Actual source code: mg.c
petsc-3.13.6 2020-09-29
2: /*
3: Defines the multigrid preconditioner interface.
4: */
5: #include <petsc/private/pcmgimpl.h>
6: #include <petscdm.h>
7: PETSC_INTERN PetscErrorCode PCPreSolveChangeRHS(PC,PetscBool*);
9: PetscErrorCode PCMGMCycle_Private(PC pc,PC_MG_Levels **mglevelsin,PCRichardsonConvergedReason *reason)
10: {
11: PC_MG *mg = (PC_MG*)pc->data;
12: PC_MG_Levels *mgc,*mglevels = *mglevelsin;
14: PetscInt cycles = (mglevels->level == 1) ? 1 : (PetscInt) mglevels->cycles;
17: if (mglevels->eventsmoothsolve) {PetscLogEventBegin(mglevels->eventsmoothsolve,0,0,0,0);}
18: KSPSolve(mglevels->smoothd,mglevels->b,mglevels->x); /* pre-smooth */
19: KSPCheckSolve(mglevels->smoothd,pc,mglevels->x);
20: if (mglevels->eventsmoothsolve) {PetscLogEventEnd(mglevels->eventsmoothsolve,0,0,0,0);}
21: if (mglevels->level) { /* not the coarsest grid */
22: if (mglevels->eventresidual) {PetscLogEventBegin(mglevels->eventresidual,0,0,0,0);}
23: (*mglevels->residual)(mglevels->A,mglevels->b,mglevels->x,mglevels->r);
24: if (mglevels->eventresidual) {PetscLogEventEnd(mglevels->eventresidual,0,0,0,0);}
26: /* if on finest level and have convergence criteria set */
27: if (mglevels->level == mglevels->levels-1 && mg->ttol && reason) {
28: PetscReal rnorm;
29: VecNorm(mglevels->r,NORM_2,&rnorm);
30: if (rnorm <= mg->ttol) {
31: if (rnorm < mg->abstol) {
32: *reason = PCRICHARDSON_CONVERGED_ATOL;
33: PetscInfo2(pc,"Linear solver has converged. Residual norm %g is less than absolute tolerance %g\n",(double)rnorm,(double)mg->abstol);
34: } else {
35: *reason = PCRICHARDSON_CONVERGED_RTOL;
36: PetscInfo2(pc,"Linear solver has converged. Residual norm %g is less than relative tolerance times initial residual norm %g\n",(double)rnorm,(double)mg->ttol);
37: }
38: return(0);
39: }
40: }
42: mgc = *(mglevelsin - 1);
43: if (mglevels->eventinterprestrict) {PetscLogEventBegin(mglevels->eventinterprestrict,0,0,0,0);}
44: MatRestrict(mglevels->restrct,mglevels->r,mgc->b);
45: if (mglevels->eventinterprestrict) {PetscLogEventEnd(mglevels->eventinterprestrict,0,0,0,0);}
46: VecSet(mgc->x,0.0);
47: while (cycles--) {
48: PCMGMCycle_Private(pc,mglevelsin-1,reason);
49: }
50: if (mglevels->eventinterprestrict) {PetscLogEventBegin(mglevels->eventinterprestrict,0,0,0,0);}
51: MatInterpolateAdd(mglevels->interpolate,mgc->x,mglevels->x,mglevels->x);
52: if (mglevels->eventinterprestrict) {PetscLogEventEnd(mglevels->eventinterprestrict,0,0,0,0);}
53: if (mglevels->eventsmoothsolve) {PetscLogEventBegin(mglevels->eventsmoothsolve,0,0,0,0);}
54: KSPSolve(mglevels->smoothu,mglevels->b,mglevels->x); /* post smooth */
55: KSPCheckSolve(mglevels->smoothu,pc,mglevels->x);
56: if (mglevels->eventsmoothsolve) {PetscLogEventEnd(mglevels->eventsmoothsolve,0,0,0,0);}
57: }
58: return(0);
59: }
61: 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)
62: {
63: PC_MG *mg = (PC_MG*)pc->data;
64: PC_MG_Levels **mglevels = mg->levels;
66: PC tpc;
67: PetscBool changeu,changed;
68: PetscInt levels = mglevels[0]->levels,i;
71: /* When the DM is supplying the matrix then it will not exist until here */
72: for (i=0; i<levels; i++) {
73: if (!mglevels[i]->A) {
74: KSPGetOperators(mglevels[i]->smoothu,&mglevels[i]->A,NULL);
75: PetscObjectReference((PetscObject)mglevels[i]->A);
76: }
77: }
79: KSPGetPC(mglevels[levels-1]->smoothd,&tpc);
80: PCPreSolveChangeRHS(tpc,&changed);
81: KSPGetPC(mglevels[levels-1]->smoothu,&tpc);
82: PCPreSolveChangeRHS(tpc,&changeu);
83: if (!changed && !changeu) {
84: VecDestroy(&mglevels[levels-1]->b);
85: mglevels[levels-1]->b = b;
86: } else { /* if the smoother changes the rhs during PreSolve, we cannot use the input vector */
87: if (!mglevels[levels-1]->b) {
88: Vec *vec;
90: KSPCreateVecs(mglevels[levels-1]->smoothd,1,&vec,0,NULL);
91: mglevels[levels-1]->b = *vec;
92: PetscFree(vec);
93: }
94: VecCopy(b,mglevels[levels-1]->b);
95: }
96: mglevels[levels-1]->x = x;
98: mg->rtol = rtol;
99: mg->abstol = abstol;
100: mg->dtol = dtol;
101: if (rtol) {
102: /* compute initial residual norm for relative convergence test */
103: PetscReal rnorm;
104: if (zeroguess) {
105: VecNorm(b,NORM_2,&rnorm);
106: } else {
107: (*mglevels[levels-1]->residual)(mglevels[levels-1]->A,b,x,w);
108: VecNorm(w,NORM_2,&rnorm);
109: }
110: mg->ttol = PetscMax(rtol*rnorm,abstol);
111: } else if (abstol) mg->ttol = abstol;
112: else mg->ttol = 0.0;
114: /* since smoother is applied to full system, not just residual we need to make sure that smoothers don't
115: stop prematurely due to small residual */
116: for (i=1; i<levels; i++) {
117: KSPSetTolerances(mglevels[i]->smoothu,0,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT);
118: if (mglevels[i]->smoothu != mglevels[i]->smoothd) {
119: /* For Richardson the initial guess is nonzero since it is solving in each cycle the original system not just applying as a preconditioner */
120: KSPSetInitialGuessNonzero(mglevels[i]->smoothd,PETSC_TRUE);
121: KSPSetTolerances(mglevels[i]->smoothd,0,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT);
122: }
123: }
125: *reason = (PCRichardsonConvergedReason)0;
126: for (i=0; i<its; i++) {
127: PCMGMCycle_Private(pc,mglevels+levels-1,reason);
128: if (*reason) break;
129: }
130: if (!*reason) *reason = PCRICHARDSON_CONVERGED_ITS;
131: *outits = i;
132: if (!changed && !changeu) mglevels[levels-1]->b = NULL;
133: return(0);
134: }
136: PetscErrorCode PCReset_MG(PC pc)
137: {
138: PC_MG *mg = (PC_MG*)pc->data;
139: PC_MG_Levels **mglevels = mg->levels;
141: PetscInt i,n;
144: if (mglevels) {
145: n = mglevels[0]->levels;
146: for (i=0; i<n-1; i++) {
147: VecDestroy(&mglevels[i+1]->r);
148: VecDestroy(&mglevels[i]->b);
149: VecDestroy(&mglevels[i]->x);
150: MatDestroy(&mglevels[i+1]->restrct);
151: MatDestroy(&mglevels[i+1]->interpolate);
152: MatDestroy(&mglevels[i+1]->inject);
153: VecDestroy(&mglevels[i+1]->rscale);
154: }
155: /* this is not null only if the smoother on the finest level
156: changes the rhs during PreSolve */
157: VecDestroy(&mglevels[n-1]->b);
159: for (i=0; i<n; i++) {
160: MatDestroy(&mglevels[i]->A);
161: if (mglevels[i]->smoothd != mglevels[i]->smoothu) {
162: KSPReset(mglevels[i]->smoothd);
163: }
164: KSPReset(mglevels[i]->smoothu);
165: }
166: }
167: return(0);
168: }
170: PetscErrorCode PCMGSetLevels_MG(PC pc,PetscInt levels,MPI_Comm *comms)
171: {
173: PC_MG *mg = (PC_MG*)pc->data;
174: MPI_Comm comm;
175: PC_MG_Levels **mglevels = mg->levels;
176: PCMGType mgtype = mg->am;
177: PetscInt mgctype = (PetscInt) PC_MG_CYCLE_V;
178: PetscInt i;
179: PetscMPIInt size;
180: const char *prefix;
181: PC ipc;
182: PetscInt n;
187: if (mg->nlevels == levels) return(0);
188: PetscObjectGetComm((PetscObject)pc,&comm);
189: if (mglevels) {
190: mgctype = mglevels[0]->cycles;
191: /* changing the number of levels so free up the previous stuff */
192: PCReset_MG(pc);
193: n = mglevels[0]->levels;
194: for (i=0; i<n; i++) {
195: if (mglevels[i]->smoothd != mglevels[i]->smoothu) {
196: KSPDestroy(&mglevels[i]->smoothd);
197: }
198: KSPDestroy(&mglevels[i]->smoothu);
199: PetscFree(mglevels[i]);
200: }
201: PetscFree(mg->levels);
202: }
204: mg->nlevels = levels;
206: PetscMalloc1(levels,&mglevels);
207: PetscLogObjectMemory((PetscObject)pc,levels*(sizeof(PC_MG*)));
209: PCGetOptionsPrefix(pc,&prefix);
211: mg->stageApply = 0;
212: for (i=0; i<levels; i++) {
213: PetscNewLog(pc,&mglevels[i]);
215: mglevels[i]->level = i;
216: mglevels[i]->levels = levels;
217: mglevels[i]->cycles = mgctype;
218: mg->default_smoothu = 2;
219: mg->default_smoothd = 2;
220: mglevels[i]->eventsmoothsetup = 0;
221: mglevels[i]->eventsmoothsolve = 0;
222: mglevels[i]->eventresidual = 0;
223: mglevels[i]->eventinterprestrict = 0;
225: if (comms) comm = comms[i];
226: KSPCreate(comm,&mglevels[i]->smoothd);
227: KSPSetErrorIfNotConverged(mglevels[i]->smoothd,pc->erroriffailure);
228: PetscObjectIncrementTabLevel((PetscObject)mglevels[i]->smoothd,(PetscObject)pc,levels-i);
229: KSPSetOptionsPrefix(mglevels[i]->smoothd,prefix);
230: PetscObjectComposedDataSetInt((PetscObject) mglevels[i]->smoothd, PetscMGLevelId, mglevels[i]->level);
231: if (i || levels == 1) {
232: char tprefix[128];
234: KSPSetType(mglevels[i]->smoothd,KSPCHEBYSHEV);
235: KSPSetConvergenceTest(mglevels[i]->smoothd,KSPConvergedSkip,NULL,NULL);
236: KSPSetNormType(mglevels[i]->smoothd,KSP_NORM_NONE);
237: KSPGetPC(mglevels[i]->smoothd,&ipc);
238: PCSetType(ipc,PCSOR);
239: KSPSetTolerances(mglevels[i]->smoothd,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT, mg->default_smoothd);
241: sprintf(tprefix,"mg_levels_%d_",(int)i);
242: KSPAppendOptionsPrefix(mglevels[i]->smoothd,tprefix);
243: } else {
244: KSPAppendOptionsPrefix(mglevels[0]->smoothd,"mg_coarse_");
246: /* coarse solve is (redundant) LU by default; set shifttype NONZERO to avoid annoying zero-pivot in LU preconditioner */
247: KSPSetType(mglevels[0]->smoothd,KSPPREONLY);
248: KSPGetPC(mglevels[0]->smoothd,&ipc);
249: MPI_Comm_size(comm,&size);
250: if (size > 1) {
251: PCSetType(ipc,PCREDUNDANT);
252: } else {
253: PCSetType(ipc,PCLU);
254: }
255: PCFactorSetShiftType(ipc,MAT_SHIFT_INBLOCKS);
256: }
257: PetscLogObjectParent((PetscObject)pc,(PetscObject)mglevels[i]->smoothd);
259: mglevels[i]->smoothu = mglevels[i]->smoothd;
260: mg->rtol = 0.0;
261: mg->abstol = 0.0;
262: mg->dtol = 0.0;
263: mg->ttol = 0.0;
264: mg->cyclesperpcapply = 1;
265: }
266: mg->levels = mglevels;
267: PCMGSetType(pc,mgtype);
268: return(0);
269: }
271: /*@C
272: PCMGSetLevels - Sets the number of levels to use with MG.
273: Must be called before any other MG routine.
275: Logically Collective on PC
277: Input Parameters:
278: + pc - the preconditioner context
279: . levels - the number of levels
280: - comms - optional communicators for each level; this is to allow solving the coarser problems
281: on smaller sets of processors.
283: Level: intermediate
285: Notes:
286: If the number of levels is one then the multigrid uses the -mg_levels prefix
287: for setting the level options rather than the -mg_coarse prefix.
289: .seealso: PCMGSetType(), PCMGGetLevels()
290: @*/
291: PetscErrorCode PCMGSetLevels(PC pc,PetscInt levels,MPI_Comm *comms)
292: {
298: PetscTryMethod(pc,"PCMGSetLevels_C",(PC,PetscInt,MPI_Comm*),(pc,levels,comms));
299: return(0);
300: }
303: PetscErrorCode PCDestroy_MG(PC pc)
304: {
306: PC_MG *mg = (PC_MG*)pc->data;
307: PC_MG_Levels **mglevels = mg->levels;
308: PetscInt i,n;
311: PCReset_MG(pc);
312: if (mglevels) {
313: n = mglevels[0]->levels;
314: for (i=0; i<n; i++) {
315: if (mglevels[i]->smoothd != mglevels[i]->smoothu) {
316: KSPDestroy(&mglevels[i]->smoothd);
317: }
318: KSPDestroy(&mglevels[i]->smoothu);
319: PetscFree(mglevels[i]);
320: }
321: PetscFree(mg->levels);
322: }
323: PetscFree(pc->data);
324: PetscObjectComposeFunction((PetscObject)pc,"PCGetInterpolations_C",NULL);
325: PetscObjectComposeFunction((PetscObject)pc,"PCGetCoarseOperators_C",NULL);
326: return(0);
327: }
331: extern PetscErrorCode PCMGACycle_Private(PC,PC_MG_Levels**);
332: extern PetscErrorCode PCMGFCycle_Private(PC,PC_MG_Levels**);
333: extern PetscErrorCode PCMGKCycle_Private(PC,PC_MG_Levels**);
335: /*
336: PCApply_MG - Runs either an additive, multiplicative, Kaskadic
337: or full cycle of multigrid.
339: Note:
340: A simple wrapper which calls PCMGMCycle(),PCMGACycle(), or PCMGFCycle().
341: */
342: static PetscErrorCode PCApply_MG(PC pc,Vec b,Vec x)
343: {
344: PC_MG *mg = (PC_MG*)pc->data;
345: PC_MG_Levels **mglevels = mg->levels;
347: PC tpc;
348: PetscInt levels = mglevels[0]->levels,i;
349: PetscBool changeu,changed;
352: if (mg->stageApply) {PetscLogStagePush(mg->stageApply);}
353: /* When the DM is supplying the matrix then it will not exist until here */
354: for (i=0; i<levels; i++) {
355: if (!mglevels[i]->A) {
356: KSPGetOperators(mglevels[i]->smoothu,&mglevels[i]->A,NULL);
357: PetscObjectReference((PetscObject)mglevels[i]->A);
358: }
359: }
361: KSPGetPC(mglevels[levels-1]->smoothd,&tpc);
362: PCPreSolveChangeRHS(tpc,&changed);
363: KSPGetPC(mglevels[levels-1]->smoothu,&tpc);
364: PCPreSolveChangeRHS(tpc,&changeu);
365: if (!changeu && !changed) {
366: VecDestroy(&mglevels[levels-1]->b);
367: mglevels[levels-1]->b = b;
368: } else { /* if the smoother changes the rhs during PreSolve, we cannot use the input vector */
369: if (!mglevels[levels-1]->b) {
370: Vec *vec;
372: KSPCreateVecs(mglevels[levels-1]->smoothd,1,&vec,0,NULL);
373: mglevels[levels-1]->b = *vec;
374: PetscFree(vec);
375: }
376: VecCopy(b,mglevels[levels-1]->b);
377: }
378: mglevels[levels-1]->x = x;
380: if (mg->am == PC_MG_MULTIPLICATIVE) {
381: VecSet(x,0.0);
382: for (i=0; i<mg->cyclesperpcapply; i++) {
383: PCMGMCycle_Private(pc,mglevels+levels-1,NULL);
384: }
385: } else if (mg->am == PC_MG_ADDITIVE) {
386: PCMGACycle_Private(pc,mglevels);
387: } else if (mg->am == PC_MG_KASKADE) {
388: PCMGKCycle_Private(pc,mglevels);
389: } else {
390: PCMGFCycle_Private(pc,mglevels);
391: }
392: if (mg->stageApply) {PetscLogStagePop();}
393: if (!changeu && !changed) mglevels[levels-1]->b = NULL;
394: return(0);
395: }
398: PetscErrorCode PCSetFromOptions_MG(PetscOptionItems *PetscOptionsObject,PC pc)
399: {
400: PetscErrorCode ierr;
401: PetscInt levels,cycles;
402: PetscBool flg;
403: PC_MG *mg = (PC_MG*)pc->data;
404: PC_MG_Levels **mglevels;
405: PCMGType mgtype;
406: PCMGCycleType mgctype;
407: PCMGGalerkinType gtype;
410: levels = PetscMax(mg->nlevels,1);
411: PetscOptionsHead(PetscOptionsObject,"Multigrid options");
412: PetscOptionsInt("-pc_mg_levels","Number of Levels","PCMGSetLevels",levels,&levels,&flg);
413: if (!flg && !mg->levels && pc->dm) {
414: DMGetRefineLevel(pc->dm,&levels);
415: levels++;
416: mg->usedmfornumberoflevels = PETSC_TRUE;
417: }
418: PCMGSetLevels(pc,levels,NULL);
419: mglevels = mg->levels;
421: mgctype = (PCMGCycleType) mglevels[0]->cycles;
422: PetscOptionsEnum("-pc_mg_cycle_type","V cycle or for W-cycle","PCMGSetCycleType",PCMGCycleTypes,(PetscEnum)mgctype,(PetscEnum*)&mgctype,&flg);
423: if (flg) {
424: PCMGSetCycleType(pc,mgctype);
425: }
426: gtype = mg->galerkin;
427: PetscOptionsEnum("-pc_mg_galerkin","Use Galerkin process to compute coarser operators","PCMGSetGalerkin",PCMGGalerkinTypes,(PetscEnum)gtype,(PetscEnum*)>ype,&flg);
428: if (flg) {
429: PCMGSetGalerkin(pc,gtype);
430: }
431: flg = PETSC_FALSE;
432: PetscOptionsBool("-pc_mg_distinct_smoothup","Create separate smoothup KSP and append the prefix _up","PCMGSetDistinctSmoothUp",PETSC_FALSE,&flg,NULL);
433: if (flg) {
434: PCMGSetDistinctSmoothUp(pc);
435: }
436: mgtype = mg->am;
437: PetscOptionsEnum("-pc_mg_type","Multigrid type","PCMGSetType",PCMGTypes,(PetscEnum)mgtype,(PetscEnum*)&mgtype,&flg);
438: if (flg) {
439: PCMGSetType(pc,mgtype);
440: }
441: if (mg->am == PC_MG_MULTIPLICATIVE) {
442: PetscOptionsInt("-pc_mg_multiplicative_cycles","Number of cycles for each preconditioner step","PCMGMultiplicativeSetCycles",mg->cyclesperpcapply,&cycles,&flg);
443: if (flg) {
444: PCMGMultiplicativeSetCycles(pc,cycles);
445: }
446: }
447: flg = PETSC_FALSE;
448: PetscOptionsBool("-pc_mg_log","Log times for each multigrid level","None",flg,&flg,NULL);
449: if (flg) {
450: PetscInt i;
451: char eventname[128];
453: levels = mglevels[0]->levels;
454: for (i=0; i<levels; i++) {
455: sprintf(eventname,"MGSetup Level %d",(int)i);
456: PetscLogEventRegister(eventname,((PetscObject)pc)->classid,&mglevels[i]->eventsmoothsetup);
457: sprintf(eventname,"MGSmooth Level %d",(int)i);
458: PetscLogEventRegister(eventname,((PetscObject)pc)->classid,&mglevels[i]->eventsmoothsolve);
459: if (i) {
460: sprintf(eventname,"MGResid Level %d",(int)i);
461: PetscLogEventRegister(eventname,((PetscObject)pc)->classid,&mglevels[i]->eventresidual);
462: sprintf(eventname,"MGInterp Level %d",(int)i);
463: PetscLogEventRegister(eventname,((PetscObject)pc)->classid,&mglevels[i]->eventinterprestrict);
464: }
465: }
467: #if defined(PETSC_USE_LOG)
468: {
469: const char *sname = "MG Apply";
470: PetscStageLog stageLog;
471: PetscInt st;
473: PetscLogGetStageLog(&stageLog);
474: for (st = 0; st < stageLog->numStages; ++st) {
475: PetscBool same;
477: PetscStrcmp(stageLog->stageInfo[st].name, sname, &same);
478: if (same) mg->stageApply = st;
479: }
480: if (!mg->stageApply) {
481: PetscLogStageRegister(sname, &mg->stageApply);
482: }
483: }
484: #endif
485: }
486: PetscOptionsTail();
487: return(0);
488: }
490: const char *const PCMGTypes[] = {"MULTIPLICATIVE","ADDITIVE","FULL","KASKADE","PCMGType","PC_MG",0};
491: const char *const PCMGCycleTypes[] = {"invalid","v","w","PCMGCycleType","PC_MG_CYCLE",0};
492: const char *const PCMGGalerkinTypes[] = {"both","pmat","mat","none","external","PCMGGalerkinType","PC_MG_GALERKIN",0};
494: #include <petscdraw.h>
495: PetscErrorCode PCView_MG(PC pc,PetscViewer viewer)
496: {
497: PC_MG *mg = (PC_MG*)pc->data;
498: PC_MG_Levels **mglevels = mg->levels;
500: PetscInt levels = mglevels ? mglevels[0]->levels : 0,i;
501: PetscBool iascii,isbinary,isdraw;
504: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);
505: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERBINARY,&isbinary);
506: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERDRAW,&isdraw);
507: if (iascii) {
508: const char *cyclename = levels ? (mglevels[0]->cycles == PC_MG_CYCLE_V ? "v" : "w") : "unknown";
509: PetscViewerASCIIPrintf(viewer," type is %s, levels=%D cycles=%s\n", PCMGTypes[mg->am],levels,cyclename);
510: if (mg->am == PC_MG_MULTIPLICATIVE) {
511: PetscViewerASCIIPrintf(viewer," Cycles per PCApply=%d\n",mg->cyclesperpcapply);
512: }
513: if (mg->galerkin == PC_MG_GALERKIN_BOTH) {
514: PetscViewerASCIIPrintf(viewer," Using Galerkin computed coarse grid matrices\n");
515: } else if (mg->galerkin == PC_MG_GALERKIN_PMAT) {
516: PetscViewerASCIIPrintf(viewer," Using Galerkin computed coarse grid matrices for pmat\n");
517: } else if (mg->galerkin == PC_MG_GALERKIN_MAT) {
518: PetscViewerASCIIPrintf(viewer," Using Galerkin computed coarse grid matrices for mat\n");
519: } else if (mg->galerkin == PC_MG_GALERKIN_EXTERNAL) {
520: PetscViewerASCIIPrintf(viewer," Using externally compute Galerkin coarse grid matrices\n");
521: } else {
522: PetscViewerASCIIPrintf(viewer," Not using Galerkin computed coarse grid matrices\n");
523: }
524: if (mg->view){
525: (*mg->view)(pc,viewer);
526: }
527: for (i=0; i<levels; i++) {
528: if (!i) {
529: PetscViewerASCIIPrintf(viewer,"Coarse grid solver -- level -------------------------------\n",i);
530: } else {
531: PetscViewerASCIIPrintf(viewer,"Down solver (pre-smoother) on level %D -------------------------------\n",i);
532: }
533: PetscViewerASCIIPushTab(viewer);
534: KSPView(mglevels[i]->smoothd,viewer);
535: PetscViewerASCIIPopTab(viewer);
536: if (i && mglevels[i]->smoothd == mglevels[i]->smoothu) {
537: PetscViewerASCIIPrintf(viewer,"Up solver (post-smoother) same as down solver (pre-smoother)\n");
538: } else if (i) {
539: PetscViewerASCIIPrintf(viewer,"Up solver (post-smoother) on level %D -------------------------------\n",i);
540: PetscViewerASCIIPushTab(viewer);
541: KSPView(mglevels[i]->smoothu,viewer);
542: PetscViewerASCIIPopTab(viewer);
543: }
544: }
545: } else if (isbinary) {
546: for (i=levels-1; i>=0; i--) {
547: KSPView(mglevels[i]->smoothd,viewer);
548: if (i && mglevels[i]->smoothd != mglevels[i]->smoothu) {
549: KSPView(mglevels[i]->smoothu,viewer);
550: }
551: }
552: } else if (isdraw) {
553: PetscDraw draw;
554: PetscReal x,w,y,bottom,th;
555: PetscViewerDrawGetDraw(viewer,0,&draw);
556: PetscDrawGetCurrentPoint(draw,&x,&y);
557: PetscDrawStringGetSize(draw,NULL,&th);
558: bottom = y - th;
559: for (i=levels-1; i>=0; i--) {
560: if (!mglevels[i]->smoothu || (mglevels[i]->smoothu == mglevels[i]->smoothd)) {
561: PetscDrawPushCurrentPoint(draw,x,bottom);
562: KSPView(mglevels[i]->smoothd,viewer);
563: PetscDrawPopCurrentPoint(draw);
564: } else {
565: w = 0.5*PetscMin(1.0-x,x);
566: PetscDrawPushCurrentPoint(draw,x+w,bottom);
567: KSPView(mglevels[i]->smoothd,viewer);
568: PetscDrawPopCurrentPoint(draw);
569: PetscDrawPushCurrentPoint(draw,x-w,bottom);
570: KSPView(mglevels[i]->smoothu,viewer);
571: PetscDrawPopCurrentPoint(draw);
572: }
573: PetscDrawGetBoundingBox(draw,NULL,&bottom,NULL,NULL);
574: bottom -= th;
575: }
576: }
577: return(0);
578: }
580: #include <petsc/private/dmimpl.h>
581: #include <petsc/private/kspimpl.h>
583: /*
584: Calls setup for the KSP on each level
585: */
586: PetscErrorCode PCSetUp_MG(PC pc)
587: {
588: PC_MG *mg = (PC_MG*)pc->data;
589: PC_MG_Levels **mglevels = mg->levels;
591: PetscInt i,n;
592: PC cpc;
593: PetscBool dump = PETSC_FALSE,opsset,use_amat,missinginterpolate = PETSC_FALSE;
594: Mat dA,dB;
595: Vec tvec;
596: DM *dms;
597: PetscViewer viewer = 0;
598: PetscBool dAeqdB = PETSC_FALSE, needRestricts = PETSC_FALSE;
601: if (!mglevels) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels with PCMGSetLevels() before setting up");
602: n = mglevels[0]->levels;
603: /* FIX: Move this to PCSetFromOptions_MG? */
604: if (mg->usedmfornumberoflevels) {
605: PetscInt levels;
606: DMGetRefineLevel(pc->dm,&levels);
607: levels++;
608: if (levels > n) { /* the problem is now being solved on a finer grid */
609: PCMGSetLevels(pc,levels,NULL);
610: n = levels;
611: PCSetFromOptions(pc); /* it is bad to call this here, but otherwise will never be called for the new hierarchy */
612: mglevels = mg->levels;
613: }
614: }
615: KSPGetPC(mglevels[0]->smoothd,&cpc);
618: /* If user did not provide fine grid operators OR operator was not updated since last global KSPSetOperators() */
619: /* so use those from global PC */
620: /* Is this what we always want? What if user wants to keep old one? */
621: KSPGetOperatorsSet(mglevels[n-1]->smoothd,NULL,&opsset);
622: if (opsset) {
623: Mat mmat;
624: KSPGetOperators(mglevels[n-1]->smoothd,NULL,&mmat);
625: if (mmat == pc->pmat) opsset = PETSC_FALSE;
626: }
628: if (!opsset) {
629: PCGetUseAmat(pc,&use_amat);
630: if (use_amat) {
631: PetscInfo(pc,"Using outer operators to define finest grid operator \n because PCMGGetSmoother(pc,nlevels-1,&ksp);KSPSetOperators(ksp,...); was not called.\n");
632: KSPSetOperators(mglevels[n-1]->smoothd,pc->mat,pc->pmat);
633: } else {
634: PetscInfo(pc,"Using matrix (pmat) operators to define finest grid operator \n because PCMGGetSmoother(pc,nlevels-1,&ksp);KSPSetOperators(ksp,...); was not called.\n");
635: KSPSetOperators(mglevels[n-1]->smoothd,pc->pmat,pc->pmat);
636: }
637: }
639: for (i=n-1; i>0; i--) {
640: if (!(mglevels[i]->interpolate || mglevels[i]->restrct)) {
641: missinginterpolate = PETSC_TRUE;
642: continue;
643: }
644: }
646: KSPGetOperators(mglevels[n-1]->smoothd,&dA,&dB);
647: if (dA == dB) dAeqdB = PETSC_TRUE;
648: if ((mg->galerkin == PC_MG_GALERKIN_NONE) || (((mg->galerkin == PC_MG_GALERKIN_PMAT) || (mg->galerkin == PC_MG_GALERKIN_MAT)) && !dAeqdB)) {
649: needRestricts = PETSC_TRUE; /* user must compute either mat, pmat, or both so must restrict x to coarser levels */
650: }
653: /*
654: Skipping if user has provided all interpolation/restriction needed (since DM might not be able to produce them (when coming from SNES/TS)
655: Skipping for galerkin==2 (externally managed hierarchy such as ML and GAMG). Cleaner logic here would be great. Wrap ML/GAMG as DMs?
656: */
657: if (missinginterpolate && pc->dm && mg->galerkin != PC_MG_GALERKIN_EXTERNAL && !pc->setupcalled) {
658: /* construct the interpolation from the DMs */
659: Mat p;
660: Vec rscale;
661: PetscMalloc1(n,&dms);
662: dms[n-1] = pc->dm;
663: /* Separately create them so we do not get DMKSP interference between levels */
664: for (i=n-2; i>-1; i--) {DMCoarsen(dms[i+1],MPI_COMM_NULL,&dms[i]);}
665: /*
666: Force the mat type of coarse level operator to be AIJ because usually we want to use LU for coarse level.
667: Notice that it can be overwritten by -mat_type because KSPSetUp() reads command line options.
668: But it is safe to use -dm_mat_type.
670: The mat type should not be hardcoded like this, we need to find a better way.
671: DMSetMatType(dms[0],MATAIJ);
672: */
673: for (i=n-2; i>-1; i--) {
674: DMKSP kdm;
675: PetscBool dmhasrestrict, dmhasinject;
676: KSPSetDM(mglevels[i]->smoothd,dms[i]);
677: if (!needRestricts) {KSPSetDMActive(mglevels[i]->smoothd,PETSC_FALSE);}
678: if (mglevels[i]->smoothd != mglevels[i]->smoothu) {
679: KSPSetDM(mglevels[i]->smoothu,dms[i]);
680: if (!needRestricts) {KSPSetDMActive(mglevels[i]->smoothu,PETSC_FALSE);}
681: }
682: DMGetDMKSPWrite(dms[i],&kdm);
683: /* Ugly hack so that the next KSPSetUp() will use the RHS that we set. A better fix is to change dmActive to take
684: * a bitwise OR of computing the matrix, RHS, and initial iterate. */
685: kdm->ops->computerhs = NULL;
686: kdm->rhsctx = NULL;
687: if (!mglevels[i+1]->interpolate) {
688: DMCreateInterpolation(dms[i],dms[i+1],&p,&rscale);
689: PCMGSetInterpolation(pc,i+1,p);
690: if (rscale) {PCMGSetRScale(pc,i+1,rscale);}
691: VecDestroy(&rscale);
692: MatDestroy(&p);
693: }
694: DMHasCreateRestriction(dms[i],&dmhasrestrict);
695: if (dmhasrestrict && !mglevels[i+1]->restrct){
696: DMCreateRestriction(dms[i],dms[i+1],&p);
697: PCMGSetRestriction(pc,i+1,p);
698: MatDestroy(&p);
699: }
700: DMHasCreateInjection(dms[i],&dmhasinject);
701: if (dmhasinject && !mglevels[i+1]->inject){
702: DMCreateInjection(dms[i],dms[i+1],&p);
703: PCMGSetInjection(pc,i+1,p);
704: MatDestroy(&p);
705: }
706: }
708: for (i=n-2; i>-1; i--) {DMDestroy(&dms[i]);}
709: PetscFree(dms);
710: }
712: if (pc->dm && !pc->setupcalled) {
713: /* finest smoother also gets DM but it is not active, independent of whether galerkin==PC_MG_GALERKIN_EXTERNAL */
714: KSPSetDM(mglevels[n-1]->smoothd,pc->dm);
715: KSPSetDMActive(mglevels[n-1]->smoothd,PETSC_FALSE);
716: if (mglevels[n-1]->smoothd != mglevels[n-1]->smoothu) {
717: KSPSetDM(mglevels[n-1]->smoothu,pc->dm);
718: KSPSetDMActive(mglevels[n-1]->smoothu,PETSC_FALSE);
719: }
720: }
722: if (mg->galerkin < PC_MG_GALERKIN_NONE) {
723: Mat A,B;
724: PetscBool doA = PETSC_FALSE,doB = PETSC_FALSE;
725: MatReuse reuse = MAT_INITIAL_MATRIX;
727: if ((mg->galerkin == PC_MG_GALERKIN_PMAT) || (mg->galerkin == PC_MG_GALERKIN_BOTH)) doB = PETSC_TRUE;
728: if ((mg->galerkin == PC_MG_GALERKIN_MAT) || ((mg->galerkin == PC_MG_GALERKIN_BOTH) && (dA != dB))) doA = PETSC_TRUE;
729: if (pc->setupcalled) reuse = MAT_REUSE_MATRIX;
730: for (i=n-2; i>-1; i--) {
731: if (!mglevels[i+1]->restrct && !mglevels[i+1]->interpolate) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_ARG_WRONGSTATE,"Must provide interpolation or restriction for each MG level except level 0");
732: if (!mglevels[i+1]->interpolate) {
733: PCMGSetInterpolation(pc,i+1,mglevels[i+1]->restrct);
734: }
735: if (!mglevels[i+1]->restrct) {
736: PCMGSetRestriction(pc,i+1,mglevels[i+1]->interpolate);
737: }
738: if (reuse == MAT_REUSE_MATRIX) {
739: KSPGetOperators(mglevels[i]->smoothd,&A,&B);
740: }
741: if (doA) {
742: MatGalerkin(mglevels[i+1]->restrct,dA,mglevels[i+1]->interpolate,reuse,1.0,&A);
743: }
744: if (doB) {
745: MatGalerkin(mglevels[i+1]->restrct,dB,mglevels[i+1]->interpolate,reuse,1.0,&B);
746: }
747: /* the management of the PetscObjectReference() and PetscObjecDereference() below is rather delicate */
748: if (!doA && dAeqdB) {
749: if (reuse == MAT_INITIAL_MATRIX) {PetscObjectReference((PetscObject)B);}
750: A = B;
751: } else if (!doA && reuse == MAT_INITIAL_MATRIX ) {
752: KSPGetOperators(mglevels[i]->smoothd,&A,NULL);
753: PetscObjectReference((PetscObject)A);
754: }
755: if (!doB && dAeqdB) {
756: if (reuse == MAT_INITIAL_MATRIX) {PetscObjectReference((PetscObject)A);}
757: B = A;
758: } else if (!doB && reuse == MAT_INITIAL_MATRIX) {
759: KSPGetOperators(mglevels[i]->smoothd,NULL,&B);
760: PetscObjectReference((PetscObject)B);
761: }
762: if (reuse == MAT_INITIAL_MATRIX) {
763: KSPSetOperators(mglevels[i]->smoothd,A,B);
764: PetscObjectDereference((PetscObject)A);
765: PetscObjectDereference((PetscObject)B);
766: }
767: dA = A;
768: dB = B;
769: }
770: }
771: if (needRestricts && pc->dm) {
772: for (i=n-2; i>=0; i--) {
773: DM dmfine,dmcoarse;
774: Mat Restrict,Inject;
775: Vec rscale;
776: KSPGetDM(mglevels[i+1]->smoothd,&dmfine);
777: KSPGetDM(mglevels[i]->smoothd,&dmcoarse);
778: PCMGGetRestriction(pc,i+1,&Restrict);
779: PCMGGetRScale(pc,i+1,&rscale);
780: PCMGGetInjection(pc,i+1,&Inject);
781: DMRestrict(dmfine,Restrict,rscale,Inject,dmcoarse);
782: }
783: }
785: if (!pc->setupcalled) {
786: for (i=0; i<n; i++) {
787: KSPSetFromOptions(mglevels[i]->smoothd);
788: }
789: for (i=1; i<n; i++) {
790: if (mglevels[i]->smoothu && (mglevels[i]->smoothu != mglevels[i]->smoothd)) {
791: KSPSetFromOptions(mglevels[i]->smoothu);
792: }
793: }
794: /* insure that if either interpolation or restriction is set the other other one is set */
795: for (i=1; i<n; i++) {
796: PCMGGetInterpolation(pc,i,NULL);
797: PCMGGetRestriction(pc,i,NULL);
798: }
799: for (i=0; i<n-1; i++) {
800: if (!mglevels[i]->b) {
801: Vec *vec;
802: KSPCreateVecs(mglevels[i]->smoothd,1,&vec,0,NULL);
803: PCMGSetRhs(pc,i,*vec);
804: VecDestroy(vec);
805: PetscFree(vec);
806: }
807: if (!mglevels[i]->r && i) {
808: VecDuplicate(mglevels[i]->b,&tvec);
809: PCMGSetR(pc,i,tvec);
810: VecDestroy(&tvec);
811: }
812: if (!mglevels[i]->x) {
813: VecDuplicate(mglevels[i]->b,&tvec);
814: PCMGSetX(pc,i,tvec);
815: VecDestroy(&tvec);
816: }
817: }
818: if (n != 1 && !mglevels[n-1]->r) {
819: /* PCMGSetR() on the finest level if user did not supply it */
820: Vec *vec;
821: KSPCreateVecs(mglevels[n-1]->smoothd,1,&vec,0,NULL);
822: PCMGSetR(pc,n-1,*vec);
823: VecDestroy(vec);
824: PetscFree(vec);
825: }
826: }
828: if (pc->dm) {
829: /* need to tell all the coarser levels to rebuild the matrix using the DM for that level */
830: for (i=0; i<n-1; i++) {
831: if (mglevels[i]->smoothd->setupstage != KSP_SETUP_NEW) mglevels[i]->smoothd->setupstage = KSP_SETUP_NEWMATRIX;
832: }
833: }
835: for (i=1; i<n; i++) {
836: if (mglevels[i]->smoothu == mglevels[i]->smoothd || mg->am == PC_MG_FULL || mg->am == PC_MG_KASKADE || mg->cyclesperpcapply > 1){
837: /* if doing only down then initial guess is zero */
838: KSPSetInitialGuessNonzero(mglevels[i]->smoothd,PETSC_TRUE);
839: }
840: if (mglevels[i]->eventsmoothsetup) {PetscLogEventBegin(mglevels[i]->eventsmoothsetup,0,0,0,0);}
841: KSPSetUp(mglevels[i]->smoothd);
842: if (mglevels[i]->smoothd->reason == KSP_DIVERGED_PC_FAILED) {
843: pc->failedreason = PC_SUBPC_ERROR;
844: }
845: if (mglevels[i]->eventsmoothsetup) {PetscLogEventEnd(mglevels[i]->eventsmoothsetup,0,0,0,0);}
846: if (!mglevels[i]->residual) {
847: Mat mat;
848: KSPGetOperators(mglevels[i]->smoothd,&mat,NULL);
849: PCMGSetResidual(pc,i,PCMGResidualDefault,mat);
850: }
851: }
852: for (i=1; i<n; i++) {
853: if (mglevels[i]->smoothu && mglevels[i]->smoothu != mglevels[i]->smoothd) {
854: Mat downmat,downpmat;
856: /* check if operators have been set for up, if not use down operators to set them */
857: KSPGetOperatorsSet(mglevels[i]->smoothu,&opsset,NULL);
858: if (!opsset) {
859: KSPGetOperators(mglevels[i]->smoothd,&downmat,&downpmat);
860: KSPSetOperators(mglevels[i]->smoothu,downmat,downpmat);
861: }
863: KSPSetInitialGuessNonzero(mglevels[i]->smoothu,PETSC_TRUE);
864: if (mglevels[i]->eventsmoothsetup) {PetscLogEventBegin(mglevels[i]->eventsmoothsetup,0,0,0,0);}
865: KSPSetUp(mglevels[i]->smoothu);
866: if (mglevels[i]->smoothu->reason == KSP_DIVERGED_PC_FAILED) {
867: pc->failedreason = PC_SUBPC_ERROR;
868: }
869: if (mglevels[i]->eventsmoothsetup) {PetscLogEventEnd(mglevels[i]->eventsmoothsetup,0,0,0,0);}
870: }
871: }
873: if (mglevels[0]->eventsmoothsetup) {PetscLogEventBegin(mglevels[0]->eventsmoothsetup,0,0,0,0);}
874: KSPSetUp(mglevels[0]->smoothd);
875: if (mglevels[0]->smoothd->reason == KSP_DIVERGED_PC_FAILED) {
876: pc->failedreason = PC_SUBPC_ERROR;
877: }
878: if (mglevels[0]->eventsmoothsetup) {PetscLogEventEnd(mglevels[0]->eventsmoothsetup,0,0,0,0);}
880: /*
881: Dump the interpolation/restriction matrices plus the
882: Jacobian/stiffness on each level. This allows MATLAB users to
883: easily check if the Galerkin condition A_c = R A_f R^T is satisfied.
885: Only support one or the other at the same time.
886: */
887: #if defined(PETSC_USE_SOCKET_VIEWER)
888: PetscOptionsGetBool(((PetscObject)pc)->options,((PetscObject)pc)->prefix,"-pc_mg_dump_matlab",&dump,NULL);
889: if (dump) viewer = PETSC_VIEWER_SOCKET_(PetscObjectComm((PetscObject)pc));
890: dump = PETSC_FALSE;
891: #endif
892: PetscOptionsGetBool(((PetscObject)pc)->options,((PetscObject)pc)->prefix,"-pc_mg_dump_binary",&dump,NULL);
893: if (dump) viewer = PETSC_VIEWER_BINARY_(PetscObjectComm((PetscObject)pc));
895: if (viewer) {
896: for (i=1; i<n; i++) {
897: MatView(mglevels[i]->restrct,viewer);
898: }
899: for (i=0; i<n; i++) {
900: KSPGetPC(mglevels[i]->smoothd,&pc);
901: MatView(pc->mat,viewer);
902: }
903: }
904: return(0);
905: }
907: /* -------------------------------------------------------------------------------------*/
909: PetscErrorCode PCMGGetLevels_MG(PC pc, PetscInt *levels)
910: {
911: PC_MG *mg = (PC_MG *) pc->data;
914: *levels = mg->nlevels;
915: return(0);
916: }
918: /*@
919: PCMGGetLevels - Gets the number of levels to use with MG.
921: Not Collective
923: Input Parameter:
924: . pc - the preconditioner context
926: Output parameter:
927: . levels - the number of levels
929: Level: advanced
931: .seealso: PCMGSetLevels()
932: @*/
933: PetscErrorCode PCMGGetLevels(PC pc,PetscInt *levels)
934: {
940: *levels = 0;
941: PetscTryMethod(pc,"PCMGGetLevels_C",(PC,PetscInt*),(pc,levels));
942: return(0);
943: }
945: /*@
946: PCMGSetType - Determines the form of multigrid to use:
947: multiplicative, additive, full, or the Kaskade algorithm.
949: Logically Collective on PC
951: Input Parameters:
952: + pc - the preconditioner context
953: - form - multigrid form, one of PC_MG_MULTIPLICATIVE, PC_MG_ADDITIVE,
954: PC_MG_FULL, PC_MG_KASKADE
956: Options Database Key:
957: . -pc_mg_type <form> - Sets <form>, one of multiplicative,
958: additive, full, kaskade
960: Level: advanced
962: .seealso: PCMGSetLevels()
963: @*/
964: PetscErrorCode PCMGSetType(PC pc,PCMGType form)
965: {
966: PC_MG *mg = (PC_MG*)pc->data;
971: mg->am = form;
972: if (form == PC_MG_MULTIPLICATIVE) pc->ops->applyrichardson = PCApplyRichardson_MG;
973: else pc->ops->applyrichardson = NULL;
974: return(0);
975: }
977: /*@
978: PCMGGetType - Determines the form of multigrid to use:
979: multiplicative, additive, full, or the Kaskade algorithm.
981: Logically Collective on PC
983: Input Parameter:
984: . pc - the preconditioner context
986: Output Parameter:
987: . type - one of PC_MG_MULTIPLICATIVE, PC_MG_ADDITIVE,PC_MG_FULL, PC_MG_KASKADE
990: Level: advanced
992: .seealso: PCMGSetLevels()
993: @*/
994: PetscErrorCode PCMGGetType(PC pc,PCMGType *type)
995: {
996: PC_MG *mg = (PC_MG*)pc->data;
1000: *type = mg->am;
1001: return(0);
1002: }
1004: /*@
1005: PCMGSetCycleType - Sets the type cycles to use. Use PCMGSetCycleTypeOnLevel() for more
1006: complicated cycling.
1008: Logically Collective on PC
1010: Input Parameters:
1011: + pc - the multigrid context
1012: - n - either PC_MG_CYCLE_V or PC_MG_CYCLE_W
1014: Options Database Key:
1015: . -pc_mg_cycle_type <v,w> - provide the cycle desired
1017: Level: advanced
1019: .seealso: PCMGSetCycleTypeOnLevel()
1020: @*/
1021: PetscErrorCode PCMGSetCycleType(PC pc,PCMGCycleType n)
1022: {
1023: PC_MG *mg = (PC_MG*)pc->data;
1024: PC_MG_Levels **mglevels = mg->levels;
1025: PetscInt i,levels;
1030: if (!mglevels) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_ORDER,"Must set MG levels with PCMGSetLevels() before calling");
1031: levels = mglevels[0]->levels;
1032: for (i=0; i<levels; i++) mglevels[i]->cycles = n;
1033: return(0);
1034: }
1036: /*@
1037: PCMGMultiplicativeSetCycles - Sets the number of cycles to use for each preconditioner step
1038: of multigrid when PCMGType of PC_MG_MULTIPLICATIVE is used
1040: Logically Collective on PC
1042: Input Parameters:
1043: + pc - the multigrid context
1044: - n - number of cycles (default is 1)
1046: Options Database Key:
1047: . -pc_mg_multiplicative_cycles n
1049: Level: advanced
1051: Notes:
1052: This is not associated with setting a v or w cycle, that is set with PCMGSetCycleType()
1054: .seealso: PCMGSetCycleTypeOnLevel(), PCMGSetCycleType()
1055: @*/
1056: PetscErrorCode PCMGMultiplicativeSetCycles(PC pc,PetscInt n)
1057: {
1058: PC_MG *mg = (PC_MG*)pc->data;
1063: mg->cyclesperpcapply = n;
1064: return(0);
1065: }
1067: PetscErrorCode PCMGSetGalerkin_MG(PC pc,PCMGGalerkinType use)
1068: {
1069: PC_MG *mg = (PC_MG*)pc->data;
1072: mg->galerkin = use;
1073: return(0);
1074: }
1076: /*@
1077: PCMGSetGalerkin - Causes the coarser grid matrices to be computed from the
1078: finest grid via the Galerkin process: A_i-1 = r_i * A_i * p_i
1080: Logically Collective on PC
1082: Input Parameters:
1083: + pc - the multigrid context
1084: - use - one of PC_MG_GALERKIN_BOTH,PC_MG_GALERKIN_PMAT,PC_MG_GALERKIN_MAT, or PC_MG_GALERKIN_NONE
1086: Options Database Key:
1087: . -pc_mg_galerkin <both,pmat,mat,none>
1089: Level: intermediate
1091: Notes:
1092: Some codes that use PCMG such as PCGAMG use Galerkin internally while constructing the hierarchy and thus do not
1093: use the PCMG construction of the coarser grids.
1095: .seealso: PCMGGetGalerkin(), PCMGGalerkinType
1097: @*/
1098: PetscErrorCode PCMGSetGalerkin(PC pc,PCMGGalerkinType use)
1099: {
1104: PetscTryMethod(pc,"PCMGSetGalerkin_C",(PC,PCMGGalerkinType),(pc,use));
1105: return(0);
1106: }
1108: /*@
1109: PCMGGetGalerkin - Checks if Galerkin multigrid is being used, i.e.
1110: A_i-1 = r_i * A_i * p_i
1112: Not Collective
1114: Input Parameter:
1115: . pc - the multigrid context
1117: Output Parameter:
1118: . galerkin - one of PC_MG_GALERKIN_BOTH,PC_MG_GALERKIN_PMAT,PC_MG_GALERKIN_MAT, PC_MG_GALERKIN_NONE, or PC_MG_GALERKIN_EXTERNAL
1120: Level: intermediate
1122: .seealso: PCMGSetGalerkin(), PCMGGalerkinType
1124: @*/
1125: PetscErrorCode PCMGGetGalerkin(PC pc,PCMGGalerkinType *galerkin)
1126: {
1127: PC_MG *mg = (PC_MG*)pc->data;
1131: *galerkin = mg->galerkin;
1132: return(0);
1133: }
1135: /*@
1136: PCMGSetNumberSmooth - Sets the number of pre and post-smoothing steps to use
1137: on all levels. Use PCMGDistinctSmoothUp() to create separate up and down smoothers if you want different numbers of
1138: pre- and post-smoothing steps.
1140: Logically Collective on PC
1142: Input Parameters:
1143: + mg - the multigrid context
1144: - n - the number of smoothing steps
1146: Options Database Key:
1147: . -mg_levels_ksp_max_it <n> - Sets number of pre and post-smoothing steps
1149: Level: advanced
1151: Notes:
1152: this does not set a value on the coarsest grid, since we assume that
1153: there is no separate smooth up on the coarsest grid.
1155: .seealso: PCMGSetDistinctSmoothUp()
1156: @*/
1157: PetscErrorCode PCMGSetNumberSmooth(PC pc,PetscInt n)
1158: {
1159: PC_MG *mg = (PC_MG*)pc->data;
1160: PC_MG_Levels **mglevels = mg->levels;
1162: PetscInt i,levels;
1167: if (!mglevels) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_ORDER,"Must set MG levels with PCMGSetLevels() before calling");
1168: levels = mglevels[0]->levels;
1170: for (i=1; i<levels; i++) {
1171: KSPSetTolerances(mglevels[i]->smoothu,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT,n);
1172: KSPSetTolerances(mglevels[i]->smoothd,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT,n);
1173: mg->default_smoothu = n;
1174: mg->default_smoothd = n;
1175: }
1176: return(0);
1177: }
1179: /*@
1180: PCMGSetDistinctSmoothUp - sets the up (post) smoother to be a separate KSP from the down (pre) smoother on all levels
1181: and adds the suffix _up to the options name
1183: Logically Collective on PC
1185: Input Parameters:
1186: . pc - the preconditioner context
1188: Options Database Key:
1189: . -pc_mg_distinct_smoothup
1191: Level: advanced
1193: Notes:
1194: this does not set a value on the coarsest grid, since we assume that
1195: there is no separate smooth up on the coarsest grid.
1197: .seealso: PCMGSetNumberSmooth()
1198: @*/
1199: PetscErrorCode PCMGSetDistinctSmoothUp(PC pc)
1200: {
1201: PC_MG *mg = (PC_MG*)pc->data;
1202: PC_MG_Levels **mglevels = mg->levels;
1204: PetscInt i,levels;
1205: KSP subksp;
1209: if (!mglevels) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_ORDER,"Must set MG levels with PCMGSetLevels() before calling");
1210: levels = mglevels[0]->levels;
1212: for (i=1; i<levels; i++) {
1213: const char *prefix = NULL;
1214: /* make sure smoother up and down are different */
1215: PCMGGetSmootherUp(pc,i,&subksp);
1216: KSPGetOptionsPrefix(mglevels[i]->smoothd,&prefix);
1217: KSPSetOptionsPrefix(subksp,prefix);
1218: KSPAppendOptionsPrefix(subksp,"up_");
1219: }
1220: return(0);
1221: }
1223: /* No new matrices are created, and the coarse operator matrices are the references to the original ones */
1224: PetscErrorCode PCGetInterpolations_MG(PC pc,PetscInt *num_levels,Mat *interpolations[])
1225: {
1226: PC_MG *mg = (PC_MG*)pc->data;
1227: PC_MG_Levels **mglevels = mg->levels;
1228: Mat *mat;
1229: PetscInt l;
1233: if (!mglevels) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling");
1234: PetscMalloc1(mg->nlevels,&mat);
1235: for (l=1; l< mg->nlevels; l++) {
1236: mat[l-1] = mglevels[l]->interpolate;
1237: PetscObjectReference((PetscObject)mat[l-1]);
1238: }
1239: *num_levels = mg->nlevels;
1240: *interpolations = mat;
1241: return(0);
1242: }
1244: /* No new matrices are created, and the coarse operator matrices are the references to the original ones */
1245: PetscErrorCode PCGetCoarseOperators_MG(PC pc,PetscInt *num_levels,Mat *coarseOperators[])
1246: {
1247: PC_MG *mg = (PC_MG*)pc->data;
1248: PC_MG_Levels **mglevels = mg->levels;
1249: PetscInt l;
1250: Mat *mat;
1254: if (!mglevels) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling");
1255: PetscMalloc1(mg->nlevels,&mat);
1256: for (l=0; l<mg->nlevels-1; l++) {
1257: KSPGetOperators(mglevels[l]->smoothd,NULL,&(mat[l]));
1258: PetscObjectReference((PetscObject)mat[l]);
1259: }
1260: *num_levels = mg->nlevels;
1261: *coarseOperators = mat;
1262: return(0);
1263: }
1265: /* ----------------------------------------------------------------------------------------*/
1267: /*MC
1268: PCMG - Use multigrid preconditioning. This preconditioner requires you provide additional
1269: information about the coarser grid matrices and restriction/interpolation operators.
1271: Options Database Keys:
1272: + -pc_mg_levels <nlevels> - number of levels including finest
1273: . -pc_mg_cycle_type <v,w> - provide the cycle desired
1274: . -pc_mg_type <additive,multiplicative,full,kaskade> - multiplicative is the default
1275: . -pc_mg_log - log information about time spent on each level of the solver
1276: . -pc_mg_distinct_smoothup - configure up (after interpolation) and down (before restriction) smoothers separately (with different options prefixes)
1277: . -pc_mg_galerkin <both,pmat,mat,none> - use Galerkin process to compute coarser operators, i.e. Acoarse = R A R'
1278: . -pc_mg_multiplicative_cycles - number of cycles to use as the preconditioner (defaults to 1)
1279: . -pc_mg_dump_matlab - dumps the matrices for each level and the restriction/interpolation matrices
1280: to the Socket viewer for reading from MATLAB.
1281: - -pc_mg_dump_binary - dumps the matrices for each level and the restriction/interpolation matrices
1282: to the binary output file called binaryoutput
1284: Notes:
1285: 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
1287: When run with a single level the smoother options are used on that level NOT the coarse grid solver options
1289: When run with KSPRICHARDSON the convergence test changes slightly if monitor is turned on. The iteration count may change slightly. This
1290: is because without monitoring the residual norm is computed WITHIN each multigrid cycle on the finest level after the pre-smoothing
1291: (because the residual has just been computed for the multigrid algorithm and is hence available for free) while with monitoring the
1292: residual is computed at the end of each cycle.
1294: Level: intermediate
1296: .seealso: PCCreate(), PCSetType(), PCType (for list of available types), PC, PCMGType, PCEXOTIC, PCGAMG, PCML, PCHYPRE
1297: PCMGSetLevels(), PCMGGetLevels(), PCMGSetType(), PCMGSetCycleType(),
1298: PCMGSetDistinctSmoothUp(), PCMGGetCoarseSolve(), PCMGSetResidual(), PCMGSetInterpolation(),
1299: PCMGSetRestriction(), PCMGGetSmoother(), PCMGGetSmootherUp(), PCMGGetSmootherDown(),
1300: PCMGSetCycleTypeOnLevel(), PCMGSetRhs(), PCMGSetX(), PCMGSetR()
1301: M*/
1303: PETSC_EXTERN PetscErrorCode PCCreate_MG(PC pc)
1304: {
1305: PC_MG *mg;
1309: PetscNewLog(pc,&mg);
1310: pc->data = (void*)mg;
1311: mg->nlevels = -1;
1312: mg->am = PC_MG_MULTIPLICATIVE;
1313: mg->galerkin = PC_MG_GALERKIN_NONE;
1315: pc->useAmat = PETSC_TRUE;
1317: pc->ops->apply = PCApply_MG;
1318: pc->ops->setup = PCSetUp_MG;
1319: pc->ops->reset = PCReset_MG;
1320: pc->ops->destroy = PCDestroy_MG;
1321: pc->ops->setfromoptions = PCSetFromOptions_MG;
1322: pc->ops->view = PCView_MG;
1324: PetscObjectComposeFunction((PetscObject)pc,"PCMGSetGalerkin_C",PCMGSetGalerkin_MG);
1325: PetscObjectComposeFunction((PetscObject)pc,"PCMGGetLevels_C",PCMGGetLevels_MG);
1326: PetscObjectComposeFunction((PetscObject)pc,"PCMGSetLevels_C",PCMGSetLevels_MG);
1327: PetscObjectComposeFunction((PetscObject)pc,"PCGetInterpolations_C",PCGetInterpolations_MG);
1328: PetscObjectComposeFunction((PetscObject)pc,"PCGetCoarseOperators_C",PCGetCoarseOperators_MG);
1329: return(0);
1330: }