xref: /petsc/src/ksp/pc/impls/mg/mg.c (revision cf50294291ee912251d43f98e0337904404b8dbf)
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 #undef __FUNCT__
10 #define __FUNCT__ "PCMGMCycle_Private"
11 PetscErrorCode PCMGMCycle_Private(PC_MG **mglevels,PetscTruth *converged)
12 {
13   PC_MG          *mg = *mglevels,*mgc;
14   PetscErrorCode ierr;
15   PetscInt       cycles = mg->cycles;
16 
17   PetscFunctionBegin;
18   if (converged) *converged = PETSC_FALSE;
19 
20   if (mg->eventsolve) {ierr = PetscLogEventBegin(mg->eventsolve,0,0,0,0);CHKERRQ(ierr);}
21   ierr = KSPSolve(mg->smoothd,mg->b,mg->x);CHKERRQ(ierr);
22   if (mg->eventsolve) {ierr = PetscLogEventEnd(mg->eventsolve,0,0,0,0);CHKERRQ(ierr);}
23   if (mg->level) {  /* not the coarsest grid */
24     ierr = (*mg->residual)(mg->A,mg->b,mg->x,mg->r);CHKERRQ(ierr);
25 
26     /* if on finest level and have convergence criteria set */
27     if (mg->level == mg->levels-1 && mg->ttol) {
28       PetscReal rnorm;
29       ierr = VecNorm(mg->r,NORM_2,&rnorm);CHKERRQ(ierr);
30       if (rnorm <= mg->ttol) {
31         *converged = PETSC_TRUE;
32         if (rnorm < mg->abstol) {
33           ierr = PetscInfo2(0,"Linear solver has converged. Residual norm %G is less than absolute tolerance %G\n",rnorm,mg->abstol);CHKERRQ(ierr);
34         } else {
35           ierr = PetscInfo2(0,"Linear solver has converged. Residual norm %G is less than relative tolerance times initial residual norm %G\n",rnorm,mg->ttol);CHKERRQ(ierr);
36         }
37         PetscFunctionReturn(0);
38       }
39     }
40 
41     mgc = *(mglevels - 1);
42     ierr = MatRestrict(mg->restrct,mg->r,mgc->b);CHKERRQ(ierr);
43     ierr = VecSet(mgc->x,0.0);CHKERRQ(ierr);
44     while (cycles--) {
45       ierr = PCMGMCycle_Private(mglevels-1,converged);CHKERRQ(ierr);
46     }
47     ierr = MatInterpolateAdd(mg->interpolate,mgc->x,mg->x,mg->x);CHKERRQ(ierr);
48     if (mg->eventsolve) {ierr = PetscLogEventBegin(mg->eventsolve,0,0,0,0);CHKERRQ(ierr);}
49     ierr = KSPSolve(mg->smoothu,mg->b,mg->x);CHKERRQ(ierr);
50     if (mg->eventsolve) {ierr = PetscLogEventEnd(mg->eventsolve,0,0,0,0);CHKERRQ(ierr);}
51   }
52   PetscFunctionReturn(0);
53 }
54 
55 /*
56        PCMGCreate_Private - Creates a PC_MG structure for use with the
57                multigrid code. Level 0 is the coarsest. (But the
58                finest level is stored first in the array).
59 
60 */
61 #undef __FUNCT__
62 #define __FUNCT__ "PCMGCreate_Private"
63 static PetscErrorCode PCMGCreate_Private(MPI_Comm comm,PetscInt levels,PC pc,MPI_Comm *comms,PC_MG ***result)
64 {
65   PC_MG          **mg;
66   PetscErrorCode ierr;
67   PetscInt       i;
68   PetscMPIInt    size;
69   const char     *prefix;
70   PC             ipc;
71 
72   PetscFunctionBegin;
73   ierr = PetscMalloc(levels*sizeof(PC_MG*),&mg);CHKERRQ(ierr);
74   ierr = PetscLogObjectMemory(pc,levels*(sizeof(PC_MG*)+sizeof(PC_MG)));CHKERRQ(ierr);
75 
76   ierr = PCGetOptionsPrefix(pc,&prefix);CHKERRQ(ierr);
77 
78   for (i=0; i<levels; i++) {
79     ierr = PetscNew(PC_MG,&mg[i]);CHKERRQ(ierr);
80     mg[i]->level           = i;
81     mg[i]->levels          = levels;
82     mg[i]->cycles          = 1;
83     mg[i]->galerkin        = PETSC_FALSE;
84     mg[i]->galerkinused    = PETSC_FALSE;
85     mg[i]->default_smoothu = 1;
86     mg[i]->default_smoothd = 1;
87 
88     if (comms) comm = comms[i];
89     ierr = KSPCreate(comm,&mg[i]->smoothd);CHKERRQ(ierr);
90     ierr = KSPSetTolerances(mg[i]->smoothd,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT, mg[i]->default_smoothd);CHKERRQ(ierr);
91     ierr = KSPSetOptionsPrefix(mg[i]->smoothd,prefix);CHKERRQ(ierr);
92 
93     /* do special stuff for coarse grid */
94     if (!i && levels > 1) {
95       ierr = KSPAppendOptionsPrefix(mg[0]->smoothd,"mg_coarse_");CHKERRQ(ierr);
96 
97       /* coarse solve is (redundant) LU by default */
98       ierr = KSPSetType(mg[0]->smoothd,KSPPREONLY);CHKERRQ(ierr);
99       ierr = KSPGetPC(mg[0]->smoothd,&ipc);CHKERRQ(ierr);
100       ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr);
101       if (size > 1) {
102         ierr = PCSetType(ipc,PCREDUNDANT);CHKERRQ(ierr);
103         ierr = PCRedundantGetPC(ipc,&ipc);CHKERRQ(ierr);
104       }
105       ierr = PCSetType(ipc,PCLU);CHKERRQ(ierr);
106 
107     } else {
108       char tprefix[128];
109       sprintf(tprefix,"mg_levels_%d_",(int)i);
110       ierr = KSPAppendOptionsPrefix(mg[i]->smoothd,tprefix);CHKERRQ(ierr);
111     }
112     ierr = PetscLogObjectParent(pc,mg[i]->smoothd);CHKERRQ(ierr);
113     mg[i]->smoothu         = mg[i]->smoothd;
114     mg[i]->rtol = 0.0;
115     mg[i]->abstol = 0.0;
116     mg[i]->dtol = 0.0;
117     mg[i]->ttol = 0.0;
118     mg[i]->eventsetup = 0;
119     mg[i]->eventsolve = 0;
120   }
121   *result = mg;
122   PetscFunctionReturn(0);
123 }
124 
125 #undef __FUNCT__
126 #define __FUNCT__ "PCDestroy_MG"
127 static PetscErrorCode PCDestroy_MG(PC pc)
128 {
129   PC_MG          **mg = (PC_MG**)pc->data;
130   PetscErrorCode ierr;
131   PetscInt       i,n = mg[0]->levels;
132 
133   PetscFunctionBegin;
134   if (mg[0]->galerkinused) {
135     Mat B;
136     for (i=0; i<n-1; i++) {
137       ierr = KSPGetOperators(mg[i]->smoothd,0,&B,0);CHKERRQ(ierr);
138       ierr = MatDestroy(B);CHKERRQ(ierr);
139     }
140   }
141 
142   for (i=0; i<n-1; i++) {
143     if (mg[i+1]->r) {ierr = VecDestroy(mg[i+1]->r);CHKERRQ(ierr);}
144     if (mg[i]->b) {ierr = VecDestroy(mg[i]->b);CHKERRQ(ierr);}
145     if (mg[i]->x) {ierr = VecDestroy(mg[i]->x);CHKERRQ(ierr);}
146     if (mg[i+1]->restrct) {ierr = MatDestroy(mg[i+1]->restrct);CHKERRQ(ierr);}
147     if (mg[i+1]->interpolate) {ierr = MatDestroy(mg[i+1]->interpolate);CHKERRQ(ierr);}
148   }
149 
150   for (i=0; i<n; i++) {
151     if (mg[i]->smoothd != mg[i]->smoothu) {
152       ierr = KSPDestroy(mg[i]->smoothd);CHKERRQ(ierr);
153     }
154     ierr = KSPDestroy(mg[i]->smoothu);CHKERRQ(ierr);
155     ierr = PetscFree(mg[i]);CHKERRQ(ierr);
156   }
157   ierr = PetscFree(mg);CHKERRQ(ierr);
158   PetscFunctionReturn(0);
159 }
160 
161 
162 
163 EXTERN PetscErrorCode PCMGACycle_Private(PC_MG**);
164 EXTERN PetscErrorCode PCMGFCycle_Private(PC_MG**);
165 EXTERN PetscErrorCode PCMGKCycle_Private(PC_MG**);
166 
167 /*
168    PCApply_MG - Runs either an additive, multiplicative, Kaskadic
169              or full cycle of multigrid.
170 
171   Note:
172   A simple wrapper which calls PCMGMCycle(),PCMGACycle(), or PCMGFCycle().
173 */
174 #undef __FUNCT__
175 #define __FUNCT__ "PCApply_MG"
176 static PetscErrorCode PCApply_MG(PC pc,Vec b,Vec x)
177 {
178   PC_MG          **mg = (PC_MG**)pc->data;
179   PetscErrorCode ierr;
180   PetscInt       levels = mg[0]->levels;
181 
182   PetscFunctionBegin;
183   mg[levels-1]->b = b;
184   mg[levels-1]->x = x;
185   if (!mg[levels-1]->r && mg[0]->am != PC_MG_ADDITIVE && levels > 1) {
186     Vec tvec;
187     ierr = VecDuplicate(mg[levels-1]->b,&tvec);CHKERRQ(ierr);
188     ierr = PCMGSetR(pc,levels-1,tvec);CHKERRQ(ierr);
189     ierr = VecDestroy(tvec);CHKERRQ(ierr);
190   }
191   if (mg[0]->am == PC_MG_MULTIPLICATIVE) {
192     ierr = VecSet(x,0.0);CHKERRQ(ierr);
193     ierr = PCMGMCycle_Private(mg+levels-1,PETSC_NULL);CHKERRQ(ierr);
194   }
195   else if (mg[0]->am == PC_MG_ADDITIVE) {
196     ierr = PCMGACycle_Private(mg);CHKERRQ(ierr);
197   }
198   else if (mg[0]->am == PC_MG_KASKADE) {
199     ierr = PCMGKCycle_Private(mg);CHKERRQ(ierr);
200   }
201   else {
202     ierr = PCMGFCycle_Private(mg);CHKERRQ(ierr);
203   }
204   PetscFunctionReturn(0);
205 }
206 
207 #undef __FUNCT__
208 #define __FUNCT__ "PCApplyRichardson_MG"
209 static PetscErrorCode PCApplyRichardson_MG(PC pc,Vec b,Vec x,Vec w,PetscReal rtol,PetscReal abstol, PetscReal dtol,PetscInt its)
210 {
211   PC_MG          **mg = (PC_MG**)pc->data;
212   PetscErrorCode ierr;
213   PetscInt       levels = mg[0]->levels;
214   PetscTruth     converged = PETSC_FALSE;
215 
216   PetscFunctionBegin;
217   mg[levels-1]->b    = b;
218   mg[levels-1]->x    = x;
219 
220   mg[levels-1]->rtol = rtol;
221   mg[levels-1]->abstol = abstol;
222   mg[levels-1]->dtol = dtol;
223   if (rtol) {
224     /* compute initial residual norm for relative convergence test */
225     PetscReal rnorm;
226     ierr               = (*mg[levels-1]->residual)(mg[levels-1]->A,b,x,w);CHKERRQ(ierr);
227     ierr               = VecNorm(w,NORM_2,&rnorm);CHKERRQ(ierr);
228     mg[levels-1]->ttol = PetscMax(rtol*rnorm,abstol);
229   } else if (abstol) {
230     mg[levels-1]->ttol = abstol;
231   } else {
232     mg[levels-1]->ttol = 0.0;
233   }
234 
235   while (its-- && !converged) {
236     ierr = PCMGMCycle_Private(mg+levels-1,&converged);CHKERRQ(ierr);
237   }
238   PetscFunctionReturn(0);
239 }
240 
241 #undef __FUNCT__
242 #define __FUNCT__ "PCSetFromOptions_MG"
243 PetscErrorCode PCSetFromOptions_MG(PC pc)
244 {
245   PetscErrorCode ierr;
246   PetscInt       m,levels = 1;
247   PetscTruth     flg;
248   PC_MG          **mg = (PC_MG**)pc->data;
249   PCMGType       mgtype;
250 
251   PetscFunctionBegin;
252 
253   ierr = PetscOptionsHead("Multigrid options");CHKERRQ(ierr);
254     if (!pc->data) {
255       ierr = PetscOptionsInt("-pc_mg_levels","Number of Levels","PCMGSetLevels",levels,&levels,&flg);CHKERRQ(ierr);
256       ierr = PCMGSetLevels(pc,levels,PETSC_NULL);CHKERRQ(ierr);
257       mg = (PC_MG**)pc->data;
258     }
259     mgtype = mg[0]->am;
260     ierr = PetscOptionsInt("-pc_mg_cycles","1 for V cycle, 2 for W-cycle","PCMGSetCycles",1,&m,&flg);CHKERRQ(ierr);
261     if (flg) {
262       ierr = PCMGSetCycles(pc,m);CHKERRQ(ierr);
263     }
264     ierr = PetscOptionsName("-pc_mg_galerkin","Use Galerkin process to compute coarser operators","PCMGSetGalerkin",&flg);CHKERRQ(ierr);
265     if (flg) {
266       ierr = PCMGSetGalerkin(pc);CHKERRQ(ierr);
267     }
268     ierr = PetscOptionsInt("-pc_mg_smoothup","Number of post-smoothing steps","PCMGSetNumberSmoothUp",1,&m,&flg);CHKERRQ(ierr);
269     if (flg) {
270       ierr = PCMGSetNumberSmoothUp(pc,m);CHKERRQ(ierr);
271     }
272     ierr = PetscOptionsInt("-pc_mg_smoothdown","Number of pre-smoothing steps","PCMGSetNumberSmoothDown",1,&m,&flg);CHKERRQ(ierr);
273     if (flg) {
274       ierr = PCMGSetNumberSmoothDown(pc,m);CHKERRQ(ierr);
275     }
276     ierr = PetscOptionsEnum("-pc_mg_type","Multigrid type","PCMGSetType",PCMGTypes,(PetscEnum)mgtype,(PetscEnum*)&mgtype,&flg);CHKERRQ(ierr);
277     if (flg) {ierr = PCMGSetType(pc,mgtype);CHKERRQ(ierr);}
278     ierr = PetscOptionsName("-pc_mg_log","Log times for each multigrid level","None",&flg);CHKERRQ(ierr);
279     if (flg) {
280       PetscInt i;
281       char     eventname[128];
282       if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling");
283       levels = mg[0]->levels;
284       for (i=0; i<levels; i++) {
285         sprintf(eventname,"MSetup Level %d",(int)i);
286         ierr = PetscLogEventRegister(&mg[i]->eventsetup,eventname,pc->cookie);CHKERRQ(ierr);
287         sprintf(eventname,"MGSolve Level %d to 0",(int)i);
288         ierr = PetscLogEventRegister(&mg[i]->eventsolve,eventname,pc->cookie);CHKERRQ(ierr);
289       }
290     }
291   ierr = PetscOptionsTail();CHKERRQ(ierr);
292   PetscFunctionReturn(0);
293 }
294 
295 const char *PCMGTypes[] = {"MULTIPLICATIVE","ADDITIVE","FULL","KASKADE","PCMGType","PC_MG",0};
296 
297 #undef __FUNCT__
298 #define __FUNCT__ "PCView_MG"
299 static PetscErrorCode PCView_MG(PC pc,PetscViewer viewer)
300 {
301   PC_MG          **mg = (PC_MG**)pc->data;
302   PetscErrorCode ierr;
303   PetscInt       levels = mg[0]->levels,i;
304   PetscTruth     iascii;
305 
306   PetscFunctionBegin;
307   ierr = PetscTypeCompare((PetscObject)viewer,PETSC_VIEWER_ASCII,&iascii);CHKERRQ(ierr);
308   if (iascii) {
309     ierr = PetscViewerASCIIPrintf(viewer,"  MG: type is %s, levels=%D cycles=%D, pre-smooths=%D, post-smooths=%D\n",
310                       PCMGTypes[mg[0]->am],levels,mg[0]->cycles,mg[0]->default_smoothd,mg[0]->default_smoothu);CHKERRQ(ierr);
311     if (mg[0]->galerkin) {
312       ierr = PetscViewerASCIIPrintf(viewer,"    Using Galerkin computed coarse grid matrices\n");CHKERRQ(ierr);
313     }
314     for (i=0; i<levels; i++) {
315       if (!i) {
316         ierr = PetscViewerASCIIPrintf(viewer,"Coarse gride solver -- level %D -------------------------------\n",i);CHKERRQ(ierr);
317       } else {
318         ierr = PetscViewerASCIIPrintf(viewer,"Down solver (pre-smoother) on level %D -------------------------------\n",i);CHKERRQ(ierr);
319       }
320       ierr = PetscViewerASCIIPushTab(viewer);CHKERRQ(ierr);
321       ierr = KSPView(mg[i]->smoothd,viewer);CHKERRQ(ierr);
322       ierr = PetscViewerASCIIPopTab(viewer);CHKERRQ(ierr);
323       if (i && mg[i]->smoothd == mg[i]->smoothu) {
324         ierr = PetscViewerASCIIPrintf(viewer,"Up solver (post-smoother) same as down solver (pre-smoother)\n");CHKERRQ(ierr);
325       } else if (i){
326         ierr = PetscViewerASCIIPrintf(viewer,"Up solver (post-smoother) on level %D -------------------------------\n",i);CHKERRQ(ierr);
327         ierr = PetscViewerASCIIPushTab(viewer);CHKERRQ(ierr);
328         ierr = KSPView(mg[i]->smoothu,viewer);CHKERRQ(ierr);
329         ierr = PetscViewerASCIIPopTab(viewer);CHKERRQ(ierr);
330       }
331     }
332   } else {
333     SETERRQ1(PETSC_ERR_SUP,"Viewer type %s not supported for PCMG",((PetscObject)viewer)->type_name);
334   }
335   PetscFunctionReturn(0);
336 }
337 
338 /*
339     Calls setup for the KSP on each level
340 */
341 #undef __FUNCT__
342 #define __FUNCT__ "PCSetUp_MG"
343 static PetscErrorCode PCSetUp_MG(PC pc)
344 {
345   PC_MG          **mg = (PC_MG**)pc->data;
346   PetscErrorCode ierr;
347   PetscInt       i,n = mg[0]->levels;
348   PC             cpc;
349   PetscTruth     preonly,lu,redundant,cholesky,monitor = PETSC_FALSE,dump;
350   PetscViewer    ascii;
351   MPI_Comm       comm;
352   Mat            dA,dB;
353   MatStructure   uflag;
354   Vec            tvec;
355 
356   PetscFunctionBegin;
357   if (!pc->setupcalled) {
358     ierr = PetscOptionsHasName(0,"-pc_mg_monitor",&monitor);CHKERRQ(ierr);
359 
360     for (i=0; i<n; i++) {
361       if (monitor) {
362         ierr = PetscObjectGetComm((PetscObject)mg[i]->smoothd,&comm);CHKERRQ(ierr);
363         ierr = PetscViewerASCIIOpen(comm,"stdout",&ascii);CHKERRQ(ierr);
364         ierr = PetscViewerASCIISetTab(ascii,n-i);CHKERRQ(ierr);
365         ierr = KSPSetMonitor(mg[i]->smoothd,KSPDefaultMonitor,ascii,(PetscErrorCode(*)(void*))PetscViewerDestroy);CHKERRQ(ierr);
366       }
367       ierr = KSPSetFromOptions(mg[i]->smoothd);CHKERRQ(ierr);
368     }
369     for (i=1; i<n; i++) {
370       if (mg[i]->smoothu && (mg[i]->smoothu != mg[i]->smoothd)) {
371         if (monitor) {
372           ierr = PetscObjectGetComm((PetscObject)mg[i]->smoothu,&comm);CHKERRQ(ierr);
373           ierr = PetscViewerASCIIOpen(comm,"stdout",&ascii);CHKERRQ(ierr);
374           ierr = PetscViewerASCIISetTab(ascii,n-i);CHKERRQ(ierr);
375           ierr = KSPSetMonitor(mg[i]->smoothu,KSPDefaultMonitor,ascii,(PetscErrorCode(*)(void*))PetscViewerDestroy);CHKERRQ(ierr);
376         }
377         ierr = KSPSetFromOptions(mg[i]->smoothu);CHKERRQ(ierr);
378       }
379     }
380     for (i=1; i<n; i++) {
381       if (!mg[i]->residual) {
382         Mat mat;
383         ierr = KSPGetOperators(mg[i]->smoothd,PETSC_NULL,&mat,PETSC_NULL);CHKERRQ(ierr);
384         ierr = PCMGSetResidual(pc,i,PCMGDefaultResidual,mat);CHKERRQ(ierr);
385       }
386       if (mg[i]->restrct && !mg[i]->interpolate) {
387         ierr = PCMGSetInterpolate(pc,i,mg[i]->restrct);CHKERRQ(ierr);
388       }
389       if (!mg[i]->restrct && mg[i]->interpolate) {
390         ierr = PCMGSetRestriction(pc,i,mg[i]->interpolate);CHKERRQ(ierr);
391       }
392 #if defined(PETSC_USE_DEBUG)
393       if (!mg[i]->restrct || !mg[i]->interpolate) {
394         SETERRQ1(PETSC_ERR_ARG_WRONGSTATE,"Need to set restriction or interpolation on level %d",(int)i);
395       }
396 #endif
397     }
398     for (i=0; i<n-1; i++) {
399       if (!mg[i]->b) {
400         Mat mat;
401         Vec vec;
402         ierr = KSPGetOperators(mg[i]->smoothd,PETSC_NULL,&mat,PETSC_NULL);CHKERRQ(ierr);
403         ierr = MatGetVecs(mat,&vec,PETSC_NULL);CHKERRQ(ierr);
404         ierr = PCMGSetRhs(pc,i,vec);CHKERRQ(ierr);
405       }
406       if (!mg[i]->r && i) {
407         ierr = VecDuplicate(mg[i]->b,&tvec);CHKERRQ(ierr);
408         ierr = PCMGSetR(pc,i,tvec);CHKERRQ(ierr);
409         ierr = VecDestroy(tvec);CHKERRQ(ierr);
410       }
411       if (!mg[i]->x) {
412         ierr = VecDuplicate(mg[i]->b,&tvec);CHKERRQ(ierr);
413         ierr = PCMGSetX(pc,i,tvec);CHKERRQ(ierr);
414         ierr = VecDestroy(tvec);CHKERRQ(ierr);
415       }
416     }
417   }
418 
419   /* If user did not provide fine grid operators, use those from PC */
420   /* BUG BUG BUG This will work ONLY the first time called: hence if the user changes
421      the PC matrices between solves PCMG will continue to use first set provided */
422   ierr = KSPGetOperators(mg[n-1]->smoothd,&dA,&dB,&uflag);CHKERRQ(ierr);
423   if (!dA  && !dB) {
424     ierr = PetscInfo(pc,"Using outer operators to define finest grid operator \n  because PCMGGetSmoother(pc,nlevels-1,&ksp);KSPSetOperators(ksp,...); was not called.\n");CHKERRQ(ierr);
425     ierr = KSPSetOperators(mg[n-1]->smoothd,pc->mat,pc->pmat,uflag);CHKERRQ(ierr);
426   }
427 
428   if (mg[0]->galerkin) {
429     Mat B;
430     mg[0]->galerkinused = PETSC_TRUE;
431     /* currently only handle case where mat and pmat are the same on coarser levels */
432     ierr = KSPGetOperators(mg[n-1]->smoothd,&dA,&dB,&uflag);CHKERRQ(ierr);
433     if (!pc->setupcalled) {
434       for (i=n-2; i>-1; i--) {
435         ierr = MatPtAP(dB,mg[i+1]->interpolate,MAT_INITIAL_MATRIX,1.0,&B);CHKERRQ(ierr);
436         ierr = KSPSetOperators(mg[i]->smoothd,B,B,uflag);CHKERRQ(ierr);
437         dB   = B;
438       }
439     } else {
440       for (i=n-2; i>-1; i--) {
441         ierr = KSPGetOperators(mg[i]->smoothd,0,&B,0);CHKERRQ(ierr);
442         ierr = MatPtAP(dB,mg[i+1]->interpolate,MAT_REUSE_MATRIX,1.0,&B);CHKERRQ(ierr);
443         ierr = KSPSetOperators(mg[i]->smoothd,B,B,uflag);CHKERRQ(ierr);
444         dB   = B;
445       }
446     }
447   }
448 
449   for (i=1; i<n; i++) {
450     if (mg[i]->smoothu == mg[i]->smoothd) {
451       /* if doing only down then initial guess is zero */
452       ierr = KSPSetInitialGuessNonzero(mg[i]->smoothd,PETSC_TRUE);CHKERRQ(ierr);
453     }
454     if (mg[i]->eventsetup) {ierr = PetscLogEventBegin(mg[i]->eventsetup,0,0,0,0);CHKERRQ(ierr);}
455     ierr = KSPSetUp(mg[i]->smoothd);CHKERRQ(ierr);
456     if (mg[i]->eventsetup) {ierr = PetscLogEventEnd(mg[i]->eventsetup,0,0,0,0);CHKERRQ(ierr);}
457   }
458   for (i=1; i<n; i++) {
459     if (mg[i]->smoothu && mg[i]->smoothu != mg[i]->smoothd) {
460       PC           uppc,downpc;
461       Mat          downmat,downpmat,upmat,uppmat;
462       MatStructure matflag;
463 
464       /* check if operators have been set for up, if not use down operators to set them */
465       ierr = KSPGetPC(mg[i]->smoothu,&uppc);CHKERRQ(ierr);
466       ierr = PCGetOperators(uppc,&upmat,&uppmat,PETSC_NULL);CHKERRQ(ierr);
467       if (!upmat) {
468         ierr = KSPGetPC(mg[i]->smoothd,&downpc);CHKERRQ(ierr);
469         ierr = PCGetOperators(downpc,&downmat,&downpmat,&matflag);CHKERRQ(ierr);
470         ierr = KSPSetOperators(mg[i]->smoothu,downmat,downpmat,matflag);CHKERRQ(ierr);
471       }
472 
473       ierr = KSPSetInitialGuessNonzero(mg[i]->smoothu,PETSC_TRUE);CHKERRQ(ierr);
474       if (mg[i]->eventsetup) {ierr = PetscLogEventBegin(mg[i]->eventsetup,0,0,0,0);CHKERRQ(ierr);}
475       ierr = KSPSetUp(mg[i]->smoothu);CHKERRQ(ierr);
476       if (mg[i]->eventsetup) {ierr = PetscLogEventEnd(mg[i]->eventsetup,0,0,0,0);CHKERRQ(ierr);}
477     }
478   }
479 
480   /*
481       If coarse solver is not direct method then DO NOT USE preonly
482   */
483   ierr = PetscTypeCompare((PetscObject)mg[0]->smoothd,KSPPREONLY,&preonly);CHKERRQ(ierr);
484   if (preonly) {
485     ierr = KSPGetPC(mg[0]->smoothd,&cpc);CHKERRQ(ierr);
486     ierr = PetscTypeCompare((PetscObject)cpc,PCLU,&lu);CHKERRQ(ierr);
487     ierr = PetscTypeCompare((PetscObject)cpc,PCREDUNDANT,&redundant);CHKERRQ(ierr);
488     ierr = PetscTypeCompare((PetscObject)cpc,PCCHOLESKY,&cholesky);CHKERRQ(ierr);
489     if (!lu && !redundant && !cholesky) {
490       ierr = KSPSetType(mg[0]->smoothd,KSPGMRES);CHKERRQ(ierr);
491     }
492   }
493 
494   if (!pc->setupcalled) {
495     if (monitor) {
496       ierr = PetscObjectGetComm((PetscObject)mg[0]->smoothd,&comm);CHKERRQ(ierr);
497       ierr = PetscViewerASCIIOpen(comm,"stdout",&ascii);CHKERRQ(ierr);
498       ierr = PetscViewerASCIISetTab(ascii,n);CHKERRQ(ierr);
499       ierr = KSPSetMonitor(mg[0]->smoothd,KSPDefaultMonitor,ascii,(PetscErrorCode(*)(void*))PetscViewerDestroy);CHKERRQ(ierr);
500     }
501     ierr = KSPSetFromOptions(mg[0]->smoothd);CHKERRQ(ierr);
502   }
503 
504   if (mg[0]->eventsetup) {ierr = PetscLogEventBegin(mg[0]->eventsetup,0,0,0,0);CHKERRQ(ierr);}
505   ierr = KSPSetUp(mg[0]->smoothd);CHKERRQ(ierr);
506   if (mg[0]->eventsetup) {ierr = PetscLogEventEnd(mg[0]->eventsetup,0,0,0,0);CHKERRQ(ierr);}
507 
508 #if defined(PETSC_USE_SOCKET_VIEWER)
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 #endif
524 
525   ierr = PetscOptionsHasName(pc->prefix,"-pc_mg_dump_binary",&dump);CHKERRQ(ierr);
526   if (dump) {
527     for (i=1; i<n; i++) {
528       ierr = MatView(mg[i]->restrct,PETSC_VIEWER_BINARY_(pc->comm));CHKERRQ(ierr);
529     }
530     for (i=0; i<n; i++) {
531       ierr = KSPGetPC(mg[i]->smoothd,&pc);CHKERRQ(ierr);
532       ierr = MatView(pc->mat,PETSC_VIEWER_BINARY_(pc->comm));CHKERRQ(ierr);
533     }
534   }
535   PetscFunctionReturn(0);
536 }
537 
538 /* -------------------------------------------------------------------------------------*/
539 
540 #undef __FUNCT__
541 #define __FUNCT__ "PCMGSetLevels"
542 /*@C
543    PCMGSetLevels - Sets the number of levels to use with MG.
544    Must be called before any other MG routine.
545 
546    Collective on PC
547 
548    Input Parameters:
549 +  pc - the preconditioner context
550 .  levels - the number of levels
551 -  comms - optional communicators for each level; this is to allow solving the coarser problems
552            on smaller sets of processors. Use PETSC_NULL_OBJECT for default in Fortran
553 
554    Level: intermediate
555 
556    Notes:
557      If the number of levels is one then the multigrid uses the -mg_levels prefix
558   for setting the level options rather than the -mg_coarse prefix.
559 
560 .keywords: MG, set, levels, multigrid
561 
562 .seealso: PCMGSetType(), PCMGGetLevels()
563 @*/
564 PetscErrorCode PETSCKSP_DLLEXPORT PCMGSetLevels(PC pc,PetscInt levels,MPI_Comm *comms)
565 {
566   PetscErrorCode ierr;
567   PC_MG          **mg=0;
568 
569   PetscFunctionBegin;
570   PetscValidHeaderSpecific(pc,PC_COOKIE,1);
571 
572   if (pc->data) {
573     SETERRQ(PETSC_ERR_ORDER,"Number levels already set for MG\n\
574     make sure that you call PCMGSetLevels() before KSPSetFromOptions()");
575   }
576   ierr                     = PCMGCreate_Private(pc->comm,levels,pc,comms,&mg);CHKERRQ(ierr);
577   mg[0]->am                = PC_MG_MULTIPLICATIVE;
578   pc->data                 = (void*)mg;
579   pc->ops->applyrichardson = PCApplyRichardson_MG;
580   PetscFunctionReturn(0);
581 }
582 
583 #undef __FUNCT__
584 #define __FUNCT__ "PCMGGetLevels"
585 /*@
586    PCMGGetLevels - Gets the number of levels to use with MG.
587 
588    Not Collective
589 
590    Input Parameter:
591 .  pc - the preconditioner context
592 
593    Output parameter:
594 .  levels - the number of levels
595 
596    Level: advanced
597 
598 .keywords: MG, get, levels, multigrid
599 
600 .seealso: PCMGSetLevels()
601 @*/
602 PetscErrorCode PETSCKSP_DLLEXPORT PCMGGetLevels(PC pc,PetscInt *levels)
603 {
604   PC_MG  **mg;
605 
606   PetscFunctionBegin;
607   PetscValidHeaderSpecific(pc,PC_COOKIE,1);
608   PetscValidIntPointer(levels,2);
609 
610   mg      = (PC_MG**)pc->data;
611   *levels = mg[0]->levels;
612   PetscFunctionReturn(0);
613 }
614 
615 #undef __FUNCT__
616 #define __FUNCT__ "PCMGSetType"
617 /*@
618    PCMGSetType - Determines the form of multigrid to use:
619    multiplicative, additive, full, or the Kaskade algorithm.
620 
621    Collective on PC
622 
623    Input Parameters:
624 +  pc - the preconditioner context
625 -  form - multigrid form, one of PC_MG_MULTIPLICATIVE, PC_MG_ADDITIVE,
626    PC_MG_FULL, PC_MG_KASKADE
627 
628    Options Database Key:
629 .  -pc_mg_type <form> - Sets <form>, one of multiplicative,
630    additive, full, kaskade
631 
632    Level: advanced
633 
634 .keywords: MG, set, method, multiplicative, additive, full, Kaskade, multigrid
635 
636 .seealso: PCMGSetLevels()
637 @*/
638 PetscErrorCode PETSCKSP_DLLEXPORT PCMGSetType(PC pc,PCMGType form)
639 {
640   PC_MG **mg;
641 
642   PetscFunctionBegin;
643   PetscValidHeaderSpecific(pc,PC_COOKIE,1);
644   mg = (PC_MG**)pc->data;
645 
646   if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling");
647   mg[0]->am = form;
648   if (form == PC_MG_MULTIPLICATIVE) pc->ops->applyrichardson = PCApplyRichardson_MG;
649   else pc->ops->applyrichardson = 0;
650   PetscFunctionReturn(0);
651 }
652 
653 #undef __FUNCT__
654 #define __FUNCT__ "PCMGSetCycles"
655 /*@
656    PCMGSetCycles - Sets the type cycles to use.  Use PCMGSetCyclesOnLevel() for more
657    complicated cycling.
658 
659    Collective on PC
660 
661    Input Parameters:
662 +  pc - the multigrid context
663 -  n - the number of cycles
664 
665    Options Database Key:
666 $  -pc_mg_cycles n - 1 denotes a V-cycle; 2 denotes a W-cycle.
667 
668    Level: advanced
669 
670 .keywords: MG, set, cycles, V-cycle, W-cycle, multigrid
671 
672 .seealso: PCMGSetCyclesOnLevel()
673 @*/
674 PetscErrorCode PETSCKSP_DLLEXPORT PCMGSetCycles(PC pc,PetscInt n)
675 {
676   PC_MG    **mg;
677   PetscInt i,levels;
678 
679   PetscFunctionBegin;
680   PetscValidHeaderSpecific(pc,PC_COOKIE,1);
681   mg     = (PC_MG**)pc->data;
682   if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling");
683   levels = mg[0]->levels;
684 
685   for (i=0; i<levels; i++) {
686     mg[i]->cycles  = n;
687   }
688   PetscFunctionReturn(0);
689 }
690 
691 #undef __FUNCT__
692 #define __FUNCT__ "PCMGSetGalerkin"
693 /*@
694    PCMGSetGalerkin - Causes the coarser grid matrices to be computed from the
695       finest grid via the Galerkin process: A_i-1 = r_i * A_i * r_i^t
696 
697    Collective on PC
698 
699    Input Parameters:
700 .  pc - the multigrid context
701 
702    Options Database Key:
703 $  -pc_mg_galerkin
704 
705    Level: intermediate
706 
707 .keywords: MG, set, Galerkin
708 
709 .seealso: PCMGGetGalerkin()
710 
711 @*/
712 PetscErrorCode PETSCKSP_DLLEXPORT PCMGSetGalerkin(PC pc)
713 {
714   PC_MG    **mg;
715   PetscInt i,levels;
716 
717   PetscFunctionBegin;
718   PetscValidHeaderSpecific(pc,PC_COOKIE,1);
719   mg     = (PC_MG**)pc->data;
720   if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling");
721   levels = mg[0]->levels;
722 
723   for (i=0; i<levels; i++) {
724     mg[i]->galerkin = PETSC_TRUE;
725   }
726   PetscFunctionReturn(0);
727 }
728 
729 #undef __FUNCT__
730 #define __FUNCT__ "PCMGGetGalerkin"
731 /*@
732    PCMGGetGalerkin - Checks if Galerkin multigrid is being used, i.e.
733       A_i-1 = r_i * A_i * r_i^t
734 
735    Not Collective
736 
737    Input Parameter:
738 .  pc - the multigrid context
739 
740    Output Parameter:
741 .  gelerkin - PETSC_TRUE or PETSC_FALSE
742 
743    Options Database Key:
744 $  -pc_mg_galerkin
745 
746    Level: intermediate
747 
748 .keywords: MG, set, Galerkin
749 
750 .seealso: PCMGSetGalerkin()
751 
752 @*/
753 PetscErrorCode PETSCKSP_DLLEXPORT PCMGGetGalerkin(PC pc,PetscTruth *galerkin)
754 {
755   PC_MG    **mg;
756 
757   PetscFunctionBegin;
758   PetscValidHeaderSpecific(pc,PC_COOKIE,1);
759   mg     = (PC_MG**)pc->data;
760   if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling");
761   *galerkin = mg[0]->galerkin;
762   PetscFunctionReturn(0);
763 }
764 
765 #undef __FUNCT__
766 #define __FUNCT__ "PCMGSetNumberSmoothDown"
767 /*@
768    PCMGSetNumberSmoothDown - Sets the number of pre-smoothing steps to
769    use on all levels. Use PCMGGetSmootherDown() to set different
770    pre-smoothing steps on different levels.
771 
772    Collective on PC
773 
774    Input Parameters:
775 +  mg - the multigrid context
776 -  n - the number of smoothing steps
777 
778    Options Database Key:
779 .  -pc_mg_smoothdown <n> - Sets number of pre-smoothing steps
780 
781    Level: advanced
782 
783 .keywords: MG, smooth, down, pre-smoothing, steps, multigrid
784 
785 .seealso: PCMGSetNumberSmoothUp()
786 @*/
787 PetscErrorCode PETSCKSP_DLLEXPORT PCMGSetNumberSmoothDown(PC pc,PetscInt n)
788 {
789   PC_MG          **mg;
790   PetscErrorCode ierr;
791   PetscInt       i,levels;
792 
793   PetscFunctionBegin;
794   PetscValidHeaderSpecific(pc,PC_COOKIE,1);
795   mg     = (PC_MG**)pc->data;
796   if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling");
797   levels = mg[0]->levels;
798 
799   for (i=1; i<levels; i++) {
800     /* make sure smoother up and down are different */
801     ierr = PCMGGetSmootherUp(pc,i,PETSC_NULL);CHKERRQ(ierr);
802     ierr = KSPSetTolerances(mg[i]->smoothd,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT,n);CHKERRQ(ierr);
803     mg[i]->default_smoothd = n;
804   }
805   PetscFunctionReturn(0);
806 }
807 
808 #undef __FUNCT__
809 #define __FUNCT__ "PCMGSetNumberSmoothUp"
810 /*@
811    PCMGSetNumberSmoothUp - Sets the number of post-smoothing steps to use
812    on all levels. Use PCMGGetSmootherUp() to set different numbers of
813    post-smoothing steps on different levels.
814 
815    Collective on PC
816 
817    Input Parameters:
818 +  mg - the multigrid context
819 -  n - the number of smoothing steps
820 
821    Options Database Key:
822 .  -pc_mg_smoothup <n> - Sets number of post-smoothing steps
823 
824    Level: advanced
825 
826    Note: this does not set a value on the coarsest grid, since we assume that
827     there is no separate smooth up on the coarsest grid.
828 
829 .keywords: MG, smooth, up, post-smoothing, steps, multigrid
830 
831 .seealso: PCMGSetNumberSmoothDown()
832 @*/
833 PetscErrorCode PETSCKSP_DLLEXPORT PCMGSetNumberSmoothUp(PC pc,PetscInt n)
834 {
835   PC_MG          **mg;
836   PetscErrorCode ierr;
837   PetscInt       i,levels;
838 
839   PetscFunctionBegin;
840   PetscValidHeaderSpecific(pc,PC_COOKIE,1);
841   mg     = (PC_MG**)pc->data;
842   if (!mg) SETERRQ(PETSC_ERR_ARG_WRONGSTATE,"Must set MG levels before calling");
843   levels = mg[0]->levels;
844 
845   for (i=1; i<levels; i++) {
846     /* make sure smoother up and down are different */
847     ierr = PCMGGetSmootherUp(pc,i,PETSC_NULL);CHKERRQ(ierr);
848     ierr = KSPSetTolerances(mg[i]->smoothu,PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT,n);CHKERRQ(ierr);
849     mg[i]->default_smoothu = n;
850   }
851   PetscFunctionReturn(0);
852 }
853 
854 /* ----------------------------------------------------------------------------------------*/
855 
856 /*MC
857    PCMG - Use geometric multigrid preconditioning. This preconditioner requires you provide additional
858     information about the coarser grid matrices and restriction/interpolation operators.
859 
860    Options Database Keys:
861 +  -pc_mg_levels <nlevels> - number of levels including finest
862 .  -pc_mg_cycles 1 or 2 - for V or W-cycle
863 .  -pc_mg_smoothup <n> - number of smoothing steps after interpolation
864 .  -pc_mg_smoothdown <n> - number of smoothing steps before applying restriction operator
865 .  -pc_mg_type <additive,multiplicative,full,cascade> - multiplicative is the default
866 .  -pc_mg_log - log information about time spent on each level of the solver
867 .  -pc_mg_monitor - print information on the multigrid convergence
868 .  -pc_mg_galerkin - use Galerkin process to compute coarser operators
869 -  -pc_mg_dump_matlab - dumps the matrices for each level and the restriction/interpolation matrices
870                         to the Socket viewer for reading from Matlab.
871 
872    Notes:
873 
874    Level: intermediate
875 
876    Concepts: multigrid
877 
878 .seealso:  PCCreate(), PCSetType(), PCType (for list of available types), PC, PCMGType,
879            PCMGSetLevels(), PCMGGetLevels(), PCMGSetType(), PCMGSetCycles(), PCMGSetNumberSmoothDown(),
880            PCMGSetNumberSmoothUp(), PCMGGetCoarseSolve(), PCMGSetResidual(), PCMGSetInterpolation(),
881            PCMGSetRestriction(), PCMGGetSmoother(), PCMGGetSmootherUp(), PCMGGetSmootherDown(),
882            PCMGSetCyclesOnLevel(), PCMGSetRhs(), PCMGSetX(), PCMGSetR()
883 M*/
884 
885 EXTERN_C_BEGIN
886 #undef __FUNCT__
887 #define __FUNCT__ "PCCreate_MG"
888 PetscErrorCode PETSCKSP_DLLEXPORT PCCreate_MG(PC pc)
889 {
890   PetscFunctionBegin;
891   pc->ops->apply          = PCApply_MG;
892   pc->ops->setup          = PCSetUp_MG;
893   pc->ops->destroy        = PCDestroy_MG;
894   pc->ops->setfromoptions = PCSetFromOptions_MG;
895   pc->ops->view           = PCView_MG;
896 
897   pc->data                = (void*)0;
898   PetscFunctionReturn(0);
899 }
900 EXTERN_C_END
901