Actual source code: hierarchical.c
petsc-3.11.4 2019-09-28
2: #include <../src/mat/impls/adj/mpi/mpiadj.h>
3: #include <petscsf.h>
4: #include <petsc/private/matimpl.h>
6: /*
7: It is a hierarchical partitioning. The partitioner has two goals:
8: (1) Most of current partitioners fail at a large scale. The hierarchical partitioning
9: strategy is trying to produce large number of subdomains when number of processor cores is large.
10: (2) PCGASM needs one 'big' subdomain across multi-cores. The partitioner provides two
11: consistent partitions, coarse parts and fine parts. A coarse part is a 'big' subdomain consisting
12: of several small subdomains.
13: */
15: PetscErrorCode MatPartitioningHierarchical_DetermineDestination(MatPartitioning,IS,PetscInt,PetscInt,IS*);
16: PetscErrorCode MatPartitioningHierarchical_AssembleSubdomain(Mat,IS,IS,IS*,Mat*,ISLocalToGlobalMapping*);
17: PetscErrorCode MatPartitioningHierarchical_ReassembleFineparts(Mat,IS,ISLocalToGlobalMapping,IS*);
19: typedef struct {
20: char* fineparttype; /* partitioner on fine level */
21: char* coarseparttype; /* partitioner on coarse level */
22: PetscInt nfineparts; /* number of fine parts on each coarse subdomain */
23: PetscInt ncoarseparts; /* number of coarse parts */
24: IS coarseparts; /* partitioning on coarse level */
25: IS fineparts; /* partitioning on fine level */
26: MatPartitioning coarseMatPart; /* MatPartititioning on coarse level (first level) */
27: MatPartitioning fineMatPart; /* MatPartitioning on fine level (second level) */
28: MatPartitioning improver; /* Improve the quality of a partition */
29: } MatPartitioning_Hierarchical;
31: /*
32: Uses a hierarchical partitioning strategy to partition the matrix in parallel.
33: Use this interface to make the partitioner consistent with others
34: */
35: static PetscErrorCode MatPartitioningApply_Hierarchical(MatPartitioning part,IS *partitioning)
36: {
37: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
38: const PetscInt *fineparts_indices, *coarseparts_indices;
39: PetscInt *fineparts_indices_tmp;
40: PetscInt *parts_indices,i,j,mat_localsize, *offsets;
41: Mat mat = part->adj,adj,sadj;
42: PetscReal *part_weights;
43: PetscBool flg;
44: PetscInt bs = 1;
45: PetscInt *coarse_vertex_weights = 0;
46: PetscMPIInt size,rank;
47: MPI_Comm comm,scomm;
48: IS destination,fineparts_temp, vweights, svweights;
49: PetscInt nsvwegihts,*fp_vweights;
50: const PetscInt *svweights_indices;
51: ISLocalToGlobalMapping mapping;
52: const char *prefix;
53: PetscErrorCode ierr;
56: PetscObjectGetComm((PetscObject)part,&comm);
57: MPI_Comm_size(comm,&size);
58: MPI_Comm_rank(comm,&rank);
59: PetscObjectTypeCompare((PetscObject)mat,MATMPIADJ,&flg);
60: if (flg) {
61: adj = mat;
62: PetscObjectReference((PetscObject)adj);
63: }else {
64: /* bs indicates if the converted matrix is "reduced" from the original and hence the
65: resulting partition results need to be stretched to match the original matrix */
66: MatConvert(mat,MATMPIADJ,MAT_INITIAL_MATRIX,&adj);
67: if (adj->rmap->n > 0) bs = mat->rmap->n/adj->rmap->n;
68: }
69: /* local size of mat */
70: mat_localsize = adj->rmap->n;
71: /* check parameters */
72: /* how many small subdomains we want from a given 'big' suddomain */
73: if(!hpart->nfineparts) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG," must set number of small subdomains for each big subdomain \n");
74: if(!hpart->ncoarseparts && !part->n) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE," did not either set number of coarse parts or total number of parts \n");
76: /* Partitioning the domain into one single subdomain is a trivial case, and we should just return */
77: if (part->n==1) {
78: PetscCalloc1(bs*adj->rmap->n,&parts_indices);
79: ISCreateGeneral(comm,bs*adj->rmap->n,parts_indices,PETSC_OWN_POINTER,partitioning);
80: hpart->ncoarseparts = 1;
81: hpart->nfineparts = 1;
82: PetscStrallocpy("NONE",&hpart->coarseparttype);
83: PetscStrallocpy("NONE",&hpart->fineparttype);
84: MatDestroy(&adj);
85: return(0);
86: }
88: if(part->n){
89: hpart->ncoarseparts = part->n/hpart->nfineparts;
91: if (part->n%hpart->nfineparts != 0) hpart->ncoarseparts++;
92: }else{
93: part->n = hpart->ncoarseparts*hpart->nfineparts;
94: }
96: PetscCalloc1(hpart->ncoarseparts+1, &offsets);
97: PetscCalloc1(hpart->ncoarseparts, &part_weights);
100: if (part->n%hpart->nfineparts != 0) offsets[1] = part->n%hpart->nfineparts;
101: else offsets[1] = hpart->nfineparts;
103: part_weights[0] = ((PetscReal)offsets[1])/part->n;
105: for (i=2; i<=hpart->ncoarseparts; i++) {
106: offsets[i] = hpart->nfineparts;
107: part_weights[i-1] = ((PetscReal)offsets[i])/part->n;
108: }
110: offsets[0] = 0;
111: for (i=1;i<=hpart->ncoarseparts; i++)
112: offsets[i] += offsets[i-1];
114: /* If these exists a mat partitioner, we should delete it */
115: MatPartitioningDestroy(&hpart->coarseMatPart);
116: MatPartitioningCreate(comm,&hpart->coarseMatPart);
117: PetscObjectGetOptionsPrefix((PetscObject)part,&prefix);
118: PetscObjectSetOptionsPrefix((PetscObject)hpart->coarseMatPart,prefix);
119: PetscObjectAppendOptionsPrefix((PetscObject)hpart->coarseMatPart,"hierarch_coarse_");
120: /* if did not set partitioning type yet, use parmetis by default */
121: if (!hpart->coarseparttype){
122: #if defined(PETSC_HAVE_PARMETIS)
123: MatPartitioningSetType(hpart->coarseMatPart,MATPARTITIONINGPARMETIS);
124: PetscStrallocpy(MATPARTITIONINGPARMETIS,&hpart->coarseparttype);
125: #elif defined(PETSC_HAVE_PTSCOTCH)
126: MatPartitioningSetType(hpart->coarseMatPart,MATPARTITIONINGPTSCOTCH);
127: PetscStrallocpy(MATPARTITIONINGPTSCOTCH,&hpart->coarseparttype);
128: #else
129: SETERRQ(PetscObjectComm((PetscObject)mat),PETSC_ERR_SUP,"Requires PETSc be installed with ParMetis or run with -mat_partitioning_hierarchical_coarseparttype partitiontype");
130: #endif
131: } else {
132: MatPartitioningSetType(hpart->coarseMatPart,hpart->coarseparttype);
133: }
134: MatPartitioningSetAdjacency(hpart->coarseMatPart,adj);
135: MatPartitioningSetNParts(hpart->coarseMatPart, hpart->ncoarseparts);
136: /* copy over vertex weights */
137: if(part->vertex_weights){
138: PetscMalloc1(mat_localsize,&coarse_vertex_weights);
139: PetscMemcpy(coarse_vertex_weights,part->vertex_weights,sizeof(PetscInt)*mat_localsize);
140: MatPartitioningSetVertexWeights(hpart->coarseMatPart,coarse_vertex_weights);
141: }
143: MatPartitioningSetPartitionWeights(hpart->coarseMatPart, part_weights);
144: MatPartitioningApply(hpart->coarseMatPart,&hpart->coarseparts);
146: PetscCalloc1(mat_localsize, &fineparts_indices_tmp);
148: /* Wrap the original vertex weights into an index set so that we can extract the corresponding
149: * vertex weights for each big subdomain using ISCreateSubIS().
150: * */
151: if (part->vertex_weights) {
152: ISCreateGeneral(comm,mat_localsize,part->vertex_weights,PETSC_COPY_VALUES,&vweights);
153: }
155: for(i=0; i<hpart->ncoarseparts; i+=size){
156: /* Determine where we want to send big subdomains */
157: MatPartitioningHierarchical_DetermineDestination(part,hpart->coarseparts,i,i+size,&destination);
158: /* Assemble a submatrix and its vertex weights for partitioning subdomains */
159: MatPartitioningHierarchical_AssembleSubdomain(adj,part->vertex_weights? vweights:NULL,destination,part->vertex_weights? &svweights:NULL,&sadj,&mapping);
160: /* We have to create a new array to hold vertex weights since coarse partitioner needs to own the vertex-weights array */
161: if (part->vertex_weights) {
162: ISGetLocalSize(svweights,&nsvwegihts);
163: PetscMalloc1(nsvwegihts,&fp_vweights);
164: ISGetIndices(svweights,&svweights_indices);
165: PetscMemcpy(fp_vweights,svweights_indices,nsvwegihts*sizeof(PetscInt));
166: ISRestoreIndices(svweights,&svweights_indices);
167: ISDestroy(&svweights);
168: }
170: ISDestroy(&destination);
171: PetscObjectGetComm((PetscObject)sadj,&scomm);
173: /*
174: * If the number of big subdomains is smaller than the number of processor cores, the higher ranks do not
175: * need to do partitioning
176: * */
177: if((i+rank)<hpart->ncoarseparts) {
178: MatPartitioningDestroy(&hpart->fineMatPart);
179: /* create a fine partitioner */
180: MatPartitioningCreate(scomm,&hpart->fineMatPart);
181: PetscObjectSetOptionsPrefix((PetscObject)hpart->fineMatPart,prefix);
182: PetscObjectAppendOptionsPrefix((PetscObject)hpart->fineMatPart,"hierarch_fine_");
183: /* if do not set partitioning type, use parmetis by default */
184: if(!hpart->fineparttype){
185: #if defined(PETSC_HAVE_PARMETIS)
186: MatPartitioningSetType(hpart->fineMatPart,MATPARTITIONINGPARMETIS);
187: PetscStrallocpy(MATPARTITIONINGPARMETIS,&hpart->fineparttype);
188: #elif defined(PETSC_HAVE_PTSCOTCH)
189: MatPartitioningSetType(hpart->fineMatPart,MATPARTITIONINGPTSCOTCH);
190: PetscStrallocpy(MATPARTITIONINGPTSCOTCH,&hpart->fineparttype);
191: #elif defined(PETSC_HAVE_CHACO)
192: MatPartitioningSetType(hpart->fineMatPart,MATPARTITIONINGCHACO);
193: PetscStrallocpy(MATPARTITIONINGCHACO,&hpart->fineparttype);
194: #elif defined(PETSC_HAVE_PARTY)
195: MatPartitioningSetType(hpart->fineMatPart,MATPARTITIONINGPARTY);
196: PetscStrallocpy(PETSC_HAVE_PARTY,&hpart->fineparttype);
197: #else
198: SETERRQ(PetscObjectComm((PetscObject)mat),PETSC_ERR_SUP,"Requires PETSc be installed with ParMetis or run with -mat_partitioning_hierarchical_coarseparttype partitiontype");
199: #endif
200: } else {
201: MatPartitioningSetType(hpart->fineMatPart,hpart->fineparttype);
202: }
203: MatPartitioningSetAdjacency(hpart->fineMatPart,sadj);
204: MatPartitioningSetNParts(hpart->fineMatPart, offsets[rank+1+i]-offsets[rank+i]);
205: if (part->vertex_weights) {
206: MatPartitioningSetVertexWeights(hpart->fineMatPart,fp_vweights);
207: }
208: MatPartitioningApply(hpart->fineMatPart,&fineparts_temp);
209: } else {
210: ISCreateGeneral(scomm,0,NULL,PETSC_OWN_POINTER,&fineparts_temp);
211: }
213: MatDestroy(&sadj);
215: /* Send partition back to the original owners */
216: MatPartitioningHierarchical_ReassembleFineparts(adj,fineparts_temp,mapping,&hpart->fineparts);
217: ISGetIndices(hpart->fineparts,&fineparts_indices);
218: for (j=0;j<mat_localsize;j++)
219: if (fineparts_indices[j] >=0) fineparts_indices_tmp[j] = fineparts_indices[j];
221: ISRestoreIndices(hpart->fineparts,&fineparts_indices);
222: ISDestroy(&hpart->fineparts);
223: ISDestroy(&fineparts_temp);
224: ISLocalToGlobalMappingDestroy(&mapping);
225: }
227: if (part->vertex_weights) {
228: ISDestroy(&vweights);
229: }
231: ISCreateGeneral(comm,mat_localsize,fineparts_indices_tmp,PETSC_OWN_POINTER,&hpart->fineparts);
232: ISGetIndices(hpart->fineparts,&fineparts_indices);
233: ISGetIndices(hpart->coarseparts,&coarseparts_indices);
234: PetscMalloc1(bs*adj->rmap->n,&parts_indices);
235: /* Modify the local indices to the global indices by combing the coarse partition and the fine partitions */
236: for(i=0; i<adj->rmap->n; i++){
237: for(j=0; j<bs; j++){
238: parts_indices[bs*i+j] = fineparts_indices[i]+offsets[coarseparts_indices[i]];
239: }
240: }
241: ISRestoreIndices(hpart->fineparts,&fineparts_indices);
242: ISRestoreIndices(hpart->coarseparts,&coarseparts_indices);
243: PetscFree(offsets);
244: ISCreateGeneral(comm,bs*adj->rmap->n,parts_indices,PETSC_OWN_POINTER,partitioning);
245: MatDestroy(&adj);
246: return(0);
247: }
250: PetscErrorCode MatPartitioningHierarchical_ReassembleFineparts(Mat adj, IS fineparts, ISLocalToGlobalMapping mapping, IS *sfineparts)
251: {
252: PetscInt *local_indices, *global_indices,*owners,*sfineparts_indices,localsize,i;
253: const PetscInt *ranges,*fineparts_indices;
254: PetscMPIInt rank;
255: MPI_Comm comm;
256: PetscLayout rmap;
257: PetscSFNode *remote;
258: PetscSF sf;
259: PetscErrorCode ierr;
263: PetscObjectGetComm((PetscObject)adj,&comm);
264: MPI_Comm_rank(comm,&rank);
265: MatGetLayouts(adj,&rmap,NULL);
266: ISGetLocalSize(fineparts,&localsize);
267: PetscCalloc2(localsize,&global_indices,localsize,&local_indices);
268: for(i=0; i<localsize; i++){
269: local_indices[i] = i;
270: }
271: /* map local indices back to global so that we can permulate data globally */
272: ISLocalToGlobalMappingApply(mapping,localsize,local_indices,global_indices);
273: PetscCalloc1(localsize,&owners);
274: /* find owners for global indices */
275: for(i=0; i<localsize; i++){
276: PetscLayoutFindOwner(rmap,global_indices[i],&owners[i]);
277: }
278: PetscLayoutGetRanges(rmap,&ranges);
279: PetscCalloc1(ranges[rank+1]-ranges[rank],&sfineparts_indices);
281: for (i=0; i<(ranges[rank+1]-ranges[rank]); i++) {
282: sfineparts_indices[i] = -1;
283: }
285: ISGetIndices(fineparts,&fineparts_indices);
286: PetscSFCreate(comm,&sf);
287: PetscCalloc1(localsize,&remote);
288: for(i=0; i<localsize; i++){
289: remote[i].rank = owners[i];
290: remote[i].index = global_indices[i]-ranges[owners[i]];
291: }
292: PetscSFSetType(sf,PETSCSFBASIC);
293: /* not sure how to add prefix to sf */
294: PetscSFSetFromOptions(sf);
295: PetscSFSetGraph(sf,localsize,localsize,NULL,PETSC_OWN_POINTER,remote,PETSC_OWN_POINTER);
296: PetscSFReduceBegin(sf,MPIU_INT,fineparts_indices,sfineparts_indices,MPIU_REPLACE);
297: PetscSFReduceEnd(sf,MPIU_INT,fineparts_indices,sfineparts_indices,MPIU_REPLACE);
298: PetscSFDestroy(&sf);
299: ISRestoreIndices(fineparts,&fineparts_indices);
300: ISCreateGeneral(comm,ranges[rank+1]-ranges[rank],sfineparts_indices,PETSC_OWN_POINTER,sfineparts);
301: PetscFree2(global_indices,local_indices);
302: PetscFree(owners);
303: return(0);
304: }
307: PetscErrorCode MatPartitioningHierarchical_AssembleSubdomain(Mat adj,IS vweights, IS destination,IS *svweights,Mat *sadj,ISLocalToGlobalMapping *mapping)
308: {
309: IS irows,icols;
310: PetscInt irows_ln;
311: PetscMPIInt rank;
312: const PetscInt *irows_indices;
313: MPI_Comm comm;
314: PetscErrorCode ierr;
317: PetscObjectGetComm((PetscObject)adj,&comm);
318: MPI_Comm_rank(comm,&rank);
319: /* figure out where data comes from */
320: ISBuildTwoSided(destination,NULL,&irows);
321: ISDuplicate(irows,&icols);
322: ISGetLocalSize(irows,&irows_ln);
323: ISGetIndices(irows,&irows_indices);
324: ISLocalToGlobalMappingCreate(comm,1,irows_ln,irows_indices,PETSC_COPY_VALUES,mapping);
325: ISRestoreIndices(irows,&irows_indices);
326: MatCreateSubMatrices(adj,1,&irows,&icols,MAT_INITIAL_MATRIX,&sadj);
327: if (vweights && svweights) {
328: ISCreateSubIS(vweights,irows,svweights);
329: }
330: ISDestroy(&irows);
331: ISDestroy(&icols);
332: return(0);
333: }
336: PetscErrorCode MatPartitioningHierarchical_DetermineDestination(MatPartitioning part, IS partitioning, PetscInt pstart, PetscInt pend, IS *destination)
337: {
338: MPI_Comm comm;
339: PetscMPIInt rank,size,target;
340: PetscInt plocalsize,*dest_indices,i;
341: const PetscInt *part_indices;
342: PetscErrorCode ierr;
345: PetscObjectGetComm((PetscObject)part,&comm);
346: MPI_Comm_rank(comm,&rank);
347: MPI_Comm_size(comm,&size);
348: if((pend-pstart)>size) SETERRQ3(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,"range [%D, %D] should be smaller than or equal to size %D",pstart,pend,size);
349: if(pstart>pend) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP," pstart %D should be smaller than pend %D",pstart,pend);
350: ISGetLocalSize(partitioning,&plocalsize);
351: PetscCalloc1(plocalsize,&dest_indices);
352: ISGetIndices(partitioning,&part_indices);
353: for(i=0; i<plocalsize; i++){
354: /* compute target */
355: target = part_indices[i]-pstart;
356: /* mark out of range entity as -1 */
357: if(part_indices[i]<pstart || part_indices[i]>=pend) target = -1;
358: dest_indices[i] = target;
359: }
360: ISCreateGeneral(comm,plocalsize,dest_indices,PETSC_OWN_POINTER,destination);
361: return(0);
362: }
365: PetscErrorCode MatPartitioningView_Hierarchical(MatPartitioning part,PetscViewer viewer)
366: {
367: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
368: PetscErrorCode ierr;
369: PetscMPIInt rank;
370: PetscBool iascii;
371: PetscViewer sviewer;
374: MPI_Comm_rank(PetscObjectComm((PetscObject)part),&rank);
375: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);
376: if(iascii){
377: PetscViewerASCIIPrintf(viewer," Number of coarse parts: %D\n",hpart->ncoarseparts);
378: PetscViewerASCIIPrintf(viewer," Coarse partitioner: %s\n",hpart->coarseparttype);
379: if (hpart->coarseMatPart) {
380: PetscViewerASCIIPushTab(viewer);
381: MatPartitioningView(hpart->coarseMatPart,viewer);
382: PetscViewerASCIIPopTab(viewer);
383: }
384: PetscViewerASCIIPrintf(viewer," Number of fine parts: %D\n",hpart->nfineparts);
385: PetscViewerASCIIPrintf(viewer," Fine partitioner: %s\n",hpart->fineparttype);
386: PetscViewerGetSubViewer(viewer,PETSC_COMM_SELF,&sviewer);
387: if (!rank && hpart->fineMatPart) {
388: PetscViewerASCIIPushTab(viewer);
389: MatPartitioningView(hpart->fineMatPart,sviewer);
390: PetscViewerASCIIPopTab(viewer);
391: }
392: PetscViewerRestoreSubViewer(viewer,PETSC_COMM_SELF,&sviewer);
393: }
394: return(0);
395: }
398: PetscErrorCode MatPartitioningHierarchicalGetFineparts(MatPartitioning part,IS *fineparts)
399: {
400: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
401: PetscErrorCode ierr;
404: *fineparts = hpart->fineparts;
405: PetscObjectReference((PetscObject)hpart->fineparts);
406: return(0);
407: }
409: PetscErrorCode MatPartitioningHierarchicalGetCoarseparts(MatPartitioning part,IS *coarseparts)
410: {
411: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
412: PetscErrorCode ierr;
415: *coarseparts = hpart->coarseparts;
416: PetscObjectReference((PetscObject)hpart->coarseparts);
417: return(0);
418: }
420: PetscErrorCode MatPartitioningHierarchicalSetNcoarseparts(MatPartitioning part, PetscInt ncoarseparts)
421: {
422: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
425: hpart->ncoarseparts = ncoarseparts;
426: return(0);
427: }
429: PetscErrorCode MatPartitioningHierarchicalSetNfineparts(MatPartitioning part, PetscInt nfineparts)
430: {
431: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
434: hpart->nfineparts = nfineparts;
435: return(0);
436: }
438: PetscErrorCode MatPartitioningSetFromOptions_Hierarchical(PetscOptionItems *PetscOptionsObject,MatPartitioning part)
439: {
440: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
442: char value[1024];
443: PetscBool flag = PETSC_FALSE;
446: PetscOptionsHead(PetscOptionsObject,"Set hierarchical partitioning options");
447: PetscOptionsString("-mat_partitioning_hierarchical_coarseparttype","coarse part type",NULL,NULL,value,1024,&flag);
448: if(flag){
449: PetscCalloc1(1024,&hpart->coarseparttype);
450: PetscStrcpy(hpart->coarseparttype,value);
451: }
452: PetscOptionsString("-mat_partitioning_hierarchical_fineparttype","fine part type",NULL,NULL,value,1024,&flag);
453: if(flag){
454: PetscCalloc1(1024,&hpart->fineparttype);
455: PetscStrcpy(hpart->fineparttype,value);
456: }
457: PetscOptionsInt("-mat_partitioning_hierarchical_ncoarseparts","number of coarse parts",NULL,hpart->ncoarseparts,&hpart->ncoarseparts,&flag);
458: PetscOptionsInt("-mat_partitioning_hierarchical_nfineparts","number of fine parts",NULL,hpart->nfineparts,&hpart->nfineparts,&flag);
459: PetscOptionsTail();
460: return(0);
461: }
464: PetscErrorCode MatPartitioningDestroy_Hierarchical(MatPartitioning part)
465: {
466: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
467: PetscErrorCode ierr;
470: if(hpart->coarseparttype) {PetscFree(hpart->coarseparttype);}
471: if(hpart->fineparttype) {PetscFree(hpart->fineparttype);}
472: ISDestroy(&hpart->fineparts);
473: ISDestroy(&hpart->coarseparts);
474: MatPartitioningDestroy(&hpart->coarseMatPart);
475: MatPartitioningDestroy(&hpart->fineMatPart);
476: MatPartitioningDestroy(&hpart->improver);
477: PetscFree(hpart);
478: return(0);
479: }
481: /*
482: Improves the quality of a partition
483: */
484: static PetscErrorCode MatPartitioningImprove_Hierarchical(MatPartitioning part, IS *partitioning)
485: {
486: PetscErrorCode ierr;
487: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
488: Mat mat = part->adj, adj;
489: PetscBool flg;
490: PetscInt *vertex_weights;
491: const char *prefix;
494: PetscObjectTypeCompare((PetscObject)mat,MATMPIADJ,&flg);
495: if (flg) {
496: adj = mat;
497: PetscObjectReference((PetscObject)adj);
498: }else {
499: /* bs indicates if the converted matrix is "reduced" from the original and hence the
500: resulting partition results need to be stretched to match the original matrix */
501: MatConvert(mat,MATMPIADJ,MAT_INITIAL_MATRIX,&adj);
502: }
504: /* If there exists a mat partitioner, we should delete it */
505: MatPartitioningDestroy(&hpart->improver);
506: MatPartitioningCreate(PetscObjectComm((PetscObject)part),&hpart->improver);
507: PetscObjectGetOptionsPrefix((PetscObject)part,&prefix);
508: PetscObjectSetOptionsPrefix((PetscObject)hpart->improver,prefix);
509: PetscObjectAppendOptionsPrefix((PetscObject)hpart->improver,"hierarch_improver_");
510: /* Only parmetis supports to refine a partition */
511: #if defined(PETSC_HAVE_PARMETIS)
512: MatPartitioningSetType(hpart->improver,MATPARTITIONINGPARMETIS);
513: #else
514: SETERRQ(PetscObjectComm((PetscObject)adj),PETSC_ERR_SUP,"Requires PETSc be installed with ParMetis\n");
515: #endif
517: MatPartitioningSetAdjacency(hpart->improver,adj);
518: MatPartitioningSetNParts(hpart->improver, part->n);
519: /* copy over vertex weights */
520: if(part->vertex_weights){
521: PetscMalloc1(adj->rmap->n,&vertex_weights);
522: PetscMemcpy(vertex_weights,part->vertex_weights,sizeof(PetscInt)*adj->rmap->n);
523: MatPartitioningSetVertexWeights(hpart->improver,vertex_weights);
524: }
525: MatPartitioningImprove(hpart->improver,partitioning);
526: MatDestroy(&adj);
527: return(0);
528: }
531: /*MC
532: MATPARTITIONINGHIERARCH - Creates a partitioning context via hierarchical partitioning strategy.
533: The graph is partitioned into a number of subgraphs, and each subgraph is further split into a few smaller
534: subgraphs. The idea can be applied in a recursive manner. It is useful when you want to partition the graph
535: into a large number of subgraphs (often more than 10K) since partitions obtained with existing partitioners
536: such as ParMETIS and PTScotch are far from ideal. The hierarchical partitioning also tries to avoid off-node
537: communication as much as possible for multi-core processor. Another user case for the hierarchical partitioning
538: is to improve PCGASM convergence by generating multi-rank connected subdomain.
540: Collective on MPI_Comm
542: Input Parameter:
543: . part - the partitioning context
545: Options Database Keys:
546: + -mat_partitioning_hierarchical_coarseparttype - partitioner type at the first level and parmetis is used by default
547: . -mat_partitioning_hierarchical_fineparttype - partitioner type at the second level and parmetis is used by default
548: . -mat_partitioning_hierarchical_ncoarseparts - number of subgraphs is required at the first level, which is often the number of compute nodes
549: - -mat_partitioning_hierarchical_nfineparts - number of smaller subgraphs for each subgraph, which is often the number of cores per compute node
551: Level: beginner
553: References:
554: 1. Fande Kong, Xiao-Chuan Cai, A highly scalable multilevel Schwarz method with boundary geometry preserving coarse spaces for 3D elasticity
555: problems on domains with complex geometry, SIAM Journal on Scientific Computing 38 (2), C73-C95, 2016
556: 2. Fande Kong, Roy H. Stogner, Derek Gaston, John W. Peterson, Cody J. Permann, Andrew E. Slaughter, and Richard C. Martineau,
557: A general-purpose hierarchical mesh partitioning method with node balancing strategies for large-scale numerical simulations,
558: arXiv preprint arXiv:1809.02666CoRR, 2018.
560: .keywords: Partitioning, create, context
562: .seealso: MatPartitioningSetType(), MatPartitioningType
564: M*/
566: PETSC_EXTERN PetscErrorCode MatPartitioningCreate_Hierarchical(MatPartitioning part)
567: {
568: PetscErrorCode ierr;
569: MatPartitioning_Hierarchical *hpart;
572: PetscNewLog(part,&hpart);
573: part->data = (void*)hpart;
575: hpart->fineparttype = 0; /* fine level (second) partitioner */
576: hpart->coarseparttype = 0; /* coarse level (first) partitioner */
577: hpart->nfineparts = 1; /* we do not further partition coarse partition any more by default */
578: hpart->ncoarseparts = 0; /* number of coarse parts (first level) */
579: hpart->coarseparts = 0;
580: hpart->fineparts = 0;
581: hpart->coarseMatPart = 0;
582: hpart->fineMatPart = 0;
584: part->ops->apply = MatPartitioningApply_Hierarchical;
585: part->ops->view = MatPartitioningView_Hierarchical;
586: part->ops->destroy = MatPartitioningDestroy_Hierarchical;
587: part->ops->setfromoptions = MatPartitioningSetFromOptions_Hierarchical;
588: part->ops->improve = MatPartitioningImprove_Hierarchical;
589: return(0);
590: }