Actual source code: ml.c
petsc-3.7.3 2016-08-01
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 <petsc/private/pcmgimpl.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: {
97: PetscErrorCode ierr;
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: const 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: VecGetArrayRead(a->lvec,&array);
114: for (i=in_length; i<out_length; i++) p[i] = array[i-in_length];
115: VecRestoreArrayRead(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: {
150: PetscErrorCode ierr;
151: Mat_MLShell *shell;
152: PetscScalar *yarray;
153: const PetscScalar *xarray;
154: PetscInt x_length,y_length;
157: MatShellGetContext(A,(void**)&shell);
158: VecGetArrayRead(x,&xarray);
159: VecGetArray(y,&yarray);
160: x_length = shell->mlmat->invec_leng;
161: y_length = shell->mlmat->outvec_leng;
162: PetscStackCall("ML_Operator_Apply",ML_Operator_Apply(shell->mlmat,x_length,(PetscScalar*)xarray,y_length,yarray));
163: VecRestoreArrayRead(x,&xarray);
164: VecRestoreArray(y,&yarray);
165: return(0);
166: }
170: /* Computes y = w + A * x
171: It is possible that w == y, but not x == y
172: */
173: static PetscErrorCode MatMultAdd_ML(Mat A,Vec x,Vec w,Vec y)
174: {
175: Mat_MLShell *shell;
176: PetscScalar *yarray;
177: const PetscScalar *xarray;
178: PetscInt x_length,y_length;
179: PetscErrorCode ierr;
182: MatShellGetContext(A, (void**) &shell);
183: if (y == w) {
184: if (!shell->work) {
185: VecDuplicate(y, &shell->work);
186: }
187: VecGetArrayRead(x, &xarray);
188: VecGetArray(shell->work, &yarray);
189: x_length = shell->mlmat->invec_leng;
190: y_length = shell->mlmat->outvec_leng;
191: PetscStackCall("ML_Operator_Apply",ML_Operator_Apply(shell->mlmat, x_length, (PetscScalar*)xarray, y_length, yarray));
192: VecRestoreArrayRead(x, &xarray);
193: VecRestoreArray(shell->work, &yarray);
194: VecAXPY(y, 1.0, shell->work);
195: } else {
196: VecGetArrayRead(x, &xarray);
197: VecGetArray(y, &yarray);
198: x_length = shell->mlmat->invec_leng;
199: y_length = shell->mlmat->outvec_leng;
200: PetscStackCall("ML_Operator_Apply",ML_Operator_Apply(shell->mlmat, x_length, (PetscScalar *)xarray, y_length, yarray));
201: VecRestoreArrayRead(x, &xarray);
202: VecRestoreArray(y, &yarray);
203: VecAXPY(y, 1.0, w);
204: }
205: return(0);
206: }
208: /* newtype is ignored since only handles one case */
211: static PetscErrorCode MatConvert_MPIAIJ_ML(Mat A,MatType newtype,MatReuse scall,Mat *Aloc)
212: {
214: Mat_MPIAIJ *mpimat=(Mat_MPIAIJ*)A->data;
215: Mat_SeqAIJ *mat,*a=(Mat_SeqAIJ*)(mpimat->A)->data,*b=(Mat_SeqAIJ*)(mpimat->B)->data;
216: PetscInt *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j;
217: PetscScalar *aa=a->a,*ba=b->a,*ca;
218: PetscInt am =A->rmap->n,an=A->cmap->n,i,j,k;
219: PetscInt *ci,*cj,ncols;
222: if (am != an) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"A must have a square diagonal portion, am: %d != an: %d",am,an);
224: if (scall == MAT_INITIAL_MATRIX) {
225: PetscMalloc1(1+am,&ci);
226: ci[0] = 0;
227: for (i=0; i<am; i++) ci[i+1] = ci[i] + (ai[i+1] - ai[i]) + (bi[i+1] - bi[i]);
228: PetscMalloc1(1+ci[am],&cj);
229: PetscMalloc1(1+ci[am],&ca);
231: k = 0;
232: for (i=0; i<am; i++) {
233: /* diagonal portion of A */
234: ncols = ai[i+1] - ai[i];
235: for (j=0; j<ncols; j++) {
236: cj[k] = *aj++;
237: ca[k++] = *aa++;
238: }
239: /* off-diagonal portion of A */
240: ncols = bi[i+1] - bi[i];
241: for (j=0; j<ncols; j++) {
242: cj[k] = an + (*bj); bj++;
243: ca[k++] = *ba++;
244: }
245: }
246: if (k != ci[am]) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"k: %d != ci[am]: %d",k,ci[am]);
248: /* put together the new matrix */
249: an = mpimat->A->cmap->n+mpimat->B->cmap->n;
250: MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,am,an,ci,cj,ca,Aloc);
252: /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */
253: /* Since these are PETSc arrays, change flags to free them as necessary. */
254: mat = (Mat_SeqAIJ*)(*Aloc)->data;
255: mat->free_a = PETSC_TRUE;
256: mat->free_ij = PETSC_TRUE;
258: mat->nonew = 0;
259: } else if (scall == MAT_REUSE_MATRIX) {
260: mat=(Mat_SeqAIJ*)(*Aloc)->data;
261: ci = mat->i; cj = mat->j; ca = mat->a;
262: for (i=0; i<am; i++) {
263: /* diagonal portion of A */
264: ncols = ai[i+1] - ai[i];
265: for (j=0; j<ncols; j++) *ca++ = *aa++;
266: /* off-diagonal portion of A */
267: ncols = bi[i+1] - bi[i];
268: for (j=0; j<ncols; j++) *ca++ = *ba++;
269: }
270: } else SETERRQ1(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Invalid MatReuse %d",(int)scall);
271: return(0);
272: }
276: static PetscErrorCode MatDestroy_ML(Mat A)
277: {
279: Mat_MLShell *shell;
282: MatShellGetContext(A,(void**)&shell);
283: VecDestroy(&shell->y);
284: if (shell->work) {VecDestroy(&shell->work);}
285: PetscFree(shell);
286: return(0);
287: }
291: static PetscErrorCode MatWrapML_SeqAIJ(ML_Operator *mlmat,MatReuse reuse,Mat *newmat)
292: {
293: struct ML_CSR_MSRdata *matdata = (struct ML_CSR_MSRdata*)mlmat->data;
294: PetscErrorCode ierr;
295: PetscInt m =mlmat->outvec_leng,n=mlmat->invec_leng,*nnz = NULL,nz_max;
296: PetscInt *ml_cols=matdata->columns,*ml_rowptr=matdata->rowptr,*aj,i;
297: PetscScalar *ml_vals=matdata->values,*aa;
300: if (!mlmat->getrow) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_NULL,"mlmat->getrow = NULL");
301: if (m != n) { /* ML Pmat and Rmat are in CSR format. Pass array pointers into SeqAIJ matrix */
302: if (reuse) {
303: Mat_SeqAIJ *aij= (Mat_SeqAIJ*)(*newmat)->data;
304: aij->i = ml_rowptr;
305: aij->j = ml_cols;
306: aij->a = ml_vals;
307: } else {
308: /* sort ml_cols and ml_vals */
309: PetscMalloc1(m+1,&nnz);
310: for (i=0; i<m; i++) nnz[i] = ml_rowptr[i+1] - ml_rowptr[i];
311: aj = ml_cols; aa = ml_vals;
312: for (i=0; i<m; i++) {
313: PetscSortIntWithScalarArray(nnz[i],aj,aa);
314: aj += nnz[i]; aa += nnz[i];
315: }
316: MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,m,n,ml_rowptr,ml_cols,ml_vals,newmat);
317: PetscFree(nnz);
318: }
319: return(0);
320: }
322: nz_max = PetscMax(1,mlmat->max_nz_per_row);
323: PetscMalloc2(nz_max,&aa,nz_max,&aj);
324: if (!reuse) {
325: MatCreate(PETSC_COMM_SELF,newmat);
326: MatSetSizes(*newmat,m,n,PETSC_DECIDE,PETSC_DECIDE);
327: MatSetType(*newmat,MATSEQAIJ);
328: /* keep track of block size for A matrices */
329: MatSetBlockSize (*newmat, mlmat->num_PDEs);
331: PetscMalloc1(m,&nnz);
332: for (i=0; i<m; i++) {
333: PetscStackCall("ML_Operator_Getrow",ML_Operator_Getrow(mlmat,1,&i,nz_max,aj,aa,&nnz[i]));
334: }
335: MatSeqAIJSetPreallocation(*newmat,0,nnz);
336: }
337: for (i=0; i<m; i++) {
338: PetscInt ncols;
340: PetscStackCall("ML_Operator_Getrow",ML_Operator_Getrow(mlmat,1,&i,nz_max,aj,aa,&ncols));
341: MatSetValues(*newmat,1,&i,ncols,aj,aa,INSERT_VALUES);
342: }
343: MatAssemblyBegin(*newmat,MAT_FINAL_ASSEMBLY);
344: MatAssemblyEnd(*newmat,MAT_FINAL_ASSEMBLY);
346: PetscFree2(aa,aj);
347: PetscFree(nnz);
348: return(0);
349: }
353: static PetscErrorCode MatWrapML_SHELL(ML_Operator *mlmat,MatReuse reuse,Mat *newmat)
354: {
356: PetscInt m,n;
357: ML_Comm *MLcomm;
358: Mat_MLShell *shellctx;
361: m = mlmat->outvec_leng;
362: n = mlmat->invec_leng;
364: if (reuse) {
365: MatShellGetContext(*newmat,(void**)&shellctx);
366: shellctx->mlmat = mlmat;
367: return(0);
368: }
370: MLcomm = mlmat->comm;
372: PetscNew(&shellctx);
373: MatCreateShell(MLcomm->USR_comm,m,n,PETSC_DETERMINE,PETSC_DETERMINE,shellctx,newmat);
374: MatShellSetOperation(*newmat,MATOP_MULT,(void(*)(void))MatMult_ML);
375: MatShellSetOperation(*newmat,MATOP_MULT_ADD,(void(*)(void))MatMultAdd_ML);
376: MatShellSetOperation(*newmat,MATOP_DESTROY,(void(*)(void))MatDestroy_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: VecSetType(shellctx->y,VECSTANDARD);
385: return(0);
386: }
390: static PetscErrorCode MatWrapML_MPIAIJ(ML_Operator *mlmat,MatReuse reuse,Mat *newmat)
391: {
392: PetscInt *aj;
393: PetscScalar *aa;
395: PetscInt i,j,*gordering;
396: PetscInt m=mlmat->outvec_leng,n,nz_max,row;
397: Mat A;
400: if (!mlmat->getrow) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_NULL,"mlmat->getrow = NULL");
401: n = mlmat->invec_leng;
402: if (m != n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"m %d must equal to n %d",m,n);
404: /* create global row numbering for a ML_Operator */
405: PetscStackCall("ML_build_global_numbering",ML_build_global_numbering(mlmat,&gordering,"rows"));
407: nz_max = PetscMax(1,mlmat->max_nz_per_row) + 1;
408: PetscMalloc2(nz_max,&aa,nz_max,&aj);
409: if (reuse) {
410: A = *newmat;
411: } else {
412: PetscInt *nnzA,*nnzB,*nnz;
413: PetscInt rstart;
414: MatCreate(mlmat->comm->USR_comm,&A);
415: MatSetSizes(A,m,n,PETSC_DECIDE,PETSC_DECIDE);
416: MatSetType(A,MATMPIAIJ);
417: /* keep track of block size for A matrices */
418: MatSetBlockSize (A,mlmat->num_PDEs);
419: PetscMalloc3(m,&nnzA,m,&nnzB,m,&nnz);
420: MPI_Scan(&m,&rstart,1,MPIU_INT,MPI_SUM,mlmat->comm->USR_comm);
421: rstart -= m;
423: for (i=0; i<m; i++) {
424: row = gordering[i] - rstart;
425: PetscStackCall("ML_Operator_Getrow",ML_Operator_Getrow(mlmat,1,&i,nz_max,aj,aa,&nnz[i]));
426: nnzA[row] = 0;
427: for (j=0; j<nnz[i]; j++) {
428: if (aj[j] < m) nnzA[row]++;
429: }
430: nnzB[row] = nnz[i] - nnzA[row];
431: }
432: MatMPIAIJSetPreallocation(A,0,nnzA,0,nnzB);
433: PetscFree3(nnzA,nnzB,nnz);
434: }
435: for (i=0; i<m; i++) {
436: PetscInt ncols;
437: row = gordering[i];
439: PetscStackCall(",ML_Operator_Getrow",ML_Operator_Getrow(mlmat,1,&i,nz_max,aj,aa,&ncols));
440: for (j = 0; j < ncols; j++) aj[j] = gordering[aj[j]];
441: MatSetValues(A,1,&row,ncols,aj,aa,INSERT_VALUES);
442: }
443: PetscStackCall("ML_free",ML_free(gordering));
444: MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);
445: MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);
446: *newmat = A;
448: PetscFree2(aa,aj);
449: return(0);
450: }
452: /* -------------------------------------------------------------------------- */
453: /*
454: PCSetCoordinates_ML
456: Input Parameter:
457: . pc - the preconditioner context
458: */
461: static PetscErrorCode PCSetCoordinates_ML(PC pc, PetscInt ndm, PetscInt a_nloc, PetscReal *coords)
462: {
463: PC_MG *mg = (PC_MG*)pc->data;
464: PC_ML *pc_ml = (PC_ML*)mg->innerctx;
466: PetscInt arrsz,oldarrsz,bs,my0,kk,ii,nloc,Iend;
467: Mat Amat = pc->pmat;
469: /* this function copied and modified from PCSetCoordinates_GEO -TGI */
472: MatGetBlockSize(Amat, &bs);
474: MatGetOwnershipRange(Amat, &my0, &Iend);
475: nloc = (Iend-my0)/bs;
477: if (nloc!=a_nloc) SETERRQ2(PetscObjectComm((PetscObject)Amat),PETSC_ERR_ARG_WRONG, "Number of local blocks must locations = %d %d.",a_nloc,nloc);
478: if ((Iend-my0)%bs!=0) SETERRQ1(PetscObjectComm((PetscObject)Amat),PETSC_ERR_ARG_WRONG, "Bad local size %d.",nloc);
480: oldarrsz = pc_ml->dim * pc_ml->nloc;
481: pc_ml->dim = ndm;
482: pc_ml->nloc = a_nloc;
483: arrsz = ndm * a_nloc;
485: /* create data - syntactic sugar that should be refactored at some point */
486: if (pc_ml->coords==0 || (oldarrsz != arrsz)) {
487: PetscFree(pc_ml->coords);
488: PetscMalloc1(arrsz, &pc_ml->coords);
489: }
490: for (kk=0; kk<arrsz; kk++) pc_ml->coords[kk] = -999.;
491: /* copy data in - column oriented */
492: for (kk = 0; kk < nloc; kk++) {
493: for (ii = 0; ii < ndm; ii++) {
494: pc_ml->coords[ii*nloc + kk] = coords[kk*ndm + ii];
495: }
496: }
497: return(0);
498: }
500: /* -----------------------------------------------------------------------------*/
501: extern PetscErrorCode PCReset_MG(PC);
504: PetscErrorCode PCReset_ML(PC pc)
505: {
507: PC_MG *mg = (PC_MG*)pc->data;
508: PC_ML *pc_ml = (PC_ML*)mg->innerctx;
509: PetscInt level,fine_level=pc_ml->Nlevels-1,dim=pc_ml->dim;
512: if (dim) {
513: ML_Aggregate_Viz_Stats * grid_info = (ML_Aggregate_Viz_Stats*) pc_ml->ml_object->Grid[0].Grid;
515: for (level=0; level<=fine_level; level++) {
516: VecDestroy(&pc_ml->gridctx[level].coords);
517: }
519: grid_info->x = 0; /* do this so ML doesn't try to free coordinates */
520: grid_info->y = 0;
521: grid_info->z = 0;
523: PetscStackCall("ML_Operator_Getrow",ML_Aggregate_VizAndStats_Clean(pc_ml->ml_object));
524: }
525: PetscStackCall("ML_Aggregate_Destroy",ML_Aggregate_Destroy(&pc_ml->agg_object));
526: PetscStackCall("ML_Aggregate_Destroy",ML_Destroy(&pc_ml->ml_object));
528: if (pc_ml->PetscMLdata) {
529: PetscFree(pc_ml->PetscMLdata->pwork);
530: MatDestroy(&pc_ml->PetscMLdata->Aloc);
531: VecDestroy(&pc_ml->PetscMLdata->x);
532: VecDestroy(&pc_ml->PetscMLdata->y);
533: }
534: PetscFree(pc_ml->PetscMLdata);
536: if (pc_ml->gridctx) {
537: for (level=0; level<fine_level; level++) {
538: if (pc_ml->gridctx[level].A) {MatDestroy(&pc_ml->gridctx[level].A);}
539: if (pc_ml->gridctx[level].P) {MatDestroy(&pc_ml->gridctx[level].P);}
540: if (pc_ml->gridctx[level].R) {MatDestroy(&pc_ml->gridctx[level].R);}
541: if (pc_ml->gridctx[level].x) {VecDestroy(&pc_ml->gridctx[level].x);}
542: if (pc_ml->gridctx[level].b) {VecDestroy(&pc_ml->gridctx[level].b);}
543: if (pc_ml->gridctx[level+1].r) {VecDestroy(&pc_ml->gridctx[level+1].r);}
544: }
545: }
546: PetscFree(pc_ml->gridctx);
547: PetscFree(pc_ml->coords);
549: pc_ml->dim = 0;
550: pc_ml->nloc = 0;
551: PCReset_MG(pc);
552: return(0);
553: }
554: /* -------------------------------------------------------------------------- */
555: /*
556: PCSetUp_ML - Prepares for the use of the ML preconditioner
557: by setting data structures and options.
559: Input Parameter:
560: . pc - the preconditioner context
562: Application Interface Routine: PCSetUp()
564: Notes:
565: The interface routine PCSetUp() is not usually called directly by
566: the user, but instead is called by PCApply() if necessary.
567: */
568: extern PetscErrorCode PCSetFromOptions_MG(PetscOptionItems *PetscOptionsObject,PC);
569: extern PetscErrorCode PCReset_MG(PC);
573: PetscErrorCode PCSetUp_ML(PC pc)
574: {
575: PetscErrorCode ierr;
576: PetscMPIInt size;
577: FineGridCtx *PetscMLdata;
578: ML *ml_object;
579: ML_Aggregate *agg_object;
580: ML_Operator *mlmat;
581: PetscInt nlocal_allcols,Nlevels,mllevel,level,level1,m,fine_level,bs;
582: Mat A,Aloc;
583: GridCtx *gridctx;
584: PC_MG *mg = (PC_MG*)pc->data;
585: PC_ML *pc_ml = (PC_ML*)mg->innerctx;
586: PetscBool isSeq, isMPI;
587: KSP smoother;
588: PC subpc;
589: PetscInt mesh_level, old_mesh_level;
590: MatInfo info;
591: static PetscBool cite = PETSC_FALSE;
594: PetscCitationsRegister("@TechReport{ml_users_guide,\n author = {M. Sala and J.J. Hu and R.S. Tuminaro},\n title = {{ML}3.1 {S}moothed {A}ggregation {U}ser's {G}uide},\n institution = {Sandia National Laboratories},\n number = {SAND2004-4821},\n year = 2004\n}\n",&cite);
595: A = pc->pmat;
596: MPI_Comm_size(PetscObjectComm((PetscObject)A),&size);
598: if (pc->setupcalled) {
599: if (pc->flag == SAME_NONZERO_PATTERN && pc_ml->reuse_interpolation) {
600: /*
601: Reuse interpolaton instead of recomputing aggregates and updating the whole hierarchy. This is less expensive for
602: multiple solves in which the matrix is not changing too quickly.
603: */
604: ml_object = pc_ml->ml_object;
605: gridctx = pc_ml->gridctx;
606: Nlevels = pc_ml->Nlevels;
607: fine_level = Nlevels - 1;
608: gridctx[fine_level].A = A;
610: PetscObjectTypeCompare((PetscObject) A, MATSEQAIJ, &isSeq);
611: PetscObjectTypeCompare((PetscObject) A, MATMPIAIJ, &isMPI);
612: if (isMPI) {
613: MatConvert_MPIAIJ_ML(A,NULL,MAT_INITIAL_MATRIX,&Aloc);
614: } else if (isSeq) {
615: Aloc = A;
616: PetscObjectReference((PetscObject)Aloc);
617: } 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);
619: MatGetSize(Aloc,&m,&nlocal_allcols);
620: PetscMLdata = pc_ml->PetscMLdata;
621: MatDestroy(&PetscMLdata->Aloc);
622: PetscMLdata->A = A;
623: PetscMLdata->Aloc = Aloc;
624: PetscStackCall("ML_Aggregate_Destroy",ML_Init_Amatrix(ml_object,0,m,m,PetscMLdata));
625: PetscStackCall("ML_Set_Amatrix_Matvec",ML_Set_Amatrix_Matvec(ml_object,0,PetscML_matvec));
627: mesh_level = ml_object->ML_finest_level;
628: while (ml_object->SingleLevel[mesh_level].Rmat->to) {
629: old_mesh_level = mesh_level;
630: mesh_level = ml_object->SingleLevel[mesh_level].Rmat->to->levelnum;
632: /* clean and regenerate A */
633: mlmat = &(ml_object->Amat[mesh_level]);
634: PetscStackCall("ML_Operator_Clean",ML_Operator_Clean(mlmat));
635: PetscStackCall("ML_Operator_Init",ML_Operator_Init(mlmat,ml_object->comm));
636: PetscStackCall("ML_Gen_AmatrixRAP",ML_Gen_AmatrixRAP(ml_object, old_mesh_level, mesh_level));
637: }
639: level = fine_level - 1;
640: if (size == 1) { /* convert ML P, R and A into seqaij format */
641: for (mllevel=1; mllevel<Nlevels; mllevel++) {
642: mlmat = &(ml_object->Amat[mllevel]);
643: MatWrapML_SeqAIJ(mlmat,MAT_REUSE_MATRIX,&gridctx[level].A);
644: level--;
645: }
646: } else { /* convert ML P and R into shell format, ML A into mpiaij format */
647: for (mllevel=1; mllevel<Nlevels; mllevel++) {
648: mlmat = &(ml_object->Amat[mllevel]);
649: MatWrapML_MPIAIJ(mlmat,MAT_REUSE_MATRIX,&gridctx[level].A);
650: level--;
651: }
652: }
654: for (level=0; level<fine_level; level++) {
655: if (level > 0) {
656: PCMGSetResidual(pc,level,PCMGResidualDefault,gridctx[level].A);
657: }
658: KSPSetOperators(gridctx[level].ksp,gridctx[level].A,gridctx[level].A);
659: }
660: PCMGSetResidual(pc,fine_level,PCMGResidualDefault,gridctx[fine_level].A);
661: KSPSetOperators(gridctx[fine_level].ksp,gridctx[level].A,gridctx[fine_level].A);
663: PCSetUp_MG(pc);
664: return(0);
665: } else {
666: /* since ML can change the size of vectors/matrices at any level we must destroy everything */
667: PCReset_ML(pc);
668: }
669: }
671: /* setup special features of PCML */
672: /*--------------------------------*/
673: /* covert A to Aloc to be used by ML at fine grid */
674: pc_ml->size = size;
675: PetscObjectTypeCompare((PetscObject) A, MATSEQAIJ, &isSeq);
676: PetscObjectTypeCompare((PetscObject) A, MATMPIAIJ, &isMPI);
677: if (isMPI) {
678: MatConvert_MPIAIJ_ML(A,NULL,MAT_INITIAL_MATRIX,&Aloc);
679: } else if (isSeq) {
680: Aloc = A;
681: PetscObjectReference((PetscObject)Aloc);
682: } 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);
684: /* create and initialize struct 'PetscMLdata' */
685: PetscNewLog(pc,&PetscMLdata);
686: pc_ml->PetscMLdata = PetscMLdata;
687: PetscMalloc1(Aloc->cmap->n+1,&PetscMLdata->pwork);
689: VecCreate(PETSC_COMM_SELF,&PetscMLdata->x);
690: VecSetSizes(PetscMLdata->x,Aloc->cmap->n,Aloc->cmap->n);
691: VecSetType(PetscMLdata->x,VECSEQ);
693: VecCreate(PETSC_COMM_SELF,&PetscMLdata->y);
694: VecSetSizes(PetscMLdata->y,A->rmap->n,PETSC_DECIDE);
695: VecSetType(PetscMLdata->y,VECSEQ);
696: PetscMLdata->A = A;
697: PetscMLdata->Aloc = Aloc;
698: if (pc_ml->dim) { /* create vecs around the coordinate data given */
699: PetscInt i,j,dim=pc_ml->dim;
700: PetscInt nloc = pc_ml->nloc,nlocghost;
701: PetscReal *ghostedcoords;
703: MatGetBlockSize(A,&bs);
704: nlocghost = Aloc->cmap->n / bs;
705: PetscMalloc1(dim*nlocghost,&ghostedcoords);
706: for (i = 0; i < dim; i++) {
707: /* copy coordinate values into first component of pwork */
708: for (j = 0; j < nloc; j++) {
709: PetscMLdata->pwork[bs * j] = pc_ml->coords[nloc * i + j];
710: }
711: /* get the ghost values */
712: PetscML_comm(PetscMLdata->pwork,PetscMLdata);
713: /* write into the vector */
714: for (j = 0; j < nlocghost; j++) {
715: ghostedcoords[i * nlocghost + j] = PetscMLdata->pwork[bs * j];
716: }
717: }
718: /* replace the original coords with the ghosted coords, because these are
719: * what ML needs */
720: PetscFree(pc_ml->coords);
721: pc_ml->coords = ghostedcoords;
722: }
724: /* create ML discretization matrix at fine grid */
725: /* ML requires input of fine-grid matrix. It determines nlevels. */
726: MatGetSize(Aloc,&m,&nlocal_allcols);
727: MatGetBlockSize(A,&bs);
728: PetscStackCall("ML_Create",ML_Create(&ml_object,pc_ml->MaxNlevels));
729: PetscStackCall("ML_Comm_Set_UsrComm",ML_Comm_Set_UsrComm(ml_object->comm,PetscObjectComm((PetscObject)A)));
730: pc_ml->ml_object = ml_object;
731: PetscStackCall("ML_Init_Amatrix",ML_Init_Amatrix(ml_object,0,m,m,PetscMLdata));
732: PetscStackCall("ML_Set_Amatrix_Getrow",ML_Set_Amatrix_Getrow(ml_object,0,PetscML_getrow,PetscML_comm,nlocal_allcols));
733: PetscStackCall("ML_Set_Amatrix_Matvec",ML_Set_Amatrix_Matvec(ml_object,0,PetscML_matvec));
735: PetscStackCall("ML_Set_Symmetrize",ML_Set_Symmetrize(ml_object,pc_ml->Symmetrize ? ML_YES : ML_NO));
737: /* aggregation */
738: PetscStackCall("ML_Aggregate_Create",ML_Aggregate_Create(&agg_object));
739: pc_ml->agg_object = agg_object;
741: {
742: MatNullSpace mnull;
743: MatGetNearNullSpace(A,&mnull);
744: if (pc_ml->nulltype == PCML_NULLSPACE_AUTO) {
745: if (mnull) pc_ml->nulltype = PCML_NULLSPACE_USER;
746: else if (bs > 1) pc_ml->nulltype = PCML_NULLSPACE_BLOCK;
747: else pc_ml->nulltype = PCML_NULLSPACE_SCALAR;
748: }
749: switch (pc_ml->nulltype) {
750: case PCML_NULLSPACE_USER: {
751: PetscScalar *nullvec;
752: const PetscScalar *v;
753: PetscBool has_const;
754: PetscInt i,j,mlocal,nvec,M;
755: const Vec *vecs;
757: if (!mnull) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_USER,"Must provide explicit null space using MatSetNearNullSpace() to use user-specified null space");
758: MatGetSize(A,&M,NULL);
759: MatGetLocalSize(Aloc,&mlocal,NULL);
760: MatNullSpaceGetVecs(mnull,&has_const,&nvec,&vecs);
761: PetscMalloc1((nvec+!!has_const)*mlocal,&nullvec);
762: if (has_const) for (i=0; i<mlocal; i++) nullvec[i] = 1.0/M;
763: for (i=0; i<nvec; i++) {
764: VecGetArrayRead(vecs[i],&v);
765: for (j=0; j<mlocal; j++) nullvec[(i+!!has_const)*mlocal + j] = v[j];
766: VecRestoreArrayRead(vecs[i],&v);
767: }
768: PetscStackCall("ML_Aggregate_Create",ML_Aggregate_Set_NullSpace(agg_object,bs,nvec+!!has_const,nullvec,mlocal);CHKERRQ(ierr));
769: PetscFree(nullvec);
770: } break;
771: case PCML_NULLSPACE_BLOCK:
772: PetscStackCall("ML_Aggregate_Set_NullSpace",ML_Aggregate_Set_NullSpace(agg_object,bs,bs,0,0);CHKERRQ(ierr));
773: break;
774: case PCML_NULLSPACE_SCALAR:
775: break;
776: default: SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_SUP,"Unknown null space type");
777: }
778: }
779: PetscStackCall("ML_Aggregate_Set_MaxCoarseSize",ML_Aggregate_Set_MaxCoarseSize(agg_object,pc_ml->MaxCoarseSize));
780: /* set options */
781: switch (pc_ml->CoarsenScheme) {
782: case 1:
783: PetscStackCall("ML_Aggregate_Set_CoarsenScheme_Coupled",ML_Aggregate_Set_CoarsenScheme_Coupled(agg_object));break;
784: case 2:
785: PetscStackCall("ML_Aggregate_Set_CoarsenScheme_MIS",ML_Aggregate_Set_CoarsenScheme_MIS(agg_object));break;
786: case 3:
787: PetscStackCall("ML_Aggregate_Set_CoarsenScheme_METIS",ML_Aggregate_Set_CoarsenScheme_METIS(agg_object));break;
788: }
789: PetscStackCall("ML_Aggregate_Set_Threshold",ML_Aggregate_Set_Threshold(agg_object,pc_ml->Threshold));
790: PetscStackCall("ML_Aggregate_Set_DampingFactor",ML_Aggregate_Set_DampingFactor(agg_object,pc_ml->DampingFactor));
791: if (pc_ml->SpectralNormScheme_Anorm) {
792: PetscStackCall("ML_Set_SpectralNormScheme_Anorm",ML_Set_SpectralNormScheme_Anorm(ml_object));
793: }
794: agg_object->keep_agg_information = (int)pc_ml->KeepAggInfo;
795: agg_object->keep_P_tentative = (int)pc_ml->Reusable;
796: agg_object->block_scaled_SA = (int)pc_ml->BlockScaling;
797: agg_object->minimizing_energy = (int)pc_ml->EnergyMinimization;
798: agg_object->minimizing_energy_droptol = (double)pc_ml->EnergyMinimizationDropTol;
799: agg_object->cheap_minimizing_energy = (int)pc_ml->EnergyMinimizationCheap;
801: if (pc_ml->Aux) {
802: if (!pc_ml->dim) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_USER,"Auxiliary matrix requires coordinates");
803: ml_object->Amat[0].aux_data->threshold = pc_ml->AuxThreshold;
804: ml_object->Amat[0].aux_data->enable = 1;
805: ml_object->Amat[0].aux_data->max_level = 10;
806: ml_object->Amat[0].num_PDEs = bs;
807: }
809: MatGetInfo(A,MAT_LOCAL,&info);
810: ml_object->Amat[0].N_nonzeros = (int) info.nz_used;
812: if (pc_ml->dim) {
813: PetscInt i,dim = pc_ml->dim;
814: ML_Aggregate_Viz_Stats *grid_info;
815: PetscInt nlocghost;
817: MatGetBlockSize(A,&bs);
818: nlocghost = Aloc->cmap->n / bs;
820: PetscStackCall("ML_Aggregate_VizAndStats_Setup(",ML_Aggregate_VizAndStats_Setup(ml_object)); /* create ml info for coords */
821: grid_info = (ML_Aggregate_Viz_Stats*) ml_object->Grid[0].Grid;
822: for (i = 0; i < dim; i++) {
823: /* set the finest level coordinates to point to the column-order array
824: * in pc_ml */
825: /* NOTE: must point away before VizAndStats_Clean so ML doesn't free */
826: switch (i) {
827: case 0: grid_info->x = pc_ml->coords + nlocghost * i; break;
828: case 1: grid_info->y = pc_ml->coords + nlocghost * i; break;
829: case 2: grid_info->z = pc_ml->coords + nlocghost * i; break;
830: default: SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_ARG_SIZ,"PCML coordinate dimension must be <= 3");
831: }
832: }
833: grid_info->Ndim = dim;
834: }
836: /* repartitioning */
837: if (pc_ml->Repartition) {
838: PetscStackCall("ML_Repartition_Activate",ML_Repartition_Activate(ml_object));
839: PetscStackCall("ML_Repartition_Set_LargestMinMaxRatio",ML_Repartition_Set_LargestMinMaxRatio(ml_object,pc_ml->MaxMinRatio));
840: PetscStackCall("ML_Repartition_Set_MinPerProc",ML_Repartition_Set_MinPerProc(ml_object,pc_ml->MinPerProc));
841: PetscStackCall("ML_Repartition_Set_PutOnSingleProc",ML_Repartition_Set_PutOnSingleProc(ml_object,pc_ml->PutOnSingleProc));
842: #if 0 /* Function not yet defined in ml-6.2 */
843: /* I'm not sure what compatibility issues might crop up if we partitioned
844: * on the finest level, so to be safe repartition starts on the next
845: * finest level (reflection default behavior in
846: * ml_MultiLevelPreconditioner) */
847: PetscStackCall("ML_Repartition_Set_StartLevel",ML_Repartition_Set_StartLevel(ml_object,1));
848: #endif
850: if (!pc_ml->RepartitionType) {
851: PetscInt i;
853: if (!pc_ml->dim) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_USER,"ML Zoltan repartitioning requires coordinates");
854: PetscStackCall("ML_Repartition_Set_Partitioner",ML_Repartition_Set_Partitioner(ml_object,ML_USEZOLTAN));
855: PetscStackCall("ML_Aggregate_Set_Dimensions",ML_Aggregate_Set_Dimensions(agg_object, pc_ml->dim));
857: for (i = 0; i < ml_object->ML_num_levels; i++) {
858: ML_Aggregate_Viz_Stats *grid_info = (ML_Aggregate_Viz_Stats*)ml_object->Grid[i].Grid;
859: grid_info->zoltan_type = pc_ml->ZoltanScheme + 1; /* ml numbers options 1, 2, 3 */
860: /* defaults from ml_agg_info.c */
861: grid_info->zoltan_estimated_its = 40; /* only relevant to hypergraph / fast hypergraph */
862: grid_info->zoltan_timers = 0;
863: grid_info->smoothing_steps = 4; /* only relevant to hypergraph / fast hypergraph */
864: }
865: } else {
866: PetscStackCall("ML_Repartition_Set_Partitioner",ML_Repartition_Set_Partitioner(ml_object,ML_USEPARMETIS));
867: }
868: }
870: if (pc_ml->OldHierarchy) {
871: PetscStackCall("ML_Gen_MGHierarchy_UsingAggregation",Nlevels = ML_Gen_MGHierarchy_UsingAggregation(ml_object,0,ML_INCREASING,agg_object));
872: } else {
873: PetscStackCall("ML_Gen_MultiLevelHierarchy_UsingAggregation",Nlevels = ML_Gen_MultiLevelHierarchy_UsingAggregation(ml_object,0,ML_INCREASING,agg_object));
874: }
875: if (Nlevels<=0) SETERRQ1(PetscObjectComm((PetscObject)pc),PETSC_ERR_ARG_OUTOFRANGE,"Nlevels %d must > 0",Nlevels);
876: pc_ml->Nlevels = Nlevels;
877: fine_level = Nlevels - 1;
879: PCMGSetLevels(pc,Nlevels,NULL);
880: /* set default smoothers */
881: for (level=1; level<=fine_level; level++) {
882: PCMGGetSmoother(pc,level,&smoother);
883: KSPSetType(smoother,KSPRICHARDSON);
884: KSPGetPC(smoother,&subpc);
885: PCSetType(subpc,PCSOR);
886: }
887: PetscObjectOptionsBegin((PetscObject)pc);
888: PCSetFromOptions_MG(PetscOptionsObject,pc); /* should be called in PCSetFromOptions_ML(), but cannot be called prior to PCMGSetLevels() */
889: PetscOptionsEnd();
891: PetscMalloc1(Nlevels,&gridctx);
893: pc_ml->gridctx = gridctx;
895: /* wrap ML matrices by PETSc shell matrices at coarsened grids.
896: Level 0 is the finest grid for ML, but coarsest for PETSc! */
897: gridctx[fine_level].A = A;
899: level = fine_level - 1;
900: if (size == 1) { /* convert ML P, R and A into seqaij format */
901: for (mllevel=1; mllevel<Nlevels; mllevel++) {
902: mlmat = &(ml_object->Pmat[mllevel]);
903: MatWrapML_SeqAIJ(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].P);
904: mlmat = &(ml_object->Rmat[mllevel-1]);
905: MatWrapML_SeqAIJ(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].R);
907: mlmat = &(ml_object->Amat[mllevel]);
908: MatWrapML_SeqAIJ(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].A);
909: level--;
910: }
911: } else { /* convert ML P and R into shell format, ML A into mpiaij format */
912: for (mllevel=1; mllevel<Nlevels; mllevel++) {
913: mlmat = &(ml_object->Pmat[mllevel]);
914: MatWrapML_SHELL(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].P);
915: mlmat = &(ml_object->Rmat[mllevel-1]);
916: MatWrapML_SHELL(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].R);
918: mlmat = &(ml_object->Amat[mllevel]);
919: MatWrapML_MPIAIJ(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].A);
920: level--;
921: }
922: }
924: /* create vectors and ksp at all levels */
925: for (level=0; level<fine_level; level++) {
926: level1 = level + 1;
927: VecCreate(((PetscObject)gridctx[level].A)->comm,&gridctx[level].x);
928: VecSetSizes(gridctx[level].x,gridctx[level].A->cmap->n,PETSC_DECIDE);
929: VecSetType(gridctx[level].x,VECMPI);
930: PCMGSetX(pc,level,gridctx[level].x);
932: VecCreate(((PetscObject)gridctx[level].A)->comm,&gridctx[level].b);
933: VecSetSizes(gridctx[level].b,gridctx[level].A->rmap->n,PETSC_DECIDE);
934: VecSetType(gridctx[level].b,VECMPI);
935: PCMGSetRhs(pc,level,gridctx[level].b);
937: VecCreate(((PetscObject)gridctx[level1].A)->comm,&gridctx[level1].r);
938: VecSetSizes(gridctx[level1].r,gridctx[level1].A->rmap->n,PETSC_DECIDE);
939: VecSetType(gridctx[level1].r,VECMPI);
940: PCMGSetR(pc,level1,gridctx[level1].r);
942: if (level == 0) {
943: PCMGGetCoarseSolve(pc,&gridctx[level].ksp);
944: } else {
945: PCMGGetSmoother(pc,level,&gridctx[level].ksp);
946: }
947: }
948: PCMGGetSmoother(pc,fine_level,&gridctx[fine_level].ksp);
950: /* create coarse level and the interpolation between the levels */
951: for (level=0; level<fine_level; level++) {
952: level1 = level + 1;
953: PCMGSetInterpolation(pc,level1,gridctx[level].P);
954: PCMGSetRestriction(pc,level1,gridctx[level].R);
955: if (level > 0) {
956: PCMGSetResidual(pc,level,PCMGResidualDefault,gridctx[level].A);
957: }
958: KSPSetOperators(gridctx[level].ksp,gridctx[level].A,gridctx[level].A);
959: }
960: PCMGSetResidual(pc,fine_level,PCMGResidualDefault,gridctx[fine_level].A);
961: KSPSetOperators(gridctx[fine_level].ksp,gridctx[level].A,gridctx[fine_level].A);
963: /* put coordinate info in levels */
964: if (pc_ml->dim) {
965: PetscInt i,j,dim = pc_ml->dim;
966: PetscInt bs, nloc;
967: PC subpc;
968: PetscReal *array;
970: level = fine_level;
971: for (mllevel = 0; mllevel < Nlevels; mllevel++) {
972: ML_Aggregate_Viz_Stats *grid_info = (ML_Aggregate_Viz_Stats*)ml_object->Amat[mllevel].to->Grid->Grid;
973: MPI_Comm comm = ((PetscObject)gridctx[level].A)->comm;
975: MatGetBlockSize (gridctx[level].A, &bs);
976: MatGetLocalSize (gridctx[level].A, NULL, &nloc);
977: nloc /= bs; /* number of local nodes */
979: VecCreate(comm,&gridctx[level].coords);
980: VecSetSizes(gridctx[level].coords,dim * nloc,PETSC_DECIDE);
981: VecSetType(gridctx[level].coords,VECMPI);
982: VecGetArray(gridctx[level].coords,&array);
983: for (j = 0; j < nloc; j++) {
984: for (i = 0; i < dim; i++) {
985: switch (i) {
986: case 0: array[dim * j + i] = grid_info->x[j]; break;
987: case 1: array[dim * j + i] = grid_info->y[j]; break;
988: case 2: array[dim * j + i] = grid_info->z[j]; break;
989: default: SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_ARG_SIZ,"PCML coordinate dimension must be <= 3");
990: }
991: }
992: }
994: /* passing coordinates to smoothers/coarse solver, should they need them */
995: KSPGetPC(gridctx[level].ksp,&subpc);
996: PCSetCoordinates(subpc,dim,nloc,array);
997: VecRestoreArray(gridctx[level].coords,&array);
998: level--;
999: }
1000: }
1002: /* setupcalled is set to 0 so that MG is setup from scratch */
1003: pc->setupcalled = 0;
1004: PCSetUp_MG(pc);
1005: return(0);
1006: }
1008: /* -------------------------------------------------------------------------- */
1009: /*
1010: PCDestroy_ML - Destroys the private context for the ML preconditioner
1011: that was created with PCCreate_ML().
1013: Input Parameter:
1014: . pc - the preconditioner context
1016: Application Interface Routine: PCDestroy()
1017: */
1020: PetscErrorCode PCDestroy_ML(PC pc)
1021: {
1023: PC_MG *mg = (PC_MG*)pc->data;
1024: PC_ML *pc_ml= (PC_ML*)mg->innerctx;
1027: PCReset_ML(pc);
1028: PetscFree(pc_ml);
1029: PCDestroy_MG(pc);
1030: PetscObjectComposeFunction((PetscObject)pc,"PCSetCoordinates_C",NULL);
1031: return(0);
1032: }
1036: PetscErrorCode PCSetFromOptions_ML(PetscOptionItems *PetscOptionsObject,PC pc)
1037: {
1039: PetscInt indx,PrintLevel,partindx;
1040: const char *scheme[] = {"Uncoupled","Coupled","MIS","METIS"};
1041: const char *part[] = {"Zoltan","ParMETIS"};
1042: #if defined(HAVE_ML_ZOLTAN)
1043: const char *zscheme[] = {"RCB","hypergraph","fast_hypergraph"};
1044: #endif
1045: PC_MG *mg = (PC_MG*)pc->data;
1046: PC_ML *pc_ml = (PC_ML*)mg->innerctx;
1047: PetscMPIInt size;
1048: MPI_Comm comm;
1051: PetscObjectGetComm((PetscObject)pc,&comm);
1052: MPI_Comm_size(comm,&size);
1053: PetscOptionsHead(PetscOptionsObject,"ML options");
1055: PrintLevel = 0;
1056: indx = 0;
1057: partindx = 0;
1059: PetscOptionsInt("-pc_ml_PrintLevel","Print level","ML_Set_PrintLevel",PrintLevel,&PrintLevel,NULL);
1060: PetscStackCall("ML_Set_PrintLeve",ML_Set_PrintLevel(PrintLevel));
1061: PetscOptionsInt("-pc_ml_maxNlevels","Maximum number of levels","None",pc_ml->MaxNlevels,&pc_ml->MaxNlevels,NULL);
1062: PetscOptionsInt("-pc_ml_maxCoarseSize","Maximum coarsest mesh size","ML_Aggregate_Set_MaxCoarseSize",pc_ml->MaxCoarseSize,&pc_ml->MaxCoarseSize,NULL);
1063: PetscOptionsEList("-pc_ml_CoarsenScheme","Aggregate Coarsen Scheme","ML_Aggregate_Set_CoarsenScheme_*",scheme,4,scheme[0],&indx,NULL);
1065: pc_ml->CoarsenScheme = indx;
1067: PetscOptionsReal("-pc_ml_DampingFactor","P damping factor","ML_Aggregate_Set_DampingFactor",pc_ml->DampingFactor,&pc_ml->DampingFactor,NULL);
1068: PetscOptionsReal("-pc_ml_Threshold","Smoother drop tol","ML_Aggregate_Set_Threshold",pc_ml->Threshold,&pc_ml->Threshold,NULL);
1069: PetscOptionsBool("-pc_ml_SpectralNormScheme_Anorm","Method used for estimating spectral radius","ML_Set_SpectralNormScheme_Anorm",pc_ml->SpectralNormScheme_Anorm,&pc_ml->SpectralNormScheme_Anorm,NULL);
1070: PetscOptionsBool("-pc_ml_Symmetrize","Symmetrize aggregation","ML_Set_Symmetrize",pc_ml->Symmetrize,&pc_ml->Symmetrize,NULL);
1071: PetscOptionsBool("-pc_ml_BlockScaling","Scale all dofs at each node together","None",pc_ml->BlockScaling,&pc_ml->BlockScaling,NULL);
1072: PetscOptionsEnum("-pc_ml_nullspace","Which type of null space information to use","None",PCMLNullSpaceTypes,(PetscEnum)pc_ml->nulltype,(PetscEnum*)&pc_ml->nulltype,NULL);
1073: 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);
1074: 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);
1075: /*
1076: The following checks a number of conditions. If we let this stuff slip by, then ML's error handling will take over.
1077: This is suboptimal because it amounts to calling exit(1) so we check for the most common conditions.
1079: We also try to set some sane defaults when energy minimization is activated, otherwise it's hard to find a working
1080: combination of options and ML's exit(1) explanations don't help matters.
1081: */
1082: if (pc_ml->EnergyMinimization < -1 || pc_ml->EnergyMinimization > 4) SETERRQ(comm,PETSC_ERR_ARG_OUTOFRANGE,"EnergyMinimization must be in range -1..4");
1083: if (pc_ml->EnergyMinimization == 4 && size > 1) SETERRQ(comm,PETSC_ERR_SUP,"Energy minimization type 4 does not work in parallel");
1084: if (pc_ml->EnergyMinimization == 4) {PetscInfo(pc,"Mandel's energy minimization scheme is experimental and broken in ML-6.2\n");}
1085: if (pc_ml->EnergyMinimization) {
1086: PetscOptionsReal("-pc_ml_EnergyMinimizationDropTol","Energy minimization drop tolerance","None",pc_ml->EnergyMinimizationDropTol,&pc_ml->EnergyMinimizationDropTol,NULL);
1087: }
1088: if (pc_ml->EnergyMinimization == 2) {
1089: /* According to ml_MultiLevelPreconditioner.cpp, this option is only meaningful for norm type (2) */
1090: PetscOptionsBool("-pc_ml_EnergyMinimizationCheap","Use cheaper variant of norm type 2","None",pc_ml->EnergyMinimizationCheap,&pc_ml->EnergyMinimizationCheap,NULL);
1091: }
1092: /* energy minimization sometimes breaks if this is turned off, the more classical stuff should be okay without it */
1093: if (pc_ml->EnergyMinimization) pc_ml->KeepAggInfo = PETSC_TRUE;
1094: PetscOptionsBool("-pc_ml_KeepAggInfo","Allows the preconditioner to be reused, or auxilliary matrices to be generated","None",pc_ml->KeepAggInfo,&pc_ml->KeepAggInfo,NULL);
1095: /* Option (-1) doesn't work at all (calls exit(1)) if the tentative restriction operator isn't stored. */
1096: if (pc_ml->EnergyMinimization == -1) pc_ml->Reusable = PETSC_TRUE;
1097: PetscOptionsBool("-pc_ml_Reusable","Store intermedaiate data structures so that the multilevel hierarchy is reusable","None",pc_ml->Reusable,&pc_ml->Reusable,NULL);
1098: /*
1099: ML's C API is severely underdocumented and lacks significant functionality. The C++ API calls
1100: ML_Gen_MultiLevelHierarchy_UsingAggregation() which is a modified copy (!?) of the documented function
1101: ML_Gen_MGHierarchy_UsingAggregation(). This modification, however, does not provide a strict superset of the
1102: functionality in the old function, so some users may still want to use it. Note that many options are ignored in
1103: this context, but ML doesn't provide a way to find out which ones.
1104: */
1105: PetscOptionsBool("-pc_ml_OldHierarchy","Use old routine to generate hierarchy","None",pc_ml->OldHierarchy,&pc_ml->OldHierarchy,NULL);
1106: PetscOptionsBool("-pc_ml_repartition", "Allow ML to repartition levels of the heirarchy","ML_Repartition_Activate",pc_ml->Repartition,&pc_ml->Repartition,NULL);
1107: if (pc_ml->Repartition) {
1108: PetscOptionsReal("-pc_ml_repartitionMaxMinRatio", "Acceptable ratio of repartitioned sizes","ML_Repartition_Set_LargestMinMaxRatio",pc_ml->MaxMinRatio,&pc_ml->MaxMinRatio,NULL);
1109: PetscOptionsInt("-pc_ml_repartitionMinPerProc", "Smallest repartitioned size","ML_Repartition_Set_MinPerProc",pc_ml->MinPerProc,&pc_ml->MinPerProc,NULL);
1110: PetscOptionsInt("-pc_ml_repartitionPutOnSingleProc", "Problem size automatically repartitioned to one processor","ML_Repartition_Set_PutOnSingleProc",pc_ml->PutOnSingleProc,&pc_ml->PutOnSingleProc,NULL);
1111: #if defined(HAVE_ML_ZOLTAN)
1112: partindx = 0;
1113: PetscOptionsEList("-pc_ml_repartitionType", "Repartitioning library to use","ML_Repartition_Set_Partitioner",part,2,part[0],&partindx,NULL);
1115: pc_ml->RepartitionType = partindx;
1116: if (!partindx) {
1117: PetscInt zindx = 0;
1119: PetscOptionsEList("-pc_ml_repartitionZoltanScheme", "Repartitioning scheme to use","None",zscheme,3,zscheme[0],&zindx,NULL);
1121: pc_ml->ZoltanScheme = zindx;
1122: }
1123: #else
1124: partindx = 1;
1125: PetscOptionsEList("-pc_ml_repartitionType", "Repartitioning library to use","ML_Repartition_Set_Partitioner",part,2,part[1],&partindx,NULL);
1126: pc_ml->RepartitionType = partindx;
1127: if (!partindx) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_SUP_SYS,"ML not compiled with Zoltan");
1128: #endif
1129: PetscOptionsBool("-pc_ml_Aux","Aggregate using auxiliary coordinate-based laplacian","None",pc_ml->Aux,&pc_ml->Aux,NULL);
1130: PetscOptionsReal("-pc_ml_AuxThreshold","Auxiliary smoother drop tol","None",pc_ml->AuxThreshold,&pc_ml->AuxThreshold,NULL);
1131: }
1132: PetscOptionsTail();
1133: return(0);
1134: }
1136: /* -------------------------------------------------------------------------- */
1137: /*
1138: PCCreate_ML - Creates a ML preconditioner context, PC_ML,
1139: and sets this as the private data within the generic preconditioning
1140: context, PC, that was created within PCCreate().
1142: Input Parameter:
1143: . pc - the preconditioner context
1145: Application Interface Routine: PCCreate()
1146: */
1148: /*MC
1149: PCML - Use algebraic multigrid preconditioning. This preconditioner requires you provide
1150: fine grid discretization matrix. The coarser grid matrices and restriction/interpolation
1151: operators are computed by ML, with the matrices coverted to PETSc matrices in aij format
1152: and the restriction/interpolation operators wrapped as PETSc shell matrices.
1154: Options Database Key:
1155: Multigrid options(inherited):
1156: + -pc_mg_cycles <1>: 1 for V cycle, 2 for W-cycle (MGSetCycles)
1157: . -pc_mg_smoothup <1>: Number of post-smoothing steps (MGSetNumberSmoothUp)
1158: . -pc_mg_smoothdown <1>: Number of pre-smoothing steps (MGSetNumberSmoothDown)
1159: - -pc_mg_type <multiplicative>: (one of) additive multiplicative full kascade
1160: ML options:
1161: + -pc_ml_PrintLevel <0>: Print level (ML_Set_PrintLevel)
1162: . -pc_ml_maxNlevels <10>: Maximum number of levels (None)
1163: . -pc_ml_maxCoarseSize <1>: Maximum coarsest mesh size (ML_Aggregate_Set_MaxCoarseSize)
1164: . -pc_ml_CoarsenScheme <Uncoupled>: (one of) Uncoupled Coupled MIS METIS
1165: . -pc_ml_DampingFactor <1.33333>: P damping factor (ML_Aggregate_Set_DampingFactor)
1166: . -pc_ml_Threshold <0>: Smoother drop tol (ML_Aggregate_Set_Threshold)
1167: . -pc_ml_SpectralNormScheme_Anorm <false>: Method used for estimating spectral radius (ML_Set_SpectralNormScheme_Anorm)
1168: . -pc_ml_repartition <false>: Allow ML to repartition levels of the heirarchy (ML_Repartition_Activate)
1169: . -pc_ml_repartitionMaxMinRatio <1.3>: Acceptable ratio of repartitioned sizes (ML_Repartition_Set_LargestMinMaxRatio)
1170: . -pc_ml_repartitionMinPerProc <512>: Smallest repartitioned size (ML_Repartition_Set_MinPerProc)
1171: . -pc_ml_repartitionPutOnSingleProc <5000>: Problem size automatically repartitioned to one processor (ML_Repartition_Set_PutOnSingleProc)
1172: . -pc_ml_repartitionType <Zoltan>: Repartitioning library to use (ML_Repartition_Set_Partitioner)
1173: . -pc_ml_repartitionZoltanScheme <RCB>: Repartitioning scheme to use (None)
1174: . -pc_ml_Aux <false>: Aggregate using auxiliary coordinate-based laplacian (None)
1175: - -pc_ml_AuxThreshold <0.0>: Auxiliary smoother drop tol (None)
1177: Level: intermediate
1179: Concepts: multigrid
1181: .seealso: PCCreate(), PCSetType(), PCType (for list of available types), PC, PCMGType,
1182: PCMGSetLevels(), PCMGGetLevels(), PCMGSetType(), MPSetCycles(), PCMGSetNumberSmoothDown(),
1183: PCMGSetNumberSmoothUp(), PCMGGetCoarseSolve(), PCMGSetResidual(), PCMGSetInterpolation(),
1184: PCMGSetRestriction(), PCMGGetSmoother(), PCMGGetSmootherUp(), PCMGGetSmootherDown(),
1185: PCMGSetCycleTypeOnLevel(), PCMGSetRhs(), PCMGSetX(), PCMGSetR()
1186: M*/
1190: PETSC_EXTERN PetscErrorCode PCCreate_ML(PC pc)
1191: {
1193: PC_ML *pc_ml;
1194: PC_MG *mg;
1197: /* PCML is an inherited class of PCMG. Initialize pc as PCMG */
1198: PCSetType(pc,PCMG); /* calls PCCreate_MG() and MGCreate_Private() */
1199: PetscObjectChangeTypeName((PetscObject)pc,PCML);
1200: /* Since PCMG tries to use DM assocated with PC must delete it */
1201: DMDestroy(&pc->dm);
1202: mg = (PC_MG*)pc->data;
1203: mg->galerkin = 2; /* Use Galerkin, but it is computed externally */
1205: /* create a supporting struct and attach it to pc */
1206: PetscNewLog(pc,&pc_ml);
1207: mg->innerctx = pc_ml;
1209: pc_ml->ml_object = 0;
1210: pc_ml->agg_object = 0;
1211: pc_ml->gridctx = 0;
1212: pc_ml->PetscMLdata = 0;
1213: pc_ml->Nlevels = -1;
1214: pc_ml->MaxNlevels = 10;
1215: pc_ml->MaxCoarseSize = 1;
1216: pc_ml->CoarsenScheme = 1;
1217: pc_ml->Threshold = 0.0;
1218: pc_ml->DampingFactor = 4.0/3.0;
1219: pc_ml->SpectralNormScheme_Anorm = PETSC_FALSE;
1220: pc_ml->size = 0;
1221: pc_ml->dim = 0;
1222: pc_ml->nloc = 0;
1223: pc_ml->coords = 0;
1224: pc_ml->Repartition = PETSC_FALSE;
1225: pc_ml->MaxMinRatio = 1.3;
1226: pc_ml->MinPerProc = 512;
1227: pc_ml->PutOnSingleProc = 5000;
1228: pc_ml->RepartitionType = 0;
1229: pc_ml->ZoltanScheme = 0;
1230: pc_ml->Aux = PETSC_FALSE;
1231: pc_ml->AuxThreshold = 0.0;
1233: /* allow for coordinates to be passed */
1234: PetscObjectComposeFunction((PetscObject)pc,"PCSetCoordinates_C",PCSetCoordinates_ML);
1236: /* overwrite the pointers of PCMG by the functions of PCML */
1237: pc->ops->setfromoptions = PCSetFromOptions_ML;
1238: pc->ops->setup = PCSetUp_ML;
1239: pc->ops->reset = PCReset_ML;
1240: pc->ops->destroy = PCDestroy_ML;
1241: return(0);
1242: }