu) {
501: KSPDestroy(&mglevels[i]->smoothd);
502: }
503: KSPDestroy(&mglevels[i]->smoothu);
504: KSPDestroy(&mglevels[i]->cr);
505: PetscFree(mglevels[i]);
506: }
507: PetscFree(mg->levels);
508: }
509: PetscFree(pc->data);
510: PetscObjectComposeFunction((PetscObject)pc,"PCGetInterpolations_C",NULL);
511: PetscObjectComposeFunction((PetscObject)pc,"PCGetCoarseOperators_C",NULL);
512: return 0;
513: }
515: /*
516: PCApply_MG - Runs either an additive, multiplicative, Kaskadic
517: or full cycle of multigrid.
519: Note:
520: A simple wrapper which calls PCMGMCycle(),PCMGACycle(), or PCMGFCycle().
521: */
522: static PetscErrorCode PCApply_MG_Internal(PC pc,Vec b,Vec x,Mat B,Mat X,PetscBool transpose)
523: {
524: PC_MG *mg = (PC_MG*)pc->data;
525: PC_MG_Levels **mglevels = mg->levels;
526: PC tpc;
527: PetscInt levels = mglevels[0]->levels,i;
528: PetscBool changeu,changed,matapp;
530: matapp = (PetscBool)(B && X);
531: if (mg->stageApply) PetscLogStagePush(mg->stageApply);
532: /* When the DM is supplying the matrix then it will not exist until here */
533: for (i=0; i<levels; i++) {
534: if (!mglevels[i]->A) {
535: KSPGetOperators(mglevels[i]->smoothu,&mglevels[i]->A,NULL);
536: PetscObjectReference((PetscObject)mglevels[i]->A);
537: }
538: }
540: KSPGetPC(mglevels[levels-1]->smoothd,&tpc);
541: PCPreSolveChangeRHS(tpc,&changed);
542: KSPGetPC(mglevels[levels-1]->smoothu,&tpc);
543: PCPreSolveChangeRHS(tpc,&changeu);
544: if (!changeu && !changed) {
545: if (matapp) {
546: MatDestroy(&mglevels[levels-1]->B);
547: mglevels[levels-1]->B = B;
548: } else {
549: VecDestroy(&mglevels[levels-1]->b);
550: mglevels[levels-1]->b = b;
551: }
552: } else { /* if the smoother changes the rhs during PreSolve, we cannot use the input vector */
553: if (matapp) {
554: if (mglevels[levels-1]->B) {
555: PetscInt N1,N2;
556: PetscBool flg;
558: MatGetSize(mglevels[levels-1]->B,NULL,&N1);
559: MatGetSize(B,NULL,&N2);
560: PetscObjectTypeCompare((PetscObject)mglevels[levels-1]->B,((PetscObject)B)->type_name,&flg);
561: if (N1 != N2 || !flg) {
562: MatDestroy(&mglevels[levels-1]->B);
563: }
564: }
565: if (!mglevels[levels-1]->B) {
566: MatDuplicate(B,MAT_COPY_VALUES,&mglevels[levels-1]->B);
567: } else {
568: MatCopy(B,mglevels[levels-1]->B,SAME_NONZERO_PATTERN);
569: }
570: } else {
571: if (!mglevels[levels-1]->b) {
572: Vec *vec;
574: KSPCreateVecs(mglevels[levels-1]->smoothd,1,&vec,0,NULL);
575: mglevels[levels-1]->b = *vec;
576: PetscFree(vec);
577: }
578: VecCopy(b,mglevels[levels-1]->b);
579: }
580: }
581: if (matapp) { mglevels[levels-1]->X = X; }
582: else { mglevels[levels-1]->x = x; }
584: /* If coarser Xs are present, it means we have already block applied the PC at least once
585: Reset operators if sizes/type do no match */
586: if (matapp && levels > 1 && mglevels[levels-2]->X) {
587: PetscInt Xc,Bc;
588: PetscBool flg;
590: MatGetSize(mglevels[levels-2]->X,NULL,&Xc);
591: MatGetSize(mglevels[levels-1]->B,NULL,&Bc);
592: PetscObjectTypeCompare((PetscObject)mglevels[levels-2]->X,((PetscObject)mglevels[levels-1]->X)->type_name,&flg);
593: if (Xc != Bc || !flg) {
594: MatDestroy(&mglevels[levels-1]->R);
595: for (i=0;i<levels-1;i++) {
596: MatDestroy(&mglevels[i]->R);
597: MatDestroy(&mglevels[i]->B);
598: MatDestroy(&mglevels[i]->X);
599: }
600: }
601: }
603: if (mg->am == PC_MG_MULTIPLICATIVE) {
604: if (matapp) MatZeroEntries(X);
605: else VecZeroEntries(x);
606: for (i=0; i<mg->cyclesperpcapply; i++) {
607: PCMGMCycle_Private(pc,mglevels+levels-1,transpose,matapp,NULL);
608: }
609: } else if (mg->am == PC_MG_ADDITIVE) {
610: PCMGACycle_Private(pc,mglevels,transpose,matapp);
611: } else if (mg->am == PC_MG_KASKADE) {
612: PCMGKCycle_Private(pc,mglevels,transpose,matapp);
613: } else {
614: PCMGFCycle_Private(pc,mglevels,transpose,matapp);
615: }
616: if (mg->stageApply) PetscLogStagePop();
617: if (!changeu && !changed) {
618: if (matapp) { mglevels[levels-1]->B = NULL; }
619: else { mglevels[levels-1]->b = NULL; }
620: }
621: return 0;
622: }
624: static PetscErrorCode PCApply_MG(PC pc,Vec b,Vec x)
625: {
626: PCApply_MG_Internal(pc,b,x,NULL,NULL,PETSC_FALSE);
627: return 0;
628: }
630: static PetscErrorCode PCApplyTranspose_MG(PC pc,Vec b,Vec x)
631: {
632: PCApply_MG_Internal(pc,b,x,NULL,NULL,PETSC_TRUE);
633: return 0;
634: }
636: static PetscErrorCode PCMatApply_MG(PC pc,Mat b,Mat x)
637: {
638: PCApply_MG_Internal(pc,NULL,NULL,b,x,PETSC_FALSE);
639: return 0;
640: }
642: PetscErrorCode PCSetFromOptions_MG(PetscOptionItems *PetscOptionsObject,PC pc)
643: {
644: PetscInt levels,cycles;
645: PetscBool flg, flg2;
646: PC_MG *mg = (PC_MG*)pc->data;
647: PC_MG_Levels **mglevels;
648: PCMGType mgtype;
649: PCMGCycleType mgctype;
650: PCMGGalerkinType gtype;
652: levels = PetscMax(mg->nlevels,1);
653: PetscOptionsHead(PetscOptionsObject,"Multigrid options");
654: PetscOptionsInt("-pc_mg_levels","Number of Levels","PCMGSetLevels",levels,&levels,&flg);
655: if (!flg && !mg->levels && pc->dm) {
656: DMGetRefineLevel(pc->dm,&levels);
657: levels++;
658: mg->usedmfornumberoflevels = PETSC_TRUE;
659: }
660: PCMGSetLevels(pc,levels,NULL);
661: mglevels = mg->levels;
663: mgctype = (PCMGCycleType) mglevels[0]->cycles;
664: PetscOptionsEnum("-pc_mg_cycle_type","V cycle or for W-cycle","PCMGSetCycleType",PCMGCycleTypes,(PetscEnum)mgctype,(PetscEnum*)&mgctype,&flg);
665: if (flg) {
666: PCMGSetCycleType(pc,mgctype);
667: }
668: gtype = mg->galerkin;
669: PetscOptionsEnum("-pc_mg_galerkin","Use Galerkin process to compute coarser operators","PCMGSetGalerkin",PCMGGalerkinTypes,(PetscEnum)gtype,(PetscEnum*)>ype,&flg);
670: if (flg) {
671: PCMGSetGalerkin(pc,gtype);
672: }
673: flg2 = PETSC_FALSE;
674: PetscOptionsBool("-pc_mg_adapt_interp","Adapt interpolation using some coarse space","PCMGSetAdaptInterpolation",PETSC_FALSE,&flg2,&flg);
675: if (flg) PCMGSetAdaptInterpolation(pc, flg2);
676: PetscOptionsInt("-pc_mg_adapt_interp_n","Size of the coarse space for adaptive interpolation","PCMGSetCoarseSpace",mg->Nc,&mg->Nc,&flg);
677: PetscOptionsEnum("-pc_mg_adapt_interp_coarse_space","Type of coarse space: polynomial, harmonic, eigenvector, generalized_eigenvector","PCMGSetAdaptCoarseSpaceType",PCMGCoarseSpaceTypes,(PetscEnum)mg->coarseSpaceType,(PetscEnum*)&mg->coarseSpaceType,&flg);
678: PetscOptionsBool("-pc_mg_mesp_monitor","Monitor the multilevel eigensolver","PCMGSetAdaptInterpolation",PETSC_FALSE,&mg->mespMonitor,&flg);
679: flg2 = PETSC_FALSE;
680: PetscOptionsBool("-pc_mg_adapt_cr","Monitor coarse space quality using Compatible Relaxation (CR)","PCMGSetAdaptCR",PETSC_FALSE,&flg2,&flg);
681: if (flg) PCMGSetAdaptCR(pc, flg2);
682: flg = PETSC_FALSE;
683: PetscOptionsBool("-pc_mg_distinct_smoothup","Create separate smoothup KSP and append the prefix _up","PCMGSetDistinctSmoothUp",PETSC_FALSE,&flg,NULL);
684: if (flg) {
685: PCMGSetDistinctSmoothUp(pc);
686: }
687: mgtype = mg->am;
688: PetscOptionsEnum("-pc_mg_type","Multigrid type","PCMGSetType",PCMGTypes,(PetscEnum)mgtype,(PetscEnum*)&mgtype,&flg);
689: if (flg) {
690: PCMGSetType(pc,mgtype);
691: }
692: if (mg->am == PC_MG_MULTIPLICATIVE) {
693: PetscOptionsInt("-pc_mg_multiplicative_cycles","Number of cycles for each preconditioner step","PCMGMultiplicativeSetCycles",mg->cyclesperpcapply,&cycles,&flg);
694: if (flg) {
695: PCMGMultiplicativeSetCycles(pc,cycles);
696: }
697: }
698: flg = PETSC_FALSE;
699: PetscOptionsBool("-pc_mg_log","Log times for each multigrid level","None",flg,&flg,NULL);
700: if (flg) {
701: PetscInt i;
702: char eventname[128];
704: levels = mglevels[0]->levels;
705: for (i=0; i<levels; i++) {
706: sprintf(eventname,"MGSetup Level %d",(int)i);
707: PetscLogEventRegister(eventname,((PetscObject)pc)->classid,&mglevels[i]->eventsmoothsetup);
708: sprintf(eventname,"MGSmooth Level %d",(int)i);
709: PetscLogEventRegister(eventname,((PetscObject)pc)->classid,&mglevels[i]->eventsmoothsolve);
710: if (i) {
711: sprintf(eventname,"MGResid Level %d",(int)i);
712: PetscLogEventRegister(eventname,((PetscObject)pc)->classid,&mglevels[i]->eventresidual);
713: sprintf(eventname,"MGInterp Level %d",(int)i);
714: PetscLogEventRegister(eventname,((PetscObject)pc)->classid,&mglevels[i]->eventinterprestrict);
715: }
716: }
718: #if defined(PETSC_USE_LOG)
719: {
720: const char *sname = "MG Apply";
721: PetscStageLog stageLog;
722: PetscInt st;
724: PetscLogGetStageLog(&stageLog);
725: for (st = 0; st < stageLog->numStages; ++st) {
726: PetscBool same;
728: PetscStrcmp(stageLog->stageInfo[st].name, sname, &same);
729: if (same) mg->stageApply = st;
730: }
731: if (!mg->stageApply) {
732: PetscLogStageRegister(sname, &mg->stageApply);
733: }
734: }
735: #endif
736: }
737: PetscOptionsTail();
738: /* Check option consistency */
739: PCMGGetGalerkin(pc, >ype);
740: PCMGGetAdaptInterpolation(pc, &flg);
742: return 0;
743: }
745: const char *const PCMGTypes[] = {"MULTIPLICATIVE","ADDITIVE","FULL","KASKADE","PCMGType","PC_MG",NULL};
746: const char *const PCMGCycleTypes[] = {"invalid","v","w","PCMGCycleType","PC_MG_CYCLE",NULL};
747: const char *const PCMGGalerkinTypes[] = {"both","pmat","mat","none","external","PCMGGalerkinType","PC_MG_GALERKIN",NULL};
748: const char *const PCMGCoarseSpaceTypes[] = {"polynomial","harmonic","eigenvector","generalized_eigenvector","PCMGCoarseSpaceType","PCMG_POLYNOMIAL",NULL};
750: #include <petscdraw.h>
751: PetscErrorCode PCView_MG(PC pc,PetscViewer viewer)
752: {
753: PC_MG *mg = (PC_MG*)pc->data;
754: PC_MG_Levels **mglevels = mg->levels;
755: PetscInt levels = mglevels ? mglevels[0]->levels : 0,i;
756: PetscBool iascii,isbinary,isdraw;
758: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);
759: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);
760: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);
761: if (iascii) {
762: const char *cyclename = levels ? (mglevels[0]->cycles == PC_MG_CYCLE_V ? "v" : "w") : "unknown";
763: PetscViewerASCIIPrintf(viewer," type is %s, levels=%D cycles=%s\n", PCMGTypes[mg->am],levels,cyclename);
764: if (mg->am == PC_MG_MULTIPLICATIVE) {
765: PetscViewerASCIIPrintf(viewer," Cycles per PCApply=%d\n",mg->cyclesperpcapply);
766: }
767: if (mg->galerkin == PC_MG_GALERKIN_BOTH) {
768: PetscViewerASCIIPrintf(viewer," Using Galerkin computed coarse grid matrices\n");
769: } else if (mg->galerkin == PC_MG_GALERKIN_PMAT) {
770: PetscViewerASCIIPrintf(viewer," Using Galerkin computed coarse grid matrices for pmat\n");
771: } else if (mg->galerkin == PC_MG_GALERKIN_MAT) {
772: PetscViewerASCIIPrintf(viewer," Using Galerkin computed coarse grid matrices for mat\n");
773: } else if (mg->galerkin == PC_MG_GALERKIN_EXTERNAL) {
774: PetscViewerASCIIPrintf(viewer," Using externally compute Galerkin coarse grid matrices\n");
775: } else {
776: PetscViewerASCIIPrintf(viewer," Not using Galerkin computed coarse grid matrices\n");
777: }
778: if (mg->view) {
779: (*mg->view)(pc,viewer);
780: }
781: for (i=0; i<levels; i++) {
782: if (!i) {
783: PetscViewerASCIIPrintf(viewer,"Coarse grid solver -- level -------------------------------\n",i);
784: } else {
785: PetscViewerASCIIPrintf(viewer,"Down solver (pre-smoother) on level %D -------------------------------\n",i);
786: }
787: PetscViewerASCIIPushTab(viewer);
788: KSPView(mglevels[i]->smoothd,viewer);
789: PetscViewerASCIIPopTab(viewer);
790: if (i && mglevels[i]->smoothd == mglevels[i]->smoothu) {
791: PetscViewerASCIIPrintf(viewer,"Up solver (post-smoother) same as down solver (pre-smoother)\n");
792: } else if (i) {
793: PetscViewerASCIIPrintf(viewer,"Up solver (post-smoother) on level %D -------------------------------\n",i);
794: PetscViewerASCIIPushTab(viewer);
795: KSPView(mglevels[i]->smoothu,viewer);
796: PetscViewerASCIIPopTab(viewer);
797: }
798: if (i && mglevels[i]->cr) {
799: PetscViewerASCIIPrintf(viewer,"CR solver on level %D -------------------------------\n",i);
800: PetscViewerASCIIPushTab(viewer);
801: KSPView(mglevels[i]->cr,viewer);
802: PetscViewerASCIIPopTab(viewer);
803: }
804: }
805: } else if (isbinary) {
806: for (i=levels-1; i>=0; i--) {
807: KSPView(mglevels[i]->smoothd,viewer);
808: if (i && mglevels[i]->smoothd != mglevels[i]->smoothu) {
809: KSPView(mglevels[i]->smoothu,viewer);
810: }
811: }
812: } else if (isdraw) {
813: PetscDraw draw;
814: PetscReal x,w,y,bottom,th;
815: PetscViewerDrawGetDraw(viewer,0,&draw);
816: PetscDrawGetCurrentPoint(draw,&x,&y);
817: PetscDrawStringGetSize(draw,NULL,&th);
818: bottom = y - th;
819: for (i=levels-1; i>=0; i--) {
820: if (!mglevels[i]->smoothu || (mglevels[i]->smoothu == mglevels[i]->smoothd)) {
821: PetscDrawPushCurrentPoint(draw,x,bottom);
822: KSPView(mglevels[i]->smoothd,viewer);
823: PetscDrawPopCurrentPoint(draw);
824: } else {
825: w = 0.5*PetscMin(1.0-x,x);
826: PetscDrawPushCurrentPoint(draw,x+w,bottom);
827: KSPView(mglevels[i]->smoothd,viewer);
828: PetscDrawPopCurrentPoint(draw);
829: PetscDrawPushCurrentPoint(draw,x-w,bottom);
830: KSPView(mglevels[i]->smoothu,viewer);
831: PetscDrawPopCurrentPoint(draw);
832: }
833: PetscDrawGetBoundingBox(draw,NULL,&bottom,NULL,NULL);
834: bottom -= th;
835: }
836: }
837: return 0;
838: }
840: #include <petsc/private/kspimpl.h>
842: /*
843: Calls setup for the KSP on each level
844: */
845: PetscErrorCode PCSetUp_MG(PC pc)
846: {
847: PC_MG *mg = (PC_MG*)pc->data;
848: PC_MG_Levels **mglevels = mg->levels;
849: PetscInt i,n;
850: PC cpc;
851: PetscBool dump = PETSC_FALSE,opsset,use_amat,missinginterpolate = PETSC_FALSE;
852: Mat dA,dB;
853: Vec tvec;
854: DM *dms;
855: PetscViewer viewer = NULL;
856: PetscBool dAeqdB = PETSC_FALSE, needRestricts = PETSC_FALSE, doCR = PETSC_FALSE;
859: n = mglevels[0]->levels;
860: /* FIX: Move this to PCSetFromOptions_MG? */
861: if (mg->usedmfornumberoflevels) {
862: PetscInt levels;
863: DMGetRefineLevel(pc->dm,&levels);
864: levels++;
865: if (levels > n) { /* the problem is now being solved on a finer grid */
866: PCMGSetLevels(pc,levels,NULL);
867: n = levels;
868: PCSetFromOptions(pc); /* it is bad to call this here, but otherwise will never be called for the new hierarchy */
869: mglevels = mg->levels;
870: }
871: }
872: KSPGetPC(mglevels[0]->smoothd,&cpc);
874: /* If user did not provide fine grid operators OR operator was not updated since last global KSPSetOperators() */
875: /* so use those from global PC */
876: /* Is this what we always want? What if user wants to keep old one? */
877: KSPGetOperatorsSet(mglevels[n-1]->smoothd,NULL,&opsset);
878: if (opsset) {
879: Mat mmat;
880: KSPGetOperators(mglevels[n-1]->smoothd,NULL,&mmat);
881: if (mmat == pc->pmat) opsset = PETSC_FALSE;
882: }
884: /* Create CR solvers */
885: PCMGGetAdaptCR(pc, &doCR);
886: if (doCR) {
887: const char *prefix;
889: PCGetOptionsPrefix(pc, &prefix);
890: for (i = 1; i < n; ++i) {
891: PC ipc, cr;
892: char crprefix[128];
894: KSPCreate(PetscObjectComm((PetscObject) pc), &mglevels[i]->cr);
895: KSPSetErrorIfNotConverged(mglevels[i]->cr, PETSC_FALSE);
896: PetscObjectIncrementTabLevel((PetscObject) mglevels[i]->cr, (PetscObject) pc, n-i);
897: KSPSetOptionsPrefix(mglevels[i]->cr, prefix);
898: PetscObjectComposedDataSetInt((PetscObject) mglevels[i]->cr, PetscMGLevelId, mglevels[i]->level);
899: KSPSetType(mglevels[i]->cr, KSPCHEBYSHEV);
900: KSPSetConvergenceTest(mglevels[i]->cr, KSPConvergedSkip, NULL, NULL);
901: KSPSetNormType(mglevels[i]->cr, KSP_NORM_PRECONDITIONED);
902: KSPGetPC(mglevels[i]->cr, &ipc);
904: PCSetType(ipc, PCCOMPOSITE);
905: PCCompositeSetType(ipc, PC_COMPOSITE_MULTIPLICATIVE);
906: PCCompositeAddPCType(ipc, PCSOR);
907: CreateCR_Private(pc, i, &cr);
908: PCCompositeAddPC(ipc, cr);
909: PCDestroy(&cr);
911: KSPSetTolerances(mglevels[i]->cr, PETSC_DEFAULT, PETSC_DEFAULT, PETSC_DEFAULT, mg->default_smoothd);
912: KSPSetInitialGuessNonzero(mglevels[i]->cr, PETSC_TRUE);
913: PetscSNPrintf(crprefix, 128, "mg_levels_%d_cr_", (int) i);
914: KSPAppendOptionsPrefix(mglevels[i]->cr, crprefix);
915: PetscLogObjectParent((PetscObject) pc, (PetscObject) mglevels[i]->cr);
916: }
917: }
919: if (!opsset) {
920: PCGetUseAmat(pc,&use_amat);
921: if (use_amat) {
922: PetscInfo(pc,"Using outer operators to define finest grid operator \n because PCMGGetSmoother(pc,nlevels-1,&ksp);KSPSetOperators(ksp,...); was not called.\n");
923: KSPSetOperators(mglevels[n-1]->smoothd,pc->mat,pc->pmat);
924: } else {
925: PetscInfo(pc,"Using matrix (pmat) operators to define finest grid operator \n because PCMGGetSmoother(pc,nlevels-1,&ksp);KSPSetOperators(ksp,...); was not called.\n");
926: KSPSetOperators(mglevels[n-1]->smoothd,pc->pmat,pc->pmat);
927: }
928: }
930: for (i=n-1; i>0; i--) {
931: if (!(mglevels[i]->interpolate || mglevels[i]->restrct)) {
932: missinginterpolate = PETSC_TRUE;
933: continue;
934: }
935: }
937: KSPGetOperators(mglevels[n-1]->smoothd,&dA,&dB);
938: if (dA == dB) dAeqdB = PETSC_TRUE;
939: if ((mg->galerkin == PC_MG_GALERKIN_NONE) || (((mg->galerkin == PC_MG_GALERKIN_PMAT) || (mg->galerkin == PC_MG_GALERKIN_MAT)) && !dAeqdB)) {
940: needRestricts = PETSC_TRUE; /* user must compute either mat, pmat, or both so must restrict x to coarser levels */
941: }
943: /*
944: Skipping if user has provided all interpolation/restriction needed (since DM might not be able to produce them (when coming from SNES/TS)
945: Skipping for galerkin==2 (externally managed hierarchy such as ML and GAMG). Cleaner logic here would be great. Wrap ML/GAMG as DMs?
946: */
947: if (missinginterpolate && pc->dm && mg->galerkin != PC_MG_GALERKIN_EXTERNAL && !pc->setupcalled) {
948: /* construct the interpolation from the DMs */
949: Mat p;
950: Vec rscale;
951: PetscMalloc1(n,&dms);
952: dms[n-1] = pc->dm;
953: /* Separately create them so we do not get DMKSP interference between levels */
954: for (i=n-2; i>-1; i--) DMCoarsen(dms[i+1],MPI_COMM_NULL,&dms[i]);
955: /*
956: Force the mat type of coarse level operator to be AIJ because usually we want to use LU for coarse level.
957: Notice that it can be overwritten by -mat_type because KSPSetUp() reads command line options.
958: But it is safe to use -dm_mat_type.
960: The mat type should not be hardcoded like this, we need to find a better way.
961: DMSetMatType(dms[0],MATAIJ);
962: */
963: for (i=n-2; i>-1; i--) {
964: DMKSP kdm;
965: PetscBool dmhasrestrict, dmhasinject;
966: KSPSetDM(mglevels[i]->smoothd,dms[i]);
967: if (!needRestricts) KSPSetDMActive(mglevels[i]->smoothd,PETSC_FALSE);
968: if (mglevels[i]->smoothd != mglevels[i]->smoothu) {
969: KSPSetDM(mglevels[i]->smoothu,dms[i]);
970: if (!needRestricts) KSPSetDMActive(mglevels[i]->smoothu,PETSC_FALSE);
971: }
972: if (mglevels[i]->cr) {
973: KSPSetDM(mglevels[i]->cr,dms[i]);
974: if (!needRestricts) KSPSetDMActive(mglevels[i]->cr,PETSC_FALSE);
975: }
976: DMGetDMKSPWrite(dms[i],&kdm);
977: /* Ugly hack so that the next KSPSetUp() will use the RHS that we set. A better fix is to change dmActive to take
978: * a bitwise OR of computing the matrix, RHS, and initial iterate. */
979: kdm->ops->computerhs = NULL;
980: kdm->rhsctx = NULL;
981: if (!mglevels[i+1]->interpolate) {
982: DMCreateInterpolation(dms[i],dms[i+1],&p,&rscale);
983: PCMGSetInterpolation(pc,i+1,p);
984: if (rscale) PCMGSetRScale(pc,i+1,rscale);
985: VecDestroy(&rscale);
986: MatDestroy(&p);
987: }
988: DMHasCreateRestriction(dms[i],&dmhasrestrict);
989: if (dmhasrestrict && !mglevels[i+1]->restrct) {
990: DMCreateRestriction(dms[i],dms[i+1],&p);
991: PCMGSetRestriction(pc,i+1,p);
992: MatDestroy(&p);
993: }
994: DMHasCreateInjection(dms[i],&dmhasinject);
995: if (dmhasinject && !mglevels[i+1]->inject) {
996: DMCreateInjection(dms[i],dms[i+1],&p);
997: PCMGSetInjection(pc,i+1,p);
998: MatDestroy(&p);
999: }
1000: }
1002: for (i=n-2; i>-1; i--) DMDestroy(&dms[i]);
1003: PetscFree(dms);
1004: }
1006: if (pc->dm && !pc->setupcalled) {
1007: /* finest smoother also gets DM but it is not active, independent of whether galerkin==PC_MG_GALERKIN_EXTERNAL */
1008: KSPSetDM(mglevels[n-1]->smoothd,pc->dm);
1009: KSPSetDMActive(mglevels[n-1]->smoothd,PETSC_FALSE);
1010: if (mglevels[n-1]->smoothd != mglevels[n-1]->smoothu) {
1011: KSPSetDM(mglevels[n-1]->smoothu,pc->dm);
1012: KSPSetDMActive(mglevels[n-1]->smoothu,PETSC_FALSE);
1013: }
1014: if (mglevels[n-1]->cr) {
1015: KSPSetDM(mglevels[n-1]->cr,pc->dm);
1016: KSPSetDMActive(mglevels[n-1]->cr,PETSC_FALSE);
1017: }
1018: }
1020: if (mg->galerkin < PC_MG_GALERKIN_NONE) {
1021: Mat A,B;
1022: PetscBool doA = PETSC_FALSE,doB = PETSC_FALSE;
1023: MatReuse reuse = MAT_INITIAL_MATRIX;
1025: if ((mg->galerkin == PC_MG_GALERKIN_PMAT) || (mg->galerkin == PC_MG_GALERKIN_BOTH)) doB = PETSC_TRUE;
1026: if ((mg->galerkin == PC_MG_GALERKIN_MAT) || ((mg->galerkin == PC_MG_GALERKIN_BOTH) && (dA != dB))) doA = PETSC_TRUE;
1027: if (pc->setupcalled) reuse = MAT_REUSE_MATRIX;
1028: for (i=n-2; i>-1; i--) {
1030: if (!mglevels[i+1]->interpolate) {
1031: PCMGSetInterpolation(pc,i+1,mglevels[i+1]->restrct);
1032: }
1033: if (!mglevels[i+1]->restrct) {
1034: PCMGSetRestriction(pc,i+1,mglevels[i+1]->interpolate);
1035: }
1036: if (reuse == MAT_REUSE_MATRIX) {
1037: KSPGetOperators(mglevels[i]->smoothd,&A,&B);
1038: }
1039: if (doA) {
1040: MatGalerkin(mglevels[i+1]->restrct,dA,mglevels[i+1]->interpolate,reuse,1.0,&A);
1041: }
1042: if (doB) {
1043: MatGalerkin(mglevels[i+1]->restrct,dB,mglevels[i+1]->interpolate,reuse,1.0,&B);
1044: }
1045: /* the management of the PetscObjectReference() and PetscObjecDereference() below is rather delicate */
1046: if (!doA && dAeqdB) {
1047: if (reuse == MAT_INITIAL_MATRIX) PetscObjectReference((PetscObject)B);
1048: A = B;
1049: } else if (!doA && reuse == MAT_INITIAL_MATRIX) {
1050: KSPGetOperators(mglevels[i]->smoothd,&A,NULL);
1051: PetscObjectReference((PetscObject)A);
1052: }
1053: if (!doB && dAeqdB) {
1054: if (reuse == MAT_INITIAL_MATRIX) PetscObjectReference((PetscObject)A);
1055: B = A;
1056: } else if (!doB && reuse == MAT_INITIAL_MATRIX) {
1057: KSPGetOperators(mglevels[i]->smoothd,NULL,&B);
1058: PetscObjectReference((PetscObject)B);
1059: }
1060: if (reuse == MAT_INITIAL_MATRIX) {
1061: KSPSetOperators(mglevels[i]->smoothd,A,B);
1062: PetscObjectDereference((PetscObject)A);
1063: PetscObjectDereference((PetscObject)B);
1064: }
1065: dA = A;
1066: dB = B;
1067: }
1068: }
1070: /* Adapt interpolation matrices */
1071: if (mg->adaptInterpolation) {
1072: mg->Nc = mg->Nc < 0 ? 6 : mg->Nc; /* Default to 6 modes */
1073: for (i = 0; i < n; ++i) {
1074: PCMGComputeCoarseSpace_Internal(pc, i, mg->coarseSpaceType, mg->Nc, !i ? NULL : mglevels[i-1]->coarseSpace, &mglevels[i]->coarseSpace);
1075: if (i) PCMGAdaptInterpolator_Internal(pc, i, mglevels[i-1]->smoothu, mglevels[i]->smoothu, mg->Nc, mglevels[i-1]->coarseSpace, mglevels[i]->coarseSpace);
1076: }
1077: for (i = n-2; i > -1; --i) {
1078: PCMGRecomputeLevelOperators_Internal(pc, i);
1079: }
1080: }
1082: if (needRestricts && pc->dm) {
1083: for (i=n-2; i>=0; i--) {
1084: DM dmfine,dmcoarse;
1085: Mat Restrict,Inject;
1086: Vec rscale;
1087: KSPGetDM(mglevels[i+1]->smoothd,&dmfine);
1088: KSPGetDM(mglevels[i]->smoothd,&dmcoarse);
1089: PCMGGetRestriction(pc,i+1,&Restrict);
1090: PCMGGetRScale(pc,i+1,&rscale);
1091: PCMGGetInjection(pc,i+1,&Inject);
1092: DMRestrict(dmfine,Restrict,rscale,Inject,dmcoarse);
1093: }
1094: }
1096: if (!pc->setupcalled) {
1097: for (i=0; i<n; i++) {
1098: KSPSetFromOptions(mglevels[i]->smoothd);
1099: }
1100: for (i=1; i<n; i++) {
1101: if (mglevels[i]->smoothu && (mglevels[i]->smoothu != mglevels[i]->smoothd)) {
1102: KSPSetFromOptions(mglevels[i]->smoothu);
1103: }
1104: if (mglevels[i]->cr) {
1105: KSPSetFromOptions(mglevels[i]->cr);
1106: }
1107: }
1108: /* insure that if either interpolation or restriction is set the other other one is set */
1109: for (i=1; i<n; i++) {
1110: PCMGGetInterpolation(pc,i,NULL);
1111: PCMGGetRestriction(pc,i,NULL);
1112: }
1113: for (i=0; i<n-1; i++) {
1114: if (!mglevels[i]->b) {
1115: Vec *vec;
1116: KSPCreateVecs(mglevels[i]->smoothd,1,&vec,0,NULL);
1117: PCMGSetRhs(pc,i,*vec);
1118: VecDestroy(vec);
1119: PetscFree(vec);
1120: }
1121: if (!mglevels[i]->r && i) {
1122: VecDuplicate(mglevels[i]->b,&tvec);
1123: PCMGSetR(pc,i,tvec);
1124: VecDestroy(&tvec);
1125: }
1126: if (!mglevels[i]->x) {
1127: VecDuplicate(mglevels[i]->b,&tvec);
1128: PCMGSetX(pc,i,tvec);
1129: VecDestroy(&tvec);
1130: }
1131: if (doCR) {
1132: VecDuplicate(mglevels[i]->b,&mglevels[i]->crx);
1133: VecDuplicate(mglevels[i]->b,&mglevels[i]->crb);
1134: VecZeroEntries(mglevels[i]->crb);
1135: }
1136: }
1137: if (n != 1 && !mglevels[n-1]->r) {
1138: /* PCMGSetR() on the finest level if user did not supply it */
1139: Vec *vec;
1140: KSPCreateVecs(mglevels[n-1]->smoothd,1,&vec,0,NULL);
1141: PCMGSetR(pc,n-1,*vec);
1142: VecDestroy(vec);
1143: PetscFree(vec);
1144: }
1145: if (doCR) {
1146: VecDuplicate(mglevels[n-1]->r, &mglevels[n-1]->crx);
1147: VecDuplicate(mglevels[n-1]->r, &mglevels[n-1]->crb);
1148: VecZeroEntries(mglevels[n-1]->crb);
1149: }
1150: }
1152: if (pc->dm) {
1153: /* need to tell all the coarser levels to rebuild the matrix using the DM for that level */
1154: for (i=0; i<n-1; i++) {
1155: if (mglevels[i]->smoothd->setupstage != KSP_SETUP_NEW) mglevels[i]->smoothd->setupstage = KSP_SETUP_NEWMATRIX;
1156: }
1157: }
1158: // We got here (PCSetUp_MG) because the matrix has changed, which means the smoother needs to be set up again (e.g.,
1159: // new diagonal for Jacobi). Setting it here allows it to be logged under PCSetUp rather than deep inside a PCApply.
1160: if (mglevels[n-1]->smoothd->setupstage != KSP_SETUP_NEW) mglevels[n-1]->smoothd->setupstage = KSP_SETUP_NEWMATRIX;
1162: for (i=1; i<n; i++) {
1163: if (mglevels[i]->smoothu == mglevels[i]->smoothd || mg->am == PC_MG_FULL || mg->am == PC_MG_KASKADE || mg->cyclesperpcapply > 1) {
1164: /* if doing only down then initial guess is zero */
1165: KSPSetInitialGuessNonzero(mglevels[i]->smoothd,PETSC_TRUE);
1166: }
1167: if (mglevels[i]->cr) KSPSetInitialGuessNonzero(mglevels[i]->cr,PETSC_TRUE);
1168: if (mglevels[i]->eventsmoothsetup) PetscLogEventBegin(mglevels[i]->eventsmoothsetup,0,0,0,0);
1169: KSPSetUp(mglevels[i]->smoothd);
1170: if (mglevels[i]->smoothd->reason == KSP_DIVERGED_PC_FAILED) {
1171: pc->failedreason = PC_SUBPC_ERROR;
1172: }
1173: if (mglevels[i]->eventsmoothsetup) PetscLogEventEnd(mglevels[i]->eventsmoothsetup,0,0,0,0);
1174: if (!mglevels[i]->residual) {
1175: Mat mat;
1176: KSPGetOperators(mglevels[i]->smoothd,&mat,NULL);
1177: PCMGSetResidual(pc,i,PCMGResidualDefault,mat);
1178: }
1179: if (!mglevels[i]->residualtranspose) {
1180: Mat mat;
1181: KSPGetOperators(mglevels[i]->smoothd,&mat,NULL);
1182: PCMGSetResidualTranspose(pc,i,PCMGResidualTransposeDefault,mat);
1183: }
1184: }
1185: for (i=1; i<n; i++) {
1186: if (mglevels[i]->smoothu && mglevels[i]->smoothu != mglevels[i]->smoothd) {
1187: Mat downmat,downpmat;
1189: /* check if operators have been set for up, if not use down operators to set them */
1190: KSPGetOperatorsSet(mglevels[i]->smoothu,&opsset,NULL);
1191: if (!opsset) {
1192: KSPGetOperators(mglevels[i]->smoothd,&downmat,&downpmat);
1193: KSPSetOperators(mglevels[i]->smoothu,downmat,downpmat);
1194: }
1196: KSPSetInitialGuessNonzero(mglevels[i]->smoothu,PETSC_TRUE);
1197: if (mglevels[i]->eventsmoothsetup) PetscLogEventBegin(mglevels[i]->eventsmoothsetup,0,0,0,0);
1198: KSPSetUp(mglevels[i]->smoothu);
1199: if (mglevels[i]->smoothu->reason == KSP_DIVERGED_PC_FAILED) {
1200: pc->failedreason = PC_SUBPC_ERROR;
1201: }
1202: if (mglevels[i]->eventsmoothsetup) PetscLogEventEnd(mglevels[i]->eventsmoothsetup,0,0,0,0);
1203: }
1204: if (mglevels[i]->cr) {
1205: Mat downmat,downpmat;
1207: /* check if operators have been set for up, if not use down operators to set them */
1208: KSPGetOperatorsSet(mglevels[i]->cr,&opsset,NULL);
1209: if (!opsset) {
1210: KSPGetOperators(mglevels[i]->smoothd,&downmat,&downpmat);
1211: KSPSetOperators(mglevels[i]->cr,downmat,downpmat);
1212: }
1214: KSPSetInitialGuessNonzero(mglevels[i]->cr,PETSC_TRUE);
1215: if (mglevels[i]->eventsmoothsetup) PetscLogEventBegin(mglevels[i]->eventsmoothsetup,0,0,0,0);
1216: KSPSetUp(mglevels[i]->cr);
1217: if (mglevels[i]->cr->reason == KSP_DIVERGED_PC_FAILED) {
1218: pc->failedreason = PC_SUBPC_ERROR;
1219: }
1220: if (mglevels[i]->eventsmoothsetup) PetscLogEventEnd(mglevels[i]->eventsmoothsetup,0,0,0,0);
1221: }
1222: }
1224: if (mglevels[0]->eventsmoothsetup) PetscLogEventBegin(mglevels[0]->eventsmoothsetup,0,0,0,0);
1225: KSPSetUp(mglevels[0]->smoothd);
1226: if (mglevels[0]->smoothd->reason == KSP_DIVERGED_PC_FAILED) {
1227: pc->failedreason = PC_SUBPC_ERROR;
1228: }
1229: if (mglevels[0]->eventsmoothsetup) PetscLogEventEnd(mglevels[0]->eventsmoothsetup,0,0,0,0);
1231: /*
1232: Dump the interpolation/restriction matrices plus the
1233: Jacobian/stiffness on each level. This allows MATLAB users to
1234: easily check if the Galerkin condition A_c = R A_f R^T is satisfied.
1236: Only support one or the other at the same time.
1237: */
1238: #if defined(PETSC_USE_SOCKET_VIEWER)
1239: PetscOptionsGetBool(((PetscObject)pc)->options,((PetscObject)pc)->prefix,"-pc_mg_dump_matlab",&dump,NULL);
1240: if (dump) viewer = PETSC_VIEWER_SOCKET_(PetscObjectComm((PetscObject)pc));
1241: dump = PETSC_FALSE;
1242: #endif
1243: PetscOptionsGetBool(((PetscObject)pc)->options,((PetscObject)pc)->prefix,"-pc_mg_dump_binary",&dump,NULL);
1244: if (dump) viewer = PETSC_VIEWER_BINARY_(PetscObjectComm((PetscObject)pc));
1246: if (viewer) {
1247: for (i=1; i<n; i++) {
1248: MatView(mglevels[i]->restrct,viewer);
1249: }
1250: for (i=0; i<n; i++) {
1251: KSPGetPC(mglevels[i]->smoothd,&pc);
1252: MatView(pc->mat,viewer);
1253: }
1254: }
1255: return 0;
1256: }
1258: /* -------------------------------------------------------------------------------------*/
1260: PetscErrorCode PCMGGetLevels_MG(PC pc, PetscInt *levels)
1261: {
1262: PC_MG *mg = (PC_MG *) pc->data;
1264: *levels = mg->nlevels;
1265: return 0;
1266: }
1268: /*@
1269: PCMGGetLevels - Gets the number of levels to use with MG.
1271: Not Collective
1273: Input Parameter:
1274: . pc - the preconditioner context
1276: Output parameter:
1277: . levels - the number of levels
1279: Level: advanced
1281: .seealso: PCMGSetLevels()
1282: @*/
1283: PetscErrorCode PCMGGetLevels(PC pc,PetscInt *levels)
1284: {
1287: *levels = 0;
1288: PetscTryMethod(pc,"PCMGGetLevels_C",(PC,PetscInt*),(pc,levels));
1289: return 0;
1290: }
1292: /*@
1293: PCMGGetGridComplexity - compute operator and grid complexity of MG hierarchy
1295: Input Parameter:
1296: . pc - the preconditioner context
1298: Output Parameters:
1299: + gc - grid complexity = sum_i(n_i) / n_0
1300: - oc - operator complexity = sum_i(nnz_i) / nnz_0
1302: Level: advanced
1304: .seealso: PCMGGetLevels()
1305: @*/
1306: PetscErrorCode PCMGGetGridComplexity(PC pc, PetscReal *gc, PetscReal *oc)
1307: {
1308: PC_MG *mg = (PC_MG*)pc->data;
1309: PC_MG_Levels **mglevels = mg->levels;
1310: PetscInt lev,N;
1311: PetscLogDouble nnz0 = 0, sgc = 0, soc = 0, n0 = 0;
1312: MatInfo info;
1317: if (!pc->setupcalled) {
1318: if (gc) *gc = 0;
1319: if (oc) *oc = 0;
1320: return 0;
1321: }
1323: for (lev=0; lev<mg->nlevels; lev++) {
1324: Mat dB;
1325: KSPGetOperators(mglevels[lev]->smoothd,NULL,&dB);
1326: MatGetInfo(dB,MAT_GLOBAL_SUM,&info); /* global reduction */
1327: MatGetSize(dB,&N,NULL);
1328: sgc += N;
1329: soc += info.nz_used;
1330: if (lev==mg->nlevels-1) {nnz0 = info.nz_used; n0 = N;}
1331: }
1332: if (n0 > 0 && gc) *gc = (PetscReal)(sgc/n0);
1333: else SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Number for grid points on finest level is not available");
1334: if (nnz0 > 0 && oc) *oc = (PetscReal)(soc/nnz0);
1335: return 0;
1336: }
1338: /*@
1339: PCMGSetType - Determines the form of multigrid to use:
1340: multiplicative, additive, full, or the Kaskade algorithm.
1342: Logically Collective on PC
1344: Input Parameters:
1345: + pc - the preconditioner context
1346: - form - multigrid form, one of PC_MG_MULTIPLICATIVE, PC_MG_ADDITIVE,
1347: PC_MG_FULL, PC_MG_KASKADE
1349: Options Database Key:
1350: . -pc_mg_type <form> - Sets <form>, one of multiplicative,
1351: additive, full, kaskade
1353: Level: advanced
1355: .seealso: PCMGSetLevels()
1356: @*/
1357: PetscErrorCode PCMGSetType(PC pc,PCMGType form)
1358: {
1359: PC_MG *mg = (PC_MG*)pc->data;
1363: mg->am = form;
1364: if (form == PC_MG_MULTIPLICATIVE) pc->ops->applyrichardson = PCApplyRichardson_MG;
1365: else pc->ops->applyrichardson = NULL;
1366: return 0;
1367: }
1369: /*@
1370: PCMGGetType - Determines the form of multigrid to use:
1371: multiplicative, additive, full, or the Kaskade algorithm.
1373: Logically Collective on PC
1375: Input Parameter:
1376: . pc - the preconditioner context
1378: Output Parameter:
1379: . type - one of PC_MG_MULTIPLICATIVE, PC_MG_ADDITIVE,PC_MG_FULL, PC_MG_KASKADE
1381: Level: advanced
1383: .seealso: PCMGSetLevels()
1384: @*/
1385: PetscErrorCode PCMGGetType(PC pc,PCMGType *type)
1386: {
1387: PC_MG *mg = (PC_MG*)pc->data;
1390: *type = mg->am;
1391: return 0;
1392: }
1394: /*@
1395: PCMGSetCycleType - Sets the type cycles to use. Use PCMGSetCycleTypeOnLevel() for more
1396: complicated cycling.
1398: Logically Collective on PC
1400: Input Parameters:
1401: + pc - the multigrid context
1402: - n - either PC_MG_CYCLE_V or PC_MG_CYCLE_W
1404: Options Database Key:
1405: . -pc_mg_cycle_type <v,w> - provide the cycle desired
1407: Level: advanced
1409: .seealso: PCMGSetCycleTypeOnLevel()
1410: @*/
1411: PetscErrorCode PCMGSetCycleType(PC pc,PCMGCycleType n)
1412: {
1413: PC_MG *mg = (PC_MG*)pc->data;
1414: PC_MG_Levels **mglevels = mg->levels;
1415: PetscInt i,levels;
1420: levels = mglevels[0]->levels;
1421: for (i=0; i<levels; i++) mglevels[i]->cycles = n;
1422: return 0;
1423: }
1425: /*@
1426: PCMGMultiplicativeSetCycles - Sets the number of cycles to use for each preconditioner step
1427: of multigrid when PCMGType of PC_MG_MULTIPLICATIVE is used
1429: Logically Collective on PC
1431: Input Parameters:
1432: + pc - the multigrid context
1433: - n - number of cycles (default is 1)
1435: Options Database Key:
1436: . -pc_mg_multiplicative_cycles n - set the number of cycles
1438: Level: advanced
1440: Notes:
1441: This is not associated with setting a v or w cycle, that is set with PCMGSetCycleType()
1443: .seealso: PCMGSetCycleTypeOnLevel(), PCMGSetCycleType()
1444: @*/
1445: PetscErrorCode PCMGMultiplicativeSetCycles(PC pc,PetscInt n)
1446: {
1447: PC_MG *mg = (PC_MG*)pc->data;
1451: mg->cyclesperpcapply = n;
1452: return 0;
1453: }
1455: PetscErrorCode PCMGSetGalerkin_MG(PC pc,PCMGGalerkinType use)
1456: {
1457: PC_MG *mg = (PC_MG*)pc->data;
1459: mg->galerkin = use;
1460: return 0;
1461: }
1463: /*@
1464: PCMGSetGalerkin - Causes the coarser grid matrices to be computed from the
1465: finest grid via the Galerkin process: A_i-1 = r_i * A_i * p_i
1467: Logically Collective on PC
1469: Input Parameters:
1470: + pc - the multigrid context
1471: - use - one of PC_MG_GALERKIN_BOTH,PC_MG_GALERKIN_PMAT,PC_MG_GALERKIN_MAT, or PC_MG_GALERKIN_NONE
1473: Options Database Key:
1474: . -pc_mg_galerkin <both,pmat,mat,none> - set the matrices to form via the Galerkin process
1476: Level: intermediate
1478: Notes:
1479: Some codes that use PCMG such as PCGAMG use Galerkin internally while constructing the hierarchy and thus do not
1480: use the PCMG construction of the coarser grids.
1482: .seealso: PCMGGetGalerkin(), PCMGGalerkinType
1484: @*/
1485: PetscErrorCode PCMGSetGalerkin(PC pc,PCMGGalerkinType use)
1486: {
1488: PetscTryMethod(pc,"PCMGSetGalerkin_C",(PC,PCMGGalerkinType),(pc,use));
1489: return 0;
1490: }
1492: /*@
1493: PCMGGetGalerkin - Checks if Galerkin multigrid is being used, i.e.
1494: A_i-1 = r_i * A_i * p_i
1496: Not Collective
1498: Input Parameter:
1499: . pc - the multigrid context
1501: Output Parameter:
1502: . galerkin - one of PC_MG_GALERKIN_BOTH,PC_MG_GALERKIN_PMAT,PC_MG_GALERKIN_MAT, PC_MG_GALERKIN_NONE, or PC_MG_GALERKIN_EXTERNAL
1504: Level: intermediate
1506: .seealso: PCMGSetGalerkin(), PCMGGalerkinType
1508: @*/
1509: PetscErrorCode PCMGGetGalerkin(PC pc,PCMGGalerkinType *galerkin)
1510: {
1511: PC_MG *mg = (PC_MG*)pc->data;
1514: *galerkin = mg->galerkin;
1515: return 0;
1516: }
1518: PetscErrorCode PCMGSetAdaptInterpolation_MG(PC pc, PetscBool adapt)
1519: {
1520: PC_MG *mg = (PC_MG *) pc->data;
1522: mg->adaptInterpolation = adapt;
1523: return 0;
1524: }
1526: PetscErrorCode PCMGGetAdaptInterpolation_MG(PC pc, PetscBool *adapt)
1527: {
1528: PC_MG *mg = (PC_MG *) pc->data;
1530: *adapt = mg->adaptInterpolation;
1531: return 0;
1532: }
1534: PetscErrorCode PCMGSetAdaptCR_MG(PC pc, PetscBool cr)
1535: {
1536: PC_MG *mg = (PC_MG *) pc->data;
1538: mg->compatibleRelaxation = cr;
1539: return 0;
1540: }
1542: PetscErrorCode PCMGGetAdaptCR_MG(PC pc, PetscBool *cr)
1543: {
1544: PC_MG *mg = (PC_MG *) pc->data;
1546: *cr = mg->compatibleRelaxation;
1547: return 0;
1548: }
1550: /*@
1551: PCMGSetAdaptInterpolation - Adapt the interpolator based upon a vector space which should be accurately captured by the next coarser mesh, and thus accurately interpolated.
1553: Logically Collective on PC
1555: Input Parameters:
1556: + pc - the multigrid context
1557: - adapt - flag for adaptation of the interpolator
1559: Options Database Keys:
1560: + -pc_mg_adapt_interp - Turn on adaptation
1561: . -pc_mg_adapt_interp_n <int> - The number of modes to use, should be divisible by dimension
1562: - -pc_mg_adapt_interp_coarse_space <type> - The type of coarse space: polynomial, harmonic, eigenvector, generalized_eigenvector
1564: Level: intermediate
1566: .keywords: MG, set, Galerkin
1567: .seealso: PCMGGetAdaptInterpolation(), PCMGSetGalerkin()
1568: @*/
1569: PetscErrorCode PCMGSetAdaptInterpolation(PC pc, PetscBool adapt)
1570: {
1572: PetscTryMethod(pc,"PCMGSetAdaptInterpolation_C",(PC,PetscBool),(pc,adapt));
1573: return 0;
1574: }
1576: /*@
1577: PCMGGetAdaptInterpolation - Get the flag to adapt the interpolator based upon a vector space which should be accurately captured by the next coarser mesh, and thus accurately interpolated.
1579: Not collective
1581: Input Parameter:
1582: . pc - the multigrid context
1584: Output Parameter:
1585: . adapt - flag for adaptation of the interpolator
1587: Level: intermediate
1589: .keywords: MG, set, Galerkin
1590: .seealso: PCMGSetAdaptInterpolation(), PCMGSetGalerkin()
1591: @*/
1592: PetscErrorCode PCMGGetAdaptInterpolation(PC pc, PetscBool *adapt)
1593: {
1596: PetscUseMethod(pc,"PCMGGetAdaptInterpolation_C",(PC,PetscBool*),(pc,adapt));
1597: return 0;
1598: }
1600: /*@
1601: PCMGSetAdaptCR - Monitor the coarse space quality using an auxiliary solve with compatible relaxation.
1603: Logically Collective on PC
1605: Input Parameters:
1606: + pc - the multigrid context
1607: - cr - flag for compatible relaxation
1609: Options Database Keys:
1610: . -pc_mg_adapt_cr - Turn on compatible relaxation
1612: Level: intermediate
1614: .keywords: MG, set, Galerkin
1615: .seealso: PCMGGetAdaptCR(), PCMGSetAdaptInterpolation(), PCMGSetGalerkin()
1616: @*/
1617: PetscErrorCode PCMGSetAdaptCR(PC pc, PetscBool cr)
1618: {
1620: PetscTryMethod(pc,"PCMGSetAdaptCR_C",(PC,PetscBool),(pc,cr));
1621: return 0;
1622: }
1624: /*@
1625: PCMGGetAdaptCR - Get the flag to monitor coarse space quality using an auxiliary solve with compatible relaxation.
1627: Not collective
1629: Input Parameter:
1630: . pc - the multigrid context
1632: Output Parameter:
1633: . cr - flag for compatible relaxaion
1635: Level: intermediate
1637: .keywords: MG, set, Galerkin
1638: .seealso: PCMGSetAdaptCR(), PCMGGetAdaptInterpolation(), PCMGSetGalerkin()
1639: @*/
1640: PetscErrorCode PCMGGetAdaptCR(PC pc, PetscBool *cr)
1641: {
1644: PetscUseMethod(pc,"PCMGGetAdaptCR_C",(PC,PetscBool*),(pc,cr));
1645: return 0;
1646: }
1648: /*@
1649: PCMGSetNumberSmooth - Sets the number of pre and post-smoothing steps to use
1650: on all levels. Use PCMGDistinctSmoothUp() to create separate up and down smoothers if you want different numbers of
1651: pre- and post-smoothing steps.
1653: Logically Collective on PC
1655: Input Parameters:
1656: + mg - the multigrid context
1657: - n - the number of smoothing steps
1659: Options Database Key:
1660: . -mg_levels_ksp_max_it <n> - Sets number of pre and post-smoothing steps
1662: Level: advanced
1664: Notes:
1665: this does not set a value on the coarsest grid, since we assume that
1666: there is no separate smooth up on the coarsest grid.
1668: .seealso: PCMGSetDistinctSmoothUp()
1669: @*/
1670: PetscErrorCode PCMGSetNumberSmooth(PC pc,PetscInt n)
1671: {
1672: PC_MG *mg = (PC_MG*)pc->data;
1673: PC_MG_Levels **mglevels = mg->levels;
1674: PetscInt i,levels;
1679: levels = mglevels[0]->levels;
1681: for (i=1; i<levels; i++) {
1682: KSPSetTolerances(mglevels[i]->smoothu,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT,n);
1683: KSPSetTolerances(mglevels[i]->smoothd,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT,n);
1684: mg->default_smoothu = n;
1685: mg->default_smoothd = n;
1686: }
1687: return 0;
1688: }
1690: /*@
1691: PCMGSetDistinctSmoothUp - sets the up (post) smoother to be a separate KSP from the down (pre) smoother on all levels
1692: and adds the suffix _up to the options name
1694: Logically Collective on PC
1696: Input Parameters:
1697: . pc - the preconditioner context
1699: Options Database Key:
1700: . -pc_mg_distinct_smoothup <bool> - use distinct smoothing objects
1702: Level: advanced
1704: Notes:
1705: this does not set a value on the coarsest grid, since we assume that
1706: there is no separate smooth up on the coarsest grid.
1708: .seealso: PCMGSetNumberSmooth()
1709: @*/
1710: PetscErrorCode PCMGSetDistinctSmoothUp(PC pc)
1711: {
1712: PC_MG *mg = (PC_MG*)pc->data;
1713: PC_MG_Levels **mglevels = mg->levels;
1714: PetscInt i,levels;
1715: KSP subksp;
1719: levels = mglevels[0]->levels;
1721: for (i=1; i<levels; i++) {
1722: const char *prefix = NULL;
1723: /* make sure smoother up and down are different */
1724: PCMGGetSmootherUp(pc,i,&subksp);
1725: KSPGetOptionsPrefix(mglevels[i]->smoothd,&prefix);
1726: KSPSetOptionsPrefix(subksp,prefix);
1727: KSPAppendOptionsPrefix(subksp,"up_");
1728: }
1729: return 0;
1730: }
1732: /* No new matrices are created, and the coarse operator matrices are the references to the original ones */
1733: PetscErrorCode PCGetInterpolations_MG(PC pc,PetscInt *num_levels,Mat *interpolations[])
1734: {
1735: PC_MG *mg = (PC_MG*)pc->data;
1736: PC_MG_Levels **mglevels = mg->levels;
1737: Mat *mat;
1738: PetscInt l;
1741: PetscMalloc1(mg->nlevels,&mat);
1742: for (l=1; l< mg->nlevels; l++) {
1743: mat[l-1] = mglevels[l]->interpolate;
1744: PetscObjectReference((PetscObject)mat[l-1]);
1745: }
1746: *num_levels = mg->nlevels;
1747: *interpolations = mat;
1748: return 0;
1749: }
1751: /* No new matrices are created, and the coarse operator matrices are the references to the original ones */
1752: PetscErrorCode PCGetCoarseOperators_MG(PC pc,PetscInt *num_levels,Mat *coarseOperators[])
1753: {
1754: PC_MG *mg = (PC_MG*)pc->data;
1755: PC_MG_Levels **mglevels = mg->levels;
1756: PetscInt l;
1757: Mat *mat;
1760: PetscMalloc1(mg->nlevels,&mat);
1761: for (l=0; l<mg->nlevels-1; l++) {
1762: KSPGetOperators(mglevels[l]->smoothd,NULL,&(mat[l]));
1763: PetscObjectReference((PetscObject)mat[l]);
1764: }
1765: *num_levels = mg->nlevels;
1766: *coarseOperators = mat;
1767: return 0;
1768: }
1770: /*@C
1771: PCMGRegisterCoarseSpaceConstructor - Adds a method to the PCMG package for coarse space construction.
1773: Not collective
1775: Input Parameters:
1776: + name - name of the constructor
1777: - function - constructor routine
1779: Notes:
1780: Calling sequence for the routine:
1781: $ my_csp(PC pc, PetscInt l, DM dm, KSP smooth, PetscInt Nc, const Vec initGuess[], Vec **coarseSp)
1782: $ pc - The PC object
1783: $ l - The multigrid level, 0 is the coarse level
1784: $ dm - The DM for this level
1785: $ smooth - The level smoother
1786: $ Nc - The size of the coarse space
1787: $ initGuess - Basis for an initial guess for the space
1788: $ coarseSp - A basis for the computed coarse space
1790: Level: advanced
1792: .seealso: PCMGGetCoarseSpaceConstructor(), PCRegister()
1793: @*/
1794: PetscErrorCode PCMGRegisterCoarseSpaceConstructor(const char name[], PetscErrorCode (*function)(PC, PetscInt, DM, KSP, PetscInt, const Vec[], Vec **))
1795: {
1796: PCInitializePackage();
1797: PetscFunctionListAdd(&PCMGCoarseList,name,function);
1798: return 0;
1799: }
1801: /*@C
1802: PCMGGetCoarseSpaceConstructor - Returns the given coarse space construction method.
1804: Not collective
1806: Input Parameter:
1807: . name - name of the constructor
1809: Output Parameter:
1810: . function - constructor routine
1812: Notes:
1813: Calling sequence for the routine:
1814: $ my_csp(PC pc, PetscInt l, DM dm, KSP smooth, PetscInt Nc, const Vec initGuess[], Vec **coarseSp)
1815: $ pc - The PC object
1816: $ l - The multigrid level, 0 is the coarse level
1817: $ dm - The DM for this level
1818: $ smooth - The level smoother
1819: $ Nc - The size of the coarse space
1820: $ initGuess - Basis for an initial guess for the space
1821: $ coarseSp - A basis for the computed coarse space
1823: Level: advanced
1825: .seealso: PCMGRegisterCoarseSpaceConstructor(), PCRegister()
1826: @*/
1827: PetscErrorCode PCMGGetCoarseSpaceConstructor(const char name[], PetscErrorCode (**function)(PC, PetscInt, DM, KSP, PetscInt, const Vec[], Vec **))
1828: {
1829: PetscFunctionListFind(PCMGCoarseList,name,function);
1830: return 0;
1831: }
1833: /* ----------------------------------------------------------------------------------------*/
1835: /*MC
1836: PCMG - Use multigrid preconditioning. This preconditioner requires you provide additional
1837: information about the coarser grid matrices and restriction/interpolation operators.
1839: Options Database Keys:
1840: + -pc_mg_levels <nlevels> - number of levels including finest
1841: . -pc_mg_cycle_type <v,w> - provide the cycle desired
1842: . -pc_mg_type <additive,multiplicative,full,kaskade> - multiplicative is the default
1843: . -pc_mg_log - log information about time spent on each level of the solver
1844: . -pc_mg_distinct_smoothup - configure up (after interpolation) and down (before restriction) smoothers separately (with different options prefixes)
1845: . -pc_mg_galerkin <both,pmat,mat,none> - use Galerkin process to compute coarser operators, i.e. Acoarse = R A R'
1846: . -pc_mg_multiplicative_cycles - number of cycles to use as the preconditioner (defaults to 1)
1847: . -pc_mg_dump_matlab - dumps the matrices for each level and the restriction/interpolation matrices
1848: to the Socket viewer for reading from MATLAB.
1849: - -pc_mg_dump_binary - dumps the matrices for each level and the restriction/interpolation matrices
1850: to the binary output file called binaryoutput
1852: Notes:
1853: If one uses a Krylov method such GMRES or CG as the smoother then one must use KSPFGMRES, KSPGCR, or KSPRICHARDSON as the outer Krylov method
1855: When run with a single level the smoother options are used on that level NOT the coarse grid solver options
1857: When run with KSPRICHARDSON the convergence test changes slightly if monitor is turned on. The iteration count may change slightly. This
1858: is because without monitoring the residual norm is computed WITHIN each multigrid cycle on the finest level after the pre-smoothing
1859: (because the residual has just been computed for the multigrid algorithm and is hence available for free) while with monitoring the
1860: residual is computed at the end of each cycle.
1862: Level: intermediate
1864: .seealso: PCCreate(), PCSetType(), PCType (for list of available types), PC, PCMGType, PCEXOTIC, PCGAMG, PCML, PCHYPRE
1865: PCMGSetLevels(), PCMGGetLevels(), PCMGSetType(), PCMGSetCycleType(),
1866: PCMGSetDistinctSmoothUp(), PCMGGetCoarseSolve(), PCMGSetResidual(), PCMGSetInterpolation(),
1867: PCMGSetRestriction(), PCMGGetSmoother(), PCMGGetSmootherUp(), PCMGGetSmootherDown(),
1868: PCMGSetCycleTypeOnLevel(), PCMGSetRhs(), PCMGSetX(), PCMGSetR()
1869: M*/
1871: PETSC_EXTERN PetscErrorCode PCCreate_MG(PC pc)
1872: {
1873: PC_MG *mg;
1875: PetscNewLog(pc,&mg);
1876: pc->data = mg;
1877: mg->nlevels = -1;
1878: mg->am = PC_MG_MULTIPLICATIVE;
1879: mg->galerkin = PC_MG_GALERKIN_NONE;
1880: mg->adaptInterpolation = PETSC_FALSE;
1881: mg->Nc = -1;
1882: mg->eigenvalue = -1;
1884: pc->useAmat = PETSC_TRUE;
1886: pc->ops->apply = PCApply_MG;
1887: pc->ops->applytranspose = PCApplyTranspose_MG;
1888: pc->ops->matapply = PCMatApply_MG;
1889: pc->ops->setup = PCSetUp_MG;
1890: pc->ops->reset = PCReset_MG;
1891: pc->ops->destroy = PCDestroy_MG;
1892: pc->ops->setfromoptions = PCSetFromOptions_MG;
1893: pc->ops->view = PCView_MG;
1895: PetscObjectComposedDataRegister(&mg->eigenvalue);
1896: PetscObjectComposeFunction((PetscObject)pc,"PCMGSetGalerkin_C",PCMGSetGalerkin_MG);
1897: PetscObjectComposeFunction((PetscObject)pc,"PCMGGetLevels_C",PCMGGetLevels_MG);
1898: PetscObjectComposeFunction((PetscObject)pc,"PCMGSetLevels_C",PCMGSetLevels_MG);
1899: PetscObjectComposeFunction((PetscObject)pc,"PCGetInterpolations_C",PCGetInterpolations_MG);
1900: PetscObjectComposeFunction((PetscObject)pc,"PCGetCoarseOperators_C",PCGetCoarseOperators_MG);
1901: PetscObjectComposeFunction((PetscObject)pc,"PCMGSetAdaptInterpolation_C",PCMGSetAdaptInterpolation_MG);
1902: PetscObjectComposeFunction((PetscObject)pc,"PCMGGetAdaptInterpolation_C",PCMGGetAdaptInterpolation_MG);
1903: PetscObjectComposeFunction((PetscObject)pc,"PCMGSetAdaptCR_C",PCMGSetAdaptCR_MG);
1904: PetscObjectComposeFunction((PetscObject)pc,"PCMGGetAdaptCR_C",PCMGGetAdaptCR_MG);
1905: return 0;
1906: }