xref: /petsc/src/ksp/pc/impls/mg/mg.c (revision fccaa45e966e97669b5dcb5b296f1f67be099458)
1 #define PETSCKSP_DLL
2 
3 /*
4     Defines the multigrid preconditioner interface.
5 */
6 #include "src/ksp/pc/impls/mg/mgimpl.h"                    /*I "petscmg.h" I*/
7 
8 
9 /*
10        MGMCycle_Private - Given an MG structure created with MGCreate() runs
11                   one multiplicative cycle down through the levels and
12                   back up.
13 
14     Input Parameter:
15 .   mg - structure created with  MGCreate().
16 */
17 #undef __FUNCT__
18 #define __FUNCT__ "MGMCycle_Private"
19 PetscErrorCode MGMCycle_Private(MG *mglevels,PetscTruth *converged)
20 {
21   MG             mg = *mglevels,mgc;
22   PetscErrorCode ierr;
23   PetscInt       cycles = mg->cycles;
24   PetscScalar    zero = 0.0;
25 
26   PetscFunctionBegin;
27   if (converged) *converged = PETSC_FALSE;
28 
29   if (mg->eventsolve) {ierr = PetscLogEventBegin(mg->eventsolve,0,0,0,0);CHKERRQ(ierr);}
30   ierr = KSPSolve(mg->smoothd,mg->b,mg->x);CHKERRQ(ierr);
31   if (mg->eventsolve) {ierr = PetscLogEventEnd(mg->eventsolve,0,0,0,0);CHKERRQ(ierr);}
32   if (mg->level) {  /* not the coarsest grid */
33     ierr = (*mg->residual)(mg->A,mg->b,mg->x,mg->r);CHKERRQ(ierr);
34 
35     /* if on finest level and have convergence criteria set */
36     if (mg->level == mg->levels-1 && mg->ttol) {
37       PetscReal rnorm;
38       ierr = VecNorm(mg->r,NORM_2,&rnorm);CHKERRQ(ierr);
39       if (rnorm <= mg->ttol) {
40         *converged = PETSC_TRUE;
41         if (rnorm < mg->abstol) {
42           ierr = PetscLogInfo((0,"MGMCycle_Private:Linear solver has converged. Residual norm %g is less than absolute tolerance %g\n",rnorm,mg->abstol));CHKERRQ(ierr);
43         } else {
44           ierr = PetscLogInfo((0,"MGMCycle_Private:Linear solver has converged. Residual norm %g is less than relative tolerance times initial residual norm %g\n",rnorm,mg->ttol));CHKERRQ(ierr);
45         }
46         PetscFunctionReturn(0);
47       }
48     }
49 
50     mgc = *(mglevels - 1);
51     ierr = MatRestrict(mg->restrct,mg->r,mgc->b);CHKERRQ(ierr);
52     ierr = VecSet(&zero,mgc->x);CHKERRQ(ierr);
53     while (cycles--) {
54       ierr = MGMCycle_Private(mglevels-1,converged);CHKERRQ(ierr);
55     }
56     ierr = MatInterpolateAdd(mg->interpolate,mgc->x,mg->x,mg->x);CHKERRQ(ierr);
57     if (mg->eventsolve) {ierr = PetscLogEventBegin(mg->eventsolve,0,0,0,0);CHKERRQ(ierr);}
58     ierr = KSPSolve(mg->smoothu,mg->b,mg->x);CHKERRQ(ierr);
59     if (mg->eventsolve) {ierr = PetscLogEventEnd(mg->eventsolve,0,0,0,0);CHKERRQ(ierr);}
60   }
61   PetscFunctionReturn(0);
62 }
63 
64 /*
65        MGCreate_Private - Creates a MG structure for use with the
66                multigrid code. Level 0 is the coarsest. (But the
67                finest level is stored first in the array).
68 
69 */
70 #undef __FUNCT__
71 #define __FUNCT__ "MGCreate_Private"
72 static PetscErrorCode MGCreate_Private(MPI_Comm comm,PetscInt levels,PC pc,MPI_Comm *comms,MG **result)
73 {
74   MG             *mg;
75   PetscErrorCode ierr;
76   PetscInt       i;
77   PetscMPIInt    size;
78   char           *prefix;
79   PC             ipc;
80 
81   PetscFunctionBegin;
82   ierr = PetscMalloc(levels*sizeof(MG),&mg);CHKERRQ(ierr);
83   ierr = PetscLogObjectMemory(pc,levels*(sizeof(MG)+sizeof(struct _MG)));CHKERRQ(ierr);
84 
85   ierr = PCGetOptionsPrefix(pc,&prefix);CHKERRQ(ierr);
86 
87   for (i=0; i<levels; i++) {
88     ierr = PetscNew(struct _MG,&mg[i]);CHKERRQ(ierr);
89     mg[i]->level           = i;
90     mg[i]->levels          = levels;
91     mg[i]->cycles          = 1;
92     mg[i]->galerkin        = PETSC_FALSE;
93     mg[i]->galerkinused    = PETSC_FALSE;
94     mg[i]->default_smoothu = 1;
95     mg[i]->default_smoothd = 1;
96 
97     if (comms) comm = comms[i];
98     ierr = KSPCreate(comm,&mg[i]->smoothd);CHKERRQ(ierr);
99     ierr = KSPSetTolerances(mg[i]->smoothd,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT, mg[i]->default_smoothd);CHKERRQ(ierr);
100     ierr = KSPSetOptionsPrefix(mg[i]->smoothd,prefix);CHKERRQ(ierr);
101 
102     /* do special stuff for coarse grid */
103     if (!i && levels > 1) {
104       ierr = KSPAppendOptionsPrefix(mg[0]->smoothd,"mg_coarse_");CHKERRQ(ierr);
105 
106       /* coarse solve is (redundant) LU by default */
107       ierr = KSPSetType(mg[0]->smoothd,KSPPREONLY);CHKERRQ(ierr);
108       ierr = KSPGetPC(mg[0]->smoothd,&ipc);CHKERRQ(ierr);
109       ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr);
110       if (size > 1) {
111         ierr = PCSetType(ipc,PCREDUNDANT);CHKERRQ(ierr);
112         ierr = PCRedundantGetPC(ipc,&ipc);CHKERRQ(ierr);
113       }
114       ierr = PCSetType(ipc,PCLU);CHKERRQ(ierr);
115 
116     } else {
117       char tprefix[128];
118       sprintf(tprefix,"mg_levels_%d_",(int)i);
119       ierr = KSPAppendOptionsPrefix(mg[i]->smoothd,tprefix);CHKERRQ(ierr);
120     }
121     ierr = PetscLogObjectParent(pc,mg[i]->smoothd);CHKERRQ(ierr);
122     mg[i]->smoothu         = mg[i]->smoothd;
123     mg[i]->rtol = 0.0;
124     mg[i]->abstol = 0.0;
125     mg[i]->dtol = 0.0;
126     mg[i]->ttol = 0.0;
127     mg[i]->eventsetup = 0;
128     mg[i]->eventsolve = 0;
129   }
130   *result = mg;
131   PetscFunctionReturn(0);
132 }
133 
134 #undef __FUNCT__
135 #define __FUNCT__ "PCDestroy_MG"
136 static PetscErrorCode PCDestroy_MG(PC pc)
137 {
138   MG             *mg = (MG*)pc->data;
139   PetscErrorCode ierr;
140   PetscInt       i,n = mg[0]->levels;
141 
142   PetscFunctionBegin;
143   if (mg[0]->galerkinused) {
144     Mat B;
145     for (i=0; i<n-1; i++) {
146       ierr = KSPGetOperators(mg[i]->smoothd,0,&B,0);CHKERRQ(ierr);
147       ierr = MatDestroy(B);CHKERRQ(ierr);
148     }
149   }
150 
151   for (i=0; i<n-1; i++) {
152     if (mg[i]->r) {ierr = VecDestroy(mg[i]->r);CHKERRQ(ierr);}
153     if (mg[i]->b) {ierr = VecDestroy(mg[i]->b);CHKERRQ(ierr);}
154     if (mg[i]->x) {ierr = VecDestroy(mg[i]->x);CHKERRQ(ierr);}
155     if (mg[i+1]->restrct) {ierr = MatDestroy(mg[i+1]->restrct);CHKERRQ(ierr);}
156     if (mg[i+1]->interpolate) {ierr = MatDestroy(mg[i+1]->interpolate);CHKERRQ(ierr);}
157   }
158   if (mg[n-1]->r) {ierr = VecDestroy(mg[n-1]->r);CHKERRQ(ierr);}
159 
160   for (i=0; i<n; i++) {
161     if (mg[i]->smoothd != mg[i]->smoothu) {
162       ierr = KSPDestroy(mg[i]->smoothd);CHKERRQ(ierr);
163     }
164     ierr = KSPDestroy(mg[i]->smoothu);CHKERRQ(ierr);
165     ierr = PetscFree(mg[i]);CHKERRQ(ierr);
166   }
167   ierr = PetscFree(mg);CHKERRQ(ierr);
168   PetscFunctionReturn(0);
169 }
170 
171 
172 
173 EXTERN PetscErrorCode MGACycle_Private(MG*);
174 EXTERN PetscErrorCode MGFCycle_Private(MG*);
175 EXTERN PetscErrorCode MGKCycle_Private(MG*);
176 
177 /*
178    PCApply_MG - Runs either an additive, multiplicative, Kaskadic
179              or full cycle of multigrid.
180 
181   Note:
182   A simple wrapper which calls MGMCycle(),MGACycle(), or MGFCycle().
183 */
184 #undef __FUNCT__
185 #define __FUNCT__ "PCApply_MG"
186 static PetscErrorCode PCApply_MG(PC pc,Vec b,Vec x)
187 {
188   MG             *mg = (MG*)pc->data;
189   PetscScalar    zero = 0.0;
190   PetscErrorCode ierr;
191   PetscInt       levels = mg[0]->levels;
192 
193   PetscFunctionBegin;
194   mg[levels-1]->b = b;
195   mg[levels-1]->x = x;
196   if (mg[0]->am == MGMULTIPLICATIVE) {
197     ierr = VecSet(&zero,x);CHKERRQ(ierr);
198     ierr = MGMCycle_Private(mg+levels-1,PETSC_NULL);CHKERRQ(ierr);
199   }
200   else if (mg[0]->am == MGADDITIVE) {
201     ierr = MGACycle_Private(mg);CHKERRQ(ierr);
202   }
203   else if (mg[0]->am == MGKASKADE) {
204     ierr = MGKCycle_Private(mg);CHKERRQ(ierr);
205   }
206   else {
207     ierr = MGFCycle_Private(mg);CHKERRQ(ierr);
208   }
209   PetscFunctionReturn(0);
210 }
211 
212 #undef __FUNCT__
213 #define __FUNCT__ "PCApplyRichardson_MG"
214 static PetscErrorCode PCApplyRichardson_MG(PC pc,Vec b,Vec x,Vec w,PetscReal rtol,PetscReal abstol, PetscReal dtol,PetscInt its)
215 {
216   MG             *mg = (MG*)pc->data;
217   PetscErrorCode ierr;
218   PetscInt       levels = mg[0]->levels;
219   PetscTruth     converged = PETSC_FALSE;
220 
221   PetscFunctionBegin;
222   mg[levels-1]->b    = b;
223   mg[levels-1]->x    = x;
224 
225   mg[levels-1]->rtol = rtol;
226   mg[levels-1]->abstol = abstol;
227   mg[levels-1]->dtol = dtol;
228   if (rtol) {
229     /* compute initial residual norm for relative convergence test */
230     PetscReal rnorm;
231     ierr               = (*mg[levels-1]->residual)(mg[levels-1]->A,b,x,w);CHKERRQ(ierr);
232     ierr               = VecNorm(w,NORM_2,&rnorm);CHKERRQ(ierr);
233     mg[levels-1]->ttol = PetscMax(rtol*rnorm,abstol);
234   } else if (abstol) {
235     mg[levels-1]->ttol = abstol;
236   } else {
237     mg[levels-1]->ttol = 0.0;
238   }
239 
240   while (its-- && !converged) {
241     ierr = MGMCycle_Private(mg+levels-1,&converged);CHKERRQ(ierr);
242   }
243   PetscFunctionReturn(0);
244 }
245 
246 #undef __FUNCT__
247 #define __FUNCT__ "PCSetFromOptions_MG"
248 static PetscErrorCode PCSetFromOptions_MG(PC pc)
249 {
250   PetscErrorCode ierr;
251   PetscInt       indx,m,levels = 1;
252   PetscTruth     flg;
253   const char     *type[] = {"additive","multiplicative","full","cascade","kascade"};
254 
255   PetscFunctionBegin;
256 
257   ierr = PetscOptionsHead("Multigrid options");CHKERRQ(ierr);
258     if (!pc->data) {
259       ierr = PetscOptionsInt("-pc_mg_levels","Number of Levels","MGSetLevels",levels,&levels,&flg);CHKERRQ(ierr);
260       ierr = MGSetLevels(pc,levels,PETSC_NULL);CHKERRQ(ierr);
261     }
262     ierr = PetscOptionsInt("-pc_mg_cycles","1 for V cycle, 2 for W-cycle","MGSetCycles",1,&m,&flg);CHKERRQ(ierr);
263     if (flg) {
264       ierr = MGSetCycles(pc,m);CHKERRQ(ierr);
265     }
266     ierr = PetscOptionsName("-pc_mg_galerkin","Use Galerkin process to compute coarser operators","MGSetGalerkin",&flg);CHKERRQ(ierr);
267     if (flg) {
268       ierr = MGSetGalerkin(pc);CHKERRQ(ierr);
269     }
270     ierr = PetscOptionsInt("-pc_mg_smoothup","Number of post-smoothing steps","MGSetNumberSmoothUp",1,&m,&flg);CHKERRQ(ierr);
271     if (flg) {
272       ierr = MGSetNumberSmoothUp(pc,m);CHKERRQ(ierr);
273     }
274     ierr = PetscOptionsInt("-pc_mg_smoothdown","Number of pre-smoothing steps","MGSetNumberSmoothDown",1,&m,&flg);CHKERRQ(ierr);
275     if (flg) {
276       ierr = MGSetNumberSmoothDown(pc,m);CHKERRQ(ierr);
277     }
278     ierr = PetscOptionsEList("-pc_mg_type","Multigrid type","MGSetType",type,5,type[1],&indx,&flg);CHKERRQ(ierr);
279     if (flg) {
280       MGType mg = (MGType) 0;
281       switch (indx) {
282       case 0:
283         mg = MGADDITIVE;
284         break;
285       case 1:
286         mg = MGMULTIPLICATIVE;
287         break;
288       case 2:
289         mg = MGFULL;
290         break;
291       case 3:
292         mg = MGKASKADE;
293         break;
294       case 4:
295         mg = MGKASKADE;
296         break;
297       }
298       ierr = MGSetType(pc,mg);CHKERRQ(ierr);
299     }
300     ierr = PetscOptionsName("-pc_mg_log","Log times for each multigrid level","None",&flg);CHKERRQ(ierr);
301     if (flg) {
302       MG   *mg = (MG*)pc->data;
303       PetscInt i;
304       char eventname[128];
305       if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling");
306       levels = mg[0]->levels;
307       for (i=0; i<levels; i++) {
308         sprintf(eventname,"MSetup Level %d",(int)i);
309         ierr = PetscLogEventRegister(&mg[i]->eventsetup,eventname,pc->cookie);CHKERRQ(ierr);
310         sprintf(eventname,"MGSolve Level %d to 0",(int)i);
311         ierr = PetscLogEventRegister(&mg[i]->eventsolve,eventname,pc->cookie);CHKERRQ(ierr);
312       }
313     }
314   ierr = PetscOptionsTail();CHKERRQ(ierr);
315   PetscFunctionReturn(0);
316 }
317 
318 #undef __FUNCT__
319 #define __FUNCT__ "PCView_MG"
320 static PetscErrorCode PCView_MG(PC pc,PetscViewer viewer)
321 {
322   MG             *mg = (MG*)pc->data;
323   PetscErrorCode ierr;
324   PetscInt       levels = mg[0]->levels,i;
325   const char     *cstring;
326   PetscTruth     iascii;
327 
328   PetscFunctionBegin;
329   ierr = PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_ASCII,&iascii);CHKERRQ(ierr);
330   if (iascii) {
331     if (mg[0]->am == MGMULTIPLICATIVE) cstring = "multiplicative";
332     else if (mg[0]->am == MGADDITIVE)  cstring = "additive";
333     else if (mg[0]->am == MGFULL)      cstring = "full";
334     else if (mg[0]->am == MGKASKADE)   cstring = "Kaskade";
335     else cstring = "unknown";
336     ierr = PetscViewerASCIIPrintf(viewer,"  MG: type is %s, levels=%D cycles=%D, pre-smooths=%D, post-smooths=%D\n",
337                       cstring,levels,mg[0]->cycles,mg[0]->default_smoothd,mg[0]->default_smoothu);CHKERRQ(ierr);
338     if (mg[0]->galerkin) {
339       ierr = PetscViewerASCIIPrintf(viewer,"    Using Galerkin computed coarse grid matrices\n");CHKERRQ(ierr);
340     }
341     for (i=0; i<levels; i++) {
342       if (!i) {
343         ierr = PetscViewerASCIIPrintf(viewer,"Coarse gride solver -- level %D -------------------------------\n",i);CHKERRQ(ierr);
344       } else {
345         ierr = PetscViewerASCIIPrintf(viewer,"Down solver (pre-smoother) on level %D -------------------------------\n",i);CHKERRQ(ierr);
346       }
347       ierr = PetscViewerASCIIPushTab(viewer);CHKERRQ(ierr);
348       ierr = KSPView(mg[i]->smoothd,viewer);CHKERRQ(ierr);
349       ierr = PetscViewerASCIIPopTab(viewer);CHKERRQ(ierr);
350       if (i && mg[i]->smoothd == mg[i]->smoothu) {
351         ierr = PetscViewerASCIIPrintf(viewer,"Up solver (post-smoother) same as down solver (pre-smoother)\n");CHKERRQ(ierr);
352       } else if (i){
353         ierr = PetscViewerASCIIPrintf(viewer,"Up solver (post-smoother) on level %D -------------------------------\n",i);CHKERRQ(ierr);
354         ierr = PetscViewerASCIIPushTab(viewer);CHKERRQ(ierr);
355         ierr = KSPView(mg[i]->smoothu,viewer);CHKERRQ(ierr);
356         ierr = PetscViewerASCIIPopTab(viewer);CHKERRQ(ierr);
357       }
358     }
359   } else {
360     SETERRQ1(PETSC_ERR_SUP,"Viewer type %s not supported for PCMG",((PetscObject)viewer)->type_name);
361   }
362   PetscFunctionReturn(0);
363 }
364 
365 /*
366     Calls setup for the KSP on each level
367 */
368 #undef __FUNCT__
369 #define __FUNCT__ "PCSetUp_MG"
370 static PetscErrorCode PCSetUp_MG(PC pc)
371 {
372   MG             *mg = (MG*)pc->data;
373   PetscErrorCode ierr;
374   PetscInt       i,n = mg[0]->levels;
375   PC             cpc;
376   PetscTruth     preonly,lu,redundant,monitor = PETSC_FALSE,dump;
377   PetscViewer    ascii;
378   MPI_Comm       comm;
379   Mat            dA,dB;
380   MatStructure   uflag;
381 
382   PetscFunctionBegin;
383   if (!pc->setupcalled) {
384     ierr = PetscOptionsHasName(0,"-pc_mg_monitor",&monitor);CHKERRQ(ierr);
385 
386     for (i=0; i<n; i++) {
387       if (monitor) {
388         ierr = PetscObjectGetComm((PetscObject)mg[i]->smoothd,&comm);CHKERRQ(ierr);
389         ierr = PetscViewerASCIIOpen(comm,"stdout",&ascii);CHKERRQ(ierr);
390         ierr = PetscViewerASCIISetTab(ascii,n-i);CHKERRQ(ierr);
391         ierr = KSPSetMonitor(mg[i]->smoothd,KSPDefaultMonitor,ascii,(PetscErrorCode(*)(void*))PetscViewerDestroy);CHKERRQ(ierr);
392       }
393       ierr = KSPSetFromOptions(mg[i]->smoothd);CHKERRQ(ierr);
394     }
395     for (i=1; i<n; i++) {
396       if (mg[i]->smoothu && mg[i]->smoothu != mg[i]->smoothd) {
397         if (monitor) {
398           ierr = PetscObjectGetComm((PetscObject)mg[i]->smoothu,&comm);CHKERRQ(ierr);
399           ierr = PetscViewerASCIIOpen(comm,"stdout",&ascii);CHKERRQ(ierr);
400           ierr = PetscViewerASCIISetTab(ascii,n-i);CHKERRQ(ierr);
401           ierr = KSPSetMonitor(mg[i]->smoothu,KSPDefaultMonitor,ascii,(PetscErrorCode(*)(void*))PetscViewerDestroy);CHKERRQ(ierr);
402         }
403         ierr = KSPSetFromOptions(mg[i]->smoothu);CHKERRQ(ierr);
404       }
405     }
406     for (i=1; i<n; i++) {
407       if (mg[i]->restrct && !mg[i]->interpolate) {
408         ierr = MGSetInterpolate(pc,i,mg[i]->restrct);CHKERRQ(ierr);
409       }
410       if (!mg[i]->restrct && mg[i]->interpolate) {
411         ierr = MGSetRestriction(pc,i,mg[i]->interpolate);CHKERRQ(ierr);
412       }
413 #if defined(PETSC_USE_DEBUG)
414       if (!mg[i]->restrct || !mg[i]->interpolate) {
415         SETERRQ1(PETSC_ERR_ARG_WRONGSTATE,"Need to set restriction or interpolation on level %d",(int)i);
416       }
417 #endif
418     }
419   }
420 
421   /* If user did not provide fine grid operators, use those from PC */
422   /* BUG BUG BUG This will work ONLY the first time called: hence if the user changes
423      the PC matrices between solves PCMG will continue to use first set provided */
424   ierr = KSPGetOperators(mg[n-1]->smoothd,&dA,&dB,&uflag);CHKERRQ(ierr);
425   if (!dA  && !dB) {
426     ierr = PetscLogInfo((pc,"PCSetUp_MG: Using outer operators to define finest grid operator \n  because MGGetSmoother(pc,nlevels-1,&ksp);KSPSetOperators(ksp,...); was not called.\n"));
427     ierr = KSPSetOperators(mg[n-1]->smoothd,pc->mat,pc->pmat,uflag);CHKERRQ(ierr);
428   }
429 
430   if (mg[0]->galerkin) {
431     mg[0]->galerkinused = PETSC_TRUE;
432     /* currently only handle case where mat and pmat are the same on coarser levels */
433     Mat B;
434     ierr = KSPGetOperators(mg[n-1]->smoothd,&dA,&dB,&uflag);CHKERRQ(ierr);
435     if (!pc->setupcalled) {
436       for (i=n-2; i>-1; i--) {
437         ierr = MatPtAP(dB,mg[i+1]->interpolate,MAT_INITIAL_MATRIX,1.0,&B);CHKERRQ(ierr);
438         ierr = KSPSetOperators(mg[i]->smoothd,B,B,uflag);CHKERRQ(ierr);
439         dB   = B;
440       }
441     } else {
442       for (i=n-2; i>-1; i--) {
443         ierr = KSPGetOperators(mg[i]->smoothd,0,&B,0);CHKERRQ(ierr);
444         ierr = MatPtAP(dB,mg[i]->interpolate,MAT_REUSE_MATRIX,1.0,&B);CHKERRQ(ierr);
445         ierr = KSPSetOperators(mg[i]->smoothd,B,B,uflag);CHKERRQ(ierr);
446         dB   = B;
447       }
448     }
449   }
450 
451   for (i=1; i<n; i++) {
452     if (mg[i]->smoothu == mg[i]->smoothd) {
453       /* if doing only down then initial guess is zero */
454       ierr = KSPSetInitialGuessNonzero(mg[i]->smoothd,PETSC_TRUE);CHKERRQ(ierr);
455     }
456     if (mg[i]->eventsetup) {ierr = PetscLogEventBegin(mg[i]->eventsetup,0,0,0,0);CHKERRQ(ierr);}
457     ierr = KSPSetUp(mg[i]->smoothd);CHKERRQ(ierr);
458     if (mg[i]->eventsetup) {ierr = PetscLogEventEnd(mg[i]->eventsetup,0,0,0,0);CHKERRQ(ierr);}
459   }
460   for (i=1; i<n; i++) {
461     if (mg[i]->smoothu && mg[i]->smoothu != mg[i]->smoothd) {
462       PC           uppc,downpc;
463       Mat          downmat,downpmat,upmat,uppmat;
464       MatStructure matflag;
465 
466       /* check if operators have been set for up, if not use down operators to set them */
467       ierr = KSPGetPC(mg[i]->smoothu,&uppc);CHKERRQ(ierr);
468       ierr = PCGetOperators(uppc,&upmat,&uppmat,PETSC_NULL);CHKERRQ(ierr);
469       if (!upmat) {
470         ierr = KSPGetPC(mg[i]->smoothd,&downpc);CHKERRQ(ierr);
471         ierr = PCGetOperators(downpc,&downmat,&downpmat,&matflag);CHKERRQ(ierr);
472         ierr = KSPSetOperators(mg[i]->smoothu,downmat,downpmat,matflag);CHKERRQ(ierr);
473       }
474 
475       ierr = KSPSetInitialGuessNonzero(mg[i]->smoothu,PETSC_TRUE);CHKERRQ(ierr);
476       if (mg[i]->eventsetup) {ierr = PetscLogEventBegin(mg[i]->eventsetup,0,0,0,0);CHKERRQ(ierr);}
477       ierr = KSPSetUp(mg[i]->smoothu);CHKERRQ(ierr);
478       if (mg[i]->eventsetup) {ierr = PetscLogEventEnd(mg[i]->eventsetup,0,0,0,0);CHKERRQ(ierr);}
479     }
480   }
481 
482   /*
483       If coarse solver is not direct method then DO NOT USE preonly
484   */
485   ierr = PetscTypeCompare((PetscObject)mg[0]->smoothd,KSPPREONLY,&preonly);CHKERRQ(ierr);
486   if (preonly) {
487     ierr = KSPGetPC(mg[0]->smoothd,&cpc);CHKERRQ(ierr);
488     ierr = PetscTypeCompare((PetscObject)cpc,PCLU,&lu);CHKERRQ(ierr);
489     ierr = PetscTypeCompare((PetscObject)cpc,PCREDUNDANT,&redundant);CHKERRQ(ierr);
490     if (!lu && !redundant) {
491       ierr = KSPSetType(mg[0]->smoothd,KSPGMRES);CHKERRQ(ierr);
492     }
493   }
494 
495   if (!pc->setupcalled) {
496     if (monitor) {
497       ierr = PetscObjectGetComm((PetscObject)mg[0]->smoothd,&comm);CHKERRQ(ierr);
498       ierr = PetscViewerASCIIOpen(comm,"stdout",&ascii);CHKERRQ(ierr);
499       ierr = PetscViewerASCIISetTab(ascii,n);CHKERRQ(ierr);
500       ierr = KSPSetMonitor(mg[0]->smoothd,KSPDefaultMonitor,ascii,(PetscErrorCode(*)(void*))PetscViewerDestroy);CHKERRQ(ierr);
501     }
502     ierr = KSPSetFromOptions(mg[0]->smoothd);CHKERRQ(ierr);
503   }
504 
505   if (mg[0]->eventsetup) {ierr = PetscLogEventBegin(mg[0]->eventsetup,0,0,0,0);CHKERRQ(ierr);}
506   ierr = KSPSetUp(mg[0]->smoothd);CHKERRQ(ierr);
507   if (mg[0]->eventsetup) {ierr = PetscLogEventEnd(mg[0]->eventsetup,0,0,0,0);CHKERRQ(ierr);}
508 
509   /*
510      Dump the interpolation/restriction matrices to matlab plus the
511    Jacobian/stiffness on each level. This allows Matlab users to
512    easily check if the Galerkin condition A_c = R A_f R^T is satisfied */
513   ierr = PetscOptionsHasName(pc->prefix,"-pc_mg_dump_matlab",&dump);CHKERRQ(ierr);
514   if (dump) {
515     for (i=1; i<n; i++) {
516       ierr = MatView(mg[i]->restrct,PETSC_VIEWER_SOCKET_(pc->comm));CHKERRQ(ierr);
517     }
518     for (i=0; i<n; i++) {
519       ierr = KSPGetPC(mg[i]->smoothd,&pc);CHKERRQ(ierr);
520       ierr = MatView(pc->mat,PETSC_VIEWER_SOCKET_(pc->comm));CHKERRQ(ierr);
521     }
522   }
523   ierr = PetscOptionsHasName(pc->prefix,"-pc_mg_dump_binary",&dump);CHKERRQ(ierr);
524   if (dump) {
525     for (i=1; i<n; i++) {
526       ierr = MatView(mg[i]->restrct,PETSC_VIEWER_BINARY_(pc->comm));CHKERRQ(ierr);
527     }
528     for (i=0; i<n; i++) {
529       ierr = KSPGetPC(mg[i]->smoothd,&pc);CHKERRQ(ierr);
530       ierr = MatView(pc->mat,PETSC_VIEWER_BINARY_(pc->comm));CHKERRQ(ierr);
531     }
532   }
533   PetscFunctionReturn(0);
534 }
535 
536 /* -------------------------------------------------------------------------------------*/
537 
538 #undef __FUNCT__
539 #define __FUNCT__ "MGSetLevels"
540 /*@C
541    MGSetLevels - Sets the number of levels to use with MG.
542    Must be called before any other MG routine.
543 
544    Collective on PC
545 
546    Input Parameters:
547 +  pc - the preconditioner context
548 .  levels - the number of levels
549 -  comms - optional communicators for each level; this is to allow solving the coarser problems
550            on smaller sets of processors. Use PETSC_NULL_OBJECT for default in Fortran
551 
552    Level: intermediate
553 
554    Notes:
555      If the number of levels is one then the multigrid uses the -mg_levels prefix
556   for setting the level options rather than the -mg_coarse prefix.
557 
558 .keywords: MG, set, levels, multigrid
559 
560 .seealso: MGSetType(), MGGetLevels()
561 @*/
562 PetscErrorCode PETSCKSP_DLLEXPORT MGSetLevels(PC pc,PetscInt levels,MPI_Comm *comms)
563 {
564   PetscErrorCode ierr;
565   MG             *mg;
566 
567   PetscFunctionBegin;
568   PetscValidHeaderSpecific(pc,PC_COOKIE,1);
569 
570   if (pc->data) {
571     SETERRQ(PETSC_ERR_ORDER,"Number levels already set for MG\n\
572     make sure that you call MGSetLevels() before KSPSetFromOptions()");
573   }
574   ierr                     = MGCreate_Private(pc->comm,levels,pc,comms,&mg);CHKERRQ(ierr);
575   mg[0]->am                = MGMULTIPLICATIVE;
576   pc->data                 = (void*)mg;
577   pc->ops->applyrichardson = PCApplyRichardson_MG;
578   PetscFunctionReturn(0);
579 }
580 
581 #undef __FUNCT__
582 #define __FUNCT__ "MGGetLevels"
583 /*@
584    MGGetLevels - Gets the number of levels to use with MG.
585 
586    Not Collective
587 
588    Input Parameter:
589 .  pc - the preconditioner context
590 
591    Output parameter:
592 .  levels - the number of levels
593 
594    Level: advanced
595 
596 .keywords: MG, get, levels, multigrid
597 
598 .seealso: MGSetLevels()
599 @*/
600 PetscErrorCode PETSCKSP_DLLEXPORT MGGetLevels(PC pc,PetscInt *levels)
601 {
602   MG  *mg;
603 
604   PetscFunctionBegin;
605   PetscValidHeaderSpecific(pc,PC_COOKIE,1);
606   PetscValidIntPointer(levels,2);
607 
608   mg      = (MG*)pc->data;
609   *levels = mg[0]->levels;
610   PetscFunctionReturn(0);
611 }
612 
613 #undef __FUNCT__
614 #define __FUNCT__ "MGSetType"
615 /*@
616    MGSetType - Determines the form of multigrid to use:
617    multiplicative, additive, full, or the Kaskade algorithm.
618 
619    Collective on PC
620 
621    Input Parameters:
622 +  pc - the preconditioner context
623 -  form - multigrid form, one of MGMULTIPLICATIVE, MGADDITIVE,
624    MGFULL, MGKASKADE
625 
626    Options Database Key:
627 .  -pc_mg_type <form> - Sets <form>, one of multiplicative,
628    additive, full, kaskade
629 
630    Level: advanced
631 
632 .keywords: MG, set, method, multiplicative, additive, full, Kaskade, multigrid
633 
634 .seealso: MGSetLevels()
635 @*/
636 PetscErrorCode PETSCKSP_DLLEXPORT MGSetType(PC pc,MGType form)
637 {
638   MG *mg;
639 
640   PetscFunctionBegin;
641   PetscValidHeaderSpecific(pc,PC_COOKIE,1);
642   mg = (MG*)pc->data;
643 
644   if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling");
645   mg[0]->am = form;
646   if (form == MGMULTIPLICATIVE) pc->ops->applyrichardson = PCApplyRichardson_MG;
647   else pc->ops->applyrichardson = 0;
648   PetscFunctionReturn(0);
649 }
650 
651 #undef __FUNCT__
652 #define __FUNCT__ "MGSetCycles"
653 /*@
654    MGSetCycles - Sets the type cycles to use.  Use MGSetCyclesOnLevel() for more
655    complicated cycling.
656 
657    Collective on PC
658 
659    Input Parameters:
660 +  pc - the multigrid context
661 -  n - the number of cycles
662 
663    Options Database Key:
664 $  -pc_mg_cycles n - 1 denotes a V-cycle; 2 denotes a W-cycle.
665 
666    Level: advanced
667 
668 .keywords: MG, set, cycles, V-cycle, W-cycle, multigrid
669 
670 .seealso: MGSetCyclesOnLevel()
671 @*/
672 PetscErrorCode PETSCKSP_DLLEXPORT MGSetCycles(PC pc,PetscInt n)
673 {
674   MG       *mg;
675   PetscInt i,levels;
676 
677   PetscFunctionBegin;
678   PetscValidHeaderSpecific(pc,PC_COOKIE,1);
679   mg     = (MG*)pc->data;
680   if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling");
681   levels = mg[0]->levels;
682 
683   for (i=0; i<levels; i++) {
684     mg[i]->cycles  = n;
685   }
686   PetscFunctionReturn(0);
687 }
688 
689 #undef __FUNCT__
690 #define __FUNCT__ "MGSetGalerkin"
691 /*@
692    MGSetGalerkin - Causes the coarser grid matrices to be computed from the
693       finest grid via the Galerkin process: A_i-1 = r_i * A_i * r_i^t
694 
695    Collective on PC
696 
697    Input Parameters:
698 +  pc - the multigrid context
699 -  n - the number of cycles
700 
701    Options Database Key:
702 $  -pc_mg_galerkin
703 
704    Level: intermediate
705 
706 .keywords: MG, set, Galerkin
707 
708 @*/
709 PetscErrorCode PETSCKSP_DLLEXPORT MGSetGalerkin(PC pc)
710 {
711   MG       *mg;
712   PetscInt i,levels;
713 
714   PetscFunctionBegin;
715   PetscValidHeaderSpecific(pc,PC_COOKIE,1);
716   mg     = (MG*)pc->data;
717   if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling");
718   levels = mg[0]->levels;
719 
720   for (i=0; i<levels; i++) {
721     mg[i]->galerkin = PETSC_TRUE;
722   }
723   PetscFunctionReturn(0);
724 }
725 
726 #undef __FUNCT__
727 #define __FUNCT__ "MGCheck"
728 /*@
729    MGCheck - Checks that all components of the MG structure have
730    been set.
731 
732    Collective on PC
733 
734    Input Parameters:
735 .  mg - the MG structure
736 
737    Level: advanced
738 
739 .keywords: MG, check, set, multigrid
740 @*/
741 PetscErrorCode PETSCKSP_DLLEXPORT MGCheck(PC pc)
742 {
743   MG       *mg;
744   PetscInt i,n,count = 0;
745 
746   PetscFunctionBegin;
747   PetscValidHeaderSpecific(pc,PC_COOKIE,1);
748   mg = (MG*)pc->data;
749 
750   if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling");
751 
752   n = mg[0]->levels;
753 
754   for (i=1; i<n; i++) {
755     if (!mg[i]->restrct) {
756       (*PetscErrorPrintf)("No restrict set level %D \n",n-i); count++;
757     }
758     if (!mg[i]->interpolate) {
759       (*PetscErrorPrintf)("No interpolate set level %D \n",n-i); count++;
760     }
761     if (!mg[i]->r) {
762       (*PetscErrorPrintf)("No r set level %D \n",n-i); count++;
763     }
764     if (!mg[i-1]->x) {
765       (*PetscErrorPrintf)("No x set level %D \n",n-i); count++;
766     }
767     if (!mg[i-1]->b) {
768       (*PetscErrorPrintf)("No b set level %D \n",n-i); count++;
769     }
770   }
771   PetscFunctionReturn(count);
772 }
773 
774 
775 #undef __FUNCT__
776 #define __FUNCT__ "MGSetNumberSmoothDown"
777 /*@
778    MGSetNumberSmoothDown - Sets the number of pre-smoothing steps to
779    use on all levels. Use MGGetSmootherDown() to set different
780    pre-smoothing steps on different levels.
781 
782    Collective on PC
783 
784    Input Parameters:
785 +  mg - the multigrid context
786 -  n - the number of smoothing steps
787 
788    Options Database Key:
789 .  -pc_mg_smoothdown <n> - Sets number of pre-smoothing steps
790 
791    Level: advanced
792 
793 .keywords: MG, smooth, down, pre-smoothing, steps, multigrid
794 
795 .seealso: MGSetNumberSmoothUp()
796 @*/
797 PetscErrorCode PETSCKSP_DLLEXPORT MGSetNumberSmoothDown(PC pc,PetscInt n)
798 {
799   MG             *mg;
800   PetscErrorCode ierr;
801   PetscInt       i,levels;
802 
803   PetscFunctionBegin;
804   PetscValidHeaderSpecific(pc,PC_COOKIE,1);
805   mg     = (MG*)pc->data;
806   if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling");
807   levels = mg[0]->levels;
808 
809   for (i=0; i<levels; i++) {
810     /* make sure smoother up and down are different */
811     ierr = MGGetSmootherUp(pc,i,PETSC_NULL);CHKERRQ(ierr);
812     ierr = KSPSetTolerances(mg[i]->smoothd,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT,n);CHKERRQ(ierr);
813     mg[i]->default_smoothd = n;
814   }
815   PetscFunctionReturn(0);
816 }
817 
818 #undef __FUNCT__
819 #define __FUNCT__ "MGSetNumberSmoothUp"
820 /*@
821    MGSetNumberSmoothUp - Sets the number of post-smoothing steps to use
822    on all levels. Use MGGetSmootherUp() to set different numbers of
823    post-smoothing steps on different levels.
824 
825    Collective on PC
826 
827    Input Parameters:
828 +  mg - the multigrid context
829 -  n - the number of smoothing steps
830 
831    Options Database Key:
832 .  -pc_mg_smoothup <n> - Sets number of post-smoothing steps
833 
834    Level: advanced
835 
836    Note: this does not set a value on the coarsest grid, since we assume that
837     there is no seperate smooth up on the coarsest grid.
838 
839 .keywords: MG, smooth, up, post-smoothing, steps, multigrid
840 
841 .seealso: MGSetNumberSmoothDown()
842 @*/
843 PetscErrorCode PETSCKSP_DLLEXPORT MGSetNumberSmoothUp(PC pc,PetscInt n)
844 {
845   MG             *mg;
846   PetscErrorCode ierr;
847   PetscInt       i,levels;
848 
849   PetscFunctionBegin;
850   PetscValidHeaderSpecific(pc,PC_COOKIE,1);
851   mg     = (MG*)pc->data;
852   if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling");
853   levels = mg[0]->levels;
854 
855   for (i=1; i<levels; i++) {
856     /* make sure smoother up and down are different */
857     ierr = MGGetSmootherUp(pc,i,PETSC_NULL);CHKERRQ(ierr);
858     ierr = KSPSetTolerances(mg[i]->smoothu,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT,n);CHKERRQ(ierr);
859     mg[i]->default_smoothu = n;
860   }
861   PetscFunctionReturn(0);
862 }
863 
864 /* ----------------------------------------------------------------------------------------*/
865 
866 /*MC
867    PCMG - Use geometric multigrid preconditioning. This preconditioner requires you provide additional
868     information about the coarser grid matrices and restriction/interpolation operators.
869 
870    Options Database Keys:
871 +  -pc_mg_levels <nlevels> - number of levels including finest
872 .  -pc_mg_cycles 1 or 2 - for V or W-cycle
873 .  -pc_mg_smoothup <n> - number of smoothing steps after interpolation
874 .  -pc_mg_smoothdown <n> - number of smoothing steps before applying restriction operator
875 .  -pc_mg_type <additive,multiplicative,full,cascade> - multiplicative is the default
876 .  -pc_mg_log - log information about time spent on each level of the solver
877 .  -pc_mg_monitor - print information on the multigrid convergence
878 -  -pc_mg_dump_matlab - dumps the matrices for each level and the restriction/interpolation matrices
879                         to the Socket viewer for reading from Matlab.
880 
881    Notes:
882 
883    Level: intermediate
884 
885    Concepts: multigrid
886 
887 .seealso:  PCCreate(), PCSetType(), PCType (for list of available types), PC, PCMGType,
888            MGSetLevels(), MGGetLevels(), MGSetType(), MPSetCycles(), MGSetNumberSmoothDown(),
889            MGSetNumberSmoothUp(), MGGetCoarseSolve(), MGSetResidual(), MGSetInterpolation(),
890            MGSetRestriction(), MGGetSmoother(), MGGetSmootherUp(), MGGetSmootherDown(),
891            MGSetCyclesOnLevel(), MGSetRhs(), MGSetX(), MGSetR()
892 M*/
893 
894 EXTERN_C_BEGIN
895 #undef __FUNCT__
896 #define __FUNCT__ "PCCreate_MG"
897 PetscErrorCode PETSCKSP_DLLEXPORT PCCreate_MG(PC pc)
898 {
899   PetscFunctionBegin;
900   pc->ops->apply          = PCApply_MG;
901   pc->ops->setup          = PCSetUp_MG;
902   pc->ops->destroy        = PCDestroy_MG;
903   pc->ops->setfromoptions = PCSetFromOptions_MG;
904   pc->ops->view           = PCView_MG;
905 
906   pc->data                = (void*)0;
907   PetscFunctionReturn(0);
908 }
909 EXTERN_C_END
910