Actual source code: gamg.c
petsc-3.6.4 2016-04-12
1: /*
2: GAMG geometric-algebric multigrid PC - Mark Adams 2011
3: */
4: #include <petsc/private/matimpl.h>
5: #include <../src/ksp/pc/impls/gamg/gamg.h> /*I "petscpc.h" I*/
6: #include <petsc/private/kspimpl.h>
7: #include <../src/ksp/pc/impls/bjacobi/bjacobi.h> /* Hack to access same_local_solves */
9: #if defined PETSC_GAMG_USE_LOG
10: PetscLogEvent petsc_gamg_setup_events[NUM_SET];
11: #endif
13: #if defined PETSC_USE_LOG
14: PetscLogEvent PC_GAMGGraph_AGG;
15: PetscLogEvent PC_GAMGGraph_GEO;
16: PetscLogEvent PC_GAMGCoarsen_AGG;
17: PetscLogEvent PC_GAMGCoarsen_GEO;
18: PetscLogEvent PC_GAMGProlongator_AGG;
19: PetscLogEvent PC_GAMGProlongator_GEO;
20: PetscLogEvent PC_GAMGOptProlongator_AGG;
21: #endif
23: #define GAMG_MAXLEVELS 30
25: /* #define GAMG_STAGES */
26: #if (defined PETSC_GAMG_USE_LOG && defined GAMG_STAGES)
27: static PetscLogStage gamg_stages[GAMG_MAXLEVELS];
28: #endif
30: static PetscFunctionList GAMGList = 0;
31: static PetscBool PCGAMGPackageInitialized;
33: /* ----------------------------------------------------------------------------- */
36: PetscErrorCode PCReset_GAMG(PC pc)
37: {
39: PC_MG *mg = (PC_MG*)pc->data;
40: PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx;
43: if (pc_gamg->data) { /* this should not happen, cleaned up in SetUp */
44: PetscPrintf(PetscObjectComm((PetscObject)pc),"***[%d]%s this should not happen, cleaned up in SetUp\n",0,__FUNCT__);
45: PetscFree(pc_gamg->data);
46: }
47: pc_gamg->data_sz = 0;
48: PetscFree(pc_gamg->orig_data);
49: return(0);
50: }
52: /* -------------------------------------------------------------------------- */
53: /*
54: PCGAMGCreateLevel_GAMG: create coarse op with RAP. repartition and/or reduce number
55: of active processors.
57: Input Parameter:
58: . pc - parameters + side effect: coarse data in 'pc_gamg->data' and
59: 'pc_gamg->data_sz' are changed via repartitioning/reduction.
60: . Amat_fine - matrix on this fine (k) level
61: . cr_bs - coarse block size
62: In/Output Parameter:
63: . a_P_inout - prolongation operator to the next level (k-->k-1)
64: . a_nactive_proc - number of active procs
65: Output Parameter:
66: . a_Amat_crs - coarse matrix that is created (k-1)
67: */
71: static PetscErrorCode PCGAMGCreateLevel_GAMG(PC pc,Mat Amat_fine,PetscInt cr_bs,
72: Mat *a_P_inout,Mat *a_Amat_crs,PetscMPIInt *a_nactive_proc,
73: IS * Pcolumnperm)
74: {
75: PetscErrorCode ierr;
76: PC_MG *mg = (PC_MG*)pc->data;
77: PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx;
78: Mat Cmat,Pold=*a_P_inout;
79: MPI_Comm comm;
80: PetscMPIInt rank,size,new_size,nactive=*a_nactive_proc;
81: PetscInt ncrs_eq,ncrs,f_bs;
84: PetscObjectGetComm((PetscObject)Amat_fine,&comm);
85: MPI_Comm_rank(comm, &rank);
86: MPI_Comm_size(comm, &size);
87: MatGetBlockSize(Amat_fine, &f_bs);
88: MatPtAP(Amat_fine, Pold, MAT_INITIAL_MATRIX, 2.0, &Cmat);
90: /* set 'ncrs' (nodes), 'ncrs_eq' (equations)*/
91: MatGetLocalSize(Cmat, &ncrs_eq, NULL);
92: if (pc_gamg->data_cell_rows>0) {
93: ncrs = pc_gamg->data_sz/pc_gamg->data_cell_cols/pc_gamg->data_cell_rows;
94: } else {
95: PetscInt bs;
96: MatGetBlockSize(Cmat, &bs);
97: ncrs = ncrs_eq/bs;
98: }
100: /* get number of PEs to make active 'new_size', reduce, can be any integer 1-P */
101: {
102: PetscInt ncrs_eq_glob;
103: MatGetSize(Cmat, &ncrs_eq_glob, NULL);
104: new_size = (PetscMPIInt)((float)ncrs_eq_glob/(float)pc_gamg->min_eq_proc + 0.5); /* hardwire min. number of eq/proc */
105: if (new_size == 0) new_size = 1; /* not likely, posible? */
106: else if (new_size >= nactive) new_size = nactive; /* no change, rare */
107: }
109: if (Pcolumnperm) *Pcolumnperm = NULL;
111: if (!pc_gamg->repart && new_size==nactive) *a_Amat_crs = Cmat; /* output - no repartitioning or reduction - could bail here */
112: else {
113: PetscInt *counts,*newproc_idx,ii,jj,kk,strideNew,*tidx,ncrs_new,ncrs_eq_new,nloc_old;
114: IS is_eq_newproc,is_eq_num,is_eq_num_prim,new_eq_indices;
116: nloc_old = ncrs_eq/cr_bs;
117: if (ncrs_eq % cr_bs) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"ncrs_eq %D not divisible by cr_bs %D",ncrs_eq,cr_bs);
118: #if defined PETSC_GAMG_USE_LOG
119: PetscLogEventBegin(petsc_gamg_setup_events[SET12],0,0,0,0);
120: #endif
121: /* make 'is_eq_newproc' */
122: PetscMalloc1(size, &counts);
123: if (pc_gamg->repart) {
124: /* Repartition Cmat_{k} and move colums of P^{k}_{k-1} and coordinates of primal part accordingly */
125: Mat adj;
127: PetscInfo3(pc,"Repartition: size (active): %D --> %D, neq = %D\n",*a_nactive_proc,new_size,ncrs_eq);
129: /* get 'adj' */
130: if (cr_bs == 1) {
131: MatConvert(Cmat, MATMPIADJ, MAT_INITIAL_MATRIX, &adj);
132: } else {
133: /* make a scalar matrix to partition (no Stokes here) */
134: Mat tMat;
135: PetscInt Istart_crs,Iend_crs,ncols,jj,Ii;
136: const PetscScalar *vals;
137: const PetscInt *idx;
138: PetscInt *d_nnz, *o_nnz, M, N;
139: static PetscInt llev = 0;
140: MatType mtype;
142: PetscMalloc2(ncrs, &d_nnz,ncrs, &o_nnz);
143: MatGetOwnershipRange(Cmat, &Istart_crs, &Iend_crs);
144: MatGetSize(Cmat, &M, &N);
145: for (Ii = Istart_crs, jj = 0; Ii < Iend_crs; Ii += cr_bs, jj++) {
146: MatGetRow(Cmat,Ii,&ncols,0,0);
147: d_nnz[jj] = ncols/cr_bs;
148: o_nnz[jj] = ncols/cr_bs;
149: MatRestoreRow(Cmat,Ii,&ncols,0,0);
150: if (d_nnz[jj] > ncrs) d_nnz[jj] = ncrs;
151: if (o_nnz[jj] > (M/cr_bs-ncrs)) o_nnz[jj] = M/cr_bs-ncrs;
152: }
154: MatGetType(Amat_fine,&mtype);
155: MatCreate(comm, &tMat);
156: MatSetSizes(tMat, ncrs, ncrs,PETSC_DETERMINE, PETSC_DETERMINE);
157: MatSetType(tMat,mtype);
158: MatSeqAIJSetPreallocation(tMat,0,d_nnz);
159: MatMPIAIJSetPreallocation(tMat,0,d_nnz,0,o_nnz);
160: PetscFree2(d_nnz,o_nnz);
162: for (ii = Istart_crs; ii < Iend_crs; ii++) {
163: PetscInt dest_row = ii/cr_bs;
164: MatGetRow(Cmat,ii,&ncols,&idx,&vals);
165: for (jj = 0; jj < ncols; jj++) {
166: PetscInt dest_col = idx[jj]/cr_bs;
167: PetscScalar v = 1.0;
168: MatSetValues(tMat,1,&dest_row,1,&dest_col,&v,ADD_VALUES);
169: }
170: MatRestoreRow(Cmat,ii,&ncols,&idx,&vals);
171: }
172: MatAssemblyBegin(tMat,MAT_FINAL_ASSEMBLY);
173: MatAssemblyEnd(tMat,MAT_FINAL_ASSEMBLY);
175: if (llev++ == -1) {
176: PetscViewer viewer; char fname[32];
177: PetscSNPrintf(fname,sizeof(fname),"part_mat_%D.mat",llev);
178: PetscViewerBinaryOpen(comm,fname,FILE_MODE_WRITE,&viewer);
179: MatView(tMat, viewer);
180: PetscViewerDestroy(&viewer);
181: }
183: MatConvert(tMat, MATMPIADJ, MAT_INITIAL_MATRIX, &adj);
185: MatDestroy(&tMat);
186: } /* create 'adj' */
188: { /* partition: get newproc_idx */
189: char prefix[256];
190: const char *pcpre;
191: const PetscInt *is_idx;
192: MatPartitioning mpart;
193: IS proc_is;
194: PetscInt targetPE;
196: MatPartitioningCreate(comm, &mpart);
197: MatPartitioningSetAdjacency(mpart, adj);
198: PCGetOptionsPrefix(pc, &pcpre);
199: PetscSNPrintf(prefix,sizeof(prefix),"%spc_gamg_",pcpre ? pcpre : "");
200: PetscObjectSetOptionsPrefix((PetscObject)mpart,prefix);
201: MatPartitioningSetFromOptions(mpart);
202: MatPartitioningSetNParts(mpart, new_size);
203: MatPartitioningApply(mpart, &proc_is);
204: MatPartitioningDestroy(&mpart);
206: /* collect IS info */
207: PetscMalloc1(ncrs_eq, &newproc_idx);
208: ISGetIndices(proc_is, &is_idx);
209: targetPE = 1; /* bring to "front" of machine */
210: /*targetPE = size/new_size;*/ /* spread partitioning across machine */
211: for (kk = jj = 0 ; kk < nloc_old ; kk++) {
212: for (ii = 0 ; ii < cr_bs ; ii++, jj++) {
213: newproc_idx[jj] = is_idx[kk] * targetPE; /* distribution */
214: }
215: }
216: ISRestoreIndices(proc_is, &is_idx);
217: ISDestroy(&proc_is);
218: }
219: MatDestroy(&adj);
221: ISCreateGeneral(comm, ncrs_eq, newproc_idx, PETSC_COPY_VALUES, &is_eq_newproc);
222: PetscFree(newproc_idx);
223: } else { /* simple aggreagtion of parts -- 'is_eq_newproc' */
225: PetscInt rfactor,targetPE;
226: /* find factor */
227: if (new_size == 1) rfactor = size; /* easy */
228: else {
229: PetscReal best_fact = 0.;
230: jj = -1;
231: for (kk = 1 ; kk <= size ; kk++) {
232: if (size%kk==0) { /* a candidate */
233: PetscReal nactpe = (PetscReal)size/(PetscReal)kk, fact = nactpe/(PetscReal)new_size;
234: if (fact > 1.0) fact = 1./fact; /* keep fact < 1 */
235: if (fact > best_fact) {
236: best_fact = fact; jj = kk;
237: }
238: }
239: }
240: if (jj != -1) rfactor = jj;
241: else rfactor = 1; /* does this happen .. a prime */
242: }
243: new_size = size/rfactor;
245: if (new_size==nactive) {
246: *a_Amat_crs = Cmat; /* output - no repartitioning or reduction, bail out because nested here */
247: PetscFree(counts);
248: PetscInfo2(pc,"Aggregate processors noop: new_size=%D, neq(loc)=%D\n",new_size,ncrs_eq);
249: return(0);
250: }
252: PetscInfo1(pc,"Number of equations (loc) %D with simple aggregation\n",ncrs_eq);
253: targetPE = rank/rfactor;
254: ISCreateStride(comm, ncrs_eq, targetPE, 0, &is_eq_newproc);
255: } /* end simple 'is_eq_newproc' */
257: /*
258: Create an index set from the is_eq_newproc index set to indicate the mapping TO
259: */
260: ISPartitioningToNumbering(is_eq_newproc, &is_eq_num);
261: is_eq_num_prim = is_eq_num;
262: /*
263: Determine how many equations/vertices are assigned to each processor
264: */
265: ISPartitioningCount(is_eq_newproc, size, counts);
266: ncrs_eq_new = counts[rank];
267: ISDestroy(&is_eq_newproc);
268: ncrs_new = ncrs_eq_new/cr_bs; /* eqs */
270: PetscFree(counts);
271: #if defined PETSC_GAMG_USE_LOG
272: PetscLogEventEnd(petsc_gamg_setup_events[SET12],0,0,0,0);
273: #endif
274: /* data movement scope -- this could be moved to subclasses so that we don't try to cram all auxilary data into some complex abstracted thing */
275: {
276: Vec src_crd, dest_crd;
277: const PetscInt *idx,ndata_rows=pc_gamg->data_cell_rows,ndata_cols=pc_gamg->data_cell_cols,node_data_sz=ndata_rows*ndata_cols;
278: VecScatter vecscat;
279: PetscScalar *array;
280: IS isscat;
282: /* move data (for primal equations only) */
283: /* Create a vector to contain the newly ordered element information */
284: VecCreate(comm, &dest_crd);
285: VecSetSizes(dest_crd, node_data_sz*ncrs_new, PETSC_DECIDE);
286: VecSetType(dest_crd,VECSTANDARD); /* this is needed! */
287: /*
288: There are 'ndata_rows*ndata_cols' data items per node, (one can think of the vectors of having
289: a block size of ...). Note, ISs are expanded into equation space by 'cr_bs'.
290: */
291: PetscMalloc1(ncrs*node_data_sz, &tidx);
292: ISGetIndices(is_eq_num_prim, &idx);
293: for (ii=0,jj=0; ii<ncrs; ii++) {
294: PetscInt id = idx[ii*cr_bs]/cr_bs; /* get node back */
295: for (kk=0; kk<node_data_sz; kk++, jj++) tidx[jj] = id*node_data_sz + kk;
296: }
297: ISRestoreIndices(is_eq_num_prim, &idx);
298: ISCreateGeneral(comm, node_data_sz*ncrs, tidx, PETSC_COPY_VALUES, &isscat);
299: PetscFree(tidx);
300: /*
301: Create a vector to contain the original vertex information for each element
302: */
303: VecCreateSeq(PETSC_COMM_SELF, node_data_sz*ncrs, &src_crd);
304: for (jj=0; jj<ndata_cols; jj++) {
305: const PetscInt stride0=ncrs*pc_gamg->data_cell_rows;
306: for (ii=0; ii<ncrs; ii++) {
307: for (kk=0; kk<ndata_rows; kk++) {
308: PetscInt ix = ii*ndata_rows + kk + jj*stride0, jx = ii*node_data_sz + kk*ndata_cols + jj;
309: PetscScalar tt = (PetscScalar)pc_gamg->data[ix];
310: VecSetValues(src_crd, 1, &jx, &tt, INSERT_VALUES);
311: }
312: }
313: }
314: VecAssemblyBegin(src_crd);
315: VecAssemblyEnd(src_crd);
316: /*
317: Scatter the element vertex information (still in the original vertex ordering)
318: to the correct processor
319: */
320: VecScatterCreate(src_crd, NULL, dest_crd, isscat, &vecscat);
321: ISDestroy(&isscat);
322: VecScatterBegin(vecscat,src_crd,dest_crd,INSERT_VALUES,SCATTER_FORWARD);
323: VecScatterEnd(vecscat,src_crd,dest_crd,INSERT_VALUES,SCATTER_FORWARD);
324: VecScatterDestroy(&vecscat);
325: VecDestroy(&src_crd);
326: /*
327: Put the element vertex data into a new allocation of the gdata->ele
328: */
329: PetscFree(pc_gamg->data);
330: PetscMalloc1(node_data_sz*ncrs_new, &pc_gamg->data);
332: pc_gamg->data_sz = node_data_sz*ncrs_new;
333: strideNew = ncrs_new*ndata_rows;
335: VecGetArray(dest_crd, &array);
336: for (jj=0; jj<ndata_cols; jj++) {
337: for (ii=0; ii<ncrs_new; ii++) {
338: for (kk=0; kk<ndata_rows; kk++) {
339: PetscInt ix = ii*ndata_rows + kk + jj*strideNew, jx = ii*node_data_sz + kk*ndata_cols + jj;
340: pc_gamg->data[ix] = PetscRealPart(array[jx]);
341: }
342: }
343: }
344: VecRestoreArray(dest_crd, &array);
345: VecDestroy(&dest_crd);
346: }
347: /* move A and P (columns) with new layout */
348: #if defined PETSC_GAMG_USE_LOG
349: PetscLogEventBegin(petsc_gamg_setup_events[SET13],0,0,0,0);
350: #endif
352: /*
353: Invert for MatGetSubMatrix
354: */
355: ISInvertPermutation(is_eq_num, ncrs_eq_new, &new_eq_indices);
356: ISSort(new_eq_indices); /* is this needed? */
357: ISSetBlockSize(new_eq_indices, cr_bs);
358: if (is_eq_num != is_eq_num_prim) {
359: ISDestroy(&is_eq_num_prim); /* could be same as 'is_eq_num' */
360: }
361: if (Pcolumnperm) {
362: PetscObjectReference((PetscObject)new_eq_indices);
363: *Pcolumnperm = new_eq_indices;
364: }
365: ISDestroy(&is_eq_num);
366: #if defined PETSC_GAMG_USE_LOG
367: PetscLogEventEnd(petsc_gamg_setup_events[SET13],0,0,0,0);
368: PetscLogEventBegin(petsc_gamg_setup_events[SET14],0,0,0,0);
369: #endif
370: /* 'a_Amat_crs' output */
371: {
372: Mat mat;
373: MatGetSubMatrix(Cmat, new_eq_indices, new_eq_indices, MAT_INITIAL_MATRIX, &mat);
374: *a_Amat_crs = mat;
376: if (!PETSC_TRUE) {
377: PetscInt cbs, rbs;
378: MatGetBlockSizes(Cmat, &rbs, &cbs);
379: PetscPrintf(MPI_COMM_SELF,"[%d]%s Old Mat rbs=%d cbs=%d\n",rank,__FUNCT__,rbs,cbs);
380: MatGetBlockSizes(mat, &rbs, &cbs);
381: PetscPrintf(MPI_COMM_SELF,"[%d]%s New Mat rbs=%d cbs=%d cr_bs=%d\n",rank,__FUNCT__,rbs,cbs,cr_bs);
382: }
383: }
384: MatDestroy(&Cmat);
386: #if defined PETSC_GAMG_USE_LOG
387: PetscLogEventEnd(petsc_gamg_setup_events[SET14],0,0,0,0);
388: #endif
389: /* prolongator */
390: {
391: IS findices;
392: PetscInt Istart,Iend;
393: Mat Pnew;
394: MatGetOwnershipRange(Pold, &Istart, &Iend);
395: #if defined PETSC_GAMG_USE_LOG
396: PetscLogEventBegin(petsc_gamg_setup_events[SET15],0,0,0,0);
397: #endif
398: ISCreateStride(comm,Iend-Istart,Istart,1,&findices);
399: ISSetBlockSize(findices,f_bs);
400: MatGetSubMatrix(Pold, findices, new_eq_indices, MAT_INITIAL_MATRIX, &Pnew);
401: ISDestroy(&findices);
403: if (!PETSC_TRUE) {
404: PetscInt cbs, rbs;
405: MatGetBlockSizes(Pold, &rbs, &cbs);
406: PetscPrintf(MPI_COMM_SELF,"[%d]%s Pold rbs=%d cbs=%d\n",rank,__FUNCT__,rbs,cbs);
407: MatGetBlockSizes(Pnew, &rbs, &cbs);
408: PetscPrintf(MPI_COMM_SELF,"[%d]%s Pnew rbs=%d cbs=%d\n",rank,__FUNCT__,rbs,cbs);
409: }
410: #if defined PETSC_GAMG_USE_LOG
411: PetscLogEventEnd(petsc_gamg_setup_events[SET15],0,0,0,0);
412: #endif
413: MatDestroy(a_P_inout);
415: /* output - repartitioned */
416: *a_P_inout = Pnew;
417: }
418: ISDestroy(&new_eq_indices);
420: *a_nactive_proc = new_size; /* output */
421: }
423: /* outout matrix data */
424: if (!PETSC_TRUE) {
425: PetscViewer viewer; char fname[32]; static int llev=0; Cmat = *a_Amat_crs;
426: if (llev==0) {
427: sprintf(fname,"Cmat_%d.m",llev++);
428: PetscViewerASCIIOpen(comm,fname,&viewer);
429: PetscViewerSetFormat(viewer, PETSC_VIEWER_ASCII_MATLAB);
430: MatView(Amat_fine, viewer);
431: PetscViewerDestroy(&viewer);
432: }
433: sprintf(fname,"Cmat_%d.m",llev++);
434: PetscViewerASCIIOpen(comm,fname,&viewer);
435: PetscViewerSetFormat(viewer, PETSC_VIEWER_ASCII_MATLAB);
436: MatView(Cmat, viewer);
437: PetscViewerDestroy(&viewer);
438: }
439: return(0);
440: }
442: /* -------------------------------------------------------------------------- */
443: /*
444: PCSetUp_GAMG - Prepares for the use of the GAMG preconditioner
445: by setting data structures and options.
447: Input Parameter:
448: . pc - the preconditioner context
450: */
453: PetscErrorCode PCSetUp_GAMG(PC pc)
454: {
456: PC_MG *mg = (PC_MG*)pc->data;
457: PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx;
458: Mat Pmat = pc->pmat;
459: PetscInt fine_level,level,level1,bs,M,qq,lidx,nASMBlocksArr[GAMG_MAXLEVELS];
460: MPI_Comm comm;
461: PetscMPIInt rank,size,nactivepe;
462: Mat Aarr[GAMG_MAXLEVELS],Parr[GAMG_MAXLEVELS];
463: PetscReal emaxs[GAMG_MAXLEVELS];
464: IS *ASMLocalIDsArr[GAMG_MAXLEVELS];
465: PetscLogDouble nnz0=0.,nnztot=0.;
466: MatInfo info;
469: PetscObjectGetComm((PetscObject)pc,&comm);
470: MPI_Comm_rank(comm,&rank);
471: MPI_Comm_size(comm,&size);
473: if (pc_gamg->setup_count++ > 0) {
474: if ((PetscBool)(!pc_gamg->reuse_prol)) {
475: /* reset everything */
476: PCReset_MG(pc);
477: pc->setupcalled = 0;
478: } else {
479: PC_MG_Levels **mglevels = mg->levels;
480: /* just do Galerkin grids */
481: Mat B,dA,dB;
483: if (!pc->setupcalled) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"PCSetUp() has not been called yet");
484: if (pc_gamg->Nlevels > 1) {
485: /* currently only handle case where mat and pmat are the same on coarser levels */
486: KSPGetOperators(mglevels[pc_gamg->Nlevels-1]->smoothd,&dA,&dB);
487: /* (re)set to get dirty flag */
488: KSPSetOperators(mglevels[pc_gamg->Nlevels-1]->smoothd,dA,dB);
490: for (level=pc_gamg->Nlevels-2; level>=0; level--) {
491: /* the first time through the matrix structure has changed from repartitioning */
492: if (pc_gamg->setup_count==2) {
493: MatPtAP(dB,mglevels[level+1]->interpolate,MAT_INITIAL_MATRIX,1.0,&B);
494: MatDestroy(&mglevels[level]->A);
496: mglevels[level]->A = B;
497: } else {
498: KSPGetOperators(mglevels[level]->smoothd,NULL,&B);
499: MatPtAP(dB,mglevels[level+1]->interpolate,MAT_REUSE_MATRIX,1.0,&B);
500: }
501: KSPSetOperators(mglevels[level]->smoothd,B,B);
502: dB = B;
503: }
504: }
506: PCSetUp_MG(pc);
508: return(0);
509: }
510: }
512: if (!pc_gamg->data) {
513: if (pc_gamg->orig_data) {
514: MatGetBlockSize(Pmat, &bs);
515: MatGetLocalSize(Pmat, &qq, NULL);
517: pc_gamg->data_sz = (qq/bs)*pc_gamg->orig_data_cell_rows*pc_gamg->orig_data_cell_cols;
518: pc_gamg->data_cell_rows = pc_gamg->orig_data_cell_rows;
519: pc_gamg->data_cell_cols = pc_gamg->orig_data_cell_cols;
521: PetscMalloc1(pc_gamg->data_sz, &pc_gamg->data);
522: for (qq=0; qq<pc_gamg->data_sz; qq++) pc_gamg->data[qq] = pc_gamg->orig_data[qq];
523: } else {
524: if (!pc_gamg->ops->createdefaultdata) SETERRQ(comm,PETSC_ERR_PLIB,"'createdefaultdata' not set(?) need to support NULL data");
525: pc_gamg->ops->createdefaultdata(pc,Pmat);
526: }
527: }
529: /* cache original data for reuse */
530: if (!pc_gamg->orig_data && (PetscBool)(!pc_gamg->reuse_prol)) {
531: PetscMalloc1(pc_gamg->data_sz, &pc_gamg->orig_data);
532: for (qq=0; qq<pc_gamg->data_sz; qq++) pc_gamg->orig_data[qq] = pc_gamg->data[qq];
533: pc_gamg->orig_data_cell_rows = pc_gamg->data_cell_rows;
534: pc_gamg->orig_data_cell_cols = pc_gamg->data_cell_cols;
535: }
537: /* get basic dims */
538: MatGetBlockSize(Pmat, &bs);
539: MatGetSize(Pmat, &M, &qq);
541: MatGetInfo(Pmat,MAT_GLOBAL_SUM,&info); /* global reduction */
542: nnz0 = info.nz_used;
543: nnztot = info.nz_used;
544: PetscInfo6(pc,"level %d) N=%D, n data rows=%d, n data cols=%d, nnz/row (ave)=%d, np=%d\n",
545: 0,M,pc_gamg->data_cell_rows,pc_gamg->data_cell_cols,
546: (int)(nnz0/(PetscReal)M+0.5),size);
547:
549: /* Get A_i and R_i */
550: for (level=0, Aarr[0]=Pmat, nactivepe = size; /* hard wired stopping logic */
551: level < (pc_gamg->Nlevels-1) && (level==0 || M>pc_gamg->coarse_eq_limit);
552: level++) {
553: pc_gamg->current_level = level;
554: level1 = level + 1;
555: #if defined PETSC_GAMG_USE_LOG
556: PetscLogEventBegin(petsc_gamg_setup_events[SET1],0,0,0,0);
557: #if (defined GAMG_STAGES)
558: PetscLogStagePush(gamg_stages[level]);
559: #endif
560: #endif
561: { /* construct prolongator */
562: Mat Gmat;
563: PetscCoarsenData *agg_lists;
564: Mat Prol11;
566: pc_gamg->ops->graph(pc,Aarr[level], &Gmat);
567: pc_gamg->ops->coarsen(pc, &Gmat, &agg_lists);
568: pc_gamg->ops->prolongator(pc,Aarr[level],Gmat,agg_lists,&Prol11);
570: /* could have failed to create new level */
571: if (Prol11) {
572: /* get new block size of coarse matrices */
573: MatGetBlockSizes(Prol11, NULL, &bs);
575: if (pc_gamg->ops->optprolongator) {
576: /* smooth */
577: pc_gamg->ops->optprolongator(pc, Aarr[level], &Prol11);
578: }
580: Parr[level1] = Prol11;
581: } else Parr[level1] = NULL;
583: if (pc_gamg->use_aggs_in_gasm) {
584: PetscInt bs;
585: MatGetBlockSizes(Prol11, &bs, NULL);
586: PetscCDGetASMBlocks(agg_lists, bs, &nASMBlocksArr[level], &ASMLocalIDsArr[level]);
587: }
589: MatDestroy(&Gmat);
590: PetscCDDestroy(agg_lists);
591: } /* construct prolongator scope */
592: #if defined PETSC_GAMG_USE_LOG
593: PetscLogEventEnd(petsc_gamg_setup_events[SET1],0,0,0,0);
594: #endif
595: /* cache eigen estimate */
596: if (pc_gamg->emax_id != -1) {
597: PetscBool flag;
598: PetscObjectComposedDataGetReal((PetscObject)Aarr[level], pc_gamg->emax_id, emaxs[level], flag);
599: if (!flag) emaxs[level] = -1.;
600: } else emaxs[level] = -1.;
601: if (level==0) Aarr[0] = Pmat; /* use Pmat for finest level setup */
602: if (!Parr[level1]) {
603: PetscInfo1(pc,"Stop gridding, level %D\n",level);
604: #if (defined PETSC_GAMG_USE_LOG && defined GAMG_STAGES)
605: PetscLogStagePop();
606: #endif
607: break;
608: }
609: #if defined PETSC_GAMG_USE_LOG
610: PetscLogEventBegin(petsc_gamg_setup_events[SET2],0,0,0,0);
611: #endif
613: pc_gamg->ops->createlevel(pc, Aarr[level], bs,&Parr[level1], &Aarr[level1], &nactivepe, NULL);
615: #if defined PETSC_GAMG_USE_LOG
616: PetscLogEventEnd(petsc_gamg_setup_events[SET2],0,0,0,0);
617: #endif
618: MatGetSize(Aarr[level1], &M, &qq);
620: MatGetInfo(Aarr[level1], MAT_GLOBAL_SUM, &info);
621: nnztot += info.nz_used;
622: PetscInfo5(pc,"%d) N=%D, n data cols=%d, nnz/row (ave)=%d, %d active pes\n",level1,M,pc_gamg->data_cell_cols,(int)(info.nz_used/(PetscReal)M),nactivepe);
624: #if (defined PETSC_GAMG_USE_LOG && defined GAMG_STAGES)
625: PetscLogStagePop();
626: #endif
627: /* stop if one node or one proc -- could pull back for singular problems */
628: if ( (pc_gamg->data_cell_cols && M/pc_gamg->data_cell_cols < 2) || (!pc_gamg->data_cell_cols && M/bs < 2) ) {
629: PetscInfo2(pc,"HARD stop of coarsening on level %D. Grid too small: %D block nodes\n",level,M/bs);
630: level++;
631: break;
632: }
633: } /* levels */
634: PetscFree(pc_gamg->data);
636: PetscInfo2(pc,"%D levels, grid complexity = %g\n",level+1,nnztot/nnz0);
637: pc_gamg->Nlevels = level + 1;
638: fine_level = level;
639: PCMGSetLevels(pc,pc_gamg->Nlevels,NULL);
641: /* simple setup */
642: if (!PETSC_TRUE) {
643: PC_MG_Levels **mglevels = mg->levels;
644: for (lidx=0,level=pc_gamg->Nlevels-1;
645: lidx<fine_level;
646: lidx++, level--) {
647: PCMGSetInterpolation(pc, lidx+1, Parr[level]);
648: KSPSetOperators(mglevels[lidx]->smoothd, Aarr[level], Aarr[level]);
649: MatDestroy(&Parr[level]);
650: MatDestroy(&Aarr[level]);
651: }
652: KSPSetOperators(mglevels[fine_level]->smoothd, Aarr[0], Aarr[0]);
654: PCSetUp_MG(pc);
655: } else if (pc_gamg->Nlevels > 1) { /* don't setup MG if one level */
656: /* set default smoothers & set operators */
657: for (lidx = 1, level = pc_gamg->Nlevels-2;
658: lidx <= fine_level;
659: lidx++, level--) {
660: KSP smoother;
661: PC subpc;
663: PCMGGetSmoother(pc, lidx, &smoother);
664: KSPGetPC(smoother, &subpc);
666: KSPSetNormType(smoother, KSP_NORM_NONE);
667: /* set ops */
668: KSPSetOperators(smoother, Aarr[level], Aarr[level]);
669: PCMGSetInterpolation(pc, lidx, Parr[level+1]);
671: /* set defaults */
672: KSPSetType(smoother, KSPCHEBYSHEV);
674: /* set blocks for GASM smoother that uses the 'aggregates' */
675: if (pc_gamg->use_aggs_in_gasm) {
676: PetscInt sz;
677: IS *is;
679: sz = nASMBlocksArr[level];
680: is = ASMLocalIDsArr[level];
681: PCSetType(subpc, PCGASM);
682: PCGASMSetOverlap(subpc, 0);
683: if (sz==0) {
684: IS is;
685: PetscInt my0,kk;
686: MatGetOwnershipRange(Aarr[level], &my0, &kk);
687: ISCreateGeneral(PETSC_COMM_SELF, 1, &my0, PETSC_COPY_VALUES, &is);
688: PCGASMSetSubdomains(subpc, 1, &is, NULL);
689: ISDestroy(&is);
690: } else {
691: PetscInt kk;
692: PCGASMSetSubdomains(subpc, sz, is, NULL);
693: for (kk=0; kk<sz; kk++) {
694: ISDestroy(&is[kk]);
695: }
696: PetscFree(is);
697: }
698: ASMLocalIDsArr[level] = NULL;
699: nASMBlocksArr[level] = 0;
700: PCGASMSetType(subpc, PC_GASM_BASIC);
701: } else {
702: PCSetType(subpc, PCSOR);
703: }
704: }
705: {
706: /* coarse grid */
707: KSP smoother,*k2; PC subpc,pc2; PetscInt ii,first;
708: Mat Lmat = Aarr[(level=pc_gamg->Nlevels-1)]; lidx = 0;
709: PCMGGetSmoother(pc, lidx, &smoother);
710: KSPSetOperators(smoother, Lmat, Lmat);
711: KSPSetNormType(smoother, KSP_NORM_NONE);
712: KSPGetPC(smoother, &subpc);
713: PCSetType(subpc, PCBJACOBI);
714: PCSetUp(subpc);
715: PCBJacobiGetSubKSP(subpc,&ii,&first,&k2);
716: if (ii != 1) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_PLIB,"ii %D is not one",ii);
717: KSPGetPC(k2[0],&pc2);
718: PCSetType(pc2, PCLU);
719: PCFactorSetShiftType(pc2,MAT_SHIFT_INBLOCKS);
720: KSPSetTolerances(k2[0],PETSC_DEFAULT,PETSC_DEFAULT,PETSC_DEFAULT,1);
721: /* This flag gets reset by PCBJacobiGetSubKSP(), but our BJacobi really does the same algorithm everywhere (and in
722: * fact, all but one process will have zero dofs), so we reset the flag to avoid having PCView_BJacobi attempt to
723: * view every subdomain as though they were different. */
724: ((PC_BJacobi*)subpc->data)->same_local_solves = PETSC_TRUE;
725: }
727: /* should be called in PCSetFromOptions_GAMG(), but cannot be called prior to PCMGSetLevels() */
728: PetscObjectOptionsBegin((PetscObject)pc);
729: PCSetFromOptions_MG(PetscOptionsObject,pc);
730: PetscOptionsEnd();
731: if (!mg->galerkin) SETERRQ(comm,PETSC_ERR_USER,"PCGAMG must use Galerkin for coarse operators.");
732: if (mg->galerkin == 1) mg->galerkin = 2;
734: /* create cheby smoothers */
735: for (lidx = 1, level = pc_gamg->Nlevels-2; lidx <= fine_level; lidx++, level--) {
736: KSP smoother;
737: PetscBool flag,flag2;
738: PC subpc;
740: PCMGGetSmoother(pc, lidx, &smoother);
741: KSPGetPC(smoother, &subpc);
743: /* do my own cheby */
744: PetscObjectTypeCompare((PetscObject)smoother, KSPCHEBYSHEV, &flag);
745: if (0 && flag) {
746: PetscReal emax, emin;
747: PetscObjectTypeCompare((PetscObject)subpc, PCJACOBI, &flag);
748: PetscObjectTypeCompare((PetscObject)subpc, PCSOR, &flag2);
749: /* eigen estimate only for diagnal PC but lets acccept SOR because it is close and safe (always lower) */
750: if ((flag||flag2) && (emax=emaxs[level]) > 0.0) {
751: PetscInt N1, N0;
752: emax=emaxs[level];
753: MatGetSize(Aarr[level], &N1, NULL);
754: MatGetSize(Aarr[level+1], &N0, NULL);
755: emin = emax * pc_gamg->eigtarget[0];
756: emax *= pc_gamg->eigtarget[1];
757: KSPChebyshevSetEigenvalues(smoother, emax, emin);
758: }
759: } /* setup checby flag */
760: } /* non-coarse levels */
762: /* clean up */
763: for (level=1; level<pc_gamg->Nlevels; level++) {
764: MatDestroy(&Parr[level]);
765: MatDestroy(&Aarr[level]);
766: }
768: PCSetUp_MG(pc);
769: } else {
770: KSP smoother;
771: PetscInfo(pc,"One level solver used (system is seen as DD). Using default solver.\n");
772: PCMGGetSmoother(pc, 0, &smoother);
773: KSPSetOperators(smoother, Aarr[0], Aarr[0]);
774: KSPSetType(smoother, KSPPREONLY);
775: PCSetUp_MG(pc);
776: }
777: return(0);
778: }
780: /* ------------------------------------------------------------------------- */
781: /*
782: PCDestroy_GAMG - Destroys the private context for the GAMG preconditioner
783: that was created with PCCreate_GAMG().
785: Input Parameter:
786: . pc - the preconditioner context
788: Application Interface Routine: PCDestroy()
789: */
792: PetscErrorCode PCDestroy_GAMG(PC pc)
793: {
795: PC_MG *mg = (PC_MG*)pc->data;
796: PC_GAMG *pc_gamg= (PC_GAMG*)mg->innerctx;
799: PCReset_GAMG(pc);
800: if (pc_gamg->ops->destroy) {
801: (*pc_gamg->ops->destroy)(pc);
802: }
803: PetscFree(pc_gamg->ops);
804: PetscFree(pc_gamg->gamg_type_name);
805: PetscFree(pc_gamg);
806: PCDestroy_MG(pc);
807: return(0);
808: }
813: /*@
814: PCGAMGSetProcEqLim - Set number of equations to aim for on coarse grids via processor reduction.
816: Logically Collective on PC
818: Input Parameters:
819: + pc - the preconditioner context
820: - n - the number of equations
823: Options Database Key:
824: . -pc_gamg_process_eq_limit <limit>
826: Level: intermediate
828: Concepts: Unstructured multigrid preconditioner
830: .seealso: PCGAMGSetCoarseEqLim()
831: @*/
832: PetscErrorCode PCGAMGSetProcEqLim(PC pc, PetscInt n)
833: {
838: PetscTryMethod(pc,"PCGAMGSetProcEqLim_C",(PC,PetscInt),(pc,n));
839: return(0);
840: }
844: static PetscErrorCode PCGAMGSetProcEqLim_GAMG(PC pc, PetscInt n)
845: {
846: PC_MG *mg = (PC_MG*)pc->data;
847: PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx;
850: if (n>0) pc_gamg->min_eq_proc = n;
851: return(0);
852: }
856: /*@
857: PCGAMGSetCoarseEqLim - Set max number of equations on coarse grids.
859: Collective on PC
861: Input Parameters:
862: + pc - the preconditioner context
863: - n - maximum number of equations to aim for
865: Options Database Key:
866: . -pc_gamg_coarse_eq_limit <limit>
868: Level: intermediate
870: Concepts: Unstructured multigrid preconditioner
872: .seealso: PCGAMGSetProcEqLim()
873: @*/
874: PetscErrorCode PCGAMGSetCoarseEqLim(PC pc, PetscInt n)
875: {
880: PetscTryMethod(pc,"PCGAMGSetCoarseEqLim_C",(PC,PetscInt),(pc,n));
881: return(0);
882: }
886: static PetscErrorCode PCGAMGSetCoarseEqLim_GAMG(PC pc, PetscInt n)
887: {
888: PC_MG *mg = (PC_MG*)pc->data;
889: PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx;
892: if (n>0) pc_gamg->coarse_eq_limit = n;
893: return(0);
894: }
898: /*@
899: PCGAMGSetRepartitioning - Repartition the coarse grids
901: Collective on PC
903: Input Parameters:
904: + pc - the preconditioner context
905: - n - PETSC_TRUE or PETSC_FALSE
907: Options Database Key:
908: . -pc_gamg_repartition <true,false>
910: Level: intermediate
912: Concepts: Unstructured multigrid preconditioner
914: .seealso: ()
915: @*/
916: PetscErrorCode PCGAMGSetRepartitioning(PC pc, PetscBool n)
917: {
922: PetscTryMethod(pc,"PCGAMGSetRepartitioning_C",(PC,PetscBool),(pc,n));
923: return(0);
924: }
928: static PetscErrorCode PCGAMGSetRepartitioning_GAMG(PC pc, PetscBool n)
929: {
930: PC_MG *mg = (PC_MG*)pc->data;
931: PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx;
934: pc_gamg->repart = n;
935: return(0);
936: }
940: /*@
941: PCGAMGSetReuseInterpolation - Reuse prolongation when rebuilding preconditioner
943: Collective on PC
945: Input Parameters:
946: + pc - the preconditioner context
947: - n - PETSC_TRUE or PETSC_FALSE
949: Options Database Key:
950: . -pc_gamg_reuse_interpolation <true,false>
952: Level: intermediate
954: Concepts: Unstructured multigrid preconditioner
956: .seealso: ()
957: @*/
958: PetscErrorCode PCGAMGSetReuseInterpolation(PC pc, PetscBool n)
959: {
964: PetscTryMethod(pc,"PCGAMGSetReuseInterpolation_C",(PC,PetscBool),(pc,n));
965: return(0);
966: }
970: static PetscErrorCode PCGAMGSetReuseInterpolation_GAMG(PC pc, PetscBool n)
971: {
972: PC_MG *mg = (PC_MG*)pc->data;
973: PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx;
976: pc_gamg->reuse_prol = n;
977: return(0);
978: }
982: /*@
983: PCGAMGSetUseASMAggs -
985: Collective on PC
987: Input Parameters:
988: . pc - the preconditioner context
991: Options Database Key:
992: . -pc_gamg_use_agg_gasm
994: Level: intermediate
996: Concepts: Unstructured multigrid preconditioner
998: .seealso: ()
999: @*/
1000: PetscErrorCode PCGAMGSetUseASMAggs(PC pc, PetscBool n)
1001: {
1006: PetscTryMethod(pc,"PCGAMGSetUseASMAggs_C",(PC,PetscBool),(pc,n));
1007: return(0);
1008: }
1012: static PetscErrorCode PCGAMGSetUseASMAggs_GAMG(PC pc, PetscBool n)
1013: {
1014: PC_MG *mg = (PC_MG*)pc->data;
1015: PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx;
1018: pc_gamg->use_aggs_in_gasm = n;
1019: return(0);
1020: }
1024: /*@
1025: PCGAMGSetNlevels - Sets the maximum number of levels PCGAMG will use
1027: Not collective on PC
1029: Input Parameters:
1030: + pc - the preconditioner
1031: - n - the maximum number of levels to use
1033: Options Database Key:
1034: . -pc_mg_levels
1036: Level: intermediate
1038: Concepts: Unstructured multigrid preconditioner
1040: .seealso: ()
1041: @*/
1042: PetscErrorCode PCGAMGSetNlevels(PC pc, PetscInt n)
1043: {
1048: PetscTryMethod(pc,"PCGAMGSetNlevels_C",(PC,PetscInt),(pc,n));
1049: return(0);
1050: }
1054: static PetscErrorCode PCGAMGSetNlevels_GAMG(PC pc, PetscInt n)
1055: {
1056: PC_MG *mg = (PC_MG*)pc->data;
1057: PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx;
1060: pc_gamg->Nlevels = n;
1061: return(0);
1062: }
1066: /*@
1067: PCGAMGSetThreshold - Relative threshold to use for dropping edges in aggregation graph
1069: Not collective on PC
1071: Input Parameters:
1072: + pc - the preconditioner context
1073: - threshold - the threshold value, 0.0 means keep all nonzero entries in the graph; negative means keep even zero entries in the graph
1075: Options Database Key:
1076: . -pc_gamg_threshold <threshold>
1078: Level: intermediate
1080: Concepts: Unstructured multigrid preconditioner
1082: .seealso: ()
1083: @*/
1084: PetscErrorCode PCGAMGSetThreshold(PC pc, PetscReal n)
1085: {
1090: PetscTryMethod(pc,"PCGAMGSetThreshold_C",(PC,PetscReal),(pc,n));
1091: return(0);
1092: }
1096: static PetscErrorCode PCGAMGSetThreshold_GAMG(PC pc, PetscReal n)
1097: {
1098: PC_MG *mg = (PC_MG*)pc->data;
1099: PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx;
1102: pc_gamg->threshold = n;
1103: return(0);
1104: }
1108: /*@
1109: PCGAMGSetType - Set solution method
1111: Collective on PC
1113: Input Parameters:
1114: + pc - the preconditioner context
1115: - type - PCGAMGAGG, PCGAMGGEO, or PCGAMGCLASSICAL
1117: Options Database Key:
1118: . -pc_gamg_type <agg,geo,classical>
1120: Level: intermediate
1122: Concepts: Unstructured multigrid preconditioner
1124: .seealso: PCGAMGGetType(), PCGAMG
1125: @*/
1126: PetscErrorCode PCGAMGSetType(PC pc, PCGAMGType type)
1127: {
1132: PetscTryMethod(pc,"PCGAMGSetType_C",(PC,PCGAMGType),(pc,type));
1133: return(0);
1134: }
1138: /*@
1139: PCGAMGGetType - Get solution method
1141: Collective on PC
1143: Input Parameter:
1144: . pc - the preconditioner context
1146: Output Parameter:
1147: . type - the type of algorithm used
1149: Level: intermediate
1151: Concepts: Unstructured multigrid preconditioner
1153: .seealso: PCGAMGSetType(), PCGAMGType
1154: @*/
1155: PetscErrorCode PCGAMGGetType(PC pc, PCGAMGType *type)
1156: {
1161: PetscUseMethod(pc,"PCGAMGGetType_C",(PC,PCGAMGType*),(pc,type));
1162: return(0);
1163: }
1167: static PetscErrorCode PCGAMGGetType_GAMG(PC pc, PCGAMGType *type)
1168: {
1169: PC_MG *mg = (PC_MG*)pc->data;
1170: PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx;
1173: *type = pc_gamg->type;
1174: return(0);
1175: }
1179: static PetscErrorCode PCGAMGSetType_GAMG(PC pc, PCGAMGType type)
1180: {
1181: PetscErrorCode ierr,(*r)(PC);
1182: PC_MG *mg = (PC_MG*)pc->data;
1183: PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx;
1186: pc_gamg->type = type;
1187: PetscFunctionListFind(GAMGList,type,&r);
1188: if (!r) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_UNKNOWN_TYPE,"Unknown GAMG type %s given",type);
1189: if (pc_gamg->ops->destroy) {
1190: (*pc_gamg->ops->destroy)(pc);
1191: PetscMemzero(pc_gamg->ops,sizeof(struct _PCGAMGOps));
1192: pc_gamg->ops->createlevel = PCGAMGCreateLevel_GAMG;
1193: /* cleaning up common data in pc_gamg - this should disapear someday */
1194: pc_gamg->data_cell_cols = 0;
1195: pc_gamg->data_cell_rows = 0;
1196: pc_gamg->orig_data_cell_cols = 0;
1197: pc_gamg->orig_data_cell_rows = 0;
1198: PetscFree(pc_gamg->data);
1199: pc_gamg->data_sz = 0;
1200: }
1201: PetscFree(pc_gamg->gamg_type_name);
1202: PetscStrallocpy(type,&pc_gamg->gamg_type_name);
1203: (*r)(pc);
1204: return(0);
1205: }
1209: static PetscErrorCode PCView_GAMG(PC pc,PetscViewer viewer)
1210: {
1212: PC_MG *mg = (PC_MG*)pc->data;
1213: PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx;
1216: PetscViewerASCIIPrintf(viewer," GAMG specific options\n");
1217: PetscViewerASCIIPrintf(viewer," Threshold for dropping small values from graph %g\n",(double)pc_gamg->threshold);
1218: if (pc_gamg->ops->view) {
1219: (*pc_gamg->ops->view)(pc,viewer);
1220: }
1221: return(0);
1222: }
1226: PetscErrorCode PCSetFromOptions_GAMG(PetscOptions *PetscOptionsObject,PC pc)
1227: {
1229: PC_MG *mg = (PC_MG*)pc->data;
1230: PC_GAMG *pc_gamg = (PC_GAMG*)mg->innerctx;
1231: PetscBool flag;
1232: PetscInt two = 2;
1233: MPI_Comm comm;
1236: PetscObjectGetComm((PetscObject)pc,&comm);
1237: PetscOptionsHead(PetscOptionsObject,"GAMG options");
1238: {
1239: char tname[256];
1240: PetscOptionsFList("-pc_gamg_type","Type of AMG method","PCGAMGSetType",GAMGList, pc_gamg->gamg_type_name, tname, sizeof(tname), &flag);
1241: if (flag) {
1242: PCGAMGSetType(pc,tname);
1243: }
1244: PetscOptionsBool("-pc_gamg_repartition","Repartion coarse grids","PCGAMGRepartitioning",pc_gamg->repart,&pc_gamg->repart,NULL);
1245: PetscOptionsBool("-pc_gamg_reuse_interpolation","Reuse prolongation operator","PCGAMGReuseInterpolation",pc_gamg->reuse_prol,&pc_gamg->reuse_prol,NULL);
1246: PetscOptionsBool("-pc_gamg_use_agg_gasm","Use aggregation agragates for GASM smoother","PCGAMGUseASMAggs",pc_gamg->use_aggs_in_gasm,&pc_gamg->use_aggs_in_gasm,NULL);
1247: PetscOptionsInt("-pc_gamg_process_eq_limit","Limit (goal) on number of equations per process on coarse grids","PCGAMGSetProcEqLim",pc_gamg->min_eq_proc,&pc_gamg->min_eq_proc,NULL);
1248: PetscOptionsInt("-pc_gamg_coarse_eq_limit","Limit on number of equations for the coarse grid","PCGAMGSetCoarseEqLim",pc_gamg->coarse_eq_limit,&pc_gamg->coarse_eq_limit,NULL);
1249: PetscOptionsReal("-pc_gamg_threshold","Relative threshold to use for dropping edges in aggregation graph","PCGAMGSetThreshold",pc_gamg->threshold,&pc_gamg->threshold,&flag);
1250: PetscOptionsRealArray("-pc_gamg_eigtarget","Target eigenvalue range as fraction of estimated maximum eigenvalue","PCGAMGSetEigTarget",pc_gamg->eigtarget,&two,NULL);
1251: PetscOptionsInt("-pc_mg_levels","Set number of MG levels","PCGAMGSetNlevels",pc_gamg->Nlevels,&pc_gamg->Nlevels,NULL);
1253: /* set options for subtype */
1254: if (pc_gamg->ops->setfromoptions) {(*pc_gamg->ops->setfromoptions)(PetscOptionsObject,pc);}
1255: }
1256: PetscOptionsTail();
1257: return(0);
1258: }
1260: /* -------------------------------------------------------------------------- */
1261: /*MC
1262: PCGAMG - Geometric algebraic multigrid (AMG) preconditioner
1264: Options Database Keys:
1265: Multigrid options(inherited)
1266: + -pc_mg_cycles <v>: v or w (PCMGSetCycleType())
1267: . -pc_mg_smoothup <1>: Number of post-smoothing steps (PCMGSetNumberSmoothUp)
1268: . -pc_mg_smoothdown <1>: Number of pre-smoothing steps (PCMGSetNumberSmoothDown)
1269: - -pc_mg_type <multiplicative>: (one of) additive multiplicative full kascade
1272: Notes: In order to obtain good performance for PCGAMG for vector valued problems you must
1273: $ Call MatSetBlockSize() to indicate the number of degrees of freedom per grid point
1274: $ Call MatSetNearNullSpace() (or PCSetCoordinates() if solving the equations of elasticity) to indicate the near null space of the operator
1275: $ See the Users Manual Chapter 4 for more details
1277: Level: intermediate
1279: Concepts: algebraic multigrid
1281: .seealso: PCCreate(), PCSetType(), MatSetBlockSize(), PCMGType, PCSetCoordinates(), MatSetNearNullSpace(), PCGAMGSetType(), PCGAMGAGG, PCGAMGGEO, PCGAMGCLASSICAL, PCGAMGSetProcEqLim(),
1282: PCGAMGSetCoarseEqLim(), PCGAMGSetRepartitioning(), PCGAMGRegister(), PCGAMGSetReuseInterpolation(), PCGAMGSetUseASMAggs(), PCGAMGSetNlevels(), PCGAMGSetThreshold(), PCGAMGGetType()
1283: M*/
1287: PETSC_EXTERN PetscErrorCode PCCreate_GAMG(PC pc)
1288: {
1290: PC_GAMG *pc_gamg;
1291: PC_MG *mg;
1294: /* register AMG type */
1295: PCGAMGInitializePackage();
1297: /* PCGAMG is an inherited class of PCMG. Initialize pc as PCMG */
1298: PCSetType(pc, PCMG);
1299: PetscObjectChangeTypeName((PetscObject)pc, PCGAMG);
1301: /* create a supporting struct and attach it to pc */
1302: PetscNewLog(pc,&pc_gamg);
1303: mg = (PC_MG*)pc->data;
1304: mg->galerkin = 2; /* Use Galerkin, but it is computed externally from PCMG by GAMG code */
1305: mg->innerctx = pc_gamg;
1307: PetscNewLog(pc,&pc_gamg->ops);
1309: pc_gamg->setup_count = 0;
1310: /* these should be in subctx but repartitioning needs simple arrays */
1311: pc_gamg->data_sz = 0;
1312: pc_gamg->data = 0;
1314: /* overwrite the pointers of PCMG by the functions of base class PCGAMG */
1315: pc->ops->setfromoptions = PCSetFromOptions_GAMG;
1316: pc->ops->setup = PCSetUp_GAMG;
1317: pc->ops->reset = PCReset_GAMG;
1318: pc->ops->destroy = PCDestroy_GAMG;
1319: mg->view = PCView_GAMG;
1321: PetscObjectComposeFunction((PetscObject)pc,"PCGAMGSetProcEqLim_C",PCGAMGSetProcEqLim_GAMG);
1322: PetscObjectComposeFunction((PetscObject)pc,"PCGAMGSetCoarseEqLim_C",PCGAMGSetCoarseEqLim_GAMG);
1323: PetscObjectComposeFunction((PetscObject)pc,"PCGAMGSetRepartitioning_C",PCGAMGSetRepartitioning_GAMG);
1324: PetscObjectComposeFunction((PetscObject)pc,"PCGAMGSetReuseInterpolation_C",PCGAMGSetReuseInterpolation_GAMG);
1325: PetscObjectComposeFunction((PetscObject)pc,"PCGAMGSetUseASMAggs_C",PCGAMGSetUseASMAggs_GAMG);
1326: PetscObjectComposeFunction((PetscObject)pc,"PCGAMGSetThreshold_C",PCGAMGSetThreshold_GAMG);
1327: PetscObjectComposeFunction((PetscObject)pc,"PCGAMGSetType_C",PCGAMGSetType_GAMG);
1328: PetscObjectComposeFunction((PetscObject)pc,"PCGAMGGetType_C",PCGAMGGetType_GAMG);
1329: PetscObjectComposeFunction((PetscObject)pc,"PCGAMGSetNlevels_C",PCGAMGSetNlevels_GAMG);
1330: pc_gamg->repart = PETSC_FALSE;
1331: pc_gamg->reuse_prol = PETSC_FALSE;
1332: pc_gamg->use_aggs_in_gasm = PETSC_FALSE;
1333: pc_gamg->min_eq_proc = 50;
1334: pc_gamg->coarse_eq_limit = 50;
1335: pc_gamg->threshold = 0.;
1336: pc_gamg->Nlevels = GAMG_MAXLEVELS;
1337: pc_gamg->emax_id = -1;
1338: pc_gamg->current_level = 0; /* don't need to init really */
1339: pc_gamg->eigtarget[0] = 0.05;
1340: pc_gamg->eigtarget[1] = 1.05;
1341: pc_gamg->ops->createlevel = PCGAMGCreateLevel_GAMG;
1343: /* PCSetUp_GAMG assumes that the type has been set, so set it to the default now */
1344: PCGAMGSetType(pc,PCGAMGAGG);
1345: return(0);
1346: }
1350: /*@C
1351: PCGAMGInitializePackage - This function initializes everything in the PCGAMG package. It is called
1352: from PetscDLLibraryRegister() when using dynamic libraries, and on the first call to PCCreate_GAMG()
1353: when using static libraries.
1355: Level: developer
1357: .keywords: PC, PCGAMG, initialize, package
1358: .seealso: PetscInitialize()
1359: @*/
1360: PetscErrorCode PCGAMGInitializePackage(void)
1361: {
1365: if (PCGAMGPackageInitialized) return(0);
1366: PCGAMGPackageInitialized = PETSC_TRUE;
1367: PetscFunctionListAdd(&GAMGList,PCGAMGGEO,PCCreateGAMG_GEO);
1368: PetscFunctionListAdd(&GAMGList,PCGAMGAGG,PCCreateGAMG_AGG);
1369: PetscFunctionListAdd(&GAMGList,PCGAMGCLASSICAL,PCCreateGAMG_Classical);
1370: PetscRegisterFinalize(PCGAMGFinalizePackage);
1372: /* general events */
1373: PetscLogEventRegister("PCGAMGGraph_AGG", 0, &PC_GAMGGraph_AGG);
1374: PetscLogEventRegister("PCGAMGGraph_GEO", PC_CLASSID, &PC_GAMGGraph_GEO);
1375: PetscLogEventRegister("PCGAMGCoarse_AGG", PC_CLASSID, &PC_GAMGCoarsen_AGG);
1376: PetscLogEventRegister("PCGAMGCoarse_GEO", PC_CLASSID, &PC_GAMGCoarsen_GEO);
1377: PetscLogEventRegister("PCGAMGProl_AGG", PC_CLASSID, &PC_GAMGProlongator_AGG);
1378: PetscLogEventRegister("PCGAMGProl_GEO", PC_CLASSID, &PC_GAMGProlongator_GEO);
1379: PetscLogEventRegister("PCGAMGPOpt_AGG", PC_CLASSID, &PC_GAMGOptProlongator_AGG);
1381: #if defined PETSC_GAMG_USE_LOG
1382: PetscLogEventRegister("GAMG: createProl", PC_CLASSID, &petsc_gamg_setup_events[SET1]);
1383: PetscLogEventRegister(" Graph", PC_CLASSID, &petsc_gamg_setup_events[GRAPH]);
1384: /* PetscLogEventRegister(" G.Mat", PC_CLASSID, &petsc_gamg_setup_events[GRAPH_MAT]); */
1385: /* PetscLogEventRegister(" G.Filter", PC_CLASSID, &petsc_gamg_setup_events[GRAPH_FILTER]); */
1386: /* PetscLogEventRegister(" G.Square", PC_CLASSID, &petsc_gamg_setup_events[GRAPH_SQR]); */
1387: PetscLogEventRegister(" MIS/Agg", PC_CLASSID, &petsc_gamg_setup_events[SET4]);
1388: PetscLogEventRegister(" geo: growSupp", PC_CLASSID, &petsc_gamg_setup_events[SET5]);
1389: PetscLogEventRegister(" geo: triangle", PC_CLASSID, &petsc_gamg_setup_events[SET6]);
1390: PetscLogEventRegister(" search&set", PC_CLASSID, &petsc_gamg_setup_events[FIND_V]);
1391: PetscLogEventRegister(" SA: col data", PC_CLASSID, &petsc_gamg_setup_events[SET7]);
1392: PetscLogEventRegister(" SA: frmProl0", PC_CLASSID, &petsc_gamg_setup_events[SET8]);
1393: PetscLogEventRegister(" SA: smooth", PC_CLASSID, &petsc_gamg_setup_events[SET9]);
1394: PetscLogEventRegister("GAMG: partLevel", PC_CLASSID, &petsc_gamg_setup_events[SET2]);
1395: PetscLogEventRegister(" repartition", PC_CLASSID, &petsc_gamg_setup_events[SET12]);
1396: PetscLogEventRegister(" Invert-Sort", PC_CLASSID, &petsc_gamg_setup_events[SET13]);
1397: PetscLogEventRegister(" Move A", PC_CLASSID, &petsc_gamg_setup_events[SET14]);
1398: PetscLogEventRegister(" Move P", PC_CLASSID, &petsc_gamg_setup_events[SET15]);
1400: /* PetscLogEventRegister(" PL move data", PC_CLASSID, &petsc_gamg_setup_events[SET13]); */
1401: /* PetscLogEventRegister("GAMG: fix", PC_CLASSID, &petsc_gamg_setup_events[SET10]); */
1402: /* PetscLogEventRegister("GAMG: set levels", PC_CLASSID, &petsc_gamg_setup_events[SET11]); */
1403: /* create timer stages */
1404: #if defined GAMG_STAGES
1405: {
1406: char str[32];
1407: PetscInt lidx;
1408: sprintf(str,"MG Level %d (finest)",0);
1409: PetscLogStageRegister(str, &gamg_stages[0]);
1410: for (lidx=1; lidx<9; lidx++) {
1411: sprintf(str,"MG Level %d",lidx);
1412: PetscLogStageRegister(str, &gamg_stages[lidx]);
1413: }
1414: }
1415: #endif
1416: #endif
1417: return(0);
1418: }
1422: /*@C
1423: PCGAMGFinalizePackage - This function frees everything from the PCGAMG package. It is
1424: called from PetscFinalize() automatically.
1426: Level: developer
1428: .keywords: Petsc, destroy, package
1429: .seealso: PetscFinalize()
1430: @*/
1431: PetscErrorCode PCGAMGFinalizePackage(void)
1432: {
1436: PCGAMGPackageInitialized = PETSC_FALSE;
1437: PetscFunctionListDestroy(&GAMGList);
1438: return(0);
1439: }
1443: /*@C
1444: PCGAMGRegister - Register a PCGAMG implementation.
1446: Input Parameters:
1447: + type - string that will be used as the name of the GAMG type.
1448: - create - function for creating the gamg context.
1450: Level: advanced
1452: .seealso: PCGAMGType, PCGAMG, PCGAMGSetType()
1453: @*/
1454: PetscErrorCode PCGAMGRegister(PCGAMGType type, PetscErrorCode (*create)(PC))
1455: {
1459: PCGAMGInitializePackage();
1460: PetscFunctionListAdd(&GAMGList,type,create);
1461: return(0);
1462: }