Actual source code: hierarchical.c
petsc-3.7.7 2017-09-25
2: #include <../src/mat/impls/adj/mpi/mpiadj.h> /*I "petscmat.h" I*/
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 part, IS partitioning, PetscInt pstart, PetscInt pend, IS *destination);
16: PetscErrorCode MatPartitioningHierarchical_AssembleSubdomain(Mat adj,IS destination,Mat *sadj, ISLocalToGlobalMapping *mapping);
17: PetscErrorCode MatPartitioningHierarchical_ReassembleFineparts(Mat adj, IS fineparts, ISLocalToGlobalMapping mapping, IS *sfineparts);
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_Hierarchical;
28: /*
29: Uses a hierarchical partitioning strategy to partition the matrix in parallel.
30: Use this interface to make the partitioner consistent with others
31: */
34: static PetscErrorCode MatPartitioningApply_Hierarchical(MatPartitioning part,IS *partitioning)
35: {
36: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
37: const PetscInt *fineparts_indices, *coarseparts_indices;
38: PetscInt *parts_indices,i,j,mat_localsize;
39: Mat mat = part->adj,adj,sadj;
40: PetscBool flg;
41: PetscInt bs = 1;
42: MatPartitioning finePart, coarsePart;
43: PetscInt *coarse_vertex_weights = 0;
44: PetscMPIInt size,rank;
45: MPI_Comm comm,scomm;
46: IS destination,fineparts_temp;
47: ISLocalToGlobalMapping mapping;
48: PetscErrorCode ierr;
51: PetscObjectGetComm((PetscObject)part,&comm);
52: MPI_Comm_size(comm,&size);
53: MPI_Comm_rank(comm,&rank);
54: PetscObjectTypeCompare((PetscObject)mat,MATMPIADJ,&flg);
55: if (flg) {
56: adj = mat;
57: PetscObjectReference((PetscObject)adj);
58: }else {
59: /* bs indicates if the converted matrix is "reduced" from the original and hence the
60: resulting partition results need to be stretched to match the original matrix */
61: MatConvert(mat,MATMPIADJ,MAT_INITIAL_MATRIX,&adj);
62: if (adj->rmap->n > 0) bs = mat->rmap->n/adj->rmap->n;
63: }
64: /* local size of mat */
65: mat_localsize = adj->rmap->n;
66: /* check parameters */
67: /* how many small subdomains we want from a given 'big' suddomain */
68: if(!hpart->Nfineparts) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG," must set number of small subdomains for each big subdomain \n");
69: 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");
70: if(part->n && part->n%hpart->Nfineparts!=0) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP,
71: " total number of parts %D can not be divided by number of fine parts %D\n",part->n,hpart->Nfineparts);
72: if(part->n){
73: hpart->Ncoarseparts = part->n/hpart->Nfineparts;
74: }else{
75: part->n = hpart->Ncoarseparts*hpart->Nfineparts;
76: }
77: /* we do not support this case currently, but this restriction should be
78: * removed in the further
79: * */
80: if(hpart->Ncoarseparts>size) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_SUP," we do not support number of coarse parts %D > size %D \n",hpart->Ncoarseparts,size);
81: MatPartitioningCreate(comm,&coarsePart);
82: /* if did not set partitioning type yet, use parmetis by default */
83: if(!hpart->coarseparttype){
84: MatPartitioningSetType(coarsePart,MATPARTITIONINGPARMETIS);
85: }else{
86: MatPartitioningSetType(coarsePart,hpart->coarseparttype);
87: }
88: MatPartitioningSetAdjacency(coarsePart,adj);
89: MatPartitioningSetNParts(coarsePart, hpart->Ncoarseparts);
90: /* copy over vertex weights */
91: if(part->vertex_weights){
92: PetscMalloc(sizeof(PetscInt)*mat_localsize,&coarse_vertex_weights);
93: PetscMemcpy(coarse_vertex_weights,part->vertex_weights,sizeof(PetscInt)*mat_localsize);
94: MatPartitioningSetVertexWeights(coarsePart,coarse_vertex_weights);
95: }
96: /* It looks nontrivial to support part weights,
97: * I will return back to implement it when have
98: * an idea.
99: * */
100: MatPartitioningApply(coarsePart,&hpart->coarseparts);
101: MatPartitioningDestroy(&coarsePart);
102: /* In the current implementation, destination should be the same as hpart->coarseparts,
103: * and this interface is preserved to deal with the case hpart->coarseparts>size in the
104: * future.
105: * */
106: MatPartitioningHierarchical_DetermineDestination(part,hpart->coarseparts,0,hpart->Ncoarseparts,&destination);
107: /* assemble a submatrix for partitioning subdomains */
108: MatPartitioningHierarchical_AssembleSubdomain(adj,destination,&sadj,&mapping);
109: ISDestroy(&destination);
110: PetscObjectGetComm((PetscObject)sadj,&scomm);
111: /* create a fine partitioner */
112: MatPartitioningCreate(scomm,&finePart);
113: /* if do not set partitioning type, use parmetis by default */
114: if(!hpart->fineparttype){
115: MatPartitioningSetType(finePart,MATPARTITIONINGPARMETIS);
116: }else{
117: MatPartitioningSetType(finePart,hpart->fineparttype);
118: }
119: MatPartitioningSetAdjacency(finePart,sadj);
120: MatPartitioningSetNParts(finePart, hpart->Nfineparts);
121: MatPartitioningApply(finePart,&fineparts_temp);
122: MatDestroy(&sadj);
123: MatPartitioningDestroy(&finePart);
124: MatPartitioningHierarchical_ReassembleFineparts(adj,fineparts_temp,mapping,&hpart->fineparts);
125: ISDestroy(&fineparts_temp);
126: ISLocalToGlobalMappingDestroy(&mapping);
128: ISGetIndices(hpart->fineparts,&fineparts_indices);
129: ISGetIndices(hpart->coarseparts,&coarseparts_indices);
130: PetscMalloc1(bs*adj->rmap->n,&parts_indices);
131: for(i=0; i<adj->rmap->n; i++){
132: for(j=0; j<bs; j++){
133: parts_indices[bs*i+j] = fineparts_indices[i]+coarseparts_indices[i]*hpart->Nfineparts;
134: }
135: }
136: ISCreateGeneral(comm,bs*adj->rmap->n,parts_indices,PETSC_OWN_POINTER,partitioning);
137: MatDestroy(&adj);
138: return(0);
139: }
144: PetscErrorCode MatPartitioningHierarchical_ReassembleFineparts(Mat adj, IS fineparts, ISLocalToGlobalMapping mapping, IS *sfineparts)
145: {
146: PetscInt *local_indices, *global_indices,*owners,*sfineparts_indices,localsize,i;
147: const PetscInt *ranges,*fineparts_indices;
148: PetscMPIInt rank;
149: MPI_Comm comm;
150: PetscLayout rmap;
151: PetscSFNode *remote;
152: PetscSF sf;
153: PetscErrorCode ierr;
156: PetscObjectGetComm((PetscObject)adj,&comm);
157: MPI_Comm_rank(comm,&rank);
158: MatGetLayouts(adj,&rmap,PETSC_NULL);
159: ISGetLocalSize(fineparts,&localsize);
160: PetscCalloc2(localsize,&global_indices,localsize,&local_indices);
161: for(i=0; i<localsize; i++){
162: local_indices[i] = i;
163: }
164: /* map local indices back to global so that we can permulate data globally */
165: ISLocalToGlobalMappingApply(mapping,localsize,local_indices,global_indices);
166: PetscCalloc1(localsize,&owners);
167: /* find owners for global indices */
168: for(i=0; i<localsize; i++){
169: PetscLayoutFindOwner(rmap,global_indices[i],&owners[i]);
170: }
171: PetscLayoutGetRanges(rmap,&ranges);
172: PetscCalloc1(ranges[rank+1]-ranges[rank],&sfineparts_indices);
173: ISGetIndices(fineparts,&fineparts_indices);
174: PetscSFCreate(comm,&sf);
175: PetscCalloc1(localsize,&remote);
176: for(i=0; i<localsize; i++){
177: remote[i].rank = owners[i];
178: remote[i].index = global_indices[i]-ranges[owners[i]];
179: }
180: PetscSFSetType(sf,PETSCSFBASIC);
181: /* not sure how to add prefix to sf */
182: PetscSFSetFromOptions(sf);
183: PetscSFSetGraph(sf,localsize,localsize,PETSC_NULL,PETSC_OWN_POINTER,remote,PETSC_OWN_POINTER);
184: PetscSFReduceBegin(sf,MPIU_INT,fineparts_indices,sfineparts_indices,MPIU_REPLACE);
185: PetscSFReduceEnd(sf,MPIU_INT,fineparts_indices,sfineparts_indices,MPIU_REPLACE);
186: PetscSFDestroy(&sf);
187: ISRestoreIndices(fineparts,&fineparts_indices);
188: ISCreateGeneral(comm,ranges[rank+1]-ranges[rank],sfineparts_indices,PETSC_OWN_POINTER,sfineparts);
189: PetscFree2(global_indices,local_indices);
190: PetscFree(owners);
191: return(0);
192: }
197: PetscErrorCode MatPartitioningHierarchical_AssembleSubdomain(Mat adj,IS destination,Mat *sadj, ISLocalToGlobalMapping *mapping)
198: {
199: IS irows,icols;
200: PetscInt irows_ln;
201: PetscMPIInt rank;
202: const PetscInt *irows_indices;
203: MPI_Comm comm;
204: PetscErrorCode ierr;
207: PetscObjectGetComm((PetscObject)adj,&comm);
208: MPI_Comm_rank(comm,&rank);
209: /* figure out where data comes from */
210: ISBuildTwoSided(destination,NULL,&irows);
211: ISDuplicate(irows,&icols);
212: ISGetLocalSize(irows,&irows_ln);
213: ISGetIndices(irows,&irows_indices);
214: ISLocalToGlobalMappingCreate(comm,1,irows_ln,irows_indices,PETSC_COPY_VALUES,mapping);
215: ISRestoreIndices(irows,&irows_indices);
216: MatGetSubMatrices(adj,1,&irows,&icols,MAT_INITIAL_MATRIX,&sadj);
217: ISDestroy(&irows);
218: ISDestroy(&icols);
219: return(0);
220: }
225: PetscErrorCode MatPartitioningHierarchical_DetermineDestination(MatPartitioning part, IS partitioning, PetscInt pstart, PetscInt pend, IS *destination)
226: {
227: MPI_Comm comm;
228: PetscMPIInt rank,size,target;
229: PetscInt plocalsize,*dest_indices,i;
230: const PetscInt *part_indices;
231: PetscErrorCode ierr;
234: PetscObjectGetComm((PetscObject)part,&comm);
235: MPI_Comm_rank(comm,&rank);
236: MPI_Comm_size(comm,&size);
237: 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);
238: if(pstart>pend) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP," pstart %D should be smaller than pend %D",pstart,pend);
239: ISGetLocalSize(partitioning,&plocalsize);
240: PetscCalloc1(plocalsize,&dest_indices);
241: ISGetIndices(partitioning,&part_indices);
242: for(i=0; i<plocalsize; i++){
243: /* compute target */
244: target = part_indices[i]-pstart;
245: /* mark out of range entity as -1 */
246: if(part_indices[i]<pstart || part_indices[i]>pend) target = -1;
247: dest_indices[i] = target;
248: }
249: ISCreateGeneral(comm,plocalsize,dest_indices,PETSC_OWN_POINTER,destination);
250: return(0);
251: }
256: PetscErrorCode MatPartitioningView_Hierarchical(MatPartitioning part,PetscViewer viewer)
257: {
258: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
259: PetscErrorCode ierr;
260: PetscMPIInt rank;
261: PetscBool iascii;
264: MPI_Comm_rank(PetscObjectComm((PetscObject)part),&rank);
265: PetscObjectTypeCompare((PetscObject)viewer,PETSCVIEWERASCII,&iascii);
266: if(iascii){
267: PetscViewerASCIIPrintf(viewer," Fine partitioner %s \n",hpart->fineparttype);
268: PetscViewerASCIIPrintf(viewer," Coarse partitioner %s \n",hpart->coarseparttype);
269: PetscViewerASCIIPrintf(viewer," Number of coarse parts %D \n",hpart->Ncoarseparts);
270: PetscViewerASCIIPrintf(viewer," Number of fine parts %D \n",hpart->Nfineparts);
271: }
272: return(0);
273: }
278: PetscErrorCode MatPartitioningHierarchicalGetFineparts(MatPartitioning part,IS *fineparts)
279: {
280: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
281: PetscErrorCode ierr;
284: *fineparts = hpart->fineparts;
285: PetscObjectReference((PetscObject)hpart->fineparts);
286: return(0);
287: }
291: PetscErrorCode MatPartitioningHierarchicalGetCoarseparts(MatPartitioning part,IS *coarseparts)
292: {
293: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
294: PetscErrorCode ierr;
297: *coarseparts = hpart->coarseparts;
298: PetscObjectReference((PetscObject)hpart->coarseparts);
299: return(0);
300: }
304: PetscErrorCode MatPartitioningHierarchicalSetNcoarseparts(MatPartitioning part, PetscInt Ncoarseparts)
305: {
306: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
309: hpart->Ncoarseparts = Ncoarseparts;
310: return(0);
311: }
315: PetscErrorCode MatPartitioningHierarchicalSetNfineparts(MatPartitioning part, PetscInt Nfineparts)
316: {
317: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
320: hpart->Nfineparts = Nfineparts;
321: return(0);
322: }
326: PetscErrorCode MatPartitioningSetFromOptions_Hierarchical(PetscOptionItems *PetscOptionsObject,MatPartitioning part)
327: {
328: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
330: char value[1024];
331: PetscBool flag = PETSC_FALSE;
334: PetscOptionsHead(PetscOptionsObject,"Set hierarchical partitioning options");
335: PetscOptionsString("-mat_partitioning_hierarchical_coarseparttype","coarse part type",PETSC_NULL,PETSC_NULL,value,1024,&flag);
336: if(flag){
337: PetscCalloc1(1024,&hpart->coarseparttype);
338: PetscStrcpy(hpart->coarseparttype,value);
339: }
340: PetscOptionsString("-mat_partitioning_hierarchical_fineparttype","fine part type",PETSC_NULL,PETSC_NULL,value,1024,&flag);
341: if(flag){
342: PetscCalloc1(1024,&hpart->fineparttype);
343: PetscStrcpy(hpart->fineparttype,value);
344: }
345: PetscOptionsInt("-mat_partitioning_hierarchical_Ncoarseparts","number of coarse parts",PETSC_NULL,0,&hpart->Ncoarseparts,&flag);
346: PetscOptionsInt("-mat_partitioning_hierarchical_Nfineparts","number of fine parts",PETSC_NULL,1,&hpart->Nfineparts,&flag);
347: PetscOptionsTail();
348: return(0);
349: }
354: PetscErrorCode MatPartitioningDestroy_Hierarchical(MatPartitioning part)
355: {
356: MatPartitioning_Hierarchical *hpart = (MatPartitioning_Hierarchical*)part->data;
357: PetscErrorCode ierr;
360: if(hpart->coarseparttype) {PetscFree(hpart->coarseparttype);}
361: if(hpart->fineparttype) {PetscFree(hpart->fineparttype);}
362: ISDestroy(&hpart->fineparts);
363: ISDestroy(&hpart->coarseparts);
364: PetscFree(hpart);
365: return(0);
366: }
369: /*MC
370: MATPARTITIONINGHIERARCHPART - Creates a partitioning context via hierarchical partitioning strategy.
372: Collective on MPI_Comm
374: Input Parameter:
375: . part - the partitioning context
377: Options Database Keys:
379: Level: beginner
381: .keywords: Partitioning, create, context
383: .seealso: MatPartitioningSetType(), MatPartitioningType
385: M*/
389: PETSC_EXTERN PetscErrorCode MatPartitioningCreate_Hierarchical(MatPartitioning part)
390: {
391: PetscErrorCode ierr;
392: MatPartitioning_Hierarchical *hpart;
395: PetscNewLog(part,&hpart);
396: part->data = (void*)hpart;
398: hpart->fineparttype = 0; /* fine level partitioner */
399: hpart->coarseparttype = 0; /* coarse level partitioner */
400: hpart->Nfineparts = 1; /* we do not further partition coarse partition any more by default */
401: hpart->Ncoarseparts = 0; /* number of coarse parts (first level) */
402: hpart->coarseparts = 0;
403: hpart->fineparts = 0;
405: part->ops->apply = MatPartitioningApply_Hierarchical;
406: part->ops->view = MatPartitioningView_Hierarchical;
407: part->ops->destroy = MatPartitioningDestroy_Hierarchical;
408: part->ops->setfromoptions = MatPartitioningSetFromOptions_Hierarchical;
409: return(0);
410: }