Actual source code: ml.c

petsc-3.4.4 2014-03-13
  2: /*
  3:    Provides an interface to the ML smoothed Aggregation
  4:    Note: Something non-obvious breaks -pc_mg_type ADDITIVE for parallel runs
  5:                                     Jed Brown, see [PETSC #18321, #18449].
  6: */
  7: #include <petsc-private/pcimpl.h>   /*I "petscpc.h" I*/
  8: #include <../src/ksp/pc/impls/mg/mgimpl.h>                    /*I "petscksp.h" I*/
  9: #include <../src/mat/impls/aij/seq/aij.h>
 10: #include <../src/mat/impls/aij/mpi/mpiaij.h>
 11: #include <petscdm.h>            /* for DMDestroy(&pc->mg) hack */

 13: EXTERN_C_BEGIN
 14: /* HAVE_CONFIG_H flag is required by ML include files */
 15: #if !defined(HAVE_CONFIG_H)
 16: #define HAVE_CONFIG_H
 17: #endif
 18: #include <ml_include.h>
 19: #include <ml_viz_stats.h>
 20: EXTERN_C_END

 22: typedef enum {PCML_NULLSPACE_AUTO,PCML_NULLSPACE_USER,PCML_NULLSPACE_BLOCK,PCML_NULLSPACE_SCALAR} PCMLNullSpaceType;
 23: static const char *const PCMLNullSpaceTypes[] = {"AUTO","USER","BLOCK","SCALAR","PCMLNullSpaceType","PCML_NULLSPACE_",0};

 25: /* The context (data structure) at each grid level */
 26: typedef struct {
 27:   Vec x,b,r;                  /* global vectors */
 28:   Mat A,P,R;
 29:   KSP ksp;
 30:   Vec coords;                 /* projected by ML, if PCSetCoordinates is called; values packed by node */
 31: } GridCtx;

 33: /* The context used to input PETSc matrix into ML at fine grid */
 34: typedef struct {
 35:   Mat         A;       /* Petsc matrix in aij format */
 36:   Mat         Aloc;    /* local portion of A to be used by ML */
 37:   Vec         x,y;
 38:   ML_Operator *mlmat;
 39:   PetscScalar *pwork;  /* tmp array used by PetscML_comm() */
 40: } FineGridCtx;

 42: /* The context associates a ML matrix with a PETSc shell matrix */
 43: typedef struct {
 44:   Mat         A;        /* PETSc shell matrix associated with mlmat */
 45:   ML_Operator *mlmat;   /* ML matrix assorciated with A */
 46:   Vec         y, work;
 47: } Mat_MLShell;

 49: /* Private context for the ML preconditioner */
 50: typedef struct {
 51:   ML                *ml_object;
 52:   ML_Aggregate      *agg_object;
 53:   GridCtx           *gridctx;
 54:   FineGridCtx       *PetscMLdata;
 55:   PetscInt          Nlevels,MaxNlevels,MaxCoarseSize,CoarsenScheme,EnergyMinimization,MinPerProc,PutOnSingleProc,RepartitionType,ZoltanScheme;
 56:   PetscReal         Threshold,DampingFactor,EnergyMinimizationDropTol,MaxMinRatio,AuxThreshold;
 57:   PetscBool         SpectralNormScheme_Anorm,BlockScaling,EnergyMinimizationCheap,Symmetrize,OldHierarchy,KeepAggInfo,Reusable,Repartition,Aux;
 58:   PetscBool         reuse_interpolation;
 59:   PCMLNullSpaceType nulltype;
 60:   PetscMPIInt       size; /* size of communicator for pc->pmat */
 61:   PetscInt          dim;  /* data from PCSetCoordinates(_ML) */
 62:   PetscInt          nloc;
 63:   PetscReal         *coords; /* ML has a grid object for each level: the finest grid will point into coords */
 64: } PC_ML;

 68: static int PetscML_getrow(ML_Operator *ML_data, int N_requested_rows, int requested_rows[],int allocated_space, int columns[], double values[], int row_lengths[])
 69: {
 71:   PetscInt       m,i,j,k=0,row,*aj;
 72:   PetscScalar    *aa;
 73:   FineGridCtx    *ml=(FineGridCtx*)ML_Get_MyGetrowData(ML_data);
 74:   Mat_SeqAIJ     *a = (Mat_SeqAIJ*)ml->Aloc->data;

 76:   MatGetSize(ml->Aloc,&m,NULL); if (ierr) return(0);
 77:   for (i = 0; i<N_requested_rows; i++) {
 78:     row            = requested_rows[i];
 79:     row_lengths[i] = a->ilen[row];
 80:     if (allocated_space < k+row_lengths[i]) return(0);
 81:     if ((row >= 0) || (row <= (m-1))) {
 82:       aj = a->j + a->i[row];
 83:       aa = a->a + a->i[row];
 84:       for (j=0; j<row_lengths[i]; j++) {
 85:         columns[k]  = aj[j];
 86:         values[k++] = aa[j];
 87:       }
 88:     }
 89:   }
 90:   return(1);
 91: }

 95: static PetscErrorCode PetscML_comm(double p[],void *ML_data)
 96: {
 98:   FineGridCtx    *ml = (FineGridCtx*)ML_data;
 99:   Mat            A   = ml->A;
100:   Mat_MPIAIJ     *a  = (Mat_MPIAIJ*)A->data;
101:   PetscMPIInt    size;
102:   PetscInt       i,in_length=A->rmap->n,out_length=ml->Aloc->cmap->n;
103:   PetscScalar    *array;

106:   MPI_Comm_size(PetscObjectComm((PetscObject)A),&size);
107:   if (size == 1) return 0;

109:   VecPlaceArray(ml->y,p);
110:   VecScatterBegin(a->Mvctx,ml->y,a->lvec,INSERT_VALUES,SCATTER_FORWARD);
111:   VecScatterEnd(a->Mvctx,ml->y,a->lvec,INSERT_VALUES,SCATTER_FORWARD);
112:   VecResetArray(ml->y);
113:   VecGetArray(a->lvec,&array);
114:   for (i=in_length; i<out_length; i++) p[i] = array[i-in_length];
115:   VecRestoreArray(a->lvec,&array);
116:   return(0);
117: }

121: static int PetscML_matvec(ML_Operator *ML_data,int in_length,double p[],int out_length,double ap[])
122: {
124:   FineGridCtx    *ml = (FineGridCtx*)ML_Get_MyMatvecData(ML_data);
125:   Mat            A   = ml->A, Aloc=ml->Aloc;
126:   PetscMPIInt    size;
127:   PetscScalar    *pwork=ml->pwork;
128:   PetscInt       i;

131:   MPI_Comm_size(PetscObjectComm((PetscObject)A),&size);
132:   if (size == 1) {
133:     VecPlaceArray(ml->x,p);
134:   } else {
135:     for (i=0; i<in_length; i++) pwork[i] = p[i];
136:     PetscML_comm(pwork,ml);
137:     VecPlaceArray(ml->x,pwork);
138:   }
139:   VecPlaceArray(ml->y,ap);
140:   MatMult(Aloc,ml->x,ml->y);
141:   VecResetArray(ml->x);
142:   VecResetArray(ml->y);
143:   return(0);
144: }

148: static PetscErrorCode MatMult_ML(Mat A,Vec x,Vec y)
149: {
151:   Mat_MLShell    *shell;
152:   PetscScalar    *xarray,*yarray;
153:   PetscInt       x_length,y_length;

156:   MatShellGetContext(A,(void**)&shell);
157:   VecGetArray(x,&xarray);
158:   VecGetArray(y,&yarray);
159:   x_length = shell->mlmat->invec_leng;
160:   y_length = shell->mlmat->outvec_leng;
161:   PetscStackCall("ML_Operator_Apply",ML_Operator_Apply(shell->mlmat,x_length,xarray,y_length,yarray));
162:   VecRestoreArray(x,&xarray);
163:   VecRestoreArray(y,&yarray);
164:   return(0);
165: }

169: /* Computes y = w + A * x
170:    It is possible that w == y, but not x == y
171: */
172: static PetscErrorCode MatMultAdd_ML(Mat A,Vec x,Vec w,Vec y)
173: {
174:   Mat_MLShell    *shell;
175:   PetscScalar    *xarray,*yarray;
176:   PetscInt       x_length,y_length;

180:   MatShellGetContext(A, (void**) &shell);
181:   if (y == w) {
182:     if (!shell->work) {
183:       VecDuplicate(y, &shell->work);
184:     }
185:     VecGetArray(x,           &xarray);
186:     VecGetArray(shell->work, &yarray);
187:     x_length = shell->mlmat->invec_leng;
188:     y_length = shell->mlmat->outvec_leng;
189:     PetscStackCall("ML_Operator_Apply",ML_Operator_Apply(shell->mlmat, x_length, xarray, y_length, yarray));
190:     VecRestoreArray(x,           &xarray);
191:     VecRestoreArray(shell->work, &yarray);
192:     VecAXPY(y, 1.0, shell->work);
193:   } else {
194:     VecGetArray(x, &xarray);
195:     VecGetArray(y, &yarray);
196:     x_length = shell->mlmat->invec_leng;
197:     y_length = shell->mlmat->outvec_leng;
198:     PetscStackCall("ML_Operator_Apply",ML_Operator_Apply(shell->mlmat, x_length, xarray, y_length, yarray));
199:     VecRestoreArray(x, &xarray);
200:     VecRestoreArray(y, &yarray);
201:     VecAXPY(y, 1.0, w);
202:   }
203:   return(0);
204: }

206: /* newtype is ignored since only handles one case */
209: static PetscErrorCode MatConvert_MPIAIJ_ML(Mat A,MatType newtype,MatReuse scall,Mat *Aloc)
210: {
212:   Mat_MPIAIJ     *mpimat=(Mat_MPIAIJ*)A->data;
213:   Mat_SeqAIJ     *mat,*a=(Mat_SeqAIJ*)(mpimat->A)->data,*b=(Mat_SeqAIJ*)(mpimat->B)->data;
214:   PetscInt       *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j;
215:   PetscScalar    *aa=a->a,*ba=b->a,*ca;
216:   PetscInt       am =A->rmap->n,an=A->cmap->n,i,j,k;
217:   PetscInt       *ci,*cj,ncols;

220:   if (am != an) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"A must have a square diagonal portion, am: %d != an: %d",am,an);

222:   if (scall == MAT_INITIAL_MATRIX) {
223:     PetscMalloc((1+am)*sizeof(PetscInt),&ci);
224:     ci[0] = 0;
225:     for (i=0; i<am; i++) ci[i+1] = ci[i] + (ai[i+1] - ai[i]) + (bi[i+1] - bi[i]);
226:     PetscMalloc((1+ci[am])*sizeof(PetscInt),&cj);
227:     PetscMalloc((1+ci[am])*sizeof(PetscScalar),&ca);

229:     k = 0;
230:     for (i=0; i<am; i++) {
231:       /* diagonal portion of A */
232:       ncols = ai[i+1] - ai[i];
233:       for (j=0; j<ncols; j++) {
234:         cj[k]   = *aj++;
235:         ca[k++] = *aa++;
236:       }
237:       /* off-diagonal portion of A */
238:       ncols = bi[i+1] - bi[i];
239:       for (j=0; j<ncols; j++) {
240:         cj[k]   = an + (*bj); bj++;
241:         ca[k++] = *ba++;
242:       }
243:     }
244:     if (k != ci[am]) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"k: %d != ci[am]: %d",k,ci[am]);

246:     /* put together the new matrix */
247:     an   = mpimat->A->cmap->n+mpimat->B->cmap->n;
248:     MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,am,an,ci,cj,ca,Aloc);

250:     /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */
251:     /* Since these are PETSc arrays, change flags to free them as necessary. */
252:     mat          = (Mat_SeqAIJ*)(*Aloc)->data;
253:     mat->free_a  = PETSC_TRUE;
254:     mat->free_ij = PETSC_TRUE;

256:     mat->nonew = 0;
257:   } else if (scall == MAT_REUSE_MATRIX) {
258:     mat=(Mat_SeqAIJ*)(*Aloc)->data;
259:     ci = mat->i; cj = mat->j; ca = mat->a;
260:     for (i=0; i<am; i++) {
261:       /* diagonal portion of A */
262:       ncols = ai[i+1] - ai[i];
263:       for (j=0; j<ncols; j++) *ca++ = *aa++;
264:       /* off-diagonal portion of A */
265:       ncols = bi[i+1] - bi[i];
266:       for (j=0; j<ncols; j++) *ca++ = *ba++;
267:     }
268:   } else SETERRQ1(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Invalid MatReuse %d",(int)scall);
269:   return(0);
270: }

272: extern PetscErrorCode MatDestroy_Shell(Mat);
275: static PetscErrorCode MatDestroy_ML(Mat A)
276: {
278:   Mat_MLShell    *shell;

281:   MatShellGetContext(A,(void**)&shell);
282:   VecDestroy(&shell->y);
283:   if (shell->work) {VecDestroy(&shell->work);}
284:   PetscFree(shell);
285:   MatDestroy_Shell(A);
286:   PetscObjectChangeTypeName((PetscObject)A,0);
287:   return(0);
288: }

292: static PetscErrorCode MatWrapML_SeqAIJ(ML_Operator *mlmat,MatReuse reuse,Mat *newmat)
293: {
294:   struct ML_CSR_MSRdata *matdata = (struct ML_CSR_MSRdata*)mlmat->data;
295:   PetscErrorCode        ierr;
296:   PetscInt              m       =mlmat->outvec_leng,n=mlmat->invec_leng,*nnz = NULL,nz_max;
297:   PetscInt              *ml_cols=matdata->columns,*ml_rowptr=matdata->rowptr,*aj,i;
298:   PetscScalar           *ml_vals=matdata->values,*aa;

301:   if (!mlmat->getrow) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_NULL,"mlmat->getrow = NULL");
302:   if (m != n) { /* ML Pmat and Rmat are in CSR format. Pass array pointers into SeqAIJ matrix */
303:     if (reuse) {
304:       Mat_SeqAIJ *aij= (Mat_SeqAIJ*)(*newmat)->data;
305:       aij->i = ml_rowptr;
306:       aij->j = ml_cols;
307:       aij->a = ml_vals;
308:     } else {
309:       /* sort ml_cols and ml_vals */
310:       PetscMalloc((m+1)*sizeof(PetscInt),&nnz);
311:       for (i=0; i<m; i++) nnz[i] = ml_rowptr[i+1] - ml_rowptr[i];
312:       aj = ml_cols; aa = ml_vals;
313:       for (i=0; i<m; i++) {
314:         PetscSortIntWithScalarArray(nnz[i],aj,aa);
315:         aj  += nnz[i]; aa += nnz[i];
316:       }
317:       MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,m,n,ml_rowptr,ml_cols,ml_vals,newmat);
318:       PetscFree(nnz);
319:     }
320:     return(0);
321:   }

323:   nz_max = PetscMax(1,mlmat->max_nz_per_row);
324:   PetscMalloc2(nz_max,PetscScalar,&aa,nz_max,PetscInt,&aj);
325:   if (!reuse) {
326:     MatCreate(PETSC_COMM_SELF,newmat);
327:     MatSetSizes(*newmat,m,n,PETSC_DECIDE,PETSC_DECIDE);
328:     MatSetType(*newmat,MATSEQAIJ);
329:     /* keep track of block size for A matrices */
330:     MatSetBlockSize (*newmat, mlmat->num_PDEs);

332:     PetscMalloc(m*sizeof(PetscInt),&nnz);
333:     for (i=0; i<m; i++) {
334:       PetscStackCall("ML_Operator_Getrow",ML_Operator_Getrow(mlmat,1,&i,nz_max,aj,aa,&nnz[i]));
335:     }
336:     MatSeqAIJSetPreallocation(*newmat,0,nnz);
337:   }
338:   for (i=0; i<m; i++) {
339:     PetscInt ncols;

341:     PetscStackCall("ML_Operator_Getrow",ML_Operator_Getrow(mlmat,1,&i,nz_max,aj,aa,&ncols));
342:     MatSetValues(*newmat,1,&i,ncols,aj,aa,INSERT_VALUES);
343:   }
344:   MatAssemblyBegin(*newmat,MAT_FINAL_ASSEMBLY);
345:   MatAssemblyEnd(*newmat,MAT_FINAL_ASSEMBLY);

347:   PetscFree2(aa,aj);
348:   PetscFree(nnz);
349:   return(0);
350: }

354: static PetscErrorCode MatWrapML_SHELL(ML_Operator *mlmat,MatReuse reuse,Mat *newmat)
355: {
357:   PetscInt       m,n;
358:   ML_Comm        *MLcomm;
359:   Mat_MLShell    *shellctx;

362:   m = mlmat->outvec_leng;
363:   n = mlmat->invec_leng;

365:   if (reuse) {
366:     MatShellGetContext(*newmat,(void**)&shellctx);
367:     shellctx->mlmat = mlmat;
368:     return(0);
369:   }

371:   MLcomm = mlmat->comm;

373:   PetscNew(Mat_MLShell,&shellctx);
374:   MatCreateShell(MLcomm->USR_comm,m,n,PETSC_DETERMINE,PETSC_DETERMINE,shellctx,newmat);
375:   MatShellSetOperation(*newmat,MATOP_MULT,(void(*)(void))MatMult_ML);
376:   MatShellSetOperation(*newmat,MATOP_MULT_ADD,(void(*)(void))MatMultAdd_ML);

378:   shellctx->A         = *newmat;
379:   shellctx->mlmat     = mlmat;
380:   shellctx->work      = NULL;

382:   VecCreate(MLcomm->USR_comm,&shellctx->y);
383:   VecSetSizes(shellctx->y,m,PETSC_DECIDE);
384:   VecSetFromOptions(shellctx->y);

386:   (*newmat)->ops->destroy = MatDestroy_ML;
387:   return(0);
388: }

392: static PetscErrorCode MatWrapML_MPIAIJ(ML_Operator *mlmat,MatReuse reuse,Mat *newmat)
393: {
394:   PetscInt       *aj;
395:   PetscScalar    *aa;
397:   PetscInt       i,j,*gordering;
398:   PetscInt       m=mlmat->outvec_leng,n,nz_max,row;
399:   Mat            A;

402:   if (!mlmat->getrow) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_NULL,"mlmat->getrow = NULL");
403:   n = mlmat->invec_leng;
404:   if (m != n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"m %d must equal to n %d",m,n);

406:   nz_max = PetscMax(1,mlmat->max_nz_per_row);
407:   PetscMalloc2(nz_max,PetscScalar,&aa,nz_max,PetscInt,&aj);
408:   if (reuse) A = *newmat;
409:   else {
410:     PetscInt *nnzA,*nnzB,*nnz;
411:     MatCreate(mlmat->comm->USR_comm,&A);
412:     MatSetSizes(A,m,n,PETSC_DECIDE,PETSC_DECIDE);
413:     MatSetType(A,MATMPIAIJ);
414:     /* keep track of block size for A matrices */
415:     MatSetBlockSize (A,mlmat->num_PDEs);
416:     PetscMalloc3(m,PetscInt,&nnzA,m,PetscInt,&nnzB,m,PetscInt,&nnz);

418:     for (i=0; i<m; i++) {
419:       PetscStackCall("ML_Operator_Getrow",ML_Operator_Getrow(mlmat,1,&i,nz_max,aj,aa,&nnz[i]));
420:       nnzA[i] = 0;
421:       for (j=0; j<nnz[i]; j++) {
422:         if (aj[j] < m) nnzA[i]++;
423:       }
424:       nnzB[i] = nnz[i] - nnzA[i];
425:     }
426:     MatMPIAIJSetPreallocation(A,0,nnzA,0,nnzB);
427:     PetscFree3(nnzA,nnzB,nnz);
428:   }
429:   /* create global row numbering for a ML_Operator */
430:   PetscStackCall("ML_build_global_numbering",ML_build_global_numbering(mlmat,&gordering,"rows"));
431:   for (i=0; i<m; i++) {
432:     PetscInt ncols;
433:     row = gordering[i];

435:     PetscStackCall(",ML_Operator_Getrow",ML_Operator_Getrow(mlmat,1,&i,nz_max,aj,aa,&ncols));
436:     for (j = 0; j < ncols; j++) aj[j] = gordering[aj[j]];
437:     MatSetValues(A,1,&row,ncols,aj,aa,INSERT_VALUES);
438:   }
439:   PetscStackCall("ML_Operator_Getrow",ML_free(gordering));
440:   MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);
441:   MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);
442:   *newmat = A;

444:   PetscFree2(aa,aj);
445:   return(0);
446: }

448: /* -------------------------------------------------------------------------- */
449: /*
450:    PCSetCoordinates_ML

452:    Input Parameter:
453:    .  pc - the preconditioner context
454: */
457: static PetscErrorCode PCSetCoordinates_ML(PC pc, PetscInt ndm, PetscInt a_nloc, PetscReal *coords)
458: {
459:   PC_MG          *mg    = (PC_MG*)pc->data;
460:   PC_ML          *pc_ml = (PC_ML*)mg->innerctx;
462:   PetscInt       arrsz,oldarrsz,bs,my0,kk,ii,nloc,Iend;
463:   Mat            Amat = pc->pmat;

465:   /* this function copied and modified from PCSetCoordinates_GEO -TGI */
468:   MatGetBlockSize(Amat, &bs);

470:   MatGetOwnershipRange(Amat, &my0, &Iend);
471:   nloc = (Iend-my0)/bs;

473:   if (nloc!=a_nloc) SETERRQ2(PetscObjectComm((PetscObject)Amat),PETSC_ERR_ARG_WRONG, "Number of local blocks must locations = %d %d.",a_nloc,nloc);
474:   if ((Iend-my0)%bs!=0) SETERRQ1(PetscObjectComm((PetscObject)Amat),PETSC_ERR_ARG_WRONG, "Bad local size %d.",nloc);

476:   oldarrsz    = pc_ml->dim * pc_ml->nloc;
477:   pc_ml->dim  = ndm;
478:   pc_ml->nloc = a_nloc;
479:   arrsz       = ndm * a_nloc;

481:   /* create data - syntactic sugar that should be refactored at some point */
482:   if (pc_ml->coords==0 || (oldarrsz != arrsz)) {
483:     PetscFree(pc_ml->coords);
484:     PetscMalloc((arrsz)*sizeof(PetscReal), &pc_ml->coords);
485:   }
486:   for (kk=0; kk<arrsz; kk++) pc_ml->coords[kk] = -999.;
487:   /* copy data in - column oriented */
488:   for (kk = 0; kk < nloc; kk++) {
489:     for (ii = 0; ii < ndm; ii++) {
490:       pc_ml->coords[ii*nloc + kk] =  coords[kk*ndm + ii];
491:     }
492:   }
493:   return(0);
494: }

496: /* -----------------------------------------------------------------------------*/
499: PetscErrorCode PCReset_ML(PC pc)
500: {
502:   PC_MG          *mg    = (PC_MG*)pc->data;
503:   PC_ML          *pc_ml = (PC_ML*)mg->innerctx;
504:   PetscInt       level,fine_level=pc_ml->Nlevels-1,dim=pc_ml->dim;

507:   if (dim) {
508:     ML_Aggregate_Viz_Stats * grid_info = (ML_Aggregate_Viz_Stats*) pc_ml->ml_object->Grid[0].Grid;

510:     for (level=0; level<=fine_level; level++) {
511:       VecDestroy(&pc_ml->gridctx[level].coords);
512:     }

514:     grid_info->x = 0; /* do this so ML doesn't try to free coordinates */
515:     grid_info->y = 0;
516:     grid_info->z = 0;

518:     PetscStackCall("ML_Operator_Getrow",ML_Aggregate_VizAndStats_Clean(pc_ml->ml_object));
519:   }
520:   PetscStackCall("ML_Aggregate_Destroy",ML_Aggregate_Destroy(&pc_ml->agg_object));
521:   PetscStackCall("ML_Aggregate_Destroy",ML_Destroy(&pc_ml->ml_object));

523:   if (pc_ml->PetscMLdata) {
524:     PetscFree(pc_ml->PetscMLdata->pwork);
525:     MatDestroy(&pc_ml->PetscMLdata->Aloc);
526:     VecDestroy(&pc_ml->PetscMLdata->x);
527:     VecDestroy(&pc_ml->PetscMLdata->y);
528:   }
529:   PetscFree(pc_ml->PetscMLdata);

531:   if (pc_ml->gridctx) {
532:     for (level=0; level<fine_level; level++) {
533:       if (pc_ml->gridctx[level].A) {MatDestroy(&pc_ml->gridctx[level].A);}
534:       if (pc_ml->gridctx[level].P) {MatDestroy(&pc_ml->gridctx[level].P);}
535:       if (pc_ml->gridctx[level].R) {MatDestroy(&pc_ml->gridctx[level].R);}
536:       if (pc_ml->gridctx[level].x) {VecDestroy(&pc_ml->gridctx[level].x);}
537:       if (pc_ml->gridctx[level].b) {VecDestroy(&pc_ml->gridctx[level].b);}
538:       if (pc_ml->gridctx[level+1].r) {VecDestroy(&pc_ml->gridctx[level+1].r);}
539:     }
540:   }
541:   PetscFree(pc_ml->gridctx);
542:   PetscFree(pc_ml->coords);

544:   pc_ml->dim  = 0;
545:   pc_ml->nloc = 0;
546:   return(0);
547: }
548: /* -------------------------------------------------------------------------- */
549: /*
550:    PCSetUp_ML - Prepares for the use of the ML preconditioner
551:                     by setting data structures and options.

553:    Input Parameter:
554: .  pc - the preconditioner context

556:    Application Interface Routine: PCSetUp()

558:    Notes:
559:    The interface routine PCSetUp() is not usually called directly by
560:    the user, but instead is called by PCApply() if necessary.
561: */
562: extern PetscErrorCode PCSetFromOptions_MG(PC);
563: extern PetscErrorCode PCReset_MG(PC);

567: PetscErrorCode PCSetUp_ML(PC pc)
568: {
570:   PetscMPIInt    size;
571:   FineGridCtx    *PetscMLdata;
572:   ML             *ml_object;
573:   ML_Aggregate   *agg_object;
574:   ML_Operator    *mlmat;
575:   PetscInt       nlocal_allcols,Nlevels,mllevel,level,level1,m,fine_level,bs;
576:   Mat            A,Aloc;
577:   GridCtx        *gridctx;
578:   PC_MG          *mg    = (PC_MG*)pc->data;
579:   PC_ML          *pc_ml = (PC_ML*)mg->innerctx;
580:   PetscBool      isSeq, isMPI;
581:   KSP            smoother;
582:   PC             subpc;
583:   PetscInt       mesh_level, old_mesh_level;
584:   MatInfo        info;

587:   A    = pc->pmat;
588:   MPI_Comm_size(PetscObjectComm((PetscObject)A),&size);

590:   if (pc->setupcalled) {
591:     if (pc->flag == SAME_NONZERO_PATTERN && pc_ml->reuse_interpolation) {
592:       /*
593:        Reuse interpolaton instead of recomputing aggregates and updating the whole hierarchy. This is less expensive for
594:        multiple solves in which the matrix is not changing too quickly.
595:        */
596:       ml_object             = pc_ml->ml_object;
597:       gridctx               = pc_ml->gridctx;
598:       Nlevels               = pc_ml->Nlevels;
599:       fine_level            = Nlevels - 1;
600:       gridctx[fine_level].A = A;

602:       PetscObjectTypeCompare((PetscObject) A, MATSEQAIJ, &isSeq);
603:       PetscObjectTypeCompare((PetscObject) A, MATMPIAIJ, &isMPI);
604:       if (isMPI) {
605:         MatConvert_MPIAIJ_ML(A,NULL,MAT_INITIAL_MATRIX,&Aloc);
606:       } else if (isSeq) {
607:         Aloc = A;
608:         PetscObjectReference((PetscObject)Aloc);
609:       } else SETERRQ1(PetscObjectComm((PetscObject)pc),PETSC_ERR_ARG_WRONG, "Matrix type '%s' cannot be used with ML. ML can only handle AIJ matrices.",((PetscObject)A)->type_name);

611:       MatGetSize(Aloc,&m,&nlocal_allcols);
612:       PetscMLdata       = pc_ml->PetscMLdata;
613:       MatDestroy(&PetscMLdata->Aloc);
614:       PetscMLdata->A    = A;
615:       PetscMLdata->Aloc = Aloc;
616:       PetscStackCall("ML_Aggregate_Destroy",ML_Init_Amatrix(ml_object,0,m,m,PetscMLdata));
617:       PetscStackCall("ML_Set_Amatrix_Matvec",ML_Set_Amatrix_Matvec(ml_object,0,PetscML_matvec));

619:       mesh_level = ml_object->ML_finest_level;
620:       while (ml_object->SingleLevel[mesh_level].Rmat->to) {
621:         old_mesh_level = mesh_level;
622:         mesh_level     = ml_object->SingleLevel[mesh_level].Rmat->to->levelnum;

624:         /* clean and regenerate A */
625:         mlmat = &(ml_object->Amat[mesh_level]);
626:         PetscStackCall("ML_Operator_Clean",ML_Operator_Clean(mlmat));
627:         PetscStackCall("ML_Operator_Init",ML_Operator_Init(mlmat,ml_object->comm));
628:         PetscStackCall("ML_Gen_AmatrixRAP",ML_Gen_AmatrixRAP(ml_object, old_mesh_level, mesh_level));
629:       }

631:       level = fine_level - 1;
632:       if (size == 1) { /* convert ML P, R and A into seqaij format */
633:         for (mllevel=1; mllevel<Nlevels; mllevel++) {
634:           mlmat = &(ml_object->Amat[mllevel]);
635:           MatWrapML_SeqAIJ(mlmat,MAT_REUSE_MATRIX,&gridctx[level].A);
636:           level--;
637:         }
638:       } else { /* convert ML P and R into shell format, ML A into mpiaij format */
639:         for (mllevel=1; mllevel<Nlevels; mllevel++) {
640:           mlmat  = &(ml_object->Amat[mllevel]);
641:           MatWrapML_MPIAIJ(mlmat,MAT_REUSE_MATRIX,&gridctx[level].A);
642:           level--;
643:         }
644:       }

646:       for (level=0; level<fine_level; level++) {
647:         if (level > 0) {
648:           PCMGSetResidual(pc,level,PCMGResidual_Default,gridctx[level].A);
649:         }
650:         KSPSetOperators(gridctx[level].ksp,gridctx[level].A,gridctx[level].A,SAME_NONZERO_PATTERN);
651:       }
652:       PCMGSetResidual(pc,fine_level,PCMGResidual_Default,gridctx[fine_level].A);
653:       KSPSetOperators(gridctx[fine_level].ksp,gridctx[level].A,gridctx[fine_level].A,SAME_NONZERO_PATTERN);

655:       PCSetUp_MG(pc);
656:       return(0);
657:     } else {
658:       /* since ML can change the size of vectors/matrices at any level we must destroy everything */
659:       PCReset_ML(pc);
660:       PCReset_MG(pc);
661:     }
662:   }

664:   /* setup special features of PCML */
665:   /*--------------------------------*/
666:   /* covert A to Aloc to be used by ML at fine grid */
667:   pc_ml->size = size;
668:   PetscObjectTypeCompare((PetscObject) A, MATSEQAIJ, &isSeq);
669:   PetscObjectTypeCompare((PetscObject) A, MATMPIAIJ, &isMPI);
670:   if (isMPI) {
671:     MatConvert_MPIAIJ_ML(A,NULL,MAT_INITIAL_MATRIX,&Aloc);
672:   } else if (isSeq) {
673:     Aloc = A;
674:     PetscObjectReference((PetscObject)Aloc);
675:   } else SETERRQ1(PetscObjectComm((PetscObject)pc),PETSC_ERR_ARG_WRONG, "Matrix type '%s' cannot be used with ML. ML can only handle AIJ matrices.",((PetscObject)A)->type_name);

677:   /* create and initialize struct 'PetscMLdata' */
678:   PetscNewLog(pc,FineGridCtx,&PetscMLdata);
679:   pc_ml->PetscMLdata = PetscMLdata;
680:   PetscMalloc((Aloc->cmap->n+1)*sizeof(PetscScalar),&PetscMLdata->pwork);

682:   VecCreate(PETSC_COMM_SELF,&PetscMLdata->x);
683:   VecSetSizes(PetscMLdata->x,Aloc->cmap->n,Aloc->cmap->n);
684:   VecSetType(PetscMLdata->x,VECSEQ);

686:   VecCreate(PETSC_COMM_SELF,&PetscMLdata->y);
687:   VecSetSizes(PetscMLdata->y,A->rmap->n,PETSC_DECIDE);
688:   VecSetType(PetscMLdata->y,VECSEQ);
689:   PetscMLdata->A    = A;
690:   PetscMLdata->Aloc = Aloc;
691:   if (pc_ml->dim) { /* create vecs around the coordinate data given */
692:     PetscInt  i,j,dim=pc_ml->dim;
693:     PetscInt  nloc = pc_ml->nloc,nlocghost;
694:     PetscReal *ghostedcoords;

696:     MatGetBlockSize(A,&bs);
697:     nlocghost = Aloc->cmap->n / bs;
698:     PetscMalloc(dim*nlocghost*sizeof(PetscReal),&ghostedcoords);
699:     for (i = 0; i < dim; i++) {
700:       /* copy coordinate values into first component of pwork */
701:       for (j = 0; j < nloc; j++) {
702:         PetscMLdata->pwork[bs * j] = pc_ml->coords[nloc * i + j];
703:       }
704:       /* get the ghost values */
705:       PetscML_comm(PetscMLdata->pwork,PetscMLdata);
706:       /* write into the vector */
707:       for (j = 0; j < nlocghost; j++) {
708:         ghostedcoords[i * nlocghost + j] = PetscMLdata->pwork[bs * j];
709:       }
710:     }
711:     /* replace the original coords with the ghosted coords, because these are
712:      * what ML needs */
713:     PetscFree(pc_ml->coords);
714:     pc_ml->coords = ghostedcoords;
715:   }

717:   /* create ML discretization matrix at fine grid */
718:   /* ML requires input of fine-grid matrix. It determines nlevels. */
719:   MatGetSize(Aloc,&m,&nlocal_allcols);
720:   MatGetBlockSize(A,&bs);
721:   PetscStackCall("ML_Create",ML_Create(&ml_object,pc_ml->MaxNlevels));
722:   PetscStackCall("ML_Comm_Set_UsrComm",ML_Comm_Set_UsrComm(ml_object->comm,PetscObjectComm((PetscObject)A)));
723:   pc_ml->ml_object = ml_object;
724:   PetscStackCall("ML_Init_Amatrix",ML_Init_Amatrix(ml_object,0,m,m,PetscMLdata));
725:   PetscStackCall("ML_Set_Amatrix_Getrow",ML_Set_Amatrix_Getrow(ml_object,0,PetscML_getrow,PetscML_comm,nlocal_allcols));
726:   PetscStackCall("ML_Set_Amatrix_Matvec",ML_Set_Amatrix_Matvec(ml_object,0,PetscML_matvec));

728:   PetscStackCall("ML_Set_Symmetrize",ML_Set_Symmetrize(ml_object,pc_ml->Symmetrize ? ML_YES : ML_NO));

730:   /* aggregation */
731:   PetscStackCall("ML_Aggregate_Create",ML_Aggregate_Create(&agg_object));
732:   pc_ml->agg_object = agg_object;

734:   {
735:     MatNullSpace mnull;
736:     MatGetNearNullSpace(A,&mnull);
737:     if (pc_ml->nulltype == PCML_NULLSPACE_AUTO) {
738:       if (mnull) pc_ml->nulltype = PCML_NULLSPACE_USER;
739:       else if (bs > 1) pc_ml->nulltype = PCML_NULLSPACE_BLOCK;
740:       else pc_ml->nulltype = PCML_NULLSPACE_SCALAR;
741:     }
742:     switch (pc_ml->nulltype) {
743:     case PCML_NULLSPACE_USER: {
744:       PetscScalar       *nullvec;
745:       const PetscScalar *v;
746:       PetscBool         has_const;
747:       PetscInt          i,j,mlocal,nvec,M;
748:       const Vec         *vecs;

750:       if (!mnull) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_USER,"Must provide explicit null space using MatSetNearNullSpace() to use user-specified null space");
751:       MatGetSize(A,&M,NULL);
752:       MatGetLocalSize(Aloc,&mlocal,NULL);
753:       MatNullSpaceGetVecs(mnull,&has_const,&nvec,&vecs);
754:       PetscMalloc((nvec+!!has_const)*mlocal*sizeof(*nullvec),&nullvec);
755:       if (has_const) for (i=0; i<mlocal; i++) nullvec[i] = 1.0/M;
756:       for (i=0; i<nvec; i++) {
757:         VecGetArrayRead(vecs[i],&v);
758:         for (j=0; j<mlocal; j++) nullvec[(i+!!has_const)*mlocal + j] = v[j];
759:         VecRestoreArrayRead(vecs[i],&v);
760:       }
761:       PetscStackCall("ML_Aggregate_Create",ML_Aggregate_Set_NullSpace(agg_object,bs,nvec+!!has_const,nullvec,mlocal);CHKERRQ(ierr));
762:       PetscFree(nullvec);
763:     } break;
764:     case PCML_NULLSPACE_BLOCK:
765:       PetscStackCall("ML_Aggregate_Set_NullSpace",ML_Aggregate_Set_NullSpace(agg_object,bs,bs,0,0);CHKERRQ(ierr));
766:       break;
767:     case PCML_NULLSPACE_SCALAR:
768:       break;
769:     default: SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_SUP,"Unknown null space type");
770:     }
771:   }
772:   PetscStackCall("ML_Aggregate_Set_MaxCoarseSize",ML_Aggregate_Set_MaxCoarseSize(agg_object,pc_ml->MaxCoarseSize));
773:   /* set options */
774:   switch (pc_ml->CoarsenScheme) {
775:   case 1:
776:     PetscStackCall("ML_Aggregate_Set_CoarsenScheme_Coupled",ML_Aggregate_Set_CoarsenScheme_Coupled(agg_object));break;
777:   case 2:
778:     PetscStackCall("ML_Aggregate_Set_CoarsenScheme_MIS",ML_Aggregate_Set_CoarsenScheme_MIS(agg_object));break;
779:   case 3:
780:     PetscStackCall("ML_Aggregate_Set_CoarsenScheme_METIS",ML_Aggregate_Set_CoarsenScheme_METIS(agg_object));break;
781:   }
782:   PetscStackCall("ML_Aggregate_Set_Threshold",ML_Aggregate_Set_Threshold(agg_object,pc_ml->Threshold));
783:   PetscStackCall("ML_Aggregate_Set_DampingFactor",ML_Aggregate_Set_DampingFactor(agg_object,pc_ml->DampingFactor));
784:   if (pc_ml->SpectralNormScheme_Anorm) {
785:     PetscStackCall("ML_Set_SpectralNormScheme_Anorm",ML_Set_SpectralNormScheme_Anorm(ml_object));
786:   }
787:   agg_object->keep_agg_information      = (int)pc_ml->KeepAggInfo;
788:   agg_object->keep_P_tentative          = (int)pc_ml->Reusable;
789:   agg_object->block_scaled_SA           = (int)pc_ml->BlockScaling;
790:   agg_object->minimizing_energy         = (int)pc_ml->EnergyMinimization;
791:   agg_object->minimizing_energy_droptol = (double)pc_ml->EnergyMinimizationDropTol;
792:   agg_object->cheap_minimizing_energy   = (int)pc_ml->EnergyMinimizationCheap;

794:   if (pc_ml->Aux) {
795:     if (!pc_ml->dim) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_USER,"Auxiliary matrix requires coordinates");
796:     ml_object->Amat[0].aux_data->threshold = pc_ml->AuxThreshold;
797:     ml_object->Amat[0].aux_data->enable    = 1;
798:     ml_object->Amat[0].aux_data->max_level = 10;
799:     ml_object->Amat[0].num_PDEs            = bs;
800:   }

802:   MatGetInfo(A,MAT_LOCAL,&info);
803:   ml_object->Amat[0].N_nonzeros = (int) info.nz_used;

805:   if (pc_ml->dim) {
806:     PetscInt               i,dim = pc_ml->dim;
807:     ML_Aggregate_Viz_Stats *grid_info;
808:     PetscInt               nlocghost;

810:     MatGetBlockSize(A,&bs);
811:     nlocghost = Aloc->cmap->n / bs;

813:     PetscStackCall("ML_Aggregate_VizAndStats_Setup(",ML_Aggregate_VizAndStats_Setup(ml_object)); /* create ml info for coords */
814:     grid_info = (ML_Aggregate_Viz_Stats*) ml_object->Grid[0].Grid;
815:     for (i = 0; i < dim; i++) {
816:       /* set the finest level coordinates to point to the column-order array
817:        * in pc_ml */
818:       /* NOTE: must point away before VizAndStats_Clean so ML doesn't free */
819:       switch (i) {
820:       case 0: grid_info->x = pc_ml->coords + nlocghost * i; break;
821:       case 1: grid_info->y = pc_ml->coords + nlocghost * i; break;
822:       case 2: grid_info->z = pc_ml->coords + nlocghost * i; break;
823:       default: SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_ARG_SIZ,"PCML coordinate dimension must be <= 3");
824:       }
825:     }
826:     grid_info->Ndim = dim;
827:   }

829:   /* repartitioning */
830:   if (pc_ml->Repartition) {
831:     PetscStackCall("ML_Repartition_Activate",ML_Repartition_Activate(ml_object));
832:     PetscStackCall("ML_Repartition_Set_LargestMinMaxRatio",ML_Repartition_Set_LargestMinMaxRatio(ml_object,pc_ml->MaxMinRatio));
833:     PetscStackCall("ML_Repartition_Set_MinPerProc",ML_Repartition_Set_MinPerProc(ml_object,pc_ml->MinPerProc));
834:     PetscStackCall("ML_Repartition_Set_PutOnSingleProc",ML_Repartition_Set_PutOnSingleProc(ml_object,pc_ml->PutOnSingleProc));
835: #if 0                           /* Function not yet defined in ml-6.2 */
836:     /* I'm not sure what compatibility issues might crop up if we partitioned
837:      * on the finest level, so to be safe repartition starts on the next
838:      * finest level (reflection default behavior in
839:      * ml_MultiLevelPreconditioner) */
840:     PetscStackCall("ML_Repartition_Set_StartLevel",ML_Repartition_Set_StartLevel(ml_object,1));
841: #endif

843:     if (!pc_ml->RepartitionType) {
844:       PetscInt i;

846:       if (!pc_ml->dim) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_USER,"ML Zoltan repartitioning requires coordinates");
847:       PetscStackCall("ML_Repartition_Set_Partitioner",ML_Repartition_Set_Partitioner(ml_object,ML_USEZOLTAN));
848:       PetscStackCall("ML_Aggregate_Set_Dimensions",ML_Aggregate_Set_Dimensions(agg_object, pc_ml->dim));

850:       for (i = 0; i < ml_object->ML_num_levels; i++) {
851:         ML_Aggregate_Viz_Stats *grid_info = (ML_Aggregate_Viz_Stats*)ml_object->Grid[i].Grid;
852:         grid_info->zoltan_type = pc_ml->ZoltanScheme + 1; /* ml numbers options 1, 2, 3 */
853:         /* defaults from ml_agg_info.c */
854:         grid_info->zoltan_estimated_its = 40; /* only relevant to hypergraph / fast hypergraph */
855:         grid_info->zoltan_timers        = 0;
856:         grid_info->smoothing_steps      = 4;  /* only relevant to hypergraph / fast hypergraph */
857:       }
858:     } else {
859:       PetscStackCall("ML_Repartition_Set_Partitioner",ML_Repartition_Set_Partitioner(ml_object,ML_USEPARMETIS));
860:     }
861:   }

863:   if (pc_ml->OldHierarchy) {
864:     PetscStackCall("ML_Gen_MGHierarchy_UsingAggregation",Nlevels = ML_Gen_MGHierarchy_UsingAggregation(ml_object,0,ML_INCREASING,agg_object));
865:   } else {
866:     PetscStackCall("ML_Gen_MultiLevelHierarchy_UsingAggregation",Nlevels = ML_Gen_MultiLevelHierarchy_UsingAggregation(ml_object,0,ML_INCREASING,agg_object));
867:   }
868:   if (Nlevels<=0) SETERRQ1(PetscObjectComm((PetscObject)pc),PETSC_ERR_ARG_OUTOFRANGE,"Nlevels %d must > 0",Nlevels);
869:   pc_ml->Nlevels = Nlevels;
870:   fine_level     = Nlevels - 1;

872:   PCMGSetLevels(pc,Nlevels,NULL);
873:   /* set default smoothers */
874:   for (level=1; level<=fine_level; level++) {
875:     PCMGGetSmoother(pc,level,&smoother);
876:     KSPSetType(smoother,KSPRICHARDSON);
877:     KSPGetPC(smoother,&subpc);
878:     PCSetType(subpc,PCSOR);
879:   }
880:   PetscObjectOptionsBegin((PetscObject)pc);
881:   PCSetFromOptions_MG(pc); /* should be called in PCSetFromOptions_ML(), but cannot be called prior to PCMGSetLevels() */
882:   PetscOptionsEnd();

884:   PetscMalloc(Nlevels*sizeof(GridCtx),&gridctx);

886:   pc_ml->gridctx = gridctx;

888:   /* wrap ML matrices by PETSc shell matrices at coarsened grids.
889:      Level 0 is the finest grid for ML, but coarsest for PETSc! */
890:   gridctx[fine_level].A = A;

892:   level = fine_level - 1;
893:   if (size == 1) { /* convert ML P, R and A into seqaij format */
894:     for (mllevel=1; mllevel<Nlevels; mllevel++) {
895:       mlmat = &(ml_object->Pmat[mllevel]);
896:       MatWrapML_SeqAIJ(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].P);
897:       mlmat = &(ml_object->Rmat[mllevel-1]);
898:       MatWrapML_SeqAIJ(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].R);

900:       mlmat = &(ml_object->Amat[mllevel]);
901:       MatWrapML_SeqAIJ(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].A);
902:       level--;
903:     }
904:   } else { /* convert ML P and R into shell format, ML A into mpiaij format */
905:     for (mllevel=1; mllevel<Nlevels; mllevel++) {
906:       mlmat  = &(ml_object->Pmat[mllevel]);
907:       MatWrapML_SHELL(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].P);
908:       mlmat  = &(ml_object->Rmat[mllevel-1]);
909:       MatWrapML_SHELL(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].R);

911:       mlmat  = &(ml_object->Amat[mllevel]);
912:       MatWrapML_MPIAIJ(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].A);
913:       level--;
914:     }
915:   }

917:   /* create vectors and ksp at all levels */
918:   for (level=0; level<fine_level; level++) {
919:     level1 = level + 1;
920:     VecCreate(((PetscObject)gridctx[level].A)->comm,&gridctx[level].x);
921:     VecSetSizes(gridctx[level].x,gridctx[level].A->cmap->n,PETSC_DECIDE);
922:     VecSetType(gridctx[level].x,VECMPI);
923:     PCMGSetX(pc,level,gridctx[level].x);

925:     VecCreate(((PetscObject)gridctx[level].A)->comm,&gridctx[level].b);
926:     VecSetSizes(gridctx[level].b,gridctx[level].A->rmap->n,PETSC_DECIDE);
927:     VecSetType(gridctx[level].b,VECMPI);
928:     PCMGSetRhs(pc,level,gridctx[level].b);

930:     VecCreate(((PetscObject)gridctx[level1].A)->comm,&gridctx[level1].r);
931:     VecSetSizes(gridctx[level1].r,gridctx[level1].A->rmap->n,PETSC_DECIDE);
932:     VecSetType(gridctx[level1].r,VECMPI);
933:     PCMGSetR(pc,level1,gridctx[level1].r);

935:     if (level == 0) {
936:       PCMGGetCoarseSolve(pc,&gridctx[level].ksp);
937:     } else {
938:       PCMGGetSmoother(pc,level,&gridctx[level].ksp);
939:     }
940:   }
941:   PCMGGetSmoother(pc,fine_level,&gridctx[fine_level].ksp);

943:   /* create coarse level and the interpolation between the levels */
944:   for (level=0; level<fine_level; level++) {
945:     level1 = level + 1;
946:     PCMGSetInterpolation(pc,level1,gridctx[level].P);
947:     PCMGSetRestriction(pc,level1,gridctx[level].R);
948:     if (level > 0) {
949:       PCMGSetResidual(pc,level,PCMGResidual_Default,gridctx[level].A);
950:     }
951:     KSPSetOperators(gridctx[level].ksp,gridctx[level].A,gridctx[level].A,DIFFERENT_NONZERO_PATTERN);
952:   }
953:   PCMGSetResidual(pc,fine_level,PCMGResidual_Default,gridctx[fine_level].A);
954:   KSPSetOperators(gridctx[fine_level].ksp,gridctx[level].A,gridctx[fine_level].A,DIFFERENT_NONZERO_PATTERN);

956:   /* put coordinate info in levels */
957:   if (pc_ml->dim) {
958:     PetscInt  i,j,dim = pc_ml->dim;
959:     PetscInt  bs, nloc;
960:     PC        subpc;
961:     PetscReal *array;

963:     level = fine_level;
964:     for (mllevel = 0; mllevel < Nlevels; mllevel++) {
965:       ML_Aggregate_Viz_Stats *grid_info = (ML_Aggregate_Viz_Stats*)ml_object->Amat[mllevel].to->Grid->Grid;
966:       MPI_Comm               comm       = ((PetscObject)gridctx[level].A)->comm;

968:       MatGetBlockSize (gridctx[level].A, &bs);
969:       MatGetLocalSize (gridctx[level].A, NULL, &nloc);
970:       nloc /= bs; /* number of local nodes */

972:       VecCreate(comm,&gridctx[level].coords);
973:       VecSetSizes(gridctx[level].coords,dim * nloc,PETSC_DECIDE);
974:       VecSetType(gridctx[level].coords,VECMPI);
975:       VecGetArray(gridctx[level].coords,&array);
976:       for (j = 0; j < nloc; j++) {
977:         for (i = 0; i < dim; i++) {
978:           switch (i) {
979:           case 0: array[dim * j + i] = grid_info->x[j]; break;
980:           case 1: array[dim * j + i] = grid_info->y[j]; break;
981:           case 2: array[dim * j + i] = grid_info->z[j]; break;
982:           default: SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_ARG_SIZ,"PCML coordinate dimension must be <= 3");
983:           }
984:         }
985:       }

987:       /* passing coordinates to smoothers/coarse solver, should they need them */
988:       KSPGetPC(gridctx[level].ksp,&subpc);
989:       PCSetCoordinates(subpc,dim,nloc,array);
990:       VecRestoreArray(gridctx[level].coords,&array);
991:       level--;
992:     }
993:   }

995:   /* setupcalled is set to 0 so that MG is setup from scratch */
996:   pc->setupcalled = 0;
997:   PCSetUp_MG(pc);
998:   return(0);
999: }

1001: /* -------------------------------------------------------------------------- */
1002: /*
1003:    PCDestroy_ML - Destroys the private context for the ML preconditioner
1004:    that was created with PCCreate_ML().

1006:    Input Parameter:
1007: .  pc - the preconditioner context

1009:    Application Interface Routine: PCDestroy()
1010: */
1013: PetscErrorCode PCDestroy_ML(PC pc)
1014: {
1016:   PC_MG          *mg   = (PC_MG*)pc->data;
1017:   PC_ML          *pc_ml= (PC_ML*)mg->innerctx;

1020:   PCReset_ML(pc);
1021:   PetscFree(pc_ml);
1022:   PCDestroy_MG(pc);
1023:   PetscObjectComposeFunction((PetscObject)pc,"PCSetCoordinates_C",NULL);
1024:   return(0);
1025: }

1029: PetscErrorCode PCSetFromOptions_ML(PC pc)
1030: {
1032:   PetscInt       indx,PrintLevel,partindx;
1033:   const char     *scheme[] = {"Uncoupled","Coupled","MIS","METIS"};
1034:   const char     *part[]   = {"Zoltan","ParMETIS"};
1035: #if defined(HAVE_ML_ZOLTAN)
1036:   PetscInt   zidx;
1037:   const char *zscheme[] = {"RCB","hypergraph","fast_hypergraph"};
1038: #endif
1039:   PC_MG       *mg    = (PC_MG*)pc->data;
1040:   PC_ML       *pc_ml = (PC_ML*)mg->innerctx;
1041:   PetscMPIInt size;
1042:   MPI_Comm    comm;

1045:   PetscObjectGetComm((PetscObject)pc,&comm);
1046:   MPI_Comm_size(comm,&size);
1047:   PetscOptionsHead("ML options");

1049:   PrintLevel = 0;
1050:   indx       = 0;
1051:   partindx   = 0;

1053:   PetscOptionsInt("-pc_ml_PrintLevel","Print level","ML_Set_PrintLevel",PrintLevel,&PrintLevel,NULL);
1054:   PetscStackCall("ML_Set_PrintLeve",ML_Set_PrintLevel(PrintLevel));
1055:   PetscOptionsInt("-pc_ml_maxNlevels","Maximum number of levels","None",pc_ml->MaxNlevels,&pc_ml->MaxNlevels,NULL);
1056:   PetscOptionsInt("-pc_ml_maxCoarseSize","Maximum coarsest mesh size","ML_Aggregate_Set_MaxCoarseSize",pc_ml->MaxCoarseSize,&pc_ml->MaxCoarseSize,NULL);
1057:   PetscOptionsEList("-pc_ml_CoarsenScheme","Aggregate Coarsen Scheme","ML_Aggregate_Set_CoarsenScheme_*",scheme,4,scheme[0],&indx,NULL);

1059:   pc_ml->CoarsenScheme = indx;

1061:   PetscOptionsReal("-pc_ml_DampingFactor","P damping factor","ML_Aggregate_Set_DampingFactor",pc_ml->DampingFactor,&pc_ml->DampingFactor,NULL);
1062:   PetscOptionsReal("-pc_ml_Threshold","Smoother drop tol","ML_Aggregate_Set_Threshold",pc_ml->Threshold,&pc_ml->Threshold,NULL);
1063:   PetscOptionsBool("-pc_ml_SpectralNormScheme_Anorm","Method used for estimating spectral radius","ML_Set_SpectralNormScheme_Anorm",pc_ml->SpectralNormScheme_Anorm,&pc_ml->SpectralNormScheme_Anorm,NULL);
1064:   PetscOptionsBool("-pc_ml_Symmetrize","Symmetrize aggregation","ML_Set_Symmetrize",pc_ml->Symmetrize,&pc_ml->Symmetrize,NULL);
1065:   PetscOptionsBool("-pc_ml_BlockScaling","Scale all dofs at each node together","None",pc_ml->BlockScaling,&pc_ml->BlockScaling,NULL);
1066:   PetscOptionsEnum("-pc_ml_nullspace","Which type of null space information to use","None",PCMLNullSpaceTypes,(PetscEnum)pc_ml->nulltype,(PetscEnum*)&pc_ml->nulltype,NULL);
1067:   PetscOptionsInt("-pc_ml_EnergyMinimization","Energy minimization norm type (0=no minimization; see ML manual for 1,2,3; -1 and 4 undocumented)","None",pc_ml->EnergyMinimization,&pc_ml->EnergyMinimization,NULL);
1068:   PetscOptionsBool("-pc_ml_reuse_interpolation","Reuse the interpolation operators when possible (cheaper, weaker when matrix entries change a lot)","None",pc_ml->reuse_interpolation,&pc_ml->reuse_interpolation,NULL);
1069:   /*
1070:     The following checks a number of conditions.  If we let this stuff slip by, then ML's error handling will take over.
1071:     This is suboptimal because it amounts to calling exit(1) so we check for the most common conditions.

1073:     We also try to set some sane defaults when energy minimization is activated, otherwise it's hard to find a working
1074:     combination of options and ML's exit(1) explanations don't help matters.
1075:   */
1076:   if (pc_ml->EnergyMinimization < -1 || pc_ml->EnergyMinimization > 4) SETERRQ(comm,PETSC_ERR_ARG_OUTOFRANGE,"EnergyMinimization must be in range -1..4");
1077:   if (pc_ml->EnergyMinimization == 4 && size > 1) SETERRQ(comm,PETSC_ERR_SUP,"Energy minimization type 4 does not work in parallel");
1078:   if (pc_ml->EnergyMinimization == 4) {PetscInfo(pc,"Mandel's energy minimization scheme is experimental and broken in ML-6.2");}
1079:   if (pc_ml->EnergyMinimization) {
1080:     PetscOptionsReal("-pc_ml_EnergyMinimizationDropTol","Energy minimization drop tolerance","None",pc_ml->EnergyMinimizationDropTol,&pc_ml->EnergyMinimizationDropTol,NULL);
1081:   }
1082:   if (pc_ml->EnergyMinimization == 2) {
1083:     /* According to ml_MultiLevelPreconditioner.cpp, this option is only meaningful for norm type (2) */
1084:     PetscOptionsBool("-pc_ml_EnergyMinimizationCheap","Use cheaper variant of norm type 2","None",pc_ml->EnergyMinimizationCheap,&pc_ml->EnergyMinimizationCheap,NULL);
1085:   }
1086:   /* energy minimization sometimes breaks if this is turned off, the more classical stuff should be okay without it */
1087:   if (pc_ml->EnergyMinimization) pc_ml->KeepAggInfo = PETSC_TRUE;
1088:   PetscOptionsBool("-pc_ml_KeepAggInfo","Allows the preconditioner to be reused, or auxilliary matrices to be generated","None",pc_ml->KeepAggInfo,&pc_ml->KeepAggInfo,NULL);
1089:   /* Option (-1) doesn't work at all (calls exit(1)) if the tentative restriction operator isn't stored. */
1090:   if (pc_ml->EnergyMinimization == -1) pc_ml->Reusable = PETSC_TRUE;
1091:   PetscOptionsBool("-pc_ml_Reusable","Store intermedaiate data structures so that the multilevel hierarchy is reusable","None",pc_ml->Reusable,&pc_ml->Reusable,NULL);
1092:   /*
1093:     ML's C API is severely underdocumented and lacks significant functionality.  The C++ API calls
1094:     ML_Gen_MultiLevelHierarchy_UsingAggregation() which is a modified copy (!?) of the documented function
1095:     ML_Gen_MGHierarchy_UsingAggregation().  This modification, however, does not provide a strict superset of the
1096:     functionality in the old function, so some users may still want to use it.  Note that many options are ignored in
1097:     this context, but ML doesn't provide a way to find out which ones.
1098:    */
1099:   PetscOptionsBool("-pc_ml_OldHierarchy","Use old routine to generate hierarchy","None",pc_ml->OldHierarchy,&pc_ml->OldHierarchy,NULL);
1100:   PetscOptionsBool("-pc_ml_repartition", "Allow ML to repartition levels of the heirarchy","ML_Repartition_Activate",pc_ml->Repartition,&pc_ml->Repartition,NULL);
1101:   if (pc_ml->Repartition) {
1102:     PetscOptionsReal("-pc_ml_repartitionMaxMinRatio", "Acceptable ratio of repartitioned sizes","ML_Repartition_Set_LargestMinMaxRatio",pc_ml->MaxMinRatio,&pc_ml->MaxMinRatio,NULL);
1103:     PetscOptionsInt("-pc_ml_repartitionMinPerProc", "Smallest repartitioned size","ML_Repartition_Set_MinPerProc",pc_ml->MinPerProc,&pc_ml->MinPerProc,NULL);
1104:     PetscOptionsInt("-pc_ml_repartitionPutOnSingleProc", "Problem size automatically repartitioned to one processor","ML_Repartition_Set_PutOnSingleProc",pc_ml->PutOnSingleProc,&pc_ml->PutOnSingleProc,NULL);
1105: #if defined(HAVE_ML_ZOLTAN)
1106:     partindx = 0;
1107:     PetscOptionsEList("-pc_ml_repartitionType", "Repartitioning library to use","ML_Repartition_Set_Partitioner",part,2,part[0],&partindx,NULL);

1109:     pc_ml->RepartitionType = partindx;
1110:     if (!partindx) {
1111:       PetscInt zindx = 0;

1113:       PetscOptionsEList("-pc_ml_repartitionZoltanScheme", "Repartitioning scheme to use","None",zscheme,3,zscheme[0],&zindx,NULL);

1115:       pc_ml->ZoltanScheme = zindx;
1116:     }
1117: #else
1118:     partindx = 1;
1119:     PetscOptionsEList("-pc_ml_repartitionType", "Repartitioning library to use","ML_Repartition_Set_Partitioner",part,2,part[1],&partindx,NULL);
1120:     if (!partindx) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_SUP_SYS,"ML not compiled with Zoltan");
1121: #endif
1122:     PetscOptionsBool("-pc_ml_Aux","Aggregate using auxiliary coordinate-based laplacian","None",pc_ml->Aux,&pc_ml->Aux,NULL);
1123:     PetscOptionsReal("-pc_ml_AuxThreshold","Auxiliary smoother drop tol","None",pc_ml->AuxThreshold,&pc_ml->AuxThreshold,NULL);
1124:   }
1125:   PetscOptionsTail();
1126:   return(0);
1127: }

1129: /* -------------------------------------------------------------------------- */
1130: /*
1131:    PCCreate_ML - Creates a ML preconditioner context, PC_ML,
1132:    and sets this as the private data within the generic preconditioning
1133:    context, PC, that was created within PCCreate().

1135:    Input Parameter:
1136: .  pc - the preconditioner context

1138:    Application Interface Routine: PCCreate()
1139: */

1141: /*MC
1142:      PCML - Use algebraic multigrid preconditioning. This preconditioner requires you provide
1143:        fine grid discretization matrix. The coarser grid matrices and restriction/interpolation
1144:        operators are computed by ML, with the matrices coverted to PETSc matrices in aij format
1145:        and the restriction/interpolation operators wrapped as PETSc shell matrices.

1147:    Options Database Key:
1148:    Multigrid options(inherited)
1149: +  -pc_mg_cycles <1>: 1 for V cycle, 2 for W-cycle (MGSetCycles)
1150: .  -pc_mg_smoothup <1>: Number of post-smoothing steps (MGSetNumberSmoothUp)
1151: .  -pc_mg_smoothdown <1>: Number of pre-smoothing steps (MGSetNumberSmoothDown)
1152:    -pc_mg_type <multiplicative>: (one of) additive multiplicative full kascade
1153:    ML options:
1154: .  -pc_ml_PrintLevel <0>: Print level (ML_Set_PrintLevel)
1155: .  -pc_ml_maxNlevels <10>: Maximum number of levels (None)
1156: .  -pc_ml_maxCoarseSize <1>: Maximum coarsest mesh size (ML_Aggregate_Set_MaxCoarseSize)
1157: .  -pc_ml_CoarsenScheme <Uncoupled>: (one of) Uncoupled Coupled MIS METIS
1158: .  -pc_ml_DampingFactor <1.33333>: P damping factor (ML_Aggregate_Set_DampingFactor)
1159: .  -pc_ml_Threshold <0>: Smoother drop tol (ML_Aggregate_Set_Threshold)
1160: .  -pc_ml_SpectralNormScheme_Anorm <false>: Method used for estimating spectral radius (ML_Set_SpectralNormScheme_Anorm)
1161: .  -pc_ml_repartition <false>: Allow ML to repartition levels of the heirarchy (ML_Repartition_Activate)
1162: .  -pc_ml_repartitionMaxMinRatio <1.3>: Acceptable ratio of repartitioned sizes (ML_Repartition_Set_LargestMinMaxRatio)
1163: .  -pc_ml_repartitionMinPerProc <512>: Smallest repartitioned size (ML_Repartition_Set_MinPerProc)
1164: .  -pc_ml_repartitionPutOnSingleProc <5000>: Problem size automatically repartitioned to one processor (ML_Repartition_Set_PutOnSingleProc)
1165: .  -pc_ml_repartitionType <Zoltan>: Repartitioning library to use (ML_Repartition_Set_Partitioner)
1166: .  -pc_ml_repartitionZoltanScheme <RCB>: Repartitioning scheme to use (None)
1167: .  -pc_ml_Aux <false>: Aggregate using auxiliary coordinate-based laplacian (None)
1168: -  -pc_ml_AuxThreshold <0.0>: Auxiliary smoother drop tol (None)

1170:    Level: intermediate

1172:   Concepts: multigrid

1174: .seealso:  PCCreate(), PCSetType(), PCType (for list of available types), PC, PCMGType,
1175:            PCMGSetLevels(), PCMGGetLevels(), PCMGSetType(), MPSetCycles(), PCMGSetNumberSmoothDown(),
1176:            PCMGSetNumberSmoothUp(), PCMGGetCoarseSolve(), PCMGSetResidual(), PCMGSetInterpolation(),
1177:            PCMGSetRestriction(), PCMGGetSmoother(), PCMGGetSmootherUp(), PCMGGetSmootherDown(),
1178:            PCMGSetCyclesOnLevel(), PCMGSetRhs(), PCMGSetX(), PCMGSetR()
1179: M*/

1183: PETSC_EXTERN PetscErrorCode PCCreate_ML(PC pc)
1184: {
1186:   PC_ML          *pc_ml;
1187:   PC_MG          *mg;

1190:   /* PCML is an inherited class of PCMG. Initialize pc as PCMG */
1191:   PCSetType(pc,PCMG); /* calls PCCreate_MG() and MGCreate_Private() */
1192:   PetscObjectChangeTypeName((PetscObject)pc,PCML);
1193:   /* Since PCMG tries to use DM assocated with PC must delete it */
1194:   DMDestroy(&pc->dm);
1195:   mg           = (PC_MG*)pc->data;
1196:   mg->galerkin = 2;             /* Use Galerkin, but it is computed externally */

1198:   /* create a supporting struct and attach it to pc */
1199:   PetscNewLog(pc,PC_ML,&pc_ml);
1200:   mg->innerctx = pc_ml;

1202:   pc_ml->ml_object                = 0;
1203:   pc_ml->agg_object               = 0;
1204:   pc_ml->gridctx                  = 0;
1205:   pc_ml->PetscMLdata              = 0;
1206:   pc_ml->Nlevels                  = -1;
1207:   pc_ml->MaxNlevels               = 10;
1208:   pc_ml->MaxCoarseSize            = 1;
1209:   pc_ml->CoarsenScheme            = 1;
1210:   pc_ml->Threshold                = 0.0;
1211:   pc_ml->DampingFactor            = 4.0/3.0;
1212:   pc_ml->SpectralNormScheme_Anorm = PETSC_FALSE;
1213:   pc_ml->size                     = 0;
1214:   pc_ml->dim                      = 0;
1215:   pc_ml->nloc                     = 0;
1216:   pc_ml->coords                   = 0;
1217:   pc_ml->Repartition              = PETSC_FALSE;
1218:   pc_ml->MaxMinRatio              = 1.3;
1219:   pc_ml->MinPerProc               = 512;
1220:   pc_ml->PutOnSingleProc          = 5000;
1221:   pc_ml->RepartitionType          = 0;
1222:   pc_ml->ZoltanScheme             = 0;
1223:   pc_ml->Aux                      = PETSC_FALSE;
1224:   pc_ml->AuxThreshold             = 0.0;

1226:   /* allow for coordinates to be passed */
1227:   PetscObjectComposeFunction((PetscObject)pc,"PCSetCoordinates_C",PCSetCoordinates_ML);

1229:   /* overwrite the pointers of PCMG by the functions of PCML */
1230:   pc->ops->setfromoptions = PCSetFromOptions_ML;
1231:   pc->ops->setup          = PCSetUp_ML;
1232:   pc->ops->reset          = PCReset_ML;
1233:   pc->ops->destroy        = PCDestroy_ML;
1234:   return(0);
1235: }