Actual source code: partition.c
2: #include <petsc/private/matimpl.h>
4: /* Logging support */
5: PetscClassId MAT_PARTITIONING_CLASSID;
7: /*
8: Simplest partitioning, keeps the current partitioning.
9: */
10: static PetscErrorCode MatPartitioningApply_Current(MatPartitioning part, IS *partitioning)
11: {
12: PetscInt m;
13: PetscMPIInt rank, size;
15: PetscFunctionBegin;
16: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)part), &size));
17: if (part->n != size) {
18: const char *prefix;
19: PetscCall(PetscObjectGetOptionsPrefix((PetscObject)part, &prefix));
20: SETERRQ(PetscObjectComm((PetscObject)part), PETSC_ERR_SUP, "This is the DEFAULT NO-OP partitioner, it currently only supports one domain per processor\nuse -%smat_partitioning_type parmetis or chaco or ptscotch for more than one subdomain per processor", prefix ? prefix : "");
21: }
22: PetscCallMPI(MPI_Comm_rank(PetscObjectComm((PetscObject)part), &rank));
24: PetscCall(MatGetLocalSize(part->adj, &m, NULL));
25: PetscCall(ISCreateStride(PetscObjectComm((PetscObject)part), m, rank, 0, partitioning));
26: PetscFunctionReturn(PETSC_SUCCESS);
27: }
29: /*
30: partition an index to rebalance the computation
31: */
32: static PetscErrorCode MatPartitioningApply_Average(MatPartitioning part, IS *partitioning)
33: {
34: PetscInt m, M, nparts, *indices, r, d, *parts, i, start, end, loc;
36: PetscFunctionBegin;
37: PetscCall(MatGetSize(part->adj, &M, NULL));
38: PetscCall(MatGetLocalSize(part->adj, &m, NULL));
39: nparts = part->n;
40: PetscCall(PetscMalloc1(nparts, &parts));
41: d = M / nparts;
42: for (i = 0; i < nparts; i++) parts[i] = d;
43: r = M % nparts;
44: for (i = 0; i < r; i++) parts[i] += 1;
45: for (i = 1; i < nparts; i++) parts[i] += parts[i - 1];
46: PetscCall(PetscMalloc1(m, &indices));
47: PetscCall(MatGetOwnershipRange(part->adj, &start, &end));
48: for (i = start; i < end; i++) {
49: PetscCall(PetscFindInt(i, nparts, parts, &loc));
50: if (loc < 0) loc = -(loc + 1);
51: else loc = loc + 1;
52: indices[i - start] = loc;
53: }
54: PetscCall(PetscFree(parts));
55: PetscCall(ISCreateGeneral(PetscObjectComm((PetscObject)part), m, indices, PETSC_OWN_POINTER, partitioning));
56: PetscFunctionReturn(PETSC_SUCCESS);
57: }
59: static PetscErrorCode MatPartitioningApply_Square(MatPartitioning part, IS *partitioning)
60: {
61: PetscInt cell, n, N, p, rstart, rend, *color;
62: PetscMPIInt size;
64: PetscFunctionBegin;
65: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)part), &size));
66: PetscCheck(part->n == size, PetscObjectComm((PetscObject)part), PETSC_ERR_SUP, "Currently only supports one domain per processor");
67: p = (PetscInt)PetscSqrtReal((PetscReal)part->n);
68: PetscCheck(p * p == part->n, PetscObjectComm((PetscObject)part), PETSC_ERR_SUP, "Square partitioning requires \"perfect square\" number of domains");
70: PetscCall(MatGetSize(part->adj, &N, NULL));
71: n = (PetscInt)PetscSqrtReal((PetscReal)N);
72: PetscCheck(n * n == N, PetscObjectComm((PetscObject)part), PETSC_ERR_SUP, "Square partitioning requires square domain");
73: PetscCheck(n % p == 0, PETSC_COMM_SELF, PETSC_ERR_SUP, "Square partitioning requires p to divide n");
74: PetscCall(MatGetOwnershipRange(part->adj, &rstart, &rend));
75: PetscCall(PetscMalloc1(rend - rstart, &color));
76: /* for (int cell=rstart; cell<rend; cell++) color[cell-rstart] = ((cell%n) < (n/2)) + 2 * ((cell/n) < (n/2)); */
77: for (cell = rstart; cell < rend; cell++) color[cell - rstart] = ((cell % n) / (n / p)) + p * ((cell / n) / (n / p));
78: PetscCall(ISCreateGeneral(PetscObjectComm((PetscObject)part), rend - rstart, color, PETSC_OWN_POINTER, partitioning));
79: PetscFunctionReturn(PETSC_SUCCESS);
80: }
82: PETSC_EXTERN PetscErrorCode MatPartitioningCreate_Current(MatPartitioning part)
83: {
84: PetscFunctionBegin;
85: part->ops->apply = MatPartitioningApply_Current;
86: part->ops->view = NULL;
87: part->ops->destroy = NULL;
88: PetscFunctionReturn(PETSC_SUCCESS);
89: }
91: PETSC_EXTERN PetscErrorCode MatPartitioningCreate_Average(MatPartitioning part)
92: {
93: PetscFunctionBegin;
94: part->ops->apply = MatPartitioningApply_Average;
95: part->ops->view = NULL;
96: part->ops->destroy = NULL;
97: PetscFunctionReturn(PETSC_SUCCESS);
98: }
100: PETSC_EXTERN PetscErrorCode MatPartitioningCreate_Square(MatPartitioning part)
101: {
102: PetscFunctionBegin;
103: part->ops->apply = MatPartitioningApply_Square;
104: part->ops->view = NULL;
105: part->ops->destroy = NULL;
106: PetscFunctionReturn(PETSC_SUCCESS);
107: }
109: /* gets as input the "sizes" array computed by ParMetis_*_NodeND and returns
110: seps[ 0 : 2*p) : the start and end node of each subdomain
111: seps[2*p : 2*p+2*(p-1)) : the start and end node of each separator
112: levels[ 0 : p-1) : level in the tree for each separator (-1 root, -2 and -3 first level and so on)
113: The arrays must be large enough
114: */
115: PETSC_INTERN PetscErrorCode MatPartitioningSizesToSep_Private(PetscInt p, PetscInt sizes[], PetscInt seps[], PetscInt level[])
116: {
117: PetscInt l2p, i, pTree, pStartTree;
119: PetscFunctionBegin;
120: l2p = PetscLog2Real(p);
121: PetscCheck(!(l2p - (PetscInt)PetscLog2Real(p)), PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "%" PetscInt_FMT " is not a power of 2", p);
122: if (!p) PetscFunctionReturn(PETSC_SUCCESS);
123: PetscCall(PetscArrayzero(seps, 2 * p - 2));
124: PetscCall(PetscArrayzero(level, p - 1));
125: seps[2 * p - 2] = sizes[2 * p - 2];
126: pTree = p;
127: pStartTree = 0;
128: while (pTree != 1) {
129: for (i = pStartTree; i < pStartTree + pTree; i++) {
130: seps[i] += sizes[i];
131: seps[pStartTree + pTree + (i - pStartTree) / 2] += seps[i];
132: }
133: pStartTree += pTree;
134: pTree = pTree / 2;
135: }
136: seps[2 * p - 2] -= sizes[2 * p - 2];
138: pStartTree = 2 * p - 2;
139: pTree = 1;
140: while (pStartTree > 0) {
141: for (i = pStartTree; i < pStartTree + pTree; i++) {
142: PetscInt k = 2 * i - (pStartTree + 2 * pTree);
143: PetscInt n = seps[k + 1];
145: seps[k + 1] = seps[i] - sizes[k + 1];
146: seps[k] = seps[k + 1] + sizes[k + 1] - n - sizes[k];
147: level[i - p] = -pTree - i + pStartTree;
148: }
149: pTree *= 2;
150: pStartTree -= pTree;
151: }
152: /* I know there should be a formula */
153: PetscCall(PetscSortIntWithArrayPair(p - 1, seps + p, sizes + p, level));
154: for (i = 2 * p - 2; i >= 0; i--) {
155: seps[2 * i] = seps[i];
156: seps[2 * i + 1] = seps[i] + PetscMax(sizes[i] - 1, 0);
157: }
158: PetscFunctionReturn(PETSC_SUCCESS);
159: }
161: PetscFunctionList MatPartitioningList = NULL;
162: PetscBool MatPartitioningRegisterAllCalled = PETSC_FALSE;
164: /*@C
165: MatPartitioningRegister - Adds a new sparse matrix partitioning to the matrix package.
167: Not Collective
169: Input Parameters:
170: + sname - name of partitioning (for example `MATPARTITIONINGCURRENT`) or `MATPARTITIONINGPARMETIS`
171: - function - function pointer that creates the partitioning type
173: Level: developer
175: Sample usage:
176: .vb
177: MatPartitioningRegister("my_part", MyPartCreate);
178: .ve
180: Then, your partitioner can be chosen with the procedural interface via
181: $ MatPartitioningSetType(part, "my_part")
182: or at runtime via the option
183: $ -mat_partitioning_type my_part
185: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `MatPartitioningType`, `MatPartitioningCreate()`, `MatPartitioningRegisterDestroy()`, `MatPartitioningRegisterAll()`
186: @*/
187: PetscErrorCode MatPartitioningRegister(const char sname[], PetscErrorCode (*function)(MatPartitioning))
188: {
189: PetscFunctionBegin;
190: PetscCall(MatInitializePackage());
191: PetscCall(PetscFunctionListAdd(&MatPartitioningList, sname, function));
192: PetscFunctionReturn(PETSC_SUCCESS);
193: }
195: /*@C
196: MatPartitioningGetType - Gets the Partitioning method type and name (as a string)
197: from the partitioning context.
199: Not Collective
201: Input Parameter:
202: . partitioning - the partitioning context
204: Output Parameter:
205: . type - partitioner type
207: Level: intermediate
209: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `MatPartitioningType`, `MatPartitioningCreate()`, `MatPartitioningRegisterDestroy()`, `MatPartitioningRegisterAll()`
210: @*/
211: PetscErrorCode MatPartitioningGetType(MatPartitioning partitioning, MatPartitioningType *type)
212: {
213: PetscFunctionBegin;
216: *type = ((PetscObject)partitioning)->type_name;
217: PetscFunctionReturn(PETSC_SUCCESS);
218: }
220: /*@C
221: MatPartitioningSetNParts - Set how many partitions need to be created;
222: by default this is one per processor. Certain partitioning schemes may
223: in fact only support that option.
225: Collective
227: Input Parameters:
228: + partitioning - the partitioning context
229: - n - the number of partitions
231: Level: intermediate
233: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `MatPartitioningCreate()`, `MatPartitioningApply()`
234: @*/
235: PetscErrorCode MatPartitioningSetNParts(MatPartitioning part, PetscInt n)
236: {
237: PetscFunctionBegin;
238: part->n = n;
239: PetscFunctionReturn(PETSC_SUCCESS);
240: }
242: /*@
243: MatPartitioningApplyND - Gets a nested dissection partitioning for a matrix.
245: Collective
247: Input Parameter:
248: . matp - the matrix partitioning object
250: Output Parameter:
251: . partitioning - the partitioning. For each local node, a positive value indicates the processor
252: number the node has been assigned to. Negative x values indicate the separator level -(x+1).
254: Level: intermediate
256: Note:
257: The user can define additional partitionings; see `MatPartitioningRegister()`.
259: .seealso: [](ch_matrices), `Mat`, `MatPartitioningApplyND()`, `MatPartitioningRegister()`, `MatPartitioningCreate()`,
260: `MatPartitioningDestroy()`, `MatPartitioningSetAdjacency()`, `ISPartitioningToNumbering()`,
261: `ISPartitioningCount()`
262: @*/
263: PetscErrorCode MatPartitioningApplyND(MatPartitioning matp, IS *partitioning)
264: {
265: PetscFunctionBegin;
268: PetscCheck(matp->adj->assembled, PetscObjectComm((PetscObject)matp), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
269: PetscCheck(!matp->adj->factortype, PetscObjectComm((PetscObject)matp), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
270: PetscCall(PetscLogEventBegin(MAT_PartitioningND, matp, 0, 0, 0));
271: PetscUseTypeMethod(matp, applynd, partitioning);
272: PetscCall(PetscLogEventEnd(MAT_PartitioningND, matp, 0, 0, 0));
274: PetscCall(MatPartitioningViewFromOptions(matp, NULL, "-mat_partitioning_view"));
275: PetscCall(ISViewFromOptions(*partitioning, NULL, "-mat_partitioning_view"));
276: PetscFunctionReturn(PETSC_SUCCESS);
277: }
279: /*@
280: MatPartitioningApply - Gets a partitioning for the graph represented by a sparse matrix.
282: Collective
284: Input Parameter:
285: . matp - the matrix partitioning object
287: Output Parameter:
288: . partitioning - the partitioning. For each local node this tells the processor
289: number that that node is assigned to.
291: Options Database Keys:
292: + -mat_partitioning_type <type> - set the partitioning package or algorithm to use
293: - -mat_partitioning_view - display information about the partitioning object
295: Level: beginner
297: The user can define additional partitionings; see `MatPartitioningRegister()`.
299: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `MatPartitioningType`, `MatPartitioningRegister()`, `MatPartitioningCreate()`,
300: `MatPartitioningDestroy()`, `MatPartitioningSetAdjacency()`, `ISPartitioningToNumbering()`,
301: `ISPartitioningCount()`
302: @*/
303: PetscErrorCode MatPartitioningApply(MatPartitioning matp, IS *partitioning)
304: {
305: PetscBool viewbalance, improve;
307: PetscFunctionBegin;
310: PetscCheck(matp->adj->assembled, PetscObjectComm((PetscObject)matp), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
311: PetscCheck(!matp->adj->factortype, PetscObjectComm((PetscObject)matp), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
312: PetscCall(PetscLogEventBegin(MAT_Partitioning, matp, 0, 0, 0));
313: PetscUseTypeMethod(matp, apply, partitioning);
314: PetscCall(PetscLogEventEnd(MAT_Partitioning, matp, 0, 0, 0));
316: PetscCall(MatPartitioningViewFromOptions(matp, NULL, "-mat_partitioning_view"));
317: PetscCall(ISViewFromOptions(*partitioning, NULL, "-mat_partitioning_view"));
319: PetscObjectOptionsBegin((PetscObject)matp);
320: viewbalance = PETSC_FALSE;
321: PetscCall(PetscOptionsBool("-mat_partitioning_view_imbalance", "Display imbalance information of a partition", NULL, PETSC_FALSE, &viewbalance, NULL));
322: improve = PETSC_FALSE;
323: PetscCall(PetscOptionsBool("-mat_partitioning_improve", "Improve the quality of a partition", NULL, PETSC_FALSE, &improve, NULL));
324: PetscOptionsEnd();
326: if (improve) PetscCall(MatPartitioningImprove(matp, partitioning));
328: if (viewbalance) PetscCall(MatPartitioningViewImbalance(matp, *partitioning));
329: PetscFunctionReturn(PETSC_SUCCESS);
330: }
332: /*@
333: MatPartitioningImprove - Improves the quality of a given partition.
335: Collective
337: Input Parameters:
338: + matp - the matrix partitioning object
339: - partitioning - the original partitioning. For each local node this tells the processor
340: number that that node is assigned to.
342: Options Database Key:
343: . -mat_partitioning_improve - improve the quality of the given partition
345: Level: beginner
347: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `MatPartitioningType`, `MatPartitioningApply()`, `MatPartitioningCreate()`,
348: `MatPartitioningDestroy()`, `MatPartitioningSetAdjacency()`, `ISPartitioningToNumbering()`,
349: `ISPartitioningCount()`
350: @*/
351: PetscErrorCode MatPartitioningImprove(MatPartitioning matp, IS *partitioning)
352: {
353: PetscFunctionBegin;
356: PetscCheck(matp->adj->assembled, PetscObjectComm((PetscObject)matp), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
357: PetscCheck(!matp->adj->factortype, PetscObjectComm((PetscObject)matp), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
358: PetscCall(PetscLogEventBegin(MAT_Partitioning, matp, 0, 0, 0));
359: PetscTryTypeMethod(matp, improve, partitioning);
360: PetscCall(PetscLogEventEnd(MAT_Partitioning, matp, 0, 0, 0));
361: PetscFunctionReturn(PETSC_SUCCESS);
362: }
364: /*@
365: MatPartitioningViewImbalance - Display partitioning imbalance information.
367: Collective
369: Input Parameters:
370: + matp - the matrix partitioning object
371: - partitioning - the partitioning. For each local node this tells the processor
372: number that that node is assigned to.
374: Options Database Key:
375: . -mat_partitioning_view_balance - view the balance information from the last partitioning
377: Level: beginner
379: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `MatPartitioningType`, `MatPartitioningApply()`, `MatPartitioningView()`
380: @*/
381: PetscErrorCode MatPartitioningViewImbalance(MatPartitioning matp, IS partitioning)
382: {
383: PetscInt nparts, *subdomainsizes, *subdomainsizes_tmp, nlocal, i, maxsub, minsub, avgsub;
384: const PetscInt *indices;
385: PetscViewer viewer;
387: PetscFunctionBegin;
390: nparts = matp->n;
391: PetscCall(PetscCalloc2(nparts, &subdomainsizes, nparts, &subdomainsizes_tmp));
392: PetscCall(ISGetLocalSize(partitioning, &nlocal));
393: PetscCall(ISGetIndices(partitioning, &indices));
394: for (i = 0; i < nlocal; i++) subdomainsizes_tmp[indices[i]] += matp->vertex_weights ? matp->vertex_weights[i] : 1;
395: PetscCall(MPIU_Allreduce(subdomainsizes_tmp, subdomainsizes, nparts, MPIU_INT, MPI_SUM, PetscObjectComm((PetscObject)matp)));
396: PetscCall(ISRestoreIndices(partitioning, &indices));
397: minsub = PETSC_MAX_INT, maxsub = PETSC_MIN_INT, avgsub = 0;
398: for (i = 0; i < nparts; i++) {
399: minsub = PetscMin(minsub, subdomainsizes[i]);
400: maxsub = PetscMax(maxsub, subdomainsizes[i]);
401: avgsub += subdomainsizes[i];
402: }
403: avgsub /= nparts;
404: PetscCall(PetscFree2(subdomainsizes, subdomainsizes_tmp));
405: PetscCall(PetscViewerASCIIGetStdout(PetscObjectComm((PetscObject)matp), &viewer));
406: PetscCall(MatPartitioningView(matp, viewer));
407: PetscCall(PetscViewerASCIIPrintf(viewer, "Partitioning Imbalance Info: Max %" PetscInt_FMT ", Min %" PetscInt_FMT ", Avg %" PetscInt_FMT ", R %g\n", maxsub, minsub, avgsub, (double)(maxsub / (PetscReal)minsub)));
408: PetscFunctionReturn(PETSC_SUCCESS);
409: }
411: /*@
412: MatPartitioningSetAdjacency - Sets the adjacency graph (matrix) of the thing to be
413: partitioned.
415: Collective
417: Input Parameters:
418: + part - the partitioning context
419: - adj - the adjacency matrix, this can be any `MatType` but the natural representation is `MATMPIADJ`
421: Level: beginner
423: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `MatPartitioningType`, `MatPartitioningCreate()`
424: @*/
425: PetscErrorCode MatPartitioningSetAdjacency(MatPartitioning part, Mat adj)
426: {
427: PetscFunctionBegin;
430: part->adj = adj;
431: PetscFunctionReturn(PETSC_SUCCESS);
432: }
434: /*@
435: MatPartitioningDestroy - Destroys the partitioning context.
437: Collective
439: Input Parameter:
440: . part - the partitioning context
442: Level: beginner
444: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `MatPartitioningType`, `MatPartitioningCreate()`
445: @*/
446: PetscErrorCode MatPartitioningDestroy(MatPartitioning *part)
447: {
448: PetscFunctionBegin;
449: if (!*part) PetscFunctionReturn(PETSC_SUCCESS);
451: if (--((PetscObject)(*part))->refct > 0) {
452: *part = NULL;
453: PetscFunctionReturn(PETSC_SUCCESS);
454: }
456: if ((*part)->ops->destroy) PetscCall((*(*part)->ops->destroy)((*part)));
457: PetscCall(PetscFree((*part)->vertex_weights));
458: PetscCall(PetscFree((*part)->part_weights));
459: PetscCall(PetscHeaderDestroy(part));
460: PetscFunctionReturn(PETSC_SUCCESS);
461: }
463: /*@C
464: MatPartitioningSetVertexWeights - Sets the weights for vertices for a partitioning.
466: Logically Collective
468: Input Parameters:
469: + part - the partitioning context
470: - weights - the weights, on each process this array must have the same size as the number of local rows times the value passed with `MatPartitioningSetNumberVertexWeights()` or
471: 1 if that is not provided
473: Level: beginner
475: Notes:
476: The array weights is freed by PETSc so the user should not free the array. In C/C++
477: the array must be obtained with a call to `PetscMalloc()`, not malloc().
479: The weights may not be used by some partitioners
481: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `MatPartitioningCreate()`, `MatPartitioningSetType()`, `MatPartitioningSetPartitionWeights()`, `MatPartitioningSetNumberVertexWeights()`
482: @*/
483: PetscErrorCode MatPartitioningSetVertexWeights(MatPartitioning part, const PetscInt weights[])
484: {
485: PetscFunctionBegin;
487: PetscCall(PetscFree(part->vertex_weights));
488: part->vertex_weights = (PetscInt *)weights;
489: PetscFunctionReturn(PETSC_SUCCESS);
490: }
492: /*@C
493: MatPartitioningSetPartitionWeights - Sets the weights for each partition.
495: Logically Collective
497: Input Parameters:
498: + part - the partitioning context
499: - weights - An array of size nparts that is used to specify the fraction of
500: vertex weight that should be distributed to each sub-domain for
501: the balance constraint. If all of the sub-domains are to be of
502: the same size, then each of the nparts elements should be set
503: to a value of 1/nparts. Note that the sum of all of the weights
504: should be one.
506: Level: beginner
508: Note:
509: The array weights is freed by PETSc so the user should not free the array. In C/C++
510: the array must be obtained with a call to `PetscMalloc()`, not malloc().
512: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `MatPartitioningSetVertexWeights()`, `MatPartitioningCreate()`, `MatPartitioningSetType()`, `MatPartitioningSetVertexWeights()`
513: @*/
514: PetscErrorCode MatPartitioningSetPartitionWeights(MatPartitioning part, const PetscReal weights[])
515: {
516: PetscFunctionBegin;
518: PetscCall(PetscFree(part->part_weights));
519: part->part_weights = (PetscReal *)weights;
520: PetscFunctionReturn(PETSC_SUCCESS);
521: }
523: /*@
524: MatPartitioningSetUseEdgeWeights - Set a flag to indicate whether or not to use edge weights.
526: Logically Collective
528: Input Parameters:
529: + part - the partitioning context
530: - use_edge_weights - the flag indicateing whether or not to use edge weights. By default no edge weights will be used,
531: that is, use_edge_weights is set to FALSE. If set use_edge_weights to TRUE, users need to make sure legal
532: edge weights are stored in an ADJ matrix.
534: Options Database Key:
535: . -mat_partitioning_use_edge_weights - (true or false)
537: Level: beginner
539: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `MatPartitioningCreate()`, `MatPartitioningSetType()`, `MatPartitioningSetVertexWeights()`, `MatPartitioningSetPartitionWeights()`
540: @*/
541: PetscErrorCode MatPartitioningSetUseEdgeWeights(MatPartitioning part, PetscBool use_edge_weights)
542: {
543: PetscFunctionBegin;
545: part->use_edge_weights = use_edge_weights;
546: PetscFunctionReturn(PETSC_SUCCESS);
547: }
549: /*@
550: MatPartitioningGetUseEdgeWeights - Get a flag that indicates whether or not to edge weights are used.
552: Logically Collective
554: Input Parameter:
555: . part - the partitioning context
557: Output Parameter:
558: . use_edge_weights - the flag indicateing whether or not to edge weights are used.
560: Level: beginner
562: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `MatPartitioningCreate()`, `MatPartitioningSetType()`, `MatPartitioningSetVertexWeights()`, `MatPartitioningSetPartitionWeights()`,
563: `MatPartitioningSetUseEdgeWeights`
564: @*/
565: PetscErrorCode MatPartitioningGetUseEdgeWeights(MatPartitioning part, PetscBool *use_edge_weights)
566: {
567: PetscFunctionBegin;
570: *use_edge_weights = part->use_edge_weights;
571: PetscFunctionReturn(PETSC_SUCCESS);
572: }
574: /*@
575: MatPartitioningCreate - Creates a partitioning context.
577: Collective
579: Input Parameter:
580: . comm - MPI communicator
582: Output Parameter:
583: . newp - location to put the context
585: Level: beginner
587: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `MatPartitioningSetType()`, `MatPartitioningApply()`, `MatPartitioningDestroy()`,
588: `MatPartitioningSetAdjacency()`
589: @*/
590: PetscErrorCode MatPartitioningCreate(MPI_Comm comm, MatPartitioning *newp)
591: {
592: MatPartitioning part;
593: PetscMPIInt size;
595: PetscFunctionBegin;
596: *newp = NULL;
598: PetscCall(MatInitializePackage());
599: PetscCall(PetscHeaderCreate(part, MAT_PARTITIONING_CLASSID, "MatPartitioning", "Matrix/graph partitioning", "MatOrderings", comm, MatPartitioningDestroy, MatPartitioningView));
600: part->vertex_weights = NULL;
601: part->part_weights = NULL;
602: part->use_edge_weights = PETSC_FALSE; /* By default we don't use edge weights */
604: PetscCallMPI(MPI_Comm_size(comm, &size));
605: part->n = (PetscInt)size;
606: part->ncon = 1;
608: *newp = part;
609: PetscFunctionReturn(PETSC_SUCCESS);
610: }
612: /*@C
613: MatPartitioningViewFromOptions - View a partitioning context from the options database
615: Collective
617: Input Parameters:
618: + A - the partitioning context
619: . obj - Optional object that provides the prefix used in the options database check
620: - name - command line option
622: Options Database Key:
623: . -mat_partitioning_view [viewertype]:... - the viewer and its options
625: Level: intermediate
627: Note:
628: .vb
629: If no value is provided ascii:stdout is used
630: ascii[:[filename][:[format][:append]]] defaults to stdout - format can be one of ascii_info, ascii_info_detail, or ascii_matlab,
631: for example ascii::ascii_info prints just the information about the object not all details
632: unless :append is given filename opens in write mode, overwriting what was already there
633: binary[:[filename][:[format][:append]]] defaults to the file binaryoutput
634: draw[:drawtype[:filename]] for example, draw:tikz, draw:tikz:figure.tex or draw:x
635: socket[:port] defaults to the standard output port
636: saws[:communicatorname] publishes object to the Scientific Application Webserver (SAWs)
637: .ve
639: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `MatPartitioningView()`, `PetscObjectViewFromOptions()`, `MatPartitioningCreate()`
640: @*/
641: PetscErrorCode MatPartitioningViewFromOptions(MatPartitioning A, PetscObject obj, const char name[])
642: {
643: PetscFunctionBegin;
645: PetscCall(PetscObjectViewFromOptions((PetscObject)A, obj, name));
646: PetscFunctionReturn(PETSC_SUCCESS);
647: }
649: /*@C
650: MatPartitioningView - Prints the partitioning data structure.
652: Collective
654: Input Parameters:
655: + part - the partitioning context
656: - viewer - optional visualization context
658: Level: intermediate
660: Note:
661: The available visualization contexts include
662: + `PETSC_VIEWER_STDOUT_SELF` - standard output (default)
663: - `PETSC_VIEWER_STDOUT_WORLD` - synchronized standard
664: output where only the first processor opens
665: the file. All other processors send their
666: data to the first processor to print.
668: The user can open alternative visualization contexts with
669: . `PetscViewerASCIIOpen()` - output to a specified file
671: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `PetscViewer`, `PetscViewerASCIIOpen()`
672: @*/
673: PetscErrorCode MatPartitioningView(MatPartitioning part, PetscViewer viewer)
674: {
675: PetscBool iascii;
677: PetscFunctionBegin;
679: if (!viewer) PetscCall(PetscViewerASCIIGetStdout(PetscObjectComm((PetscObject)part), &viewer));
681: PetscCheckSameComm(part, 1, viewer, 2);
683: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &iascii));
684: if (iascii) {
685: PetscCall(PetscObjectPrintClassNamePrefixType((PetscObject)part, viewer));
686: if (part->vertex_weights) PetscCall(PetscViewerASCIIPrintf(viewer, " Using vertex weights\n"));
687: }
688: PetscCall(PetscViewerASCIIPushTab(viewer));
689: PetscTryTypeMethod(part, view, viewer);
690: PetscCall(PetscViewerASCIIPopTab(viewer));
691: PetscFunctionReturn(PETSC_SUCCESS);
692: }
694: /*@C
695: MatPartitioningSetType - Sets the type of partitioner to use
697: Collective
699: Input Parameters:
700: + part - the partitioning context.
701: - type - a known method
703: Options Database Key:
704: . -mat_partitioning_type <type> - (for instance, parmetis), use -help for a list of available methods or see `MatPartitioningType`
706: Level: intermediate
708: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `MatPartitioningCreate()`, `MatPartitioningApply()`, `MatPartitioningType`
709: @*/
710: PetscErrorCode MatPartitioningSetType(MatPartitioning part, MatPartitioningType type)
711: {
712: PetscBool match;
713: PetscErrorCode (*r)(MatPartitioning);
715: PetscFunctionBegin;
719: PetscCall(PetscObjectTypeCompare((PetscObject)part, type, &match));
720: if (match) PetscFunctionReturn(PETSC_SUCCESS);
722: PetscTryTypeMethod(part, destroy);
723: part->ops->destroy = NULL;
725: part->setupcalled = 0;
726: part->data = NULL;
727: PetscCall(PetscMemzero(part->ops, sizeof(struct _MatPartitioningOps)));
729: PetscCall(PetscFunctionListFind(MatPartitioningList, type, &r));
730: PetscCheck(r, PetscObjectComm((PetscObject)part), PETSC_ERR_ARG_UNKNOWN_TYPE, "Unknown partitioning type %s", type);
732: PetscCall((*r)(part));
734: PetscCall(PetscFree(((PetscObject)part)->type_name));
735: PetscCall(PetscStrallocpy(type, &((PetscObject)part)->type_name));
736: PetscFunctionReturn(PETSC_SUCCESS);
737: }
739: /*@
740: MatPartitioningSetFromOptions - Sets various partitioning options from the
741: options database for the partitioning object
743: Collective
745: Input Parameter:
746: . part - the partitioning context.
748: Options Database Keys:
749: + -mat_partitioning_type <type> - (for instance, parmetis), use -help for a list of available methods
750: - -mat_partitioning_nparts - number of subgraphs
752: Level: beginner
754: Note:
755: If the partitioner has not been set by the user it uses one of the installed partitioner such as ParMetis. If there are
756: no installed partitioners it does no repartioning.
758: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`
759: @*/
760: PetscErrorCode MatPartitioningSetFromOptions(MatPartitioning part)
761: {
762: PetscBool flag;
763: char type[256];
764: const char *def;
766: PetscFunctionBegin;
767: PetscObjectOptionsBegin((PetscObject)part);
768: if (!((PetscObject)part)->type_name) {
769: #if defined(PETSC_HAVE_PARMETIS)
770: def = MATPARTITIONINGPARMETIS;
771: #elif defined(PETSC_HAVE_CHACO)
772: def = MATPARTITIONINGCHACO;
773: #elif defined(PETSC_HAVE_PARTY)
774: def = MATPARTITIONINGPARTY;
775: #elif defined(PETSC_HAVE_PTSCOTCH)
776: def = MATPARTITIONINGPTSCOTCH;
777: #else
778: def = MATPARTITIONINGCURRENT;
779: #endif
780: } else {
781: def = ((PetscObject)part)->type_name;
782: }
783: PetscCall(PetscOptionsFList("-mat_partitioning_type", "Type of partitioner", "MatPartitioningSetType", MatPartitioningList, def, type, 256, &flag));
784: if (flag) PetscCall(MatPartitioningSetType(part, type));
786: PetscCall(PetscOptionsInt("-mat_partitioning_nparts", "number of fine parts", NULL, part->n, &part->n, &flag));
788: PetscCall(PetscOptionsBool("-mat_partitioning_use_edge_weights", "whether or not to use edge weights", NULL, part->use_edge_weights, &part->use_edge_weights, &flag));
790: /*
791: Set the type if it was never set.
792: */
793: if (!((PetscObject)part)->type_name) PetscCall(MatPartitioningSetType(part, def));
795: PetscTryTypeMethod(part, setfromoptions, PetscOptionsObject);
796: PetscOptionsEnd();
797: PetscFunctionReturn(PETSC_SUCCESS);
798: }
800: /*@C
801: MatPartitioningSetNumberVertexWeights - Sets the number of weights per vertex
803: Not Collective
805: Input Parameters:
806: + partitioning - the partitioning context
807: - ncon - the number of weights
809: Level: intermediate
811: .seealso: [](ch_matrices), `Mat`, `MatPartitioning`, `MatPartitioningSetVertexWeights()`
812: @*/
813: PetscErrorCode MatPartitioningSetNumberVertexWeights(MatPartitioning partitioning, PetscInt ncon)
814: {
815: PetscFunctionBegin;
817: partitioning->ncon = ncon;
818: PetscFunctionReturn(PETSC_SUCCESS);
819: }