Actual source code: ml.c
petsc-3.11.4 2019-09-28
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>
8: #include <petsc/private/pcmgimpl.h>
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;
66: 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[])
67: {
69: PetscInt m,i,j,k=0,row,*aj;
70: PetscScalar *aa;
71: FineGridCtx *ml=(FineGridCtx*)ML_Get_MyGetrowData(ML_data);
72: Mat_SeqAIJ *a = (Mat_SeqAIJ*)ml->Aloc->data;
74: MatGetSize(ml->Aloc,&m,NULL); if (ierr) return(0);
75: for (i = 0; i<N_requested_rows; i++) {
76: row = requested_rows[i];
77: row_lengths[i] = a->ilen[row];
78: if (allocated_space < k+row_lengths[i]) return(0);
79: if ((row >= 0) || (row <= (m-1))) {
80: aj = a->j + a->i[row];
81: aa = a->a + a->i[row];
82: for (j=0; j<row_lengths[i]; j++) {
83: columns[k] = aj[j];
84: values[k++] = aa[j];
85: }
86: }
87: }
88: return(1);
89: }
91: static PetscErrorCode PetscML_comm(double p[],void *ML_data)
92: {
93: PetscErrorCode ierr;
94: FineGridCtx *ml = (FineGridCtx*)ML_data;
95: Mat A = ml->A;
96: Mat_MPIAIJ *a = (Mat_MPIAIJ*)A->data;
97: PetscMPIInt size;
98: PetscInt i,in_length=A->rmap->n,out_length=ml->Aloc->cmap->n;
99: const PetscScalar *array;
102: MPI_Comm_size(PetscObjectComm((PetscObject)A),&size);
103: if (size == 1) return 0;
105: VecPlaceArray(ml->y,p);
106: VecScatterBegin(a->Mvctx,ml->y,a->lvec,INSERT_VALUES,SCATTER_FORWARD);
107: VecScatterEnd(a->Mvctx,ml->y,a->lvec,INSERT_VALUES,SCATTER_FORWARD);
108: VecResetArray(ml->y);
109: VecGetArrayRead(a->lvec,&array);
110: for (i=in_length; i<out_length; i++) p[i] = array[i-in_length];
111: VecRestoreArrayRead(a->lvec,&array);
112: return(0);
113: }
115: static int PetscML_matvec(ML_Operator *ML_data,int in_length,double p[],int out_length,double ap[])
116: {
118: FineGridCtx *ml = (FineGridCtx*)ML_Get_MyMatvecData(ML_data);
119: Mat A = ml->A, Aloc=ml->Aloc;
120: PetscMPIInt size;
121: PetscScalar *pwork=ml->pwork;
122: PetscInt i;
125: MPI_Comm_size(PetscObjectComm((PetscObject)A),&size);
126: if (size == 1) {
127: VecPlaceArray(ml->x,p);
128: } else {
129: for (i=0; i<in_length; i++) pwork[i] = p[i];
130: PetscML_comm(pwork,ml);
131: VecPlaceArray(ml->x,pwork);
132: }
133: VecPlaceArray(ml->y,ap);
134: MatMult(Aloc,ml->x,ml->y);
135: VecResetArray(ml->x);
136: VecResetArray(ml->y);
137: return(0);
138: }
140: static PetscErrorCode MatMult_ML(Mat A,Vec x,Vec y)
141: {
142: PetscErrorCode ierr;
143: Mat_MLShell *shell;
144: PetscScalar *yarray;
145: const PetscScalar *xarray;
146: PetscInt x_length,y_length;
149: MatShellGetContext(A,(void**)&shell);
150: VecGetArrayRead(x,&xarray);
151: VecGetArray(y,&yarray);
152: x_length = shell->mlmat->invec_leng;
153: y_length = shell->mlmat->outvec_leng;
154: PetscStackCall("ML_Operator_Apply",ML_Operator_Apply(shell->mlmat,x_length,(PetscScalar*)xarray,y_length,yarray));
155: VecRestoreArrayRead(x,&xarray);
156: VecRestoreArray(y,&yarray);
157: return(0);
158: }
160: /* Computes y = w + A * x
161: It is possible that w == y, but not x == y
162: */
163: static PetscErrorCode MatMultAdd_ML(Mat A,Vec x,Vec w,Vec y)
164: {
165: Mat_MLShell *shell;
166: PetscScalar *yarray;
167: const PetscScalar *xarray;
168: PetscInt x_length,y_length;
169: PetscErrorCode ierr;
172: MatShellGetContext(A, (void**) &shell);
173: if (y == w) {
174: if (!shell->work) {
175: VecDuplicate(y, &shell->work);
176: }
177: VecGetArrayRead(x, &xarray);
178: VecGetArray(shell->work, &yarray);
179: x_length = shell->mlmat->invec_leng;
180: y_length = shell->mlmat->outvec_leng;
181: PetscStackCall("ML_Operator_Apply",ML_Operator_Apply(shell->mlmat, x_length, (PetscScalar*)xarray, y_length, yarray));
182: VecRestoreArrayRead(x, &xarray);
183: VecRestoreArray(shell->work, &yarray);
184: VecAXPY(y, 1.0, shell->work);
185: } else {
186: VecGetArrayRead(x, &xarray);
187: VecGetArray(y, &yarray);
188: x_length = shell->mlmat->invec_leng;
189: y_length = shell->mlmat->outvec_leng;
190: PetscStackCall("ML_Operator_Apply",ML_Operator_Apply(shell->mlmat, x_length, (PetscScalar *)xarray, y_length, yarray));
191: VecRestoreArrayRead(x, &xarray);
192: VecRestoreArray(y, &yarray);
193: VecAXPY(y, 1.0, w);
194: }
195: return(0);
196: }
198: /* newtype is ignored since only handles one case */
199: static PetscErrorCode MatConvert_MPIAIJ_ML(Mat A,MatType newtype,MatReuse scall,Mat *Aloc)
200: {
202: Mat_MPIAIJ *mpimat=(Mat_MPIAIJ*)A->data;
203: Mat_SeqAIJ *mat,*a=(Mat_SeqAIJ*)(mpimat->A)->data,*b=(Mat_SeqAIJ*)(mpimat->B)->data;
204: PetscInt *ai=a->i,*aj=a->j,*bi=b->i,*bj=b->j;
205: PetscScalar *aa=a->a,*ba=b->a,*ca;
206: PetscInt am =A->rmap->n,an=A->cmap->n,i,j,k;
207: PetscInt *ci,*cj,ncols;
210: if (am != an) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"A must have a square diagonal portion, am: %d != an: %d",am,an);
212: if (scall == MAT_INITIAL_MATRIX) {
213: PetscMalloc1(1+am,&ci);
214: ci[0] = 0;
215: for (i=0; i<am; i++) ci[i+1] = ci[i] + (ai[i+1] - ai[i]) + (bi[i+1] - bi[i]);
216: PetscMalloc1(1+ci[am],&cj);
217: PetscMalloc1(1+ci[am],&ca);
219: k = 0;
220: for (i=0; i<am; i++) {
221: /* diagonal portion of A */
222: ncols = ai[i+1] - ai[i];
223: for (j=0; j<ncols; j++) {
224: cj[k] = *aj++;
225: ca[k++] = *aa++;
226: }
227: /* off-diagonal portion of A */
228: ncols = bi[i+1] - bi[i];
229: for (j=0; j<ncols; j++) {
230: cj[k] = an + (*bj); bj++;
231: ca[k++] = *ba++;
232: }
233: }
234: if (k != ci[am]) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"k: %d != ci[am]: %d",k,ci[am]);
236: /* put together the new matrix */
237: an = mpimat->A->cmap->n+mpimat->B->cmap->n;
238: MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,am,an,ci,cj,ca,Aloc);
240: /* MatCreateSeqAIJWithArrays flags matrix so PETSc doesn't free the user's arrays. */
241: /* Since these are PETSc arrays, change flags to free them as necessary. */
242: mat = (Mat_SeqAIJ*)(*Aloc)->data;
243: mat->free_a = PETSC_TRUE;
244: mat->free_ij = PETSC_TRUE;
246: mat->nonew = 0;
247: } else if (scall == MAT_REUSE_MATRIX) {
248: mat=(Mat_SeqAIJ*)(*Aloc)->data;
249: ci = mat->i; cj = mat->j; ca = mat->a;
250: for (i=0; i<am; i++) {
251: /* diagonal portion of A */
252: ncols = ai[i+1] - ai[i];
253: for (j=0; j<ncols; j++) *ca++ = *aa++;
254: /* off-diagonal portion of A */
255: ncols = bi[i+1] - bi[i];
256: for (j=0; j<ncols; j++) *ca++ = *ba++;
257: }
258: } else SETERRQ1(PetscObjectComm((PetscObject)A),PETSC_ERR_ARG_WRONG,"Invalid MatReuse %d",(int)scall);
259: return(0);
260: }
262: static PetscErrorCode MatDestroy_ML(Mat A)
263: {
265: Mat_MLShell *shell;
268: MatShellGetContext(A,(void**)&shell);
269: VecDestroy(&shell->y);
270: if (shell->work) {VecDestroy(&shell->work);}
271: PetscFree(shell);
272: return(0);
273: }
275: static PetscErrorCode MatWrapML_SeqAIJ(ML_Operator *mlmat,MatReuse reuse,Mat *newmat)
276: {
277: struct ML_CSR_MSRdata *matdata = (struct ML_CSR_MSRdata*)mlmat->data;
278: PetscErrorCode ierr;
279: PetscInt m =mlmat->outvec_leng,n=mlmat->invec_leng,*nnz = NULL,nz_max;
280: PetscInt *ml_cols=matdata->columns,*ml_rowptr=matdata->rowptr,*aj,i;
281: PetscScalar *ml_vals=matdata->values,*aa;
284: if (!mlmat->getrow) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_NULL,"mlmat->getrow = NULL");
285: if (m != n) { /* ML Pmat and Rmat are in CSR format. Pass array pointers into SeqAIJ matrix */
286: if (reuse) {
287: Mat_SeqAIJ *aij= (Mat_SeqAIJ*)(*newmat)->data;
288: aij->i = ml_rowptr;
289: aij->j = ml_cols;
290: aij->a = ml_vals;
291: } else {
292: /* sort ml_cols and ml_vals */
293: PetscMalloc1(m+1,&nnz);
294: for (i=0; i<m; i++) nnz[i] = ml_rowptr[i+1] - ml_rowptr[i];
295: aj = ml_cols; aa = ml_vals;
296: for (i=0; i<m; i++) {
297: PetscSortIntWithScalarArray(nnz[i],aj,aa);
298: aj += nnz[i]; aa += nnz[i];
299: }
300: MatCreateSeqAIJWithArrays(PETSC_COMM_SELF,m,n,ml_rowptr,ml_cols,ml_vals,newmat);
301: PetscFree(nnz);
302: }
303: return(0);
304: }
306: nz_max = PetscMax(1,mlmat->max_nz_per_row);
307: PetscMalloc2(nz_max,&aa,nz_max,&aj);
308: if (!reuse) {
309: MatCreate(PETSC_COMM_SELF,newmat);
310: MatSetSizes(*newmat,m,n,PETSC_DECIDE,PETSC_DECIDE);
311: MatSetType(*newmat,MATSEQAIJ);
312: /* keep track of block size for A matrices */
313: MatSetBlockSize (*newmat, mlmat->num_PDEs);
315: PetscMalloc1(m,&nnz);
316: for (i=0; i<m; i++) {
317: PetscStackCall("ML_Operator_Getrow",ML_Operator_Getrow(mlmat,1,&i,nz_max,aj,aa,&nnz[i]));
318: }
319: MatSeqAIJSetPreallocation(*newmat,0,nnz);
320: }
321: for (i=0; i<m; i++) {
322: PetscInt ncols;
324: PetscStackCall("ML_Operator_Getrow",ML_Operator_Getrow(mlmat,1,&i,nz_max,aj,aa,&ncols));
325: MatSetValues(*newmat,1,&i,ncols,aj,aa,INSERT_VALUES);
326: }
327: MatAssemblyBegin(*newmat,MAT_FINAL_ASSEMBLY);
328: MatAssemblyEnd(*newmat,MAT_FINAL_ASSEMBLY);
330: PetscFree2(aa,aj);
331: PetscFree(nnz);
332: return(0);
333: }
335: static PetscErrorCode MatWrapML_SHELL(ML_Operator *mlmat,MatReuse reuse,Mat *newmat)
336: {
338: PetscInt m,n;
339: ML_Comm *MLcomm;
340: Mat_MLShell *shellctx;
343: m = mlmat->outvec_leng;
344: n = mlmat->invec_leng;
346: if (reuse) {
347: MatShellGetContext(*newmat,(void**)&shellctx);
348: shellctx->mlmat = mlmat;
349: return(0);
350: }
352: MLcomm = mlmat->comm;
354: PetscNew(&shellctx);
355: MatCreateShell(MLcomm->USR_comm,m,n,PETSC_DETERMINE,PETSC_DETERMINE,shellctx,newmat);
356: MatShellSetOperation(*newmat,MATOP_MULT,(void(*)(void))MatMult_ML);
357: MatShellSetOperation(*newmat,MATOP_MULT_ADD,(void(*)(void))MatMultAdd_ML);
358: MatShellSetOperation(*newmat,MATOP_DESTROY,(void(*)(void))MatDestroy_ML);
360: shellctx->A = *newmat;
361: shellctx->mlmat = mlmat;
362: shellctx->work = NULL;
364: VecCreate(MLcomm->USR_comm,&shellctx->y);
365: VecSetSizes(shellctx->y,m,PETSC_DECIDE);
366: VecSetType(shellctx->y,VECSTANDARD);
367: return(0);
368: }
370: static PetscErrorCode MatWrapML_MPIAIJ(ML_Operator *mlmat,MatReuse reuse,Mat *newmat)
371: {
372: PetscInt *aj;
373: PetscScalar *aa;
375: PetscInt i,j,*gordering;
376: PetscInt m=mlmat->outvec_leng,n,nz_max,row;
377: Mat A;
380: if (!mlmat->getrow) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_NULL,"mlmat->getrow = NULL");
381: n = mlmat->invec_leng;
382: if (m != n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_OUTOFRANGE,"m %d must equal to n %d",m,n);
384: /* create global row numbering for a ML_Operator */
385: PetscStackCall("ML_build_global_numbering",ML_build_global_numbering(mlmat,&gordering,"rows"));
387: nz_max = PetscMax(1,mlmat->max_nz_per_row) + 1;
388: PetscMalloc2(nz_max,&aa,nz_max,&aj);
389: if (reuse) {
390: A = *newmat;
391: } else {
392: PetscInt *nnzA,*nnzB,*nnz;
393: PetscInt rstart;
394: MatCreate(mlmat->comm->USR_comm,&A);
395: MatSetSizes(A,m,n,PETSC_DECIDE,PETSC_DECIDE);
396: MatSetType(A,MATMPIAIJ);
397: /* keep track of block size for A matrices */
398: MatSetBlockSize (A,mlmat->num_PDEs);
399: PetscMalloc3(m,&nnzA,m,&nnzB,m,&nnz);
400: MPI_Scan(&m,&rstart,1,MPIU_INT,MPI_SUM,mlmat->comm->USR_comm);
401: rstart -= m;
403: for (i=0; i<m; i++) {
404: row = gordering[i] - rstart;
405: PetscStackCall("ML_Operator_Getrow",ML_Operator_Getrow(mlmat,1,&i,nz_max,aj,aa,&nnz[i]));
406: nnzA[row] = 0;
407: for (j=0; j<nnz[i]; j++) {
408: if (aj[j] < m) nnzA[row]++;
409: }
410: nnzB[row] = nnz[i] - nnzA[row];
411: }
412: MatMPIAIJSetPreallocation(A,0,nnzA,0,nnzB);
413: PetscFree3(nnzA,nnzB,nnz);
414: }
415: for (i=0; i<m; i++) {
416: PetscInt ncols;
417: row = gordering[i];
419: PetscStackCall(",ML_Operator_Getrow",ML_Operator_Getrow(mlmat,1,&i,nz_max,aj,aa,&ncols));
420: for (j = 0; j < ncols; j++) aj[j] = gordering[aj[j]];
421: MatSetValues(A,1,&row,ncols,aj,aa,INSERT_VALUES);
422: }
423: PetscStackCall("ML_free",ML_free(gordering));
424: MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);
425: MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);
426: *newmat = A;
428: PetscFree2(aa,aj);
429: return(0);
430: }
432: /* -------------------------------------------------------------------------- */
433: /*
434: PCSetCoordinates_ML
436: Input Parameter:
437: . pc - the preconditioner context
438: */
439: static PetscErrorCode PCSetCoordinates_ML(PC pc, PetscInt ndm, PetscInt a_nloc, PetscReal *coords)
440: {
441: PC_MG *mg = (PC_MG*)pc->data;
442: PC_ML *pc_ml = (PC_ML*)mg->innerctx;
444: PetscInt arrsz,oldarrsz,bs,my0,kk,ii,nloc,Iend,aloc;
445: Mat Amat = pc->pmat;
447: /* this function copied and modified from PCSetCoordinates_GEO -TGI */
450: MatGetBlockSize(Amat, &bs);
452: MatGetOwnershipRange(Amat, &my0, &Iend);
453: aloc = (Iend-my0);
454: nloc = (Iend-my0)/bs;
456: if (nloc!=a_nloc && aloc != a_nloc) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Number of local blocks %D must be %D or %D.",a_nloc,nloc,aloc);
458: oldarrsz = pc_ml->dim * pc_ml->nloc;
459: pc_ml->dim = ndm;
460: pc_ml->nloc = nloc;
461: arrsz = ndm * nloc;
463: /* create data - syntactic sugar that should be refactored at some point */
464: if (pc_ml->coords==0 || (oldarrsz != arrsz)) {
465: PetscFree(pc_ml->coords);
466: PetscMalloc1(arrsz, &pc_ml->coords);
467: }
468: for (kk=0; kk<arrsz; kk++) pc_ml->coords[kk] = -999.;
469: /* copy data in - column oriented */
470: if (nloc == a_nloc) {
471: for (kk = 0; kk < nloc; kk++) {
472: for (ii = 0; ii < ndm; ii++) {
473: pc_ml->coords[ii*nloc + kk] = coords[kk*ndm + ii];
474: }
475: }
476: } else { /* assumes the coordinates are blocked */
477: for (kk = 0; kk < nloc; kk++) {
478: for (ii = 0; ii < ndm; ii++) {
479: pc_ml->coords[ii*nloc + kk] = coords[bs*kk*ndm + ii];
480: }
481: }
482: }
483: return(0);
484: }
486: /* -----------------------------------------------------------------------------*/
487: extern PetscErrorCode PCReset_MG(PC);
488: PetscErrorCode PCReset_ML(PC pc)
489: {
491: PC_MG *mg = (PC_MG*)pc->data;
492: PC_ML *pc_ml = (PC_ML*)mg->innerctx;
493: PetscInt level,fine_level=pc_ml->Nlevels-1,dim=pc_ml->dim;
496: if (dim) {
497: for (level=0; level<=fine_level; level++) {
498: VecDestroy(&pc_ml->gridctx[level].coords);
499: }
500: if (pc_ml->ml_object && pc_ml->ml_object->Grid) {
501: ML_Aggregate_Viz_Stats * grid_info = (ML_Aggregate_Viz_Stats*) pc_ml->ml_object->Grid[0].Grid;
502: grid_info->x = 0; /* do this so ML doesn't try to free coordinates */
503: grid_info->y = 0;
504: grid_info->z = 0;
505: PetscStackCall("ML_Operator_Getrow",ML_Aggregate_VizAndStats_Clean(pc_ml->ml_object));
506: }
507: }
508: PetscStackCall("ML_Aggregate_Destroy",ML_Aggregate_Destroy(&pc_ml->agg_object));
509: PetscStackCall("ML_Aggregate_Destroy",ML_Destroy(&pc_ml->ml_object));
511: if (pc_ml->PetscMLdata) {
512: PetscFree(pc_ml->PetscMLdata->pwork);
513: MatDestroy(&pc_ml->PetscMLdata->Aloc);
514: VecDestroy(&pc_ml->PetscMLdata->x);
515: VecDestroy(&pc_ml->PetscMLdata->y);
516: }
517: PetscFree(pc_ml->PetscMLdata);
519: if (pc_ml->gridctx) {
520: for (level=0; level<fine_level; level++) {
521: if (pc_ml->gridctx[level].A) {MatDestroy(&pc_ml->gridctx[level].A);}
522: if (pc_ml->gridctx[level].P) {MatDestroy(&pc_ml->gridctx[level].P);}
523: if (pc_ml->gridctx[level].R) {MatDestroy(&pc_ml->gridctx[level].R);}
524: if (pc_ml->gridctx[level].x) {VecDestroy(&pc_ml->gridctx[level].x);}
525: if (pc_ml->gridctx[level].b) {VecDestroy(&pc_ml->gridctx[level].b);}
526: if (pc_ml->gridctx[level+1].r) {VecDestroy(&pc_ml->gridctx[level+1].r);}
527: }
528: }
529: PetscFree(pc_ml->gridctx);
530: PetscFree(pc_ml->coords);
532: pc_ml->dim = 0;
533: pc_ml->nloc = 0;
534: PCReset_MG(pc);
535: return(0);
536: }
537: /* -------------------------------------------------------------------------- */
538: /*
539: PCSetUp_ML - Prepares for the use of the ML preconditioner
540: by setting data structures and options.
542: Input Parameter:
543: . pc - the preconditioner context
545: Application Interface Routine: PCSetUp()
547: Notes:
548: The interface routine PCSetUp() is not usually called directly by
549: the user, but instead is called by PCApply() if necessary.
550: */
551: extern PetscErrorCode PCSetFromOptions_MG(PetscOptionItems *PetscOptionsObject,PC);
552: extern PetscErrorCode PCReset_MG(PC);
554: PetscErrorCode PCSetUp_ML(PC pc)
555: {
556: PetscErrorCode ierr;
557: PetscMPIInt size;
558: FineGridCtx *PetscMLdata;
559: ML *ml_object;
560: ML_Aggregate *agg_object;
561: ML_Operator *mlmat;
562: PetscInt nlocal_allcols,Nlevels,mllevel,level,level1,m,fine_level,bs;
563: Mat A,Aloc;
564: GridCtx *gridctx;
565: PC_MG *mg = (PC_MG*)pc->data;
566: PC_ML *pc_ml = (PC_ML*)mg->innerctx;
567: PetscBool isSeq, isMPI;
568: KSP smoother;
569: PC subpc;
570: PetscInt mesh_level, old_mesh_level;
571: MatInfo info;
572: static PetscBool cite = PETSC_FALSE;
575: 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);
576: A = pc->pmat;
577: MPI_Comm_size(PetscObjectComm((PetscObject)A),&size);
579: if (pc->setupcalled) {
580: if (pc->flag == SAME_NONZERO_PATTERN && pc_ml->reuse_interpolation) {
581: /*
582: Reuse interpolaton instead of recomputing aggregates and updating the whole hierarchy. This is less expensive for
583: multiple solves in which the matrix is not changing too quickly.
584: */
585: ml_object = pc_ml->ml_object;
586: gridctx = pc_ml->gridctx;
587: Nlevels = pc_ml->Nlevels;
588: fine_level = Nlevels - 1;
589: gridctx[fine_level].A = A;
591: PetscObjectTypeCompare((PetscObject) A, MATSEQAIJ, &isSeq);
592: PetscObjectTypeCompare((PetscObject) A, MATMPIAIJ, &isMPI);
593: if (isMPI) {
594: MatConvert_MPIAIJ_ML(A,NULL,MAT_INITIAL_MATRIX,&Aloc);
595: } else if (isSeq) {
596: Aloc = A;
597: PetscObjectReference((PetscObject)Aloc);
598: } 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);
600: MatGetSize(Aloc,&m,&nlocal_allcols);
601: PetscMLdata = pc_ml->PetscMLdata;
602: MatDestroy(&PetscMLdata->Aloc);
603: PetscMLdata->A = A;
604: PetscMLdata->Aloc = Aloc;
605: PetscStackCall("ML_Aggregate_Destroy",ML_Init_Amatrix(ml_object,0,m,m,PetscMLdata));
606: PetscStackCall("ML_Set_Amatrix_Matvec",ML_Set_Amatrix_Matvec(ml_object,0,PetscML_matvec));
608: mesh_level = ml_object->ML_finest_level;
609: while (ml_object->SingleLevel[mesh_level].Rmat->to) {
610: old_mesh_level = mesh_level;
611: mesh_level = ml_object->SingleLevel[mesh_level].Rmat->to->levelnum;
613: /* clean and regenerate A */
614: mlmat = &(ml_object->Amat[mesh_level]);
615: PetscStackCall("ML_Operator_Clean",ML_Operator_Clean(mlmat));
616: PetscStackCall("ML_Operator_Init",ML_Operator_Init(mlmat,ml_object->comm));
617: PetscStackCall("ML_Gen_AmatrixRAP",ML_Gen_AmatrixRAP(ml_object, old_mesh_level, mesh_level));
618: }
620: level = fine_level - 1;
621: if (size == 1) { /* convert ML P, R and A into seqaij format */
622: for (mllevel=1; mllevel<Nlevels; mllevel++) {
623: mlmat = &(ml_object->Amat[mllevel]);
624: MatWrapML_SeqAIJ(mlmat,MAT_REUSE_MATRIX,&gridctx[level].A);
625: level--;
626: }
627: } else { /* convert ML P and R into shell format, ML A into mpiaij format */
628: for (mllevel=1; mllevel<Nlevels; mllevel++) {
629: mlmat = &(ml_object->Amat[mllevel]);
630: MatWrapML_MPIAIJ(mlmat,MAT_REUSE_MATRIX,&gridctx[level].A);
631: level--;
632: }
633: }
635: for (level=0; level<fine_level; level++) {
636: if (level > 0) {
637: PCMGSetResidual(pc,level,PCMGResidualDefault,gridctx[level].A);
638: }
639: KSPSetOperators(gridctx[level].ksp,gridctx[level].A,gridctx[level].A);
640: }
641: PCMGSetResidual(pc,fine_level,PCMGResidualDefault,gridctx[fine_level].A);
642: KSPSetOperators(gridctx[fine_level].ksp,gridctx[level].A,gridctx[fine_level].A);
644: PCSetUp_MG(pc);
645: return(0);
646: } else {
647: /* since ML can change the size of vectors/matrices at any level we must destroy everything */
648: PCReset_ML(pc);
649: }
650: }
652: /* setup special features of PCML */
653: /*--------------------------------*/
654: /* covert A to Aloc to be used by ML at fine grid */
655: pc_ml->size = size;
656: PetscObjectTypeCompare((PetscObject) A, MATSEQAIJ, &isSeq);
657: PetscObjectTypeCompare((PetscObject) A, MATMPIAIJ, &isMPI);
658: if (isMPI) {
659: MatConvert_MPIAIJ_ML(A,NULL,MAT_INITIAL_MATRIX,&Aloc);
660: } else if (isSeq) {
661: Aloc = A;
662: PetscObjectReference((PetscObject)Aloc);
663: } 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);
665: /* create and initialize struct 'PetscMLdata' */
666: PetscNewLog(pc,&PetscMLdata);
667: pc_ml->PetscMLdata = PetscMLdata;
668: PetscMalloc1(Aloc->cmap->n+1,&PetscMLdata->pwork);
670: VecCreate(PETSC_COMM_SELF,&PetscMLdata->x);
671: VecSetSizes(PetscMLdata->x,Aloc->cmap->n,Aloc->cmap->n);
672: VecSetType(PetscMLdata->x,VECSEQ);
674: VecCreate(PETSC_COMM_SELF,&PetscMLdata->y);
675: VecSetSizes(PetscMLdata->y,A->rmap->n,PETSC_DECIDE);
676: VecSetType(PetscMLdata->y,VECSEQ);
677: PetscMLdata->A = A;
678: PetscMLdata->Aloc = Aloc;
679: if (pc_ml->dim) { /* create vecs around the coordinate data given */
680: PetscInt i,j,dim=pc_ml->dim;
681: PetscInt nloc = pc_ml->nloc,nlocghost;
682: PetscReal *ghostedcoords;
684: MatGetBlockSize(A,&bs);
685: nlocghost = Aloc->cmap->n / bs;
686: PetscMalloc1(dim*nlocghost,&ghostedcoords);
687: for (i = 0; i < dim; i++) {
688: /* copy coordinate values into first component of pwork */
689: for (j = 0; j < nloc; j++) {
690: PetscMLdata->pwork[bs * j] = pc_ml->coords[nloc * i + j];
691: }
692: /* get the ghost values */
693: PetscML_comm(PetscMLdata->pwork,PetscMLdata);
694: /* write into the vector */
695: for (j = 0; j < nlocghost; j++) {
696: ghostedcoords[i * nlocghost + j] = PetscMLdata->pwork[bs * j];
697: }
698: }
699: /* replace the original coords with the ghosted coords, because these are
700: * what ML needs */
701: PetscFree(pc_ml->coords);
702: pc_ml->coords = ghostedcoords;
703: }
705: /* create ML discretization matrix at fine grid */
706: /* ML requires input of fine-grid matrix. It determines nlevels. */
707: MatGetSize(Aloc,&m,&nlocal_allcols);
708: MatGetBlockSize(A,&bs);
709: PetscStackCall("ML_Create",ML_Create(&ml_object,pc_ml->MaxNlevels));
710: PetscStackCall("ML_Comm_Set_UsrComm",ML_Comm_Set_UsrComm(ml_object->comm,PetscObjectComm((PetscObject)A)));
711: pc_ml->ml_object = ml_object;
712: PetscStackCall("ML_Init_Amatrix",ML_Init_Amatrix(ml_object,0,m,m,PetscMLdata));
713: PetscStackCall("ML_Set_Amatrix_Getrow",ML_Set_Amatrix_Getrow(ml_object,0,PetscML_getrow,PetscML_comm,nlocal_allcols));
714: PetscStackCall("ML_Set_Amatrix_Matvec",ML_Set_Amatrix_Matvec(ml_object,0,PetscML_matvec));
716: PetscStackCall("ML_Set_Symmetrize",ML_Set_Symmetrize(ml_object,pc_ml->Symmetrize ? ML_YES : ML_NO));
718: /* aggregation */
719: PetscStackCall("ML_Aggregate_Create",ML_Aggregate_Create(&agg_object));
720: pc_ml->agg_object = agg_object;
722: {
723: MatNullSpace mnull;
724: MatGetNearNullSpace(A,&mnull);
725: if (pc_ml->nulltype == PCML_NULLSPACE_AUTO) {
726: if (mnull) pc_ml->nulltype = PCML_NULLSPACE_USER;
727: else if (bs > 1) pc_ml->nulltype = PCML_NULLSPACE_BLOCK;
728: else pc_ml->nulltype = PCML_NULLSPACE_SCALAR;
729: }
730: switch (pc_ml->nulltype) {
731: case PCML_NULLSPACE_USER: {
732: PetscScalar *nullvec;
733: const PetscScalar *v;
734: PetscBool has_const;
735: PetscInt i,j,mlocal,nvec,M;
736: const Vec *vecs;
738: if (!mnull) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_USER,"Must provide explicit null space using MatSetNearNullSpace() to use user-specified null space");
739: MatGetSize(A,&M,NULL);
740: MatGetLocalSize(Aloc,&mlocal,NULL);
741: MatNullSpaceGetVecs(mnull,&has_const,&nvec,&vecs);
742: PetscMalloc1((nvec+!!has_const)*mlocal,&nullvec);
743: if (has_const) for (i=0; i<mlocal; i++) nullvec[i] = 1.0/M;
744: for (i=0; i<nvec; i++) {
745: VecGetArrayRead(vecs[i],&v);
746: for (j=0; j<mlocal; j++) nullvec[(i+!!has_const)*mlocal + j] = v[j];
747: VecRestoreArrayRead(vecs[i],&v);
748: }
749: PetscStackCall("ML_Aggregate_Create",ML_Aggregate_Set_NullSpace(agg_object,bs,nvec+!!has_const,nullvec,mlocal);CHKERRQ(ierr));
750: PetscFree(nullvec);
751: } break;
752: case PCML_NULLSPACE_BLOCK:
753: PetscStackCall("ML_Aggregate_Set_NullSpace",ML_Aggregate_Set_NullSpace(agg_object,bs,bs,0,0);CHKERRQ(ierr));
754: break;
755: case PCML_NULLSPACE_SCALAR:
756: break;
757: default: SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_SUP,"Unknown null space type");
758: }
759: }
760: PetscStackCall("ML_Aggregate_Set_MaxCoarseSize",ML_Aggregate_Set_MaxCoarseSize(agg_object,pc_ml->MaxCoarseSize));
761: /* set options */
762: switch (pc_ml->CoarsenScheme) {
763: case 1:
764: PetscStackCall("ML_Aggregate_Set_CoarsenScheme_Coupled",ML_Aggregate_Set_CoarsenScheme_Coupled(agg_object));break;
765: case 2:
766: PetscStackCall("ML_Aggregate_Set_CoarsenScheme_MIS",ML_Aggregate_Set_CoarsenScheme_MIS(agg_object));break;
767: case 3:
768: PetscStackCall("ML_Aggregate_Set_CoarsenScheme_METIS",ML_Aggregate_Set_CoarsenScheme_METIS(agg_object));break;
769: }
770: PetscStackCall("ML_Aggregate_Set_Threshold",ML_Aggregate_Set_Threshold(agg_object,pc_ml->Threshold));
771: PetscStackCall("ML_Aggregate_Set_DampingFactor",ML_Aggregate_Set_DampingFactor(agg_object,pc_ml->DampingFactor));
772: if (pc_ml->SpectralNormScheme_Anorm) {
773: PetscStackCall("ML_Set_SpectralNormScheme_Anorm",ML_Set_SpectralNormScheme_Anorm(ml_object));
774: }
775: agg_object->keep_agg_information = (int)pc_ml->KeepAggInfo;
776: agg_object->keep_P_tentative = (int)pc_ml->Reusable;
777: agg_object->block_scaled_SA = (int)pc_ml->BlockScaling;
778: agg_object->minimizing_energy = (int)pc_ml->EnergyMinimization;
779: agg_object->minimizing_energy_droptol = (double)pc_ml->EnergyMinimizationDropTol;
780: agg_object->cheap_minimizing_energy = (int)pc_ml->EnergyMinimizationCheap;
782: if (pc_ml->Aux) {
783: if (!pc_ml->dim) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_USER,"Auxiliary matrix requires coordinates");
784: ml_object->Amat[0].aux_data->threshold = pc_ml->AuxThreshold;
785: ml_object->Amat[0].aux_data->enable = 1;
786: ml_object->Amat[0].aux_data->max_level = 10;
787: ml_object->Amat[0].num_PDEs = bs;
788: }
790: MatGetInfo(A,MAT_LOCAL,&info);
791: ml_object->Amat[0].N_nonzeros = (int) info.nz_used;
793: if (pc_ml->dim) {
794: PetscInt i,dim = pc_ml->dim;
795: ML_Aggregate_Viz_Stats *grid_info;
796: PetscInt nlocghost;
798: MatGetBlockSize(A,&bs);
799: nlocghost = Aloc->cmap->n / bs;
801: PetscStackCall("ML_Aggregate_VizAndStats_Setup(",ML_Aggregate_VizAndStats_Setup(ml_object)); /* create ml info for coords */
802: grid_info = (ML_Aggregate_Viz_Stats*) ml_object->Grid[0].Grid;
803: for (i = 0; i < dim; i++) {
804: /* set the finest level coordinates to point to the column-order array
805: * in pc_ml */
806: /* NOTE: must point away before VizAndStats_Clean so ML doesn't free */
807: switch (i) {
808: case 0: grid_info->x = pc_ml->coords + nlocghost * i; break;
809: case 1: grid_info->y = pc_ml->coords + nlocghost * i; break;
810: case 2: grid_info->z = pc_ml->coords + nlocghost * i; break;
811: default: SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_ARG_SIZ,"PCML coordinate dimension must be <= 3");
812: }
813: }
814: grid_info->Ndim = dim;
815: }
817: /* repartitioning */
818: if (pc_ml->Repartition) {
819: PetscStackCall("ML_Repartition_Activate",ML_Repartition_Activate(ml_object));
820: PetscStackCall("ML_Repartition_Set_LargestMinMaxRatio",ML_Repartition_Set_LargestMinMaxRatio(ml_object,pc_ml->MaxMinRatio));
821: PetscStackCall("ML_Repartition_Set_MinPerProc",ML_Repartition_Set_MinPerProc(ml_object,pc_ml->MinPerProc));
822: PetscStackCall("ML_Repartition_Set_PutOnSingleProc",ML_Repartition_Set_PutOnSingleProc(ml_object,pc_ml->PutOnSingleProc));
823: #if 0 /* Function not yet defined in ml-6.2 */
824: /* I'm not sure what compatibility issues might crop up if we partitioned
825: * on the finest level, so to be safe repartition starts on the next
826: * finest level (reflection default behavior in
827: * ml_MultiLevelPreconditioner) */
828: PetscStackCall("ML_Repartition_Set_StartLevel",ML_Repartition_Set_StartLevel(ml_object,1));
829: #endif
831: if (!pc_ml->RepartitionType) {
832: PetscInt i;
834: if (!pc_ml->dim) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_USER,"ML Zoltan repartitioning requires coordinates");
835: PetscStackCall("ML_Repartition_Set_Partitioner",ML_Repartition_Set_Partitioner(ml_object,ML_USEZOLTAN));
836: PetscStackCall("ML_Aggregate_Set_Dimensions",ML_Aggregate_Set_Dimensions(agg_object, pc_ml->dim));
838: for (i = 0; i < ml_object->ML_num_levels; i++) {
839: ML_Aggregate_Viz_Stats *grid_info = (ML_Aggregate_Viz_Stats*)ml_object->Grid[i].Grid;
840: grid_info->zoltan_type = pc_ml->ZoltanScheme + 1; /* ml numbers options 1, 2, 3 */
841: /* defaults from ml_agg_info.c */
842: grid_info->zoltan_estimated_its = 40; /* only relevant to hypergraph / fast hypergraph */
843: grid_info->zoltan_timers = 0;
844: grid_info->smoothing_steps = 4; /* only relevant to hypergraph / fast hypergraph */
845: }
846: } else {
847: PetscStackCall("ML_Repartition_Set_Partitioner",ML_Repartition_Set_Partitioner(ml_object,ML_USEPARMETIS));
848: }
849: }
851: if (pc_ml->OldHierarchy) {
852: PetscStackCall("ML_Gen_MGHierarchy_UsingAggregation",Nlevels = ML_Gen_MGHierarchy_UsingAggregation(ml_object,0,ML_INCREASING,agg_object));
853: } else {
854: PetscStackCall("ML_Gen_MultiLevelHierarchy_UsingAggregation",Nlevels = ML_Gen_MultiLevelHierarchy_UsingAggregation(ml_object,0,ML_INCREASING,agg_object));
855: }
856: if (Nlevels<=0) SETERRQ1(PetscObjectComm((PetscObject)pc),PETSC_ERR_ARG_OUTOFRANGE,"Nlevels %d must > 0",Nlevels);
857: pc_ml->Nlevels = Nlevels;
858: fine_level = Nlevels - 1;
860: PCMGSetLevels(pc,Nlevels,NULL);
861: /* set default smoothers */
862: for (level=1; level<=fine_level; level++) {
863: PCMGGetSmoother(pc,level,&smoother);
864: KSPSetType(smoother,KSPRICHARDSON);
865: KSPGetPC(smoother,&subpc);
866: PCSetType(subpc,PCSOR);
867: }
868: PetscObjectOptionsBegin((PetscObject)pc);
869: PCSetFromOptions_MG(PetscOptionsObject,pc); /* should be called in PCSetFromOptions_ML(), but cannot be called prior to PCMGSetLevels() */
870: PetscOptionsEnd();
872: PetscMalloc1(Nlevels,&gridctx);
874: pc_ml->gridctx = gridctx;
876: /* wrap ML matrices by PETSc shell matrices at coarsened grids.
877: Level 0 is the finest grid for ML, but coarsest for PETSc! */
878: gridctx[fine_level].A = A;
880: level = fine_level - 1;
881: if (size == 1) { /* convert ML P, R and A into seqaij format */
882: for (mllevel=1; mllevel<Nlevels; mllevel++) {
883: mlmat = &(ml_object->Pmat[mllevel]);
884: MatWrapML_SeqAIJ(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].P);
885: mlmat = &(ml_object->Rmat[mllevel-1]);
886: MatWrapML_SeqAIJ(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].R);
888: mlmat = &(ml_object->Amat[mllevel]);
889: MatWrapML_SeqAIJ(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].A);
890: level--;
891: }
892: } else { /* convert ML P and R into shell format, ML A into mpiaij format */
893: for (mllevel=1; mllevel<Nlevels; mllevel++) {
894: mlmat = &(ml_object->Pmat[mllevel]);
895: MatWrapML_SHELL(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].P);
896: mlmat = &(ml_object->Rmat[mllevel-1]);
897: MatWrapML_SHELL(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].R);
899: mlmat = &(ml_object->Amat[mllevel]);
900: MatWrapML_MPIAIJ(mlmat,MAT_INITIAL_MATRIX,&gridctx[level].A);
901: level--;
902: }
903: }
905: /* create vectors and ksp at all levels */
906: for (level=0; level<fine_level; level++) {
907: level1 = level + 1;
908: VecCreate(((PetscObject)gridctx[level].A)->comm,&gridctx[level].x);
909: VecSetSizes(gridctx[level].x,gridctx[level].A->cmap->n,PETSC_DECIDE);
910: VecSetType(gridctx[level].x,VECMPI);
911: PCMGSetX(pc,level,gridctx[level].x);
913: VecCreate(((PetscObject)gridctx[level].A)->comm,&gridctx[level].b);
914: VecSetSizes(gridctx[level].b,gridctx[level].A->rmap->n,PETSC_DECIDE);
915: VecSetType(gridctx[level].b,VECMPI);
916: PCMGSetRhs(pc,level,gridctx[level].b);
918: VecCreate(((PetscObject)gridctx[level1].A)->comm,&gridctx[level1].r);
919: VecSetSizes(gridctx[level1].r,gridctx[level1].A->rmap->n,PETSC_DECIDE);
920: VecSetType(gridctx[level1].r,VECMPI);
921: PCMGSetR(pc,level1,gridctx[level1].r);
923: if (level == 0) {
924: PCMGGetCoarseSolve(pc,&gridctx[level].ksp);
925: } else {
926: PCMGGetSmoother(pc,level,&gridctx[level].ksp);
927: }
928: }
929: PCMGGetSmoother(pc,fine_level,&gridctx[fine_level].ksp);
931: /* create coarse level and the interpolation between the levels */
932: for (level=0; level<fine_level; level++) {
933: level1 = level + 1;
934: PCMGSetInterpolation(pc,level1,gridctx[level].P);
935: PCMGSetRestriction(pc,level1,gridctx[level].R);
936: if (level > 0) {
937: PCMGSetResidual(pc,level,PCMGResidualDefault,gridctx[level].A);
938: }
939: KSPSetOperators(gridctx[level].ksp,gridctx[level].A,gridctx[level].A);
940: }
941: PCMGSetResidual(pc,fine_level,PCMGResidualDefault,gridctx[fine_level].A);
942: KSPSetOperators(gridctx[fine_level].ksp,gridctx[level].A,gridctx[fine_level].A);
944: /* put coordinate info in levels */
945: if (pc_ml->dim) {
946: PetscInt i,j,dim = pc_ml->dim;
947: PetscInt bs, nloc;
948: PC subpc;
949: PetscReal *array;
951: level = fine_level;
952: for (mllevel = 0; mllevel < Nlevels; mllevel++) {
953: ML_Aggregate_Viz_Stats *grid_info = (ML_Aggregate_Viz_Stats*)ml_object->Amat[mllevel].to->Grid->Grid;
954: MPI_Comm comm = ((PetscObject)gridctx[level].A)->comm;
956: MatGetBlockSize (gridctx[level].A, &bs);
957: MatGetLocalSize (gridctx[level].A, NULL, &nloc);
958: nloc /= bs; /* number of local nodes */
960: VecCreate(comm,&gridctx[level].coords);
961: VecSetSizes(gridctx[level].coords,dim * nloc,PETSC_DECIDE);
962: VecSetType(gridctx[level].coords,VECMPI);
963: VecGetArray(gridctx[level].coords,&array);
964: for (j = 0; j < nloc; j++) {
965: for (i = 0; i < dim; i++) {
966: switch (i) {
967: case 0: array[dim * j + i] = grid_info->x[j]; break;
968: case 1: array[dim * j + i] = grid_info->y[j]; break;
969: case 2: array[dim * j + i] = grid_info->z[j]; break;
970: default: SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_ARG_SIZ,"PCML coordinate dimension must be <= 3");
971: }
972: }
973: }
975: /* passing coordinates to smoothers/coarse solver, should they need them */
976: KSPGetPC(gridctx[level].ksp,&subpc);
977: PCSetCoordinates(subpc,dim,nloc,array);
978: VecRestoreArray(gridctx[level].coords,&array);
979: level--;
980: }
981: }
983: /* setupcalled is set to 0 so that MG is setup from scratch */
984: pc->setupcalled = 0;
985: PCSetUp_MG(pc);
986: return(0);
987: }
989: /* -------------------------------------------------------------------------- */
990: /*
991: PCDestroy_ML - Destroys the private context for the ML preconditioner
992: that was created with PCCreate_ML().
994: Input Parameter:
995: . pc - the preconditioner context
997: Application Interface Routine: PCDestroy()
998: */
999: PetscErrorCode PCDestroy_ML(PC pc)
1000: {
1002: PC_MG *mg = (PC_MG*)pc->data;
1003: PC_ML *pc_ml= (PC_ML*)mg->innerctx;
1006: PCReset_ML(pc);
1007: PetscFree(pc_ml);
1008: PCDestroy_MG(pc);
1009: PetscObjectComposeFunction((PetscObject)pc,"PCSetCoordinates_C",NULL);
1010: return(0);
1011: }
1013: PetscErrorCode PCSetFromOptions_ML(PetscOptionItems *PetscOptionsObject,PC pc)
1014: {
1016: PetscInt indx,PrintLevel,partindx;
1017: const char *scheme[] = {"Uncoupled","Coupled","MIS","METIS"};
1018: const char *part[] = {"Zoltan","ParMETIS"};
1019: #if defined(HAVE_ML_ZOLTAN)
1020: const char *zscheme[] = {"RCB","hypergraph","fast_hypergraph"};
1021: #endif
1022: PC_MG *mg = (PC_MG*)pc->data;
1023: PC_ML *pc_ml = (PC_ML*)mg->innerctx;
1024: PetscMPIInt size;
1025: MPI_Comm comm;
1028: PetscObjectGetComm((PetscObject)pc,&comm);
1029: MPI_Comm_size(comm,&size);
1030: PetscOptionsHead(PetscOptionsObject,"ML options");
1032: PrintLevel = 0;
1033: indx = 0;
1034: partindx = 0;
1036: PetscOptionsInt("-pc_ml_PrintLevel","Print level","ML_Set_PrintLevel",PrintLevel,&PrintLevel,NULL);
1037: PetscStackCall("ML_Set_PrintLevel",ML_Set_PrintLevel(PrintLevel));
1038: PetscOptionsInt("-pc_ml_maxNlevels","Maximum number of levels","None",pc_ml->MaxNlevels,&pc_ml->MaxNlevels,NULL);
1039: PetscOptionsInt("-pc_ml_maxCoarseSize","Maximum coarsest mesh size","ML_Aggregate_Set_MaxCoarseSize",pc_ml->MaxCoarseSize,&pc_ml->MaxCoarseSize,NULL);
1040: PetscOptionsEList("-pc_ml_CoarsenScheme","Aggregate Coarsen Scheme","ML_Aggregate_Set_CoarsenScheme_*",scheme,4,scheme[0],&indx,NULL);
1042: pc_ml->CoarsenScheme = indx;
1044: PetscOptionsReal("-pc_ml_DampingFactor","P damping factor","ML_Aggregate_Set_DampingFactor",pc_ml->DampingFactor,&pc_ml->DampingFactor,NULL);
1045: PetscOptionsReal("-pc_ml_Threshold","Smoother drop tol","ML_Aggregate_Set_Threshold",pc_ml->Threshold,&pc_ml->Threshold,NULL);
1046: PetscOptionsBool("-pc_ml_SpectralNormScheme_Anorm","Method used for estimating spectral radius","ML_Set_SpectralNormScheme_Anorm",pc_ml->SpectralNormScheme_Anorm,&pc_ml->SpectralNormScheme_Anorm,NULL);
1047: PetscOptionsBool("-pc_ml_Symmetrize","Symmetrize aggregation","ML_Set_Symmetrize",pc_ml->Symmetrize,&pc_ml->Symmetrize,NULL);
1048: PetscOptionsBool("-pc_ml_BlockScaling","Scale all dofs at each node together","None",pc_ml->BlockScaling,&pc_ml->BlockScaling,NULL);
1049: PetscOptionsEnum("-pc_ml_nullspace","Which type of null space information to use","None",PCMLNullSpaceTypes,(PetscEnum)pc_ml->nulltype,(PetscEnum*)&pc_ml->nulltype,NULL);
1050: 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);
1051: 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);
1052: /*
1053: The following checks a number of conditions. If we let this stuff slip by, then ML's error handling will take over.
1054: This is suboptimal because it amounts to calling exit(1) so we check for the most common conditions.
1056: We also try to set some sane defaults when energy minimization is activated, otherwise it's hard to find a working
1057: combination of options and ML's exit(1) explanations don't help matters.
1058: */
1059: if (pc_ml->EnergyMinimization < -1 || pc_ml->EnergyMinimization > 4) SETERRQ(comm,PETSC_ERR_ARG_OUTOFRANGE,"EnergyMinimization must be in range -1..4");
1060: if (pc_ml->EnergyMinimization == 4 && size > 1) SETERRQ(comm,PETSC_ERR_SUP,"Energy minimization type 4 does not work in parallel");
1061: if (pc_ml->EnergyMinimization == 4) {PetscInfo(pc,"Mandel's energy minimization scheme is experimental and broken in ML-6.2\n");}
1062: if (pc_ml->EnergyMinimization) {
1063: PetscOptionsReal("-pc_ml_EnergyMinimizationDropTol","Energy minimization drop tolerance","None",pc_ml->EnergyMinimizationDropTol,&pc_ml->EnergyMinimizationDropTol,NULL);
1064: }
1065: if (pc_ml->EnergyMinimization == 2) {
1066: /* According to ml_MultiLevelPreconditioner.cpp, this option is only meaningful for norm type (2) */
1067: PetscOptionsBool("-pc_ml_EnergyMinimizationCheap","Use cheaper variant of norm type 2","None",pc_ml->EnergyMinimizationCheap,&pc_ml->EnergyMinimizationCheap,NULL);
1068: }
1069: /* energy minimization sometimes breaks if this is turned off, the more classical stuff should be okay without it */
1070: if (pc_ml->EnergyMinimization) pc_ml->KeepAggInfo = PETSC_TRUE;
1071: PetscOptionsBool("-pc_ml_KeepAggInfo","Allows the preconditioner to be reused, or auxilliary matrices to be generated","None",pc_ml->KeepAggInfo,&pc_ml->KeepAggInfo,NULL);
1072: /* Option (-1) doesn't work at all (calls exit(1)) if the tentative restriction operator isn't stored. */
1073: if (pc_ml->EnergyMinimization == -1) pc_ml->Reusable = PETSC_TRUE;
1074: PetscOptionsBool("-pc_ml_Reusable","Store intermedaiate data structures so that the multilevel hierarchy is reusable","None",pc_ml->Reusable,&pc_ml->Reusable,NULL);
1075: /*
1076: ML's C API is severely underdocumented and lacks significant functionality. The C++ API calls
1077: ML_Gen_MultiLevelHierarchy_UsingAggregation() which is a modified copy (!?) of the documented function
1078: ML_Gen_MGHierarchy_UsingAggregation(). This modification, however, does not provide a strict superset of the
1079: functionality in the old function, so some users may still want to use it. Note that many options are ignored in
1080: this context, but ML doesn't provide a way to find out which ones.
1081: */
1082: PetscOptionsBool("-pc_ml_OldHierarchy","Use old routine to generate hierarchy","None",pc_ml->OldHierarchy,&pc_ml->OldHierarchy,NULL);
1083: PetscOptionsBool("-pc_ml_repartition", "Allow ML to repartition levels of the heirarchy","ML_Repartition_Activate",pc_ml->Repartition,&pc_ml->Repartition,NULL);
1084: if (pc_ml->Repartition) {
1085: PetscOptionsReal("-pc_ml_repartitionMaxMinRatio", "Acceptable ratio of repartitioned sizes","ML_Repartition_Set_LargestMinMaxRatio",pc_ml->MaxMinRatio,&pc_ml->MaxMinRatio,NULL);
1086: PetscOptionsInt("-pc_ml_repartitionMinPerProc", "Smallest repartitioned size","ML_Repartition_Set_MinPerProc",pc_ml->MinPerProc,&pc_ml->MinPerProc,NULL);
1087: PetscOptionsInt("-pc_ml_repartitionPutOnSingleProc", "Problem size automatically repartitioned to one processor","ML_Repartition_Set_PutOnSingleProc",pc_ml->PutOnSingleProc,&pc_ml->PutOnSingleProc,NULL);
1088: #if defined(HAVE_ML_ZOLTAN)
1089: partindx = 0;
1090: PetscOptionsEList("-pc_ml_repartitionType", "Repartitioning library to use","ML_Repartition_Set_Partitioner",part,2,part[0],&partindx,NULL);
1092: pc_ml->RepartitionType = partindx;
1093: if (!partindx) {
1094: PetscInt zindx = 0;
1096: PetscOptionsEList("-pc_ml_repartitionZoltanScheme", "Repartitioning scheme to use","None",zscheme,3,zscheme[0],&zindx,NULL);
1098: pc_ml->ZoltanScheme = zindx;
1099: }
1100: #else
1101: partindx = 1;
1102: PetscOptionsEList("-pc_ml_repartitionType", "Repartitioning library to use","ML_Repartition_Set_Partitioner",part,2,part[1],&partindx,NULL);
1103: pc_ml->RepartitionType = partindx;
1104: if (!partindx) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_SUP_SYS,"ML not compiled with Zoltan");
1105: #endif
1106: PetscOptionsBool("-pc_ml_Aux","Aggregate using auxiliary coordinate-based laplacian","None",pc_ml->Aux,&pc_ml->Aux,NULL);
1107: PetscOptionsReal("-pc_ml_AuxThreshold","Auxiliary smoother drop tol","None",pc_ml->AuxThreshold,&pc_ml->AuxThreshold,NULL);
1108: }
1109: PetscOptionsTail();
1110: return(0);
1111: }
1113: /* -------------------------------------------------------------------------- */
1114: /*
1115: PCCreate_ML - Creates a ML preconditioner context, PC_ML,
1116: and sets this as the private data within the generic preconditioning
1117: context, PC, that was created within PCCreate().
1119: Input Parameter:
1120: . pc - the preconditioner context
1122: Application Interface Routine: PCCreate()
1123: */
1125: /*MC
1126: PCML - Use algebraic multigrid preconditioning. This preconditioner requires you provide
1127: fine grid discretization matrix. The coarser grid matrices and restriction/interpolation
1128: operators are computed by ML, with the matrices coverted to PETSc matrices in aij format
1129: and the restriction/interpolation operators wrapped as PETSc shell matrices.
1131: Options Database Key:
1132: Multigrid options(inherited):
1133: + -pc_mg_cycles <1>: 1 for V cycle, 2 for W-cycle (MGSetCycles)
1134: . -pc_mg_distinct_smoothup: Should one configure the up and down smoothers separately (PCMGSetDistinctSmoothUp)
1135: - -pc_mg_type <multiplicative>: (one of) additive multiplicative full kascade
1136: ML options:
1137: + -pc_ml_PrintLevel <0>: Print level (ML_Set_PrintLevel)
1138: . -pc_ml_maxNlevels <10>: Maximum number of levels (None)
1139: . -pc_ml_maxCoarseSize <1>: Maximum coarsest mesh size (ML_Aggregate_Set_MaxCoarseSize)
1140: . -pc_ml_CoarsenScheme <Uncoupled>: (one of) Uncoupled Coupled MIS METIS
1141: . -pc_ml_DampingFactor <1.33333>: P damping factor (ML_Aggregate_Set_DampingFactor)
1142: . -pc_ml_Threshold <0>: Smoother drop tol (ML_Aggregate_Set_Threshold)
1143: . -pc_ml_SpectralNormScheme_Anorm <false>: Method used for estimating spectral radius (ML_Set_SpectralNormScheme_Anorm)
1144: . -pc_ml_repartition <false>: Allow ML to repartition levels of the heirarchy (ML_Repartition_Activate)
1145: . -pc_ml_repartitionMaxMinRatio <1.3>: Acceptable ratio of repartitioned sizes (ML_Repartition_Set_LargestMinMaxRatio)
1146: . -pc_ml_repartitionMinPerProc <512>: Smallest repartitioned size (ML_Repartition_Set_MinPerProc)
1147: . -pc_ml_repartitionPutOnSingleProc <5000>: Problem size automatically repartitioned to one processor (ML_Repartition_Set_PutOnSingleProc)
1148: . -pc_ml_repartitionType <Zoltan>: Repartitioning library to use (ML_Repartition_Set_Partitioner)
1149: . -pc_ml_repartitionZoltanScheme <RCB>: Repartitioning scheme to use (None)
1150: . -pc_ml_Aux <false>: Aggregate using auxiliary coordinate-based laplacian (None)
1151: - -pc_ml_AuxThreshold <0.0>: Auxiliary smoother drop tol (None)
1153: Level: intermediate
1155: Concepts: multigrid
1157: .seealso: PCCreate(), PCSetType(), PCType (for list of available types), PC, PCMGType,
1158: PCMGSetLevels(), PCMGGetLevels(), PCMGSetType(), MPSetCycles(), PCMGSetDistinctSmoothUp(),
1159: PCMGGetCoarseSolve(), PCMGSetResidual(), PCMGSetInterpolation(),
1160: PCMGSetRestriction(), PCMGGetSmoother(), PCMGGetSmootherUp(), PCMGGetSmootherDown(),
1161: PCMGSetCycleTypeOnLevel(), PCMGSetRhs(), PCMGSetX(), PCMGSetR()
1162: M*/
1164: PETSC_EXTERN PetscErrorCode PCCreate_ML(PC pc)
1165: {
1167: PC_ML *pc_ml;
1168: PC_MG *mg;
1171: /* PCML is an inherited class of PCMG. Initialize pc as PCMG */
1172: PCSetType(pc,PCMG); /* calls PCCreate_MG() and MGCreate_Private() */
1173: PetscObjectChangeTypeName((PetscObject)pc,PCML);
1174: /* Since PCMG tries to use DM assocated with PC must delete it */
1175: DMDestroy(&pc->dm);
1176: PCMGSetGalerkin(pc,PC_MG_GALERKIN_EXTERNAL);
1177: mg = (PC_MG*)pc->data;
1179: /* create a supporting struct and attach it to pc */
1180: PetscNewLog(pc,&pc_ml);
1181: mg->innerctx = pc_ml;
1183: pc_ml->ml_object = 0;
1184: pc_ml->agg_object = 0;
1185: pc_ml->gridctx = 0;
1186: pc_ml->PetscMLdata = 0;
1187: pc_ml->Nlevels = -1;
1188: pc_ml->MaxNlevels = 10;
1189: pc_ml->MaxCoarseSize = 1;
1190: pc_ml->CoarsenScheme = 1;
1191: pc_ml->Threshold = 0.0;
1192: pc_ml->DampingFactor = 4.0/3.0;
1193: pc_ml->SpectralNormScheme_Anorm = PETSC_FALSE;
1194: pc_ml->size = 0;
1195: pc_ml->dim = 0;
1196: pc_ml->nloc = 0;
1197: pc_ml->coords = 0;
1198: pc_ml->Repartition = PETSC_FALSE;
1199: pc_ml->MaxMinRatio = 1.3;
1200: pc_ml->MinPerProc = 512;
1201: pc_ml->PutOnSingleProc = 5000;
1202: pc_ml->RepartitionType = 0;
1203: pc_ml->ZoltanScheme = 0;
1204: pc_ml->Aux = PETSC_FALSE;
1205: pc_ml->AuxThreshold = 0.0;
1207: /* allow for coordinates to be passed */
1208: PetscObjectComposeFunction((PetscObject)pc,"PCSetCoordinates_C",PCSetCoordinates_ML);
1210: /* overwrite the pointers of PCMG by the functions of PCML */
1211: pc->ops->setfromoptions = PCSetFromOptions_ML;
1212: pc->ops->setup = PCSetUp_ML;
1213: pc->ops->reset = PCReset_ML;
1214: pc->ops->destroy = PCDestroy_ML;
1215: return(0);
1216: }