Actual source code: matrix.c
1: /*
2: This is where the abstract matrix operations are defined
3: Portions of this code are under:
4: Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved.
5: */
7: #include <petsc/private/matimpl.h>
8: #include <petsc/private/isimpl.h>
9: #include <petsc/private/vecimpl.h>
11: /* Logging support */
12: PetscClassId MAT_CLASSID;
13: PetscClassId MAT_COLORING_CLASSID;
14: PetscClassId MAT_FDCOLORING_CLASSID;
15: PetscClassId MAT_TRANSPOSECOLORING_CLASSID;
17: PetscLogEvent MAT_Mult, MAT_MultAdd, MAT_MultTranspose;
18: PetscLogEvent MAT_ADot, MAT_ANorm;
19: PetscLogEvent MAT_MultTransposeAdd, MAT_Solve, MAT_Solves, MAT_SolveAdd, MAT_SolveTranspose, MAT_MatSolve, MAT_MatTrSolve;
20: PetscLogEvent MAT_SolveTransposeAdd, MAT_SOR, MAT_ForwardSolve, MAT_BackwardSolve, MAT_LUFactor, MAT_LUFactorSymbolic;
21: PetscLogEvent MAT_LUFactorNumeric, MAT_CholeskyFactor, MAT_CholeskyFactorSymbolic, MAT_CholeskyFactorNumeric, MAT_ILUFactor;
22: PetscLogEvent MAT_ILUFactorSymbolic, MAT_ICCFactorSymbolic, MAT_Copy, MAT_Convert, MAT_Scale, MAT_AssemblyBegin;
23: PetscLogEvent MAT_QRFactorNumeric, MAT_QRFactorSymbolic, MAT_QRFactor;
24: PetscLogEvent MAT_AssemblyEnd, MAT_SetValues, MAT_GetValues, MAT_GetRow, MAT_GetRowIJ, MAT_CreateSubMats, MAT_GetOrdering, MAT_RedundantMat, MAT_GetSeqNonzeroStructure;
25: PetscLogEvent MAT_IncreaseOverlap, MAT_Partitioning, MAT_PartitioningND, MAT_Coarsen, MAT_ZeroEntries, MAT_Load, MAT_View, MAT_AXPY, MAT_FDColoringCreate;
26: PetscLogEvent MAT_FDColoringSetUp, MAT_FDColoringApply, MAT_Transpose, MAT_FDColoringFunction, MAT_CreateSubMat;
27: PetscLogEvent MAT_TransposeColoringCreate;
28: PetscLogEvent MAT_MatMult, MAT_MatMultSymbolic, MAT_MatMultNumeric;
29: PetscLogEvent MAT_PtAP, MAT_PtAPSymbolic, MAT_PtAPNumeric, MAT_RARt, MAT_RARtSymbolic, MAT_RARtNumeric;
30: PetscLogEvent MAT_MatTransposeMult, MAT_MatTransposeMultSymbolic, MAT_MatTransposeMultNumeric;
31: PetscLogEvent MAT_TransposeMatMult, MAT_TransposeMatMultSymbolic, MAT_TransposeMatMultNumeric;
32: PetscLogEvent MAT_MatMatMult, MAT_MatMatMultSymbolic, MAT_MatMatMultNumeric;
33: PetscLogEvent MAT_MultHermitianTranspose, MAT_MultHermitianTransposeAdd;
34: PetscLogEvent MAT_Getsymtransreduced, MAT_GetBrowsOfAcols;
35: PetscLogEvent MAT_GetBrowsOfAocols, MAT_Getlocalmat, MAT_Getlocalmatcondensed, MAT_Seqstompi, MAT_Seqstompinum, MAT_Seqstompisym;
36: PetscLogEvent MAT_GetMultiProcBlock;
37: PetscLogEvent MAT_CUSPARSECopyToGPU, MAT_CUSPARSECopyFromGPU, MAT_CUSPARSEGenerateTranspose, MAT_CUSPARSESolveAnalysis;
38: PetscLogEvent MAT_HIPSPARSECopyToGPU, MAT_HIPSPARSECopyFromGPU, MAT_HIPSPARSEGenerateTranspose, MAT_HIPSPARSESolveAnalysis;
39: PetscLogEvent MAT_PreallCOO, MAT_SetVCOO;
40: PetscLogEvent MAT_CreateGraph;
41: PetscLogEvent MAT_SetValuesBatch;
42: PetscLogEvent MAT_ViennaCLCopyToGPU;
43: PetscLogEvent MAT_CUDACopyToGPU, MAT_HIPCopyToGPU;
44: PetscLogEvent MAT_DenseCopyToGPU, MAT_DenseCopyFromGPU;
45: PetscLogEvent MAT_Merge, MAT_Residual, MAT_SetRandom;
46: PetscLogEvent MAT_FactorFactS, MAT_FactorInvS;
47: PetscLogEvent MATCOLORING_Apply, MATCOLORING_Comm, MATCOLORING_Local, MATCOLORING_ISCreate, MATCOLORING_SetUp, MATCOLORING_Weights;
48: PetscLogEvent MAT_H2Opus_Build, MAT_H2Opus_Compress, MAT_H2Opus_Orthog, MAT_H2Opus_LR;
50: const char *const MatFactorTypes[] = {"NONE", "LU", "CHOLESKY", "ILU", "ICC", "ILUDT", "QR", "MatFactorType", "MAT_FACTOR_", NULL};
52: /*@
53: MatSetRandom - Sets all components of a matrix to random numbers.
55: Logically Collective
57: Input Parameters:
58: + x - the matrix
59: - rctx - the `PetscRandom` object, formed by `PetscRandomCreate()`, or `NULL` and
60: it will create one internally.
62: Level: intermediate
64: Notes:
65: For sparse matrices that have been preallocated but not been assembled, it randomly selects appropriate locations,
67: for sparse matrices that already have nonzero locations, it fills the locations with random numbers.
69: It generates an error if used on unassembled sparse matrices that have not been preallocated.
71: If values have been inserted with `MatSetValues()`, call `MatAssemblyBegin()` and `MatAssemblyEnd()` before calling this routine.
72: The matrix is assembled on return.
74: Example Usage:
75: .vb
76: PetscRandomCreate(PETSC_COMM_WORLD, &rctx);
77: MatSetRandom(x, rctx);
78: PetscRandomDestroy(&rctx);
79: .ve
81: Developer Notes:
82: Implementations must leave the matrix assembled with increased object state, assembling it themselves when it is not already the case.
83: Implementations that insert only local entries can temporarily set `MAT_NO_OFF_PROC_ENTRIES` to `PETSC_TRUE` during assembly, restoring its previous value afterwards.
85: .seealso: [](ch_matrices), `Mat`, `PetscRandom`, `PetscRandomCreate()`, `MatZeroEntries()`, `MatSetValues()`, `PetscRandomDestroy()`
86: @*/
87: PetscErrorCode MatSetRandom(Mat x, PetscRandom rctx)
88: {
89: PetscRandom randObj = NULL;
91: PetscFunctionBegin;
95: PetscCheck(x->insertmode == NOT_SET_VALUES, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for matrices where you have set values but not yet assembled");
96: MatCheckPreallocated(x, 1);
98: if (!rctx) {
99: MPI_Comm comm;
100: PetscCall(PetscObjectGetComm((PetscObject)x, &comm));
101: PetscCall(PetscRandomCreate(comm, &randObj));
102: PetscCall(PetscRandomSetType(randObj, x->defaultrandtype));
103: PetscCall(PetscRandomSetFromOptions(randObj));
104: rctx = randObj;
105: }
106: PetscCall(PetscLogEventBegin(MAT_SetRandom, x, rctx, 0, 0));
107: PetscUseTypeMethod(x, setrandom, rctx);
108: PetscCall(PetscLogEventEnd(MAT_SetRandom, x, rctx, 0, 0));
110: PetscCall(PetscRandomDestroy(&randObj));
111: PetscFunctionReturn(PETSC_SUCCESS);
112: }
114: /*@
115: MatCopyHashToXAIJ - copy hash table entries into an XAIJ matrix type
117: Logically Collective
119: Input Parameter:
120: . A - A matrix in unassembled, hash table form
122: Output Parameter:
123: . B - The XAIJ matrix. This can either be `A` or some matrix of equivalent size, e.g. obtained from `A` via `MatDuplicate()`
125: Example:
126: .vb
127: PetscCall(MatDuplicate(A, MAT_DO_NOT_COPY_VALUES, &B));
128: PetscCall(MatCopyHashToXAIJ(A, B));
129: .ve
131: Level: advanced
133: Notes:
134: If `B` is `A`, then the hash table data structure will be destroyed. `B` is assembled
136: .seealso: [](ch_matrices), `Mat`, `MAT_USE_HASH_TABLE`
137: @*/
138: PetscErrorCode MatCopyHashToXAIJ(Mat A, Mat B)
139: {
140: PetscFunctionBegin;
142: PetscUseTypeMethod(A, copyhashtoxaij, B);
143: PetscFunctionReturn(PETSC_SUCCESS);
144: }
146: /*@
147: MatFactorGetErrorZeroPivot - returns the pivot value that was determined to be zero and the row it occurred in
149: Logically Collective
151: Input Parameter:
152: . mat - the factored matrix
154: Output Parameters:
155: + pivot - the pivot value computed
156: - row - the row that the zero pivot occurred. This row value must be interpreted carefully due to row reorderings and which processes
157: the share the matrix
159: Level: advanced
161: Notes:
162: This routine does not work for factorizations done with external packages.
164: This routine should only be called if `MatGetFactorError()` returns a value of `MAT_FACTOR_NUMERIC_ZEROPIVOT`
166: This can also be called on non-factored matrices that come from, for example, matrices used in SOR.
168: .seealso: [](ch_matrices), `Mat`, `MatZeroEntries()`, `MatFactor()`, `MatGetFactor()`,
169: `MatLUFactorSymbolic()`, `MatCholeskyFactorSymbolic()`, `MatFactorClearError()`,
170: `MAT_FACTOR_NUMERIC_ZEROPIVOT`
171: @*/
172: PetscErrorCode MatFactorGetErrorZeroPivot(Mat mat, PetscReal *pivot, PetscInt *row)
173: {
174: PetscFunctionBegin;
176: PetscAssertPointer(pivot, 2);
177: PetscAssertPointer(row, 3);
178: *pivot = mat->factorerror_zeropivot_value;
179: *row = mat->factorerror_zeropivot_row;
180: PetscFunctionReturn(PETSC_SUCCESS);
181: }
183: /*@
184: MatFactorGetError - gets the error code from a factorization
186: Logically Collective
188: Input Parameter:
189: . mat - the factored matrix
191: Output Parameter:
192: . err - the error code
194: Level: advanced
196: Note:
197: This can also be called on non-factored matrices that come from, for example, matrices used in SOR.
199: .seealso: [](ch_matrices), `Mat`, `MatZeroEntries()`, `MatFactor()`, `MatGetFactor()`, `MatLUFactorSymbolic()`, `MatCholeskyFactorSymbolic()`,
200: `MatFactorClearError()`, `MatFactorGetErrorZeroPivot()`, `MatFactorError`
201: @*/
202: PetscErrorCode MatFactorGetError(Mat mat, MatFactorError *err)
203: {
204: PetscFunctionBegin;
206: PetscAssertPointer(err, 2);
207: *err = mat->factorerrortype;
208: PetscFunctionReturn(PETSC_SUCCESS);
209: }
211: /*@
212: MatFactorClearError - clears the error code in a factorization
214: Logically Collective
216: Input Parameter:
217: . mat - the factored matrix
219: Level: developer
221: Note:
222: This can also be called on non-factored matrices that come from, for example, matrices used in SOR.
224: .seealso: [](ch_matrices), `Mat`, `MatZeroEntries()`, `MatFactor()`, `MatGetFactor()`, `MatLUFactorSymbolic()`, `MatCholeskyFactorSymbolic()`, `MatFactorGetError()`, `MatFactorGetErrorZeroPivot()`,
225: `MatGetErrorCode()`, `MatFactorError`
226: @*/
227: PetscErrorCode MatFactorClearError(Mat mat)
228: {
229: PetscFunctionBegin;
231: mat->factorerrortype = MAT_FACTOR_NOERROR;
232: mat->factorerror_zeropivot_value = 0.0;
233: mat->factorerror_zeropivot_row = 0;
234: PetscFunctionReturn(PETSC_SUCCESS);
235: }
237: PetscErrorCode MatFindNonzeroRowsOrCols_Basic(Mat mat, PetscBool cols, PetscReal tol, IS *nonzero)
238: {
239: Vec r, l;
240: const PetscScalar *al;
241: PetscInt i, nz, gnz, N, n, st;
243: PetscFunctionBegin;
244: PetscCall(MatCreateVecs(mat, &r, &l));
245: if (!cols) { /* nonzero rows */
246: PetscCall(MatGetOwnershipRange(mat, &st, NULL));
247: PetscCall(MatGetSize(mat, &N, NULL));
248: PetscCall(MatGetLocalSize(mat, &n, NULL));
249: PetscCall(VecSetRandom(r, NULL));
250: PetscCall(MatMult(mat, r, l));
251: PetscCall(VecGetArrayRead(l, &al));
252: } else { /* nonzero columns */
253: PetscCall(MatGetOwnershipRangeColumn(mat, &st, NULL));
254: PetscCall(MatGetSize(mat, NULL, &N));
255: PetscCall(MatGetLocalSize(mat, NULL, &n));
256: PetscCall(VecSet(r, 0.0));
257: PetscCall(VecSetRandom(l, NULL));
258: PetscCall(MatMultTranspose(mat, l, r));
259: PetscCall(VecGetArrayRead(r, &al));
260: }
261: if (tol <= 0.0) {
262: for (i = 0, nz = 0; i < n; i++)
263: if (al[i] != 0.0) nz++;
264: } else {
265: for (i = 0, nz = 0; i < n; i++)
266: if (PetscAbsScalar(al[i]) > tol) nz++;
267: }
268: PetscCallMPI(MPIU_Allreduce(&nz, &gnz, 1, MPIU_INT, MPI_SUM, PetscObjectComm((PetscObject)mat)));
269: if (gnz != N) {
270: PetscInt *nzr;
271: PetscCall(PetscMalloc1(nz, &nzr));
272: if (nz) {
273: if (tol < 0) {
274: for (i = 0, nz = 0; i < n; i++)
275: if (al[i] != 0.0) nzr[nz++] = i + st;
276: } else {
277: for (i = 0, nz = 0; i < n; i++)
278: if (PetscAbsScalar(al[i]) > tol) nzr[nz++] = i + st;
279: }
280: }
281: PetscCall(ISCreateGeneral(PetscObjectComm((PetscObject)mat), nz, nzr, PETSC_OWN_POINTER, nonzero));
282: } else *nonzero = NULL;
283: if (!cols) { /* nonzero rows */
284: PetscCall(VecRestoreArrayRead(l, &al));
285: } else {
286: PetscCall(VecRestoreArrayRead(r, &al));
287: }
288: PetscCall(VecDestroy(&l));
289: PetscCall(VecDestroy(&r));
290: PetscFunctionReturn(PETSC_SUCCESS);
291: }
293: /*@
294: MatFindNonzeroRows - Locate all rows that are not completely zero in the matrix
296: Input Parameter:
297: . mat - the matrix
299: Output Parameter:
300: . keptrows - the rows that are not completely zero
302: Level: intermediate
304: Note:
305: `keptrows` is set to `NULL` if all rows are nonzero.
307: Developer Note:
308: If `keptrows` is not `NULL`, it must be sorted.
310: .seealso: [](ch_matrices), `Mat`, `MatFindZeroRows()`
311: @*/
312: PetscErrorCode MatFindNonzeroRows(Mat mat, IS *keptrows)
313: {
314: PetscFunctionBegin;
317: PetscAssertPointer(keptrows, 2);
318: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
319: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
320: if (mat->ops->findnonzerorows) PetscUseTypeMethod(mat, findnonzerorows, keptrows);
321: else PetscCall(MatFindNonzeroRowsOrCols_Basic(mat, PETSC_FALSE, 0.0, keptrows));
322: if (keptrows && *keptrows) PetscCall(ISSetInfo(*keptrows, IS_SORTED, IS_GLOBAL, PETSC_FALSE, PETSC_TRUE));
323: PetscFunctionReturn(PETSC_SUCCESS);
324: }
326: /*@
327: MatFindZeroRows - Locate all rows that are completely zero in the matrix
329: Input Parameter:
330: . mat - the matrix
332: Output Parameter:
333: . zerorows - the rows that are completely zero
335: Level: intermediate
337: Note:
338: `zerorows` is set to `NULL` if no rows are zero.
340: .seealso: [](ch_matrices), `Mat`, `MatFindNonzeroRows()`
341: @*/
342: PetscErrorCode MatFindZeroRows(Mat mat, IS *zerorows)
343: {
344: IS keptrows;
345: PetscInt m, n;
347: PetscFunctionBegin;
350: PetscAssertPointer(zerorows, 2);
351: PetscCall(MatFindNonzeroRows(mat, &keptrows));
352: /* MatFindNonzeroRows sets keptrows to NULL if there are no zero rows.
353: In keeping with this convention, we set zerorows to NULL if there are no zero
354: rows. */
355: if (keptrows == NULL) {
356: *zerorows = NULL;
357: } else {
358: PetscCall(MatGetOwnershipRange(mat, &m, &n));
359: PetscCall(ISComplement(keptrows, m, n, zerorows));
360: PetscCall(ISDestroy(&keptrows));
361: }
362: PetscFunctionReturn(PETSC_SUCCESS);
363: }
365: /*@
366: MatGetDiagonalBlock - Returns the part of the matrix associated with the on-process coupling
368: Not Collective
370: Input Parameter:
371: . A - the matrix
373: Output Parameter:
374: . a - the diagonal part (which is a SEQUENTIAL matrix)
376: Level: advanced
378: Notes:
379: See `MatCreateAIJ()` for more information on the "diagonal part" of the matrix.
381: Use caution, as the reference count on the returned matrix is not incremented and it is used as part of `A`'s normal operation.
383: .seealso: [](ch_matrices), `Mat`, `MatCreateAIJ()`, `MATAIJ`, `MATBAIJ`, `MATSBAIJ`
384: @*/
385: PetscErrorCode MatGetDiagonalBlock(Mat A, Mat *a)
386: {
387: PetscFunctionBegin;
390: PetscAssertPointer(a, 2);
391: PetscCheck(!A->factortype, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
392: if (A->ops->getdiagonalblock) PetscUseTypeMethod(A, getdiagonalblock, a);
393: else {
394: PetscMPIInt size;
396: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)A), &size));
397: PetscCheck(size == 1, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Not for parallel matrix type %s", ((PetscObject)A)->type_name);
398: *a = A;
399: }
400: PetscFunctionReturn(PETSC_SUCCESS);
401: }
403: /*@
404: MatGetMultPetscSF - Returns the `PetscSF` that communicates to each MPI process the values held on other MPI processes that are coupled to it by the `Mat`
406: Not Collective
408: Input Parameter:
409: . A - the matrix
411: Output Parameter:
412: . sf - the `PetscSF`.
414: Level: advanced
416: Notes:
417: The returned `PetscSF` is owned by the matrix; do not destroy it.
419: It is only valid if this function is called after the matrix has been assembled
420: (and for `MATMPIDENSE` after a `MatMult()` has been additionally called).
422: This is only implemented for the matrix types listed below; calling it on a sequential matrix or a type that does not
423: build such a `PetscSF` raises an error.
425: For `MATMPIAIJ`, `MATMPIBAIJ`, `MATMPIDENSE`, and `MATMPISELL`, this `PetscSF` is used within
426: `MatMult()` to provide the contribution of vector entries that are not local to each MPI process to the matrix-vector product.
427: For `MATMPISBAIJ` the returned `PetscSF` is instead the off-process column gather used by operations such as
428: `MatDiagonalScale()`; `MatMult_MPISBAIJ()` uses a separate, augmented scatter context.
429: In all cases the `PetscSF` maps the global vector layout (the matrix column layout) onto the off-process columns that the local rows couple to,
430: so it can be reused to communicate any per-column data, for example with `PetscSFBcastBegin()`.
432: For `MATMPIDENSE` this `PetscSF` is an allgather: every process gathers all columns, including its own, in the natural global
433: order rather than a sparse `garray` order. The leaf set and ordering therefore differ across matrix types, so callers should use
434: `PetscSFGetGraph()` rather than assume a particular leaf layout.
436: .seealso: [](ch_matrices), `Mat`, `PetscSF`, `MatGetDiagonalBlock()`, `MatMPIAIJGetSeqAIJ()`, `PetscSFBcastBegin()`, `MatMult()`
437: @*/
438: PetscErrorCode MatGetMultPetscSF(Mat A, PetscSF *sf)
439: {
440: PetscFunctionBegin;
443: PetscAssertPointer(sf, 2);
444: PetscCheck(A->assembled, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
445: *sf = NULL;
446: PetscUseMethod(A, "MatGetMultPetscSF_C", (Mat, PetscSF *), (A, sf));
447: PetscFunctionReturn(PETSC_SUCCESS);
448: }
450: /*@
451: MatGetTrace - Gets the trace of a matrix. The sum of the diagonal entries.
453: Collective
455: Input Parameter:
456: . mat - the matrix
458: Output Parameter:
459: . trace - the sum of the diagonal entries
461: Level: advanced
463: .seealso: [](ch_matrices), `Mat`
464: @*/
465: PetscErrorCode MatGetTrace(Mat mat, PetscScalar *trace)
466: {
467: Vec diag;
469: PetscFunctionBegin;
471: PetscAssertPointer(trace, 2);
472: PetscCall(MatCreateVecs(mat, &diag, NULL));
473: PetscCall(MatGetDiagonal(mat, diag));
474: PetscCall(VecSum(diag, trace));
475: PetscCall(VecDestroy(&diag));
476: PetscFunctionReturn(PETSC_SUCCESS);
477: }
479: /*@
480: MatRealPart - Zeros out the imaginary part of the matrix
482: Logically Collective
484: Input Parameter:
485: . mat - the matrix
487: Level: advanced
489: .seealso: [](ch_matrices), `Mat`, `MatImaginaryPart()`
490: @*/
491: PetscErrorCode MatRealPart(Mat mat)
492: {
493: PetscFunctionBegin;
496: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
497: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
498: MatCheckPreallocated(mat, 1);
499: PetscUseTypeMethod(mat, realpart);
500: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
501: PetscFunctionReturn(PETSC_SUCCESS);
502: }
504: /*@
505: MatGetGhosts - Get the global indices of all ghost nodes defined by the sparse matrix
507: Collective
509: Input Parameter:
510: . mat - the matrix
512: Output Parameters:
513: + nghosts - number of ghosts (for `MATBAIJ` and `MATSBAIJ` matrices there is one ghost for each matrix block)
514: - ghosts - the global indices of the ghost points
516: Level: advanced
518: Note:
519: `nghosts` and `ghosts` are suitable to pass into `VecCreateGhost()` or `VecCreateGhostBlock()`
521: .seealso: [](ch_matrices), `Mat`, `VecCreateGhost()`, `VecCreateGhostBlock()`
522: @*/
523: PetscErrorCode MatGetGhosts(Mat mat, PetscInt *nghosts, const PetscInt *ghosts[])
524: {
525: PetscFunctionBegin;
528: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
529: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
530: if (mat->ops->getghosts) PetscUseTypeMethod(mat, getghosts, nghosts, ghosts);
531: else {
532: if (nghosts) *nghosts = 0;
533: if (ghosts) *ghosts = NULL;
534: }
535: PetscFunctionReturn(PETSC_SUCCESS);
536: }
538: /*@
539: MatImaginaryPart - Moves the imaginary part of the matrix to the real part and zeros the imaginary part
541: Logically Collective
543: Input Parameter:
544: . mat - the matrix
546: Level: advanced
548: .seealso: [](ch_matrices), `Mat`, `MatRealPart()`
549: @*/
550: PetscErrorCode MatImaginaryPart(Mat mat)
551: {
552: PetscFunctionBegin;
555: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
556: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
557: MatCheckPreallocated(mat, 1);
558: PetscUseTypeMethod(mat, imaginarypart);
559: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
560: PetscFunctionReturn(PETSC_SUCCESS);
561: }
563: // PetscClangLinter pragma disable: -fdoc-section-header-unknown
564: /*@
565: MatGetRow - Gets a row of a matrix. You MUST call `MatRestoreRow()`
566: for each row that you get to ensure that your application does
567: not bleed memory.
569: Not Collective
571: Input Parameters:
572: + mat - the matrix
573: - row - the row to get
575: Output Parameters:
576: + ncols - if not `NULL`, the number of nonzeros in `row`
577: . cols - if not `NULL`, the column numbers
578: - vals - if not `NULL`, the numerical values
580: Level: advanced
582: Notes:
583: This routine is provided for people who need to have direct access
584: to the structure of a matrix. We hope that we provide enough
585: high-level matrix routines that few users will need it.
587: `MatGetRow()` always returns 0-based column indices, regardless of
588: whether the internal representation is 0-based (default) or 1-based.
590: For better efficiency, set `cols` and/or `vals` to `NULL` if you do
591: not wish to extract these quantities. `vals` must be `NULL` for a matrix with
592: the `MAT_STRUCTURE_ONLY` option set to true, since no numerical values are stored.
594: The user can only examine the values extracted with `MatGetRow()`;
595: the values CANNOT be altered. To change the matrix entries, one
596: must use `MatSetValues()`.
598: You can only have one call to `MatGetRow()` outstanding for a particular
599: matrix at a time, per process. `MatGetRow()` can only obtain rows
600: associated with the given process, it cannot get rows from the
601: other processes; for that we suggest using `MatCreateSubMatrices()`, then
602: `MatGetRow()` on the submatrix. The row index passed to `MatGetRow()`
603: is in the global number of rows.
605: Use `MatGetRowIJ()` and `MatRestoreRowIJ()` to access all the local indices of the sparse matrix.
607: Use `MatSeqAIJGetArray()` and similar functions to access the numerical values for certain matrix types directly.
609: Fortran Note:
610: .vb
611: PetscInt, pointer :: cols(:)
612: PetscScalar, pointer :: vals(:)
613: .ve
615: .seealso: [](ch_matrices), `Mat`, `MatRestoreRow()`, `MatSetValues()`, `MatGetValues()`, `MatCreateSubMatrices()`, `MatGetDiagonal()`, `MatGetRowIJ()`, `MatRestoreRowIJ()`
616: @*/
617: PetscErrorCode MatGetRow(Mat mat, PetscInt row, PetscInt *ncols, const PetscInt *cols[], const PetscScalar *vals[])
618: {
619: PetscInt incols;
621: PetscFunctionBegin;
624: PetscCheck(mat->assembled, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
625: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
626: PetscCheck(!mat->structure_only || vals == NULL, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for matrix with MAT_STRUCTURE_ONLY");
627: MatCheckPreallocated(mat, 1);
628: PetscCheck(row >= mat->rmap->rstart && row < mat->rmap->rend, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Only for local rows, %" PetscInt_FMT " not in [%" PetscInt_FMT ",%" PetscInt_FMT ")", row, mat->rmap->rstart, mat->rmap->rend);
629: PetscCall(PetscLogEventBegin(MAT_GetRow, mat, 0, 0, 0));
630: PetscUseTypeMethod(mat, getrow, row, &incols, (PetscInt **)cols, (PetscScalar **)vals);
631: if (ncols) *ncols = incols;
632: PetscCall(PetscLogEventEnd(MAT_GetRow, mat, 0, 0, 0));
633: PetscFunctionReturn(PETSC_SUCCESS);
634: }
636: /*@
637: MatConjugate - replaces the matrix values with their complex conjugates
639: Logically Collective
641: Input Parameter:
642: . mat - the matrix
644: Level: advanced
646: .seealso: [](ch_matrices), `Mat`, `MatRealPart()`, `MatImaginaryPart()`, `VecConjugate()`, `MatTranspose()`
647: @*/
648: PetscErrorCode MatConjugate(Mat mat)
649: {
650: PetscFunctionBegin;
652: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
653: if (PetscDefined(USE_COMPLEX) && !(mat->symmetric == PETSC_BOOL3_TRUE && mat->hermitian == PETSC_BOOL3_TRUE)) {
654: PetscUseTypeMethod(mat, conjugate);
655: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
656: }
657: PetscFunctionReturn(PETSC_SUCCESS);
658: }
660: /*@
661: MatRestoreRow - Frees any temporary space allocated by `MatGetRow()`.
663: Not Collective
665: Input Parameters:
666: + mat - the matrix
667: . row - the row to get
668: . ncols - the number of nonzeros
669: . cols - the columns of the nonzeros
670: - vals - if nonzero the column values
672: Level: advanced
674: Notes:
675: This routine should be called after you have finished examining the entries.
677: This routine zeros out `ncols`, `cols`, and `vals`. This is to prevent accidental
678: us of the array after it has been restored. If you pass `NULL`, it will
679: not zero the pointers. Use of `cols` or `vals` after `MatRestoreRow()` is invalid.
681: Fortran Note:
682: .vb
683: PetscInt, pointer :: cols(:)
684: PetscScalar, pointer :: vals(:)
685: .ve
687: .seealso: [](ch_matrices), `Mat`, `MatGetRow()`
688: @*/
689: PetscErrorCode MatRestoreRow(Mat mat, PetscInt row, PetscInt *ncols, const PetscInt *cols[], const PetscScalar *vals[])
690: {
691: PetscFunctionBegin;
693: if (ncols) PetscAssertPointer(ncols, 3);
694: PetscCheck(mat->assembled, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
695: PetscTryTypeMethod(mat, restorerow, row, ncols, (PetscInt **)cols, (PetscScalar **)vals);
696: if (ncols) *ncols = 0;
697: if (cols) *cols = NULL;
698: if (vals) *vals = NULL;
699: PetscFunctionReturn(PETSC_SUCCESS);
700: }
702: /*@
703: MatGetRowUpperTriangular - Sets a flag to enable calls to `MatGetRow()` for matrix in `MATSBAIJ` format.
704: You should call `MatRestoreRowUpperTriangular()` after calling` MatGetRow()` and `MatRestoreRow()` to disable the flag.
706: Not Collective
708: Input Parameter:
709: . mat - the matrix
711: Level: advanced
713: Note:
714: The flag is to ensure that users are aware that `MatGetRow()` only provides the upper triangular part of the row for the matrices in `MATSBAIJ` format.
716: .seealso: [](ch_matrices), `Mat`, `MATSBAIJ`, `MatRestoreRowUpperTriangular()`
717: @*/
718: PetscErrorCode MatGetRowUpperTriangular(Mat mat)
719: {
720: PetscFunctionBegin;
723: PetscCheck(mat->assembled, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
724: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
725: MatCheckPreallocated(mat, 1);
726: PetscTryTypeMethod(mat, getrowuppertriangular);
727: PetscFunctionReturn(PETSC_SUCCESS);
728: }
730: /*@
731: MatRestoreRowUpperTriangular - Disable calls to `MatGetRow()` for matrix in `MATSBAIJ` format.
733: Not Collective
735: Input Parameter:
736: . mat - the matrix
738: Level: advanced
740: Note:
741: This routine should be called after you have finished calls to `MatGetRow()` and `MatRestoreRow()`.
743: .seealso: [](ch_matrices), `Mat`, `MATSBAIJ`, `MatGetRowUpperTriangular()`
744: @*/
745: PetscErrorCode MatRestoreRowUpperTriangular(Mat mat)
746: {
747: PetscFunctionBegin;
750: PetscCheck(mat->assembled, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
751: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
752: MatCheckPreallocated(mat, 1);
753: PetscTryTypeMethod(mat, restorerowuppertriangular);
754: PetscFunctionReturn(PETSC_SUCCESS);
755: }
757: /*@
758: MatSetOptionsPrefix - Sets the prefix used for searching for all
759: `Mat` options in the database.
761: Logically Collective
763: Input Parameters:
764: + A - the matrix
765: - prefix - the prefix to prepend to all option names
767: Level: advanced
769: Notes:
770: A hyphen (-) must NOT be given at the beginning of the prefix name.
771: The first character of all runtime options is AUTOMATICALLY the hyphen.
773: This is NOT used for options for the factorization of the matrix. Normally the
774: prefix is automatically passed in from the PC calling the factorization. To set
775: it directly use `MatSetOptionsPrefixFactor()`
777: .seealso: [](ch_matrices), `Mat`, `MatSetFromOptions()`, `MatSetOptionsPrefixFactor()`
778: @*/
779: PetscErrorCode MatSetOptionsPrefix(Mat A, const char prefix[])
780: {
781: PetscFunctionBegin;
783: PetscCall(PetscObjectSetOptionsPrefix((PetscObject)A, prefix));
784: PetscTryMethod(A, "MatSetOptionsPrefix_C", (Mat, const char[]), (A, prefix));
785: PetscFunctionReturn(PETSC_SUCCESS);
786: }
788: /*@
789: MatSetOptionsPrefixFactor - Sets the prefix used for searching for all matrix factor options in the database for
790: for matrices created with `MatGetFactor()`
792: Logically Collective
794: Input Parameters:
795: + A - the matrix
796: - prefix - the prefix to prepend to all option names for the factored matrix
798: Level: developer
800: Notes:
801: A hyphen (-) must NOT be given at the beginning of the prefix name.
802: The first character of all runtime options is AUTOMATICALLY the hyphen.
804: Normally the prefix is automatically passed in from the `PC` calling the factorization. To set
805: it directly when not using `KSP`/`PC` use `MatSetOptionsPrefixFactor()`
807: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatSetFromOptions()`, `MatSetOptionsPrefix()`, `MatAppendOptionsPrefixFactor()`
808: @*/
809: PetscErrorCode MatSetOptionsPrefixFactor(Mat A, const char prefix[])
810: {
811: PetscFunctionBegin;
813: if (prefix) {
814: PetscAssertPointer(prefix, 2);
815: PetscCheck(prefix[0] != '-', PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONG, "Options prefix should not begin with a hyphen");
816: if (prefix != A->factorprefix) {
817: PetscCall(PetscFree(A->factorprefix));
818: PetscCall(PetscStrallocpy(prefix, &A->factorprefix));
819: }
820: } else PetscCall(PetscFree(A->factorprefix));
821: PetscFunctionReturn(PETSC_SUCCESS);
822: }
824: /*@
825: MatAppendOptionsPrefixFactor - Appends to the prefix used for searching for all matrix factor options in the database for
826: for matrices created with `MatGetFactor()`
828: Logically Collective
830: Input Parameters:
831: + A - the matrix
832: - prefix - the prefix to prepend to all option names for the factored matrix
834: Level: developer
836: Notes:
837: A hyphen (-) must NOT be given at the beginning of the prefix name.
838: The first character of all runtime options is AUTOMATICALLY the hyphen.
840: Normally the prefix is automatically passed in from the `PC` calling the factorization. To set
841: it directly when not using `KSP`/`PC` use `MatAppendOptionsPrefixFactor()`
843: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `PetscOptionsCreate()`, `PetscOptionsDestroy()`, `PetscObjectSetOptionsPrefix()`, `PetscObjectPrependOptionsPrefix()`,
844: `PetscObjectGetOptionsPrefix()`, `TSAppendOptionsPrefix()`, `SNESAppendOptionsPrefix()`, `KSPAppendOptionsPrefix()`, `MatSetOptionsPrefixFactor()`,
845: `MatSetOptionsPrefix()`
846: @*/
847: PetscErrorCode MatAppendOptionsPrefixFactor(Mat A, const char prefix[])
848: {
849: size_t len1, len2, new_len;
851: PetscFunctionBegin;
853: if (!prefix) PetscFunctionReturn(PETSC_SUCCESS);
854: if (!A->factorprefix) {
855: PetscCall(MatSetOptionsPrefixFactor(A, prefix));
856: PetscFunctionReturn(PETSC_SUCCESS);
857: }
858: PetscCheck(prefix[0] != '-', PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONG, "Options prefix should not begin with a hyphen");
860: PetscCall(PetscStrlen(A->factorprefix, &len1));
861: PetscCall(PetscStrlen(prefix, &len2));
862: new_len = len1 + len2 + 1;
863: PetscCall(PetscRealloc(new_len * sizeof(*A->factorprefix), &A->factorprefix));
864: PetscCall(PetscStrncpy(A->factorprefix + len1, prefix, len2 + 1));
865: PetscFunctionReturn(PETSC_SUCCESS);
866: }
868: /*@
869: MatAppendOptionsPrefix - Appends to the prefix used for searching for all
870: matrix options in the database.
872: Logically Collective
874: Input Parameters:
875: + A - the matrix
876: - prefix - the prefix to prepend to all option names
878: Level: advanced
880: Note:
881: A hyphen (-) must NOT be given at the beginning of the prefix name.
882: The first character of all runtime options is AUTOMATICALLY the hyphen.
884: .seealso: [](ch_matrices), `Mat`, `MatGetOptionsPrefix()`, `MatAppendOptionsPrefixFactor()`, `MatSetOptionsPrefix()`
885: @*/
886: PetscErrorCode MatAppendOptionsPrefix(Mat A, const char prefix[])
887: {
888: PetscFunctionBegin;
890: PetscCall(PetscObjectAppendOptionsPrefix((PetscObject)A, prefix));
891: PetscTryMethod(A, "MatAppendOptionsPrefix_C", (Mat, const char[]), (A, prefix));
892: PetscFunctionReturn(PETSC_SUCCESS);
893: }
895: /*@
896: MatGetOptionsPrefix - Gets the prefix used for searching for all
897: matrix options in the database.
899: Not Collective
901: Input Parameter:
902: . A - the matrix
904: Output Parameter:
905: . prefix - pointer to the prefix string used
907: Level: advanced
909: .seealso: [](ch_matrices), `Mat`, `MatAppendOptionsPrefix()`, `MatSetOptionsPrefix()`, `MatAppendOptionsPrefixFactor()`, `MatSetOptionsPrefixFactor()`
910: @*/
911: PetscErrorCode MatGetOptionsPrefix(Mat A, const char *prefix[])
912: {
913: PetscFunctionBegin;
915: PetscAssertPointer(prefix, 2);
916: PetscCall(PetscObjectGetOptionsPrefix((PetscObject)A, prefix));
917: PetscFunctionReturn(PETSC_SUCCESS);
918: }
920: /*@
921: MatGetState - Gets a snapshot of the state of a `Mat`
923: Not Collective, No Fortran Support
925: Input Parameter:
926: . A - the matrix
928: Output Parameter:
929: . state - the matrix state
931: Level: developer
933: Notes:
934: The snapshot includes the matrix identity, object state, and nonzero state. Use `MatStateCompare()` to determine whether two snapshots are the same, or `MatStateCompareUpdate()` to compare and update a saved snapshot.
936: .seealso: [](ch_matrices), `Mat`, `MatState`, `MatStateCompare()`, `MatStateCompareUpdate()`, `MatStateInvalidate()`, `PetscObjectStateGet()`, `MatGetNonzeroState()`
937: @*/
938: PetscErrorCode MatGetState(Mat A, MatState *state)
939: {
940: PetscFunctionBegin;
942: PetscAssertPointer(state, 2);
943: state->id = ((PetscObject)A)->id;
944: state->state = ((PetscObject)A)->state;
945: state->nonzerostate = A->nonzerostate;
946: PetscFunctionReturn(PETSC_SUCCESS);
947: }
949: /*@
950: MatStateCompare - Compares two matrix state snapshots
952: Not Collective, No Fortran Support
954: Input Parameters:
955: + state1 - the first matrix state
956: - state2 - the second matrix state
958: Output Parameter:
959: . same - `PETSC_TRUE` if the matrix identity, object state, and nonzero state are the same, `PETSC_FALSE` otherwise
961: Level: developer
963: .seealso: [](ch_matrices), `Mat`, `MatState`, `MatGetState()`, `MatStateCompareUpdate()`, `MatStateInvalidate()`
964: @*/
965: PetscErrorCode MatStateCompare(MatState state1, MatState state2, PetscBool *same)
966: {
967: PetscFunctionBegin;
968: PetscAssertPointer(same, 3);
969: *same = (PetscBool)(state1.id == state2.id && state1.state == state2.state && state1.nonzerostate == state2.nonzerostate);
970: PetscFunctionReturn(PETSC_SUCCESS);
971: }
973: /*@
974: MatStateCompareUpdate - Compares a matrix with a state snapshot, then updates the snapshot
976: Not Collective, No Fortran Support
978: Input Parameter:
979: . A - the matrix
981: Input/Output Parameter:
982: . state - the matrix state snapshot to compare with and update
984: Output Parameter:
985: . same - `PETSC_TRUE` if the matrix state matched the snapshot before it was updated, `PETSC_FALSE` otherwise
987: Level: developer
989: .seealso: [](ch_matrices), `Mat`, `MatState`, `MatGetState()`, `MatStateCompare()`, `MatStateInvalidate()`
990: @*/
991: PetscErrorCode MatStateCompareUpdate(Mat A, MatState *state, PetscBool *same)
992: {
993: MatState current;
995: PetscFunctionBegin;
997: PetscAssertPointer(state, 2);
998: PetscAssertPointer(same, 3);
999: PetscCall(MatGetState(A, ¤t));
1000: PetscCall(MatStateCompare(current, *state, same));
1001: *state = current;
1002: PetscFunctionReturn(PETSC_SUCCESS);
1003: }
1005: /*@
1006: MatResetPreallocation - Reset matrix to use the original preallocation values provided by the user, for example with `MatXAIJSetPreallocation()`
1008: Collective
1010: Input Parameter:
1011: . A - the matrix
1013: Level: beginner
1015: Notes:
1016: After calling `MatAssemblyBegin()` and `MatAssemblyEnd()` with `MAT_FINAL_ASSEMBLY` the matrix data structures represent the nonzeros assigned to the
1017: matrix. If that space is less than the preallocated space that extra preallocated space is no longer available to take on new values. `MatResetPreallocation()`
1018: makes all of the preallocation space available
1020: Current values in the matrix are lost in this call
1022: Currently only supported for `MATAIJ` matrices.
1024: .seealso: [](ch_matrices), `Mat`, `MatSeqAIJSetPreallocation()`, `MatMPIAIJSetPreallocation()`, `MatXAIJSetPreallocation()`
1025: @*/
1026: PetscErrorCode MatResetPreallocation(Mat A)
1027: {
1028: PetscFunctionBegin;
1031: PetscUseMethod(A, "MatResetPreallocation_C", (Mat), (A));
1032: PetscFunctionReturn(PETSC_SUCCESS);
1033: }
1035: /*@
1036: MatResetHash - Reset the matrix so that it will use a hash table for the next round of `MatSetValues()` and `MatAssemblyBegin()`/`MatAssemblyEnd()`.
1038: Collective
1040: Input Parameter:
1041: . A - the matrix
1043: Level: intermediate
1045: Notes:
1046: The matrix will again delete the hash table data structures after following calls to `MatAssemblyBegin()`/`MatAssemblyEnd()` with `MAT_FINAL_ASSEMBLY`.
1048: Currently only supported for `MATAIJ` matrices.
1050: .seealso: [](ch_matrices), `Mat`, `MatResetPreallocation()`
1051: @*/
1052: PetscErrorCode MatResetHash(Mat A)
1053: {
1054: PetscFunctionBegin;
1057: PetscCheck(A->insertmode == NOT_SET_VALUES, PETSC_COMM_SELF, PETSC_ERR_SUP, "Cannot reset to hash state after setting some values but not yet calling MatAssemblyBegin()/MatAssemblyEnd()");
1058: if (A->num_ass == 0) PetscFunctionReturn(PETSC_SUCCESS);
1059: PetscUseMethod(A, "MatResetHash_C", (Mat), (A));
1060: /* These flags are used to determine whether certain setups occur */
1061: A->was_assembled = PETSC_FALSE;
1062: A->assembled = PETSC_FALSE;
1063: /* Log that the state of this object has changed; this will help guarantee that preconditioners get re-setup */
1064: PetscCall(PetscObjectStateIncrease((PetscObject)A));
1065: PetscFunctionReturn(PETSC_SUCCESS);
1066: }
1068: /*@
1069: MatSetUp - Sets up the internal matrix data structures for later use by the matrix
1071: Collective
1073: Input Parameter:
1074: . A - the matrix
1076: Level: advanced
1078: Notes:
1079: If no preallocation has been provided for a `MATAIJ`, `MATBAIJ`, or `MATSBAIJ` matrix, `MatSetUp()` makes it record the
1080: values given to `MatSetValues()` in a hash table, and the first assembly determines the nonzero pattern from it. Providing
1081: preallocation instead, for example with `MatXAIJSetPreallocation()` or `MatSetPreallocationCOO()`, disables hash-table mode.
1083: This routine is called internally by other `Mat` functions when needed so rarely needs to be called by users
1085: .seealso: [](ch_matrices), `Mat`, `MatMult()`, `MatCreate()`, `MatDestroy()`, `MatXAIJSetPreallocation()`
1086: @*/
1087: PetscErrorCode MatSetUp(Mat A)
1088: {
1089: PetscFunctionBegin;
1091: if (!((PetscObject)A)->type_name) {
1092: PetscMPIInt size;
1094: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)A), &size));
1095: PetscCall(MatSetType(A, size == 1 ? MATSEQAIJ : MATMPIAIJ));
1096: }
1097: if (!A->preallocated) PetscTryTypeMethod(A, setup);
1098: PetscCall(PetscLayoutSetUp(A->rmap));
1099: PetscCall(PetscLayoutSetUp(A->cmap));
1100: A->preallocated = PETSC_TRUE;
1101: PetscFunctionReturn(PETSC_SUCCESS);
1102: }
1104: #if PetscDefined(HAVE_SAWS)
1105: #include <petscviewersaws.h>
1106: #endif
1108: /*
1109: If threadsafety is on extraneous matrices may be printed
1111: This flag cannot be stored in the matrix because the original matrix in MatView() may assemble a new matrix which is passed into MatViewFromOptions()
1112: */
1113: #if !PetscDefined(HAVE_THREADSAFETY)
1114: static PetscInt insidematview = 0;
1115: #endif
1117: /*@
1118: MatViewFromOptions - View properties of the matrix based on options set in the options database
1120: Collective
1122: Input Parameters:
1123: + A - the matrix
1124: . obj - optional additional object that provides the options prefix to use
1125: - name - command line option
1127: Options Database Key:
1128: . -name viewer_specification - See `PetscOptionsCreateViewer()` for the values of `viewer_specification`
1130: Level: intermediate
1132: Note:
1133: This checks the options database, creates the viewer on-the-fly, uses it and then destroys it. Hence it should not be called in heavily used routines,
1134: rather `PetscOptionsCreateViewer()` should be used to construct the viewer once which can then be utilized in the heavily used routine.
1136: .seealso: [](ch_matrices), `Mat`, `MatView()`, `PetscObjectViewFromOptions()`, `MatCreate()`, `PetscOptionsCreateViewer()`
1137: @*/
1138: PetscErrorCode MatViewFromOptions(Mat A, PetscObject obj, const char name[])
1139: {
1140: PetscFunctionBegin;
1142: #if !PetscDefined(HAVE_THREADSAFETY)
1143: if (insidematview) PetscFunctionReturn(PETSC_SUCCESS);
1144: #endif
1145: PetscCall(PetscObjectViewFromOptions((PetscObject)A, obj, name));
1146: PetscFunctionReturn(PETSC_SUCCESS);
1147: }
1149: /*@
1150: MatView - display information about a matrix in a variety ways
1152: Collective on viewer
1154: Input Parameters:
1155: + mat - the matrix
1156: - viewer - visualization context
1158: Options Database Key:
1159: . -mat_view viewer_specification - Call `MatView()` at the conclusion of `MatAssemblyEnd()` or other routines that have changed the matrix values.
1160: See `PetscOptionsCreateViewer()` for the values of `viewer_specification`.
1162: Level: beginner
1164: Notes:
1165: The available visualization contexts include
1166: + `PETSC_VIEWER_STDOUT_SELF` - for sequential matrices
1167: . `PETSC_VIEWER_STDOUT_WORLD` - for parallel matrices created on `PETSC_COMM_WORLD`
1168: . `PETSC_VIEWER_STDOUT_`(comm) - for matrices created on MPI communicator comm
1169: - `PETSC_VIEWER_DRAW_WORLD` - graphical display of nonzero structure
1171: The user can open alternative visualization contexts with
1172: + `PetscViewerASCIIOpen()` - Outputs matrix to a specified file
1173: . `PetscViewerBinaryOpen()` - Outputs matrix in binary to a specified file; corresponding input uses `MatLoad()`
1174: . `PetscViewerDrawOpen()` - Outputs nonzero matrix nonzero structure to an X window display
1175: - `PetscViewerSocketOpen()` - Outputs matrix to Socket viewer, `PETSCVIEWERSOCKET`. Only the `MATSEQDENSE` and `MATAIJ` types support this viewer.
1177: The user can call `PetscViewerPushFormat()` to specify the output
1178: format of ASCII printed objects (when using `PETSC_VIEWER_STDOUT_SELF`,
1179: `PETSC_VIEWER_STDOUT_WORLD` and `PetscViewerASCIIOpen()`). Available formats include
1180: + `PETSC_VIEWER_DEFAULT` - default, prints matrix contents
1181: . `PETSC_VIEWER_ASCII_MATLAB` - prints matrix contents in MATLAB format
1182: . `PETSC_VIEWER_ASCII_DENSE` - prints entire matrix including zeros
1183: . `PETSC_VIEWER_ASCII_COMMON` - prints matrix contents, using a sparse format common among all matrix types
1184: . `PETSC_VIEWER_ASCII_IMPL` - prints matrix contents, using an implementation-specific format (which is in many cases the same as the default)
1185: . `PETSC_VIEWER_ASCII_INFO` - prints basic information about the matrix size and structure (not the matrix entries)
1186: - `PETSC_VIEWER_ASCII_INFO_DETAIL` - prints more detailed information about the matrix nonzero structure (still not vector or matrix entries)
1188: The ASCII viewers are only recommended for small matrices on at most a moderate number of processes,
1189: the program will seemingly hang and take hours for larger matrices, for larger matrices one should use the binary format.
1191: In the debugger you can do "call MatView(mat,0)" to display the matrix. (The same holds for any PETSc object viewer).
1193: See the manual page for `MatLoad()` for the exact format of the binary file when the binary
1194: viewer is used.
1196: `MatViewFromOptions()` provides an alternative to this routine that only views the matrix if the requested value
1197: is provided in the options database.
1199: See `share/petsc/matlab/PetscBinaryRead.m` for a MATLAB code that can read in the binary file when the binary
1200: viewer is used and `lib/petsc/bin/PetscBinaryIO.py` for loading them into Python.
1202: One can use `-mat_view draw -draw_pause -1` to pause the graphical display of matrix nonzero structure,
1203: and then use the following mouse functions.
1204: .vb
1205: left mouse: zoom in
1206: middle mouse: zoom out
1207: right mouse: continue with the simulation
1208: .ve
1210: .seealso: [](ch_matrices), `Mat`, `PetscViewerPushFormat()`, `PetscViewerASCIIOpen()`, `PetscViewerDrawOpen()`, `PetscViewer`,
1211: `PetscViewerSocketOpen()`, `PetscViewerBinaryOpen()`, `MatLoad()`, `MatViewFromOptions()`, `PetscOptionsCreateViewer()`
1212: @*/
1213: PetscErrorCode MatView(Mat mat, PetscViewer viewer)
1214: {
1215: PetscInt rows, cols, rbs, cbs;
1216: PetscBool isascii, isstring, issaws;
1217: PetscViewerFormat format;
1218: PetscMPIInt size;
1220: PetscFunctionBegin;
1223: if (!viewer) PetscCall(PetscViewerASCIIGetStdout(PetscObjectComm((PetscObject)mat), &viewer));
1226: PetscCall(PetscViewerGetFormat(viewer, &format));
1227: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)viewer), &size));
1228: if (size == 1 && format == PETSC_VIEWER_LOAD_BALANCE) PetscFunctionReturn(PETSC_SUCCESS);
1230: #if !PetscDefined(HAVE_THREADSAFETY)
1231: insidematview++;
1232: #endif
1233: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERSTRING, &isstring));
1234: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isascii));
1235: PetscCall(PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERSAWS, &issaws));
1236: PetscCheck((isascii && (format == PETSC_VIEWER_ASCII_INFO || format == PETSC_VIEWER_ASCII_INFO_DETAIL)) || !mat->factortype, PetscObjectComm((PetscObject)viewer), PETSC_ERR_ARG_WRONGSTATE, "No viewers for factored matrix except ASCII, info, or info_detail");
1238: PetscCall(PetscLogEventBegin(MAT_View, mat, viewer, 0, 0));
1239: if (isascii) {
1240: if (!mat->preallocated) {
1241: PetscCall(PetscViewerASCIIPrintf(viewer, "Matrix has not been preallocated yet\n"));
1242: #if !PetscDefined(HAVE_THREADSAFETY)
1243: insidematview--;
1244: #endif
1245: PetscCall(PetscLogEventEnd(MAT_View, mat, viewer, 0, 0));
1246: PetscFunctionReturn(PETSC_SUCCESS);
1247: }
1248: if (!mat->assembled) {
1249: PetscCall(PetscViewerASCIIPrintf(viewer, "Matrix has not been assembled yet\n"));
1250: #if !PetscDefined(HAVE_THREADSAFETY)
1251: insidematview--;
1252: #endif
1253: PetscCall(PetscLogEventEnd(MAT_View, mat, viewer, 0, 0));
1254: PetscFunctionReturn(PETSC_SUCCESS);
1255: }
1256: PetscCall(PetscObjectPrintClassNamePrefixType((PetscObject)mat, viewer));
1257: if (format == PETSC_VIEWER_ASCII_INFO || format == PETSC_VIEWER_ASCII_INFO_DETAIL) {
1258: MatNullSpace nullsp, transnullsp;
1259: PetscBool nz_factor = PETSC_TRUE;
1261: PetscCall(PetscViewerASCIIPushTab(viewer));
1262: PetscCall(MatGetSize(mat, &rows, &cols));
1263: PetscCall(MatGetBlockSizes(mat, &rbs, &cbs));
1264: if (rbs != 1 || cbs != 1) {
1265: if (rbs != cbs) PetscCall(PetscViewerASCIIPrintf(viewer, "rows=%" PetscInt_FMT ", cols=%" PetscInt_FMT ", rbs=%" PetscInt_FMT ", cbs=%" PetscInt_FMT "%s\n", rows, cols, rbs, cbs, mat->bsizes ? " variable blocks set" : ""));
1266: else PetscCall(PetscViewerASCIIPrintf(viewer, "rows=%" PetscInt_FMT ", cols=%" PetscInt_FMT ", bs=%" PetscInt_FMT "%s\n", rows, cols, rbs, mat->bsizes ? " variable blocks set" : ""));
1267: } else PetscCall(PetscViewerASCIIPrintf(viewer, "rows=%" PetscInt_FMT ", cols=%" PetscInt_FMT "\n", rows, cols));
1268: if (mat->factortype) {
1269: MatSolverType solver;
1271: PetscCall(MatFactorGetSolverType(mat, &solver));
1272: PetscCall(PetscViewerASCIIPrintf(viewer, "package used to perform factorization: %s\n", solver));
1273: PetscCall(PetscStrcmpAny(solver, &nz_factor, MATSOLVERUMFPACK, MATSOLVERCHOLMOD, MATSOLVERSUPERLU, MATSOLVERSUPERLU_DIST, MATSOLVERSTRUMPACK, MATSOLVERHTOOL, ""));
1274: nz_factor = !nz_factor;
1275: }
1276: if (mat->ops->getinfo) {
1277: PetscBool is_constant_or_diagonal;
1279: // Don't print nonzero information for constant or diagonal matrices, it just adds noise to the output
1280: PetscCall(PetscObjectTypeCompareAny((PetscObject)mat, &is_constant_or_diagonal, MATCONSTANTDIAGONAL, MATDIAGONAL, ""));
1281: if (!is_constant_or_diagonal && nz_factor) {
1282: MatInfo info;
1284: PetscCall(MatGetInfo(mat, MAT_GLOBAL_SUM, &info));
1285: PetscCall(PetscViewerASCIIPrintf(viewer, "total: nonzeros=%.f, allocated nonzeros=%.f\n", info.nz_used, info.nz_allocated));
1286: if (!mat->factortype) PetscCall(PetscViewerASCIIPrintf(viewer, "total number of mallocs used during MatSetValues calls=%" PetscInt_FMT "\n", (PetscInt)info.mallocs));
1287: }
1288: }
1289: PetscCall(MatGetNullSpace(mat, &nullsp));
1290: PetscCall(MatGetTransposeNullSpace(mat, &transnullsp));
1291: if (nullsp) PetscCall(PetscViewerASCIIPrintf(viewer, " has attached null space\n"));
1292: if (transnullsp && transnullsp != nullsp) PetscCall(PetscViewerASCIIPrintf(viewer, " has attached transposed null space\n"));
1293: PetscCall(MatGetNearNullSpace(mat, &nullsp));
1294: if (nullsp) PetscCall(PetscViewerASCIIPrintf(viewer, " has attached near null space\n"));
1295: PetscCall(PetscViewerASCIIPushTab(viewer));
1296: PetscCall(MatProductView(mat, viewer));
1297: PetscCall(PetscViewerASCIIPopTab(viewer));
1298: if (mat->bsizes && format == PETSC_VIEWER_ASCII_INFO_DETAIL) {
1299: IS tmp;
1301: PetscCall(ISCreateGeneral(PetscObjectComm((PetscObject)viewer), mat->nblocks, mat->bsizes, PETSC_USE_POINTER, &tmp));
1302: PetscCall(PetscObjectSetName((PetscObject)tmp, "Block Sizes"));
1303: PetscCall(PetscViewerASCIIPushTab(viewer));
1304: PetscCall(ISView(tmp, viewer));
1305: PetscCall(PetscViewerASCIIPopTab(viewer));
1306: PetscCall(ISDestroy(&tmp));
1307: }
1308: }
1309: } else if (issaws) {
1310: #if PetscDefined(HAVE_SAWS)
1311: PetscMPIInt rank;
1313: PetscCall(PetscObjectName((PetscObject)mat));
1314: PetscCallMPI(MPI_Comm_rank(PETSC_COMM_WORLD, &rank));
1315: if (!((PetscObject)mat)->amsmem && rank == 0) PetscCall(PetscObjectViewSAWs((PetscObject)mat, viewer));
1316: #endif
1317: } else if (isstring) {
1318: const char *type;
1319: PetscCall(MatGetType(mat, &type));
1320: PetscCall(PetscViewerStringSPrintf(viewer, " MatType: %-7.7s", type));
1321: PetscTryTypeMethod(mat, view, viewer);
1322: }
1323: if ((format == PETSC_VIEWER_NATIVE || format == PETSC_VIEWER_LOAD_BALANCE) && mat->ops->viewnative) {
1324: PetscCall(PetscViewerASCIIPushTab(viewer));
1325: PetscUseTypeMethod(mat, viewnative, viewer);
1326: PetscCall(PetscViewerASCIIPopTab(viewer));
1327: } else if (mat->ops->view) {
1328: PetscCall(PetscViewerASCIIPushTab(viewer));
1329: PetscUseTypeMethod(mat, view, viewer);
1330: PetscCall(PetscViewerASCIIPopTab(viewer));
1331: }
1332: if (isascii) {
1333: PetscCall(PetscViewerGetFormat(viewer, &format));
1334: if (format == PETSC_VIEWER_ASCII_INFO || format == PETSC_VIEWER_ASCII_INFO_DETAIL) PetscCall(PetscViewerASCIIPopTab(viewer));
1335: }
1336: PetscCall(PetscLogEventEnd(MAT_View, mat, viewer, 0, 0));
1337: #if !PetscDefined(HAVE_THREADSAFETY)
1338: insidematview--;
1339: #endif
1340: PetscFunctionReturn(PETSC_SUCCESS);
1341: }
1343: #if PetscDefined(USE_DEBUG)
1344: #include <../src/sys/totalview/tv_data_display.h>
1345: PETSC_UNUSED static int TV_display_type(const struct _p_Mat *mat)
1346: {
1347: TV_add_row("Local rows", "int", &mat->rmap->n);
1348: TV_add_row("Local columns", "int", &mat->cmap->n);
1349: TV_add_row("Global rows", "int", &mat->rmap->N);
1350: TV_add_row("Global columns", "int", &mat->cmap->N);
1351: TV_add_row("Typename", TV_ascii_string_type, ((PetscObject)mat)->type_name);
1352: return TV_format_OK;
1353: }
1354: #endif
1356: /*@
1357: MatLoad - Loads a matrix that has been stored in binary/HDF5 format
1358: with `MatView()`. The matrix format is determined from the options database.
1359: Generates a parallel MPI matrix if the communicator has more than one
1360: process. The default matrix type is `MATAIJ`.
1362: Collective
1364: Input Parameters:
1365: + mat - the newly loaded matrix, this needs to have been created with `MatCreate()`
1366: or some related function before a call to `MatLoad()`
1367: - viewer - `PETSCVIEWERBINARY`/`PETSCVIEWERHDF5` file viewer
1369: Options Database Key:
1370: . -matload_block_size bs - set block size
1372: Level: beginner
1374: Notes:
1375: If the `Mat` type has not yet been given then `MATAIJ` is used, call `MatSetFromOptions()` on the
1376: `Mat` before calling this routine if you wish to set it from the options database.
1378: `MatLoad()` automatically loads into the options database any options
1379: given in the file filename.info where filename is the name of the file
1380: that was passed to the `PetscViewerBinaryOpen()`. The options in the info
1381: file will be ignored if you use the `-viewer_binary_skip_info` option.
1383: If the type or size of mat is not set before a call to `MatLoad()`, PETSc
1384: sets the default matrix type AIJ and sets the local and global sizes.
1385: If type and/or size is already set, then the same are used.
1387: In parallel, each process can load a subset of rows (or the
1388: entire matrix). This routine is especially useful when a large
1389: matrix is stored on disk and only part of it is desired on each
1390: process. For example, a parallel solver may access only some of
1391: the rows from each process. The algorithm used here reads
1392: relatively small blocks of data rather than reading the entire
1393: matrix and then subsetting it.
1395: Viewer's `PetscViewerType` must be either `PETSCVIEWERBINARY` or `PETSCVIEWERHDF5`.
1396: Such viewer can be created using `PetscViewerBinaryOpen()` or `PetscViewerHDF5Open()`,
1397: or the sequence like
1398: .vb
1399: PetscViewer v;
1400: PetscViewerCreate(PETSC_COMM_WORLD, &v);
1401: PetscViewerSetType(v, PETSCVIEWERBINARY);
1402: PetscViewerSetFromOptions(v);
1403: PetscViewerFileSetMode(v, FILE_MODE_READ);
1404: PetscViewerFileSetName(v, "datafile");
1405: .ve
1406: The optional `PetscViewerSetFromOptions()` call allows overriding `PetscViewerSetType()` using the option
1407: .vb
1408: -viewer_type (binary|hdf5)
1409: .ve
1411: See the example src/ksp/ksp/tutorials/ex27.c with the first approach,
1412: and src/mat/tutorials/ex10.c with the second approach.
1414: In case of `PETSCVIEWERBINARY`, a native PETSc binary format is used. Each of the blocks
1415: is read onto MPI rank 0 and then shipped to its destination MPI rank, one after another.
1416: Multiple objects, both matrices and vectors, can be stored within the same file.
1417: Their `PetscObject` name is ignored; they are loaded in the order of their storage.
1419: Most users should not need to know the details of the binary storage
1420: format, since `MatLoad()` and `MatView()` completely hide these details.
1421: But for anyone who is interested, the standard binary matrix storage
1422: format is
1424: .vb
1425: PetscInt MAT_FILE_CLASSID
1426: PetscInt number of rows
1427: PetscInt number of columns
1428: PetscInt total number of nonzeros
1429: PetscInt *number nonzeros in each row
1430: PetscInt *column indices of all nonzeros (starting index is zero)
1431: PetscScalar *values of all nonzeros
1432: .ve
1433: If PETSc was not configured with `--with-64-bit-indices` then only `MATMPIAIJ` matrices with more than `PETSC_INT_MAX` non-zeros can be
1434: stored or loaded (each MPI process part of the matrix must have less than `PETSC_INT_MAX` nonzeros). Since the total nonzero count in this
1435: case will not fit in a (32-bit) `PetscInt` the value `PETSC_INT_MAX` is used for the header entry `total number of nonzeros`.
1437: PETSc automatically does the byte swapping for
1438: machines that store the bytes reversed. Thus if you write your own binary
1439: read/write routines you have to swap the bytes; see `PetscBinaryRead()`
1440: and `PetscBinaryWrite()` to see how this may be done.
1442: In case of `PETSCVIEWERHDF5`, a parallel HDF5 reader is used.
1443: Each process's chunk is loaded independently by its owning MPI process.
1444: Multiple objects, both matrices and vectors, can be stored within the same file.
1445: They are looked up by their PetscObject name.
1447: As the MATLAB MAT-File Version 7.3 format is also a HDF5 flavor, we decided to use
1448: by default the same structure and naming of the AIJ arrays and column count
1449: within the HDF5 file. This means that a MAT file saved with -v7.3 flag, e.g.
1450: .vb
1451: save example.mat A b -v7.3
1452: .ve
1453: can be directly read by this routine (see Reference 1 for details).
1455: Depending on your MATLAB version, this format might be a default,
1456: otherwise you can set it as default in Preferences.
1458: Unless `-nocompression` flag is used to save the file in MATLAB,
1459: PETSc must be configured with ZLIB package.
1461: See also examples `src/mat/tutorials/ex10.c` and `src/ksp/ksp/tutorials/ex27.c`
1463: This reader currently supports only real `MATSEQAIJ`, `MATMPIAIJ`, `MATSEQDENSE`, and `MATMPIDENSE` matrices for `PETSCVIEWERHDF5`
1465: Corresponding `MatView()` is not yet implemented.
1467: The loaded matrix is actually a transpose of the original one in MATLAB,
1468: unless you push `PETSC_VIEWER_HDF5_MAT` format (see examples above).
1469: With this format, matrix is automatically transposed by PETSc,
1470: unless the matrix is marked as SPD or symmetric
1471: (see `MatSetOption()`, `MAT_SPD`, `MAT_SYMMETRIC`).
1473: See MATLAB Documentation on `save()`, <https://www.mathworks.com/help/matlab/ref/save.html#btox10b-1-version>
1475: .seealso: [](ch_matrices), `Mat`, `PetscViewerBinaryOpen()`, `PetscViewerSetType()`, `MatView()`, `VecLoad()`
1476: @*/
1477: PetscErrorCode MatLoad(Mat mat, PetscViewer viewer)
1478: {
1479: PetscBool flg;
1481: PetscFunctionBegin;
1485: if (!((PetscObject)mat)->type_name) PetscCall(MatSetType(mat, MATAIJ));
1487: flg = PETSC_FALSE;
1488: PetscCall(PetscOptionsGetBool(((PetscObject)mat)->options, ((PetscObject)mat)->prefix, "-matload_symmetric", &flg, NULL));
1489: if (flg) {
1490: PetscCall(MatSetOption(mat, MAT_SYMMETRIC, PETSC_TRUE));
1491: PetscCall(MatSetOption(mat, MAT_SYMMETRY_ETERNAL, PETSC_TRUE));
1492: }
1493: flg = PETSC_FALSE;
1494: PetscCall(PetscOptionsGetBool(((PetscObject)mat)->options, ((PetscObject)mat)->prefix, "-matload_spd", &flg, NULL));
1495: if (flg) PetscCall(MatSetOption(mat, MAT_SPD, PETSC_TRUE));
1497: PetscCall(PetscLogEventBegin(MAT_Load, mat, viewer, 0, 0));
1498: PetscUseTypeMethod(mat, load, viewer);
1499: PetscCall(PetscLogEventEnd(MAT_Load, mat, viewer, 0, 0));
1500: PetscFunctionReturn(PETSC_SUCCESS);
1501: }
1503: static PetscErrorCode MatDestroy_Redundant(Mat_Redundant **redundant)
1504: {
1505: Mat_Redundant *redund = *redundant;
1507: PetscFunctionBegin;
1508: if (redund) {
1509: if (redund->matseq) { /* via MatCreateSubMatrices() */
1510: PetscCall(ISDestroy(&redund->isrow));
1511: PetscCall(ISDestroy(&redund->iscol));
1512: PetscCall(MatDestroySubMatrices(1, &redund->matseq));
1513: } else {
1514: PetscCall(PetscFree2(redund->send_rank, redund->recv_rank));
1515: PetscCall(PetscFree(redund->sbuf_j));
1516: PetscCall(PetscFree(redund->sbuf_a));
1517: for (PetscInt i = 0; i < redund->nrecvs; i++) {
1518: PetscCall(PetscFree(redund->rbuf_j[i]));
1519: PetscCall(PetscFree(redund->rbuf_a[i]));
1520: }
1521: PetscCall(PetscFree4(redund->sbuf_nz, redund->rbuf_nz, redund->rbuf_j, redund->rbuf_a));
1522: }
1524: PetscCall(PetscCommDestroy(&redund->subcomm));
1525: PetscCall(PetscFree(redund));
1526: }
1527: PetscFunctionReturn(PETSC_SUCCESS);
1528: }
1530: /*@
1531: MatDestroy - Frees space taken by a matrix.
1533: Collective
1535: Input Parameter:
1536: . A - the matrix
1538: Level: beginner
1540: Developer Note:
1541: Some special arrays of matrices are not destroyed in this routine but instead by the routines called by
1542: `MatDestroySubMatrices()`. Thus one must be sure that any changes here must also be made in those routines.
1543: `MatHeaderMerge()` and `MatHeaderReplace()` also manipulate the data in the `Mat` object and likely need changes
1544: if changes are needed here.
1546: .seealso: [](ch_matrices), `Mat`, `MatCreate()`
1547: @*/
1548: PetscErrorCode MatDestroy(Mat *A)
1549: {
1550: PetscFunctionBegin;
1551: if (!*A) PetscFunctionReturn(PETSC_SUCCESS);
1553: if (--((PetscObject)*A)->refct > 0) {
1554: *A = NULL;
1555: PetscFunctionReturn(PETSC_SUCCESS);
1556: }
1558: /* if memory was published with SAWs then destroy it */
1559: PetscCall(PetscObjectSAWsViewOff((PetscObject)*A));
1560: PetscTryTypeMethod(*A, destroy);
1562: PetscCall(PetscFree((*A)->factorprefix));
1563: PetscCall(PetscFree((*A)->defaultvectype));
1564: PetscCall(PetscFree((*A)->defaultrandtype));
1565: PetscCall(PetscFree((*A)->bsizes));
1566: PetscCall(PetscFree((*A)->solvertype));
1567: for (PetscInt i = 0; i < MAT_FACTOR_NUM_TYPES; i++) PetscCall(PetscFree((*A)->preferredordering[i]));
1568: if ((*A)->redundant && (*A)->redundant->matseq[0] == *A) (*A)->redundant->matseq[0] = NULL;
1569: PetscCall(MatDestroy_Redundant(&(*A)->redundant));
1570: PetscCall(MatProductClear(*A));
1571: PetscCall(MatNullSpaceDestroy(&(*A)->nullsp));
1572: PetscCall(MatNullSpaceDestroy(&(*A)->transnullsp));
1573: PetscCall(MatNullSpaceDestroy(&(*A)->nearnullsp));
1574: PetscCall(MatDestroy(&(*A)->schur));
1575: PetscCall(VecDestroy(&(*A)->dot_vec));
1576: PetscCall(PetscLayoutDestroy(&(*A)->rmap));
1577: PetscCall(PetscLayoutDestroy(&(*A)->cmap));
1578: PetscCall(PetscHeaderDestroy(A));
1579: PetscFunctionReturn(PETSC_SUCCESS);
1580: }
1582: // PetscClangLinter pragma disable: -fdoc-section-header-unknown
1583: /*@
1584: MatSetValues - Inserts or adds a block of values into a matrix.
1585: These values may be cached, so `MatAssemblyBegin()` and `MatAssemblyEnd()`
1586: MUST be called after all calls to `MatSetValues()` have been completed.
1588: Not Collective
1590: Input Parameters:
1591: + mat - the matrix
1592: . m - the number of rows
1593: . idxm - the global indices of the rows
1594: . n - the number of columns
1595: . idxn - the global indices of the columns
1596: . v - a one-dimensional array that contains the values implicitly stored as a two-dimensional array, by default in row-major order.
1597: See `MAT_ROW_ORIENTED` in `MatSetOption()` for how to use column-major order.
1598: - addv - either `ADD_VALUES` to add values to any existing entries, or `INSERT_VALUES` to replace existing entries with new values
1600: Level: beginner
1602: Notes:
1603: Calls to `MatSetValues()` with the `INSERT_VALUES` and `ADD_VALUES`
1604: options cannot be mixed without intervening calls to the assembly
1605: routines.
1607: `MatSetValues()` uses 0-based row and column numbers in Fortran
1608: as well as in C.
1610: Negative indices may be passed in `idxm` and `idxn`, these rows and columns are simply ignored. This allows easily inserting element stiffness matrices
1611: with homogeneous Dirichlet boundary conditions that you don't want represented
1612: in the matrix.
1614: Efficiency Alert:
1615: The routine `MatSetValuesBlocked()` may offer much better efficiency
1616: for users of block sparse formats (`MATSEQBAIJ` and `MATMPIBAIJ`).
1618: Fortran Notes:
1619: If any of `idxm`, `idxn`, and `v` are scalars pass them using, for example,
1620: .vb
1621: call MatSetValues(mat, one, [idxm], one, [idxn], [v], INSERT_VALUES, ierr)
1622: .ve
1624: If `v` is a two-dimensional array make sure to first call `MatSetOption(mat, MAT_ROW_ORIENTED, PETSC_FALSE, ierr)` before using this function,
1625: otherwise the transpose of `v` will seemingly be inserted in the matrix, since Fortran passes two-dimensional arrays with column orientation.
1627: .seealso: [](ch_matrices), `Mat`, `MatSetOption()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValuesBlocked()`, `MatSetValuesLocal()`,
1628: `InsertMode`, `INSERT_VALUES`, `ADD_VALUES`
1629: @*/
1630: PetscErrorCode MatSetValues(Mat mat, PetscInt m, const PetscInt idxm[], PetscInt n, const PetscInt idxn[], const PetscScalar v[], InsertMode addv)
1631: {
1632: PetscFunctionBeginHot;
1635: if (!m || !n) PetscFunctionReturn(PETSC_SUCCESS); /* no values to insert */
1636: PetscAssertPointer(idxm, 3);
1637: PetscAssertPointer(idxn, 5);
1638: MatCheckPreallocated(mat, 1);
1640: if (mat->insertmode == NOT_SET_VALUES) mat->insertmode = addv;
1641: else PetscCheck(mat->insertmode == addv, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Cannot mix add values and insert values");
1643: if (PetscDefined(USE_DEBUG)) {
1644: PetscInt i, j;
1646: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
1647: if (v) {
1648: for (i = 0; i < m; i++) {
1649: for (j = 0; j < n; j++) {
1650: if (mat->erroriffailure && PetscIsInfOrNanScalar(v[i * n + j]))
1651: #if PetscDefined(USE_COMPLEX)
1652: SETERRQ(PETSC_COMM_SELF, PETSC_ERR_FP, "Inserting %g+i%g at matrix entry (%" PetscInt_FMT ",%" PetscInt_FMT ")", (double)PetscRealPart(v[i * n + j]), (double)PetscImaginaryPart(v[i * n + j]), idxm[i], idxn[j]);
1653: #else
1654: SETERRQ(PETSC_COMM_SELF, PETSC_ERR_FP, "Inserting %g at matrix entry (%" PetscInt_FMT ",%" PetscInt_FMT ")", (double)v[i * n + j], idxm[i], idxn[j]);
1655: #endif
1656: }
1657: }
1658: }
1659: for (i = 0; i < m; i++) PetscCheck(idxm[i] < mat->rmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Cannot insert in row %" PetscInt_FMT ", maximum is %" PetscInt_FMT, idxm[i], mat->rmap->N - 1);
1660: for (i = 0; i < n; i++) PetscCheck(idxn[i] < mat->cmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Cannot insert in column %" PetscInt_FMT ", maximum is %" PetscInt_FMT, idxn[i], mat->cmap->N - 1);
1661: }
1663: if (mat->assembled) {
1664: mat->was_assembled = PETSC_TRUE;
1665: mat->assembled = PETSC_FALSE;
1666: }
1667: PetscCall(PetscLogEventBegin(MAT_SetValues, mat, 0, 0, 0));
1668: PetscUseTypeMethod(mat, setvalues, m, idxm, n, idxn, v, addv);
1669: PetscCall(PetscLogEventEnd(MAT_SetValues, mat, 0, 0, 0));
1670: PetscFunctionReturn(PETSC_SUCCESS);
1671: }
1673: // PetscClangLinter pragma disable: -fdoc-section-header-unknown
1674: /*@
1675: MatSetValuesIS - Inserts or adds a block of values into a matrix using an `IS` to indicate the rows and columns
1676: These values may be cached, so `MatAssemblyBegin()` and `MatAssemblyEnd()`
1677: MUST be called after all calls to `MatSetValues()` have been completed.
1679: Not Collective
1681: Input Parameters:
1682: + mat - the matrix
1683: . ism - the rows to provide
1684: . isn - the columns to provide
1685: . v - a one-dimensional array that contains the values implicitly stored as a two-dimensional array, by default in row-major order.
1686: See `MAT_ROW_ORIENTED` in `MatSetOption()` for how to use column-major order.
1687: - addv - either `ADD_VALUES` to add values to any existing entries, or `INSERT_VALUES` to replace existing entries with new values
1689: Level: beginner
1691: Notes:
1692: By default, the values, `v`, are stored in row-major order. See `MAT_ROW_ORIENTED` in `MatSetOption()` for how to use column-major order.
1694: Calls to `MatSetValues()` with the `INSERT_VALUES` and `ADD_VALUES`
1695: options cannot be mixed without intervening calls to the assembly
1696: routines.
1698: `MatSetValues()` uses 0-based row and column numbers in Fortran
1699: as well as in C.
1701: Negative indices may be passed in `ism` and `isn`, these rows and columns are
1702: simply ignored. This allows easily inserting element stiffness matrices
1703: with homogeneous Dirichlet boundary conditions that you don't want represented
1704: in the matrix.
1706: Fortran Note:
1707: If `v` is a two-dimensional array make sure to first call `MatSetOption(mat, MAT_ROW_ORIENTED, PETSC_FALSE, ierr)` before using this function,
1708: otherwise the transpose of `v` will seemingly be inserted in the matrix, since Fortran passes two-dimensional arrays with column orientation.
1710: Efficiency Alert:
1711: The routine `MatSetValuesBlocked()` may offer much better efficiency
1712: for users of block sparse formats (`MATSEQBAIJ` and `MATMPIBAIJ`).
1714: This is currently not optimized for any particular `ISType`
1716: .seealso: [](ch_matrices), `Mat`, `MatSetOption()`, `MatSetValues()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValuesBlocked()`, `MatSetValuesLocal()`,
1717: `InsertMode`, `INSERT_VALUES`, `ADD_VALUES`
1718: @*/
1719: PetscErrorCode MatSetValuesIS(Mat mat, IS ism, IS isn, const PetscScalar v[], InsertMode addv)
1720: {
1721: PetscInt m, n;
1722: const PetscInt *rows, *cols;
1724: PetscFunctionBeginHot;
1726: PetscCall(ISGetIndices(ism, &rows));
1727: PetscCall(ISGetIndices(isn, &cols));
1728: PetscCall(ISGetLocalSize(ism, &m));
1729: PetscCall(ISGetLocalSize(isn, &n));
1730: PetscCall(MatSetValues(mat, m, rows, n, cols, v, addv));
1731: PetscCall(ISRestoreIndices(ism, &rows));
1732: PetscCall(ISRestoreIndices(isn, &cols));
1733: PetscFunctionReturn(PETSC_SUCCESS);
1734: }
1736: /*@
1737: MatSetValuesRowLocal - Inserts a row of nonzero values into a matrix
1739: Not Collective
1741: Input Parameters:
1742: + mat - the matrix
1743: . row - the row to set
1744: - v - a one-dimensional array that contains the values
1746: Level: intermediate
1748: Notes:
1749: Currently only supported for `MATAIJ`.
1751: All the nonzero values in `row` must be provided
1753: The matrix must have previously had its column indices set, likely by having been assembled.
1755: `row` must belong to this MPI process
1757: .seealso: [](ch_matrices), `Mat`, `MatSetOption()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValuesBlocked()`, `MatSetValuesLocal()`,
1758: `InsertMode`, `INSERT_VALUES`, `ADD_VALUES`, `MatSetValues()`, `MatSetValuesRow()`, `MatSetLocalToGlobalMapping()`, `MATAIJ`
1759: @*/
1760: PetscErrorCode MatSetValuesRowLocal(Mat mat, PetscInt row, const PetscScalar v[])
1761: {
1762: PetscInt globalrow;
1764: PetscFunctionBegin;
1767: PetscAssertPointer(v, 3);
1768: PetscCall(ISLocalToGlobalMappingApply(mat->rmap->mapping, 1, &row, &globalrow));
1769: PetscCall(MatSetValuesRow(mat, globalrow, v));
1770: PetscFunctionReturn(PETSC_SUCCESS);
1771: }
1773: /*@
1774: MatSetValuesRow - Inserts a row of nonzero values into a matrix
1776: Not Collective
1778: Input Parameters:
1779: + mat - the matrix
1780: . row - the row to set
1781: - v - a one dimensional array of values
1783: Level: advanced
1785: Notes:
1786: Currently only supported for `MATAIJ`.
1788: All the nonzeros in `row` must be provided
1790: The matrix must have previously had its column indices set, likely by having been assembled.
1792: `row` must belong to this process
1794: .seealso: [](ch_matrices), `Mat`, `MatSetValues()`, `MatSetOption()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValuesBlocked()`, `MatSetValuesLocal()`,
1795: `InsertMode`, `INSERT_VALUES`, `ADD_VALUES`, `MATAIJ`
1796: @*/
1797: PetscErrorCode MatSetValuesRow(Mat mat, PetscInt row, const PetscScalar v[])
1798: {
1799: PetscFunctionBeginHot;
1802: MatCheckPreallocated(mat, 1);
1803: PetscAssertPointer(v, 3);
1804: PetscCheck(mat->insertmode != ADD_VALUES, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Cannot mix add and insert values");
1805: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
1806: mat->insertmode = INSERT_VALUES;
1808: if (mat->assembled) {
1809: mat->was_assembled = PETSC_TRUE;
1810: mat->assembled = PETSC_FALSE;
1811: }
1812: PetscCall(PetscLogEventBegin(MAT_SetValues, mat, 0, 0, 0));
1813: PetscUseTypeMethod(mat, setvaluesrow, row, v);
1814: PetscCall(PetscLogEventEnd(MAT_SetValues, mat, 0, 0, 0));
1815: PetscFunctionReturn(PETSC_SUCCESS);
1816: }
1818: // PetscClangLinter pragma disable: -fdoc-section-header-unknown
1819: /*@
1820: MatSetValuesStencil - Inserts or adds a block of values into a matrix.
1821: Using structured grid indexing
1823: Not Collective
1825: Input Parameters:
1826: + mat - the matrix
1827: . m - number of rows being entered
1828: . idxm - grid coordinates (and component number when dof > 1) for matrix rows being entered
1829: . n - number of columns being entered
1830: . idxn - grid coordinates (and component number when dof > 1) for matrix columns being entered
1831: . v - a one-dimensional array that contains the values implicitly stored as a two-dimensional array, by default in row-major order.
1832: See `MAT_ROW_ORIENTED` in `MatSetOption()` for how to use column-major order.
1833: - addv - either `ADD_VALUES` to add to existing entries at that location or `INSERT_VALUES` to replace existing entries with new values
1835: Level: beginner
1837: Notes:
1838: By default the values, `v`, are row-oriented. See `MatSetOption()` for other options.
1840: Calls to `MatSetValuesStencil()` with the `INSERT_VALUES` and `ADD_VALUES`
1841: options cannot be mixed without intervening calls to the assembly
1842: routines.
1844: The grid coordinates are across the entire grid, not just the local portion
1846: `MatSetValuesStencil()` uses 0-based row and column numbers in Fortran
1847: as well as in C.
1849: For setting/accessing vector values via array coordinates you can use the `DMDAVecGetArray()` routine
1851: In order to use this routine you must either obtain the matrix with `DMCreateMatrix()`
1852: or call `MatSetLocalToGlobalMapping()` and `MatSetStencil()` first.
1854: The columns and rows in the stencil passed in MUST be contained within the
1855: ghost region of the given process as set with DMDACreateXXX() or `MatSetStencil()`. For example,
1856: if you create a `DMDA` with an overlap of one grid level and on a particular process its first
1857: local nonghost x logical coordinate is 6 (so its first ghost x logical coordinate is 5) the
1858: first i index you can use in your column and row indices in `MatSetStencil()` is 5.
1860: For periodic boundary conditions use negative indices for values to the left (below 0; that are to be
1861: obtained by wrapping values from right edge). For values to the right of the last entry using that index plus one
1862: etc to obtain values that obtained by wrapping the values from the left edge. This does not work for anything but the
1863: `DM_BOUNDARY_PERIODIC` boundary type.
1865: For indices that don't mean anything for your case (like the k index when working in 2d) or the c index when you have
1866: a single value per point) you can skip filling those indices.
1868: Inspired by the structured grid interface to the HYPRE package
1869: (https://computation.llnl.gov/projects/hypre-scalable-linear-solvers-multigrid-methods)
1871: Fortran Notes:
1872: If any of `idxm`, `idxn`, and `v` are scalars pass them using, for example,
1873: .vb
1874: call MatSetValuesStencil(mat, one, [idxm], one, [idxn], [v], INSERT_VALUES, ierr)
1875: .ve
1877: If `v` is a two-dimensional array make sure to first call `MatSetOption(mat, MAT_ROW_ORIENTED, PETSC_FALSE, ierr)` before using this function,
1878: otherwise the transpose of `v` will seemingly be inserted in the matrix, since Fortran passes two-dimensional arrays with column orientation.
1880: Efficiency Alert:
1881: The routine `MatSetValuesBlockedStencil()` may offer much better efficiency
1882: for users of block sparse formats (`MATSEQBAIJ` and `MATMPIBAIJ`).
1884: .seealso: [](ch_matrices), `Mat`, `DMDA`, `MatSetOption()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValuesBlocked()`, `MatSetValuesLocal()`,
1885: `MatSetValues()`, `MatSetValuesBlockedStencil()`, `MatSetStencil()`, `DMCreateMatrix()`, `DMDAVecGetArray()`, `MatStencil`
1886: @*/
1887: PetscErrorCode MatSetValuesStencil(Mat mat, PetscInt m, const MatStencil idxm[], PetscInt n, const MatStencil idxn[], const PetscScalar v[], InsertMode addv)
1888: {
1889: PetscInt buf[8192], *bufm = NULL, *bufn = NULL, *jdxm, *jdxn;
1890: PetscInt j, i, dim = mat->stencil.dim, *dims = mat->stencil.dims + 1, tmp;
1891: PetscInt *starts = mat->stencil.starts, *dxm = (PetscInt *)idxm, *dxn = (PetscInt *)idxn, sdim = dim - (1 - (PetscInt)mat->stencil.noc);
1893: PetscFunctionBegin;
1894: if (!m || !n) PetscFunctionReturn(PETSC_SUCCESS); /* no values to insert */
1897: PetscAssertPointer(idxm, 3);
1898: PetscAssertPointer(idxn, 5);
1900: if ((m + n) <= (PetscInt)PETSC_STATIC_ARRAY_LENGTH(buf)) {
1901: jdxm = buf;
1902: jdxn = buf + m;
1903: } else {
1904: PetscCall(PetscMalloc2(m, &bufm, n, &bufn));
1905: jdxm = bufm;
1906: jdxn = bufn;
1907: }
1908: for (i = 0; i < m; i++) {
1909: for (j = 0; j < 3 - sdim; j++) dxm++;
1910: tmp = *dxm++ - starts[0];
1911: for (j = 0; j < dim - 1; j++) {
1912: if ((*dxm++ - starts[j + 1]) < 0 || tmp < 0) tmp = -1;
1913: else tmp = tmp * dims[j] + *(dxm - 1) - starts[j + 1];
1914: }
1915: if (mat->stencil.noc) dxm++;
1916: jdxm[i] = tmp;
1917: }
1918: for (i = 0; i < n; i++) {
1919: for (j = 0; j < 3 - sdim; j++) dxn++;
1920: tmp = *dxn++ - starts[0];
1921: for (j = 0; j < dim - 1; j++) {
1922: if ((*dxn++ - starts[j + 1]) < 0 || tmp < 0) tmp = -1;
1923: else tmp = tmp * dims[j] + *(dxn - 1) - starts[j + 1];
1924: }
1925: if (mat->stencil.noc) dxn++;
1926: jdxn[i] = tmp;
1927: }
1928: PetscCall(MatSetValuesLocal(mat, m, jdxm, n, jdxn, v, addv));
1929: PetscCall(PetscFree2(bufm, bufn));
1930: PetscFunctionReturn(PETSC_SUCCESS);
1931: }
1933: /*@
1934: MatSetValuesBlockedStencil - Inserts or adds a block of values into a matrix.
1935: Using structured grid indexing
1937: Not Collective
1939: Input Parameters:
1940: + mat - the matrix
1941: . m - number of rows being entered
1942: . idxm - grid coordinates for matrix rows being entered
1943: . n - number of columns being entered
1944: . idxn - grid coordinates for matrix columns being entered
1945: . v - a one-dimensional array that contains the values implicitly stored as a two-dimensional array, by default in row-major order.
1946: See `MAT_ROW_ORIENTED` in `MatSetOption()` for how to use column-major order.
1947: - addv - either `ADD_VALUES` to add to existing entries or `INSERT_VALUES` to replace existing entries with new values
1949: Level: beginner
1951: Notes:
1952: By default the values, `v`, are row-oriented and unsorted.
1953: See `MatSetOption()` for other options.
1955: Calls to `MatSetValuesBlockedStencil()` with the `INSERT_VALUES` and `ADD_VALUES`
1956: options cannot be mixed without intervening calls to the assembly
1957: routines.
1959: The grid coordinates are across the entire grid, not just the local portion
1961: `MatSetValuesBlockedStencil()` uses 0-based row and column numbers in Fortran
1962: as well as in C.
1964: For setting/accessing vector values via array coordinates you can use the `DMDAVecGetArray()` routine
1966: In order to use this routine you must either obtain the matrix with `DMCreateMatrix()`
1967: or call `MatSetBlockSize()`, `MatSetLocalToGlobalMapping()` and `MatSetStencil()` first.
1969: The columns and rows in the stencil passed in MUST be contained within the
1970: ghost region of the given process as set with DMDACreateXXX() or `MatSetStencil()`. For example,
1971: if you create a `DMDA` with an overlap of one grid level and on a particular process its first
1972: local nonghost x logical coordinate is 6 (so its first ghost x logical coordinate is 5) the
1973: first i index you can use in your column and row indices in `MatSetStencil()` is 5.
1975: Negative indices may be passed in `idxm` and `idxn`, these rows and columns are
1976: simply ignored. This allows easily inserting element stiffness matrices
1977: with homogeneous Dirichlet boundary conditions that you don't want represented
1978: in the matrix.
1980: Inspired by the structured grid interface to the HYPRE package
1981: (https://computation.llnl.gov/projects/hypre-scalable-linear-solvers-multigrid-methods)
1983: Fortran Notes:
1984: If any of `idxm`, `idxn`, and `v` are scalars pass them using, for example,
1985: .vb
1986: call MatSetValuesBlockedStencil(mat, one, [idxm], one, [idxn], [v], INSERT_VALUES, ierr)
1987: .ve
1989: If `v` is a two-dimensional array make sure to first call `MatSetOption(mat, MAT_ROW_ORIENTED, PETSC_FALSE, ierr)` before using this function,
1990: otherwise the transpose of `v` will seemingly be inserted in the matrix, since Fortran passes two-dimensional arrays with column orientation.
1992: .seealso: [](ch_matrices), `Mat`, `DMDA`, `MatSetOption()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValuesBlocked()`, `MatSetValuesLocal()`,
1993: `MatSetValues()`, `MatSetValuesStencil()`, `MatSetStencil()`, `DMCreateMatrix()`, `DMDAVecGetArray()`, `MatStencil`,
1994: `MatSetBlockSize()`, `MatSetLocalToGlobalMapping()`
1995: @*/
1996: PetscErrorCode MatSetValuesBlockedStencil(Mat mat, PetscInt m, const MatStencil idxm[], PetscInt n, const MatStencil idxn[], const PetscScalar v[], InsertMode addv)
1997: {
1998: PetscInt buf[8192], *bufm = NULL, *bufn = NULL, *jdxm, *jdxn;
1999: PetscInt j, i, dim = mat->stencil.dim, *dims = mat->stencil.dims + 1, tmp;
2000: PetscInt *starts = mat->stencil.starts, *dxm = (PetscInt *)idxm, *dxn = (PetscInt *)idxn, sdim = dim - (1 - (PetscInt)mat->stencil.noc);
2002: PetscFunctionBegin;
2003: if (!m || !n) PetscFunctionReturn(PETSC_SUCCESS); /* no values to insert */
2006: PetscAssertPointer(idxm, 3);
2007: PetscAssertPointer(idxn, 5);
2008: PetscAssertPointer(v, 6);
2010: if ((m + n) <= (PetscInt)PETSC_STATIC_ARRAY_LENGTH(buf)) {
2011: jdxm = buf;
2012: jdxn = buf + m;
2013: } else {
2014: PetscCall(PetscMalloc2(m, &bufm, n, &bufn));
2015: jdxm = bufm;
2016: jdxn = bufn;
2017: }
2018: for (i = 0; i < m; i++) {
2019: for (j = 0; j < 3 - sdim; j++) dxm++;
2020: tmp = *dxm++ - starts[0];
2021: for (j = 0; j < sdim - 1; j++) {
2022: if ((*dxm++ - starts[j + 1]) < 0 || tmp < 0) tmp = -1;
2023: else tmp = tmp * dims[j] + *(dxm - 1) - starts[j + 1];
2024: }
2025: dxm++;
2026: jdxm[i] = tmp;
2027: }
2028: for (i = 0; i < n; i++) {
2029: for (j = 0; j < 3 - sdim; j++) dxn++;
2030: tmp = *dxn++ - starts[0];
2031: for (j = 0; j < sdim - 1; j++) {
2032: if ((*dxn++ - starts[j + 1]) < 0 || tmp < 0) tmp = -1;
2033: else tmp = tmp * dims[j] + *(dxn - 1) - starts[j + 1];
2034: }
2035: dxn++;
2036: jdxn[i] = tmp;
2037: }
2038: PetscCall(MatSetValuesBlockedLocal(mat, m, jdxm, n, jdxn, v, addv));
2039: PetscCall(PetscFree2(bufm, bufn));
2040: PetscFunctionReturn(PETSC_SUCCESS);
2041: }
2043: /*@
2044: MatSetStencil - Sets the grid information for setting values into a matrix via
2045: `MatSetValuesStencil()`
2047: Not Collective
2049: Input Parameters:
2050: + mat - the matrix
2051: . dim - dimension of the grid 1, 2, or 3
2052: . dims - number of grid points in x, y, and z direction, including ghost points on your process
2053: . starts - starting point of ghost nodes on your process in x, y, and z direction
2054: - dof - number of degrees of freedom per node
2056: Level: beginner
2058: Notes:
2059: Inspired by the structured grid interface to the HYPRE package
2060: (www.llnl.gov/CASC/hyper)
2062: For matrices generated with `DMCreateMatrix()` this routine is automatically called and so not needed by the
2063: user.
2065: .seealso: [](ch_matrices), `Mat`, `MatStencil`, `MatSetOption()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValuesBlocked()`, `MatSetValuesLocal()`,
2066: `MatSetValues()`, `MatSetValuesBlockedStencil()`, `MatSetValuesStencil()`
2067: @*/
2068: PetscErrorCode MatSetStencil(Mat mat, PetscInt dim, const PetscInt dims[], const PetscInt starts[], PetscInt dof)
2069: {
2070: PetscFunctionBegin;
2072: PetscAssertPointer(dims, 3);
2073: PetscAssertPointer(starts, 4);
2075: mat->stencil.dim = dim + (dof > 1);
2076: for (PetscInt i = 0; i < dim; i++) {
2077: mat->stencil.dims[i] = dims[dim - i - 1]; /* copy the values in backwards */
2078: mat->stencil.starts[i] = starts[dim - i - 1];
2079: }
2080: mat->stencil.dims[dim] = dof;
2081: mat->stencil.starts[dim] = 0;
2082: mat->stencil.noc = (PetscBool)(dof == 1);
2083: PetscFunctionReturn(PETSC_SUCCESS);
2084: }
2086: /*@
2087: MatSetValuesBlocked - Inserts or adds a block of values into a matrix.
2089: Not Collective
2091: Input Parameters:
2092: + mat - the matrix
2093: . m - the number of block rows
2094: . idxm - the global block indices
2095: . n - the number of block columns
2096: . idxn - the global block indices
2097: . v - a one-dimensional array that contains the values implicitly stored as a two-dimensional array, by default in row-major order.
2098: See `MAT_ROW_ORIENTED` in `MatSetOption()` for how to use column-major order.
2099: - addv - either `ADD_VALUES` to add values to any existing entries, or `INSERT_VALUES` replaces existing entries with new values
2101: Level: intermediate
2103: Notes:
2104: If you create the matrix yourself (that is not with a call to `DMCreateMatrix()`) then you MUST call
2105: MatXXXXSetPreallocation() or `MatSetUp()` before using this routine.
2107: The `m` and `n` count the NUMBER of blocks in the row direction and column direction,
2108: NOT the total number of rows/columns; for example, if the block size is 2 and
2109: you are passing in values for rows 2,3,4,5 then `m` would be 2 (not 4).
2110: The values in `idxm` would be 1 2; that is the first index for each block divided by
2111: the block size.
2113: You must call `MatSetBlockSize()` when constructing this matrix (before
2114: preallocating it).
2116: By default, the values, `v`, are stored in row-major order. See `MAT_ROW_ORIENTED` in `MatSetOption()` for how to use column-major order.
2118: Calls to `MatSetValuesBlocked()` with the `INSERT_VALUES` and `ADD_VALUES`
2119: options cannot be mixed without intervening calls to the assembly
2120: routines.
2122: `MatSetValuesBlocked()` uses 0-based row and column numbers in Fortran
2123: as well as in C.
2125: Negative indices may be passed in `idxm` and `idxn`, these rows and columns are
2126: simply ignored. This allows easily inserting element stiffness matrices
2127: with homogeneous Dirichlet boundary conditions that you don't want represented
2128: in the matrix.
2130: Each time an entry is set within a sparse matrix via `MatSetValues()`,
2131: internal searching must be done to determine where to place the
2132: data in the matrix storage space. By instead inserting blocks of
2133: entries via `MatSetValuesBlocked()`, the overhead of matrix assembly is
2134: reduced.
2136: Example:
2137: .vb
2138: Suppose m=n=2 and block size(bs) = 2 The array is
2140: 1 2 | 3 4
2141: 5 6 | 7 8
2142: - - - | - - -
2143: 9 10 | 11 12
2144: 13 14 | 15 16
2146: v[] should be passed in like
2147: v[] = [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16]
2149: If you are not using row-oriented storage of v (that is you called MatSetOption(mat,MAT_ROW_ORIENTED,PETSC_FALSE)) then
2150: v[] = [1,5,9,13,2,6,10,14,3,7,11,15,4,8,12,16]
2151: .ve
2153: Fortran Notes:
2154: If any of `idmx`, `idxn`, and `v` are scalars pass them using, for example,
2155: .vb
2156: call MatSetValuesBlocked(mat, one, [idxm], one, [idxn], [v], INSERT_VALUES, ierr)
2157: .ve
2159: If `v` is a two-dimensional array make sure to first call `MatSetOption(mat, MAT_ROW_ORIENTED, PETSC_FALSE, ierr)` before using this function,
2160: otherwise the transpose of `v` will seemingly be inserted in the matrix, since Fortran passes two-dimensional arrays with column orientation.
2162: .seealso: [](ch_matrices), `Mat`, `MatSetBlockSize()`, `MatSetOption()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValues()`, `MatSetValuesBlockedLocal()`
2163: @*/
2164: PetscErrorCode MatSetValuesBlocked(Mat mat, PetscInt m, const PetscInt idxm[], PetscInt n, const PetscInt idxn[], const PetscScalar v[], InsertMode addv)
2165: {
2166: PetscFunctionBeginHot;
2169: if (!m || !n) PetscFunctionReturn(PETSC_SUCCESS); /* no values to insert */
2170: PetscAssertPointer(idxm, 3);
2171: PetscAssertPointer(idxn, 5);
2172: MatCheckPreallocated(mat, 1);
2173: if (mat->insertmode == NOT_SET_VALUES) mat->insertmode = addv;
2174: else PetscCheck(mat->insertmode == addv, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Cannot mix add values and insert values");
2175: if (PetscDefined(USE_DEBUG)) {
2176: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2177: PetscCheck(mat->ops->setvaluesblocked || mat->ops->setvalues, PETSC_COMM_SELF, PETSC_ERR_SUP, "Mat type %s", ((PetscObject)mat)->type_name);
2178: }
2179: if (PetscDefined(USE_DEBUG)) {
2180: PetscInt rbs, cbs, M, N, i;
2181: PetscCall(MatGetBlockSizes(mat, &rbs, &cbs));
2182: PetscCall(MatGetSize(mat, &M, &N));
2183: for (i = 0; i < m; i++) PetscCheck(idxm[i] * rbs < M, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Row block %" PetscInt_FMT " contains an index %" PetscInt_FMT "*%" PetscInt_FMT " greater than row length %" PetscInt_FMT, i, idxm[i], rbs, M);
2184: for (i = 0; i < n; i++)
2185: PetscCheck(idxn[i] * cbs < N, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Column block %" PetscInt_FMT " contains an index %" PetscInt_FMT "*%" PetscInt_FMT " greater than column length %" PetscInt_FMT, i, idxn[i], cbs, N);
2186: }
2187: if (mat->assembled) {
2188: mat->was_assembled = PETSC_TRUE;
2189: mat->assembled = PETSC_FALSE;
2190: }
2191: PetscCall(PetscLogEventBegin(MAT_SetValues, mat, 0, 0, 0));
2192: if (mat->ops->setvaluesblocked) PetscUseTypeMethod(mat, setvaluesblocked, m, idxm, n, idxn, v, addv);
2193: else {
2194: PetscInt buf[8192], *bufr = NULL, *bufc = NULL, *iidxm, *iidxn;
2195: PetscInt i, j, bs, cbs;
2197: PetscCall(MatGetBlockSizes(mat, &bs, &cbs));
2198: if ((m * bs + n * cbs) <= (PetscInt)PETSC_STATIC_ARRAY_LENGTH(buf)) {
2199: iidxm = buf;
2200: iidxn = buf + m * bs;
2201: } else {
2202: PetscCall(PetscMalloc2(m * bs, &bufr, n * cbs, &bufc));
2203: iidxm = bufr;
2204: iidxn = bufc;
2205: }
2206: for (i = 0; i < m; i++) {
2207: for (j = 0; j < bs; j++) iidxm[i * bs + j] = bs * idxm[i] + j;
2208: }
2209: if (m != n || bs != cbs || idxm != idxn) {
2210: for (i = 0; i < n; i++) {
2211: for (j = 0; j < cbs; j++) iidxn[i * cbs + j] = cbs * idxn[i] + j;
2212: }
2213: } else iidxn = iidxm;
2214: PetscCall(MatSetValues(mat, m * bs, iidxm, n * cbs, iidxn, v, addv));
2215: PetscCall(PetscFree2(bufr, bufc));
2216: }
2217: PetscCall(PetscLogEventEnd(MAT_SetValues, mat, 0, 0, 0));
2218: PetscFunctionReturn(PETSC_SUCCESS);
2219: }
2221: /*@
2222: MatGetValues - Gets a block of local values from a matrix.
2224: Not Collective; can only return values that are owned by the give process
2226: Input Parameters:
2227: + mat - the matrix
2228: . v - a logically two-dimensional array for storing the values
2229: . m - the number of rows
2230: . idxm - the global indices of the rows
2231: . n - the number of columns
2232: - idxn - the global indices of the columns
2234: Level: advanced
2236: Notes:
2237: The user must allocate space (m*n `PetscScalar`s) for the values, `v`.
2239: The values, `v`, are returned in a row-oriented format, analogous to that used by default in `MatSetValues()`,
2240: unless `MatSetOption(mat, MAT_ROW_ORIENTED, PETSC_FALSE)` is called in which case they are returned column oriented.
2242: `MatGetValues()` uses 0-based row and column numbers in
2243: Fortran as well as in C.
2245: For `MATSBAIJ` matrices only the block upper triangular entries will be set.
2247: `MatGetValues()` requires that the matrix has been assembled
2248: with `MatAssemblyBegin()`/`MatAssemblyEnd()`. Thus, calls to
2249: `MatSetValues()` and `MatGetValues()` CANNOT be made in succession
2250: without intermediate matrix assembly.
2252: Negative row or column indices will be ignored and those locations in `v` will be
2253: left unchanged.
2255: For the standard row-based matrix formats, `idxm` can only contain rows owned by the requesting MPI process.
2256: That is, rows with global index greater than or equal to `rstart` and less than `rend` where `rstart` and `rend` are obtainable
2257: from `MatGetOwnershipRange`(mat,&rstart,&rend).
2259: .seealso: [](ch_matrices), `Mat`, `MatGetRow()`, `MatCreateSubMatrices()`, `MatSetValues()`, `MatGetOwnershipRange()`, `MatGetValuesLocal()`, `MatGetValue()`
2260: @*/
2261: PetscErrorCode MatGetValues(Mat mat, PetscInt m, const PetscInt idxm[], PetscInt n, const PetscInt idxn[], PetscScalar v[])
2262: {
2263: PetscFunctionBegin;
2266: if (!m || !n) PetscFunctionReturn(PETSC_SUCCESS);
2267: PetscAssertPointer(idxm, 3);
2268: PetscAssertPointer(idxn, 5);
2269: PetscAssertPointer(v, 6);
2270: PetscCheck(mat->assembled, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2271: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2272: MatCheckPreallocated(mat, 1);
2274: PetscCall(PetscLogEventBegin(MAT_GetValues, mat, 0, 0, 0));
2275: PetscUseTypeMethod(mat, getvalues, m, idxm, n, idxn, v);
2276: PetscCall(PetscLogEventEnd(MAT_GetValues, mat, 0, 0, 0));
2277: PetscFunctionReturn(PETSC_SUCCESS);
2278: }
2280: /*@
2281: MatGetValuesLocal - retrieves values from certain locations in a matrix using the local numbering of the indices
2282: defined previously by `MatSetLocalToGlobalMapping()`
2284: Not Collective
2286: Input Parameters:
2287: + mat - the matrix
2288: . nrow - number of rows
2289: . irow - the row local indices
2290: . ncol - number of columns
2291: - icol - the column local indices
2293: Output Parameter:
2294: . y - a one-dimensional array that contains the values implicitly stored as a two-dimensional array, by default in row-major order.
2295: See `MAT_ROW_ORIENTED` in `MatSetOption()` for how to use column-major order.
2297: Level: advanced
2299: Notes:
2300: If you create the matrix yourself (that is not with a call to `DMCreateMatrix()`) then you MUST call `MatSetLocalToGlobalMapping()` before using this routine.
2302: This routine can only return values that are owned by the requesting MPI process. That is, for standard matrix formats, rows that, in the global numbering,
2303: are greater than or equal to rstart and less than rend where rstart and rend are obtainable from `MatGetOwnershipRange`(mat,&rstart,&rend). One can
2304: determine if the resulting global row associated with the local row r is owned by the requesting MPI process by applying the `ISLocalToGlobalMapping` set
2305: with `MatSetLocalToGlobalMapping()`.
2307: .seealso: [](ch_matrices), `Mat`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValues()`, `MatSetLocalToGlobalMapping()`,
2308: `MatSetValuesLocal()`, `MatGetValues()`
2309: @*/
2310: PetscErrorCode MatGetValuesLocal(Mat mat, PetscInt nrow, const PetscInt irow[], PetscInt ncol, const PetscInt icol[], PetscScalar y[])
2311: {
2312: PetscFunctionBeginHot;
2315: MatCheckPreallocated(mat, 1);
2316: if (!nrow || !ncol) PetscFunctionReturn(PETSC_SUCCESS); /* no values to retrieve */
2317: PetscAssertPointer(irow, 3);
2318: PetscAssertPointer(icol, 5);
2319: if (PetscDefined(USE_DEBUG)) {
2320: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2321: PetscCheck(mat->ops->getvalueslocal || mat->ops->getvalues, PETSC_COMM_SELF, PETSC_ERR_SUP, "Mat type %s", ((PetscObject)mat)->type_name);
2322: }
2323: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2324: PetscCall(PetscLogEventBegin(MAT_GetValues, mat, 0, 0, 0));
2325: if (mat->ops->getvalueslocal) PetscUseTypeMethod(mat, getvalueslocal, nrow, irow, ncol, icol, y);
2326: else {
2327: PetscInt buf[8192], *bufr = NULL, *bufc = NULL, *irowm, *icolm;
2328: if ((nrow + ncol) <= (PetscInt)PETSC_STATIC_ARRAY_LENGTH(buf)) {
2329: irowm = buf;
2330: icolm = buf + nrow;
2331: } else {
2332: PetscCall(PetscMalloc2(nrow, &bufr, ncol, &bufc));
2333: irowm = bufr;
2334: icolm = bufc;
2335: }
2336: PetscCheck(mat->rmap->mapping, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "MatGetValuesLocal() cannot proceed without local-to-global row mapping (See MatSetLocalToGlobalMapping()).");
2337: PetscCheck(mat->cmap->mapping, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "MatGetValuesLocal() cannot proceed without local-to-global column mapping (See MatSetLocalToGlobalMapping()).");
2338: PetscCall(ISLocalToGlobalMappingApply(mat->rmap->mapping, nrow, irow, irowm));
2339: PetscCall(ISLocalToGlobalMappingApply(mat->cmap->mapping, ncol, icol, icolm));
2340: PetscCall(MatGetValues(mat, nrow, irowm, ncol, icolm, y));
2341: PetscCall(PetscFree2(bufr, bufc));
2342: }
2343: PetscCall(PetscLogEventEnd(MAT_GetValues, mat, 0, 0, 0));
2344: PetscFunctionReturn(PETSC_SUCCESS);
2345: }
2347: /*@
2348: MatSetValuesBatch - Adds (`ADD_VALUES`) many blocks of values into a matrix at once. The blocks must all be square and
2349: the same size. Currently, this can only be called once and creates the given matrix.
2351: Not Collective
2353: Input Parameters:
2354: + mat - the matrix
2355: . nb - the number of blocks
2356: . bs - the number of rows (and columns) in each block
2357: . rows - a concatenation of the rows for each block
2358: - v - a concatenation of logically two-dimensional arrays of values
2360: Level: advanced
2362: Notes:
2363: `MatSetPreallocationCOO()` and `MatSetValuesCOO()` may be a better way to provide the values
2365: In the future, we will extend this routine to handle rectangular blocks, and to allow multiple calls for a given matrix.
2367: .seealso: [](ch_matrices), `Mat`, `MatSetOption()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValuesBlocked()`, `MatSetValuesLocal()`,
2368: `InsertMode`, `INSERT_VALUES`, `ADD_VALUES`, `MatSetValues()`, `MatSetPreallocationCOO()`, `MatSetValuesCOO()`
2369: @*/
2370: PetscErrorCode MatSetValuesBatch(Mat mat, PetscInt nb, PetscInt bs, PetscInt rows[], const PetscScalar v[])
2371: {
2372: PetscFunctionBegin;
2375: PetscAssertPointer(rows, 4);
2376: PetscAssertPointer(v, 5);
2377: PetscAssert(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2379: PetscCall(PetscLogEventBegin(MAT_SetValuesBatch, mat, 0, 0, 0));
2380: for (PetscInt b = 0; b < nb; ++b) PetscCall(MatSetValues(mat, bs, &rows[b * bs], bs, &rows[b * bs], &v[b * bs * bs], ADD_VALUES));
2381: PetscCall(PetscLogEventEnd(MAT_SetValuesBatch, mat, 0, 0, 0));
2382: PetscFunctionReturn(PETSC_SUCCESS);
2383: }
2385: /*@
2386: MatSetLocalToGlobalMapping - Sets a local-to-global numbering for use by
2387: the routine `MatSetValuesLocal()` to allow users to insert matrix entries
2388: using a local (per-process) numbering.
2390: Not Collective
2392: Input Parameters:
2393: + x - the matrix
2394: . rmapping - row mapping created with `ISLocalToGlobalMappingCreate()` or `ISLocalToGlobalMappingCreateIS()`
2395: - cmapping - column mapping
2397: Level: intermediate
2399: Note:
2400: If the matrix is obtained with `DMCreateMatrix()` then this may already have been called on the matrix
2402: .seealso: [](ch_matrices), `Mat`, `DM`, `DMCreateMatrix()`, `MatGetLocalToGlobalMapping()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValues()`, `MatSetValuesLocal()`, `MatGetValuesLocal()`
2403: @*/
2404: PetscErrorCode MatSetLocalToGlobalMapping(Mat x, ISLocalToGlobalMapping rmapping, ISLocalToGlobalMapping cmapping)
2405: {
2406: PetscFunctionBegin;
2411: if (x->ops->setlocaltoglobalmapping) PetscUseTypeMethod(x, setlocaltoglobalmapping, rmapping, cmapping);
2412: else {
2413: PetscCall(PetscLayoutSetISLocalToGlobalMapping(x->rmap, rmapping));
2414: PetscCall(PetscLayoutSetISLocalToGlobalMapping(x->cmap, cmapping));
2415: }
2416: PetscFunctionReturn(PETSC_SUCCESS);
2417: }
2419: /*@
2420: MatGetLocalToGlobalMapping - Gets the local-to-global numbering set by `MatSetLocalToGlobalMapping()`
2422: Not Collective
2424: Input Parameter:
2425: . A - the matrix
2427: Output Parameters:
2428: + rmapping - row mapping
2429: - cmapping - column mapping
2431: Level: advanced
2433: .seealso: [](ch_matrices), `Mat`, `MatSetLocalToGlobalMapping()`, `MatSetValuesLocal()`
2434: @*/
2435: PetscErrorCode MatGetLocalToGlobalMapping(Mat A, ISLocalToGlobalMapping *rmapping, ISLocalToGlobalMapping *cmapping)
2436: {
2437: PetscFunctionBegin;
2440: if (rmapping) {
2441: PetscAssertPointer(rmapping, 2);
2442: *rmapping = A->rmap->mapping;
2443: }
2444: if (cmapping) {
2445: PetscAssertPointer(cmapping, 3);
2446: *cmapping = A->cmap->mapping;
2447: }
2448: PetscFunctionReturn(PETSC_SUCCESS);
2449: }
2451: /*@
2452: MatSetLayouts - Sets the `PetscLayout` objects for rows and columns of a matrix
2454: Logically Collective
2456: Input Parameters:
2457: + A - the matrix
2458: . rmap - row layout
2459: - cmap - column layout
2461: Level: advanced
2463: Note:
2464: The `PetscLayout` objects are usually created automatically for the matrix so this routine rarely needs to be called.
2466: .seealso: [](ch_matrices), `Mat`, `PetscLayout`, `MatCreateVecs()`, `MatGetLocalToGlobalMapping()`, `MatGetLayouts()`
2467: @*/
2468: PetscErrorCode MatSetLayouts(Mat A, PetscLayout rmap, PetscLayout cmap)
2469: {
2470: PetscFunctionBegin;
2472: PetscCall(PetscLayoutReference(rmap, &A->rmap));
2473: PetscCall(PetscLayoutReference(cmap, &A->cmap));
2474: PetscFunctionReturn(PETSC_SUCCESS);
2475: }
2477: /*@
2478: MatGetLayouts - Gets the `PetscLayout` objects for rows and columns
2480: Not Collective
2482: Input Parameter:
2483: . A - the matrix
2485: Output Parameters:
2486: + rmap - row layout
2487: - cmap - column layout
2489: Level: advanced
2491: .seealso: [](ch_matrices), `Mat`, [Matrix Layouts](sec_matlayout), `PetscLayout`, `MatCreateVecs()`, `MatGetLocalToGlobalMapping()`, `MatSetLayouts()`
2492: @*/
2493: PetscErrorCode MatGetLayouts(Mat A, PetscLayout *rmap, PetscLayout *cmap)
2494: {
2495: PetscFunctionBegin;
2498: if (rmap) {
2499: PetscAssertPointer(rmap, 2);
2500: *rmap = A->rmap;
2501: }
2502: if (cmap) {
2503: PetscAssertPointer(cmap, 3);
2504: *cmap = A->cmap;
2505: }
2506: PetscFunctionReturn(PETSC_SUCCESS);
2507: }
2509: /*@
2510: MatSetValuesLocal - Inserts or adds values into certain locations of a matrix,
2511: using a local numbering of the rows and columns.
2513: Not Collective
2515: Input Parameters:
2516: + mat - the matrix
2517: . nrow - number of rows
2518: . irow - the row local indices
2519: . ncol - number of columns
2520: . icol - the column local indices
2521: . v - a one-dimensional array that contains the values implicitly stored as a two-dimensional array, by default in row-major order.
2522: See `MAT_ROW_ORIENTED` in `MatSetOption()` for how to use column-major order.
2523: - addv - either `ADD_VALUES` to add values to any existing entries, or `INSERT_VALUES` to replace existing entries with new values
2525: Level: intermediate
2527: Notes:
2528: If you create the matrix yourself (that is not with a call to `DMCreateMatrix()`) then you MUST call `MatSetLocalToGlobalMapping()` before using this routine
2530: Calls to `MatSetValuesLocal()` with the `INSERT_VALUES` and `ADD_VALUES`
2531: options cannot be mixed without intervening calls to the assembly
2532: routines.
2534: These values may be cached, so `MatAssemblyBegin()` and `MatAssemblyEnd()`
2535: MUST be called after all calls to `MatSetValuesLocal()` have been completed.
2537: Fortran Notes:
2538: If any of `irow`, `icol`, and `v` are scalars pass them using, for example,
2539: .vb
2540: call MatSetValuesLocal(mat, one, [irow], one, [icol], [v], INSERT_VALUES, ierr)
2541: .ve
2543: If `v` is a two-dimensional array make sure to first call `MatSetOption(mat, MAT_ROW_ORIENTED, PETSC_FALSE, ierr)` before using this function,
2544: otherwise the transpose of `v` will seemingly be inserted in the matrix, since Fortran passes two-dimensional arrays with column orientation.
2546: .seealso: [](ch_matrices), `Mat`, `MatAssemblyBegin()`, `MatAssemblyEnd()`, `MatSetValues()`, `MatSetLocalToGlobalMapping()`,
2547: `MatGetValuesLocal()`
2548: @*/
2549: PetscErrorCode MatSetValuesLocal(Mat mat, PetscInt nrow, const PetscInt irow[], PetscInt ncol, const PetscInt icol[], const PetscScalar v[], InsertMode addv)
2550: {
2551: PetscFunctionBeginHot;
2554: MatCheckPreallocated(mat, 1);
2555: if (!nrow || !ncol) PetscFunctionReturn(PETSC_SUCCESS); /* no values to insert */
2556: PetscAssertPointer(irow, 3);
2557: PetscAssertPointer(icol, 5);
2558: if (mat->insertmode == NOT_SET_VALUES) mat->insertmode = addv;
2559: else PetscCheck(mat->insertmode == addv, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Cannot mix add values and insert values");
2560: if (PetscDefined(USE_DEBUG)) {
2561: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2562: PetscCheck(mat->ops->setvalueslocal || mat->ops->setvalues, PETSC_COMM_SELF, PETSC_ERR_SUP, "Mat type %s", ((PetscObject)mat)->type_name);
2563: }
2565: if (mat->assembled) {
2566: mat->was_assembled = PETSC_TRUE;
2567: mat->assembled = PETSC_FALSE;
2568: }
2569: PetscCall(PetscLogEventBegin(MAT_SetValues, mat, 0, 0, 0));
2570: if (mat->ops->setvalueslocal) PetscUseTypeMethod(mat, setvalueslocal, nrow, irow, ncol, icol, v, addv);
2571: else {
2572: PetscInt buf[8192], *bufr = NULL, *bufc = NULL;
2573: const PetscInt *irowm, *icolm;
2575: if ((!mat->rmap->mapping && !mat->cmap->mapping) || (nrow + ncol) <= (PetscInt)PETSC_STATIC_ARRAY_LENGTH(buf)) {
2576: bufr = buf;
2577: bufc = buf + nrow;
2578: irowm = bufr;
2579: icolm = bufc;
2580: } else {
2581: PetscCall(PetscMalloc2(nrow, &bufr, ncol, &bufc));
2582: irowm = bufr;
2583: icolm = bufc;
2584: }
2585: if (mat->rmap->mapping) PetscCall(ISLocalToGlobalMappingApply(mat->rmap->mapping, nrow, irow, bufr));
2586: else irowm = irow;
2587: if (mat->cmap->mapping) {
2588: if (mat->cmap->mapping != mat->rmap->mapping || ncol != nrow || icol != irow) PetscCall(ISLocalToGlobalMappingApply(mat->cmap->mapping, ncol, icol, bufc));
2589: else icolm = irowm;
2590: } else icolm = icol;
2591: PetscCall(MatSetValues(mat, nrow, irowm, ncol, icolm, v, addv));
2592: if (bufr != buf) PetscCall(PetscFree2(bufr, bufc));
2593: }
2594: PetscCall(PetscLogEventEnd(MAT_SetValues, mat, 0, 0, 0));
2595: PetscFunctionReturn(PETSC_SUCCESS);
2596: }
2598: /*@
2599: MatSetValuesBlockedLocal - Inserts or adds values into certain locations of a matrix,
2600: using a local ordering of the nodes a block at a time.
2602: Not Collective
2604: Input Parameters:
2605: + mat - the matrix
2606: . nrow - number of rows
2607: . irow - the row local indices
2608: . ncol - number of columns
2609: . icol - the column local indices
2610: . v - a one-dimensional array that contains the values implicitly stored as a two-dimensional array, by default in row-major order.
2611: See `MAT_ROW_ORIENTED` in `MatSetOption()` for how to use column-major order.
2612: - addv - either `ADD_VALUES` to add values to any existing entries, or `INSERT_VALUES` to replace existing entries with new values
2614: Level: intermediate
2616: Notes:
2617: If you create the matrix yourself (that is not with a call to `DMCreateMatrix()`) then you MUST call `MatSetBlockSize()` and `MatSetLocalToGlobalMapping()`
2618: before using this routineBefore calling `MatSetValuesLocal()`, the user must first set the
2620: Calls to `MatSetValuesBlockedLocal()` with the `INSERT_VALUES` and `ADD_VALUES`
2621: options cannot be mixed without intervening calls to the assembly
2622: routines.
2624: These values may be cached, so `MatAssemblyBegin()` and `MatAssemblyEnd()`
2625: MUST be called after all calls to `MatSetValuesBlockedLocal()` have been completed.
2627: Fortran Notes:
2628: If any of `irow`, `icol`, and `v` are scalars pass them using, for example,
2629: .vb
2630: call MatSetValuesBlockedLocal(mat, one, [irow], one, [icol], [v], INSERT_VALUES, ierr)
2631: .ve
2633: If `v` is a two-dimensional array make sure to first call `MatSetOption(mat, MAT_ROW_ORIENTED, PETSC_FALSE, ierr)` before using this function,
2634: otherwise the transpose of `v` will seemingly be inserted in the matrix, since Fortran passes two-dimensional arrays with column orientation.
2636: .seealso: [](ch_matrices), `Mat`, `MatSetBlockSize()`, `MatSetLocalToGlobalMapping()`, `MatAssemblyBegin()`, `MatAssemblyEnd()`,
2637: `MatSetValuesLocal()`, `MatSetValuesBlocked()`
2638: @*/
2639: PetscErrorCode MatSetValuesBlockedLocal(Mat mat, PetscInt nrow, const PetscInt irow[], PetscInt ncol, const PetscInt icol[], const PetscScalar v[], InsertMode addv)
2640: {
2641: PetscFunctionBeginHot;
2644: MatCheckPreallocated(mat, 1);
2645: if (!nrow || !ncol) PetscFunctionReturn(PETSC_SUCCESS); /* no values to insert */
2646: PetscAssertPointer(irow, 3);
2647: PetscAssertPointer(icol, 5);
2648: if (mat->insertmode == NOT_SET_VALUES) mat->insertmode = addv;
2649: else PetscCheck(mat->insertmode == addv, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Cannot mix add values and insert values");
2650: if (PetscDefined(USE_DEBUG)) {
2651: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2652: PetscCheck(mat->ops->setvaluesblockedlocal || mat->ops->setvaluesblocked || mat->ops->setvalueslocal || mat->ops->setvalues, PETSC_COMM_SELF, PETSC_ERR_SUP, "Mat type %s", ((PetscObject)mat)->type_name);
2653: }
2655: if (mat->assembled) {
2656: mat->was_assembled = PETSC_TRUE;
2657: mat->assembled = PETSC_FALSE;
2658: }
2659: if (PetscUnlikelyDebug(mat->rmap->mapping)) { /* Condition on the mapping existing, because MatSetValuesBlockedLocal_IS does not require it to be set. */
2660: PetscInt irbs, rbs;
2661: PetscCall(MatGetBlockSizes(mat, &rbs, NULL));
2662: PetscCall(ISLocalToGlobalMappingGetBlockSize(mat->rmap->mapping, &irbs));
2663: PetscCheck(rbs == irbs, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Different row block sizes! mat %" PetscInt_FMT ", row l2g map %" PetscInt_FMT, rbs, irbs);
2664: }
2665: if (PetscUnlikelyDebug(mat->cmap->mapping)) {
2666: PetscInt icbs, cbs;
2667: PetscCall(MatGetBlockSizes(mat, NULL, &cbs));
2668: PetscCall(ISLocalToGlobalMappingGetBlockSize(mat->cmap->mapping, &icbs));
2669: PetscCheck(cbs == icbs, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Different col block sizes! mat %" PetscInt_FMT ", col l2g map %" PetscInt_FMT, cbs, icbs);
2670: }
2671: PetscCall(PetscLogEventBegin(MAT_SetValues, mat, 0, 0, 0));
2672: if (mat->ops->setvaluesblockedlocal) PetscUseTypeMethod(mat, setvaluesblockedlocal, nrow, irow, ncol, icol, v, addv);
2673: else {
2674: PetscInt buf[8192], *bufr = NULL, *bufc = NULL;
2675: const PetscInt *irowm, *icolm;
2677: if ((!mat->rmap->mapping && !mat->cmap->mapping) || (nrow + ncol) <= ((PetscInt)PETSC_STATIC_ARRAY_LENGTH(buf))) {
2678: bufr = buf;
2679: bufc = buf + nrow;
2680: irowm = bufr;
2681: icolm = bufc;
2682: } else {
2683: PetscCall(PetscMalloc2(nrow, &bufr, ncol, &bufc));
2684: irowm = bufr;
2685: icolm = bufc;
2686: }
2687: if (mat->rmap->mapping) PetscCall(ISLocalToGlobalMappingApplyBlock(mat->rmap->mapping, nrow, irow, bufr));
2688: else irowm = irow;
2689: if (mat->cmap->mapping) {
2690: if (mat->cmap->mapping != mat->rmap->mapping || ncol != nrow || icol != irow) PetscCall(ISLocalToGlobalMappingApplyBlock(mat->cmap->mapping, ncol, icol, bufc));
2691: else icolm = irowm;
2692: } else icolm = icol;
2693: PetscCall(MatSetValuesBlocked(mat, nrow, irowm, ncol, icolm, v, addv));
2694: if (bufr != buf) PetscCall(PetscFree2(bufr, bufc));
2695: }
2696: PetscCall(PetscLogEventEnd(MAT_SetValues, mat, 0, 0, 0));
2697: PetscFunctionReturn(PETSC_SUCCESS);
2698: }
2700: /*@
2701: MatMultDiagonalBlock - Computes the matrix-vector product, $y = Dx$. Where `D` is defined by the inode or block structure of the diagonal
2703: Collective
2705: Input Parameters:
2706: + mat - the matrix
2707: - x - the vector to be multiplied
2709: Output Parameter:
2710: . y - the result
2712: Level: developer
2714: Note:
2715: The vectors `x` and `y` cannot be the same. I.e., one cannot
2716: call `MatMultDiagonalBlock`(A,y,y).
2718: .seealso: [](ch_matrices), `Mat`, `MatMult()`, `MatMultTranspose()`, `MatMultAdd()`, `MatMultTransposeAdd()`
2719: @*/
2720: PetscErrorCode MatMultDiagonalBlock(Mat mat, Vec x, Vec y)
2721: {
2722: PetscFunctionBegin;
2728: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2729: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2730: PetscCheck(x != y, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "x and y must be different vectors");
2731: MatCheckPreallocated(mat, 1);
2733: PetscUseTypeMethod(mat, multdiagonalblock, x, y);
2734: PetscCall(PetscObjectStateIncrease((PetscObject)y));
2735: PetscFunctionReturn(PETSC_SUCCESS);
2736: }
2738: /*@
2739: MatMult - Computes the matrix-vector product, $y = Ax$.
2741: Neighbor-wise Collective
2743: Input Parameters:
2744: + mat - the matrix
2745: - x - the vector to be multiplied
2747: Output Parameter:
2748: . y - the result
2750: Level: beginner
2752: Note:
2753: The vectors `x` and `y` cannot be the same. I.e., one cannot
2754: call `MatMult`(A,y,y).
2756: .seealso: [](ch_matrices), `Mat`, `MatMultTranspose()`, `MatMultAdd()`, `MatMultTransposeAdd()`
2757: @*/
2758: PetscErrorCode MatMult(Mat mat, Vec x, Vec y)
2759: {
2760: PetscFunctionBegin;
2764: VecCheckAssembled(x);
2766: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2767: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2768: PetscCheck(x != y, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "x and y must be different vectors");
2769: PetscCheck(mat->cmap->N == x->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, x->map->N);
2770: PetscCheck(mat->rmap->N == y->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec y: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->N, y->map->N);
2771: PetscCheck(mat->cmap->n == x->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->n, x->map->n);
2772: PetscCheck(mat->rmap->n == y->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec y: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->n, y->map->n);
2773: PetscCall(VecSetErrorIfLocked(y, 3));
2774: if (mat->erroriffailure) PetscCall(VecValidValues_Internal(x, 2, PETSC_TRUE));
2775: MatCheckPreallocated(mat, 1);
2777: PetscCall(VecLockReadPush(x));
2778: PetscCall(PetscLogEventBegin(MAT_Mult, mat, x, y, 0));
2779: PetscUseTypeMethod(mat, mult, x, y);
2780: PetscCall(PetscLogEventEnd(MAT_Mult, mat, x, y, 0));
2781: if (mat->erroriffailure) PetscCall(VecValidValues_Internal(y, 3, PETSC_FALSE));
2782: PetscCall(VecLockReadPop(x));
2783: PetscFunctionReturn(PETSC_SUCCESS);
2784: }
2786: /*@
2787: MatMultTranspose - Computes matrix transpose times a vector $y = A^T * x$.
2789: Neighbor-wise Collective
2791: Input Parameters:
2792: + mat - the matrix
2793: - x - the vector to be multiplied
2795: Output Parameter:
2796: . y - the result
2798: Level: beginner
2800: Notes:
2801: The vectors `x` and `y` cannot be the same. I.e., one cannot
2802: call `MatMultTranspose`(A,y,y).
2804: For complex numbers this does NOT compute the Hermitian (complex conjugate) transpose multiple,
2805: use `MatMultHermitianTranspose()`
2807: .seealso: [](ch_matrices), `Mat`, `MatMult()`, `MatMultAdd()`, `MatMultTransposeAdd()`, `MatMultHermitianTranspose()`, `MatTranspose()`
2808: @*/
2809: PetscErrorCode MatMultTranspose(Mat mat, Vec x, Vec y)
2810: {
2811: PetscErrorCode (*op)(Mat, Vec, Vec) = NULL;
2813: PetscFunctionBegin;
2817: VecCheckAssembled(x);
2820: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2821: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2822: PetscCheck(x != y, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "x and y must be different vectors");
2823: PetscCheck(mat->cmap->N == y->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec y: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, y->map->N);
2824: PetscCheck(mat->rmap->N == x->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->N, x->map->N);
2825: PetscCheck(mat->cmap->n == y->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec y: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->n, y->map->n);
2826: PetscCheck(mat->rmap->n == x->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->n, x->map->n);
2827: if (mat->erroriffailure) PetscCall(VecValidValues_Internal(x, 2, PETSC_TRUE));
2828: MatCheckPreallocated(mat, 1);
2830: if (!mat->ops->multtranspose) {
2831: if (mat->symmetric == PETSC_BOOL3_TRUE && mat->ops->mult) op = mat->ops->mult;
2832: PetscCheck(op, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Matrix type %s does not have a multiply transpose defined or is symmetric and does not have a multiply defined", ((PetscObject)mat)->type_name);
2833: } else op = mat->ops->multtranspose;
2834: PetscCall(PetscLogEventBegin(MAT_MultTranspose, mat, x, y, 0));
2835: PetscCall(VecLockReadPush(x));
2836: PetscCall((*op)(mat, x, y));
2837: PetscCall(VecLockReadPop(x));
2838: PetscCall(PetscLogEventEnd(MAT_MultTranspose, mat, x, y, 0));
2839: PetscCall(PetscObjectStateIncrease((PetscObject)y));
2840: if (mat->erroriffailure) PetscCall(VecValidValues_Internal(y, 3, PETSC_FALSE));
2841: PetscFunctionReturn(PETSC_SUCCESS);
2842: }
2844: /*@
2845: MatMultHermitianTranspose - Computes matrix Hermitian-transpose times a vector $y = A^H * x$.
2847: Neighbor-wise Collective
2849: Input Parameters:
2850: + mat - the matrix
2851: - x - the vector to be multiplied
2853: Output Parameter:
2854: . y - the result
2856: Level: beginner
2858: Notes:
2859: The vectors `x` and `y` cannot be the same. I.e., one cannot
2860: call `MatMultHermitianTranspose`(A,y,y).
2862: Also called the conjugate transpose, complex conjugate transpose, or adjoint.
2864: For real numbers `MatMultTranspose()` and `MatMultHermitianTranspose()` are identical.
2866: .seealso: [](ch_matrices), `Mat`, `MatMult()`, `MatMultAdd()`, `MatMultHermitianTransposeAdd()`, `MatMultTranspose()`
2867: @*/
2868: PetscErrorCode MatMultHermitianTranspose(Mat mat, Vec x, Vec y)
2869: {
2870: PetscFunctionBegin;
2876: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2877: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2878: PetscCheck(x != y, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "x and y must be different vectors");
2879: PetscCheck(mat->cmap->N == y->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec y: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, y->map->N);
2880: PetscCheck(mat->rmap->N == x->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->N, x->map->N);
2881: PetscCheck(mat->cmap->n == y->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec y: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->n, y->map->n);
2882: PetscCheck(mat->rmap->n == x->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->n, x->map->n);
2883: MatCheckPreallocated(mat, 1);
2885: PetscCall(PetscLogEventBegin(MAT_MultHermitianTranspose, mat, x, y, 0));
2886: if (PetscDefined(USE_COMPLEX)) {
2887: if (mat->ops->multhermitiantranspose || (mat->hermitian == PETSC_BOOL3_TRUE && mat->ops->mult)) {
2888: PetscCall(VecLockReadPush(x));
2889: if (mat->ops->multhermitiantranspose) PetscUseTypeMethod(mat, multhermitiantranspose, x, y);
2890: else PetscUseTypeMethod(mat, mult, x, y);
2891: PetscCall(VecLockReadPop(x));
2892: } else {
2893: Vec w;
2894: PetscCall(VecDuplicate(x, &w));
2895: PetscCall(VecCopy(x, w));
2896: PetscCall(VecConjugate(w));
2897: PetscCall(MatMultTranspose(mat, w, y));
2898: PetscCall(VecDestroy(&w));
2899: PetscCall(VecConjugate(y));
2900: }
2901: PetscCall(PetscObjectStateIncrease((PetscObject)y));
2902: } else PetscCall(MatMultTranspose(mat, x, y));
2903: PetscCall(PetscLogEventEnd(MAT_MultHermitianTranspose, mat, x, y, 0));
2904: PetscFunctionReturn(PETSC_SUCCESS);
2905: }
2907: /*@
2908: MatMultAdd - Computes $v3 = v2 + A * v1$.
2910: Neighbor-wise Collective
2912: Input Parameters:
2913: + mat - the matrix
2914: . v1 - the vector to be multiplied by `mat`
2915: - v2 - the vector to be added to the result
2917: Output Parameter:
2918: . v3 - the result
2920: Level: beginner
2922: Note:
2923: The vectors `v1` and `v3` cannot be the same. I.e., one cannot
2924: call `MatMultAdd`(A,v1,v2,v1).
2926: .seealso: [](ch_matrices), `Mat`, `MatMultTranspose()`, `MatMult()`, `MatMultTransposeAdd()`
2927: @*/
2928: PetscErrorCode MatMultAdd(Mat mat, Vec v1, Vec v2, Vec v3)
2929: {
2930: PetscFunctionBegin;
2937: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2938: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2939: PetscCheck(mat->cmap->N == v1->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec v1: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, v1->map->N);
2940: /* PetscCheck(mat->rmap->N == v2->map->N,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Mat mat,Vec v2: global dim %" PetscInt_FMT " %" PetscInt_FMT,mat->rmap->N,v2->map->N);
2941: PetscCheck(mat->rmap->N == v3->map->N,PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ,"Mat mat,Vec v3: global dim %" PetscInt_FMT " %" PetscInt_FMT,mat->rmap->N,v3->map->N); */
2942: PetscCheck(mat->rmap->n == v3->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec v3: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->n, v3->map->n);
2943: PetscCheck(mat->rmap->n == v2->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec v2: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->n, v2->map->n);
2944: PetscCheck(v1 != v3, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "v1 and v3 must be different vectors");
2945: MatCheckPreallocated(mat, 1);
2947: PetscCall(PetscLogEventBegin(MAT_MultAdd, mat, v1, v2, v3));
2948: PetscCall(VecLockReadPush(v1));
2949: PetscUseTypeMethod(mat, multadd, v1, v2, v3);
2950: PetscCall(VecLockReadPop(v1));
2951: PetscCall(PetscLogEventEnd(MAT_MultAdd, mat, v1, v2, v3));
2952: PetscCall(PetscObjectStateIncrease((PetscObject)v3));
2953: PetscFunctionReturn(PETSC_SUCCESS);
2954: }
2956: /*@
2957: MatMultTransposeAdd - Computes $v3 = v2 + A^T * v1$.
2959: Neighbor-wise Collective
2961: Input Parameters:
2962: + mat - the matrix
2963: . v1 - the vector to be multiplied by the transpose of the matrix
2964: - v2 - the vector to be added to the result
2966: Output Parameter:
2967: . v3 - the result
2969: Level: beginner
2971: Note:
2972: The vectors `v1` and `v3` cannot be the same. I.e., one cannot
2973: call `MatMultTransposeAdd`(A,v1,v2,v1).
2975: .seealso: [](ch_matrices), `Mat`, `MatMultTranspose()`, `MatMultAdd()`, `MatMult()`
2976: @*/
2977: PetscErrorCode MatMultTransposeAdd(Mat mat, Vec v1, Vec v2, Vec v3)
2978: {
2979: PetscErrorCode (*op)(Mat, Vec, Vec, Vec) = (!mat->ops->multtransposeadd && mat->symmetric) ? mat->ops->multadd : mat->ops->multtransposeadd;
2981: PetscFunctionBegin;
2988: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
2989: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
2990: PetscCheck(mat->rmap->N == v1->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec v1: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->N, v1->map->N);
2991: PetscCheck(mat->cmap->N == v2->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec v2: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, v2->map->N);
2992: PetscCheck(mat->cmap->N == v3->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec v3: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, v3->map->N);
2993: PetscCheck(v1 != v3, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "v1 and v3 must be different vectors");
2994: PetscCheck(op, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Mat type %s", ((PetscObject)mat)->type_name);
2995: MatCheckPreallocated(mat, 1);
2997: PetscCall(PetscLogEventBegin(MAT_MultTransposeAdd, mat, v1, v2, v3));
2998: PetscCall(VecLockReadPush(v1));
2999: PetscCall((*op)(mat, v1, v2, v3));
3000: PetscCall(VecLockReadPop(v1));
3001: PetscCall(PetscLogEventEnd(MAT_MultTransposeAdd, mat, v1, v2, v3));
3002: PetscCall(PetscObjectStateIncrease((PetscObject)v3));
3003: PetscFunctionReturn(PETSC_SUCCESS);
3004: }
3006: /*@
3007: MatMultHermitianTransposeAdd - Computes $v3 = v2 + A^H * v1$.
3009: Neighbor-wise Collective
3011: Input Parameters:
3012: + mat - the matrix
3013: . v1 - the vector to be multiplied by the Hermitian transpose
3014: - v2 - the vector to be added to the result
3016: Output Parameter:
3017: . v3 - the result
3019: Level: beginner
3021: Note:
3022: The vectors `v1` and `v3` cannot be the same. I.e., one cannot
3023: call `MatMultHermitianTransposeAdd`(A,v1,v2,v1).
3025: .seealso: [](ch_matrices), `Mat`, `MatMultHermitianTranspose()`, `MatMultTranspose()`, `MatMultAdd()`, `MatMult()`
3026: @*/
3027: PetscErrorCode MatMultHermitianTransposeAdd(Mat mat, Vec v1, Vec v2, Vec v3)
3028: {
3029: PetscFunctionBegin;
3036: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3037: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
3038: PetscCheck(v1 != v3, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "v1 and v3 must be different vectors");
3039: PetscCheck(mat->rmap->N == v1->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec v1: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->N, v1->map->N);
3040: PetscCheck(mat->cmap->N == v2->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec v2: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, v2->map->N);
3041: PetscCheck(mat->cmap->N == v3->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec v3: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, v3->map->N);
3042: MatCheckPreallocated(mat, 1);
3044: PetscCall(PetscLogEventBegin(MAT_MultHermitianTransposeAdd, mat, v1, v2, v3));
3045: PetscCall(VecLockReadPush(v1));
3046: if (mat->ops->multhermitiantransposeadd) PetscUseTypeMethod(mat, multhermitiantransposeadd, v1, v2, v3);
3047: else {
3048: Vec w, z;
3049: PetscCall(VecDuplicate(v1, &w));
3050: PetscCall(VecCopy(v1, w));
3051: PetscCall(VecConjugate(w));
3052: PetscCall(VecDuplicate(v3, &z));
3053: PetscCall(MatMultTranspose(mat, w, z));
3054: PetscCall(VecDestroy(&w));
3055: PetscCall(VecConjugate(z));
3056: if (v2 != v3) PetscCall(VecWAXPY(v3, 1.0, v2, z));
3057: else PetscCall(VecAXPY(v3, 1.0, z));
3058: PetscCall(VecDestroy(&z));
3059: }
3060: PetscCall(VecLockReadPop(v1));
3061: PetscCall(PetscLogEventEnd(MAT_MultHermitianTransposeAdd, mat, v1, v2, v3));
3062: PetscCall(PetscObjectStateIncrease((PetscObject)v3));
3063: PetscFunctionReturn(PETSC_SUCCESS);
3064: }
3066: static PetscErrorCode MatADot_Default(Mat mat, Vec x, Vec y, PetscScalar *val)
3067: {
3068: PetscFunctionBegin;
3069: if (!mat->dot_vec) PetscCall(MatCreateVecs(mat, NULL, &mat->dot_vec));
3070: PetscCall(MatMult(mat, x, mat->dot_vec));
3071: PetscCall(VecDot(mat->dot_vec, y, val));
3072: PetscFunctionReturn(PETSC_SUCCESS);
3073: }
3075: static PetscErrorCode MatANorm_Default(Mat mat, Vec x, PetscReal *val)
3076: {
3077: PetscScalar sval;
3079: PetscFunctionBegin;
3080: PetscCall(MatADot(mat, x, x, &sval));
3081: PetscCheck(PetscRealPart(sval) >= 0.0, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONG, "Matrix argument is not positive definite");
3082: PetscCheck(PetscAbsReal(PetscImaginaryPart(sval)) <= 100 * PETSC_MACHINE_EPSILON * PetscMax(1.0, PetscAbsScalar(sval)), PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONG, "Matrix argument is not Hermitian");
3083: *val = PetscSqrtReal(PetscRealPart(sval));
3084: PetscFunctionReturn(PETSC_SUCCESS);
3085: }
3087: /*@
3088: MatADot - Computes the inner product with respect to a matrix, i.e., $(x, y)_A = y^H A x$ where $A$ is symmetric (Hermitian when using complex)
3089: positive definite.
3091: Collective
3093: Input Parameters:
3094: + mat - matrix used to define the inner product
3095: . x - first vector
3096: - y - second vector
3098: Output Parameter:
3099: . val - the dot product with respect to `A`
3101: Level: intermediate
3103: Note:
3104: For complex vectors, `MatADot()` computes
3105: $$
3106: val = (x,y)_A = y^H A x,
3107: $$
3108: where $y^H$ denotes the conjugate transpose of `y`. Note that this corresponds to the "mathematicians" complex
3109: inner product where the SECOND argument gets the complex conjugate.
3111: .seealso: [](ch_matrices), `Mat`, `MatANorm()`, `VecDot()`, `VecNorm()`, `MatMult()`, `MatMultAdd()`, `MatMultTransposeAdd()`
3112: @*/
3113: PetscErrorCode MatADot(Mat mat, Vec x, Vec y, PetscScalar *val)
3114: {
3115: PetscFunctionBegin;
3119: VecCheckAssembled(x);
3121: VecCheckAssembled(y);
3124: PetscAssertPointer(val, 4);
3125: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3126: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
3127: PetscCheck(mat->cmap->N == x->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, x->map->N);
3128: PetscCheck(mat->rmap->N == y->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec y: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->N, y->map->N);
3129: PetscCheck(mat->cmap->n == x->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->n, x->map->n);
3130: PetscCheck(mat->rmap->n == y->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec y: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->n, y->map->n);
3131: if (mat->erroriffailure) PetscCall(VecValidValues_Internal(x, 2, PETSC_TRUE));
3132: if (mat->erroriffailure) PetscCall(VecValidValues_Internal(y, 3, PETSC_TRUE));
3133: MatCheckPreallocated(mat, 1);
3135: PetscCall(VecLockReadPush(x));
3136: PetscCall(VecLockReadPush(y));
3137: PetscCall(PetscLogEventBegin(MAT_ADot, mat, x, y, 0));
3138: if (mat->ops->adot) PetscUseTypeMethod(mat, adot, x, y, val);
3139: else PetscCall(MatADot_Default(mat, x, y, val));
3140: PetscCall(PetscLogEventEnd(MAT_ADot, mat, x, y, 0));
3141: PetscCall(VecLockReadPop(y));
3142: PetscCall(VecLockReadPop(x));
3143: PetscFunctionReturn(PETSC_SUCCESS);
3144: }
3146: /*@
3147: MatANorm - Computes the norm with respect to a matrix, i.e., $(x, x)_A^{1/2} = (x^H A x)^{1/2}$ where $A$ is symmetric (Hermitian when using complex)
3148: positive definite.
3150: Collective
3152: Input Parameters:
3153: + mat - matrix used to define norm
3154: - x - the vector to compute the norm of
3156: Output Parameter:
3157: . val - the norm with respect to `A`
3159: Level: intermediate
3161: Note:
3162: For complex vectors, `MatANorm()` computes
3163: $$
3164: val = (x,x)_A^{1/2} = (x^H A x)^{1/2},
3165: $$
3166: where $x^H$ denotes the conjugate transpose of `x`.
3168: .seealso: [](ch_matrices), `Mat`, `MatADot()`, `VecDot()`, `VecNorm()`, `MatMult()`, `MatMultAdd()`, `MatMultTransposeAdd()`
3169: @*/
3170: PetscErrorCode MatANorm(Mat mat, Vec x, PetscReal *val)
3171: {
3172: PetscFunctionBegin;
3176: VecCheckAssembled(x);
3178: PetscAssertPointer(val, 3);
3179: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3180: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
3181: PetscCheck(mat->cmap->N == x->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, x->map->N);
3182: PetscCheck(mat->rmap->N == x->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->N, x->map->N);
3183: PetscCheck(mat->cmap->n == x->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->n, x->map->n);
3184: PetscCheck(mat->rmap->n == x->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->n, x->map->n);
3185: if (mat->erroriffailure) PetscCall(VecValidValues_Internal(x, 2, PETSC_TRUE));
3186: MatCheckPreallocated(mat, 1);
3188: PetscCall(VecLockReadPush(x));
3189: PetscCall(PetscLogEventBegin(MAT_ANorm, mat, x, 0, 0));
3190: if (mat->ops->anorm) PetscUseTypeMethod(mat, anorm, x, val);
3191: else PetscCall(MatANorm_Default(mat, x, val));
3192: PetscCall(PetscLogEventEnd(MAT_ANorm, mat, x, 0, 0));
3193: PetscCall(VecLockReadPop(x));
3194: PetscFunctionReturn(PETSC_SUCCESS);
3195: }
3197: /*@
3198: MatGetFactorType - gets the type of factorization a matrix is
3200: Not Collective
3202: Input Parameter:
3203: . mat - the matrix
3205: Output Parameter:
3206: . t - the type, one of `MAT_FACTOR_NONE`, `MAT_FACTOR_LU`, `MAT_FACTOR_CHOLESKY`, `MAT_FACTOR_ILU`, `MAT_FACTOR_ICC,MAT_FACTOR_ILUDT`, `MAT_FACTOR_QR`
3208: Level: intermediate
3210: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatFactorType`, `MatGetFactor()`, `MatSetFactorType()`, `MAT_FACTOR_NONE`, `MAT_FACTOR_LU`, `MAT_FACTOR_CHOLESKY`, `MAT_FACTOR_ILU`,
3211: `MAT_FACTOR_ICC`, `MAT_FACTOR_ILUDT`, `MAT_FACTOR_QR`
3212: @*/
3213: PetscErrorCode MatGetFactorType(Mat mat, MatFactorType *t)
3214: {
3215: PetscFunctionBegin;
3218: PetscAssertPointer(t, 2);
3219: *t = mat->factortype;
3220: PetscFunctionReturn(PETSC_SUCCESS);
3221: }
3223: /*@
3224: MatSetFactorType - sets the type of factorization a matrix is
3226: Logically Collective
3228: Input Parameters:
3229: + mat - the matrix
3230: - t - the type, one of `MAT_FACTOR_NONE`, `MAT_FACTOR_LU`, `MAT_FACTOR_CHOLESKY`, `MAT_FACTOR_ILU`, `MAT_FACTOR_ICC,MAT_FACTOR_ILUDT`, `MAT_FACTOR_QR`
3232: Level: intermediate
3234: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatFactorType`, `MatGetFactor()`, `MatGetFactorType()`, `MAT_FACTOR_NONE`, `MAT_FACTOR_LU`, `MAT_FACTOR_CHOLESKY`, `MAT_FACTOR_ILU`,
3235: `MAT_FACTOR_ICC`, `MAT_FACTOR_ILUDT`, `MAT_FACTOR_QR`
3236: @*/
3237: PetscErrorCode MatSetFactorType(Mat mat, MatFactorType t)
3238: {
3239: PetscFunctionBegin;
3242: mat->factortype = t;
3243: PetscFunctionReturn(PETSC_SUCCESS);
3244: }
3246: /*@
3247: MatGetInfo - Returns information about matrix storage (number of
3248: nonzeros, memory, etc.).
3250: Collective if `MAT_GLOBAL_MAX` or `MAT_GLOBAL_SUM` is used as the flag
3252: Input Parameters:
3253: + mat - the matrix
3254: - flag - flag indicating the type of parameters to be returned (`MAT_LOCAL` - local matrix, `MAT_GLOBAL_MAX` - maximum over all processes, `MAT_GLOBAL_SUM` - sum over all processes)
3256: Output Parameter:
3257: . info - matrix information context
3259: Options Database Key:
3260: . -mat_view :[filename]:ascii_info - print the matrix information to `filename` or `stdout`, see `MatView()`
3262: Level: intermediate
3264: Notes:
3265: The `MatInfo` context contains a variety of matrix data, including
3266: number of nonzeros allocated and used, number of mallocs during
3267: matrix assembly, etc. Additional information for factored matrices
3268: is provided (such as the fill ratio, number of mallocs during
3269: factorization, etc.).
3271: Example:
3272: See the file `${PETSC_DIR}/include/petscmat.h` for a complete list of
3273: data within the `MatInfo` context. For example,
3274: .vb
3275: MatInfo info;
3276: Mat A;
3277: double mal, nz_a, nz_u;
3279: MatGetInfo(A, MAT_LOCAL, &info);
3280: mal = info.mallocs;
3281: nz_a = info.nz_allocated;
3282: .ve
3284: .seealso: [](ch_matrices), `Mat`, `MatInfo`, `MatStashGetInfo()`
3285: @*/
3286: PetscErrorCode MatGetInfo(Mat mat, MatInfoType flag, MatInfo *info)
3287: {
3288: PetscFunctionBegin;
3291: PetscAssertPointer(info, 3);
3292: MatCheckPreallocated(mat, 1);
3293: PetscUseTypeMethod(mat, getinfo, flag, info);
3294: PetscFunctionReturn(PETSC_SUCCESS);
3295: }
3297: /*
3298: This is used by external packages where it is not easy to get the info from the actual
3299: matrix factorization.
3300: */
3301: PetscErrorCode MatGetInfo_External(Mat A, MatInfoType flag, MatInfo *info)
3302: {
3303: PetscFunctionBegin;
3304: PetscCall(PetscMemzero(info, sizeof(MatInfo)));
3305: PetscFunctionReturn(PETSC_SUCCESS);
3306: }
3308: /*@
3309: MatLUFactor - Performs in-place LU factorization of matrix.
3311: Collective
3313: Input Parameters:
3314: + mat - the matrix
3315: . row - row permutation
3316: . col - column permutation
3317: - info - options for factorization, includes
3318: .vb
3319: fill - expected fill as ratio of original fill.
3320: dtcol - pivot tolerance (0 no pivot, 1 full column pivoting)
3321: Run with the option -info to determine an optimal value to use
3322: .ve
3324: Level: developer
3326: Notes:
3327: Most users should employ the `KSP` interface for linear solvers
3328: instead of working directly with matrix algebra routines such as this.
3329: See, e.g., `KSPCreate()`.
3331: This changes the state of the matrix to a factored matrix; it cannot be used
3332: for example with `MatSetValues()` unless one first calls `MatSetUnfactored()`.
3334: This is really in-place only for dense matrices, the preferred approach is to use `MatGetFactor()`, `MatLUFactorSymbolic()`, and `MatLUFactorNumeric()`
3335: when not using `KSP`.
3337: Fortran Note:
3338: A valid (non-null) `info` argument must be provided
3340: .seealso: [](ch_matrices), [Matrix Factorization](sec_matfactor), `Mat`, `MatFactorType`, `MatLUFactorSymbolic()`, `MatLUFactorNumeric()`, `MatCholeskyFactor()`,
3341: `MatGetOrdering()`, `MatSetUnfactored()`, `MatFactorInfo`, `MatGetFactor()`
3342: @*/
3343: PetscErrorCode MatLUFactor(Mat mat, IS row, IS col, const MatFactorInfo *info)
3344: {
3345: MatFactorInfo tinfo;
3347: PetscFunctionBegin;
3351: if (info) PetscAssertPointer(info, 4);
3353: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3354: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
3355: MatCheckPreallocated(mat, 1);
3356: if (!info) {
3357: PetscCall(MatFactorInfoInitialize(&tinfo));
3358: info = &tinfo;
3359: }
3361: PetscCall(PetscLogEventBegin(MAT_LUFactor, mat, row, col, 0));
3362: PetscUseTypeMethod(mat, lufactor, row, col, info);
3363: PetscCall(PetscLogEventEnd(MAT_LUFactor, mat, row, col, 0));
3364: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
3365: PetscFunctionReturn(PETSC_SUCCESS);
3366: }
3368: /*@
3369: MatILUFactor - Performs in-place ILU factorization of matrix.
3371: Collective
3373: Input Parameters:
3374: + mat - the matrix
3375: . row - row permutation
3376: . col - column permutation
3377: - info - structure containing
3378: .vb
3379: levels - number of levels of fill.
3380: expected fill - as ratio of original fill.
3381: 1 or 0 - indicating force fill on diagonal (improves robustness for matrices
3382: missing diagonal entries)
3383: .ve
3385: Level: developer
3387: Notes:
3388: Most users should employ the `KSP` interface for linear solvers
3389: instead of working directly with matrix algebra routines such as this.
3390: See, e.g., `KSPCreate()`.
3392: Probably really in-place only when level of fill is zero, otherwise allocates
3393: new space to store factored matrix and deletes previous memory. The preferred approach is to use `MatGetFactor()`, `MatILUFactorSymbolic()`, and `MatLUFactorNumeric()`
3394: when not using `KSP`.
3396: Fortran Note:
3397: A valid (non-null) `info` argument must be provided
3399: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatILUFactorSymbolic()`, `MatLUFactorNumeric()`, `MatCholeskyFactor()`, `MatFactorInfo`
3400: @*/
3401: PetscErrorCode MatILUFactor(Mat mat, IS row, IS col, const MatFactorInfo *info)
3402: {
3403: PetscFunctionBegin;
3407: PetscAssertPointer(info, 4);
3409: PetscCheck(mat->rmap->N == mat->cmap->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONG, "matrix must be square");
3410: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3411: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
3412: MatCheckPreallocated(mat, 1);
3414: PetscCall(PetscLogEventBegin(MAT_ILUFactor, mat, row, col, 0));
3415: PetscUseTypeMethod(mat, ilufactor, row, col, info);
3416: PetscCall(PetscLogEventEnd(MAT_ILUFactor, mat, row, col, 0));
3417: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
3418: PetscFunctionReturn(PETSC_SUCCESS);
3419: }
3421: /*@
3422: MatLUFactorSymbolic - Performs symbolic LU factorization of matrix.
3423: Call this routine before calling `MatLUFactorNumeric()` and after `MatGetFactor()`.
3425: Collective
3427: Input Parameters:
3428: + fact - the factor matrix obtained with `MatGetFactor()`
3429: . mat - the matrix
3430: . row - the row permutation
3431: . col - the column permutation
3432: - info - options for factorization, includes
3433: .vb
3434: fill - expected fill as ratio of original fill. Run with the option -info to determine an optimal value to use
3435: dtcol - pivot tolerance (0 no pivot, 1 full column pivoting)
3436: .ve
3438: Level: developer
3440: Notes:
3441: See [Matrix Factorization](sec_matfactor) for additional information about factorizations
3443: Most users should employ the simplified `KSP` interface for linear solvers
3444: instead of working directly with matrix algebra routines such as this.
3445: See, e.g., `KSPCreate()`.
3447: Fortran Note:
3448: A valid (non-null) `info` argument must be provided
3450: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatLUFactor()`, `MatLUFactorNumeric()`, `MatCholeskyFactor()`, `MatFactorInfo`, `MatFactorInfoInitialize()`
3451: @*/
3452: PetscErrorCode MatLUFactorSymbolic(Mat fact, Mat mat, IS row, IS col, const MatFactorInfo *info)
3453: {
3454: MatFactorInfo tinfo;
3456: PetscFunctionBegin;
3461: if (info) PetscAssertPointer(info, 5);
3464: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3465: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
3466: MatCheckPreallocated(mat, 2);
3467: if (!info) {
3468: PetscCall(MatFactorInfoInitialize(&tinfo));
3469: info = &tinfo;
3470: }
3472: if (!fact->trivialsymbolic) PetscCall(PetscLogEventBegin(MAT_LUFactorSymbolic, mat, row, col, 0));
3473: PetscUseTypeMethod(fact, lufactorsymbolic, mat, row, col, info);
3474: if (!fact->trivialsymbolic) PetscCall(PetscLogEventEnd(MAT_LUFactorSymbolic, mat, row, col, 0));
3475: PetscCall(PetscObjectStateIncrease((PetscObject)fact));
3476: PetscFunctionReturn(PETSC_SUCCESS);
3477: }
3479: /*@
3480: MatLUFactorNumeric - Performs numeric LU factorization of a matrix.
3481: Call this routine after first calling `MatLUFactorSymbolic()` and `MatGetFactor()`.
3483: Collective
3485: Input Parameters:
3486: + fact - the factor matrix obtained with `MatGetFactor()`
3487: . mat - the matrix
3488: - info - options for factorization
3490: Level: developer
3492: Notes:
3493: See `MatLUFactor()` for in-place factorization. See
3494: `MatCholeskyFactorNumeric()` for the symmetric, positive definite case.
3496: Most users should employ the `KSP` interface for linear solvers
3497: instead of working directly with matrix algebra routines such as this.
3498: See, e.g., `KSPCreate()`.
3500: Fortran Note:
3501: A valid (non-null) `info` argument must be provided
3503: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatFactorInfo`, `MatLUFactorSymbolic()`, `MatLUFactor()`, `MatCholeskyFactor()`
3504: @*/
3505: PetscErrorCode MatLUFactorNumeric(Mat fact, Mat mat, const MatFactorInfo *info)
3506: {
3507: MatFactorInfo tinfo;
3509: PetscFunctionBegin;
3514: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3515: PetscCheck(mat->rmap->N == (fact)->rmap->N && mat->cmap->N == (fact)->cmap->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Mat fact: global dimensions are different %" PetscInt_FMT " should = %" PetscInt_FMT " %" PetscInt_FMT " should = %" PetscInt_FMT,
3516: mat->rmap->N, (fact)->rmap->N, mat->cmap->N, (fact)->cmap->N);
3518: MatCheckPreallocated(mat, 2);
3519: if (!info) {
3520: PetscCall(MatFactorInfoInitialize(&tinfo));
3521: info = &tinfo;
3522: }
3524: if (!fact->trivialsymbolic) PetscCall(PetscLogEventBegin(MAT_LUFactorNumeric, mat, fact, 0, 0));
3525: else PetscCall(PetscLogEventBegin(MAT_LUFactor, mat, fact, 0, 0));
3526: PetscUseTypeMethod(fact, lufactornumeric, mat, info);
3527: if (!fact->trivialsymbolic) PetscCall(PetscLogEventEnd(MAT_LUFactorNumeric, mat, fact, 0, 0));
3528: else PetscCall(PetscLogEventEnd(MAT_LUFactor, mat, fact, 0, 0));
3529: PetscCall(MatViewFromOptions(fact, NULL, "-mat_factor_view"));
3530: PetscCall(PetscObjectStateIncrease((PetscObject)fact));
3531: PetscFunctionReturn(PETSC_SUCCESS);
3532: }
3534: /*@
3535: MatCholeskyFactor - Performs in-place Cholesky factorization of a
3536: symmetric matrix.
3538: Collective
3540: Input Parameters:
3541: + mat - the matrix
3542: . perm - row and column permutations
3543: - info - expected fill as ratio of original fill
3545: Level: developer
3547: Notes:
3548: See `MatLUFactor()` for the nonsymmetric case. See also `MatGetFactor()`,
3549: `MatCholeskyFactorSymbolic()`, and `MatCholeskyFactorNumeric()`.
3551: Most users should employ the `KSP` interface for linear solvers
3552: instead of working directly with matrix algebra routines such as this.
3553: See, e.g., `KSPCreate()`.
3555: Fortran Note:
3556: A valid (non-null) `info` argument must be provided
3558: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatFactorInfo`, `MatLUFactor()`, `MatCholeskyFactorSymbolic()`, `MatCholeskyFactorNumeric()`,
3559: `MatGetOrdering()`
3560: @*/
3561: PetscErrorCode MatCholeskyFactor(Mat mat, IS perm, const MatFactorInfo *info)
3562: {
3563: MatFactorInfo tinfo;
3565: PetscFunctionBegin;
3568: if (info) PetscAssertPointer(info, 3);
3570: PetscCheck(mat->rmap->N == mat->cmap->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONG, "Matrix must be square");
3571: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3572: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
3573: MatCheckPreallocated(mat, 1);
3574: if (!info) {
3575: PetscCall(MatFactorInfoInitialize(&tinfo));
3576: info = &tinfo;
3577: }
3579: PetscCall(PetscLogEventBegin(MAT_CholeskyFactor, mat, perm, 0, 0));
3580: PetscUseTypeMethod(mat, choleskyfactor, perm, info);
3581: PetscCall(PetscLogEventEnd(MAT_CholeskyFactor, mat, perm, 0, 0));
3582: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
3583: PetscFunctionReturn(PETSC_SUCCESS);
3584: }
3586: /*@
3587: MatCholeskyFactorSymbolic - Performs symbolic Cholesky factorization
3588: of a symmetric matrix.
3590: Collective
3592: Input Parameters:
3593: + fact - the factor matrix obtained with `MatGetFactor()`
3594: . mat - the matrix
3595: . perm - row and column permutations
3596: - info - options for factorization, includes
3597: .vb
3598: fill - expected fill as ratio of original fill.
3599: dtcol - pivot tolerance (0 no pivot, 1 full column pivoting)
3600: Run with the option -info to determine an optimal value to use
3601: .ve
3603: Level: developer
3605: Notes:
3606: See `MatLUFactorSymbolic()` for the nonsymmetric case. See also
3607: `MatCholeskyFactor()` and `MatCholeskyFactorNumeric()`.
3609: Most users should employ the `KSP` interface for linear solvers
3610: instead of working directly with matrix algebra routines such as this.
3611: See, e.g., `KSPCreate()`.
3613: Fortran Note:
3614: A valid (non-null) `info` argument must be provided
3616: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatFactorInfo`, `MatGetFactor()`, `MatLUFactorSymbolic()`, `MatCholeskyFactor()`, `MatCholeskyFactorNumeric()`,
3617: `MatGetOrdering()`
3618: @*/
3619: PetscErrorCode MatCholeskyFactorSymbolic(Mat fact, Mat mat, IS perm, const MatFactorInfo *info)
3620: {
3621: MatFactorInfo tinfo;
3623: PetscFunctionBegin;
3627: if (info) PetscAssertPointer(info, 4);
3630: PetscCheck(mat->rmap->N == mat->cmap->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONG, "Matrix must be square");
3631: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3632: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
3633: MatCheckPreallocated(mat, 2);
3634: if (!info) {
3635: PetscCall(MatFactorInfoInitialize(&tinfo));
3636: info = &tinfo;
3637: }
3639: if (!fact->trivialsymbolic) PetscCall(PetscLogEventBegin(MAT_CholeskyFactorSymbolic, mat, perm, 0, 0));
3640: PetscUseTypeMethod(fact, choleskyfactorsymbolic, mat, perm, info);
3641: if (!fact->trivialsymbolic) PetscCall(PetscLogEventEnd(MAT_CholeskyFactorSymbolic, mat, perm, 0, 0));
3642: PetscCall(PetscObjectStateIncrease((PetscObject)fact));
3643: PetscFunctionReturn(PETSC_SUCCESS);
3644: }
3646: /*@
3647: MatCholeskyFactorNumeric - Performs numeric Cholesky factorization
3648: of a symmetric matrix. Call this routine after first calling `MatGetFactor()` and
3649: `MatCholeskyFactorSymbolic()`.
3651: Collective
3653: Input Parameters:
3654: + fact - the factor matrix obtained with `MatGetFactor()`, where the factored values are stored
3655: . mat - the initial matrix that is to be factored
3656: - info - options for factorization
3658: Level: developer
3660: Note:
3661: Most users should employ the `KSP` interface for linear solvers
3662: instead of working directly with matrix algebra routines such as this.
3663: See, e.g., `KSPCreate()`.
3665: Fortran Note:
3666: A valid (non-null) `info` argument must be provided
3668: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatFactorInfo`, `MatGetFactor()`, `MatCholeskyFactorSymbolic()`, `MatCholeskyFactor()`, `MatLUFactorNumeric()`
3669: @*/
3670: PetscErrorCode MatCholeskyFactorNumeric(Mat fact, Mat mat, const MatFactorInfo *info)
3671: {
3672: MatFactorInfo tinfo;
3674: PetscFunctionBegin;
3679: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3680: PetscCheck(mat->rmap->N == (fact)->rmap->N && mat->cmap->N == (fact)->cmap->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Mat fact: global dim %" PetscInt_FMT " should = %" PetscInt_FMT " %" PetscInt_FMT " should = %" PetscInt_FMT,
3681: mat->rmap->N, (fact)->rmap->N, mat->cmap->N, (fact)->cmap->N);
3682: MatCheckPreallocated(mat, 2);
3683: if (!info) {
3684: PetscCall(MatFactorInfoInitialize(&tinfo));
3685: info = &tinfo;
3686: }
3688: if (!fact->trivialsymbolic) PetscCall(PetscLogEventBegin(MAT_CholeskyFactorNumeric, mat, fact, 0, 0));
3689: else PetscCall(PetscLogEventBegin(MAT_CholeskyFactor, mat, fact, 0, 0));
3690: PetscUseTypeMethod(fact, choleskyfactornumeric, mat, info);
3691: if (!fact->trivialsymbolic) PetscCall(PetscLogEventEnd(MAT_CholeskyFactorNumeric, mat, fact, 0, 0));
3692: else PetscCall(PetscLogEventEnd(MAT_CholeskyFactor, mat, fact, 0, 0));
3693: PetscCall(MatViewFromOptions(fact, NULL, "-mat_factor_view"));
3694: PetscCall(PetscObjectStateIncrease((PetscObject)fact));
3695: PetscFunctionReturn(PETSC_SUCCESS);
3696: }
3698: /*@
3699: MatQRFactor - Performs in-place QR factorization of matrix.
3701: Collective
3703: Input Parameters:
3704: + mat - the matrix
3705: . col - column permutation
3706: - info - options for factorization, includes
3707: .vb
3708: fill - expected fill as ratio of original fill.
3709: dtcol - pivot tolerance (0 no pivot, 1 full column pivoting)
3710: Run with the option -info to determine an optimal value to use
3711: .ve
3713: Level: developer
3715: Notes:
3716: Most users should employ the `KSP` interface for linear solvers
3717: instead of working directly with matrix algebra routines such as this.
3718: See, e.g., `KSPCreate()`.
3720: This changes the state of the matrix to a factored matrix; it cannot be used
3721: for example with `MatSetValues()` unless one first calls `MatSetUnfactored()`.
3723: Fortran Note:
3724: A valid (non-null) `info` argument must be provided
3726: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatFactorInfo`, `MatGetFactor()`, `MatQRFactorSymbolic()`, `MatQRFactorNumeric()`, `MatLUFactor()`,
3727: `MatSetUnfactored()`
3728: @*/
3729: PetscErrorCode MatQRFactor(Mat mat, IS col, const MatFactorInfo *info)
3730: {
3731: PetscFunctionBegin;
3734: if (info) PetscAssertPointer(info, 3);
3736: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3737: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
3738: MatCheckPreallocated(mat, 1);
3739: PetscCall(PetscLogEventBegin(MAT_QRFactor, mat, col, 0, 0));
3740: PetscUseMethod(mat, "MatQRFactor_C", (Mat, IS, const MatFactorInfo *), (mat, col, info));
3741: PetscCall(PetscLogEventEnd(MAT_QRFactor, mat, col, 0, 0));
3742: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
3743: PetscFunctionReturn(PETSC_SUCCESS);
3744: }
3746: /*@
3747: MatQRFactorSymbolic - Performs symbolic QR factorization of matrix.
3748: Call this routine after `MatGetFactor()` but before calling `MatQRFactorNumeric()`.
3750: Collective
3752: Input Parameters:
3753: + fact - the factor matrix obtained with `MatGetFactor()`
3754: . mat - the matrix
3755: . col - column permutation
3756: - info - options for factorization, includes
3757: .vb
3758: fill - expected fill as ratio of original fill.
3759: dtcol - pivot tolerance (0 no pivot, 1 full column pivoting)
3760: Run with the option -info to determine an optimal value to use
3761: .ve
3763: Level: developer
3765: Note:
3766: Most users should employ the `KSP` interface for linear solvers
3767: instead of working directly with matrix algebra routines such as this.
3768: See, e.g., `KSPCreate()`.
3770: Fortran Note:
3771: A valid (non-null) `info` argument must be provided
3773: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatFactorInfo`, `MatQRFactor()`, `MatQRFactorNumeric()`, `MatLUFactor()`, `MatFactorInfoInitialize()`
3774: @*/
3775: PetscErrorCode MatQRFactorSymbolic(Mat fact, Mat mat, IS col, const MatFactorInfo *info)
3776: {
3777: MatFactorInfo tinfo;
3779: PetscFunctionBegin;
3783: if (info) PetscAssertPointer(info, 4);
3786: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3787: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
3788: MatCheckPreallocated(mat, 2);
3789: if (!info) {
3790: PetscCall(MatFactorInfoInitialize(&tinfo));
3791: info = &tinfo;
3792: }
3794: if (!fact->trivialsymbolic) PetscCall(PetscLogEventBegin(MAT_QRFactorSymbolic, fact, mat, col, 0));
3795: PetscUseMethod(fact, "MatQRFactorSymbolic_C", (Mat, Mat, IS, const MatFactorInfo *), (fact, mat, col, info));
3796: if (!fact->trivialsymbolic) PetscCall(PetscLogEventEnd(MAT_QRFactorSymbolic, fact, mat, col, 0));
3797: PetscCall(PetscObjectStateIncrease((PetscObject)fact));
3798: PetscFunctionReturn(PETSC_SUCCESS);
3799: }
3801: /*@
3802: MatQRFactorNumeric - Performs numeric QR factorization of a matrix.
3803: Call this routine after first calling `MatGetFactor()`, and `MatQRFactorSymbolic()`.
3805: Collective
3807: Input Parameters:
3808: + fact - the factor matrix obtained with `MatGetFactor()`
3809: . mat - the matrix
3810: - info - options for factorization
3812: Level: developer
3814: Notes:
3815: See `MatQRFactor()` for in-place factorization.
3817: Most users should employ the `KSP` interface for linear solvers
3818: instead of working directly with matrix algebra routines such as this.
3819: See, e.g., `KSPCreate()`.
3821: Fortran Note:
3822: A valid (non-null) `info` argument must be provided
3824: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatFactorInfo`, `MatGetFactor()`, `MatQRFactor()`, `MatQRFactorSymbolic()`, `MatLUFactor()`
3825: @*/
3826: PetscErrorCode MatQRFactorNumeric(Mat fact, Mat mat, const MatFactorInfo *info)
3827: {
3828: MatFactorInfo tinfo;
3830: PetscFunctionBegin;
3835: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
3836: PetscCheck(mat->rmap->N == fact->rmap->N && mat->cmap->N == fact->cmap->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Mat fact: global dimensions are different %" PetscInt_FMT " should = %" PetscInt_FMT " %" PetscInt_FMT " should = %" PetscInt_FMT,
3837: mat->rmap->N, (fact)->rmap->N, mat->cmap->N, (fact)->cmap->N);
3839: MatCheckPreallocated(mat, 2);
3840: if (!info) {
3841: PetscCall(MatFactorInfoInitialize(&tinfo));
3842: info = &tinfo;
3843: }
3845: if (!fact->trivialsymbolic) PetscCall(PetscLogEventBegin(MAT_QRFactorNumeric, mat, fact, 0, 0));
3846: else PetscCall(PetscLogEventBegin(MAT_QRFactor, mat, fact, 0, 0));
3847: PetscUseMethod(fact, "MatQRFactorNumeric_C", (Mat, Mat, const MatFactorInfo *), (fact, mat, info));
3848: if (!fact->trivialsymbolic) PetscCall(PetscLogEventEnd(MAT_QRFactorNumeric, mat, fact, 0, 0));
3849: else PetscCall(PetscLogEventEnd(MAT_QRFactor, mat, fact, 0, 0));
3850: PetscCall(MatViewFromOptions(fact, NULL, "-mat_factor_view"));
3851: PetscCall(PetscObjectStateIncrease((PetscObject)fact));
3852: PetscFunctionReturn(PETSC_SUCCESS);
3853: }
3855: /*@
3856: MatSolve - Solves $A x = b$, given a factored matrix.
3858: Neighbor-wise Collective
3860: Input Parameters:
3861: + mat - the factored matrix
3862: - b - the right-hand-side vector
3864: Output Parameter:
3865: . x - the result vector
3867: Level: developer
3869: Notes:
3870: The vectors `b` and `x` cannot be the same. I.e., one cannot
3871: call `MatSolve`(A,x,x).
3873: Most users should employ the `KSP` interface for linear solvers
3874: instead of working directly with matrix algebra routines such as this.
3875: See, e.g., `KSPCreate()`.
3877: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatLUFactor()`, `MatSolveAdd()`, `MatSolveTranspose()`, `MatSolveTransposeAdd()`
3878: @*/
3879: PetscErrorCode MatSolve(Mat mat, Vec b, Vec x)
3880: {
3881: PetscFunctionBegin;
3886: PetscCheckSameComm(mat, 1, b, 2);
3887: PetscCheckSameComm(mat, 1, x, 3);
3888: PetscCheck(x != b, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "x and b must be different vectors");
3889: PetscCheck(mat->cmap->N == x->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, x->map->N);
3890: PetscCheck(mat->rmap->N == b->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec b: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->N, b->map->N);
3891: PetscCheck(mat->rmap->n == b->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec b: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->n, b->map->n);
3892: if (!mat->rmap->N && !mat->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
3893: MatCheckPreallocated(mat, 1);
3895: PetscCall(PetscLogEventBegin(MAT_Solve, mat, b, x, 0));
3896: PetscCall(VecFlag(x, mat->factorerrortype));
3897: if (mat->factorerrortype) PetscCall(PetscInfo(mat, "MatFactorError %d\n", mat->factorerrortype));
3898: else PetscUseTypeMethod(mat, solve, b, x);
3899: PetscCall(PetscLogEventEnd(MAT_Solve, mat, b, x, 0));
3900: PetscCall(PetscObjectStateIncrease((PetscObject)x));
3901: PetscFunctionReturn(PETSC_SUCCESS);
3902: }
3904: static PetscErrorCode MatMatSolve_Basic(Mat A, Mat B, Mat X, PetscBool trans)
3905: {
3906: Vec b, x;
3907: PetscInt N;
3908: PetscErrorCode (*f)(Mat, Vec, Vec);
3909: PetscBool Abound, Bneedconv = PETSC_FALSE, Xneedconv = PETSC_FALSE;
3911: PetscFunctionBegin;
3912: f = (!trans || (!A->ops->solvetranspose && A->symmetric)) ? A->ops->solve : A->ops->solvetranspose;
3913: PetscCheck(f, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Mat type %s", ((PetscObject)A)->type_name);
3914: PetscCall(MatBoundToCPU(A, &Abound));
3915: if (!Abound) {
3916: PetscCall(PetscObjectTypeCompareAny((PetscObject)B, &Bneedconv, MATSEQDENSE, MATMPIDENSE, ""));
3917: PetscCall(PetscObjectTypeCompareAny((PetscObject)X, &Xneedconv, MATSEQDENSE, MATMPIDENSE, ""));
3918: }
3919: #if PetscDefined(HAVE_CUDA)
3920: if (Bneedconv) PetscCall(MatConvert(B, MATDENSECUDA, MAT_INPLACE_MATRIX, &B));
3921: if (Xneedconv) PetscCall(MatConvert(X, MATDENSECUDA, MAT_INPLACE_MATRIX, &X));
3922: #elif PetscDefined(HAVE_HIP)
3923: if (Bneedconv) PetscCall(MatConvert(B, MATDENSEHIP, MAT_INPLACE_MATRIX, &B));
3924: if (Xneedconv) PetscCall(MatConvert(X, MATDENSEHIP, MAT_INPLACE_MATRIX, &X));
3925: #endif
3926: PetscCall(MatGetSize(B, NULL, &N));
3927: for (PetscInt i = 0; i < N; i++) {
3928: PetscCall(MatDenseGetColumnVecRead(B, i, &b));
3929: PetscCall(MatDenseGetColumnVecWrite(X, i, &x));
3930: PetscCall((*f)(A, b, x));
3931: PetscCall(MatDenseRestoreColumnVecWrite(X, i, &x));
3932: PetscCall(MatDenseRestoreColumnVecRead(B, i, &b));
3933: }
3934: if (Bneedconv) PetscCall(MatConvert(B, MATDENSE, MAT_INPLACE_MATRIX, &B));
3935: if (Xneedconv) PetscCall(MatConvert(X, MATDENSE, MAT_INPLACE_MATRIX, &X));
3936: PetscFunctionReturn(PETSC_SUCCESS);
3937: }
3939: /*@
3940: MatMatSolve - Solves $A X = B$, given a factored matrix.
3942: Neighbor-wise Collective
3944: Input Parameters:
3945: + A - the factored matrix
3946: - B - the right-hand-side matrix `MATDENSE` (or sparse `MATAIJ`-- when using MUMPS)
3948: Output Parameter:
3949: . X - the result matrix (dense matrix)
3951: Level: developer
3953: Note:
3954: If `B` is a `MATDENSE` matrix then one can call `MatMatSolve`(A,B,B) except with `MATSOLVERMKL_CPARDISO`;
3955: otherwise, `B` and `X` cannot be the same.
3957: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatSolve()`, `MatMatSolveTranspose()`, `MatLUFactor()`, `MatCholeskyFactor()`
3958: @*/
3959: PetscErrorCode MatMatSolve(Mat A, Mat B, Mat X)
3960: {
3961: PetscFunctionBegin;
3966: PetscCheckSameComm(A, 1, B, 2);
3967: PetscCheckSameComm(A, 1, X, 3);
3968: PetscCheck(A->cmap->N == X->rmap->N, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_SIZ, "Mat A,Mat X: global dim %" PetscInt_FMT " %" PetscInt_FMT, A->cmap->N, X->rmap->N);
3969: PetscCheck(A->rmap->N == B->rmap->N, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_SIZ, "Mat A,Mat B: global dim %" PetscInt_FMT " %" PetscInt_FMT, A->rmap->N, B->rmap->N);
3970: PetscCheck(X->cmap->N == B->cmap->N, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_SIZ, "Solution matrix must have same number of columns as rhs matrix");
3971: if (!A->rmap->N && !A->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
3972: MatCheckPreallocated(A, 1);
3974: PetscCall(PetscLogEventBegin(MAT_MatSolve, A, B, X, 0));
3975: if (A->factorerrortype) {
3976: PetscCall(PetscInfo(A, "MatFactorError %d\n", A->factorerrortype));
3977: PetscCall(MatFlag(X, 1));
3978: } else if (!A->ops->matsolve) {
3979: PetscCall(PetscInfo(A, "Mat type %s using basic MatMatSolve\n", ((PetscObject)A)->type_name));
3980: PetscCall(MatMatSolve_Basic(A, B, X, PETSC_FALSE));
3981: } else PetscUseTypeMethod(A, matsolve, B, X);
3982: PetscCall(PetscLogEventEnd(MAT_MatSolve, A, B, X, 0));
3983: PetscCall(PetscObjectStateIncrease((PetscObject)X));
3984: PetscFunctionReturn(PETSC_SUCCESS);
3985: }
3987: /*@
3988: MatMatSolveTranspose - Solves $A^T X = B $, given a factored matrix.
3990: Neighbor-wise Collective
3992: Input Parameters:
3993: + A - the factored matrix
3994: - B - the right-hand-side matrix (`MATDENSE` matrix)
3996: Output Parameter:
3997: . X - the result matrix (dense matrix)
3999: Level: developer
4001: Note:
4002: The matrices `B` and `X` cannot be the same. I.e., one cannot
4003: call `MatMatSolveTranspose`(A,X,X).
4005: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatSolveTranspose()`, `MatMatSolve()`, `MatLUFactor()`, `MatCholeskyFactor()`
4006: @*/
4007: PetscErrorCode MatMatSolveTranspose(Mat A, Mat B, Mat X)
4008: {
4009: PetscFunctionBegin;
4014: PetscCheckSameComm(A, 1, B, 2);
4015: PetscCheckSameComm(A, 1, X, 3);
4016: PetscCheck(X != B, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_IDN, "X and B must be different matrices");
4017: PetscCheck(A->cmap->N == X->rmap->N, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_SIZ, "Mat A,Mat X: global dim %" PetscInt_FMT " %" PetscInt_FMT, A->cmap->N, X->rmap->N);
4018: PetscCheck(A->rmap->N == B->rmap->N, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_SIZ, "Mat A,Mat B: global dim %" PetscInt_FMT " %" PetscInt_FMT, A->rmap->N, B->rmap->N);
4019: PetscCheck(A->rmap->n == B->rmap->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat A,Mat B: local dim %" PetscInt_FMT " %" PetscInt_FMT, A->rmap->n, B->rmap->n);
4020: PetscCheck(X->cmap->N >= B->cmap->N, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Solution matrix must have same number of columns as rhs matrix");
4021: if (!A->rmap->N && !A->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
4022: MatCheckPreallocated(A, 1);
4024: PetscCall(PetscLogEventBegin(MAT_MatSolve, A, B, X, 0));
4025: if (A->factorerrortype) {
4026: PetscCall(PetscInfo(A, "MatFactorError %d\n", A->factorerrortype));
4027: PetscCall(MatFlag(X, 1));
4028: } else if (!A->ops->matsolvetranspose) {
4029: PetscCall(PetscInfo(A, "Mat type %s using basic MatMatSolveTranspose\n", ((PetscObject)A)->type_name));
4030: PetscCall(MatMatSolve_Basic(A, B, X, PETSC_TRUE));
4031: } else PetscUseTypeMethod(A, matsolvetranspose, B, X);
4032: PetscCall(PetscLogEventEnd(MAT_MatSolve, A, B, X, 0));
4033: PetscCall(PetscObjectStateIncrease((PetscObject)X));
4034: PetscFunctionReturn(PETSC_SUCCESS);
4035: }
4037: /*@
4038: MatMatTransposeSolve - Solves $A X = B^T$, given a factored matrix.
4040: Neighbor-wise Collective
4042: Input Parameters:
4043: + A - the factored matrix
4044: - Bt - the transpose of right-hand-side matrix as a `MATDENSE`
4046: Output Parameter:
4047: . X - the result matrix (dense matrix)
4049: Level: developer
4051: Note:
4052: For MUMPS, it only supports centralized sparse compressed column format on the host process for right-hand side matrix. User must create `Bt` in sparse compressed row
4053: format on the host process and call `MatMatTransposeSolve()` to implement MUMPS' `MatMatSolve()`.
4055: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatMatSolve()`, `MatMatSolveTranspose()`, `MatLUFactor()`, `MatCholeskyFactor()`
4056: @*/
4057: PetscErrorCode MatMatTransposeSolve(Mat A, Mat Bt, Mat X)
4058: {
4059: PetscFunctionBegin;
4064: PetscCheckSameComm(A, 1, Bt, 2);
4065: PetscCheckSameComm(A, 1, X, 3);
4067: PetscCheck(X != Bt, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_IDN, "X and B must be different matrices");
4068: PetscCheck(A->cmap->N == X->rmap->N, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_SIZ, "Mat A,Mat X: global dim %" PetscInt_FMT " %" PetscInt_FMT, A->cmap->N, X->rmap->N);
4069: PetscCheck(A->rmap->N == Bt->cmap->N, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_SIZ, "Mat A,Mat Bt: global dim %" PetscInt_FMT " %" PetscInt_FMT, A->rmap->N, Bt->cmap->N);
4070: PetscCheck(X->cmap->N >= Bt->rmap->N, PetscObjectComm((PetscObject)X), PETSC_ERR_ARG_SIZ, "Solution matrix must have same number of columns as row number of the rhs matrix");
4071: if (!A->rmap->N && !A->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
4072: PetscCheck(A->factortype, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Unfactored matrix");
4073: MatCheckPreallocated(A, 1);
4075: PetscCall(PetscLogEventBegin(MAT_MatTrSolve, A, Bt, X, 0));
4076: if (A->factorerrortype) {
4077: PetscCall(PetscInfo(A, "MatFactorError %d\n", A->factorerrortype));
4078: PetscCall(MatFlag(X, 1));
4079: } else PetscUseTypeMethod(A, mattransposesolve, Bt, X);
4080: PetscCall(PetscLogEventEnd(MAT_MatTrSolve, A, Bt, X, 0));
4081: PetscCall(PetscObjectStateIncrease((PetscObject)X));
4082: PetscFunctionReturn(PETSC_SUCCESS);
4083: }
4085: /*@
4086: MatForwardSolve - Solves $ L x = b $, given a factored matrix, $A = LU $, or
4087: $U^T*D^(1/2) x = b$, given a factored symmetric matrix, $A = U^T*D*U$,
4089: Neighbor-wise Collective
4091: Input Parameters:
4092: + mat - the factored matrix
4093: - b - the right-hand-side vector
4095: Output Parameter:
4096: . x - the result vector
4098: Level: developer
4100: Notes:
4101: `MatSolve()` should be used for most applications, as it performs
4102: a forward solve followed by a backward solve.
4104: The vectors `b` and `x` cannot be the same, i.e., one cannot
4105: call `MatForwardSolve`(A,x,x).
4107: For matrix in `MATSEQBAIJ` format with block size larger than 1,
4108: the diagonal blocks are not implemented as $D = D^(1/2) * D^(1/2)$ yet.
4109: `MatForwardSolve()` solves $U^T*D y = b$, and
4110: `MatBackwardSolve()` solves $U x = y$.
4111: Thus they do not provide a symmetric preconditioner.
4113: .seealso: [](ch_matrices), `Mat`, `MatBackwardSolve()`, `MatGetFactor()`, `MatSolve()`
4114: @*/
4115: PetscErrorCode MatForwardSolve(Mat mat, Vec b, Vec x)
4116: {
4117: PetscFunctionBegin;
4122: PetscCheckSameComm(mat, 1, b, 2);
4123: PetscCheckSameComm(mat, 1, x, 3);
4124: PetscCheck(x != b, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "x and b must be different vectors");
4125: PetscCheck(mat->cmap->N == x->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, x->map->N);
4126: PetscCheck(mat->rmap->N == b->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec b: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->N, b->map->N);
4127: PetscCheck(mat->rmap->n == b->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec b: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->n, b->map->n);
4128: if (!mat->rmap->N && !mat->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
4129: MatCheckPreallocated(mat, 1);
4131: PetscCall(PetscLogEventBegin(MAT_ForwardSolve, mat, b, x, 0));
4132: PetscUseTypeMethod(mat, forwardsolve, b, x);
4133: PetscCall(PetscLogEventEnd(MAT_ForwardSolve, mat, b, x, 0));
4134: PetscCall(PetscObjectStateIncrease((PetscObject)x));
4135: PetscFunctionReturn(PETSC_SUCCESS);
4136: }
4138: /*@
4139: MatBackwardSolve - Solves $U x = b$, given a factored matrix, $A = LU$.
4140: $D^(1/2) U x = b$, given a factored symmetric matrix, $A = U^T*D*U$,
4142: Neighbor-wise Collective
4144: Input Parameters:
4145: + mat - the factored matrix
4146: - b - the right-hand-side vector
4148: Output Parameter:
4149: . x - the result vector
4151: Level: developer
4153: Notes:
4154: `MatSolve()` should be used for most applications, as it performs
4155: a forward solve followed by a backward solve.
4157: The vectors `b` and `x` cannot be the same. I.e., one cannot
4158: call `MatBackwardSolve`(A,x,x).
4160: For matrix in `MATSEQBAIJ` format with block size larger than 1,
4161: the diagonal blocks are not implemented as $D = D^(1/2) * D^(1/2)$ yet.
4162: `MatForwardSolve()` solves $U^T*D y = b$, and
4163: `MatBackwardSolve()` solves $U x = y$.
4164: Thus they do not provide a symmetric preconditioner.
4166: .seealso: [](ch_matrices), `Mat`, `MatForwardSolve()`, `MatGetFactor()`, `MatSolve()`
4167: @*/
4168: PetscErrorCode MatBackwardSolve(Mat mat, Vec b, Vec x)
4169: {
4170: PetscFunctionBegin;
4175: PetscCheckSameComm(mat, 1, b, 2);
4176: PetscCheckSameComm(mat, 1, x, 3);
4177: PetscCheck(x != b, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "x and b must be different vectors");
4178: PetscCheck(mat->cmap->N == x->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, x->map->N);
4179: PetscCheck(mat->rmap->N == b->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec b: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->N, b->map->N);
4180: PetscCheck(mat->rmap->n == b->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec b: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->n, b->map->n);
4181: if (!mat->rmap->N && !mat->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
4182: MatCheckPreallocated(mat, 1);
4184: PetscCall(PetscLogEventBegin(MAT_BackwardSolve, mat, b, x, 0));
4185: PetscUseTypeMethod(mat, backwardsolve, b, x);
4186: PetscCall(PetscLogEventEnd(MAT_BackwardSolve, mat, b, x, 0));
4187: PetscCall(PetscObjectStateIncrease((PetscObject)x));
4188: PetscFunctionReturn(PETSC_SUCCESS);
4189: }
4191: /*@
4192: MatSolveAdd - Computes $x = y + A^{-1}*b$, given a factored matrix.
4194: Neighbor-wise Collective
4196: Input Parameters:
4197: + mat - the factored matrix
4198: . b - the right-hand-side vector
4199: - y - the vector to be added to
4201: Output Parameter:
4202: . x - the result vector
4204: Level: developer
4206: Note:
4207: The vectors `b` and `x` cannot be the same. I.e., one cannot
4208: call `MatSolveAdd`(A,x,y,x).
4210: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatSolve()`, `MatGetFactor()`, `MatSolveTranspose()`, `MatSolveTransposeAdd()`
4211: @*/
4212: PetscErrorCode MatSolveAdd(Mat mat, Vec b, Vec y, Vec x)
4213: {
4214: PetscScalar one = 1.0;
4215: Vec tmp;
4217: PetscFunctionBegin;
4223: PetscCheckSameComm(mat, 1, b, 2);
4224: PetscCheckSameComm(mat, 1, y, 3);
4225: PetscCheckSameComm(mat, 1, x, 4);
4226: PetscCheck(x != b, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "x and b must be different vectors");
4227: PetscCheck(mat->cmap->N == x->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, x->map->N);
4228: PetscCheck(mat->rmap->N == b->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec b: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->N, b->map->N);
4229: PetscCheck(mat->rmap->N == y->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec y: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->N, y->map->N);
4230: PetscCheck(mat->rmap->n == b->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec b: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->n, b->map->n);
4231: PetscCheck(x->map->n == y->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Vec x,Vec y: local dim %" PetscInt_FMT " %" PetscInt_FMT, x->map->n, y->map->n);
4232: if (!mat->rmap->N && !mat->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
4233: MatCheckPreallocated(mat, 1);
4235: PetscCall(PetscLogEventBegin(MAT_SolveAdd, mat, b, x, y));
4236: PetscCall(VecFlag(x, mat->factorerrortype));
4237: if (mat->factorerrortype) {
4238: PetscCall(PetscInfo(mat, "MatFactorError %d\n", mat->factorerrortype));
4239: } else if (mat->ops->solveadd) {
4240: PetscUseTypeMethod(mat, solveadd, b, y, x);
4241: } else {
4242: /* do the solve then the add manually */
4243: if (x != y) {
4244: PetscCall(MatSolve(mat, b, x));
4245: PetscCall(VecAXPY(x, one, y));
4246: } else {
4247: PetscCall(VecDuplicate(x, &tmp));
4248: PetscCall(VecCopy(x, tmp));
4249: PetscCall(MatSolve(mat, b, x));
4250: PetscCall(VecAXPY(x, one, tmp));
4251: PetscCall(VecDestroy(&tmp));
4252: }
4253: }
4254: PetscCall(PetscLogEventEnd(MAT_SolveAdd, mat, b, x, y));
4255: PetscCall(PetscObjectStateIncrease((PetscObject)x));
4256: PetscFunctionReturn(PETSC_SUCCESS);
4257: }
4259: /*@
4260: MatSolveTranspose - Solves $A^T x = b$, given a factored matrix.
4262: Neighbor-wise Collective
4264: Input Parameters:
4265: + mat - the factored matrix
4266: - b - the right-hand-side vector
4268: Output Parameter:
4269: . x - the result vector
4271: Level: developer
4273: Notes:
4274: The vectors `b` and `x` cannot be the same. I.e., one cannot
4275: call `MatSolveTranspose`(A,x,x).
4277: Most users should employ the `KSP` interface for linear solvers
4278: instead of working directly with matrix algebra routines such as this.
4279: See, e.g., `KSPCreate()`.
4281: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `KSP`, `MatSolve()`, `MatSolveAdd()`, `MatSolveTransposeAdd()`
4282: @*/
4283: PetscErrorCode MatSolveTranspose(Mat mat, Vec b, Vec x)
4284: {
4285: PetscErrorCode (*f)(Mat, Vec, Vec) = (!mat->ops->solvetranspose && mat->symmetric) ? mat->ops->solve : mat->ops->solvetranspose;
4287: PetscFunctionBegin;
4292: PetscCheckSameComm(mat, 1, b, 2);
4293: PetscCheckSameComm(mat, 1, x, 3);
4294: PetscCheck(x != b, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "x and b must be different vectors");
4295: PetscCheck(mat->rmap->N == x->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->N, x->map->N);
4296: PetscCheck(mat->cmap->N == b->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec b: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, b->map->N);
4297: if (!mat->rmap->N && !mat->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
4298: MatCheckPreallocated(mat, 1);
4299: PetscCall(PetscLogEventBegin(MAT_SolveTranspose, mat, b, x, 0));
4300: PetscCall(VecFlag(x, mat->factorerrortype));
4301: if (mat->factorerrortype) {
4302: PetscCall(PetscInfo(mat, "MatFactorError %d\n", mat->factorerrortype));
4303: } else {
4304: PetscCheck(f, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Matrix type %s", ((PetscObject)mat)->type_name);
4305: PetscCall((*f)(mat, b, x));
4306: }
4307: PetscCall(PetscLogEventEnd(MAT_SolveTranspose, mat, b, x, 0));
4308: PetscCall(PetscObjectStateIncrease((PetscObject)x));
4309: PetscFunctionReturn(PETSC_SUCCESS);
4310: }
4312: /*@
4313: MatSolveTransposeAdd - Computes $x = y + A^{-T} b$
4314: factored matrix.
4316: Neighbor-wise Collective
4318: Input Parameters:
4319: + mat - the factored matrix
4320: . b - the right-hand-side vector
4321: - y - the vector to be added to
4323: Output Parameter:
4324: . x - the result vector
4326: Level: developer
4328: Note:
4329: The vectors `b` and `x` cannot be the same. I.e., one cannot
4330: call `MatSolveTransposeAdd`(A,x,y,x).
4332: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatSolve()`, `MatSolveAdd()`, `MatSolveTranspose()`
4333: @*/
4334: PetscErrorCode MatSolveTransposeAdd(Mat mat, Vec b, Vec y, Vec x)
4335: {
4336: PetscScalar one = 1.0;
4337: Vec tmp;
4338: PetscErrorCode (*f)(Mat, Vec, Vec, Vec) = (!mat->ops->solvetransposeadd && mat->symmetric) ? mat->ops->solveadd : mat->ops->solvetransposeadd;
4340: PetscFunctionBegin;
4346: PetscCheckSameComm(mat, 1, b, 2);
4347: PetscCheckSameComm(mat, 1, y, 3);
4348: PetscCheckSameComm(mat, 1, x, 4);
4349: PetscCheck(x != b, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "x and b must be different vectors");
4350: PetscCheck(mat->rmap->N == x->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->N, x->map->N);
4351: PetscCheck(mat->cmap->N == b->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec b: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, b->map->N);
4352: PetscCheck(mat->cmap->N == y->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec y: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, y->map->N);
4353: PetscCheck(x->map->n == y->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Vec x,Vec y: local dim %" PetscInt_FMT " %" PetscInt_FMT, x->map->n, y->map->n);
4354: if (!mat->rmap->N && !mat->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
4355: MatCheckPreallocated(mat, 1);
4357: PetscCall(PetscLogEventBegin(MAT_SolveTransposeAdd, mat, b, x, y));
4358: PetscCall(VecFlag(x, mat->factorerrortype));
4359: if (mat->factorerrortype) {
4360: PetscCall(PetscInfo(mat, "MatFactorError %d\n", mat->factorerrortype));
4361: } else if (f) {
4362: PetscCall((*f)(mat, b, y, x));
4363: } else {
4364: /* do the solve then the add manually */
4365: if (x != y) {
4366: PetscCall(MatSolveTranspose(mat, b, x));
4367: PetscCall(VecAXPY(x, one, y));
4368: } else {
4369: PetscCall(VecDuplicate(x, &tmp));
4370: PetscCall(VecCopy(x, tmp));
4371: PetscCall(MatSolveTranspose(mat, b, x));
4372: PetscCall(VecAXPY(x, one, tmp));
4373: PetscCall(VecDestroy(&tmp));
4374: }
4375: }
4376: PetscCall(PetscLogEventEnd(MAT_SolveTransposeAdd, mat, b, x, y));
4377: PetscCall(PetscObjectStateIncrease((PetscObject)x));
4378: PetscFunctionReturn(PETSC_SUCCESS);
4379: }
4381: // PetscClangLinter pragma disable: -fdoc-section-header-unknown
4382: /*@
4383: MatSOR - Computes relaxation (SOR, Gauss-Seidel) sweeps.
4385: Neighbor-wise Collective
4387: Input Parameters:
4388: + mat - the matrix
4389: . b - the right-hand side
4390: . omega - the relaxation factor
4391: . flag - flag indicating the type of SOR (see below)
4392: . shift - diagonal shift
4393: . its - the number of iterations
4394: - lits - the number of local iterations
4396: Output Parameter:
4397: . x - the solution (can contain an initial guess, use option `SOR_ZERO_INITIAL_GUESS` to indicate no guess)
4399: SOR Flags:
4400: + `SOR_FORWARD_SWEEP` - forward SOR
4401: . `SOR_BACKWARD_SWEEP` - backward SOR
4402: . `SOR_SYMMETRIC_SWEEP` - SSOR (symmetric SOR)
4403: . `SOR_LOCAL_FORWARD_SWEEP` - local forward SOR
4404: . `SOR_LOCAL_BACKWARD_SWEEP` - local forward SOR
4405: . `SOR_LOCAL_SYMMETRIC_SWEEP` - local SSOR
4406: . `SOR_EISENSTAT` - SOR with Eisenstat trick
4407: . `SOR_APPLY_UPPER`, `SOR_APPLY_LOWER` - applies upper/lower triangular part of matrix to vector (with `omega`)
4408: - `SOR_ZERO_INITIAL_GUESS` - zero initial guess
4410: Level: developer
4412: Notes:
4413: `SOR_LOCAL_FORWARD_SWEEP`, `SOR_LOCAL_BACKWARD_SWEEP`, and
4414: `SOR_LOCAL_SYMMETRIC_SWEEP` perform separate independent smoothings
4415: on each process.
4417: Application programmers will not generally use `MatSOR()` directly,
4418: but instead will employ `PCSOR` or `PCEISENSTAT`
4420: For `MATBAIJ`, `MATSBAIJ`, and `MATAIJ` matrices with inodes, this does a block SOR smoothing, otherwise it does a pointwise smoothing.
4421: For `MATAIJ` matrices with inodes, the block sizes are determined by the inode sizes, not the block size set with `MatSetBlockSize()`
4423: Vectors `x` and `b` CANNOT be the same
4425: The flags are implemented as bitwise inclusive or operations.
4426: For example, use (`SOR_ZERO_INITIAL_GUESS` | `SOR_SYMMETRIC_SWEEP`)
4427: to specify a zero initial guess for SSOR.
4429: Developer Note:
4430: We should add block SOR support for `MATAIJ` matrices with block size set to greater than one and no inodes
4432: .seealso: [](ch_matrices), `Mat`, `MatMult()`, `KSP`, `PC`, `MatGetFactor()`
4433: @*/
4434: PetscErrorCode MatSOR(Mat mat, Vec b, PetscReal omega, MatSORType flag, PetscReal shift, PetscInt its, PetscInt lits, Vec x)
4435: {
4436: PetscFunctionBegin;
4441: PetscCheckSameComm(mat, 1, b, 2);
4442: PetscCheckSameComm(mat, 1, x, 8);
4443: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
4444: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
4445: PetscCheck(mat->cmap->N == x->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec x: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->cmap->N, x->map->N);
4446: PetscCheck(mat->rmap->N == b->map->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_SIZ, "Mat mat,Vec b: global dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->N, b->map->N);
4447: PetscCheck(mat->rmap->n == b->map->n, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Mat mat,Vec b: local dim %" PetscInt_FMT " %" PetscInt_FMT, mat->rmap->n, b->map->n);
4448: PetscCheck(its > 0, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Relaxation requires global its %" PetscInt_FMT " positive", its);
4449: PetscCheck(lits > 0, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "Relaxation requires local its %" PetscInt_FMT " positive", lits);
4450: PetscCheck(b != x, PETSC_COMM_SELF, PETSC_ERR_ARG_IDN, "b and x vector cannot be the same");
4452: MatCheckPreallocated(mat, 1);
4453: PetscCall(PetscLogEventBegin(MAT_SOR, mat, b, x, 0));
4454: PetscUseTypeMethod(mat, sor, b, omega, flag, shift, its, lits, x);
4455: PetscCall(PetscLogEventEnd(MAT_SOR, mat, b, x, 0));
4456: PetscCall(PetscObjectStateIncrease((PetscObject)x));
4457: PetscFunctionReturn(PETSC_SUCCESS);
4458: }
4460: /*
4461: Default matrix copy routine.
4462: */
4463: PetscErrorCode MatCopy_Basic(Mat A, Mat B, MatStructure str)
4464: {
4465: PetscInt i, rstart = 0, rend = 0, nz;
4466: const PetscInt *cwork;
4467: const PetscScalar *vwork;
4469: PetscFunctionBegin;
4470: if (B->assembled) PetscCall(MatZeroEntries(B));
4471: if (str == SAME_NONZERO_PATTERN) {
4472: PetscCall(MatGetOwnershipRange(A, &rstart, &rend));
4473: for (i = rstart; i < rend; i++) {
4474: PetscCall(MatGetRow(A, i, &nz, &cwork, &vwork));
4475: PetscCall(MatSetValues(B, 1, &i, nz, cwork, vwork, INSERT_VALUES));
4476: PetscCall(MatRestoreRow(A, i, &nz, &cwork, &vwork));
4477: }
4478: } else {
4479: PetscCall(MatAYPX(B, 0.0, A, str));
4480: }
4481: PetscCall(MatAssemblyBegin(B, MAT_FINAL_ASSEMBLY));
4482: PetscCall(MatAssemblyEnd(B, MAT_FINAL_ASSEMBLY));
4483: PetscFunctionReturn(PETSC_SUCCESS);
4484: }
4486: /*@
4487: MatCopy - Copies a matrix to another matrix.
4489: Collective
4491: Input Parameters:
4492: + A - the matrix
4493: - str - `SAME_NONZERO_PATTERN` or `DIFFERENT_NONZERO_PATTERN`
4495: Output Parameter:
4496: . B - where the copy is put
4498: Level: intermediate
4500: Notes:
4501: If you use `SAME_NONZERO_PATTERN`, then the two matrices must have the same nonzero pattern or the routine will crash.
4503: `MatCopy()` copies the matrix entries of a matrix to another existing
4504: matrix (after first zeroing the second matrix). A related routine is
4505: `MatConvert()`, which first creates a new matrix and then copies the data.
4507: .seealso: [](ch_matrices), `Mat`, `MatConvert()`, `MatDuplicate()`
4508: @*/
4509: PetscErrorCode MatCopy(Mat A, Mat B, MatStructure str)
4510: {
4511: PetscInt i;
4513: PetscFunctionBegin;
4518: PetscCheckSameComm(A, 1, B, 2);
4519: MatCheckPreallocated(B, 2);
4520: PetscCheck(A->assembled, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
4521: PetscCheck(!A->factortype, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
4522: PetscCheck(A->rmap->N == B->rmap->N && A->cmap->N == B->cmap->N, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_SIZ, "Mat A,Mat B: global dim (%" PetscInt_FMT ",%" PetscInt_FMT ") (%" PetscInt_FMT ",%" PetscInt_FMT ")", A->rmap->N, B->rmap->N,
4523: A->cmap->N, B->cmap->N);
4524: MatCheckPreallocated(A, 1);
4525: if (A == B) PetscFunctionReturn(PETSC_SUCCESS);
4527: PetscCall(PetscLogEventBegin(MAT_Copy, A, B, 0, 0));
4528: if (A->ops->copy) PetscUseTypeMethod(A, copy, B, str);
4529: else PetscCall(MatCopy_Basic(A, B, str));
4531: B->stencil.dim = A->stencil.dim;
4532: B->stencil.noc = A->stencil.noc;
4533: for (i = 0; i <= A->stencil.dim + (A->stencil.noc ? 0 : -1); i++) {
4534: B->stencil.dims[i] = A->stencil.dims[i];
4535: B->stencil.starts[i] = A->stencil.starts[i];
4536: }
4538: PetscCall(PetscLogEventEnd(MAT_Copy, A, B, 0, 0));
4539: PetscCall(PetscObjectStateIncrease((PetscObject)B));
4540: PetscFunctionReturn(PETSC_SUCCESS);
4541: }
4543: /*@
4544: MatConvert - Converts a matrix to another matrix, either of the same
4545: or different type.
4547: Collective
4549: Input Parameters:
4550: + mat - the matrix
4551: . newtype - new matrix type. Use `MATSAME` to create a new matrix of the
4552: same type as the original matrix.
4553: - reuse - denotes if the destination matrix is to be created or reused.
4554: Use `MAT_INPLACE_MATRIX` for inplace conversion (that is when you want the input `Mat` to be changed to contain the matrix in the new format), otherwise use
4555: `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX` (can only be used after the first call was made with `MAT_INITIAL_MATRIX`, causes the matrix space in M to be reused).
4557: Output Parameter:
4558: . M - pointer to place new matrix
4560: Level: intermediate
4562: Notes:
4563: `MatConvert()` first creates a new matrix and then copies the data from
4564: the first matrix. A related routine is `MatCopy()`, which copies the matrix
4565: entries of one matrix to another already existing matrix context.
4567: Cannot be used to convert a sequential matrix to parallel or parallel to sequential,
4568: the MPI communicator of the generated matrix is always the same as the communicator
4569: of the input matrix.
4571: .seealso: [](ch_matrices), `Mat`, `MatCopy()`, `MatDuplicate()`, `MAT_INITIAL_MATRIX`, `MAT_REUSE_MATRIX`, `MAT_INPLACE_MATRIX`
4572: @*/
4573: PetscErrorCode MatConvert(Mat mat, MatType newtype, MatReuse reuse, Mat *M)
4574: {
4575: PetscBool sametype, issame, flg;
4576: PetscBool3 issymmetric, ishermitian, isspd;
4577: char convname[256], mtype[256];
4578: Mat B;
4580: PetscFunctionBegin;
4583: PetscAssertPointer(M, 4);
4584: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
4585: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
4586: MatCheckPreallocated(mat, 1);
4588: PetscCall(PetscOptionsGetString(((PetscObject)mat)->options, ((PetscObject)mat)->prefix, "-matconvert_type", mtype, sizeof(mtype), &flg));
4589: if (flg) newtype = mtype;
4591: PetscCall(PetscObjectTypeCompare((PetscObject)mat, newtype, &sametype));
4592: PetscCall(PetscStrcmp(newtype, "same", &issame));
4593: PetscCheck(!(reuse == MAT_INPLACE_MATRIX) || !(mat != *M), PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "MAT_INPLACE_MATRIX requires same input and output matrix");
4594: if (reuse == MAT_REUSE_MATRIX) {
4596: PetscCheck(mat != *M, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "MAT_REUSE_MATRIX means reuse matrix in final argument, perhaps you mean MAT_INPLACE_MATRIX");
4597: }
4599: if ((reuse == MAT_INPLACE_MATRIX) && (issame || sametype)) {
4600: PetscCall(PetscInfo(mat, "Early return for inplace %s %d %d\n", ((PetscObject)mat)->type_name, sametype, issame));
4601: PetscFunctionReturn(PETSC_SUCCESS);
4602: }
4604: /* Cache Mat options because some converters use MatHeaderReplace() */
4605: issymmetric = mat->symmetric;
4606: ishermitian = mat->hermitian;
4607: isspd = mat->spd;
4609: if ((sametype || issame) && (reuse == MAT_INITIAL_MATRIX) && mat->ops->duplicate) {
4610: PetscCall(PetscInfo(mat, "Calling duplicate for initial matrix %s %d %d\n", ((PetscObject)mat)->type_name, sametype, issame));
4611: PetscUseTypeMethod(mat, duplicate, MAT_COPY_VALUES, M);
4612: } else {
4613: PetscErrorCode (*conv)(Mat, MatType, MatReuse, Mat *) = NULL;
4614: const char *prefix[3] = {"seq", "mpi", ""};
4615: PetscInt i;
4616: /*
4617: Order of precedence:
4618: 0) See if newtype is a superclass of the current matrix.
4619: 1) See if a specialized converter is known to the current matrix.
4620: 2) See if a specialized converter is known to the desired matrix class.
4621: 3) See if a good general converter is registered for the desired class
4622: (as of 6/27/03 only MATMPIADJ falls into this category).
4623: 4) See if a good general converter is known for the current matrix.
4624: 5) Use a really basic converter.
4625: */
4627: /* 0) See if newtype is a superclass of the current matrix.
4628: i.e mat is mpiaij and newtype is aij */
4629: for (i = 0; i < (PetscInt)PETSC_STATIC_ARRAY_LENGTH(prefix); i++) {
4630: PetscCall(PetscStrncpy(convname, prefix[i], sizeof(convname)));
4631: PetscCall(PetscStrlcat(convname, newtype, sizeof(convname)));
4632: PetscCall(PetscStrcmp(convname, ((PetscObject)mat)->type_name, &flg));
4633: PetscCall(PetscInfo(mat, "Check superclass %s %s -> %d\n", convname, ((PetscObject)mat)->type_name, flg));
4634: if (flg) {
4635: if (reuse == MAT_INPLACE_MATRIX) {
4636: PetscCall(PetscInfo(mat, "Early return\n"));
4637: PetscFunctionReturn(PETSC_SUCCESS);
4638: } else if (reuse == MAT_INITIAL_MATRIX && mat->ops->duplicate) {
4639: PetscCall(PetscInfo(mat, "Calling MatDuplicate\n"));
4640: PetscUseTypeMethod(mat, duplicate, MAT_COPY_VALUES, M);
4641: PetscFunctionReturn(PETSC_SUCCESS);
4642: } else if (reuse == MAT_REUSE_MATRIX && mat->ops->copy) {
4643: PetscCall(PetscInfo(mat, "Calling MatCopy\n"));
4644: PetscCall(MatCopy(mat, *M, SAME_NONZERO_PATTERN));
4645: PetscFunctionReturn(PETSC_SUCCESS);
4646: }
4647: }
4648: }
4649: /* 1) See if a specialized converter is known to the current matrix and the desired class */
4650: for (i = 0; i < (PetscInt)PETSC_STATIC_ARRAY_LENGTH(prefix); i++) {
4651: PetscCall(PetscStrncpy(convname, "MatConvert_", sizeof(convname)));
4652: PetscCall(PetscStrlcat(convname, ((PetscObject)mat)->type_name, sizeof(convname)));
4653: PetscCall(PetscStrlcat(convname, "_", sizeof(convname)));
4654: PetscCall(PetscStrlcat(convname, prefix[i], sizeof(convname)));
4655: PetscCall(PetscStrlcat(convname, issame ? ((PetscObject)mat)->type_name : newtype, sizeof(convname)));
4656: PetscCall(PetscStrlcat(convname, "_C", sizeof(convname)));
4657: PetscCall(PetscObjectQueryFunction((PetscObject)mat, convname, &conv));
4658: PetscCall(PetscInfo(mat, "Check specialized (1) %s (%s) -> %d\n", convname, ((PetscObject)mat)->type_name, !!conv));
4659: if (conv) goto foundconv;
4660: }
4662: /* 2) See if a specialized converter is known to the desired matrix class. */
4663: PetscCall(MatCreate(PetscObjectComm((PetscObject)mat), &B));
4664: PetscCall(MatSetSizes(B, mat->rmap->n, mat->cmap->n, mat->rmap->N, mat->cmap->N));
4665: PetscCall(MatSetType(B, newtype));
4666: for (i = 0; i < (PetscInt)PETSC_STATIC_ARRAY_LENGTH(prefix); i++) {
4667: PetscCall(PetscStrncpy(convname, "MatConvert_", sizeof(convname)));
4668: PetscCall(PetscStrlcat(convname, ((PetscObject)mat)->type_name, sizeof(convname)));
4669: PetscCall(PetscStrlcat(convname, "_", sizeof(convname)));
4670: PetscCall(PetscStrlcat(convname, prefix[i], sizeof(convname)));
4671: PetscCall(PetscStrlcat(convname, newtype, sizeof(convname)));
4672: PetscCall(PetscStrlcat(convname, "_C", sizeof(convname)));
4673: PetscCall(PetscObjectQueryFunction((PetscObject)B, convname, &conv));
4674: PetscCall(PetscInfo(mat, "Check specialized (2) %s (%s) -> %d\n", convname, ((PetscObject)B)->type_name, !!conv));
4675: if (conv) {
4676: PetscCall(MatDestroy(&B));
4677: goto foundconv;
4678: }
4679: }
4681: /* 3) See if a good general converter is registered for the desired class */
4682: conv = B->ops->convertfrom;
4683: PetscCall(PetscInfo(mat, "Check convertfrom (%s) -> %d\n", ((PetscObject)B)->type_name, !!conv));
4684: PetscCall(MatDestroy(&B));
4685: if (conv) goto foundconv;
4687: /* 4) See if a good general converter is known for the current matrix */
4688: if (mat->ops->convert) conv = mat->ops->convert;
4689: PetscCall(PetscInfo(mat, "Check general convert (%s) -> %d\n", ((PetscObject)mat)->type_name, !!conv));
4690: if (conv) goto foundconv;
4692: /* 5) Use a really basic converter. */
4693: PetscCall(PetscInfo(mat, "Using MatConvert_Basic\n"));
4694: conv = MatConvert_Basic;
4696: foundconv:
4697: PetscCall(PetscLogEventBegin(MAT_Convert, mat, 0, 0, 0));
4698: PetscCall((*conv)(mat, newtype, reuse, M));
4699: if (mat->rmap->mapping && mat->cmap->mapping && !(*M)->rmap->mapping && !(*M)->cmap->mapping) {
4700: /* the block sizes must be same if the mappings are copied over */
4701: (*M)->rmap->bs = mat->rmap->bs;
4702: (*M)->cmap->bs = mat->cmap->bs;
4703: PetscCall(PetscObjectReference((PetscObject)mat->rmap->mapping));
4704: PetscCall(PetscObjectReference((PetscObject)mat->cmap->mapping));
4705: (*M)->rmap->mapping = mat->rmap->mapping;
4706: (*M)->cmap->mapping = mat->cmap->mapping;
4707: }
4708: (*M)->stencil.dim = mat->stencil.dim;
4709: (*M)->stencil.noc = mat->stencil.noc;
4710: for (i = 0; i <= mat->stencil.dim + (mat->stencil.noc ? 0 : -1); i++) {
4711: (*M)->stencil.dims[i] = mat->stencil.dims[i];
4712: (*M)->stencil.starts[i] = mat->stencil.starts[i];
4713: }
4714: PetscCall(PetscLogEventEnd(MAT_Convert, mat, 0, 0, 0));
4715: }
4716: PetscCall(PetscObjectStateIncrease((PetscObject)*M));
4718: /* Reset Mat options */
4719: if (issymmetric != PETSC_BOOL3_UNKNOWN) PetscCall(MatSetOption(*M, MAT_SYMMETRIC, PetscBool3ToBool(issymmetric)));
4720: if (ishermitian != PETSC_BOOL3_UNKNOWN) PetscCall(MatSetOption(*M, MAT_HERMITIAN, PetscBool3ToBool(ishermitian)));
4721: if (isspd != PETSC_BOOL3_UNKNOWN) PetscCall(MatSetOption(*M, MAT_SPD, PetscBool3ToBool(isspd)));
4722: PetscFunctionReturn(PETSC_SUCCESS);
4723: }
4725: /*@
4726: MatFactorGetSolverType - Returns name of the package providing the factorization routines
4728: Not Collective
4730: Input Parameter:
4731: . mat - the matrix, must be a factored matrix
4733: Output Parameter:
4734: . type - the string name of the package (do not free this string)
4736: Level: intermediate
4738: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatSolverType`, `MatCopy()`, `MatDuplicate()`, `MatGetFactorAvailable()`
4739: @*/
4740: PetscErrorCode MatFactorGetSolverType(Mat mat, MatSolverType *type)
4741: {
4742: PetscErrorCode (*conv)(Mat, MatSolverType *);
4744: PetscFunctionBegin;
4747: PetscAssertPointer(type, 2);
4748: PetscCheck(mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Only for factored matrix");
4749: PetscCall(PetscObjectQueryFunction((PetscObject)mat, "MatFactorGetSolverType_C", &conv));
4750: if (conv) PetscCall((*conv)(mat, type));
4751: else *type = MATSOLVERPETSC;
4752: PetscFunctionReturn(PETSC_SUCCESS);
4753: }
4755: typedef struct _MatSolverTypeForSpecifcType *MatSolverTypeForSpecifcType;
4756: struct _MatSolverTypeForSpecifcType {
4757: MatType mtype;
4758: /* no entry for MAT_FACTOR_NONE */
4759: PetscErrorCode (*createfactor[MAT_FACTOR_NUM_TYPES - 1])(Mat, MatFactorType, Mat *);
4760: MatSolverTypeForSpecifcType next;
4761: };
4763: typedef struct _MatSolverTypeHolder *MatSolverTypeHolder;
4764: struct _MatSolverTypeHolder {
4765: char *name;
4766: MatSolverTypeForSpecifcType handlers;
4767: MatSolverTypeHolder next;
4768: };
4770: static MatSolverTypeHolder MatSolverTypeHolders = NULL;
4772: /*@
4773: MatSolverTypeRegister - Registers a `MatSolverType` that works for a particular matrix type
4775: Logically Collective, No Fortran Support
4777: Input Parameters:
4778: + package - name of the package, for example `petsc` or `superlu`
4779: . mtype - the matrix type that works with this package
4780: . ftype - the type of factorization supported by the package
4781: - createfactor - routine that will create the factored matrix ready to be used
4783: Level: developer
4785: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatFactorGetSolverType()`, `MatCopy()`, `MatDuplicate()`, `MatGetFactorAvailable()`,
4786: `MatGetFactor()`
4787: @*/
4788: PetscErrorCode MatSolverTypeRegister(MatSolverType package, MatType mtype, MatFactorType ftype, PetscErrorCode (*createfactor)(Mat, MatFactorType, Mat *))
4789: {
4790: MatSolverTypeHolder next = MatSolverTypeHolders, prev = NULL;
4791: PetscBool flg;
4792: MatSolverTypeForSpecifcType inext, iprev = NULL;
4794: PetscFunctionBegin;
4795: PetscCall(MatInitializePackage());
4796: if (!next) {
4797: PetscCall(PetscNew(&MatSolverTypeHolders));
4798: PetscCall(PetscStrallocpy(package, &MatSolverTypeHolders->name));
4799: PetscCall(PetscNew(&MatSolverTypeHolders->handlers));
4800: PetscCall(PetscStrallocpy(mtype, (char **)&MatSolverTypeHolders->handlers->mtype));
4801: MatSolverTypeHolders->handlers->createfactor[(int)ftype - 1] = createfactor;
4802: PetscFunctionReturn(PETSC_SUCCESS);
4803: }
4804: while (next) {
4805: PetscCall(PetscStrcasecmp(package, next->name, &flg));
4806: if (flg) {
4807: PetscCheck(next->handlers, PETSC_COMM_SELF, PETSC_ERR_PLIB, "MatSolverTypeHolder is missing handlers");
4808: inext = next->handlers;
4809: while (inext) {
4810: PetscCall(PetscStrcasecmp(mtype, inext->mtype, &flg));
4811: if (flg) {
4812: inext->createfactor[(int)ftype - 1] = createfactor;
4813: PetscFunctionReturn(PETSC_SUCCESS);
4814: }
4815: iprev = inext;
4816: inext = inext->next;
4817: }
4818: PetscCall(PetscNew(&iprev->next));
4819: PetscCall(PetscStrallocpy(mtype, (char **)&iprev->next->mtype));
4820: iprev->next->createfactor[(int)ftype - 1] = createfactor;
4821: PetscFunctionReturn(PETSC_SUCCESS);
4822: }
4823: prev = next;
4824: next = next->next;
4825: }
4826: PetscCall(PetscNew(&prev->next));
4827: PetscCall(PetscStrallocpy(package, &prev->next->name));
4828: PetscCall(PetscNew(&prev->next->handlers));
4829: PetscCall(PetscStrallocpy(mtype, (char **)&prev->next->handlers->mtype));
4830: prev->next->handlers->createfactor[(int)ftype - 1] = createfactor;
4831: PetscFunctionReturn(PETSC_SUCCESS);
4832: }
4834: /*@
4835: MatSolverTypeGet - Gets the function that creates the factor matrix if it exist
4837: Input Parameters:
4838: + type - name of the package, for example `petsc` or `superlu`, if this is `NULL`, then the first result that satisfies the other criteria is returned
4839: . ftype - the type of factorization supported by the type
4840: - mtype - the matrix type that works with this type
4842: Output Parameters:
4843: + foundtype - `PETSC_TRUE` if the type was registered
4844: . foundmtype - `PETSC_TRUE` if the type supports the requested mtype
4845: - createfactor - routine that will create the factored matrix ready to be used or `NULL` if not found
4847: Calling sequence of `createfactor`:
4848: + A - the matrix providing the factor matrix
4849: . ftype - the `MatFactorType` of the factor requested
4850: - B - the new factor matrix that responds to MatXXFactorSymbolic,Numeric() functions, such as `MatLUFactorSymbolic()`
4852: Level: developer
4854: Note:
4855: When `type` is `NULL` the available functions are searched for based on the order of the calls to `MatSolverTypeRegister()` in `MatInitializePackage()`.
4856: Since different PETSc configurations may have different external solvers, seemingly identical runs with different PETSc configurations may use a different solver.
4857: For example if one configuration had `--download-mumps` while a different one had `--download-superlu_dist`.
4859: .seealso: [](ch_matrices), `Mat`, `MatFactorType`, `MatType`, `MatCopy()`, `MatDuplicate()`, `MatGetFactorAvailable()`, `MatSolverTypeRegister()`, `MatGetFactor()`,
4860: `MatInitializePackage()`
4861: @*/
4862: PetscErrorCode MatSolverTypeGet(MatSolverType type, MatType mtype, MatFactorType ftype, PetscBool *foundtype, PetscBool *foundmtype, PetscErrorCode (**createfactor)(Mat A, MatFactorType ftype, Mat *B))
4863: {
4864: MatSolverTypeHolder next = MatSolverTypeHolders;
4865: PetscBool flg;
4866: MatSolverTypeForSpecifcType inext;
4868: PetscFunctionBegin;
4869: if (foundtype) *foundtype = PETSC_FALSE;
4870: if (foundmtype) *foundmtype = PETSC_FALSE;
4871: if (createfactor) *createfactor = NULL;
4873: if (type) {
4874: while (next) {
4875: PetscCall(PetscStrcasecmp(type, next->name, &flg));
4876: if (flg) {
4877: if (foundtype) *foundtype = PETSC_TRUE;
4878: inext = next->handlers;
4879: while (inext) {
4880: PetscCall(PetscStrbeginswith(mtype, inext->mtype, &flg));
4881: if (flg) {
4882: if (foundmtype) *foundmtype = PETSC_TRUE;
4883: if (createfactor) *createfactor = inext->createfactor[(int)ftype - 1];
4884: PetscFunctionReturn(PETSC_SUCCESS);
4885: }
4886: inext = inext->next;
4887: }
4888: }
4889: next = next->next;
4890: }
4891: } else {
4892: while (next) {
4893: inext = next->handlers;
4894: while (inext) {
4895: PetscCall(PetscStrcmp(mtype, inext->mtype, &flg));
4896: if (flg && inext->createfactor[(int)ftype - 1]) {
4897: if (foundtype) *foundtype = PETSC_TRUE;
4898: if (foundmtype) *foundmtype = PETSC_TRUE;
4899: if (createfactor) *createfactor = inext->createfactor[(int)ftype - 1];
4900: PetscFunctionReturn(PETSC_SUCCESS);
4901: }
4902: inext = inext->next;
4903: }
4904: next = next->next;
4905: }
4906: /* try with base classes inext->mtype */
4907: next = MatSolverTypeHolders;
4908: while (next) {
4909: inext = next->handlers;
4910: while (inext) {
4911: PetscCall(PetscStrbeginswith(mtype, inext->mtype, &flg));
4912: if (flg && inext->createfactor[(int)ftype - 1]) {
4913: if (foundtype) *foundtype = PETSC_TRUE;
4914: if (foundmtype) *foundmtype = PETSC_TRUE;
4915: if (createfactor) *createfactor = inext->createfactor[(int)ftype - 1];
4916: PetscFunctionReturn(PETSC_SUCCESS);
4917: }
4918: inext = inext->next;
4919: }
4920: next = next->next;
4921: }
4922: }
4923: PetscFunctionReturn(PETSC_SUCCESS);
4924: }
4926: PetscErrorCode MatSolverTypeDestroy(void)
4927: {
4928: MatSolverTypeHolder next = MatSolverTypeHolders, prev;
4929: MatSolverTypeForSpecifcType inext, iprev;
4931: PetscFunctionBegin;
4932: while (next) {
4933: PetscCall(PetscFree(next->name));
4934: inext = next->handlers;
4935: while (inext) {
4936: PetscCall(PetscFree(inext->mtype));
4937: iprev = inext;
4938: inext = inext->next;
4939: PetscCall(PetscFree(iprev));
4940: }
4941: prev = next;
4942: next = next->next;
4943: PetscCall(PetscFree(prev));
4944: }
4945: MatSolverTypeHolders = NULL;
4946: PetscFunctionReturn(PETSC_SUCCESS);
4947: }
4949: static PetscErrorCode MatGetFactor_Private(Mat mat, MatFactorType ftype, PetscBool exact, PetscBool *found, Mat *f)
4950: {
4951: MatSolverTypeHolder next = MatSolverTypeHolders;
4952: MatSolverTypeForSpecifcType inext;
4953: PetscBool flg, same;
4955: PetscFunctionBegin;
4956: *found = PETSC_FALSE;
4957: *f = NULL;
4958: /* When no solver type is requested, MatGetFactor() must honor registration order, but a registered
4959: MatSolverType may only be able to reject a particular MatType at runtime by returning NULL in *f.
4960: Keep walking the registry until a matching backend actually creates a factor. */
4961: while (next) {
4962: inext = next->handlers;
4963: while (inext) {
4964: PetscCall(PetscStrcmp(((PetscObject)mat)->type_name, inext->mtype, &same));
4965: if (exact) flg = same;
4966: else {
4967: /* Do the base-type pass separately from the exact pass so exact registrations for the MatType
4968: are all tried before broader registrations such as implementation base classes. */
4969: PetscCall(PetscStrbeginswith(((PetscObject)mat)->type_name, inext->mtype, &flg));
4970: flg = (PetscBool)(flg && !same);
4971: }
4972: if (flg && inext->createfactor[(int)ftype - 1]) {
4973: *found = PETSC_TRUE;
4974: PetscCall((*inext->createfactor[(int)ftype - 1])(mat, ftype, f));
4975: if (*f) PetscFunctionReturn(PETSC_SUCCESS);
4976: }
4977: inext = inext->next;
4978: }
4979: next = next->next;
4980: }
4981: PetscFunctionReturn(PETSC_SUCCESS);
4982: }
4984: /*@
4985: MatFactorGetCanUseOrdering - Indicates if the factorization can use the ordering provided in `MatLUFactorSymbolic()`, `MatCholeskyFactorSymbolic()`
4987: Logically Collective
4989: Input Parameter:
4990: . mat - the matrix
4992: Output Parameter:
4993: . flg - `PETSC_TRUE` if uses the ordering
4995: Level: developer
4997: Note:
4998: Most internal PETSc factorizations use the ordering passed to the factorization routine but external
4999: packages do not, thus we want to skip generating the ordering when it is not needed or used.
5001: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatCopy()`, `MatDuplicate()`, `MatGetFactorAvailable()`, `MatGetFactor()`, `MatLUFactorSymbolic()`, `MatCholeskyFactorSymbolic()`
5002: @*/
5003: PetscErrorCode MatFactorGetCanUseOrdering(Mat mat, PetscBool *flg)
5004: {
5005: PetscFunctionBegin;
5006: *flg = mat->canuseordering;
5007: PetscFunctionReturn(PETSC_SUCCESS);
5008: }
5010: /*@
5011: MatFactorGetPreferredOrdering - The preferred ordering for a particular matrix factor object
5013: Logically Collective
5015: Input Parameters:
5016: + mat - the matrix obtained with `MatGetFactor()`
5017: - ftype - the factorization type to be used
5019: Output Parameter:
5020: . otype - the preferred ordering type
5022: Level: developer
5024: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatFactorType`, `MatOrderingType`, `MatCopy()`, `MatDuplicate()`, `MatGetFactorAvailable()`, `MatGetFactor()`, `MatLUFactorSymbolic()`, `MatCholeskyFactorSymbolic()`
5025: @*/
5026: PetscErrorCode MatFactorGetPreferredOrdering(Mat mat, MatFactorType ftype, MatOrderingType *otype)
5027: {
5028: PetscFunctionBegin;
5029: *otype = mat->preferredordering[ftype];
5030: PetscCheck(*otype, PETSC_COMM_SELF, PETSC_ERR_PLIB, "MatFactor did not have a preferred ordering");
5031: PetscFunctionReturn(PETSC_SUCCESS);
5032: }
5034: /*@
5035: MatGetFactor - Returns a matrix suitable to calls to routines such as `MatLUFactorSymbolic()`, `MatCholeskyFactorSymbolic()`, `MatILUFactorSymbolic()`,
5036: `MatICCFactorSymbolic()`, `MatLUFactorNumeric()`, and `MatCholeskyFactorNumeric()`
5038: Collective
5040: Input Parameters:
5041: + mat - the matrix
5042: . type - name of solver type, for example, `superlu_dist`, `petsc` (to use PETSc's solver if it is available), if this is `NULL`, then the first result that satisfies
5043: the other criteria is returned
5044: - ftype - factor type, `MAT_FACTOR_LU`, `MAT_FACTOR_CHOLESKY`, `MAT_FACTOR_ICC`, `MAT_FACTOR_ILU`, `MAT_FACTOR_QR`
5046: Output Parameter:
5047: . f - the factor matrix used with MatXXFactorSymbolic,Numeric() calls. Can be `NULL` in some cases, see notes below.
5049: Options Database Keys:
5050: + -pc_factor_mat_solver_type type - choose the type at run time. When using `KSP` solvers
5051: . -pc_factor_mat_factor_on_host (true|false) - do matrix factorization on host (with device matrices). Default is doing it on device
5052: - -pc_factor_mat_solve_on_host (true|false) - do matrix solve on host (with device matrices). Default is doing it on device
5054: Level: intermediate
5056: Notes:
5057: Some of the packages, such as MUMPS, have options for controlling the factorization, these are in the form `-prefix_mat_packagename_packageoption`
5058: (for example, `-mat_mumps_icntl_6 1`) where `prefix` is normally set automatically from the calling `KSP`/`PC`. If `MatGetFactor()` is called directly,
5059: without using a `PC`, one can set the prefix by
5060: calling `MatSetOptionsPrefixFactor()` on the originating matrix or `MatSetOptionsPrefix()` on the resulting factor matrix.
5062: Some PETSc matrix formats have alternative solvers available that are provided by alternative packages
5063: such as PaStiX, SuperLU_DIST, MUMPS etc. PETSc must have been configured to use the external solver,
5064: using the corresponding `./configure` option such as `--download-package` or `--with-package-dir`.
5066: When `type` is `NULL` the available results are searched for based on the order of the calls to `MatSolverTypeRegister()` in `MatInitializePackage()`.
5067: Since different PETSc configurations may have different external solvers, seemingly identical runs with different PETSc configurations may use a different solver.
5068: For example if one configuration had `--download-mumps` while a different one had `--download-superlu_dist`.
5070: The return matrix can be `NULL` if the requested factorization is not available, since some combinations of matrix types and factorization
5071: types registered with `MatSolverTypeRegister()` cannot be fully tested if not at runtime.
5073: Developer Note:
5074: This should actually be called `MatCreateFactor()` since it creates a new factor object
5076: The `MatGetFactor()` implementations should not be accessing the PETSc options database or making other decisions about solver options,
5077: that should be delayed until the later operations. This is to ensure the correct options prefix has been set in the factor matrix.
5079: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `KSP`, `MatSolverType`, `MatFactorType`, `MatCopy()`, `MatDuplicate()`,
5080: `MatGetFactorAvailable()`, `MatFactorGetCanUseOrdering()`, `MatSolverTypeRegister()`, `MatSolverTypeGet()`,
5081: `MAT_FACTOR_LU`, `MAT_FACTOR_CHOLESKY`, `MAT_FACTOR_ICC`, `MAT_FACTOR_ILU`, `MAT_FACTOR_QR`, `MatInitializePackage()`,
5082: `MatLUFactorSymbolic()`, `MatCholeskyFactorSymbolic()`, `MatILUFactorSymbolic()`,
5083: `MatICCFactorSymbolic()`, `MatLUFactorNumeric()`, `MatCholeskyFactorNumeric()`
5084: @*/
5085: PetscErrorCode MatGetFactor(Mat mat, MatSolverType type, MatFactorType ftype, Mat *f)
5086: {
5087: PetscBool foundtype, foundmtype, shell, hasop = PETSC_FALSE;
5088: PetscErrorCode (*conv)(Mat, MatFactorType, Mat *);
5090: PetscFunctionBegin;
5094: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
5095: MatCheckPreallocated(mat, 1);
5097: PetscCall(MatIsShell(mat, &shell));
5098: if (shell) PetscCall(MatHasOperation(mat, MATOP_GET_FACTOR, &hasop));
5099: if (hasop) {
5100: PetscUseTypeMethod(mat, getfactor, type, ftype, f);
5101: PetscFunctionReturn(PETSC_SUCCESS);
5102: }
5104: if (!type) {
5105: PetscBool foundbase;
5107: /* First try exact MatType registrations in solver registration order. If all matching backends
5108: decline this matrix instance by returning NULL, then try base-type registrations. */
5109: PetscCall(MatGetFactor_Private(mat, ftype, PETSC_TRUE, &foundtype, f));
5110: if (!*f) {
5111: PetscCall(MatGetFactor_Private(mat, ftype, PETSC_FALSE, &foundbase, f));
5112: foundtype = (PetscBool)(foundtype || foundbase);
5113: }
5114: PetscCheck(foundtype, PetscObjectComm((PetscObject)mat), PETSC_ERR_MISSING_FACTOR, "Could not locate a solver type for factorization type %s and matrix type %s.", MatFactorTypes[ftype], ((PetscObject)mat)->type_name);
5115: if (mat->factorprefix) PetscCall(MatSetOptionsPrefix(*f, mat->factorprefix));
5116: PetscFunctionReturn(PETSC_SUCCESS);
5117: }
5119: PetscCall(MatSolverTypeGet(type, ((PetscObject)mat)->type_name, ftype, &foundtype, &foundmtype, &conv));
5120: PetscCheck(foundtype, PetscObjectComm((PetscObject)mat), PETSC_ERR_MISSING_FACTOR, "Could not locate%s solver type%s%s for factorization type %s and matrix type %s.%s%s", !type ? " a" : "", type ? " " : "", type ? type : "", MatFactorTypes[ftype],
5121: ((PetscObject)mat)->type_name, type ? " Perhaps you must ./configure with --download-" : "", type ? type : "");
5122: PetscCheck(foundmtype, PetscObjectComm((PetscObject)mat), PETSC_ERR_MISSING_FACTOR, "MatSolverType %s does not support matrix type %s", type, ((PetscObject)mat)->type_name);
5123: PetscCheck(conv, PetscObjectComm((PetscObject)mat), PETSC_ERR_MISSING_FACTOR, "MatSolverType %s does not support factorization type %s for matrix type %s", type, MatFactorTypes[ftype], ((PetscObject)mat)->type_name);
5125: PetscCall((*conv)(mat, ftype, f));
5126: if (mat->factorprefix) PetscCall(MatSetOptionsPrefix(*f, mat->factorprefix));
5127: PetscFunctionReturn(PETSC_SUCCESS);
5128: }
5130: /*@
5131: MatGetFactorAvailable - Returns a flag if matrix supports particular type and factor type
5133: Not Collective
5135: Input Parameters:
5136: + mat - the matrix
5137: . type - name of solver type, for example, `superlu`, `petsc` (to use PETSc's default)
5138: - ftype - factor type, `MAT_FACTOR_LU`, `MAT_FACTOR_CHOLESKY`, `MAT_FACTOR_ICC`, `MAT_FACTOR_ILU`, `MAT_FACTOR_QR`
5140: Output Parameter:
5141: . flg - `PETSC_TRUE` if the factorization is available
5143: Level: intermediate
5145: Notes:
5146: Some PETSc matrix formats have alternative solvers available that are contained in alternative packages
5147: such as `pastix`, `superlu`, `mumps`, etc.
5149: PETSc must have been configured with `./configure` to use the external solver using the option `--download-package` where package is the name of the package
5151: Developer Note:
5152: This should actually be called `MatCreateFactorAvailable()` since `MatGetFactor()` creates a new factor object
5154: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatSolverType`, `MatFactorType`, `MatGetFactor()`, `MatCopy()`, `MatDuplicate()`, `MatSolverTypeRegister()`,
5155: `MAT_FACTOR_LU`, `MAT_FACTOR_CHOLESKY`, `MAT_FACTOR_ICC`, `MAT_FACTOR_ILU`, `MAT_FACTOR_QR`, `MatSolverTypeGet()`
5156: @*/
5157: PetscErrorCode MatGetFactorAvailable(Mat mat, MatSolverType type, MatFactorType ftype, PetscBool *flg)
5158: {
5159: PetscErrorCode (*gconv)(Mat, MatFactorType, Mat *);
5161: PetscFunctionBegin;
5163: PetscAssertPointer(flg, 4);
5165: *flg = PETSC_FALSE;
5166: if (!((PetscObject)mat)->type_name) PetscFunctionReturn(PETSC_SUCCESS);
5168: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
5169: MatCheckPreallocated(mat, 1);
5171: PetscCall(MatSolverTypeGet(type, ((PetscObject)mat)->type_name, ftype, NULL, NULL, &gconv));
5172: *flg = gconv ? PETSC_TRUE : PETSC_FALSE;
5173: PetscFunctionReturn(PETSC_SUCCESS);
5174: }
5176: /*@
5177: MatDuplicate - Duplicates a matrix including the non-zero structure.
5179: Collective
5181: Input Parameters:
5182: + mat - the matrix
5183: - op - One of `MAT_DO_NOT_COPY_VALUES`, `MAT_COPY_VALUES`, or `MAT_SHARE_NONZERO_PATTERN`.
5184: See the manual page for `MatDuplicateOption()` for an explanation of these options.
5186: Output Parameter:
5187: . M - pointer to place new matrix
5189: Level: intermediate
5191: Notes:
5192: You cannot change the nonzero pattern for the parent or child matrix later if you use `MAT_SHARE_NONZERO_PATTERN`.
5194: If `op` is not `MAT_COPY_VALUES` the numerical values in the new matrix are zeroed.
5196: May be called with an unassembled input `Mat` if `MAT_DO_NOT_COPY_VALUES` is used, in which case the output `Mat` is unassembled as well.
5198: When original mat is a product of matrix operation, e.g., an output of `MatMatMult()` or `MatCreateSubMatrix()`, only the matrix data structure of `mat`
5199: is duplicated and the internal data structures created for the reuse of previous matrix operations are not duplicated.
5200: User should not use `MatDuplicate()` to create new matrix `M` if `M` is intended to be reused as the product of matrix operation.
5202: .seealso: [](ch_matrices), `Mat`, `MatCopy()`, `MatConvert()`, `MatDuplicateOption`
5203: @*/
5204: PetscErrorCode MatDuplicate(Mat mat, MatDuplicateOption op, Mat *M)
5205: {
5206: Mat B;
5207: VecType vtype;
5208: PetscInt i;
5209: PetscObject dm, container_h, container_d;
5210: PetscErrorCodeFn *viewf;
5212: PetscFunctionBegin;
5215: PetscAssertPointer(M, 3);
5216: PetscCheck(op != MAT_COPY_VALUES || mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "MAT_COPY_VALUES not allowed for unassembled matrix");
5217: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
5218: MatCheckPreallocated(mat, 1);
5220: PetscCall(PetscLogEventBegin(MAT_Convert, mat, 0, 0, 0));
5221: PetscUseTypeMethod(mat, duplicate, op, M);
5222: PetscCall(PetscLogEventEnd(MAT_Convert, mat, 0, 0, 0));
5223: B = *M;
5225: PetscCall(MatGetOperation(mat, MATOP_VIEW, &viewf));
5226: if (viewf) PetscCall(MatSetOperation(B, MATOP_VIEW, viewf));
5227: PetscCall(MatGetVecType(mat, &vtype));
5228: PetscCall(MatSetVecType(B, vtype));
5230: B->stencil.dim = mat->stencil.dim;
5231: B->stencil.noc = mat->stencil.noc;
5232: for (i = 0; i <= mat->stencil.dim + (mat->stencil.noc ? 0 : -1); i++) {
5233: B->stencil.dims[i] = mat->stencil.dims[i];
5234: B->stencil.starts[i] = mat->stencil.starts[i];
5235: }
5237: B->nooffproczerorows = mat->nooffproczerorows;
5238: B->nooffprocentries = mat->nooffprocentries;
5240: PetscCall(PetscObjectQuery((PetscObject)mat, "__PETSc_dm", &dm));
5241: if (dm) PetscCall(PetscObjectCompose((PetscObject)B, "__PETSc_dm", dm));
5242: PetscCall(PetscObjectQuery((PetscObject)mat, "__PETSc_MatCOOStruct_Host", &container_h));
5243: if (container_h) PetscCall(PetscObjectCompose((PetscObject)B, "__PETSc_MatCOOStruct_Host", container_h));
5244: PetscCall(PetscObjectQuery((PetscObject)mat, "__PETSc_MatCOOStruct_Device", &container_d));
5245: if (container_d) PetscCall(PetscObjectCompose((PetscObject)B, "__PETSc_MatCOOStruct_Device", container_d));
5246: if (op == MAT_COPY_VALUES) PetscCall(MatPropagateSymmetryOptions(mat, B));
5247: PetscCall(PetscObjectStateIncrease((PetscObject)B));
5248: PetscFunctionReturn(PETSC_SUCCESS);
5249: }
5251: /*@
5252: MatGetDiagonal - Gets the diagonal of a matrix as a `Vec`
5254: Logically Collective
5256: Input Parameter:
5257: . mat - the matrix
5259: Output Parameter:
5260: . v - the diagonal of the matrix
5262: Level: intermediate
5264: Note:
5265: If `mat` has local sizes `n` x `m`, this routine fills the first `ndiag = min(n, m)` entries
5266: of `v` with the diagonal values. Thus `v` must have local size of at least `ndiag`. If `v`
5267: is larger than `ndiag`, the values of the remaining entries are unspecified.
5269: Currently only correct in parallel for square matrices.
5271: .seealso: [](ch_matrices), `Mat`, `Vec`, `MatGetRow()`, `MatCreateSubMatrices()`, `MatCreateSubMatrix()`, `MatGetRowMaxAbs()`
5272: @*/
5273: PetscErrorCode MatGetDiagonal(Mat mat, Vec v)
5274: {
5275: PetscFunctionBegin;
5279: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5280: MatCheckPreallocated(mat, 1);
5281: if (PetscDefined(USE_DEBUG)) {
5282: PetscInt nv, row, col, ndiag;
5284: PetscCall(VecGetLocalSize(v, &nv));
5285: PetscCall(MatGetLocalSize(mat, &row, &col));
5286: ndiag = PetscMin(row, col);
5287: PetscCheck(nv >= ndiag, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Nonconforming Mat and Vec. Vec local size %" PetscInt_FMT " < Mat local diagonal length %" PetscInt_FMT, nv, ndiag);
5288: }
5290: PetscUseTypeMethod(mat, getdiagonal, v);
5291: PetscCall(PetscObjectStateIncrease((PetscObject)v));
5292: PetscFunctionReturn(PETSC_SUCCESS);
5293: }
5295: /*@
5296: MatGetRowMin - Gets the minimum value (of the real part) of each
5297: row of the matrix
5299: Logically Collective
5301: Input Parameter:
5302: . mat - the matrix
5304: Output Parameters:
5305: + v - the vector for storing the maximums
5306: - idx - the indices of the column found for each row (optional, pass `NULL` if not needed)
5308: Level: intermediate
5310: Note:
5311: The result of this call are the same as if one converted the matrix to dense format
5312: and found the minimum value in each row (i.e. the implicit zeros are counted as zeros).
5314: This code is only implemented for a couple of matrix formats.
5316: .seealso: [](ch_matrices), `Mat`, `MatGetDiagonal()`, `MatCreateSubMatrices()`, `MatCreateSubMatrix()`, `MatGetRowMaxAbs()`, `MatGetRowMinAbs()`,
5317: `MatGetRowMax()`
5318: @*/
5319: PetscErrorCode MatGetRowMin(Mat mat, Vec v, PetscInt idx[])
5320: {
5321: PetscFunctionBegin;
5325: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5327: if (!mat->cmap->N) {
5328: PetscCall(VecSet(v, PETSC_MAX_REAL));
5329: if (idx) {
5330: PetscInt i, m = mat->rmap->n;
5331: for (i = 0; i < m; i++) idx[i] = -1;
5332: }
5333: } else {
5334: MatCheckPreallocated(mat, 1);
5335: }
5336: PetscUseTypeMethod(mat, getrowmin, v, idx);
5337: PetscCall(PetscObjectStateIncrease((PetscObject)v));
5338: PetscFunctionReturn(PETSC_SUCCESS);
5339: }
5341: /*@
5342: MatGetRowMinAbs - Gets the minimum value (in absolute value) of each
5343: row of the matrix
5345: Logically Collective
5347: Input Parameter:
5348: . mat - the matrix
5350: Output Parameters:
5351: + v - the vector for storing the minimums
5352: - idx - the indices of the column found for each row (or `NULL` if not needed)
5354: Level: intermediate
5356: Notes:
5357: if a row is completely empty or has only 0.0 values, then the `idx` value for that
5358: row is 0 (the first column).
5360: This code is only implemented for a couple of matrix formats.
5362: .seealso: [](ch_matrices), `Mat`, `MatGetDiagonal()`, `MatCreateSubMatrices()`, `MatCreateSubMatrix()`, `MatGetRowMax()`, `MatGetRowMaxAbs()`, `MatGetRowMin()`
5363: @*/
5364: PetscErrorCode MatGetRowMinAbs(Mat mat, Vec v, PetscInt idx[])
5365: {
5366: PetscFunctionBegin;
5370: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5371: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
5373: if (!mat->cmap->N) {
5374: PetscCall(VecSet(v, 0.0));
5375: if (idx) {
5376: PetscInt i, m = mat->rmap->n;
5377: for (i = 0; i < m; i++) idx[i] = -1;
5378: }
5379: } else {
5380: MatCheckPreallocated(mat, 1);
5381: if (idx) PetscCall(PetscArrayzero(idx, mat->rmap->n));
5382: PetscUseTypeMethod(mat, getrowminabs, v, idx);
5383: }
5384: PetscCall(PetscObjectStateIncrease((PetscObject)v));
5385: PetscFunctionReturn(PETSC_SUCCESS);
5386: }
5388: /*@
5389: MatGetRowMax - Gets the maximum value (of the real part) of each
5390: row of the matrix
5392: Logically Collective
5394: Input Parameter:
5395: . mat - the matrix
5397: Output Parameters:
5398: + v - the vector for storing the maximums
5399: - idx - the indices of the column found for each row (optional, otherwise pass `NULL`)
5401: Level: intermediate
5403: Notes:
5404: The result of this call are the same as if one converted the matrix to dense format
5405: and found the minimum value in each row (i.e. the implicit zeros are counted as zeros).
5407: This code is only implemented for a couple of matrix formats.
5409: .seealso: [](ch_matrices), `Mat`, `MatGetDiagonal()`, `MatCreateSubMatrices()`, `MatCreateSubMatrix()`, `MatGetRowMaxAbs()`, `MatGetRowMin()`, `MatGetRowMinAbs()`
5410: @*/
5411: PetscErrorCode MatGetRowMax(Mat mat, Vec v, PetscInt idx[])
5412: {
5413: PetscFunctionBegin;
5417: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5419: if (!mat->cmap->N) {
5420: PetscCall(VecSet(v, PETSC_MIN_REAL));
5421: if (idx) {
5422: PetscInt i, m = mat->rmap->n;
5423: for (i = 0; i < m; i++) idx[i] = -1;
5424: }
5425: } else {
5426: MatCheckPreallocated(mat, 1);
5427: PetscUseTypeMethod(mat, getrowmax, v, idx);
5428: }
5429: PetscCall(PetscObjectStateIncrease((PetscObject)v));
5430: PetscFunctionReturn(PETSC_SUCCESS);
5431: }
5433: /*@
5434: MatGetRowMaxAbs - Gets the maximum value (in absolute value) of each
5435: row of the matrix
5437: Logically Collective
5439: Input Parameter:
5440: . mat - the matrix
5442: Output Parameters:
5443: + v - the vector for storing the maximums
5444: - idx - the indices of the column found for each row (or `NULL` if not needed)
5446: Level: intermediate
5448: Notes:
5449: if a row is completely empty or has only 0.0 values, then the `idx` value for that
5450: row is 0 (the first column).
5452: This code is only implemented for a couple of matrix formats.
5454: .seealso: [](ch_matrices), `Mat`, `MatGetDiagonal()`, `MatCreateSubMatrices()`, `MatCreateSubMatrix()`, `MatGetRowSum()`, `MatGetRowMin()`, `MatGetRowMinAbs()`
5455: @*/
5456: PetscErrorCode MatGetRowMaxAbs(Mat mat, Vec v, PetscInt idx[])
5457: {
5458: PetscFunctionBegin;
5462: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5464: if (!mat->cmap->N) {
5465: PetscCall(VecSet(v, 0.0));
5466: if (idx) {
5467: PetscInt i, m = mat->rmap->n;
5468: for (i = 0; i < m; i++) idx[i] = -1;
5469: }
5470: } else {
5471: MatCheckPreallocated(mat, 1);
5472: if (idx) PetscCall(PetscArrayzero(idx, mat->rmap->n));
5473: PetscUseTypeMethod(mat, getrowmaxabs, v, idx);
5474: }
5475: PetscCall(PetscObjectStateIncrease((PetscObject)v));
5476: PetscFunctionReturn(PETSC_SUCCESS);
5477: }
5479: /*@
5480: MatGetRowSumAbs - Gets the sum value (in absolute value) of each row of the matrix
5482: Logically Collective
5484: Input Parameter:
5485: . mat - the matrix
5487: Output Parameter:
5488: . v - the vector for storing the sum
5490: Level: intermediate
5492: This code is only implemented for a couple of matrix formats.
5494: .seealso: [](ch_matrices), `Mat`, `MatGetDiagonal()`, `MatCreateSubMatrices()`, `MatCreateSubMatrix()`, `MatGetRowMax()`, `MatGetRowMin()`, `MatGetRowMinAbs()`
5495: @*/
5496: PetscErrorCode MatGetRowSumAbs(Mat mat, Vec v)
5497: {
5498: PetscFunctionBegin;
5502: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5504: if (!mat->cmap->N) PetscCall(VecSet(v, 0.0));
5505: else {
5506: MatCheckPreallocated(mat, 1);
5507: PetscUseTypeMethod(mat, getrowsumabs, v);
5508: }
5509: PetscCall(PetscObjectStateIncrease((PetscObject)v));
5510: PetscFunctionReturn(PETSC_SUCCESS);
5511: }
5513: /*@
5514: MatGetRowSum - Gets the sum of each row of the matrix
5516: Logically or Neighborhood Collective
5518: Input Parameter:
5519: . mat - the matrix
5521: Output Parameter:
5522: . v - the vector for storing the sum of rows
5524: Level: intermediate
5526: Note:
5527: This code is slow since it is not currently specialized for different formats
5529: .seealso: [](ch_matrices), `Mat`, `MatGetDiagonal()`, `MatCreateSubMatrices()`, `MatCreateSubMatrix()`, `MatGetRowMax()`, `MatGetRowMin()`, `MatGetRowMaxAbs()`, `MatGetRowMinAbs()`, `MatGetRowSumAbs()`
5530: @*/
5531: PetscErrorCode MatGetRowSum(Mat mat, Vec v)
5532: {
5533: Vec ones;
5535: PetscFunctionBegin;
5539: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5540: MatCheckPreallocated(mat, 1);
5541: PetscCall(MatCreateVecs(mat, &ones, NULL));
5542: PetscCall(VecSet(ones, 1.));
5543: PetscCall(MatMult(mat, ones, v));
5544: PetscCall(VecDestroy(&ones));
5545: PetscFunctionReturn(PETSC_SUCCESS);
5546: }
5548: /*@
5549: MatTransposeSetPrecursor - Set the matrix from which the second matrix will receive numerical transpose data with a call to `MatTranspose`(A,`MAT_REUSE_MATRIX`,&B)
5550: when B was not obtained with `MatTranspose`(A,`MAT_INITIAL_MATRIX`,&B)
5552: Collective
5554: Input Parameter:
5555: . mat - the matrix to provide the transpose
5557: Output Parameter:
5558: . B - the matrix to contain the transpose; it MUST have the nonzero structure of the transpose of A or the code will crash or generate incorrect results
5560: Level: advanced
5562: Note:
5563: Normally the use of `MatTranspose`(A, `MAT_REUSE_MATRIX`, &B) requires that `B` was obtained with a call to `MatTranspose`(A, `MAT_INITIAL_MATRIX`, &B). This
5564: routine allows bypassing that call.
5566: .seealso: [](ch_matrices), `Mat`, `MatTransposeSymbolic()`, `MatTranspose()`, `MatMultTranspose()`, `MatMultTransposeAdd()`, `MatIsTranspose()`, `MatReuse`, `MAT_INITIAL_MATRIX`, `MAT_REUSE_MATRIX`, `MAT_INPLACE_MATRIX`
5567: @*/
5568: PetscErrorCode MatTransposeSetPrecursor(Mat mat, Mat B)
5569: {
5570: MatState *rb = NULL;
5572: PetscFunctionBegin;
5573: PetscCall(PetscNew(&rb));
5574: rb->id = ((PetscObject)mat)->id;
5575: rb->state = 0;
5576: PetscCall(MatGetNonzeroState(mat, &rb->nonzerostate));
5577: PetscCall(PetscObjectContainerCompose((PetscObject)B, "MatTransposeParent", rb, PetscCtxDestroyDefault));
5578: PetscFunctionReturn(PETSC_SUCCESS);
5579: }
5581: static PetscErrorCode MatTranspose_Private(Mat mat, MatReuse reuse, Mat *B, PetscBool conjugate)
5582: {
5583: PetscContainer rB = NULL;
5584: MatState *rb = NULL;
5585: PetscErrorCode (*f)(Mat, MatReuse, Mat *) = NULL;
5587: PetscFunctionBegin;
5590: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5591: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
5592: PetscCheck(reuse != MAT_INPLACE_MATRIX || mat == *B, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "MAT_INPLACE_MATRIX requires last matrix to match first");
5593: PetscCheck(reuse != MAT_REUSE_MATRIX || mat != *B, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Perhaps you mean MAT_INPLACE_MATRIX");
5594: MatCheckPreallocated(mat, 1);
5595: if (reuse == MAT_REUSE_MATRIX) {
5596: PetscCall(PetscObjectQuery((PetscObject)*B, "MatTransposeParent", (PetscObject *)&rB));
5597: PetscCheck(rB, PetscObjectComm((PetscObject)*B), PETSC_ERR_ARG_WRONG, "Reuse matrix used was not generated from call to MatTranspose(). Suggest MatTransposeSetPrecursor().");
5598: PetscCall(PetscContainerGetPointer(rB, &rb));
5599: PetscCheck(rb->id == ((PetscObject)mat)->id, PetscObjectComm((PetscObject)*B), PETSC_ERR_ARG_WRONG, "Reuse matrix used was not generated from input matrix");
5600: if (rb->state == ((PetscObject)mat)->state) PetscFunctionReturn(PETSC_SUCCESS);
5601: }
5603: if (conjugate) {
5604: f = mat->ops->hermitiantranspose;
5605: if (f) PetscCall((*f)(mat, reuse, B));
5606: }
5607: if (!f && !(reuse == MAT_INPLACE_MATRIX && mat->hermitian == PETSC_BOOL3_TRUE && conjugate)) {
5608: PetscCall(PetscLogEventBegin(MAT_Transpose, mat, 0, 0, 0));
5609: if (reuse != MAT_INPLACE_MATRIX || mat->symmetric != PETSC_BOOL3_TRUE) {
5610: PetscUseTypeMethod(mat, transpose, reuse, B);
5611: PetscCall(PetscObjectStateIncrease((PetscObject)*B));
5612: }
5613: PetscCall(PetscLogEventEnd(MAT_Transpose, mat, 0, 0, 0));
5614: if (conjugate) PetscCall(MatConjugate(*B));
5615: }
5617: if (reuse == MAT_INITIAL_MATRIX) PetscCall(MatTransposeSetPrecursor(mat, *B));
5618: if (reuse != MAT_INPLACE_MATRIX) {
5619: PetscCall(PetscObjectQuery((PetscObject)*B, "MatTransposeParent", (PetscObject *)&rB));
5620: PetscCall(PetscContainerGetPointer(rB, &rb));
5621: rb->state = ((PetscObject)mat)->state;
5622: rb->nonzerostate = mat->nonzerostate;
5623: }
5624: PetscFunctionReturn(PETSC_SUCCESS);
5625: }
5627: /*@
5628: MatTranspose - Computes the transpose of a matrix, either in-place or out-of-place.
5630: Collective
5632: Input Parameters:
5633: + mat - the matrix to transpose
5634: - reuse - either `MAT_INITIAL_MATRIX`, `MAT_REUSE_MATRIX`, or `MAT_INPLACE_MATRIX`
5636: Output Parameter:
5637: . B - the transpose of the matrix
5639: Level: intermediate
5641: Notes:
5642: If you use `MAT_INPLACE_MATRIX` then you must pass in `&mat` for `B`
5644: `MAT_REUSE_MATRIX` uses the `B` matrix obtained from a previous call to this function with `MAT_INITIAL_MATRIX` to store the transpose. If you already have a matrix to contain the
5645: transpose, call `MatTransposeSetPrecursor(mat, B)` before calling this routine.
5647: If the nonzero structure of `mat` changed from the previous call to this function with the same matrices an error will be generated for some matrix types.
5649: Consider using `MatCreateTranspose()` instead if you only need a matrix that behaves like the transpose but don't need the storage to be changed.
5650: For example, the result of `MatCreateTranspose()` will compute the transpose of the given matrix times a vector for matrix-vector products computed with `MatMult()`.
5652: If `mat` is unchanged from the last call this function returns immediately without recomputing the result
5654: If you only need the symbolic transpose of a matrix, and not the numerical values, use `MatTransposeSymbolic()`
5656: .seealso: [](ch_matrices), `Mat`, `MatTransposeSetPrecursor()`, `MatMultTranspose()`, `MatMultTransposeAdd()`, `MatIsTranspose()`, `MatReuse`, `MAT_INITIAL_MATRIX`, `MAT_REUSE_MATRIX`, `MAT_INPLACE_MATRIX`,
5657: `MatTransposeSymbolic()`, `MatCreateTranspose()`
5658: @*/
5659: PetscErrorCode MatTranspose(Mat mat, MatReuse reuse, Mat *B)
5660: {
5661: PetscFunctionBegin;
5662: PetscCall(MatTranspose_Private(mat, reuse, B, PETSC_FALSE));
5663: PetscFunctionReturn(PETSC_SUCCESS);
5664: }
5666: /*@
5667: MatTransposeSymbolic - Computes the symbolic part of the transpose of a matrix.
5669: Collective
5671: Input Parameter:
5672: . A - the matrix to transpose
5674: Output Parameter:
5675: . B - the transpose. This is a complete matrix but the numerical portion is invalid. One can call `MatTranspose`(A,`MAT_REUSE_MATRIX`,&B) to compute the
5676: numerical portion.
5678: Level: intermediate
5680: Note:
5681: This is not supported for many matrix types, use `MatTranspose()` in those cases
5683: .seealso: [](ch_matrices), `Mat`, `MatTransposeSetPrecursor()`, `MatTranspose()`, `MatMultTranspose()`, `MatMultTransposeAdd()`, `MatIsTranspose()`, `MatReuse`, `MAT_INITIAL_MATRIX`, `MAT_REUSE_MATRIX`, `MAT_INPLACE_MATRIX`
5684: @*/
5685: PetscErrorCode MatTransposeSymbolic(Mat A, Mat *B)
5686: {
5687: PetscFunctionBegin;
5690: PetscCheck(A->assembled, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5691: PetscCheck(!A->factortype, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
5692: PetscCall(PetscLogEventBegin(MAT_Transpose, A, 0, 0, 0));
5693: PetscUseTypeMethod(A, transposesymbolic, B);
5694: PetscCall(PetscLogEventEnd(MAT_Transpose, A, 0, 0, 0));
5696: PetscCall(MatTransposeSetPrecursor(A, *B));
5697: PetscFunctionReturn(PETSC_SUCCESS);
5698: }
5700: PetscErrorCode MatTransposeCheckNonzeroState_Private(Mat A, Mat B)
5701: {
5702: PetscContainer rB;
5703: MatState *rb;
5705: PetscFunctionBegin;
5708: PetscCheck(A->assembled, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5709: PetscCheck(!A->factortype, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
5710: PetscCall(PetscObjectQuery((PetscObject)B, "MatTransposeParent", (PetscObject *)&rB));
5711: PetscCheck(rB, PetscObjectComm((PetscObject)B), PETSC_ERR_ARG_WRONG, "Reuse matrix used was not generated from call to MatTranspose()");
5712: PetscCall(PetscContainerGetPointer(rB, &rb));
5713: PetscCheck(rb->id == ((PetscObject)A)->id, PetscObjectComm((PetscObject)B), PETSC_ERR_ARG_WRONG, "Reuse matrix used was not generated from input matrix");
5714: PetscCheck(rb->nonzerostate == A->nonzerostate, PetscObjectComm((PetscObject)B), PETSC_ERR_ARG_WRONGSTATE, "Reuse matrix has changed nonzero structure");
5715: PetscFunctionReturn(PETSC_SUCCESS);
5716: }
5718: /*@
5719: MatIsTranspose - Test whether a matrix is another one's transpose,
5720: or its own, in which case it tests symmetry.
5722: Collective
5724: Input Parameters:
5725: + A - the matrix to test
5726: . B - the matrix to test against, this can equal the first parameter
5727: - tol - tolerance, differences between entries smaller than this are counted as zero
5729: Output Parameter:
5730: . flg - the result
5732: Level: intermediate
5734: Notes:
5735: The sequential algorithm has a running time of the order of the number of nonzeros; the parallel
5736: test involves parallel copies of the block off-diagonal parts of the matrix.
5738: .seealso: [](ch_matrices), `Mat`, `MatTranspose()`, `MatIsSymmetric()`, `MatIsHermitian()`
5739: @*/
5740: PetscErrorCode MatIsTranspose(Mat A, Mat B, PetscReal tol, PetscBool *flg)
5741: {
5742: PetscErrorCode (*f)(Mat, Mat, PetscReal, PetscBool *), (*g)(Mat, Mat, PetscReal, PetscBool *);
5744: PetscFunctionBegin;
5747: PetscAssertPointer(flg, 4);
5748: PetscCall(PetscObjectQueryFunction((PetscObject)A, "MatIsTranspose_C", &f));
5749: PetscCall(PetscObjectQueryFunction((PetscObject)B, "MatIsTranspose_C", &g));
5750: *flg = PETSC_FALSE;
5751: if (f && g) {
5752: PetscCheck(f == g, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_NOTSAMETYPE, "Matrices do not have the same comparator for symmetry test");
5753: PetscCall((*f)(A, B, tol, flg));
5754: } else {
5755: MatType mattype;
5757: PetscCall(MatGetType(f ? B : A, &mattype));
5758: SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "Matrix of type %s does not support checking for transpose", mattype);
5759: }
5760: PetscFunctionReturn(PETSC_SUCCESS);
5761: }
5763: /*@
5764: MatHermitianTranspose - Computes an in-place or out-of-place Hermitian transpose of a matrix in complex conjugate.
5766: Collective
5768: Input Parameters:
5769: + mat - the matrix to transpose and complex conjugate
5770: - reuse - either `MAT_INITIAL_MATRIX`, `MAT_REUSE_MATRIX`, or `MAT_INPLACE_MATRIX`
5772: Output Parameter:
5773: . B - the Hermitian transpose
5775: Level: intermediate
5777: .seealso: [](ch_matrices), `Mat`, `MatTranspose()`, `MatMultTranspose()`, `MatMultTransposeAdd()`, `MatIsTranspose()`, `MatReuse`
5778: @*/
5779: PetscErrorCode MatHermitianTranspose(Mat mat, MatReuse reuse, Mat *B)
5780: {
5781: PetscFunctionBegin;
5782: PetscCall(MatTranspose_Private(mat, reuse, B, PetscDefined(USE_COMPLEX) ? PETSC_TRUE : PETSC_FALSE));
5783: PetscFunctionReturn(PETSC_SUCCESS);
5784: }
5786: /*@
5787: MatIsHermitianTranspose - Test whether a matrix is another one's Hermitian transpose,
5789: Collective
5791: Input Parameters:
5792: + A - the matrix to test
5793: . B - the matrix to test against, this can equal the first parameter
5794: - tol - tolerance, differences between entries smaller than this are counted as zero
5796: Output Parameter:
5797: . flg - the result
5799: Level: intermediate
5801: Notes:
5802: Only available for `MATAIJ` matrices.
5804: The sequential algorithm
5805: has a running time of the order of the number of nonzeros; the parallel
5806: test involves parallel copies of the block off-diagonal parts of the matrix.
5808: .seealso: [](ch_matrices), `Mat`, `MatTranspose()`, `MatIsSymmetric()`, `MatIsHermitian()`, `MatIsTranspose()`
5809: @*/
5810: PetscErrorCode MatIsHermitianTranspose(Mat A, Mat B, PetscReal tol, PetscBool *flg)
5811: {
5812: PetscErrorCode (*f)(Mat, Mat, PetscReal, PetscBool *), (*g)(Mat, Mat, PetscReal, PetscBool *);
5814: PetscFunctionBegin;
5817: PetscAssertPointer(flg, 4);
5818: PetscCall(PetscObjectQueryFunction((PetscObject)A, "MatIsHermitianTranspose_C", &f));
5819: PetscCall(PetscObjectQueryFunction((PetscObject)B, "MatIsHermitianTranspose_C", &g));
5820: if (f && g) {
5821: PetscCheck(f == g, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_NOTSAMETYPE, "Matrices do not have the same comparator for Hermitian test");
5822: PetscCall((*f)(A, B, tol, flg));
5823: } else {
5824: MatType mattype;
5826: PetscCall(MatGetType(f ? B : A, &mattype));
5827: SETERRQ(PETSC_COMM_SELF, PETSC_ERR_SUP, "Matrix of type %s does not support checking for Hermitian transpose", mattype);
5828: }
5829: PetscFunctionReturn(PETSC_SUCCESS);
5830: }
5832: /*@
5833: MatPermute - Creates a new matrix with rows and columns permuted from the
5834: original.
5836: Collective
5838: Input Parameters:
5839: + mat - the matrix to permute
5840: . row - row permutation, each process supplies only the permutation for its rows
5841: - col - column permutation, each process supplies only the permutation for its columns
5843: Output Parameter:
5844: . B - the permuted matrix
5846: Level: advanced
5848: Note:
5849: The index sets map from `row`/`col` of permuted matrix to `row`/`col` of original matrix.
5850: The index sets should be on the same communicator as mat and have the same local sizes.
5851: `MATSEQSBAIJ` inputs may produce a `MATSEQBAIJ` matrix when the permutation does not preserve symmetry.
5853: Developer Note:
5854: If you want to implement `MatPermute()` for a matrix type, and your approach doesn't
5855: exploit the fact that `row` and `col` are permutations, consider implementing the
5856: more general `MatCreateSubMatrix()` instead.
5858: .seealso: [](ch_matrices), `Mat`, `MatGetOrdering()`, `ISAllGather()`, `MatCreateSubMatrix()`
5859: @*/
5860: PetscErrorCode MatPermute(Mat mat, IS row, IS col, Mat *B)
5861: {
5862: PetscFunctionBegin;
5867: PetscAssertPointer(B, 4);
5868: PetscCheckSameComm(mat, 1, row, 2);
5869: if (row != col) PetscCheckSameComm(row, 2, col, 3);
5870: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5871: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
5872: PetscCheck(mat->ops->permute || mat->ops->createsubmatrix, PETSC_COMM_SELF, PETSC_ERR_SUP, "MatPermute not available for Mat type %s", ((PetscObject)mat)->type_name);
5873: MatCheckPreallocated(mat, 1);
5875: if (mat->ops->permute) {
5876: PetscUseTypeMethod(mat, permute, row, col, B);
5877: PetscCall(PetscObjectStateIncrease((PetscObject)*B));
5878: } else {
5879: PetscCall(MatCreateSubMatrix(mat, row, col, MAT_INITIAL_MATRIX, B));
5880: }
5881: PetscFunctionReturn(PETSC_SUCCESS);
5882: }
5884: /*@
5885: MatEqual - Compares two matrices.
5887: Collective
5889: Input Parameters:
5890: + A - the first matrix
5891: - B - the second matrix
5893: Output Parameter:
5894: . flg - `PETSC_TRUE` if the matrices are equal; `PETSC_FALSE` otherwise.
5896: Level: intermediate
5898: Note:
5899: If either of the matrix is "matrix-free", meaning the matrix entries are not stored explicitly then equality is determined by comparing
5900: the results of several matrix-vector product using randomly created vectors, see `MatMultEqual()`.
5902: .seealso: [](ch_matrices), `Mat`, `MatMultEqual()`
5903: @*/
5904: PetscErrorCode MatEqual(Mat A, Mat B, PetscBool *flg)
5905: {
5906: PetscFunctionBegin;
5911: PetscAssertPointer(flg, 3);
5912: PetscCheckSameComm(A, 1, B, 2);
5913: MatCheckPreallocated(A, 1);
5914: MatCheckPreallocated(B, 2);
5915: PetscCheck(A->assembled, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5916: PetscCheck(B->assembled, PetscObjectComm((PetscObject)B), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5917: PetscCheck(A->rmap->N == B->rmap->N && A->cmap->N == B->cmap->N, PetscObjectComm((PetscObject)A), PETSC_ERR_ARG_SIZ, "Mat A,Mat B: global dim %" PetscInt_FMT " %" PetscInt_FMT " %" PetscInt_FMT " %" PetscInt_FMT, A->rmap->N, B->rmap->N, A->cmap->N,
5918: B->cmap->N);
5919: if (A->ops->equal && A->ops->equal == B->ops->equal) PetscUseTypeMethod(A, equal, B, flg);
5920: else PetscCall(MatMultEqual(A, B, 10, flg));
5921: PetscFunctionReturn(PETSC_SUCCESS);
5922: }
5924: /*@
5925: MatDiagonalScale - Scales a matrix on the left and right by diagonal
5926: matrices that are stored as vectors. Either of the two scaling
5927: matrices can be `NULL`.
5929: Collective
5931: Input Parameters:
5932: + mat - the matrix to be scaled
5933: . l - the left scaling vector (or `NULL`)
5934: - r - the right scaling vector (or `NULL`)
5936: Level: intermediate
5938: Note:
5939: `MatDiagonalScale()` computes $A = LAR$, where
5940: L = a diagonal matrix (stored as a vector), R = a diagonal matrix (stored as a vector)
5941: The L scales the rows of the matrix, the R scales the columns of the matrix.
5942: For `MATSEQSBAIJ`, if `l` and `r` are different `Vec` objects, `mat` changes to type `MATSEQBAIJ` because the result is not necessarily symmetric.
5944: .seealso: [](ch_matrices), `Mat`, `MatScale()`, `MatShift()`, `MatDiagonalSet()`
5945: @*/
5946: PetscErrorCode MatDiagonalScale(Mat mat, Vec l, Vec r)
5947: {
5948: PetscBool flg = PETSC_FALSE;
5950: PetscFunctionBegin;
5953: if (l) {
5955: PetscCheckSameComm(mat, 1, l, 2);
5956: }
5957: if (r) {
5959: PetscCheckSameComm(mat, 1, r, 3);
5960: }
5961: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
5962: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
5963: MatCheckPreallocated(mat, 1);
5964: if (!l && !r) PetscFunctionReturn(PETSC_SUCCESS);
5966: PetscCall(PetscLogEventBegin(MAT_Scale, mat, 0, 0, 0));
5967: PetscUseTypeMethod(mat, diagonalscale, l, r);
5968: PetscCall(PetscLogEventEnd(MAT_Scale, mat, 0, 0, 0));
5969: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
5970: if (l != r && (PetscBool3ToBool(mat->symmetric) || PetscBool3ToBool(mat->hermitian))) {
5971: if (!PetscDefined(USE_COMPLEX) || PetscBool3ToBool(mat->symmetric)) {
5972: if (l && r) PetscCall(VecEqual(l, r, &flg));
5973: if (!flg) {
5974: PetscCall(PetscObjectTypeCompare((PetscObject)mat, MATMPISBAIJ, &flg));
5975: PetscCheck(!flg, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_OUTOFRANGE, "For MATMPISBAIJ, left and right scaling vectors must be the same");
5976: mat->symmetric = mat->spd = PETSC_BOOL3_FALSE;
5977: if (!PetscDefined(USE_COMPLEX)) mat->hermitian = PETSC_BOOL3_FALSE;
5978: else mat->hermitian = PETSC_BOOL3_UNKNOWN;
5979: }
5980: }
5981: if (PetscDefined(USE_COMPLEX) && PetscBool3ToBool(mat->hermitian)) {
5982: flg = PETSC_FALSE;
5983: if (l && r) {
5984: Vec conjugate;
5986: PetscCall(VecDuplicate(l, &conjugate));
5987: PetscCall(VecCopy(l, conjugate));
5988: PetscCall(VecConjugate(conjugate));
5989: PetscCall(VecEqual(conjugate, r, &flg));
5990: PetscCall(VecDestroy(&conjugate));
5991: }
5992: if (!flg) {
5993: PetscCall(PetscObjectTypeCompare((PetscObject)mat, MATMPISBAIJ, &flg));
5994: PetscCheck(!flg, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_OUTOFRANGE, "For Hermitian MATMPISBAIJ, left and right scaling vectors must be conjugate one of the other");
5995: mat->hermitian = PETSC_BOOL3_FALSE;
5996: mat->symmetric = mat->spd = PETSC_BOOL3_UNKNOWN;
5997: }
5998: }
5999: }
6000: PetscFunctionReturn(PETSC_SUCCESS);
6001: }
6003: /*@
6004: MatScale - Scales all elements of a matrix by a given number.
6006: Logically Collective
6008: Input Parameters:
6009: + mat - the matrix to be scaled
6010: - a - the scaling value
6012: Level: intermediate
6014: .seealso: [](ch_matrices), `Mat`, `MatDiagonalScale()`
6015: @*/
6016: PetscErrorCode MatScale(Mat mat, PetscScalar a)
6017: {
6018: PetscFunctionBegin;
6021: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
6022: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
6024: MatCheckPreallocated(mat, 1);
6026: PetscCall(PetscLogEventBegin(MAT_Scale, mat, 0, 0, 0));
6027: if (a != (PetscScalar)1.0) {
6028: PetscUseTypeMethod(mat, scale, a);
6029: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
6030: }
6031: PetscCall(PetscLogEventEnd(MAT_Scale, mat, 0, 0, 0));
6032: PetscFunctionReturn(PETSC_SUCCESS);
6033: }
6035: /*@
6036: MatNorm - Calculates various norms of a matrix.
6038: Collective
6040: Input Parameters:
6041: + mat - the matrix
6042: - type - the type of norm, `NORM_1`, `NORM_FROBENIUS`, `NORM_INFINITY`
6044: Output Parameter:
6045: . nrm - the resulting norm
6047: Level: intermediate
6049: .seealso: [](ch_matrices), `Mat`, `MatNormApproximate()`
6050: @*/
6051: PetscErrorCode MatNorm(Mat mat, NormType type, PetscReal *nrm)
6052: {
6053: PetscFunctionBegin;
6057: PetscAssertPointer(nrm, 3);
6059: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
6060: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
6061: MatCheckPreallocated(mat, 1);
6063: PetscUseTypeMethod(mat, norm, type, nrm);
6064: PetscFunctionReturn(PETSC_SUCCESS);
6065: }
6067: static PetscErrorCode VecSetFinalNormApp_Private(Vec x)
6068: {
6069: PetscScalar *ax, nm1;
6070: PetscInt st, en, n;
6072: PetscFunctionBegin;
6073: PetscCall(VecGetSize(x, &n));
6074: if (n < 2) PetscFunctionReturn(PETSC_SUCCESS);
6075: nm1 = n - 1;
6076: PetscCall(VecGetOwnershipRange(x, &st, &en));
6077: PetscCall(VecGetArrayWrite(x, &ax));
6078: for (PetscInt i = st; i < en; i++) {
6079: const PetscInt ii = i - st;
6080: const PetscScalar s = i % 2 ? -1.0 : 1.0;
6082: ax[ii] = s * (1.0 + i / nm1);
6083: }
6084: PetscCall(VecRestoreArrayWrite(x, &ax));
6085: PetscFunctionReturn(PETSC_SUCCESS);
6086: }
6088: static PetscErrorCode MatNormApproximateForwardOnly_Private(Mat A, NormType normtype, PetscInt maxit, PetscBool boundtocpu, PetscReal *n)
6089: {
6090: Vec x, y;
6091: PetscReal normx, normy;
6092: PetscInt i, N;
6093: PetscRandom rnd;
6095: PetscFunctionBegin;
6096: if (maxit < 0) maxit = 1;
6097: PetscCall(PetscRandomCreate(PetscObjectComm((PetscObject)A), &rnd));
6098: PetscCall(PetscRandomSetFromOptions(rnd));
6099: PetscCall(MatCreateVecs(A, &x, &y));
6100: PetscCall(VecBindToCPU(x, boundtocpu));
6101: PetscCall(VecBindToCPU(y, boundtocpu));
6102: PetscCall(VecGetSize(x, &N));
6103: *n = 0.0;
6104: for (i = 0; i < maxit; i++) {
6105: PetscCall(VecSetRandom(x, rnd));
6106: switch (normtype) {
6107: case NORM_1:
6108: PetscCall(VecNorm(x, NORM_1, &normx));
6109: if (normx > 0.0) PetscCall(VecScale(x, 1.0 / normx));
6110: break;
6111: case NORM_INFINITY:
6112: PetscCall(VecShift(x, -0.5));
6113: PetscCall(VecPointwiseSign(x, x, VEC_SIGN_ZERO_TO_SIGNED_UNIT));
6114: break;
6115: case NORM_2:
6116: PetscCall(VecNormalize(x, NULL));
6117: break;
6118: default:
6119: PetscUnreachable();
6120: }
6121: PetscCall(MatMult(A, x, y));
6122: PetscCall(VecNorm(y, normtype, &normy));
6123: *n = PetscMax(*n, normy);
6124: PetscCall(PetscInfo(A, "%s norm forward-only sample %" PetscInt_FMT " -> %g\n", NormTypes[normtype], i, (double)normy));
6125: }
6126: PetscCall(VecDestroy(&x));
6127: PetscCall(VecDestroy(&y));
6128: PetscCall(PetscRandomDestroy(&rnd));
6129: PetscFunctionReturn(PETSC_SUCCESS);
6130: }
6132: /*@
6133: MatNormApproximate - Approximate the norm of a matrix.
6135: Collective
6137: Input Parameters:
6138: + A - the matrix
6139: . normtype - the `NormType`
6140: - maxit - maximum number of iterations to use
6142: Output Parameter:
6143: . n - the norm estimate
6145: Level: intermediate
6147: Notes:
6148: Does not need access to the matrix entries; it just performs matrix-vector and transposed matrix-vector products {cite}`Higham1992`, {cite}`doi:10.1137/S0895479899356080`.
6150: If `maxit` is negative, a default number of iterations (10 for `NORM_1` and `NORM_INFINITY` and 20 for `NORM_2`) is performed.
6152: .seealso: [](ch_matrices), `Mat`, `MatNorm()`
6153: @*/
6154: PetscErrorCode MatNormApproximate(Mat A, NormType normtype, PetscInt maxit, PetscReal *n)
6155: {
6156: Vec x, y, w, z;
6157: PetscReal normz, adot;
6158: PetscScalar dot;
6159: PetscInt i, j, N, jold = -1;
6160: PetscBool boundtocpu = PETSC_TRUE, setherm, isherm, hasop;
6162: PetscFunctionBegin;
6167: PetscAssertPointer(n, 4);
6168: #if PetscDefined(HAVE_DEVICE)
6169: boundtocpu = A->boundtocpu;
6170: #endif
6171: PetscCall(MatHasOperation(A, MATOP_MULT_HERMITIAN_TRANSPOSE, &hasop));
6172: switch (normtype) {
6173: case NORM_INFINITY:
6174: case NORM_1:
6175: if (!hasop) {
6176: PetscCall(MatNormApproximateForwardOnly_Private(A, normtype, maxit, boundtocpu, n));
6177: i = maxit;
6178: break;
6179: } else {
6180: PetscCall(MatIsHermitianKnown(A, &setherm, &isherm));
6181: if ((setherm && isherm) || normtype == NORM_1) PetscCall(PetscObjectReference((PetscObject)A));
6182: else {
6183: Mat B;
6185: PetscCall(MatCreateHermitianTranspose(A, &B));
6186: A = B;
6187: }
6188: }
6189: if (maxit < 0) maxit = 10; /* pure guess */
6190: PetscCall(MatCreateVecs(A, &x, &y));
6191: PetscCall(MatCreateVecs(A, &z, &w));
6192: PetscCall(VecBindToCPU(x, boundtocpu));
6193: PetscCall(VecBindToCPU(y, boundtocpu));
6194: PetscCall(VecBindToCPU(z, boundtocpu));
6195: PetscCall(VecBindToCPU(w, boundtocpu));
6196: PetscCall(VecGetSize(x, &N));
6197: PetscCall(VecSet(x, 1. / N));
6198: *n = 0.0;
6199: for (i = 0; i < maxit; i++) {
6200: PetscCall(MatMult(A, x, y));
6201: PetscCall(VecNorm(y, NORM_1, n));
6202: if (PetscDefined(USE_COMPLEX)) {
6203: PetscCall(VecCopy(y, w));
6204: PetscCall(VecAbs(w));
6205: PetscCall(VecPointwiseDivide(w, y, w));
6206: } else PetscCall(VecPointwiseSign(w, y, VEC_SIGN_ZERO_TO_SIGNED_UNIT));
6207: PetscCall(MatMultHermitianTranspose(A, w, z));
6208: PetscCall(VecRealPart(z));
6209: PetscCall(VecNorm(z, NORM_INFINITY, &normz));
6210: PetscCall(VecDot(x, z, &dot));
6211: adot = PetscAbsScalar(dot);
6212: PetscCall(PetscInfo(A, "%s norm it %" PetscInt_FMT " -> %g (%g %g)\n", NormTypes[normtype], i, (double)*n, (double)normz, (double)adot));
6213: if (normz <= adot && i > 0) {
6214: PetscCall(PetscInfo(A, "%s norm converged\n", NormTypes[normtype]));
6215: break;
6216: }
6217: PetscCall(VecAbs(z));
6218: PetscCall(VecMax(z, &j, &normz));
6219: if (j == jold) {
6220: PetscCall(PetscInfo(A, "%s norm it %" PetscInt_FMT " -> breakdown (j==jold)\n", NormTypes[normtype], i));
6221: break;
6222: }
6223: jold = j;
6224: if (i < maxit - 1) PetscCall(VecSetStdBasis(x, j));
6225: }
6226: /* last check */
6227: if (N > 1) {
6228: PetscReal ny;
6230: PetscCall(VecSetFinalNormApp_Private(x));
6231: PetscCall(MatMult(A, x, y));
6232: PetscCall(VecNorm(y, NORM_1, &ny));
6233: ny = 2 * ny / (3 * N);
6234: PetscCall(PetscInfo(A, "%s norm final check: current %g test %g\n", NormTypes[normtype], (double)*n, (double)ny));
6235: *n = PetscMax(*n, ny);
6236: }
6237: PetscCall(MatDestroy(&A));
6238: PetscCall(VecDestroy(&x));
6239: PetscCall(VecDestroy(&w));
6240: PetscCall(VecDestroy(&y));
6241: PetscCall(VecDestroy(&z));
6242: break;
6243: case NORM_2:
6244: if (!hasop) {
6245: PetscCall(MatNormApproximateForwardOnly_Private(A, normtype, maxit, boundtocpu, n));
6246: i = maxit;
6247: break;
6248: }
6249: if (maxit < 0) maxit = 20; /* pure guess */
6250: PetscCall(MatCreateVecs(A, &x, &y));
6251: PetscCall(MatCreateVecs(A, &z, NULL));
6252: PetscCall(VecBindToCPU(x, boundtocpu));
6253: PetscCall(VecBindToCPU(y, boundtocpu));
6254: PetscCall(VecBindToCPU(z, boundtocpu));
6255: PetscCall(VecSetRandom(x, NULL));
6256: PetscCall(VecNormalize(x, NULL));
6257: *n = 0.0;
6258: for (i = 0; i < maxit; i++) {
6259: PetscCall(MatMult(A, x, y));
6260: PetscCall(VecNormalize(y, n));
6261: PetscCall(MatMultHermitianTranspose(A, y, z));
6262: PetscCall(VecNorm(z, NORM_2, &normz));
6263: PetscCall(VecDot(x, z, &dot));
6264: adot = PetscAbsScalar(dot);
6265: PetscCall(PetscInfo(A, "%s norm it %" PetscInt_FMT " -> %g (%g %g)\n", NormTypes[normtype], i, (double)*n, (double)normz, (double)adot));
6266: if (normz <= adot) {
6267: PetscCall(PetscInfo(A, "%s norm converged\n", NormTypes[normtype]));
6268: break;
6269: }
6270: if (i < maxit - 1) {
6271: Vec t;
6273: PetscCall(VecNormalize(z, NULL));
6274: t = x;
6275: x = z;
6276: z = t;
6277: }
6278: }
6279: PetscCall(VecDestroy(&x));
6280: PetscCall(VecDestroy(&y));
6281: PetscCall(VecDestroy(&z));
6282: break;
6283: default:
6284: SETERRQ(PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "%s norm not supported", NormTypes[normtype]);
6285: }
6286: PetscCall(PetscInfo(A, "%s norm %g computed in %" PetscInt_FMT " iterations\n", NormTypes[normtype], (double)*n, i));
6287: PetscFunctionReturn(PETSC_SUCCESS);
6288: }
6290: /*
6291: This variable is used to prevent counting of MatAssemblyBegin() that
6292: are called from within a MatAssemblyEnd().
6293: */
6294: static PetscInt MatAssemblyEnd_InUse = 0;
6295: /*@
6296: MatAssemblyBegin - Begins assembling the matrix. This routine should
6297: be called after completing all calls to `MatSetValues()`.
6299: Collective
6301: Input Parameters:
6302: + mat - the matrix
6303: - type - type of assembly, either `MAT_FLUSH_ASSEMBLY` or `MAT_FINAL_ASSEMBLY`
6305: Level: beginner
6307: Notes:
6308: `MatSetValues()` generally caches the values that belong to other MPI processes. The matrix is ready to
6309: use only after `MatAssemblyBegin()` and `MatAssemblyEnd()` have been called.
6311: Use `MAT_FLUSH_ASSEMBLY` when switching between `ADD_VALUES` and `INSERT_VALUES`
6312: in `MatSetValues()`; use `MAT_FINAL_ASSEMBLY` for the final assembly before
6313: using the matrix.
6315: ALL processes that share a matrix MUST call `MatAssemblyBegin()` and `MatAssemblyEnd()` the SAME NUMBER of times, and each time with the
6316: same flag of `MAT_FLUSH_ASSEMBLY` or `MAT_FINAL_ASSEMBLY` for all processes. Thus you CANNOT locally change from `ADD_VALUES` to `INSERT_VALUES`, that is
6317: a global collective operation requiring all processes that share the matrix.
6319: Space for preallocated nonzeros that is not filled by a call to `MatSetValues()` or a related routine are compressed
6320: out by assembly. If you intend to use that extra space on a subsequent assembly, be sure to insert explicit zeros
6321: before `MAT_FINAL_ASSEMBLY` so the space is not compressed out.
6323: .seealso: [](ch_matrices), `Mat`, `MatAssemblyEnd()`, `MatSetValues()`, `MatAssembled()`
6324: @*/
6325: PetscErrorCode MatAssemblyBegin(Mat mat, MatAssemblyType type)
6326: {
6327: PetscFunctionBegin;
6330: MatCheckPreallocated(mat, 1);
6331: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix. Did you forget to call MatSetUnfactored()?");
6332: if (mat->assembled) {
6333: mat->was_assembled = PETSC_TRUE;
6334: mat->assembled = PETSC_FALSE;
6335: }
6337: if (!MatAssemblyEnd_InUse) {
6338: PetscCall(PetscLogEventBegin(MAT_AssemblyBegin, mat, 0, 0, 0));
6339: PetscTryTypeMethod(mat, assemblybegin, type);
6340: PetscCall(PetscLogEventEnd(MAT_AssemblyBegin, mat, 0, 0, 0));
6341: } else PetscTryTypeMethod(mat, assemblybegin, type);
6342: PetscFunctionReturn(PETSC_SUCCESS);
6343: }
6345: /*@
6346: MatAssembled - Indicates if a matrix has been assembled and is ready for
6347: use; for example, in matrix-vector product.
6349: Not Collective
6351: Input Parameter:
6352: . mat - the matrix
6354: Output Parameter:
6355: . assembled - `PETSC_TRUE` or `PETSC_FALSE`
6357: Level: advanced
6359: .seealso: [](ch_matrices), `Mat`, `MatAssemblyEnd()`, `MatSetValues()`, `MatAssemblyBegin()`
6360: @*/
6361: PetscErrorCode MatAssembled(Mat mat, PetscBool *assembled)
6362: {
6363: PetscFunctionBegin;
6365: PetscAssertPointer(assembled, 2);
6366: *assembled = mat->assembled;
6367: PetscFunctionReturn(PETSC_SUCCESS);
6368: }
6370: /*@
6371: MatAssemblyEnd - Completes assembling the matrix. This routine should
6372: be called after `MatAssemblyBegin()`.
6374: Collective
6376: Input Parameters:
6377: + mat - the matrix
6378: - type - type of assembly, either `MAT_FLUSH_ASSEMBLY` or `MAT_FINAL_ASSEMBLY`
6380: Options Database Key:
6381: . -mat_view viewer_specification - Displays the matrix during this function call. See `PetscOptionsCreateViewer()` for the values of `viewer_specification`
6383: Level: beginner
6385: .seealso: [](ch_matrices), `Mat`, `MatAssemblyBegin()`, `MatSetValues()`, `PetscDrawOpenX()`, `PetscDrawCreate()`, `MatView()`, `MatAssembled()`, `PetscViewerSocketOpen()`,
6386: `MatViewFromOptions()`, `PetscObjectViewFromOptions()`, `PetscOptionsCreateViewer()`
6387: @*/
6388: PetscErrorCode MatAssemblyEnd(Mat mat, MatAssemblyType type)
6389: {
6390: static PetscInt inassm = 0;
6391: PetscBool flg = PETSC_FALSE;
6393: PetscFunctionBegin;
6397: inassm++;
6398: MatAssemblyEnd_InUse++;
6399: if (MatAssemblyEnd_InUse == 1) { /* Do the logging only the first time through */
6400: PetscCall(PetscLogEventBegin(MAT_AssemblyEnd, mat, 0, 0, 0));
6401: PetscTryTypeMethod(mat, assemblyend, type);
6402: PetscCall(PetscLogEventEnd(MAT_AssemblyEnd, mat, 0, 0, 0));
6403: } else PetscTryTypeMethod(mat, assemblyend, type);
6405: /* Flush assembly is not a true assembly */
6406: if (type != MAT_FLUSH_ASSEMBLY) {
6407: if (mat->num_ass) {
6408: if (!mat->symmetry_eternal) {
6409: mat->symmetric = PETSC_BOOL3_UNKNOWN;
6410: mat->hermitian = PETSC_BOOL3_UNKNOWN;
6411: }
6412: if (!mat->structural_symmetry_eternal && mat->ass_nonzerostate != mat->nonzerostate) mat->structurally_symmetric = PETSC_BOOL3_UNKNOWN;
6413: if (!mat->spd_eternal) mat->spd = PETSC_BOOL3_UNKNOWN;
6414: }
6415: mat->num_ass++;
6416: mat->assembled = PETSC_TRUE;
6417: mat->ass_nonzerostate = mat->nonzerostate;
6418: }
6420: mat->insertmode = NOT_SET_VALUES;
6421: MatAssemblyEnd_InUse--;
6422: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
6423: if (inassm == 1 && type != MAT_FLUSH_ASSEMBLY) {
6424: PetscCall(MatViewFromOptions(mat, NULL, "-mat_view"));
6426: if (mat->checksymmetryonassembly) {
6427: PetscCall(MatIsSymmetric(mat, mat->checksymmetrytol, &flg));
6428: if (flg) {
6429: PetscCall(PetscPrintf(PetscObjectComm((PetscObject)mat), "Matrix is symmetric (tolerance %g)\n", (double)mat->checksymmetrytol));
6430: } else {
6431: PetscCall(PetscPrintf(PetscObjectComm((PetscObject)mat), "Matrix is not symmetric (tolerance %g)\n", (double)mat->checksymmetrytol));
6432: }
6433: }
6434: if (mat->nullsp && mat->checknullspaceonassembly) PetscCall(MatNullSpaceTest(mat->nullsp, mat, NULL));
6435: }
6436: inassm--;
6437: PetscFunctionReturn(PETSC_SUCCESS);
6438: }
6440: // PetscClangLinter pragma disable: -fdoc-section-header-unknown
6441: /*@
6442: MatSetOption - Sets a parameter option for a matrix. Some options
6443: may be specific to certain storage formats. Some options
6444: determine how values will be inserted (or added). Sorted,
6445: row-oriented input will generally assemble the fastest. The default
6446: is row-oriented.
6448: Logically Collective for certain operations, such as `MAT_SPD`, not collective for `MAT_ROW_ORIENTED`, see `MatOption`
6450: Input Parameters:
6451: + mat - the matrix
6452: . op - the option, one of those listed below (and possibly others),
6453: - flg - turn the option on (`PETSC_TRUE`) or off (`PETSC_FALSE`)
6455: Options Describing Matrix Structure:
6456: + `MAT_SPD` - symmetric positive definite
6457: . `MAT_SYMMETRIC` - symmetric in terms of both structure and value
6458: . `MAT_HERMITIAN` - transpose is the complex conjugation
6459: . `MAT_STRUCTURALLY_SYMMETRIC` - symmetric nonzero structure
6460: . `MAT_SYMMETRY_ETERNAL` - indicates the symmetry (or Hermitian structure) or its absence will persist through any changes to the matrix
6461: . `MAT_STRUCTURAL_SYMMETRY_ETERNAL` - indicates the structural symmetry or its absence will persist through any changes to the matrix
6462: . `MAT_SPD_ETERNAL` - indicates the value of `MAT_SPD` (true or false) will persist through any changes to the matrix
6464: These are not really options of the matrix, they are knowledge about the structure of the matrix that users may provide so that they
6465: do not need to be computed (usually at a high cost)
6467: Options For Use with `MatSetValues()` and `MatGetValues()`:
6468: Insert a logically dense subblock, which can be
6469: . `MAT_ROW_ORIENTED` - row-oriented (default)
6471: These options reflect the data you pass in with `MatSetValues()` or receive with `MatGetValues()`; it has
6472: nothing to do with how the data is stored internally in the matrix
6473: data structure.
6475: When (re)assembling a matrix, we can restrict the input for
6476: efficiency/debugging purposes. These options include
6477: . `MAT_NEW_NONZERO_LOCATIONS` - additional insertions will be allowed if they generate a new nonzero (slow)
6478: . `MAT_FORCE_DIAGONAL_ENTRIES` - forces diagonal entries to be allocated
6479: . `MAT_IGNORE_OFF_PROC_ENTRIES` - drops off-process entries
6480: . `MAT_NEW_NONZERO_LOCATION_ERR` - generates an error for new matrix entry
6481: . `MAT_USE_HASH_TABLE` - uses a hash table to speed up matrix assembly
6482: . `MAT_NO_OFF_PROC_ENTRIES` - you know each process will only set values for its own rows, will generate an error if
6483: any process sets values for another process. This avoids all reductions in the MatAssembly routines and thus improves
6484: performance for very large process counts.
6485: - `MAT_SUBSET_OFF_PROC_ENTRIES` - you know that the first assembly after setting this flag will set a superset
6486: of the off-process entries required for all subsequent assemblies. This avoids a rendezvous step in the MatAssembly
6487: functions, instead sending only neighbor messages.
6489: Level: intermediate
6491: Notes:
6492: Except for `MAT_UNUSED_NONZERO_LOCATION_ERR` and `MAT_ROW_ORIENTED` all processes that share the matrix must pass the same value in flg!
6494: Some options are relevant only for particular matrix types and
6495: are thus ignored by others. Other options are not supported by
6496: certain matrix types and will generate an error message if set.
6498: Once `MAT_STRUCTURE_ONLY` has been set to `PETSC_TRUE`, it cannot be set back to `PETSC_FALSE`.
6500: The matrix properties must be consistent with one another. With real scalars, `MAT_SPD` implies `MAT_SYMMETRIC`, and
6501: both `MAT_SYMMETRIC` and `MAT_HERMITIAN` imply `MAT_STRUCTURALLY_SYMMETRIC`. Setting a property that contradicts
6502: another known property generates an error. Still with real scalars, setting either `MAT_SYMMETRIC` or `MAT_HERMITIAN`
6503: also sets the other to the same value.
6505: If using Fortran to compute a matrix, one may need to use the column-oriented
6506: option (or convert to the row-oriented format).
6508: `MAT_NEW_NONZERO_LOCATIONS` set to `PETSC_FALSE` indicates that any add or insertion
6509: that would generate a new entry in the nonzero structure is instead
6510: ignored. Thus, if memory has not already been allocated for this particular
6511: data, then the insertion is ignored. For dense matrices, in which
6512: the entire array is allocated, no entries are ever ignored.
6513: Set after the first `MatAssemblyEnd()`. If this option is set, then the `MatAssemblyBegin()`/`MatAssemblyEnd()` processes has one less global reduction
6515: `MAT_NEW_NONZERO_LOCATION_ERR` set to `PETSC_TRUE` indicates that any add or insertion
6516: that would generate a new entry in the nonzero structure instead produces
6517: an error. (Currently supported for `MATAIJ` and `MATBAIJ` formats only.) If this option is set, then the `MatAssemblyBegin()`/`MatAssemblyEnd()` processes has one less global reduction
6519: `MAT_NEW_NONZERO_ALLOCATION_ERR` set to `PETSC_TRUE` indicates that any add or insertion
6520: that would generate a new entry that has not been preallocated will
6521: instead produce an error. (Currently supported for `MATAIJ` and `MATBAIJ` formats
6522: only.) This is a useful flag when debugging matrix memory preallocation.
6523: If this option is set, then the `MatAssemblyBegin()`/`MatAssemblyEnd()` processes has one less global reduction
6525: `MAT_IGNORE_OFF_PROC_ENTRIES` set to `PETSC_TRUE` indicates entries destined for
6526: other processes should be dropped, rather than stashed.
6527: This is useful if you know that the "owning" process is also
6528: always generating the correct matrix entries, so that PETSc need
6529: not transfer duplicate entries generated on another process.
6531: `MAT_USE_HASH_TABLE` indicates that a hash table be used to improve the
6532: searches during matrix assembly. When this flag is set, the hash table
6533: is created during the first matrix assembly. This hash table is
6534: used the next time through, during `MatSetValues()`/`MatSetValuesBlocked()`
6535: to improve the searching of indices. `MAT_NEW_NONZERO_LOCATIONS` flag
6536: should be used with `MAT_USE_HASH_TABLE` flag. This option is currently
6537: supported by `MATMPIBAIJ` format only.
6539: `MAT_KEEP_NONZERO_PATTERN` indicates when `MatZeroRows()` is called the zeroed entries
6540: are kept in the nonzero structure. This flag is not used for `MatZeroRowsColumns()`
6542: `MAT_IGNORE_ZERO_ENTRIES` - for `MATAIJ`, `MATSELL`, and `MATIS` matrices this will stop zero values
6543: from creating a zero location in the matrix. A zero on the diagonal is exempt and still creates its
6544: location, so that operations needing a full diagonal, such as `MatSOR()`, keep working. The exemption
6545: covers the diagonal portion of the owning process's rows; a zero added with `ADD_VALUES` to a row owned
6546: by another process is dropped as it is stashed and never reaches that test. A matrix assembled without
6547: preallocation, through `MatSetUp()`, decides the exemption from process-local indices, so it does not
6548: hold there when the row and column layouts differ
6550: `MAT_USE_INODES` - indicates using inode version of the code - works with `MATAIJ` matrix types
6552: `MAT_NO_OFF_PROC_ZERO_ROWS` - you know each process will only zero its own rows. This avoids all reductions in the
6553: zero row routines and thus improves performance for very large process counts.
6555: `MAT_IGNORE_LOWER_TRIANGULAR` - For `MATSBAIJ` matrices will ignore any insertions you make in the lower triangular
6556: part of the matrix (since they should match the upper triangular part).
6558: `MAT_SORTED_FULL` - each process provides exactly its local rows; all column indices for a given row are passed in a
6559: single call to `MatSetValues()`, preallocation is perfect, row-oriented, `INSERT_VALUES` is used. Common
6560: with finite difference schemes with non-periodic boundary conditions.
6562: Developer Note:
6563: `MAT_SYMMETRY_ETERNAL`, `MAT_STRUCTURAL_SYMMETRY_ETERNAL`, and `MAT_SPD_ETERNAL` are used by `MatAssemblyEnd()` and in other
6564: places where otherwise the value of `MAT_SYMMETRIC`, `MAT_STRUCTURALLY_SYMMETRIC` or `MAT_SPD` would need to be changed back
6565: to `PETSC_BOOL3_UNKNOWN` because the matrix values had changed so the code cannot be certain that the related property had
6566: not changed.
6568: .seealso: [](ch_matrices), `MatOption`, `Mat`, `MatGetOption()`
6569: @*/
6570: PetscErrorCode MatSetOption(Mat mat, MatOption op, PetscBool flg)
6571: {
6572: PetscFunctionBegin;
6574: if (op > 0) {
6577: }
6579: PetscCheck(((int)op) > MAT_OPTION_MIN && ((int)op) < MAT_OPTION_MAX, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_OUTOFRANGE, "Options %d is out of range", (int)op);
6581: switch (op) {
6582: case MAT_FORCE_DIAGONAL_ENTRIES:
6583: mat->force_diagonals = flg;
6584: PetscFunctionReturn(PETSC_SUCCESS);
6585: case MAT_NO_OFF_PROC_ENTRIES:
6586: mat->nooffprocentries = flg;
6587: PetscFunctionReturn(PETSC_SUCCESS);
6588: case MAT_SUBSET_OFF_PROC_ENTRIES:
6589: mat->assembly_subset = flg;
6590: if (!mat->assembly_subset) { /* See the same logic in VecAssembly wrt VEC_SUBSET_OFF_PROC_ENTRIES */
6591: #if !PetscDefined(HAVE_MPIUNI)
6592: PetscCall(MatStashScatterDestroy_BTS(&mat->stash));
6593: #endif
6594: mat->stash.first_assembly_done = PETSC_FALSE;
6595: }
6596: PetscFunctionReturn(PETSC_SUCCESS);
6597: case MAT_NO_OFF_PROC_ZERO_ROWS:
6598: mat->nooffproczerorows = flg;
6599: PetscFunctionReturn(PETSC_SUCCESS);
6600: case MAT_SPD:
6601: PetscCheck(PetscDefined(USE_COMPLEX) || !flg || mat->symmetric != PETSC_BOOL3_FALSE, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_INCOMP, "Cannot set MAT_SPD to PETSC_TRUE when MAT_SYMMETRIC is PETSC_FALSE");
6602: PetscCheck(!flg || mat->structurally_symmetric != PETSC_BOOL3_FALSE, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_INCOMP, "Cannot set MAT_SPD to PETSC_TRUE when MAT_STRUCTURALLY_SYMMETRIC is PETSC_FALSE");
6603: if (flg) {
6604: mat->spd = PETSC_BOOL3_TRUE;
6605: mat->symmetric = PETSC_BOOL3_TRUE;
6606: mat->structurally_symmetric = PETSC_BOOL3_TRUE;
6607: if (!PetscDefined(USE_COMPLEX)) mat->hermitian = PETSC_BOOL3_TRUE;
6608: } else mat->spd = PETSC_BOOL3_FALSE;
6609: break;
6610: case MAT_SYMMETRIC:
6611: PetscCheck(PetscDefined(USE_COMPLEX) || flg || mat->spd != PETSC_BOOL3_TRUE, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_INCOMP, "Cannot set MAT_SYMMETRIC to PETSC_FALSE when MAT_SPD is PETSC_TRUE");
6612: PetscCheck(!flg || mat->structurally_symmetric != PETSC_BOOL3_FALSE, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_INCOMP, "Cannot set MAT_SYMMETRIC to PETSC_TRUE when MAT_STRUCTURALLY_SYMMETRIC is PETSC_FALSE");
6613: mat->symmetric = PetscBoolToBool3(flg);
6614: if (flg) mat->structurally_symmetric = PETSC_BOOL3_TRUE;
6615: if (!PetscDefined(USE_COMPLEX)) mat->hermitian = PetscBoolToBool3(flg);
6616: break;
6617: case MAT_HERMITIAN:
6618: PetscCheck(PetscDefined(USE_COMPLEX) || flg || mat->spd != PETSC_BOOL3_TRUE, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_INCOMP, "Cannot set MAT_HERMITIAN to PETSC_FALSE when MAT_SPD is PETSC_TRUE");
6619: PetscCheck(!flg || mat->structurally_symmetric != PETSC_BOOL3_FALSE, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_INCOMP, "Cannot set MAT_HERMITIAN to PETSC_TRUE when MAT_STRUCTURALLY_SYMMETRIC is PETSC_FALSE");
6620: mat->hermitian = PetscBoolToBool3(flg);
6621: if (flg) mat->structurally_symmetric = PETSC_BOOL3_TRUE;
6622: if (!PetscDefined(USE_COMPLEX)) mat->symmetric = PetscBoolToBool3(flg);
6623: break;
6624: case MAT_STRUCTURALLY_SYMMETRIC:
6625: PetscCheck(flg || mat->spd != PETSC_BOOL3_TRUE, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_INCOMP, "Cannot set MAT_STRUCTURALLY_SYMMETRIC to PETSC_FALSE when MAT_SPD is PETSC_TRUE");
6626: PetscCheck(flg || mat->symmetric != PETSC_BOOL3_TRUE, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_INCOMP, "Cannot set MAT_STRUCTURALLY_SYMMETRIC to PETSC_FALSE when MAT_SYMMETRIC is PETSC_TRUE");
6627: PetscCheck(flg || mat->hermitian != PETSC_BOOL3_TRUE, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_INCOMP, "Cannot set MAT_STRUCTURALLY_SYMMETRIC to PETSC_FALSE when MAT_HERMITIAN is PETSC_TRUE");
6628: mat->structurally_symmetric = PetscBoolToBool3(flg);
6629: break;
6630: case MAT_SYMMETRY_ETERNAL:
6631: PetscCheck(mat->symmetric != PETSC_BOOL3_UNKNOWN, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Cannot set MAT_SYMMETRY_ETERNAL without first setting MAT_SYMMETRIC to PETSC_TRUE or PETSC_FALSE");
6632: mat->symmetry_eternal = flg;
6633: if (flg) mat->structural_symmetry_eternal = PETSC_TRUE;
6634: break;
6635: case MAT_STRUCTURAL_SYMMETRY_ETERNAL:
6636: PetscCheck(mat->structurally_symmetric != PETSC_BOOL3_UNKNOWN, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Cannot set MAT_STRUCTURAL_SYMMETRY_ETERNAL without first setting MAT_STRUCTURALLY_SYMMETRIC to PETSC_TRUE or PETSC_FALSE");
6637: mat->structural_symmetry_eternal = flg;
6638: break;
6639: case MAT_SPD_ETERNAL:
6640: PetscCheck(mat->spd != PETSC_BOOL3_UNKNOWN, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Cannot set MAT_SPD_ETERNAL without first setting MAT_SPD to PETSC_TRUE or PETSC_FALSE");
6641: mat->spd_eternal = flg;
6642: if (flg) {
6643: mat->structural_symmetry_eternal = PETSC_TRUE;
6644: mat->symmetry_eternal = PETSC_TRUE;
6645: }
6646: break;
6647: case MAT_STRUCTURE_ONLY:
6648: PetscCheck(flg || !mat->structure_only, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Cannot set MAT_STRUCTURE_ONLY to PETSC_FALSE after it has been set to PETSC_TRUE");
6649: mat->structure_only = flg;
6650: break;
6651: case MAT_SORTED_FULL:
6652: mat->sortedfull = flg;
6653: break;
6654: default:
6655: break;
6656: }
6657: PetscCheck((op != MAT_ROW_ORIENTED) || ((PetscObject)mat)->type_name, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "The matrix type must be set (MatSetType()) before setting this option");
6658: PetscTryTypeMethod(mat, setoption, op, flg);
6659: PetscFunctionReturn(PETSC_SUCCESS);
6660: }
6662: /*@
6663: MatGetOption - Gets a parameter option that has been set for a matrix.
6665: Logically Collective
6667: Input Parameters:
6668: + mat - the matrix
6669: - op - the option, this only responds to certain options, check the code for which ones
6671: Output Parameter:
6672: . flg - turn the option on (`PETSC_TRUE`) or off (`PETSC_FALSE`)
6674: Level: intermediate
6676: Notes:
6677: Can only be called after `MatSetSizes()` and `MatSetType()` have been set.
6679: Certain option values may be unknown, for those use the routines `MatIsSymmetric()`, `MatIsHermitian()`, `MatIsStructurallySymmetric()`, or
6680: `MatIsSymmetricKnown()`, `MatIsHermitianKnown()`, `MatIsStructurallySymmetricKnown()`
6682: .seealso: [](ch_matrices), `Mat`, `MatOption`, `MatSetOption()`, `MatIsSymmetric()`, `MatIsHermitian()`, `MatIsStructurallySymmetric()`,
6683: `MatIsSymmetricKnown()`, `MatIsHermitianKnown()`, `MatIsStructurallySymmetricKnown()`
6684: @*/
6685: PetscErrorCode MatGetOption(Mat mat, MatOption op, PetscBool *flg)
6686: {
6687: PetscFunctionBegin;
6691: PetscCheck(((int)op) > MAT_OPTION_MIN && ((int)op) < MAT_OPTION_MAX, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_OUTOFRANGE, "Options %d is out of range", (int)op);
6692: PetscCheck(((PetscObject)mat)->type_name, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_TYPENOTSET, "Cannot get options until type and size have been set, see MatSetType() and MatSetSizes()");
6694: switch (op) {
6695: case MAT_NO_OFF_PROC_ENTRIES:
6696: *flg = mat->nooffprocentries;
6697: break;
6698: case MAT_NO_OFF_PROC_ZERO_ROWS:
6699: *flg = mat->nooffproczerorows;
6700: break;
6701: case MAT_SYMMETRIC:
6702: SETERRQ(PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Use MatIsSymmetric() or MatIsSymmetricKnown()");
6703: break;
6704: case MAT_HERMITIAN:
6705: SETERRQ(PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Use MatIsHermitian() or MatIsHermitianKnown()");
6706: break;
6707: case MAT_STRUCTURALLY_SYMMETRIC:
6708: SETERRQ(PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Use MatIsStructurallySymmetric() or MatIsStructurallySymmetricKnown()");
6709: break;
6710: case MAT_SPD:
6711: SETERRQ(PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "Use MatIsSPDKnown()");
6712: break;
6713: case MAT_SYMMETRY_ETERNAL:
6714: *flg = mat->symmetry_eternal;
6715: break;
6716: case MAT_STRUCTURAL_SYMMETRY_ETERNAL:
6717: *flg = mat->symmetry_eternal;
6718: break;
6719: default:
6720: break;
6721: }
6722: PetscFunctionReturn(PETSC_SUCCESS);
6723: }
6725: /*@
6726: MatZeroEntries - Zeros all entries of a matrix. For sparse matrices
6727: this routine retains the old nonzero structure.
6729: Logically Collective
6731: Input Parameter:
6732: . mat - the matrix
6734: Level: intermediate
6736: Note:
6737: If the matrix was not preallocated then a default, likely poor preallocation will be set in the matrix, so this should be called after the preallocation phase.
6738: See the Performance chapter of the users manual for information on preallocating matrices.
6739: For matrices with the `MAT_STRUCTURE_ONLY` option set to true, this routine leaves the structure and object state unchanged because no numerical values are stored.
6741: .seealso: [](ch_matrices), `Mat`, `MatZeroRows()`, `MatZeroRowsColumns()`
6742: @*/
6743: PetscErrorCode MatZeroEntries(Mat mat)
6744: {
6745: PetscFunctionBegin;
6748: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
6749: PetscCheck(mat->insertmode == NOT_SET_VALUES, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for matrices where you have set values but not yet assembled");
6750: MatCheckPreallocated(mat, 1);
6752: if (mat->structure_only == PETSC_FALSE) {
6753: PetscCall(PetscLogEventBegin(MAT_ZeroEntries, mat, 0, 0, 0));
6754: PetscUseTypeMethod(mat, zeroentries);
6755: PetscCall(PetscLogEventEnd(MAT_ZeroEntries, mat, 0, 0, 0));
6756: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
6757: }
6758: PetscFunctionReturn(PETSC_SUCCESS);
6759: }
6761: /*@
6762: MatZeroRowsColumns - Zeros all entries (except possibly the main diagonal)
6763: of a set of rows and columns of a matrix.
6765: Collective
6767: Input Parameters:
6768: + mat - the matrix
6769: . numRows - the number of rows/columns to zero
6770: . rows - the global row indices
6771: . diag - value put in the diagonal of the eliminated rows
6772: . x - optional vector of the solution for zeroed rows (other entries in vector are not used), these must be set before this call
6773: - b - optional vector of the right-hand side, that will be adjusted by provided solution entries
6775: Level: intermediate
6777: Notes:
6778: This routine, along with `MatZeroRows()`, is typically used to eliminate known Dirichlet boundary conditions from a linear system.
6780: For each zeroed row, the value of the corresponding `b` is set to diag times the value of the corresponding `x`.
6781: The other entries of `b` will be adjusted by the known values of `x` times the corresponding matrix entries in the columns that are being eliminated
6783: If the resulting linear system is to be solved with `KSP` then one can (but does not have to) call `KSPSetInitialGuessNonzero()` to allow the
6784: Krylov method to take advantage of the known solution on the zeroed rows.
6786: For the parallel case, all processes that share the matrix (i.e.,
6787: those in the communicator used for matrix creation) MUST call this
6788: routine, regardless of whether any rows being zeroed are owned by
6789: them.
6791: Unlike `MatZeroRows()`, this ignores the `MAT_KEEP_NONZERO_PATTERN` option value set with `MatSetOption()`, it merely zeros those entries in the matrix, but never
6792: removes them from the nonzero pattern. The nonzero pattern of the matrix can still change if a nonzero needs to be inserted on a diagonal entry that was previously
6793: missing.
6795: Each process can indicate any rows in the entire matrix to be zeroed (i.e. each process does NOT have to
6796: list only rows local to itself).
6798: The option `MAT_NO_OFF_PROC_ZERO_ROWS` does not apply to this routine.
6800: .seealso: [](ch_matrices), `Mat`, `MatZeroRowsIS()`, `MatZeroRows()`, `MatZeroRowsLocalIS()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
6801: `MatZeroRowsColumnsLocal()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRowsColumnsIS()`, `MatZeroRowsColumnsStencil()`
6802: @*/
6803: PetscErrorCode MatZeroRowsColumns(Mat mat, PetscInt numRows, const PetscInt rows[], PetscScalar diag, Vec x, Vec b)
6804: {
6805: PetscFunctionBegin;
6808: if (numRows) PetscAssertPointer(rows, 3);
6809: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
6810: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
6811: MatCheckPreallocated(mat, 1);
6813: PetscUseTypeMethod(mat, zerorowscolumns, numRows, rows, diag, x, b);
6814: PetscCall(MatViewFromOptions(mat, NULL, "-mat_view"));
6815: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
6816: PetscFunctionReturn(PETSC_SUCCESS);
6817: }
6819: /*@
6820: MatZeroRowsColumnsIS - Zeros all entries (except possibly the main diagonal)
6821: of a set of rows and columns of a matrix.
6823: Collective
6825: Input Parameters:
6826: + mat - the matrix
6827: . is - the rows to zero
6828: . diag - value put in all diagonals of eliminated rows (0.0 will even eliminate diagonal entry)
6829: . x - optional vector of solutions for zeroed rows (other entries in vector are not used)
6830: - b - optional vector of right-hand side, that will be adjusted by provided solution
6832: Level: intermediate
6834: Note:
6835: See `MatZeroRowsColumns()` for details on how this routine operates.
6837: .seealso: [](ch_matrices), `Mat`, `MatZeroRowsIS()`, `MatZeroRowsColumns()`, `MatZeroRowsLocalIS()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
6838: `MatZeroRowsColumnsLocal()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRows()`, `MatZeroRowsColumnsStencil()`
6839: @*/
6840: PetscErrorCode MatZeroRowsColumnsIS(Mat mat, IS is, PetscScalar diag, Vec x, Vec b)
6841: {
6842: PetscInt numRows;
6843: const PetscInt *rows;
6845: PetscFunctionBegin;
6850: PetscCall(ISGetLocalSize(is, &numRows));
6851: PetscCall(ISGetIndices(is, &rows));
6852: PetscCall(MatZeroRowsColumns(mat, numRows, rows, diag, x, b));
6853: PetscCall(ISRestoreIndices(is, &rows));
6854: PetscFunctionReturn(PETSC_SUCCESS);
6855: }
6857: /*@
6858: MatZeroRows - Zeros all entries (except possibly the main diagonal)
6859: of a set of rows of a matrix.
6861: Collective
6863: Input Parameters:
6864: + mat - the matrix
6865: . numRows - the number of rows to zero
6866: . rows - the global row indices
6867: . diag - value put in the diagonal of the zeroed rows
6868: . x - optional vector of solutions for zeroed rows (other entries in vector are not used), these must be set before this call
6869: - b - optional vector of right-hand side, that will be adjusted by provided solution entries
6871: Level: intermediate
6873: Notes:
6874: This routine, along with `MatZeroRowsColumns()`, is typically used to eliminate known Dirichlet boundary conditions from a linear system.
6876: For each zeroed row, the value of the corresponding `b` is set to `diag` times the value of the corresponding `x`.
6878: If the resulting linear system is to be solved with `KSP` then one can (but does not have to) call `KSPSetInitialGuessNonzero()` to allow the
6879: Krylov method to take advantage of the known solution on the zeroed rows.
6881: May be followed by using a `PC` of type `PCREDISTRIBUTE` to solve the reduced problem (`PCDISTRIBUTE` completely eliminates the zeroed rows and their corresponding columns)
6882: from the matrix.
6884: Unlike `MatZeroRowsColumns()` for the `MATAIJ` and `MATBAIJ` matrix formats this removes the old nonzero structure, from the eliminated rows of the matrix
6885: but does not release memory. Because of this removal matrix-vector products with the adjusted matrix will be a bit faster. For the dense
6886: formats this does not alter the nonzero structure.
6888: If the option `MatSetOption`(mat,`MAT_KEEP_NONZERO_PATTERN`,`PETSC_TRUE`) the nonzero structure
6889: of the matrix is not changed the values are
6890: merely zeroed.
6892: The user can set a value in the diagonal entry (or for the `MATAIJ` format
6893: formats can optionally remove the main diagonal entry from the
6894: nonzero structure as well, by passing 0.0 as the final argument).
6896: For the parallel case, all processes that share the matrix (i.e.,
6897: those in the communicator used for matrix creation) MUST call this
6898: routine, regardless of whether any rows being zeroed are owned by
6899: them.
6901: Each process can indicate any rows in the entire matrix to be zeroed (i.e. each process does NOT have to
6902: list only rows local to itself).
6904: You can call `MatSetOption`(mat,`MAT_NO_OFF_PROC_ZERO_ROWS`,`PETSC_TRUE`) if each process indicates only rows it
6905: owns that are to be zeroed. This saves a global synchronization in the implementation.
6907: .seealso: [](ch_matrices), `Mat`, `MatZeroRowsIS()`, `MatZeroRowsColumns()`, `MatZeroRowsLocalIS()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
6908: `MatZeroRowsColumnsLocal()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRowsColumnsIS()`, `MatZeroRowsColumnsStencil()`, `PCREDISTRIBUTE`, `MAT_KEEP_NONZERO_PATTERN`
6909: @*/
6910: PetscErrorCode MatZeroRows(Mat mat, PetscInt numRows, const PetscInt rows[], PetscScalar diag, Vec x, Vec b)
6911: {
6912: PetscFunctionBegin;
6915: if (numRows) PetscAssertPointer(rows, 3);
6916: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
6917: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
6918: MatCheckPreallocated(mat, 1);
6920: PetscUseTypeMethod(mat, zerorows, numRows, rows, diag, x, b);
6921: PetscCall(MatViewFromOptions(mat, NULL, "-mat_view"));
6922: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
6923: PetscFunctionReturn(PETSC_SUCCESS);
6924: }
6926: /*@
6927: MatZeroRowsIS - Zeros all entries (except possibly the main diagonal)
6928: of a set of rows of a matrix indicated by an `IS`
6930: Collective
6932: Input Parameters:
6933: + mat - the matrix
6934: . is - index set, `IS`, of rows to remove (if `NULL` then no row is removed)
6935: . diag - value put in all diagonals of eliminated rows
6936: . x - optional vector of solutions for zeroed rows (other entries in vector are not used)
6937: - b - optional vector of right-hand side, that will be adjusted by provided solution
6939: Level: intermediate
6941: Note:
6942: See `MatZeroRows()` for details on how this routine operates.
6944: .seealso: [](ch_matrices), `Mat`, `MatZeroRows()`, `MatZeroRowsColumns()`, `MatZeroRowsLocalIS()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
6945: `MatZeroRowsColumnsLocal()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRowsColumnsIS()`, `MatZeroRowsColumnsStencil()`, `IS`
6946: @*/
6947: PetscErrorCode MatZeroRowsIS(Mat mat, IS is, PetscScalar diag, Vec x, Vec b)
6948: {
6949: PetscInt numRows = 0;
6950: const PetscInt *rows = NULL;
6952: PetscFunctionBegin;
6955: if (is) {
6957: PetscCall(ISGetLocalSize(is, &numRows));
6958: PetscCall(ISGetIndices(is, &rows));
6959: }
6960: PetscCall(MatZeroRows(mat, numRows, rows, diag, x, b));
6961: if (is) PetscCall(ISRestoreIndices(is, &rows));
6962: PetscFunctionReturn(PETSC_SUCCESS);
6963: }
6965: /*@
6966: MatZeroRowsStencil - Zeros all entries (except possibly the main diagonal)
6967: of a set of rows of a matrix indicated by a `MatStencil`. These rows must be local to the process.
6969: Collective
6971: Input Parameters:
6972: + mat - the matrix
6973: . numRows - the number of rows to remove
6974: . rows - the grid coordinates (and component number when dof > 1) for matrix rows indicated by an array of `MatStencil`
6975: . diag - value put in all diagonals of eliminated rows (0.0 will even eliminate diagonal entry)
6976: . x - optional vector of solutions for zeroed rows (other entries in vector are not used)
6977: - b - optional vector of right-hand side, that will be adjusted by provided solution
6979: Level: intermediate
6981: Notes:
6982: See `MatZeroRows()` for details on how this routine operates.
6984: The grid coordinates are across the entire grid, not just the local portion
6986: For periodic boundary conditions use negative indices for values to the left (below 0; that are to be
6987: obtained by wrapping values from right edge). For values to the right of the last entry using that index plus one
6988: etc to obtain values that obtained by wrapping the values from the left edge. This does not work for anything but the
6989: `DM_BOUNDARY_PERIODIC` boundary type.
6991: For indices that don't mean anything for your case (like the `k` index when working in 2d) or the `c` index when you have
6992: a single value per point) you can skip filling those indices.
6994: Fortran Note:
6995: `idxm` and `idxn` should be declared as
6996: .vb
6997: MatStencil idxm(4, m)
6998: .ve
6999: and the values inserted using
7000: .vb
7001: idxm(MatStencil_i, 1) = i
7002: idxm(MatStencil_j, 1) = j
7003: idxm(MatStencil_k, 1) = k
7004: idxm(MatStencil_c, 1) = c
7005: etc
7006: .ve
7008: .seealso: [](ch_matrices), `Mat`, `MatStencil`, `MatZeroRowsIS()`, `MatZeroRowsColumns()`, `MatZeroRowsLocalIS()`, `MatZeroRows()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
7009: `MatZeroRowsColumnsLocal()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRowsColumnsIS()`, `MatZeroRowsColumnsStencil()`
7010: @*/
7011: PetscErrorCode MatZeroRowsStencil(Mat mat, PetscInt numRows, const MatStencil rows[], PetscScalar diag, Vec x, Vec b)
7012: {
7013: PetscInt dim = mat->stencil.dim;
7014: PetscInt sdim = dim - (1 - (PetscInt)mat->stencil.noc);
7015: PetscInt *dims = mat->stencil.dims + 1;
7016: PetscInt *starts = mat->stencil.starts;
7017: PetscInt *dxm = (PetscInt *)rows;
7018: PetscInt *jdxm, i, j, tmp, numNewRows = 0;
7020: PetscFunctionBegin;
7023: if (numRows) PetscAssertPointer(rows, 3);
7025: PetscCall(PetscMalloc1(numRows, &jdxm));
7026: for (i = 0; i < numRows; ++i) {
7027: /* Skip unused dimensions (they are ordered k, j, i, c) */
7028: for (j = 0; j < 3 - sdim; ++j) dxm++;
7029: /* Local index in X dir */
7030: tmp = *dxm++ - starts[0];
7031: /* Loop over remaining dimensions */
7032: for (j = 0; j < dim - 1; ++j) {
7033: /* If nonlocal, set index to be negative */
7034: if ((*dxm++ - starts[j + 1]) < 0 || tmp < 0) tmp = PETSC_INT_MIN;
7035: /* Update local index */
7036: else tmp = tmp * dims[j] + *(dxm - 1) - starts[j + 1];
7037: }
7038: /* Skip component slot if necessary */
7039: if (mat->stencil.noc) dxm++;
7040: /* Local row number */
7041: if (tmp >= 0) jdxm[numNewRows++] = tmp;
7042: }
7043: PetscCall(MatZeroRowsLocal(mat, numNewRows, jdxm, diag, x, b));
7044: PetscCall(PetscFree(jdxm));
7045: PetscFunctionReturn(PETSC_SUCCESS);
7046: }
7048: /*@
7049: MatZeroRowsColumnsStencil - Zeros all row and column entries (except possibly the main diagonal)
7050: of a set of rows and columns of a matrix.
7052: Collective
7054: Input Parameters:
7055: + mat - the matrix
7056: . numRows - the number of rows/columns to remove
7057: . rows - the grid coordinates (and component number when dof > 1) for matrix rows
7058: . diag - value put in all diagonals of eliminated rows (0.0 will even eliminate diagonal entry)
7059: . x - optional vector of solutions for zeroed rows (other entries in vector are not used)
7060: - b - optional vector of right-hand side, that will be adjusted by provided solution
7062: Level: intermediate
7064: Notes:
7065: See `MatZeroRowsColumns()` for details on how this routine operates.
7067: The grid coordinates are across the entire grid, not just the local portion
7069: For periodic boundary conditions use negative indices for values to the left (below 0; that are to be
7070: obtained by wrapping values from right edge). For values to the right of the last entry using that index plus one
7071: etc to obtain values that obtained by wrapping the values from the left edge. This does not work for anything but the
7072: `DM_BOUNDARY_PERIODIC` boundary type.
7074: For indices that don't mean anything for your case (like the `k` index when working in 2d) or the `c` index when you have
7075: a single value per point) you can skip filling those indices.
7077: Fortran Note:
7078: `idxm` and `idxn` should be declared as
7079: .vb
7080: MatStencil idxm(4, m)
7081: .ve
7082: and the values inserted using
7083: .vb
7084: idxm(MatStencil_i, 1) = i
7085: idxm(MatStencil_j, 1) = j
7086: idxm(MatStencil_k, 1) = k
7087: idxm(MatStencil_c, 1) = c
7088: etc
7089: .ve
7091: .seealso: [](ch_matrices), `Mat`, `MatZeroRowsIS()`, `MatZeroRowsColumns()`, `MatZeroRowsLocalIS()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
7092: `MatZeroRowsColumnsLocal()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRowsColumnsIS()`, `MatZeroRows()`
7093: @*/
7094: PetscErrorCode MatZeroRowsColumnsStencil(Mat mat, PetscInt numRows, const MatStencil rows[], PetscScalar diag, Vec x, Vec b)
7095: {
7096: PetscInt dim = mat->stencil.dim;
7097: PetscInt sdim = dim - (1 - (PetscInt)mat->stencil.noc);
7098: PetscInt *dims = mat->stencil.dims + 1;
7099: PetscInt *starts = mat->stencil.starts;
7100: PetscInt *dxm = (PetscInt *)rows;
7101: PetscInt *jdxm, i, j, tmp, numNewRows = 0;
7103: PetscFunctionBegin;
7106: if (numRows) PetscAssertPointer(rows, 3);
7108: PetscCall(PetscMalloc1(numRows, &jdxm));
7109: for (i = 0; i < numRows; ++i) {
7110: /* Skip unused dimensions (they are ordered k, j, i, c) */
7111: for (j = 0; j < 3 - sdim; ++j) dxm++;
7112: /* Local index in X dir */
7113: tmp = *dxm++ - starts[0];
7114: /* Loop over remaining dimensions */
7115: for (j = 0; j < dim - 1; ++j) {
7116: /* If nonlocal, set index to be negative */
7117: if ((*dxm++ - starts[j + 1]) < 0 || tmp < 0) tmp = PETSC_INT_MIN;
7118: /* Update local index */
7119: else tmp = tmp * dims[j] + *(dxm - 1) - starts[j + 1];
7120: }
7121: /* Skip component slot if necessary */
7122: if (mat->stencil.noc) dxm++;
7123: /* Local row number */
7124: if (tmp >= 0) jdxm[numNewRows++] = tmp;
7125: }
7126: PetscCall(MatZeroRowsColumnsLocal(mat, numNewRows, jdxm, diag, x, b));
7127: PetscCall(PetscFree(jdxm));
7128: PetscFunctionReturn(PETSC_SUCCESS);
7129: }
7131: /*@
7132: MatZeroRowsLocal - Zeros all entries (except possibly the main diagonal)
7133: of a set of rows of a matrix; using local numbering of rows.
7135: Collective
7137: Input Parameters:
7138: + mat - the matrix
7139: . numRows - the number of rows to remove
7140: . rows - the local row indices
7141: . diag - value put in all diagonals of eliminated rows
7142: . x - optional vector of solutions for zeroed rows (other entries in vector are not used)
7143: - b - optional vector of right-hand side, that will be adjusted by provided solution
7145: Level: intermediate
7147: Notes:
7148: Before calling `MatZeroRowsLocal()`, the user must first set the
7149: local-to-global mapping by calling MatSetLocalToGlobalMapping(), this is often already set for matrices obtained with `DMCreateMatrix()`.
7151: See `MatZeroRows()` for details on how this routine operates.
7153: .seealso: [](ch_matrices), `Mat`, `MatZeroRowsIS()`, `MatZeroRowsColumns()`, `MatZeroRowsLocalIS()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRows()`, `MatSetOption()`,
7154: `MatZeroRowsColumnsLocal()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRowsColumnsIS()`, `MatZeroRowsColumnsStencil()`
7155: @*/
7156: PetscErrorCode MatZeroRowsLocal(Mat mat, PetscInt numRows, const PetscInt rows[], PetscScalar diag, Vec x, Vec b)
7157: {
7158: PetscFunctionBegin;
7161: if (numRows) PetscAssertPointer(rows, 3);
7162: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
7163: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
7164: MatCheckPreallocated(mat, 1);
7166: if (mat->ops->zerorowslocal) {
7167: PetscUseTypeMethod(mat, zerorowslocal, numRows, rows, diag, x, b);
7168: } else {
7169: IS is, newis;
7170: PetscInt *newRows, nl = 0;
7172: PetscCheck(mat->rmap->mapping, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Need to provide local to global mapping to matrix first");
7173: PetscCall(ISCreateGeneral(PETSC_COMM_SELF, numRows, rows, PETSC_USE_POINTER, &is));
7174: PetscCall(ISLocalToGlobalMappingApplyIS(mat->rmap->mapping, is, &newis));
7175: PetscCall(ISGetIndices(newis, (const PetscInt **)&newRows));
7176: for (PetscInt i = 0; i < numRows; i++)
7177: if (newRows[i] > -1) newRows[nl++] = newRows[i];
7178: PetscUseTypeMethod(mat, zerorows, nl, newRows, diag, x, b);
7179: PetscCall(ISRestoreIndices(newis, (const PetscInt **)&newRows));
7180: PetscCall(ISDestroy(&newis));
7181: PetscCall(ISDestroy(&is));
7182: }
7183: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
7184: PetscFunctionReturn(PETSC_SUCCESS);
7185: }
7187: /*@
7188: MatZeroRowsLocalIS - Zeros all entries (except possibly the main diagonal)
7189: of a set of rows of a matrix; using local numbering of rows.
7191: Collective
7193: Input Parameters:
7194: + mat - the matrix
7195: . is - index set of rows to remove
7196: . diag - value put in all diagonals of eliminated rows
7197: . x - optional vector of solutions for zeroed rows (other entries in vector are not used)
7198: - b - optional vector of right-hand side, that will be adjusted by provided solution
7200: Level: intermediate
7202: Notes:
7203: Before calling `MatZeroRowsLocalIS()`, the user must first set the
7204: local-to-global mapping by calling `MatSetLocalToGlobalMapping()`, this is often already set for matrices obtained with `DMCreateMatrix()`.
7206: See `MatZeroRows()` for details on how this routine operates.
7208: .seealso: [](ch_matrices), `Mat`, `MatZeroRowsIS()`, `MatZeroRowsColumns()`, `MatZeroRows()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
7209: `MatZeroRowsColumnsLocal()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRowsColumnsIS()`, `MatZeroRowsColumnsStencil()`
7210: @*/
7211: PetscErrorCode MatZeroRowsLocalIS(Mat mat, IS is, PetscScalar diag, Vec x, Vec b)
7212: {
7213: PetscInt numRows;
7214: const PetscInt *rows;
7216: PetscFunctionBegin;
7220: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
7221: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
7222: MatCheckPreallocated(mat, 1);
7224: PetscCall(ISGetLocalSize(is, &numRows));
7225: PetscCall(ISGetIndices(is, &rows));
7226: PetscCall(MatZeroRowsLocal(mat, numRows, rows, diag, x, b));
7227: PetscCall(ISRestoreIndices(is, &rows));
7228: PetscFunctionReturn(PETSC_SUCCESS);
7229: }
7231: /*@
7232: MatZeroRowsColumnsLocal - Zeros all entries (except possibly the main diagonal)
7233: of a set of rows and columns of a matrix; using local numbering of rows.
7235: Collective
7237: Input Parameters:
7238: + mat - the matrix
7239: . numRows - the number of rows to remove
7240: . rows - the global row indices
7241: . diag - value put in all diagonals of eliminated rows
7242: . x - optional vector of solutions for zeroed rows (other entries in vector are not used)
7243: - b - optional vector of right-hand side, that will be adjusted by provided solution
7245: Level: intermediate
7247: Notes:
7248: Before calling `MatZeroRowsColumnsLocal()`, the user must first set the
7249: local-to-global mapping by calling `MatSetLocalToGlobalMapping()`, this is often already set for matrices obtained with `DMCreateMatrix()`.
7251: See `MatZeroRowsColumns()` for details on how this routine operates.
7253: .seealso: [](ch_matrices), `Mat`, `MatZeroRowsIS()`, `MatZeroRowsColumns()`, `MatZeroRowsLocalIS()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
7254: `MatZeroRows()`, `MatZeroRowsColumnsLocalIS()`, `MatZeroRowsColumnsIS()`, `MatZeroRowsColumnsStencil()`
7255: @*/
7256: PetscErrorCode MatZeroRowsColumnsLocal(Mat mat, PetscInt numRows, const PetscInt rows[], PetscScalar diag, Vec x, Vec b)
7257: {
7258: PetscFunctionBegin;
7261: if (numRows) PetscAssertPointer(rows, 3);
7262: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
7263: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
7264: MatCheckPreallocated(mat, 1);
7266: if (mat->ops->zerorowscolumnslocal) {
7267: PetscUseTypeMethod(mat, zerorowscolumnslocal, numRows, rows, diag, x, b);
7268: } else {
7269: IS is, newis;
7270: PetscInt *newRows, nl = 0;
7272: PetscCheck(mat->rmap->mapping, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Need to provide local to global mapping to matrix first");
7273: PetscCall(ISCreateGeneral(PETSC_COMM_SELF, numRows, rows, PETSC_USE_POINTER, &is));
7274: PetscCall(ISLocalToGlobalMappingApplyIS(mat->rmap->mapping, is, &newis));
7275: PetscCall(ISGetIndices(newis, (const PetscInt **)&newRows));
7276: for (PetscInt i = 0; i < numRows; i++)
7277: if (newRows[i] > -1) newRows[nl++] = newRows[i];
7278: PetscUseTypeMethod(mat, zerorowscolumns, nl, newRows, diag, x, b);
7279: PetscCall(ISRestoreIndices(newis, (const PetscInt **)&newRows));
7280: PetscCall(ISDestroy(&newis));
7281: PetscCall(ISDestroy(&is));
7282: }
7283: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
7284: PetscFunctionReturn(PETSC_SUCCESS);
7285: }
7287: /*@
7288: MatZeroRowsColumnsLocalIS - Zeros all entries (except possibly the main diagonal)
7289: of a set of rows and columns of a matrix; using local numbering of rows.
7291: Collective
7293: Input Parameters:
7294: + mat - the matrix
7295: . is - index set of rows to remove
7296: . diag - value put in all diagonals of eliminated rows
7297: . x - optional vector of solutions for zeroed rows (other entries in vector are not used)
7298: - b - optional vector of right-hand side, that will be adjusted by provided solution
7300: Level: intermediate
7302: Notes:
7303: Before calling `MatZeroRowsColumnsLocalIS()`, the user must first set the
7304: local-to-global mapping by calling `MatSetLocalToGlobalMapping()`, this is often already set for matrices obtained with `DMCreateMatrix()`.
7306: See `MatZeroRowsColumns()` for details on how this routine operates.
7308: .seealso: [](ch_matrices), `Mat`, `MatZeroRowsIS()`, `MatZeroRowsColumns()`, `MatZeroRowsLocalIS()`, `MatZeroRowsStencil()`, `MatZeroEntries()`, `MatZeroRowsLocal()`, `MatSetOption()`,
7309: `MatZeroRowsColumnsLocal()`, `MatZeroRows()`, `MatZeroRowsColumnsIS()`, `MatZeroRowsColumnsStencil()`
7310: @*/
7311: PetscErrorCode MatZeroRowsColumnsLocalIS(Mat mat, IS is, PetscScalar diag, Vec x, Vec b)
7312: {
7313: PetscInt numRows;
7314: const PetscInt *rows;
7316: PetscFunctionBegin;
7320: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
7321: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
7322: MatCheckPreallocated(mat, 1);
7324: PetscCall(ISGetLocalSize(is, &numRows));
7325: PetscCall(ISGetIndices(is, &rows));
7326: PetscCall(MatZeroRowsColumnsLocal(mat, numRows, rows, diag, x, b));
7327: PetscCall(ISRestoreIndices(is, &rows));
7328: PetscFunctionReturn(PETSC_SUCCESS);
7329: }
7331: /*@
7332: MatGetSize - Returns the numbers of rows and columns in a matrix.
7334: Not Collective
7336: Input Parameter:
7337: . mat - the matrix
7339: Output Parameters:
7340: + m - the number of global rows
7341: - n - the number of global columns
7343: Level: beginner
7345: Note:
7346: Both output parameters can be `NULL` on input.
7348: .seealso: [](ch_matrices), `Mat`, `MatSetSizes()`, `MatGetLocalSize()`
7349: @*/
7350: PetscErrorCode MatGetSize(Mat mat, PetscInt *m, PetscInt *n)
7351: {
7352: PetscFunctionBegin;
7354: if (m) *m = mat->rmap->N;
7355: if (n) *n = mat->cmap->N;
7356: PetscFunctionReturn(PETSC_SUCCESS);
7357: }
7359: /*@
7360: MatGetLocalSize - For most matrix formats, excluding `MATELEMENTAL` and `MATSCALAPACK`, Returns the number of local rows and local columns
7361: of a matrix. For all matrices this is the local size of the left and right vectors as returned by `MatCreateVecs()`.
7363: Not Collective
7365: Input Parameter:
7366: . mat - the matrix
7368: Output Parameters:
7369: + m - the number of local rows, use `NULL` to not obtain this value
7370: - n - the number of local columns, use `NULL` to not obtain this value
7372: Level: beginner
7374: .seealso: [](ch_matrices), `Mat`, `MatSetSizes()`, `MatGetSize()`
7375: @*/
7376: PetscErrorCode MatGetLocalSize(Mat mat, PetscInt *m, PetscInt *n)
7377: {
7378: PetscFunctionBegin;
7380: if (m) PetscAssertPointer(m, 2);
7381: if (n) PetscAssertPointer(n, 3);
7382: if (m) *m = mat->rmap->n;
7383: if (n) *n = mat->cmap->n;
7384: PetscFunctionReturn(PETSC_SUCCESS);
7385: }
7387: /*@
7388: MatGetOwnershipRangeColumn - Returns the range of matrix columns associated with rows of a
7389: vector one multiplies this matrix by that are owned by this process.
7391: Not Collective, unless matrix has not been allocated, then collective
7393: Input Parameter:
7394: . mat - the matrix
7396: Output Parameters:
7397: + m - the global index of the first local column, use `NULL` to not obtain this value
7398: - n - one more than the global index of the last local column, use `NULL` to not obtain this value
7400: Level: developer
7402: Notes:
7403: If the `Mat` was obtained from a `DM` with `DMCreateMatrix()`, then the range values are determined by the specific `DM`.
7405: If the `Mat` was created directly the range values are determined by the local size passed to `MatSetSizes()` or `MatCreateAIJ()`.
7406: If `PETSC_DECIDE` was passed as the local size, then the vector uses default values for the range using `PetscSplitOwnership()`.
7408: For certain `DM`, such as `DMDA`, it is better to use `DM` specific routines, such as `DMDAGetGhostCorners()`, to determine
7409: the local values in the matrix.
7411: Returns the columns of the "diagonal block" for most sparse matrix formats. See [Matrix
7412: Layouts](sec_matlayout) for details on matrix layouts.
7414: .seealso: [](ch_matrices), `Mat`, `MatGetOwnershipRange()`, `MatGetOwnershipRanges()`, `MatGetOwnershipRangesColumn()`, `PetscLayout`,
7415: `MatSetSizes()`, `MatCreateAIJ()`, `DMDAGetGhostCorners()`, `DM`
7416: @*/
7417: PetscErrorCode MatGetOwnershipRangeColumn(Mat mat, PetscInt *m, PetscInt *n)
7418: {
7419: PetscFunctionBegin;
7422: if (m) PetscAssertPointer(m, 2);
7423: if (n) PetscAssertPointer(n, 3);
7424: MatCheckPreallocated(mat, 1);
7425: if (m) *m = mat->cmap->rstart;
7426: if (n) *n = mat->cmap->rend;
7427: PetscFunctionReturn(PETSC_SUCCESS);
7428: }
7430: /*@
7431: MatGetOwnershipRange - For matrices that own values by row, excludes `MATELEMENTAL` and `MATSCALAPACK`, returns the range of matrix rows owned by
7432: this MPI process.
7434: Not Collective
7436: Input Parameter:
7437: . mat - the matrix
7439: Output Parameters:
7440: + m - the global index of the first local row, use `NULL` to not obtain this value
7441: - n - one more than the global index of the last local row, use `NULL` to not obtain this value
7443: Level: beginner
7445: Notes:
7446: If the `Mat` was obtained from a `DM` with `DMCreateMatrix()`, then the range values are determined by the specific `DM`.
7448: If the `Mat` was created directly the range values are determined by the local size passed to `MatSetSizes()` or `MatCreateAIJ()`.
7449: If `PETSC_DECIDE` was passed as the local size, then the vector uses default values for the range using `PetscSplitOwnership()`.
7451: For certain `DM`, such as `DMDA`, it is better to use `DM` specific routines, such as `DMDAGetGhostCorners()`, to determine
7452: the local values in the matrix.
7454: The high argument is one more than the last element stored locally.
7456: For all matrices it returns the range of matrix rows associated with rows of a vector that
7457: would contain the result of a matrix vector product with this matrix. See [Matrix
7458: Layouts](sec_matlayout) for details on matrix layouts.
7460: .seealso: [](ch_matrices), `Mat`, `MatGetOwnershipRanges()`, `MatGetOwnershipRangeColumn()`, `MatGetOwnershipRangesColumn()`, `PetscSplitOwnership()`,
7461: `PetscSplitOwnershipBlock()`, `PetscLayout`, `MatSetSizes()`, `MatCreateAIJ()`, `DMDAGetGhostCorners()`, `DM`
7462: @*/
7463: PetscErrorCode MatGetOwnershipRange(Mat mat, PetscInt *m, PetscInt *n)
7464: {
7465: PetscFunctionBegin;
7468: if (m) PetscAssertPointer(m, 2);
7469: if (n) PetscAssertPointer(n, 3);
7470: MatCheckPreallocated(mat, 1);
7471: if (m) *m = mat->rmap->rstart;
7472: if (n) *n = mat->rmap->rend;
7473: PetscFunctionReturn(PETSC_SUCCESS);
7474: }
7476: /*@
7477: MatGetOwnershipRanges - For matrices that own values by row, excludes `MATELEMENTAL` and
7478: `MATSCALAPACK`, returns the range of matrix rows owned by each process.
7480: Not Collective, unless matrix has not been allocated
7482: Input Parameter:
7483: . mat - the matrix
7485: Output Parameter:
7486: . ranges - start of each process's portion plus one more than the total length at the end, of length `size` + 1
7487: where `size` is the number of MPI processes used by `mat`
7489: Level: beginner
7491: Notes:
7492: If the `Mat` was obtained from a `DM` with `DMCreateMatrix()`, then the range values are determined by the specific `DM`.
7494: If the `Mat` was created directly the range values are determined by the local size passed to `MatSetSizes()` or `MatCreateAIJ()`.
7495: If `PETSC_DECIDE` was passed as the local size, then the vector uses default values for the range using `PetscSplitOwnership()`.
7497: For certain `DM`, such as `DMDA`, it is better to use `DM` specific routines, such as `DMDAGetGhostCorners()`, to determine
7498: the local values in the matrix.
7500: For all matrices it returns the ranges of matrix rows associated with rows of a vector that
7501: would contain the result of a matrix vector product with this matrix. See [Matrix
7502: Layouts](sec_matlayout) for details on matrix layouts.
7504: .seealso: [](ch_matrices), `Mat`, `MatGetOwnershipRange()`, `MatGetOwnershipRangeColumn()`, `MatGetOwnershipRangesColumn()`, `PetscLayout`,
7505: `PetscSplitOwnership()`, `PetscSplitOwnershipBlock()`, `MatSetSizes()`, `MatCreateAIJ()`,
7506: `DMDAGetGhostCorners()`, `DM`
7507: @*/
7508: PetscErrorCode MatGetOwnershipRanges(Mat mat, const PetscInt *ranges[])
7509: {
7510: PetscFunctionBegin;
7513: MatCheckPreallocated(mat, 1);
7514: PetscCall(PetscLayoutGetRanges(mat->rmap, ranges));
7515: PetscFunctionReturn(PETSC_SUCCESS);
7516: }
7518: /*@
7519: MatGetOwnershipRangesColumn - Returns the ranges of matrix columns associated with rows of a
7520: vector one multiplies this vector by that are owned by each process.
7522: Not Collective, unless matrix has not been allocated
7524: Input Parameter:
7525: . mat - the matrix
7527: Output Parameter:
7528: . ranges - start of each process's portion plus one more than the total length at the end
7530: Level: beginner
7532: Notes:
7533: If the `Mat` was obtained from a `DM` with `DMCreateMatrix()`, then the range values are determined by the specific `DM`.
7535: If the `Mat` was created directly the range values are determined by the local size passed to `MatSetSizes()` or `MatCreateAIJ()`.
7536: If `PETSC_DECIDE` was passed as the local size, then the vector uses default values for the range using `PetscSplitOwnership()`.
7538: For certain `DM`, such as `DMDA`, it is better to use `DM` specific routines, such as `DMDAGetGhostCorners()`, to determine
7539: the local values in the matrix.
7541: Returns the columns of the "diagonal blocks", for most sparse matrix formats. See [Matrix
7542: Layouts](sec_matlayout) for details on matrix layouts.
7544: .seealso: [](ch_matrices), `Mat`, `MatGetOwnershipRange()`, `MatGetOwnershipRangeColumn()`, `MatGetOwnershipRanges()`,
7545: `PetscSplitOwnership()`, `PetscSplitOwnershipBlock()`, `PetscLayout`, `MatSetSizes()`, `MatCreateAIJ()`,
7546: `DMDAGetGhostCorners()`, `DM`
7547: @*/
7548: PetscErrorCode MatGetOwnershipRangesColumn(Mat mat, const PetscInt *ranges[])
7549: {
7550: PetscFunctionBegin;
7553: MatCheckPreallocated(mat, 1);
7554: PetscCall(PetscLayoutGetRanges(mat->cmap, ranges));
7555: PetscFunctionReturn(PETSC_SUCCESS);
7556: }
7558: /*@
7559: MatGetOwnershipIS - Get row and column ownership of a matrices' values as index sets.
7561: Not Collective
7563: Input Parameter:
7564: . A - matrix
7566: Output Parameters:
7567: + rows - rows in which this process owns elements, , use `NULL` to not obtain this value
7568: - cols - columns in which this process owns elements, use `NULL` to not obtain this value
7570: Level: intermediate
7572: Note:
7573: You should call `ISDestroy()` on the returned `IS`
7575: For most matrices, excluding `MATELEMENTAL` and `MATSCALAPACK`, this corresponds to values
7576: returned by `MatGetOwnershipRange()`, `MatGetOwnershipRangeColumn()`. For `MATELEMENTAL` and
7577: `MATSCALAPACK` the ownership is more complicated. See [Matrix Layouts](sec_matlayout) for
7578: details on matrix layouts.
7580: .seealso: [](ch_matrices), `IS`, `Mat`, `MatGetOwnershipRanges()`, `MatSetValues()`, `MATELEMENTAL`, `MATSCALAPACK`
7581: @*/
7582: PetscErrorCode MatGetOwnershipIS(Mat A, IS *rows, IS *cols)
7583: {
7584: PetscErrorCode (*f)(Mat, IS *, IS *);
7586: PetscFunctionBegin;
7589: MatCheckPreallocated(A, 1);
7590: PetscCall(PetscObjectQueryFunction((PetscObject)A, "MatGetOwnershipIS_C", &f));
7591: if (f) {
7592: PetscCall((*f)(A, rows, cols));
7593: } else { /* Create a standard row-based partition, each process is responsible for ALL columns in their row block */
7594: if (rows) PetscCall(ISCreateStride(PETSC_COMM_SELF, A->rmap->n, A->rmap->rstart, 1, rows));
7595: if (cols) PetscCall(ISCreateStride(PETSC_COMM_SELF, A->cmap->N, 0, 1, cols));
7596: }
7597: PetscFunctionReturn(PETSC_SUCCESS);
7598: }
7600: /*@
7601: MatILUFactorSymbolic - Performs symbolic ILU factorization of a matrix obtained with `MatGetFactor()`
7602: Uses levels of fill only, not drop tolerance. Use `MatLUFactorNumeric()`
7603: to complete the factorization.
7605: Collective
7607: Input Parameters:
7608: + fact - the factorized matrix obtained with `MatGetFactor()`
7609: . mat - the matrix
7610: . row - row permutation
7611: . col - column permutation
7612: - info - structure containing
7613: .vb
7614: levels - number of levels of fill.
7615: expected fill - as ratio of original fill.
7616: 1 or 0 - indicating force fill on diagonal (improves robustness for matrices
7617: missing diagonal entries)
7618: .ve
7620: Level: developer
7622: Notes:
7623: See [Matrix Factorization](sec_matfactor) for additional information.
7625: Most users should employ the `KSP` interface for linear solvers
7626: instead of working directly with matrix algebra routines such as this.
7627: See, e.g., `KSPCreate()`.
7629: Uses the definition of level of fill as in Y. Saad, {cite}`saad2003`
7631: Fortran Note:
7632: A valid (non-null) `info` argument must be provided
7634: .seealso: [](ch_matrices), `Mat`, [Matrix Factorization](sec_matfactor), `MatGetFactor()`, `MatLUFactorSymbolic()`, `MatLUFactorNumeric()`, `MatCholeskyFactor()`,
7635: `MatGetOrdering()`, `MatFactorInfo`
7636: @*/
7637: PetscErrorCode MatILUFactorSymbolic(Mat fact, Mat mat, IS row, IS col, const MatFactorInfo *info)
7638: {
7639: PetscFunctionBegin;
7644: PetscAssertPointer(info, 5);
7645: PetscAssertPointer(fact, 1);
7646: PetscCheck(info->levels >= 0, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_OUTOFRANGE, "Levels of fill negative %" PetscInt_FMT, (PetscInt)info->levels);
7647: PetscCheck(info->fill >= 1.0, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_OUTOFRANGE, "Expected fill less than 1.0 %g", (double)info->fill);
7648: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
7649: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
7650: MatCheckPreallocated(mat, 2);
7652: if (!fact->trivialsymbolic) PetscCall(PetscLogEventBegin(MAT_ILUFactorSymbolic, mat, row, col, 0));
7653: PetscUseTypeMethod(fact, ilufactorsymbolic, mat, row, col, info);
7654: if (!fact->trivialsymbolic) PetscCall(PetscLogEventEnd(MAT_ILUFactorSymbolic, mat, row, col, 0));
7655: PetscFunctionReturn(PETSC_SUCCESS);
7656: }
7658: /*@
7659: MatICCFactorSymbolic - Performs symbolic incomplete
7660: Cholesky factorization for a symmetric matrix. Use
7661: `MatCholeskyFactorNumeric()` to complete the factorization.
7663: Collective
7665: Input Parameters:
7666: + fact - the factorized matrix obtained with `MatGetFactor()`
7667: . mat - the matrix to be factored
7668: . perm - row and column permutation
7669: - info - structure containing
7670: .vb
7671: levels - number of levels of fill.
7672: expected fill - as ratio of original fill.
7673: .ve
7675: Level: developer
7677: Notes:
7678: Most users should employ the `KSP` interface for linear solvers
7679: instead of working directly with matrix algebra routines such as this.
7680: See, e.g., `KSPCreate()`.
7682: This uses the definition of level of fill as in Y. Saad {cite}`saad2003`
7684: Fortran Note:
7685: A valid (non-null) `info` argument must be provided
7687: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatCholeskyFactorNumeric()`, `MatCholeskyFactor()`, `MatFactorInfo`
7688: @*/
7689: PetscErrorCode MatICCFactorSymbolic(Mat fact, Mat mat, IS perm, const MatFactorInfo *info)
7690: {
7691: PetscFunctionBegin;
7695: PetscAssertPointer(info, 4);
7696: PetscAssertPointer(fact, 1);
7697: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
7698: PetscCheck(info->levels >= 0, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_OUTOFRANGE, "Levels negative %" PetscInt_FMT, (PetscInt)info->levels);
7699: PetscCheck(info->fill >= 1.0, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_OUTOFRANGE, "Expected fill less than 1.0 %g", (double)info->fill);
7700: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
7701: MatCheckPreallocated(mat, 2);
7703: if (!fact->trivialsymbolic) PetscCall(PetscLogEventBegin(MAT_ICCFactorSymbolic, mat, perm, 0, 0));
7704: PetscUseTypeMethod(fact, iccfactorsymbolic, mat, perm, info);
7705: if (!fact->trivialsymbolic) PetscCall(PetscLogEventEnd(MAT_ICCFactorSymbolic, mat, perm, 0, 0));
7706: PetscFunctionReturn(PETSC_SUCCESS);
7707: }
7709: /*@
7710: MatCreateSubMatrices - Extracts several submatrices from a matrix. If submat
7711: points to an array of valid matrices, they may be reused to store the new
7712: submatrices.
7714: Collective
7716: Input Parameters:
7717: + mat - the matrix
7718: . n - the number of submatrixes to be extracted (on this process, may be zero)
7719: . irow - index set of rows to extract
7720: . icol - index set of columns to extract
7721: - scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
7723: Output Parameter:
7724: . submat - the array of submatrices
7726: Level: advanced
7728: Notes:
7729: `MatCreateSubMatrices()` can extract ONLY sequential submatrices
7730: (from both sequential and parallel matrices). Use `MatCreateSubMatrix()`
7731: to extract a parallel submatrix.
7733: Some matrix types place restrictions on the row and column
7734: indices, such as that they be sorted or that they be equal to each other.
7735: `MATSEQSBAIJ` inputs may produce `MATSEQBAIJ` submatrices when the row and column index sets do not preserve symmetry.
7737: The index sets may not have duplicate entries.
7739: When extracting submatrices from a parallel matrix, each process can
7740: form a different submatrix by setting the rows and columns of its
7741: individual index sets according to the local submatrix desired.
7743: When finished using the submatrices, the user should destroy
7744: them with `MatDestroySubMatrices()`.
7746: `MAT_REUSE_MATRIX` can only be used when the nonzero structure of the
7747: original matrix has not changed from that last call to `MatCreateSubMatrices()`.
7749: This routine creates the matrices in submat; you should NOT create them before
7750: calling it. It also allocates the array of matrix pointers submat.
7752: For `MATBAIJ` matrices the index sets must respect the block structure, that is if they
7753: request one row/column in a block, they must request all rows/columns that are in
7754: that block. For example, if the block size is 2 you cannot request just row 0 and
7755: column 0.
7757: Fortran Note:
7758: .vb
7759: Mat, pointer :: submat(:)
7760: .ve
7762: .seealso: [](ch_matrices), `Mat`, `MatDestroySubMatrices()`, `MatCreateSubMatrix()`, `MatGetRow()`, `MatGetDiagonal()`, `MatReuse`
7763: @*/
7764: PetscErrorCode MatCreateSubMatrices(Mat mat, PetscInt n, const IS irow[], const IS icol[], MatReuse scall, Mat *submat[])
7765: {
7766: PetscInt i;
7767: PetscBool eq;
7769: PetscFunctionBegin;
7772: if (n) {
7773: PetscAssertPointer(irow, 3);
7775: PetscAssertPointer(icol, 4);
7777: }
7778: PetscAssertPointer(submat, 6);
7779: if (n && scall == MAT_REUSE_MATRIX) {
7780: PetscAssertPointer(*submat, 6);
7782: }
7783: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
7784: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
7785: MatCheckPreallocated(mat, 1);
7786: PetscCall(PetscLogEventBegin(MAT_CreateSubMats, mat, 0, 0, 0));
7787: PetscUseTypeMethod(mat, createsubmatrices, n, irow, icol, scall, submat);
7788: PetscCall(PetscLogEventEnd(MAT_CreateSubMats, mat, 0, 0, 0));
7789: for (i = 0; i < n; i++) {
7790: (*submat)[i]->factortype = MAT_FACTOR_NONE; /* in case in place factorization was previously done on submatrix */
7791: PetscCall(ISEqualUnsorted(irow[i], icol[i], &eq));
7792: if (eq) PetscCall(MatPropagateSymmetryOptions(mat, (*submat)[i]));
7793: #if PetscDefined(HAVE_VIENNACL) || PetscDefined(HAVE_CUDA) || PetscDefined(HAVE_HIP)
7794: if (mat->boundtocpu && mat->bindingpropagates) {
7795: PetscCall(MatBindToCPU((*submat)[i], PETSC_TRUE));
7796: PetscCall(MatSetBindingPropagates((*submat)[i], PETSC_TRUE));
7797: }
7798: #endif
7799: }
7800: PetscFunctionReturn(PETSC_SUCCESS);
7801: }
7803: /*@
7804: MatCreateSubMatricesMPI - Extracts MPI submatrices across a sub communicator of `mat` (by pairs of `IS` that may live on subcomms).
7806: Collective
7808: Input Parameters:
7809: + mat - the matrix
7810: . n - the number of submatrixes to be extracted
7811: . irow - index set of rows to extract
7812: . icol - index set of columns to extract
7813: - scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
7815: Output Parameter:
7816: . submat - the array of submatrices
7818: Level: advanced
7820: Note:
7821: This is used by `PCGASM`
7823: .seealso: [](ch_matrices), `Mat`, `PCGASM`, `MatCreateSubMatrices()`, `MatCreateSubMatrix()`, `MatGetRow()`, `MatGetDiagonal()`, `MatReuse`
7824: @*/
7825: PetscErrorCode MatCreateSubMatricesMPI(Mat mat, PetscInt n, const IS irow[], const IS icol[], MatReuse scall, Mat *submat[])
7826: {
7827: PetscInt i;
7828: PetscBool eq;
7830: PetscFunctionBegin;
7833: if (n) {
7834: PetscAssertPointer(irow, 3);
7836: PetscAssertPointer(icol, 4);
7838: }
7839: PetscAssertPointer(submat, 6);
7840: if (n && scall == MAT_REUSE_MATRIX) {
7841: PetscAssertPointer(*submat, 6);
7843: }
7844: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
7845: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
7846: MatCheckPreallocated(mat, 1);
7848: PetscCall(PetscLogEventBegin(MAT_CreateSubMats, mat, 0, 0, 0));
7849: PetscUseTypeMethod(mat, createsubmatricesmpi, n, irow, icol, scall, submat);
7850: PetscCall(PetscLogEventEnd(MAT_CreateSubMats, mat, 0, 0, 0));
7851: for (i = 0; i < n; i++) {
7852: PetscCall(ISEqualUnsorted(irow[i], icol[i], &eq));
7853: if (eq) PetscCall(MatPropagateSymmetryOptions(mat, (*submat)[i]));
7854: }
7855: PetscFunctionReturn(PETSC_SUCCESS);
7856: }
7858: /*@
7859: MatDestroyMatrices - Destroys an array of matrices
7861: Collective
7863: Input Parameters:
7864: + n - the number of local matrices
7865: - mat - the matrices (this is a pointer to the array of matrices)
7867: Level: advanced
7869: Notes:
7870: Frees not only the matrices, but also the array that contains the matrices
7872: For matrices obtained with `MatCreateSubMatrices()` use `MatDestroySubMatrices()`
7874: .seealso: [](ch_matrices), `Mat`, `MatCreateSubMatrices()`, `MatDestroySubMatrices()`
7875: @*/
7876: PetscErrorCode MatDestroyMatrices(PetscInt n, Mat *mat[])
7877: {
7878: PetscInt i;
7880: PetscFunctionBegin;
7881: if (!*mat) PetscFunctionReturn(PETSC_SUCCESS);
7882: PetscCheck(n >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Trying to destroy negative number of matrices %" PetscInt_FMT, n);
7883: PetscAssertPointer(mat, 2);
7885: for (i = 0; i < n; i++) PetscCall(MatDestroy(&(*mat)[i]));
7887: /* memory is allocated even if n = 0 */
7888: PetscCall(PetscFree(*mat));
7889: PetscFunctionReturn(PETSC_SUCCESS);
7890: }
7892: /*@
7893: MatDestroySubMatrices - Destroys a set of matrices obtained with `MatCreateSubMatrices()`.
7895: Collective
7897: Input Parameters:
7898: + n - the number of local matrices
7899: - mat - the matrices (this is a pointer to the array of matrices, to match the calling sequence of `MatCreateSubMatrices()`)
7901: Level: advanced
7903: Note:
7904: Frees not only the matrices, but also the array that contains the matrices
7906: .seealso: [](ch_matrices), `Mat`, `MatCreateSubMatrices()`, `MatDestroyMatrices()`
7907: @*/
7908: PetscErrorCode MatDestroySubMatrices(PetscInt n, Mat *mat[])
7909: {
7910: Mat mat0;
7912: PetscFunctionBegin;
7913: if (!*mat) PetscFunctionReturn(PETSC_SUCCESS);
7914: /* mat[] is an array of length n+1, see MatCreateSubMatrices_xxx() */
7915: PetscCheck(n >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Trying to destroy negative number of matrices %" PetscInt_FMT, n);
7916: PetscAssertPointer(mat, 2);
7918: mat0 = (*mat)[0];
7919: if (mat0 && mat0->ops->destroysubmatrices) {
7920: PetscCall((*mat0->ops->destroysubmatrices)(n, mat));
7921: } else {
7922: PetscCall(MatDestroyMatrices(n, mat));
7923: }
7924: PetscFunctionReturn(PETSC_SUCCESS);
7925: }
7927: /*@
7928: MatGetSeqNonzeroStructure - Extracts the nonzero structure from a matrix and stores it, in its entirety, on each process
7930: Collective
7932: Input Parameter:
7933: . mat - the matrix
7935: Output Parameter:
7936: . matstruct - the sequential matrix with the nonzero structure of `mat`
7938: Level: developer
7940: .seealso: [](ch_matrices), `Mat`, `MatDestroySeqNonzeroStructure()`, `MatCreateSubMatrices()`, `MatDestroyMatrices()`
7941: @*/
7942: PetscErrorCode MatGetSeqNonzeroStructure(Mat mat, Mat *matstruct)
7943: {
7944: PetscFunctionBegin;
7946: PetscAssertPointer(matstruct, 2);
7949: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
7950: MatCheckPreallocated(mat, 1);
7952: PetscCall(PetscLogEventBegin(MAT_GetSeqNonzeroStructure, mat, 0, 0, 0));
7953: PetscUseTypeMethod(mat, getseqnonzerostructure, matstruct);
7954: PetscCall(PetscLogEventEnd(MAT_GetSeqNonzeroStructure, mat, 0, 0, 0));
7955: PetscFunctionReturn(PETSC_SUCCESS);
7956: }
7958: /*@
7959: MatDestroySeqNonzeroStructure - Destroys matrix obtained with `MatGetSeqNonzeroStructure()`.
7961: Collective
7963: Input Parameter:
7964: . mat - the matrix
7966: Level: advanced
7968: Note:
7969: This is not needed, one can just call `MatDestroy()`
7971: .seealso: [](ch_matrices), `Mat`, `MatGetSeqNonzeroStructure()`
7972: @*/
7973: PetscErrorCode MatDestroySeqNonzeroStructure(Mat *mat)
7974: {
7975: PetscFunctionBegin;
7976: PetscAssertPointer(mat, 1);
7977: PetscCall(MatDestroy(mat));
7978: PetscFunctionReturn(PETSC_SUCCESS);
7979: }
7981: /*@
7982: MatIncreaseOverlap - Given a set of submatrices indicated by index sets,
7983: replaces the index sets by larger ones that represent submatrices with
7984: additional overlap.
7986: Collective
7988: Input Parameters:
7989: + mat - the matrix
7990: . n - the number of index sets
7991: . is - the array of index sets (these index sets will changed during the call)
7992: - ov - the additional overlap requested
7994: Options Database Key:
7995: . -mat_increase_overlap_scalable - use a scalable algorithm to compute the overlap (supported by MPIAIJ matrix)
7997: Level: developer
7999: Note:
8000: The computed overlap preserves the matrix block sizes when the blocks are square.
8001: That is: if a matrix nonzero for a given block would increase the overlap all columns associated with
8002: that block are included in the overlap regardless of whether each specific column would increase the overlap.
8004: .seealso: [](ch_matrices), `Mat`, `PCASM`, `MatSetBlockSize()`, `MatIncreaseOverlapSplit()`, `MatCreateSubMatrices()`
8005: @*/
8006: PetscErrorCode MatIncreaseOverlap(Mat mat, PetscInt n, IS is[], PetscInt ov)
8007: {
8008: PetscInt i, bs, cbs;
8010: PetscFunctionBegin;
8014: PetscCheck(n >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Must have one or more domains, you have %" PetscInt_FMT, n);
8015: if (n) {
8016: PetscAssertPointer(is, 3);
8018: }
8019: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
8020: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
8021: MatCheckPreallocated(mat, 1);
8023: if (!ov || !n) PetscFunctionReturn(PETSC_SUCCESS);
8024: PetscCall(PetscLogEventBegin(MAT_IncreaseOverlap, mat, 0, 0, 0));
8025: PetscUseTypeMethod(mat, increaseoverlap, n, is, ov);
8026: PetscCall(PetscLogEventEnd(MAT_IncreaseOverlap, mat, 0, 0, 0));
8027: PetscCall(MatGetBlockSizes(mat, &bs, &cbs));
8028: if (bs == cbs) {
8029: for (i = 0; i < n; i++) PetscCall(ISSetBlockSize(is[i], bs));
8030: }
8031: PetscFunctionReturn(PETSC_SUCCESS);
8032: }
8034: PetscErrorCode MatIncreaseOverlapSplit_Single(Mat, IS *, PetscInt);
8036: /*@
8037: MatIncreaseOverlapSplit - Given a set of submatrices indicated by index sets across
8038: a sub communicator, replaces the index sets by larger ones that represent submatrices with
8039: additional overlap.
8041: Collective
8043: Input Parameters:
8044: + mat - the matrix
8045: . n - the number of index sets
8046: . is - the array of index sets (these index sets will changed during the call)
8047: - ov - the additional overlap requested
8049: ` Options Database Key:
8050: . -mat_increase_overlap_scalable - use a scalable algorithm to compute the overlap (supported by MPIAIJ matrix)
8052: Level: developer
8054: .seealso: [](ch_matrices), `Mat`, `MatCreateSubMatrices()`, `MatIncreaseOverlap()`
8055: @*/
8056: PetscErrorCode MatIncreaseOverlapSplit(Mat mat, PetscInt n, IS is[], PetscInt ov)
8057: {
8058: PetscInt i;
8060: PetscFunctionBegin;
8063: PetscCheck(n >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Must have one or more domains, you have %" PetscInt_FMT, n);
8064: if (n) {
8065: PetscAssertPointer(is, 3);
8067: }
8068: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
8069: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
8070: MatCheckPreallocated(mat, 1);
8071: if (!ov) PetscFunctionReturn(PETSC_SUCCESS);
8072: PetscCall(PetscLogEventBegin(MAT_IncreaseOverlap, mat, 0, 0, 0));
8073: for (i = 0; i < n; i++) PetscCall(MatIncreaseOverlapSplit_Single(mat, &is[i], ov));
8074: PetscCall(PetscLogEventEnd(MAT_IncreaseOverlap, mat, 0, 0, 0));
8075: PetscFunctionReturn(PETSC_SUCCESS);
8076: }
8078: /*@
8079: MatGetBlockSize - Returns the matrix block size.
8081: Not Collective
8083: Input Parameter:
8084: . mat - the matrix
8086: Output Parameter:
8087: . bs - block size
8089: Level: intermediate
8091: Notes:
8092: Block row formats are `MATBAIJ` and `MATSBAIJ` ALWAYS have square block storage in the matrix.
8094: If the block size has not been set yet this routine returns 1.
8096: .seealso: [](ch_matrices), `Mat`, `MATBAIJ`, `MATSBAIJ`, `MatCreateSeqBAIJ()`, `MatCreateBAIJ()`, `MatGetBlockSizes()`
8097: @*/
8098: PetscErrorCode MatGetBlockSize(Mat mat, PetscInt *bs)
8099: {
8100: PetscFunctionBegin;
8102: PetscAssertPointer(bs, 2);
8103: *bs = mat->rmap->bs;
8104: PetscFunctionReturn(PETSC_SUCCESS);
8105: }
8107: /*@
8108: MatGetBlockSizes - Returns the matrix block row and column sizes.
8110: Not Collective
8112: Input Parameter:
8113: . mat - the matrix
8115: Output Parameters:
8116: + rbs - row block size
8117: - cbs - column block size
8119: Level: intermediate
8121: Notes:
8122: Block row formats are `MATBAIJ` and `MATSBAIJ` ALWAYS have square block storage in the matrix.
8123: If you pass a different block size for the columns than the rows, the row block size determines the square block storage.
8125: If a block size has not been set yet this routine returns 1.
8127: .seealso: [](ch_matrices), `Mat`, `MATBAIJ`, `MATSBAIJ`, `MatCreateSeqBAIJ()`, `MatCreateBAIJ()`, `MatGetBlockSize()`, `MatSetBlockSize()`, `MatSetBlockSizes()`
8128: @*/
8129: PetscErrorCode MatGetBlockSizes(Mat mat, PetscInt *rbs, PetscInt *cbs)
8130: {
8131: PetscFunctionBegin;
8133: if (rbs) PetscAssertPointer(rbs, 2);
8134: if (cbs) PetscAssertPointer(cbs, 3);
8135: if (rbs) *rbs = mat->rmap->bs;
8136: if (cbs) *cbs = mat->cmap->bs;
8137: PetscFunctionReturn(PETSC_SUCCESS);
8138: }
8140: /*@
8141: MatSetBlockSize - Sets the matrix block size.
8143: Logically Collective
8145: Input Parameters:
8146: + mat - the matrix
8147: - bs - block size
8149: Level: intermediate
8151: Notes:
8152: Block row formats are `MATBAIJ` and `MATSBAIJ` formats ALWAYS have square block storage in the matrix.
8153: This must be called before `MatSetUp()` or MatXXXSetPreallocation() (or will default to 1) and the block size cannot be changed later.
8155: For `MATAIJ` matrix format, this function can be called at a later stage, provided that the specified block size
8156: is compatible with the matrix local sizes.
8158: .seealso: [](ch_matrices), `Mat`, `MATBAIJ`, `MATSBAIJ`, `MATAIJ`, `MatCreateSeqBAIJ()`, `MatCreateBAIJ()`, `MatGetBlockSize()`, `MatSetBlockSizes()`, `MatGetBlockSizes()`
8159: @*/
8160: PetscErrorCode MatSetBlockSize(Mat mat, PetscInt bs)
8161: {
8162: PetscFunctionBegin;
8165: PetscCall(MatSetBlockSizes(mat, bs, bs));
8166: PetscFunctionReturn(PETSC_SUCCESS);
8167: }
8169: typedef struct {
8170: PetscInt n;
8171: IS *is;
8172: Mat *mat;
8173: PetscObjectState nonzerostate;
8174: Mat C;
8175: } EnvelopeData;
8177: static PetscErrorCode EnvelopeDataDestroy(PetscCtxRt ptr)
8178: {
8179: EnvelopeData *edata = *(EnvelopeData **)ptr;
8181: PetscFunctionBegin;
8182: for (PetscInt i = 0; i < edata->n; i++) PetscCall(ISDestroy(&edata->is[i]));
8183: PetscCall(PetscFree(edata->is));
8184: PetscCall(PetscFree(edata));
8185: PetscFunctionReturn(PETSC_SUCCESS);
8186: }
8188: /*@
8189: MatComputeVariableBlockEnvelope - Given a matrix whose nonzeros are in blocks along the diagonal this computes and stores
8190: the sizes of these blocks in the matrix. An individual block may lie over several processes.
8192: Collective
8194: Input Parameter:
8195: . mat - the matrix
8197: Level: intermediate
8199: Notes:
8200: There can be zeros within the blocks
8202: The blocks can overlap between processes, including laying on more than two processes
8204: .seealso: [](ch_matrices), `Mat`, `MatInvertVariableBlockEnvelope()`, `MatSetVariableBlockSizes()`
8205: @*/
8206: PetscErrorCode MatComputeVariableBlockEnvelope(Mat mat)
8207: {
8208: PetscInt n, *sizes, *starts, i = 0, env = 0, tbs = 0, lblocks = 0, rstart, II, ln = 0, cnt = 0, cstart, cend;
8209: PetscInt *diag, *odiag, sc;
8210: VecScatter scatter;
8211: PetscScalar *seqv;
8212: const PetscScalar *parv;
8213: const PetscInt *ia, *ja;
8214: PetscBool set, flag, done;
8215: Mat AA = mat, A;
8216: MPI_Comm comm;
8217: PetscMPIInt rank, size, tag;
8218: MPI_Status status;
8219: PetscContainer container;
8220: EnvelopeData *edata;
8221: Vec seq, par;
8222: IS isglobal;
8224: PetscFunctionBegin;
8226: PetscCall(MatIsSymmetricKnown(mat, &set, &flag));
8227: if (!set || !flag) {
8228: /* TODO: only needs nonzero structure of transpose */
8229: PetscCall(MatTranspose(mat, MAT_INITIAL_MATRIX, &AA));
8230: PetscCall(MatAXPY(AA, 1.0, mat, DIFFERENT_NONZERO_PATTERN));
8231: }
8232: PetscCall(MatAIJGetLocalMat(AA, &A));
8233: PetscCall(MatGetRowIJ(A, 0, PETSC_FALSE, PETSC_FALSE, &n, &ia, &ja, &done));
8234: PetscCheck(done, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Unable to get IJ structure from matrix");
8236: PetscCall(MatGetLocalSize(mat, &n, NULL));
8237: PetscCall(PetscObjectGetNewTag((PetscObject)mat, &tag));
8238: PetscCall(PetscObjectGetComm((PetscObject)mat, &comm));
8239: PetscCallMPI(MPI_Comm_size(comm, &size));
8240: PetscCallMPI(MPI_Comm_rank(comm, &rank));
8242: PetscCall(PetscMalloc2(n, &sizes, n, &starts));
8244: if (rank > 0) {
8245: PetscCallMPI(MPI_Recv(&env, 1, MPIU_INT, rank - 1, tag, comm, &status));
8246: PetscCallMPI(MPI_Recv(&tbs, 1, MPIU_INT, rank - 1, tag, comm, &status));
8247: }
8248: PetscCall(MatGetOwnershipRange(mat, &rstart, NULL));
8249: for (i = 0; i < n; i++) {
8250: env = PetscMax(env, ja[ia[i + 1] - 1]);
8251: II = rstart + i;
8252: if (env == II) {
8253: starts[lblocks] = tbs;
8254: sizes[lblocks++] = 1 + II - tbs;
8255: tbs = 1 + II;
8256: }
8257: }
8258: if (rank < size - 1) {
8259: PetscCallMPI(MPI_Send(&env, 1, MPIU_INT, rank + 1, tag, comm));
8260: PetscCallMPI(MPI_Send(&tbs, 1, MPIU_INT, rank + 1, tag, comm));
8261: }
8263: PetscCall(MatRestoreRowIJ(A, 0, PETSC_FALSE, PETSC_FALSE, &n, &ia, &ja, &done));
8264: if (!set || !flag) PetscCall(MatDestroy(&AA));
8265: PetscCall(MatDestroy(&A));
8267: PetscCall(PetscNew(&edata));
8268: PetscCall(MatGetNonzeroState(mat, &edata->nonzerostate));
8269: edata->n = lblocks;
8270: /* create IS needed for extracting blocks from the original matrix */
8271: PetscCall(PetscMalloc1(lblocks, &edata->is));
8272: for (PetscInt i = 0; i < lblocks; i++) PetscCall(ISCreateStride(PETSC_COMM_SELF, sizes[i], starts[i], 1, &edata->is[i]));
8274: /* Create the resulting inverse matrix nonzero structure with preallocation information */
8275: PetscCall(MatCreate(PetscObjectComm((PetscObject)mat), &edata->C));
8276: PetscCall(MatSetSizes(edata->C, mat->rmap->n, mat->cmap->n, mat->rmap->N, mat->cmap->N));
8277: PetscCall(MatSetBlockSizesFromMats(edata->C, mat, mat));
8278: PetscCall(MatSetType(edata->C, MATAIJ));
8280: /* Communicate the start and end of each row, from each block to the correct rank */
8281: /* TODO: Use PetscSF instead of VecScatter */
8282: for (PetscInt i = 0; i < lblocks; i++) ln += sizes[i];
8283: PetscCall(VecCreateSeq(PETSC_COMM_SELF, 2 * ln, &seq));
8284: PetscCall(VecGetArrayWrite(seq, &seqv));
8285: for (PetscInt i = 0; i < lblocks; i++) {
8286: for (PetscInt j = 0; j < sizes[i]; j++) {
8287: seqv[cnt] = starts[i];
8288: seqv[cnt + 1] = starts[i] + sizes[i];
8289: cnt += 2;
8290: }
8291: }
8292: PetscCall(VecRestoreArrayWrite(seq, &seqv));
8293: PetscCallMPI(MPI_Scan(&cnt, &sc, 1, MPIU_INT, MPI_SUM, PetscObjectComm((PetscObject)mat)));
8294: sc -= cnt;
8295: PetscCall(VecCreateMPI(PetscObjectComm((PetscObject)mat), 2 * mat->rmap->n, 2 * mat->rmap->N, &par));
8296: PetscCall(ISCreateStride(PETSC_COMM_SELF, cnt, sc, 1, &isglobal));
8297: PetscCall(VecScatterCreate(seq, NULL, par, isglobal, &scatter));
8298: PetscCall(ISDestroy(&isglobal));
8299: PetscCall(VecScatterBegin(scatter, seq, par, INSERT_VALUES, SCATTER_FORWARD));
8300: PetscCall(VecScatterEnd(scatter, seq, par, INSERT_VALUES, SCATTER_FORWARD));
8301: PetscCall(VecScatterDestroy(&scatter));
8302: PetscCall(VecDestroy(&seq));
8303: PetscCall(MatGetOwnershipRangeColumn(mat, &cstart, &cend));
8304: PetscCall(PetscMalloc2(mat->rmap->n, &diag, mat->rmap->n, &odiag));
8305: PetscCall(VecGetArrayRead(par, &parv));
8306: cnt = 0;
8307: PetscCall(MatGetSize(mat, NULL, &n));
8308: for (PetscInt i = 0; i < mat->rmap->n; i++) {
8309: PetscInt start, end, d = 0, od = 0;
8311: start = (PetscInt)PetscRealPart(parv[cnt]);
8312: end = (PetscInt)PetscRealPart(parv[cnt + 1]);
8313: cnt += 2;
8315: if (start < cstart) {
8316: od += cstart - start + n - cend;
8317: d += cend - cstart;
8318: } else if (start < cend) {
8319: od += n - cend;
8320: d += cend - start;
8321: } else od += n - start;
8322: if (end <= cstart) {
8323: od -= cstart - end + n - cend;
8324: d -= cend - cstart;
8325: } else if (end < cend) {
8326: od -= n - cend;
8327: d -= cend - end;
8328: } else od -= n - end;
8330: odiag[i] = od;
8331: diag[i] = d;
8332: }
8333: PetscCall(VecRestoreArrayRead(par, &parv));
8334: PetscCall(VecDestroy(&par));
8335: PetscCall(MatXAIJSetPreallocation(edata->C, mat->rmap->bs, diag, odiag, NULL, NULL));
8336: PetscCall(PetscFree2(diag, odiag));
8337: PetscCall(PetscFree2(sizes, starts));
8339: PetscCall(PetscContainerCreate(PETSC_COMM_SELF, &container));
8340: PetscCall(PetscContainerSetPointer(container, edata));
8341: PetscCall(PetscContainerSetCtxDestroy(container, EnvelopeDataDestroy));
8342: PetscCall(PetscObjectCompose((PetscObject)mat, "EnvelopeData", (PetscObject)container));
8343: PetscCall(PetscObjectDereference((PetscObject)container));
8344: PetscFunctionReturn(PETSC_SUCCESS);
8345: }
8347: /*@
8348: MatInvertVariableBlockEnvelope - set matrix C to be the inverted block diagonal of matrix A
8350: Collective
8352: Input Parameters:
8353: + A - the matrix
8354: - reuse - indicates if the `C` matrix was obtained from a previous call to this routine
8356: Output Parameter:
8357: . C - matrix with inverted block diagonal of `A`
8359: Level: advanced
8361: Note:
8362: For efficiency the matrix `A` should have all the nonzero entries clustered in smallish blocks along the diagonal.
8364: .seealso: [](ch_matrices), `Mat`, `MatInvertBlockDiagonal()`, `MatComputeBlockDiagonal()`
8365: @*/
8366: PetscErrorCode MatInvertVariableBlockEnvelope(Mat A, MatReuse reuse, Mat *C)
8367: {
8368: PetscContainer container;
8369: EnvelopeData *edata;
8370: PetscObjectState nonzerostate;
8372: PetscFunctionBegin;
8373: PetscCall(PetscObjectQuery((PetscObject)A, "EnvelopeData", (PetscObject *)&container));
8374: if (!container) {
8375: PetscCall(MatComputeVariableBlockEnvelope(A));
8376: PetscCall(PetscObjectQuery((PetscObject)A, "EnvelopeData", (PetscObject *)&container));
8377: }
8378: PetscCall(PetscContainerGetPointer(container, &edata));
8379: PetscCall(MatGetNonzeroState(A, &nonzerostate));
8380: PetscCheck(nonzerostate <= edata->nonzerostate, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Cannot handle changes to matrix nonzero structure");
8381: PetscCheck(reuse != MAT_REUSE_MATRIX || *C == edata->C, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "C matrix must be the same as previously output");
8383: PetscCall(MatCreateSubMatrices(A, edata->n, edata->is, edata->is, MAT_INITIAL_MATRIX, &edata->mat));
8384: *C = edata->C;
8386: for (PetscInt i = 0; i < edata->n; i++) {
8387: Mat D;
8388: PetscScalar *dvalues;
8390: PetscCall(MatConvert(edata->mat[i], MATSEQDENSE, MAT_INITIAL_MATRIX, &D));
8391: PetscCall(MatSetOption(*C, MAT_ROW_ORIENTED, PETSC_FALSE));
8392: PetscCall(MatSeqDenseInvert(D));
8393: PetscCall(MatDenseGetArray(D, &dvalues));
8394: PetscCall(MatSetValuesIS(*C, edata->is[i], edata->is[i], dvalues, INSERT_VALUES));
8395: PetscCall(MatDestroy(&D));
8396: }
8397: PetscCall(MatDestroySubMatrices(edata->n, &edata->mat));
8398: PetscCall(MatAssemblyBegin(*C, MAT_FINAL_ASSEMBLY));
8399: PetscCall(MatAssemblyEnd(*C, MAT_FINAL_ASSEMBLY));
8400: PetscFunctionReturn(PETSC_SUCCESS);
8401: }
8403: /*@
8404: MatSetVariableBlockSizes - Sets diagonal point-blocks of the matrix that need not be of the same size
8406: Not Collective
8408: Input Parameters:
8409: + mat - the matrix
8410: . nblocks - the number of blocks on this process, each block can only exist on a single MPI process
8411: - bsizes - the block sizes
8413: Level: intermediate
8415: Note:
8416: Currently used by `PCVPBJACOBI` for `MATAIJ` matrices
8418: .seealso: [](ch_matrices), `Mat`, `MatCreateSeqBAIJ()`, `MatCreateBAIJ()`, `MatGetBlockSize()`, `MatSetBlockSizes()`, `MatGetBlockSizes()`, `MatGetVariableBlockSizes()`,
8419: `MatComputeVariableBlockEnvelope()`, `PCVPBJACOBI`
8420: @*/
8421: PetscErrorCode MatSetVariableBlockSizes(Mat mat, PetscInt nblocks, const PetscInt bsizes[])
8422: {
8423: PetscInt ncnt = 0, nlocal;
8425: PetscFunctionBegin;
8427: PetscCall(MatGetLocalSize(mat, &nlocal, NULL));
8428: PetscCheck(nblocks >= 0 && nblocks <= nlocal, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Number of local blocks %" PetscInt_FMT " is not in [0, %" PetscInt_FMT "]", nblocks, nlocal);
8429: for (PetscInt i = 0; i < nblocks; i++) ncnt += bsizes[i];
8430: PetscCheck(ncnt == nlocal, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Sum of local block sizes %" PetscInt_FMT " does not equal local size of matrix %" PetscInt_FMT, ncnt, nlocal);
8431: PetscCall(PetscFree(mat->bsizes));
8432: mat->nblocks = nblocks;
8433: PetscCall(PetscMalloc1(nblocks, &mat->bsizes));
8434: PetscCall(PetscArraycpy(mat->bsizes, bsizes, nblocks));
8435: PetscFunctionReturn(PETSC_SUCCESS);
8436: }
8438: /*@
8439: MatGetVariableBlockSizes - Gets a diagonal blocks of the matrix that need not be of the same size
8441: Not Collective; No Fortran Support
8443: Input Parameter:
8444: . mat - the matrix
8446: Output Parameters:
8447: + nblocks - the number of blocks on this process
8448: - bsizes - the block sizes
8450: Level: intermediate
8452: .seealso: [](ch_matrices), `Mat`, `MatCreateSeqBAIJ()`, `MatCreateBAIJ()`, `MatGetBlockSize()`, `MatSetBlockSizes()`, `MatGetBlockSizes()`, `MatSetVariableBlockSizes()`, `MatComputeVariableBlockEnvelope()`
8453: @*/
8454: PetscErrorCode MatGetVariableBlockSizes(Mat mat, PetscInt *nblocks, const PetscInt *bsizes[])
8455: {
8456: PetscFunctionBegin;
8458: if (nblocks) *nblocks = mat->nblocks;
8459: if (bsizes) *bsizes = mat->bsizes;
8460: PetscFunctionReturn(PETSC_SUCCESS);
8461: }
8463: /*@
8464: MatSelectVariableBlockSizes - When creating a submatrix, pass on the variable block sizes
8466: Not Collective
8468: Input Parameters:
8469: + subA - the submatrix
8470: . A - the original matrix
8471: - isrow - The `IS` of selected rows for the submatrix, must be sorted
8473: Level: developer
8475: Note:
8476: If the index set is not sorted or contains off-process entries, this function will do nothing.
8478: .seealso: [](ch_matrices), `Mat`, `MatSetVariableBlockSizes()`, `MatComputeVariableBlockEnvelope()`
8479: @*/
8480: PetscErrorCode MatSelectVariableBlockSizes(Mat subA, Mat A, IS isrow)
8481: {
8482: const PetscInt *rows;
8483: PetscInt n, rStart, rEnd, Nb = 0;
8484: PetscBool flg = A->bsizes ? PETSC_TRUE : PETSC_FALSE;
8486: PetscFunctionBegin;
8487: // The code for block size extraction does not support an unsorted IS
8488: if (flg) PetscCall(ISSorted(isrow, &flg));
8489: // We don't support originally off-diagonal blocks
8490: if (flg) {
8491: PetscCall(MatGetOwnershipRange(A, &rStart, &rEnd));
8492: PetscCall(ISGetLocalSize(isrow, &n));
8493: PetscCall(ISGetIndices(isrow, &rows));
8494: for (PetscInt i = 0; i < n && flg; ++i) {
8495: if (rows[i] < rStart || rows[i] >= rEnd) flg = PETSC_FALSE;
8496: }
8497: PetscCall(ISRestoreIndices(isrow, &rows));
8498: }
8499: // quiet return if we can't extract block size
8500: PetscCallMPI(MPIU_Allreduce(MPI_IN_PLACE, &flg, 1, MPI_C_BOOL, MPI_LAND, PetscObjectComm((PetscObject)subA)));
8501: if (!flg) PetscFunctionReturn(PETSC_SUCCESS);
8503: // extract block sizes
8504: PetscCall(ISGetIndices(isrow, &rows));
8505: for (PetscInt b = 0, gr = rStart, i = 0; b < A->nblocks; ++b) {
8506: PetscBool occupied = PETSC_FALSE;
8508: for (PetscInt br = 0; br < A->bsizes[b]; ++br) {
8509: const PetscInt row = gr + br;
8511: if (i == n) break;
8512: if (rows[i] == row) {
8513: occupied = PETSC_TRUE;
8514: ++i;
8515: }
8516: while (i < n && rows[i] < row) ++i;
8517: }
8518: gr += A->bsizes[b];
8519: if (occupied) ++Nb;
8520: }
8521: subA->nblocks = Nb;
8522: PetscCall(PetscFree(subA->bsizes));
8523: PetscCall(PetscMalloc1(subA->nblocks, &subA->bsizes));
8524: PetscInt sb = 0;
8525: for (PetscInt b = 0, gr = rStart, i = 0; b < A->nblocks; ++b) {
8526: if (sb < subA->nblocks) subA->bsizes[sb] = 0;
8527: for (PetscInt br = 0; br < A->bsizes[b]; ++br) {
8528: const PetscInt row = gr + br;
8530: if (i == n) break;
8531: if (rows[i] == row) {
8532: ++subA->bsizes[sb];
8533: ++i;
8534: }
8535: while (i < n && rows[i] < row) ++i;
8536: }
8537: gr += A->bsizes[b];
8538: if (sb < subA->nblocks && subA->bsizes[sb]) ++sb;
8539: }
8540: PetscCheck(sb == subA->nblocks, PETSC_COMM_SELF, PETSC_ERR_PLIB, "Invalid number of blocks %" PetscInt_FMT " != %" PetscInt_FMT, sb, subA->nblocks);
8541: PetscInt nlocal, ncnt = 0;
8542: PetscCall(MatGetLocalSize(subA, &nlocal, NULL));
8543: PetscCheck(subA->nblocks >= 0 && subA->nblocks <= nlocal, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Number of local blocks %" PetscInt_FMT " is not in [0, %" PetscInt_FMT "]", subA->nblocks, nlocal);
8544: for (PetscInt i = 0; i < subA->nblocks; i++) ncnt += subA->bsizes[i];
8545: PetscCheck(ncnt == nlocal, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "Sum of local block sizes %" PetscInt_FMT " does not equal local size of matrix %" PetscInt_FMT, ncnt, nlocal);
8546: PetscCall(ISRestoreIndices(isrow, &rows));
8547: PetscFunctionReturn(PETSC_SUCCESS);
8548: }
8550: /*@
8551: MatSetBlockSizes - Sets the matrix block row and column sizes.
8553: Logically Collective
8555: Input Parameters:
8556: + mat - the matrix
8557: . rbs - row block size
8558: - cbs - column block size
8560: Level: intermediate
8562: Notes:
8563: Block row formats are `MATBAIJ` and `MATSBAIJ`. These formats ALWAYS have square block storage in the matrix.
8564: If you pass a different block size for the columns than the rows, the row block size determines the square block storage.
8565: This must be called before `MatSetUp()` or MatXXXSetPreallocation() (or will default to 1) and the block size cannot be changed later.
8567: For `MATAIJ` matrix this function can be called at a later stage, provided that the specified block sizes
8568: are compatible with the matrix local sizes.
8570: The row and column block size determine the blocksize of the "row" and "column" vectors returned by `MatCreateVecs()`.
8572: .seealso: [](ch_matrices), `Mat`, `MatCreateSeqBAIJ()`, `MatCreateBAIJ()`, `MatGetBlockSize()`, `MatSetBlockSize()`, `MatGetBlockSizes()`
8573: @*/
8574: PetscErrorCode MatSetBlockSizes(Mat mat, PetscInt rbs, PetscInt cbs)
8575: {
8576: PetscFunctionBegin;
8580: PetscTryTypeMethod(mat, setblocksizes, rbs, cbs);
8581: if (mat->rmap->refcnt) {
8582: ISLocalToGlobalMapping l2g = NULL;
8583: PetscLayout nmap = NULL;
8585: PetscCall(PetscLayoutDuplicate(mat->rmap, &nmap));
8586: if (mat->rmap->mapping) PetscCall(ISLocalToGlobalMappingDuplicate(mat->rmap->mapping, &l2g));
8587: PetscCall(PetscLayoutDestroy(&mat->rmap));
8588: mat->rmap = nmap;
8589: mat->rmap->mapping = l2g;
8590: }
8591: if (mat->cmap->refcnt) {
8592: ISLocalToGlobalMapping l2g = NULL;
8593: PetscLayout nmap = NULL;
8595: PetscCall(PetscLayoutDuplicate(mat->cmap, &nmap));
8596: if (mat->cmap->mapping) PetscCall(ISLocalToGlobalMappingDuplicate(mat->cmap->mapping, &l2g));
8597: PetscCall(PetscLayoutDestroy(&mat->cmap));
8598: mat->cmap = nmap;
8599: mat->cmap->mapping = l2g;
8600: }
8601: PetscCall(PetscLayoutSetBlockSize(mat->rmap, rbs));
8602: PetscCall(PetscLayoutSetBlockSize(mat->cmap, cbs));
8603: PetscFunctionReturn(PETSC_SUCCESS);
8604: }
8606: /*@
8607: MatSetBlockSizesFromMats - Sets the matrix block row and column sizes to match a pair of matrices
8609: Logically Collective
8611: Input Parameters:
8612: + mat - the matrix
8613: . fromRow - matrix from which to copy row block size
8614: - fromCol - matrix from which to copy column block size (can be same as `fromRow`)
8616: Level: developer
8618: .seealso: [](ch_matrices), `Mat`, `MatCreateSeqBAIJ()`, `MatCreateBAIJ()`, `MatGetBlockSize()`, `MatSetBlockSizes()`
8619: @*/
8620: PetscErrorCode MatSetBlockSizesFromMats(Mat mat, Mat fromRow, Mat fromCol)
8621: {
8622: PetscFunctionBegin;
8626: PetscTryTypeMethod(mat, setblocksizes, fromRow->rmap->bs, fromCol->cmap->bs);
8627: PetscCall(PetscLayoutSetBlockSize(mat->rmap, fromRow->rmap->bs));
8628: PetscCall(PetscLayoutSetBlockSize(mat->cmap, fromCol->cmap->bs));
8629: PetscFunctionReturn(PETSC_SUCCESS);
8630: }
8632: /*@
8633: MatResidual - Default routine to calculate the residual $r = b - Ax$
8635: Collective
8637: Input Parameters:
8638: + mat - the matrix
8639: . b - the right-hand-side
8640: - x - the approximate solution
8642: Output Parameter:
8643: . r - location to store the residual
8645: Level: developer
8647: .seealso: [](ch_matrices), `Mat`, `MatMult()`, `MatMultAdd()`, `PCMGSetResidual()`
8648: @*/
8649: PetscErrorCode MatResidual(Mat mat, Vec b, Vec x, Vec r)
8650: {
8651: PetscFunctionBegin;
8657: MatCheckPreallocated(mat, 1);
8658: PetscCall(PetscLogEventBegin(MAT_Residual, mat, 0, 0, 0));
8659: if (!mat->ops->residual) {
8660: PetscCall(MatMult(mat, x, r));
8661: PetscCall(VecAYPX(r, -1.0, b));
8662: } else {
8663: PetscUseTypeMethod(mat, residual, b, x, r);
8664: }
8665: PetscCall(PetscLogEventEnd(MAT_Residual, mat, 0, 0, 0));
8666: PetscFunctionReturn(PETSC_SUCCESS);
8667: }
8669: /*@
8670: MatGetRowIJ - Returns the compressed row storage i and j indices for the local rows of a sparse matrix
8672: Collective
8674: Input Parameters:
8675: + mat - the matrix
8676: . shift - 0 or 1 indicating we want the indices starting at 0 or 1
8677: . symmetric - `PETSC_TRUE` or `PETSC_FALSE` indicating the matrix data structure should be symmetrized
8678: - inodecompressed - `PETSC_TRUE` or `PETSC_FALSE` indicating if the nonzero structure of the
8679: inodes or the nonzero elements is wanted. For `MATBAIJ` matrices the compressed version is
8680: always used.
8682: Output Parameters:
8683: + n - number of local rows in the (possibly compressed) matrix, use `NULL` if not needed
8684: . ia - the row pointers; that is ia[0] = 0, ia[row] = ia[row-1] + number of elements in that row of the matrix, use `NULL` if not needed
8685: . ja - the column indices, use `NULL` if not needed
8686: - done - indicates if the routine actually worked and returned appropriate `ia` and `ja` arrays; callers
8687: are responsible for handling the case when done is `PETSC_FALSE` and `ia` and `ja` are not provided
8689: Level: developer
8691: Notes:
8692: You CANNOT change any of the `ia` or `ja` values.
8694: Use `MatRestoreRowIJ()` when you are finished accessing the `ia` and `ja` values.
8696: Fortran Notes:
8697: Use
8698: .vb
8699: PetscInt, pointer :: ia(:),ja(:)
8700: call MatGetRowIJ(mat,shift,symmetric,inodecompressed,n,ia,ja,done,ierr)
8701: ! Access the ith and jth entries via ia(i) and ja(j)
8702: .ve
8704: .seealso: [](ch_matrices), `Mat`, `MATAIJ`, `MatGetColumnIJ()`, `MatRestoreRowIJ()`, `MatSeqAIJGetArray()`
8705: @*/
8706: PetscErrorCode MatGetRowIJ(Mat mat, PetscInt shift, PetscBool symmetric, PetscBool inodecompressed, PetscInt *n, const PetscInt *ia[], const PetscInt *ja[], PetscBool *done)
8707: {
8708: PetscFunctionBegin;
8711: if (n) PetscAssertPointer(n, 5);
8712: if (ia) PetscAssertPointer(ia, 6);
8713: if (ja) PetscAssertPointer(ja, 7);
8714: if (done) PetscAssertPointer(done, 8);
8715: MatCheckPreallocated(mat, 1);
8716: if (!mat->ops->getrowij && done) *done = PETSC_FALSE;
8717: else {
8718: if (done) *done = PETSC_TRUE;
8719: PetscCall(PetscLogEventBegin(MAT_GetRowIJ, mat, 0, 0, 0));
8720: PetscUseTypeMethod(mat, getrowij, shift, symmetric, inodecompressed, n, ia, ja, done);
8721: PetscCall(PetscLogEventEnd(MAT_GetRowIJ, mat, 0, 0, 0));
8722: }
8723: PetscFunctionReturn(PETSC_SUCCESS);
8724: }
8726: /*@
8727: MatGetColumnIJ - Returns the compressed column storage i and j indices for sequential matrices.
8729: Collective
8731: Input Parameters:
8732: + mat - the matrix
8733: . shift - 1 or zero indicating we want the indices starting at 0 or 1
8734: . symmetric - `PETSC_TRUE` or `PETSC_FALSE` indicating the matrix data structure should be
8735: symmetrized
8736: - inodecompressed - `PETSC_TRUE` or `PETSC_FALSE` indicating if the nonzero structure of the
8737: inodes or the nonzero elements is wanted. For `MATBAIJ` matrices the compressed version is
8738: always used.
8740: Output Parameters:
8741: + n - number of columns in the (possibly compressed) matrix
8742: . ia - the column pointers; that is ia[0] = 0, ia[col] = i[col-1] + number of elements in that col of the matrix
8743: . ja - the row indices
8744: - done - `PETSC_TRUE` or `PETSC_FALSE`, indicating whether the values have been returned
8746: Level: developer
8748: .seealso: [](ch_matrices), `Mat`, `MatGetRowIJ()`, `MatRestoreColumnIJ()`
8749: @*/
8750: PetscErrorCode MatGetColumnIJ(Mat mat, PetscInt shift, PetscBool symmetric, PetscBool inodecompressed, PetscInt *n, const PetscInt *ia[], const PetscInt *ja[], PetscBool *done)
8751: {
8752: PetscFunctionBegin;
8755: PetscAssertPointer(n, 5);
8756: if (ia) PetscAssertPointer(ia, 6);
8757: if (ja) PetscAssertPointer(ja, 7);
8758: PetscAssertPointer(done, 8);
8759: MatCheckPreallocated(mat, 1);
8760: if (!mat->ops->getcolumnij) *done = PETSC_FALSE;
8761: else {
8762: *done = PETSC_TRUE;
8763: PetscUseTypeMethod(mat, getcolumnij, shift, symmetric, inodecompressed, n, ia, ja, done);
8764: }
8765: PetscFunctionReturn(PETSC_SUCCESS);
8766: }
8768: /*@
8769: MatRestoreRowIJ - Call after you are completed with the ia,ja indices obtained with `MatGetRowIJ()`.
8771: Collective
8773: Input Parameters:
8774: + mat - the matrix
8775: . shift - 1 or zero indicating we want the indices starting at 0 or 1
8776: . symmetric - `PETSC_TRUE` or `PETSC_FALSE` indicating the matrix data structure should be symmetrized
8777: . inodecompressed - `PETSC_TRUE` or `PETSC_FALSE` indicating if the nonzero structure of the
8778: inodes or the nonzero elements is wanted. For `MATBAIJ` matrices the compressed version is
8779: always used.
8780: . n - size of (possibly compressed) matrix, or `NULL`
8781: . ia - the row pointers, or `NULL`
8782: - ja - the column indices, or `NULL`
8784: Output Parameter:
8785: . done - `PETSC_TRUE` or `PETSC_FALSE` indicated that the values have been returned
8787: Level: developer
8789: Note:
8790: This routine zeros out `n`, `ia`, and `ja` if they are provided. Use of `ia` or `ja` after `MatRestoreRowIJ()` is always invalid.
8792: .seealso: [](ch_matrices), `Mat`, `MatGetRowIJ()`, `MatRestoreColumnIJ()`
8793: @*/
8794: PetscErrorCode MatRestoreRowIJ(Mat mat, PetscInt shift, PetscBool symmetric, PetscBool inodecompressed, PetscInt *n, const PetscInt *ia[], const PetscInt *ja[], PetscBool *done)
8795: {
8796: PetscFunctionBegin;
8799: if (ia) PetscAssertPointer(ia, 6);
8800: if (ja) PetscAssertPointer(ja, 7);
8801: if (done) PetscAssertPointer(done, 8);
8802: MatCheckPreallocated(mat, 1);
8804: if (!mat->ops->restorerowij && done) *done = PETSC_FALSE;
8805: else {
8806: if (done) *done = PETSC_TRUE;
8807: PetscUseTypeMethod(mat, restorerowij, shift, symmetric, inodecompressed, n, ia, ja, done);
8808: if (n) *n = 0;
8809: if (ia) *ia = NULL;
8810: if (ja) *ja = NULL;
8811: }
8812: PetscFunctionReturn(PETSC_SUCCESS);
8813: }
8815: /*@
8816: MatRestoreColumnIJ - Call after you are completed with the ia,ja indices obtained with `MatGetColumnIJ()`.
8818: Collective
8820: Input Parameters:
8821: + mat - the matrix
8822: . shift - 1 or zero indicating we want the indices starting at 0 or 1
8823: . symmetric - `PETSC_TRUE` or `PETSC_FALSE` indicating the matrix data structure should be symmetrized
8824: - inodecompressed - `PETSC_TRUE` or `PETSC_FALSE` indicating if the nonzero structure of the
8825: inodes or the nonzero elements is wanted. For `MATBAIJ` matrices the compressed version is
8826: always used.
8828: Output Parameters:
8829: + n - size of (possibly compressed) matrix
8830: . ia - the column pointers
8831: . ja - the row indices
8832: - done - `PETSC_TRUE` or `PETSC_FALSE` indicated that the values have been returned
8834: Level: developer
8836: .seealso: [](ch_matrices), `Mat`, `MatGetColumnIJ()`, `MatRestoreRowIJ()`
8837: @*/
8838: PetscErrorCode MatRestoreColumnIJ(Mat mat, PetscInt shift, PetscBool symmetric, PetscBool inodecompressed, PetscInt *n, const PetscInt *ia[], const PetscInt *ja[], PetscBool *done)
8839: {
8840: PetscFunctionBegin;
8843: if (ia) PetscAssertPointer(ia, 6);
8844: if (ja) PetscAssertPointer(ja, 7);
8845: PetscAssertPointer(done, 8);
8846: MatCheckPreallocated(mat, 1);
8848: if (!mat->ops->restorecolumnij) *done = PETSC_FALSE;
8849: else {
8850: *done = PETSC_TRUE;
8851: PetscUseTypeMethod(mat, restorecolumnij, shift, symmetric, inodecompressed, n, ia, ja, done);
8852: if (n) *n = 0;
8853: if (ia) *ia = NULL;
8854: if (ja) *ja = NULL;
8855: }
8856: PetscFunctionReturn(PETSC_SUCCESS);
8857: }
8859: /*@
8860: MatColoringPatch - Utility routine used inside matrix coloring routines that use `MatGetRowIJ()` and/or
8861: `MatGetColumnIJ()`.
8863: Collective
8865: Input Parameters:
8866: + mat - the matrix
8867: . ncolors - maximum color value
8868: . n - number of entries in `colorarray`
8869: - colorarray - array indicating color for each column
8871: Output Parameter:
8872: . iscoloring - coloring generated using `colorarray` information
8874: Level: developer
8876: .seealso: [](ch_matrices), `Mat`, `MatGetRowIJ()`, `MatGetColumnIJ()`
8877: @*/
8878: PetscErrorCode MatColoringPatch(Mat mat, PetscInt ncolors, PetscInt n, ISColoringValue colorarray[], ISColoring *iscoloring)
8879: {
8880: PetscFunctionBegin;
8883: PetscAssertPointer(colorarray, 4);
8884: PetscAssertPointer(iscoloring, 5);
8885: MatCheckPreallocated(mat, 1);
8887: if (!mat->ops->coloringpatch) {
8888: PetscCall(ISColoringCreate(PetscObjectComm((PetscObject)mat), ncolors, n, colorarray, PETSC_OWN_POINTER, iscoloring));
8889: } else {
8890: PetscUseTypeMethod(mat, coloringpatch, ncolors, n, colorarray, iscoloring);
8891: }
8892: PetscFunctionReturn(PETSC_SUCCESS);
8893: }
8895: /*@
8896: MatSetUnfactored - Resets a factored matrix to be treated as unfactored.
8898: Logically Collective
8900: Input Parameter:
8901: . mat - the factored matrix to be reset
8903: Level: developer
8905: Notes:
8906: This routine should be used only with factored matrices formed by in-place
8907: factorization via ILU(0) (or by in-place LU factorization for the `MATSEQDENSE`
8908: format). This option can save memory, for example, when solving nonlinear
8909: systems with a matrix-free Newton-Krylov method and a matrix-based, in-place
8910: ILU(0) preconditioner.
8912: One can specify in-place ILU(0) factorization by calling
8913: .vb
8914: PCType(pc,PCILU);
8915: PCFactorSeUseInPlace(pc);
8916: .ve
8917: or by using the options -pc_type ilu -pc_factor_in_place
8919: In-place factorization ILU(0) can also be used as a local
8920: solver for the blocks within the block Jacobi or additive Schwarz
8921: methods (runtime option: -sub_pc_factor_in_place). See Users-Manual: ch_pc
8922: for details on setting local solver options.
8924: Most users should employ the `KSP` interface for linear solvers
8925: instead of working directly with matrix algebra routines such as this.
8926: See, e.g., `KSPCreate()`.
8928: .seealso: [](ch_matrices), `Mat`, `PCFactorSetUseInPlace()`, `PCFactorGetUseInPlace()`
8929: @*/
8930: PetscErrorCode MatSetUnfactored(Mat mat)
8931: {
8932: PetscFunctionBegin;
8935: MatCheckPreallocated(mat, 1);
8936: mat->factortype = MAT_FACTOR_NONE;
8937: if (!mat->ops->setunfactored) PetscFunctionReturn(PETSC_SUCCESS);
8938: PetscUseTypeMethod(mat, setunfactored);
8939: PetscFunctionReturn(PETSC_SUCCESS);
8940: }
8942: /*@
8943: MatCreateSubMatrix - Gets a single submatrix on the same number of processes
8944: as the original matrix.
8946: Collective
8948: Input Parameters:
8949: + mat - the original matrix
8950: . isrow - parallel `IS` containing the rows this process should obtain
8951: . iscol - parallel `IS` containing all columns you wish to keep. Each process should list the columns that will be in IT's "diagonal part" in the new matrix.
8952: - cll - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
8954: Output Parameter:
8955: . newmat - the new submatrix, of the same type as the original matrix (except potentially for `MATSEQSBAIJ`)
8957: Level: advanced
8959: Notes:
8960: The submatrix will be able to be multiplied with vectors using the same layout as `iscol`.
8962: Some matrix types place restrictions on the row and column indices, such
8963: as that they be sorted or that they be equal to each other. For `MATBAIJ` and `MATSBAIJ` matrices the indices must include all rows/columns of a block;
8964: for example, if the block size is 3 one cannot select the 0 and 2 rows without selecting the 1 row.
8965: `MATSEQSBAIJ` inputs may produce a `MATSEQBAIJ` matrix when the row and column index sets do not preserve symmetry.
8967: The index sets may not have duplicate entries.
8969: The first time this is called you should use a `cll` of `MAT_INITIAL_MATRIX`,
8970: the `MatCreateSubMatrix()` routine will create the newmat for you. Any additional calls
8971: to this routine with a mat of the same nonzero structure and with a call of `MAT_REUSE_MATRIX`
8972: will reuse the matrix generated the first time. You should call `MatDestroy()` on `newmat` when
8973: you are finished using it.
8975: The communicator of the newly obtained matrix is ALWAYS the same as the communicator of
8976: the input matrix.
8978: If `iscol` is `NULL` then all columns are obtained (not supported in Fortran).
8980: If `isrow` and `iscol` have a nontrivial block-size, then the resulting matrix has this block-size as well. This feature
8981: is used by `PCFIELDSPLIT` to allow easy nesting of its use.
8983: Example usage:
8984: Consider the following 8x8 matrix with 34 non-zero values, that is
8985: assembled across 3 processes. Let's assume that proc0 owns 3 rows,
8986: proc1 owns 3 rows, proc2 owns 2 rows. This division can be shown
8987: as follows
8988: .vb
8989: 1 2 0 | 0 3 0 | 0 4
8990: Proc0 0 5 6 | 7 0 0 | 8 0
8991: 9 0 10 | 11 0 0 | 12 0
8992: -------------------------------------
8993: 13 0 14 | 15 16 17 | 0 0
8994: Proc1 0 18 0 | 19 20 21 | 0 0
8995: 0 0 0 | 22 23 0 | 24 0
8996: -------------------------------------
8997: Proc2 25 26 27 | 0 0 28 | 29 0
8998: 30 0 0 | 31 32 33 | 0 34
8999: .ve
9001: Suppose `isrow` = [0 1 | 4 | 6 7] and `iscol` = [1 2 | 3 4 5 | 6]. The resulting submatrix is
9003: .vb
9004: 2 0 | 0 3 0 | 0
9005: Proc0 5 6 | 7 0 0 | 8
9006: -------------------------------
9007: Proc1 18 0 | 19 20 21 | 0
9008: -------------------------------
9009: Proc2 26 27 | 0 0 28 | 29
9010: 0 0 | 31 32 33 | 0
9011: .ve
9013: .seealso: [](ch_matrices), `Mat`, `MatCreateSubMatrices()`, `MatCreateSubMatricesMPI()`, `MatCreateSubMatrixVirtual()`, `MatSubMatrixVirtualUpdate()`
9014: @*/
9015: PetscErrorCode MatCreateSubMatrix(Mat mat, IS isrow, IS iscol, MatReuse cll, Mat *newmat)
9016: {
9017: PetscMPIInt size;
9018: Mat *local;
9019: IS iscoltmp;
9020: PetscBool flg;
9022: PetscFunctionBegin;
9026: PetscAssertPointer(newmat, 5);
9029: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
9030: PetscCheck(cll != MAT_IGNORE_MATRIX, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Cannot use MAT_IGNORE_MATRIX");
9031: PetscCheck(cll != MAT_INPLACE_MATRIX, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Cannot use MAT_INPLACE_MATRIX");
9033: MatCheckPreallocated(mat, 1);
9034: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)mat), &size));
9036: if (!iscol || isrow == iscol) {
9037: PetscBool stride;
9038: PetscMPIInt grab = 0;
9039: PetscCall(PetscObjectTypeCompare((PetscObject)isrow, ISSTRIDE, &stride));
9040: if (stride) {
9041: PetscInt first, step, n, rstart, rend;
9042: PetscCall(ISStrideGetInfo(isrow, &first, &step));
9043: if (step == 1) {
9044: PetscCall(MatGetOwnershipRange(mat, &rstart, &rend));
9045: if (rstart == first) {
9046: PetscCall(ISGetLocalSize(isrow, &n));
9047: if (n == rend - rstart) grab = 1;
9048: }
9049: }
9050: }
9051: PetscCallMPI(MPIU_Allreduce(MPI_IN_PLACE, &grab, 1, MPI_INT, MPI_MIN, PetscObjectComm((PetscObject)mat)));
9052: if (grab) {
9053: PetscCall(PetscInfo(mat, "Getting entire matrix as submatrix\n"));
9054: if (cll == MAT_INITIAL_MATRIX) {
9055: *newmat = mat;
9056: PetscCall(PetscObjectReference((PetscObject)mat));
9057: }
9058: PetscFunctionReturn(PETSC_SUCCESS);
9059: }
9060: }
9062: if (!iscol) {
9063: PetscCall(ISCreateStride(PetscObjectComm((PetscObject)mat), mat->cmap->n, mat->cmap->rstart, 1, &iscoltmp));
9064: } else {
9065: iscoltmp = iscol;
9066: }
9068: /* if original matrix is on just one process then use submatrix generated */
9069: if (mat->ops->createsubmatrices && !mat->ops->createsubmatrix && size == 1 && cll == MAT_REUSE_MATRIX) {
9070: PetscCall(MatCreateSubMatrices(mat, 1, &isrow, &iscoltmp, MAT_REUSE_MATRIX, &newmat));
9071: goto setproperties;
9072: } else if (mat->ops->createsubmatrices && !mat->ops->createsubmatrix && size == 1) {
9073: PetscCall(MatCreateSubMatrices(mat, 1, &isrow, &iscoltmp, MAT_INITIAL_MATRIX, &local));
9074: *newmat = *local;
9075: PetscCall(PetscFree(local));
9076: goto setproperties;
9077: } else if (!mat->ops->createsubmatrix) {
9078: /* Create a new matrix type that implements the operation using the full matrix */
9079: PetscCall(PetscLogEventBegin(MAT_CreateSubMat, mat, 0, 0, 0));
9080: switch (cll) {
9081: case MAT_INITIAL_MATRIX:
9082: PetscCall(MatCreateSubMatrixVirtual(mat, isrow, iscoltmp, newmat));
9083: break;
9084: case MAT_REUSE_MATRIX:
9085: PetscCall(MatSubMatrixVirtualUpdate(*newmat, mat, isrow, iscoltmp));
9086: break;
9087: default:
9088: SETERRQ(PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_OUTOFRANGE, "Invalid MatReuse, must be either MAT_INITIAL_MATRIX or MAT_REUSE_MATRIX");
9089: }
9090: PetscCall(PetscLogEventEnd(MAT_CreateSubMat, mat, 0, 0, 0));
9091: goto setproperties;
9092: }
9094: PetscCall(PetscLogEventBegin(MAT_CreateSubMat, mat, 0, 0, 0));
9095: PetscUseTypeMethod(mat, createsubmatrix, isrow, iscoltmp, cll, newmat);
9096: PetscCall(PetscLogEventEnd(MAT_CreateSubMat, mat, 0, 0, 0));
9098: setproperties:
9099: if ((*newmat)->symmetric == PETSC_BOOL3_UNKNOWN && (*newmat)->structurally_symmetric == PETSC_BOOL3_UNKNOWN && (*newmat)->spd == PETSC_BOOL3_UNKNOWN && (*newmat)->hermitian == PETSC_BOOL3_UNKNOWN) {
9100: PetscCall(ISEqualUnsorted(isrow, iscoltmp, &flg));
9101: if (flg) PetscCall(MatPropagateSymmetryOptions(mat, *newmat));
9102: }
9103: if (!iscol) PetscCall(ISDestroy(&iscoltmp));
9104: if (*newmat && cll == MAT_INITIAL_MATRIX) PetscCall(PetscObjectStateIncrease((PetscObject)*newmat));
9105: if (!iscol || isrow == iscol) PetscCall(MatSelectVariableBlockSizes(*newmat, mat, isrow));
9106: PetscFunctionReturn(PETSC_SUCCESS);
9107: }
9109: /*@
9110: MatPropagateSymmetryOptions - Propagates symmetry properties set on a matrix to another matrix
9112: Not Collective
9114: Input Parameters:
9115: + A - the matrix we wish to propagate properties from
9116: - B - the matrix we wish to propagate properties to
9118: Level: beginner
9120: Note:
9121: Propagates the properties associated to `MAT_SYMMETRY_ETERNAL`, `MAT_STRUCTURALLY_SYMMETRIC`, `MAT_HERMITIAN`, `MAT_SPD`, `MAT_SYMMETRIC`, and `MAT_STRUCTURAL_SYMMETRY_ETERNAL`
9123: .seealso: [](ch_matrices), `Mat`, `MatSetOption()`, `MatIsSymmetricKnown()`, `MatIsSPDKnown()`, `MatIsHermitianKnown()`, `MatIsStructurallySymmetricKnown()`
9124: @*/
9125: PetscErrorCode MatPropagateSymmetryOptions(Mat A, Mat B)
9126: {
9127: PetscFunctionBegin;
9130: B->symmetry_eternal = A->symmetry_eternal;
9131: B->structural_symmetry_eternal = A->structural_symmetry_eternal;
9132: B->symmetric = A->symmetric;
9133: B->structurally_symmetric = A->structurally_symmetric;
9134: B->spd = A->spd;
9135: B->hermitian = A->hermitian;
9136: PetscFunctionReturn(PETSC_SUCCESS);
9137: }
9139: /*@
9140: MatStashSetInitialSize - sets the sizes of the matrix stash, that is
9141: used during the assembly process to store values that belong to
9142: other processes.
9144: Not Collective
9146: Input Parameters:
9147: + mat - the matrix
9148: . size - the initial size of the stash.
9149: - bsize - the initial size of the block-stash(if used).
9151: Options Database Key:
9152: . -matstash_initial_size (size|size0,size1,...,sizep-1) - set initial size of stash for all or each of the MPI processes, sets both block and non-block stash sizes
9154: Level: intermediate
9156: Notes:
9157: The block-stash is used for values set with `MatSetValuesBlocked()` while
9158: the stash is used for values set with `MatSetValues()`
9160: Run with the option `-info` and look for output of the form
9161: MatAssemblyBegin_MPIXXX:Stash has MM entries, uses nn mallocs.
9162: to determine the appropriate value, MM, to use for size and
9163: MatAssemblyBegin_MPIXXX:Block-Stash has BMM entries, uses nn mallocs.
9164: to determine the value, BMM to use for bsize
9166: .seealso: [](ch_matrices), `MatAssemblyBegin()`, `MatAssemblyEnd()`, `Mat`, `MatStashGetInfo()`
9167: @*/
9168: PetscErrorCode MatStashSetInitialSize(Mat mat, PetscInt size, PetscInt bsize)
9169: {
9170: PetscFunctionBegin;
9173: PetscCall(MatStashSetInitialSize_Private(&mat->stash, size));
9174: PetscCall(MatStashSetInitialSize_Private(&mat->bstash, bsize));
9175: PetscFunctionReturn(PETSC_SUCCESS);
9176: }
9178: /*@
9179: MatInterpolateAdd - $w = y + A*x$ or $A^T*x$ depending on the shape of
9180: the matrix
9182: Neighbor-wise Collective
9184: Input Parameters:
9185: + A - the matrix
9186: . x - the vector to be multiplied by the interpolation operator
9187: - y - the vector to be added to the result
9189: Output Parameter:
9190: . w - the resulting vector
9192: Level: intermediate
9194: Notes:
9195: `w` may be the same vector as `y`.
9197: This allows one to use either the restriction or interpolation (its transpose)
9198: matrix to do the interpolation
9200: .seealso: [](ch_matrices), `Mat`, `MatMultAdd()`, `MatMultTransposeAdd()`, `MatRestrict()`, `PCMG`
9201: @*/
9202: PetscErrorCode MatInterpolateAdd(Mat A, Vec x, Vec y, Vec w)
9203: {
9204: PetscInt M, N, Ny;
9206: PetscFunctionBegin;
9211: PetscCall(MatGetSize(A, &M, &N));
9212: PetscCall(VecGetSize(y, &Ny));
9213: if (M == Ny) PetscCall(MatMultAdd(A, x, y, w));
9214: else PetscCall(MatMultTransposeAdd(A, x, y, w));
9215: PetscFunctionReturn(PETSC_SUCCESS);
9216: }
9218: /*@
9219: MatInterpolate - $y = A*x$ or $A^T*x$ depending on the shape of
9220: the matrix
9222: Neighbor-wise Collective
9224: Input Parameters:
9225: + A - the matrix
9226: - x - the vector to be interpolated
9228: Output Parameter:
9229: . y - the resulting vector
9231: Level: intermediate
9233: Note:
9234: This allows one to use either the restriction or interpolation (its transpose)
9235: matrix to do the interpolation
9237: .seealso: [](ch_matrices), `Mat`, `MatMultAdd()`, `MatMultTransposeAdd()`, `MatRestrict()`, `PCMG`
9238: @*/
9239: PetscErrorCode MatInterpolate(Mat A, Vec x, Vec y)
9240: {
9241: PetscInt M, N, Ny;
9243: PetscFunctionBegin;
9247: PetscCall(MatGetSize(A, &M, &N));
9248: PetscCall(VecGetSize(y, &Ny));
9249: if (M == Ny) PetscCall(MatMult(A, x, y));
9250: else PetscCall(MatMultTranspose(A, x, y));
9251: PetscFunctionReturn(PETSC_SUCCESS);
9252: }
9254: /*@
9255: MatRestrict - $y = A*x$ or $A^T*x$
9257: Neighbor-wise Collective
9259: Input Parameters:
9260: + A - the matrix
9261: - x - the vector to be restricted
9263: Output Parameter:
9264: . y - the resulting vector
9266: Level: intermediate
9268: Note:
9269: This allows one to use either the restriction or interpolation (its transpose)
9270: matrix to do the restriction
9272: .seealso: [](ch_matrices), `Mat`, `MatMultAdd()`, `MatMultTransposeAdd()`, `MatInterpolate()`, `PCMG`
9273: @*/
9274: PetscErrorCode MatRestrict(Mat A, Vec x, Vec y)
9275: {
9276: PetscInt M, N, Nx;
9278: PetscFunctionBegin;
9282: PetscCall(MatGetSize(A, &M, &N));
9283: PetscCall(VecGetSize(x, &Nx));
9284: if (M == Nx) PetscCall(MatMultTranspose(A, x, y));
9285: else PetscCall(MatMult(A, x, y));
9286: PetscFunctionReturn(PETSC_SUCCESS);
9287: }
9289: /*@
9290: MatMatInterpolateAdd - $Y = W + A*X$ or $W + A^T*X$ depending on the shape of `A`
9292: Neighbor-wise Collective
9294: Input Parameters:
9295: + A - the matrix
9296: . x - the input dense matrix to be multiplied
9297: - w - the input dense matrix to be added to the result
9299: Output Parameter:
9300: . y - the output dense matrix
9302: Level: intermediate
9304: Note:
9305: This allows one to use either the restriction or interpolation (its transpose)
9306: matrix to do the interpolation. `y` matrix can be reused if already created with the proper sizes,
9307: otherwise it will be recreated. `y` must be initialized to `NULL` if not supplied.
9309: .seealso: [](ch_matrices), `Mat`, `MatInterpolateAdd()`, `MatMatInterpolate()`, `MatMatRestrict()`, `PCMG`
9310: @*/
9311: PetscErrorCode MatMatInterpolateAdd(Mat A, Mat x, Mat w, Mat *y)
9312: {
9313: PetscInt M, N, Mx, Nx, Mo, My = 0, Ny = 0;
9314: PetscBool trans = PETSC_TRUE;
9315: MatReuse reuse = MAT_INITIAL_MATRIX;
9317: PetscFunctionBegin;
9323: PetscCall(MatGetSize(A, &M, &N));
9324: PetscCall(MatGetSize(x, &Mx, &Nx));
9325: if (N == Mx) trans = PETSC_FALSE;
9326: else PetscCheck(M == Mx, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Size mismatch: A %" PetscInt_FMT "x%" PetscInt_FMT ", X %" PetscInt_FMT "x%" PetscInt_FMT, M, N, Mx, Nx);
9327: Mo = trans ? N : M;
9328: if (*y) {
9329: PetscCall(MatGetSize(*y, &My, &Ny));
9330: if (Mo == My && Nx == Ny) reuse = MAT_REUSE_MATRIX;
9331: else {
9332: PetscCheck(w || *y != w, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Cannot reuse y and w, size mismatch: A %" PetscInt_FMT "x%" PetscInt_FMT ", X %" PetscInt_FMT "x%" PetscInt_FMT ", Y %" PetscInt_FMT "x%" PetscInt_FMT, M, N, Mx, Nx, My, Ny);
9333: PetscCall(MatDestroy(y));
9334: }
9335: }
9337: if (w && *y == w) { /* this is to minimize changes in PCMG */
9338: PetscBool flg;
9340: PetscCall(PetscObjectQuery((PetscObject)*y, "__MatMatIntAdd_w", (PetscObject *)&w));
9341: if (w) {
9342: PetscInt My, Ny, Mw, Nw;
9344: PetscCall(PetscObjectTypeCompare((PetscObject)*y, ((PetscObject)w)->type_name, &flg));
9345: PetscCall(MatGetSize(*y, &My, &Ny));
9346: PetscCall(MatGetSize(w, &Mw, &Nw));
9347: if (!flg || My != Mw || Ny != Nw) w = NULL;
9348: }
9349: if (!w) {
9350: PetscCall(MatDuplicate(*y, MAT_COPY_VALUES, &w));
9351: PetscCall(PetscObjectCompose((PetscObject)*y, "__MatMatIntAdd_w", (PetscObject)w));
9352: PetscCall(PetscObjectDereference((PetscObject)w));
9353: } else PetscCall(MatCopy(*y, w, UNKNOWN_NONZERO_PATTERN));
9354: }
9355: if (!trans) PetscCall(MatMatMult(A, x, reuse, PETSC_DETERMINE, y));
9356: else PetscCall(MatTransposeMatMult(A, x, reuse, PETSC_DETERMINE, y));
9357: if (w) PetscCall(MatAXPY(*y, 1.0, w, UNKNOWN_NONZERO_PATTERN));
9358: PetscFunctionReturn(PETSC_SUCCESS);
9359: }
9361: /*@
9362: MatMatInterpolate - $Y = A*X$ or $A^T*X$ depending on the shape of `A`
9364: Neighbor-wise Collective
9366: Input Parameters:
9367: + A - the matrix
9368: - x - the input dense matrix
9370: Output Parameter:
9371: . y - the output dense matrix
9373: Level: intermediate
9375: Note:
9376: This allows one to use either the restriction or interpolation (its transpose)
9377: matrix to do the interpolation. `y` matrix can be reused if already created with the proper sizes,
9378: otherwise it will be recreated. `y` must be initialized to `NULL` if not supplied.
9380: .seealso: [](ch_matrices), `Mat`, `MatInterpolate()`, `MatRestrict()`, `MatMatRestrict()`, `PCMG`
9381: @*/
9382: PetscErrorCode MatMatInterpolate(Mat A, Mat x, Mat *y)
9383: {
9384: PetscFunctionBegin;
9385: PetscCall(MatMatInterpolateAdd(A, x, NULL, y));
9386: PetscFunctionReturn(PETSC_SUCCESS);
9387: }
9389: /*@
9390: MatMatRestrict - $Y = A*X$ or $A^T*X$ depending on the shape of `A`
9392: Neighbor-wise Collective
9394: Input Parameters:
9395: + A - the matrix
9396: - x - the input dense matrix
9398: Output Parameter:
9399: . y - the output dense matrix
9401: Level: intermediate
9403: Note:
9404: This allows one to use either the restriction or interpolation (its transpose)
9405: matrix to do the restriction. `y` matrix can be reused if already created with the proper sizes,
9406: otherwise it will be recreated. `y` must be initialized to `NULL` if not supplied.
9408: .seealso: [](ch_matrices), `Mat`, `MatRestrict()`, `MatInterpolate()`, `MatMatInterpolate()`, `PCMG`
9409: @*/
9410: PetscErrorCode MatMatRestrict(Mat A, Mat x, Mat *y)
9411: {
9412: PetscFunctionBegin;
9413: PetscCall(MatMatInterpolateAdd(A, x, NULL, y));
9414: PetscFunctionReturn(PETSC_SUCCESS);
9415: }
9417: /*@
9418: MatGetNullSpace - retrieves the null space of a matrix that was provided with `MatSetNullSpace()`
9420: Logically Collective
9422: Input Parameters:
9423: + mat - the matrix
9424: - nullsp - the null space object
9426: Level: developer
9428: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `MatNullSpaceCreate()`, `MatSetNearNullSpace()`, `MatSetNullSpace()`, `MatNullSpace`
9429: @*/
9430: PetscErrorCode MatGetNullSpace(Mat mat, MatNullSpace *nullsp)
9431: {
9432: PetscFunctionBegin;
9434: PetscAssertPointer(nullsp, 2);
9435: *nullsp = (mat->symmetric == PETSC_BOOL3_TRUE && !mat->nullsp) ? mat->transnullsp : mat->nullsp;
9436: PetscFunctionReturn(PETSC_SUCCESS);
9437: }
9439: /*@
9440: MatGetNullSpaces - gets the null spaces, transpose null spaces, and near null spaces from an array of matrices that were supplied with `MatSetNullSpace()`,
9441: `MatSetTransposeNullSpace()`, and `MatSetNearNullSpace()`
9443: Logically Collective
9445: Input Parameters:
9446: + n - the number of matrices
9447: - mat - the array of matrices
9449: Output Parameters:
9450: . nullsp - an array of null spaces, `NULL` will be inserted for each matrix that does not have a null space, length 3 * `n`
9452: Level: developer
9454: Note:
9455: Call `MatRestoreNullspaces()` to provide these to another array of matrices
9457: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `MatNullSpaceCreate()`, `MatSetNearNullSpace()`, `MatGetNullSpace()`, `MatSetTransposeNullSpace()`, `MatGetTransposeNullSpace()`,
9458: `MatNullSpaceRemove()`, `MatRestoreNullSpaces()`, `MatNullSpace`
9459: @*/
9460: PetscErrorCode MatGetNullSpaces(PetscInt n, Mat mat[], MatNullSpace *nullsp[])
9461: {
9462: PetscFunctionBegin;
9463: PetscCheck(n >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Number of matrices %" PetscInt_FMT " must be non-negative", n);
9464: PetscAssertPointer(mat, 2);
9465: PetscAssertPointer(nullsp, 3);
9467: PetscCall(PetscCalloc1(3 * n, nullsp));
9468: for (PetscInt i = 0; i < n; i++) {
9470: (*nullsp)[i] = mat[i]->nullsp;
9471: PetscCall(PetscObjectReference((PetscObject)(*nullsp)[i]));
9472: (*nullsp)[n + i] = mat[i]->nearnullsp;
9473: PetscCall(PetscObjectReference((PetscObject)(*nullsp)[n + i]));
9474: (*nullsp)[2 * n + i] = mat[i]->transnullsp;
9475: PetscCall(PetscObjectReference((PetscObject)(*nullsp)[2 * n + i]));
9476: }
9477: PetscFunctionReturn(PETSC_SUCCESS);
9478: }
9480: /*@
9481: MatRestoreNullSpaces - sets the null spaces, transpose null spaces, and near null spaces obtained with `MatGetNullSpaces()` for an array of matrices
9483: Logically Collective
9485: Input Parameters:
9486: + n - the number of matrices
9487: . mat - the array of matrices
9488: - nullsp - an array of null spaces, of length 3 * `n`
9490: Level: developer
9492: Notes:
9493: Call `MatGetNullSpaces()` to create `nullsp`.
9495: Frees `nullsp`.
9497: Developer Note:
9498: The name of this function is confusing. Traditionally in PETSc, a restore operation undoes something that was previously done on an object (or objects) with a get operation.
9499: This restore routine does something to a new set of objects using the results of a get operation on a previous set of objects. Perhaps this routine
9500: should have simply been called `MatSetNullSpaces()`
9502: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `MatNullSpaceCreate()`, `MatSetNearNullSpace()`, `MatGetNullSpace()`, `MatSetTransposeNullSpace()`, `MatGetTransposeNullSpace()`,
9503: `MatNullSpaceRemove()`, `MatGetNullSpaces()`, `MatNullSpace`
9504: @*/
9505: PetscErrorCode MatRestoreNullSpaces(PetscInt n, Mat mat[], MatNullSpace *nullsp[])
9506: {
9507: PetscFunctionBegin;
9508: PetscCheck(n >= 0, PETSC_COMM_SELF, PETSC_ERR_ARG_OUTOFRANGE, "Number of matrices %" PetscInt_FMT " must be non-negative", n);
9509: PetscAssertPointer(mat, 2);
9510: PetscAssertPointer(nullsp, 3);
9511: PetscAssertPointer(*nullsp, 3);
9513: for (PetscInt i = 0; i < n; i++) {
9515: PetscCall(MatSetNullSpace(mat[i], (*nullsp)[i]));
9516: PetscCall(PetscObjectDereference((PetscObject)(*nullsp)[i]));
9517: PetscCall(MatSetNearNullSpace(mat[i], (*nullsp)[n + i]));
9518: PetscCall(PetscObjectDereference((PetscObject)(*nullsp)[n + i]));
9519: PetscCall(MatSetTransposeNullSpace(mat[i], (*nullsp)[2 * n + i]));
9520: PetscCall(PetscObjectDereference((PetscObject)(*nullsp)[2 * n + i]));
9521: }
9522: PetscCall(PetscFree(*nullsp));
9523: PetscFunctionReturn(PETSC_SUCCESS);
9524: }
9526: /*@
9527: MatSetNullSpace - attaches a null space to a matrix.
9529: Logically Collective
9531: Input Parameters:
9532: + mat - the matrix
9533: - nullsp - the null space object
9535: Level: advanced
9537: Notes:
9538: This null space is used by the `KSP` linear solvers to solve singular systems.
9540: Overwrites any previous null space that may have been attached. You can remove the null space from the matrix object by calling this routine with an nullsp of `NULL`
9542: For inconsistent singular systems (linear systems where the right-hand side is not in the range of the operator) the `KSP` residuals will not converge
9543: to zero but the linear system will still be solved in a least squares sense.
9545: The fundamental theorem of linear algebra (Gilbert Strang, Introduction to Applied Mathematics, page 72) states that
9546: the domain of a matrix $A$ (from $R^n$ to $R^m$ ($m$ rows, $n$ columns) $R^n$ = the direct sum of the null space of $A$, $n(A)$, plus the range of $A^T$, $R(A^T)$.
9547: Similarly $R^m$ = direct sum $n(A^T) + R(A)$. Hence the linear system $A x = b$ has a solution only if $b$ in $R(A)$ (or correspondingly $b$ is orthogonal to
9548: $n(A^T))$ and if $x$ is a solution then $x + \alpha n(A)$ is a solution for any $\alpha$. The minimum norm solution is orthogonal to $n(A)$. For problems without a solution
9549: the solution that minimizes the norm of the residual (the least squares solution) can be obtained by solving $A x = \hat{b}$ where $\hat{b}$ is $b$ orthogonalized to the $n(A^T)$.
9550: This $\hat{b}$ can be obtained by calling `MatNullSpaceRemove()` with the null space of the transpose of the matrix.
9552: If the matrix is known to be symmetric because it is an `MATSBAIJ` matrix or one has called
9553: `MatSetOption`(mat,`MAT_SYMMETRIC` or possibly `MAT_SYMMETRY_ETERNAL`,`PETSC_TRUE`); this
9554: routine also automatically calls `MatSetTransposeNullSpace()`.
9556: The user should call `MatNullSpaceDestroy()`.
9558: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `MatNullSpaceCreate()`, `MatSetNearNullSpace()`, `MatGetNullSpace()`, `MatSetTransposeNullSpace()`, `MatGetTransposeNullSpace()`, `MatNullSpaceRemove()`,
9559: `KSPSetPCSide()`, `MatNullSpace`
9560: @*/
9561: PetscErrorCode MatSetNullSpace(Mat mat, MatNullSpace nullsp)
9562: {
9563: PetscFunctionBegin;
9566: PetscCall(PetscObjectReference((PetscObject)nullsp));
9567: PetscCall(MatNullSpaceDestroy(&mat->nullsp));
9568: mat->nullsp = nullsp;
9569: if (mat->symmetric == PETSC_BOOL3_TRUE) PetscCall(MatSetTransposeNullSpace(mat, nullsp));
9570: PetscFunctionReturn(PETSC_SUCCESS);
9571: }
9573: /*@
9574: MatGetTransposeNullSpace - retrieves the null space of the transpose of a matrix that was set with `MatSetTransposeNullSpace()`
9576: Logically Collective
9578: Input Parameters:
9579: + mat - the matrix
9580: - nullsp - the null space object
9582: Level: developer
9584: .seealso: [](ch_matrices), `Mat`, `MatNullSpace`, `MatCreate()`, `MatNullSpaceCreate()`, `MatSetNearNullSpace()`, `MatSetTransposeNullSpace()`, `MatSetNullSpace()`, `MatGetNullSpace()`
9585: @*/
9586: PetscErrorCode MatGetTransposeNullSpace(Mat mat, MatNullSpace *nullsp)
9587: {
9588: PetscFunctionBegin;
9591: PetscAssertPointer(nullsp, 2);
9592: *nullsp = (mat->symmetric == PETSC_BOOL3_TRUE && !mat->transnullsp) ? mat->nullsp : mat->transnullsp;
9593: PetscFunctionReturn(PETSC_SUCCESS);
9594: }
9596: /*@
9597: MatSetTransposeNullSpace - attaches the null space of a transpose of a matrix to the matrix
9599: Logically Collective
9601: Input Parameters:
9602: + mat - the matrix
9603: - nullsp - the null space object
9605: Level: advanced
9607: Notes:
9608: This allows solving singular linear systems defined by the transpose of the matrix using `KSP` solvers with left preconditioning.
9610: See `MatSetNullSpace()`
9612: .seealso: [](ch_matrices), `Mat`, `MatNullSpace`, `MatCreate()`, `MatNullSpaceCreate()`, `MatSetNearNullSpace()`, `MatGetNullSpace()`, `MatSetNullSpace()`, `MatGetTransposeNullSpace()`, `MatNullSpaceRemove()`, `KSPSetPCSide()`
9613: @*/
9614: PetscErrorCode MatSetTransposeNullSpace(Mat mat, MatNullSpace nullsp)
9615: {
9616: PetscFunctionBegin;
9619: PetscCall(PetscObjectReference((PetscObject)nullsp));
9620: PetscCall(MatNullSpaceDestroy(&mat->transnullsp));
9621: mat->transnullsp = nullsp;
9622: PetscFunctionReturn(PETSC_SUCCESS);
9623: }
9625: /*@
9626: MatSetNearNullSpace - attaches a null space to a matrix, which is often the null space (rigid body modes) of the operator without boundary conditions
9627: This null space will be used to provide near null space vectors to a multigrid preconditioner built from this matrix.
9629: Logically Collective
9631: Input Parameters:
9632: + mat - the matrix
9633: - nullsp - the null space object
9635: Level: advanced
9637: Notes:
9638: Overwrites any previous near null space that may have been attached
9640: You can remove the null space by calling this routine with an `nullsp` of `NULL`
9642: .seealso: [](ch_matrices), `Mat`, `MatNullSpace`, `MatCreate()`, `MatNullSpaceCreate()`, `MatSetNullSpace()`, `MatNullSpaceCreateRigidBody()`, `MatGetNearNullSpace()`
9643: @*/
9644: PetscErrorCode MatSetNearNullSpace(Mat mat, MatNullSpace nullsp)
9645: {
9646: PetscFunctionBegin;
9650: MatCheckPreallocated(mat, 1);
9651: PetscCall(PetscObjectReference((PetscObject)nullsp));
9652: PetscCall(MatNullSpaceDestroy(&mat->nearnullsp));
9653: mat->nearnullsp = nullsp;
9654: PetscFunctionReturn(PETSC_SUCCESS);
9655: }
9657: /*@
9658: MatGetNearNullSpace - Get null space from a matrix that was attached with `MatSetNearNullSpace()`
9660: Not Collective
9662: Input Parameter:
9663: . mat - the matrix
9665: Output Parameter:
9666: . nullsp - the null space object, `NULL` if not set
9668: Level: advanced
9670: .seealso: [](ch_matrices), `Mat`, `MatNullSpace`, `MatSetNearNullSpace()`, `MatGetNullSpace()`, `MatNullSpaceCreate()`
9671: @*/
9672: PetscErrorCode MatGetNearNullSpace(Mat mat, MatNullSpace *nullsp)
9673: {
9674: PetscFunctionBegin;
9677: PetscAssertPointer(nullsp, 2);
9678: MatCheckPreallocated(mat, 1);
9679: *nullsp = mat->nearnullsp;
9680: PetscFunctionReturn(PETSC_SUCCESS);
9681: }
9683: /*@
9684: MatICCFactor - Performs in-place incomplete Cholesky factorization of matrix.
9686: Collective
9688: Input Parameters:
9689: + mat - the matrix
9690: . row - row/column permutation
9691: - info - information on desired factorization process
9693: Level: developer
9695: Notes:
9696: Probably really in-place only when level of fill is zero, otherwise allocates
9697: new space to store factored matrix and deletes previous memory.
9699: Most users should employ the `KSP` interface for linear solvers
9700: instead of working directly with matrix algebra routines such as this.
9701: See, e.g., `KSPCreate()`.
9703: Fortran Note:
9704: A valid (non-null) `info` argument must be provided
9706: .seealso: [](ch_matrices), `Mat`, `MatFactorInfo`, `MatGetFactor()`, `MatICCFactorSymbolic()`, `MatLUFactorNumeric()`, `MatCholeskyFactor()`
9707: @*/
9708: PetscErrorCode MatICCFactor(Mat mat, IS row, const MatFactorInfo *info)
9709: {
9710: PetscFunctionBegin;
9714: PetscAssertPointer(info, 3);
9715: PetscCheck(mat->rmap->N == mat->cmap->N, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONG, "matrix must be square");
9716: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
9717: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
9718: MatCheckPreallocated(mat, 1);
9719: PetscUseTypeMethod(mat, iccfactor, row, info);
9720: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
9721: PetscFunctionReturn(PETSC_SUCCESS);
9722: }
9724: /*@
9725: MatDiagonalScaleLocal - Scales columns of a matrix given the scaling values including the
9726: ghosted ones.
9728: Not Collective
9730: Input Parameters:
9731: + mat - the matrix
9732: - diag - the diagonal values, including ghost ones
9734: Level: developer
9736: Notes:
9737: Works only for `MATMPIAIJ` and `MATMPIBAIJ` matrices
9739: `diag` is a sequential vector that has a length which is the same as the local (ghosted) length of the vector associated with
9740: the matrix's `ISLocalToGlobalMapping` set with `MatSetLocalToGlobalMapping()`.
9742: This allows one to avoid during communication to perform the scaling that must be done with `MatDiagonalScale()`.
9744: .seealso: [](ch_matrices), `Mat`, `MatDiagonalScale()`, `MatSetLocalToGlobalMapping()`, `ISLocalToGlobalMapping`
9745: @*/
9746: PetscErrorCode MatDiagonalScaleLocal(Mat mat, Vec diag)
9747: {
9748: PetscMPIInt size;
9750: PetscFunctionBegin;
9755: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Matrix must be already assembled");
9756: PetscCall(PetscLogEventBegin(MAT_Scale, mat, 0, 0, 0));
9757: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)mat), &size));
9758: if (size == 1) {
9759: PetscInt n, m;
9760: PetscCall(VecGetSize(diag, &n));
9761: PetscCall(MatGetSize(mat, NULL, &m));
9762: PetscCheck(m == n, PETSC_COMM_SELF, PETSC_ERR_SUP, "Only supported for sequential matrices when no ghost points/periodic conditions");
9763: PetscCall(MatDiagonalScale(mat, NULL, diag));
9764: } else PetscUseMethod(mat, "MatDiagonalScaleLocal_C", (Mat, Vec), (mat, diag));
9765: PetscCall(PetscLogEventEnd(MAT_Scale, mat, 0, 0, 0));
9766: PetscCall(PetscObjectStateIncrease((PetscObject)mat));
9767: PetscFunctionReturn(PETSC_SUCCESS);
9768: }
9770: /*@
9771: MatGetInertia - Gets the inertia from a factored matrix
9773: Collective
9775: Input Parameter:
9776: . mat - the matrix
9778: Output Parameters:
9779: + nneg - number of negative eigenvalues
9780: . nzero - number of zero eigenvalues
9781: - npos - number of positive eigenvalues
9783: Level: advanced
9785: Note:
9786: Matrix must have been factored by `MatCholeskyFactor()`
9788: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatCholeskyFactor()`
9789: @*/
9790: PetscErrorCode MatGetInertia(Mat mat, PetscInt *nneg, PetscInt *nzero, PetscInt *npos)
9791: {
9792: PetscFunctionBegin;
9795: PetscCheck(mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Unfactored matrix");
9796: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Numeric factor mat is not assembled");
9797: PetscUseTypeMethod(mat, getinertia, nneg, nzero, npos);
9798: PetscFunctionReturn(PETSC_SUCCESS);
9799: }
9801: /*@
9802: MatSolves - Solves $A x = b$, given a factored matrix, for a collection of vectors
9804: Neighbor-wise Collective
9806: Input Parameters:
9807: + mat - the factored matrix obtained with `MatGetFactor()`
9808: - b - the right-hand-side vectors
9810: Output Parameter:
9811: . x - the result vectors
9813: Level: developer
9815: Note:
9816: The vectors `b` and `x` cannot be the same. I.e., one cannot
9817: call `MatSolves`(A,x,x).
9819: .seealso: [](ch_matrices), `Mat`, `Vecs`, `MatGetFactor()`, `MatSolveAdd()`, `MatSolveTranspose()`, `MatSolveTransposeAdd()`, `MatSolve()`
9820: @*/
9821: PetscErrorCode MatSolves(Mat mat, Vecs b, Vecs x)
9822: {
9823: PetscFunctionBegin;
9826: PetscCheck(x != b, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_IDN, "x and b must be different vectors");
9827: PetscCheck(mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Unfactored matrix");
9828: if (!mat->rmap->N && !mat->cmap->N) PetscFunctionReturn(PETSC_SUCCESS);
9830: MatCheckPreallocated(mat, 1);
9831: PetscCall(PetscLogEventBegin(MAT_Solves, mat, 0, 0, 0));
9832: PetscUseTypeMethod(mat, solves, b, x);
9833: PetscCall(PetscLogEventEnd(MAT_Solves, mat, 0, 0, 0));
9834: PetscFunctionReturn(PETSC_SUCCESS);
9835: }
9837: /*@
9838: MatIsSymmetric - Test whether a matrix is symmetric
9840: Collective
9842: Input Parameters:
9843: + A - the matrix to test
9844: - tol - difference between value and its transpose less than this amount counts as equal (use 0.0 for exact transpose)
9846: Output Parameter:
9847: . flg - the result
9849: Level: intermediate
9851: Notes:
9852: For real numbers `MatIsSymmetric()` and `MatIsHermitian()` return identical results
9854: If the matrix does not yet know if it is symmetric or not this can be an expensive operation, also available `MatIsSymmetricKnown()`
9856: One can declare that a matrix is symmetric with `MatSetOption`(mat,`MAT_SYMMETRIC`,`PETSC_TRUE`) and if it is known to remain symmetric
9857: after changes to the matrices values one can call `MatSetOption`(mat,`MAT_SYMMETRY_ETERNAL`,`PETSC_TRUE`). If these properties
9858: have been set then `MatIsSymmetric()` does not need to perform any computations and returns immediately with the result.
9860: .seealso: [](ch_matrices), `Mat`, `MatTranspose()`, `MatIsTranspose()`, `MatIsHermitian()`, `MatIsStructurallySymmetric()`, `MatSetOption()`, `MatIsSymmetricKnown()`,
9861: `MAT_SYMMETRIC`, `MAT_SYMMETRY_ETERNAL`
9862: @*/
9863: PetscErrorCode MatIsSymmetric(Mat A, PetscReal tol, PetscBool *flg)
9864: {
9865: PetscFunctionBegin;
9867: PetscAssertPointer(flg, 3);
9868: if (A->symmetric != PETSC_BOOL3_UNKNOWN && !tol) *flg = PetscBool3ToBool(A->symmetric);
9869: else {
9870: if (A->ops->issymmetric) PetscUseTypeMethod(A, issymmetric, tol, flg);
9871: else PetscCall(MatIsTranspose(A, A, tol, flg));
9872: if (!tol) PetscCall(MatSetOption(A, MAT_SYMMETRIC, *flg));
9873: }
9874: PetscFunctionReturn(PETSC_SUCCESS);
9875: }
9877: /*@
9878: MatIsHermitian - Test whether a matrix is Hermitian
9880: Collective
9882: Input Parameters:
9883: + A - the matrix to test
9884: - tol - difference between value and its transpose less than this amount counts as equal (use 0.0 for exact Hermitian)
9886: Output Parameter:
9887: . flg - the result
9889: Level: intermediate
9891: Notes:
9892: For real numbers `MatIsSymmetric()` and `MatIsHermitian()` return identical results
9894: If the matrix does not yet know if it is Hermitian or not this can be an expensive operation, also available `MatIsHermitianKnown()`
9896: One can declare that a matrix is Hermitian with `MatSetOption`(mat,`MAT_HERMITIAN`,`PETSC_TRUE`) and if it is known to remain Hermitian
9897: after changes to the matrices values one can call `MatSetOption`(mat,`MAT_SYMMETRY_ETERNAL`,`PETSC_TRUE`). If these properties
9898: have been set then `MatIsHermitian()` does not need to perform any computations and returns immediately with the result.
9900: .seealso: [](ch_matrices), `Mat`, `MatTranspose()`, `MatIsTranspose()`, `MatIsHermitianKnown()`, `MatIsStructurallySymmetric()`, `MatSetOption()`,
9901: `MatIsSymmetricKnown()`, `MatIsSymmetric()`, `MAT_HERMITIAN`, `MAT_SYMMETRY_ETERNAL`
9902: @*/
9903: PetscErrorCode MatIsHermitian(Mat A, PetscReal tol, PetscBool *flg)
9904: {
9905: PetscFunctionBegin;
9907: PetscAssertPointer(flg, 3);
9908: if (A->hermitian != PETSC_BOOL3_UNKNOWN && !tol) *flg = PetscBool3ToBool(A->hermitian);
9909: else {
9910: if (A->ops->ishermitian) PetscUseTypeMethod(A, ishermitian, tol, flg);
9911: else PetscCall(MatIsHermitianTranspose(A, A, tol, flg));
9912: if (!tol) PetscCall(MatSetOption(A, MAT_HERMITIAN, *flg));
9913: }
9914: PetscFunctionReturn(PETSC_SUCCESS);
9915: }
9917: /*@
9918: MatIsSymmetricKnown - Checks if a matrix knows if it is symmetric or not and its symmetric state
9920: Not Collective
9922: Input Parameter:
9923: . A - the matrix to check
9925: Output Parameters:
9926: + set - `PETSC_TRUE` if the matrix knows its symmetry state (this tells you if the next flag is valid)
9927: - flg - the result (only valid if set is `PETSC_TRUE`)
9929: Level: advanced
9931: Notes:
9932: Does not check the matrix values directly, so this may return unknown (set = `PETSC_FALSE`). Use `MatIsSymmetric()`
9933: if you want it explicitly checked
9935: One can declare that a matrix is symmetric with `MatSetOption`(mat,`MAT_SYMMETRIC`,`PETSC_TRUE`) and if it is known to remain symmetric
9936: after changes to the matrices values one can call `MatSetOption`(mat,`MAT_SYMMETRY_ETERNAL`,`PETSC_TRUE`)
9938: .seealso: [](ch_matrices), `Mat`, `MAT_SYMMETRY_ETERNAL`, `MatTranspose()`, `MatIsTranspose()`, `MatIsHermitian()`, `MatIsStructurallySymmetric()`, `MatSetOption()`, `MatIsSymmetric()`, `MatIsHermitianKnown()`
9939: @*/
9940: PetscErrorCode MatIsSymmetricKnown(Mat A, PetscBool *set, PetscBool *flg)
9941: {
9942: PetscFunctionBegin;
9944: PetscAssertPointer(set, 2);
9945: PetscAssertPointer(flg, 3);
9946: if (A->symmetric != PETSC_BOOL3_UNKNOWN) {
9947: *set = PETSC_TRUE;
9948: *flg = PetscBool3ToBool(A->symmetric);
9949: } else *set = PETSC_FALSE;
9950: PetscFunctionReturn(PETSC_SUCCESS);
9951: }
9953: /*@
9954: MatIsSPDKnown - Checks if a matrix knows if it is symmetric positive definite or not and its symmetric positive definite state
9956: Not Collective
9958: Input Parameter:
9959: . A - the matrix to check
9961: Output Parameters:
9962: + set - `PETSC_TRUE` if the matrix knows its symmetric positive definite state (this tells you if the next flag is valid)
9963: - flg - the result (only valid if set is `PETSC_TRUE`)
9965: Level: advanced
9967: Notes:
9968: Does not check the matrix values directly, so this may return unknown (set = `PETSC_FALSE`).
9970: One can declare that a matrix is SPD with `MatSetOption`(mat,`MAT_SPD`,`PETSC_TRUE`) and if it is known to remain SPD
9971: after changes to the matrices values one can call `MatSetOption`(mat,`MAT_SPD_ETERNAL`,`PETSC_TRUE`)
9973: .seealso: [](ch_matrices), `Mat`, `MAT_SPD_ETERNAL`, `MAT_SPD`, `MatTranspose()`, `MatIsTranspose()`, `MatIsHermitian()`, `MatIsStructurallySymmetric()`, `MatSetOption()`, `MatIsSymmetric()`, `MatIsHermitianKnown()`
9974: @*/
9975: PetscErrorCode MatIsSPDKnown(Mat A, PetscBool *set, PetscBool *flg)
9976: {
9977: PetscFunctionBegin;
9979: PetscAssertPointer(set, 2);
9980: PetscAssertPointer(flg, 3);
9981: if (A->spd != PETSC_BOOL3_UNKNOWN) {
9982: *set = PETSC_TRUE;
9983: *flg = PetscBool3ToBool(A->spd);
9984: } else *set = PETSC_FALSE;
9985: PetscFunctionReturn(PETSC_SUCCESS);
9986: }
9988: /*@
9989: MatIsHermitianKnown - Checks if a matrix knows if it is Hermitian or not and its Hermitian state
9991: Not Collective
9993: Input Parameter:
9994: . A - the matrix to check
9996: Output Parameters:
9997: + set - `PETSC_TRUE` if the matrix knows its Hermitian state (this tells you if the next flag is valid)
9998: - flg - the result (only valid if set is `PETSC_TRUE`)
10000: Level: advanced
10002: Notes:
10003: Does not check the matrix values directly, so this may return unknown (set = `PETSC_FALSE`). Use `MatIsHermitian()`
10004: if you want it explicitly checked
10006: One can declare that a matrix is Hermitian with `MatSetOption`(mat,`MAT_HERMITIAN`,`PETSC_TRUE`) and if it is known to remain Hermitian
10007: after changes to the matrices values one can call `MatSetOption`(mat,`MAT_SYMMETRY_ETERNAL`,`PETSC_TRUE`)
10009: .seealso: [](ch_matrices), `Mat`, `MAT_SYMMETRY_ETERNAL`, `MAT_HERMITIAN`, `MatTranspose()`, `MatIsTranspose()`, `MatIsHermitian()`, `MatIsStructurallySymmetric()`, `MatSetOption()`, `MatIsSymmetric()`
10010: @*/
10011: PetscErrorCode MatIsHermitianKnown(Mat A, PetscBool *set, PetscBool *flg)
10012: {
10013: PetscFunctionBegin;
10015: PetscAssertPointer(set, 2);
10016: PetscAssertPointer(flg, 3);
10017: if (A->hermitian != PETSC_BOOL3_UNKNOWN) {
10018: *set = PETSC_TRUE;
10019: *flg = PetscBool3ToBool(A->hermitian);
10020: } else *set = PETSC_FALSE;
10021: PetscFunctionReturn(PETSC_SUCCESS);
10022: }
10024: /*@
10025: MatIsStructurallySymmetric - Test whether a matrix is structurally symmetric
10027: Collective
10029: Input Parameter:
10030: . A - the matrix to test
10032: Output Parameter:
10033: . flg - the result
10035: Level: intermediate
10037: Notes:
10038: If the matrix does yet know it is structurally symmetric this can be an expensive operation, also available `MatIsStructurallySymmetricKnown()`
10040: One can declare that a matrix is structurally symmetric with `MatSetOption`(mat,`MAT_STRUCTURALLY_SYMMETRIC`,`PETSC_TRUE`) and if it is known to remain structurally
10041: symmetric after changes to the matrices values one can call `MatSetOption`(mat,`MAT_STRUCTURAL_SYMMETRY_ETERNAL`,`PETSC_TRUE`). If these properties
10042: have been set then `MatIsStructurallySymmetric()` does not need to perform any computations and returns immediately with the result.
10044: .seealso: [](ch_matrices), `Mat`, `MAT_STRUCTURALLY_SYMMETRIC`, `MAT_STRUCTURAL_SYMMETRY_ETERNAL`, `MatTranspose()`, `MatIsTranspose()`, `MatIsHermitian()`, `MatIsSymmetric()`, `MatSetOption()`, `MatIsStructurallySymmetricKnown()`
10045: @*/
10046: PetscErrorCode MatIsStructurallySymmetric(Mat A, PetscBool *flg)
10047: {
10048: PetscFunctionBegin;
10050: PetscAssertPointer(flg, 2);
10051: if (A->structurally_symmetric != PETSC_BOOL3_UNKNOWN) *flg = PetscBool3ToBool(A->structurally_symmetric);
10052: else {
10053: PetscUseTypeMethod(A, isstructurallysymmetric, flg);
10054: PetscCall(MatSetOption(A, MAT_STRUCTURALLY_SYMMETRIC, *flg));
10055: }
10056: PetscFunctionReturn(PETSC_SUCCESS);
10057: }
10059: /*@
10060: MatIsStructurallySymmetricKnown - Checks if a matrix knows if it is structurally symmetric or not and its structurally symmetric state
10062: Not Collective
10064: Input Parameter:
10065: . A - the matrix to check
10067: Output Parameters:
10068: + set - `PETSC_TRUE` if the matrix knows its structurally symmetric state (this tells you if the next flag is valid)
10069: - flg - the result (only valid if set is `PETSC_TRUE`)
10071: Level: advanced
10073: Notes:
10074: One can declare that a matrix is structurally symmetric with `MatSetOption`(mat,`MAT_STRUCTURALLY_SYMMETRIC`,`PETSC_TRUE`) and if it is known to remain structurally
10075: symmetric after changes to the matrices values one can call `MatSetOption`(mat,`MAT_STRUCTURAL_SYMMETRY_ETERNAL`,`PETSC_TRUE`)
10077: Use `MatIsStructurallySymmetric()` to explicitly check if a matrix is structurally symmetric (this is an expensive operation)
10079: .seealso: [](ch_matrices), `Mat`, `MAT_STRUCTURALLY_SYMMETRIC`, `MatTranspose()`, `MatIsTranspose()`, `MatIsHermitian()`, `MatIsStructurallySymmetric()`, `MatSetOption()`, `MatIsSymmetric()`, `MatIsHermitianKnown()`
10080: @*/
10081: PetscErrorCode MatIsStructurallySymmetricKnown(Mat A, PetscBool *set, PetscBool *flg)
10082: {
10083: PetscFunctionBegin;
10085: PetscAssertPointer(set, 2);
10086: PetscAssertPointer(flg, 3);
10087: if (A->structurally_symmetric != PETSC_BOOL3_UNKNOWN) {
10088: *set = PETSC_TRUE;
10089: *flg = PetscBool3ToBool(A->structurally_symmetric);
10090: } else *set = PETSC_FALSE;
10091: PetscFunctionReturn(PETSC_SUCCESS);
10092: }
10094: /*@
10095: MatStashGetInfo - Gets how many values are currently in the matrix stash, i.e. need
10096: to be communicated to other processes during the `MatAssemblyBegin()`/`MatAssemblyEnd()` process
10098: Not Collective
10100: Input Parameter:
10101: . mat - the matrix
10103: Output Parameters:
10104: + nstash - the size of the stash
10105: . reallocs - the number of additional mallocs incurred.
10106: . bnstash - the size of the block stash
10107: - breallocs - the number of additional mallocs incurred.in the block stash
10109: Level: advanced
10111: .seealso: [](ch_matrices), `MatAssemblyBegin()`, `MatAssemblyEnd()`, `Mat`, `MatStashSetInitialSize()`
10112: @*/
10113: PetscErrorCode MatStashGetInfo(Mat mat, PetscInt *nstash, PetscInt *reallocs, PetscInt *bnstash, PetscInt *breallocs)
10114: {
10115: PetscFunctionBegin;
10116: PetscCall(MatStashGetInfo_Private(&mat->stash, nstash, reallocs));
10117: PetscCall(MatStashGetInfo_Private(&mat->bstash, bnstash, breallocs));
10118: PetscFunctionReturn(PETSC_SUCCESS);
10119: }
10121: /*@
10122: MatCreateVecs - Get vector(s) compatible with the matrix, i.e. with the same
10123: parallel layout, `PetscLayout` for rows and columns
10125: Collective
10127: Input Parameter:
10128: . mat - the matrix
10130: Output Parameters:
10131: + right - (optional) vector that the matrix can be multiplied against
10132: - left - (optional) vector that the matrix vector product can be stored in
10134: Options Database Key:
10135: . -mat_vec_type type - set the `VecType` of the created vectors during `MatSetFromOptions()`
10137: Level: advanced
10139: Notes:
10140: The blocksize of the returned vectors is determined by the row and column block sizes set with `MatSetBlockSizes()` or the single blocksize (same for both) set by `MatSetBlockSize()`.
10142: The `VecType` of the created vectors is determined by the `MatType` of `mat`. This can be overridden by using `MatSetVecType()` or the option `-mat_vec_type`.
10144: These are new vectors which are not owned by the `mat`, they should be destroyed with `VecDestroy()` when no longer needed.
10146: PETSc `Vec` always have all zero entries when created with `MatCreateVecs()` until routines such as `VecSet()` or `VecSetValues()`
10147: are used to change the values. There is no reason to call `VecZeroEntries()` after creation.
10149: .seealso: [](ch_matrices), `Mat`, `Vec`, `VecCreate()`, `VecDestroy()`, `DMCreateGlobalVector()`, `MatSetVecType()`
10150: @*/
10151: PetscErrorCode MatCreateVecs(Mat mat, Vec *right, Vec *left)
10152: {
10153: PetscFunctionBegin;
10156: if (mat->ops->getvecs) {
10157: PetscUseTypeMethod(mat, getvecs, right, left);
10158: } else {
10159: if (right) {
10160: PetscCheck(mat->cmap->n >= 0, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "PetscLayout for columns not yet setup");
10161: PetscCall(VecCreateWithLayout_Private(mat->cmap, right));
10162: PetscCall(VecSetType(*right, mat->defaultvectype));
10163: #if PetscDefined(HAVE_VIENNACL) || PetscDefined(HAVE_CUDA) || PetscDefined(HAVE_HIP)
10164: if (mat->boundtocpu && mat->bindingpropagates) {
10165: PetscCall(VecSetBindingPropagates(*right, PETSC_TRUE));
10166: PetscCall(VecBindToCPU(*right, PETSC_TRUE));
10167: }
10168: #endif
10169: }
10170: if (left) {
10171: PetscCheck(mat->rmap->n >= 0, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "PetscLayout for rows not yet setup");
10172: PetscCall(VecCreateWithLayout_Private(mat->rmap, left));
10173: PetscCall(VecSetType(*left, mat->defaultvectype));
10174: #if PetscDefined(HAVE_VIENNACL) || PetscDefined(HAVE_CUDA) || PetscDefined(HAVE_HIP)
10175: if (mat->boundtocpu && mat->bindingpropagates) {
10176: PetscCall(VecSetBindingPropagates(*left, PETSC_TRUE));
10177: PetscCall(VecBindToCPU(*left, PETSC_TRUE));
10178: }
10179: #endif
10180: }
10181: }
10182: PetscFunctionReturn(PETSC_SUCCESS);
10183: }
10185: /*@
10186: MatFactorInfoInitialize - Initializes a `MatFactorInfo` data structure
10187: with default values.
10189: Not Collective
10191: Input Parameter:
10192: . info - the `MatFactorInfo` data structure
10194: Level: developer
10196: Notes:
10197: The solvers are generally used through the `KSP` and `PC` objects, for example
10198: `PCLU`, `PCILU`, `PCCHOLESKY`, `PCICC`
10200: Once the data structure is initialized one may change certain entries as desired for the particular factorization to be performed
10202: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorInfo`
10203: @*/
10204: PetscErrorCode MatFactorInfoInitialize(MatFactorInfo *info)
10205: {
10206: PetscFunctionBegin;
10207: PetscCall(PetscMemzero(info, sizeof(MatFactorInfo)));
10208: PetscFunctionReturn(PETSC_SUCCESS);
10209: }
10211: /*@
10212: MatFactorSetSchurIS - Set indices corresponding to the Schur complement you wish to have computed
10214: Collective
10216: Input Parameters:
10217: + mat - the factored matrix
10218: - is - the index set defining the Schur indices (0-based)
10220: Level: advanced
10222: Notes:
10223: Call `MatFactorSolveSchurComplement()` or `MatFactorSolveSchurComplementTranspose()` after this call to solve a Schur complement system.
10225: You can call `MatFactorGetSchurComplement()` or `MatFactorCreateSchurComplement()` after this call.
10227: This functionality is only supported for `MATSOLVERMUMPS` and `MATSOLVERMKL_PARDISO`
10229: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorGetSchurComplement()`, `MatFactorRestoreSchurComplement()`, `MatFactorCreateSchurComplement()`, `MatFactorSolveSchurComplement()`,
10230: `MatFactorSolveSchurComplementTranspose()`, `MATSOLVERMUMPS`, `MATSOLVERMKL_PARDISO`
10231: @*/
10232: PetscErrorCode MatFactorSetSchurIS(Mat mat, IS is)
10233: {
10234: PetscErrorCode (*f)(Mat, IS);
10236: PetscFunctionBegin;
10241: PetscCheckSameComm(mat, 1, is, 2);
10242: PetscCheck(mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Only for factored matrix");
10243: PetscCall(PetscObjectQueryFunction((PetscObject)mat, "MatFactorSetSchurIS_C", &f));
10244: PetscCheck(f, PetscObjectComm((PetscObject)mat), PETSC_ERR_SUP, "The selected MatSolverType does not support Schur complement computation. You should use MATSOLVERMUMPS or MATSOLVERMKL_PARDISO");
10245: PetscCall((*f)(mat, is));
10246: PetscCheck(mat->schur, PetscObjectComm((PetscObject)mat), PETSC_ERR_PLIB, "Schur complement has not been created");
10247: PetscFunctionReturn(PETSC_SUCCESS);
10248: }
10250: /*@
10251: MatFactorCreateSchurComplement - Create a Schur complement matrix object using Schur data computed during the factorization step
10253: Logically Collective
10255: Input Parameters:
10256: + F - the factored matrix obtained by calling `MatGetFactor()`
10257: . S - location where to return the Schur complement, can be `NULL`
10258: - status - the status of the Schur complement matrix, can be `NULL`
10260: Level: advanced
10262: Notes:
10263: You must call `MatFactorSetSchurIS()` before calling this routine.
10265: This functionality is only supported for `MATSOLVERMUMPS` and `MATSOLVERMKL_PARDISO`
10267: The routine provides a copy of the Schur matrix stored within the solver data structures.
10268: The caller must destroy the object when it is no longer needed.
10269: If `MatFactorInvertSchurComplement()` has been called, the routine gets back the inverse.
10271: Use `MatFactorGetSchurComplement()` to get access to the Schur complement matrix inside the factored matrix instead of making a copy of it (which this function does)
10273: See `MatCreateSchurComplement()` or `MatGetSchurComplement()` for ways to create virtual or approximate Schur complements.
10275: Developer Note:
10276: The reason this routine exists is because the representation of the Schur complement within the factor matrix may be different than a standard PETSc
10277: matrix representation and we normally do not want to use the time or memory to make a copy as a regular PETSc matrix.
10279: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorSetSchurIS()`, `MatFactorGetSchurComplement()`, `MatFactorSchurStatus`, `MATSOLVERMUMPS`, `MATSOLVERMKL_PARDISO`
10280: @*/
10281: PetscErrorCode MatFactorCreateSchurComplement(Mat F, Mat *S, MatFactorSchurStatus *status)
10282: {
10283: PetscFunctionBegin;
10285: if (S) PetscAssertPointer(S, 2);
10286: if (status) PetscAssertPointer(status, 3);
10287: if (S) {
10288: PetscErrorCode (*f)(Mat, Mat *);
10290: PetscCall(PetscObjectQueryFunction((PetscObject)F, "MatFactorCreateSchurComplement_C", &f));
10291: if (f) PetscCall((*f)(F, S));
10292: else PetscCall(MatDuplicate(F->schur, MAT_COPY_VALUES, S));
10293: }
10294: if (status) *status = F->schur_status;
10295: PetscFunctionReturn(PETSC_SUCCESS);
10296: }
10298: /*@
10299: MatFactorGetSchurComplement - Gets access to a Schur complement matrix using the current Schur data within a factored matrix
10301: Logically Collective
10303: Input Parameters:
10304: + F - the factored matrix obtained by calling `MatGetFactor()`
10305: . S - location where to return the Schur complement, can be `NULL`
10306: - status - the status of the Schur complement matrix, can be `NULL`
10308: Level: advanced
10310: Notes:
10311: You must call `MatFactorSetSchurIS()` before calling this routine.
10313: Schur complement mode is currently implemented for sequential matrices with factor type of `MATSOLVERMUMPS`
10315: The routine returns a the Schur Complement stored within the data structures of the solver.
10317: If `MatFactorInvertSchurComplement()` has previously been called, the returned matrix is actually the inverse of the Schur complement.
10319: The returned matrix should not be destroyed; the caller should call `MatFactorRestoreSchurComplement()` when the object is no longer needed.
10321: Use `MatFactorCreateSchurComplement()` to create a copy of the Schur complement matrix that is within a factored matrix
10323: See `MatCreateSchurComplement()` or `MatGetSchurComplement()` for ways to create virtual or approximate Schur complements.
10325: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorSetSchurIS()`, `MatFactorRestoreSchurComplement()`, `MatFactorCreateSchurComplement()`, `MatFactorSchurStatus`
10326: @*/
10327: PetscErrorCode MatFactorGetSchurComplement(Mat F, Mat *S, MatFactorSchurStatus *status)
10328: {
10329: PetscFunctionBegin;
10331: if (S) {
10332: PetscAssertPointer(S, 2);
10333: *S = F->schur;
10334: }
10335: if (status) {
10336: PetscAssertPointer(status, 3);
10337: *status = F->schur_status;
10338: }
10339: PetscFunctionReturn(PETSC_SUCCESS);
10340: }
10342: static PetscErrorCode MatFactorUpdateSchurStatus_Private(Mat F)
10343: {
10344: Mat S = F->schur;
10346: PetscFunctionBegin;
10347: switch (F->schur_status) {
10348: case MAT_FACTOR_SCHUR_UNFACTORED: // fall-through
10349: case MAT_FACTOR_SCHUR_INVERTED:
10350: if (S) {
10351: S->ops->solve = NULL;
10352: S->ops->matsolve = NULL;
10353: S->ops->solvetranspose = NULL;
10354: S->ops->matsolvetranspose = NULL;
10355: S->ops->solveadd = NULL;
10356: S->ops->solvetransposeadd = NULL;
10357: S->factortype = MAT_FACTOR_NONE;
10358: PetscCall(PetscFree(S->solvertype));
10359: }
10360: case MAT_FACTOR_SCHUR_FACTORED: // fall-through
10361: break;
10362: default:
10363: SETERRQ(PetscObjectComm((PetscObject)F), PETSC_ERR_SUP, "Unhandled MatFactorSchurStatus %d", F->schur_status);
10364: }
10365: PetscFunctionReturn(PETSC_SUCCESS);
10366: }
10368: /*@
10369: MatFactorRestoreSchurComplement - Restore the Schur complement matrix object obtained from a call to `MatFactorGetSchurComplement()`
10371: Logically Collective
10373: Input Parameters:
10374: + F - the factored matrix obtained by calling `MatGetFactor()`
10375: . S - location where the Schur complement is stored
10376: - status - the status of the Schur complement matrix (see `MatFactorSchurStatus`)
10378: Level: advanced
10380: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorSetSchurIS()`, `MatFactorCreateSchurComplement()`, `MatFactorSchurStatus`
10381: @*/
10382: PetscErrorCode MatFactorRestoreSchurComplement(Mat F, Mat *S, MatFactorSchurStatus status)
10383: {
10384: PetscFunctionBegin;
10386: if (S) {
10388: *S = NULL;
10389: }
10390: F->schur_status = status;
10391: PetscCall(MatFactorUpdateSchurStatus_Private(F));
10392: PetscFunctionReturn(PETSC_SUCCESS);
10393: }
10395: /*@
10396: MatFactorSolveSchurComplementTranspose - Solve the transpose of the Schur complement system computed during the factorization step
10398: Logically Collective
10400: Input Parameters:
10401: + F - the factored matrix obtained by calling `MatGetFactor()`
10402: . rhs - location where the right-hand side of the Schur complement system is stored
10403: - sol - location where the solution of the Schur complement system has to be returned
10405: Level: advanced
10407: Notes:
10408: The sizes of the vectors should match the size of the Schur complement
10410: Must be called after `MatFactorSetSchurIS()`
10412: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorSetSchurIS()`, `MatFactorSolveSchurComplement()`
10413: @*/
10414: PetscErrorCode MatFactorSolveSchurComplementTranspose(Mat F, Vec rhs, Vec sol)
10415: {
10416: PetscFunctionBegin;
10423: PetscCheckSameComm(F, 1, rhs, 2);
10424: PetscCheckSameComm(F, 1, sol, 3);
10425: PetscCall(MatFactorFactorizeSchurComplement(F));
10426: switch (F->schur_status) {
10427: case MAT_FACTOR_SCHUR_FACTORED:
10428: PetscCall(MatSolveTranspose(F->schur, rhs, sol));
10429: break;
10430: case MAT_FACTOR_SCHUR_INVERTED:
10431: PetscCall(MatMultTranspose(F->schur, rhs, sol));
10432: break;
10433: default:
10434: SETERRQ(PetscObjectComm((PetscObject)F), PETSC_ERR_SUP, "Unhandled MatFactorSchurStatus %d", F->schur_status);
10435: }
10436: PetscFunctionReturn(PETSC_SUCCESS);
10437: }
10439: /*@
10440: MatFactorSolveSchurComplement - Solve the Schur complement system computed during the factorization step
10442: Logically Collective
10444: Input Parameters:
10445: + F - the factored matrix obtained by calling `MatGetFactor()`
10446: . rhs - location where the right-hand side of the Schur complement system is stored
10447: - sol - location where the solution of the Schur complement system has to be returned
10449: Level: advanced
10451: Notes:
10452: The sizes of the vectors should match the size of the Schur complement
10454: Must be called after `MatFactorSetSchurIS()`
10456: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorSetSchurIS()`, `MatFactorSolveSchurComplementTranspose()`
10457: @*/
10458: PetscErrorCode MatFactorSolveSchurComplement(Mat F, Vec rhs, Vec sol)
10459: {
10460: PetscFunctionBegin;
10467: PetscCheckSameComm(F, 1, rhs, 2);
10468: PetscCheckSameComm(F, 1, sol, 3);
10469: PetscCall(MatFactorFactorizeSchurComplement(F));
10470: switch (F->schur_status) {
10471: case MAT_FACTOR_SCHUR_FACTORED:
10472: PetscCall(MatSolve(F->schur, rhs, sol));
10473: break;
10474: case MAT_FACTOR_SCHUR_INVERTED:
10475: PetscCall(MatMult(F->schur, rhs, sol));
10476: break;
10477: default:
10478: SETERRQ(PetscObjectComm((PetscObject)F), PETSC_ERR_SUP, "Unhandled MatFactorSchurStatus %d", F->schur_status);
10479: }
10480: PetscFunctionReturn(PETSC_SUCCESS);
10481: }
10483: PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode MatSeqDenseInvertFactors_Private(Mat);
10484: #if PetscDefined(HAVE_CUDA)
10485: PETSC_SINGLE_LIBRARY_INTERN PetscErrorCode MatSeqDenseCUDAInvertFactors_Internal(Mat);
10486: #endif
10488: /* Schur status updated in the interface */
10489: static PetscErrorCode MatFactorInvertSchurComplement_Private(Mat F)
10490: {
10491: Mat S = F->schur;
10493: PetscFunctionBegin;
10494: if (S) {
10495: PetscMPIInt size;
10496: PetscBool isdense, isdensecuda;
10498: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)S), &size));
10499: PetscCheck(size <= 1, PetscObjectComm((PetscObject)S), PETSC_ERR_SUP, "Not yet implemented");
10500: PetscCall(PetscObjectTypeCompare((PetscObject)S, MATSEQDENSE, &isdense));
10501: PetscCall(PetscObjectTypeCompare((PetscObject)S, MATSEQDENSECUDA, &isdensecuda));
10502: PetscCheck(isdense || isdensecuda, PetscObjectComm((PetscObject)S), PETSC_ERR_SUP, "Not implemented for type %s", ((PetscObject)S)->type_name);
10503: PetscCall(PetscLogEventBegin(MAT_FactorInvS, F, 0, 0, 0));
10504: if (isdense) {
10505: PetscCall(MatSeqDenseInvertFactors_Private(S));
10506: } else if (isdensecuda) {
10507: #if PetscDefined(HAVE_CUDA)
10508: PetscCall(MatSeqDenseCUDAInvertFactors_Internal(S));
10509: #endif
10510: }
10511: // HIP??????????????
10512: PetscCall(PetscLogEventEnd(MAT_FactorInvS, F, 0, 0, 0));
10513: }
10514: PetscFunctionReturn(PETSC_SUCCESS);
10515: }
10517: /*@
10518: MatFactorInvertSchurComplement - Invert the Schur complement matrix computed during the factorization step
10520: Logically Collective
10522: Input Parameter:
10523: . F - the factored matrix obtained by calling `MatGetFactor()`
10525: Level: advanced
10527: Notes:
10528: Must be called after `MatFactorSetSchurIS()`.
10530: Call `MatFactorGetSchurComplement()` or `MatFactorCreateSchurComplement()` AFTER this call to actually compute the inverse and get access to it.
10532: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorSetSchurIS()`, `MatFactorGetSchurComplement()`, `MatFactorCreateSchurComplement()`
10533: @*/
10534: PetscErrorCode MatFactorInvertSchurComplement(Mat F)
10535: {
10536: PetscFunctionBegin;
10539: if (F->schur_status == MAT_FACTOR_SCHUR_INVERTED) PetscFunctionReturn(PETSC_SUCCESS);
10540: PetscCall(MatFactorFactorizeSchurComplement(F));
10541: PetscCall(MatFactorInvertSchurComplement_Private(F));
10542: F->schur_status = MAT_FACTOR_SCHUR_INVERTED;
10543: PetscFunctionReturn(PETSC_SUCCESS);
10544: }
10546: /*@
10547: MatFactorFactorizeSchurComplement - Factorize the Schur complement matrix computed during the factorization step
10549: Logically Collective
10551: Input Parameter:
10552: . F - the factored matrix obtained by calling `MatGetFactor()`
10554: Level: advanced
10556: Note:
10557: Must be called after `MatFactorSetSchurIS()`
10559: .seealso: [](ch_matrices), `Mat`, `MatGetFactor()`, `MatFactorSetSchurIS()`, `MatFactorInvertSchurComplement()`
10560: @*/
10561: PetscErrorCode MatFactorFactorizeSchurComplement(Mat F)
10562: {
10563: MatFactorInfo info;
10565: PetscFunctionBegin;
10568: if (F->schur_status == MAT_FACTOR_SCHUR_INVERTED || F->schur_status == MAT_FACTOR_SCHUR_FACTORED) PetscFunctionReturn(PETSC_SUCCESS);
10569: PetscCall(PetscLogEventBegin(MAT_FactorFactS, F, 0, 0, 0));
10570: PetscCall(PetscMemzero(&info, sizeof(MatFactorInfo)));
10571: if (F->factortype == MAT_FACTOR_CHOLESKY) { /* LDL^t regarded as Cholesky */
10572: PetscCall(MatCholeskyFactor(F->schur, NULL, &info));
10573: } else {
10574: PetscCall(MatLUFactor(F->schur, NULL, NULL, &info));
10575: }
10576: PetscCall(PetscLogEventEnd(MAT_FactorFactS, F, 0, 0, 0));
10577: F->schur_status = MAT_FACTOR_SCHUR_FACTORED;
10578: PetscFunctionReturn(PETSC_SUCCESS);
10579: }
10581: /*@
10582: MatPtAP - Creates the matrix product $C = P^T * A * P$
10584: Neighbor-wise Collective
10586: Input Parameters:
10587: + A - the matrix
10588: . P - the projection matrix
10589: . scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
10590: - fill - expected fill as ratio of nnz(C)/(nnz(A) + nnz(P)), use `PETSC_DETERMINE` or `PETSC_CURRENT` if you do not have a good estimate
10591: if the result is a dense matrix this is irrelevant
10593: Output Parameter:
10594: . C - the product matrix
10596: Level: intermediate
10598: Notes:
10599: `C` will be created and must be destroyed by the user with `MatDestroy()`.
10601: This is a convenience routine that wraps the use of the `MatProductCreate()` with a `MatProductType` of `MATPRODUCT_PtAP`
10602: functionality into a single function call. For more involved matrix-matrix operations see `MatProductCreate()`.
10604: The deprecated `PETSC_DEFAULT` in `fill` also means use the current value
10606: Developer Note:
10607: For matrix types without special implementation the function fallbacks to `MatMatMult()` followed by `MatTransposeMatMult()`.
10609: .seealso: [](ch_matrices), `Mat`, `MatProductCreate()`, `MatMatMult()`, `MatRARt()`
10610: @*/
10611: PetscErrorCode MatPtAP(Mat A, Mat P, MatReuse scall, PetscReal fill, Mat *C)
10612: {
10613: PetscFunctionBegin;
10614: if (scall == MAT_REUSE_MATRIX) MatCheckProduct(*C, 5);
10615: PetscCheck(scall != MAT_INPLACE_MATRIX, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Inplace product not supported");
10617: if (scall == MAT_INITIAL_MATRIX) {
10618: PetscCall(MatProductCreate(A, P, NULL, C));
10619: PetscCall(MatProductSetType(*C, MATPRODUCT_PtAP));
10620: PetscCall(MatProductSetAlgorithm(*C, "default"));
10621: PetscCall(MatProductSetFill(*C, fill));
10623: (*C)->product->api_user = PETSC_TRUE;
10624: PetscCall(MatProductSetFromOptions(*C));
10625: PetscCheck((*C)->ops->productsymbolic, PetscObjectComm((PetscObject)*C), PETSC_ERR_SUP, "MatProduct %s not supported for A %s and P %s", MatProductTypes[MATPRODUCT_PtAP], ((PetscObject)A)->type_name, ((PetscObject)P)->type_name);
10626: PetscCall(MatProductSymbolic(*C));
10627: } else { /* scall == MAT_REUSE_MATRIX */
10628: PetscCall(MatProductReplaceMats(A, P, NULL, *C));
10629: }
10631: PetscCall(MatProductNumeric(*C));
10632: if (A->symmetric == PETSC_BOOL3_TRUE) {
10633: PetscCall(MatSetOption(*C, MAT_SYMMETRIC, PETSC_TRUE));
10634: (*C)->spd = A->spd;
10635: }
10636: PetscFunctionReturn(PETSC_SUCCESS);
10637: }
10639: /*@
10640: MatRARt - Creates the matrix product $C = R * A * R^T$
10642: Neighbor-wise Collective
10644: Input Parameters:
10645: + A - the matrix
10646: . R - the projection matrix
10647: . scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
10648: - fill - expected fill as ratio of nnz(C)/nnz(A), use `PETSC_DETERMINE` or `PETSC_CURRENT` if you do not have a good estimate
10649: if the result is a dense matrix this is irrelevant
10651: Output Parameter:
10652: . C - the product matrix
10654: Level: intermediate
10656: Notes:
10657: `C` will be created and must be destroyed by the user with `MatDestroy()`.
10659: This is a convenience routine that wraps the use of the `MatProductCreate()` with a `MatProductType` of `MATPRODUCT_RARt`
10660: functionality into a single function call. For more involved matrix-matrix operations see `MatProductCreate()`.
10662: This routine is currently only implemented for pairs of `MATAIJ` matrices and classes
10663: which inherit from `MATAIJ`. Due to PETSc sparse matrix block row distribution among processes,
10664: the parallel `MatRARt()` is implemented computing the explicit transpose of `R`, which can be very expensive.
10665: We recommend using `MatPtAP()` when possible.
10667: The deprecated `PETSC_DEFAULT` in `fill` also means use the current value
10669: .seealso: [](ch_matrices), `Mat`, `MatProductCreate()`, `MatMatMult()`, `MatPtAP()`
10670: @*/
10671: PetscErrorCode MatRARt(Mat A, Mat R, MatReuse scall, PetscReal fill, Mat *C)
10672: {
10673: PetscFunctionBegin;
10674: if (scall == MAT_REUSE_MATRIX) MatCheckProduct(*C, 5);
10675: PetscCheck(scall != MAT_INPLACE_MATRIX, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Inplace product not supported");
10677: if (scall == MAT_INITIAL_MATRIX) {
10678: PetscCall(MatProductCreate(A, R, NULL, C));
10679: PetscCall(MatProductSetType(*C, MATPRODUCT_RARt));
10680: PetscCall(MatProductSetAlgorithm(*C, "default"));
10681: PetscCall(MatProductSetFill(*C, fill));
10683: (*C)->product->api_user = PETSC_TRUE;
10684: PetscCall(MatProductSetFromOptions(*C));
10685: PetscCheck((*C)->ops->productsymbolic, PetscObjectComm((PetscObject)*C), PETSC_ERR_SUP, "MatProduct %s not supported for A %s and R %s", MatProductTypes[MATPRODUCT_RARt], ((PetscObject)A)->type_name, ((PetscObject)R)->type_name);
10686: PetscCall(MatProductSymbolic(*C));
10687: } else { /* scall == MAT_REUSE_MATRIX */
10688: PetscCall(MatProductReplaceMats(A, R, NULL, *C));
10689: }
10691: PetscCall(MatProductNumeric(*C));
10692: if (A->symmetric == PETSC_BOOL3_TRUE) PetscCall(MatSetOption(*C, MAT_SYMMETRIC, PETSC_TRUE));
10693: PetscFunctionReturn(PETSC_SUCCESS);
10694: }
10696: /* MatProductComputeWithMat_Private - Computes D = A * B or D = A^T * B into an existing matrix D, keeping the product on a dense D
10698: Collective
10700: Input Parameters:
10701: + A - the left matrix
10702: . B - the right matrix
10703: . ptype - MATPRODUCT_AB or MATPRODUCT_AtB
10704: - D - the result, of the size of the product
10706: Level: developer
10708: Notes:
10709: For a dense D, this routine keeps the product on D between calls, so only the first call runs the symbolic phase. The product is created
10710: again only when ptype, A, or the nonzero pattern of A changes; a different B of the same size and type reuses it. The kept product holds
10711: references to A and B until MatProductClear() is called on D or D is destroyed. In contrast, MatMatMult() with MAT_REUSE_MATRIX creates the
10712: product and destroys it again on every call into a dense D that does not already have a product.
10714: For any other D, which must already have a product, this calls MatMatMult() or MatTransposeMatMult() with MAT_REUSE_MATRIX.
10716: Developer Note:
10717: The nonzero state of A is compared without a reduction, since the parallel AIJ and BAIJ types make it the same on all processes during
10718: assembly.
10720: .seealso: `MatMatMult()`, `MatTransposeMatMult()`, `MatProductReplaceMats()`, `PCMGMatResidualDefault()`
10721: */
10722: PetscErrorCode MatProductComputeWithMat_Private(Mat A, Mat B, MatProductType ptype, Mat D)
10723: {
10724: Mat_Product *product = D->product;
10725: PetscObjectState state;
10726: PetscBool flg;
10728: PetscFunctionBegin;
10729: PetscCall(PetscObjectBaseTypeCompareAny((PetscObject)D, &flg, MATSEQDENSE, MATMPIDENSE, ""));
10730: if (!flg) { /* only a dense result can be given without a product, MatMatMult() and MatTransposeMatMult() reuse the product of any other */
10731: if (ptype == MATPRODUCT_AB) PetscCall(MatMatMult(A, B, MAT_REUSE_MATRIX, PETSC_CURRENT, &D));
10732: else PetscCall(MatTransposeMatMult(A, B, MAT_REUSE_MATRIX, PETSC_CURRENT, &D));
10733: PetscFunctionReturn(PETSC_SUCCESS);
10734: }
10735: PetscCall(MatGetNonzeroState(A, &state));
10736: if (product == NULL || product->type != ptype || product->A != A || product->Anonzerostate != state) {
10737: PetscCall(PetscInfo(D, "Setting up the %s product into the supplied dense matrix\n", MatProductTypes[ptype]));
10738: PetscCall(MatProductCreateWithMat(A, B, NULL, D));
10739: D->product->api_user = PETSC_TRUE;
10740: PetscCall(MatProductSetType(D, ptype));
10741: PetscCall(MatProductSetFromOptions(D));
10742: PetscCall(MatProductSymbolic(D));
10743: } else {
10744: PetscCall(PetscInfo(D, "Reusing the %s product attached to the supplied dense matrix\n", MatProductTypes[ptype]));
10745: if (product->B != B) PetscCall(MatProductReplaceMats(NULL, B, NULL, D));
10746: }
10747: PetscCall(MatProductNumeric(D));
10748: PetscFunctionReturn(PETSC_SUCCESS);
10749: }
10751: static PetscErrorCode MatProduct_Private(Mat A, Mat B, MatReuse scall, PetscReal fill, MatProductType ptype, Mat *C)
10752: {
10753: PetscBool flg = PETSC_TRUE;
10755: PetscFunctionBegin;
10756: PetscCheck(scall != MAT_INPLACE_MATRIX, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "MAT_INPLACE_MATRIX product not supported");
10757: if (scall == MAT_INITIAL_MATRIX) {
10758: PetscCall(PetscInfo(A, "Calling MatProduct API with MAT_INITIAL_MATRIX and product type %s\n", MatProductTypes[ptype]));
10759: PetscCall(MatProductCreate(A, B, NULL, C));
10760: PetscCall(MatProductSetAlgorithm(*C, MATPRODUCTALGORITHMDEFAULT));
10761: PetscCall(MatProductSetFill(*C, fill));
10762: } else { /* scall == MAT_REUSE_MATRIX */
10763: Mat_Product *product = (*C)->product;
10765: PetscCall(PetscObjectBaseTypeCompareAny((PetscObject)*C, &flg, MATSEQDENSE, MATMPIDENSE, ""));
10766: if (flg && product && product->type != ptype) {
10767: PetscCall(MatProductClear(*C));
10768: product = NULL;
10769: }
10770: PetscCall(PetscInfo(A, "Calling MatProduct API with MAT_REUSE_MATRIX %s product present and product type %s\n", product ? "with" : "without", MatProductTypes[ptype]));
10771: if (!product) { /* user provide the dense matrix *C without calling MatProductCreate() or reusing it from previous calls */
10772: PetscCheck(flg, PetscObjectComm((PetscObject)*C), PETSC_ERR_SUP, "Call MatProductCreate() first");
10773: PetscCall(MatProductCreate_Private(A, B, NULL, *C));
10774: product = (*C)->product;
10775: product->fill = fill;
10776: product->clear = PETSC_TRUE;
10777: } else { /* user may change input matrices A or B when MAT_REUSE_MATRIX */
10778: flg = PETSC_FALSE;
10779: PetscCall(MatProductReplaceMats(A, B, NULL, *C));
10780: }
10781: }
10782: if (flg) {
10783: (*C)->product->api_user = PETSC_TRUE;
10784: PetscCall(MatProductSetType(*C, ptype));
10785: PetscCall(MatProductSetFromOptions(*C));
10786: PetscCall(MatProductSymbolic(*C));
10787: }
10788: PetscCall(MatProductNumeric(*C));
10789: PetscFunctionReturn(PETSC_SUCCESS);
10790: }
10792: /*@
10793: MatMatMult - Performs matrix-matrix multiplication $ C=A*B $.
10795: Neighbor-wise Collective
10797: Input Parameters:
10798: + A - the left matrix
10799: . B - the right matrix
10800: . scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
10801: - fill - expected fill as ratio of nnz(C)/(nnz(A) + nnz(B)), use `PETSC_DETERMINE` or `PETSC_CURRENT` if you do not have a good estimate
10802: if the result is a dense matrix this is irrelevant
10804: Output Parameter:
10805: . C - the product matrix
10807: Notes:
10808: Unless scall is `MAT_REUSE_MATRIX` C will be created.
10810: `MAT_REUSE_MATRIX` can only be used if the matrices A and B have the same nonzero pattern as in the previous call and C was obtained from a previous
10811: call to this function with `MAT_INITIAL_MATRIX`.
10813: To determine the correct fill value, run with `-info` and search for the string "Fill ratio" to see the value actually needed.
10815: In the special case where matrix `B` (and hence `C`) are dense you can create the correctly sized matrix `C` yourself and then call this routine with `MAT_REUSE_MATRIX`,
10816: rather than first having `MatMatMult()` create it for you. You can NEVER do this if the matrix `C` is sparse.
10818: The deprecated `PETSC_DEFAULT` in `fill` also means use the current value
10820: This is a convenience routine that wraps the use of the `MatProductCreate()` with a `MatProductType` of `MATPRODUCT_AB`
10821: functionality into a single function call. For more involved matrix-matrix operations see `MatProductCreate()`.
10823: Example of Usage:
10824: .vb
10825: MatProductCreate(A,B,NULL,&C);
10826: MatProductSetType(C,MATPRODUCT_AB);
10827: MatProductSymbolic(C);
10828: MatProductNumeric(C); // compute C=A * B
10829: MatProductReplaceMats(A1,B1,NULL,C); // compute C=A1 * B1
10830: MatProductNumeric(C);
10831: MatProductReplaceMats(A2,NULL,NULL,C); // compute C=A2 * B1
10832: MatProductNumeric(C);
10833: .ve
10835: Level: intermediate
10837: .seealso: [](ch_matrices), `Mat`, `MatProductType`, `MATPRODUCT_AB`, `MatTransposeMatMult()`, `MatMatTransposeMult()`, `MatPtAP()`, `MatProductCreate()`, `MatProductSymbolic()`, `MatProductReplaceMats()`, `MatProductNumeric()`
10838: @*/
10839: PetscErrorCode MatMatMult(Mat A, Mat B, MatReuse scall, PetscReal fill, Mat *C)
10840: {
10841: PetscFunctionBegin;
10842: PetscCall(MatProduct_Private(A, B, scall, fill, MATPRODUCT_AB, C));
10843: PetscFunctionReturn(PETSC_SUCCESS);
10844: }
10846: /*@
10847: MatMatTransposeMult - Performs matrix-matrix multiplication $C = A*B^T$.
10849: Neighbor-wise Collective
10851: Input Parameters:
10852: + A - the left matrix
10853: . B - the right matrix
10854: . scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
10855: - fill - expected fill as ratio of nnz(C)/(nnz(A) + nnz(B)), use `PETSC_DETERMINE` or `PETSC_CURRENT` if not known
10857: Output Parameter:
10858: . C - the product matrix
10860: Options Database Key:
10861: . -matmattransmult_mpidense_mpidense_via {allgatherv,cyclic} - Choose between algorithms for `MATMPIDENSE` matrices: the
10862: first redundantly copies the transposed `B` matrix on each process and requires O(log P) communication complexity;
10863: the second never stores more than one portion of the `B` matrix at a time but requires O(P) communication complexity.
10865: Level: intermediate
10867: Notes:
10868: C will be created if `MAT_INITIAL_MATRIX` and must be destroyed by the user with `MatDestroy()`.
10870: `MAT_REUSE_MATRIX` can only be used if the matrices A and B have the same nonzero pattern as in the previous call
10872: To determine the correct fill value, run with -info and search for the string "Fill ratio" to see the value
10873: actually needed.
10875: This routine is currently only implemented for pairs of `MATSEQAIJ` matrices, for the `MATSEQDENSE` class,
10876: and for pairs of `MATMPIDENSE` matrices.
10878: This is a convenience routine that wraps the use of the `MatProductCreate()` with a `MatProductType` of `MATPRODUCT_ABt`
10879: functionality into a single function call. For more involved matrix-matrix operations see `MatProductCreate()`.
10881: The deprecated `PETSC_DEFAULT` in `fill` also means use the current value
10883: .seealso: [](ch_matrices), `Mat`, `MatProductCreate()`, `MATPRODUCT_ABt`, `MatMatMult()`, `MatTransposeMatMult()`, `MatPtAP()`, `MatProductAlgorithm`, `MatProductType`
10884: @*/
10885: PetscErrorCode MatMatTransposeMult(Mat A, Mat B, MatReuse scall, PetscReal fill, Mat *C)
10886: {
10887: PetscFunctionBegin;
10888: PetscCall(MatProduct_Private(A, B, scall, fill, MATPRODUCT_ABt, C));
10889: if (A == B) PetscCall(MatSetOption(*C, MAT_SYMMETRIC, PETSC_TRUE));
10890: PetscFunctionReturn(PETSC_SUCCESS);
10891: }
10893: /*@
10894: MatTransposeMatMult - Performs matrix-matrix multiplication $C = A^T*B$.
10896: Neighbor-wise Collective
10898: Input Parameters:
10899: + A - the left matrix
10900: . B - the right matrix
10901: . scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
10902: - fill - expected fill as ratio of nnz(C)/(nnz(A) + nnz(B)), use `PETSC_DETERMINE` or `PETSC_CURRENT` if not known
10904: Output Parameter:
10905: . C - the product matrix
10907: Level: intermediate
10909: Notes:
10910: `C` will be created if `MAT_INITIAL_MATRIX` and must be destroyed by the user with `MatDestroy()`.
10912: `MAT_REUSE_MATRIX` can only be used if `A` and `B` have the same nonzero pattern as in the previous call.
10914: This is a convenience routine that wraps the use of `MatProductCreate()` with a `MatProductType` of `MATPRODUCT_AtB`
10915: functionality into a single function call. For more involved matrix-matrix operations see `MatProductCreate()`.
10917: To determine the correct fill value, run with -info and search for the string "Fill ratio" to see the value
10918: actually needed.
10920: This routine is currently implemented for pairs of `MATAIJ` matrices and pairs of `MATSEQDENSE` matrices and classes
10921: which inherit from `MATSEQAIJ`. `C` will be of the same type as the input matrices.
10923: The deprecated `PETSC_DEFAULT` in `fill` also means use the current value
10925: .seealso: [](ch_matrices), `Mat`, `MatProductCreate()`, `MATPRODUCT_AtB`, `MatMatMult()`, `MatMatTransposeMult()`, `MatPtAP()`
10926: @*/
10927: PetscErrorCode MatTransposeMatMult(Mat A, Mat B, MatReuse scall, PetscReal fill, Mat *C)
10928: {
10929: PetscFunctionBegin;
10930: PetscCall(MatProduct_Private(A, B, scall, fill, MATPRODUCT_AtB, C));
10931: PetscFunctionReturn(PETSC_SUCCESS);
10932: }
10934: /*@
10935: MatMatMatMult - Performs matrix-matrix-matrix multiplication D=A*B*C.
10937: Neighbor-wise Collective
10939: Input Parameters:
10940: + A - the left matrix
10941: . B - the middle matrix
10942: . C - the right matrix
10943: . scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
10944: - fill - expected fill as ratio of nnz(D)/(nnz(A) + nnz(B)+nnz(C)), use `PETSC_DETERMINE` or `PETSC_CURRENT` if you do not have a good estimate
10945: if the result is a dense matrix this is irrelevant
10947: Output Parameter:
10948: . D - the product matrix
10950: Level: intermediate
10952: Notes:
10953: Unless `scall` is `MAT_REUSE_MATRIX` `D` will be created.
10955: `MAT_REUSE_MATRIX` can only be used if the matrices `A`, `B`, and `C` have the same nonzero pattern as in the previous call
10957: This is a convenience routine that wraps the use of the `MatProductCreate()` with a `MatProductType` of `MATPRODUCT_ABC`
10958: functionality into a single function call. For more involved matrix-matrix operations see `MatProductCreate()`.
10960: To determine the correct fill value, run with `-info` and search for the string "Fill ratio" to see the value
10961: actually needed.
10963: If you have many matrices with the same non-zero structure to multiply, you
10964: should use `MAT_REUSE_MATRIX` in all calls but the first
10966: The deprecated `PETSC_DEFAULT` in `fill` also means use the current value
10968: .seealso: [](ch_matrices), `Mat`, `MatProductCreate()`, `MATPRODUCT_ABC`, `MatMatMult`, `MatPtAP()`, `MatMatTransposeMult()`, `MatTransposeMatMult()`
10969: @*/
10970: PetscErrorCode MatMatMatMult(Mat A, Mat B, Mat C, MatReuse scall, PetscReal fill, Mat *D)
10971: {
10972: PetscFunctionBegin;
10973: if (scall == MAT_REUSE_MATRIX) MatCheckProduct(*D, 6);
10974: PetscCheck(scall != MAT_INPLACE_MATRIX, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Inplace product not supported");
10976: if (scall == MAT_INITIAL_MATRIX) {
10977: PetscCall(MatProductCreate(A, B, C, D));
10978: PetscCall(MatProductSetType(*D, MATPRODUCT_ABC));
10979: PetscCall(MatProductSetAlgorithm(*D, "default"));
10980: PetscCall(MatProductSetFill(*D, fill));
10982: (*D)->product->api_user = PETSC_TRUE;
10983: PetscCall(MatProductSetFromOptions(*D));
10984: PetscCheck((*D)->ops->productsymbolic, PetscObjectComm((PetscObject)*D), PETSC_ERR_SUP, "MatProduct %s not supported for A %s, B %s and C %s", MatProductTypes[MATPRODUCT_ABC], ((PetscObject)A)->type_name, ((PetscObject)B)->type_name,
10985: ((PetscObject)C)->type_name);
10986: PetscCall(MatProductSymbolic(*D));
10987: } else { /* user may change input matrices when REUSE */
10988: PetscCall(MatProductReplaceMats(A, B, C, *D));
10989: }
10990: PetscCall(MatProductNumeric(*D));
10991: PetscFunctionReturn(PETSC_SUCCESS);
10992: }
10994: /*@
10995: MatCreateRedundantMatrix - Create redundant matrices and put them into processes of subcommunicators.
10997: Collective
10999: Input Parameters:
11000: + mat - the matrix
11001: . nsubcomm - the number of subcommunicators (= number of redundant parallel or sequential matrices)
11002: . subcomm - MPI communicator split from the communicator where mat resides in (or `MPI_COMM_NULL` if nsubcomm is used)
11003: - reuse - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
11005: Output Parameter:
11006: . matredundant - redundant matrix
11008: Level: advanced
11010: Notes:
11011: `MAT_REUSE_MATRIX` can only be used when the nonzero structure of the
11012: original matrix has not changed from that last call to `MatCreateRedundantMatrix()`.
11014: This routine creates the duplicated matrices in the subcommunicators; you should NOT create them before
11015: calling it.
11017: `PetscSubcommCreate()` can be used to manage the creation of the subcomm but need not be.
11019: .seealso: [](ch_matrices), `Mat`, `MatDestroy()`, `PetscSubcommCreate()`, `PetscSubcomm`
11020: @*/
11021: PetscErrorCode MatCreateRedundantMatrix(Mat mat, PetscInt nsubcomm, MPI_Comm subcomm, MatReuse reuse, Mat *matredundant)
11022: {
11023: MPI_Comm comm;
11024: PetscMPIInt size;
11025: PetscInt mloc_sub, nloc_sub, rstart, rend, M = mat->rmap->N, N = mat->cmap->N, bs = mat->rmap->bs;
11026: Mat_Redundant *redund = NULL;
11027: PetscSubcomm psubcomm = NULL;
11028: MPI_Comm subcomm_in = subcomm;
11029: Mat *matseq;
11030: IS isrow, iscol;
11031: PetscBool newsubcomm = PETSC_FALSE;
11033: PetscFunctionBegin;
11035: if (nsubcomm && reuse == MAT_REUSE_MATRIX) {
11036: PetscAssertPointer(*matredundant, 5);
11038: }
11040: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)mat), &size));
11041: if (size == 1 || nsubcomm == 1) {
11042: if (reuse == MAT_INITIAL_MATRIX) {
11043: PetscCall(MatDuplicate(mat, MAT_COPY_VALUES, matredundant));
11044: } else {
11045: PetscCheck(*matredundant != mat, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "MAT_REUSE_MATRIX means reuse the matrix passed in as the final argument, not the original matrix");
11046: PetscCall(MatCopy(mat, *matredundant, SAME_NONZERO_PATTERN));
11047: }
11048: PetscFunctionReturn(PETSC_SUCCESS);
11049: }
11051: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
11052: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
11053: MatCheckPreallocated(mat, 1);
11055: PetscCall(PetscLogEventBegin(MAT_RedundantMat, mat, 0, 0, 0));
11056: if (subcomm_in == MPI_COMM_NULL && reuse == MAT_INITIAL_MATRIX) { /* get subcomm if user does not provide subcomm */
11057: /* create psubcomm, then get subcomm */
11058: PetscCall(PetscObjectGetComm((PetscObject)mat, &comm));
11059: PetscCallMPI(MPI_Comm_size(comm, &size));
11060: PetscCheck(nsubcomm >= 1 && nsubcomm <= size, PETSC_COMM_SELF, PETSC_ERR_ARG_SIZ, "nsubcomm must between 1 and %d", size);
11062: PetscCall(PetscSubcommCreate(comm, &psubcomm));
11063: PetscCall(PetscSubcommSetNumber(psubcomm, nsubcomm));
11064: PetscCall(PetscSubcommSetType(psubcomm, PETSC_SUBCOMM_CONTIGUOUS));
11065: PetscCall(PetscSubcommSetFromOptions(psubcomm));
11066: PetscCall(PetscCommDuplicate(PetscSubcommChild(psubcomm), &subcomm, NULL));
11067: newsubcomm = PETSC_TRUE;
11068: PetscCall(PetscSubcommDestroy(&psubcomm));
11069: }
11071: /* get isrow, iscol and a local sequential matrix matseq[0] */
11072: if (reuse == MAT_INITIAL_MATRIX) {
11073: mloc_sub = PETSC_DECIDE;
11074: nloc_sub = PETSC_DECIDE;
11075: if (bs < 1) {
11076: PetscCall(PetscSplitOwnership(subcomm, &mloc_sub, &M));
11077: PetscCall(PetscSplitOwnership(subcomm, &nloc_sub, &N));
11078: } else {
11079: PetscCall(PetscSplitOwnershipBlock(subcomm, bs, &mloc_sub, &M));
11080: PetscCall(PetscSplitOwnershipBlock(subcomm, bs, &nloc_sub, &N));
11081: }
11082: PetscCallMPI(MPI_Scan(&mloc_sub, &rend, 1, MPIU_INT, MPI_SUM, subcomm));
11083: rstart = rend - mloc_sub;
11084: PetscCall(ISCreateStride(PETSC_COMM_SELF, mloc_sub, rstart, 1, &isrow));
11085: PetscCall(ISCreateStride(PETSC_COMM_SELF, N, 0, 1, &iscol));
11086: PetscCall(ISSetIdentity(iscol));
11087: } else { /* reuse == MAT_REUSE_MATRIX */
11088: PetscCheck(*matredundant != mat, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "MAT_REUSE_MATRIX means reuse the matrix passed in as the final argument, not the original matrix");
11089: /* retrieve subcomm */
11090: PetscCall(PetscObjectGetComm((PetscObject)*matredundant, &subcomm));
11091: redund = (*matredundant)->redundant;
11092: isrow = redund->isrow;
11093: iscol = redund->iscol;
11094: matseq = redund->matseq;
11095: }
11096: PetscCall(MatCreateSubMatrices(mat, 1, &isrow, &iscol, reuse, &matseq));
11098: /* get matredundant over subcomm */
11099: if (reuse == MAT_INITIAL_MATRIX) {
11100: PetscCall(MatCreateMPIMatConcatenateSeqMat(subcomm, matseq[0], nloc_sub, reuse, matredundant));
11102: /* create a supporting struct and attach it to C for reuse */
11103: PetscCall(PetscNew(&redund));
11104: (*matredundant)->redundant = redund;
11105: redund->isrow = isrow;
11106: redund->iscol = iscol;
11107: redund->matseq = matseq;
11108: if (newsubcomm) {
11109: redund->subcomm = subcomm;
11110: } else {
11111: redund->subcomm = MPI_COMM_NULL;
11112: }
11113: } else {
11114: PetscCall(MatCreateMPIMatConcatenateSeqMat(subcomm, matseq[0], PETSC_DECIDE, reuse, matredundant));
11115: }
11116: #if PetscDefined(HAVE_VIENNACL) || PetscDefined(HAVE_CUDA) || PetscDefined(HAVE_HIP)
11117: if (matseq[0]->boundtocpu && matseq[0]->bindingpropagates) {
11118: PetscCall(MatBindToCPU(*matredundant, PETSC_TRUE));
11119: PetscCall(MatSetBindingPropagates(*matredundant, PETSC_TRUE));
11120: }
11121: #endif
11122: PetscCall(PetscLogEventEnd(MAT_RedundantMat, mat, 0, 0, 0));
11123: PetscFunctionReturn(PETSC_SUCCESS);
11124: }
11126: /*@
11127: MatGetMultiProcBlock - Create multiple 'parallel submatrices' from
11128: a given `Mat`. Each submatrix can span multiple procs.
11130: Collective
11132: Input Parameters:
11133: + mat - the matrix
11134: . subComm - the sub communicator obtained as if by `MPI_Comm_split(PetscObjectComm((PetscObject)mat))`
11135: - scall - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
11137: Output Parameter:
11138: . subMat - parallel sub-matrices each spanning a given `subcomm`
11140: Level: advanced
11142: Notes:
11143: The submatrix partition across processes is dictated by `subComm` a
11144: communicator obtained by `MPI_comm_split()` or via `PetscSubcommCreate()`. The `subComm`
11145: is not restricted to be grouped with consecutive original MPI processes.
11147: Due the `MPI_Comm_split()` usage, the parallel layout of the submatrices
11148: map directly to the layout of the original matrix [wrt the local
11149: row,col partitioning]. So the original 'DiagonalMat' naturally maps
11150: into the 'DiagonalMat' of the `subMat`, hence it is used directly from
11151: the `subMat`. However the offDiagMat looses some columns - and this is
11152: reconstructed with `MatSetValues()`
11154: This is used by `PCBJACOBI` when a single block spans multiple MPI processes.
11156: .seealso: [](ch_matrices), `Mat`, `MatCreateRedundantMatrix()`, `MatCreateSubMatrices()`, `PCBJACOBI`
11157: @*/
11158: PetscErrorCode MatGetMultiProcBlock(Mat mat, MPI_Comm subComm, MatReuse scall, Mat *subMat)
11159: {
11160: PetscMPIInt commsize, subCommSize;
11162: PetscFunctionBegin;
11163: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)mat), &commsize));
11164: PetscCallMPI(MPI_Comm_size(subComm, &subCommSize));
11165: PetscCheck(subCommSize <= commsize, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_OUTOFRANGE, "CommSize %d < SubCommZize %d", commsize, subCommSize);
11167: PetscCheck(scall != MAT_REUSE_MATRIX || *subMat != mat, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "MAT_REUSE_MATRIX means reuse the matrix passed in as the final argument, not the original matrix");
11168: PetscCall(PetscLogEventBegin(MAT_GetMultiProcBlock, mat, 0, 0, 0));
11169: PetscUseTypeMethod(mat, getmultiprocblock, subComm, scall, subMat);
11170: PetscCall(PetscLogEventEnd(MAT_GetMultiProcBlock, mat, 0, 0, 0));
11171: PetscFunctionReturn(PETSC_SUCCESS);
11172: }
11174: /*@
11175: MatGetLocalSubMatrix - Gets a reference to a submatrix specified in local numbering
11177: Not Collective
11179: Input Parameters:
11180: + mat - matrix to extract local submatrix from
11181: . isrow - local row indices for submatrix
11182: - iscol - local column indices for submatrix
11184: Output Parameter:
11185: . submat - the submatrix
11187: Level: intermediate
11189: Notes:
11190: `submat` should be disposed of with `MatRestoreLocalSubMatrix()`.
11192: Depending on the format of `mat`, the returned `submat` may not implement `MatMult()`. Its communicator may be
11193: the same as `mat`, it may be `PETSC_COMM_SELF`, or some other sub-communictor of `mat`'s.
11195: `submat` always implements `MatSetValuesLocal()`. If `isrow` and `iscol` have the same block size, then
11196: `MatSetValuesBlockedLocal()` will also be implemented.
11198: `mat` must have had a `ISLocalToGlobalMapping` provided to it with `MatSetLocalToGlobalMapping()`.
11199: Matrices obtained with `DMCreateMatrix()` generally already have the local to global mapping provided.
11201: .seealso: [](ch_matrices), `Mat`, `MatRestoreLocalSubMatrix()`, `MatCreateLocalRef()`, `MatSetLocalToGlobalMapping()`
11202: @*/
11203: PetscErrorCode MatGetLocalSubMatrix(Mat mat, IS isrow, IS iscol, Mat *submat)
11204: {
11205: PetscFunctionBegin;
11209: PetscCheckSameComm(isrow, 2, iscol, 3);
11210: PetscAssertPointer(submat, 4);
11211: PetscCheck(mat->rmap->mapping, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Matrix must have local to global mapping provided before this call");
11213: if (mat->ops->getlocalsubmatrix) {
11214: PetscUseTypeMethod(mat, getlocalsubmatrix, isrow, iscol, submat);
11215: } else {
11216: PetscCall(MatCreateLocalRef(mat, isrow, iscol, submat));
11217: }
11218: (*submat)->assembled = mat->assembled;
11219: PetscFunctionReturn(PETSC_SUCCESS);
11220: }
11222: /*@
11223: MatRestoreLocalSubMatrix - Restores a reference to a submatrix specified in local numbering obtained with `MatGetLocalSubMatrix()`
11225: Not Collective
11227: Input Parameters:
11228: + mat - matrix to extract local submatrix from
11229: . isrow - local row indices for submatrix
11230: . iscol - local column indices for submatrix
11231: - submat - the submatrix
11233: Level: intermediate
11235: .seealso: [](ch_matrices), `Mat`, `MatGetLocalSubMatrix()`
11236: @*/
11237: PetscErrorCode MatRestoreLocalSubMatrix(Mat mat, IS isrow, IS iscol, Mat *submat)
11238: {
11239: PetscFunctionBegin;
11243: PetscCheckSameComm(isrow, 2, iscol, 3);
11244: PetscAssertPointer(submat, 4);
11247: if (mat->ops->restorelocalsubmatrix) {
11248: PetscUseTypeMethod(mat, restorelocalsubmatrix, isrow, iscol, submat);
11249: } else {
11250: PetscCall(MatDestroy(submat));
11251: }
11252: *submat = NULL;
11253: PetscFunctionReturn(PETSC_SUCCESS);
11254: }
11256: /*@
11257: MatFindZeroDiagonals - Finds all the rows of a matrix that have zero or no diagonal entry in the matrix
11259: Collective
11261: Input Parameter:
11262: . mat - the matrix
11264: Output Parameter:
11265: . is - if any rows have zero diagonals this contains the list of them
11267: Level: developer
11269: .seealso: [](ch_matrices), `Mat`, `MatMultTranspose()`, `MatMultAdd()`, `MatMultTransposeAdd()`
11270: @*/
11271: PetscErrorCode MatFindZeroDiagonals(Mat mat, IS *is)
11272: {
11273: PetscFunctionBegin;
11276: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
11277: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
11279: if (!mat->ops->findzerodiagonals) {
11280: Vec diag;
11281: const PetscScalar *a;
11282: PetscInt *rows;
11283: PetscInt rStart, rEnd, r, nrow = 0;
11285: PetscCall(MatCreateVecs(mat, &diag, NULL));
11286: PetscCall(MatGetDiagonal(mat, diag));
11287: PetscCall(MatGetOwnershipRange(mat, &rStart, &rEnd));
11288: PetscCall(VecGetArrayRead(diag, &a));
11289: for (r = 0; r < rEnd - rStart; ++r)
11290: if (a[r] == 0.0) ++nrow;
11291: PetscCall(PetscMalloc1(nrow, &rows));
11292: nrow = 0;
11293: for (r = 0; r < rEnd - rStart; ++r)
11294: if (a[r] == 0.0) rows[nrow++] = r + rStart;
11295: PetscCall(VecRestoreArrayRead(diag, &a));
11296: PetscCall(VecDestroy(&diag));
11297: PetscCall(ISCreateGeneral(PetscObjectComm((PetscObject)mat), nrow, rows, PETSC_OWN_POINTER, is));
11298: } else {
11299: PetscUseTypeMethod(mat, findzerodiagonals, is);
11300: }
11301: PetscFunctionReturn(PETSC_SUCCESS);
11302: }
11304: /*@
11305: MatFindOffBlockDiagonalEntries - Finds all the rows of a matrix that have entries outside of the main diagonal block (defined by the matrix block size)
11307: Collective
11309: Input Parameter:
11310: . mat - the matrix
11312: Output Parameter:
11313: . is - contains the list of rows with off block diagonal entries
11315: Level: developer
11317: .seealso: [](ch_matrices), `Mat`, `MatMultTranspose()`, `MatMultAdd()`, `MatMultTransposeAdd()`
11318: @*/
11319: PetscErrorCode MatFindOffBlockDiagonalEntries(Mat mat, IS *is)
11320: {
11321: PetscFunctionBegin;
11324: PetscCheck(mat->assembled, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
11325: PetscCheck(!mat->factortype, PetscObjectComm((PetscObject)mat), PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
11327: PetscUseTypeMethod(mat, findoffblockdiagonalentries, is);
11328: PetscFunctionReturn(PETSC_SUCCESS);
11329: }
11331: /*@
11332: MatInvertBlockDiagonal - Inverts the block diagonal entries.
11334: Collective; No Fortran Support
11336: Input Parameter:
11337: . mat - the matrix
11339: Output Parameter:
11340: . values - the block inverses in column major order (FORTRAN-like)
11342: Level: advanced
11344: Notes:
11345: The size of the blocks is determined by the block size of the matrix.
11347: The blocks never overlap between two MPI processes, use `MatInvertVariableBlockEnvelope()` for that case
11349: The blocks all have the same size, use `MatInvertVariableBlockDiagonal()` for variable block size
11351: .seealso: [](ch_matrices), `Mat`, `MatInvertVariableBlockEnvelope()`, `MatInvertBlockDiagonalMat()`
11352: @*/
11353: PetscErrorCode MatInvertBlockDiagonal(Mat mat, const PetscScalar *values[])
11354: {
11355: PetscFunctionBegin;
11357: PetscCheck(mat->assembled, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
11358: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
11359: PetscUseTypeMethod(mat, invertblockdiagonal, values);
11360: PetscFunctionReturn(PETSC_SUCCESS);
11361: }
11363: /*@
11364: MatInvertVariableBlockDiagonal - Inverts the point block diagonal entries.
11366: Collective; No Fortran Support
11368: Input Parameters:
11369: + mat - the matrix
11370: . nblocks - the number of blocks on the process, set with `MatSetVariableBlockSizes()`
11371: - bsizes - the size of each block on the process, set with `MatSetVariableBlockSizes()`
11373: Output Parameter:
11374: . values - the block inverses in column major order (FORTRAN-like)
11376: Level: advanced
11378: Notes:
11379: Use `MatInvertBlockDiagonal()` if all blocks have the same size
11381: The blocks never overlap between two MPI processes, use `MatInvertVariableBlockEnvelope()` for that case
11383: .seealso: [](ch_matrices), `Mat`, `MatInvertBlockDiagonal()`, `MatSetVariableBlockSizes()`, `MatInvertVariableBlockEnvelope()`
11384: @*/
11385: PetscErrorCode MatInvertVariableBlockDiagonal(Mat mat, PetscInt nblocks, const PetscInt bsizes[], PetscScalar values[])
11386: {
11387: PetscFunctionBegin;
11389: PetscCheck(mat->assembled, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for unassembled matrix");
11390: PetscCheck(!mat->factortype, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONGSTATE, "Not for factored matrix");
11391: PetscUseTypeMethod(mat, invertvariableblockdiagonal, nblocks, bsizes, values);
11392: PetscFunctionReturn(PETSC_SUCCESS);
11393: }
11395: /*@
11396: MatInvertBlockDiagonalMat - set the values of matrix C to be the inverted block diagonal of matrix A
11398: Collective
11400: Input Parameters:
11401: + A - the matrix
11402: - C - matrix with inverted block diagonal of `A`. This matrix should be created and may have its type set.
11404: Level: advanced
11406: Note:
11407: The blocksize of the matrix is used to determine the blocks on the diagonal of `C`
11409: .seealso: [](ch_matrices), `Mat`, `MatInvertBlockDiagonal()`
11410: @*/
11411: PetscErrorCode MatInvertBlockDiagonalMat(Mat A, Mat C)
11412: {
11413: const PetscScalar *vals;
11414: PetscInt *dnnz;
11415: PetscInt m, rstart, rend, bs, i, j;
11417: PetscFunctionBegin;
11418: PetscCall(MatInvertBlockDiagonal(A, &vals));
11419: PetscCall(MatGetBlockSize(A, &bs));
11420: PetscCall(MatGetLocalSize(A, &m, NULL));
11421: PetscCall(MatSetLayouts(C, A->rmap, A->cmap));
11422: PetscCall(MatSetBlockSizes(C, A->rmap->bs, A->cmap->bs));
11423: PetscCall(PetscMalloc1(m / bs, &dnnz));
11424: for (j = 0; j < m / bs; j++) dnnz[j] = 1;
11425: PetscCall(MatXAIJSetPreallocation(C, bs, dnnz, NULL, NULL, NULL));
11426: PetscCall(PetscFree(dnnz));
11427: PetscCall(MatGetOwnershipRange(C, &rstart, &rend));
11428: PetscCall(MatSetOption(C, MAT_ROW_ORIENTED, PETSC_FALSE));
11429: for (i = rstart / bs; i < rend / bs; i++) PetscCall(MatSetValuesBlocked(C, 1, &i, 1, &i, &vals[(i - rstart / bs) * bs * bs], INSERT_VALUES));
11430: PetscCall(MatSetOption(C, MAT_NO_OFF_PROC_ENTRIES, PETSC_TRUE));
11431: PetscCall(MatAssemblyBegin(C, MAT_FINAL_ASSEMBLY));
11432: PetscCall(MatAssemblyEnd(C, MAT_FINAL_ASSEMBLY));
11433: PetscCall(MatSetOption(C, MAT_NO_OFF_PROC_ENTRIES, PETSC_FALSE));
11434: PetscCall(MatSetOption(C, MAT_ROW_ORIENTED, PETSC_TRUE));
11435: PetscFunctionReturn(PETSC_SUCCESS);
11436: }
11438: /*@
11439: MatTransposeColoringDestroy - Destroys a coloring context for matrix product $C = A*B^T$ that was created
11440: via `MatTransposeColoringCreate()`.
11442: Collective
11444: Input Parameter:
11445: . c - coloring context
11447: Level: intermediate
11449: .seealso: [](ch_matrices), `Mat`, `MatTransposeColoringCreate()`
11450: @*/
11451: PetscErrorCode MatTransposeColoringDestroy(MatTransposeColoring *c)
11452: {
11453: MatTransposeColoring matcolor = *c;
11455: PetscFunctionBegin;
11456: if (!matcolor) PetscFunctionReturn(PETSC_SUCCESS);
11457: if (--((PetscObject)matcolor)->refct > 0) {
11458: matcolor = NULL;
11459: PetscFunctionReturn(PETSC_SUCCESS);
11460: }
11462: PetscCall(PetscFree3(matcolor->ncolumns, matcolor->nrows, matcolor->colorforrow));
11463: PetscCall(PetscFree(matcolor->rows));
11464: PetscCall(PetscFree(matcolor->den2sp));
11465: PetscCall(PetscFree(matcolor->colorforcol));
11466: PetscCall(PetscFree(matcolor->columns));
11467: if (matcolor->brows > 0) PetscCall(PetscFree(matcolor->lstart));
11468: PetscCall(PetscHeaderDestroy(c));
11469: PetscFunctionReturn(PETSC_SUCCESS);
11470: }
11472: /*@
11473: MatTransColoringApplySpToDen - Given a symbolic matrix product $C = A*B^T$ for which
11474: a `MatTransposeColoring` context has been created, computes a dense $B^T$ by applying
11475: `MatTransposeColoring` to sparse `B`.
11477: Collective
11479: Input Parameters:
11480: + coloring - coloring context created with `MatTransposeColoringCreate()`
11481: - B - sparse matrix
11483: Output Parameter:
11484: . Btdense - dense matrix $B^T$
11486: Level: developer
11488: Note:
11489: These are used internally for some implementations of `MatRARt()`
11491: .seealso: [](ch_matrices), `Mat`, `MatTransposeColoringCreate()`, `MatTransposeColoringDestroy()`, `MatTransColoringApplyDenToSp()`
11492: @*/
11493: PetscErrorCode MatTransColoringApplySpToDen(MatTransposeColoring coloring, Mat B, Mat Btdense)
11494: {
11495: PetscFunctionBegin;
11500: PetscCall((*B->ops->transcoloringapplysptoden)(coloring, B, Btdense));
11501: PetscFunctionReturn(PETSC_SUCCESS);
11502: }
11504: /*@
11505: MatTransColoringApplyDenToSp - Given a symbolic matrix product $C_{sp} = A*B^T$ for which
11506: a `MatTransposeColoring` context has been created and a dense matrix $C_{den} = A*B^T_{dense}$
11507: in which `B^T_{dens}` is obtained from `MatTransColoringApplySpToDen()`, recover sparse matrix
11508: $C_{sp}$ from $C_{den}$.
11510: Collective
11512: Input Parameters:
11513: + matcoloring - coloring context created with `MatTransposeColoringCreate()`
11514: - Cden - matrix product of a sparse matrix and a dense matrix Btdense
11516: Output Parameter:
11517: . Csp - sparse matrix
11519: Level: developer
11521: Note:
11522: These are used internally for some implementations of `MatRARt()`
11524: .seealso: [](ch_matrices), `Mat`, `MatTransposeColoringCreate()`, `MatTransposeColoringDestroy()`, `MatTransColoringApplySpToDen()`
11525: @*/
11526: PetscErrorCode MatTransColoringApplyDenToSp(MatTransposeColoring matcoloring, Mat Cden, Mat Csp)
11527: {
11528: PetscFunctionBegin;
11533: PetscCall((*Csp->ops->transcoloringapplydentosp)(matcoloring, Cden, Csp));
11534: PetscCall(MatAssemblyBegin(Csp, MAT_FINAL_ASSEMBLY));
11535: PetscCall(MatAssemblyEnd(Csp, MAT_FINAL_ASSEMBLY));
11536: PetscFunctionReturn(PETSC_SUCCESS);
11537: }
11539: /*@
11540: MatTransposeColoringCreate - Creates a matrix coloring context for the matrix product $C = A*B^T$.
11542: Collective
11544: Input Parameters:
11545: + mat - the matrix product C
11546: - iscoloring - the coloring of the matrix; usually obtained with `MatColoringCreate()` or `DMCreateColoring()`
11548: Output Parameter:
11549: . color - the new coloring context
11551: Level: intermediate
11553: .seealso: [](ch_matrices), `Mat`, `MatTransposeColoringDestroy()`, `MatTransColoringApplySpToDen()`,
11554: `MatTransColoringApplyDenToSp()`
11555: @*/
11556: PetscErrorCode MatTransposeColoringCreate(Mat mat, ISColoring iscoloring, MatTransposeColoring *color)
11557: {
11558: MatTransposeColoring c;
11559: MPI_Comm comm;
11561: PetscFunctionBegin;
11562: PetscAssertPointer(color, 3);
11564: PetscCall(PetscLogEventBegin(MAT_TransposeColoringCreate, mat, 0, 0, 0));
11565: PetscCall(PetscObjectGetComm((PetscObject)mat, &comm));
11566: PetscCall(PetscHeaderCreate(c, MAT_TRANSPOSECOLORING_CLASSID, "MatTransposeColoring", "Matrix product C=A*B^T via coloring", "Mat", comm, MatTransposeColoringDestroy, NULL));
11567: c->ctype = iscoloring->ctype;
11568: PetscUseTypeMethod(mat, transposecoloringcreate, iscoloring, c);
11569: *color = c;
11570: PetscCall(PetscLogEventEnd(MAT_TransposeColoringCreate, mat, 0, 0, 0));
11571: PetscFunctionReturn(PETSC_SUCCESS);
11572: }
11574: /*@
11575: MatGetNonzeroState - Returns a 64-bit integer representing the current state of nonzeros in the matrix. If the
11576: matrix has had new nonzero locations added to (or removed from) the matrix since the previous call, the value will be larger.
11578: Not Collective
11580: Input Parameter:
11581: . mat - the matrix
11583: Output Parameter:
11584: . state - the current state
11586: Level: intermediate
11588: Notes:
11589: You can only compare states from two different calls to the SAME matrix, you cannot compare calls between
11590: different matrices
11592: Use `PetscObjectStateGet()` to check for changes to the numerical values in a matrix
11594: Use the result of `PetscObjectGetId()` to compare if a previously checked matrix is the same as the current matrix, do not compare object pointers.
11596: .seealso: [](ch_matrices), `Mat`, `PetscObjectStateGet()`, `PetscObjectGetId()`
11597: @*/
11598: PetscErrorCode MatGetNonzeroState(Mat mat, PetscObjectState *state)
11599: {
11600: PetscFunctionBegin;
11602: *state = mat->nonzerostate;
11603: PetscFunctionReturn(PETSC_SUCCESS);
11604: }
11606: /*@
11607: MatCreateMPIMatConcatenateSeqMat - Creates a single large PETSc matrix by concatenating sequential
11608: matrices from each process
11610: Collective
11612: Input Parameters:
11613: + comm - the communicators the parallel matrix will live on
11614: . seqmat - the input sequential matrices
11615: . n - number of local columns (or `PETSC_DECIDE`)
11616: - reuse - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
11618: Output Parameter:
11619: . mpimat - the parallel matrix generated
11621: Level: developer
11623: Note:
11624: The number of columns of the matrix in EACH process MUST be the same.
11626: .seealso: [](ch_matrices), `Mat`
11627: @*/
11628: PetscErrorCode MatCreateMPIMatConcatenateSeqMat(MPI_Comm comm, Mat seqmat, PetscInt n, MatReuse reuse, Mat *mpimat)
11629: {
11630: PetscMPIInt size;
11632: PetscFunctionBegin;
11633: PetscCallMPI(MPI_Comm_size(comm, &size));
11634: if (size == 1) {
11635: if (reuse == MAT_INITIAL_MATRIX) {
11636: PetscCall(MatDuplicate(seqmat, MAT_COPY_VALUES, mpimat));
11637: } else {
11638: PetscCall(MatCopy(seqmat, *mpimat, SAME_NONZERO_PATTERN));
11639: }
11640: PetscFunctionReturn(PETSC_SUCCESS);
11641: }
11643: PetscCheck(reuse != MAT_REUSE_MATRIX || seqmat != *mpimat, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "MAT_REUSE_MATRIX means reuse the matrix passed in as the final argument, not the original matrix");
11645: PetscCall(PetscLogEventBegin(MAT_Merge, seqmat, 0, 0, 0));
11646: PetscCall((*seqmat->ops->creatempimatconcatenateseqmat)(comm, seqmat, n, reuse, mpimat));
11647: PetscCall(PetscLogEventEnd(MAT_Merge, seqmat, 0, 0, 0));
11648: PetscFunctionReturn(PETSC_SUCCESS);
11649: }
11651: /*@
11652: MatSubdomainsCreateCoalesce - Creates index subdomains by coalescing adjacent MPI processes' ownership ranges.
11654: Collective
11656: Input Parameters:
11657: + A - the matrix to create subdomains from
11658: - N - requested number of subdomains
11660: Output Parameters:
11661: + n - number of subdomains resulting on this MPI process
11662: - iss - `IS` list with indices of subdomains on this MPI process
11664: Level: advanced
11666: Note:
11667: The number of subdomains must be smaller than the communicator size
11669: .seealso: [](ch_matrices), `Mat`, `IS`
11670: @*/
11671: PetscErrorCode MatSubdomainsCreateCoalesce(Mat A, PetscInt N, PetscInt *n, IS *iss[])
11672: {
11673: MPI_Comm comm, subcomm;
11674: PetscMPIInt size, rank, color;
11675: PetscInt rstart, rend, k;
11677: PetscFunctionBegin;
11678: PetscCall(PetscObjectGetComm((PetscObject)A, &comm));
11679: PetscCallMPI(MPI_Comm_size(comm, &size));
11680: PetscCallMPI(MPI_Comm_rank(comm, &rank));
11681: PetscCheck(N >= 1 && N < size, PETSC_COMM_SELF, PETSC_ERR_ARG_WRONG, "number of subdomains must be > 0 and < %d, got N = %" PetscInt_FMT, size, N);
11682: *n = 1;
11683: k = size / N + (size % N > 0); /* There are up to k ranks to a color */
11684: color = rank / k;
11685: PetscCallMPI(MPI_Comm_split(comm, color, rank, &subcomm));
11686: PetscCall(PetscMalloc1(1, iss));
11687: PetscCall(MatGetOwnershipRange(A, &rstart, &rend));
11688: PetscCall(ISCreateStride(subcomm, rend - rstart, rstart, 1, iss[0]));
11689: PetscCallMPI(MPI_Comm_free(&subcomm));
11690: PetscFunctionReturn(PETSC_SUCCESS);
11691: }
11693: /*@
11694: MatGalerkin - Constructs the coarse grid problem matrix via Galerkin projection.
11696: If the interpolation and restriction operators are the same, uses `MatPtAP()`.
11697: If they are not the same, uses `MatMatMatMult()`.
11699: Once the coarse grid problem is constructed, correct for interpolation operators
11700: that are not of full rank, which can legitimately happen in the case of non-nested
11701: geometric multigrid.
11703: Input Parameters:
11704: + restrct - restriction operator
11705: . dA - fine grid matrix
11706: . interpolate - interpolation operator
11707: . reuse - either `MAT_INITIAL_MATRIX` or `MAT_REUSE_MATRIX`
11708: - fill - expected fill, use `PETSC_DETERMINE` or `PETSC_DETERMINE` if you do not have a good estimate
11710: Output Parameter:
11711: . A - the Galerkin coarse matrix
11713: Options Database Key:
11714: . -pc_mg_galerkin (both|pmat|mat|none) - for what matrices the Galerkin process should be used
11716: Level: developer
11718: Note:
11719: The deprecated `PETSC_DEFAULT` in `fill` also means use the current value
11721: .seealso: [](ch_matrices), `Mat`, `MatPtAP()`, `MatMatMatMult()`
11722: @*/
11723: PetscErrorCode MatGalerkin(Mat restrct, Mat dA, Mat interpolate, MatReuse reuse, PetscReal fill, Mat *A)
11724: {
11725: IS zerorows;
11726: Vec diag;
11728: PetscFunctionBegin;
11729: PetscCheck(reuse != MAT_INPLACE_MATRIX, PetscObjectComm((PetscObject)A), PETSC_ERR_SUP, "Inplace product not supported");
11730: /* Construct the coarse grid matrix */
11731: if (interpolate == restrct) {
11732: PetscCall(MatPtAP(dA, interpolate, reuse, fill, A));
11733: } else {
11734: PetscCall(MatMatMatMult(restrct, dA, interpolate, reuse, fill, A));
11735: }
11737: /* If the interpolation matrix is not of full rank, A will have zero rows.
11738: This can legitimately happen in the case of non-nested geometric multigrid.
11739: In that event, we set the rows of the matrix to the rows of the identity,
11740: ignoring the equations (as the RHS will also be zero). */
11742: PetscCall(MatFindZeroRows(*A, &zerorows));
11744: if (zerorows != NULL) { /* if there are any zero rows */
11745: PetscCall(MatCreateVecs(*A, &diag, NULL));
11746: PetscCall(MatGetDiagonal(*A, diag));
11747: PetscCall(VecISSet(diag, zerorows, 1.0));
11748: PetscCall(MatDiagonalSet(*A, diag, INSERT_VALUES));
11749: PetscCall(VecDestroy(&diag));
11750: PetscCall(ISDestroy(&zerorows));
11751: }
11752: PetscFunctionReturn(PETSC_SUCCESS);
11753: }
11755: /*@
11756: MatSetOperation - Allows user to set a matrix operation for any matrix type
11758: Logically Collective
11760: Input Parameters:
11761: + mat - the matrix
11762: . op - the name of the operation
11763: - f - the function that provides the operation
11765: Level: developer
11767: Example Usage:
11768: .vb
11769: extern PetscErrorCode usermult(Mat, Vec, Vec);
11771: PetscCall(MatCreateXXX(comm, ..., &A));
11772: PetscCall(MatSetOperation(A, MATOP_MULT, (PetscErrorCodeFn *)usermult));
11773: .ve
11775: Notes:
11776: See the file `include/petscmat.h` for a complete list of matrix
11777: operations, which all have the form MATOP_<OPERATION>, where
11778: <OPERATION> is the name (in all capital letters) of the
11779: user interface routine (e.g., `MatMult()` -> `MATOP_MULT`).
11781: All user-provided functions (except for `MATOP_DESTROY`) should have the same calling
11782: sequence as the usual matrix interface routines, since they
11783: are intended to be accessed via the usual matrix interface
11784: routines, e.g.,
11785: .vb
11786: MatMult(Mat, Vec, Vec) -> usermult(Mat, Vec, Vec)
11787: .ve
11789: In particular each function MUST return `PETSC_SUCCESS` on success and
11790: nonzero on failure.
11792: This routine is distinct from `MatShellSetOperation()` in that it can be called on any matrix type.
11794: .seealso: [](ch_matrices), `Mat`, `MatGetOperation()`, `MatCreateShell()`, `MatShellSetContext()`, `MatShellSetOperation()`
11795: @*/
11796: PetscErrorCode MatSetOperation(Mat mat, MatOperation op, PetscErrorCodeFn *f)
11797: {
11798: PetscFunctionBegin;
11801: if (op == MATOP_VIEW && !mat->ops->viewnative && f != (PetscErrorCodeFn *)mat->ops->view) mat->ops->viewnative = mat->ops->view;
11802: #if !PetscDefined(USE_COMPLEX)
11803: if (op == MATOP_MULT_HERMITIAN_TRANSPOSE) op = MATOP_MULT_TRANSPOSE;
11804: else if (op == MATOP_MULT_HERMITIAN_TRANS_ADD) op = MATOP_MULT_TRANSPOSE_ADD;
11805: else if (op == MATOP_HERMITIAN_TRANSPOSE) op = MATOP_TRANSPOSE;
11806: #endif
11807: ((PetscErrorCodeFn **)mat->ops)[op] = f;
11808: PetscFunctionReturn(PETSC_SUCCESS);
11809: }
11811: /*@
11812: MatGetOperation - Gets a matrix operation for any matrix type.
11814: Not Collective
11816: Input Parameters:
11817: + mat - the matrix
11818: - op - the name of the operation
11820: Output Parameter:
11821: . f - the function that provides the operation
11823: Level: developer
11825: Example Usage:
11826: .vb
11827: PetscErrorCode (*usermult)(Mat, Vec, Vec);
11829: MatGetOperation(A, MATOP_MULT, (PetscErrorCodeFn **)&usermult);
11830: .ve
11832: Notes:
11833: See the file `include/petscmat.h` for a complete list of matrix
11834: operations, which all have the form MATOP_<OPERATION>, where
11835: <OPERATION> is the name (in all capital letters) of the
11836: user interface routine (e.g., `MatMult()` -> `MATOP_MULT`).
11838: This routine is distinct from `MatShellGetOperation()` in that it can be called on any matrix type.
11840: .seealso: [](ch_matrices), `Mat`, `MatSetOperation()`, `MatCreateShell()`, `MatShellGetContext()`, `MatShellGetOperation()`
11841: @*/
11842: PetscErrorCode MatGetOperation(Mat mat, MatOperation op, PetscErrorCodeFn **f)
11843: {
11844: PetscFunctionBegin;
11846: PetscAssertPointer(f, 3);
11847: #if !PetscDefined(USE_COMPLEX)
11848: if (op == MATOP_MULT_HERMITIAN_TRANSPOSE) op = MATOP_MULT_TRANSPOSE;
11849: else if (op == MATOP_MULT_HERMITIAN_TRANS_ADD) op = MATOP_MULT_TRANSPOSE_ADD;
11850: else if (op == MATOP_HERMITIAN_TRANSPOSE) op = MATOP_TRANSPOSE;
11851: #endif
11852: *f = ((PetscErrorCodeFn **)mat->ops)[op];
11853: PetscFunctionReturn(PETSC_SUCCESS);
11854: }
11856: /*@
11857: MatHasOperation - Determines whether the given matrix supports the particular operation.
11859: Not Collective
11861: Input Parameters:
11862: + mat - the matrix
11863: - op - the operation, for example, `MATOP_GET_DIAGONAL`
11865: Output Parameter:
11866: . has - either `PETSC_TRUE` or `PETSC_FALSE`
11868: Level: advanced
11870: Note:
11871: See `MatSetOperation()` for additional discussion on naming convention and usage of `op`.
11873: .seealso: [](ch_matrices), `Mat`, `MatCreateShell()`, `MatGetOperation()`, `MatSetOperation()`
11874: @*/
11875: PetscErrorCode MatHasOperation(Mat mat, MatOperation op, PetscBool *has)
11876: {
11877: PetscFunctionBegin;
11879: PetscAssertPointer(has, 3);
11880: #if !PetscDefined(USE_COMPLEX)
11881: if (op == MATOP_MULT_HERMITIAN_TRANSPOSE) op = MATOP_MULT_TRANSPOSE;
11882: else if (op == MATOP_MULT_HERMITIAN_TRANS_ADD) op = MATOP_MULT_TRANSPOSE_ADD;
11883: else if (op == MATOP_HERMITIAN_TRANSPOSE) op = MATOP_TRANSPOSE;
11884: #endif
11885: if (op == MATOP_ADOT || op == MATOP_ANORM) {
11886: /* MatADot() and MatANorm() fall back to MatMult() when the type has no method */
11887: if (((void **)mat->ops)[op]) *has = PETSC_TRUE;
11888: else PetscCall(MatHasOperation(mat, MATOP_MULT, has));
11889: PetscFunctionReturn(PETSC_SUCCESS);
11890: }
11891: if (mat->ops->hasoperation) {
11892: PetscUseTypeMethod(mat, hasoperation, op, has);
11893: } else {
11894: if (((void **)mat->ops)[op]) *has = PETSC_TRUE;
11895: else {
11896: *has = PETSC_FALSE;
11897: if (op == MATOP_CREATE_SUBMATRIX) {
11898: PetscMPIInt size;
11900: PetscCallMPI(MPI_Comm_size(PetscObjectComm((PetscObject)mat), &size));
11901: if (size == 1) PetscCall(MatHasOperation(mat, MATOP_CREATE_SUBMATRICES, has));
11902: }
11903: }
11904: }
11905: PetscFunctionReturn(PETSC_SUCCESS);
11906: }
11908: /*@
11909: MatHasCongruentLayouts - Determines whether the rows and columns layouts of the matrix are congruent
11911: Collective
11913: Input Parameter:
11914: . mat - the matrix
11916: Output Parameter:
11917: . cong - either `PETSC_TRUE` or `PETSC_FALSE`
11919: Level: beginner
11921: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `MatSetSizes()`, `PetscLayout`
11922: @*/
11923: PetscErrorCode MatHasCongruentLayouts(Mat mat, PetscBool *cong)
11924: {
11925: PetscFunctionBegin;
11928: PetscAssertPointer(cong, 2);
11929: if (!mat->rmap || !mat->cmap) {
11930: *cong = mat->rmap == mat->cmap ? PETSC_TRUE : PETSC_FALSE;
11931: PetscFunctionReturn(PETSC_SUCCESS);
11932: }
11933: if (mat->congruentlayouts == PETSC_DECIDE) { /* first time we compare rows and cols layouts */
11934: PetscCall(PetscLayoutSetUp(mat->rmap));
11935: PetscCall(PetscLayoutSetUp(mat->cmap));
11936: PetscCall(PetscLayoutCompare(mat->rmap, mat->cmap, cong));
11937: if (*cong) mat->congruentlayouts = 1;
11938: else mat->congruentlayouts = 0;
11939: } else *cong = mat->congruentlayouts ? PETSC_TRUE : PETSC_FALSE;
11940: PetscFunctionReturn(PETSC_SUCCESS);
11941: }
11943: /*@
11944: MatFlag - set infinity into the local part of the matrix on any subset of MPI processes
11946: Logically Collective
11948: Input Parameters:
11949: + A - the matrix, can be `NULL` but only if on all processes
11950: - flg - indicates if this processes portion of the matrix should be set to infinity
11952: Level: developer
11954: Notes:
11955: This is used to flag a block of solutions that a linear solver failed to compute, as `VecFlag()` does for a single solution.
11957: The state of `A` is increased on all processes, whether or not their portion is flagged, so an outer solver that tracks it detects the failure even when the entries were already infinite.
11959: Only the dense types (`MATSEQDENSE`, `MATMPIDENSE`, and their device variants) currently implement this operation.
11961: .seealso: [](ch_matrices), `Mat`, `VecFlag()`, `MatZeroEntries()`, `MatSetValues()`
11962: @*/
11963: PetscErrorCode MatFlag(Mat A, PetscInt flg)
11964: {
11965: PetscFunctionBegin;
11966: if (!A) PetscFunctionReturn(PETSC_SUCCESS);
11969: MatCheckPreallocated(A, 1);
11970: PetscCall(PetscObjectStateIncrease((PetscObject)A));
11971: if (flg) PetscUseTypeMethod(A, setinf);
11972: PetscFunctionReturn(PETSC_SUCCESS);
11973: }
11975: /*@
11976: MatCreateGraph - create a scalar matrix (that is a matrix with one vertex for each block vertex in the original matrix), for use in graph algorithms
11977: and possibly removes small values from the graph structure.
11979: Collective
11981: Input Parameters:
11982: + A - the matrix
11983: . sym - `PETSC_TRUE` indicates that the graph should be symmetrized
11984: . scale - `PETSC_TRUE` indicates that the graph edge weights should be symmetrically scaled with the diagonal entry
11985: . filter - filter value - < 0: does nothing; == 0: removes only 0.0 entries; otherwise: removes entries with $|entries| \le filter$
11986: . num_idx - size of `index` array
11987: - index - array of block indices to use for graph strength of connection weight
11989: Output Parameter:
11990: . graph - the resulting graph
11992: Level: advanced
11994: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `PCGAMG`
11995: @*/
11996: PetscErrorCode MatCreateGraph(Mat A, PetscBool sym, PetscBool scale, PetscReal filter, PetscInt num_idx, PetscInt index[], Mat *graph)
11997: {
11998: PetscFunctionBegin;
12002: PetscAssertPointer(graph, 7);
12003: PetscCall(PetscLogEventBegin(MAT_CreateGraph, A, 0, 0, 0));
12004: PetscUseTypeMethod(A, creategraph, sym, scale, filter, num_idx, index, graph);
12005: PetscCall(PetscLogEventEnd(MAT_CreateGraph, A, 0, 0, 0));
12006: PetscFunctionReturn(PETSC_SUCCESS);
12007: }
12009: /*@
12010: MatEliminateZeros - eliminate the nondiagonal zero entries in place from the nonzero structure of a sparse `Mat` in place,
12011: meaning the same memory is used for the matrix, and no new memory is allocated.
12013: Collective
12015: Input Parameters:
12016: + A - the matrix
12017: - keep - if for a given row of `A`, the diagonal coefficient is zero, indicates whether it should be left in the structure or eliminated as well
12019: Level: intermediate
12021: Developer Note:
12022: The entries in the sparse matrix data structure are shifted to fill in the unneeded locations in the data. Thus the end
12023: of the arrays in the data structure may be no longer needed to represent the matrix.
12025: .seealso: [](ch_matrices), `Mat`, `MatCreate()`, `MatCreateGraph()`, `MatFilter()`
12026: @*/
12027: PetscErrorCode MatEliminateZeros(Mat A, PetscBool keep)
12028: {
12029: PetscFunctionBegin;
12031: PetscUseTypeMethod(A, eliminatezeros, keep);
12032: PetscFunctionReturn(PETSC_SUCCESS);
12033: }
12035: /*@
12036: MatGetCurrentMemType - Get the memory location of the matrix
12038: Not Collective, but the result will be the same on all MPI processes
12040: Input Parameter:
12041: . A - the matrix whose memory type we are checking
12043: Output Parameter:
12044: . m - the memory type, see `PetscMemType`
12046: Level: intermediate
12048: .seealso: [](ch_matrices), `Mat`, `MatBoundToCPU()`, `PetscMemType`
12049: @*/
12050: PetscErrorCode MatGetCurrentMemType(Mat A, PetscMemType *m)
12051: {
12052: PetscFunctionBegin;
12054: PetscAssertPointer(m, 2);
12055: if (A->ops->getcurrentmemtype) PetscUseTypeMethod(A, getcurrentmemtype, m);
12056: else *m = PETSC_MEMTYPE_HOST;
12057: PetscFunctionReturn(PETSC_SUCCESS);
12058: }